Jinamoy et al.: Predictive distribution modelling for rufous-necked hornbill in Western Forest Complex Conservation & Ecology RAFFLES BULLETIN OF ZOOLOGY 62: 12–20 Date of publication: 21 February 2014 http://zoobank.org/urn:lsid:zoobank.org:pub:6A289E8C-422C-4CE6-AA25-C2945039C613 Predictive distribution modelling for rufous-necked hornbill Aceros nipalensis (Hodgson, 1829) in the core area of the Western Forest Complex, Thailand Sitthichai Jinamoy1, Yongyut Trisurat1*, Anak Pattanavibool2, Chatchawan Pisdamkham3, Sompong Thongsikem3, Vittaya Veerasamphan3, Pilai Poonswad4 & Alan Kemp5 Abstract. The rufous-necked hornbill, Aceros nipalensis (Hodgson, 1829), is listed as vulnerable and is found only in the Western Forest Complex. The objectives of this research were: 1) to estimate the geographical distribution for the rufous-necked hornbill at the Thung Yai-Huai Kha Khaeng World Heritage Site in both the breeding and non-breeding seasons; and 2) to determine seasonal changes in its habitat use. We collated the occurrence records of the rufous-necked hornbill from long-term monitoring data and conducted additional surveys during 2004–2008. In addition, spatial layers for potential environmental variables that might affect hornbill distribution were developed and Maximum Entropy (MaxEnt) modelling technique was used to generate potential distributions. The results indicated that MaxEnt models performed very well and the overall accuracies of the predicted maps in breeding and non-breeding seasons derived from the contingency matrix were 81% and 85% respectively. In addition, altitude and land cover were considered significant variables in the species distribution model. Suitable habitats for the rufous-necked hornbill were predicted in the high altitude evergreen forest and were clustered into three patches in the center of Thung Yai Naresuan West, Huai Kha Khaeng, and along the western boundary of Huai Kha Khaeng adjoining Thung Yai Naresuan East. Suitable habitats covered 11.7% of the world heritage site, of which 6.6% and 9.2% were in the breeding and non-breeding seasons respectively, owing to the fact that the home range during breeding season is smaller compared to non-breeding season. Future conservation efforts should focus on enhancing the connectivity between suitable large and small patches within the distribution range, the installation of artificial nests, and patrolling to minimise poaching. Key words. MaxEnt, maximum entropy modelling, species distribution modelling, Thung Yai–Huai Kha Khaeng World Heritage Site INTRODUCTION the many forests they inhabit, where they have a significant role in seed dispersal (Kemp, 1995). Large hornbills are also good indicators of forest condition and human disturbance because they require large tracts of contiguous primary forest with large trees for nesting sites and food resources, and they are targeted for hunting (Poonswad & Kemp, 1993). Meanwhile, hornbills also control populations of small seed predators such as insects and mice, thus helping to maintain forest structure and the productivity of forest ecosystems (Poonswad & Kemp, 1993). Hornbills are recognised as the largest bird species in the tropical forests of Asia and are classified in the family Bucerotidae of the order Bucerotiformes (Sibley & Monroe, 1990). Thirty species of hornbills are found in tropical Asia, with a further 22 species in Africa. Besides being a flagship species for conservation because of their striking appearance and intriguing nesting behaviour, with the female imprisoning herself inside a cavity in a tree (Poonswad et al., 1987), hornbills are among the primary frugivores of However, due to severe deterioration of tropical forests, eight out of 31 Asian hornbill species are threatened species (BirdLife International, 2001; Kinnaird & O’ Brien, 2007). The rufous-necked hornbill (RNH, Aceros nipalensis [Hodgson, 1829]) was originally distributed in the mountainous regions in Bhutan, north-eastern India, Myanmar, Thailand, Laos, Vietnam, southern Yunnan, and south-eastern Tibet in China. Its range has declined dramatically and it is now very rare across much of its historical range (BirdLife International, 2001). The RNH is classified as a vulnerable species at the global level and an endangered species in Thailand (Vidhidharm et al., 1995; Round, 2000; Sanguansombat, 2005) because its original habitat has disappeared from many areas, and it is now 1 Department of Forest Biology, Faculty of Forestry, Kasetsart University, Chatuchak, Bangkok 10900, Thailand; Email: fforyyt@ku.ac.th, Tel: +662 579 