Prediction and assessment of local stream habitat

Freshwater Biology (2000) 45, 343–369

APPLIED ISSUES

Prediction and assessment of local stream habitat features using large-scale catchment characteristics NERIDA M. DAVIES, RICHARD H. NORRIS and MARTIN C. THOMS Cooperative Research Centre for Freshwater Ecology, University of Canberra, ACT 2601, Australia

SUMMARY 1. Knowledge of what a habitat should be like, in the absence of the effects of human activities, is fundamental to local stream habitat assessment. It has been suggested that stream habitats are influenced by large-scale catchment features. This study aimed to identify these relationships so that local-scale habitat features could be predicted from larger-scale characteristics. 2. Fifty-one reference sites from the Upper Murrumbidgee River catchment, south-eastern Australia, were classified on the basis of the local features of their stream habitat. Large-scale variables, namely catchment area, stream length, relief ratio, alkalinity, percentage of volcanic rocks, percentage of metasediments, dominant geology and dominant soil type, provided sufficient information for classifying 69% of reference sites into appropriate reference site groups. 3. A model created using these large-scale catchment variables was able to predict the local habitat features that were expected (E) to occur at a site in the absence of the effects of human activities. These were compared with observed (O) local habitat features to provide an observed-to-expected (O/E) ratio, an assessment score of the habitat at a site. The departure of this ratio from 1 enables identification of those sites that may be impacted. A list of habitat features that are expected at a site can provide targets for habitat restoration or enhancement. 4. For impacted sites, when habitat assessment from the habitat predictive model was compared with biological assessment from the Australian River Assessment System (AUSRIVAS) predictive model, it was possible to identify whether habitat degradation or water quality degradation was the cause of biological impairment. Such assessment may make it possible to identify rehabilitation goals relevant to the biota. Keywords: assessment, AUSRIVAS, catchment scale, macroinvertebrate habitat, predictive model, river condition

Introduction Stream habitat forms an essential component of river ‘health’ (Maddock, 1999) that can be used to evaluate

Correspondence: Nerida M. Davies, Cooperative Research Centre for Freshwater Ecology, University of Canberra, ACT 2601, Australia. E-mail: [email protected] © 2000 Blackwell Science Ltd

the overall ecological integrity of a river system (Muhar & Jungwirth, 1998). The condition of local stream habitat, otherwise known as the habitat templet, influences the structure and organization of biological communities (Hynes, 1970; Southwood, 1977; Swanson, 1980; Minshall, 1984; Sweeney, 1984; Southwood, 1988; Townsend & Hildrew, 1994; Downes, Lake & Schreiber, 1995). In the absence of water quality impairment, the local-scale physical habitat 343

344

N.M. Davies et al.

will have a major influence over the biotic assemblages at a site. If water quality is of a high standard, diverse and abundant assemblages of stream biota are likely to exist if the local stream habitat is supportive (Plafkin et al., 1989; Barbour et al., 1999; Simpson & Norris, 2000). Habitat assessment can be used to determine the potential of the stream to support and maintain biota that are comparable to those found in natural habitats (Plafkin et al., 1989; Muhar & Jungwirth, 1998; Barbour, 1991). Contemporary assessment of river health often involves some form of rapid biological assessment (Reynoldson et al., 1997; Norris & Thoms, 1999). Rapid assessment approaches have focused on aquatic organisms. Empirical models have been developed that predict the occurrence of macroinvertebrate taxa, based on their association with environmental variables at reference sites (Wright, 1995; Reynoldson et al., 1997; Simpson & Norris, 2000). This approach provides an independent way of matching new sites with reference sites, enabling predictions to be made. Thus, the philosophy of this approach seems appropriate. However, these techniques have not been applied to features of streams other than their biota. Stream managers often require information about physical features of a river that need improving to enhance biological condition. Most of the rapid assessment approaches have limited ability to determine whether biological impairment results from poor water quality or from poor habitat. Therefore, the ability to predict local stream habitat features may be useful for distinguishing between the effects of water quality and the effects of habitat on biological condition, and may assist in river management. Before a habitat can be identified as damaged, it is vital to know what the habitat should be like in the absence of effects from humans. Stream habitat may be influenced by a variety of factors operating at numerous spatial and temporal scales (Frissell et al., 1986; Richards, Johnson & Host, 1996). Hynes (1975, p. 12) eloquently argued that ‘…in every respect the valley rules the stream’. Frissell et al. (1986) formalized Hynes’s view and presented a framework that divided the catchment into five spatially nested scales (stream, segment, reach, pool –riffle and microhabitat), in which each scale is constrained to some extent by the levels above it. Within this framework, geology and climate can be viewed as ‘independent’

(Schumm & Lichty, 1965; Schumm, 1977), or ‘state’ (Lotspeich, 1980), or ‘ultimate’ (Naiman et al., 1992) controlling factors of ecosystems. Essentially, these two factors —geology and climate —provide the stage on which physical and biological components of ecosystems interact, via large-scale controls of chemistry, hydrology and sediment delivery (Schumm & Lichty, 1965; Morisawa, 1968; Schumm, 1977; Lotspeich, 1980; Swanson, 1980; Schumm, 1991; Naiman et al., 1992; Allan & Johnson, 1997). It has been recognized that variables operating at larger spatial scales may be more important than local-scale variables in influencing or controlling stream habitat (Richards et al., 1996) and biota (Parsons & Norris, 1996Richards et al., 1996Richards et al., 1997; Roth, Allan & Erickson, 1996; Marchant et al., 1997). Despite this, few habitat assessment approaches have related conditions in the surrounding catchment to the condition of the local stream habitat. Numerous approaches have been developed for assessing stream habitat, many of which have been reviewed by Maddock (1999). However, the majority of these approaches are specific to a particular taxon (macroinvertebrates and fish, PHABSIM: Physical Habitat Simulation System; Bovee, 1996), or make measurements solely at the reach scale (PHABSIM, Habitat Quality Index; Binns & Eiserman, 1979), or are not process orientated (IBI: Karr, 1981; ICI: Ohio EPA, 1987). There is a need for a habitat assessment approach that places the local stream habitat in the context of the larger system and relates it to the surrounding ecosystem including the biota. Habitat assessment is beginning to focus on nationally applicable and ecologically based approaches that incorporate comparisons to reference or target conditions. The River Habitat Survey (Raven et al., 1997, 1998; Jeffers, 1998) and the United States Environment Protection Agency Rapid Bioassessment Protocols (Plafkin et al., 1989; Barbour et al., 1999) use reference conditions to assess stream habitat at broad scales (state-wide or nationally). Maddock (1999) has described the need for methods that can examine the existing condition of stream habitat and that can determine conditions that might have existed before impact. Jeffers (1998) determined the probability that some habitat features from the River Habitat Survey (Raven et al., 1997, 1998) should occur at a site given certain conditions (altitude, slope, distance from source and height of the source). However, this pre© 2000 Blackwell Science Ltd, Freshwater Biology, 45, 343 – 369

Prediction and assessment of local stream habitat 345 dictive ability needs to be developed further and incorporated into habitat assessment. With the exception of those of Jeffers (1998), few methods are capable of predicting the habitat features that should exist in the absence of human impact. An appropriate habitatassessment technique is required — one that is applicable at broad scales, directly related to the biota, process orientated, and capable of predicting and assessing the condition of the local stream habitat by identifying features of the habitat in need of rehabilitation. This study develops an empirical model that uses larger-scale catchment features little affected by hu-

man use to predict those local habitat features that should occur at a site in the absence of the effects of human activities. The theoretical frameworks proposed by Schumm & Lichty (1965), Schumm (1977), Lotspeich (1980) and Frissell et al. (1986) were drawn on to identify features of the catchment that may influence or control the local stream habitat.

Methods Study area The study was conducted in the upper Murrum-

Fig. 1 Sites from the Upper Murrumbidgee River catchment used in the development and testing of the habitat model. © 2000 Blackwell Science Ltd, Freshwater Biology, 45, 343–369

346

N.M. Davies et al.

bidgee River catchment, Australia, and encompassed an area of approximately 12 000 km2 (Fig. 1), incorporating the Murrumbidgee River and its tributaries upstream of Burrinjuck Dam. This area includes the Yass, Goodradigbee, Cotter, Gudgenby, Naas, Queanbeyan, Molonglo, Bredbo, and Numeralla sub-catchments. The Cotter, Molonglo, and Queanbeyan rivers, and the head-water streams of the Murrumbidgee River are all regulated. The catchment contains areas of high relief including part of the Australian Alps near Tantangara Dam and the Brindabella Ranges to the west of Canberra. Escarpments, high hills and ridges characterize the Monaro region to the southeast of Canberra (Harriman & Clifford, 1987). Over the entire catchment, rainfall is generally uniform throughout the year. The south of the catchment receives 500 – 700 mm, the north receives 600–700 mm and the ranges in the west of the catchment receive 800 –1200 mm of rain per year (Bureau of Meteorology, Canberra office, pers. comm.). Land uses within the catchment include national parks, forestry, agriculture, rural residential, recreational and urban areas. Most of the national parks and State forests are located in the ranges in the west of the catchment. Agricultural activity is concentrated in the valleys, the Monaro plains south-east of Canberra, areas around Cooma and in the north towards Yass. Parts of the Upper Murrumbidgee River catchment have been urbanized, including Canberra, Queanbeyan, Cooma and Yass.

Site selection A large dataset of habitat variables was available for 75 sites within the Upper Murrumbidgee River catchment, collected as part of the Australian Capital Territory (ACT) component of Australia’s National River Health Program (NRHP). Fifty-nine of these sites had complete datasets and were deemed appropriate for this study. The reference site dataset was originally collected with the aim of characterizing the rivers of the region on the basis of their macroinvertebrate composition (Davies, 1994). The sites were chosen to represent the biological condition of reference sites with minimal disturbance by human activities. Test-site data were collected at 21 sites located within the Upper Murrumbidgee River catchment in May–June 1998 (Fig. 1). Of these sites, 10 were resampled reference sites and the remaining 11 were

new test sites. These test sites were chosen to represent a variety of land uses and likely water quality impacts, and a range of stream types. Land uses surrounding test sites included grazing, cropping and urban development, and some of the test sites were also affected by river regulation.

