A preliminary knowledge-driven prediction model of ... - Springer Link

ARTICLES

Chinese Science Bulletin © 2008

SCIENCE IN CHINA PRESS

Springer

A preliminary knowledge-driven prediction model of snail distribution in the Poyang Lake region ZHAO An1,2,3 & BAO ShuMing1 1

The Key Lab of Poyang Lake Ecological Environment and Resource Development, Jiangxi Normal University, Nanchang 330022, China; 2 Institute of Geographic Sciences and Resources Research, Chinese Academy of Sciences, Beijing 100101, China; 3 School of Geography and Environmental Sciences, Jiangxi Normal University, Nanchang 330022, China

Present monitoring and prediction of schistosomiasis’s intermediate parasite, snail, are based on remote sensing image’s spectral signatures, and the calculation result is in fact an incomplete-constraints solution. TM image of the Poyang Lake region on October 31, 2005 was combined with GIS thematic data (DEM, boundary of the Poyang Lake, vegetation, soil and land use) to make a prediction on snail spatial distribution in the region by remote sensing, geo-informatics and knowledge-driven modeling according to mechanism of snail occurrence. Result shows that with change of overall fuzzy membership of snail occurrence from high to low, snail occurrence of the snail samples of validation group goes up to 81% within 10% high fuzzy membership range, denoting high efficiency of the model in predicting snail occurrence.

Generically speaking, present approaches of application of remote sensing to monitoring snail habitat usually employ transformed image-indices (prior principal components from principal component analysis, normalized difference vegetation indices-NDVI, tasseled green vegetation index-GVI, brightness indices-BI, land surface temperature-LST, etc.) as input bands to make unsupervised classification, and are then combined with in-situ surveyed snail sample data to arrive at snail ― habitat monitoring[1 5]. Some make statistic correlative point-blank between these indices and on-ground surveyed snail data [6 ] . These methods go through multi-node information transmission, that is “image information→environmental factor information→snail habitat information”, the impairment of attribute information and distortion of spatial information are unavoidable, and the uncertainty therein is very complicated. In fact, a snail is less than 1 cm in size, and perceptibly can not be detected directly from RS images. The spectral difference produced from the spatial difwww.scichina.com | csb.scichina.com | www.springerlink.com

ference of snail density can not overweigh that from environmental noise in snail habitat, in other words, the snail habitat can not form an obvious separate spectral cluster in RS images. Presence of vegetation is a necessary condition for snail habitat, but not a sufficiency condition, and occurrence of snails has something to do with local soil, altitude, hydrological situation, etc. Auxiliary GIS thematic data must be added to realize sensefully monitoring or predicting snail habitat. This study combines RS image with GIS thematic data to predict snail occurrence in the Poyang Lake region by means of geo-informatics and knowledge-driven model[7]. The calculated result is similar to that of a fuzzy classification, and cross validation with snail samples of validation group shows that most of the snail samples occur at high-value niche in the fuzzy member Received April 11, 2007; accepted September 18, 2007 doi: 10.1007/s11434-007-0522-4 † Corresponding author (email: [email protected]) Supported by the research project of Jiangxi Provincial Educational Bureau in 2007 (No. 137[2007]) and the National Natural Science Foundation of China (Grant No. 30590370)

Chinese Science Bulletin | January 2008 | vol. 53 | no. 1 | 115-123

GEOGRAPHY

knowledge-driven, snail distribution prediction, fuzzy classification, Poyang Lake region

ship range of snail occurrence, denoting we can rapidly find snails in the high fuzzy-membership pixel location of the calculation result, or high efficiency of the model in predicting snail occurrence.

