Rhinoceros habitat evaluation

XIII World Forestry Congress Buenos Aires, Argentina, 18 23 October 2009

1

Habitat evaluation and suitability modeling of Rhinoceros Unicornis in Chitwan National Park, Nepal: A geospatial

approach

Hemanta Kafley, Madhav Khadka, and Mandira Sharma 1

Effective Management and conservation of wildl i fe populations and their habitats largely depend on our abili ty to understand and predict species-habitat interactions. Intensive ground surveys cannot keep pace with the rate of land-use change and consequently habitat composition over large areas. We explored how effectively do Remote sensing satell i te imagery and GIS modeling technique could be used for assessing habitat suitabili ty of Rhinoceros unicornis and what are the habitat factors influencing rhinoceros distribution in lowland floodplain of Nepal. The Landsat ETM+ satell i te imagery (Path-142& Row-041) of the study area was used for classifying land use/land cover. Image processing and feature extraction was done in Erdas Imagine 8.7. We used supervised classification technique with 350 training points and 40 test points. GIS layers of habitat parameters- continuous distances from grasslands, water body, guard posts, Agriculture/ settlement and categorical land use/land cover map were used as predictor variables. The points of animal presence location were used as suitable proxies and Maximum entropy (MAXENT) modeling was run for predicting species potential geographic distribution. The most significant result of the image classification was that the proportion of pure grassland patches in the chitwan National Park is only 7 percent of the total area. Riparian forest, developed as a result of grassland succession, accounts for 8 percent of the park area which otherwise, as a grassland, served as a food source for R. unicornis. The Maxent model based on remotely sensed factors, habitat factors and rhino presence locations resulted in much larger area classified as suitable for Rhinos. The contribution of the variable distance to water (48.6 percent) was highest to impact the model. The model performance was assessed using receiver operating Characteristics (ROC) plots and Jackknife tests. The area under the training data ROC cuve (AUC) was 0.952 and that of test data ROC curve was 0.931 highly acceptable than the Random model AUC of 0.5. Therefore we concluded that Maxent modeling approach can be used to model the species geographic location for assessing habitat suitabil ity of the target wildli fe with the help of presence only datasets. The suitabil i ty map resulted from the modeling was useful to delineate the sites that required specific planning and management interventions. This result can be effectively used for enhancing suitabil i ty of different habitat types in favor of Rhinoceros and the ecosystem services the area provides for overall socio-economic and ecological sustainabil i ty that the forestry sector aims to provide. Introduction Wildlife management is a multi-disciplinary field-based applied science, essentially aimed at understanding of the relationships between the wild animals and their habitats, as influenced by human interference. The habitat themselves are complex ecosystems deriving their supportive attributes from a host of biotic and abiotic components. Habitat is a place occupied by a specific population within a community population (Smith, 1974). Habitat patterns. Roy et al., 1986 states that preservation of wild life requires a complete knowledge of their

e 1 Corresponding authors: Department of National Parks and Wildlife Conservation, Babar Mahal, Kathmandu, Nepal..


2

suitability of land (or water) as a habitat for specific wildlife species. To achieve this one needs a model to predict the suitability of land given a particular set of land conditions. Such model is called a habitat (environmental) suitability model (DeLeeuw and Albricht, 1996). In th is study we have used Maximum Entropy (MAXENT) species geographical modeling technique to model habitat suitability of Rhinoceros unicornis in Chitwan National Park, Nepal.

Greater one-horned rhinoceros (Rhinoceros unicornis) represents one of the most endangered species of large mammals in the world. Historically, the rhinos were distributed in the floodplain and forest tracts in Brahmaputra, Ganges and Indus river valleys. Formerly extensively distributed in the Gangetic p lain today it is restricted to parts of south Asia and South East Asia in Nepal and West Bengal in the north, the Duars, and Assam of India. Studies in the past have revealed that floodplain ecosystems are very dynamic. A model of landscape dynamics of chitwan showed that the fluvial action controlled the landscape organization (Lehmkul, undated). Dinesrtein (2003) argues that changing landscape organization, vegetation composition and proximity of other life requisites and disturbance factors has much to do with dispersion of rhinos.