Physical and chemical water quality variables A water sample (250 mL) was collected at each site and stored on ice until return to the laboratory where it was frozen until analysed for total nitrogen (TN), total phosphorus (TP) and nitrate or nitrite (NOx ) using flow injection analysis (APHA, 1992). Water temperature, electrical conductivity, pH, dissolved oxygen and turbidity were measured at each site using a multi-probe meter (Hydrolab model Scout 2; Hemel Hempstead, UK). Total alkalinity was measured in the field by titration to pH 4.5 (APHA, 1992). A Hydrological Services pygmy flow meter was used to measure water velocity in riffle and edge habitats (three measurements in each). Rainfall data were obtained for 26 stations throughout the study area from the Australian Bureau of Meteorology. Mean annual and median annual rainfall were calculated for the sites using the contour method (Gregory & Walling, 1973). Discharge data were obtained from the Department of Urban Services, Canberra, for 13 gauging stations within the Upper Murrumbidgee River catchment.

Invertebrate sampling Macroinvertebrates were collected separately from riffle and edge habitats during May and June 1998 and processed using standardized methods described by Davies (1994) and Parsons & Norris (1996). A triangular frame kick-net, 350 mm across the bottom with a 250 mm mesh, was used to collect macroinvertebrates in the riffle by disturbing the substratum upstream of the net using a kicking action over a distance of 10 m. The edge habitat was sampled with a similar net by sweeping the stream margins with a vigorous action for a distance of 10 m. The nets were washed thoroughly between collections. Samples were placed in labelled containers and preserved in 10% formalin buffered with calcium carbonate (CaCO3). Rose Bengal stain (100 g L − 1) was added to assist with sorting (Mason & Yevich, 1967). Samples were thoroughly rinsed, using a 250 mm © 2000 Blackwell Science Ltd, Freshwater Biology, 45, 343 – 369

Prediction and assessment of local stream habitat 347 Table 1 Physical, chemical and habitat variables measured at the test and reference sites used in the construction of the AUSRIVAS predictive models*

Table 1 (Continued)

Variable

Description

Larger-scale locational STORDERR, E ALTITUDER, E DFSR, E LATITUDE LONGITUDER, E CATCHAREAR ALKALINITYR

characteristics Stream order Altitude (m) Distance from source (km) Latitude (degrees/minutes, e.g. 3519) Longitude (degrees/minutes, e.g. 14857) Catchment Area upstream (km2) Alkalinity (mg L−1)

Habitat assessment SUBSTRATE EMBEDNES CHANNALT SCOURING PRRBRATO

Local scale habitat features RIPWIDTHR Riparian width (m) TREES\ 10E % Trees\ 10 m TREESB 10 % TreesB 10 m SHRUBVINER % Shrubs and vines GFSR, E % Grasses ferns sedges SHADING Shading of reach (Categories 1–5) STREAMWIDTH Stream width (m) BANKWIDTH Bank width (m) BANKHEIGHT Bank height (m) RIFFAREA % Riffle habitat RDEPTH Riffle depth (cm) RVELOCITY Riffle flow (m s−1) EDGEAREA % Edge habitat EDEPTHE Edge depth (cm) EVELOCITYE Edge flow (m s−1) Reach specific characteristics REBEDROCKR, E % Bedrock RECOBBLE % Cobble REPEBBLER, E % Pebble REGRAVEL % Gravel RESAND % Sand REMACRO % Macrophytes Riffle characteristics RBEDROCK RBOULDER RCOBBLER RPEBBLE RGRAVEL RSAND RMACROR Edge characteristics EBANKVEG EBEBROCK EBOULDER ECOBBLE EPEBBLEE EGRAVEL ESAND EMACROE

% % % % % % %

Bedrock Boulder Cobble Pebble Gravel Sand Macrophytes

Edge bank vegetation (Categories 1–4) % Bedrock % Boulder % Cobble % Pebble % Gravel % Sand % Macrophytes category (Categories 0–4)

© 2000 Blackwell Science Ltd, Freshwater Biology, 45, 343–369

Variable

BANKSTAB VEGSTAB VEGCOVERE HABSCORE

Description variables Substrate (Categories 0–20) Embeddedness (Categories 0–20) Channel Alteration (Categories 0–15) Scouring (Categories 0–15) Pool/riffle run/bend ratio (Categories 0–15) Bank stability (Categories 0–10) Vegetation stability (Categories 0–10) Vegetation cover (Categories 0–10) Habitat score (total of habitat assessment variables)

* Predictor variables used by the ACT region Autumn edge AUSRIVAS model are indicated by E. Predictor variables used by the ACT Autumn Riffle AUSRIVAS model are indicated by R . The local-scale habitat features were also used in the construction of the habitat model developed here.

mesh sieve to remove fine sediment and preservative, and placed in a sub-sampling box based on the one designed and tested by Marchant (1989). Sub-samples were extracted, sorted and identified until a total of 200 animals were collected (Davies, 1994; Parsons and Norris, 1996). A stereo microscope ( ×10 magnification) was used during sample sorting. Macroinvertebrates were identified to family level with the exception of Oligochaeta (class), Turbellaria (class), Hydracarina (order) and Chironomidae (sub-family), using keys recommended by Hawking (1994). Macroinvertebrate and habitat data were entered into the autumn riffle and edge AUSRIVAS predictive models (http://ausrivas.canberra.edu.au/ausrivas). The outputs of the model include the observed to expected ratio (O/E), which provides an assessment of biological impairment at a site (Simpson & Norris, 2000). A series of bands are used to indicate the level of impairment including; equivalent to reference (Band A), below reference (Band B), well below reference (Band C), impoverished (Band D) and above reference (Band X) (Simpson & Norris, 2000).

Local (reach-scale) habitat variables Habitat variables were recorded using standardized NRHP methods (Davies, 1994; Parsons & Norris, 1996; Table 1). The habitat survey used was originally based on and included the habitat assessment component of the United States Environmental Protection

348

N.M. Davies et al.

Agency rapid bioassessment protocols (Plafkin et al., 1989; Barbour et al., 1999). The habitat assessment rates the following habitat features along a continuum from excellent to fair: bottom substrate availability, embeddedness, velocity–depth, channel alteration, bottom scouring and deposition, pool – riffle : run –bend ratio, bank stability, and streamside cover (Plafkin et al., 1989; Barbour et al., 1999). Excellent scores equate to high levels of bottom substrate availability, velocity–depth, pool – riffle : run –bend ratio, bank stability, and streamside cover and low levels of embeddedness channel alteration and bottom scouring and deposition.

Formation of reference groups The Bray– Curtis similarity measure was applied to the habitat data because it is a robust measure of association for cluster analysis (Faith, Minchin and Belbin, 1987). To establish groups of reference sites, individual reference sites were classified by categorical habitat data, using flexible unweighted pair – group arithmetic averaging (UPGMA b = −0.1) following the recommendation of Belbin & McDonald (1993) (Fig. 2, step 4). The classification was examined as a dendrogram so that the level of group fusion could be selected (Belbin, 1993). Groups containing less than five sites were joined with other groups to prevent poor group representation (Wright, Furse & Armitage, 1993).

Large-scale catchment variables Large-scale catchment variables were deduced from topographic, geologic, and soil maps (Table 2). These large-scale catchment variables were measured for all the previously and currently sampled sites within the Upper Murrumbidgee River catchment.

Data analysis Development of a habitat model The methods used for developing a model that could predict stream habitat features at a site from its catchment characteristics were based on those used in the NRHP for the AUSRIVAS models (Simpson & Norris, 2000) and were originally based on those of Wright et al. (1984). The procedure involved grouping sites according to their local stream habitat characteristics, followed by discriminant function analysis to determine how well the groups described the structure in larger-scale environmental characteristics (Fig. 2). RIVPACS (Wright, 1995) and AUSRIVAS (Simpson & Norris, 2000) models predict the presence or absence of macroinvertebrate taxa. Therefore, habitat variables were converted to categories that were entered as separate entities (equivalent to taxa). Categories were selected to overcome non-linear distribution and also to enhance the predictive capabilities of the model. Most variables were divided into five categories and many followed a geometric distribution.

Predictor variables Large-scale catchment variables were related to the reference site groups formed from the classification, using a stepwise discriminant function analysis (stepwise MDFA) and the cross-validation procedure to select the large-scale catchment variables best able to discriminate among the site groups formed from the classification (Fig. 2, step 5). Before running the stepwise MDFA, the large-scale variables that were not normally distributed were transformed using the logarithmic transformation (log X+1) and re-examined for normality. To predict the probabilities of group membership for each reference site, the significant large-scale catchment variables from the stepwise MDFA were used in a multiple discriminant function analysis (MDFA) using the cross-validation procedure to find those that could predict group membership with the lowest error rate. Extra variables from the large-scale catchment list (Table 2) thought to be associated with the groups formed from their local-scale habitat features were added in a series of iterations to improve the prediction rate from the MDFA. The subset of large-scale catchment variables that produced the lowest error rate was used to predict group membership of new sites; they were also used as predictor variables in the model (Fig. 2, steps 5 and 6). The groups of reference sites based on their local habitat characteristics together with the large-scale environmental data form the basis of the predictive model. The combination of large-scale catchment © 2000 Blackwell Science Ltd, Freshwater Biology, 45, 343 – 369

C C

C R C

C

C

C

Predictor variables Catchment area (A) Relief ratio

Total stream length

Alkalinity

Dominant soil type

Geology

Dominant geology type

Remaining variables Catchment length (L)

Scale

© 2000 Blackwell Science Ltd, Freshwater Biology, 45, 343–369 km

%

mg L−1

km

km2 Ratio

Units

Category 1–4

1. = Alluvium; 2. = Volcanics; 3.= Metasediments; 4. =Limestone

1 =Duplex; 2 =Gradational; 3 =Lithosol

Rr=h/L

Formulae

Maximum length of the catchment from the site along a line approximately parallel to the main stream.

