Identification of suitable habitat for Swamp Deer Rucervus duvaucelii duvaucelii (Mammalia: Artiodactyla: Cervidae) in Chitwan National Park, Nepal

Swamp Deer is a globally threatened large-sized deer species confined within small patches of the Indian subcontinent. Historically, Swamp Deer occurred in Chitwan National Park, Nepal but was extirpated in the 1960s primarily due to widespread hunting. We assessed the habitat suitability at present for the Swamp Deer in Chitwan National Park using multi-criteria analysis in GIS and vegetation assessment using frequency, dominance, and cover. Within the 952.63km2 area of the national park, the habitat suitability analysis identified 14.57km2 as highly suitable, 134.87km2 as suitable, and 803.19km2 as moderate to least suitable area. Most of the national park’s grassland is suitable for Swamp Deer. Grassland is dominated by Saccharum spp.; Imperata cylindrica is the most widely distributed grass species followed by Saccharum spp., Narenga porphyrocoma, and Apluda mutica. Grass species of the Poaceae family are the most preferred species by Swamp Deer, which are found within short grasslands. The study revealed that Padampur Phanta could be the most suitable site for the reintroduction of Swamp Deer due to its highest proportion of short grass and availability of preferred food species and good habitat in comparison to other blocks. Invasion of swamps of Chitwan by Mikania micrantha and Eichornia crassipes could be a limiting factor for the habitat suitability of Swamp Deer.


INTRODUCTION
Swamp Deer Rucervus duvaucelii is listed as Vulnerable in the IUCN Red List and included in Appendix I of CITES. The species is distributed in isolated pockets of the Indian subcontinent with an estimated total population between 3,500 and 5,100 individuals. The altitude range of Swamp Deer is 100-300 m (Duckworth et al. 2015). Groves (1982) identified three subspecies of Swamp Deer, namely, R. d. duvaucelii, R. d. branderi, and R. d. ranjitsinhi. Rucervus duvaucelii duvaucelii is distributed in the Indo-Gangetic plain north of the Ganges, including Nepal and parts of India. Rucervus duvaucelii branderi is distributed between the Ganges and Godavari rivers in central India and Rucervus duvaucelii ranjitsinhi in the Brahmaputra floodplain in eastern India. The subspecies R. d. duvaucelii became extinct from Pakistan and R. d. ranjitsinhi from Bangaladesh (Qureshi et al. 2004). Historically, the Swamp Deer was distributed in swampy grasslands throughout the Terai of Nepal (Mishra 1982;Gurung 1983). The remaining population of Swamp Deer is now limited to Bardia National Park (BNP) supporting 106 individuals and Shuklaphanta National Park (ShNP) supporting 2,300 individuals in the western Terai. The population in ShNP holds the world's largest herd of Swamp Deer (Poudel 2007;DNPWC 2015) and establishes the site as globally important for its conservation.
In contrast, the Swamp Deer populations in BNP and ShNP face continuous food shortage, disease, and other anthropogenic stresses (Poudel 2007). Small and isolated populations are much more likely to go extinct due to demographic stochasticity, environment stochasticity, and genetic drift, or simply by chance events. Therefore, identification of potential habitats and establishing new populations through translocation are necessary for the long-term survival of this species.
Swamp Deer use water for drinking at least two times a day in winter and in monsoon but three times or more in summer (Bhatta 2008). Similarly, this species shows a high preference for grassland plots with water holes (Bhattarai 2015). On an average, Swamp Deer move 2-3 km a day (Martin 1977;Qureshi et al. 1995); therefore, water sources should be located within the daily range of the species.
The suitable habitat of a species can be assessed based on the availability of the food, water, cover, and space needed for the species. Habitat suitability index (HSI) modelling is one of the widely accepted methods used to assess habitat for the translocation/ reintroduction of species. Such assessment analyzes the relations of the species with their habitat (Guisan & Zimmermann 2000). A quantitative measurement through systematic ground surveys is a traditional and resource-intensive approach for obtaining information about the habitat. Geospatial technology can supplement such intensive survey methods (Nandy et al. 2012). Habitat suitability modelling was used in different areas throughout the world to determine potential sites for the translocation and restoration of different species. In Nepal, habitat suitability modelling was applied for the Snow Leopard Panthera uncia in western Nepal, One-horned Rhinoceros Rhinoceros unicornis in Chitwan National In this study, we used multi-criteria analysis to identify the potentially suitable habitat for the reintroduction of R. d. duvaucelii in Chitwan National Park (CNP) based on vegetation composition of grassland blocks and riverine forest patches. The study is intended to provide useful information for wildlife managers, conservation officers, and government authorities.