1 Brief introduction of study area and data preprocessing 1.1 Introduction of study area The Poyang Lake region is characterized by huge flooding in summer, and grass marshlands in winter, and the surrounding 8 counties are schistosomiasis-epidemic regions of lacustrine marshland type in China. The average water levels in lakeside hydrological stations were 13.27―15.72 m during 1953―1984, with the highest being 21.69―22.20 m, and the lowest 5.90―12.09 m. 75% of the highest water levels occur in June and July, 78.8% of the lowest water levels occur in December and January. The floods from 5 tributaries of the lake from April to June plus those from Yangtze River from July to September prolong the flooding period in the lake region for half a year. Along with vast grass beaches and marshlands on lakeside, the Poyang Lake region is a favorable snail habitat, becoming one of the most serious schistosomiasis epidemic regions in China[8]. 1.2 Data preprocessing Snail sample data come from Second Snail Survey of the Poyang Lake Region from 2002 to 2003. The survey method is as follows: sample points are arranged with point interval of 20 m along a relatively straight line, and the lines are also aligned in parallel with line interval of 20 m. Each sample point is 0.33 m×0.33 m = 0.11 m2 in size that is controlled by an iron-wire frame (also called “Kuang” in Chinese). The geographic coordinate positions of the points are recorded by GPSs. The survey modes can be divided into “Systematic Method” and “Environmental-type Method”. Systematic Method records each snail sample orderly according to sequential number of lines and points, and Environmental-type Method records snail sample orderly according to environment types of snail habitat. With respect to accuracy assessment for the prediction result, the snail samples were divided into 2 groups according to recording serial numbers: one for modeling,and the other for assessment. Bands 2―5 of TM image on October 31, 2005 were linearly stretched and applied in an unsupervised classification, whose result was intercrossed with the snail 116

samples of modeling group. The result shows that the top 3 clusters in total 12 classified clusters contain most (67.90%) of the snail samples of modeling group, and the percentages of each cluster containing the snail samples of modeling group will be used as fuzzy memberships of snail occurrence. GIS thematic data include DEM, boundary of the Poyang Lake at average water level, wetland vegetation, soil type, land use of the Poyang Lake region. All these data come from the Institute of Geographic Sciences and Resources Research, CAS and China State Survey and Mapping Bureau. They are projected in Gauss Kruger projection of 6° belt with central meridian of 117°E, and resampled in a spatial resolution of 25 m.

2 Methods Ordinary classification in remote sensing can only produce a classification result of “truth” (all conditions are satisfied) or “false” (at least one condition cannot be satisfied). Fuzzy classification can produce a fuzzy set, and is expressed as fuzzy memberships in a continuous value range from 0 (complete non-member) to 1 (complete member). Fuzzy memberships always pinpoint a certain proposition, like “favorability to gold enrichment” or “likelihood of snail occurrence” etc. A fuzzy logic represents subjectively judged fuzzy memberships to a set, and there are many membership combination methods for multi spatial data layers of the same fuzzy set[9,10]. 2.1 Fuzzy AND This is equivalent to a Boolean AND (logical intersection) operation on classical set values of (1,0). The result is “true” only when all data layers are “true”, that means the final result is dictated by most rigorous evidence spatial factors. Fuzzy AND is expressed as μcombination = MIN ( μA , μB , μC ,...) , (1) where μA is the fuzzy membership of data layer A, μB is the fuzzy membership of data layer B, MIN means minimizing. 2.2 Fuzzy OR The Fuzzy OR is something like the Boolean OR (Logical union) in that the output fuzzy membership values are controlled by the maximum values of any input data layers, denoting the evidence factors to target object are rare and difficult to find, and any evidence factors found

ZHAO An et al. Chinese Science Bulletin | January 2008 | vol. 53 | no. 1 | 115-123

where μA and μB are the same as the meanings in formula (1), MAX means maximizing. 2.3 Fuzzy PRODUCT All fuzzy memberships of data layers are multiplied, and the combined fuzzy membership tends to be decreasive because each layer membership is less than 1. All membership values of data layers will contribute to the final result. Fuzzy PRODUCT can be expressed as

(3)

i =1

3.1 Relational knowledge base of snail habitat in the Poyang Lake region

where μi is the membership of ith layer. 2.4 COMPLEMENT to Fuzzy PRODUCT

This fuzzy membership combination method is calculated by 1 minus multiplication of 1 subtracted by the fuzzy membership of each data layer, a complementary to the Fuzzy PRODUCT. The outcome is increasive, whose internal implication is that if many geo-objects are favorable to occurrences of a target object, they will reinforce each other when they co-exist, and the result will much more strongly support the proposition. The expression is: n

μcombination = 1 − Π (1 − μi ) ,

(4)

i =1

where μi is the fuzzy membership of ith layer. 2.5 Fuzzy GAMMA

Fuzzy PRODUCT and the COMPLEMENT to Fuzzy PRODUCT are combined to make the following expression:

μcombination = Pγ S 1−γ ,

(5)

where P is the COMPLEMENT to Fuzzy PRODUCT, S is Fuzzy PRODUCT, γ is intervenient between 0 and 1. When γ = 1, the result is the COMPLEMENT to Fuzzy PRODUCT; when γ = 0, the result is Fuzzy PRODUCT.

