Development of Tiger Habitat Suitability Model using Geospatial Tools – A case study in Achanakmar Wildlife Sanctuary (AMWLS)

7/29/2019 Development of Tiger Habitat Suitability Model using Geospatial Tools A case study in Achanakmar Wildlife Sanct

1/13

Environ Monit Assess (2009) 155:555567

DOI 10.1007/s10661-008-0455-7

Development of tiger habitat suitability model using

geospatial toolsa case study in Achankmar WildlifeSanctuary (AMWLS), Chhattisgarh India

R. Singh P. K. Joshi M. Kumar

P. P. Dash B. D. Joshi

Received: 29 October 2007 / Accepted: 26 June 2008 / Published online: 6 August 2008 Springer Science + Business Media B.V. 2008

Abstract Geospatial tools supported by ancillary

geo-database and extensive fieldwork regarding

the distribution of tiger and its prey in An-

chankmar Wildlife Sanctuary (AMWLS) were

used to build a tiger habitat suitability model.

This consists of a quantitative geographical infor-

mation system (GIS) based approach using field

parameters and spatial thematic information. The

estimates of tiger sightings, its prey sighting and

predicted distribution with the assistance of con-

textual environmental data including terrain, roadnetwork, settlement and drainage surfaces were

used to develop the model. Eight variables in the

dataset viz., forest cover type, forest cover density,

slope, aspect, altitude, and distance from road,

settlement and drainage were seen as suitable

R. SinghWildlife Institute of India, Dehradun 248 001, India

P. K. Joshi (B)TERI University, New Delhi 110 013, Indiae-mail: [email protected]

M. Kumar P. P. DashIndian Institute of Remote Sensing,Dehradun 248 001, India

B. D. JoshiGurukula Kangri University, Haridwar 249 404, India

proxies and were used as independent variables

in the analysis. Principal component analysis and

binomial multiple logistic regression were used

for statistical treatments of collected habitat pa-

rameters from field and independent variables re-

spectively. The assessment showed a strong expert

agreement between the predicted and observed

suitable areas. A combination of the generated

information and published literature was also used

while building a habitat suitability map for the

tiger. The modeling approach has taken the habi-tat preference parameters of the tiger and poten-

tial distribution of prey species into account. For

assessing the potential distribution of prey species,

independent suitability models were developed

and validated with the ground truth. It is envis-

aged that inclusion of the prey distribution proba-

bility strengthens the model when a key species is

under question. The results of the analysis indicate

that tiger occur throughout the sanctuary. The

results have been found to be an important input

as baseline information for population modelingand natural resource management in the wildlife

sanctuary. The development and application of

similar models can help in better management of

the protected areas of national interest.

Keywords Geospatial tools Habitat Model

Prey Suitability Tiger


2/13

556 Environ Monit Assess (2009) 155:555567

Introduction

Wildlife habitat planners require detailed infor-

mation pertaining to the spatial distribution and

abundance of species to understand the ecology

and develop management plan. Habitat may be

characterized by a description of the environ-mental features that are important for a species.

Such descriptions are often based on field experi-

ence and non-quantifiable human perceptions

(Burgman and Lindenmayer 1998). Such informa-

tion is also used to develop wildlife habitat models

(Pearce and Boyce 2006; Hirzel et al. 2006; Smith

et al. 2007; Braunisch et al. 2008). These mo-

dels are further used to assess habitat suitability,

identify potential risks to the species, understand

the implications of different land use practices

and to identify sites for the reintroduction of anendangered species (Stoms et al. 1992; Braunisch

et al. 2008; Drury and Candelaria 2008). One of

the widely used methods for these descriptions

is habitat suitability index (HSI) modeling. HSIs,

first developed by the United States Fish and

Wildlife Service in the early 1980s are conceptual

models based on evaluation procedure (USFWS

1980, 1996; Burgman et al. 2001). However, theseprocedures are generally linked by mathematical

functions (Fieberg and Jenkins 2005), which are

not able to compensate the individual variables in

case of presence and absence of all the variables

together (Hirzel et al. 2001; Zaniewski et al. 2002;

Marcot 2006). These could be managed by assign-

ing weights reflecting the importance of different

variables and are based on expert knowledge in

terms of combination of biology and life history

of a species and available data (Dzeroski et al.

1997, Venterink and Wassen 1997, Hackett andVamnclay 1998, Horst et al. 1998, Moltgen et al.

