Numerous indications.

Numerous indications. The archaeology of regional hunter-gatherer behaviour in northwest central Queensland, Australia.

Malcolm Ridges.

A thesis submitted for the degree of Doctor of Philosophy of the University of New England.

January 2003.

I certify that the substance of this thesis has not already been submitted for any degree and is not currently being submitted for any other degree or qualification.

Any help received in preparing this thesis, and all sources used, have been acknowledged in the text.

………………………………………………………………………… Malcolm Ridges.

Data Embargo.

Only a small amount of the data analysed in this study was collected through the direct work of the author. Much of the data came from previous archaeological studies, personal records of other archaeologists, or from the Queensland Environmental Protection Agency. Intellectual property of this data remains with its original custodians.

Similarly, the data analysed in this study contained precise coordinates of archaeological locations. In order to protect their cultural heritage as much as possible, the Aboriginal communities in the study region have expressed the desire to restrict access to the data unless it is for bonafide research purposes.

For these reasons, the archaeological data analysed in this project have not been included with the thesis. All the data cited in the text has been archived and is available, but requires written application stating the intended use of the data. All requests will be subject to negotiation with the relative custodians of the data. Requests should be directed to the School of Human and Environmental Studies, University of New England, Australia.

Abstract. Anthropological studies of modern hunter-gatherers have revealed that this mode of human existence involved many complex forms of behaviour. These complexities operated on many levels, and in many accounts were quite regionalised. The archaeological study of hunter-gatherers however has at times struggled to reproduce the complexity of anthropological studies due to the relatively uncomplicated form of what is preserved of hunter-gatherer behaviour in the archaeological record. As has become apparent with the growth of cultural resource management in Australia, the majority of this evidence consists of scatters of stone artefacts, with just the odd rock-art site. The primary focus of this study was to examine how such evidence can contribute to our understanding of regional hunter-gatherer behaviour. This thesis examined this issue through a spatial analysis of the archaeological evidence recovered from a region in northwest central Queensland Australia. Using spatial modelling techniques, it was found that several levels of archaeological variation could be identified in the region. These levels of archaeological variation provided insights into the complexity of behaviour that occurred in the region before it was severely disrupted by European colonisation. Such knowledge about the spatial dynamics of regional huntergatherer behaviour is not only informative about Aboriginal behaviour in the recent past, but also about the processes involved in the behavioural changes evident in Australia’s 50,000 year history. The results of this study demonstrated that much more could be gained from the archaeological evidence of hunter-gatherers if a regional spatial perspective is taken. Such a perspective cannot reproduce the kinds of complexities that have been observed through anthropological studies of hunter-gatherers, but provides complementary evidence about other levels of behavioural complexity that may not be attainable through anthropological research. The result is a more holistic understanding of hunter-gatherer behaviour. Such understanding has the potential to improve the way archaeologists understand the archaeological record of regions, and has the potential to enable more effective of management cultural resources.

Contents.

Table of Contents. TABLE OF CONTENTS................................................................................................................................ I LIST OF FIGURES........................................................................................................................................V LIST OF TABLES........................................................................................................................................ XI ACKNOWLEDGEMENTS......................................................................................................................XIII CHAPTER 1.....................................................................................................................................................1 1.1. THE ARCHAEOLOGY OF REGIONAL BEHAVIOUR IN NW CENTRAL QUEENSLAND. ................................1 1.2. LOCATION OF THE STUDY. ......................................................................................................................4 1.3. ORGANISATION OF THE THESIS...............................................................................................................6 CHAPTER 2.....................................................................................................................................................7 2.1. THE PROBLEM OF INTEGRATING THEORY. ..............................................................................................7 2.2. AN ALTERNATIVE THEORETICAL STRUCTURE. .................................................................................... 13 2.3. THEORISING SCALE. ............................................................................................................................. 20 2.4. IMPLICATIONS FOR ARCHAEOLOGY. .................................................................................................... 25 CHAPTER 3.................................................................................................................................................. 35 3.1. THE NATURAL ENVIRONMENT. ............................................................................................................ 36 3.1.1. Climate......................................................................................................................................... 36 3.1.2. Terrain. ........................................................................................................................................ 38 3.1.3. Hydrology. ................................................................................................................................... 39 3.1.4. Geology........................................................................................................................................ 42 3.1.5. Soils/landsystems/geomorphology.............................................................................................. 45 3.1.6. Vegetation.................................................................................................................................... 47 3.1.7. Fauna........................................................................................................................................... 49 3.2. NW QUEENSLAND ETHNOGRAPHY...................................................................................................... 51 3.2.1. Social groups............................................................................................................................... 51 3.2.2. Trade............................................................................................................................................ 55 3.3. NW QUEENSLAND ARCHAEOLOGY. .................................................................................................... 57 3.3.1. Chronology.................................................................................................................................. 59 3.3.1.1

Lawn Hill.......................................................................................................................................59

3.3.1.2

Selwyn. ..........................................................................................................................................62

3.3.2. Rock-art. ...................................................................................................................................... 64 3.3.2.1

Anthropomorphs- Ross’ study. .....................................................................................................67

3.3.2.2

Variation within the anthropomorphic style. ................................................................................70

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

i.

Contents.

3.3.2.3

Other motif types in the region. .................................................................................................... 72

3.3.2.4

Chemical analysis of the paints. ................................................................................................... 72

3.3.3. Open sites. ....................................................................................................................................75 3.3.3.1

Site definition. ............................................................................................................................... 76

3.3.3.2

Site locations. ................................................................................................................................ 77

3.3.3.3

Environmental Units. .................................................................................................................... 78

3.3.3.4

Site contents. ................................................................................................................................. 80

3.4. SPATIAL PROCESSES OPERATING IN THE REGION. ................................................................................81 CHAPTER 4. .................................................................................................................................................85 4.1. THE METHODOLOGICAL PROBLEM. ......................................................................................................85 4.2. METHODOLOGICAL COMPONENTS. ......................................................................................................90 4.2.1. Approaches to investigating archaeological spatial patterning................................................90 4.2.2. Spatial analysis. ...........................................................................................................................95 4.2.3. Spatial patterning of rock art. .....................................................................................................98 4.2.3.1

Figure distributions. ...................................................................................................................... 98

4.2.3.2

Similarity of rock-art between sites. ........................................................................................... 100

4.2.3.3

Links between sites with common figures.................................................................................. 101

4.2.4. Predictive modelling methodology............................................................................................102 4.2.5. Scales of analysis.......................................................................................................................113 4.3. DATA EXAMINED IN THIS STUDY........................................................................................................115 4.3.1. Archaeological database. ..........................................................................................................116 4.3.1.1

Recorded information. ................................................................................................................ 116

4.3.1.2

Positional accuracy. .................................................................................................................... 121

4.3.1.3

Error checking............................................................................................................................. 123

4.3.2. Non-archaeological data...........................................................................................................123 4.4. METHODOLOGICAL LIMITATIONS. .....................................................................................................133 4.4.1. Variables. ...................................................................................................................................133 4.4.2. Verification.................................................................................................................................135 CHAPTER 5. ...............................................................................................................................................137 5.1. REGIONAL SCALE MODELLING...........................................................................................................137 5.2. DISTRIBUTION OF ARCHAEOLOGICAL LOCATIONS.............................................................................138 5.3. NOMINAL VARIABLES.........................................................................................................................140 5.4. RATIO VARIABLES. .............................................................................................................................153 5.5. MODEL FORMATION. ..........................................................................................................................171 5.6. LOCATION CLASS DISTRIBUTIONS. .....................................................................................................180 5.7. LOCATION CLASS VARIABLES. ...........................................................................................................185 5.8. LOCATION CLASS MODELS. ................................................................................................................201 5.9. REGIONAL LOCATION MODELS SUMMARY. ........................................................................................213

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

ii.

Contents.

CHAPTER 6................................................................................................................................................ 215 6.1. MODELS DESCRIBING THE DISTRIBUTION OF ARCHAEOLOGICAL FEATURES. ................................... 215 6.2. LOCATIONS WITH ROCK-ART. ............................................................................................................ 219 6.3. FIGURE FREQUENCIES. ....................................................................................................................... 221 6.4. FIGURE DISTRIBUTIONS. .................................................................................................................... 231 6.5. FIGURE COMPOSITION AT SITES. ........................................................................................................ 238 6.6. LINKS BETWEEN ART SITES................................................................................................................ 254 6.7. LOCATIONS WITH STONE TOOLS. ....................................................................................................... 268 6.8. STONE RAW MATERIALS. ................................................................................................................... 269 6.9. TOOL TYPES. ...................................................................................................................................... 292 6.10. ARCHAEOLOGICAL FEATURE DISTRIBUTION SUMMARY. ................................................................ 310 CHAPTER 7................................................................................................................................................ 311 7.1. SUB-REGIONAL ANALYSIS. ................................................................................................................ 311 7.2. THE CALTON HILLS SUB-REGION. ..................................................................................................... 314 7.2.1. Calton Hills stone artefact profile............................................................................................ 316 7.2.2. Calton Hills location types........................................................................................................ 322 7.2.3. Calton Hills raw materials. ...................................................................................................... 329 7.2.4. Calton Hills artefact types. ....................................................................................................... 338 7.2.5. Calton Hills summary. .............................................................................................................. 343 7.3. THE SELWYN RANGES SUB-REGION. ................................................................................................. 344 7.3.1. Selwyn stone artefact profile..................................................................................................... 349 7.3.2. Selwyn location types. ............................................................................................................... 351 7.3.3. Selwyn raw materials. ............................................................................................................... 359 7.3.4. Selwyn artefact types................................................................................................................. 365 7.3.5. Selwyn chert............................................................................................................................... 368 7.3.6. Selwyn summary. ....................................................................................................................... 374 7.4. MODELLING AT THE LOCAL LEVEL SUMMARY. ................................................................................. 375 CHAPTER 8................................................................................................................................................ 377 8.1. THE ARCHAEOLOGY OF REGIONAL HUNTER-GATHERER BEHAVIOUR............................................... 377 8.2. LOCATION TYPES. .............................................................................................................................. 378 8.3. ROCK-ART SITES. ............................................................................................................................... 380 8.4. STONE ARTEFACTS............................................................................................................................. 385 8.5. UNDERSTANDING REGIONAL BEHAVIOUR. ........................................................................................ 390 CHAPTER 9................................................................................................................................................ 395 9.1. COMPLEXITY IN REGIONAL HUNTER-GATHERER SYSTEMS. .............................................................. 395 9.2. METHODOLOGICAL IMPLICATIONS.................................................................................................... 401 __________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

iii.

Contents.

9.3. IMPLICATIONS FOR THE MANAGEMENT OF CULTURAL HERITAGE. ....................................................405 9.3.1. Improving the database. ............................................................................................................406 9.3.2. Assessment of archaeological significance...............................................................................408 9.4. SCALE AND INTERPRETING REGIONAL BEHAVIOUR. ..........................................................................410 REFERENCES............................................................................................................................................415 APPENDIX 1. ..............................................................................................................................................431

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

iv.

Figures.

List of Figures. FIGURE 1.1. STUDY REGION– GREY INDICATES AREAS ABOVE 350M IN ELEVATION........................................5 FIGURE 2.1. TRIGGER'S (1989, 20) DIAGRAM. ................................................................................................ 16 FIGURE 2.2. TRIGGER'S ARCHAEOLOGICAL LEVELS, MODIFIED FOR THIS THESIS. ......................................... 17 FIGURE 2.3. MODIFYING THE MODEL TO INCORPORATE SCALE. .................................................................... 19 FIGURE 3.1. MEAN ANNUAL RAINFALL ISOHYETS (50 PERCENTILE) FOR THE STUDY REGION. CROSSED CIRCLES ARE WEATHER STATIONS.

........................................................................................................ 37

FIGURE 3.2. TERRAIN OF THE STUDY REGION................................................................................................. 39 FIGURE 3.3. DRAINAGE BASINS OF EASTERN AUSTRALIA.............................................................................. 40 FIGURE 3.4. DRAINAGE IN THE STUDY REGION MAPPED AT 1:1,000,000 SCALE. BLACK DOTS ARE RECORDED WATERHOLES . BLACK LINE IS THE WATERSHED BETWEEN CARPENTARIA AND LAKE EYRE DRAINAGE SYSTEMS. ................................................................................................................................................ 41

FIGURE 3.5. DISTRIBUTION OF MAJOR ROCK TYPES OCCURRING THE STUDY REGION. .................................. 43 FIGURE 3.6. SIMPLIFIED MAP OF THE SOILS IN THE STUDY REGION................................................................ 46 FIGURE 3.7. LANDSAT CLASSIFICATION. ..................................................................................................... 49 FIGURE 3.8. ROTH'S GROUP BOUNDARIES....................................................................................................... 52 FIGURE 3.9. T INDALE'S GROUP BOUNDARIES.................................................................................................. 54 FIGURE 3.10. TRADE ROUTES AND ITEMS FROM ROTH................................................................................... 56 FIGURE 3.11. LOCATION OF HISCOCK'S STUDY REGION. ................................................................................ 58 FIGURE 3.12. LOCATION OF ART SITES STUDIED IN PREVIOUS RESEARCH. .................................................... 67 FIGURE 3.13. SITES STUDIED BY ROSS IN THE CONTEXT OF ROTH'S (1897), SOLID LINE AND T INDALE'S (1974), DASHED LINE TRIBAL BOUNDARIES........................................................................................... 68 FIGURE 3.14. PCA RESULTS OF PIGMENT ANALYSIS. ..................................................................................... 73 FIGURE 4.1. EXAMPLE OF A ROCK ART DISTRIBUTION SUMMARY PLOT......................................................... 99 FIGURE 4.2. COMPARISON BETWEEN LINEAR AND LOGISTIC REGRESSION CURVES..................................... 105 FIGURE 4.3. DENSITY PLOTS FOR ARCHAEOLOGICAL LOCATIONS AND RANDOM POINTS............................ 106 FIGURE 4.4. COMPARISON OF DIFFERENT REGRESSION MODELS.................................................................. 107 FIGURE 4.5. COMPARISON OF SPLINE AND LOESS NON-LINEAR FUNCTIONS. ............................................... 108 FIGURE 4.6. DETERMINING THRESHOLD VALUE FOR REGRESSION MODELS. ............................................... 112 FIGURE 4.7. LOCATION OF SUB-REGIONS EXAMINED IN THIS STUDY. GREY INDICATES AREAS OVER 350 METRES IN ELEVATION. ........................................................................................................................ 115

FIGURE 4.8. SITE DATABASE DATA ENTRY FORM. ........................................................................................ 117 FIGURE 4.9. STONE TOOL DATABASE DATA ENTRY FORM. ........................................................................... 118 FIGURE 4.10. KIPPEN TRANSECT DATA ENTRY FORM. .................................................................................. 120 FIGURE 4.11. ART SITE DATABASE DATA ENTRY FORM................................................................................ 121 FIGURE 4.12. ELEVATION VARIABLE. ........................................................................................................... 127 FIGURE 4.13. SLOPE VARIABLE..................................................................................................................... 127 __________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

v.

Figures.

FIGURE 4.14. ASPECT VARIABLE. ..................................................................................................................128 FIGURE 4.15. WETNESS VARIABLE. ...............................................................................................................128 FIGURE 4.16. STREAMLINE PROXIMITY VARIABLE........................................................................................129 FIGURE 4.17. WEIGHTED STREAMLINE PROXIMITY VARIABLE. ....................................................................129 FIGURE 4.18. PROXIMITY TO WATERHOLES VARIABLE.................................................................................130 FIGURE 4.19. COST DISTANCES FROM CHALCEDONY AND CHERT. ...............................................................131 FIGURE 4.20. COST DISTANCES FROM QUARTZITE AND QUARTZ. .................................................................132 FIGURE 4.21. COST DISTANCES FROM SILCRETE AND VOLCANIC RAW MATERIAL TYPES. ...........................132 FIGURE 5.1. DISTRIBUTION OF ALL ARCHAEOLOGICAL FIND-SPOTS.............................................................139 FIGURE 5.2. COMPARISON OF RANDOM TO ARCHAEOLOGICAL FREQUENCIES IN EACH SOIL TYPE. .............147 FIGURE 5.3. COMPARISON OF RANDOM TO ARCHAEOLOGICAL FREQUENCIES IN EACH GEOLOGICAL UNIT. 147 FIGURE 5.4. COMPARISON OF RANDOM TO ARCHAEOLOGICAL FREQUENCIES IN EACH LANDSAT CLASS....148 FIGURE 5.5. RATIO OF ARCHAEOLOGICAL TO RANDOM LOCATION DENSITIES FOR SOIL TYPES...................150 FIGURE 5.6. RATIO OF ARCHAEOLOGICAL TO RANDOM LOCATION DENSITIES FOR GEOLOGICAL UNITS . ....151 FIGURE 5.7. RATIO OF ARCHAEOLOGICAL TO RANDOM LOCATION DENSITIES FOR LANDSAT CLASSES. .....152 FIGURE 5.8. HISTOGRAM OF ELEVATION FOR ALL LOCATIONS. ....................................................................155 FIGURE 5.9. REGRESSION CURVE FOR ELEVATION. .......................................................................................155 FIGURE 5.10. HISTOGRAM OF ASPECT FOR ALL LOCATIONS. ........................................................................157 FIGURE 5.11. REGRESSION CURVE FOR ASPECT. ...........................................................................................157 FIGURE 5.12. HISTOGRAM OF SLOPE FOR ALL LOCATIONS. ..........................................................................159 FIGURE 5.13. REGRESSION CURVE FOR SLOPE. .............................................................................................159 FIGURE 5.14. HISTOGRAM OF STREAM PROXIMITY FOR ALL LOCATIONS. ....................................................160 FIGURE 5.15. REGRESSION CURVE FOR STREAMLINE PROXIMITY.................................................................160 FIGURE 5.16. HISTOGRAM OF WATERHOLE PROXIMITY FOR ALL LOCATIONS. .............................................162 FIGURE 5.17. REGRESSION CURVE FOR PROXIMITY TO WATERHOLES. .........................................................162 FIGURE 5.18. HISTOGRAM OF STREAM PROXIMITY WEIGHTED BY STREAM ORDER, FOR ALL LOCATIONS. .166 FIGURE 5.19. REGRESSION CURVE FOR PROXIMITY TO STREAMLINES WEIGHTED BY STREAM ORDER. .......166 FIGURE 5.20. OCCURRENCE OF ARCHAEOLOGICAL FEATURES CAUSING THE SPIKE IN WEIGHTED STREAMLINE PROXIMITY.......................................................................................................................167

FIGURE 5.21. HISTOGRAM OF WETNESS FOR ALL LOCATIONS. .....................................................................170 FIGURE 5.22. REGRESSION CURVE FOR WETNESS. ........................................................................................170 FIGURE 5.23. PROBABILITY MODEL FOR ALL LOCATIONS.............................................................................172 FIGURE 5.24. PROBABILITY MODEL FOR ALL LOCATIONS INCORPORATING SOILS . ......................................173 FIGURE 5.25. PROBABILITY MODEL FOR ALL LOCATIONS INCORPORATING GEOLOGY. ...............................175 FIGURE 5.26. PROBABILITY MODEL FOR ALL LOCATIONS INCORPORATING LANDSAT CLASSIFICATION..177 FIGURE 5.27. DIFFERENCE INTRODUCED BY ADDING LANDSAT DATA TO MODEL....................................178 FIGURE 5.28. PROBABILITY MODEL INCORPORATING ALL NOMINAL VARIABLES. .......................................179 FIGURE 5.29. DISTRIBUTION OF LOCATION CLASSES. ...................................................................................182 FIGURE 5.30. RATIO TO RANDOM FOR LANDSAT CATEGORIES..................................................................185 __________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

vi.

Figures.

FIGURE 5.31. DISTRIBUTION OF LANDSAT CLASSES 5(WOODLAND) AND 10 (LOW WOODLAND). ........... 186 FIGURE 5.32. RATIO TO RANDOM FOR ROCK TYPES. .................................................................................... 187 FIGURE 5.33. HISTOGRAM OF ASPECT FOR LOCATION CLASSES................................................................... 189 FIGURE 5.34. SPLINE MODELS OF ASPECT FOR EACH LOCATION CLASS. ...................................................... 189 FIGURE 5.35. HISTOGRAM OF WATERHOLE PROXIMITY FOR LOCATION CLASSES........................................ 192 FIGURE 5.36. SPLINE MODELS OF WATERHOLE PROXIMITY FOR EACH LOCATION TYPE.............................. 192 FIGURE 5.37. HISTOGRAMS OF COST DISTANCE FROM STREAMLINES FOR EACH LOCATION TYPE. ............. 193 FIGURE 5.38. LOESS MODELS OF COST DISTANCE FROM STREAMLINES FOR EACH LOCATION TYPE. .......... 193 FIGURE 5.39. HISTOGRAM OF PROXIMITY TO STREAMLINES WEIGHTED BY STREAM ORDER FOR LOCATION CLASSES................................................................................................................................................ 194

FIGURE 5.40. SPLINE MODELS OF PROXIMITY TO STREAMLINES WEIGHTED BY STREAM ORDER FOR LOCATION CLASSES. ............................................................................................................................. 194

FIGURE 5.41. HISTOGRAM OF ELEVATION FOR LOCATION CLASSES. ........................................................... 197 FIGURE 5.42. SPLINE MODELS OF ELEVATION FOR EACH LOCATION CLASS................................................. 197 FIGURE 5.43. HISTOGRAM OF SLOPE FOR LOCATION CLASSES. .................................................................... 198 FIGURE 5.44. SPLINE MODELS OF SLOPE FOR EACH LOCATION TYPE. .......................................................... 198 FIGURE 5.45. HISTOGRAM OF WETNESS FOR LOCATION CLASSES................................................................ 200 FIGURE 5.46. SPLINE MODELS OF WETNESS FOR EACH LOCATION TYPE. ..................................................... 200 FIGURE 5.47. PROBABILITY MODEL FOR INDIVIDUAL FINDS. ....................................................................... 202 FIGURE 5.48. PROBABILITY MODEL FOR OPEN SCATTERS. ........................................................................... 203 FIGURE 5.49. PROBABILITY MODEL FOR OPEN SITES.................................................................................... 204 FIGURE 5.50. RGB COMPOSITE FOR OPEN LOCATION TYPES........................................................................ 207 FIGURE 5.51. PROBABILITY MODEL FOR ART SITES...................................................................................... 210 FIGURE 5.52. PROBABILITY MODEL FOR QUARRIES...................................................................................... 211 FIGURE 6.1. DISTRIBUTION OF LOCATION FEATURE CLASSES. ..................................................................... 218 FIGURE 6.2. COMPARISON OF FREQUENCIES FOR MOTIFS AND ADES. ........................................................ 225 FIGURE 6.3. MOTIF NUMBER X FIGURE NUMBER. ......................................................................................... 226 FIGURE 6.4. NUMBER OF ANTHROPOMORPH DESIGN ELEMENTS X NUMBER OF ANTHROPOMORPHS........... 227 FIGURE 6.5. BOX-PLOT FOR THE NUMBER OF MOTIFS AT EACH ART SITE TYPE. .......................................... 228 FIGURE 6.6. BOX-PLOT FOR THE NUMBER OF ADES AT EACH ART SITE TYPE. ............................................ 230 FIGURE 6.7. MOTIF DISTRIBUTIONS (FOR MOTIFS WITH SITE FREQUENCIES > 5). ........................................ 232 FIGURE 6.8. GEOGRAPHIC CENTROIDS OF SELECTED MOTIF GROUPS........................................................... 235 FIGURE 6.9. DISTRIBUTION OF ANTHROPOMORPH ATTRIBUTES ................................................................... 236 FIGURE 6.10. PCA RESULTS FOR THE MOTIF DATA SET................................................................................ 239 FIGURE 6.11. PCA RESULTS FOR THE ADE DATA SET. ................................................................................ 240 FIGURE 6.12. COMPONENT 1 (SITE SCORES) VERSUS MOTIF NUMBER.......................................................... 241 FIGURE 6.13. COMPONENT 1 (SITE SCORES) VERSUS ADE NUMBER............................................................ 241 FIGURE 6.14. GROUPS DEFINED FOR MOTIF PCA COMPONENTS 2 AND 3. ................................................... 242 FIGURE 6.15. ADE PCA GROUPS FROM CLUSTER ANALYSIS. ...................................................................... 243 __________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

vii.

Figures.

FIGURE 6.16. PROPORTION OF DEPICTION METHOD WITHIN EACH PCA GROUP FOR MOTIF DATA SET........245 FIGURE 6.17. MOTIF PCA RESULTS DEPICTED BY SITE TYPE........................................................................248 FIGURE 6.18. ADE PCA RESULTS DEPICTED BY SITE TYPE. .........................................................................249 FIGURE 6.19. REGIONAL DISTRIBUTION OF MOTIF (LEFT) AND ADE (RIGHT) GROUPS. ...............................252 FIGURE 6.20. MEAN CO-ORDINATE MAP FOR MOTIF (LEFT) AND ADE (RIGHT) PCA GROUPS. ...................253 FIGURE 6.21. LINK MAP FOR MOTIFS. ............................................................................................................258 FIGURE 6.22. LINK MAP FOR ADES...............................................................................................................259 FIGURE 6.23. COMPARISON OF THE BEARINGS FOR MOTIF AND ADE LINKS................................................260 FIGURE 6.24. MOTIFS VERSUS LINKS FOR MOTIF SITES.................................................................................261 FIGURE 6.25. ANTHROPOMORPHS VERSUS LINKS FOR ADE SITES. ..............................................................261 FIGURE 6.26. BOX PLOTS FOR LINK FREQUENCY AT SITE TYPES...................................................................263 FIGURE 6.27. NUMBER OF LINKS VERSUS AVERAGE LINK DISTANCES FOR MOTIF SITES..............................265 FIGURE 6.28. EDGE EFFECT ON AVERAGE LINK DISTANCE FOR MOTIFS. ......................................................265 FIGURE 6.29. NUMBER OF LINKS VERSUS AVERAGE LINK DISTANCE FOR ANTHROPOMORPH SITES. ...........266 FIGURE 6.30. AVERAGE LINK DISTANCE AT SITE TYPES. ..............................................................................267 FIGURE 6.31. STONE ARTEFACT MODEL USED FOR DEVELOPING RAW MATERIAL MODELS. ........................273 FIGURE 6.32. DISTRIBUTION AND MODEL FOR LOCATIONS CONTAINING CHALCEDONY STONE ARTEFACTS . ...............................................................................................................................................................275 FIGURE 6.33. DISTRIBUTION AND MODEL FOR LOCATIONS CONTAINING CHERT STONE ARTEFACTS...........276 FIGURE 6.34. DISTRIBUTION AND MODEL FOR LOCATIONS CONTAINING QUARTZ STONE ARTEFACTS. .......277 FIGURE 6.35. DISTRIBUTION AND MODEL FOR LOCATIONS CONTAINING QUARTZITE STONE ARTEFACTS...278 FIGURE 6.36. DISTRIBUTION AND MODEL FOR LOCATIONS CONTAINING SILCRETE STONE ARTEFACTS. .....279 FIGURE 6.37. DISTRIBUTION AND MODEL FOR LOCATIONS CONTAINING VOLCANIC STONE ARTEFACTS. ...280 FIGURE 6.38. RAW MATERIAL DIVERSITY. ....................................................................................................282 FIGURE 6.39. RBG COMPOSITE FOR VOLCANIC, QUARTZ AND QUARTZITE. .................................................285 FIGURE 6.40. RBG COMPOSITE FOR CHERT, CHALCEDONY AND SILCRETE. .................................................287 FIGURE 6.41. OCCURRENCE OF RAW MATERIAL TYPES WITH DISTANCE FROM GEOLOGICAL ORIGIN..........291 FIGURE 6.42. SITES MODEL USED IN FORMING TOOL TYPE MODELS. ............................................................294 FIGURE 6.43. MODELLED OCCURRENCES OF TULAS. ....................................................................................296 FIGURE 6.44. MODELLED OCCURRENCE OF RETOUCHED ARTEFACTS. .........................................................297 FIGURE 6.45. MODELLED OCCURRENCE OF GRINDSTONES. ..........................................................................298 FIGURE 6.46. MODELLED OCCURRENCE OF BACKED BLADES.......................................................................299 FIGURE 6.47. MODELLED OCCURRENCE OF CORES. ......................................................................................300 FIGURE 6.48. MODELLED OCCURRENCE OF AXES. ........................................................................................301 FIGURE 6.49. TOOL TYPE DIVERSITY.............................................................................................................303 FIGURE 6.50. RGB COMPOSITE FOR TULAS, BACKED BLADES AND CORES. .................................................305 FIGURE 6.51. RGB COMPOSITE FOR AXES, RETOUCHED, GRINDSTONES. .....................................................307 FIGURE 7.1. ARCHAEOLOGICAL LOCATIONS IN CALTON H ILLS SUB-REGION. .............................................315 FIGURE 7.2. GEOLOGY OF THE CALTON HILLS SUB-REGION. .......................................................................317 __________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

viii.

Figures.

FIGURE 7.3. LOCATION TYPE MODELS FOR CALTON HILLS SUB-REGION..................................................... 322 FIGURE 7.4. RGB COMPOSITE MAP FOR THE CALTON HILLS SUB-REGION LOCATION TYPE MODELS. ........ 325 FIGURE 7.5. DIFFERENCES BETWEEN REGIONAL AND SUB-REGIONAL MODELS FOR CALTON HILLS SUBREGION.................................................................................................................................................. 327

FIGURE 7.6. COMPARISON OF REGIONAL AND SUB-REGIONAL VOLCANIC MODELS FOR CALTON HILLS SUBREGION.................................................................................................................................................. 330

FIGURE 7.7. DIFFERENCES BETWEEN REGIONAL AND SUB-REGIONAL MODELS FOR CHERT IN THE CALTON HILLS SUB-REGION. .............................................................................................................................. 330 FIGURE 7.8. RGB COMPOSITE FOR VOLCANIC, QUARTZ AND QUARTZITE ARTEFACTS IN THE CALTON HILLS SUB-REGION.......................................................................................................................................... 333

FIGURE 7.9. PREDICTED ARTEFACT FREQUENCIES, CALTON HILLS SUB-REGION........................................ 335 FIGURE 7.10. INFERRED ZONES OF BEHAVIOUR IN THE CALTON HILLS SUB-REGION.................................. 337 FIGURE 7.11. PREDICTED ARTEFACT TYPE OCCURRENCES WITHIN THE CALTON HILLS SUB-REGION. ....... 339 FIGURE 7.12. RGB COMPOSITE FOR ARTEFACT TYPES IN THE CALTON HILLS SUB-REGION. ...................... 341 FIGURE 7.13. DISTRIBUTION OF LOCATION TYPES FOR SELWYN SUB-REGION. ........................................... 345 FIGURE 7.14. GEOLOGY OF THE SELWYN SUB-REGION. ............................................................................... 347 FIGURE 7.15. MODEL OF THE DISTRIBUTION OF INDIVIDUAL FINDS IN THE SELWYN SUB-REGION. ............ 352 FIGURE 7.16. RGB COMPOSITE FOR SELWYM SUB-REGION LOCATION TYPE

MODELS . .............................. 353

FIGURE 7.17. LOCATION TYPE MODEL DIFFERENCES FOR SELWYN SUB-REGION. ....................................... 353 FIGURE 7.18. OPEN SITE PROBABILITY IN RELATION TO DISTANCE FROM ART SITES IN THE SELWYN SUBREGION.................................................................................................................................................. 358

FIGURE 7.19. OPEN SITE PROBABILITY IN RELATION TO PROXIMITY TO ART SITES FOR CALTON HILLS SUBREGION.................................................................................................................................................. 359

FIGURE 7.20. PREDICTED DISTRIBUTIONS FOR CHERT AND QUARTZ IN THE SELWYN SUB-REGION. ........... 360 FIGURE 7.21. PREDICTED OCCURRENCES FOR SILCRETE AND CHALCEDONY IN THE SELWYN SUB-REGION. .............................................................................................................................................................. 361 FIGURE 7.22. PREDICTED OCCURRENCES FOR QUARTZITE AND VOLCANIC ARTEFACTS IN THE SELWYN SUBREGION.................................................................................................................................................. 362

FIGURE 7.23. RAW MATERIAL DIVERSITY IN THE SELWYN SUB-REGION. .................................................... 363 FIGURE 7.24. PREDICTED OCCURRENCES OF RETOUCHED ARTEFACTS AND CORES IN THE SELWYN SUBREGION.................................................................................................................................................. 365

FIGURE 7.25. PREDICTED OCCURRENCES OF BACKED BLADES AND TULAS IN THE SELWYN SUB-REGION.. 366 FIGURE 7.26. PREDICTED OCCURRENCES OF AXES AND GRINDSTONES IN THE SELWYN SUB-REGION. ....... 367 FIGURE 7.27. PREDICTED CHERT FREQUENCIES WITHIN THE SELWYN SUB-REGION. .................................. 369 FIGURE 7.28. PREDICTED CHERT LENGTHS WITHIN THE SELWYN SUB-REGION........................................... 371 FIGURE 8.1. IMPORTANT FEATURES OF THE ARCHAEOLOGY OF NW CENTRAL QUEENSLAND.................... 393

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

ix.

Tables.