3 Knowledge-driven prediction model of snail distribution in the Poyang Lake region Careful analysis on the result of unsupervised classification shows that the clusters 1, 4, 11 identified as quite likely snail habitats are distributed broadly in the study

To predict snail spatial distribution by knowledge-driven model, we must first establish the knowledge bases concerning snail habitat. Knowledge bases can be symbolized in a tree structure, containing definitions, rules and variants of middle nodes and final nodes[11]. An inference network is an important tool in simulating the logic thought processes of an expert. In expert systems, a series of fuzzy membership functions are regarded as “Knowledge Base”, and the inference network and combination rules are the “Inference Engine”. This fuzzy logic is an important tool in expert systems with uncertain evidence at present, and was applied to the problems of pattern recognition in geology[12]. This study aims at predicting “Occurrence of Snails”, and to accomplish this aim, it is necessary to first formulate judging rules, and. then try to use geographic spatial data to fulfill the requirement of the rules. (1) Knowledge base of altitudes and snail occurrence. Study shows[13,14] that 94.55% of the snail breeding habitat perches at altitudes 14―17 m of the Poyang Lake region, and the areas below 13 m and above 18 m do not have snails. Assume that fuzzy membership changes of snail occurrence have linear relation with area percentages of snail breeding marshland, and the fuzzy membership for occurrence of snails is 1.0 at altitude of 15.5 m, and 0 at altitudes below 13 m and above 18 m. Based upon these presumptions, we can get fuzzy membership functions of snail occurrence at different altitudes in the Poyang Lake region (Table 1). (2) Knowledge base of boundary distances of the Poyang Lake at average water level and snail occurrence.


117

GEOGRAPHY

n

μcombination = Πμi ,

area apart from the vicinity of the Poyang Lake. The areas supposed to be snail habitats in the Mulian Mountain, northeast of the Poyang Lake; the Lushan Mountain, northwest of the Poyang Lake; the hilly area in the lower reaches of the Xinjiang River, southeast of the Poyang Lake and the Meiling Mountain, west of the Poyang Lake look even larger than those of the vicinity of the Poyang Lake, apparently contradicting the actual reality. This accounts for again why the image spectral features can only provide the necessary conditions at most for prediction of snail occurrence other than the sufficiency conditions. To really achieve monitoring or predicting snail habitat, GIS thematic data must be applied into the approach towards this end.

ARTICLES

could indicate the existence of the target object. Fussy OR is expressed as μcombination = MAX ( μA , μB , μC ,...) , (2)

Table 1 Relational knowledge base between altitudes and snail occurrence in the Poyang Lake region Altitudes Snail breeding area Fuzzy membership of Fuzzy membership functions of snail occurrence at different Altitude range (m) (Wusong) (m) percentage (%) snail occurrence altitude range 13 0 0

x − 13 × 0.075841 0.5 x − 13.5 y= × (0.654649 − 0.075841) + 0.075841 14.5 29.78 0.654649 13.5―14.5 1 x − 14.5 y= × (1 − 0.654649) + 0.654649 15.5 45.49 1 14.5―15.5 1 x − 15.5 × (0.423829 − 1) + 1 y= 16.5 19.28 0.423829 15.5―16.5 1 x − 16.5 y= × (0.043966 − 0.423829) + 0.423829 17.5 2 0.043966 16.5―17.5 1 x − 17.5 y= × (0 − 0.043966) + 0.043966 18 0 0 17.5―18 0.5 x is the altitude marked by Wusong hydrological benchmark, y is fuzzy membership of snail occurrence. 13.5

3.45

0.075841

As the topographic slopes in water-land transition area are very small in the Poyang Lake region, and water-swelling process is relatively slow, the horizontal distances with very small altitude differences are quite expansive. On the other hand,the speeds at which the snails crawl are usually lower[15] therefore, at some distances off the average boundary of the Poyang Lake, the likelihood of snail occurrence will decline rapidly. Assume that the fuzzy membership of snail occurrence has linear descending relation with the distances off the average boundary of the Poyang Lake, and is 0 when the distance is greater than 15 km. Table 2 lists the fuzzy membership functions of snail occurrence at different horizontal distance ranges off the average boundary of the Poyang Lake based on the presumptions above. Table 2 Relational knowledge base between boundary distances of the Poyang Lake at average water level and snail occurrence Boundary distance ranges of the Fuzzy membership function of Poyang Lake at average water snail occurrence at various level (m) boundary distances 15000 − x y= 0―15000 15000 　15000

0

x is boundary distances of the Poyang Lake at average water level, y is fuzzy membership of snail occurrence.