Fig. 1 Location of study area


3/13

Environ Monit Assess (2009) 155:555567 557

1999, Yamada et al. 2003, Drury and Candelaria

2008). The incorporation of location-specific know-

ledge using GIS enhances the wildlife habitat

suitability models (Wightmann 1995; Zhu 1999).

Along with, systematic field investigations and

data analysis techniques can always improve these

models (Mackenzie and Royle 2005).This paper explains a methodology that utilizes

GIS and extensive field survey to elicit habitat

suitability of tiger with detailed knowledge of the

study site in Achankarmar Wildlife Sanctuary

(AMWLS). First, the paper presents a brief

methodology for structured elicitation of knowl-

edge that combines both quantitative (GIS) and

qualitative information. Second, the paper ex-

plains habitat analyses. Third, this information is

synthesized and combined with other data layers

in GIS to develop the habitat suitability model fortiger. The model also takes prey distribution into

account along with the environmental variables.

The research was conducted under conditions that

are common to many management scenarios in

India where there is little documented informa-

tion on the biology of the species or its habitat is

available.

Study area

Achanakmar Wildlife Sanctuary (AMWLS) is lo-

cated 90 km west from the district Bilaspur in

Chhattisgarh state. It lies between 81.34E to

81.55E longitude and 22.24N to 22.35N lati-

tude with a geographical area of around 530 km2

(Fig. 1). The altitude of area varies between 362 to

721 m. Thirty percent of the total area is plain in

which 22 forest villages are situated. The hilly area

has good Bamboo forest. A river named Maniyaridissects the sanctuary area. It has a typical trop-

ical climate consisting mainly rainy, winter and

Fig. 2 Habitat variable collection points for prey and predator


4/13


summer seasons. The average annual rainfall is

1,322 mm. The temperature varies from 4.2C to

40.2C. The vegetation is primarily tropical moist

deciduous forest followed by dry mixed forest

in less moist areas. The predominant vegetation

types are Sal forest, mixed forest, Bamboo breaks

and Teak plantation.

Material and methods

Spatial database development

GIS data layers for AMWLS were developed

to account for the spatial context. The en-

tire database was built at 1:50,000 scale with

Lambert Conformal Conic Projection. The an-cillary database including drainage, road net-

work, settlement, contours were generated from

topographic sheets. A hydrologically corrected

25 m resolution digital elevation model (DEM)

was created using TOPOGRID surface interpo-

lation in ArcInfo. It was then used to derive

other secondary environmental variables includ-

ing slope and aspect using Spatial Analyst module

of ArcInfo (ESRI 1999). Aspect was transformed

to a linear measure (south, southeast, southwest,

north, northeast, northwest, west and east), which

is indicator of illumination.

Habitat mapping using satellite data

The IRS-P6 LISS-III standard false color com-

posite of February 2004 was used to prepare the

habitat (forest cover types) and forest canopy den-

sity maps through on-screen visual interpretation.

Three canopy density classes viz., 1040% (open),

4070% (medium dense) and >70% (dense) and

non-forest classes could be interpreted from re-

motely sensed data. Image elements like tone,

texture, shape, size, shadow, location and associ-

ation were used for this purpose. Six forest cover

types and five non forest classes were identified

while preparing the forest type maps. Forest type

and density maps were evaluated for classification

accuracy using second set of field data.

Field data collection

In addition to the datasets described above, ex-

tensive field work was carried out for collection

of information related to indicator environmental

parameters while considering presence and ab-

sence of prey (sambar, chital and wild boar) and

Fig. 3 Procedure fordevelopment of tigerhabitat suitability model


5/13


predator (tiger) species (Fig. 2). The pugmarks

and pellets/droppings were used as direct evidence

of animal presence. Random sampling strategy

was applied for collecting the information. In total

97 plots were laid representing 27 for tiger (preda-

tor), 50 for prey species and another 20 random

plots. The plots for the tiger were planned withthe random sampling method. For the prey species

line transects were taken along the river stream,

trails, tracks and road network. Transects were

planned to remove the biased-ness for any par-

ticular prey species. The 20 random plots are the

sites identified on the suggestion of local people,

porters, trackers and at time convenience while

carrying out the field work. The plots were circular

in nature with 10-m diameter. For each sampling

site altitude, canopy cover (%), canopy height,

total number of trees, number of lopped stems,cut stems, saplings, shrub cover (%), shrub height,

grass cover (%), grass height, distance form wa-

ter, road, village, number of dung/pellets were

recorded.