List of Tables. TABLE 3.1. SOIL UNITS IN THE STUDY REGION. .............................................................................................. 47 TABLE 3.2. LANDSAT CLASSES AND VEGETATION UNITS............................................................................ 48 TABLE 4.1. APPROACHES TO EXAMINING ARCHAEOLOGICAL SPATIAL PATTERNS. ....................................... 91 TABLE 4.2. EXAMPLE OF AN INPUT TABLE.................................................................................................... 110 TABLE 4.3. EXAMPLE OF TABLE USED FOR PREDICTION............................................................................... 110 TABLE 4.4. EXAMPLE OF A 2X2 CHI SQUARED TEST FOR A MODEL OUTCOME. ............................................ 111 TABLE 4.5. SUMMARY OF CLASSIFICATORY LEVELS USED IN THIS STUDY. ................................................. 113 TABLE 4.6. SOURCES OF STONE ARTEFACT DATA......................................................................................... 119 TABLE 4.7. DATASETS OBTAINED FOR THE STUDY....................................................................................... 124 TABLE 4.8. CORRELATION MATRIX BETWEEN RATIO VARIABLES................................................................ 134 TABLE 5.1. ARCHAEOLOGICAL LOCATION AND RANDOM POINT SUMMARY BY SOIL TYPES. ...................... 141 TABLE 5.2. ARCHAEOLOGICAL LOCATION AND RANDOM POINT SUMMARY BY ROCK TYPE. ...................... 143 TABLE 5.3. ARCHAEOLOGICAL LOCATION AND RANDOM POINT SUMMARY BY LANDSAT CLASSES. .......... 145 TABLE 5.4. SIZES OF MAPPING UNITS IN NOMINAL VARIABLES.................................................................... 149 TABLE 5.5. PROPORTION OF WATERHOLES CONTAINING ARCHAEOLOGICAL FEATURES WITHIN SPECIFIED RADII..................................................................................................................................................... 165

TABLE 5.6. SUMMARY OF LOCATION CLASSES. ........................................................................................... 181 TABLE 6.1. SUMMARY OF LOCATION FEATURES.......................................................................................... 216 TABLE 6.2. MOTIF SITE FREQUENCIES (NUMBER OF SITES = 103)................................................................ 223 TABLE 6.3. ANTHROPOMORPH DESIGN ELEMENT FREQUENCIES (NUMBER OF SITES = 61). ........................ 224 TABLE 6.5. MOTIF GROUPS. .......................................................................................................................... 234 TABLE 6.6. SIGNIFICANT VARIABLES FOR MOTIF PCA, COMPONENTS 2 AND 3........................................... 244 TABLE 6.7. SIGNIFICANT VARIABLES FOR ADE PCA, COMPONENTS 2 AND 3............................................. 246 TABLE 6.8. MOTIF GROUPS BY SITE TYPES. .................................................................................................. 250 TABLE 6.9. ADE GROUPS BY SITE TYPE........................................................................................................ 250 TABLE 6.10. NNA RESULTS FOR PCA GROUPS. ........................................................................................... 253 TABLE 6.12. FREQUENCY OF GEOLOGICAL RAW MATERIALS IN ALL LOCATIONS........................................ 270 TABLE 6.13. FREQUENCY OF RAW MATERIALS QUARRIED........................................................................... 271 TABLE 6.14. SUMMARY OF RAW MATERIAL DISTRIBUTIONS........................................................................ 281 TABLE 6.15. AREAS OF RAW MATERIAL DIVERSITY. .................................................................................... 283 TABLE 6.16. OCCURRENCE OF TOOL TYPES IN SITES. ................................................................................... 293 TABLE 6.17. RAW MATERIAL TYPES USED TO MANUFACTURE TOOL TYPES. ............................................... 295 TABLE 6.18. SUMMARY OF TOOL TYPE MODELS........................................................................................... 302 TABLE 7.1. LOCATION TYPE FREQUENCIES FOR CALTON HILLS SUB-REGION. ............................................ 316 TABLE 7.2. RAW MATERIAL OCCURRENCE IN CALTON HILLS SUB-REGION. ............................................... 319 TABLE 7.3. QUARRY TYPE FREQUENCIES FOR CALTON HILLS SUB-REGION................................................ 320 __________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

xi.

Tables.

TABLE 7.4. STONE ARTEFACT TYPE FREQUENCIES IN CALTON HILLS SUB-REGION. ....................................321 TABLE 7.5. SITE CLASS FREQUENCIES FOR SELWYN SUB-REGION. ...............................................................345 TABLE 7.6. RAW MATERIAL OCCURRENCE IN SELWYN SUB-REGION. ..........................................................349 TABLE 7.7. QUARRY TYPE FREQUENCIES FOR SELWYN SUB-REGION...........................................................350 TABLE 7.8. STONE ARTEFACT TYPE FREQUENCIES IN SELWYN SUB-REGION. ..............................................350

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

xii.

Acknowledgements.

Acknowledgements. Professor Iain Davidson, my principal supervisor throughout my candidature, is credited with making this study possible for numerous reasons. It was he who, in 1992, invited me to undertake fieldwork with his team in the Selwyn Ranges. This field trip was not only significant for being the last of the ‘big’ field trips to Selwyn for Iain, but it was also my very first experience of archaeological fieldwork. Amongst many firsts on that trip (first excavation, first site discovery, first camping trip!), I also got to experience first hand what remains, even now, an awe inspiring place. Through subsequent trips to the region with Iain, and involvement with him on other projects, I have been lucky enough to have enjoyed numerous exchange of ideas and discussions. Of all things during my candidature, these are what I will cherish the most. As I struggled through the last throws of writing it was comforting, if somewhat unnerving, that I could return to the comments Iain wrote six months earlier and to my surprise not only understand them, but actually agree with them. Dr. Peter Grave was my secondary supervisor for much of my candidature. I would like to thank him especially for being the reliable source of creative ideas that he has proven to be time and again. Along with holding my hand through a trying period of PIXE/PIGME analysis, which didn’t make it into the thesis or print (yet!), PG has never failed in helping to reshape my thinking over a cup of coffee. As it had been with Iain, these exchanges have proven to be some of the most enjoyable and stimulating times of my candidature. This study was also made possible through the generous support and encouragement given to me by members of the Aboriginal communities in northwest central Queensland. It has been a pleasure and an honour to get to know just a little bit more of their cultural heritage. Among the many people from the communities in the region who have assisted with this study, I would especially like to thank Tom Sullivan, Bill and Dorrie Prowse, and Ken Isaacson. You have all contributed to some very enjoyable times in the bush and have opened my eyes to aspects of the region and your heritage, which would never have been realised without your involvement. I am also indebted to the generosity of several people who were willing for me to use and reanalyse their data. Iain Davidson held a wealth of material he had collected over the __________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

xiii.

Acknowledgements.

last decade. After gaining many grey hairs sorting it out, his data formed the basis of what is probably one of the more comprehensive regional datasets available on Aboriginal archaeology in Australia. Deriving from Iain’s research, the studies by June Ross, Rosalind James’ study of AG1 and ULO3, Tom Drury’s study, and two studies undertaken by Ken Kippen, represented the bulk of the data examined in this thesis. If any of this thesis is to prove useful to other archaeologists in the future, then that can be largely attributed to the excellent foundation the work of these people provided for this study. I would like to extend an extra special thanks to June and Ken in particular. Your generosity with both data and free reign over your work is a very kind and greatly appreciated act, to which I am especially grateful. Leila McAdam is credited with being a wonderful companion and great support in the field. Leila accompanied me to Mt Isa when I was looking for ochre sources. We found some, and in between got to see a lot of wonderful country. Your company during that time help keeps it memorable. These people, plus all the members of the School of Human and Environmental Studies at the University of New England, provided a reassuring network who confirmed that I would, indeed, finish some day. Thankfully, they were proved right. Funding for various aspects of this project have come from the University of New England, the Australian Aboriginal and Torres Straight Island Commission, the Australian Institute of Nuclear Science and Engineering and the Australian Research Council. I would like to thank Graham Bailey of the Australian Nuclear Science and Technology Organisation for spending the time with me to teach me the intricacies of PIXE/PIGME, and fielding happily my many questions about the software. Malcolm Connolly at the Queensland EPA assisted with gaining access to the site records of the region. It was also nice to discuss aspects of my work with someone who shared an interest in GIS and archaeological site modelling. The environmental data used in this project came from several government sources who were generous enough to lend me several data sets free of charge. In this respect, I would like to thank the Department of Natural Resources in Queensland, and Geoscience __________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

xiv.

Acknowledgements.

Australia (formally the Australian Geological Survey Organisation (AGSO) and the Australian Land Information Group (AUSLIG)). For assistance with GIS and computing matters, I would like to thank Alan Jones for helping me through the steep learning curve that is Arc/Info and Erdas Imagine. Similarly, I would like to thank Mike Roach for his help with technical issues and being my buddy in the SDC. Simon Ferrier listened patiently on several occasions to my confused ramblings about issues with generalised additive modelling, and helped clarify them form me. I would like to thank Bob Pressey, for his generosity with my time away from work. My work with Bob has expanded my world to a completely different discipline. Indirectly, this helped clarify my thinking immensely by demonstrating that many of the frustrating problems encountered in archaeology are not unique to the discipline. Most of all I would like to thank my wife, Bec, for her patience on what became a long and tedious journey. Through fieldwork, house moves, pregnancy, child rearing, my tantrums, and my procrastination, Bec has always been just wonderful. I do not think there is another person who would have coped with what she has, but that is why I cherish so deeply her companionship, and FINALLY look forward to enjoying it free of this burden.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

xv.

The Old Grey Donkey, Eeyore, stood by himself in a thistly corner of the Forest … and thought about things. Sometimes he thought sadly to himself, ‘Why?’ and sometimes he thought, ‘Wherefore?’ and sometimes he thought, ‘Inasmuch as which?’ – and sometimes he didn’t know quite what he was thinking about.

A. A. Milne. 1926. Winnie-the-pooh. p39-40

Chapter 1.

The archaeology of regional Aboriginal behaviour. We found here numerous indications of blacks having been here, but we saw nothing of them. It seems remarkable that their tracks are so plentiful…  William John Wills. Journal entry, January 12th 1861.

1.1. The archaeology of regional behaviour in NW Central Queensland. On their fateful journey across the Australian continent, the explorers Robert O'Hara Burke and William John Wills passed right through the middle of the study region. Yet, despite encountering many indications of Aboriginal behaviour, they saw little of them. In many respects, that experience captures the dilemma of this study. Despite the ample amount archaeological evidence recorded in NW central Queensland, a thorough understanding of the dynamics of past Aboriginal behaviour remains tantalisingly distant. The main aim of this study was to describe the spatial variation in the material culture of Aboriginal people in northwest central Queensland, and to explore how this variation __________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

1.

Chapter 1.

relates to the dynamic of regional hunter-gatherer behaviour. Like some other regions in Australia, northwest central Queensland now boasts a significant body of archaeological data that has accumulated from both research and cultural heritage assessment fieldwork. However, to date, this data has received only limited overviews in excavation reports (Davidson, Sutton, and Gale 1993) and regional context descriptions in archaeological consultancy reports (Davidson and Fife 1994, Davies, Grant, and Bowen 1996, Lance 1994). Consequently, the archaeological variability in the region is understood in only rudimentary terms, despite the research potential that has been noted for the region (Hiscock 1988b, 68). A problem therefore arises since the majority of archaeological work conducted in the region is now in the form of cultural heritage assessments, which continue to be undertaken with only a elementary understanding of regional context. These studies routinely outline the types and frequencies of the most common archaeological finds in the region, the antiquity of occupation, and some aspects of Aboriginal behaviour, such as trade, gleaned from ethnographic reports. Missing is the context of what people were doing in different parts of the region, how stone raw material was distributed, where particular artefact types were made and what these mean about the dynamics of subsistence in the region. Similarly the dynamics of regional behaviour are reflected at a social level also, since the region contains a large corpus of rock-art that has been shown to involve aspects of group identity and regionalisation (Ross 1997). Yet at the same time, it maintained links to central Australia, south Australia and northeast Queensland through the depiction of common motif types (David and Chant 1995, Franklin 1996, Rosenfeld, Horton, and Winter 1981). This study therefore aimed to build on this foundation by developing a regional context using the wealth of archaeological data that is now available. Undertaking such a task however, also meant addressing issues about how the dynamics of regional huntergatherer behaviour can be gleaned from archaeological evidence. For this reason, this study also examined the theory about how different perspectives of regional huntergatherer behaviour can be used to gain a more holistic understanding of the dynamics operating in a region. These dynamics operated on many levels, both in terms of the levels of archaeological finds and the way they are analysed, but also the many levels of __________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

2.

Chapter 1.

variation in the form of Aboriginal behaviour. Crucial to understanding these levels is the influence of scale. Scale forms the main theoretical concern of the study. In particular, the focus is upon how the scale of archaeological analysis shapes our understanding of the past. Anthropological studies, such as that recently conducted to the north of the study region by Pickering (1994), have demonstrated the complexity with which social levels permeate many aspects of Aboriginal behaviour. Pickering’s study highlighted the importance of being conversant with scale, since he found that these levels were closely intertwined with levels of the physical environment and levels of subsistence. However, how these may be reflected archaeologically remains poorly understood. For this reason, particularly with the most common form of archaeological evidence—open lithic scatters—it is difficult to discern anything other than subsistence behaviour. Nonsubsistence hunter-gatherer behaviour does preserve archaeologically in the form of rock-art, but how the complexity of form and distribution of rock pictures relate to the pattern of regional subsistence has proven a difficult problem for archaeologists to understand. Yet, these two spheres of archaeological evidence are almost all that remains, in a material sense, of hunter-gatherer behaviour at the regional level. Regional analysis is also important in the Australian archaeological context. Over the last twenty years, Australian archaeology has come to focus on the origins of complex regional behaviour. The origin of complex regional behaviour most likely stem from the changing environmental conditions during the Holocene (Lourandos and Ross 1994). However, they appear to have been driven by a multitude of factors that manifested in different ways archaeologically (Bird and Frankel 1991). On this issue, the position adopted in this study was that understanding the processes involved in the emergence of complex regional behaviour first requires a good understanding of their archaeological manifestation. If these are understood, then archaeological evidence of greater time depth may be able to be interpreted more meaningfully in terms of what characterises regional complexity. This study therefore stands in contrast to the suggestions of Gosden and Head (1994, 113), who argued that the most important aspect to understanding social complexity is to give it greater time depth. In contrast, this study focuses purely on the spatial dimension of behavioural complexity as a means from which chronological variation can be interpreted more productively.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

3.

Chapter 1.

Methodologically, this means examining regional spatial variability in much more detail than has been the case in the past. Fortunately, developments in the management and manipulation of spatial data have been greatly facilitated by the ready availability of spatial information and computer software with which to analyse them. This study makes extensive use of these tools in order to explore the spatial dynamics of the archaeological data collected within the region. These analyses were conducted at multiple scales and on a variety of archaeological evidence in order to construct a regional perspective developed from a range of perspectives. In doing so, some of the dynamics of the region became more easily discerned.

1.2. Location of the study. For the purpose of this study, the area north of Boulia, south of Gunpowder, East of Urandangi and west of Mckinlay defines northwest central Queensland. Figure 1.1 shows an outline of the study region in relation to these places. The study region is 48,000 square kilometres in area, and encompasses the two main (present day) regional centres of Mt Isa and Cloncurry. The shape of the study region takes its form because of the nature of the environmental data available for this part of Queensland. For the most part the study region is confined to topographic map sheet boundaries, for which environmental data sets have been compiled in a geographic information system (GIS). However, in two areas (near Gunpowder Creek in the north and the Hamilton River in the south) the study area extends out to incorporate other archaeological data sets that were available. Less environmental data was available for these extended areas, but they have been included because of the archaeological datasets available for them. It was not possible to extend the study region out to form a rectangle around these areas because of the lack of digital GIS data available for the wider area and the time it would take to generate it manually.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

4.

Chapter 1.

Figure 1.1. Study region– grey indicates areas above 350m in elevation.

The region has a semi-arid climate, low population density, and no tillage-based farming. Archaeological preservation through much of the region is therefore very good. The main threat to cultural heritage has been mining, which occurs in many parts of the region, but in individually small areas. The majority of archaeological work currently conducted in the region is for mitigation purposes. However, due to excellent archaeological visibility, the diversity of stone raw material sources, and kinds of stone artefact manufacture, along with its numerous rock-art sites, it has been the focus of several research programs also. Recording of archaeological material in the region extends back over 100 years (Roth 1897a), but only began to be systematically recorded in the 1970’s (Campbell 1984). The __________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

5.

Chapter 1.

current archaeological database contains almost 2,000 locations where archaeological features have been recorded. This makes the study region particularly suited to examining the archaeological expression of regional behavioural complexity.

1.3. Organisation of the thesis. The thesis is organised into nine chapters. Chapter 2 develops a theoretical framework for investigating archaeological and behavioural variation at the regional level. Chapter 3 outlines the environmental, ethnographic and archaeological background to the region. Chapter 4 uses Chapter 2 and Chapter 3 to outline the methodological approach adopted in this study. This approach comprised examining the archaeological record of the region at three different scales. Chapter 5 examines the archaeological data at the broadest spatial and categorical scale. Chapter 6 maintains the broad spatial scale, but delves into levels of greater classificatory detail. Chapter 7 reports similar analysis to that outlined in Chapter 5 and 6, but does so on two sub-regions within the study region. The analysis of all these scales is brought together in Chapter 8, where it is discussed in the context of regional dynamics in Aboriginal behaviour. Chapter 9 concludes the thesis with some statements about the overall achievements of the study and the implications of these for the study of hunter-gatherers in general.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

6.

Chapter 2.

Scale, archaeology and Aboriginal behaviour. The key issue here, it may be argued, is that of scale, and this applies equally to both data and theory. Harry Lourandos.

2.1. The problem of integrating theory. One of the most significant theoretical issues to have arisen in Australian archaeology has been what became known as ‘intensification’ (Lourandos and Ross 1994). During the 1980s, a considerable amount of effort was directed towards understanding the changes in the archaeology of shelter sites that were dated to the mid- to late-Holocene (Beaton 1985, Hughes and Lampert 1982). This research revealed that along with more sites, there was increasing use of particular environments, such as offshore islands (Sullivan 1982), the semi-arid zone (Ross 1984), and the plateau in SE Queensland (Hall 1986). Other work demonstrated significant changes in rock-art (Morwood 1984), extensive exchange systems (McBryde 1987), and skeletal evidence suggesting increased sedentism along the Murray River valley during this time (Webb 1984).

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

7.

Chapter 2.

Until the mid-1980s, theoretical approaches in Australia emphasised a functionalist depiction of Aborigines who passively adapted their technological and socio-economic behaviour in response to changing environmental conditions (Birdsell 1953, Bowdler 1977, Hughes and Lampert 1982, Jones and Johnson 1985). The inadequacy of these approaches to cope with the degree and extent of change encouraged many archaeologists to explore a more diverse range of ideas to explain the mid-Holocene trends. These ideas included an ad hoc mix of general level theory and middle level explanation of specific archaeological problems. These ranged from post-depositional factors (Head 1986), demography (Beaton 1990), ecological factors (Smith 1986), technological change (Hiscock 1986), and socio-demographic models (Lourandos 1985). As Lourandos (1985) argued, at issue was a high degree of social complexity evident ethnographically, but a prevailing attitude amongst archaeologists of long term behavioural stability. By the early 1990s, the debate had polarised between essentially techno-environmental approaches and those favouring more socially oriented perspectives. In this sense, the Australian debate mirrored similar debates occurring in other parts of the world at about the same time (Trigger 1995). Also as in other parts of the world, little was resolved out of the debate, apart from Australian archaeologists gaining a greater awareness that no single perspective was capable of explaining all of the archaeological trends observed during the mid-Holocene. So long as archaeologists adopted a flexible approach to theory, the diversity of theory was a good thing for the discipline. This has subsequently led some to conclude that: A range of frameworks has now been presented in Australian archaeology … to facilitate some recognition of the convergence inherent between explanatory approaches which give primacy to social relations … and those focusing on biophysical factors. It is argued that such frameworks aim to accommodate both social and ecological/evolutionary approaches. The somewhat artificial historical dichotomy has … been liberated by systems approaches that engender multi-causal explanations in a move away from any simplistic prime mover. (Veth, O'Connor, and Wallis 2000, 54)

Veth et al failed to elaborate on what they meant by systems approaches, but their statement is indicative of the naivety of many archaeologists in regards to integrating theoretical perspectives without any explicitly stated theoretical tools for doing so. __________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

8.

Chapter 2.

Although the most obvious issue with the intensification debate lay with contrasting paradigms, when there was no easy solution to choosing between them it became clear that a greater problem lay beneath much of the debate. Lourandos (1996, 15) identified this problem as the inconsistencies with the way data were used to verify theory. Repeatedly, he observed a mismatch between the scale of theoretical statements and the resolution of the data used to assess them. This problem is illustrated by the re-interpretation that Bird and Frankel (1991) offered for the region Lourandos (1977) originally studied when he outlined his intensification argument. Using a database with many more radiocarbon dates, Bird and Frankel (1991, 14) explained their more complicated chronological sequence as a series of short-term events, rather than the rapid transition that Lourandos originally proposed. In doing so, Bird and Frankel questioned the legitimacy of the change originally identified by Lourandos. In return, Lourandos and Ross (1994) criticised Bird and Frankel for ignoring a range of other less well-dated contextual data that supported his rapid transition argument in the first place. Bird and Frankel’s (1991) results, to Lourandos and Ross (1994) at least, appeared to convey essentially the same information as originally observed by Lourandos. Lourandos and Ross’s (1994) interpretation of Bird and Frankel’s (1991) results later formed the basis for Lourandos’s (1996, 15) conclusion that the greater problem surrounding the intensification debate was the way archaeologists were assessing data collected at different scales. To Lourandos, assessing broad scale models with fine scale data represented an invalid argument. Hence, he states: The main point here is that for arguments to be effective, to be internally consistent and valid, they must be pitched at the same level. That is, we must compare like with like. A general argument (or model), for example, cannot be critically assessed by data drawn from finer-grained levels, but only by general data. (Lourandos 1996, 15, emphasis in the original)

Lourandos used this argument to suggest that the trends he observed remain valid. A rather convenient conclusion since it is Lourandos’s conclusions that came into question in light of the higher resolution data. In fact though, Lourandos has succumbed to a broader problem of viewing different scales of archaeological perspective as an either/or issue. The disparity between the perspectives of Lourandos and that of Bird and __________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

9.

Chapter 2.

Frankel, dissolves when it is realised that long-term trends and short-term events are closely related to one another. A point that Lourandos and Ross even make themselves (Lourandos and Ross 1994, 59). The issue is not that like must be compared with like in order to make theory formulation more robust, but that observations at different scales reveal different but complementary aspects of the same past. In this case, it becomes clearer how a theoretical concept of relating observations of different scales is more important than trying to establish which one carries the greater significance. What Lourandos (1996) overlooked is that it is not just scale that was abused during the intensification debate, but also the problems with incorporating divergent archaeological evidence. Lourandos’ argument about social motivations for change came primarily from ethnographic evidence. Yet, critics of the approach used excavated sequences (Beaton 1983) or evidence collected in vastly different environments (Smith 1986) to assess the validity of Lourandos’ social models. With such divergent evidence, it is difficult to determine which is the better model when they are describing quite different aspects of human behaviour. Although he correctly identifies a problem with scale, Lourandos is misguided in believing that stipulating complementary levels makes assessing theories with archaeological data more robust. Archaeological and behavioural levels are not independent of one another, and although the processes at different levels may be different, they will undoubtedly be having some influence on each other, however small that may be. Similarly, it is often not possible to find compatible archaeological evidence that is at the same level as the theory of interest. Levels that are suitable for theorising do not necessarily translate into levels that are easy to study archaeologically. The point that Lourandos misses is that the problem he implied as being the result of comparing different scales, reflects a broader problem in defining archaeological and theoretical units. The issues arising from the intensification debate therefore reflect much broader issues about scale and archaeological units of analysis. Despite its growing sophistication, theory about the archaeology of hunter-gatherers remains fragmented and poorly integrated due to an out-dated theoretical structure and confusion about issues of scale. In hunter-gatherer archaeology there has been a tendency to focus on particular aspects of behaviour and to use those to articulate broader conceptions about human nature. Due __________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

10.

Chapter 2.

to their dependence on how human nature is theorised, these tendencies have made it difficult to construct holistic perspectives of hunter-gatherer behaviour. They have made it equally difficult to integrate different aspects of hunter-gatherer material culture into unified constructions of the past. This is particularly the case in Australia, where there are few regional syntheses of Aboriginal behaviour that draw upon a diversity of material culture to construct comprehensive pictures of the Aboriginal past. The problem is that the theoretical perspective adopted by hunter-gatherer archaeologists often depends on subject matter. For instance, the study of rock-art often focuses an art acting as a means of information exchange (Gamble 1980) and research examines its role as an intricate, if ambiguous, reflection of social context (Conkey 1989). In contrast, functionalist and economic perspectives dominate the study of stone tool production and use (Binford 1979, Hiscock and Clarkson 2000). The theoretical gaps between these areas of study make it difficult to theorise behaviour in a holistic fashion because of the different types of behavioural theory employed in each case. In the case of rock-art, this is often symbolic-structuralist theory, and ecological/functionalist theory in the case of stone tools. Archaeologists recognise the need to incorporate diverse forms of evidence, but a theoretical framework for doing so explicitly is lacking. There are two reasons for this situation. The first is the fragmented nature of theoretical perspectives in archaeological theory generally, and the difficulty of implementing the middle ground between them. The 1990s witnessed a diversification of theoretical perspectives loosely collected under the heading ‘post-processual’ (Hodder 1989). These approaches appeared largely in response to perceived shortcomings of processual approaches (Binford 1972) which espoused cultural process, quantification, deductive logic and the investigation of contemporary processes as a means of better understanding the archaeological record (middle range theory). The new perspectives emphasised the importance of belief systems, symbols, and mental structures as factors influencing the form and distribution of material culture (Hodder 1982a). As a response to the processual approach, the post-processualist agenda focused on expanding the theoretical frame of reference in archaeology, deconstruction of prior approaches, and a greater use of knowledge and theory from the humanities (Mithen 1989). The diversification of approaches has brought back some balance to the processual perspective that has dominated North American (Dunnell 1986) and Australian hunter__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

11.

Chapter 2.

gatherer archaeology (Huchet 1991). However, this diversity has introduced a new problem, which is the difficulty of reconciling so many different perspectives in any single archaeological study. This has been especially so in places like Australia where archaeologists have embraced post-processual theories along with established processual approaches in response to the political implications of studying the archaeology of a culture that continues into the present (Smith 1995). Bettinger (1991, 61) attributed the multiplicity of theoretical approaches in huntergatherer archaeology to a preoccupation with theories of limited sets. By limited sets, Bettinger was referring to the tendency of many American hunter-gatherer studies to articulate middle and low level theories with patterns observed archaeologically. In a review of the Australian scene, Huchet (1991) suggested that Australian Aboriginal archaeology has often shared a similar focus. However, the problem is not limited to processual approaches. In their haste to demonstrate the importance of belief systems, symbols and mental structures, many post-processual archaeologists have fallen into a similar trap. The general theory may now be Marxist (Shanks and Tilley 1987), critical (Gero, Lacy, and Blakey 1983, Leone, Potter Jr., and Shackel 1987), or symbolicstructuralist (Hodder 1982a), but the overall procedure of using archaeological analysis to articulate a particular general level perspective has not really changed. Such approaches of course represent valid research programs. However, problems arise when holistic descriptions of behaviour utilise one general theory to the exclusion of all others. Recent debate about appropriate theoretical approaches to archaeology has seen a move away from an either/or conception of theory application and instead has highlighted the need to establish an integrated middle ground (Shanks and Hodder 1995). These discussions tend to emphasise two things. One is that human behaviour is complex and composed of many determining, often contradictory, processes. The other is that in order to understand human behaviour in a systemic sense, the interaction between these factors is of paramount importance, and to discuss these explicitly requires some form of scientific reasoning. Whether the middle ground lies in a realist epistemology (Trigger 1998) or a postpositivist conception of science (VanPool and VanPool 1999), the real issue is that these discussions amount to little without a framework for applying them. Clear statements about how to make a middle ground actually work with archaeological data are yet to __________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

12.

Chapter 2.

appear in the literature. In the absence of such statements, most archaeologists proceed by demonstrating that multiple theoretical perspectives can be useful as context to understanding, or at least defining, their particular problem. However, bringing such diverse perspectives together coherently and explicitly remains a difficult issue, and one that has proved to be a continuing source of criticism of post-processual approaches (for example Bettinger 1991, Knapp 1996). For this reason, some have focused on adapting evolutionary theory to the problem of explaining temporal sociocultural change (Boyd and Richerson 1985, Lyman and O'Brien 1998). The advantages of this approach are the clarity of well established theory, but it carries the disadvantages of assimilating human behaviour with other processes in the natural world. The second reason extends from a fundamental issue about the nature of space and time. Holistic conceptions of behaviour seemingly require a detailed understanding of both spatial and temporal variability. However, as Davidson (1999, 127) has highlighted, archaeologists are becoming aware of the limitations of attempting to address both spatial and temporal variability together. Detailed analysis of one usually means sacrifices in another. Instead, to make a holistic perspective workable requires a clear set of theory that can situate spatial and temporal components of behaviour and discuss them in the context of each other. To do so requires some theory describing the linkages between components of the whole, not just detailed descriptions of individual systems. Thus, in order to avoid similar mistakes during debates in the future, a theoretical framework that is better capable of defining archaeological and theoretical units is required. With this in place, comparisons between different theoretical perspectives, scales and space/time contexts should be more readily achievable.

2.2. An alternative theoretical structure. As established in the last section, there is an acknowledged need for an integrated middle theoretical ground (Shanks and Hodder 1995), but making such a position workable requires a framework for integrating the diverse components of complex behavioural systems. At the centre of middle-ground arguments, there is a relatively consistent emphasis on the aim of archaeology being to describe human behaviour. This remains the case even when divergent theories about behaviour can produce pasts that appear __________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

13.

Chapter 2.

starkly different. But rather than give in to a reality involving multiple pasts (Hodder 1989), a more productive step is to emphasise that for the main part, different perspectives of the past are produced by focusing on different aspects of human behaviour. That is to say, there really is only one past, but because archaeologists have such vastly different ways of looking at it, it can appear as though there were different pasts. The solution to the multiple-pasts problem is to embrace what multiple perspectives reveal about the whole. The following points are important in being able to this: 1. The importance of the duality between reality and perception. 2. Realising that scale, along with perspective, determines conceptions of reality. 3. The importance of some form of science as a means of producing knowledge. 4. The consequence of simple processes producing complicated outcomes. The first of these points, about reality and perception, is about how humans interact with and understand their surroundings. Although human beings, like all other biological organisms, must adapt to the natural environment, modern humans alone are distinguishable for having evolved a unique way of doing this through the manipulation of symbols. As Trigger states: This highly efficacious form of adaptation, which permits individuals to objectify themselves, imagine the thought processes of others and play out in their minds innumerable scenarios involving real and imaginary characters and situations, vastly enhances social co-operation and often greatly reduces the physical risk to the individual organism, but has disturbing and not easily predictable side effects. (Trigger 1995)

The implication of behaving symbolically is the production of a duality between the real world and the world that humans perceive and construct culturally (Carrithers 1992, 86). It was this issue that Childe (1979) was referring to when he argued that the world people adapt to is not the real world, but the world they imagine it to be. Since every individual has experienced a different history, their perception of reality will also be individual, while also sharing common elements because of shared experiences with other people.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

14.

Chapter 2.

As well as relating to past human behaviour, perception and perspective are equally important to our ability to be objective. Science has come to accept the fallibility of human researchers and realised that truly objective observation is not possible (Boyd, Gasper, and Trout 1991). Especially in disciplines like the social sciences, scientific reasoning has had to adapt to the ambiguity of its subject matter by adopting a realist epistemology (Trigger 1998) whilst continuing to produce logical, empirical statements of interpretation (VanPool and VanPool 1999). Binford (1987) referred to this problem as the ambiguity of the archaeological record because of the uncertainty that exists in the association between behaviour and its material expression. However, he suggests that if middle range theory were developed further, then some of this ambiguity could be more readily understood. The duality between objective reality and human perceptions of it represents the most serious challenge for any theory involving human behaviour. Archaeological theories variously emphasise different components of this duality (Trigger 1998), and as outlined above, have tended to lead into either/or conceptions of the significance of different theoretical perspectives. The issue is that the way archaeologists go about integrating theory of different types needs to become the focus of theory building. Given the history and complexity of general theories in archaeology, developing a more holistic framework would appear a formidable task. As Trigger comments, the tendency is to rework theory rather than embrace different perspectives: … social scientists exhibit much ingenuity in dismissing results that do not agree with their presuppositions as exceptions and even in reinterpreting them as unexpected confirmation of what they believe. Given the complexity of human behaviour, there is considerable scope for such mental gymnastics. (Trigger 1989, 23)

Such mental gymnastics would not be necessary if the focus was switched from allencompassing theories, to more articulated theory about interacting processes. Previous attempts to constructing frameworks for integrating theory have proceeded by attempting to clarify the differences between them (Schiffer 1988, Trigger 1989), but this does little to bring theory together. Instead, it drives them further apart. It is not surprising then that in Trigger‘s (1989, 20 & see Figure 2.1) scheme all high level theories are depicted as equal in terms of explanatory power. In a strictly positivist view of science, the relative utility of these theories would be assessed on the degree to which __________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

15.

Chapter 2.

they direct more middle and lower range theory, and by consequence, give meaning to greater amounts of the archaeological record. Regardless of the specifics of each theory, they are effectively competing against one another in terms of explanatory power. Hence, the real problem with Trigger’s diagram is that there is no sense of how the different perspectives combine to produce better explanation. The reality of course is that they are not all the same, nor are they competing. To use Trigger’s (1989, 20) diagram as an illustration, technological determinism and idealism focus upon fundamentally different aspects of archaeological data and human behaviour. In the scheme proposed here, each of the general theories used by Trigger address the same problem, understanding human behaviour, but they do so from different angles. With this in mind, it is possible to refine Trigger’s diagram as shown in Figure 2.2.

Figure 2.1. Trigger's (1989, 20) diagram.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

16.

Chapter 2.

Figure 2.2. Trigger's archaeological levels, modified for this thesis.

The important differences between Figure 2.2 and Trigger’s diagram are the inclusion of behaviour in the centre of all the theoretical approaches and segmenting the circle into aspects of human behaviour. Placing behaviour in the middle reflects the idea that at the heart of all archaeological analysis is a desire to understand human behaviour. In this sense, archaeologists begin with broad notions about behaviour, or aspects of it (general theory) and employ middle and lower level theories to interpret the archaeological record on their way to describing the behaviour of people in the past. The depiction of segments in the circle denotes the point made above that in the absence of a unified theory of behaviour, archaeological general theory addresses aspects of human behaviour, rather than all of it. The conceptual advantage of this diagram over Trigger’s is that it frees the archaeologist from seeing different general theories as __________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

17.

Chapter 2.

competing with each other. It also makes behaviour the focus of theory rather than emphasising the links between it and archaeological data. An approach to how archaeologists might understand the integration of theories is therefore reflected in Figure 2.2, but a remaining limitation of this diagram is the over simplified relationship it depicts between each of the general theories and human behaviour. The degree of influence of each general theory is not constant or equal. Importantly, the relative utility of different sets of theory will depend on the context in which archaeologists examine behaviour (Conkey 1997). It is here that scale becomes important, for to specify context explicitly requires not only specifying the types of features that comprise a context, but also the extent of each feature that is relevant in that instance. Scale, as a requirement for specifying the limits and extent of ones perspective, is an inherent element to all theory (Barth 1978a). In the absence of a unified theory of behaviour, all theories refer to only a part of the totality of human behaviour. As such, it is crucial that any theoretical statement clearly specifies its limits be it time, space, or the size of the social groups concerned. Without such statements, integrating theoretical perspectives is difficult because of the vagueness of what forms the context in each case. Due to the multitude of contextual features that vary in size, there are innumerable ways of situating theories relative to one another. Differentiating theories solely on explanatory comprehensiveness (also a scalar entity), as was done by Schiffer (1988) and Trigger (1989), is therefore but one way of structuring general theories. The scale concept can be illustrated by modifying Figure 2.2 to that seen in Figure 2.3. This figure shows the relative influence of different general theories on the perception of behaviour by varying the size of the arrows and the shape encompassing human behaviour. As scale and/or context changes, so does the relative influence of each general theory. This is not a new idea (Bailey 1981, 1983, Binford 1980, Bulliet 1992, Fletcher 1977, Knapp 1992), but one that has rarely been made explicit in archaeology, despite attempts in other disciplines (Norton 1989).

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

18.

Chapter 2.

Figure 2.3. Modifying the model to incorporate scale.

