INTRODUCTION
Habitat use is the way an animal uses physical and biological resources (Krausman Reference KRAUSMAN1999). Habitat selection by a species can be considered as a multi-level hierarchical process (Johnson Reference JOHNSON1980). Selection of specific habitats by a species can be considered on multiple levels – the landscape where it is distributed and hence evolved as a species over millions of years, the site that it selects as a home range as an individual, and the microhabitats that are used for food, survival and reproduction, again at the individual level. Ungulates of different size and metabolic rates have specific requirements in terms of nutrition and to fulfil these requirements occupy various spatial and temporal niches (Jarman Reference JARMAN1974, Newing Reference NEWING2001). Ungulates also adopt specific anti-predatory strategies to protect themselves and their young ones from predators (Broom & Ruxton Reference BROOM and RUXTON2005, Caro et al. Reference CARO, GRAHAM, STONER and VARGAS2004, Leuthold Reference LEUTHOLD1977).
At the landscape level, the four-horned antelope (Tetracerus quadricornis Blainville 1816) uses tropical dry deciduous forest habitats (Baskaran et al. Reference BASKARAN, KANNAN, THIYAGESAN and DESAI2011, Berwick Reference BERWICK1974, Krishna et al. Reference KRISHNA, KRISHNASWAMY and KUMAR2008, Reference KRISHNA, CLYNE, KRISHNASWAMY and KUMAR2009; Sharma Reference SHARMA2006). Recent studies also indicate that its abundance and distribution is considerably affected by habitat changes because of local and landscape-level factors (Krishna et al. Reference KRISHNA, CLYNE, KRISHNASWAMY and KUMAR2009). However, even in forest patches typified as tropical dry deciduous, the distribution of four-horned antelope is often not uniform. For instance, in the current study the four-horned antelope was not encountered in most areas within the lower plateau of Panna Tiger Reserve (PTR), despite being a tropical dry deciduous forest.
Several environmental, anthropogenic and ecological variables were believed to govern the spatial distribution of the four-horned antelope within a tropical dry deciduous forest. While some literature suggested that the species is found in undulating terrain (Brander Reference BRANDER1923, Prater Reference PRATER1980), others suggested that its distribution is limited by water (Brander Reference BRANDER1923, Prater Reference PRATER1980) and within well-wooded forests (Prater Reference PRATER1980, Rahmani Reference RAHMANI, Mallon and Kingswood2001). Krishna et al. (Reference KRISHNA, KRISHNASWAMY and KUMAR2008) and Baskaran et al. (Reference BASKARAN, KANNAN, THIYAGESAN and DESAI2011) highlighted its preference for open habitats, whereas Sharma et al. (Reference SHARMA, RAHMANI and CHUNDAWAT2009) underscored the importance of undisturbed habitat and year-round fruit and flower availability. Although predators can also have strong indirect effects on habitat use by ungulates (Valeix et al. Reference VALEIX, LOVERIDGE, CHAMAILLE-JAMMES, DAVIDSON, MURINDAGOMO, FRITZ and MACDONALD2009), we assumed the predation pressure to be mostly uniform across the study area based on results of an ongoing study on tigers around the same period (Chundawat et al. unpubl. data).
Direct observations and cafeteria experiments indicate that the four-horned antelope mainly feeds on fruits, flowers and fresh browse (Sharma et al. Reference SHARMA, RAHMANI and CHUNDAWAT2009). In a tropical dry deciduous forest, the seasonal phenology of fruiting and flowering trees varies across species and species bloom is dependent on seasonal monsoon rain and the following dry cold and dry hot seasons. A sizeable proportion of these fruits and flowers fall naturally to the forest floor, others are dropped by primates foraging in the canopy and thereby become available to the terrestrial ungulates. We considered the overall number of tree species within a habitat patch to be a surrogate of year-round fruit and flower availability to the four-horned antelope. Our hypothesis was that tree species richness within habitat patches was the most important variable affecting the four-horned antelope distribution within a landscape. It was predicted that areas with greater tree species richness would have a greater probability of distribution for the four-horned antelope. We modelled the patchy spatial distribution probability of four-horned antelope within a tropical dry deciduous forest using six spatial variables.
