Printer friendly format

Page 2 of 3
| Previous | Next |

The rift in the lute: Rhino habitat in the Kaziranga National Park, India

Within the fields of RS and GIS, the approach taken by Expert Classifier is closer in terms of similarity to GIS analysis than to traditional classification. Erdas Imagine’s Expert Classifier is one of the advanced classification methods to deal with different types of problems from basic land cover mapping to resource management. As has been pointed out, an expert system is a computer application that solves a specific problem or makes a decision based on a series of rules, conditions or hypotheses defined by an expert in a given field (Jordan, 2000). Expert systems (or hypothesis testing) and/or rule-based system have been used with a high degree of success in different studies, including those undertaken by Stefanov et. al., (2001), Karanja, (2002), Civco et. al. (2002), Schadt et. al. (2002) and Jacquin et. al. (2005). These illustrate the increasing use of remote sensing data combined with GIS techniques for wildlife and conservation research.

In the present study, GIS and RS data were used to assess habitat loss and locate suitable areas within the park and its immediate vicinity for possible future extension.

Database and Methodology
Imageries of six different periods from different sensors with varied spatial resolutions were used for this study. These imageries were from Landsat TM and ETM+ with 30 m spatial resolution, IRS 1C LISS III with 23.5 m spatial resolution and ASTER with 15 m resolution.

A Survey of India Topographic map sheet on 1:50000 scale of 1971 was used to extract roads and settlements along with the satellite imageries, while a Digital Elevation Model (DEM) was obtained from Shuttle Radar Topography Mission (SRTM) data. Twenty known points, collected with a GPS on January 25, 2004, were used as the training set for the supervised classification of the images. Data on rhino population, causalities of human-rhino conflicts, rhino’s habitat requirements and other related data were also collected from secondary sources and direct field observation. Since the available satellite datasets for the years 1987, 1988 and 2004 were not covering the entire area of the park, mosaics were prepared. Images from 1987 and 1988 TM data were used for 1988 and for 2004, 2003 IRS data and 2004 Aster data were used. Finally, data pertaining to six periods, i.e. October-December 1988, January 1994, December 1999, December 2001, December 2002 and January-February 2004 were used to acquire a subset according to the boundary of the park including River Brahmaputra. While data preparation, geo-referencing, mosiacing and re-projecting were performed in Erdas Imagine 8.7, ArcGIS 9.1 was used in database creation and spatial analysis.

Supervised classification was conducted for each image by using both parametric (Maximum likelihood) and non-parametric (Feature space) decision rules in Erdas Imagine 8.7. Information from the GPS points and the topomap were used to identify the landuse and generate the training sets. The park was classified into five broad classes: woodlands, grasslands, scrublands, water and sand, and for each class, the area was added to the attribute table.

Figure 2: Flowchart of main steps to obtain suitable habitat

A subset of the mainland of the park excluding Brahamputra River was then created for each year using the vector layer. The resulting land use maps were analyzed and attribute values were compared to detect the changes and were used as inputs in the expert classification. The created signatures were evaluated for separability and contingency. Accuracy assessment was done for each classified image with the help of 100 randomly generated points throughout the classified image using ‘stratified random’ distribution parameters.

The suitable areas for the rhinos were evaluated using Erdas Imagine’s Expert classifier for each year. The expert classifier has three main components, namely hypothesis, rules and conditions. The hypotheses represent the output and the intermediate classes and the rules define the hypotheses based on the input data sets through different combinations of conditions. The input data set consists of user-defined variables and includes raster imageries, vector coverages, spatial

Figure 3: Flowchart of main steps to obtain suitable habitat

models, external programs, and simple scalars. A rule is a conditional statement, or list of conditional statements, about the variable's data values and/or attributes that determine a hypothesis. Multiple rules and hypotheses can be linked together into a hierarchy that ultimately describes terminal hypotheses. Confidence values associated with each condition are also combined to provide a confidence image corresponding to the final output which is a classified image. A hypothesis forms a classification based on the truth of one or more rules. (Source: Erdas Imagine Expert Classifier Overview)

Table 1 shows the hypothesis, rules and conditions with the confidence value for each rule. After setting the rules and defining the confidence levels, expert classification for each year was performed. Changes in the habitat were then determined by comparing the six outputs.

Table 1. Hypothesis, rules and conditions used in the classification

Finally, to determine suitable areas both inside and outside the park a modified version of the same decision tree was applied to the classified images of 2003 covering a large area including outside the Park. In this step, road buffer was not included, only the road feature was included, considering the creation of animal crossings over roads. Human settlement and the agricultural fields were taken as a single class without a buffer. The best suitable and suitable classes were combined to make one suitable class whereas the areas outside the one km buffer of water, with all other requirements was considered as less suitable. Since rhinos can travel up to 3 kms from the water bodies for their food, only the minimum requirements for the rhino habitat were considered to estimate suitable areas outside the park for the expansion and creation of corridors and animal crossings over-bridges.

Results and Analysis:

Grasslands occupy the maximum area, followed by scrublands and woodlands. The presence of grasslands makes the KNP the ideal home for Indian rhinos. Encouragingly, there has been a perceptible increase in the area under grasslands (Table 2) during 1988 to 2004.

Class name	Area in km²
	1988	1994	1999	2001	2002	2004
Grasslands	148.58	135.12	109.09	134.05	192.51	186.89
Water	35.80	53.68	39.44	27.28	27.10	31.52
Scrublands	82.41	55.27	121.02	91.23	57.07	70.73
Woodlands	128.91	150.79	126.38	140.28	114.43	100.02
Sand	1.74	0.37	0.12	0.30	0.23	0.17
Total	397.44	395.23	396.06	393.14	391.34	389.33

Table 2: Land Uses in Kaziranga National Park

Apart from grasslands, all other land use types decreased. In 1994, there were decreases in the area of the grasslands and the scrublands, while area of the woodlands and water increased. On the other hand, from 1994 to 1999, area in all other land uses decreased except in scrublands. Burning of grasslands for the management purpose is one of the possible reasons for this change. In 2002, the area of the grasslands increased, but from 2002-04 there was a slight decrease. At the same time, as the grasslands were increasing from 1999 to 2002, the scrublands and woodlands areas were decreasing. During 1988-2004, sand showed the least change.