(3) Knowledge base of vegetation and snail occurrence. There usually exist typical wet-feet or semi wet-feet plants in whatever environments where the snails breed. The higher the vegetation coverage and the biodiversity, the more likely the snails occur[16,17]. Considering no vegetation map of the Poyang Lake region was available, the wet-land vegetation map was taken to represent the vegetation situation in the Poyang Lake 118

13―13.5

y=

region. Based on experiential knowledge, the wet-land areas were evaluated for fuzzy memberships of snail occurrence at 0.9, and the non wet-land areas were evaluated for fuzzy membership of snail occurrence at 0.2 because we can not rule out the possibility that snail occurrence may exist in the non wet-land areas, but we do know the snails usually exist in the areas where water and land frequently alternate. (4) Knowledge base of soil type and snail occurrence. From the soil micro-environmental point of view, soluble salt, nitrite nitrogen, nitrite sulfur and phosphor in waters are obviously higher in snail breeding environment. The humus soils being of loose structure and more moist are very favorable environments for snail breeding[18]. A study[19] found the soil types of snail breeding environment in Jiangning County of Jiangsu Province of China were mainly Ochri-Aquic Cambosols and Anthrosols (refer to Chinese Soil Taxonomy, 2001). The snail densities were higher in the soils having sandy texture, light alkali and more organic substance, and were lower in the acid soils. Grounded on these existent study results, the soil types along the Poyang Lake were evaluated for the fuzzy memberships of snail occurrence (Table 3). The fuzzy memberships of snail occurrence listed in Table 3 are mainly based on the study results of the relations between soil subgroups and snail habitat. Mapping time of original soil types and the “water-land” alternation nature were considered in determination of the fuzzy memberships of snail occurrence for soil types “lakes, and reservoirs” and “rivers”, as well as the spatial relations between snail sample points and soil subgroups.


nal” and “lake”, as well as spatial relations between snail sample points and land use types. (6) Knowledge base of TM spectrum cluster and snail occurrence. The percents of snail samples calculated for individual spectrum cluster in the intercrossed table by the unsupervised classification result and the snail samples of modeling group were used as the fuzzy memberships of snail occurrence for TM image spectra (Table 5).

ARTICLES

(5) Knowledge base of land use and snail occurrence. Land use types reflect human perception and use levels to natural conditions and natural resources, which will surely provide us with informative message to locate the spots of snail occurrence. In accordance with the thematic land use map of the Poyang Lake region in 2000, the relational knowledge base between land-use types and snail occurrence is established in Table 4. The fuzzy memberships of snail occurrence listed in Table 4 are mainly based on the definitions of land use types. Seasonal changes of water levels and the “water-land” alternation nature were taken into account in formulating the fuzzy memberships of snail occurrence for land use types “reservoir and pond”, “river and ca-

3.2 Knowledge-driven prediction model of snail occurrence in the Poyang Lake region

(1) Establishment of an inference network for knowledge-driven prediction model of snail occurrence

Relational knowledge base between soil type (subgroup) and snail occurrence Fuzzy memberFuzzy memberSoil types (subgroup) in Chi- Fuzzy membership Soil types (subgroup) in ChiSoil types (subgroup) in Chinese ship of snail ocship of snail nese Soil Taxonomy, 2001 of snail occurrence nese Soil Taxonomy, 2001 Soil Taxonomy, 2001 currence occurrence Humic ali-perudic cambosols 0 Ochri-aquic cambosols 0.2 Lithic udi-orthic primosols 0

Table 3

0.3

Rivers

0.2

Dystric purpli-udic cambosols

Urban areas

0

Beaches, islands

0

Orthic primosols

0

Alluvic primosols

0.25

0

Hapli-stagnic anthrosols

0.1

Primosols

0

0

Claypani-udic argosols

0.2

Argi-udic ferrosols

0.35

Albic stagni-perudic cambosols Albic Fe-accmuli-stagnic anthrosols Gleyi-stagnic anthrosols