MATERIALS AND METHODS
Study area
The study was conducted in Panna Tiger Reserve in Central India. PTR encompasses 543 km2 of tropical dry deciduous forest (Champion & Seth Reference CHAMPION and SETH1968). At the time of the study, there were 13 villages inside Panna Tiger Reserve with a human and cattle population of 2500 and 9500 respectively. Agriculture and cattle rearing was the main occupation of people residing inside the Reserve. Intensive fieldwork was carried out within an area of approximately 350 km2. The terrain of the area is characterized by extensive plateaux and gorges. The plateaux are separated by 10–80-m high escarpments, creating several deep gorges at many places that are characterized by steep rock faces, thick forest cover and a series of caves at the base of the escarpments. The average annual rainfall is 1100 mm, most of which falls between July and September. This is followed by a long dry spell between October and June, usually broken in January and February by occasional showers. The hilly topography and long dry season make water a major limiting factor during the dry months when the temperature can regularly exceed 45 °C. The dominant vegetation types in PTR are southern tropical dry deciduous teak forest, northern tropical dry deciduous mixed forest, dry deciduous scrub forest, Anogeissus pendula forest, Boswellia forest and dry bamboo brakes (Champion & Seth Reference CHAMPION and SETH1968). The extensive dry and short-grass habitats with open thorny woodlands support antelope populations of nilgai (Boselaphus tragocamelus Pallas 1766) and chinkara (Gazella bennettii Sykes 1831). The more mesic habitats with tall grass and associated closed mixed-species forest, mainly distributed along the major seasonal drainages, support high ungulate densities including sambar (Rusa unicolor Kerr 1792), chital (Axis axis Erxleben 1777), wild pig (Sus scrofa Linnaeus 1758) and four-horned antelope. The distribution of these habitats creates a diverse and heterogeneous landscape, where ecological conditions also vary seasonally. The Tiger Reserve supports a diverse carnivore population, which includes tiger (Panthera tigris Linnaeus 1758), leopard (Panthera pardus Linnaeus 1758), sloth bear (Melursus ursinus Shaw 1791), dhole (Cuon alpinus Pallas 1811), grey wolf (Canis lupus Linnaeus 1758), striped hyena (Hyaena hyaena Linnaeus 1758), jungle cat (Felis chaus Schreber 1777) and Indian fox (Vulpes bengalensis Shaw 1800).
Study species
The four-horned antelope (Tetracerus quadricornis) is a small antelope that weighs only about 20 kg and stands 55–60 cm at shoulder height. It is endemic to the Indian subcontinent and found only in India and Nepal (Chesemore Reference CHESEMORE1970, Corbet & Hill Reference CORBET and HILL1992, Krishna et al. Reference KRISHNA, CLYNE, KRISHNASWAMY and KUMAR2009, Krishnan Reference KRISHNAN1972, Leslie & Sharma Reference LESLIE and SHARMA2009, Rahmani Reference RAHMANI, Mallon and Kingswood2001, Rice Reference RICE1991, Sharma Reference SHARMA2006, Singh & Swain Reference SINGH and SWAIN2003). The four-horned antelope is solitary (Leslie & Sharma Reference LESLIE and SHARMA2009) and its populations are fragmented within their distribution range (Krishna et al. Reference KRISHNA, CLYNE, KRISHNASWAMY and KUMAR2009, Rice Reference RICE1991, Sharma Reference SHARMA2006). As an elusive species, direct sighting of four-horned antelope is considered to be rare in most areas where it is found (Berwick Reference BERWICK1974, Jathanna et al. Reference JATHANNA, KARANTH and JOHNSINGH2003, Karanth & Sunquist Reference KARANTH and SUNQUIST1992). However, this was not the case in PTR where the mean encounter rate of four-horned antelopes was 0.17 km−1 (Sharma Reference SHARMA2006). Considering the reasonably high encounter rate, data on coordinates of sighting and other habitat parameters were collected on forest roads from vehicles and walking randomly laid line transects between 2002 and 2006. Randomly laid line transects were walked to estimate ungulate populations using distance sampling. At each sighting, distance markers were used to record the location, and perpendicular distance of the group of animals was estimated using a rangefinder and compass. Using Distance and