Study Area
CNP is the first national park of Nepal (Bhuju et al. 2007) and was established in 1973 covering an area of 952.63km 2 , 27.282-27.703 0 N & 83.839-84.773 0 E) surrounded by an additional 750km 2 buffer zone (Fig. 1).
The vegetation of CNP consists of tropical to subtropical forests with mosaics of successional communities at different stages in alluvial floodplains, including Bombax ceiba-Trewia nudiflora riverine forest to climax Shorea robusta (Thapa 2013; CNP 2015a). Most of the national park grasslands are dominated by Saccharum spp. The park is home to many rare and threatened species including Tiger Panthera tigris, Indian Rhinoceros Rhinoceros unicornis, Asian Elephant Elephas maximus, Gaur Bos gaurus, Sloth Bear Ursus ursinus, South Asian River Dolphin Platanista gangetica, Bengal Florican Houbaropsis bengalensis, and Gharial Gavialis gangeticus (CNP 2015a). The park is distributed at an altitude of 150-850 m (CNP 2015b).

Habitat Survey Sample Design
The park area was divided into three strata: grassland (including swampy areas and wetlands), riverine forest, and other forest land. Floristic composition of vegetation was assessed in the preferred habitats of Swamp Deer (grassland and riverine forest). To represent the grasslands of Chitwan adequately, the three largest patches, namely, Padampur Phanta (PP), Bhawanipur- Jarneli (BJ), and Bhimle-Sukibhar (BS), were selected for field sampling. A total of 120 random sample points were set out proportionately in these grasslands (Fig. 2). Similarly, nine sample points were laid systematically (at 400m interval) in the riverine forest in the eastern sector of CNP, which starts from Sauraha.
Vegetation assessment in each sampling plot within the grassland blocks was carried out using 1m×1m quadrats (Oosting 1956).
Frequency, dominance, and cover were used to assess the vegetation status. During the study, the team failed to study the density of grassland plots due to the dominance of tall grass species (>3m) in the park. Therefore, the ocular method was used to estimate the frequency, dominance, and cover in the grassland. In riverine forest patches, density, frequency, abundance, and cover were studied using nested quadrats of 20m×25m, 10m×10m, 5m×5m, and 1m×1m for poles having a diameter more than 29.99cm, trees having a diameter 10-29.99 cm, herbs having a diameter less than 10cm, and shrubs having heights less than 1m (CFD 2004).

GIS data collection/ preparation
The national park boundary layer was obtained from the department of survey (DoS), Nepal. Landsat-8 satellite images of March 2016 were used to obtain the updated land cover (or habitat type) of the park area. Supervised image classification using 253 training sample points collected from the field study was used with maximum likelihood approach. The image was classified into six land cover categories, namely, river, riverbank, riverine forest, forest, grassland, and bushes with 76.67% accuracy. The classification shows 692.47km 2 (72.69%) of the park area is covered by Shorea robusta and mixed forest, 112.80km 2 (11.84%) by grasslands, and 93.45km 2 (9.81%) by riverine forest. A small part of the park is covered by riverbank (32.10km 2 , 3.37%), water bodies (18.86km 2 , 1.97%), and bushes (2.95km 2 , 0.31%). The land cover classification map is shown in Fig. 3.

Suitability Analysis
The land use classified raster image was used in suitability analysis. Water source, road, settlement, and vegetation type/land use thematic layers were rated into habitat values by assigning habitat quality rating (HQR) based on their suitability on a scale in decreasing order of suitability (1 for highly suitable and 4 for least suitable) (Nandy et al. 2012). The area close to a water source was considered to be highly suitable and the area close to district/feeder road or settlement was considered to be least suitable due to high human interference. The different thematic layers were adopted from Nandy et al. (2012) and modifications were made in HQR based on findings of Bhattarai (2015). HQR for distance to water was modified with expert judgment from less than 1000m to less than 500m for highly suitable, 500m to 1000m for suitable category, 1000m to 2000m for moderately suitable, and more than 2000m for least suitable category. Similarly, the park area having an elevation greater than 300m was considered least suitable. HQR is given in Table 1.
After rating all thematic layers into 1 to 4 suitability classes, a weighted sum was done in GIS environment. Weights were assigned to different layers by using a decision-aiding tool, analytical hierarchy process (AHP) (Saaty 1987), and weights were given using Nandy et al. (2012) and modified with site condition. Pair-wise comparison matrix of variables vegetation types/ land use, water source, road, and settlement was carried out and is shown in Table 2. Vegetation type and water source are essential habitat factors and road and settlements are disturbance factors. During the evaluation process, more weight was given to vegetation type/ land cover and then to water bodies, while less weight was given to road and settlement. Between road and settlement, more weight was assigned to road as a district road (Bharatpur-Madi-Thori) crosses the national park core region from Kasara to Bankatta and all of the settlement lies outside the natural river boundary which separates the national park core region from its buffer zone.
The pair-wise matrix was then synthesized to standardize by dividing each element of the matrix by its column value total. The priority vector was obtained by averaging row in Table 3.   The consistency ratio (CR) of the pair-wise comparison matrix was 0.043. The CR less than 0.1 is accepted according to Saaty's principle (Saaty 1987). To calculate consistency ratio, consistency index (CI) was calculated and random index (RI) was derived from Saaty's index.
The weights of different variables, obtained from the above analysis, were used in the weighted suitability to evaluate the suitable habitat for Swamp Deer. The HSI is calculated as HSI = 0.558×VLI + 0.263×WI + 0.122×RI + 0.057×SI where VLI = vegetation type/land use Index, WI = water source index, RI = road index, SI = settlement index.