Compensating polar planimeter. Dividing the difference in altitude between the mouth and highest point on the drainage divide (h) by the maximum basin length (L). Total distance of all stream segments within a catchment. Measured in the field by titration to pH 4.5. Dominant soil type was determined from a 1:200 000 map. Three soil types were marked on the map. Categories 1–3 were used to allocate the soil type that dominated by percentage over the smaller catchments. 1:100 000 geologic maps of the Upper Murrumbidgee (except in Cooma region in which only 1:250 000 maps existed). Eight categories were used (later reduced to four). The area of each geologic category within the smaller catchments was measured with a polar compensating planimeter and then converted to a % of the total area for that catchment. The dominant geologic category to dominate the catchment above each site was used as a measure of the dominant geology type of that catchment.

Method

Gordon et al. (1992)

Horton (1945); Gordon, McMahon & Finlayson (1992)

Horton (1945) Schumm (1956)

Reference

Table 2 Catchment-scale (C), valley-scale (V) and reach-scale (R) variables measured from 1:100 000 topographic maps or from the field, with a description of the method of their measurement*

Prediction and assessment of local stream habitat 349

R R R R

Altitude

Conductivity

pH

Mean stream slope

mS cm−1

m

m

km

No units

km km−2

Ratio

Ratio

Units

(alt 0.75L)

Sb=(alt 0.85L)−(alt 0.1L)

RD =(tot L)/A

Dividing the catchment area (A) by the Horton (1932) squared length of the basin (L). Calculating the diameter of a circle Schumm (1956) with the same area as the area of the basin and then dividing this by the length of the catchment. Dividing total stream length (tot L) for Gordon et al. (1992) the catchment by the catchment area (A). Van Haveren (1986) Altitude at 85% of catchment length, minus altitude at 10% of the catchment length divided by the altitude at 75% of the catchment length. Exterior streams are the first order. Horton (1945) modified by Joining of two first order forms the Strahler (1952) second order. Junction of two second order streams forms a third order stream. Measuring the distance of the main stream channel from the reach to the source. Measuring the distance between the contours on each side of the stream. Taking the altitude at 1 km downstream of the site from the altitude at 1 km upstream of the site and dividing by the distance between the two. Measured at a site from 1:100 000 maps. Confirmed during sampling using Global Positioning System (GPS). Measured as altitude from contours at each site. Measured in the field using Hydrolab Scout II. Measured in the field using Hydrolab Scout II. Difference in altitude between Gordon et al. (1992) source and mouth divided by the length of the stream.

Rf=A/L 2

Reference

Method

Formulae

* Variables chosen as predictor variables for use in the habitat model are described at the start of the Table.

R

C

Distance from source

Latitude and longitude (degrees, minutes, seconds)

C

Stream order

V

C

Mean catchment slope

Valley slope

C

Drainage density

V

C

Elongation ratio (basin elongation)

Valley floor width

C

Scale

Form ratio (basin shape)

Table 2 (Continued)

350 N.M. Davies et al.

© 2000 Blackwell Science Ltd, Freshwater Biology, 45, 343 – 369


Fig. 2 Main steps in the development (left), testing and running the habitat model (right). Main statistical methods include classification using flexible UPGMA and discriminant function analysis (MDFA) in boxes with thicker border. Shaded boxes are outputs from the model. The numbers in brackets provide a guide to the order of the modelling process.

variables that produced the lowest error rate in the MDFA was used as an independent way of matching test sites to the groups of reference sites formed by classifying the local habitat characteristics. The model determines the probability that a test site belongs to each of the reference site groups based on the values of its large-scale predictor variables (Fig. 2, step 14). The frequency of occurrence of each habitat variable category in each of the classification groups was calculated (Fig. 2, step 7). The probability of a habitat © 2000 Blackwell Science Ltd, Freshwater Biology, 45, 343–369

category occurring at a site was calculated by multiplying the frequency of occurrence for each habitat category in a classification group by the probability of the test site belonging to that group, and summing the results for all the groups in the classification (Fig. 2, steps 8 and 15). These calculations follow the methods described by Wright et al. (1984). RIVPACS models consider taxa that have a probability of occurrence ] 0.01 (Wright et al., 1984) and AUSRIVAS uses those that have a probability of oc-

352

N.M. Davies et al.

currence of ] 0.5 (Simpson & Norris, 2000). The occurrences of taxa in the RIVPACS and AUSRIVAS models are considered independent of each other and there is the potential for all taxa to occur at a site. The habitat categories for each habitat variable are also exclusive and only one category for each variable can actually occur at a site. However, more than one category of the same variable can be predicted to occur (in the same way that many species in the same family might be predicted). In such cases, this may lower the probability of each category occurring at a site. The decision on the probability level to use was investigated by comparing the cumulative probability of occurrence of all the variables with their predicted occurrence. To find the habitat categories that were outside the range of those expected at similar reference sites, the 10th percentile of the reference habitat data was plotted. The cut-off probability of occurrence was chosen where it was near the mean and median for the habitat data outside the reference data, thus excluding many of the characteristics that were not representative of the reference condition. Habitat categories were only used in the habitat model developed here if their probability of occurrence was greater than or equal to this cut-off level. The sum of the probability of each habitat category occurring at a site provides the number of local habitat categories expected to be at a site in the absence of the effects of human activities (Fig. 2, steps 9 and 16). The expected (E) measure is compared to the number of local habitat categories observed (O) (Fig. 2, steps 10 and 17) to provide a measure of how different the habitat condition is at a site compared to the habitat expected (O/E ratio; Fig. 2, steps 11 and 18).

Validation of the model As an internal check of predictive ability and a measure of the distribution of O/E ratios for reference sites, the reference sites used to build the habitat model were run through their own model. Any sites for which the observed local habitat characteristics were less than 75% of those expected (O/E) were reviewed. Eight sites were found to have unusual habitat conditions, potential pollution, land use problems, or poor habitat assessment categories not meeting a priori criteria and were therefore removed because they were inappropriate reference sites.

Some sites may have a lower O/E ratio than 0.75 if their site type is under-represented in the reference site range. This problem can be rectified by extra sampling of any site types not adequately covered in the reference range, and then rebuilding of the model (Simpson & Norris, 2000). The constraints of the present study did not allow additional sampling; thus these sites were also discarded. Further validation of the model was achieved by testing the 10 reference sites re-sampled in this study and determining if the model would assess them as having habitat equivalent to the reference condition. The mean O/E ratio and two standard deviations from the mean (2 ×SD) of the O/E ratios were used to determine if sites lay within the range of the reference condition used to create the habitat model. The 11 test sites sampled in this study with known impacts were also run through the model to see if it could detect habitat degradation.

Separating water quality and habitat effects in biotic impairment To determine if it was possible to separate the effects of habitat and water quality degradation on biotic impairment, the outputs for the habitat and the AUSRIVAS models were compared. The comparisons were assisted by a list of missing habitat features from the habitat model, water quality measurements, notes and variables from the original survey sheets (e.g. presence of non-point source pollution and catchment erosion). A list of taxa that were expected but missing at sites in the study also helped us determine whether biological impairment had occurred because of habitat or water quality effects.

Results Discharge Below-average rainfall in seven of the eight months before sampling (Bureau of Meteorology, Canberra office, pers. comm.) resulted in low flows in the catchment. In the month before sampling, the mean discharges for the majority of rivers sampled were below long-term values (Table 3). Although sampling took place after recent rainfall, low flows still occurred at many sites. Sampling was completed before the high flows occurred, in June (Table 3). © 2000 Blackwell Science Ltd, Freshwater Biology, 45, 343 – 369

Prediction and assessment of local stream habitat 353 Water quality characteristics With the exception of turbidity, water quality characteristics at the reference sites were within the guidelines specified for maintaining ecosystem health (Table 4; Maher et al., 1994). Turbidity levels at the reference sites 26, 28 and 33 exceeded the recommended maximum (10 NTU) for rural rivers and streams (Maher et al., 1994). Water quality characteristics at the test sites were generally within the guidelines specified for maintaining ecosystem health (Table 4; Maher et al., 1994). The exception was turbidity levels at sites 193 and 272, which exceeded the recommended maximum (10 NTU; Maher et al., 1994). Alkalinity and conductivity levels at site 193 were greater than the recommended levels for domestic water supply.

Biotic site assessments (AUSRIVAS) The biotic condition of all but two of the 10 re-sampled reference sites (28 and 138) and one of the 11 test

sites (291), was assessed as below reference condition for one or both habitats (Table 5). Prolonged low flows are the likely cause of lower than expected macroinvertebrate richness at 80% of the reference sites and it is possible that a similar percentage of test sites was also affected by drought (Table 5). Differences in biological assessment between the edge and riffle habitats may have resulted from the different effects of drought on each habitat. Many of the sites (e.g. 3, 33, 129, 130, 139, 193, 267, 274, 283 and 286) had a large percentage of cover by periphyton (Table 6), while others were highly embedded (e.g. 19, 26, 282 and 283; Table 6). Numerically dominant taxa such as Chironominae (at reference sites 3, 8, 129, 130, 139, and test sites 273, 282 and 283) and Oligochaeta (at test sites 193, 263, 267, 272, 274 and 286) may be indicative of organic enrichment (Table 6). Other sites (19, 26, 272 and 293) were numerically dominated by Simuliidae and Gripopterygidae (sites 8, 33, 267 and 274; Table 6).