0.1

Typic purpli-udic cambosols

0.1

Ali-udic cambosols

0

Argic rhodic alliti-udic ferrosols

0

Cambosols

0.1

Lakes, reservoirs

0.6

Ochri-aquic cambosols

0.85

Ali-udic argosols

0.15

Ferri-udic argosols

0.2

Argosols

0

Brown carbonati-udic argosols

0

Xanthic ali-udic cambosols

0

Anthrosols

0.25

Fe-accumuli-stagnic anthrosols

0.1

Ali-perudic argosols

0

Lithic udi-orthic primosols

0

Relational knowledge base between land use types and snail occurrence Fuzzy membership Fuzzy membership Land use type Land use type Land use type of snail occurrence of snail occurrence Urban area 0 River and canal 0.3 Reservoir and pond Grass land of lower vegeta0 Lake 0.35 Plain of paddy land tion coverage Grass land of high vegetation 0 Barren area 0 Hilly forest with paddy field coverage Shrubs 0 Naked rock and gravel 0 Mountain with paddy field

0

GEOGRAPHY

Ochri-aquic Cambosols

Table 4

Dry land of steep slope

0

Rural resident area

Plain with dry land

0

Other built-up area

Hills with dry land

0

Other forest

Maintain with dry land

0

Sparse forest

0

Table 5

0

Fuzzy membership of snail occurrence 0.3 0.15 0.1 0

Beach

0.65

0

Forest land

0

0

Marshland Grass land of middle vegetation coverage

0.8 0

Relational knowledge base between TM unsupervised classification’s spectral clusters and snail occurrence

Spectrum clusters

Snail-sample percents

Spectrum clusters


Spectrum clusters


Cluster 10

0.0210

Cluster 7

0.0290

Cluster 12

0.0080

Cluster 3

0.0370

Cluster 6

0.0270

Cluster 9

0.0300

Cluster 2

0.0660

Cluster 11

0.2200

Cluster 8

0.0700

Cluster 4

0.1990

Cluster 5

0.0340

Cluster 1

0.2600


119

Figure 1

Knowledge-driven snail prediction model in the Poyang Lake region).

in the Poyang Lake region. The key to the knowledge-driven geographic modeling is to transform the cognitions and the previous study achievements of a specific geographic issue into a mathematical model that tallies with the nature of the issue itself. The prior knowledge of the problems studied and their mechanism should be understood, meanwhile people have to take account of the selection of geographic factors, organization and expression of the spatial data layers, weighting of the geographic variants, and determination of the arithmetic operators for the variants as well. A knowledge-driven prediction model of snail occurrence in the Poyang Lake region is founded herein on the knowledge bases of snail habitat established previously (Figure 1). (2) Calculation of the knowledge-driven prediction model of snail occurrence 1) Incorporation of geographic factors. “Altitudes” and “boundary distances of the Poyang Lake” convey message of vertical distances and horizontal distances to the most favorable snail habitat, and they are mutually complementary and reinforceful to each other, therefore we consider to incorporate both factors into one, called “spatial factor” by arithmetic operator COMPLEMENT to FUZZY PRODUCT. Another fact was also noticed, that is the vertical distances directly control the numbers of emerging days and submerging days of the marshlands in the Poyang Lake region, and the horizontal distances take effect only on the precondition 120

tion of existence of vertical distances. Thus the weights of the fuzzy memberships were given by 1 and 0.5 for the factors “altitudes” and “boundary distances” respectively. The incorporating formula is as follows: Wspace = 1−(1−AltitudeW)×(1−0.5×DistanceW), (6) where “Wspace” is the fuzzy membership of “spatial factor” of snail occurrence; “altitudeW ” is the fuzzy membership of “altitudes” of snail occurrence; “DistanceW ” is the fuzzy membership of “boundary distance” of snail occurrence. From geographic point of view, soil and land-use are closely related to each other, and there would be much redundant information if they were calculated in the final model in parallel. Thus the two factors were incorporated into one, called “soil&land” by Fuzzy AND operator as follows: Wsoil&land = MIN(SoilW, LanduseW), (7) where “Wsoil&land” is the fuzzy membership of “soil&land” of snail occurrence; “SoilW ” is the fuzzy membership of “soil” of snail occurrence; “LanduseW ” is the fuzzy membership of “land use” of snail occurrence; “MIN” means minimizing. 2) Calculation of the assembly model. Fuzzy Gamma operator was adopted in the final assembly model, and the overall fuzzy membership of snail occurrence is computed by expression as follows: Wtotal = [(1−(1−Wspace)×(1−Wsoilland) ×(1−WetlandW)×(1−TMw))0.85×[(Wspace