Clarke’s (1977) description of micro, semi-micro and macro archaeological levels illustrates how this concept might operate in a real scenario. By varying spatial extent only, Clarke (1977, 11-13) hypothesised that the driving factors changed from being primarily social at the micro and semi-micro levels to predominantly economic at the macro. Correspondingly, the types of theory brought to bear at each level would necessarily be different. The ideas that Clarke developed have influenced a lot of archaeological research that has utilised a spatial context. However, space is but one of the features that can be used to define context. It is not very difficult to imagine that varying the scale of multiple features comprising archaeological context can quite quickly generate very complex relationships between general theory and the perception of past behaviour. Nonetheless, as others have also argued (Davidson 1999), without explicit statements about scale, debate about the nature of past behaviour will suffer from misconceptions about the nature and cause of archaeological variation.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

19.

Chapter 2.

2.3. Theorising scale. Scale is inherently important in comparing different kinds of archaeological data, but it does not explain archaeological observations. Rather, it establishes their context. Scale is a concept that permits explicit statements about context. Most commonly, archaeologists use scale to establish spatial context. For instance, map scale defines a level of abstraction within the boundless limits of space. The map itself is not reality, but a representation of it. As a representation it sacrifices complexity in order to summarise the essential elements of reality at a given extent for a particular purpose (Robinson et al. 1984). The same also applies to archaeological theory. Theories describe reality at given levels of generalisation, where the levels can represent any observable entity that varies in size. Scale need not be restricted to explicit whole number ratios as in map scale. Statements such as regional, local, long term, and short term all imply levels of generalisation. Clarke (1977) has made scale explicit in spatial terms, although this is but one archaeological component that is scale dependent. Apart from its common application to time and space, scale is also applicable to almost any kind of feature that can be measured or theorised. For example, the difference between focusing on the individual, a community, or a whole society, is also an issue of scale. Cowgill (1974, 509) makes a related point about demographics. He suggested that what might appear to be population increase at one spatial scale, at a larger or smaller scale, can appear to be merely cyclical transhumance or a process of agglomeration or dispersal. Barth makes the point in a slightly different way: … we expect a political system of 500 persons to show different properties from one of 500,000; but a system of 500 can also be expected to show different properties depending on whether its members are concentrated, or spread over 500km2. The special character of the concept of scale lies in entailing a comparison and a judgement of significance on the dimension of size … (Barth 1978a, 254)

Many of the features (in both its material sense and as process) that archaeologists describe and theorise are subject to the effects of scale by virtue of them having or referring to some sizeable quantity. However, perhaps in trying too hard to make

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

20.

Chapter 2.

uniformitarian assumptions work, the effect of scale is poorly articulated in many theories. An example serves to illustrate this point. Optimal foraging theory is one example that inherently uses scale to define its entire research agenda. This theory, stated simply, assumes that hunter-gatherers seek to optimise the rate of return (in some currency) for a given expenditure as part of undertaking their daily activities (Winterhalder 1981). As such, it has been possible for anthropologists to record how individual hunter-gatherers rationalise activity in costbenefit terms. Optimal foraging describes well the behaviour of individuals or small groups. However, it is not clear how it applies to larger groups of people acting collectively during congregations (Conkey 1980) or co-operatively when dispersed throughout a region (Ingold 1986). What might be rational and optimal at the level of individuals may not be rational or optimal at the group level. More recent developments in human behavioural ecology and related fields have addressed this issue by focusing on areas other than foraging (Winterhalder and Smith 2000), but the concern here is only with optimal foraging theory. Similarly, Ingold (1986) has theorised how hunting and gathering people co-operatively appropriated resources in a regional setting by considering the concept of ownership rather than foraging strategy. However, what is important about scale in the context of optimal foraging theory, is that much of the archaeological record for hunter-gatherers reflects the accumulation of the products of behaviour of people, sometimes acting individually, and sometimes collectively. When we examine site distribution for example, we are examining an amalgam of behavioural processes. Important to understanding site distribution from archaeological evidence is understanding what happens to optimal foraging concepts when the level of analysis is changed from an individual to archaeological evidence of many individuals. This problem is in no way a fault of the theory of optimal foraging, rather it relates to how the specifics of a particular behavioural problem apply to archaeological problems where the organisational level may be quite different. The optimal foraging example illustrates two aspects of the research strategy archaeologists commonly employ- often in conjunction. The first is the investigation of processes. Understanding how individual hunters and gatherers forage in a landscape, and the way they manufacture stone tools, are examples of investigating processes. In __________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

21.

Chapter 2.

contrast, archaeologists also want to explain systems, either in terms of how systems may have changed through time, or as they operated at particular points in time. The move from processes to systems is not a simple one, and one that is often oversimplified. Systems are comprised of complex interactions between processes, which serve to manipulate the effect of any individual process. This precludes the straightforward application of theory about individual processes to real archaeological examples, where not only are different processes interacting, but they are also interacting on multiple levels. For this reason, it is important to understand what happens to processes when they are viewed at different levels. The issue of process transformation due to alterations in scale has received increasing attention in ecology as the discipline comes to terms with addressing such diverse problems as global climate change and responses of species to local habitat destruction. There are an increasing number of examples that illustrate the dynamic nature of individual processes with changes in scale. For instance, Brunsden and Thornes (1979) have illustrated the sensitivity of landscape changes to many scaled geological and ecological processes. Subtle differences in combination of factors can produce vastly different outcomes in terms of erosion. Other work has highlighted the effects of temporal scale also. For example, Phillips (1995) demonstrates that if the interval between disturbances is equal to or less than the recovery time, successional or transient plant communities will dominate a landscape. Such spatial and temporal dynamics are also evident in predator/prey relationships (Rose and Legget 1990). At small spatial scales predator and prey dynamics appear to be negatively correlated. Whereas at large spatial scales the correlation becomes positive as both respond to the same set of background environmental processes. Consideration of scale has also changed the way ecologists look at basic ecosystem properties such as stability and equilibrium (Turner et al. 1993). If the disturbances to an ecosystem are large and rapid, compared to the cycles in the ecosystem of interest, then the ecosystem is likely to become unstable. However, if the same sets of disturbances are considered at larger spatial scales, ecosystems appear to respond in a stable manner. Thus, although individual stands of forest may come and go, the total forest cover for a region can remain relatively constant. O'Neill and King (1998, 5-6) summarise these issues by stating: __________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

22.

Chapter 2.

These studies … lead to one inescapable conclusion: if you move far enough across scale, the dominant processes change. It is not just that things get bigger or smaller, but the phenomena themselves change. Unstable systems now seem stable…Bottomup control turns into top-down control …

O'Neill and King (1998) flag the definition of scale and level as important to clarifying descriptions of complex systems. However, it is important to realise that comparing different descriptions of systems requires understanding the levels imposed by the describer, versus that that may exist inherently in the system. Hence, a second related point is that there are grounds for questioning the description of levels, or at least their inherent existence. Consequently, a further point concerns the role of hierarchy theory in scale research. O’Neill and King distinguish between scale and level because of the confusion between these terms in the ecology literature, but the same holds true for the archaeological literature also. They define scale by stating: Scale refers to physical dimensions of observed entities and phenomena. Scale is recorded as a quantity and involves (or at least implies) measurement and measurement units. Things, objects, processes, and events can be characterized and distinguished from others by their scale, such as the size of an object or the frequency of a process. As a rule of thumb, when we use the term scale, we should be able to assign or identify dimensions and units of measurement. (O'Neill and King 1998, 7).

To adhere to O’Neill and King’s definition, the common archaeological practice of referring to regions and regional scales is meaningless because of the vagueness of what the limits of a region are (Smith 1976b). O’Neill and King extend their definition of scale to include the scale of observation, that is the temporal and spatial dimensions at which phenomena are observed. They suggest scale of observation has two components: grain and extent. Grain refers to the smallest unitary element in an observation set. Extent refers to the total area or length of time over which observations of a particular grain are made. For example, counting individual artefacts in a 1 metre square every 100m in a transect across a site fixes the grain of observation. The length of the transect and the number of sites establish an extent. Calculating means coarsens the grain, whilst subsampling (see Orton 2000, 148) reduces the extent. __________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

23.

Chapter 2.

Unlike scale, levels are states within dynamic systems. O’Neill and King (1998, 7) suggest ‘level refers to level of organisation in a hierarchically organized system’. Levels describe states emerging out of processes, flows, interactions, and rates. O’Neill and King argue that it is rate that causes the structure of levels. Within a hierarchy, higher levels have slower rates and lower levels react more quickly. Levels are characterised by their rank within a hierarchy—subsystems with similar rates occupy the same level. Levels of the same rank in different hierarchies can have different rates and can be defined by different processes, flows, or interactions. Any single system can also be described in terms of multiple hierarchies. The confusion between scale and level lies in that a level of organisation is not a scale, but a level of organisation can have a scale. For example, the spatial extent over which processes operate characterise the spatial scale of that level. In archaeological terms, sites can come in many different forms and sizes. Site, as a class of archaeological information, is independent of scale. However, particular examples of sites can be characterised by their scale. More difficult is how the level of organisation that emerges from interactions between behaviour at sites, can be characterised by their scale—that is the spatial extent the interactions occupy, or the time during which they existed. There is, therefore, a definable association between level and scale, but is not a causal link. Importantly, the form of the association is such that scale (as a function of grain and extent) limits the observable types and characteristics of levels. In archaeology, Clarke (1977) is perhaps the best known example of utilising levels for distinguishing micro, semi-micro and macro levels based on the delineation of archaeological observable features. However, it is here that O’Neill and King’s second point becomes important. There is a growing ecological literature that demonstrates the existence of levels within natural systems. In these cases, levels can be defined by statistical methods that identify discontinuities as scale changes. It was on this basis for instance that fractals were defined (Gleick 1987, 161). Problems arise when levels arbitrarily imposed upon data do not correspond with naturally occurring levels in the phenomena of interest. Although Clarke has quite rightly identified discontinuities in the form of archaeological evidence in order to define his spatial levels, it remains unclear (and untested) whether these archaeological levels correspond with similar levels in human behaviour.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

24.

Chapter 2.

Crossing

naturally

occurring

scale

discontinuities

can

create

problems

with

interpretation. The general phenomenon, called transmutation of aggregation error (O'Neill 1979), exists when correlation between processes become muddled or even reversed due to the influence of observations undertaken at different levels. The influence of other levels, which are driven by different sets of processes, complicates interpretation due to the difficulty in identifying which processes produce pattern that is specific to the level of interest. Much of the literature on transmutation error comes from mathematical systems modelling. Solutions to the problem are available (Schneider 1994), but they require establishing the characteristics of process interaction within the system—a practice which has barely begun in archaeology. An example of this problem in archaeology is the natural occurrence of stratigraphic levels and the excavation units imposed by the archaeologist. Detailed geomorphological analysis, or large numbers of radiocarbon dates may be necessary to get correspondence between the two sets of levels (Davidson, Sutton, and Gale 1993, Hiscock 1990). Unfortunately, due to the interchangeable use of scale and level, it is often unclear what are the subjectively imposed levels of the researcher and the delineated structure of the dataset. This has a critical bearing on the comparability of archaeologists’ conclusions and how complementary conclusions on any particular archaeological problem can be defined as such. These are difficult problems to resolve because of how much archaeology, and science generally, has to learn about how processes interact within systems. Even well established, but admittedly subjective schemes, like those used to classify life forms in biology, have come into question since the development of phylogenetics (Keeton and Gould 1986, 1017). Thus, archaeological theory can benefit from being more explicit about scale and levels when describing archaeological data and theorising human behaviour.

2.4. Implications for archaeology. By now it is hopefully clear that: a) there is a middle ground to archaeological theory but that it requires an integration of perspectives, and b) scale is an important concept for discussing interacting processes within complex systems. From here, the issue is about applying these ideas to archaeological examples, and in doing so highlighting the __________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

25.

Chapter 2.

research problems they present for archaeology. To illustrate these issues the following discussion uses Pickering’s (1994) ethnoarchaeological study of the Garawa. Pickering’s study is important because the Garawa live approximately 450km to the northwest of the area examined in this thesis. However, the most important feature of the study (in the context of this discussion) was the use of scale to demonstrate the interaction between different aspects of behaviour mediated through levels of landscapes. As such, it is a good illustration of how various behavioural factors interact on several levels in huntergatherer societies. The purpose of Pickering’s (1994, 149) study was to ‘… define, precisely, what social landscapes are, the variables and processes that structured them and if and how they are likely to be reflected in the archaeological landscape’. To explore this problem, Pickering delineated physical, subsistence-settlement and social landscapes within Garawa society. The physical landscape comprised ‘…the spatial and temporal arrangement of environmental phenomena, such as climate, topography, hydrology, geology, vegetation, fauna and so on’ (Pickering 1994, 150). Whereas the subsistence-settlement landscape comprised ‘…the spatial and temporal distribution of domestic occupation sites, activity areas, and resource locales’ (Pickering 1994, 150). Defining the social landscape was less straightforward. Pickering suggested that social landscapes are rarely incorporated into the concept of cultural landscapes, the latter usually being defined by the combination and interaction between the physical and subsistence-settlement landscapes. Unlike the physical and subsistence-settlement landscapes, social landscapes do not consist of a set of easily quantifiable entities. Instead, social landscapes: …have symbolic and material dimensions. The ‘symbolic’ dimensions are primarily intellectual, and manifest themselves as ideas, principles, and institutions of social organisation. The ‘material’ dimension of a social landscape is primarily spatial, emphasising the location characteristics of social phenomena, and can be represented through maps of the on-the-ground distribution of social relationships, institutions, social and religious territories, and activities. (Pickering 1994, 150)

In setting out these landscapes, Pickering stresses that the relationships between them are not simple, and depend on the scale at which the relationships are observed. Pickering distinguishes between micro- and macro-scale social landscapes. Micro-scale __________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

26.

Chapter 2.

social landscapes concern the behaviour of individuals or small groups of people, and because of this, they are spatially and temporally dynamic due to the continually fluid sense of micro-scale social behaviour. Each individual simultaneously behaves within the context of multiple social landscapes, which may be defined by local and regional distribution of family members, religious affiliations, conflicts with other individuals and so on. Due to the high degree of variability, pattern is difficult to discern at this scale. Unlike the micro-scale, Pickering (1994, 151) outlines how macro-scale social landscapes are tied to fixed features in the environment, and are more stable and absolute than micro-scale social landscapes. The spatial distribution of social groups over the long-term determines the pattern in macro-scale social landscapes. Pickering suggests that populations maintained formal and relatively stable macro-scale social landscapes through shared social and religious institutions. Both micro- and macro-scales have material expressions in the spatial dimension, and Pickering (1994, 151) suggests these can be used to interpret different aspects of the archaeological record. He suggests micro-scales are most appropriate for understanding the landscapes of individuals at archaeological sites for instance, whilst the macro-scale is better suited to examining the landscapes of populations and corporate groups comprising a regional archaeological record. As evidence for the distinction between micro- and macro-scales in Aboriginal societies, Pickering cites previous Australian research that has highlighted the correspondence between macro-scale social landscapes (‘tribes’) and macro-environmental structure. Peterson (1976) and Sutton (1990) have both emphasised the association between drainage divisions and regions of cultural similarity for example. The nature of the correspondence between the environment and macro-social landscapes is that resourcerich river corridors tended to be the focus of most activity, and were where a population’s social and economic landscapes became concentrated. Social and economic boundaries tend to lie in resource-poor areas, such as hills, ridges, waterless plains, or homogeneous vegetation. Pickering also suggests that similar relationships between social and physical landscapes exist at the micro-scale, particularly in relation to watercourses and drainage areas.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

27.

Chapter 2.

Pickering described the settlement-subsistence landscape using Garawa informants who maintain an intimate knowledge of a traditional lifestyle, and through studying the archaeology of fifty-four historically occupied sites. The settlement system was highly seasonal and through the year the Garawa went from being highly dispersed, to being congregated, as water availability gradually became more constricted during the dryer months. Pickering quantified this pattern by measuring the distance of occupied sites to drainage channels during each season. In the wetter months, 50% of sites were within two kilometres of a drainage line. In the drier months, the same percentage was located within 500m of a drainage line. A migration toward the lower channels of the Robinson River accompanied the contraction towards drainage lines, as the river contained the most reliable sources of water in the drier months. At the macro-scale, Pickering emphasises that the seasonal climatic cycle largely determined Garawa settlement patterns. In contrast, social factors such as affiliation to home estate, religious rites, family, and friends had a dominant bearing on the composition of groups occupying a site at any time during the year. These social factors formed an enveloping social landscape that interacted with the broader scale processes operating on a seasonal basis. Linking the social landscape to the physical landscape was the composition of a local descent group whose 10 estates divided the Garawa area. These estates were congruent with fourth order tributaries of the Robinson River, and were managed on average by an estimated 10 people (Pickering 1994, 156). Each estate had access to the river, whilst also extending into the periphery areas on the upper tributaries. This tended to tie each estate not only to a patch of the physical landscape, but an elementary unit of the subsistence landscape as well. Associated with each estate is a group of people who share a common affiliation with the physical landscape within that estate. As Pickering explains: Each estate is associated with a ‘local descent group’ which is also the primary ‘land owning group’ of that estate. The members of this group are patrilineally related individuals who share a common ancestry and primary social and spiritual affiliations. This land-owning group is considered to have primary rights to a particular estate. The estate of each land-owning group embraces sites of particular

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

28.

Chapter 2.

spiritual significance to that group. The estate is, therefore, the social, religious, and geographic home of the land owning group. (Pickering 1994, 155)

The land-owning group within each estate is related, but not identical to the group of people residing and subsisting within each estate. Estates were not exclusive zones, although as Peterson (1986) has suggested, there may have been particularly significant places within each estate where all but the land owning group were excluded. The landowning group is therefore distinguishable from land-using groups. However, land using groups were not a random selection from the remainder of the Garawa group. The religious affiliation between each land-owning group and its estate was also tied to a set of religious affiliations defining the Garawa as a group. These macro-scale affiliations determined things like suitable marriages, and strongly influenced the movement of people between estates. The social landscape was therefore not a passive overlay on the physical landscape, it also played an active role in the operation of the subsistence landscape. Although it is clear that there is a close spatial correspondence between social, physical and settlement/subsistence landscapes, these links are not causal. As Pickering explains: The pattern of correspondence between the social landscapes…the land owning groups and the subsistence-settlement landscapes…could result from imposing a deliberate strategy of patterning estates … Alternatively it may just reflect simple spatial correspondence between the spatial attributes of social and subsistencesettlement landscapes defined entirely by independent processes. The latter hypothesis is preferred as it avoids the suggestion that the symbolic dimensions of the social and religious institutions … were generated by the environment. At the same time it accepts that the physical processes and variables that structure the material spatial characteristics of both social and economic territories may be similar. (Pickering 1994, 156)

Whether it is ultimately cultural or physical processes that determine the spatial correspondence, the issue is irrelevant to the Garawa who see the estates as the embodiment of symbolic social institutions. The two are not separable. This illustrates the importance of understanding how processes integrate, rather than attempting to determine which is most significant. It is quite clear from the Garawa example that any attempt to separate behaviour into isolated parts or to view the parts operating on purely __________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

29.

Chapter 2.

causal connections is grossly under-realising the intricate web of interactions between them. Pickering summarises this issue by stating: The problem, therefore, is how to articulate social, religious, economic, and domestic strategies so that symbolic and material needs can be satisfied simultaneously. For the Garawa, such articulation was achieved through a corporate model of colocation1, in space and time, between the physical, subsistence and social landscapes. (Pickering 1994, 157)

Each of the three realms identified by Pickering are dependent upon each other for their existence. Symbolic and social institutions cannot function without nutrition provided by the natural environment. Efficient and sustainable subsistence within the environment requires a system of regulated human action, of which social and symbolic behaviour form crucial components. This would all come to nothing were it not for the fixed spatial connection the system has to the physical landscape and which the settlement system implements. The Garawa study illustrates the need to understand the past with a multiple perspective approach. However, the much thornier issue is how each of these perspectives (landscapes) contributes to the pattern observed in the archaeological record. Pickering’s position on this issue is somewhat oversimplified. The model he presents is quite complicated because scale, co-location, and the characteristics of each of the landscape types potentially produce many different archaeological outcomes. Yet, in his conclusion, Pickering chooses to focus on a narrow archaeological problem (hunter-gatherer settlement patterns), at a fixed scale (macro-scale) and eliminates complexity brought about by the integration of social and subsistence behaviour by suggesting it is not archaeologically visible. In summing up the Garawa study, Pickering suggests that macro-scale settlement pattern was predominantly conditioned by the environment. On this point, he is in good company. Several prominent archaeologists have argued the same thing, but have not

1

Co-location is a term developed by Carlstein (1982, 71-3) to refer to the requirement of dependent entities to be

proximately located in space and time in order to facilitate the exchange of information or energy necessary for the operation of that relationship. __________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

30.

Chapter 2.

arrived at their conclusions through developing such an intricate social context (Binford 1980, Clarke 1977, Foley 1981a, Gamble and Boismier 1991, Jochim 1976). The path taken by Pickering distinguishes his argument for it does not also suggest that the environment determined the nature of the social landscape. Rather, because the subsistencesettlement and social landscape share such a dependent relationship, ‘significant social landscapes will thus often articulate spatially and temporally with the subsistencesettlement landscape’ (Pickering 1994, 159). In this instance then, at the macro-scale, colocation occurs. The important implication of co-location of the subsistence-settlement and social landscape is that it renders the social landscape archaeologically invisible. This is because at the macro-scale, the structure of the subsistence-settlement and social landscape are embedded within the same environmental structure. Thus: The

symbolic

dimension

of social

institutions operated

independently

of

environmental pressures. The environment did, however, structure the distribution of the locales from which these institutions could comfortably operate through structuring the subsistence-settlement landscape. … While macro-scale social landscapes are therefore likely to be archaeologically invisible, they are, nonetheless inherent in the patterning displayed by the regional archaeological landscape. (Pickering 1994, 159)

To some extent, this may account for so many (mostly processual) archaeologists reporting such success in describing the settlement patterns of hunter-gatherers via environmental variables. Social landscapes probably do often co-locate with the subsistence landscapes as Pickering suggests, but there remains the need to incorporate social factors into the analysis of settlement patterns, since they are inherent in the archaeological pattern at regional scales. There are therefore two reasons why environmental generalisations cannot be relied upon, as processualists have historically been inclined to do. The first reason is that the relationship between the social landscape and the physical landscape has received such little archaeological attention that it has not been established whether this is the only aspect driving regional archaeological pattern. Pickering also makes this point:

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

31.

Chapter 2.

Anthropological studies have demonstrated the association between the total area and spatial characteristics of ‘tribal’ territories and general environmental phenomena. The relationship between the physical landscape and internal social landscapes has, however, been largely neglected. (Pickering 1994, 151)

The archaeological conclusion made by Pickering is limited to the kind of archaeological problem he addressed, that is settlement patterns. It is clear from studies examining different kinds of archaeological evidence that other kinds of social/subsistencesettlement patterns potentially exist. One obvious example is with rock-art. In the same volume as Pickering’s article, Taçon (1994) illustrates how there are many ways in which people socialised landscapes, and only some of these relate to subsistence. For instance, the frequencies with which various animal species were depicted in parietal rock-art were shown by Altuna (1983) to be very different from the frequency with which bones of these animals were recovered archaeologically. Hence, it would seem unlikely that rock-art would demonstrate the same kind of co-location observed by Pickering for settlement patterns. But again, such statements remain untested archaeologically. The few studies that have explored the physical/economic/social relationship for huntergatherers have noted a similar spatial correspondence, giving some support to Pickering’s argument. For instance Sutton and Rigsby (1982, 157) have noted that in the Princess Charlotte Bay area in Queensland, political behaviour within Aboriginal society involved rights of access and exploitations of resources in clan-owned estates. In this example, the political landscape was closely tied to the estate and subsistence landscape. Similarly, Myers (1986) highlights how the Pintupi did not simply follow a least cost subsistence pattern, but instead used their pattern of movement to facilitate social action: Pintupi did not simply move to where the food was, but rather scheduled their movements so as to use available resources in pursuit of their own social values—to initiate a boy or to exchange ceremonies. (Myers 1986, 292)

Superficially, this may look as if Pintupi subsistence followed resource availability, but such statements ignore alternative opportunities offered by brief times of resource abundance, like after rain for example. Deciding between competing subsistence opportunities, at least for the Pintupi, was resolved through the relative social advantages attached to each one. It is not surprising then that in such circumstances, the __________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

32.

Chapter 2.

social and economic base became spatially correlated. This was not because the link represented a strictly dependent relationship, but because there were social advantages for it being that way at certain times. However, neither Pickering’s or Myers’ studies, or others like them (Chase and Sutton 1987, Layton 1983, Memmott 1983, Williams 1982), explore differences brought about due to scale. Each of these studies examines the relationship between various social formations and their spatial expression and find correlation with aspects of the environment. However, the relationships are all defined at the level of a single tribal or linguistic group, and variation in the nature of the relationship, either within the group or between neighbouring groups, was not explored. The important issue is whether the co-location is such that it always masks any archaeological expression of the social landscape. Pickering stresses that many relationships between landscape types are scale dependent because the interaction between the processes involved transforms, sometimes fundamentally, with changes in spatial or temporal scale. As Pickering states: The key symbolic and material characteristics of social, subsistence-settlement, and physical landscapes will differ at different scales. The processes that structured these landscapes can be expected to differ correspondingly. Similarly, the archaeological reflection of these landscapes will be prescribed by considerations of scale. (Pickering 1994, 158)

Assessing this issue is made difficult because of Pickering’s vague description of what constitutes the macro-scale. The micro-scale, which Pickering associated with the action of individuals, is keyed to individual sites. The macro-scale, in contrast, is linked to the regional archaeological database. However, both site and region are not scale-specific terms, and can be defined in many different ways (Smith 1976a). The scalar limits of the existence of a co-location relationship between the social and subsistence-settlement landscapes are therefore also vague. For this reason, it would be unsound to form any more specific inferences from Pickering’s argument unless it could be established that the physical dimensions, numbers of people, social, subsistence and ecological systems were all comparable in size and complexity. Probably for this reason, Pickering (1994, 158) adopts a similar position to that of Lourandos (1996) by suggesting that like must be compared with like. __________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

33.

Chapter 2.

It has only been through studies like Pickering’s that the dynamics of hunter-gatherer behaviour at increasingly larger spatial scales have become apparent. Such work highlights that there are indeed differences brought about by scale. Previous ethnoarchaeological studies have suffered from the limitation of the ‘observer’ method, as Pickering (1994, 158) put it. The conclusions reached by an anthropologist will depend on how penetrating a relationship was formed with informants, and the kinds of questions that were asked. Pickering (1994, 151) suggests, as Peterson (1986, iv-v) has also, that it has only been since the introduction of Aboriginal land claims in Australia that anthropologists began to change the scale of the questions they ask. For land rights cases, this meant establishing in detail the kinds of broad scale land delineation observed by large groups of Aboriginal people. The result has been a realisation of the dynamics of Aboriginal spatial behaviour across scales, and these are only just beginning to be explored archaeologically. Pickering’s archaeological argument is logically coherent. However, the archaeological evidence he presents is insufficient to support or deny the social implications he attributes to it. All it does is establish some aspects of the settlement base. This is a weakness of the argument presented by Pickering. Pickering’s argument is best considered an archaeological prediction based on very sound anthropological evidence. As such, it presents the crux of the archaeological problem examined in this thesis. For whilst it is possible, with the aid of detailed anthropological evidence, to identify multiple landscapes and some of the interactions between them, the same cannot be said of archaeological evidence separated from ongoing ethnography given current practice. However, this in no way means that the archaeological evidence does not preserve such information. It only means that approaches to the analysis of the hunter-gatherer archaeological data have been insufficient to establish whether landscape relationships exist and what form they may have taken in the past. Given the need to understand scale, the mandate for this study is to manipulate spatial scale in order to examine the kinds of changes that take place within a single archaeological system. In doing so, it can be demonstrated how different factors interact to produce the variability within the archaeological record, and that single dominant factors, whilst important in structuring the pattern, are not the sole drivers of variability.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

34.

Chapter 3.

Regional context. When we left such campsites on the Burke, Iain Davidson asked Tom Sullivan2 how he felt when he saw such sites. His answer was ‘I feel proud’. He said his pride was because these places showed how his people had lived in the area. Davidson. 1991. Burke River Pipeline. A survey on the impact on archaeological sites. Appendix C-6

The purpose of this chapter is to identify those environmental, sociocultural, and archaeological issues that are pertinent to the current understanding of Aboriginal behaviour in the study region. With this knowledge established, the spatial distribution of Aboriginal material culture can be more meaningfully analysed in later chapters. Many of the features of the physical geography described in the next few sections were mapped in as much detail as possible or was appropriate given the limitations of the scale they are presented in each figure. Although all the detail of this information is difficult to represent in these diagrams, all of the detail was used as input to the analysis using a

2

Iain Davidson is the author’s doctoral supervisor. Tom Sullivan is an Aboriginal elder living in Mt Isa who has been an

invaluable companion and source of information about the Aboriginal heritage in the region.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

35.

Chapter 3.

computer based Geographic Information System (GIS), using methods described in Chapter 4.

3.1. The natural environment. The study region is located across the junction of two important environmental boundaries. The first of these is the transition from the fully arid to tropical north regions in Australia. The other is the boundary between water draining north in the Gulf of Carpentaria and south to Lake Eyre and Lake Frome. These characteristics, together with the varied geology, make the region a complex one in terms of the natural resources available to people subsisting by hunting and gathering.

3.1.1. Climate. The climate of the study region has been classified as BShw using the Koppen scheme by Dick (1975, 37). The BShw class is a semi-arid, tropical desert with a winter drought. The climatic unit extends in a band across northern Australia at a latitude of 20°S. In Australia the BShw or semi-arid zone is generally recognised to lie between the 250 and 500mm mean annual rainfall isohyet (Dick 1975, 38). As can be seen in Figure 3.1, the 250mm isohyet is located just to the south of the study region and the 500mm crosses the northern part of the region. Characteristic of this environment is low and unreliable rainfall, and high rates of evaporation. Temperatures demonstrate large seasonal and diurnal fluctuations. January is the hottest month, with Mt Isa experiencing a mean maximum monthly temperature of 38°C and a mean minimum temperature of 9°C during the coolest month of July. The wind direction in winter is generally from the south or southeast, changing to the northwest during summer when monsoon conditions push warm moist air south from the tropics. It is during this time that the region receives most of its rainfall in the form of thunderstorms.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

36.

Chapter 3.

Figure 3.1. Mean annual rainfall isohyets (50 percentile) for the study region. Crossed circles are weather stations.

The most important feature of the climate of northwest Queensland is its extreme variability. Since the region relies primarily on storms as a source of rainfall, the occurrence and volume of rainfall is highly unpredictable. Due to the limited extent of most summer storms, rainfall is also highly variable spatially. Surface water, although generally more abundant during the hotter months, is usually only available for short periods due to high rates of evaporation. Evaporation far exceeds the amount of rainfall received each year, reaching 300-350mm per month during the summer months and annually more than 3000mm (Meteorology 1988).

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

37.

Chapter 3.

3.1.2. Terrain. The Mt Isa uplands and the Selwyn Ranges dominate the terrain of the study region (see Figure 3.2). These two sets of ranges create a varied topography ranging from just over 600m elevation in the Mt Isa uplands, to below 200m in the southern part of the region on the Hamilton River. Despite only a small elevation range of 400m, the terrain of the region is characterised by extensive areas of dissected plateau and rugged scarps formed from the variable geology of the region. The terrain is not so rugged as to impede the movement of people in any particular direction, but demonstrates enough relief to have a strong influence on the location of activity and the distribution of resources. The Mt Isa uplands and the Selwyn Ranges are markedly different in terms of local terrain. The cause of these differences is the differing geology in each area. The Mt Isa uplands consist of a block of heavily folded and faulted set of hills and ranges, formed from Cambrian and pre-Cambrian rocks. Figure 3.2 illustrates the north-south trend of these folds in the left-hand side of the figure. Throughout the Mt Isa uplands, the terrain is very rugged. However, this ruggedness also forms the many breaks in slope, where water collects in rock holes. It is around these that many rock-art sites also occur. The Selwyn Ranges in contrast were formed by a series of granitic volcanic episodes that have intruded into horizontal bands of sedimentary sequences. This gives rise to a complex network of dissected plateaux. What is significant about the terrain of the Selwyn Ranges is the large number of scarps and mesas. It is on these features that rock shelters form and provide locations suitable for habitation and rock-art production.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

38.

Chapter 3.

Figure 3.2. Terrain of the study region.

3.1.3. Hydrology. The main feature of the hydrology in the region is the drainage divide formed by the Mt Isa Uplands and Selwyn Ranges. The divide separates water flowing into the Gulf of Carpentaria, from that flowing towards Lake Eyre in Australia’s south (Figure 3.3). The divide trends in a northwest-southeast direction, causing the majority of the drainage to flow either north/northeast or south/southwest (see Figure 3.4).

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

39.

Chapter 3.

Figure 3.3. Drainage basins of eastern Australia.

The highly variable rainfall in the region results in only intermittent flow in rivers. None of the rivers in the region flow all year, with the majority flowing for only a short period after rain. Armstrong (n.d., 42) noted that all the main rivers have waterholes in them. For two or three months after the wet season (summer), water is generally more abundant, but by the end of the dry season (winter), it becomes quite scarce, especially in the southern areas of the study region.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

40.

Chapter 3.

Figure 3.4. Drainage in the study region mapped at 1:1,000,000 scale. Black dots are recorded waterholes. Black line is the watershed between Carpentaria and Lake Eyre drainage systems.

The uplands drain surface water from areas over 350m in elevation very quickly, leaving these the least well-watered areas in the region (Stewart 1954). This also has two effects on the nature of the channels themselves. In the areas over 350m in elevation, the channels are mostly lower order streams that predominantly have a rocky bottom, consistent with the geology they are associated with, i.e. they pass through the rugged rock terrain. In the lower regions, on the plains, the channels are wider, tend to be sandy, and often braided. The braided nature of these channels results from the large volumes of water that flow after heavy rain. Between periods of rain, the rivers do not flow at all. These two types of channel provide different potential for containing water in the form of waterholes. In the upland areas, fractures and fissures in the host rock determine the location of waterholes. When there has been sufficient rain during the summer months, __________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

41.

Chapter 3.

these waterholes will hold water through most of the year. In contrast, waterholes in the lowland areas are located in local shallow depressions in the terrain and form through damming in the channels as the sandy sediments accumulate. The waterholes on the floodplains tend to be more extensive (see Figure 3.4), can contain larger volumes of water, but are less permanent throughout the year as the ground water level drops through winter. Several of the larger waterholes on the plains are semi-permanent and fixed in location where the bedrock is closer to the surface. However in many areas on the plains, smaller waterholes are dependent on the nature of sediment flow in the channels, and may disappear altogether over the course of a decade (Sullivan pers. Comm.). Due to the nature of the sandier channels in the northeast of the region, and the high rates of rainfall received in these areas, waterholes are more common on the plains in these areas (see Figure 3.4). In the southern parts of the region, not only does rainfall become less frequent, but the waterholes where water can be found are less frequent also, making the southern part of the region considerably drier and less predictable than the northern parts of the region.