Table 3 Discharge for rivers in the study area* Location

May 1998 mean

May long-term median of mean monthly flows

June 1998 mean

June long-term median of mean monthly flows

Goodradigbee River at Wee Jasper Molonglo River at Burbong Bridge Molonglo River at Coppins Crossing Molonglo River at Oaks Estate Murrumbidgee River at Billilingra Murrumbidgee River at Halls Crossing Murrumbidgee River at Lobbs Hole Numeralla River at Numeralla School Queanbeyan River at ACT Border Queanbeyan River at Tinderry Queanbeyan River at Wickerslack Sullivan’s Creek at Barry Drive Sullivan’s Creek at Southwell Park

0.817

3.659

4.462

6.742

0.004

0.254

0.923

0.439

0.189

1.146

5.536

1.546

0.296

1.264

3.011

1.397

0.663

3.468

–

6.934

1.151

8.011

13.96

14.70

0.545

3.901

9.115

5.479

0.020

0.748

–

1.299

0.244

0.269

0.569

0.638

0.175

0.932

0.914

1.107

0.201

0.259

0.266

0.692

0.042

0.067

0.399

0.083

0.018

0.026

0.178

0.027

* Mean monthly discharge was calculated from mean daily discharge (m3 s−1). © 2000 Blackwell Science Ltd, Freshwater Biology, 45, 343–369

3 8 19 26 28 33 129 130 138 139

193 263 267 272 273 274 282 283 286 291 293

Reference sites 3 8 19 26 28 33 129 130 138 139

Test sites 193 263 267 272 273 274 282 283 286 291 293 0.00 0.00 0.00 0.00 0.02 0.00 0.02 0.00 0.01 0.01 0.00

0.00 0.00 0.03 0.05 0.05 0.07 0.07 0.00 0.00 0.00

TP (mg L−1)

1.84 0.11 0.11 0.84 0.41 0.20 0.23 0.26 0.34 0.07 0.19

0.13 0.06 0.34 0.48 0.51 0.46 0.07 0.08 0.15 0.03

TN (mg L−1)

1.60 0.01 0.00 0.75 0.03 0.00 0.05 0.02 0.06 0.03 0.01

0.01 0.01 0.01 0.02 0.01 0.01 0.01 0.01 0.01 0.00

NOx (mg L−1)

240 25 23 55 115 15 60 65 50 20 50

175 30 40 20 20 15 50 30 30 15

Alkalinity (mg CaCO3 L−1)

3.52 8.63 10.14 9.63 9.74 7.67 10.7 9.59 8.93 9.21 5.99

7.7 8.43 8.39 6.41 4.37 4.61 5.64 6.14 2.87 3.96

Water temperature (°C)

917 57.8 60 221 309 33.5 127.5 216 114.1 40.9 93.7

44.9 74.1 116.9 81.8 76.3 66.4 93.1 62.9 58.4 24.8

Conductivity (mS cm−1)

*Bold values indicated that the site exceeded recommended levels for that water quality characteristic (Maher et al. 1994).

Site

Location code

Table 4 Water quality characteristics of test and reference sites in the Upper Murrumbidgee Catchment, May–June 1998*

7.83 7.22 7.16 7.24 8.36 7.7 7.51 7.58 7.51 6.87 7.67

7.37 7.16 7.41 7.22 7.64 6.96 7.16 6.74 7.68 7.46

pH

95 95 84 109 94 97 79 96 103 106

99 97 103 97 95 97 101 106 94 91

Dissolved oxygen (% sat)

10.2 3.8 5.4 36.1 5.7 2.9 3.7 5 4.6 3.2 4.5

1.5 3.7 9 17.2 17.5 18.4 5 7.7 4.4 5.2

Turbidity (NTU)



Prediction and assessment of local stream habitat 355 Development of the habitat model Formation of reference groups Four groups were defined from the classification of local-scale categorical habitat data. Site 132 was removed from initial analyses because it was not strongly affiliated with any group. Site 12 was also removed because it was surrounded by moderate catchment erosion, grazing and potential non-point source pollution and did not adequately represent the reference condition. Sites from group 1 were generally fourth or fifth order streams in large catchments, surrounded by native forest or some grazing activity and characterized by stable and vegetated banks, good streamside cover, deep riffles and pools and little embeddedness. Many of the sites placed into group 2 were located in national park, pine or native forest with large percentages of cobble particles in the riffle and excellent bottom substrate availability.

Group 3 sites were generally located in large catchments in pastoral or grazing areas with good bottom substrate availability, good pool – riffle : run– bend ratio and bank vegetative stability. Nearly all the sites from group 4 were surrounded by native grasslands or national park and characterized by excellent bank vegetative stability and low alkalinity levels.

Predictor variables Catchment area, stream length, relief ratio, alkalinity, percentage of volcanic rocks, percentage of metasediments, dominant geology and dominant soil type provided sufficient information for classifying 69% of reference sites into the appropriate reference site groups formed using their local habitat characteristics (Table 7). The highest percentage of correctly placed sites was in group 4 and the lowest was in groups 1 and 3.

Table 5 Overall biotic site assessment from the combined observed/expected (O/E) taxa assessment of the ACT Autumn Riffle and the ACT Autumn Edge AUSRIVAS models* Site band riffle habitat

O/E taxa edge habitat

Site band edge habitat

Overall site assessment

Reference sites 3 – 8 0.53 19 0.62 26 0.6 28 1.07 33 0.65 129 1.01 130 0.95 138 1.03 139 0.81

– C C C A B A A A B

0.64 0.58 0.81 – 0.88 0.8 0.83 0.64 1 0.98

B C B – A B B B A A

B C C C A B B B A B

Test sites 193 263 267 272 273 274 282 283 286 291 293

– B C B D B – C – X B

0.6 0.82 0.75 0.89 0.48 1 0.78 0.64 0.57 0.9 0.91

B B B A C A B B C A A

B B C B D B B C C X B

Site

O/E taxa riffle habitat

– 0.83 0.48 0.66 0.35 0.87 – 0.54 – 1.19 0.83

*Band A is equivalent to reference, B is below reference, C is well below reference, D is impoverished and X is above reference. The dash (–) indicates no biological assessment was made for this habitat. © 2000 Blackwell Science Ltd, Freshwater Biology, 45, 343–369


C

B

B

B

33

129

130

C

19

26

C

8

Reference sites 3 B

Site

Ancylidae, Sphaeriidae, Hydrophilidae, Elmidae, Simuliidae, Podonominae, Corixidae, Coenagrionidae, Gomphidae, Calamoceratidae

Hydrobiidae, Sphaeriidae, Hydrophilidae, Elmidae, Podonominae, Coenagrionidae, Gomphidae, Calamoceratidae

Baetidae, Caenidae, Corydalidae, Gomphidae, Hydrobiosidae, Philopotamidae, Hydropsychidae, Ecnomidae, Conoesucidae

Psephenidae, Tipulidae, Baetidae, Corydalidae, Gomphidae, Glossosomatidae, Philopotamidae, Hydropsychidae, Ecnomidae, Conoesucidae

Psephenidae, Podonominae, Tanypodinae, Baetidae, Leptophlebiidae, Corydalidae, Gomphidae, Hydroptilidae, Philopotamidae, Hydropsychidae, Conoesucidae

Psephenidae, Tipulidae, Baetidae, Caenidae, Corydalidae, Gomphidae, Glossosomatidae, Philopotamidae, Hydropsychidae, Ecnomidae, Conoesucidae

Hydrobiidae, Ancylidae, Acarina, Hydrophilidae, Simuliidae, Podonominae, Corixidae, Coenagrionidae, Gomphidae, Calamoceratidae, Leptoceridae

Missing taxa

Oligochaeta (33%), Chironominae (27%)


Gripopterygidae (39%)

Oligochaeta (30%), Simuliidae (19%)

Simuliidae (51%), Oligochaeta (25%)

Gripopterygidae (36%), Chironominae (22%)


Numerically dominant taxa (% contribution)

Low flow, 65–90% of edge covered by periphyton

\ 90% of edge covered by periphyton

Slight turbidity, barrier upstream, 35–65% riffle covered by periphtyon

Low flows, slight turbidity, 50–75% cobble, boulder particles surrounded by fine sediment

Low flows, slight turbidity, lack of stable habitat, gravel, cobble and boulder particles over 75% surrounded by fine sediment, sedimentation, moderate catchment erosion

Low flow

Low flow, \ 90% of edge covered by periphyton

Potential impact indicators

Prolonged low flows

Prolonged low flows

Prolonged low flows

Prolonged low flows, sedimentation

Prolonged low flows, sedimentation, rural runoff

Prolonged low flows

Prolonged low flows

Possible impacts

Table 6 Summary information for sites falling below reference conditions including: the overall site assessment; a list of the taxa predicted to occur (based on the AUSRIVAS model) but which were not collected; the percentage contribution of numerically dominant taxa; the potential impact indicators; and the possible impact at the site



B

139

B

D

B

D

B

B

263

267

272

273

274

282

Test sites 193 B


Site

Table 6 (Continued)

Oligochaeta (88%)

Chrionominae (43%)


© 2000 Blackwell Science Ltd, Freshwater Biology, 45, 343–369 Gripopterygidae (22%), Oligochaeta (17%)

Chironominae (70%)

Simuliidae (36%), Oligochaeta (22%)

Planorbidae, Elmidae, Corixidae, Chironominae (25%), Synlestidae, Ecnomidae, Conoesucidae Atyidae (25%) Orthocladiinae (18%)

Hydrobiidae, Podonominae, Tanypodinae, Baetidae, Leptophlebiidae

Hydrobiidae, Ancylidae, Podonominae, Tanypodinae, Chironominae, Baetidae, Leptophlebiidae, Caenidae, Gomphidae, Hydrobiosidae, Hydroptilidae, Hydropsychidae, Leptoceridae

Elmidae, Psephenidae, Baetidae, Hydrobiosidae, Hydroptilidae, Hydropsychidae, Conoesucidae

Hydrobiidae, Ancylidae, Elmidae, Gripopterygidae (46%), Podonominae, Tanypodinae, Baetidae, Oligochaeta (19%) Leptophlebiidae, Gomphidae, Hydrobiosidae, Hydropsychidae, Leptoceridae

Ancylidae, Hydrophilidae, Simuliidae, Oligochaeta (38%) Podonominae, Baetidae, Coenagrionidae, Gomphidae

Hydrophilidae, Elmidae, Baetidae, Leptophlebiidae, Corixidae, Calamoceratidae, Leptoceridae

Amphipoda, Podonominae, Hydroptilidae

Missing taxa Prolonged low flows

Possible impacts

Prolonged low flows, stock, sedimentation, rural runoff

Prolonged low flows, stock, rural runoff

Rural runoff, prolonged low flows

Rural runoff, stock, leachates from fallen bridge

Prolonged low flows

Moderate catchment erosion, gravel, Rural runoff cobble, and boulder particles between 50 and 75% surrounded by fine sediment, dam present upstream

Low flow, moderate catchment erosion, slight anaerobic condition (small black undersides of rocks), \ 90% riffle covered by periphyton

Low flow, moderate catchment erosion, slight turbidity

Low flow, moderate catchment erosion

Low flow, barrier present upstream, 65–90% of riffle covered by periphyton. Moderate catchment erosion.