3.3 Evaluation of the calculation result of the assembly model

GEOGRAPHY

This study aims at combining a remote sensing image and other GIS thematic data layers through fuzzy membership operators in light of spatial relational knowledge and research achievements concerning snail habitat and geographic factors to construct a knowledge-driven prediction model of snail occurrence to simulate the spatial distribution of the snails in the Poyang Lake region. The calculation result is very similar to that of a fuzzy classi-

fication in remote sensing (like that of the software eCognition). With regard to validation to fuzzy classification result, the optimum should be that the fuzzy memberships of the snail samples of validation group concentrate on high-value niche of the whole fuzzy membership range. In other words, the higher the fuzzy memberships, the more likely the snails will occur. The snail samples of validation group were used to assess the efficiency of the model. The overall fuzzy memberships of snail occurrence were multiplied by −1 in order to illustrate that with change of overall fuzzy memberships of snail occurrence from high to low, how high the percentage of the snail samples of validation group is and how low the percentage of the total pixels is in the whole study area. The overall fuzzy membership map was crossed by the map of the snail samples of validation group. In the intercrossed table, the snail samples of validation group were cumulated in the

ARTICLES

×Wsoilland×WetlandW×TMw)]0.15, (8) where WetlandW is the fuzzy membership of “Wetland” of snail occurrence; TMw is the fuzzy membership of “TM image” of snail occurrence; Wtotal is the overall fuzzy membership of all contributing factors of snail occurrence. Other symbols are the same as those in expressions (6) and (7) (Figure 2).

Figure 2 Fuzzy membership map of snail occurrence by knowledge-driven model in Poyang Lake region.


121

Figure 3

Relation map between cumulated snail occurrence ratio and cumulated pixel percent from high to low fuzzy membership.

number of pixels while the corresponding fuzzy memberships of snail occurrence change from high to low. Finally, the cumulated percent of the snail samples of validation group over the total snail samples of validation and the cumulated percent of the pixels over total pixels in the whole study area with fuzzy memberships from high to low were scaled in the axes X and Y respectively. A relational curve of “cumulated snail number / total snail number of validation group-cumulated pixels / total pixels with fuzzy membership from high to low” was plotted in the coordinate system (Figure 3). Figure 3 shows that with change of overall fuzzy memberships of snail occurrence from high to low, 81% of the snail samples of validation group gather around in 10% of the top high fuzzy membership range, denoting high efficiency of the model established in predicting snail occurrence. If many snail samples of validation group had lower fuzzy memberships of snail occurrence, the corresponding cumulated pixel percent to 81% of the snail samples of validation group would have become very high, and the snail distribution would show random pattern when this figure of percent amounts to 81%, in this case, all the geographic factors considered would have nothing to do with snail occurrence. From remote sensing point of view, the calculated fuzzy memberships of snail occurrence are something equivalent to those of a fuzzy classification, each pixel does not necessarily have a discrete Yes or No attribution to a category, but has a fuzzy membership belonging to a certain specific category with a continuous range from 0 to 1, and the 122

higher the membership, the more likely the pixel will fall into the category.

4 Summary and discussion Generally speaking, snail habitat cannot form a separate spectral cluster in RS images. Snail occurrences have something to do with many geographic environmental factors like soil, altitude, hydrology, vegetation etc. This study combines a TM image with GIS thematic data (DEM, boundary of the Poyang Lake, vegetation, soil and land use) to make a prediction on snail spatial distribution in the Poyang Lake region by geo-informatics and knowledge-driven modeling. Whereas the computation result is satisfactory, there still exists a certain improvement gap of the accuracy with improvement of GIS thematic data quality and adjustment of the model in the future. Therefore, this study is at initial stage in this field, appealing to the professional experts of relevance worldwide to upgrade the approach for the future. The technical noduses and heading directions of knowledge-driven geographical modeling vary depending on the specific geographic issues. In regard to the issue of snail occurrence, the authors consider the following aspects: (1) Determination of the relational knowledge base of different GIS thematic data with snail occurrence. The future research priorities should be given to raising the quality of the related GIS thematic data and quantitative relations between thematic data


combinations and snail habitat. The data of land use, wetland vegetation and soil employed in the study come from the Institute of Geographic Sciences and Resources Research, CAS, whereas soil data are provided by the Nanjing Institute of Soil Science, CAS. Lake boundary data come from Water System of Secondary Grade in China published by China State Survey and Mapping Bureau on a Webside.

ton-Dordrecht-Lancaster: Kluwer-Nijhoff Publishing, 1985. 363

Lin T, Jiang Q W, Zhang S Q, et al. Study on classification of schistosomiasis-epidemic regions of beach type by remote sensing image,

11

ERDAS Inc. ERDAS IMAGINE Tour Guides. Atlanta, Georgia, 2003.