0176, Fax: +662 942 8107 (*corresponding author) 2 Wildlife Conservation Society Thailand Program, Pak Kret District, Nonthaburi Province 11120, Thailand 3 Department of National Parks, Wildlife and Plant Conservation, Bangkok 10900, Thailand 4 Department of Microbiology, Faculty of Science, Mahidol University, Bangkok 10400, Thailand 5 Naturalists & Nomads, Postnet Suite 38, Private Bag X19, Menlo Park, 0102, South Africa © National University of Singapore ISSN 2345-7600 (electronic) | ISSN 0217-2445 (print) 12 RAFFLES BULLETIN OF ZOOLOGY 2014 1), which cover approximately 6,488 km2 and designated as a UNESCO Natural World Heritage Site in 1991. Thung Yai Naresuan has been divided into East and West parts for effective administration. Thung Yai Naresuan and Huai Kha Khaeng formed the core area of the 18,727 km2 WEFCOM, which comprise of six wildlife sanctuaries and 11 national parks. The WEFCOM is the largest forest complex in Thailand and more importantly, the largest conservation area in mainland Southeast Asia. It is especially important for its high capacity to support large and endangered mammal species (Trisurat et al., 2010) and birds. At least six hornbill species have been observed in the World Heritage Site, including the great hornbill Buceros bicornis (Linnaeus, 1758), RNH, wreathed hornbill Rhyticeros undulatus (Shaw, 1811), plain-pouched hornbill Rhyticeros subruficollis (Blyth, 1843), Tickell’s brown hornbill Ptilolaemus tickelli (Blyth, 1855), and Oriental pied hornbill Anthracoceros albirostris (Shaw & Nodder, 1807). found mainly in the Western Forest Complex (WEFCOM), particularly in the hill evergreen forest of the Huai Kha Khaeng Wildlife Sanctuary at elevations above 1,000 m (Chimchome et al., 1998; Ouithavon et al., 2005). Previous studies of the RNH in Thailand were confined to describing its general habitat, breeding biology, and ecology (Lekagul & Round, 1991; Poonswad, 1993; Poonswad & Kemp, 1993; Kemp, 1995; Chimchome et al., 1998; Poonswad et al., 1999; Ouithavon et al., 2005). Recently, Tifong (2007) studied the home range, feeding, and roosting sites of one female and two male RNHs using radio-telemetry during the breeding and non-breeding seasons. The results indicated that aspects of fruit availability, vegetation type, slope and altitude within the area, and distance from water source affected their daily movements, feeding, and roosting sites. Much research has attempted to predict species distributions in the landscape. In addition, several approaches have been applied to species-distribution modelling using presenceabsence data and a geographic information system (GIS; Agresti, 1996; Corsi et al., 2000), with popular and frequently used methods being Generalised Linear Model (GLM) or Generalised Additive Model (GAM; Guisan & Zimmermann, 2000; Pearce & Ferrier, 2000; Guisan et al., 2002; Beck et al., 2005; Trisurat et al., 2010). A problem found when using logistic regression is that true absence data were often not available and in many cases, pseudo-absence data were created. However, the determined pseudo-absence may appear in the presence localities (Brotons et al., 2004; Phillips et al., 2006) because species that are listed as near extinction (endangered or critically endangered species) are often difficult to detect (Engler et al., 2004), or the survey may not have covered the habitat(s) of the species (Trisurat et al., 2010). Species occurrence data. Occurrence records of the RNH were gathered along 14 transects, each 9 km in length (Fig. 1). Transects 1–4 were located in Huai Kha Khaeng, 5–9 in Thung Yai Naresuan West, and 10–14 in Thung Yai Naresuan East. Forty-five point-count sampling spots at 200-m intervals were designated at each transect and trained personnel spent 10 minutes at each spot to record hornbill presence based on sightings and calls (Fig. 1). Field surveys along each transect were conducted twice in each breeding and non-breeding season between January 2004 to December 2008. The breeding season commences in January and lasts until May while the non-breeding season is between June and December (Chimchome et al., 1998; Ouithavon et al., 2005; Tifong, 2007). Additional records of hornbill occurrence were obtained from long-term monitoring studies of hornbills jointly conducted by the Thailand Hornbill Project in association with the Wildlife Conservation Society–Thailand Program. Recently, there has been progress in the development of models to predict species distribution using only presence data. Most research indicated that Maximum Entropy (MaxEnt) model was superior