Low flow, barrier present upstream, 65–90% of edge covered by periphyton

Turbid, moderate catchment erosion, Urban development, prolonged barriers present upstream, obvious low flow non-point source pollution (urban development), low flow, 65–90% of edge covered by periphyton

Low flows, 35–65% riffle covered by periphyton



N.M. Davies et al.

Moderate catchment erosion Simuliidae (23%), Leptophlebiidae (23%)

The asymptote of the cumulative probability curve for categories of habitat characteristics predicted at a site was near the 2.5% probability-of-occurrence level (Fig. 3). After this level, any extra habitat categories contributed little to the list of habitat categories expected by the model because the variables had such a low probability of occurring. However, the inclusion of all habitat categories with probability of occurrence ] 2.5% allowed the prediction of poor and fair habitat categories (defined in the US EPA habitat assessment; Plafkin et al., 1989; Barbour et al., 1999) which may occur, as well as good or excellent habitat categories. To exclude many of the categories not representative of the reference condition a probability level of ] 30% was chosen, because it was near the mean for the probability of occurrence of habitat data that were outside the 10th and 90th percentiles of the reference condition. Thus, many of the poor and fair habitat features were excluded while much of the natural variation experienced within the reference condition was still represented. Comparisons between the variables that were expected to occur at a site and those that actually occurred were only made for habitat categories that had a ] 30% probability of occurrence.

Elmidae, Psephenidae, Corydalidae, Gomphidae, Hydropsychidae, Conoesucidae

The O/E ratios for the reference sites used to create the model were found to be normally distributed (mean=1.02, SD =0.136, W : Normal = 0.96, p B w = 0.167). With the exception of sites 19 and 26, all the re-sampled reference site O/E ratios for habitat were within the range of the O/E ratios for the reference data (2 ×SD; Table 8). Five of the 11 test sites had O/E ratios within the range of O/E ratios (2 ×SD) encompassed by the reference data from the model (Table 8). The O/E ratios for habitat for the remaining six test sites varied between 0.46 and 0.72 and were below the second standard deviation of the O/E ratios for the reference data from the model (Table 8), indicating that six of the test sites may have had degraded habitat conditions.

B

Model validation

Assessing impact

293

Prolonged low flows

Some catchment erosion, dam present upstream, \ 90% of edge covered by periphyton, 35–65% of edge covered by filamentous algae Oligochaeta (54%) Hydrobiidae, Ancylidae, Hydrophilidae, Elmidae, Ceratopogonidae, Simuliidae, Podonominae, Leptophlebiidae, Corixidae, Coenagrionidae, Gomphidae, Calamoceratidae 286

C

Dam influence

Low flow, \ 90% of riffle covered by Prolonged low flows, rural and periphyton, cobble and boulder particles urban (Queanbeyan) runoff between 50 and 75% surrounded by fine sediment, some deposition of sediment, 10–30% rubble, gravel or other stable habitat Orthocladiinae (33%), Chironominae (29%), Hydropsychidae (24%)

Probability of occurrence of habitat categories

Elmidae, Psephenidae, Tipulidae, Tanypodinae, Baetidae, Caenidae, Corydalidae, Gomphidae, Glossomatidae, Hydroptilidae, Philopotamidae, Conoesucidae C 283

Overall site assessment Site

Table 6 (Continued)

Missing taxa



Possible impacts

358

Generally, the habitat categories predicted at a site with a high probability of occurring were those with © 2000 Blackwell Science Ltd, Freshwater Biology, 45, 343 – 369

Prediction and assessment of local stream habitat 359 Table 7 Accuracy with which prediction using catchment features of sites agrees with groupings pre-determined by local habitat characteristics* Into group From group

1

2

3

4

Total

1

7 53.85%

2 15.38%

2 15.38%

2 15.38%

13

2

4 18.18%

14 63.64%

2 9.09%

2 9.09%

22

3

2 16.67%

3 25.00%

7 58.33%

0 0.00%

12

4

0 0.00%

0 0.00%

0 0.00%

4 100%

4

Error

46%

36%

42%

0%

Overall error: 31%

* Error is the percentage of sites incorrectly placed into the pre-determined groups using the cross-validation procedure.

Fig. 3 Habitat accretion curve showing expected number of habitat categories versus observed number of habitat categories and the probabilities of occurrence for these categories for an example site.

good and excellent scores (bottom substrate availability, embeddedness, channelization, bottom scouring, bank vegetative stability, streamside cover and overall habitat score). Other important variables were associated with the riparian zone (width and amount), channel morphology and some substratum characteristics in the reach, riffle and edge. © 2000 Blackwell Science Ltd, Freshwater Biology, 45, 343–369

The model was able to identify sites that had O/E ratios outside the range of the original reference data that had poor to fair habitat characteristics when good to excellent categories were predicted (e.g. sites 19, 193, 263, 272, 282, 283 and 293; Tables 8 and 9). This allowed specific habitat features to be identified and targeted for rehabilitation.

360

N.M. Davies et al.

The model sometimes predicted that more than one habitat category would occur at a site, although only one category could be found. For example, 10 sites (19, 26, 138, 139, 193, 263, 274, 282, 283 and 293) had both good and excellent habitat categories predicted for one or more of the following variables: bottom scouring and deposition, bank vegetative cover, pool – riffle : run –bend ratio and bank stability. If the collected data are not considered carefully, it is possible for a site to be given a lower habitat assessment (O/E ratio) if a predicted habitat category is absent even though an ecologically ‘better’ category is present. For example, habitat assessment categories such as the bottom substrate availability, bank vegetative stability and the velocity–depth categories were good or excellent at site 282 at which poor and fair categories were predicted. The effect of this on the assessment must be investigated.

Not all the predicted habitat categories have sound ecological meaning; their ‘observed’ measurement may be affected by factors other than human impact. For example, the observed depth category for edge and riffle habitats was greater than predicted at sites 263 and 272 (Table 9). Prolonged low flows followed by rain may contribute to natural fluctuations; thus, deviations in predictions and observed depth measures may not be very useful for identifying impact.

Separating water quality and habitat degradation The habitat conditions and the biological assemblages at three sites (28, 138 and 291) were within the reference condition (Tables 5 and 8). The habitats at 12 sites (3, 8, 33, 129, 130, 139, 267, 273, 274, and 286) were assessed as within the range of the reference condition, but the macroinvertebrate fauna at these

Table 8 Assessment of reference and test sites used to test the habitat model including an observed to expected (O/E) ratio* Site

Reference sites 3 8 19 26 28 33 129 130 138 139

No. of categories expected

No. of categories predicted

No. of categories collected

Observed to expected ratio

20.75 21.01 20.90 22.89 19.89 22.85 21.80 19.10 19.42 22.14

44 45 44 48 44 46 46 42 44 42

19 18 12 14 19 19 19 21 15 17

0.92 0.86 0.57 0.61 0.96 0.83 0.87 1.10 0.77 0.77

Mean O/E Test sites 193 263 267 272 273 274 282 283 286 291 293 Mean O/E

0.90 23.73 24.90 24.96 22.27 20.91 21.21 21.07 21.25 21.52 21.98 20.61

49 49 49 47 44 44 42 45 43 41 39

11 18 20 16 22 25 13 15 19 21 13

0.46 0.72 0.80 0.72 1.05 1.18 0.62 0.71 0.88 0.96 0.63 0.79

* At bold sites, the O/E ratio fell outside the range of two standard deviations from the mean of O/E ratios (\ 0.71) for the reference sites used to create the model. Predictions are based on occurrence of categories of habitat features that had ] 30% probability of occurring. © 2000 Blackwell Science Ltd, Freshwater Biology, 45, 343 – 369

Features predicted to occur that were not observed


Riparian width B 5 m, bank height \ 2 m, 10–20 cm deep riffle, 10–20% reach gravel, 40–80% riffle cobble, excellent bottom substrate, excellent pool–riffle:run–bend ratio, good bottom scouring, excellent velocity–depth, excellent bank stability, excellent habitat score

263

Riparian width B 5 m, 20–40 cm deep edge, 0–10% cobble in edge, \ 40% edge habitat, 5–10% reach gravel, 20–40% reach cobble, excellent embeddedness, excellent channel alteration, good streamside cover categories

Test sites 193 B 20% trees greater than 10 m, B 20% of trees less than 10 m, 10–20 cm deep riffle, 0–20 cm edge depth, 40–80% cobble in riffle, excellent bottom substrate availability, good bottom scouring, excellent embeddedness, excellent channel alteration, excellent velocity–depth, excellent pool–riffle:run–bend ratio, excellent bank stability, excellent streamside cover, excellent habitat score

26

Reference sites 19 Riparian width B 5 m, B 20% of trees greater than 10 m, B 20% of trees less than 10 m, stream width between 0 and 5 m, edge depth of 20–40 cm, 5–10% gravel, excellent bottom substrate, excellent embeddedness, good bottom scouring, excellent pool–riffle:run–bend ratio, good habitat score

Site

6 m riparian width, \ 40 cm deep edge, 25% edge habitat, 15% cobble in edge, 15% reach gravel, 20% reach cobble, 20% cobble in riffle, 55% pebble in riffle, good embeddedness, good channel alteration, excellent streamside cover

No riparian vegetation except grasses, no riffle, fair bottom substrate availability, poor embeddedness, fair scouring and deposition, poor channel alteration, poor velocity–depth, poor pool–riffle:run–bend ratio, fair stream side cover and good bank stability categories

0–5 m bank height, 25% gravel in reach, 20% cobble in riffle. Poor embeddedness category. Where not excellent then habitat assessment good

30% trees \ 10 m, 35% trees B 10 m. Poor bottom substrate availability, embeddedness, bottom scouring and fair channel alteration categories