3.1.4. Geology. The geological data used in this study came from a project to map the region at 1:100,000 by the Australian geological survey organisation (AGSO). This has mainly been for mineral exploration purposes, so some areas with low mineral potential were not mapped. Areas of extensive alluvial deposits were the areas overlooked, such as in the NE corner of the region and along the Burke River in the south of the region. In these cases the geological data was supplemented by digitising relevant portions of the 1:250,000 geological map sheets, and edge matching them to the 1:100,000 scale data. Some problems were encountered with this approach because of the age difference between mapping at the 1:100,00 and 1:250,000 scales. To overcome this, the geological data was re-classified into dominant rock types. The distribution of these rock types is displayed in Figure 3.5.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

42.

Chapter 3.

Figure 3.5. Distribution of major rock types occurring the study region.

Structurally, the geology of the region is known collectively as the Mt Isa inlier, which is composed of three structural elements: the Western fold belt, Leichhardt block, and the Eastern fold belt (see Figure 3.5). These elements are the oldest structural units in the region, all being comprised of Precambrian rocks (>1400 millions years old). The Western fold belt comprises some mafic volcanics, but is mostly comprised of sedimentary and metasedimentary units. The Leichhardt block consists of felsic and mafic volcanics and metavolcanics, The Eastern fold belt is comprised of chemical precipitates and clastic sedimentary sequences with some minor felsic and mafic volcanics. The deposition of each of these structural elements was accompanied by periods of deformation and intrusion. The deformation of the Western fold belt and Leichhardt block resulted from region deformation producing the north-south trending units.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

43.

Chapter 3.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

44.

Chapter 3.

Various sedimentary units have been deposited uncomformably over these structural units from the Cambrian through to the Cretaceous. These have included carbonate rich marine sediments and shallow marine siltstones and claystones. During the Tertiary these units experienced several periods of up-warping, erosion and laterisation. During the Quaternary period, the geology experienced little change apart from gentle lifting and erosion producing the extensive alluvial plains in the northeast and south. The most important aspect of the geology for archaeological investigation is the distribution of stone suitable for manufacturing stone tools. The metabasalt used to manufacture stone axes (Davidson, Cook, and Fischer 1994), predominantly occurs in the western portion of the region, as does quartzite. The laterised sediments of the Tertiary are more common in the east of the region where they occur as silcrete. In the south of the region, the carbonate rich sediments of the Cretaceous have produced several areas where chert and chalcedony occurs.

3.1.5. Soils/landsystems/geomorphology. The soil data came from the Atlas of Australian Soils, which mapped soil types at a scale of 1:1,000,000. Soils were used in this study because they most closely approximated landsystem and geomorphological mapping, and they had been mapped for the entire region. Three different authorities have divided parts of the region into landsystems, but each of these were spatial mutually exclusive, and did not edge match the other studies, or share common landsystem classes (Christian et al. 1952, Perry 1964, Wilson, Purdie, and Ahern 1990). Similarly, geomorphological studies have been conducted in the northern part of the region, but the units were not mapped (Stewart 1954, Twidale 1966). The soils dataset, although mapped at a coarser scale, are described in terms of soil and terrain units so that they represent a reasonable surrogate for land systems. However, unlike landsystems, the soil units do not take into account geology or vegetation.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

45.

Chapter 3.

Figure 3.6. Simplified map of the soils in the study region.

The soil units can be simplified into four classes, soils associated with hills, undulating areas, plains and plateau, following broadly similar classifications by Wilson (1990, 3-5). These classes are used to depict the soil units in Figure 3.6. A fuller description of each soil type is provided in Table 3.1. The main utility of soils is that, as a surrogate for landsystems and geomorphology at the regional level, they potentially inform about two important aspects of the archaeological record. The first aspect concerns the influence the local environment, as reflected by the landsystem classification, would have had on the types of activities hunter-gatherers undertook. Certainly, Drury (1996) identified significant differences in the types of stone artefacts located in different land systems in the Selwyn ranges. The second aspect concerns site preservation. Wilson (1990, 3-5) divided the landsystems in the region into destruction zones which comprised uplands and plains, and

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

46.

Chapter 3.

constructional areas comprising mostly recent alluvia. On this basis, it may be expected that some soil units, such as B2, CC, Fq, and Mz may have a capacity to obscure surface archaeological finds. These units are displayed with hatching in Figure 3.6. In contrast the other soil units may have a capacity to erode surfaces containing archaeological features, and affect their spatial integrity or archaeological visibility. Such factors could be used to understand differences in archaeological feature frequency in particular soil units. On this basis, the soil data justified its inclusion in the study.

Table 3.1. Soil units in the study region. Soil Unit B2

Description Sandy plains, dunes and stream channels

Soil Unit Mr

Bz

Steep stony sandstone hills

Ms

CC Ca Cd Fb Fq Fu Fz Il

Level Alluvial plains Undulating granite hills Steep granitic hills Dissected alluvial floodplains Alluvial fan plain Undulating hills Undulating stone hills Undulating hills with streams

Mu Mw My Mz Oa Oc Od Qa

JK Kb LK

Steep stony sandstone and quartzite hills Undulating stony plains, low basaltic hills Strongly dissected sandstone hills and mesas

Qb Ro Sl

MM MQ MR

Broad undulating clayey plains Flat plains and dunes Low, undulating plains with incised drainage

TB Ub Va

MS

Gently undulating plains with low lateritic Vc scarps Gently undulating plains Vd Basaltic hills, cones and mesas Wc

MT Mo

Description Undulating plains with some strongly dissected areas Undulating lands with broad shallow depressions Undulating dissected plains Undulating elevated plains Gently undulating plains Alluvial fans and sand filled streams Lateritic alluvial fans and plains Hilly and mountainous with gravelly surfaces Small level plains Hilly and rocky undulating and dissected plains Undulating and hilly lands Undulating and dissected lands Level plains with low, linear rocky outcrops and shallow streams Low irregular basaltic plateau Moderate or gently undulating lands Moderately undulating lands with occasional high linear stony ridges Gently undulating sandy plains Level plains with low valleys Strongly undulating lands with narrow high ridges and broad valleys

3.1.6. Vegetation. Like the landsystem data sets, no single project had mapped vegetation for the entire region. Neldner (1991) had mapped the lower half of the region, but no comparable mapping existed for the northern half of the region. To overcome this problem, LANDSAT satellite data was used to approximate a regional vegetation layer using

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

47.

Chapter 3.

Neldner’s mapping as a guide. Supervised classification3 of the LANDSAT data was attempting using Neldner’s vegetation mapping as training samples (this is discussed in more detail in Section 4.3.2). The resulting ten classes are therefore based on Neldner’s (1991, 32) structural formations, but do not necessarily coform exactly to his mapping. These are summarised in Table 3.2.

Table 3.2. LANDSAT classes and vegetation units.

LANDSAT Class 1 2 3 4 5 6 7 8 9 10

Description Open shrubland Hummock grassland Tall open shrubland Tussock grassland Woodland Open tussock grassland Low open woodland Open woodland Bare ground Low woodland

The mapped areas of each classified LANDSAT category is shown in Figure 3.7. Included in the figure is a frequency histogram for the dataset, which outlines the relative area of each class as indicated by the number of grid cells classified into each class. Neldner (1991, 39) suggests that although overgrazing by cattle has been a problem in the region, it has not had a significant effect on the distribution of the main floristic associations since Europeans arrived. He suggests that most changes have been confined to the herb layer, and then mostly around waterholes and adjacent to streamlines. Since

3

The aim of satellite data classification is to partition the image into spatial regions, which can be assigned labels

corresponding to given characteristics of these areas on the ground, such as vegetation type. Classification can be undertaken in two ways: supervised and unsupervised. Supervised classification uses training areas that are identified on the image by the user. These areas are used to statistically describe the spectral characteristics of each land cover class, which in turn, can be used to classify the entire image. In contrast, unsupervised classification involves dividing an image into a specified number of statistically derived classes based on the spectral characteristics of the entire image. The classification undertaken in this study used the supervised approach using the tools available in Erdas Imagine 8.5.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

48.

Chapter 3.

very little clearing has been undertaken in the region, he suggests that the distribution of most woody species would not be significantly different from those encountered before European settlement. Consequently, since the classification of the LANDSAT data was done using such broad categories, it is likely that they are reasonably indicative of the vegetation pattern existing in the recent past.

Figure 3.7. LANDSAT classification.

3.1.7. Fauna. Very little information about faunal distributions is available for the study region. McFarland (1992, 8-11) for example noted that very little data has been collected on faunal distributions in western Queensland. Horton (1976) provided an outline of the major species occurring in the region, but did not discuss their distributions. What faunal data was available for the region was therefore of limited value for the present study. For this reason fauna were not considered in this study as a variable for understanding the distribution of archaeological material.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

49.

Chapter 3.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

50.

Chapter 3.

3.2. NW Queensland ethnography. Encounters between European settlers and Aborigines in the study region have had a long history. On their ill-fated trip in 1860, Burke and Wills travelled right through the centre of the region (Bergin 1981). However, although they noted that there were numerous indications of Aborigines living there (Wills 1860, January 12th 1861), they came into contact with them on only a couple of occasions. A subsequent search party to find the Burke and Wills party, led by McKinlay, traversed the eastern portion of the study region. This party had even fewer encounters with Aborigines, although smoke from their fires was often noted (McKinlay 1862, 76, 78, 82, 84, 85, 88). This changed dramatically however once European settlers arrived in the region in the late 1860’s (Fysh 1933). As was the case in many regions of Australia, this led to confrontations between settlers and Aborigines on numerous occasions. However, unlike other regions, these culminated in a pitched battle in 1884 (Armstrong n.d., 144). From then on, starved and sick, the Aborigines began a process of migrating into the major homesteads and towns. It was not for another ten years, after which Aboriginal society had been irreversibly altered, that Roth (1897a) began recording details of their behaviour. Roth’s work, one of the pioneering studies in Australian anthropology, is the most extensive account of Aboriginal life in the region, and records many details about the social customs, material items, and language. As extensive as these descriptions are, only a small amount of them were devoted to describing any aspects of spatial behaviour in the region. Although there are many aspects of Roth’s work that are informative about the behaviour of Aborigines in the region, only social groupings and trade have a spatial component which is relevant to the present study.

3.2.1. Social groups. Social boundaries in the region have been outlined in studies by Roth (1897a, 41-43) and Tindale (1974). The areas outlined by Roth are shown in Figure 3.8, and those of Tindale are shown in Figure 3.9. The names in both maps are as used in the respective sources. Along with group boundaries, Roth (1897a, 41) also included the main encampments of

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

51.

Chapter 3.

the groups he knew of in the Boulia district. The locations of these camps are also shown in Figure 3.8.

Figure 3.8. Roth's group boundaries.

The social groups defined in both studies are problematic. Roth based his groups upon language districts in the region, although by his own admission ‘…the exact ethnographical limits of such a district must necessarily prove a matter of some difficulty’ (Roth 1897a, 1). Confusion arises from Roth’s definition of the term messmate: …speaking generally, these same tribes are able to render themselves pretty much mutually intelligible, and possess in common various trade routes, markets, and hunting-grounds, customs, manners, and beliefs; in other words they might as a whole, be well described as “messmates”, though in the aboriginal language there appears to be no one word which would express them collectively. (Roth 1897a, 1)

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

52.

Chapter 3.

Roth named 19 ‘tribes’ of the Boulia district, which he collectively referred to as the Pitta Pitta and messmates. Either Roth meant that the Pitta Pitta and messmates should be considered collectively as the Pitta Pitta, or Roth could not delineate the boundaries of the sub-groups of the Pitta Pitta group. Either way, it becomes unclear what Roth meant by the term ‘messmates’. For the other groups, Roth made no mention of whether he was referring to messmates or something else. Although he later goes on to describe how the Kalkadoon were messmates with the Workoboongo and Injilini; the Mitakoodi were messmates with the Nouun and Woonamurra; and the Yarroinga messmates with the Workia and Undekerebina (Roth 1897a, 42-3). The black arrows in Figure 3.8 indicate the associations between these groups. Roth also mentioned that the Mitakoodi recognised a southern and western division within the Kalkadoon group (Roth 1897a, 42). Adding to the confusion is that Roth states that there was no collective word to refer to the concept of ‘messmate’ in the languages spoken by any of the groups in the region. Yet earlier Roth (1897a, 1) mentions that ‘…there is nevertheless a portion of country known to the Pitta-Pitta aboriginals as the “ooroo-ena mie-ena”—i.e., “one-and-the-same country”’. Although different dialects were spoken throughout the Pitta Pitta tribes, Roth suggests these people were all mutually intelligible with one another. Given the time that he spent with the people in Boulia, and the detail with which he recorded their culture, Roth must be considered the authority on the region. However, the confusion over terms suggests that outside the Pitta Pitta group, Roth’s knowledge was likely to be mostly second-hand. Thus, Roth’s boundaries in Figure 3.8 should be treated with some caution.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

53.

Chapter 3.

Figure 3.9. Tindale's group boundaries.

The boundaries defined by Tindale are more easily interpreted than Roth’s, since he was clear about the way he derived them from natural topographic features, and his notion of the ‘tribe’ (Tindale 1974). However, Tindale’s notion of the tribe has been questioned on several grounds because of the difficulty of understanding what social level a tribal boundary was meant to describe (Berndt 1976). For this reason, although better organised than Roth’s, it remains unclear what Tindale’s boundaries actually reflect. As was the case with Roth’s map, these boundaries are indicative of some division on some social level, but do not necessarily indicate demarcation on all levels, or directly reflect divisions observed in the archaeological record.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

54.

Chapter 3.

3.2.2. Trade. Roth (1897a, 132) documented trade in the region, and his comments in this regard are summarised in Figure 3.10 by indicating those items entering and leaving the study region at different places. The dashed lines in Figure 3.10 also indicate the main routes of travel that Roth described, and the arrows indicate the direction of movement along these routes. Of equal importance, are the much more extensive long distance trade routes that crossed the study region. These have been described by McBryde (1987), Mulvaney (1976), and Davidson et al. (In Press). At this broader scale, ceremonies, the narcotic plant Pituri (Duboisia hopwoodii), ceremonies, axes, and ochre formed the important elements of a network extending throughout the Lake Eyre basin. McCarthy (1939) implicated these broader trade routes in the trunk routes that he claimed crisscrossed the Australian continent. Although there is a great deal of information about the types and distribution of individual items, many of those recorded by Roth for example do not preserve well in the archaeological record. Of those that do preserve archaeologically, for instance grindstones; stone knives; and ochre; there is little information about these items that can be used to elucidate the specifics of how they moved around the region. Sourcing studies of axes (Davidson et al. In Press), has shown that axes were distributed in finished form throughout the study region, especially those originating in the northwest. Ochres used for painting in an art site in the Selwyn Ranges came from at least two sources (Ridges, Davidson, and Tucker 2000). One located within a few kilometres of the site, and the other fifty kilometres away. Thus although there was a process of items entering and leaving the region as outlined by Roth, there were also items moving within the region.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

55.

Chapter 3.

Figure 3.10. Trade routes and items from Roth.

The most important aspect of the trade routes in the context of this study, is what it reveals about the movement of people. Roth (1897a, 132) describes the ‘walk-about’ as ‘…one of the most important institutions in vogue among the aboriginals…’. The walkabout was integral to the operation of the trade network. The movement of people during the walkabout is indicated in Figure 3.10 by the dashed lines. Roth (1897a, 134-6) described the movements of each of the tribes that he recorded. In Figure 3.10 each dashed line indicates the movement of one of Roth’s tribes along that route. Multiple lines adjacent to one another therefore indicate the movement of multiple groups. According to Roth (1897a, 132), the movement of people rigidly adhered to the waterways, where people knew they could freely travel.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

56.

Chapter 3.

Although it is clear from the routes of travel that Roth mentioned that some drainage courses were probably used more than others, it is likely that all the major drainage lines in the region facilitated the movement of people. As is indicated by the black arrows in Figure 3.10, the pattern of drainage suggests that the movement of people was divided into southwest and northeast directions. At a broad scale then, this would suggest that the occurrence of exotic items in the region would need to be interpreted in the context of these directions of movement.

3.3. NW Queensland archaeology. As was highlighted by Campbell (1984, 173), the archaeological study of Aboriginal culture in northwest Queensland was almost non-existent up until the 1980’s. Mostly due to the expansion of mining developments in the region, Aboriginal cultural heritage surveys for impact assessments rose sharply in number through the 1980’s. These surveys contributed substantially to the number of NW Queensland sites listed on the Queensland archaeological site register along with raising issues with their distribution. Cultural heritage survey still forms the majority of archaeological study presently undertaken in the region. Despite the emphasis on impact assessment, the region has been recognised as having a high potential for conducting archaeological research and examining behavioural variability on broad scales. Evidence of this potential comes from two long-term research projects conducted in the region. One of these was directed by Hughes (1983) and led to Hiscock’s (1988a) PhD project. The other was directed by Davidson (1993) and led to this project. Figure 3.11 indicates the location of Hiscock’s study region relative to the present one. Both these projects have produced research that has proved important in the context of Australian archaeology.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

57.

Chapter 3.

Figure 3.11. Location of Hiscock's study region.

The projects of Hughes and Davidson, along with other archaeological study in the region, have raised several archaeological issues. These include: 1. The chronology of Aboriginal occupation in the region. 2. The nature of stone tool use. 3. The processes leading to the complexity of trade and social relations observed at contact. 4. The role of rock-art.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

58.

Chapter 3.

5. Subsistence and settlement patterns in a resource rich semiarid environment. The main purpose of this chapter is to review how previous archaeological research has explored these issues. The chapter therefore moves through the research that has focused on chronology, rock-art, stone tools and site predictive modelling. The final section of the chapter outlines the ways the present study contributes to the archaeological research agenda for the region.

3.3.1. Chronology. In reviewing the chronology of northern Queensland in 1984, Campbell (1984, 174) began by stating that ‘we do not yet know when north Queensland was first colonised’. Fifteen years later, this question remains largely unresolved. The excavation of two sites in northwest Queensland has demonstrated that occupation of the region extends back at least to the last glacial maximum (that is 15-20,000 years ago). These sites have also shown that occupation from the last glacial maximum to late Holocene (that is until the last 5,000 years) may not have been continuous. In both instances, these gaps probably relate to environmental changes taking place since the Pleistocene. The significant sites for this discussion are Colless Creek Cave, Louie Creek Cave, and Cuckadoo 1. Colless Creek Cave and Louie Creek Cave are sites located in the Lawn Hill area to the north of the study region (part of Hughes’ project). Cuckadoo 1 is a site excavated by Davidson in the SE corner of the study region. 3.3.1.1 Lawn Hill. The Lawn Hill area is characterised by steep sided gorges cutting through jointed and dissected dolomite. The sides of the gorges form 30-45m high cliffs that contain many rock shelters. For his PhD., Hiscock excavated Colless Creek rock shelter, which is located on a tributary of Colless Creek on Lawn Hill station, about 100km NW of the study region. The creek has permanent water, and stone sources were available in the creek and on top of the plateau. Using the results from the Colless Creek excavations, Hiscock (1988a, 244) proposed four phases for the occupation of the Lawn Hill region: Phase 1: 40-18,000 years BP

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

59.

Chapter 3.

Phase 2: 18-14,000 years BP Phase 3: 14-5,000 years BP Phase 4: 5,000 years BP to the present. During the first phase, the only evidence for occupation came from the two excavated cave sites. The artefacts found in the caves at this time were small with relatively thick conchoidal platforms. Hiscock (1988a, 245) suggested that these artefacts were the product of manufacturing small retouched flakes that were rarely left in the site. The absence of completed retouched flakes indicated to Hiscock that these artefacts probably had a long use-life. Supporting such a view was the occurrence of distant stone raw materials and a variety of faunal species in the sites during this phase. A combination of tools with long use-lives and the occurrence of distant resources suggested that during this phase the economy focused on the wider landscape beyond the immediate confines of the gorges. The second phase coincides with the onset of the last glacial period when the climate was cooler and drier. Correspondingly, Hiscock found that human occupation at this time focused more on the areas within the gorges where there was presumably a reliable source of water and associated biotic resources. Evidence from the cave excavations supporting this idea were: 1. higher discard rates; 2. a greater emphasis on the river chert cobbles; 3. higher occurrence of fires; and 4. more frequent trampling of artefacts4. Hiscock also suggested that the social networks observed for Aboriginal people occupying the arid interior of Australia at the time of contact (e.g. Gould 1980) were probably not in existence 14,000 years ago. Yengoyan (1976) argued that elaborate social networks are one method for ensuring access to resources and other people in risky

4

Hiscock inferred the amount of trampling from the number broken flakes in each period. Broken flakes can also be

produced deliberately (Kippen 1992, 44), so their frequency can also reflect changes in stone technology or their application as well as trampling.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

60.

Chapter 3.

environments. Exploiting the resources in the gorge during this phase suggested people did not have a system in place for exploiting distant and presumably risky resources. Hiscock also suggested that at the height of glacial aridity, the population of the region would have been small. Given the amount of resources available in the gorges, Hiscock (1988a, 247) proposed that if these were all that people utilised, it was unlikely that more than 20-30 people lived there. It was also Hiscock’s suggestion that if populations were low, then this would have exacerbated the difficulties with maintaining social networks. Hiscock stresses that he does not believe the people of Lawn Hill were completely isolated, but that the social relations were less frequent, shorter, and and of a different nature to those seen at contact in central Australia. During the post-glacial phase, Hiscock saw a return to an economy that exploited both the plains as well as the gorges. The re-appearance of stone raw materials from the plains and tablelands indicated a rearrangement of the economic system. In addition, indicative of a trend to expand out of the gorges was the decrease in discard rates of both stone and bone in the excavated sites. The final phase of Hiscock’s sequence covers the time from mid-Holocene to the present. The 62 open sites that Hiscock examined represent the phase archaeologically. All open sites examined by Hiscock were determined to be of Holocene age (1988a, 81). Hiscock inferred the age of the open sites using several lines of evidence: 1. For one excavated open site on a levee within one of the gorges, a radiocarbon date of an unreported substance at a depth of 1.2m returned an age of 4,360 ± 100 (SUA 1878). 2. The presence of ‘implement types’, such as backed blades and unifacial points were found at many sites. In other contexts in Australia, these implements have been associated with Holocene periods of occupation. 3. Post European contact items, such as metal spear points and flaked glass, existed in some sites. 4. None of the exposed Pleistocene soils had artefacts on their surfaces. Based on his open site surveys, Hiscock (1988a, 81) concluded that virtually all the artefacts he examined in this context were probably less than 5,000 years old.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

61.

Chapter 3.

The apparent abundance of open sites suggested to Hiscock that during the most recent stage the use of both the plains and the gorges was more extensive than in any other period. However, by Hiscock’s (1988a, 250) own admission, the largely un-dateable context of the open sites means that it is impossible to make too much of such an assertion. The similar discard rates, and broad similarities between the excavated assemblages observed in this phase, and the previous phase, make such assertions even more difficult. The main supporting evidence comes from increases in the rates of exotic raw materials in the upper levels excavated at both Colless Creek and Louie Creek Caves. Hiscock did not address whether the appearance of distant stone was the result of exploiting a wider range of resources or because people transported stone over greater distances. Hiscock did stress however that the types of resources utilised most probably did change and accompanying the change were rearrangements in locations of different activities.

3.3.1.2 Selwyn. Hiscock’s analysis of chronological change for the Lawn Hill region is important because it was the first piece of research which demonstrated that significant changes in Aboriginal behaviour had taken place through time in western Queensland. Unfortunately, because the Lawn Hill region is such a unique environment, the applicability of Hiscock’s findings to other adjacent areas remained uncertain. Davidson’s (1993) excavations at Cuckadoo 1 partially aimed to address whether the trends observed by Hiscock also took place in areas with a less reliable water supply. Cuckadoo 1 is located in the southwest corner of the study region. The site itself lies on the western bank of Sandy Creek. Sandy Creek is a southerly flowing creek and drains the southern margins of the Selwyn Range. Where it passes by Cuckadoo 1, Sandy Creek cuts through a rocky barrier. At this point a permanent waterhole forms, known locally as Wildman Soak. Excavations at Cuckadoo 1 returned a radiocarbon data of 15,270 years BP (17,900 years BP calibrated) for charcoal removed from the penultimate removal. In 1989, Davidson excavated more squares, which confirmed the stratigraphy observed in the first season. Each stratigraphic unit was older than the one above it, but this was interspersed with several hearth features that complicated understanding the

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

62.

Chapter 3.

stratified sequence of the site. Dates from the new excavations indicated that sedimentation rates had not been consistent, and that at least one discontinuity existed in the sequence (Davidson, Sutton, and Gale 1993, 169). The deepest sediments in these squares were 12,500 years old. Davidson et. al. (1993, 167) also studied the sediments of Cuckadoo 1 using geomorphological techniques. These analyses revealed a strong association between mass susceptibility (a magnetic property of the sediment) and the occurrence of charcoal in different levels providing good evidence for the increased use of fires in the site between 3-4,000 years BP. Other evidence included the occurrence of iron in the sediments, which occurred at three times the concentration in the granite of the roof of the shelter, or the alluvium outside it. Davidson et. al. (1993) suggested that one explanation for the high levels of iron might be the human use of ochre in the levels dated to 3-4,000 years old. Chert and quartz stone artefacts occurred in each excavation unit. Order of magnitude changes occurred in the frequencies of these artefacts at about the same time as those found by Hiscock at Lawn Hill. Davidson (1993, 211) also highlighted the occurrence of metamorphosed basalt used for producing stone axes. The occurrence of metabasalt in the Cuckadoo 1 sequence was interpreted as evidence of long distance trade networks. Fragments of metamorphosed basalt only occurred in the upper sediments of the Cuckadoo sequence, suggesting that long distance trade networks may have only been in existence for the last 1-2,000 years. Further excavations at two other sites (Cuckadoo 3 and Anvil Creek 1) failed to identify Pleistocene occupation. Charcoal from the bottom layer of Cuckadoo 3 was radiocarbon dated to 420 ± 60 years BP (Beta 25056). Charcoal recovered from the upper layer of Anvil Creek 1 was radiocarbon dated to 230 ± 50 years BP (Beta 28450), and charcoal in the lowest layer was radiocarbon dated to 940 ± 70 years BP (Beta 28451). In addition, a small excavation was undertaken at an ochre quarry, approximately 3km to the south of Anvil Creek 1. A shallow excavation revealed an ochre rich layer just below the surface that contained charcoal, which was dated to 250 ± 70 years BP (Beta 40523). At the same ochre quarry lichen living on the surface of outcropping ochre was found to have a relatively recent age of 140 ± 100 years BP (NZA 5730). Since the occurrence of the

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

63.

Chapter 3.

lichen on the ochre, in this case, must have post-dated its quarrying from larger portions of ochre, the date indicates a minimum age of quarrying. Within the study region, researchers have also attempted to determine the age of some open sites. In 1997, as part of an Aboriginal cultural heritage assessment, Shawcross and Hughes (1997) examined hearths at five open sites, and excavated three of them. Charcoal from one of these excavations was radiocarbon dated to 860 ± 150 BP (WK 5799) (Shawcross and Hughes 1997, 13). These excavations are consistent with Hiscock’s impression that open sites in western Queensland are generally mid to late Holocene in age (that is less than 5,000 years), and probably most are less than 1000 years in age. Hiscock (1988b, 54) also inferred relatively recent ages for the sites he studied in the Boulia district to the south of the study region. Hiscock excavated one open site on Mucklandama Creek and found artefacts within the top few centimetres of the deposit. A radiocarbon date of a charcoal lens eight centimetres below the surface returned an age of 1190 ±120 (Beta-21518). In summary, there are broad similarities between Hiscock’s findings at Lawn Hill and Davidson’s in the Selwyn Ranges. These similarities suggest it is probable that significant changes in the nature of regional occupation have taken place during the Holocene. At Lawn Hill it appears that these changes resulted in the rearrangement of the resource economy, with the gorges being utilised more intensively during the last phase of glaciation. In the Selwyn Ranges, the response appears to have been periodic episodes of abandonment during the last glaciation, although it is difficult to confirm this in the absence of those sedimentary units representing the gaps. The Selwyn area also appears to have had increased levels of occupation over the last 1,000 years.

3.3.2. Rock-art. Until recently, published information about the rock-art of northwest central Queensland was sparse. However, the systematic recording of archaeological sites in the late 1970s led to several reports focusing on rock-art sites in more detail (Franklin 1988, Hill 1981, Morwood 1978). As well as providing the first quantitative analysis of the art, the publications of Franklin and Morwood also revealed some of the character of its regional variation. The two sites discussed by Morwood, Carbine Creek and Saxby Waterhole,

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

64.

Chapter 3.

showed significant differences in content. Non-figurative engravings dominated Carbine Creek, while in contrast, Saxby Waterhole was characterised by a large number of painted anthropomorphic figures. Franklin’s publication sought to extend the propositions of Morwood by analysing a further five sites with multivariate statistical techniques. Her findings were different from Morwood’s in that she was not able to show any major groupings of sites within the region. Instead, she found that some sites were different from others in terms of the relative frequencies of different motif types. Human figures dominated one site for instance, isolated pits and macropod tracks characterised another, and the third contained complex non-figurative designs and circles. Three of the sites contained common elements, although in differing proportions, forming a loosely aggregated group. To Franklin, these results suggested that there were some indications about regional structure, but the nature of this structure was not straightforward. Franklin also performed a second set of analysis that sought to examine claims that the art in northwest Queensland had elements similar to those found in other parts of Australia. Maynard (1979, 96-97) had previously suggested that early descriptions were consistent with the Panaramittee style observed in central and South Australia. In discussing the rock-art of the Laura region, Rosenfeld (1981, 89) had also suggested that the art in the two regions demonstrated some similarities, but acknowledged that overall both were distinctive. Rosenfeld attributed the distinctiveness of the northwest central Queensland sites to the region being located in between central Australia and Laura, so that the Mt Isa sites had similarities with both areas. By comparing the Mt Isa sites with art in other parts of Australia, Franklin’s purpose was ‘…to determine whether (the art sites) constitute a regional entity or style.’ (Franklin 1996, 138). Franklin found that four of the five Mt Isa sites stood out from other Australian sites because of the complex circle designs and figurative motifs they contain. Another distinguishing feature was the relative paucity of tracks compared with other regions. Through higher proportions of pits, the remaining site grouped with sites in the Laura region. On the basis of relative motif percentages, Franklin (1996, 146) characterised the four distinctive Mt Isa sites as having relatively high percentages of complex circles, crescents, simple circles and figurative motifs.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

65.

Chapter 3.

Amongst these distinctive characteristics, Franklin found a decrease in proportions of tracks and pits from central Australia to Mt Isa, followed by an increase in them at Laura. Conversely, simple and complex circles, crescents, and figurative motifs increase from central Australia to Mt Isa, and then decrease in the Laura region. Based on these trends, Franklin concluded that Rosenfeld’s impression was the right one. In regards to Morwood’s results, Franklin concurred that one of the distinguishing features of the Mt Isa art sites was the proportion of figurative motifs. Citing studies to the south of Mt Isa in the Toko Range (Kelly 1968) and along the Mulligan River (Davidson 1983b), Franklin (1996, 146) suggests that the decline of figurative motifs is observable. Although she noted that crescents, lines, and circles are similarly dominant on some more southern sites, like those at Cooper Creek (Elkin 1949/50). Her conclusion was that the region was characterised by a mix of similarity and distinctiveness between the Mt Isa art, and the art of surrounding areas. On this issue, a broad scale study by David and Cole (1990) came to a very different conclusion. Based on Morwood’s data, David and Cole (1990, 802) concluded the Mt Isa art was consistent with earlier engraved styles across western and northern Queensland. Additionally, they suggested that the painted rock-art of the Mt Isa region clustered with that of Lawn Hill, Chillagoe, Central Queensland, and Olary in a homogeneous block. The conclusion of David and Cole was that some artistic elements could be observed over vast areas of the Australian continent. Their argument being that western Queensland was a ‘risky’ (p.789) environment, requiring the maintenance of extensive long distance social networks. After Gamble (1982), they suggest art played a crucial role in the maintenance of such networks and served as a medium through which affiliation could be reinforced (David and Cole 1990, 802-3). Davidson (1997, 218) has questioned these results by suggesting that the broad similarities identified by David and Cole may have more to do with their classification of anthropomorphic motifs than broad stylistic trends across regions. As is detailed further below, Ross (1997) has demonstrated that in northwest Queensland a number of stylistic elements are brought together when depicting an anthropomorphic form, and that these are unique to the region. Davidson therefore stresses that although the referent (humans) is similar, the symbolic conventions were completely different.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

66.

Chapter 3.

3.3.2.1 Anthropomorphs- Ross’ study. To explore the issue of whether the Mt Isa art was a distinctive unit, a comprehensive study of the anthropomorphic figures occurring in the region was undertaken by Ross (1997). The purpose of her study was to examine whether there was a regionally distinctive rock-art style in northwest Queensland, and if so, to determine its spatial and temporal distribution. Motivating the study were the social issues raised by Morwood (1978), in particular the social dynamics that led to regionally distinctive art styles. One of the clear deficiencies of previous studies had been the number of sites used in each analysis– two in the case of Morwood, and five for Franklin. In Franklin’s case there was also a bias for engraved sites over painted ones. Ross addressed these problems by analysing 60 art sites, and in doing so provided the first detailed look at the art of the region. The location of the sites studied by Ross, together with those studied by Franklin and Morwood, are illustrated Figure 3.12.

Figure 3.12. Location of art sites studied in previous research.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

67.

Chapter 3.

From the analysis that Ross performed, it became clear that anthropomorphic motifs form a regionally distinctive unit. Several lines of evidence indicated this: 1. The restricted spatial distribution of the motif. The majority of sites containing the anthropomorphic figures were confined to the Kalkadoon tribal area described by Roth (1897a) and Tindale (1974). This is illustrated in Figure 3.13. 2. The repeated use of basic design elements (although with considerable variation in the arrangement and form of these elements). 3. Direct and indirect evidence indicating that the motif has a recent temporal origin, most likely to within the last 1000 years. 4. The most elaborate anthropomorphic designs were painted in prominent locations where people were most likely to congregate such as cliff faces adjacent to waterholes.

Figure 3.13. Sites studied by Ross in the context of Roth's (1897), solid line and Tindale's (1974), dashed line tribal boundaries.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

68.

Chapter 3.