Azimuth Tools (Random point generator (randpts.avx) extension for ArcView 3.x, v. 1.3. Jenness Enterprises. Available at: http://www.jennessent.com.) in ArcGIS, the sighting points were located on a map and plotted together with points denoting direct sightings. Although direct sightings certainly provide information about the physical location of an animal at the time of sighting, we were concerned that covariates such as detection probability (Buckland et al. Reference BUCKLAND, ANDERSON, BURNHAM and LAAKE1994, Burnham et al. Reference BURNHAM, ANDERSON and LAAKE1980) and systematic avoidance of roads in case of vehicular surveys (Thomas & Karanth Reference THOMAS, KARANTH, Karanth and Nichols2002; this study) may have biased the data.
Baskaran et al. (Reference BASKARAN, KANNAN, THIYAGESAN and DESAI2011) report that the four-horned antelope uses areas of low grass height. However, in absence of empirical estimation of detectability, it is likely that areas with low detectability may be misrepresented as those with low probability of distribution. To check the effect of grass height on detectability, we estimated grass height wherever a direct sighting was obtained. With the help of a stick with six sections etched on it, grass height was measured as one of the following six categories: very low (0< to <10 cm), low (10< to <25 cm), low medium (25< to <45 cm), medium (45< to <70 cm), tall medium (70< to <1 m) and tall (>1 m). We used Multiple Covariate Distance Sampling (Buckland et al. Reference Buckland, Anderson, Burnham, Laake, Borchers and Thomas2004) to test the effect of grass height as a covariate and found little evidence of interaction between grass height and overall detectability. The shape of the histogram for sightings of the four-horned antelope within different grass heights did not change even after correcting for detection probability (Appendix 1). This allowed us to directly use coordinates from sightings to develop a distribution model for four-horned antelope within PTR using maximum entropy.
Geographical data
Six spatial data layers were prepared to analyse spatial resource selection by the four-horned antelope in the study area. These included habitat type (eight forest habitat classes), tree species richness (surrogate for year-long availability of fruits and flowers), slope, isotherm (indicating temperature zones within the landscape), disturbance (in the form of grazing pressure) and distance from nearest water sources. Some of the layers were prepared from remote sensing data, whereas others were created using models based on data collected in the field.
Satellite imagery (LISSIII) taken in the month of October 2004 and ground-truthing data from 260 habitat evaluation plots (Sharma Reference SHARMA2006) were used to classify the satellite imagery into eight habitat categories. Trees were identified from saplings with greater than 20 cm girth at breast height. The habitat categories included dense miscellaneous forest (tree density higher than 1100 ha−1), gladed mixed forest (mixed vegetation with a close-knit mosaic of open areas along with clusters of thick vegetation), open forest (tree density between 250 and 550 ha−1 and sapling density between 600 and 1100 ha−1), miscellaneous forest (tree density between 350 and 550 ha−1 and sapling density between 1500 and 2600 ha−1), mixed thorn forest (with thorny Acacia and Ziziphus), mixed forest with dense understorey (sapling density>2600 ha−1), scrub (higher number of saplings and a lower tree density than grasses) and grassland (dominated by grasses). Additionally, binary data on the phenological state of 54 known tree species in the study area were collected every fortnight (fruiting: yes/no, flowering: yes/no) during field visits spread across 4 y to estimate the year-long availability of fruits and flowers to herbivores. Even if all species present within a plot are detected, it is usually difficult to design a study where all species from a particular site get represented within these plots. Estimation of detection probability of various species is considered a reliable way of estimating and comparing species richness (Chao Reference CHAO, Balakrishnan, Read and Vidakovic2005, Williams et al. Reference WILLIAMS, NICHOLS and