Chapter 27: Modelling pygmy hog habitat
293
sent to the Natural Sciences Collections Association (NATSCA),
a UK-based organization representing natural science collec-
tions, with a similar request for information on pygmy hog speci-
mens. Finally, we searched for pygmy hog specimens in online
databases such as VertNet (http://portal.vertnet.org/search),
GBIF (www.gbif.org/) and MaNIS (http://manisnet.org/). We
confirmed the coordinates for the locations of any specimens
or those recorded in the literature, by georeferencing them on
US Army Map Service Topographic Maps using QGIS 8.2.1. The
results were mapped using ArcGIS 1.2.
Habitat Suitability Modelling
We selected the historical range of the species as the study area
for the prediction of potentially suitable habitat in Maxent mod-
elling. The historic range was designated on the basis of the
records of the former distribution of pygmy hogs collected dur-
ing the study.
We parameterized the model with presence data collected by
Ecosystem India from 2010 to 2015, which included sightings
data from the three protected areas where the pygmy hogs are
found at present, i.e. Manas National Park, Sonai Rupai Wildlife
Sanctuary, and Rajiv Gandhi Orang National Park, all of which
are located in north-west Assam. On account of the rarity and
the elusiveness of the species, none of the points were direct
sightings but rather secondary sightings, i.e. evidence of pellets,
foraging marks, nests (burnt, fresh and degraded).
We recorded 337 presence points of the pygmy hogs. The
points were particularly clustered, with more than nine points
in one pixel (94.5 m) in some instances. Biased data sets, due to
observer effort being focused in particular areas, can lead
to artificial clusters of observation, which consequently violate the
assumption of independence (Dormann et al. 2007). In order to
avoid spatial correlation, the points were filtered so that no more
than one point fell in one pixel (Rödder et al. 2009; Stiels et al. 2011;
Varela et al. 2014). Thus, the sample data set was reduced from an
initial number of 337 points to 174 points to build the model.
We chose seven variables as potential predictors of the
P. salvania habitat distribution (Figure 27.4). These variables were
chosen based on their biological relevance to pygmy hog distri-
bution. A key reason was also the availability of high-resolution
data which are generally poor for the particular region. To capture
land cover, a map was obtained from the Moderate Resolution
Imaging Spectroradiometer (MODIS), land cover map
(version 5.2). The 2010 map with the International Biosphere–
Geosphere Programme legend was procured that comprises 16
classes (10 vegetation classes, three developed and mosaic land
classes and three non-vegetated land classes). Additional con-
tinuous measures of land cover and the cover from the MODIS
Vegetation Continuous Fields product was obtained from the
NEPAL
UTTAR PRADESH
APPROXIMATE KNOWN FORMER DISTRIBUTION
PRESENT KNOWN DISTRIBUTION
UNCONFIRMED RECENT REPORTS
BIHAR
BENGAL
BANGALADESH
MEGHALAYA
BURMA
0 100 200
km
SIKKIM BHUTAN
ManasBarnadi
ARUNACHAL
PRADESH
ASSAM
Ganga
River
Bra
maput
ra Rive
r
?
?
?
?
?
Figure 27.3 Knowledge of former distribution of pygmy hogs prior to this study (source: Oliver et al. 1993).
.029
12:52:16