Kaziranga national park, prime habitat of rh inoceros, after init iation of conservation programme, realized five-fo ld increase in rhino population between 1959 and 1985 and after that no significant population increase has been recorded (Kushwaha et al ., 2000). This trend has been attributed to the habitat extent and quality of the park. Chitwan population in Nepal declined to about 100 in sixties (Caughley, 1969; Pelinck and Uprety, 1972). However, effective protection measures adopted through the establishment of national park in 1973 the rhino population of the chitwan increased to about 270-310 by 1975 (Laurie, 1978). By 1988, the population rose to 358 (Dinerstein and Price, 1991), 466 in 1994 (Yonzon, 1994) and 544 in 2000 (Rhino count, 2000). Rhino census 2005 revealed total population around 375 individuals that again rose to 408 animals in 2008. Th is unpredictable population fluctuation in the dynamic floodplain habitat of Chitwan National Park demands habitat suitabilit y evaluation for identifying the key habitat factors and total suitable area for determining fate of rh inos in the park.

During the last three decades, development of remote sensing techniques has made significant contribution in the management of natural resources (Marble et. al. 1983 and Gugan, 1993.) and environmental monitoring (Kushwaha 1990, 1997). Remote sensing and GIS have been widely used in wildlife habitat studies (Roy, et al. 1995., Gratto-Trevor, 1996., Porwal et al. 1996., Verlinden and Masogo, 1997., Williams and Dowdeswell, 1998., Kushwaha et al. 2000.) Remote sensing and GIS technologies together provide vital geo-information support for relevant, reliable and timely information needed for conservation planning (Nellis et al. 1990). However, these techniques have not been widely used for wildlife habitat studies in Nepal. This study therefore used the remote sensing data and GIS technologies as well as machine based MAXENT modeling software to address the questions on habitat suitability of rhinos in Chitwan National Park. Materials and Methods We used presence-only data points of rhinoceros in the study area for modeling purpose. Opportunistic search method was used during Rhino Count 2008 for co llecting presence-only points. GPS locations of total 408 rhino-presence points were converted to UTM WGS 84 Zone 44 N project ion for subsequent GIS integration. 36 points appeared as out of bound points and were lost during data preparation phase. Therefore, 372 presence-only points were used for creating rhino location shape file. Arc GIS 9.2 software was used for creating point shapefile of the rhino presence points. This file was transformed into .CSV file for using as a dependent variable in MaxEnt program. Thus presence-only data was used as a response variable for predicting the area that could be potentially occupied by the species in the given set of conditions.

We used Erdas Imagine 8.7 software for image processing and analysis. Landsat ETM+ FCC with 30m ground resolution was used to prepare the land-use / land-cover map of the study area. We selected the bands 1,2,3,4,5 and 7 of the same resolution (30m) and layer stacked for further analysis. Thermal band (Band-6) with 60 m resolution and Panchromatic band (Band-8) with 15 m resolution were excluded to stack in the FCC prepared for data analysis. This operation was carried out to reduce data redundancy due to resampling of bands with different resolutions and insignificant contribution of thermal and panchromatic bands for vegetation analysis in this study.

We selected larger Area of Interest (AOI) beyond national park boundary to examine the modeling result in landscape level. Study area from the entire Landsat FCC was extracted by sub setting procedure


3

in Erdas Imagine 8.7 Software. We used supervised classification technique, maximum likelihood decision rule for image classification. For effect ive use of land cover map in the modeling process, we defined cover classes that were most relevant for rh inos. We established a classification sch eme to classify image into six land cover classes - Sal forest, Grass land, Riverine forest, Sand bank/ Barren land, Water bodies and Agriculture/Settlement. Water bodies are an important life requisite for rhino. It was very important to detect all the waterbodies for including in the further analysis. However, classifying 30m resolution image could not separate smaller waterbodies viz. s mall ponds, lakes, streams, rivulets etc. Moreover, eutrophication of the water bodies might have influence in rendering inseparability of small water ho les spectrally. Therefore, we on-screen digit ized the waterbodies that were not picked up by digital image classification method and then merged with the spectrally classified image using ARC INFO. Continuous surfaces of distances from all land cover types were created using Euclidean distance algorithm availab le in spatial Analysts Tools in Arc GIS. These distance layers were then converted to ASCII files using conversion tools (Raster to Ascii) available in Arc GIS Toolbox. All the distance files in Ascii format were used as continuous predictor variables and the land use/ landcover Ascii file was prepared for input as categorical predictor variables. Results and Discussions Land use / Land cover classification Supervised image classification was performed to generate Land use / land cover classification map targeted to model rhino habitat suitability study (Fig. 1).