Habitat Suitability
The habitat suitability analysis showed that about 14.57km 2 (1.53%) of CNP was found to be highly suitable, 134.87km 2 (14.16%) suitable, 203.89km 2 (21.40%) moderately suitable, and 599.30km 2 (62.91%) least suitable for Swamp Deer. Most of the grassland of CNP falls into the suitable to highly suitable categories. The distribution of the areas with varying degrees of suitability is shown in Fig. 4.
Similar proportion of highly suitable to suitable area was found in all three studied blocks; however, the highest area was found in BS followed by BJ and PP grassland blocks. Some small patches were also identified as highly suitable west of BS and east of PP (Table 4).

Habitat Composition
Grassland: Among the 120 quadrats sampled, 90% (n=108) sample points were distributed in grassland, 8.33% (n=10) in forest, and 1.67% (n=2) in bushes. Of the 108 grassland quadrats, 20.37% (n=22) were found to be of short grass (<3m) and 79.63% (n=86) of tall grass (>3m). The highest proportion of tall grassland was   found in BJ (97.06%, n=33) and the lowest in PP (70.73%, n=29). Most of the short grassland quadrats observed (n=9) at BS grassland were located on the banks of the Rewa and Rapti rivers. Frequency analysis revealed Imperata cylindrica as the most common species with a frequency of 96.13% (n=105) followed by Saccharum arundinaceum, Narenga porphyrocoma, Apluda mutica, and Cynodon dactylon. Saccharum arundinaceum (54.63%) was found to be the most dominant grass species in the study area followed by N. porphyrocoma, I. cylindrical, and Themeda spp. All of the study quadrats had areas more than 80% covered by grass species. The swampy areas and waterholes of the park were covered by Mikania micrantha and Eichornia crassipes. The block-wise dominant and co-dominant species are given in Table 5. During frequency and dominance calculations, species whose occurrence was less than 5% of the total number of quadrats were eliminated due to insignificant occurrence (Table 5). Riverine Forest: The highest density (0.01, 0.42, 0.11, and 0.89), frequency (100, 100, 100, and 77.79), and abundance (2.33, 4.22, 2.78, and 1.43) in riverine forest were found on Trewia nudiflora in nested quadrat of 20mx25m, 10mx10m, 5mx5m, and 1mx1m area. Most of the sample plots (80%) in riverine forest were found to be an association of T. nudiflora and Bombax ceiba. Similarly, Ehretia laevis and Litsea monopetala were also found frequently in the riverine forest. Murrya koenigii and Toona ciliata saplings were found on 30% and 10% plots, respectively. The seedlings of T. nudiflora, L. monopetala, M. koinegii, and E. laevis and invasive species Mikania micrantha and Lantana camara were also observed in the studied quadrats. An average of 75% crown coverage and 50% ground coverage was estimated in riverine forest patches.