Statistical processing of habitat variables

The field data was statistically analyzed to un-

derstand the habitat use and distribution pattern

by the prey and predator. This included principalcomponent analysis (PCA) and binomial multiple

logistic regression (BMLR). Statistical Package

for the Social Sciences (SPSS) has been used

for the statistical analysis (SPSS 1988). PCA in-

volves a mathematical procedure that transforms

a number of (possibly) correlated variables into a

(smaller) number of uncorrelated variables called

principal components. The succeeding component

accounts for receding variability as possible. All

cases were first filtered on the basis of the sighting

of individual species and the PCA (correlationcoefficients, varimax rotation) was run on the

dataset. Variables showing very low loadings on

the rotated component matrices were successively

dropped to reduce the noise in the dataset. PCA

was carried out on the data collected from the field

only to justify the dependence on predator of the

prey distribution.

BMLR is a form of regression, which is used

when the dependent is a dichotomy and the

independents are of any type. It is used to predict

a dependent variable on the basis of independentsand to determine the percent of variance in the de-

pendent variable explained by the independents;

to rank the relative importance of independents;

to assess interaction effects; and to understand

the impact of covariate control variables. Eight

variables in the dataset viz., forest type, density,

slope, aspect, altitude, and distance from road,

settlement and drainage were used as suitable

proxies and were used as independents in the

analysis. Individual cases of animal sightings were

considered as Boolean and BMLR were run. Out-liers in the dataset were detected using standard

deviations of residuals greater than specified cut.

The coefficients thus obtained were then used for

subsequent raster analysis (distribution mapping).

Table 1 Areadistribution in differentforest type, forestcover/density and landuse

Area (km2) Area (%) Area (km2) Area (%)

Forest cover type Forest cover/density

Sal forest 143.38 27.04 Dense (>70%) 265.62 50.09

Sal mixed forest 96.19 18.14 Density (4070%) 142.17 26.81

Sal bamboo forest 49.70 9.37 Density (1040%) 16.99 3.20

Mixed forest 89.02 16.79

Bamboo mixed 27.04 5.10

Bamboo breaks 19.45 3.67

Land use

Teak plantation 2.13 0.40

Scrub 67.50 12.73

Agriculture 24.30 4.58

Water body 0.50 0.09

Riverbed 11.05 2.08

Total 530.26 100


6/13


Table 2 Habitat variables for prey, random and tiger plot

Variables Prey plot Random plot Tiger plot F Sig.

Mean SE Mean SE Mean SE

Altitude 500.44 (10.88) 522.22 (26.09) 460.54 (19.93) 1.76 0.18

Canopy cover (%) 65.0 (2.69) 64.44 (2.94) 65.45 (3.66) 0.02 0.95

Canopy height (m) 10.28 (0.55) 8.78 (0.69) 11.45 (1.42) 1.71 0.19

Distance from road (km) 1.41 (0.18) 0.92 (0.30) 1.00 (0.22) 0.23 0.79

Distance from village (km) 2.68 (0.25) 2.853 (0.43) 2.77 (0.57) 0.06 0.94

Distance from water (km) 165.93 (35.52) 231.94 (63.60) 209.54 (118.14) 0.41 0.66

Forest density (%) 67.41 (1.87) 67.78 (1.29) 70.00 (1.91) 0.22 0.80

Grass cover (%) 26.02 (2.86) 31.67 (4.59) 29.09 (8.79) 0.47 0.63

Grass height (cm) 16.81 (1.26) 21.11 (2.00) 12.18 (3.18) 3.27 0.04

Number of cut stems 3.74 (1.03) 5.72 (1.16) 2.09 (0.61) 1.11 0.33

Number of dung pellets 0.04 (0.04) 0.000 (0.00) 0.09 (0.09) 0.46 0.63

Number of lopped stems 0.03 (0.03) 0.11 (0.11) 0.00 (0.00) 0.69 0.50

Number of saplings 12.43 (0.82) 14.17 (2.69) 13.73 (2.12) 0.41 0.67

Shrub cover (%) 36.67 (2.63) 43.33 (5.89) 48.18 (8.72) 1.57 0.21

Shrub height (m) 2.78 (1.27) 3.08 (2.17) 4.66 (3.53) 0.17 0.84

Total number of trees 20.74 (2.00) 30.05 (9.26) 16.45 (1.73) 1.68 0.19

The result was logit transformed [ P = {exp(a +

BX. . . )/(1 + (exp(a + BX. . . )))}] to obtain absolute

habitat occupancy map. For obtaining the habitat

preferences the output was rescaled to a range of

1 to 10 and was subjected to an exponential trans-

formation to produce the most conservative esti-

mates possible. For the prey species only aforesaid

eight parameters were used. However while

assessing for the tiger, probability distribution of

sambar, wild boar and chital were used in addition

to aforesaid eight parameters. The inclusion of

probability distribution of the prey probability

was based on the assumption that the predator

will be more in the abundance of prey. Thus total

eleven parameters were used in case of tiger.