[INSERT YOUR JPEG OR EXCEL IMAGE FILE HERE STYLE: WFC - IMAGE]

Figure 1. Figure title

Source: Style- Source

Sal fo rest (46%) is the dominant vegetation cover type in the landscape AOI selected for the habitat

study. However, within the park area sal forest constitutes 55% of the total area. Grassland occupies the smallest area (4%) in landscape selected and 6% in the park area (Fig. 2). The classification showed 141 km2 grasslands within the park and bufferzone area excluding numerous small waterbodies and swamps interspersed within the grasslands. Sal forest occupies 994 km2. Riverine forest coverage is 169 km2, most of the patches interspersed within grassland tracts as a result of succession.

Figure 1. Land use/land cover map of Chitwan National Park and periphery


4

Vegetation classificat ion map enabled me to analyse habitat blocks based on geometry and spatial

organization of land cover types. The grassland blocks in the western part of chitwan and the eastern part are connected by dense Sal forest with narrower and fragmented belt of riverine forest. This feature is qualitatively strong barrier for movement of rhinos between two areas. MaxEnt Modelling We used Maximum entropy modeling of species geographic distributions (MaxEnt) for pred icting probability of occurrence of rhinos. Continuous predictor variables as distances to Sal forest, grass lands, Riverine forest, water bodies, sand bank/ barren land and agriculture settlement and categorical land use / land cover map were used as independents to evaluate the habitat variables that effectively defines rhino presence. Distance from water sources showed highest (48.6%) heuristic estimate of relat ive contribut ion

Table 1. Relative contribution of the independent variables to the Maxent model S.No. Variable Contribution

1 Distance from Water sources 48.6%

2 Distance from Grasslands 20.0%

3 Land use / Land cover 8.9%

4 Distance from Guard posts 13.7%

5 Distance from Sal forest 2.1%

6 Distance from Riverine forest 0.9%

7 Distance from Settlement / Agriculture 6.2%

%

Figure 2. Percent Area under land use/land cover categories


5

The response curves (Fig. 3) for the model showed fairly accurate trend for rhino suitability. Pred icted

Probability of rhino occurrence decreased with the increase in distance from grasslands. Same was the case with distance from riverine forests and distance from water bodies. The response curve for the

these variables increases the probability of occurrence of rhinos. However, the curves showed this trend up to certain distance (approx 1km from

Figure 3. Response Curves of Variables affecting Maxent Prediction

sal distance and 10km from settlement distance) and beyond that the occurrence probability decreased. This may be due to the reason that there were no rhinos present beyond th ose distances of the respective cover types.


6

Figure 4. Jackknife of AUC for Rhinoceros unicornis

(Fig.4). This may be

because most of the rhino presence points were falling in the grassland habitat type. However, The environmental variable with highest gain when used in isolation is distance from water (Fig. 5), which therefore appears to have the most useful informat ion by itself. Analysis of Omission rate and predicted area as a function of cumulative threshold (Philips et al. 2005) showed that omission rate was close to the predicted omission (Fig. 6) depicting the model to be robust to conduct further analysis.

Figure 5. Jackknife of regularized training gain for Rhinoceros unicornis


7

Figure 6. Omission rate and predicted area as a function of the cumulative threshold

Maxent thus generated a habitat suitability map (Fig.7). The map was then reclassified based on the

different species occurrence probabilspecific probability thresholds to classify suitability map into different suitability classes.

Figure 7. Maxent habitat suitability map

Firstly, I classified areas with 0-5% occurrence probability as Unsuitable, areas with 5-20% probability

as acceptable (moderately suitable) and areas with 20-75% species occurrence probability as suitable habitat (Fig. 8)


8

Figure 8. Three categories of habitat suitability

However, the acceptable category did not fall under any specific land use classes so that specific

management intervention could be adopted for enhancing its suitability. Therefore, I reclassified the suitability map into two classes- Suitable and Unsuitable. The areas with 0-5% occurrence probability were classified as unsuitable habitat and area with 5-75% species occurrence probability as suitable habitat (Fig. 9). This unsuitable category included the areas that have least probability for rhinos to occur.