Some of the observed habitat features

Little riparian vegetation (occasional tree), no riffle, only edge habitat, requires more stable in-stream habitat, fine sediment around boulder or cobble particles (infilling of interstitial spaces), not enough habitat variety, essentially straight stream, deposition of fine material in pools, requires more riparian vegetation for inputs and stability; some catchment erosion, urban development (non-point source pollution) Some variation in substrate percentages, some fines around larger particles (infilling of interstitial spaces) although still good, runs and pools dominate, riparian vegetation dominated by willows, habitat assessment categories good or excellent

River wider than expected, willows dominate riparian vegetation, river dominated by sand and gravel, little stable substrate (bed probably moves frequently), fines surrounding larger particles (infilling of interstitial spaces), river dominated by runs, little riffle, some pool, little good edge habitat, bare banks River dominated by fine particles, riffle dominated by pebble and gravel, less stable bottom substrate, willows and grasses dominate riparian vegetation, lots of runs and pools, edge habitat scattered, fine particles surrounding larger particles (infilling of interstitial spaces)

Site description in terms of degradation

Table 9 Summary information for sites assessed as having habitat conditions outside the range of the reference condition: a list of the variables predicted to occur that were not observed at the site; some habitat features observed at the sites; site description in terms of degradation*


20–40% trees greater than 10 m, \ 80% trees less than 10 m, 5–10% reach clay, \ 10% clay in reach, 6–10% silt in reach, 40–80 m riparian vegetation, poor velocity–depth, poor habitat score, poor bank stability, fair vegetative stability 21–40% trees greater than 10 m, \ 80% trees less than 10 m, 40–80 m riparian width, 0–10% reach clay, 5–10% reach silt, \ 20% edge pebble, poor velocity–depth, excellent channel alteration, excellent pool–riffle:run–bend ratio, poor–good bank stability, fair–excellent vegetative stability, poor–good habitat score \ 80% trees less than 10 m, 40–80 m riparian width, \ 1 m s−1 riffle flow, \ 20% boulder in edge, excellent pool–riffle:run–bend ratio, poor bank stability, excellent vegetative stability, excellent streamside cover, poor–good habitat score

282

Site description in terms of degradation

Riparian vegetation dominated by willows and blackberry, runs dominate river, channel altered in high flows, scouring and deposition of fines and gravel, little stable habitat, fines surrounding larger particles (interstitial spaces)

80% trees less than 10 m, 6 m riparian zone, no boulder in Trees very close to expected percentages, some edge, fair bank stability, good bank vegetative stability erosion and scouring of banks, riparian vegetation dominated by casuarinas and blackberry, lots of riffle and run, little edge habitat, bare areas on banks, unstable banks, surrounding grazing, large flows cause erosion of banks, scouring and deposition of pebble and cobble

20% trees greater than 10 m, 60% trees less than 10 m, 3.5 m riparian zone, no reach clay, no reach silt, 20% edge pebble, good velocity–depth, fair channel alteration, fair pool–riffle:run–bend ratio, good bank stability, good vegetative stability categories

Trees dominated by poplars, dominated by runs and pools, riffle area dominated by bedrock, amount of stable bottom habitat could be improved (logs, cobble or boulder), scouring and deposition occurring (habitat changing), more bank vegetation required to stabilize bare areas, some catchment erosion, grazing surrounding. Generally not as excellent as expected but still ‘good’; habitat probably adequate 5% trees greater than 10 m, 30% trees less than 10 m, 4 m Less riparian vegetation than expected, riparian width, no clay in reach, 5% silt in reach, good substrate characteristics vary from predicted, velocity–depth, fair bank stability, good vegetative stability velocity–depth and vegetative stability better than expected

\ 20 cm deep riffle, 10% cobble in riffle, 65% bedrock, good bottom substrate availability, fair embeddedness, good velocity–depth, fair channel alteration, good bottom scouring, good pool–riffle:run–bend ratio, bank stability, vegetative stability and streamside cover

Some of the observed habitat features

* Habitat categories: excellent, good, fair and poor originate from the US EPA rapid bioassessment protocols for habitat assessment (Plafkin et al., 1989; Barbour et al., 1999). Excellent equates to a lot of the category with the exception of embeddedness and channel alteration in which excellent scores are received at sites with little embeddedness or channel alteration.

293

283

10–20 cm deep riffle, 40–80% riffle cobble, excellent bottom substrate availability, excellent embeddedness, excellent velocity–depth, excellent channel alteration, excellent bottom scouring, excellent pool–riffle:run– bend ratio, excellent bank stability, good–excellent streamside cover, excellent habitat score

Features predicted to occur that were not observed

272

Site

Table 9 (Continued)



Prediction and assessment of local stream habitat 363 sites was below reference (Tables 5 and 8). This indicated that water quality and not habitat was the cause of reduced taxon richness. High turbidity levels at four sites (28, 33, 193 and 272; Table 4) may have contributed to reduced taxon richness. No other water quality indicators were evident. Prolonged low flows and increased periphyton cover (Table 6) may also have influenced the macroinvertebrate assemblages at all sites in the study area. Both the habitat condition and the biological condition at eight sites (19, 26, 193, 263, 272, 282, 283 and 293) were assessed as below reference (Tables 5 and 8) indicating that both poor water quality and degraded habitat may be contributing to biotic impairment at these sites. The combination of the effects of the drought and damaged habitat, such as sparseness of riparian vegetation, poor embeddedness, little riffle habitat, much scouring and deposition and fair bottom substrate availability (e.g. site 193; Table 9) may be responsible for impaired biotic condition. Other sites had altered channels, only fair pool – riffle : run – bend ratios (283) fair bank stability (sites 282 and 293; Table 9) that may have contributed to the biotic impairment at these sites.

Discussion Development of a model to predict local habitat features Few studies have tested the theory that large-scale features of the catchment control local stream habitats, with the exception of Richards et al. (1996), Raven et al. (1997) and Jeffers (1998) who found that catchment-scale features such as catchment area and geological variables influenced reach-scale habitat variables. In this study, large-scale variables were able to discriminate between site groups formed according to their habitat features (Table 8). This quantifies the views of Schumm & Lichty (1965), Morisawa (1968), Schumm (1977), Lotspeich (1980), Swanson (1980), Frissell et al. (1986), Schumm (1991) and Naiman et al. (1992) that the catchment is an hierarchically organized system in which large-scale catchment features influence components of the stream such as the local stream habitat. On the basis of large-scale catchment variables (catchment area, total stream length, catchment relief, alkalinity, % volcanics, % metasediments, dominant geology and dominant soils), 69% of sites could be correctly © 2000 Blackwell Science Ltd, Freshwater Biology, 45, 343–369

placed into the appropriate reference classification groups, using discriminant function analysis (Table 7). This compares with a mean misclassification error of about 30% for the completed AUSRIVAS models (Simpson & Norris, 2000) and is lower than the mean misclassification error of 44% produced by the RIVPACS II models (Wright et al., 1991) in which sites grouped from their biotic composition are matched with environmental data. The error rates are used by the RIVPACS (Wright, 1995) and AUSRIVAS (Simpson & Norris, 2000) predictive models as a guide only and the value is not critical for prediction. The calculation of the occurrence of a habitat category at a new site incorporates the probability of the new site belonging to each of the reference site groups (Simpson & Norris, 2000). This effectively removes the artificially imposed group boundaries and reverts the prediction back to a continuum of habitat change. Therefore, it is possible for a site to be misclassified and still have adequate habitat predictions. The ability to correctly place reference sites into the appropriate reference site groups using large-scale features enables each new test site to be matched with the appropriate reference group. The low classification error indicates that the hierarchical framework suggested by Frissell et al. (1986) and other authors (Schumm & Lichty, 1965; Schumm, 1977; Lotspeich, 1980; Frissell et al., 1986; Naiman et al., 1992) is an appropriate framework for studying relationships between components of the catchment ecosystem. Therefore, this approach, based on theory (e.g. Schumm & Lichty, 1965; Schumm, 1977; Lotspeich, 1980; Frissell et al., 1986; Naiman et al., 1992) and good empirical relationships, produced a powerful technique for predicting local stream habitat features from large-scale catchment characteristics. The potential exists to apply this technique to other regions. The catchment features important in explaining the variance in variables from the River Habitat Survey (Raven et al., 1997, 1998) in the UK were all mapbased variables (Jeffers, 1998). Similarly, in this study, all the predictor variables used in the model were derived from maps, with the exception of alkalinity. Map-derived predictor variables are not affected by human activities and only change over long timescales, making them ideal for predicting the habitat features that should occur at a site in the absence of human impact. The use of map-based predictor variables may allow agencies with their own large data

364

N.M. Davies et al.

sets of habitat measurements to build their own predictive habitat assessment relevant to their area. The predictor variables can be measured for each site from a map without having to visit each site, thus reducing the costs of sampling and model development. Therefore, the use of map-based predictor variables described here has the potential to provide a widely applicable method of habitat assessment.

Model validation Validation of the predictive ability of a model can be achieved by sampling reference sites other than those used in the creation of the model, to check whether the model predicts them as reference sites (Wright, 1995; Simpson & Norris, 2000). Reference sites used by the model to test the prediction of local habitat condition were sampled independently of the original reference sites that were used to build the model (i.e. they were re-sampled 3– 4 years later). However, some of the large-scale predictor variables were measured from maps and did not change from the original data. Thus, the reference sites may not have been completely independent of the original data. Therefore, further validation of the model may be required, using known reference sites not used in the creation of the model. Nonetheless, the predictions for the reference sites were similar to the conditions observed at these sites, demonstrating that the model is capable of producing appropriate assessments of local stream habitat condition. The local habitat features of nearly all the re-sampled reference sites used for validation fell within the reference condition (Table 8). Examination of the habitat features of the reference sites not assessed as reference suggested that these sites may have been suffering from habitat degradation. For example, site 19 had less available substrate, more fines surrounding larger substrate particles, scouring and deposition and more channel alteration than predicted by the model (Table 9), indicating degraded habitat relative to expectations. Previous AUSRIVAS sampling indicated unimpaired biota at these same sites. The habitat assessment suggests that some sites can no longer be considered reference and should be reviewed. Thus, the habitat assessment proposed here might be valuable for reviewing reference sites used within biological assessment approaches that rely on refer-

ence sites (e.g. RIVPACS: Wright, 1995; AUSRIVAS: Simpson & Norris, 2000).