GIS integration

The regression coefficients obtained from the

analysis were then attributed to the respective lay-

ers. The estimated log-odds image was then logit

transformed to produce the intended probability

map. The probability maps were developed for

sambar, chital and wild boar. The transformed

Table 3 PCA of varioushabitat variables/fielddata

Variables Factor scores

PC I PC II PC III PC IV

Altitude 0.254 0.222 0.594 0.495

Canopy cover (%) 0.323 0.689 0.125 0.221

Canopy height (m) 0.440 0.494 0.299 0.197

Distance from road (km) 0.333 0.058 0.091 0.274

Distance from village (km) 0.295 0.229 0.068 0.245

Distance from water (m) 0.083 0.162 0.402 0.209

Forest density (%) 0.187 0.776 0.033 0.035

Grass cover (%) 0.154 0.523 0.217 0.497

Grass height (cm) 0.029 0.085 0.264 0.734

Number of cut stems 0.840 0.073 0.142 0.100

Number of lopped stems 0.089 0.417 0.096 0.276

Number of saplings 0.415 0.289 0.486 0.056

Shrub cover (%) 0.374 0.381 0.131 0.167

Shrub height (m) 0.857 0.081 0.242 0.021

Total number of trees 0.745 0.029 0.387 0.148


7/13


Fig. 4 Distribution of prey and predator (tiger) species

output was then equal-interval sliced into leastsuitable, moderately suitable, suitable and highly

suitable categories. The probability maps devel-

oped for the prey species, independents and the

coefficient derived from the BLMR were used to

develop a habitat suitability model for the tiger.

The procedure for development of habitat suit-

ability model is given in Fig. 3. The final model

is expressed in following form:

Y= constant+ B1

forest cover density

+ B2

forest cover type

+ B3 (distance from road)+ B4 (altitude)+ B5

aspect

+ B6 (distribution of Sambar)+ B7 (distribution of wild boar) (1)

Habitat suitability index is represented as:

P logit (Y) = log

1

1+ exp (Y)

(2)

Results

Habitat mapping using satellite data

The satellite data was transformed into thematic

forest cover type/land use map using onscreen

Table 4 Coefficient forwild boar

Variable B SE Wald df Sig. Exp(B)

Forest cover density 0.006 6 1.000

Forest cover density (1) 45.459 5,856.681 0.000 1 0.994

Forest cover density (2) 5.514 5,845.191 0.000 1 1.000 0.004

Forest cover density (3) 108

.267

6,343.308 0.000 1 0.986 0.000Forest cover density (4) 55.887 7,791.098 0.000 1 0.994 0.000



Forest cover type 0.012 5 1.000

Forest cover type(1) 27.306 1,829.081 0.000 1 0.988

Forest cover type(2) 54.864 1,964.701 0.001 1 0.978 0.000



Forest cover type(5) 177.553 2,497.604 0.005 1 0.943

Slope 12.127 106.045 0.013 1 0.909 0.000

Distance from settlement 0.021 0.259 0.006 1 0.936 0.979

Distance from road 0.059 0.637 0.009 1 0.926 1.061

Distance from drainage 0.271 6.067 0.002 1 0.964 0.763

Altitude 0.305 3.867 0.006 1 0.937 0.737

Aspect 0.007 7 1.000

Aspect(1) 80.850 1,326.429 0.004 1 0.951 0.000

Aspect(2) 5.487 409.929 0.000 1 0.989 241.507

Aspect(3) 1.634 1,012.831 0.000 1 0.999 0.195

Aspect(4) 23.881 1,290.907 0.000 1 0.985 0.000

Aspect(5) 12.681 1,891.896 0.000 1 0.995 0.000

Aspect(6) 5.641 208.614 0.001 1 0.978 281.823

Aspect(7) 32.766 1,173.318 0.001 1 0.978

Constant 163.756 5,821.031 0.001 1 0.978


8/13


Table 5 Coefficient forsambar

Variable Coeff SE Wald df Sig Exp(B)