9

Figure 9. Suitable and Unsuitable habitats for rhinos in Chitwan National Park

While suitable category included the areas currently being used by rhinos and the areas that could be

potentially used. Also it included portions of all other land cover types that were in proximity to the water sources and grasslands. The settlement and agriculture area that came under suitable category has been interpreted as the areas prone to crop raiding and human casualties. Similarly, patches of sal forest under suitable category can be viewed as the areas that can potentially harbor rhinos, if properly managed. However, these areas require intensive habitat management interventions for being used by rhinos. Based on this classificat ion scheme, 720 sq. km. in the park and bufferzone was classified as suitable and remain ing 1082 sq.km as unsuitable (Fig. 10). Among the suitable area 73sq/km agricu lture and settlement cover type, 33 sq.km water bodies and 171 sq.km Sal fo rest cover types are presently unavailable for rhinos. Therefore I subtracted these areas from total suitable area. Thus according to this model total suitable habitat available to rhinos at present worked out to be 443 sq.km. This area included grassland, riverine forest and sand bank/ Barren land cover types.


10

Fig 10. Area under suitability category

with less species occurrence probability. Therefore for enhancing its probability habitat improvement interventions have to be carried out. This model also suggests that increasing number of water sources will help increasing suitable habitat for rhinos. This result suggests that contiguous patches under different probability of species occurrence should be worked out for improving its suitability while prioritizing area for management intervention. The result of habitat suitability map shows that rhinos can occupy the 171 sq.km of sal forest that has been categorized as suitable. Thus total potentially suitable habitat for rh inos in Chitwan National park is 614 sq.km. However, for achieving this extent of area as suitable, number of water bodies has to be significantly increased and maintained. Conclusions Evaluation of potential area for Rhinoceros unicornis can be considered as one of the most important steps towards the conservation of the rhinos. Rhino, being herbivore species, has greater affin ity towards vegetation that serves as food for it. Its main food comprises of varieties of grass species and hence it prefers to spend majority of its time in grassland habitat. As the climatic condition, where it thrives, is hot and humid, it also requires waterbodies for wallowing to keep it cool in ext reme temperatures. Moreover, the grasslands it prefers are more likely to be available in the floodplains of the rivers and maintained by the periodic flood. Likely to almost all wild animals it avoids any kinds of anthropogenic disturbances. Hence availab ility of contiguous grasslands interspersed with sufficient waterbodies and sufficiently distant from factors of disturbances is considered as suitable condition where rhino thrives well. These parameters for rhino habitat suitability are detected in appropriate resolution satellite imagery with clear distinction between the features of interest. Hence, remote sensing satellite imagery was effectively used to model rhino habitat suitability.

The results of the study revealed that 443 km2 of the park is modeled as suitable including 101 km2 grasslands, 175 km2 of sand banks / barrenlands and 167 km2 riverine forest. The patches of sal forest (171 km2 ) most of them contiguous to the Sukhibhar grassland are also modeled as suitable owing its proximity to the source of water and perhaps due to the contiguity to the grasslan d with high rhino occurrence. Thus, we conclude that if enough water holes are created and thinning operation is carried out


11

for enhancing colonization by grassland community these patches of sal forest will serve as promising habitat for rh inos. The area under agriculture and settlement land cover class which is modeled as Suitable is potentially prone to crop raid ing by rhinos.

Considering the areas used by rhinos at present and the areas modeled as suitable, we conclude that suitable habitat for rhino in the Chitwan is potentially more than that used by the species at present. If managed properly it can sustain more animals than the highest 512 indiv iduals the park harbored till date. Acknowledgements The authors thank Shyam Bajimaya, Director, Department of National Parks and Wild life Conservation of the Government of Nepal for extending permission and facilities support for this work. Our grateful thanks are due to the Alcoa Foundation and Institute of International Education (IIE) for p roviding financial support to conduct the field work, World Wildlife Fund (WWF) Washington DC for supervising the technical part of this research and International Tropical Timber Organisation (ITTO) Fellowship Program for provid ing fellowship to undertake internship to the principal investigator for completing this work working with WWF, Washington DC. Our grateful thanks to Dr. Eric Dinerstein for providing overall technical guidance and supervision throughout the completion of this project.

References Caughley. G. 1969. W ild life and recreation in the Trisuli watershed and other areas in Nepal.

HMG/FAO/UNDP Trisuli Watershed and Development Project Report No. 6, Kathmandu, 56pp. De Leeuw, J. and Albricht, R.C. 1996. Habitat evaluation, land evaluation for wildlife. Extended

Abstract. Conference on the application of remotely sensed data and geographic information system (GIS) in environmental and natural resources assessment in Africa, Harare.