Probability of occurrence of local habitat features Predicted habitat features are allocated a probability of occurring at a site given the catchment characteristics of that site, in a similar manner to the RIVPACS (Wright, 1995) and AUSRIVAS (Simpson & Norris, 2000) models. Those models predict the presence or absence of discrete taxa. However, the categories predicted by the habitat model developed here are not exclusive. For example, predictions of bottom substrate consisting of 50% cobble will include 20% cobble. The inclusion of habitat categories with a probability of occurrence only ] 30% minimized this problem. Even so, two or more categories of some local habitat features were still predicted to occur at some sites. This has the potential to decrease the O/E habitat ratio and future work should aim to test this potential difficulty to ensure that site assessments are accurate. Ordination techniques were used as an alternative approach for predicting the probability of occurrence of some habitat variables from the River Habitat Survey (Jeffers, 1998). The adoption of ordination techniques capable of predicting continuous data may overcome any problems associated with categorization. However, in this study, the average O/E habitat ratio for reference sites (0.90) used to test the model was within the range of the reference condition ( \ 0.72) indicating that the prediction of more than one habitat category had little effect on the model outputs.

Assessing impact The habitat model developed here provides its assessment based on the deviation between observed and expected habitat features at a site. This deviation allows features of the habitat, such as bank erosion and vegetative cover, to be identified for rehabilitation, which is an important output of effective stream habitat assessment. For example, the habitat features predicted to occur at Site 193 included stable banks, little channel alteration, habitat variety (lots of available substrate and habitat, including riffles and pools) and good levels of scouring and deposition (Table 9). The habitat features actually observed at this site did not coincide with those predicted (Table 9): the site © 2000 Blackwell Science Ltd, Freshwater Biology, 45, 343 – 369

Prediction and assessment of local stream habitat 365 had unstable channels, poor levels of scouring and deposition, embeddedness, no riffle, and bank vegetative stability (Table 9). The probabilities of occurrence of some habitat features measured in the River Habitat Survey (Raven et al., 1997, 1998) were predicted using a few large-scale catchment features (Jeffers, 1998). However, Jeffers did not provide assessments of the sites and did not emphasize the use of these predictions in terms of habitat assessment. The data from the River Habitat Survey have been used in other studies to assess stream habitats; habitat features at a site have been compared with conditions for top quality sites, described by an expert panel, to produce a Habitat Quality Assessment (Raven et al., 1998). While this may be of relevance it does not provide an assessment weighted by the likelihood that habitat features should actually occur at a site. Despite an extensive literature review, no other methods have been found that are capable of site-specific habitat assessment based on the comparison between predicted and observed measures of the habitat. This demonstrates the importance of the habitat model developed in the current study. The inclusion of habitat variables affected by natural fluctuations such as are caused by drought and floods may affect the predictive abilities of the model when predicted features of the habitat are not observed. Depths of riffle and edge habitats will be subject to natural changes and are likely to differ between sampling periods. For example, a natural event such as heavy rain following drought may have been responsible for the higher than predicted water levels at sites 263 and 272 (Table 9). The absence of the observed measurement in the exact habitat category may contribute to lower O/E ratios, suggesting that sites may be more impacted than they really are. Therefore, it may be necessary to remove any habitat categories that are subject to natural fluctuations and hence might affect the ability of the habitat model to provide meaningful O/E ratios and predictions. Not all differences between observed and expected categories described damage to local habitat. For example, site 19 had more trees and shrubs than predicted by the model and may indicate that this component of the stream habitat is not as damaged as suggested by the model (Table 9). Thus, it is imperative that the predicted and observed habitat features are compared to determine if the difference between © 2000 Blackwell Science Ltd, Freshwater Biology, 45, 343–369

the two measures provided by the model really represents damage to the stream habitat.

Separating water quality and habitat effects on biological condition Macroinvertebrate communities are influenced by a variety of impacts to water quality, ranging from point source pollution, e.g. sewage (Chessman, 1995; Boulton & Brock, 1999), and non-point source pollution, e.g. nutrient run-off (Chessman, 1995; Boulton & Brock, 1999), through to drought (Extence, 1981; Chessman & Robinson, 1987). Sites with unimpacted habitat conditions should be supportive of a healthy macroinvertebrate population unless affected by poor water quality. Comparison of the outputs from this study’s habitat model and the AUSRIVAS predictive model identified sites with unimpacted habitat conditions and impaired biological condition. For example, the habitat at site 273 (O/E= 1.05) was equivalent to reference (Table 8) while the AUSRIVAS assessment of this site indicated it was biologically impaired (Table 5). Impaired biological condition may have resulted from poor water quality associated with drought, indicated by low flows (Table 3) and a community composition that was numerically dominated by Chironominae (Table 6; Extence, 1981). In this case, poor water quality was identified as the likely cause of biological impairment. Thus, the use of a habitat model such as the one developed here, in conjunction with biological assessment via a system such as AUSRIVAS, can assist in the separation of the effects of poor water quality and habitat on biological condition. In the absence of poor water quality, degraded stream habitat conditions can influence the biological community at a site (Plafkin et al., 1989), possibly impairing the macroinvertebrate fauna there. For example, some sites had much higher levels of scouring, deposition, channel alteration (site 193) and embedded substrate (sites 193, 272) than predicted by the model (Table 9). The habitat features and high turbidity levels at these sites (Table 4) indicate that much fine material was being transported. Fine sediment can inhibit feeding, respiration and movement, can fill substrate interstices and is unfavourable to many taxa (Swanson, 1980; Lemly, 1982; Minshall, 1984; Hogg & Norris, 1991), and may have prevented the occurrence of some members of the Elmidae,

366

N.M. Davies et al.

Baetidae, Leptophlebiidae and Leptoceridae families at these sites (Table 6). Therefore, habitat degradation may have been responsible for impaired biological condition at some sites. These findings demonstrate that by comparing habitat condition with the biological assessment, the likely causes of biological impairment can be identified. If the local stream habitat is responsible for biotic impairment at a site, specific habitat features may be targeted for rehabilitation.

Conclusions This study developed an empirical model capable of predicting local stream habitat features using largescale catchment characteristics, providing evidence that independent controls such as climate and geology strongly influence lower levels in a catchment hierarchy down to the level of local stream habitat. Based on a subset of large-scale catchment features, a new test site could be adequately classified into its appropriate reference group. This enabled determination of the features of a local stream habitat that should occur at a site in the absence of the effects of human activities. Expected habitat features were compared to those observed, the deviation between the two measures provided a measure of magnitude of habitat degradation. The habitat features expected but not observed at a site may indicate appropriate targets for rehabilitation. Thus, the habitat model provides a powerful tool for assessing local stream habitats. By identifying components of the local stream habitat in need of rehabilitation, the habitat model developed here can be used in conjunction with biological assessment to identify the most likely cause of biological impairment. Thus, the habitat model developed here provides a rapid and powerful technique for assessing local stream habitats, and for identifying specific habitat features as targets for rehabilitation. Therefore, it has much potential for assisting stream management.

Acknowledgments We acknowledge and thank the many people who contributed time, labour and data to this study, including Peter Liston, Justen Simpson, Wayne Robinson, Greg Keen, Ian Lawrence, Melissa Parsons, Julie Coysh and Troy Bray. Philip Sloane, Daniel

Mawer and Kate Roberts assisted in the collection of field data. Thanks also to Sue Nichols, Sue Cunningham, Philip Sloane and Daniel Mawer for their help in the identification of macroinvertebrates and with general queries and the CRC for Freshwater Ecology for support and provision of resources.

References Allan J.D. & Johnson L.B. (1997) Catchment-scale analysis of aquatic ecosystems. Freshwater Biology, 37, 107 – 111. APHA (1992) Standard Methods for the Examination of Water and Wastewater, 17th edn. American Public Health Association, Washington, DC. Barbour M.T. (1991) Stream surveys: the importance of the relation between habitat quality and biological condition. In: Sediment and Stream Water Quality in A Changing Environment: Trends and Explanation.. Proceedings of the Vienna Symposium, August 1991, pp. 161 – 169. IAHS Publication No. 203. Barbour M.T., Gerritsen J., Snyder B.D. & Stribling J.B. (1999) Revision to Rapid Bioassessment Protocols for Use in Streams and Rivers: Periphyton, Benthic Macroinvertebrates and Fish. US Environmental Protection Agency Assessment and Watershed Protection Division, Washington, DC. EPA 841-D-97-002 (http: www.epa.gov/OWOW/monitoring/AWPD/RBP/ bioasses.html). Belbin L. (1993) PATN — Pattern Analysis Package. CSIRO Division of Wildlife and Rangelands Research, Canberra. Belbin L. & McDonald C. (1993) Comparing three classification strategies for use in ecology. Journal of Vegetation Science, 4, 341– 348. Binns N.A. & Eiserman F.M. (1979) Quantification of fluvial trout habitat. Transactions of the American Fisheries Society, 108, 215 – 228. Boulton A.J.& & Brock M.A. (Eds) (1999) Australian Freshwater Ecology Processes and Management. Gleneagles Publishing, 300 pp, Glen Osmond, Australia. Bovee K.D. (1996) Perspectives on two-dimensional river habitat models: the PHABSIM experience. In: Proceedings of the 2nd International Symposium on Habitats and Hydraulics, Quebec, Canada (eds M. Leclerc, H. Capra, S. Valentin, A. Boudreault & I. Cote), vol. B, pp. 149 – 162. Chessman B.C. (1995) Rapid assessment of rivers using macroinvertebrates: a procedure based on © 2000 Blackwell Science Ltd, Freshwater Biology, 45, 343 – 369