Forest cover type 10.540 5 1.000

Forest cover type(1) 36.634 608.538 0.318 1 0.988 0.000





Slope 4.759 50.581 4.589 1 0.909 0.988Distance from settlement 0.020 0.236 3.459 1 0.936 0.987

Altitude 0.198 4.661 7.284 1 0.936 1.000

Aspect 5.199 7 1.000

Aspect(1) 49.976 1,297.448 0.465 1 0.636 0.559

Aspect(2) 23.049 1,786.280 2.019 1 0.495 0.167

Aspect(3) 14.256 1,276.691 0.122 1 0.155

Aspect(4) 115.942 1,788.541 3.241 1 0.727 0.087

Aspect(5) 11.689 2,520.722 0.035 1 0.072 0.000

Aspect(6) 32.369 1,536.205 2.082 1 0.853 0.178

Aspect(7) 66.204 2,225.313 3.041 1 0.149 0.008

Constant 52.879 2,077.847 2.914 1 0.081 0.002

visual interpretation technique. To maintain the

consistency the visual interpretation was carried

out at 1:50,000 scale using the visual interpreta-

tion key developed after extensive ground truth

collection. The satellite data of 2004 was classified

into 11 (seven forest cover type and four land use)

classes including viz., Sal forest, Sal mixed forest,

Sal bamboo mixed forest, mixed forest, bamboo

breaks, bamboo mixed, teak plantation, agricul-

ture, scrub, water body, and river bed. Forestcover was also classified into forest cover density.

Around 265.62 km2 area of forest come under

very dense (>70%) category which is 50.09% of

the geographic area. 142.17 km2 area of forest

are dense (4070%) covering 26.81% of the total

forest. Open forest (1040%) categories occupy

16.99 km2 of forest land (3.20%) followed by

105.48 km2 of forest area that comes under non

forest (19.79%). Agriculture is mainly confined

to plain areas, which occupy around 24.30 km2

(4.58%). It is the main occupation of the peopleliving within the boundaries of this sanctuary. The

Table 6 Coefficient forchital

Variable B S.E Wald df Sig Exp(B)

Forest cover type 34.292 10.540 5 0.061






Distance from settlement 0.001 0.001 4.589 1 0.032 0.999


Altitude 0.030 7.284 1 0.007 1.030

Aspect 5.199 1 0.636

Aspect(1) 0.823 1.207 0.465 1 0.495 0.439

Aspect(2) 1.682 1.084 2.019 1 0.155 0.186

Aspect(3) 11.937 34.184 0.122 1 0.727

Aspect(4) 2.344 1.302 3.241 1 0.072 0.096

Aspect(5) 11.550 62.130 0.035 1 0.853 0.000

Aspect(6) 1.779 1.233 2.082 1 0.149 0.169

Aspect(7) 4.908 2.814 3.041 0.081 0.007

Constant 7.114 4.167 2.914 1 0.088 0.001


9/13


Table 7 Coefficient fortiger

Variable B SE Wald df Sig Exp

Forest cover density 0.587 6 0.997

Forest cover density(1) 24.717 776.068 0.001 1 0.975 0.000



Forest cover density(4) 26.023 1,041.095 0.001 1 0.980 0.000


Forest cover density(6) 19.852 877.515 0.001 1 0.982 0.000Forest cover type 3.988 5 0.551

Forest cover type(1) 37.141 268.301 0.019 1 0.890

Forest cover type(2) 14.999 245.659 0.004 1 0.951 3,267,009.5

Forest cover type(3) 21.029 245.687 0.007 1 0.932 1,357,600,579.142

Forest cover type(4) 13.437 245.655 0.003 1 0.956 684,932.978

Forest cover type(5) 23.985 999.490 0.001 1 0.981 26,090,067,477.167


Altitude 0.047 0.021 5.134 1 0.023 0.954

Aspect 5.884 7 0.553

Aspect(1) 2.887 2.037 1.665 1 0.197 17.933

Aspect(2) 3.909 2.012 3.775 1 0.052 49.850

Aspect(3) 19.365 180.441 0.012 1 0.915 0.000Aspect(4) 3.443 1.867 3.400 1 0.065 31.270

Aspect(5) 19.076 196.732 0.009 1 0.923 192,560,342.116

Aspect(6) 6.901 3.974 3.016 1 0.082 993.650

Aspect(7) 8.854 4.358 4.128 1 0.042 7,001.200

Probability distribution 29.977 178.670 0.028 1 0.867 0.000

of sambar

Probability distribution 29.967 223.961 0.018 1 0.894 0.000

of wild boar

Constant 25.022 736.195 0.001 1 0.973 73,633,232,277.120

area distribution in different forest type, forestcover/density and land use is given in Table 1.