DISCUSSION
The GIS based multi-criteria study identified 149.44km 2 (15.69%) as suitable of which 40.57km 2 (4.26%) lies in three major grasslands. This constitutes 97.67% of the total area of these three major grasslands. BS (14km 2 ) has the largest area of suitable habitat followed by BJ (13.48km 2 ) and PP (13.10km 2 ). Habitat suitability was mostly influenced by the presence of water sources and vegetation types. In addition to the above-mentioned grasslands, most of the other grassland patches were also found suitable which includes the floodplain of three major river systems of the park, namely, Rapti, Rewa, and Narayani. The highly suitable area of both PP and BJ grasslands are distant from Rapti River and most of the BJ grassland is bordered by riverine forest. On the other hand, the highly suitable area of BS falls on the river bank of Rewa and the suitable area is extended along the bank of Rapti River. All of the suitable category habitat areas have a risk of inundation and flooding during rainy season. The BS grassland, however, has a high risk of flooding. Similarly, BS is separated from the eastern grasslands (PP and BJ) by Shorea robusta forest and the Bharatpur-Madi-Thori District Road, which possibly restricts the movement of Swamp Deer. There is, however, the possibility of joining BS with suitable grassland patches that are uniformly distributed in the western part.
The study of vegetation composition in the quadrats showed that 20.37% area of the three grassland blocks is covered by short grasses. According to previous studies, short grasslands, rivers, and riverbeds are the most preferred habitat of Swamp Deer. In the study area, I. cylindrica, C. dactylon, S. spontaneum, S. bengalensis, D. bipinnata, Phragmites karka, Cyperus spp., N. porphyrocoma, Themeda spp., Apluda mutica, Hemartha compressa, and Arundinella nepalensis were reported as food species of Swamp Deer by previous researchers. This study also found I. cylindrica as the most commonly distributed grass species followed by S. arundinaceum and N. porphyrocoma in the park. The availability of the most preferred food species I. cylindrica increases the suitability of the area which is dominant only in 16.67% of the quadrats. In contrast, most of the grasslands are dominated by S. arundinaceum (54.63%) which will be least suitable after growth in summer (April-June).
If there is a decision to reintroduce Swamp Deer in CNP, PP has the highest proportion of short grassland and dominance of I. cylindrica, and therefore could be selected based on vegetation, however, this area (PP) bears challenges of inundation and flooding due to proximity to a perennial river system. Study quadrats in BJ had 97.06% tall grassland, thus, it would be favourable to Swamp Deer only from February to April when grass sprouts start to emerge. BS grassland is dominated by tall grass species (72.73%) and most of the short grasses are distributed along the Rewa and Rapti rivers. In case of heavy flooding and inundation, there is a risk of these grassland patches being modified. Hence, for this area to be appropriate for the reintroduction of Swamp Deer, habitat management interventions are required around Bhimle and Sukibhar posts.
All of the grasslands of CNP lie in close proximity to Rapti, Rewa, and Narayani rivers. Similarly, in some areas, park management developed water holes which provide water to wildlife species. All of the suitable areas of CNP lie within a 2km-distance to a water source and are, therefore, within the daily range of the Swamp Deer.
For the viability of the species, sufficient numbers of individuals need to be reintroduced. Previous studies on Swamp Deer were mainly focused on habitat preference, food and feeding behaviour, and herd size and there is no information available on the area required for this species. The study from JJCR found that 14km 2 is insufficient for 134 Swamp Deer (Nandy et al. 2012) and 0.009km 2 area per individual is insufficient in maintaining a viable population. We believe that the suitable area identified by this study can support only a small population as described by Nandy et al. (2012) based on study in JJCR. We, however, would like to recommend a more detailed study regarding population viability.
This study and previous studies by Ram (2014) and Lamichhane et al. (2014) in CNP showed that the national park is severely affected by the invasion of Mikania micrantha, a mile-a-minute invasive species. The intensity of M. micrantha invasion is greatest in and around sources of water. Most of the swampy grassland area is invaded by this species. There is no information available on the response of Swamp Deer to M. micrantha up to now, but this invasive species could have a strong impact by limiting food availability. In addition to invasive species, several other environment factors (such as population density, social organization, preypredator relationship, ungulate species relationship, and risk assessment) could affect the suitability of an area for the reintroduction of the species and these were not assessed in this study. Further research on these topics is recommended.

CONCLUSION
Habitat suitability assessment using multi-criteria analysis in a GIS environment showed 149.44km 2 of the park area provides suitable habitat for Swamp Deer in CNP including 100.14km 2 grassland, 25.72km 2 riverbank, 16.64km 2 water bodies, 4.46km 2 forest, 1.86km 2 riverine forest, and 0.62km 2 bushes. A total of 112.80km 2 of the park is grassland. Out of this, we assessed 41.54km 2 area in three major grasslands for the vegetation suitability. Amongst the assessed blocks of grassland, 40.57km 2 was found to be suitable according to the parameters of the study. Considering suitability in terms of habitat size, food, and water availability, PP is found to be more suitable for reintroduction of Swamp Deer compared to the other two sites. Alternatively, BS grassland could be another potential habitat for reintroduction of Swamp Deer with habitat management interventions. BJ has only seasonal suitability due to the high coverage of tall grass species.