On the basis of this evidence, Ross suggests that the form and distribution of the anthropomorphic figures was consistent with a regionally distinctive social system. Her argument was that artistic styles play an important role in the mediation of social information. So that when a style is bounded and highly repetitive, as in northwest central Queensland, it indicates a social context for maintaining and portraying information regarding group affiliation. As Ross explained: By presenting a standardised image to outsiders, a distinctive rock-art style would have provided a means of emphasising the unity between the small localised bands which traditionally made up a regional group. Further, the territory of the resident group would be clearly marked so that, even in the absence of the owners, the outsiders would be notified that some modification in behaviour was required. In this way, the art acts as a mechanism which mediates between the producer group and the viewing group. … Viewed in this way, the anthropomorphic motifs can be seen as a ‘tool’ in the negotiation of social interaction. (Ross 1997, 143-4)

Ross hypothesised that an increase in extended trading networks over the last 1000 years would have resulted in greater pressure on the role of social interaction. As she states: ‘The implication of the co-occurrence of the emergence of a distinctive art style, the opening up of quarries within the region, the maintenance of a distinctive language, is that all could be interlocking components of a regional social strategy’ (Ross 1997, 143). As is consistent with other parts of Australia (e.g. David and Cole 1990), this is likely to have been a consequence of an increased regionalisation of Aboriginal behaviour. The distinctive nature of the anthropomorphic figure that characterises the region’s art assemblage then indicates that it likely functioned as a mechanism that supported the trend to regionalisation. The important finding about the anthropomorphic figures, is the contrast they form with the distribution and broad stylistic links observed for other figures occurring in the region. Ross (1997, 126) found that the most common motifs associated with anthropomorphic figures in sites were wavy lines (58% of sites), tridents (45%), parallel lines (38%), circles (35%), snakes (33%), concentric arcs (33%) and extended tridents (27%). In particular, concentric arcs, circles, and tridents are frequently represented in the art of neighbouring regions and throughout Queensland. For example, ‘series of arcs’ were observed in the Laura region, and lines and tracks were found by Franklin to be a recurring theme in many art sites across Queensland and central Australia. __________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

69.

Chapter 3.

Viewed through the distribution of anthropomorphic figures, Ross’ conclusion about the mediation of social processes lies in stark contrast to the conclusion of David and Cole (1990). David and Cole assumed that the art of the region displayed no trend to regionalisation and that the extended trade networks were evidence of this. What the analysis of David and Cole lacked however, was the knowledge of the distribution, form and distinctiveness of the anthropomorphic figures. Ross’ conclusions are therefore important in the context of David and Cole’s regionalisation argument because they demonstrate that the trend to regionalisation also took place in semi-arid as well as temperate environments like those on the Cape York Peninsula. Morwood (1985, 143) believed some engraved, non-figurative motifs were quite recent. Similarly, Ross suggested that the anthropomorphs (which are mostly painted) were probably less than a thousand years old (Ridges, Davidson, and Tucker 2000, Ross 1997, 67). If this is so, then the emergence of regionalisation appears later in the study region than in the Cape York where it occurred during the mid-Holocene (that is 4 – 5,000 years BP). Similarly, in Cape York regionalisation was associated with fragmentation of both social networks and artistic systems. On the other hand, in northwest Queensland it appears to have been a process of one area (possibly one group) seeking to distinguish itself from surrounding groups. In addition, the Mt Isa example is likely to have been closely linked with the distribution of a few important resources. Thus, the distinctive area/group maintained extensive long distance links, which the artistic system reinforced. The implication is that regionalisation, if it is a feature of the late Holocene, is likely to be more complex than originally envisaged by David and Cole.

3.3.2.2 Variation within the anthropomorphic style. In Ross’ (1997) analysis, despite the generalised features that characterise the anthropomorphic style, internally, there was a large amount of variation in terms of how the elements were brought together. For example, although many anthropomorphic figures were depicted without headdresses, those that did had many different types of headdresses. Ross also found that in some instances headdresses were depicted without anthropomorphs. Consequently, she concluded that few significant relationships appeared consistently between design elements.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

70.

Chapter 3.

No clear pattern emerged either when the number of anthropomorphic figures at different site types was examined. The average number present at waterholes, rocky outliers and rock shelters was very consistent (Ross 1997, table 4.25). A similar lack of pattern was found between the number of anthropomorphs present at a site and the range of associated motif types present (Ross 1997, 108). However, Ross found significant patterns when she categorised the anthropomorphs into two groups. The first category, what Ross (1997, 99) called Basic Motifs, were small, monochrome, and rarely contained other design elements like feet, hands or headdresses. Referred to as Detailed Motifs, the second group was large, usually bi- or polychrome, and had some or all of the decorative elements such as outline, headdress, background, hands or feet. The division of motifs into these groups was deemed to reflect the intuitive impression on the one hand that there were some motifs that displayed only the minimum visual requirements for illustrating the anthropomorphic style. On the other, some had Detailed decoration, which permitted greater freedom of expression, but within the confines of the anthropomorphic style. The implication of the complexity of the detailed figures was that they could communicate greater amounts of information. The two categories of anthropomorph showed marked differences in site context. Preferentially, Basic types were depicted in rock shelters, and Detailed ones at waterholes (Ross 1997, 109-10). This pattern was also reflected in the visibility of anthropomorphs at each site type as well. Ross found that 88% of highly visible motifs located at waterholes were Detailed figures. In addition, the presence of archaeological material was also found to be highly correlated with the presence of Detailed figures when they occurred in rock shelters. From these findings Ross (1997, 146) concluded that the representation of two categories of anthropomorph served to convey varying information in different contexts. She suggested that the occurrence of Detailed anthropomorphs at waterholes and highly visible locations indicated the portrayal of identity at places frequented by a larger number of people. Every Detailed figure is unique in some aspect of its form, suggesting that freedom in composition was used to convey additional information along with that associated with the anthropomorphic style.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

71.

Chapter 3.

3.3.2.3 Other motif types in the region. The other motif types most commonly associated with anthropomorphs in the sites Ross studied were lines (35 sites), tridents (27 sites), parallel lines (23 sites), circles (21 sites), snakes and concentric arcs (20 sites each) and extended tridents (16 sites). In this regard, Ross’ findings were similar to those of previous researchers in that geometric motifs were the most frequently depicted figures. Ross found no consistent pattern with the occurrence of motif types (other than anthropomorphs) when examined by site type. The exception to this was the occurrence of hand stencils, which occurred exclusively in rock shelters (Ross 1997, 113). Hand stencils were also an exception since they only occurred in the southeast corner of Ross’ study region (Ross 1997, 114). Ross suggested that ‘As hand stencils have not been recorded at waterholes or rocky outliers they may best be viewed as a localised practice specific to rockshelters rather than a part of the artistic system as a whole.’ (Ross 1997, 115). With the exception of one motif type, no significant association was found between the presence of motif types and reliable water. The exception was with fern figures, which were always associated with reliable water at the 15 sites in which it occurred. There has been some suggestion in other regions (Layton 1992, 155-7) that the fern motif is associated with ceremonial activity. Within the study region, the occurrence of similar fern motifs as headdresses on anthropomorphs suggests a similar interpretation. However, Ross found that the occurrence of anthropomorphs with fern headdresses did not show the association with reliable water that was observed when ferns were depicted on their own. Ross (1997, 119) offered no explanation for this trend, only suggesting that the association between the depiction of ferns and ceremonial activity sounded reasonable but was not testable.

3.3.2.4 Chemical analysis of the paints. Ross’ (1997) study showed that, while the anthropomorphs had a spatially limited distribution, many of the associated motifs were much more widespread. She also suggested that some components in the regional art assemblage such as barred circles,

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

72.

Chapter 3.

concentric arcs, panels of lines and other geometric motifs functioned differently. These motifs occur in neighbouring areas where ethnographic accounts and linguistic evidence record intermarriage and exchange. It seems likely that they served as part of a linking mechanism amongst interacting groups. If such functionality is assumed for the northwest central Queensland paintings and engravings, then this may lead into other hypotheses about the use and application of ochre. The symbolic value of ochre contrasts with other classes of traded items because it is known that it was directly linked with several forms of symbolic behaviour, most notably rock and body painting. In addition, there may have been an important connection between particular motifs and ochres obtained from particular locations. Particular ochres have been described by Aboriginal people as being more “powerful” than others (Kretchmar 1936), and in the Kimberley region, particular ochres have been linked to the power of some motifs (Clarke 1976)—the link between the motifs and the ochre source being their common mythology. Consequently, certain motif types may have been made more powerful through the use of particular ochres.

Figure 3.14. PCA results of pigment analysis.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

73.

Chapter 3.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

74.

Chapter 3.

One argument might be that a locally distinctive motif, like the anthropomorphs, would be more likely to contain ochre obtained from local sources. Other motif types—such as those which are more likely to have been associated with mythological stories—would be more likely to contain exotic ochres. To test such hypotheses, paint samples were collected from 14 motifs in the Selwyn Ranges, and others from local sources of ochre (Davidson et al. In Press). Examples of red and yellow paint were included, as were a range of motifs. The samples were analysed using the PIXE/PIGME technique (David, Clayton, and Watchman 1993), which measures the concentrations of about 25 elements. Principle components analysis (PCA) performed on the results of this analysis are shown in Figure 3.14. Figure 3.14 shows the results for both the red and yellow paints. The clustering of the paints from anthropomorphs around the local ochre sources would appear to support the hypothesis that local ochres were used for the locally distinctive motif. The compositions of paints from other motif types are outliers to the distribution of local ochres. These motifs include a snake figure, a dingo, a trident, and one sample from an anthropomorphic, but unique figure from the region. In addition, paints from single sites do not appear to cluster together. The paints from two sites located close to one source appear to contain ochre unlike that analysed from the source. Hence, proximity to a source may not be the only factor governing the choice of ochre used for painting. There are several reasons for displaying caution about these interpretations. A range of factors cannot, yet, be accounted for. Field survey confirmed that numerous ochre sources exist in the region, and that their chemistry varies, but far less than when compared to ochres from outside the region. For some sources, the chemistry is similar to the pigments found in some paints analysed here. This is not conclusive, but it begins to suggest that some paintings were done with local ochres, and some were not.

3.3.3. Open sites. Much of the archaeological work undertaken in the study region is concerned with cultural heritage assessment. The vast majority of evidence encountered in these studies __________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

75.

Chapter 3.

is in the form of stone artefact scatters. Consequently, open site locations and composition have been the focus of most archaeological discussion in the region. This section describes these reviews, and outlines the main conclusions relevant to the study of spatial distributions considered in the present study.

3.3.3.1 Site definition. Due to high archaeological visibility, ready availability of stone raw material and geomorphic stability, stone artefacts occur virtually continuously across the land surface throughout much of the study region. For example, in conducting a random sampling program in the Selwyn Ranges using 500 metre quadrats, Davidson (pers. comm.) found artefacts in every square that was surveyed. ‘Sites’ in this environment therefore represent increases in artefact density, rather than discrete units. In this context, Hiscock (1988b, 18) proposed that in western Queensland, a scatter of stone artefacts would constitute a site if: 1. More than 5 artefacts were present 2. They covered 2m2 or more in area 3. Average artefact density is more than 4x the average density of the background scatter 4. Average density was more than 1.0m2 Since the 2m2 area prevented the identification of knapping floors as sites, Davidson et. al. (1991, 29) suggested an amendment to Hiscock’s criteria such that if all the artefacts in the scatter were of the same raw material type, then Hiscock’s second criteria could be relaxed. Subsequently, Davidson and Fife (1994) modified the criteria again when using transect survey methods. In this scheme, a transect section became a site if it met Hiscock’s criteria, or if the density of artefacts was more than two standard deviations above the mean for those transect sections occurring within the same environmental unit. One of the problems with the continuous artefact distribution encountered in the study region is that there are inevitably many ways to define sites. Depending on survey strategy, the environmental context, and preference of the recorder, many different definitions of site have appeared in archaeological reports describing work conducted in

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

76.

Chapter 3.

the region. For this study, this issue was too extensive to address, since in many reports the criteria used to define sites was not even described. Consequently, for the purposes of this study, a location containing archaeological material was classified as a site if the recorder ascribed it this class. Although this inevitably introduces inconsistencies, it at least permitted sites to be distinguished from other location classes based on artefact density when these had been recorded.

3.3.3.2 Site locations. Almost universally, archaeologists who have worked in the region point out the importance of water in determining where open sites and artefacts occur (Alfredson 1991, 9, Bird 1997, 11, Brayshaw 1985, 7, Brayshaw 1995, 11, Davies, Grant, and Bowen 1996, 46, Lance 1994, 3). Border (1989, 48-50) for instance observed that in the region 50% of sites were located within 100 metres, and 80% were located within 500 metres of water. Similarly, in the Mitchell Grass Downs biogeographic zone, Border and Rowland (1990, 55) reported that 50% of all archaeological features occurred adjacent to creeks. Despite the importance of water, very little work has focused upon the specifics of variation adjacent to water, and along streamlines. What work has been done on this topic, suggests that stream junctions are important for containing larger sites (Hiscock 1988b, 58-9). Similarly, Ridges and Davidson (1996) found that sites at stream junctions were more likely to contain tulas, cores and backed blades, presumably as people were retooling. In contrast, artefact scatters adjacent to streamlines, but between junctions, were found to contain larger flakes, but at lower densities. The issue of water proximity was also found to be complicated by issues of quarrying and retooling in the south of the study region (Ridges et al. 2001). The survey area in this example comprised a broad low mesa, which contained numerous silcrete quarries (Davidson and Fife 1994). The only source of water on the mesa were gilgais. Although artefact density at the quarries, and at reduction sites near them was extremely high, sites with diverse raw material, and diverse artefact types were concentrated around the gilgais. It seemed that in this case, people were coming onto the mesa, camping at the gilgai, discarding their worn out stone tools there, and then retooling at the quarries

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

77.

Chapter 3.

before leaving the mesa. What this example showed was that distance to water was not necessarily the main driver of artefact density. It also suggested that behaviour near water was different from that away from water. This example also presented one of the few exceptions where artefact density is expected to be high away from water. Hiscock (1988b) and Lance (1994, 10) both emphasised that the dissected residual land zones were the only areas where artefact density would be high away from water due to the presence of stone quarries.

3.3.3.3 Environmental Units. The only other factor that has been reported to be important in determining the location, and density of sites and artefacts in the region has been environmental units. Lance (1994) and Davies (1996) used land systems as the basis of deriving predictive models in the western portion of the study region. Land systems were also used by Davidson and Fife (1994) for their estimates of artefact densities in the south of the region. Essentially, these models are based on summaries of existing archaeological data and information described in reports to derive estimated artefact densities, site densities, expected artefact types and raw material types in different land systems (for an example see Ridges and Davidson 1996, 14). Application of these models was found to work reasonably well at predicting the occurrence of artefacts and sites (Ridges et al. 2001), but less well at predicting their contents, which were found to be more variable (Ridges and Davidson 1996). Land units also formed the basis of Drury’s (1996) study in which he compared ten assemblages located in three land systems in the Selwyn area. Drury found that as well as varying artefact densities, the composition and form of assemblages varied depending on the land system in which they were located. The land systems comprised: a granitic landscape containing shallow soils and few shrubs; a land system with more diverse vegetation based on metasedimentary geology; and a land system located on the alluvial plains along a major river course. Drury found that across a range of flake variables, the assemblages occurring in sites in the granitic landscape demonstrated signs of more expedient behaviour. For example,

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

78.

Chapter 3.

flakes were smaller, less diverse and contained artefacts associated with extractive tasks. Each of the sites studied within the granitic landscape were classed as general artefact scatters. The sites occurring in the metasedimentary land system contained more tulas and backed blades. The assemblages in this land system also contained higher frequencies of retouched artefacts, and the artefacts were the biggest of all the three land systems. The alluvial land system contained assemblages that were generally very similar to those occurring in the metasedimentary land system. The main difference was the degree of rationing, which was higher than that observered in the metasedimentary land system. In the metasedimentary land system the characteristics of retouched and non-retouched artefacts were very different, whereas they had been similar in the alluvial land system. From these findings Drury (1996, 174) concluded that the assemblages in each land system were consistent with the kinds of subsistence possible in each of them. In the granitic country, where resources were scarcer, the assemblages reflected greater degrees of mobility, and few sites indicated extended occupation or revisiting. Sites in the other two land systems suggested extended occupation and the kinds of activities associated with base camps. Drury’s study is important in the context of spatial modelling using land systems because it demonstrated that, as well as variation in site and artefact densities in different land systems, there were also differences in assemblage composition. These assemblage differences were consistent with expectations about the nature of subsistence in different land systems. Although land units have proved useful in outlining archaeological variation in the region, they are limiting also. For instance, as was outlined above, land systems, and vegetation data were not available for the entire study region. The soil data was a possibility for using as a surrogate for land systems, but there remain problems with the generality of the approach which are discussed in more detail in the next chapter. Suffice it to say here that what has had to be assumed in the above studies is that artefact distribution and assemblage variability are more uniform within land units than between them. However, Ridges and Davidson’s (1996) findings of variation along drainage lines indicate that uniformity within land systems may not be a suitable approach to understanding regional archaeological variability.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

79.

Chapter 3.

3.3.3.4 Site contents. The most abundant forms of archaeological evidence in the region are stone artefacts. Stone artefacts have been the focus of several Honours theses in the region, and are the mainstay of many cultural heritage assessments. Despite this, no comprehensive review has been attempted in the region in order to outline where particular artefact types occur, their geographic distribution or the distribution of the raw materials that they were made from. Several case studies have been conducted in the Selwyn area that show some of the dynamics of stone artefact use and manufacture in the region (Drury 1996, Kippen 1992). However, due to the diversity of raw material sources in the broader region and the small area examined in these studies, it was not possible to project these findings to the rest of the region. Similarly, the studies by Hiscock to the north of the study region (Hiscock 1988a), and to the south (Hiscock 1988c), were too removed from the study region to provide any insight into processes operating within the region. Although a regional perspective of stone artefact distribution and occurrence is lacking, particular case studies in the region have flagged various processes that are likely to be important in understanding the regional perspective. The focus of this study is primarily regional distributions, so these processes are only presented in summary form since they are not the main focus of this discussion: •

The manufacture of axes in the northwest of the study region was involved in long distance transfer both across the region and throughout the Lake Eyre basin (Davidson, Cook, and Fischer 1992, Davidson et al. In Press).

•

Workshops for some artefacts types were apparent from the cache of tulas discovered by Hiscock (1988c). It is likely that such caches indicate the intention of manufacturing some artefacts was for trade and exchange.

•

Production of stone artefacts for intended distribution may also be reflected by standardisation of form. Hiscock (1988c) noted standardisation in the tulas that had been cached, while tulas found in other open sites have also been found to have standardised measurements also (Ridges and Davidson 1996).

•

Assemblages of stone artefacts are likely to be different in different environmental zones (Drury 1996).

•

The composition of raw material types and stone artefact types may also vary between drainage lines (Hiscock 1988b).

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

80.

Chapter 3.

•

The strategies used for manufacturing artefacts from a particular raw material are likely to be related to the distance from the source of that raw material (Kippen 1992, Ridges et al. 2001).

•

Raw material diversity is likely to reflect the extent and pattern of foraging strategies (Hiscock 1988a).

•

Occurrence of some artefact types may be discrete in the region since many retouched artefacts have been recorded in the Selwyn region (James 1993). Yet retouched artefacts were found to be comparatively rare in the west (Brayshaw 1985) and northwest (Kippen 1996) of the region. Similarly, axes are more common in the western (Knuckey 1996) and northwestern (Kippen 1996) parts of the region.

•

Metabasalt, which is most commonly associated with the production of axes, appears to have been also used for producing flakes in the northwest of the region (Kippen 1996).

•

Raw material composition of sites is likely to be primarily related to the nature of the local geology (Hiscock 1988a, Kippen 1996, 18).

•

Due to the nature of retooling near quarries, the area around quarries has been found to consist of a mixture of discarded artefacts at the end of their use-life, extraction zones and reduction areas (Davidson, Cliff, and Sullivan 1991, Ridges et al. 2001).

•

Proximity to source may not be the sole determinant of assemblage composition as Hiscock (1984) has suggested. Sites quite close to quarried outcrops of chalcedony and silcrete were still found to be dominated by chert in a survey conducted by Ridges and Davidson (1996).

•

Cores were found to be very rare in open sites (Ridges and Davidson 1996).

•

In the sites he examined in the Boulia area, Hiscock (1988b, 57) suggested that there were two types of assemblage, those dominated by blades, and those dominated by tulas.

The case studies where these processes have been identified have generally focused upon characterising how the processes operated. However, few of these ideas have been implemented into the regional pattern of stone artefact distribution or occurrence on a broad scale in the region. For this reason, one of the objectives of this study is to apply some of these ideas to understanding regional distribution patterns.

3.4. Spatial processes operating in the region. The above review of regional context raises several issues about regional scale processes in the region, which are summarised in the points below: 1. The complex conception of space evident in the rock-art. On one level, the art indicates that people identified themselves as unique- unique enough to reinforce the

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

81.

Chapter 3.

message with a proliferation of distinctive figures over the landscape. However, on a different level, the same people actively participated in the often very long distance exchange networks, and according to Roth, maintained at least economic links with all the people around them and often further afield. Similar links were observed in the rock-art as shared motif types between the study region and other artistic regions in Australia. 2. The maintenance of complex systems for managing several important resources. Some resources, such as water, seem to have been freely accessible, in most instances, to most people, and facilitated the movement of people throughout the region. Other resources such as axes, pituri and ochre appear to have become more important during the later Holocene as the social system appears to have become more complex. Certainly in the case of pituri, access and distribution was highly ritualised (Watson 1980), and particular ochres may have played important roles in rock-art depiction. 3. The movement of people. Roth mentions in his writings that by and large, so long as protocols were met, people were generally free to travel through whatever country they chose. The movement of people usually took place during winter when for several months virtually everyone went “walkabout”. This situation suggests complex spatial arrangements between “public” areas along the waterways, and areas of greater significance to local people who had knowledge about those places. Thus, although there was freedom of movement, it appears that not all space was considered equal. 4. The environmental variability of the region. The study region is not a homogeneous region in terms of resources. Stone for tool manufacture has a particular pattern of distribution, as do the places suitable for rock shelter formation and providing surfaces for painting. Similarly, the varied and complex nature of the geology means floral resources are not evenly distributed, nor were water resources. Consequently, it would appear likely that there were variations in adaptive responses in different parts of the region.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

82.

Chapter 3.

These points form the crux of the spatial context of this study. The following chapter develops a methodology for examining these issues in greater detail, which is then taken up in Chapters 5 to 7.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

83.

Chapter 4.

Methodology. Perhaps we should be less concerned with definition, and concentrate on the structure of interaction and interpretation of the variability of that structure. —Ian Hodder. 1978. The spatial organisation of culture. p. ix.

4.1. The methodological problem. In Chapter 2, it was suggested that understanding hunter-gatherer behaviour requires also understanding how the multitude of behavioural processes, and their material expressions, combine to produce archaeological variation. It was also noted that complicating the ability to make predictions about archaeological manifestations of behaviour was the many levels of variation that exist in human behaviour. Subsequently, a crucial aspect of understanding archaeological variation is the need to understand the many levels of processes that produce variation in the archaeological record. Addressing this issue is the crux of the methodological problem addressed in this study. Using anthropological evidence, Pickering (1994) has argued that some behavioural levels were not recoverable from hunter-gatherer settlement patterns because they were largely masked by the dominance of subsistence related behaviour. Pickering cites the high degree of correlation between the location of hunter-gatherer archaeological finds and aspects of the environment as evidence for the masking of anything other than

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

85.

Chapter 4.

subsistence related behaviour (Foley 1981a, Jochim 1976, Kohler and Parker 1986). However, Pickering’s analysis was based on a only a small amount of archaeological data, and was not analysed for any variation other than that relating to subsistence. In contrast, the position taken in this study was that the masking effect of subsistence is an aspect of the archaeology of hunter-gatherers that has been assumed but never adequately demonstrated (Gamble 1986, 299). Subsequently, this study focuses upon the variation and pattern within hunter-gatherer settlement patterns and explores both subsistence and other factors contributing to the variation observed. The issue of whether aspects of the social landscape that Pickering observed for the Garawa are preserved in the archaeological record comes down to an issue about how the archaeological record is believed to have formed. In assuming social behaviour is not reflected in the character of hunter-gatherer settlement patterns, many archaeologists also assume that settlement pattern is purely the result of adaptive behaviour (Binford 1968). It is tempting to make such an assumption given the success that functional interpretations of archaeological spatial patterning have achieved (Dunnell 1986, 39). However, since little research has been conducted into other factors contributing to spatial pattern in hunter-gatherer settlement patterns, it is problematic to assume that other social factors did not play a role in where activities took place. Part of the problem stems from the use of the term ‘settlement patterns’. For the mobile lifestyle that most hunter-gatherers adopted, the notion of settlement patterns would seem inappropriate. This is especially so given that the act of settling involves a sense of commitment and planning about where activities will subsequently be located. For hunters and gatherers, some degree of planning would clearly have been required in order to recover resources efficiently (Winterhalder and Smith 1981), but of equal importance was flexibility in order to cope with changing conditions (Smith 1988). Consequently, although planning may have been an aspect of where hunter-gatherers located their activities, commitment to long-term residence at those locations would appear lacking. In this regard, the need to have the location of activities carefully selected would appear to be more important in a sedentary context, since the cost of getting the decision wrong would be greater. In contrast, the flexibility necessary for the location of activities for hunter-gatherers would presumably permit a greater range of non-subsistence factors to come into play.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

86.

Chapter 4.

Certainly, Woodburn (1972, 205) has noted that the movement of people between camps was primarily motivated by social forces. In particular, changing camp involved changing the associations between people, and often this was a means of relieving tensions in the group. As Lee (1972, 181-2) also observes, although scarcity of water may draw people together to camp at waterholes, eventually it was tensions within the group that drove them apart again. Hence, although social factors were involved in where huntergatherers located their activities, we do not understand very well how they manifest themselves archaeologically. One aspect of this problem concerns focusing upon how environmental variables correlate with the location of hunter-gatherer activities. Doing so does not address any pattern of variation within the correlation. For example, if all hunter-gatherer activities were located near water, does it follow that all sources of water contain evidence of hunter-gatherer activities? Likewise, in how many places in the landscape could huntergatherer activities be undertaken with due consideration of the subsistence requirements of those activities? Given that there is potentially quite a broad area where huntergatherers may have located their activities successfully in subsistence terms, what factors then determine variation within those areas? Such questions raise a broader epistemological issue about whether any class of archaeological evidence can be assumed to reflect only restricted sets of behaviour. To be sure, some forms of behaviour will influence particular classes of archaeological evidence more than others. For example, the factors influencing the form of a painted motif will be dominated by different factors from those dominating the form of a stone artefact (Dunnell 1978). However, to assume that style has no influence on the form of a stone artefact (Wiessner 1983), or that painted figures were not functional in anyway (Vinnicombe 1972), is to deny any prospect for understanding behaviour as a system. This is not just a pedantic issue about the interpretation of archaeological variability (Binford 1972, 131-5), it goes to the very core of how archaeological methodology is derived. The problem stems from one of the important tenets of the new archaeology, which was the requirement to explain analogous similarities through empirical generalisations that account for the appearance of particular archaeological patterns under specifiable conditions (Binford 1989, 18). Even allowing for less positivist views of science (Binford

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

87.

Chapter 4.

1989, 69, VanPool and VanPool 1999), the importance of producing positive knowledge about the processes operating in the past has left an indelible impression in the minds of many archaeologists. The result has been that the derivation and verification of the driving factors of archaeological variability have become more important than the variability itself (Dunnell 1982). In Australia, this can be seen in the continued importance of regional archaeological programs that use a series of rockshelter excavations, some open site recording, and where available, rock-art analysis, to develop regional prehistories (David 1991b, Lewis 1988, Morwood 1984, Morwood and Hobbs 1995, Smith 1993, Veth 1989). These types of studies traditionally focus upon chronological evidence, through spatially sampling both sites and landscapes in a very selective ways (Davidson 1983a, Davidson 1999, 127-8). Although this provides the best available approach to understanding chronological change, it comes at the expense of understanding spatial variation in any detail. Contrary to Gosden and Head (1994, 113), who suggested that understanding social landscapes requires longer time frames, the proposition here is that any chronological change is dependent upon the spatial context of the sample in which the change is observed. Without understanding spatial variation thoroughly, chronological change in behaviour becomes indistinguishable from a spatial rearrangement of behaviour over the same period. The problem of understanding spatial variation thoroughly has also emerged as an issue in cultural resource management (Davidson 1983a). Many cultural heritage assessments are undertaken at very localised scales, and often in the context of rudimentary spatial contexts of regional archaeological variability. In such circumstances it is difficult for archaeologists to identify anything other than subsistence related trends because the spatial context necessary for assessing anything else is lacking. Consequently, Lourandos and Ross (1994) have noted that it was not until sufficiently comprehensive regional perspectives were developed in Australia that the issue of variability became a theoretical topic. In part, this is what promoted the debate in Australia about midHolocene change (Lourandos 1996, 17). Addressing these issues first requires identifying the links between the behavioural levels described anthropologically, with the levels in material evidence described archaeologically. Since the Land Rights movement began, Australian anthropologists

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

88.

Chapter 4.

have begun considering the many spatial levels in which Aboriginal behaviour varies (Peterson and Long 1986, iv-v). In contrast, archaeology continues to work with spatial levels that are derived from data (Clarke 1977). Consequently, it is often unclear how archaeological levels relate to human behaviour, and vice versa. There therefore continues to be much debate over whether archaeologists should be considering ‘sites’ or some other classificatory level when considering the behaviour of hunter-gatherers (Dunnell 1992). The methodological challenge in this context is to understand better the multiple levels that comprise the archaeological record of hunter-gatherers. If the diversity of approaches currently seen in archaeology (Trigger 1991) is any indication, then clearly the investigation of past human behaviour is a complex problem that can be addressed in many ways. Despite this, three issues emerge as being crucial to all archaeological research: chronological variation, spatial variation, and the levels at which both of these occur. No single methodology is capable of addressing all of these adequately. Archaeological methodology has tended to focus upon chronological variation, primarily because it has not always had the capability to address spatial variation comprehensively. However, there are several methodological advantages to the use of a spatial framework over a more traditional chronological framework for investigating regional dynamics of behaviour: Diversity of evidence: For the majority of archaeological evidence, location is easier to measure than antiquity, permitting the incorporation of a greater diversity and volume of archaeological evidence. Control of scale: With a set of data referenced by location, spatially sub-sampling the dataset simplifies exploration of variation due to spatial scale (Mueller 1975). Analytical tools: Geographic information systems (GIS) provide a powerful tool for managing and analysing archaeological spatial information (Green 1990). The utility of a GIS lies in its ability to manipulate the visual display of spatial data, and to describe the spatial relationships within a spatially referenced data set through an array of analytical functions (Berry 1993). Ethnographic analogues: A large volume of detailed anthropological descriptions exist for the spatial component of hunter-gatherer behaviour. Although there are problems with projecting these back into the past (Davidson 1988, 24-26), they form __________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

89.

Chapter 4.

a foundation from which to form hypotheses about archaeological spatial patterning. Understanding the ways in which hunter-gatherer behaviour produces spatial variation in archaeological finds requires clarifying how scale affects both behaviour and archaeological data. In turn, it also requires analysing spatial variation comprehensively. Therefore, scale and spatial analysis formed the cornerstones of the methodology adopted in this study.

4.2. Methodological components. This section outlines the main components of the methodology developed for this study. It begins with an outline of the most common approaches used in the study of huntergatherer spatial patterning. It then outlines how various aspects of these approaches were incorporated to address the issue of spatial variation and scale in this study.

4.2.1. Approaches to investigating archaeological spatial patterning. Archaeologists have adopted various approaches for analysing the archaeological spatial patterns of hunter-gatherers. These approaches have usually been employed to address specific issues at different spatial levels, as is summarised in Table 4.1. These levels concern: the relationships between a single activity and its immediate surroundings; the local level, which concerns the relationship between small sets of proximate activity locations; to regions, which are broad collections of activity locations. The inter-regional level concerns comparisons between regions. Not considered in this table are the spatial relationships within archaeological locations, although various methods have been employed to investigate this level also (Kroll and Price 1991). Each of the approaches listed in Table 4.1 are discussed briefly below.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

90.

Chapter 4.

Table 4.1. Approaches to examining archaeological spatial patterns. Approach

Immediate

Local

SCA

X

X

Off-site

X

X

Regional

APM

X

X

RAM

X

X

Inter-regional

CA

X

X

LA

X

X

SCA = site catchment analysis; APM = archaeological predictive modelling; RAM = resource acquisition models; CA = culture areas; LA = landscape archaeology

Site catchment analysis (SCA) is a collection of methods for exploring the relationship between human activity occurring at a place and the surrounding environment around that place (Bailey and Davidson 1983, Roper 1979, Vita-Finzi and Higgs 1970). In SCA the environment can refer to the ecological surrounds (as a radius or distance walked) of a site. Alternatively, it can refer to the wider cultural catchment from which the artefacts (and/or their raw materials) within a site originated. The distinction between these types of catchment has been referred to as site territory versus site catchment respectively (Roper 1979, 124). In hunter-gatherer terms, a site territory can be considered the equivalent of the range (Stanner 1965). In contrast, the catchment of a site can be associated with the extent of relationships maintained between an often sparsely distributed mobile society (Gould 1977). There are two kinds of objective encountered in SCA. One is to understand the relationship between site contents and the surrounding catchment (for example VitaFinzi and Higgs 1970). Such knowledge can be informative about the range of resources exploited at a site, and how the site was integrated in regional scale processes like trade. In contrast, a second objective is to distinguish the differences between assemblages brought about through variation in their respective site exploitation territories, and differences caused through other factors operating over broader spatial scales (for example Hunt 1992). With this objective, comparisons between site assemblages can be

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

91.

Chapter 4.

interpreted more meaningfully once the range of resources from which choices were made is also understood (Davidson 1983c). Off-site archaeology, like SCA, is not a single method, but a collection of methods grouped around a theoretical expectation about how hunter-gatherers discarded artefacts. The main purpose of adopting off-site approaches is to change the analytical unit from the site to the artefact (Dunnell and Dancey 1983, Ebert 1992). The use of artefacts as the analytical unit was designed to overcome the problem with identifying sites in areas where changes in artefact density form a cline rather than hard breaks (Ebert 1992, 55). As will be recalled from Chapter 3, this is a feature of the study region considered in this thesis. Based on anthropological descriptions of hunter-gatherer behaviour within a range, Foley (1981a) proposed a hypothetical archaeological model for the distribution of artefacts based on off-site principles. Many archaeologists in Australia employ this model, often implicitly, to describe the archaeology of Aboriginal behaviour at open locations within regions. However, the applicability of the model remains largely untested, especially at the regional level where it is assumed that regional artefact distributions are simply the sum of distributions within ranges. Due to the number of artefacts that can occur within a given landscape (Foley 1981a, 13), the application of offsite approaches has been mainly limited to the local level. However, the idea is implicit in dividing archaeological locations into classes other than the ‘site’. Archaeological predictive modelling attempts to describe the likelihood of archaeological features occurring at any given location (Kvamme 1988b). Although they can be applied to any scale, predictive models have found most application in cultural resource assessment at the local and regional scale (Kincaid 1988). Predictive models come in two forms depending on whether the primary logic is based on inductive or deductive statements (Altschul 1988, 63). In its inductive form, the approach involves deriving the correlation between locations containing the feature of interest and environmental variables defining where they occur. Due to their availability and general success, environmental variables are most commonly employed for this purpose (Ebert 2000, see van Leusen in Gaffney and van Leusen 1995). However, other variables such as proximity to social boundary have also been used (Whitley 2000). In their deductive form, predictive models are based upon expert knowledge and theories of hunter-gatherer behaviour.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

92.