Analysis of habitat variables

While analyzing the distribution of the prey

predator species based collected field data, it was

found that the sambar, chital and tiger are the

widely distributed species in the entire sanctuary

area. However, wild boar is only concentrated in

the central core part of the sanctuary. Chital is

distributed in entire area with good representationin the fringes, sambar is more towards the north-

ern part of the sanctuary and tiger moves in the

entire sanctuary freely with limited presence inextreme east, west and south part of the sanctuary.

Invariably all the species prefer to be in core area

of the sanctuary except sambar. The processed

field data (habitat variable for prey, predator and

random points) is given in Table 2. There were

no significant differences in the habitat variables

recorded for tiger, random and prey plots. The

variables which had higher value in tiger plots

as compared to random plots are canopy cover,

canopy height, forest, shrub cover, shrub height

and number of dung pellets whereas the variablewhich has higher values in prey plots as compared

to random plots is distance from road. The grass

Table 8 Logisticregression coefficientaccuracy for wild boar

Wild boar Accuracy

Predicted = 0 Predicted = 1

Observed = 0 75 1 98.7

Observed = 1 3 6 66.7

R2 = 0.86 95.3


10/13


Table 9 Logisticregression coefficientaccuracy for sambar

Sambar Accuracy


Observed = 0 84 0 100.0

Observed = 1 1 11 91.7

R2 = 0.97 99.0

height has shown significant difference however;

assessing it through satellite data has not been

attempted. The analysis of data highlights im-

portance of other environmental parameters viz.,

slope, aspect and elevation along with forest cover

type and forest cover density.

The principle components 1, 2, 3, and 4

explained 52% of the variation as shown in

Table 3. The PC1 explained 18.38% of the vari-

ation, PC2 13.83%, PC3 10.21% and PC4 9.69%.

The PC1 had higher positive (+ve) loading onshrub height, number of cut stems, total numbers

of trees while the higher negative values were on

canopy height, distance from the road and num-

ber of samplings. Factor one explained trees and

taller shrubs. The PC2 had higher positive loading

on forest cover density, canopy cover and shrub

cover. While higher negative values of grass/herb

cover, number of lopped stems, and distance from

water. The PC3 had higher positive loading on

number of samplings, number of trees and canopy

height while higher negative values were for slope,altitude and distance from water. The PC4 had

higher positive value on grass height, herb/grass

cover and altitude while higher negative values

are for distance from road, distance from village

and distance from water. The PC1 and PC2 were

plotted to get the scatter plot of the distribution

of animals. The Fig. 4 shows the scatter plot. It is

evident form the plot that most of the recorded lo-

cations of tiger are in and around the distribution

of prey species.

BMLR was run individually for each preyspecies and then for the Tiger species. The vari-

ables along with the coefficients for wild boar,

sambar, chital and tiger are given in Tables 4,

5, 6 and 7 respectively. The accuracy of derived

coefficients for wild boar, sambar, chital and tiger

are given in Tables 8, 9, 10 and 11 respectively.

The prey species showed a wide distribution in the

entire area with a good degree of agreement. The

accuracy for logistic coefficients for wild boar is

95.3 (R2 = 0.86), sambar 99 ( R2 = 0.97), chital 81

(R2 = 0.65) and tiger 92.9 ( R2 = 0.87).

Habitat suitability model

The regression coefficients obtained from the

analysis were then attributed to the respective

layers. The estimated log-odds image was then

logit transformed to produce the intended prob-

ability map. The probability maps were devel-

oped for sambar, chital and wild boar. Sambar

has high probability of being found in the entire

area. However, chital and wild boar have selected

area of occurrence. The probability distribution ofwild boar is defined by forest cover density, forest

cover type, slope, aspect, altitude and distance

from road, settlement and drainage; for sambar

forest cover type, slope, aspect, altitude and dis-

tance from settlements; and for chital forest cover

type, aspect, altitude, and distance from road and

settlement.