CONROY2002). Ecologists have applied capture-recapture estimators to the problem of estimating species richness (Boulinier et al. Reference BOULINIER, NICHOLS, SAUER, HINES and POLLOCK1998, Colwell & Coddington Reference COLWELL and CODDINGTON1994, Nichols & Conroy Reference NICHOLS, CONROY, Wilson, Cole, Nichols, Rudran and Foster1996) and suggest that the number of species recorded during a survey varies with survey effort and detectability of various species (Boulinier et al. Reference BOULINIER, NICHOLS, SAUER, HINES and POLLOCK1998, Williams et al. Reference WILLIAMS, NICHOLS and CONROY2002). Data were collected from an additional 501 points to estimate tree species richness. Quadrat sampling design (Williams et al. Reference WILLIAMS, NICHOLS and CONROY2002) was used representing a tight and conservatively estimated area of about 7150 ha. Software CAPTURE 2 (Rexstad & Burnham Reference REXSTAD and BURNHAM1991; revised from White et al. Reference WHITE, BURNHAM, OTIS and ANDERSON1978) was used for analysing data on species richness (Williams et al. Reference WILLIAMS, NICHOLS and CONROY2002). A program was written in Visual Basic 5.0 to incorporate the estimates of tree species richness as separate variables into the habitat map of the study area. We used multiple regression to extrapolate tree species richness across the study area using geology, slope, surface temperature, habitat/land use, grazing/disturbance pressure and distance from the nearest water body as the independent variables. A slope map was created using ArcGIS spatial analyst on the digital elevation model of the study area obtained using Shuttle Radar Topography Mission (SRTM) data. An isotherm for the landscape was created by converting the thermal infrared band (Band 6) of the Landsat imagery from spectral radiance to temperature. A grazing pressure map was prepared by modifying the grazing pressure index within the study area (Chundawat et al. unpubl. data) for each compartment (smallest administrative unit within the forest areas in India) and overlaying it on the habitat classes and the distance from nearest active settlements. Locations of all water bodies were obtained by conducting thorough surveys across the study area with support from forest department staff. Multiple buffers were prepared around these water bodies at 100-m intervals to determine distance of any point on the map from the nearest water body. All spatial layers with continuous datasets were z-transformed to have a mean value 0 and a standard deviation of 1. This allowed direct comparisons of variables whose measurements were not all at the same scale.
Modelling
The locations of four-horned antelopes on foot transects (n = 125) were treated as the primary data as these locations were not expected to have been influenced by micro-habitats such as forest roads and other geographical features such as cliffs, slopes, water bodies and settlements, given their random layout. Data from direct sightings on forest tracks and random surveys in vehicles were also pooled with those on foot transects (n = 547) and analysed separately because it was expected to improve the precision by increasing the number of locations used.
Although numerous methods are available for habitat-suitability modelling in the literature (Elith et al. Reference ELITH, GRAHAM, ANDERSON, DUDIK, FERRIER, GUISAN, HIJMANS, HUETTMANN, LEATHWICK, LEHMANN, LI, LOHMANN, LOISELLE, MANION, MORITZ, NAKAMURA, NAKAZAWA, OVERTON, PETERSON, PHILLIPS, RICHARDSON, SCACHETTI-PEREIRA, SCHAPIRE, SOBERON, WILLIAMS, WISZ and ZIMMERMANN2006), the Maximum Entropy (Maxent) species-distribution modelling algorithm (Elith et al. Reference ELITH, PHILLIPS, HASTIE, DUDIK, CHEE and YATES2010, Phillips & Dudik Reference PHILLIPS and DUDIK2008, Phillips et al. Reference PHILLIPS, ANDERSON and SCHAPIRE2006) was used to produce a habitat-suitability model for the study area. Methods of maximum entropy can be used to make predictions or inferences from incomplete information (Jaynes Reference JAYNES1957). The Maximum Entropy principle ensures that an approximation satisfies any constraints on the unknown distribution that we are aware of and, subject to these constraints, the distribution should have maximum entropy (Jaynes Reference JAYNES1957).