Chapter 4.

These are used to derive statements about the kinds of behaviour, and subsequent archaeological pattern, expected in different ecological zones (Jochim 1976). Both approaches offer advantages and disadvantages. Inductive models tend to be empirical statements of correlation with little or no inherent explanatory value. Whereas, deductive models are inherently based on explanatory statements and potentially offer more satisfactory behavioural predictions. However, deductive models are much more difficult to derive because of uncertainties about the motivations for particular behaviour and the difficulty with verifying the resulting model. In contrast, correlative models are more readily assessed in terms of goodness-of-fit measures such as predictive accuracy. Inductive and deductive statements are mutually exclusive in principle (Boyd 1991, 4). However, in practice they are complementary since inductive statements are important for describing pattern, whereas deductive statements provide meaning for pattern (Altschul 1988, 8) and research programs frequently move between the two strategies. The objective of resource acquisition models is to understand the processes through which hunter-gatherers solved the spatial problem of acquiring and using natural resources. The issue has been explored in most theoretical detail through optimal foraging models (Winterhalder 1981) and diet studies (Meehan 1982). Despite what these studies have found about the decision processes of hunter-gatherers, these approaches have had little applicability for the analysis of archaeological data. An exception to this however has been stone raw material rationing models (Byrne 1980). These models are based upon the notion that as distance from source increases, the need to conserve stone raw material also increases, and leads to the adoption of different flaking methods for extending the use life of individual artefacts. Hiscock (1986) suggested that rationing of raw material is the most important factor determining the variation between stone assemblages at different locations. Resource acquisition models are better considered as a general approach rather than a specific method. However, as is the case with stone rationing models, the approach has had an important role to play in research design. In this respect, resource acquisition models are closely related to archaeological predictive modelling since they are both about situating where activities take place. However, in resource acquisition studies greater emphasis is given to the variation in the contents of locations rather than where they occur. Since resource acquisition models examine the relationship between human

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

93.

Chapter 4.

movement and resources, they have tended to be undertaken at the local and regional level. The culture area concept was proposed by Kroeber (1939) as a means of reducing the number of entities among the American Indian groups by showing similarities in the material culture of some groups and differences in others. The main premise in the approach is that broad geographic variation in the form of archaeological finds could be associated with ‘cultures’ (Clarke 1968, 285). However, it was subsequently found that there were many factors affecting geographic variation in the form of archaeological finds, which did not simply translate into cultural groupings defined anthropologically (Hodder 1978). Furthermore, there is often much more to spatial patterning than just geographic spread, and social processes operating outside cultural groupings can affect the distribution and form of archaeological finds (White and Modjeska 1978). The culture area concept has had some influence on interpretations of the distribution of Aboriginal items in Australia (Davidson 1938, Tindale 1968). Peterson (1976) tried to introduce aspects of the culture area concept by suggesting that drainage basins would be a suitable means of grouping material culture distributions. Subsequent analysis by McAdam (2001) has shown that for the most part, this is a useful way to organise the geographic distribution of variation in boomerangs, although there were significant degrees of variation within drainage basins also. Although there remain problems in the interpretation of what geographic distributions mean in behavioural terms, this does not deny the importance of examining distributions at the regional and inter-regional level. The final approach given in Table 4.1 is landscape archaeology. As with resource acquisition models, landscape archaeology is more of a general approach than method. As Knapp and Ashmore (1999) emphasised, landscapes mean many different things to different people, and in part, this is one of the defining characteristics of landscape archaeology. However, in terms of methodology, landscape archaeology focuses upon placing archaeological evidence in a wider context. For this reason landscape archaeology was placed in the broadest categories of spatial level in Table 4.1, although the approach can be used to interpret archaeological evidence at any scale. In landscape archaeology context can include the physical landscape, such as regional geomorphic and ecological factors leading to archaeological visibility and hunter-gatherer past land-use (Rossignol 1992, 4). It can also include aspects of the way humans project culture onto their

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

94.

Chapter 4.

environment,

creating

cultural

landscapes

(Crumley

and

Marquardt

1990).

Methodologically, the contribution of landscape archaeology is the importance it places upon looking at diverse strains of archaeological evidence, at multiple scales, in order to construct a holistic context for interpreting archaeological variability. The approach taken in this study drew upon aspects of all the approaches described above. The main focus of the methodology was to elucidate the processes operating within the study region, and consequently, it concentrated upon aspects of archaeological predictive modelling and resource acquisition models. The approach can be divided into two aspects, a component that focused upon spatial patterning, and another that examined patterning resulting from changes in scale. Each of these aspects are discussed below.

4.2.2. Spatial analysis. The main issue with the spatial analysis performed in this study concerned the behavioural processes operating within the region. The purpose of choosing a regional focus for investigating the dynamics of hunter-gatherer behaviour was because it was at such a scale that the interaction between many of the important components of huntergatherer behaviour could be observed. As Gamble (1986) also noted for his review of the Palaeolithic settlement of Europe, the importance of examining hunter-gatherer regional behaviour was that it encompassed: …the determinant features of the environment, to which groups must adapt, but also the continual process of social reproduction which specifies that the habitat shall be exploited according to the principles of hunter-gatherer formation in order to sustain and reproduce social existence. (Gamble 1986, 31)

This study also shared other elements to Gamble’s study, such as the focus on two main aspects of hunter-gatherer behaviour: settlement patterns and the production of art. However, from here the methods adopted in this study diverge from Gamble’s because of the different nature of the archaeological database and the size of the region. The size of the region in this study was only one sixteenth the area considered by Gamble. In addition, Gamble’s (1986, 138) study focused upon the excavated evidence from around 170 key sites in Europe, whereas this study comprised nearly two thousand locations, the __________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

95.

Chapter 4.

majority of which were open locations containing artefact scatters. Subsequently, the focus in this study was purely spatial since there was limited scope for investigating chronological variation with the number of excavated sites in the region (see Chapter 3). Similarly, all the art data in this study comprised art depicted on rock surfaces rather than the excavated items studied by Gamble. Consequently, the methodology adopted in this study is necessarily different from that of Gamble. The excavated material that Gamble studied was taken to be indicative of the way hunter-gatherers in Europe were subsisting in their environment at particular times. In contrast, the approach taken in this study was to use the large number of open locations, and smaller area, to investigate variation in subsistence pattern throughout the region. Gamble examined variation in subsistence patterns also, but did so on the basis of environmental provinces (Gamble 1986, 72). No environmental sub-divisions were used in this study, and instead variation produced by local context was investigated as the main driver of variation in archaeological distribution. Local context in this sense included environmental variables and factors such as trade and the production of rock-art. Since the area was smaller, and the number of locations higher, this study was better suited to the incorporation of off-site principles. However, this did not extend to adopting the artefact as the primary analytical unit as Dunnell (1983) advocated. Instead, the primary analytical unit was the archaeological location. Archaeological locations were defined as places containing archaeological features. The location was considered a dimensionless point with attributes extending from the archaeological features recorded at that place. In some cases, the archaeological feature comprised a single artefact, in others it was thousands of artefacts, in others it was rock-art. It is acknowledged that representing archaeological features as dimensionless points ignores variation in the size of archaeological features. However, since the recording of the majority of locations in the region was based upon presence or absence measures of stone artefacts, points were thought to better reflect this approach at the regional level. Adopting locations as analytical units permitted a greater expanse to be modelled in offsite terms. For this purpose, the primary method employed was predictive modelling. However, whereas predictive modelling has mainly been used to model the distribution of ‘sites’ in a cultural resource management setting (Ebert 2000), the approach was used

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

96.

Chapter 4.

here to examine several different kinds of archaeological location type. Hence, predictive models were developed for: 1. sites; 2. artefact scatters not classified as sites; 3. individual finds; 4. various sub-classes of locations based on their contents such as art sites; 5. locations with particular artefact types; and 6. stone raw material types. Through the development of these separate predictive models, the particular characteristics of different aspects of behaviour could be understood within the context of an overall settlement pattern inferred from a model describing the location of all archaeological features combined. These comparisons formed the principal approach for examining variation in spatial behaviour. In this sense, the prediction maps produced using the models formed continuous surfaces that were in keeping with the concept of off-site archaeology. However, each of the mapped surfaces was slightly different due to the types of behaviour they reflected. Adopting this approach involved utilising predictive modelling in quite a different way from its more common application in cultural resource management. Its application in this study can be considered as a data visualisation tool. The predictive models developed in this study were all of the inductive type. That is, they were all based upon correlations between the position of archaeological locations and their environmental context. Hence, when the models were used to predict the probability of archaeological locations across the region, the resulting map gave an indication of the spatial relationship between archaeological location types and the variables used to produce the model. The prediction maps do not inherently explain any aspect of behaviour, but represent a useful tool for describing predicted archaeological distributions from which behaviour might be inferred. The modelling of location attributes such as stone artefact and raw material type, required other factors (along with environmental variables) since it was known a priori that non-environmental factors were important in determining their distribution. For instance, the occurrence of artefacts made from particular raw materials at Lawn Hill, to the north of the study region, was known to be dependent upon proximity to the source of

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

97.

Chapter 4.

these raw materials (Hiscock 1988a). Hence, due to the importance of raw material rationing arguments in other studies within the region (James 1993, Kippen 1992), proximity to stone raw material source was included as a variable for the stone artefact and raw material type models.

4.2.3. Spatial patterning of rock art. Unlike the analysis of stone tools, the spatial analysis of the attributes of rock-art sites necessarily involved a different approach due to the topology of how rock-art occurs in the landscape. Whereas stone tools can potentially occur at any location within the region, the occurrence of rock-art was highly constrained by the occurrence of rock surfaces suitable for depicting rock images. Away from areas where suitable rock surfaces occurred, there was no chance that rock-art sites would occur. Thus, although it was applicable to treat stone artefacts with an off-site approach, the same was not also applicable to art sites, which only occur as ‘sites’. Hence, although probability models were suitable for producing probability surfaces describing which locations with suitable rock surfaces were more likely to contain rock-art, they could not be used appropriately to describe the regional distribution of rock-art figures. Consequently, the analysis of rock-art distributions necessarily involved the analysis of point patterns. The point pattern analysis of the rock-art sites involved the analysis of three aspects of the art. One was the nature of the distribution of different figures depicted in sites. Another aspect involved comparing the collections of figures observed at different sites. The third aspect compared the position of each art site to all other art sites containing similar types of figures. Each of these aspects involved analysing different characteristics of point patterns, which are outlined below.

4.2.3.1 Figure distributions. For the distribution of different figure types, it was necessary to compare the point distribution of every figure type recorded in the region. In all, this involved comparing 160 point distributions. To simplify the comparison of these distributions, plots were composed using two pieces of information describing each distribution. The first piece of __________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

98.

Chapter 4.

information was the geographic centroid of each distribution, calculated as the mean X and Y co-ordinate of all points in the distribution. Figure 4.1 presents an example of a rock-art distribution plot (which will be presented again with full discussion in Chapter 6, Figure 6.7), where each point represents the geographic centroid of its respective distribution, relative to the geographic centroid of the whole dataset as indicated by the axes. A point which plots in the lower right hand corner of the plot indicates that figure type predominately occurs in those art sites occurring in the southeast of the region.

Figure 4.1. Example of a rock art distribution summary plot.

The second piece of information concerned the pattern of distribution, which described the degree to which the points in the distribution occurred randomly, regularly spaced, or were clustered. The derivation of spatial pattern for each distribution, involved performing a Nearest Neighbour Analysis (NNA) on the locations containing each variable (Hodder and Orton 1976, 38). NNA is calculated by dividing the average distance from each point in a sample to its nearest neighbour, by the expected average nearest neighbour distance given the number of points and the size of the region (Pinder, Shimada, and Gregory 1979, 431). The NNA in this study utilised the method developed by Clark and Evans (1954) and incorporated the adjustment for edge effects developed __________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

99.

Chapter 4.

by Donnolly (see Hooge and Eichenlaub 1997). The analysis was applied using a GIS tool developed by Hooge and Eichenlaub (1997), and for each analysis the entire study area was set as the region. NNA returns values in the order of 0 to 2.15 (Pinder, Shimada, and Gregory 1979, 431). A value approaching zero tends to be a clustered spatial pattern. Values close to one are spatially random and greater than one reflects a regularly spaced pattern. In the example plot in Figure 4.1, along with position as determined by its geographic centroid, each distribution was ascribed a symbol indicating its spatial pattern calculated by NNA. Since each distribution came from a larger population of locations (that is all rock-art sites), the NNA result of each variable was subtracted from the NNA for all sites in each dataset. It was these values that were used to assign symbols in the example in Figure 4.1. In Figure 4.1, the black dots indicate variables that had a more regular distribution, and the open circles a distribution that was more clustered than the overall data set. The stars indicate a distribution that was approximately the same as all locations in the dataset. Each figure therefore contains two pieces of information, the relative distribution of each variable and an indication of its spatial pattern relative to all sites in the respective datasets. From these two pieces of information, the pattern of distribution of many rock-art figures could be assessed from a single diagram.

4.2.3.2 Similarity of rock-art between sites. The other approach involved examining the assemblages of figures at different art sites and comparing the similarity between different art sites. This approach is used in Chapter 6, section 6.5. The comparisons were made with principal components analysis (PCA). PCA is a mathematical technique that describes the relationship of variables in multivariate space in order to estimate the latent components that underlie the data set. It is an exploratory technique designed to reduce a large set of variables into a smaller number of components that account for the majority of variance (Tabachnick and Fidell 1996, 635). Each component derived by PCA is a linear and orthogonal combination of variables, such that the first component accounts for the largest proportion of the variance, with successive components being combinations of variables that are uncorrelated with preceding components.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

100.

Chapter 4.

In theory, as many components as there are variables can be derived from a dataset. However, in this study only the first three were considered because further components accounted for only a small percentage of the total variance. On the advice given in the SPlus manual (MathSoft 1999b, 10) the PCA was performed using a covariance matrix. This was on the basis that the presence and absence data collected about the rock-art represented an unscaled data set where all variables were of the same order of magnitude. Cluster analysis was then used to determine any groupings of sites indicated by the PCA component scores. The method employed for this purpose was the partitioning around medoids algorithm available in S-Plus. This is a robust5 method which uses a dissimilarity matrix to derive medoids6 for each group (MathSoft 1999b, 74). The method assigns records to one of k groups based on their proximity to a medoid (MathSoft 1999b, 75). For the partitioning around medoids algorithm, the user specifies the number of groups the algorithm will attempt to place the data into. The choice of the number of groups to use is based on trial and error to see which number produces the largest separation between groups (MathSoft 1999b, 68). The degree of separation was assessed via silhouette plots, which indicate the average dissimilarity within each group (MathSoft 1999b, 77). For each silhouette plot, S-Plus calculates the average silhouette width for all groups. As the average silhouette width value approaches one, the degree of separation between groups is maximised (MathSoft 1999b, 78). Using this process, some of the subjective nature of how many groups to assign was avoided. Two to ten groups were examined for each data set.

4.2.3.3 Links between sites with common figures. A second aspect of the collection of figures depicted at art sites was to examine where else in the region each of the figures depicted at a site occurred. The distribution of

5

Robust in this sense refers to a method that is less sensitive to the effect of outliers.

6

A medoid is similar to the centroid for a group, but where a centroid is based on the average centre calculated using

Euclidean distances, medoids are calculated using dissimilarity distances.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

101.

Chapter 4.

figures depicted at an art site was used in this study to examine whether there was a preference for depicting figures with a particular distribution, in particular sites. The depiction of a particular figure at a site served to potentially link that site to all other art sites where that figure was also depicted. The use of links in this regard shares some common elements with Gamble’s (1986, 335) examination of stone raw material sources and Hodder’s (1976, 188-191) analysis of pottery sources. The difference in this case is that the links describe relationships between sites with similar contents rather than links between sites and the source of the items they contain. Otherwise, the methodology is essentially the same. The use of this method is taken up in Chapter 6, section 6.6. The process of examining links between art sites involved listing all the pairs of sites for each motif and anthropomorph type where the occurrence of that motif or anthropomorph type was greater than one. Each list contained a “from” and “to” site, with associated co-ordinates, from which distances and bearings were calculated. Using the co-ordinates of both sites in each pair, a unique identity number was assigned to each pair of sites. When all lists were combined from each data set, the unique identity numbers for each link could be used to determine how many links existed between each pair of sites. Likewise, using the “from” and “to” site information, a distance, bearing and frequency profile could be constructed for the links involving each site. The two datasets where link analysis was performed derived about 15,000 and 18,000 links respectively. A visual display of the links derived for each data set was derived in the GIS by connecting lines between the pair of points defining each link (see Figure 6.21 and Figure 6.22).

4.2.4. Predictive modelling methodology. Since predictive modelling involves some relatively complex statistical and GIS procedures, it is necessary to explain in some detail how these were applied in this study. As was mentioned above, the predictive modelling technique used in this study was an inductive method using the correlation between the location of archaeological finds and the particular environmental characteristics of these locations. Since most predictive models begin with environmental variables, these are used by way of example here. However, there is no reason why the variables should be limited to just environmental variables, like slope, vegetation or distance to water. Indeed, some of the models in this study also included proximity to geological sources of stone used for making artefacts. __________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

102.

Chapter 4.

There are a number of ways of using the correlation between the location of archaeological finds and their environmental context to produce predictions about the probability of archaeological finds in any part of a region (Altschul 1988). For instance, a model can be developed quite simply with a set of environmental layers and intersecting these in a GIS to extract only those environmental units overlapping with locations containing archaeological locations. Probabilities of the occurrence of archaeological features is then calculated using the density of sites (for example Rowland, Border, and Smith 1994) or artefacts (for example Hall and Lomax 1996) located in each land unit type. However, the limitation of this approach is that it requires all continuous variables, like elevation, to be converted into categorical variables, and hence the results depend upon how these categories are defined. A further problem is that the models assume that the density of archaeological features is uniform within each zone produced through the intersection in the GIS. A more useful approach is to construct a regression model describing the relationship between the environmental characteristics of where archaeological features occur and the environmental characteristics of where there are no archaeological features. The utility of forming a regression model is that it can then be used to calculate the probability of an archaeological feature occurring given any combination of environmental characteristics. A further advantage of using regression approaches is that a mixture of continuous and categorical variables7 can be used to derive the models. However, the difficulty with the approach is that it involves the use of multivariate regression, the mathematics of which is quite complicated. Another issue with the regression approach is the source of locations that do not contain archaeological features. Ideally, these should be derived from survey that has confirmed the locations contain no archaeological evidence. However, although non-archaeological location data was available in this study, they were geographically restricted to the few archaeological surveys where such information had been recorded. Hence, they did not

7

S-Plus incorporates categorical variables into regression models through the use of dummy variables (MathSoft 1999a,

42, see also Warren and Asch 2000, 19). By default, S-Plus uses the Helmert method for calculating coefficients of the linear combination of dummy variables in a regression model (MathSoft 1999a, 42). This default method was used in all models incorporating categorical variables in this study.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

103.

Chapter 4.

form a very representative sample of the total environmental variation in the region. For this reason, a random set of points was generated for use in the regression models. It is acknowledged that there is the danger that some of the random points may actually represent archaeological features. However, the purpose of using random points was to compare the selective use of particular parts of the environment by people, to the overall distribution of environmental features. Thus, random points provided an indication of environmental variability unbiased by human selection of activity location. Forming a regression model for a binary outcome like the presence or absence of archaeological evidence requires a special adaptation of the standard least squares regression technique. A normal least-squares model fits a straight line through a set of points such that the distance from the line to any point in the dataset is minimised (Shennan 1988, 124). Figure 4.2 shows an example of fitting a regression curve to a data set where the dependent variable is binary. In both panels of Figure 4.2, the X-axis represents a continuous variable of interest such as elevation. The Y-axis represents a dependent variable with a binary outcome (indicated by a value of 0 or 1), such as presence of absence of archaeological evidence. As the top panel of Figure 4.2 illustrates, a least squares model fitted to a binary data set produces values for Y that can range from minus infinity to plus infinity. However, since the outcome is binary, the outcome of Y must lie between 0 and 1. Thus, for binary datasets, a least-squares fit is inappropriate. The alternative is to fit a logistic model, which has a different mathematical form to the least squares regression, and is constrained to produce values from 0 to 1. In a least squares model, the form of the regression line is usually: Least squares regression:

Y = Xb + e

(1)

Where Y is the outcome of interest, X is a measured variable, b is a coefficient for the measured variable, and e is a constant. In logistic regression, equation (1) is modified to take the following form (where b and e remain the same): Logistic regression:

Y = exp(Xb + e) / (1 + exp(Xb + e))

(2)

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

104.

Chapter 4.

Figure 4.2. Comparison between linear and logistic regression curves.

The resulting logistic curve is illustrated in the lower panel of Figure 4.2. It can be seen in this graph that when X is less than –2, there are no observations of Y equalling 1, and the curve correspondingly approaches zero. The converse is also true for values of X greater than 2. When values of X range from –4 to 4, values of Y predicted from the logistic regression curve range from 0 to 1, as would be expected for a binary outcome. Logistic regression is therefore a more appropriate method for fitting a regression curve to data sets where the dependent variable is binary. However, rarely explored in archaeological predictive modelling, are the limitations with logistic regression brought about by the nature of archaeological data. A logistic regression works best when there is a relatively clear separation between the outcomes of Y for the values of X, as was the case in Figure 4.2. For this reason, one of the common assessments in archaeological analysis of the suitability of variables for logistic regression is the degree to which they differentiate archaeological and random locations (Warren and Asch 2000, 14). However, as was the case in this study, there can be considerable overlap between the archaeological and random distributions, as is shown in the frequency density plots in Figure 4.3. Although it is relatively clear in Figure 4.3 that __________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

105.

Chapter 4.

the archaeological locations occupy a narrower range of elevation values than random locations, the extent of overlap makes it inappropriate to fit a logistic regression to these data.

Random locations

0.008 0.006 0.004 0.002 0.000

Archaeological locations

0.008 0.006 0.004 0.002 0.000 100

200

300

400 Elevation

500

600

Figure 4.3. Density plots for archaeological locations and random points.

The result of fitting a logistic regression to the data in Figure 4.3, is shown in Figure 4.4. As the degree of overlap between the datasets increases, the form of the logistic regression becomes more like that of a least squares. At higher and lower values of elevation, the respective curves take the same form of those illustrated in Figure 4.2. Thus in this example, a logistic regression performs no better than a least squares model. Conversely, on these grounds, elevation would normally be a poor candidate as a variable for a logistic regression model.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

106.

Chapter 4.

1.0

0.9

0.8 Least squares 0.7 Location probability

Logistic 0.6

0.5

0.4

0.3 Generalised additive

0.2

0.1

0.0 100

150

200

250

300

350

400

450

500

550

Elevation

Figure 4.4. Comparison of different regression models.

The limitation of the logistic method in this example comes about because of the linearity of the relationship between the response and the predictor. Despite the visual form of the curve, the logit of Y is a linear function of X, as illustrated in equations (3) and (4). Rearranging equation (2):

Y / (1-Y) = exp(Xb + e)

(3)

Taking the logs of equation (3):

log[Y / (1 – Y)] = Xb + e

(4)

Since it is reasonably clear from Figure 4.3 that there is a difference between the range of elevation values for archaeological locations compared to random locations, it follows that a non-linear form of regression is required for this variable. The statistical package used for deriving the regression models in this study, S-Plus, offers a modified form of the logistic regression model, generalised additive modelling (GAM), which permits nonlinear regression on binary data. The form of a generalised additive model is essentially the same as in equation (4) except that rather than obtaining the logit of Y ( log[Y / (1 – Y)] ) from a linear model of X, X is first modelled using a non-parametric link function (MathSoft 1999a, 326). In S-Plus, the non-parametric function can take two forms of a non-linear model. Both are essentially the same, in that they apply regression successively over small sections of the data range

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

107.

Chapter 4.

for X, and then combined the successive regressions into a single model. The LOESS function in S-Plus applies a linear regression over each small section, whereas the SPLINE function applies a second order polynomial function. The LOESS function generally produces a more volatile curve that is responsive to small changes in Y along each section of X, whereas the SPLINE function produces a smoother curve indicating the overall trend. The differences between the functions are illustrated Figure 4.5.

1.0 Spline model Loess model

0.9 0.8

Probability

0.7 0.6 0.5 0.4 0.3 0.2 0.1 0.0 0

1

2

3

4

5

6

7

8

9

10

11

12

Slope (degrees)

Figure 4.5. Comparison of spline and loess non-linear functions.

Since the majority of the environmental variables utilised in this study demonstrated problems like that shown in Figure 4.3, all the predictive models applied in this study utilised a GAM with the spline option for logistic regression. A spline function was used in order to derive the main trends in each model, rather than the particular nuances of an individual data set potentially brought about by sampling bias. Although it has been rarely explored in archaeological modelling, the binary outcome of presence or absence of archaeological features is only one of the forms of regression from which spatial predictions can be made. Other common distributions, Gaussian (normal) and Poisson (for data based on counts) can also be applied using a GAM in S-Plus. The difference with these models lies in the observed distribution for Y (it is either expected to follow a Gaussian or Poisson distribution depending on the nature of the data). Such

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

108.

Chapter 4.

models were derived in this study for flake length (using a Gaussian distribution) and artefact frequency (using a Poisson distribution). An important difference in these models is the absence of non-archaeological data. Since they are predicting attributes of archaeological locations, not their presence or absence, a comparison dataset was not required. A second set of procedures was required to produce a map of predictions from each model, as well as deriving the input tables for each model. All these operations were performed within the GIS. The environmental variables used in this study were stored in the GIS as raster layers, where each layer is stored as collection of small contiguous square grid cells covering the extent of the region. In the GIS, about 600,000 grid cells that were each 277 metres long on a side, represented the region. Along with locational information, each grid cell had an attribute corresponding to the data set it describes (for example elevation). By intersecting the archaeological and random locations with the raster layers held in the GIS, tables were constructed containing the environmental information for all points. An example of such a table is shown in Table 4.2. Tables such as these were used as input for the regression model within S-Plus. By combining the information from all layers in the GIS, a second table was generated like that shown in Table 4.3. The number of rows in such a table equalled the number of grid cells covering the region, and the X and Y coordinates indicated the centroid of each cell. Using the model generated from a regression on a table like Table 4.2, predictions could be produced for the GIS data as listed in a table like Table 4.3. The predictions were performed within S-Plus, and produce a new field (labelled the ‘model’ field in Table 4.3) which indicated the probability that the grid cell would contain the archaeological feature of interest. With the predictions calculated, a table like Table 4.3 was then exported from S-Plus back to the GIS and used to form a new raster layer where the attribute of each grid cell was based on its predicted probability value. Using the display tools within GIS, the grid cell attributes of the resulting layer could be displayed in greyscale to produce a map of the probability of archaeological features occurring at any location.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

109.

Chapter 4.

Table 4.2. Example of an input table. Value

Type

X_coord

Y_coord

Slope

Elevation

Aspect

Soil

1

Archaeological

314674

6754320

0.321

453

25.322

3

1

Archaeological

319736

6753876

1.234

398

123.345

4

0

Random

310462

6752407

1.324

378

327.463

2

0

Random

315384

6759482

0.982

404

198.734

5

Table 4.3. Example of table used for prediction. X_coord

Y_coord

Slope

Elevation

Aspect

Soil

Model

314000

6754320

0.321

453

25.322

3

0.678

314277

6754320

1.234

398

123.345

4

0.763

314544

6754320

1.324

378

327.463

2

0.231

314831

6754320

0.982

404

198.734

5

0.316

The purpose of using predictive modelling in this study was to explore differences in behaviour. Subsequently, it was important that all models derived from a single data set were comparable. To ensure this, each model derived for a given data set utilised the same set of random locations. The number of random points combined with each data set equalled the total number of archaeological locations in that data set. Thus, when models were constructed on subsets of the database, the same random locations were incorporated regardless of how many archaeological locations were represented by the subset. Hence, a model derived to describe the occurrence of all archaeological locations in the region utilised the same set of random locations as a model derived for rock-art sites in the region. This was to ensure that the variation in the probabilities predicted by the models was due to differences in the archaeological data and not between different sets of random points. Since the main objective of this study was to compare models rather than to address the accuracy of individual models, none of the models have been tested with new or withheld

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

110.

Chapter 4.

archaeological data in the manner suggested by Rose and Altschul (1988). Instead, to ensure that all models were producing significant results, the models were tested with χ2 (Chi squared). For the binary regression models developed in this study, the χ2 were calculated using 2x2 contingency tables based on the number of archaeological and random locations predicted with a probability greater or less than a threshold probability level. An example calculation of this kind is provided in Table 4.4 where the probability threshold was 0.5. In this example, the predicted accuracy of the model is also included, which is the percentage of archaeological locations having a predicted probability greater than a threshold value of 0.5.

Table 4.4. Example of a 2x2 chi squared test for a model outcome. Archaeological locations

Random locations

TOTALS

Number > 0.5

1206

444

1650

Number < 0.5

589

1351

1940

TOTALS

1795

1795

3590

χ2 = 651.206; P < 0.001; Predicted accuracy = 67%

Since the number of random points used for all models was constant, it follows that as the size of the archaeological subset became smaller, the predicted probabilities of the resulting model would fall. This is consistent with the probability of finding rare archaeological features being correspondingly small. However, this presents a problem for model assessment since for rare archaeological features, the probability of all known archaeological features may not exceed 0.5 in any instance. The issue of what is an appropriate threshold value with which to assess model accuracy, especially with rare archaeological features, therefore becomes a problem. The threshold of 0.5 used in the example above would only be expected when the number of archaeological and random locations used to generate the model is approximately equal.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

111.

Chapter 4.

100 90

Archaeological Random

Percentage cases over threshold value

80 70 60 50 40 30 20 10 0 0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

Threshold value

Figure 4.6. Determining threshold value for regression models.

To overcome this issue, the threshold used to assess each model was derived from plotting the percentage of cases over a range of threshold values for both archaeological and random locations, as is illustrated in Figure 4.6. Where the curve for random and archaeological features intersects is taken to be the optimum threshold value for assessing the model (Kvamme 1988b, 392). In the example in Figure 4.6, it can be seen that the threshold value is slightly less than 0.5, and can be much lower when the ratio of archaeological to random locations is small. Hence, for each model, four pieces of information are presented: 1. TV : threshold value obtained from the intersection of threshold curves for archaeological and random locations; 2. χ2 : Chi squared calculated using a 2x2 contingency table based on the proportions of locations over and above the threshold value; 3. P: the probability level at which the χ2 result was significant; 4. AP : the predicted accuracy for archaeological features at the calculated threshold value. As a general rule, if the subset represented fewer than 10% of the number of locations of the entire data set, a model was not formed for that subset. None of the models

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

112.

Chapter 4.

presented in this study produced χ2 results that were not significant at the 0.01 probability level, at the threshold level calculated for the model.

4.2.5. Scales of analysis. Along with the spatial analysis, the other important component of the methodology involved the analysis performed at different scales. In the past, studies have emphasised the use of data from multiple scales (Gamble 1986, 26), or like Lourandos (1996, 15) stressed the importance of ensuring analysis was performed at a single scale. In this study, the analysis performed at different scales was an explicit component of understanding the variation of behaviour within the region. For this reason, scale was examined in this study in two ways: through classificatory scale, and spatial scale. The classificatory scale involved investigating the spatial pattern of several classes of archaeological evidence. These classes are summarised in Table 4.5.

Table 4.5. Summary of classificatory levels used in this study. Broadest category

Mutually exclusive categories

Categories based on contents of locations

All archaeological features —>

Individual find —>

Raw material type, artefact type

Open scatter —>

Raw material type, artefact type

Open Site —>

Raw material, artefact type

Art site —>

Motif types, anthropomorph form

Quarry —>

Raw material type

Three levels of classificatory scale were employed in this study. The broadest categorical scale comprised all archaeological locations undifferentiated by the type of archaeological features they contained. Following from this, the locations were divided into mutually exclusive groups, based on the presence or absence of key archaeological features. For

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

113.

Chapter 4.

example, if a location contained rock-art, then regardless of what other features it contained, it was classed as an art site. Similarly, if the location contained evidence of quarrying, it was classified as a quarry. The division of open locations into individual finds, open scatters and open sites was based on the frequency of artefacts at the given location. If it was a single find, then the location was classed as an individual find. All other open locations were classified as open scatters, unless in the recorder’s opinion or through the occurrence of features such as scarred trees, then the open scatter was classed as an open site. Along with these mutually exclusive groups, a further level was utilised, depending on the mutually exclusive group they derived from. For open locations, which were defined by the presence of stone artefacts, the division was based upon raw material type and artefact type. For art sites, the division was based upon motif types, and the design elements of a single motif type. The analysis of these categorical levels provided the opportunity to examine variations in spatial pattern occurring in each of the classificatory levels. For the open locations, this also involved examining variation in spatial patterning occurring at the regional and subregional level. Two sub-regions were defined for the region (Figure 4.7), based mainly on the concentration of archaeological data available for areas in the northwest of the study region, and in its southeast corner. For both sub-regions, the spatial analysis performed on each classificatory level at the regional level was repeated for the sub-regional data, permitting the changes brought about by classificatory and spatial scale to be examined.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

114.

Chapter 4.

Figure 4.7. Location of sub-regions examined in this study. Grey indicates areas over 350 metres in elevation.

4.3. Data examined in this study. Much of the archaeological and environmental data examined in this study has been presented and discussed in Chapter 3. The purpose of revisiting it in this section is to identify which of these datasets were used for modelling the distribution of archaeological features. In doing so, the source and derivation of each dataset is provided in this section.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

115.

Chapter 4.

4.3.1. Archaeological database. Much of the archaeological data used in this study came from previous archaeological research projects or archaeological consultants reports. As was explained in Chapter 3, there has now been a large amount of archaeological survey undertaken in the region and much of the results from of this work were able to be incorporated into the archaeological database. In total, the database contained basic information such as location, presence or absence of various archaeological features and a site type class for 1825 locations, 1795 of which occurred within the study region. The sources of information used for compiling the database are listed below: 1. Queensland EPA site records (339 locations) 2. Ken Kippen’s survey of Calton Hills station (875 locations) 3. Archaeological consultancies undertaken by Iain Davidson (212 locations) 4. Rock-art sites recorded by Iain Davidson (88 locations) 5. Other locations recorded as part of Davidson’s Selwyn Ranges project (149 locations) 6. Locations obtained from numerous other sources (132 locations).

4.3.1.1 Recorded information. The archaeological information from these sources was entered into a Microsoft Access database. When all the data had been entered and checked, it was exported and converted into an ArcView point theme that was used for spatial analysis. Figure 4.8 shows the form that was used for entering information into the Access database.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

116.

Chapter 4.

Figure 4.8. Site database data entry form.