Habitat suitability map of Tiger in Achanakmar

Wildlife Sanctuary is given in Fig. 5. As the

log-transform squashes the lower values and

exaggerates higher values and the fact that theclassification accuracies had been calculated at

Table 10 Logisticregression coefficientaccuracy for chital

Chital Accuracy


Observed = 0 44 8 84.6

Observed = 1 9 29 76.3

R2 = 0.65 81.1


11/13


Table 11 Logisticregression coefficientaccuracy for tiger

Tiger Accuracy


Observed = 0 55 3 94.8

Observed = 1 3 24 88.9

R2 = 0.87 92.9

cutoff of 0.5, the output map was sliced to least-

suitable at values lower than 0.5 and suitable at

values higher than that. Among the total 530 km2

area of the sanctuary, 24.32 km2 is found highly

suitable. Most of the suitable area falls under the

core zone of the sanctuary. This is approximately

4.59% of sanctuary area. Around 180.33 km2 area

found suitable representing 34.39% of the total

sanctuary area. Highly suitable and suitable area

together account for around 206.65 km2, which is

39.98% of sanctuary. Around 282 km2

was foundmoderately suitable for tiger which covers the

buffer zone; this is 53.35% of the total sanctuary

area and approximately 7.68% area of sanctuary is

least suitable for tiger habitat. Fringes of these ar-

eas have scattered villages and agricultural fields.

The suitable sites for the tiger are dependent

on forest cover density, forest cover type, aspect

and distance from road. Among the prey species

distribution of sambar and wild boar highly affects

the distribution and suitable sites for tiger habitat.

Conclusion

The habitat model reflects what the geostatisti-

cal tools assess with regard to the environmental

Fig. 5 Habitat suitability map for tiger


12/13


conditions suitable for Tiger in AMWLS. The

result relies on the validity of two main assump-

tions. First, there is a assumption that the spatial

maps derived from the topographic maps are of

sufficient accuracy and scale to both describe the

size and to build the habitat model. Secondly,

the pressure from the anthropogenic sources isnot more that the presence in the AMWLS and

extraction of resources in their day-to-day life.

Evaluating the results with an independent set of

field observations of tiger distribution can validate

the HSI model and map, and also some of the

underlying assumptions.

This research has presented and evaluated a

quantitative field and GIS based technique for

eliciting knowledge about habitat suitability of

tiger in Achankmar Wildlife Sanctuary. The aim

was to identify suitable habitat for tiger to builda habitat map and assist with the management of

the species in the wildlife sanctuary. Moreover,

the reclassified habitat suitability map provides

more truthful and relevant predictions. The re-

search also sought to answer questions of habitat

suitability modeling with detailed field studies and

monitoring of the species to understand its habitat

preferences. The methodology used in this paper

could be improved by consulting experts and by

improving the GIS layers to reduce error in locat-

ing tiger sites. The GIS-based approach was im-portant as it provides experts with spatial context

in a repeatable, objective and structured frame-

work. It also simplifies data management, analy-

sis and construction of spatially explicit habitat

maps. The statistical analysis of quantitative prey

sighting data helped to identify probable area for

tiger sighting in AMWLS, and this was consistent

with the outcomes of the qualitative (sighting and

field data) assessment. All ecological and ground

level specialties are found in the sanctuary area for

protected wild animals.

Acknowledgements The authors are thankful toChattisgarh State Forest Department and Officials ofAchanakmar Wildlife Sanctuary for support in the fieldwork. The work was carried out under DOS-DBT projectentitled Biodiversity Characterization at Landscape Level,which is duly acknowledged. The authors are thankful tothe anonymous reviewers for constructive suggestions andhelping in reshaping the manuscript.

References

Braunisch, C., Bullmann, K., Graf, R. F., & Hirzel, A. H.(2008). Living on the edgeModeling habitat suitabil-ity for species at the edge of their fundamental niche.Ecological Modelling, 214(24), 153167. doi:10.1016/

j.ecolmodel.2008.02.001.

Burgman, M. A., & Lindenmayer, D. B. (1998). Conser-vation biology for the Australian environment. Sydney:Surrey Beatty and Sons.

Burgman, M. A., Breininger, D. R., Duncan, B. W., &Ferson, S. (2001). Setting reliability bounds on habitatsuitability indices. Ecological Applications, 11, 7078. doi:10.1890/1051-0761(2001)011[0070:SRBOHS]2.0.CO;2.

Drury, K. L. S., & Candelaria, J. F. (2008). Using modelidentification to analyze spatially explicit data withhabitat, and temporal, variability. Ecological Mo-delling, 214(24), 305315. doi:10.1016/j.ecolmodel.2008.02.009.

Dzeroski, S., Grbovic, J., Walley, W. J., & Kompare, B.(1997). Using machine learning techniques in the con-struction of models: II data analysis with rule induc-tion. Ecological Modelling, 95, 95111. doi:10.1016/S0304-3800(96)00029-4.