Prediction and assessment of local stream habitat 367 habitat-specific sampling, family level identification and a biotic index. Australian Journal of Ecology, 20, 122 –129. Chessman B.C. & Robinson D.P. (1987) Some effects of the 1982– 1983 drought on water quality and macroinvertebrate fauna in the lower La Trobe River. Victoria. Australian Journal of Marine and Freshwater Research, 38, 289–299. Davies P.E. (Ed.) (1994) Monitoring River Health Initiative. River Bioassessment Manual. National River Processes and Management Program. Freshwater Systems, Tasmania. Land and Water Resources Research Development Corporation, Department of the Environment, Sport and Territories and Commonwealth Environment Protection Agency, Canberra. Downes B.J., Lake P.S. & Schreiber E.S.G. (1995) Habitat structure and invertebrate assemblages on stream stones: a multivariate view from the riffle. Australian Journal of Ecology, 20, 502– 514. Extence C.A. (1981) The effect of drought on benthic invertebrate communities in a lowland river. Hydrobiologia, 83, 217 –224. Faith D.P., Minchin P.R. & Belbin L. (1987) Compositional dissimilarity as a robust measure of ecological distance. Vegetation, 69, 57– 68. Frissell C.A., Liss W.J, Warren C.E. & Hurley M.D. (1986) Hierarchical framework for stream habitat classification: viewing streams in a watershed context. Environmental Management, 10, 199 – 214. Gordon N.D., McMahon T.A. & Finlayson B.L. (1992) Stream Hydrology: An Introduction for Ecologists. John Wiley and Sons, Brisbane. Gregory K.J. & Walling D.E. (1973) Drainage Basin Form and Process. Edward Arnold, London. Harriman R.J. & Clifford E.S. (1987) Atlas of New South Wales. Central Mapping Authority, Department of Lands, New South Wales. Hawking J.H. (1994) A Preliminary Guide to Keys and Zoological Information to Identify Invertebrates from Australian Fresh Waters. Cooperative Research Centre for Freshwater Ecology Identification Guide No. 2. Cooperative Research Centre for Freshwater Ecology, Albury, NSW. Hogg I.D. & Norris R.H. (1991) Effects of runoff from land clearing and urban development on the distribution and abundance of macroinvertebrates in pool areas of a river. Australian Journal of Marine and Freshwater Resources, 41, 507–518. © 2000 Blackwell Science Ltd, Freshwater Biology, 45, 343–369

Horton R.E. (1932) Drainage basin characteristics. Transactions of the American Geophysical Union, 13, 350– 361. Horton R.E. (1945) Erosional development of streams and their drainage basins: hydrophysical approach to quantitative morphology. Bulletin of the Geological Society of America, 56, 275 – 370. Hynes H.B.N. (1970) The Ecology of Running Waters. University of Toronto Press, Toronto. Hynes J.D. (1975) Annual cycles of macroinvertebrates of a river in southern Ghana. Freshwater Biology, 5, 71 – 83. Jeffers J.N.R. (1998) Characterization of river habitats and prediction of habitat features using ordination techniques. Aquatic Conservation: Marine and Freshwater Ecosystems, 8, 529 – 540. Karr J.R. (1981) Assessment of biotic integrity using fish communities. Fisheries, 6, 21 – 26. Lemly A.D. (1982) Modification of benthic insect communities in polluted streams: combined effects of sedimentation and nutrient enrichment. Hydrobiologia, 87, 229 –245. Lotspeich F.B. (1980) Watersheds as the basic ecosystem: this conceptual framework provides a basis for a natural classification system. Water Resources Bulletin, 16, 581 – 586. Maddock I. (1999) The importance of physical habitat assessment for evaluating river health. Freshwater Biology, 41, 373 – 391. Marchant R. (1989) A subsampler for samples of benthic invertebrates. Bulletin of the Australian Society for Limnology, 12, 49– 52. Marchant R., Hirst A., Norris R.H., Butcher R., Metzeling L. & Tiller D. (1997) Classification and prediction of macroinvertebrate assemblages from running waters in Victoria, Australia. Journal of the North American Benthological Society, 16, 664 –681. Maher W., Liston P., Wade A., Moore J., Cullen P. & Norris R. (1994) Review of ACT Water Quality and Ecological Guidelines. Water Research Centre, University of Canberra. Mason Jr W.T. & Yevich P.P. (1967) The use of phloxine b and rose bengal stains to facilitate sorting benthic samples. Transactions of the American Microscopic Society, 56, 221 – 223. Minshall G.W. (1984) Aquatic insect–substratum relationships. In: The Ecology of Aquatic Insects (eds V.H. Resh & D.M. Rosenberg), pp. 358 – 400. Praeger Publishers,pp, New York.

368

N.M. Davies et al.

Morisawa M. (1968) Streams: Their Dynamics and Morphology. McGraw-Hill Book Company, New York. Muhar S. & Jungwirth M. (1998) Habitat integrity of running waters —assessment criteria and their biological relevance. Hydrobiologia, 386, 195 –202. Naiman R.J., Lonzarich D.G., Beechie T.J. & Ralph S.C. (1992) General principles of classification and the assessment of conservation potential in rivers. In: River Conservation and Management (eds P.J. Boon & G.E. Petts), pp. 93–123. Wiley Publishing, Brisbane. Norris R.H. & Thoms M.C. (1999) What is river health? Freshwater Biology, 41, 197– 209. Ohio EPA (1987) Biological Criteria for the Protection of Aquatic Life. The Role of Biological Data in Water Quality Assessment, vol. 1. Division of Water Quality Monitoring and Assessment, Surface Water Section, Ohio Environment Protection Agency, Columbus, OH. Parsons M. & Norris R.H. (1996) The effect of habitatspecific sampling on biological assessment of water quality using a predictive model. Freshwater Biology, 36, 419 –434. Plafkin J.L., Barbour M.T., Porter K.D., Gross S.K. & Hughes R.M. (1989) Rapid Bioassessment Protocols for Use in Streams and Rivers: Benthic Macroinvertebrates & Fish. US Environmental Protection Agency Assessment and Watershed Protection Division, Washington, DC. EPA/440/4-89/001. Raven P.J., Fox P., Everard M., Holmes N.T.H. & Dawson F.H. (1997) River habitat survey: a new system for classifying rivers according to their habitat quality. In: Freshwater Quality: Defining the Indefinable? (eds P.J. Boon & D.L. Howell). Scottish Natural Heritage, Edinburgh. Raven P.J., Holmes N.T.H., Dawson F.H. & Everard M. (1998) Quality assessment using river habitat survey data. Aquatic Conservation: Marine and Freshwater Ecosystems, 8, 477–499. Reynoldson T.B., Norris R.H., Resh V.H., Day K.E. & Rosenberg D.M. (1997) The reference condition: a comparison of multimetric and multivariate approaches to assess water-quality impairment using benthic macroinvertebrates. Journal of the North American Benthological Society, 16, 833 –852. Richards C., Johnson L.B. & Host G.E. (1996) Landscape-scale influences on stream habitats and biota. Canadian Journal of Fisheries and Aquatic Sciences, 53(Suppl. 1), 295– 311.

Richards C., Haro R.J., Johnson L.B. & Host G.E. (1997) Catchment and reach-scale properties as indicators of macroinvertebrate species traits. Freshwater Biology, 37, 219 – 230. Roth N.E., Allan J.D. & Erickson D.L. (1996) Landscape influences on stream biotic integrity assessed at multiple spatial scales. Landscape Ecology, 11, 141 – 156. Schumm S.A. (1956) Evolution of drainage systems and slopes in Badlands at Perth Amboy, New Jersey. Geological Society of America Bulletin, 67, 597– 646. Schumm S.A. (1977) The Fluvial System. Wiley-Interscience Publication, Sydney. Schumm S.A. (1991) To Interpret the Earth, Ten Ways To Be Wrong. Cambridge University Press, Sydney. Schumm S.A. & Lichty R.W. (1965) Time, space and causality in geomorphology. American Journal of Science, 263, 110 – 119. Simpson J. & Norris R.H. (2000) Biological assessment of water quality: development of AUSRIVAS models and outputs. In: Assessing the Biological Quality of Freshwaters: RIVPACS and Similar Techniques (eds J.F. Wright, D.W. Sutcliffe & M.T. Furse), pp. 125– 142. Freshwater Biological Association and Environment Agency, U.K. Southwood T.R.E. (1977) Habitat, the templet for ecological strategies? Journal of Animal Ecology, 46, 337– 365. Southwood T.R.E. (1988) Tactics, strategies and templets. Oikos, 52, 3 – 18. Strahler A.N. (1952) Hypsometric (area– altitude) analysis of erosional topography. Bulletin of the Geological Society of America, 63, 1117 – 1147. Swanson F.J. (1980) Geomorphology and ecosystems. In: Forests: Fresh Perspectives from Ecosystem Analysis. Proceedings of the 40th Annual Biology Colloquium (ed. R.H. Waring). Oregon State University Press, Oregon. Sweeney B.W. (1984) Factors influencing life-history patterns of aquatic insects. In: The Ecology of Aquatic Insects (eds V.H. Resh & D.M. Rosenberg), pp. 56 – 100. Praeger Publishers, New York. Townsend C.R. & Hildrew A.G. (1994) Species traits in relation to a habitat templet for river systems. Freshwater Biology, 31, 265 – 275. Van Haveren B.P. (1986) Water Resources Measurements: A Handbook for Hydrologists and Engineers. American Water Works Association, Denver, CO. © 2000 Blackwell Science Ltd, Freshwater Biology, 45, 343 – 369

Prediction and assessment of local stream habitat 369 Wright J.F. (1995) Development and use of a system for predicting the macroinvertebrate fauna in flowing waters. Australian Journal of Ecology, 20, 181– 197. Wright J.F., Furse M.T. & Armitage P.D. (1993) RIVPACS — a technique for evaluating the biological quality of rivers in the UK. European Water Pollution Control, 3, 15– 25. Wright J.F., Furse M.T., Clarke R.T. & Moss D. (1991) Testing and further development of RIVPACS; interim


report R and D project 243. National Rivers Authority, Bristol. Wright J.F., Moss D., Armitage P.D. & Furse M.T. (1984) A preliminary classification of running water sites in Great Britain based on macroinvertebrate species and the prediction of community type using environmental data. Freshwater Biology, 14, 221 – 256. (Manuscript accepted 15 February 2000 )