The database was based upon the information requested on the Queensland EPA site card. The card forms the minimum standard of information recording for archaeological locations in Queensland, and as such was generally adhered to in most archaeological recording. The form was later modified to accommodate other information encountered in various location descriptions, such as stone tool types and raw materials, and subclasses of locations, such as individual finds and whether artefacts were present. Such information was not always entered onto EPA site cards, but could be found in archaeological reports or the field notes of recorders. A limitation of the EPA site card is that it focuses upon recording archaeological ‘sites’. Although the site concept (as an assessment of behavioural density) was adopted as a category in this study, for the purpose of data entry the primary analytical unit was the presence or absence of archaeological features. The classification of some locations as sites, for the most part, reflected assessments made by the recorder at the time of recording. For this reason, other fields were added to the database, such as the presence of artefacts, whether a find was an isolated artefact and the kind of material found. This afforded greater flexibility in the classification of locations based on the features they

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

117.

Chapter 4.

contained, rather than whether or not they were ‘sites’. However, it was not possible to base the database on artefact locations alone given that in most instances, such information was not recorded and it would be impractical over such an extensive region. By not forming the basic analytical unit for recording, the site classification therefore played a less important role in determining whether locations should be included in the database. Along with the database of all locations, several subsidiary databases were developed for specific datasets. These datasets included the information recorded by Iain Davidson and his students in the Selwyn Ranges, and by Kippen for his survey of Calton Hills. These datasets contained additional information about individual stone artefacts. The form used for entering this information is presented in Figure 4.9.

Figure 4.9. Stone tool database data entry form.

The stone artefact information recorded varied depending on the recorder and the purpose of the study. At a minimum, artefact raw material type, the type of artefact (flake, retouched flake, flaked piece, core, or axe), and for flakes: length; width; and thickness, were required for the information to be incorporated into the database. All of the stone artefact data incorporated into the database came from studies by people who had been students of Iain Davidson. Each of the studies had been conducted under Davidson’s supervision. This ensured there was consistency in the definition of key terms, such as retouch, and that measurements such as length were done in the same way. Full definitions of all categories used in the database can be found in the studies listed in Table 4.6. Detailed justification for the method employed in each of the studies

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

118.

Chapter 4.

from which artefact data was gathered can also be found in the references cited in Table 4.6.

Table 4.6. Sources of stone artefact data.

Archaeological study.

Source.

Davidson Selwyn Research

Unpublished

Kippen Honours

(Kippen 1992)

James Honours

(James 1993)

Drury Honours

(Drury 1996)

Osborne Consultancies

(Davidson and Fife 1994, Ridges et al. 2001)

Phosphate Hill consultancies

(Knuckey 1996, Ridges and Davidson 1996)

Limiting the utility of the stone artefact database was the amount of information recorded in different projects. For instance in one study (see Kippen 1992) only whole chert flakes were measured, whereas in another all categories were measured in all raw material types (see Drury 1996). Site sampling also varied considerably. For instance this varied from a sample of 100 artefacts encountered in a transect across a site (see Kippen 1992), to random selections from sites (see Davidson and Fife 1994, Ridges et al. 2001), to measuring all surface artefacts (see James 1993). The analysis of this data necessarily had to proceed on a selective basis. That is, only certain kinds of analysis could be formed in particular instances, depending on the information that was available. Explanation of how this was done is given in the analysis section of each sub-region in Chapter 7. A second stone artefact database was developed for the data collected by Ken Kippen on Calton Hills Station (the project is described in Kippen 1996). Kippen surveyed along transects and in each 100m segment of transect he recorded the number of artefacts he encountered, along with the type of stone tool (flake, retouched flake, core and flaked piece) and the raw material it was made from. The presence of quarries, hearths and proximate waterholes was also recorded for each transect.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

119.

Chapter 4.

Figure 4.10. Kippen transect data entry form.

One further database comprised recorded rock-art sites. Davidson and his students (1989) had recorded many rock-art sites in the Selwyn Ranges and its surrounds and the data he compiled formed the foundation of the database. This was added to by the detailed recordings made by Ross (1997) with sites containing anthropomorphs. Information located within the reports and site cards held by the Queensland EPA was also included in the database. Due to the diversity of sources from which data about the art

was

obtained,

the

information

about

individual

motifs

was

limited

to

presence/absence, and where possible, counts. The occurrence of colours was made for the entire site, as were the techniques employed. Figure 4.11 illustrates the information that was collected.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

120.

Chapter 4.

Figure 4.11. Art site database data entry form.

The quality of information for particular sites varied considerably. For this reason, a ‘complete’ field was included in the database to indicate which recordings represented complete censuses of the figures depicted in the site. Of the 181 sites included in the database, only 49 had complete censuses, mostly those recorded by Ross (1997). This places restrictions on the applications of the art site data, and in subsequent chapters, it is used for establishing motif distributions, and where possible, the associations between motif types. A second rock-art database was analysed in this project. This was the database compiled by Ross (1997). For specifics of the contents of this database, and the definition of key terms used in the database, the reader is referred to Ross’ (1997) Honours thesis. Details of the variables used in this study are provided in Appendix 1.

4.3.1.2 Positional accuracy. All position information entered into the archaeological location database was checked against topographic maps or other contextual GIS data such as drainage; roads; or __________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

121.

Chapter 4.

terrain, for consistency with location descriptions when these were given. Locations with unreliable location information were excluded from the database. A location was defined as being unreliable if it could not be confidently located within five hundred metres of contextual information (e.g. a specified topographic feature). The definition of unreliable was also employed when multiple spatially referenced descriptions referred to locations that were more than five hundred metres from one another. Such a definition was deliberately loose because of the variety of sources and methods of referencing find spots to spatial positions. In some cases, the only information available was written or drawn descriptions and/or references to named topographic features. Such information came from either EPA site cards or the journals of recorders where these were available. Problems were also encountered with co-ordinate information when this was given. In most cases, the number of significant figures in the grid reference could be used to assess positional reliability. However, this was not always possible since some older grid references had been transposed from feet to metres, and were reported to six or seven significant figures. These co-ordinates gave the impression of a precise location although they may have derived from an imprecise one. The Ellis Thorpe map library at the University of New England held topographic maps of the region dating back to 1937, and these were consulted for checking the older location references. Checking of this kind also revealed errors in transcribing co-ordinates, such as X and Y co-ordinates written back to front or referring to an incorrect map sheet. Wherever possible, co-ordinates were confirmed using features on either 1:250,000 or 1:100,000 topographic maps of the same edition given in the position description, or of the year of recording. For locations that had been recorded using transect surveys, positions were calculated using the descriptions of the recorders. Ken Kippen used transect recording exclusively during his survey of Carlton Hills, where he had adapted it from his work with Iain Davidson in the Selwyn Ranges. On both these projects, location information included the start and end positions of each transect, its bearing and the distance from the origin of the transect. Trigonometric formulae were used to convert this information into map co-ordinates. The accuracy of deriving transect co-ordinates with trigonometry introduces an unknown amount of error because it makes the assumption that each transect was walked in a straight line between the start and end points. Field survey conducted by the author has

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

122.

Chapter 4.

demonstrated that transects are rarely walked in a straight line (Ridges et al. 2001). Despite this, it is unlikely that any of the locations along a transect (which were usually one kilometre in length in both the Selwyn and Carlton Hills projects) would have varied by more than the five hundred metre tolerance applied for assessing the locational accuracy of other methods of recording.

4.3.1.3 Error checking. When all the data had been entered into the database, they were error checked by examining a mix of randomly selected and arbitrary records. To screen for spatial duplicates the records were exported to ArcView and used to build a point layer, preserving the attributes from the database. A query was then performed to select clusters of points within five hundred metres of one another. These were examined on a cluster by cluster basis. If locations within the cluster had similar attributes, N-1 duplicates were deleted and the remaining location moved to the mean position of the duplicates. Through this process 23 locations were deleted from the database after inspecting 143 clusters.

4.3.2. Non-archaeological data. One of the problems encountered with this study was the lack of environmental datasets covering the entire region. Of those that did provide a complete coverage, the mapping had been conducted at quite different scales (see Table 4.7).

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

123.

Chapter 4.

Table 4.7. Datasets obtained for the study. Dataset

Source

Scale

Coverage

9 Sec DEM¶

AUSLIG

277m

Complete

Drainage general

AUSLIG

1:5,000,000

Complete

Drainage detailed

AUSLIG

1:250,000

Complete

20m contours

AUSLIG

1:100,000

Northern half

Spot heights

AUSLIG

1:100,000

Northern half

Geology

AGSO

1:100,000

85% Complete

Soils

Soil Con.

1:5,000,000

Complete

Vegetation

Qld DPI

1:1,000,000

Southern half

Land Systems*

Qld DPI

1:1,000,000

Southern 2/3's

LANDSAT#

Geoimage

50m

90% coverage

¶ DEM= digital elevation model. * Landsystem mapping existed for the remainder of the region, but in a further two datasets. The landsystems in each map were different, and none of them had matching landsystem boundaries. The landsystems referred to in the table is the most recent mapping (Wilson, Purdie, and Ahern 1990). # The LANDSAT dataset consisted of a mosaic of eight spectrally matched Landsat scenes. Bands 1-4 were included in the dataset.

These problems were addressed in two ways. The first approach was to make the number of classes in the nominal datasets comparable. For instance, the 527 geological units were simplified into 49 rock types using the primary rock type described in the legend text associated with each unit. The second approach was to produce a surrogate for the land systems and vegetation mapping from the LANDSAT data. Supervised classification using the vegetation and/or land system mapping as a guide was used for this purpose. However, since the mapping was incomplete (see section 3.1.5), the vegetation data were simplified into 10 broad vegetation structures as outlined by Neldner (1991). These categories gave the most consistent mapping of LANDSAT classes across the region that were broadly consistent with the land system mapping (see section 3.1.5). The geological data obtained from AGSO for the project did not completely cover the entire region. To complete the coverage it was necessary to digitise the remaining areas from coarser scale maps. The mapping units were simplified to rock units, which made it easier to resolve edge-matching issues between the mapping at different scales.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

124.

Chapter 4.

Using existing digital and hardcopy data, two other datasets important to the project were digitised. Drainage basins were digitised from the 1:250,000 drainage mapping of the region. Waterholes marked on published geological and topographic maps were also manually digitised. In addition, the remaining 20m contours and spot heights were digitised from hard copy dyeline maps for the southeast corner of the region so that terrain based variables for the Selwyn area could be derived at a higher resolution. For each of the sub-regions, the only published and digitally available topographic data existed at a scale of 1:250,000. To improve the resolution of the digital elevation model (DEM) available in these areas, the digitised 20m contours and spot heights were used to derive a 100m DEM. The DEM was calculated using the ANU DEM algorithm available in Arc/Info, incorporating the 1:250,000 drainage in order to correct the resulting DEM for the direction of water flow (Hutchinson 1989). This last aspect was important for ensuring correct measures of aspect and wetness (see below) in areas of low relief. Creating a hydrologically correct DEM was also important for enabling the derivation of a stream network that could be ordered using the Strahler method. In the Strahler method of stream ordering, stream order only increases when streams of the same order intersect. Therefore the intersection of a first order and second order stream will remain a second order stream rather than create a third order stream. When two first order streams intersect, the down-slope link is assigned an order of 2. When two second order streams intersect, the down-slope stream is assigned an order of 3, and so on. The derivation of other datasets involved calculating cost distances rather than Euclidean distances from features such as streamlines, stream orders, and stone raw material sources. Cost distance mapping is a variation of Euclidean distance mapping, where the distance from a feature of interest to a location of interest is weighted by a cost factor. In all cases in this study the cost factor used was slope in order to simulate the movement of people following least cost paths across terrain. This is a concept that has also been used in site catchment analysis (Bailey and Davidson 1983).

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

125.

Chapter 4.

From the compiled datasets seven GIS layers were derived: 1. Elevation. This was the DEM layer. 2. Slope. Derived from the DEM 3. Aspect. Derived from the DEM. 4. Wetness. The wetness index measures the capacity of any location to retain water flowing over its surface based on its slope and the size of the catchment leading into that point (Moore et al. 1998, Wolock and McCabe 1995). It is derived from the DEM and slope layers. 5. Streamline proximity. Calculated as the cost distance from the streamlines digitised at 1:250,000. 6. Weighted streamline proximity. Cost distance layers were calculated from each stream order, normalised, and then summed using the stream order as a weighting. Thus: weighted streamline proximity = (cost distance from first order streams x 1) + (cost distance from second order streams x 2)… and so on. 7. Proximity to waterholes. Calculated as the cost distance from recorded waterholes in the region. Each of the final data sets used as variables in this study are summarised in the following seven figures. A map of the variable, and a frequency histogram (based on grid cell frequencies) is provided in each figure. In each figure also, the shade associated with each class in the legend corresponds with the shades used in the histogram and the mapping of each class.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

126.

Chapter 4.

Figure 4.12. Elevation variable.

Figure 4.13. Slope variable.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

127.

Chapter 4.

Figure 4.14. Aspect variable.

Figure 4.15. Wetness variable. __________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

128.

Chapter 4.

Figure 4.16. Streamline proximity variable.

Figure 4.17. Weighted streamline proximity variable.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

129.

Chapter 4.

Figure 4.18. Proximity to waterholes variable.

The environmental variables were therefore all various measures of terrain or proximity to water. Since the region is a semi-arid environment, water was clearly an important resource, which is why three different measures of proximity to streamline were employed. Each measure describes aspects of streamline proximity at different scales. For instance, the wetness index was included for differentiating every minor drainage channel in the region. The stream line proximity measured how close a location was to one of the main drainage channels. Likewise, the weighted streamline proximity measured the degree to which locations occurred at the headwaters or down stream areas of each drainage system.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

130.

Chapter 4.

Six other variables were derived using the cost-distance from geological sources used for manufacturing stone tools. These corresponded with the six stone raw material types for which quarries have been recorded in the region. The geological source areas for each rock type were derived from the geological rock units that each recorded quarry was located within. These polygons were then used as input for calculating cost distance layers. The six layers derived in this way are portrayed in figures Figure 4.19 to Figure 4.21. For these figures, areas that are black indicate the geological source, white is then proximate to the source in cost-distance terms, and dark grey distant from the source in cost-distance terms.

Figure 4.19. Cost distances from chalcedony and chert.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

131.

Chapter 4.

Figure 4.20. Cost distances from quartzite and quartz.

Figure 4.21. Cost distances from silcrete and volcanic raw material types.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

132.

Chapter 4.

4.4. Methodological limitations. As with the development of any methodology, there are limitations preventing an ideal application of methodological theory. One thing that needs stressing about this study is that it attempts to tackle a wide range of issues in a relatively broad regional framework. If any one of these issues were tackled individually, then the kinds of data used to do so would be specific to that task. The approach adopted in this study has necessarily taken a broad perspective of the constituents of regional archaeological structure, and as such has had to make sacrifices in the resolution and kind of data that has been included for analysis. Such a broad perspective has its advantages and disadvantages. On the positive side is the possibility of putting a range of issues into a more complete regional context. Crucial to the success of doing this is an explicit use of scale, and the diversity of data that was considered. On the negative side is the quality of individual datasets, and their adequacy for addressing the issues posed by the thesis. The data, as they stand, are not of a sufficient quality to draw definitive conclusions. However, although not perfect, the database is one of the more comprehensive regional databases in Australia. In this context also, the problems of data quality are by no means unique to this region. There were however, two issues about the approach taken in this study that require specific comment: •

the variables used to model archaeological distributions

•

lack of verification of the models generated

Each of these is considered below.

4.4.1. Variables. All the ratio scale variables in this study were derived from terrain. Even the costdistance layers from geological sources involved using slope as a cost variable for example. On this basis, it would have been expected that all the variables would have demonstrated a high degree of correlation. However, Table 4.8 shows that only two combinations of variables demonstrated any significant degree of correlation. These were

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

133.

Chapter 4.

slope and stream proximity and elevation and weighted streamline proximity. The correlation between these variables are indicated in bold in Table 4.8.

Table 4.8. Correlation matrix between ratio variables. Elevation 1.000

Slope

Slope

0.339

1.000

Aspect

0.117

0.016

1.000

-0.217

-0.463

0.000

1.000

0.449

0.639

0.022

-0.396

1.000

-0.726

-0.373

-0.048

0.132

-0.489

1.000

0.406

0.476

0.020

-0.218

0.489

-0.430

Elevation

Wetness Strm prox

1

Strmproxwtd2 WH prox

3

Aspect

Wetness

Strm prox

Strmproxwtd

WH prox

1.000

1 = Streamline proximity; 2 = streamline proximity weighted by stream order; 3 = waterhole proximity

The correlation between slope and stream proximity and elevation and weighted streamline proximity both occur due to the inherent character of natural terrain. For instance, it is expected that as stream order increases, elevation will fall. Similarly, due to the natural paths created by flowing water, the greatest cost-distances from streamlines will correspond with areas of greatest slope. However, from examining the spatial character of these variables, it is clear that although there is a degree of correlation, the variables are measuring different aspects of the landscape. These differences are also significant for understanding the location of archaeological finds. For instance, although elevation and weighted streamline proximity demonstrate a degree of correlation, the important distinction between them is that weighted streamline proximity shows areas where several streams coalesce producing areas that were potential junctions for the movement of people if they habitually travelled along drainage lines. Such characteristics would not be possible from elevation alone. The important feature of Table 4.8 is that it shows that although three different types of proximity to streamlines were used in this study they are not highly correlated within one another. The purpose of including all three was that in semi-arid regions, water was obviously an important resource, but previous work (see Chapter 3) has shown that its importance can take several different forms.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

134.

Chapter 4.

A final issue concerns the appropriateness of cost distance as a better measure of proximity than Euclidean distance. The cost surface used in all cases was slope in order to simulate the effect of having to move through varying terrain. However, others have questioned whether such measures are a reliable indicator of movement since that do not take into account time or barriers to movement (Zubrow 1990). For example, although it may incur a high cost to traverse a range in order to move from one valley to another, it may actually be quicker to traverse the range. The counter argument to such comments however are that it may be quicker, but did hunter-gatherers always take the quickest route? The character of movement described by cost distance would appear to be an adequate indicator of the way people moved around most of the time. The other query about barriers is not really an issue in this study since the terrain throughout the region is not so rugged as to totally restrict movement in any direction.

4.4.2. Verification. One clear limitation of this study is that a multitude of models are generated, but none of them have been tested. However, as explained earlier, the use of models in this study has more to do with data visualisation than attempting to produce accurate maps of where archaeological features occur. In cultural resource management contexts, clearly model accuracy is a more pertinent issue. However, the models in these studies usually attempt to describe the occurrence of all archaeological features, whereas the focus in this study was upon understanding the spatial patterns in various classes of archaeological feature. On this basis, it is believed that the need to rigorously examine how much improvement could be gained by adjusting which variables were used or the modelling technique applied was unnecessary. Although this would undoubtedly result in models with greater accuracy, it would complicate the comparison of different models. The approach taken in this study was to maximise the consistency between models, although it is understood that this may affect the accuracy of individual models. A further justification for not exploring the accuracy of the model is that doing so may introduce error produced by fitting an overly accurate model to inaccurate data. The data used in this study have clear spatial and categorical bias (see Chapter 5). Consequently, the application of more sophisticated modelling techniques would run the risk of producing patterns that reflect this bias rather than the trends evident in the data. __________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

135.

Chapter 4.

The most significant limitation to the accuracy of the models generated however has more to do with the issue of chronology, than model precision. It has been assumed throughout this study that the majority of open locations in the region are of recent (less than 1,000 years) antiquity, and therefore represent consistent behavioural patterns. As was outlined in Chapter 3, there are several grounds for making this assumption, but it is an assumption nevertheless. If there were one major improvement that could be made to the methodology adopted in this study, it would be to incorporate the aspect of time. However, this would require a program of determining the age of open sites which was clearly beyond the scope of this study, but which may be an issue that future research can address.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

136.

Chapter 5.

Regional Location Models. Our life is frittered away by detail...simplify, simplify. —Henry David Thoreau. Where I lived and what I live for.

5.1. Regional Scale modelling. This chapter presents the results of examining the archaeological database at the most general scale. In doing so, this chapter establishes the context for the following two chapters, which progressively step into increasing levels of categorical and spatial detail. At the level of generality examined in this chapter, the spatial scale is at its broadest with the entire region being the focus. Similarly, the categorical scale is also at its most general. The primary concern in this chapter is the distribution of locations containing archaeological features and classes of archaeological location. These classes included rock-art sites, open camp sites, open scatters and individual finds. Two aspects of the archaeological database were used to explore the dynamics of the region at its most general level. Firstly, the location of archaeological finds was used to identify those areas containing zones of concentrated activity. Point distributions, the density of find spots within various environmental units and probabilistic modelling were used to identify these areas. Secondly, the results were analysed in the context of

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

137.

Chapter 5.

different types of activity as indicated by the class of find spot, such as open camps and art sites. With models of each location class in place, it was possible to explore the context of these trends provided for structuring regional behaviour. Throughout this analysis, the emphasis was on understanding the spatial dynamics of the context in which archaeological finds occur. The approach represents a move away from simply identifying the important variables determining the majority of find spots or drawing circles around the distribution of particular items. The results presented in this chapter outline the varying role some environmental variables play in determining the probability of different types of activity occurring in a variety of contexts. In doing so, it can be illustrated, firstly, how dynamic the region was in terms of the location of activity, and secondly, how it does not conform to simple spatial patterns around which bounded spaces can be defined.

5.2. Distribution of archaeological locations. The first aspect of exploring the dynamics of Aboriginal activity in the region concerned examining the location of archaeological finds. Figure 5.1 displays the distribution of the 1795 find-spots recorded in the database. Even from a simple visual examination of the point distribution in Figure 5.1, it is apparent that the database contained considerable spatial bias. The northwest and southeast areas for instance contained many more findspots than the remainder of the region. Chapter 4 discussed how this bias reflected the differing density of survey coverage throughout the region. Consequently, the process of inferring trends in the distribution of archaeological finds was not straightforward since analysis of point patterns alone would have given a biased impression of the archaeological pattern occurring throughout the region. Nonetheless, with close to 2000 archaeological find spots, it was possible to explore some of the trends existing in the location of archaeological material, albeit through probabilistic modelling.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

138.

Chapter 5.

Figure 5.1. Distribution of all archaeological find-spots.

To understand the spatial trends in the database, it was necessary to examine the relationships between each of the find spots and the context in which they occur. There are several ways to do this, depending on the type of data involved. In the case of nominal variables, such as geology and soils, it involved comparing the frequency of finds occurring in each category to a set of points of the same number generated as random points in the GIS. For ratio variables such as elevation and slope, the analysis involved fitting a regression model, as was described in Chapter 4. The formation of the final model was a combination of these approaches, since each offered particular advantages. For instance, the use of nominal scale data was important for identifying differences in the location of find-spots brought about by the influence of zones of broadly similar context, such as particular soil types or geological units. Due to __________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

139.

Chapter 5.

the resources available in each unit or their potential for preserving archaeological finds, the occurrence of find-spots may be generally higher in some units than in others. In contrast, the use of ratio variables was useful for identifying changes that occur systematically with changes in context. For instance, the frequency of find spots may decrease dramatically with distance from streamlines. Thus, ratio scale variables were important for establishing the variability of find-spot frequency within each soil or geological unit, whereas the soil and vegetation units were useful for identifying the relative frequencies of find-spots in different parts of the region. Using the environmental datasets included in the GIS, the probability of archaeological evidence occurring was calculated for all locations using the fitted regression models incorporating both nominal and ratio variables.

5.3. Nominal variables. The first step in constructing a model of archaeological evidence was to examine the frequencies of known find-spots in relation to the nominal variables. Given the problems with some datasets, such as land systems (see Chapter 4), the nominal variables selected were soils, geology, and a classification of the LANDSAT data. Table 5.1, Table 5.2, and Table 5.3 summarise the comparison of location frequencies for archaeological locations with a random distribution. The random distribution of 1795 points was generated in ArcView, specifying that no random points would occur within two hundred and fifty metres from the boundary of the study region or to another random point. The process of generating random points was repeated twenty five times, and the mean frequency occurring in each mapping unit was calculated in order to allow for fluctuations between each random set.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

140.

7.6 1.3

1.2 2.0 2.2 13.2 1.5 1.7 2.1

14.1 3.2 0.4 0.6 75.4 2.3 1.5 5.0

Arch/ 10,000 Ha 7.1 0.4 3.4 27.5 18.4 0.6 1.05 10.37 0.37 198.5 100.96 6.43

1.897 0.118 0.913 7.563 4.727 0.158 0.000 3.409 0.878 0.109 0.150 27.083 0.670 0.411 1.266 0.000 0.383 0.449 0.619 3.356 0.413 0.395 0.601 0.000 2.093 0.353 0.000 0.000 0.000 62.39 49.31

3.51 107.7 2.94 33.84 6.90 5.20 0.40

1.00 0.32 11.69 19.76 54.41 0.03 9.53 0.49

χ2

Ratio¶

141.

_________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

Area (Ha) Random § % of random Random/ 10,000 Archaeological % of locations Ha 15441 5.8 0.3 3.8 11 0.6 53088 17.0 0.9 3.2 2 0.1 295381 108.5 6.0 3.7 99 5.5 100036 36.4 2.0 3.6 275 15.3 101471 39.6 2.2 3.9 187 10.4 33806 12.7 0.7 3.8 2 0.1 32194 11.4 0.6 3.6 2121 0.9 0.0 4.1 3 0.2 145843 53.6 3.0 3.7 47 2.6 50304 18.4 1.0 3.7 2 0.1 89554 33.4 1.9 3.7 5 0.3 6895 1.9 0.1 2.8 52 2.9 12896 4.5 0.2 3.5 3 0.2 112801 41.4 2.3 3.7 17 0.9 56381 22.1 1.2 3.9 28 1.6 6571 2.0 0.1 3.1 48815 15.7 0.9 3.2 6 0.3 953554 429.4 23.9 4.5 193 10.8 102535 37.2 2.1 3.6 23 1.3 67511 26.5 1.5 3.9 89 5.0 84780 31.4 1.8 3.7 13 0.7 53389 22.8 1.3 4.3 9 0.5 23501 8.3 0.5 3.5 5 0.3 9873 3.8 0.2 3.8 392378 143.3 8.0 3.7 300 16.7 409185 153.1 8.5 3.7 54 3.0 5900 2.6 0.1 4.4 18889 6.8 0.4 3.6 1396 1.0 0.1 7.2

Table 5.1. Archaeological location and random point summary by soil types.

Soil Unit B2 Bz CC Ca Cd Fb Fq Fu Fz Il JK Kb Lake LK MM MQ MR MS MT Mo Mr Ms Mu Mw My Mz Oa Oc Od

361457 38973 112493 557033 45213 4180 119203 15850 82343 11739

137.8 14.4 41.9 206.0 17.4 1.6 44.1 5.8 29.8 4.8

7.7 0.8 2.3 11.5 1.0 0.1 2.5 0.3 1.7 0.3

Area (Ha) Random § % of random

Random/ 10,000 Ha 3.8 3.7 3.7 3.7 3.9 3.8 3.7 3.6 3.6 4.1 14 3 14

0.8 0.2 0.8

1.2 1.9 1.7

Archaeological % of locations Arch/ 10,000 Ha 11 0.6 0.3 17 0.9 4.4 14 0.8 1.2 127 7.1 2.3 170 9.5 37.6 0.080 1.181 0.334 0.616 9.748 0.000 0.317 0.521 0.470 0.000

Ratio¶

14.81 0.37 5.06

110.95 0.08 13.15 20.14 129.39

χ2

142.

_________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

¶ The ratio was calculated as archaeological locations per 10,000 Ha divided by random locations per 10,000 Ha.

the number of points occurring in each soil unit was averaged over the twenty-five replications.

study region. The soil unit classification was then determined for each point. This process was then repeated twenty-five times, and

§ 1795 random points were generated such that no two points were less than 250 metres apart or 250 metres from the boundary of the

Soil Unit Qa Qb Ro Sl TB Ub Va Vc Vd Wc

Archaeological location and random point summary by soil types (cont).

134

31547

Mafic Dyke

Metarenite

11.0

0.0

27.9

3.1

0.0

4.8

0.4

13.0

7.8

167.9

32.7

0.4

5.8

12.4

22.5

0.7

6.1

15.0

0.0

21.6

0.0

0.1

0.0

13.1

15.3

762.2

Random Points 0.8

0.6

0.0

1.6

0.2

0.0

0.3

0.0

0.7

0.4

9.4

1.8

0.0

0.3

0.7

1.3

0.0

0.3

0.8

0.0

1.2

0.0

0.0

0.0

0.7

0.9

42.5

% of random points 0.0

3.499

0.000

3.804

3.502

0.000

3.789

3.279

3.900

4.014

3.655

3.673

2.890

3.387

4.101

3.709

2.919

3.189

3.816

0.000

3.540

0.000

4.931

0.000

3.678

3.534

3.711

Random points/ 10,000 Ha 3.288

4

0

13

6

0

1

0

0

1

90

19

0

1

28

5

1

4

22

0

8

0

0

0

3

1

776

Archaeological locations 0

0.2

0.0

0.7

0.3

0.0

0.1

0.0

0.0

0.1

5.0

1.1

0.0

0.1

1.6

0.3

0.1

0.2

1.2

0.0

0.4

0.0

0.0

0.0

0.2

0.1

43.2

% of archaeological locations 0.0

1.268

0.000

1.771

6.735

0.000

0.796

0.000

0.000

0.512

1.959

2.135

0.000

0.584

9.290

0.825

4.292

2.084

5.612

0.000

1.311

0.000

0.000

0.000

0.844

0.231

3.779

Archaeological locations/ 10,000 Ha 0.000

0.362

0.000

0.466

1.923

0.000

0.210

0.000

0.000

0.128

0.536

0.581

0.000

0.172

2.265

0.222

1.471

0.654

1.471

0.000

0.370

0.000

0.000

0.000

0.229

0.065

1.018

0.000

Ratio

3.28

0.00

5.49

0.93

0.00

2.49

0.00

0.00

5.27

25.35

3.68

0.00

3.39

6.09

11.22

0.05

0.40

1.34

0.00

6.59

0.00

0.00

0.00

6.27

12.30

0.22

0.00

χ2

143.

_________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

73399

Limestone

5

Ironstone

8909

Ignimbrite

Laterite

1220

12562

Greywacke

19531

Granite

33333

88977

459421

Gneiss

Granofels

1246

Gabbro

Granodiorite

30139

60614

Dolerite

17126

2330

Diorite

Duricrust

19192

Conglomerate

Dolomite

39204

Chert

3

61012

Breccia

Chalcedony

162

124

82

Arkose

Basalt

35565

Arenite

Banded Fe

43355

Amphibolite

2053650

2433

Adamallite

Alluvial

Area (Ha)

Rock type

Table 5.2. Archaeological location and random point summary by rock type.

229448

109181

31432

Sandstone

Schist

Shale

99

1.4

0.0

8.7

68.6

11.4

42.8

86.8

3.9

124.5

0.1

0.2

77.8

0.5

43.0

36.2

0.0

0.1 23.2

7.9

0.0

110.6

Random Points 0.7

0.1

0.0

0.5

3.8

0.6

2.4

4.8

0.2

6.9

0.0

0.0

4.3

0.0

2.4

2.0

0.0

0.0 1.3

0.4

0.0

6.2

% of random points 0.0

3.101

0.000

3.631

3.799

3.627

3.924

3.785

3.248

3.725

1.912

3.379

3.742

2.870

3.595

3.460

0.000

3.664 3.660

3.904

0.000

3.794

Random points/ 10,000 Ha 5.300

2

0

8

70

8

38

88

2

89

1

0

20

0

38

17

0

0 23

27

0

380

Archaeological locations 1

0.1

0.0

0.4

3.9

0.4

2.1

4.9

0.1

5.0

0.1

0.0

1.1

0.0

2.1

0.9

0.0

0.0 1.3

1.5

0.0

21.2

% of archaeological locations 0.1

4.560

0.000

3.346

3.879

2.545

3.480

3.835

1.657

2.663

23.894

0.000

0.962

0.000

3.177

1.623

0.000

0.000 3.634

13.376

0.000

13.029

Archaeological locations/ 10,000 Ha 7.361

1.471

0.000

0.922

1.021

0.702

0.887

1.013

0.510

0.715

12.500

0.000

0.257

0.000

0.884

0.469

0.000

0.000 0.993

3.426

0.000

3.435

1.389

Ratio

0.11

0.00

0.03

0.01

0.60

0.29

0.01

0.61

6.28

0.74

0.00

35.12

0.00

0.32

7.03

0.00

0.00 0.00

10.56

0.00

171.35

0.05

χ2

144.

_________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

4386

Tuff

Water

23906

Slate

180480

12071

Rhyolite

Siltstone

334208

Quartzite

592

207896

Porphyry

419

1673

Pegmatite

Quartz hematite

119594

Mudstone

Quartz dyke

62

104727

Metasiltstone

218 63284

20186

12

Metarhyolite

Metadolomite Metagreywacke

Metadolerite

Metaconglomerate

1359

291647

Metabasalt

Area (Ha)

Metarkose

Rock type

Archaeological location and random point summary by rock type (cont).

Random

97.0 242.0 219.1 170.3 182.6 156.0 170.6 172.1 203.5 140.7

Area (Ha)

260735 643148 591973 471574 488304 418433 454606 463460 563767 369186

5.4 13.5 12.2 9.5 10.2 8.7 9.5 9.6 11.3 7.8

% of random

Random/ 10,000 Ha 3.720 3.763 3.701 3.612 3.739 3.729 3.753 3.714 3.609 3.812

Archaeological % of locations Arch/ 10,000 Ha 92 5.1 3.528 38 2.1 0.591 188 10.5 3.176 250 13.9 5.301 213 11.9 4.362 236 13.1 5.640 143 8.0 3.146 137 7.6 2.956 132 7.4 2.341 81 4.5 2.194 0.948 0.157 0.858 1.468 1.166 1.512 0.838 0.796 0.649 0.576

Ratio 0.14 161.201 2.68 17.12 2.63 18.33 2.66 4.36 16.80 17.13

χ2

145.

_________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

supervised classification routine in Erdas Image 8.3.

§ The Landsat data were classified into 10 classes using the vegatation data as a guide. The classification was performed using the

Landsat class § 1 2 3 4 5 6 7 8 9 10

Table 5.3. Archaeological location and random point summary by Landsat classes.

Chapter 5.

The frequencies of random and archaeological locations occurring in each mapping unit were converted to frequencies per 10,000 hectares. The ratio of these values indicates where the frequency of archaeological locations is higher or lower than that occurring for random locations. The frequency of random locations per 10,000 hectares is relatively constant, averaging 3.702 locations per 10,000 hectares. This is what would be expected if the distribution were random, where for a region equalling 4,832,233 hectares, 3.715 locations would be expected in every 10,000 hectares. Chi squared was calculated for each of the environmental units also in order to determine whether the ratios were statistically significant. For geology, the mapped areas of arkose, basalt, chalcedony, ironstone, mafic dyke, metaconglomerate, metarhyolite and tuff were very small, the largest being basalt at 124 hectares. Due to their size, no random points fell within these areas in any of the random point generations. For this reason, these geological units are listed with zero points per 10,000 hectares, although their probability of containing a random location is not absolutely zero. No archaeological locations fell within these geological units either. As Table 5.1, Table 5.2, and Table 5.3 indicate, the ratio between the frequency of random and archaeological locations varied significantly. Figure 5.2, Figure 5.3, and Figure 5.4 illustrate graphically these differences by plotting the frequency of locations per 10,000 hectares for both the archaeological and random datasets. In some cases, the degree of difference was very large, especially in the soils dataset. For instance, in units Ca, Kb and Tb, the frequency of archaeological locations was seven, twenty-seven and ten times higher than that occurring for the random locations. Similarly, for soil units Bz, Fb, Il, JK and Qa, the archaeological frequencies were ten times or more lower than that observed for the random datasets. The chi squared for all these soil units was very high, indicating that these were very significant differences.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

146.