ESRI (1999). ArcView GIS vers. 3.2. Redlands, CA, USA:Environmental Systems Research Institute, Inc.

Fieberg, J., & Jenkins, K. J. (2005). Assessing uncertaintyin ecological systems using global sensitivity analy-ses: A case example of simulated wolf reintroductioneffects on elk. Ecological Modelling, 187, 259280.doi:10.1016/j.ecolmodel.2005.01.042.

Hackett, C., & Vamnclay, J. K. (1998). Mobilizing expertknowledge of tree growth with the PLANTGRO and

INFER systems. Ecological Modelling, 106, 233246.doi:10.1016/S0304-3800(97)00185-3.

Hirzel, A. H., Hausser, J., & Perrin, N. (2006). Biomap-per 3.2. Lab. For conservation biology. University ofLausanne, Lausanne. http://www.unil.ch/biomapper.

Hirzel, A. H., Helfer, V., & Metral, F. (2001). Assess-ing habitat-suitability models with a virtual species.Ecological Modelling, 145, 111121. doi:10.1016/S0304-3800(01)00396-9.

Horst, H. S., Dijkhuizen, A. A., Huirne, R. B. M., &De Leeuw, P. W. (1998). Introduction of contagiousanimal diseases into The Netherlands: Elicitation ofexpert opinions. Livestock Production Science, 53,253264. doi:10.1016/S0301-6226(97)00098-5.

Mackenzie, D. I., & Royle, J. A. (2005). Designing oc-cupancy studies: General advice and allocating sur-vey effort. Journal of Applied Ecology, 42, 11051114.doi:10.1111/j.1365-2664.2005.01098.x.

Marcot, B. G. (2006). Characterizing species at risk I: Mod-eling rare species under the Northwest Forest Plan.Ecology and Society, 11, 10.

Mltgen, J., Schmidt, B., & Kuhn, W. (1999). Landscapeediting with knowledge-based measure deductions forecological planning. In P. Agouris & A. Stefanidis(Eds.), ISD99Integrated spatial databases. Lecturenotes in computer science 1737. Berlin: Springer.


13/13


Pearce, J. L., & Boyce, M. S. (2006). Modelling distrib-ution and abundance with presence-only data. Jour-nal of Applied Ecology, 43, 405412. doi:10.1111/

j.1365-2664.2005.01112.x.Smith, C., Felderhof, L., & Bosch, O. J. H. (2007). Adaptive

management: Making it happen through participatorysystems analysis. Systems Research and Behavioral Sci-

ence, 24, 567587.SPSS (1988). SPSS-X users guide (3rd ed). Chicago:SPSS Inc.

Stoms, D. M., Davis, F. W., & Cogan, C. B. (1992). Sen-sitivity of wildlife habitat models to uncertainties inGIS data. Photogrammetric Engineering and RemoteSensing, 58, 843850.

USFWS (1980). Habitat evaluation procedures report ESM102. Washington, DC, USA: United States Fish andWildlife Service.

USFWS (1996) Habitat evaluation procedures report 870FW 1. Washington, DC, USA: United States Fish andWildlife Service.

Venterink, H. G. M. O., & Wassen, M. J. (1997). A

comparison of six models predicting vegetation

response to hydrological habitat change. EcologicalModelling, 101, 347361. doi:10.1016/S0304-3800(97)00062-8.

Wightmann, R. (1995). GIS-based forest managementplanning in New Brunswick. Proceedings of ninth an-nual symposium on geographic information systems innatural resources management (Vol. 2, pp. 503506)

Vancouver, BC, Canada. GIS World.Yamada, K., Elith, J., McCarthy, M., & Zerger, A. (2003).Eliciting and integrating expert knowledge for wildlifehabitat modeling. Environmental Modelling, 165,251264.

Zaniewski, A. E., Lehmann, A., & Overton, J. M.(2002). Predicting species spatial distributions us-ing presence-only data: A case study of native NewZealand ferns. Ecological Modelling, 157, 261280.doi:10.1016/S0304-3800(02)00199-0.

Zhu, A. X. (1999). A personal construct-based know-ledge acquisition process for natural resourcemapping. International Journal of Geographical In-formation Science, 13, 119141. doi:10.1080/1365881

99241382.

Documents

Development of Tiger Habitat Suitability Model using Geospatial Tools – A case study in Achanakmar Wildlife Sanctuary (AMWLS)