in performance (Phillips et al., 2006; Sérgio et al., 2007) to other models (e.g., Artificial Neural Networks [ANN; Pearson et al., 2002], Ecological Niche Factor Analysis [ENFA; Chefaoui et al., 2005; Peterson, 2006; Santos et al., 2006], Genetic Algorithm for Rule-set Production [GARP; Stockwell & Peters, 1999]). Phillips et al. (2006) reported that MaxEnt model performed well even for small sample sizes. Dataset. Environmental variables that may influence hornbill distribution at landscape levels were identified from The Thailand Hornbill Project conducted long-term studies of hornbills over 30 years, where only occurrence data was recorded and the MaxEnt model was adopted. The objectives of this research were: 1) to estimate the geographical distribution of the RNH in the Thung Yai–Huai Kha Khaeng World Heritage Site in both the breeding and non-breeding seasons; and 2) to determine the seasonal habitat use of the RNH. Fig. 1. Map of research locations showing the study areas where the point counts were conducted along14 line-transects, using a hand-held GPS to accurately measure the count locations or extent and position of evergreen forest where the RNH lives, in order to predict RNH distribution. MATERIAL AND METHODS Study areas. The research was conducted in the Thung Yai Naresuan and Huai Kha Khaeng wildlife sanctuaries (Fig. 13 Jinamoy et al.: Predictive distribution modelling for rufous-necked hornbill in Western Forest Complex and non-breeding season (11.16 km2) determined in Huai Kha Khaeng by Tifong (2007) as a minimum habitat size to eliminate smaller habitat areas (noise) for preparation of the final distribution maps. previous research (Kinnaird & O’ Brien, 2007; Tifong, 2007; Trisurat et al., 2013). These variables consisted of climate (minimum and maximum temperature, precipitation), land cover, topography (digital elevation model (DEM), slope), proximity to roads, proximity to streams, and proximity to ranger stations. A land cover map in 2008 and the locations of ranger stations were obtained from the Department of National Parks, Wildlife and Plant Conservation. Climatic variables were downloaded from the WORLDCLIM database (see http://www.worldclim.org/), while topographic feature layers were obtained from the WEFCOM GIS database at a scale of 1:50,000 (WEFCOM, 2004). All environmental variables were converted to a raster (grid) with a spatial resolution of 100 × 100 m and ArcGIS 9.2 software was used for all spatial analyses. RESULTS Species occurrence sampling points. In total, 330 presence data were obtained for the study areas, of which 169 were for breeding and 161 for non-breeding seasons (Table 1). In the breeding season, 127 presence records were used to generate distribution model, 42 for testing and, in the nonbreeding season, 121 and 40 records were used, respectively. In addition, 40 absence points were randomly chosen from all transects in the breeding season and 39 points were chosen for the non-breeding season. These data were used as independent absence data to validate the species distribution model accuracy. Species distribution model. We selected the MaxEnt model because it was developed specifically to model species distributions with presence-only data (Phillips et al., 2006), which is the only hornbill data available for Thailand. We ran the MaxEnt model using the following parameters: random test percentage = 25%, regularisation multiplier = 0.2, maximum iteration = 1,000, convergence threshold = 0.001, maximum number of background points = 10,000. From the surveys, we found the occurrence of the RNH in Huai Kha Khaeng over all four trails used for point counts (trails 1–4), but only along trail 4 in the non-breeding season. In Thung Yai Naresuan West, we observed RNH in both seasons over all five trails (trails 5–9). In Thung Yai Naresuan East, the RNH was detected on trails 10–12 in both seasons but was absent along trails 13 and 14 (Fig. 1). The output of the MaxEnt model was the continuous probability of occurrence of the species’ distribution model output (0.00–1.00). In this study, the more conservative cut-off value of “equal training sensitivity and specificity” was used to classify the continuous probability into a binary prediction (0–1) of presence-absence. If the probability value was equal to or greater than this threshold value, it was classified as present, if less, it was classified as absent. Presence-absence data were separated between the breeding and non-breeding seasons. Model performance and favorable habitat factors. Among the 10 possible environmental factors