12

Griffiths C M. An example of the use of fuzzy-set based pattern recog-

543―578

Chin. J Prev Med (in Chinese), 2000, 34(5): 263―265 2

3

Guo J G, Lin D D, Hu G H, et al. Study on rapid determination of snail habitat in Poyang lake region by remote sending technology, Chin. J

nition approach to the problem of strata recognition from drilling re-

Epidemiol (in Chinese), 2002, 23(2): 99―101

sponse: In Handbook of geophysical Exploration, Section I: Seismic

Jiang Q W, Lin D D, Liu J X, et al. Study on classification of vegeta-

Prospecting, Vol 20. In: Aminzadeh F, ed. Pattern recognition and image Processing. London-Amsterdam: Geophysical Press, 1987. 504―538

tion in grass beaches of the Banghu lake in the Poyang lake region, Jiangxi province by satellite image, Chin. J Epidemiol (in Chinese),

13

2001, 22(2): 114―115 4

water level in the Poyng lake region, Chin. J Schisto Control (in Chinese), 2003, 15(6): 429―433

Lin D D, Zhou X N, Liu Y M, et al. Prediction of snail distribution in the Poyang lake region after 1998 floods by Satellite image, Chin. J

14

Schisto Control (in Chinese), 2002, 14(2): 119―121 5

nese), 2002, 23(2): 90―93 15

Chin. J Prev Med (in Chinese). 2003, 37(5): 31―35

2003. 123 16

Chin. J Epidemiol (in Chinese), 2003, 24(4): 257―260 7

18(2): 62―63 17

1994. (ISBN 0 08 041867 8): 267―332 8

(2): 77―83 18

nology, 1988, ISBN 7-5323-0660-7/K·2. 15―20 9

10

Qian X H, Yang J, Chen L, et al. Study on physic-chemical property in micro environment of snail habitat Anning watershed. J Cur Prev Med (in Chinese), 2000, 27(1): 15―17

An P, Moon W M, Bonham-Carter G F. On a knowledge-based approach of integrating remote sensing, geophysical and geological in-

Wu W P. Primary survey of vegetation and snail distribution in the beaches of the Poayng lake region. J Jiangxi Univ (in Chinese), 1978,

Editorial Board of Study on the Poyang Lake. Study on the Poyang Lake (in Chinese). Shanghai: Shanghai Press of Science and Tech-

Xv F S, Qian X H, Wen S, et al. Relation between vegetation, soil and snail distribution in the Anning watershed. Sichuan Animal, 1999,

Bonham-Carter G F. Geographic Information Systems for Geoscientists: Modeling with GIS, Ontario: Delta Printing Ltd. (Hardcover),

Yuan H C, Zhang S J, Jiang Q W. Principle and Practice in Schistosomiasis Control (in Chinese). Shanghai: Fudan University Press,

Zhang B, Zhang Z Y, Xu D Z, et al. Application of MODIS image to study on the snail distribution in the beaches of Jiangning County,

Lin D D, Zhang S J. Geographic environment and schistosomiasis transmission in the Poyang lake region, Chin. J Epidemiol (in Chi-

Zhang S Q, Jiang Q W, Wang T P, et al. Monitoring snail habitat in schistosomiasis-epidemic regions of beach type by remote sensing,

6

Ning A, Chen N G, Zhong J H, et al. Relation of snail distribution and

19

Zhang Z Y, Sun Z D, Xv D Z, et al. Analysis of environmental fea-

formation. Proceedings IGARSS’92. 1992. 34―38

tures of snail habitat in Jiangning county by GIS. J Dis Control (in

Zimmermann H J. Fuzzy Set Theory and Its Applications. Bos-

Chinese), 2004, 8(1): 16―18


123

GEOGRAPHY

1

ARTICLES

and snail occurrence. (2) Adoption of proper fuzzy operators to remove redundant information between thematic data, and proper arithmetic operators to combine various GIS thematic data in assembly models. The future research priorities should be directed to integrated quantitative relational studies between thematic data