The Maxent modelling procedure uses continuous as well as categorical data and does not require information on true absence (Phillips et al. Reference PHILLIPS, ANDERSON and SCHAPIRE2006). In our data it was not possible to separate true absence from non-detection despite presence, therefore we could only use presence data for the modelling process. The Maxent model has been used widely and tested to converge to optimal probability distribution using logistic models (Phillips et al. Reference PHILLIPS, ANDERSON and SCHAPIRE2006). It generates a probability distribution over pixels in the grid, starting from uniform distribution and repeatedly improving the fit of the data. The spatial layers were used as environmental variables, and the two datasets (foot transects and all direct sightings) were used as presence locations.
The jack-knife test was used to measure the importance of specific variables in the model as it is less likely to be influenced by correlation between the environmental variables used (Baldwin Reference BALDWIN2009). For the jack-knife approach of assessing variable importance, each variable was excluded in turn and a model created with the remaining variables. A model was also created using each variable in isolation as well as using all variables. The variable that decreased the gain most when omitted is considered to have the most useful information not present in other variables. This, along with the variable with highest gain when used in isolation, helped us identify the most important variables.
The Receiver Operating Curve (ROC) was used to gauge the predictive power of the model using the same data for training and testing. Since we used data on only presence (no absence), the fraction of the total study area with predicted presence was used for specificity instead of the fraction of absences predicted as species presence on the x-axis. On the y-axis, we used omission rate as an indicator of sensitivity. The curve represents the dependence of predicted suitability on selected variables and dependencies induced by correlations between the variables and the area under the curve (AUC) varies between 0 and 1 (Fielding & Bell Reference FIELDING and BELL1997). The higher the AUC, the better the model is expected to be at predicting the presences contained in the test sample of the data.
RESULTS
Data on phenology indicated that in every season at least four species bear fruit or flowers in the study area (Figure 1). Monsoon (July–September) was the only time of year when fruit or flower availability to herbivores was low. Even though forest types were similar across the study area, the tree species richness was not uniform, and mean ± SE varied from 11 ± 0.79 species in some sites in the lower plateau to up to 79.8 ± 17.0 species at some sites in the middle plateau. No spatial variation in the fruiting and flowering patterns of trees was observed within the study area, therefore we assumed that the temporal variation in phenology was sufficient to describe year-round availability of fruits or flowers to herbivores within habitat patches. Since more than 60% of the tree species bore palatable fruits or flowers at some time during the year, species richness within these habitat patches was treated as a surrogate of constant availability of fruits or flowers to herbivores.
Figure 1. Temporal (fortnightly) availability of fruits and flowers from trees in Panna Tiger Reserve, India estimated as the number of tree species bearing fruits or flowers at any point in time.
The ROC curve for sensitivity vs. false positive rate (1-specificity) provided the AUC equal to 0.966 for foot transects, and 0.946 for all locations (Figure 2). The jack-knife test of variable importance (Figure 3a, b) revealed that in both cases tree species richness contributed most significantly to the models and was identified as the most important variable. Other variables in declining order of importance were grazing pressure, habitat, distance from water, slope and temperature. Within the habitat classes, the probability distribution was highest for dense forest, followed by open forest, miscellaneous forest, gladed forest and grasslands.
Figure 2. Receiver Operating Characteristic (ROC) curve for sensitivity vs. false positive rates (1-specificity) for randomized training data using all sightings (a) as well as sightings on foot (b) of the four-horned antelope in Panna Tiger Reserve. Both datasets provided a good fit denoted by the area under the curve with values close to 1 (AUCall = 0.946, AUCfoot = 0.966).
