The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B4. Beijing 2008
270
calculated. 2) From the Global Map Japan, the main streams of
rivers were derived in a national scale of 1:250,000. In order to
acquire the streams in a regional scale of 1:25,000, the DEM
was used to calculate river streams using the Hydrology tool in
ArcMap R 9.0’s Spatial Analyst. Next, the two river streams
were combined and the distance to streams variable was calcu
lated by the Euclidean distance in the Spatial Analyst of
ArcMap®. 3) All climatic variables were derived in ESRI grid
format at a resolution of 1 km by 1 km (30 arc-seconds) from
WorldClim database (Hijmans et al., 2005). The annual precipi
tation was calculated from the precipitation data for 12 months.
The minimum - and maximum temperature per month was also
derived, and from that data the annual minimum - and maxi
mum temperature was calculated. 4) The data of the roads were
decoded to the XML files, and then converted to shape files in
ESRI formats. Six different layers of roads were prepared based
on types and width (See Table 1 for detail categorization.) The
distance to each road was calculated by the same method as for
river streams. 5) Vegetation cover was directly derived from the
dataset for GIS on the Natural Environment, Japan (CD-Rom;
the source is indicated in Table 1) and contains 57 classes at
species level; species level is from this dataset. Four scenes of
Landsat-7 ETM+ imagery (dated 12 Oct 1999, 13 Nov 1999, 8
Nov 2000, and 24 Sep 2001) were geo-referenced with a maxi
mum error of 100m and ortho-rectified. Scenes were mosaicked
by histogram matching using overlapping areas. After creating
a subset of a mosaicked image, the NDVI was calculated and
resampled.
2.3.3 Preparation for test and train data
Plotting 1861 species records, 25 records fell outside the study
area. These records were discarded so that the total number be
came 1836. From all layers of environmental predictor vari
ables, pixel values were extracted for the geographic coordi
nates of species’ presence and absence records by ArcMap 9.0
R ’s in-built function.
Subsequently the dataset, containing species presence-absence
records, geographic coordinates, and each environmental pre
dictor’s value, was split into a train and a test dataset. The train
dataset was used to make predictive models. Then, the test data
set was used to assess the accuracy of these models. An inde
pendent dataset is ideal for testing the models. Thus for Asiatic
black bear, all records in the South Alps were used for training
and all records in Fuji area were used for testing the models. By
this method, also the model’s transferability can be tested
whether models developed on one local population, South Alps,
can be applied to another neighboring local population in Fuji.
To maintain a balance in presence and absence data, approxi
mately the same number of records was taken from absence
data for training and testing. Because no independent dataset
for the Japanese serow was available, its records were randomly
partitioned into two subsamples. One subsample was used as
the train dataset and another subsample was kept for testing
models. This method is known as “split-sample approach”
(Guisan and Zimmermann, 2000).
2.3.4 Statistical analysis for screening predictors
The choice of predictors is a major concern for building any
predictive model. Therefore a set of chosen predictors was
screened prior to creation of the predictive models using statis
tical analysis with R-software, version 2.4.0.
Inter-correlation among environmental predictors may cause
bias, such as overfit and multicollinearity (Grahama, 2003).
Because the environmental variables were not normally distrib
uted, the Spearman's rank correlation coefficient p was adopted.
In the preliminary data survey, we eliminated high collinearity
within the environmental variables (exclusion of the variables
in case of the Spearman p > 0.85). A similar approach was car
ried out in studies of habitat-models by Bonn and Schroder
(2001) and Fielding and Haworth (1995).
Jackknife tests were carried out to determine the relative impor
tance of variables by running an “experimental” model of Max-
Ent with all environmental variables. The variables which did
not have relative importance were eliminated.
The screened variables were used for building the final models
and are shown in Table 1.
2.4 Modeling
There are a number of modeling techniques and algorithms to
predict the probability of species occurrences by the environ
mental variables as limiting factors for species’ survival. Three
modeling algorithms: GARP - Genetic Algorithm for Rule-set
Production (Stockwell and Peters, 1999), MaxEnt -Maximum
Entropy (Phillips et al., 2006), and GLMs - Generalized Linear
Models (logistic regression models) (Nelder and McCullagh,
1989) were used in this research.
2.5 Validation and comparison of the predictive models
The accuracy of the predictive models was measured using the
test dataset by the Kappa statistics (Landis and Koch, 1977).
The pixel values of the predictive maps generated by different
modeling algorithms were extracted to the points of both train
and test datasets for each species respectively by ArcMap 9.0*.
A database with presence-absence data (value is either 0 or 1)
as the ground truth, and predicted values by each modeling
algorithm was prepared for each test dataset respectively.
2.6 Estimation of population size within habitat patches
Based on the comparison of the predictive models, the predic
tive maps by the best modeling algorithm for each species were
chosen to estimate the population size. First, the predictive
raster maps were reclassified into predicted presence and pre
dicted absence using optimum probability as cutoff values. The
predicted present location was considered to represent the “core
area” which may consist of the ecological networks if needed.
Then, the reclassified raster maps were converted to the
ESRI®’s shape files in order to calculate the area of core area in
km 2 using ArcMap“ 9.0’s VBA built-in function. Finally the
population of the target species was estimated based on known
population density and area in km 2 , derived by a following
equation.
N = A*PD
(Equation 1)
where N is estimated population (head-count), A is area (km 2 ),
and PD is population density (head-count/km 2 ).
2.6.1 Estimation of population for the Asiatic black bear
The home range of Asiatic black bear is known to vary between
50 km 2 and 70 km 2 for adult male and approximately 30 km 2
for adult female (Yoneda, 2001). Considering this minimum
home range, patches smaller than 30 km 2 were eliminated from
the core area and the other patches were grouped into potential
suitable habitat patches. Since their home ranges are not exclu-