Chapter 5.

100

Random/ 10,000 Ha Arch/ 10,000 Ha

Count / 10,000 Ha

10

1

O a O c O d U b W c

Fq M Q M w

Il Bz Q a

Fb JK

Va

M z R o M R

M s M r LK

V c Vd

Sl M T M u M S

Fz La ke

Q b C C

M M

M o M y B2

C a C d Fu

Kb TB

0

Soil Units

Figure 5.2. Comparison of random to archaeological frequencies in each soil type. Geological unit frequencies Frequency (n) 100

Random / 10,000 Ha Archaological / 10,000 Ha

10

1

Porphyry

Ignimbrite

Metagreywacke

Greywacke

Gneiss

Alluvial

Breccia

Arenite

Granofels

Laterite

Gabbro

Dolomite

Banded Fe

Quartz dyke

Chert

Meta arkose

Mudstone

Sandstone

Meta arenite

Diorite

Granodiorite

Granite

Rhyolite

Siltstone

Pegmatite

Shale

Amphibolite

Metasiltstone

Schist

Slate

Dolerite

Conglomerate

Limestone

Duricrust

Metabasalt

Metadolerite

Quartzite

Adamallite

0

Metadolomite

0

Geological units

Figure 5.3. Comparison of random to archaeological frequencies in each geological unit.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

147.

Chapter 5.

Landsat Class Frequencies 6 Random/ 10,000 Ha Arch/ 10,000 Ha

Frequency / 10,000 Ha

5

4

3

2

1

0 6

4

5

1

3

7

8

9

10

2

Landsat classes

Figure 5.4. Comparison of random to archaeological frequencies in each Landsat class.

Similar trends were observed in the geological and LANDSAT datasets, although the degree of difference was not as large, the Chi squared reveal that they were also significant. For geology, one unit, quartz hematite, had an archaeological frequency twelve times that of the random distributions, however this is due to the small size of the geological unit. Since the average number of random points occurring in this unit was 0.1, the single archaeological site occurring in this unit produced an artificially high ratio. Correspondingly, the Chi squared for this unit was 0.74, indicating that despite the high ratio it was not a significant difference. Except for quartz hematite, the largest deviation from random was three times greater, and ten times smaller. For the LANDSAT classification, the largest deviation greater than random was one and a half, and the largest deviation less than random was five times. The differences between these datasets are largely a product of the scale at which they were mapped, and the number of categories used to classify them. The soils data, mapped at a scale of 1:1,000,000, is the coarsest scale mapping. However, with 39 soil types, it has a similar number of classes as the geological data, which was classed into 49 rock types, but was mapped at a scale of 1:100,000. Subsequently, the difference between the soils and geology datasets lies in the relative sizes of different categories (see Table 5.4). The alluvial mapping dominates the geology dataset by taking up 42% of the region (2,000,000 __________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

148.

Chapter 5.

hectares), with the remaining units having a mean area of 58,000 hectares, nearly half that of the overall dataset. Although it had fewer categories, the average size of the soil units was smaller than for geology, by virtue of being a more consistent size. The most consistent mapping categories were found in the LANDSAT classification, although due to the small number of categories, these were the largest mapping units.

Table 5.4. Sizes of mapping units in nominal variables.

Dataset

Mean unit size (Ha)

Standard deviation of unit sizes

Soils

49,000

65,000

Geology

98,000

298,000

LANDSAT

470,000

105, 000

The significance of these differences becomes apparent when the ratios of random to archaeological point frequencies in each mapping unit are used to colour these units in the GIS. Figure 5.5, Figure 5.6, and Figure 5.7 illustrate the results of performing this procedure. The categories in these figures also take into account the degree of significance of the ratios based on the Chi squared results by multiplying the ratio by the Chi squared value. This has the result of reducing the relative value of the ratio when the difference was not statistically significant. In contrast, high ratios that were also statistically significant are exaggerated by this procedure. For the soils dataset (Figure 5.5), the bias in terms of the density of mapping in different parts of the region is readily apparent as the darker units in the northwest of the region. Although the mapping units in the soil dataset are relatively even, this acts to emphasise the spatial bias in the archaeological survey.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

149.

Chapter 5.

Figure 5.5. Ratio of archaeological to random location densities for soil types.

In contrast, the geological dataset (Figure 5.6), despite being dominated by the alluvial map unit, has a pattern of archaeological location densities which does not so readily reflect the spatial bias in the database. The geological dataset also demonstrates high archaeological frequencies in those units occurring in the northwest corner of the region, although other significantly high archaeological densities also occur in geological units throughout much of the region.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

150.

Chapter 5.

Figure 5.6. Ratio of archaeological to random location densities for geological units.

The LANDSAT classification (Figure 5.7) displays a similar trend to that seen in the geological dataset where the areas corresponding to alluvial in the geological dataset approximate the random point frequencies. In the highlands, the LANDSAT classifications are also high, indicating the significantly high densities of archaeological locations in these areas.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

151.

Chapter 5.

Figure 5.7. Ratio of archaeological to random location densities for Landsat classes.

Since the most common archaeological locations are scatters of stone artefacts, and these occur on places best suited for camping, that is flat ground near water, the lowlands would initially be expected to have higher archaeological frequencies. However, the geology and LANDSAT data suggest that a contrast existed between lowland and upland archaeological location frequencies, which reflects the large number of quarries and rockart sites occurring in the uplands. On this basis, the geological and LANDSAT data appear to be suitable for differentiating the trends in archaeological location pattern for the whole region in a more consistent way than soils do. Since soils were unduly influenced by the spatial bias in archaeological survey intensity, they were a poor candidate for describing regional trends in the location of archaeological features.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

152.

Chapter 5.

5.4. Ratio variables. For the ratio variables, archaeological locations were compared to the first of the twentyfive random distributions derived for examining the nominal variables. The same random distribution of 1795 locations was used for comparing all ratio variables. Trellis plots were employed to make the comparisons. A trellis plot permits relationships in a dataset to be observed by conditioning (MathSoft 1999c, 107). In each trellis plot a series of panels are displayed, each containing a plot of a subset of the dataset divided using a conditioning variable. In the plots below, the conditioning variable was location type, which contained two categories, archaeological and random locations. The vertical axes in the histograms indicate probability values associated with the density line included on each plot. The density line is a non-parametric density estimate that makes no assumptions about the mean or variance of the dataset (MathSoft 1999c, 74). The height of the histogram bars are scaled to the probability values derived for the density curve. The density curve closely mimics changes in frequencies in the dataset, but does so using a probability scale instead of frequency. The advantage of using a probability scale is that it enables direct comparisons between the trends within subsets of the dataset. In each of the plots that follow, the X and Y axes have the same scale in each panel of the plot, and the trends in them can be compared directly. Along with the trellis plot for each variable, a second plot displays a non-linear regression between archaeological and random locations for each variable. In each case the method for fitting the regression was a generalised additive model, using the cubic spline option (MathSoft 1999a, 326). Chapter 4 discussed the importance of adopting non-linear fits for the variables used in this study, in this instance the discussion focuses upon the implications of the form of the regressions. The first ratio variable examined was elevation, and the trellis plot derived for this variable is illustrated in Figure 5.8. Two things are apparent from the comparison of elevation values for the archaeological and random locations. One is the considerable overlap in values, as was examined in Chapter 4. The other is that despite the overlap, it is clear that archaeological sites occupy a more narrow range of elevation than the random locations. There are two reasons for this. One is due to the spatial bias of the archaeological locations. Not only do the archaeological locations clump together __________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

153.

Chapter 5.

spatially, but the clumps in the southeast and northwest (see Figure 5.1) also occupy similar elevation ranges. Exacerbating this problem is the lack of sites occurring on the plains. In Chapter 3 it was discussed how archaeological features tend to be obscured on the plains, due to the development of self-mulching soils. At higher elevations, the occurrence of fewer archaeological locations reflects human preference, since the tops of the ranges were less suitable for many of the activities that result in archaeologically observable features. However, this also reflects survey bias since the tops of the ranges were surveyed less frequently. The combination of taphonomic factors at lower elevations and human preference at higher elevations produces the narrower range of elevation values seen in Figure 5.8. This trend is apparent in the form of the regression curve in Figure 5.9. The regression suggests that the probability of archaeological feature increases over that expected from a random distribution between 275 and 420m elevation. Outside this range, the probability of archaeological features decreases rapidly. The trend in Figure 5.9 suggests that archaeological features are unlikely at low and high elevations, although it is important to realise that the majority of archaeological survey has focused on areas that were 300 to 400 metres in elevation.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

154.

Chapter 5.

Archaeological locations

Random locations

98

158

218

278

338

398

458

518

578

Elevation

Figure 5.8. Histogram of elevation for all locations.

Probability of archaeological location presence

1.00

0.75

0.50

0.25

0.00 100

150

200

250

300

350

400

450

500

550

Elevation

Figure 5.9. Regression curve for elevation.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

155.

Chapter 5.

Clear distinctions between archaeological and random locations are also observed for aspect. In Figure 5.10, the X values reflect compass bearings, ranging from zero to 360 degrees. Aspect values were derived in the GIS, using standard algorithms for calculating aspect from a DEM (see discussion in Chapter 4). The important difference between the aspect values for archaeological and random locations, are the relatively even distribution of values for the random locations and the bimodal distribution for archaeological ones. The archaeological locations have their highest frequencies at 60 and 290 degrees. On first inspection, this trend may reflect a human preference for locating activities on slopes with a northerly aspect, providing warmth in winter. However, contributing to this trend is the pattern in terrain brought about by the north-south orientation of the drainage in the region. Some of the pattern seen in Figure 5.11, may therefore be a consequence of occupying locations adjacent to rivers where aspect is constrained by the orientation of linear valleys. Support for this is given by the similar, but less prominent trend seen for the random points. The distinction between the random and archaeological locations is distinct enough to be reflected in the regression (see Figure 5.11). Slope is a variable that is often cited as being critical for predicting the location of archaeological features due to the preference of people to locate their activities on level ground. This trend is evident in Figure 5.12. However, what is also significant, is that the random locations also demonstrate a very similar trend. This is largely because places with high slope values are very rare in the region. Similarly, when slope is steep it tends to be quite localised. Thus, stating that archaeological locations occur on level terrain does not reveal a great deal about the location of archaeological features in this instance.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

156.

Chapter 5.

Archaeological locations

Random locations

0

29

57

86

115

144

172

201

230

259

287

316

345

Aspect

Figure 5.10. Histogram of aspect for all locations.

Probability of archaeological location presence

1.00

0.75

0.50

0.25

0.00 0

50

100

150

200

250

300

350

Aspect (degrees)

Figure 5.11. Regression curve for aspect.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

157.

Chapter 5.

In fact to some extent, the reverse is also true. That is, the archaeological locations have a slightly higher propensity to be located in areas of steep slope, as can be seen in the form of the regression in Figure 5.13. The explanation for this trend is the occurrence of rockshelters and art sites along cliff lines. The regression curve also illustrates that it is around one degree, not zero, that the probability of archaeological occurrence increases over random. This trend can again be related to the spatial bias in survey coverage. Much of the survey coverage occurred at intermediate elevations in the region where slope is low but not absolutely zero. Those areas that are level, and of low elevation, correspond with the black soil plains where there are issues of site visibility. Given the nature of the arid environment, it would be expected that distance to streamline would be an important determinant of the location of archaeological features. However, the relationship is not as strong as might be first supposed, and what is actually observed is that a large number of random locations also occur along streamlines. The large proportion of random locations close to streams is due to the large area that is situated proximately (in slope-cost terms) to streams. For instance, 66% of the region is situated within 2,000 cost-distance units from streams, half of which occurs within 500 cost-distance units. Superficially, the similarity of the random data in Figure 5.14 might suggest that the location of archaeological features is only marginally different from a random distribution. However, the nature of the regression curve in Figure 5.15 indicates the need to consider the pattern of probabilities far from streams, close to streams and in between.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

158.

Chapter 5.

Archaeological locations

Random locations

0

1

2

3

4

6

7

8

9

10

Slope (deg.)

Figure 5.12. Histogram of slope for all locations.

Probability of archaeological location presence

1.00

0.75

0.50

0.25

0.00 0

1

2

3

4

5

6

7

8

9

10

11

12

Slope (degrees)

Figure 5.13. Regression curve for slope.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

159.

Chapter 5.

Archaeological locations

Random locations

0

2,000

4,000

6,000

8,000

10,000

12,000

Cost distance from stream lines

Figure 5.14. Histogram of stream proximity for all locations. 1.00

Probability of archaeological location presence

spline loess

0.75

0.50

0.25

0.00 0

2000

4000

6000

8000

10000

12000

Cost distance from streamlines

Figure 5.15. Regression curve for streamline proximity.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

160.

Chapter 5.

For example the probability of archaeological features occurring at cost-distances greater than 2,000 is low, especially at cost-distances greater than 6,000. Given the scarcity of water resources in the region, such a pattern is consistent with preconceptions about the nature of foraging in arid landscapes. However, the nature of how foraging takes place at cost-distances from streamlines that are less than 2,000 reveals other features about the nature of streamline proximity. The identification of these features required the adoption of a different kind of model fitting algorithm that was more capable of identifying the dramatic change in the occurrence of archaeological features very close to streamlines. For all other variables, a spline algorithm was used to construct models of how the location of archaeological features relates to that particular variable. However, for streamline proximity, the probability of occurrence of archaeological features increases dramatically as costdistance approaches zero. The spine for this relationship indicates this trend, but the LOESS method illustrated in Figure 5.15 demonstrates how the nature of the relationship is much more dramatic. The LOESS model illustrates the sudden change in probability below cost distance values of 500. The model suggests that the probability of archaeological features increases from 0.5 to 0.6 over this range. The loess model also contains a section between a cost-distance of 500 and 2,000, where the predicted probability is 0.5. Two things are important about this section of the model. One is its uniformity, and the other is its level character at 0.5. The combination of increased probability adjacent to streams and a zone of medium probability around them may be consistent with a foraging strategy where camps are located along streamlines.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

161.

Chapter 5.

Archaeological locations

Random locations

0

2,000

4,000

6,000

8,000

10,000

12,000 14,000 16,000

18,000 20,000

Cost distance from waterholes

Figure 5.16. Histogram of waterhole proximity for all locations.

Probability of archaeological location presence

1.00

0.75

0.50

0.25

0.00 0

2,000

4,000

6,000

8,000

10,000

12,000

14,000

16,000

18,000

20,000

Cost distance from waterholes

Figure 5.17. Regression curve for proximity to waterholes.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

162.

Chapter 5.

Proximity to waterholes presents a very different pattern from that seen for proximity to streamlines. Unlike streamlines, the probability of occurrence of archaeological features does not peak as cost-distance approaches zero for waterholes. Instead, random locations are more likely to be located adjacent to waterholes than archaeological features. This would appear to be a counter-intuitive given that a waterhole represents a definite source of water whereas streamlines represent only a potential source. There are three possible reasons why the occurrence of archaeological features adjacent to waterholes is apparently so rare: 1. Disturbance from cattle. 2. The role waterholes played in a foraging round. 3. The rarity of waterholes in the region. Since pastoral development, waterholes have become an important resource for cattle as well as people. In many cases this has resulted in the area around waterholes being highly disturbed. If sites did occur adjacent to waterholes, it is possible that this disturbance hinders their recognition. This is especially the case with stone artefact scatters, but less of a problem with art sites. It is also dependent upon the nature of the waterhole. As was outlined in Chapter 5, two types of waterholes can be identified in the region; those that occur in sandy depressions within the higher stream orders, and those that formed within geological rock formations. Surrounding the former are loose riverbank deposits that are more prone to disturbance from cattle scuffing. Exacerbating the absence of archaeological features adjacent to waterholes is the relative abundance of waterholes in sandy streamlines, and their temporary nature. Thus, although it would be possible for people to regularly focus activity at waterholes, the duration of occupation may have been determined by the permanence of the waterhole. Personal communication with members of the Aboriginal community and pastoralists in the region suggests that some waterholes can appear and disappear on a decade long basis. In contrast, those waterholes associated with rock formations are generally in rougher terrain and less likely to have campsites adjacent to them due to the absence of soft level ground. Instead, the rock around such waterholes often presents suitable surfaces for producing rock-art.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

163.

Chapter 5.

Experience in the region has also found that the majority of mapping underestimates the number of waterholes (Davidson pers. comm.). In particular, pastoralists tend to nominate waterholes that are accessible to cattle, and waterholes have been observed that were inaccessible for cattle and unknown to the leaseholder. Another important feature about waterholes, in terms of the proximity of archaeological features, appears to be a broad zone around which the probability archaeological features is just above that of the random distribution. The model illustrated in Figure 5.17 reflects this pattern where the probability of archaeological features occurring between costdistance values of 2,000 and 10,000 is greater than 0.5. One possible explanation for this pattern is that it reflects the way waterholes operated as places where people could retreat to when less reliable water resources become depleted. Gould (1977) for instance reported that in the Western Desert, less reliable sources of water were preferentially exhausted first before retreating to the more reliable ones. If such a practice was in operation in the study region, then it might result in less time spent at the major waterholes. Assuming that severe drought, although unpredictable and relatively frequent was interspersed with longer periods of less severe conditions, it could result in less archaeological material accumulating adjacent to major waterholes. This does not negate the importance of waterholes as a crucial element in hunter-gatherers foraging patterns in the region. Rather, it may not necessarily follow that many archaeological features will be located adjacent to all of them. A final characteristic of the pattern seen in the proximity of archaeological features to waterholes is that the trend observed in Figure 5.16 and Figure 5.17 reflects the relative rarity of waterholes in the region. Due to their relative rarity, even if a high proportion of them did have archaeological features close by, the small number of waterholes means that the proportion of all archaeological features this would represent would still be small. In fact however, the proportion of waterholes that do have archaeological features near them is quite small. It can been seen in Table 5.5 for instance that even within 2.5km of waterholes, the occurrence of archaeological features is still very rare. This may reflect the taphonomy of cattle destroying sites, but it may equally reflect the small proportion of waterholes that have been surveyed archaeologically.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

164.

Chapter 5.

Table 5.5. Proportion of waterholes containing archaeological features within specified radii.

Radius of buffer

% waterholes containing archaeological features

500m

4%

1000m

8%

2500m

16 %

5000m

38 %

Proximity to stream lines weighted by stream order, introduced some unanticipated results, which Figure 5.18 illustrates. The most noteworthy feature in Figure 5.18 is the spike in archaeological features occurring about a value of 50. The spike is also reflected in the nature of the regression curve seen in Figure 5.19. This spike is the result of two processes. The first is the result of a high occurrence of stone artefact scatters in the middle portion of two long linear valleys in the Calton Hills sub-region where survey coverage was particularly high. Splitting the valley transversely is the drainage divide between water flowing north to the Gulf of Carpentaria and south to the Lake Eyre basin. This portion of the valley therefore occurs at the lower end of the range of values for the weighted proximity to streamline variable, as it is located at the headwaters of two drainage systems. The occurrence of archaeological features along the valley is virtually continuous, as the movement of people is constrained by the nature of the terrain. Consequently, the occurrence of archaeological features along the valley goes against the overall trend for archaeological features to be preferentially located towards higher values for the weighted streamline proximity index.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

165.

Chapter 5.

Archaeological locations

Random locations

3

13

23

33

43

53

63

73

83

93

103

Cost distance to streams weighted by stream order

Figure 5.18. Histogram of stream proximity weighted by stream order, for all locations.

Probability of archaeological location presence

1.00

0.75

0.50

0.25

0.00 0

10

20

30

40

50

60

70

80

90

100

Cost distance to streamlines weighted by stream order

Figure 5.19. Regression curve for proximity to streamlines weighted by stream order.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

166.

Chapter 5.

Figure 5.20 illustrates the nature of the terrain in the northwest part of the study region and indicates those archaeological locations causing the spike in Figure 5.18. The zone of darker grey in Figure 5.20 indicates the area where the weighted proximity to streamline variable is 50 or less. The red areas correspond to areas that are proximate to the headwaters of drainage channels. That such a pattern would occur in the middle of two long valleys seems counter-intuitive, and is something of a unique hydrological pattern.

Figure 5.20. Occurrence of archaeological features causing the spike in weighted streamline proximity.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

167.

Chapter 5.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

168.

Chapter 5.

The pattern in Figure 5.20 illustrates two important points about the distribution of archaeological features in the region. One is that despite identifiable trends, the example illustrated in Figure 5.18 shows that circumstances can arise which substantially modify the overall pattern. The second point is that this example is a good illustration of the effect of a spatially biased dataset. It is likely that even in a dataset that is spatially representative, the spike seen in this variable would still have been present, but its dominance is exaggerated by the nature of the spatial bias that was present. Since it was a real feature of the distribution of archaeological features in the Calton Hills sub-region, no attempt was made to remove the locations causing the spike from the model. Such a procedure might otherwise have been done if the locations causing the spike were treated as outliers from the overall trend. One further feature is also worthy of note about the weighted streamline proximity variable, which is the decline in the number of archaeological features occurring at values greater than 88. This portion of the variable corresponds with the higher stream orders, or major river channels. For these areas, a pattern similar to that observed for waterholes occurs due to fewer archaeological features being located on the bigger drainage channels. As with waterholes, this seems a counter-intuitive result. However, the pattern can be understood by reference to similar processes. Areas along the bigger river channels are generally poorly represented by archaeological survey in the region, and due to their occurrence on low flat country, are more susceptible to disturbance from flooding. The absence of archaeological features in this zone may not be indicative therefore about the true nature of hunter-gatherer activity in these areas since there are potentially many undiscovered archaeological features along the major drainage lines. Despite this, the important implication of the pattern observed for weighted proximity to drainage is that whilst there is an overall trend for archaeological features to be located adjacent to drainage lines this is not a consistent pattern along their entire extent. Whether it is because of the way hunter-gatherers foraged in proximity to waterholes within drainage lines, or taphonomic effects associated with flooding, it is clear that the highest probability of archaeological features occurs on the middle stream orders. Such a finding is similar to the observations of Pickering (1994, 153), who found that stream order structured clan estates.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

169.

Chapter 5.

Archaeological locations

Random locations

6

8

10

12

14

16

18

20

22

24

26

28

Wetness

Figure 5.21. Histogram of wetness for all locations.

Probabilty of archaeological location presence

1.00

0.75

0.50

0.25

0.00 5

7

9

11

13

15

17

19

21

23

25

27

Wetness

Figure 5.22. Regression curve for wetness.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

170.

Chapter 5.

The final ratio variable used in this study for spatially modelling the occurrence of archaeological features was wetness. As with many of the other variables, wetness also demonstrated a high degree of similarity between the occurrence of archaeological and random locations, as is indicated in Figure 5.21. However, Figure 5.21 also shows that the higher occurrence of archaeological features at higher wetness values, corresponds to areas with greater capacity for retaining surface water. These areas are primarily drainage channels and their surrounds. This trend is borne out in the nature of the regression curve depicted in Figure 5.22.

5.5. Model formation. As explained in the previous chapter, logistic regression was used, with a generalised additive model, to produce probabilistic predictions about the occurrence of archaeological features throughout the region. Figure 5.23 illustrates the results calculating these predictions using all locations containing archaeological features and the ratio variables already discussed. In this figure, the darker areas indicate a higher probability of an archaeological feature to occur at that location. Of particular interest in this figure, are the locations with low probability on the tops of the ranges, and in the north east of the study region. The low probability of archaeological features occurring on the tops of the ranges is readily explained through reference to the difficulty of accessing these areas and the scarcity of crucial resources, especially water. The low probability in the north east portion of the region is related to two aspects. As discussed above, visibility and an absence of sampling in this area has produced an artificially low predicted probability. In a similar environment in the south of the study region Davidson and Fife (1994) reported that the occurrence of archaeological features becomes very scarce on the plains, but isolated artefacts could still be found in this area. The main difference with the north east portion of the study region however is that it occurs at a lower elevation to that studied by Davidson and Fife. Since no recorded archaeological locations occur at this elevation, the model suggests the occurrence of archaeological features is zero. This is unlikely to be the case, but it is unverifiable until some archaeological survey is undertaken in this part of the region.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

171.

Chapter 5.

Of further note in this model, is the high probabilities predicted along the upper stream orders and valley bottoms within the ranges. Such a pattern reflects the high occurrence of art sites in these areas, particularly on the sides of these valleys, and the concentration of archaeological survey in them also. Archaeological visibility and integrity in the valley bottoms is good, making the detection of archaeological features easier also. In the south and west of the region, on the plains, the probability of archaeological features drops in accordance with the smaller amount of survey conducted in these areas, and their poor visibility and archaeological integrity.

Figure 5.23. Probability model for all locations. TV = 0.46; χ2 = 428.35; P < 0.001; AP = 76.2%.

One shortcoming with employing only the ratio scale variables is that the model does not reflect changes in archaeological location probability associated with vegetation, soils or geology. However, due to the types of resources these variables are surrogates for, it is __________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

172.

Chapter 5.

expected that the probability of archaeological feature occurrence would be different between them as people preferentially exploited some resources over others. This issue was considered above for these variables only, but here they are incorporated with the ratio variables to produce a series of composite models.

Figure 5.24. Probability model for all locations incorporating soils. TV = 0.41; χ2 = 374.21; P < 0.001; AP = 74.5%.

The first of these composite models introduces soils into the model, which is illustrated in Figure 5.24. It is readily apparent that the sampling problem encountered with soils above is also reflected in the composite model. Nonetheless, the model demonstrates important aspects about the nature of archaeological feature probability within nominal variable categories. One is that archaeological feature distribution is not uniform within individual categories. Although individual soil types have contrasting densities of __________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

173.

Chapter 5.

archaeological features, within them, variability in archaeological feature probability is brought about by variation in ratio variables. Hence, the proximity to streamlines, slope or elevation all act to modify the pattern of probability within individual units. Another noteworthy feature introduced by introducing soil types into the model is that those soil units containing no archaeological features have probabilities equal to zero. One instance of this can be seen in Figure 5.24 in the south east of the region in the Selwyn Ranges. Since this unit is small, the result of the model in this instance is more likely to reflect sampling bias. Adjacent areas with similar characteristics in terms of the ratio variables indicate that this area should have quite a high probability of containing archaeological features. Similarly, the converse is also true for the very high probabilities returned by the model in the north west of the region. The high density of survey in this area has driven up the probability just as it did in Figure 5.5. Such problematic issues with representative sampling of the soil types occurring in the region, make it difficult to justify the incorporation of soils into models of archaeological distributions. One final implication of the incorporation of soil types is that it suggests that, at least in some areas, higher probabilities may be possible in the north east portion of the study region where no archaeological features have been recorded. Many of the sites recorded around the township of Cloncurry occur on a soil unit that extends into this north east portion. The medium range probability associated with this soil type suggests that the low probability produced from the model based solely on ratio variables may not be a good reflection of archaeological potential in this area. Figure 5.25 illustrates the result of incorporating geology into the model based on ratio variables only. In this case the problems introduced by biased sampling of geological types is less prominent, but nonetheless have an affect on the model. Areas on the eastern side of the Selwyn Ranges have zero probabilities, and the probability of a band in the north of the region has been raised to almost one. Also similar to the soils composite model, the north west portion of the region produces areas with much higher probability, but again this is likely to reflect the greater density of survey undertaken in this area.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

174.

Chapter 5.

Figure 5.25. Probability model for all locations incorporating geology. TV = 0.45; χ2 = 502.97; P < 0.001; AP = 78.2%.

Despite its problems, geology introduces changes to the model that are more indicative of known archaeological distributions in the region. The dark band in the central north of the region is a good example of these changes. The geological unit associated with this band contains a rock type that was exploited for stone axe production in several other areas of the region. The rock type is particularly exploited for example in the northwestern portion of the region but is more difficult to discern due to the high probabilities predicted for these areas to begin with. A further example exists in the southern extension of the region where a broad band of higher probability corresponds with a silcrete unit where a large number of silcrete quarries have been located (Davidson and Fife 1994, Ridges et al. 2001).

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

175.

Chapter 5.

For the remainder of the region, the introduction of the geological variable has produced little change in the pattern of probability, except for those geological units without recorded archaeological features. One notable exception is the north east corner where there are now the appearance of areas with low (rather than nil) probability. This is due to the broad expanse of the alluvial geological unit, which in this case is suggesting that there are some similarities between all areas on the alluvium in terms of archaeological potential. This reinforces the need for future survey in this portion of the region. Despite the difference in the north eastern corner of the region, the generally isolated differences produced by the introduction of the geology variable indicate that its role in determining the occurrence of archaeological features is feature-type specific. For instance, geology is a useful variable only for those types of features that are highly correlated with particular geological units. In this study there are two such features types, stone quarries and art sites. Subsequent use of the geology variable is therefore restricted to models concerning these features types only. The introduction of the LANDSAT data introduces an important difference to the model, but due to the chaotic distribution of the LANDSAT categories, the difference tends to blend in with the overall pattern of probability in the model. In order to make the difference more visible Figure 5.27 shows two maps portraying the difference between the models. The left map indicates the areas where the LANDSAT model is lower than the model based on ratio variables, and the right map shows those areas where the LANDSAT map predicts higher probabilities. The darker areas in both maps indicate where the two models vary the most.

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

176.

Chapter 5.

Figure 5.26. Probability model for all locations incorporating LANDSAT classification. TV = 0.49; χ2 = 452.72; P < 0.001; AP = 71.4%.

The amount of white in both maps illustrates that the two models are quite similar. However, the areas where the LANDSAT model predicts lower probabilities generally occur at the interface between the ranges and lowlands. Whereas the places where the LANDSAT model predicts higher probabilities tend to occur on top of the ranges. Since the LANDSAT classification was classification using vegetation as a guide, the differences therefore reflect correlation between the archaeological data and biotic variation in the landscape. Those areas where the LANDSAT model predicts lower probabilities indicate areas where the abiotic variables overestimate the occurrence of archaeological features. Conversely, the areas in the LANDSAT model which are higher than the original model are places where biotic variation is an important component determining the location of archaeological features. __________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

177.

Chapter 5.

Figure 5.27. Difference introduced by adding LANDSAT data to model.

The incorporation of the LANDSAT data is therefore important in putting some controls upon the correlation between archaeological features and the abiotic variables used in the model. Importantly, the incorporation of the LANDSAT data into the model indicates that the areas where biotic variation makes a significant contribution to the location of archaeological features tend to be fairly restricted and localised. Such a pattern would be consistent with Aboriginal people exploiting particular plant resources in a way that leads to a detectable archaeological signature in certain parts of the landscape. With more detailed vegetation data in the areas where the LANDSAT model produces higher predictions, the relationship between the archaeological features and vegetation types may be discernible. However, such attempts were beyond the scope of this study. The areas where the addition of the LANDSAT variable produces lower probabilities are important for indicating variation in the occurrence of archaeological features brought about through subtle changes in areas with similar abiotic characteristics. The importance of the LANDSAT variable is that it reflects historical (for example fire or flooding events) or local variability in the abiotic variables that are undetectable in broader scale mapping. The broad areas where the LANDSAT variable suggests archaeological features are less frequent illustrate the importance of the LANDSAT data for shedding light on areas where the abiotic variables demonstrate little variability. As Figure 5.27 shows, these areas of lower probability correspond with what are probably edaphic changes undetectable at the scale of the terrain mapping. Consequently, the

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

178.

Chapter 5.

LANDSAT data is a useful variable for providing a more detailed picture of local scale variability at the interface between the ranges and lowlands.

Figure 5.28. Probability model incorporating all nominal variables. TV = 0.49; χ2 = 562.78; P < 0.001; AP = 79.5%.

Figure 5.28 presents the results of incorporating of all three nominal scale variables. For context, this figure also displays the archaeological locations used in the construction of the model. The dominant impact of the soil data is readily discernible in Figure 5.28. This is problematic for not only the bias it introduces as a result of the sampling particularities of the archaeological data, but also the contrasting predictions it makes in areas where few or no archaeological locations were recorded. To a lesser extent, the same is also true of the geological variable. Such a pattern highlights the problem with incorporating

__________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

179.

Chapter 5.

variables without careful consideration of their appropriateness to the dataset, or to what is attempting to be achieved with the model. In this case, the goal was to produce a generic model describing the trends in the occurrence of archaeological features across the region. Subsequently, differences produced by variation in the type of archaeological features, and their attributes, could be measured against this model. Although the incorporation of soils and geology undoubtedly are important determinants of the location of archaeological features, the problem of representative sampling make it difficult to assess their usefulness at this scale of analysis. Certainly, the results that Figure 5.28 illustrate suggest that they are not useful. Subsequently, the regional model incorporating the LANDSAT data was taken as the most indicative of the regional pattern of archaeological location distribution at the most generic classificatory level. The examination of changes introduced by altering the spatial and classificatory resolution of the archaeological dataset were performed using the model illustrated in Figure 5.26. The model incorporating all nominal scale variables, and the one incorporating soils were not used for subsequent analysis. The incorporation of geology into subsequent models was subsequently restricted to cases where it was judged an essential component of the distribution of particular features and/or their attributes.

5.6. Location class distributions. The models examined in the previous section provide an overview of the occurrence of archaeological location regardless of the type of feature they may contain. Although this is the most common approach adopted in the majority of predictive modelling projects (Ebert and Kohler 1988, Kvamme 1985, Warren 1990, Westcott and Brandon 2000), it nevertheless blends variation in the occurrence of many types of archaeological feature which are associated with diverse types of behaviour. For instance, the types of activities and the context in which art sites occur are very different from those for stone quarries or open campsites. Consequently, in order to understand better the function and context of different kinds of Aboriginal behaviour, it was necessary to break apart the classification of archaeological locations into the kind of features that comprise them. __________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

180.

Chapter 5.

Table 5.6 summarises the seven classes of archaeological location adopted in this study. As was explained in Chapter 4, these classifications are mutually exclusive groups. In this form of classification, a rockshelter can only occur in that class, it cannot also be a quarry or campsite. Thus, the sum of the frequencies for each class equals the total number of locations in the database. The classifications were primarily derived by the occurrence of dominant activities (for example quarrying versus camping) and the context within each class occurs (for instance rockshelters versus open campsites). The classification of open locations into individual finds, scatters and sites was based on the classifications adopted by recorders who recognised the groups primarily on the number and kind of artefacts at each location.

Table 5.6. Summary of location classes.

Site Class

Frequency

Frequency

(All locations)

(Sites only)

Open individual finds

395 (23%)

2 (