Figure 3. Jack-knife test to assess the relative importance of variables for the models based on sightings of four-horned antelope in Panna Tiger Reserve only on foot transects (a) and all sighting including foot transects and other random encounters (b). Areas under the curve were compared for model outputs with and without one of the six variables, and with all variables used together. The variables compared included habitat classes (Habitat), grazing pressure (Grazing), distance from nearest water body (Water), terrain slope (Slope), tree species richness (Specrich) and isotherm (Temp).
The response curves of each environmental variable help understand their respective effects on the distribution probability of four-horned antelopes within the study area (Figure 4). It is evident that the distribution probability decreased with increase in grazing pressures, distance from water, slope and temperature. On the other hand, the species was more likely to be distributed in areas that had greater tree species richness. Analysis of the only categorical variable, habitat type, shows that dense, open, miscellaneous and cluttered open and closed forest types were more likely to be used by the four-horned antelope as compared with the remaining classes.
Figure 4. Response curve plots reflecting the dependence of predicted habitat suitability of four-horned antelope in Panna Tiger Reserve, both on the selected variable and on dependencies induced by correlations between the selected variable and other variables. Each curve represents a Maxent model created using only the corresponding variable. In case of habitat class (a), each number represented a particular habitat class (1 = water, 2 = disturbed, 3 = gladed mixed forest, 4 = dense miscellaneous forest, 5 = open, 6 = mixed thorn forest, 7 = miscellaneous forest, 8 = mixed forest with dense understorey, 9 = scrub, 10 = grassland and 11 = cloud), whereas the units for continuous variables are the respective z-scores of grazing pressure (b), distance (m) from nearest water body (c), terrain slope in degrees (d), tree species richness (e), and isotherm in °C (f).
We present here the results of the all-locations data model rather than the foot-transect data, as there were only minor differences between the two outputs but a greater sample size in the former, thereby adding to the generality of the model. The model output (Figure 5) shows that the middle plateau of PTR had a greater four-horned antelope distribution probability, followed by pockets in the upper plateau and a small number of patches in the remaining areas of the Tiger Reserve. Only 9.5% of the 543 km2 of PTR was predicted to have high probability of distribution (>0.5) of four-horned antelope.
Figure 5. Model output highlighting the areas with high predicted probability of presence (P >0.5) of the four-horned antelope (4HA) in Panna Tiger Reserve, India, depicted in black.
DISCUSSION
Studies show that the four-horned antelope uses open habitats and deciduous forest types at the landscape level (Krishna et al. Reference KRISHNA, CLYNE, KRISHNASWAMY and KUMAR2009, Rice Reference RICE1991, Sharma Reference SHARMA2006). Despite the deciduousness and uniform patterns of forest types, the four-horned antelope is seldom uniformly distributed within landscapes. In PTR, several transects and surveys on foot as well as by vehicle reported no sighting of the species in the lower plateau despite presence of similar forest type classes and as good access to water as the other two plateaux (middle and upper).
Our study provided an opportunity to look at the four-horned antelope distribution at a finer scale within tropical dry deciduous forests. While our model generated a reasonably strong response from the open forest class, it also highlighted the insufficiency of this variable alone in successfully predicting the distribution of four-horned antelope within PTR. Our model indicated that the four-horned antelope inhabits dense as well as open forested habitat but avoided human presence and open habitats like savanna grasslands. Other variables such as water, undulation (high slope) and temperature regimes across the Reserve did not influence the distribution, therefore indicating their low relevance, especially in determining the probability distribution of four-horned antelope within a landscape. On the other hand, the inclusion of tree species richness as an environmental variable improved the output considerably. Jointly, all variables provide a replicable multiplicative model of four-horned antelope distribution within a tropical dry deciduous forest.