Dinerstein, E and Price, L. 1991. Demography and habitat use by Greater one-horned rhinoceros in Nepal. Journal of Wildlife Mangement, 55: 401-411.

Dinerstein, E. 2003. The Return of the Unicorn is, Columbia University Press, New York, USA. pp 316+XVIII

Gratto-Trevor, C.L. 1996. Use of Landsat-TM imagery in determining important shorebird habitat in the outer Mackenzie Delta, Northwest Terrotories. Arctic, 49(1): 11-22.

Gugan, D.J. 1993. Integration of remote sensing and GIS. Proc. Int. symp. Operationalization of Remote Sensing, 19-23 April, 1993, ITC, The Netherlands, pp.77-85.

Kushwaha, S.P.S. 1990. Forest type mapping and change detection from satellite imagery. ISPRS Photogramm. & Remote Sensing, 45: 175-181.

Kushwaha, S.P.S. 1997. Environmental monitoring and cyclone impact assessment on Sriharikot Island, India. Geocarto International, 12(2): 55-62

Kushwaha, S.P.S., Azeem, A., Boruah, P., Roy, P.S., Singh, S., Lahan, P. and Bonal, B.S. 2000. Remote sensing and geographic information system in wildlife habitat evaluation. Sustainable Management of Forests-India, International Book Distributors, Dehradun.

Laurie, W.A. 1978. The ecology and behavior of the greater one-horned rhinoceros. Ph.D. dissertation, University of Cambridge, Cambridge.

Lehmkul, undated. The organization and Human use of Terai Riverine Grasslands in the Royal Chitwan National Park, Nepal. Available at http://www.fs.fed.us/pnw/pubs/journals/432-Lehmkuhl.pdf

Marble, D.F., Peuquet, D.J., Boyle, A.R., Bryant, N., Calkins, H.W. and Johnson, T. 1983. Geographic informat ion system and remote sensing. In: Manual of Remote sensing (Ed. R.N. Colwell), American Soc. Of Photogramm., Falls Church, Virgin ia, pp. 200-225.

Nellis, M.D., Lu lla, K. and Lensen, J. 1990. Interfacing geographic informat ion system and remote sensing for rural land use analysis. PE & RS, 56(3): 329-331.

Pelinck, E. and Upreti, B.N. 1972. A census of rhinoceros in Chitwan National Park and Tamaspur forest, Nepal, HMG/FAO/UNDP. Nat ional Park and Wild life Conservation Project. Internal report, Kathmandu. 10p.

Philips, S.J., Anderson, R.P. and Schapire, R.E. 2005. Maximum entropy modeling of species geographic distributions. Ecological Modelling, 190(3-4): 231-259.

Porwal, M.C., Roy, P.S. and chellamuthu, V. 1996. Wild life habitat analysis for Sambar (Cervus unicolor) in Kanha National Park using remote sensing. International Journal of remote Sensing, 17(14):2683-2697.

http://www.fs.fed.us/pnw/pubs/journals/432-Lehmkuhl.pdf


12

Roy, P.S., Ravan, S.A., Rajadyana, N., Das, K.K., Jain, A. and Singh, S. 1995. Habitat suitability analysis of Nemorhaedus goral A Remote Sensing and Geographical Syatem approach. Current Science, 69(8):685-691.

Roy, P.S., Saxena, K.G., Pant, D.N. and Kotwal, P.C. 1986. Analysis of vegetation types using remote sensing techniques for wildlife evaluation in Kanha National Park. Proc. Seminar-cum-workshop: Wildlife Habitat Evaluation using Remote Sensing Techniques, 22-23 October, 1986, Dehradun. pp 6-12.

Smith, R.L. 1974. Ecology and Field Biology (New York) Harper and Row publishers. Verlinden, A. and Masogo, R. 1997. Satellite remote sensing of habitat suitability fo r ungulates and

ostrich in the Kalahari of Botswana. Arid Environment, 35(3): 563-574. Williams, M. and Dowdeswell, J.A. 1998. Mapping sea bird nesting habitats in Franz Josef Land, Russian

High Arctic using digital Landsat Thematic Mapper imagery. Po lar Res., 17(1): 15-30. Yonzon, P. 1994. Count rhino 1994. WWF Nepal program, report series. No. 10 Kathmandu.

Rhinoceros habitat evaluation

Documents