The four-horned antelope is a small-bodied antelope and is expected to have specialized dietary requirements with high nutritional value (Jarman Reference JARMAN1974). Observations from the field indicate that it relies almost entirely upon fruits, flowers, pods and fresh browse for forage (Sharma et al. Reference SHARMA, RAHMANI and CHUNDAWAT2009). This was in contrast with Baskaran et al. (Reference BASKARAN, KANNAN, THIYAGESAN and DESAI2011) who indicated that only 8% of the four-horned antelope diet comprised fruits and flowers. However, a lacuna in their study was that 48% of the plant matter in the faeces could not be identified under a microscope, which could have represented the relatively more digestible plant matter such as fruits and flowers. Given its specialized dietary requirements, the four-horned antelope depends on the fruiting and flowering of trees and this was provided for in different seasons by different species. Greater tree species richness can be treated as a surrogate variable implying a greater probability of year-long supply of food to four-horned antelopes in the form of fruits, flowers and pods dropping from the trees.
Our model helped explain the distribution of four-horned antelope, largely limited to the middle and upper plateaux of PTR by using tree species richness as a critical surrogate variable indicating forage security. The model thus also identifies the presence of four-horned antelope as an indicator of high tree diversity within tropical dry deciduous forests. However, conversely, absence of the four-horned antelope does not necessarily indicate poor tree diversity as the four-horned antelope responds to many variables. The four-horned antelope occurs in low densities at most places where it is distributed (Krishna et al. Reference KRISHNA, CLYNE, KRISHNASWAMY and KUMAR2009, Rice Reference RICE1991, Sharma Reference SHARMA2006). Its distribution across the subcontinent is limited to dry deciduous forests; this makes the species more vulnerable to alteration in habitat structure and type, forest fires, excessive harvest or collection of minor forest produce and human disturbance because these directly influence its dietary and anti-predatory requirements (Sharma et al. Reference SHARMA, RAHMANI and CHUNDAWAT2009). The model developed here provides an explanation for the low and patchy distribution of four-horned antelopes within tropical dry deciduous forests.
The four-horned antelope is listed as vulnerable by the IUCN and is included in Schedule I of India's Wildlife (Protection) Act, 1972 (amended 2006). Its population has declined in some Protected Areas, such as Pench and Sariska Tiger Reserves (Johnsingh Reference JOHNSINGH2006; B. Wright pers. obs.), but the reasons behind such declines are mostly unknown. We recommend that conservation and management plans for protected areas and forest areas with multiple use give emphasis to this species as an indicator of high floristic diversity and hence habitat quality. The tropical dry deciduous forest types have been most impacted by anthropogenic pressures, and hence are also the most vulnerable (Chundawat et al. Reference CHUNDAWAT, GOGATE, JOHNSINGH, Seidensticker, Christie and Jackson1999). Considering their vulnerability and relevance as indicators of species richness of a habitat patch within tropical dry deciduous forests, it would be valuable to monitor four-horned antelope populations and the changes in distribution over time. Such monitoring may in turn help assess the health of these forest patches.
ACKNOWLEDGEMENTS
We are grateful to the Department of Science and Technology for funding the study that resulted in this paper and to Madhya Pradesh Biodiversity Board for their support. The Chief Wildlife Warden, Field Directors of Panna Tiger Reserve, Deputy Directors and Range officers along with other forest department staff are thanked for providing necessary permits and ensuring smooth field work during the study. We thank Drs David M. Leslie, Gopi Sundar, Aparajita Datta and Yash Veer Bhatnagar for their useful comments on the manuscript. We are also thankful to the anonymous reviewers for their critical, yet constructive comments. Finally, we would like to express our gratitude and condolences to late Mr Uttam Singh Yadav, our field assistant who lost his life to tuberculosis because of lack of medication in the remote forest village where he lived – had it not been for his skills in identifying vegetation, spotting animals and cooking food, the study would have never been successful.
Appendix 1. Naïve and corrected proportion of four-horned antelope sightings across different grass heights in Panna Tiger Reserve. Corrected number of sightings obtained with the help of detection probability estimated across different grass heights using distance sampling.