contributing to RNH presence in the breeding season, the MaxEnt model indicated that DEM was the highest contributor (33%), followed by land cover (29%), and minimum temperature (13%; see Table 2 for details). The contributions of proximity to ranger stations, villages, and roads were moderate (6%), while proximity to streams, annual precipitation, and maximum temperature contributed the least (<5%). However, in the non-breeding season, while the contributions of elevation and minimum temperature were still significant, the contribution of land use had dropped from 29% in the breeding season to 9%. In both seasons, the vegetation type used by the RNH was evergreen forest, which includes hill evergreen Previous studies normally used the area under the curve (AUC) of a receiver operating characteristic (ROC; Zweig & Campbell, 1993; Trisurat et al., 2011) to determine the accuracy of the predicted distribution models. However, the AUC has several shortcomings that were pointed out by Lobo et al. (2008) and Allouche et al. (2006). In the current research, the presence data were randomly partitioned into two datasets. Seventy percent of the data were used as training data to generate the distribution model and the remaining 25% were used as independent points to validate the model. In addition, the same amount of absence data from all transects was randomly selected and used as an independent test. Then, the contingency matrix was developed to calculate omission and commission errors and the kappa coefficient for assessing the accuracy of predicted distribution maps rather than the AUC (Sim & Wright, 2005; Allouche et al., 2006). The predicted distribution maps for the RNH were generated for both the breeding and non-breeding seasons. In addition, the extents of predicted presence-absence in both seasons and estimated seasonal shifts in distribution were compared and discussed (Fig. 2). It was noted that preliminary predicted distribution maps were generalised using the average home range size for a RNH male in the breeding season (6.19 km2) Fig. 2. Diagram showing the conceptual framework used to model the distribution of the rufous-necked hornbill in the Western Forest Complex (WEFCOM), Thailand. 14 RAFFLES BULLETIN OF ZOOLOGY 2014 Table 1. Rufous-necked hornbill occurrence data in the breeding and non-breeding seasons in the core areas of the Western Forest Complex. Protected Area Name Number of Occurrences Trail No. Huai Kha Khaeng Thung Yai Naresuan West Thung Yai Naresuan East Breeding Season Non-Breeding Season 31 7 6 0 18 20 16 23 2 18 28 0 0 28 8 2 10 15 19 27 5 18 13 16 0 0 169 161 1 2 3 4 5 6 7 8 9 10 11 12 13 Total Table 2. Contributions of environmental factors to rufous-necked hornbill distribution in the breeding and non-breeding seasons, in descending order of importance. Breeding Season Variables Non-breeding Season % Contribution Variables % Contribution Altitude (m) 33.4 Altitude (m) 47.6 Land use 28.8 Minimum temperature (°C) 17.5 Minimum temperature (°C) 12.9 Land use 8.5 Proximity to ranger station (m) 9.0 Proximity to road (m) 6.9 Proximity to village (m) 6.7 Proximity to ranger station (m) 5.3 Proximity to road (m) 6.0 Maximum temperature (°C) 4.3 Proximity to stream (m) 1.2 Annual precipitation (mm) 4.2 Annual precipitation 1.0 Proximity to village 2.6 Slope (%) 0.8 Proximity to stream (m) 2.1 Maximum temperature (°C) 0.3 Slope (%) 1.0 Total 100.0 AUC values 0.98 Total 100.0 0.96 due to the lower omission and commission errors for both presence and absence classes. forests (montane), degraded hill forests, and dry evergreen forests. Hill evergreen forest occurs from about 1,000–1,700 m above mean sea level (asl). No single tree species or family is dominant, but the families of Fagaceae, Myrtaceae, Lauraceae, Theceae, and Magnoliaceae are well represented in the mountainous WEFCOM. Dry evergreen forest generally occurs from 600–1,000 m asl and is sometimes found along streams at altitudes as low as 400 m asl. Common species at canopy level include Dipterocarpus spp. (e.g., D. costatus and D. turbinatus; WEFCOM, 2003, 2004). Species distribution maps. The presence-absence maps for the RNH in the breeding and non-breeding seasons were derived from equal training sensitivity, and specificity threshold values of 0.245 in the breeding season and 0.247 in the non-breeding season. The preliminary predicted distribution areas for the RNH in the breeding and nonbreeding seasons covered 476 and 675 km2 respectively; the areas of suitable habitat gain and habitat loss were 86 and 285 km2 respectively; and the seasonal shifts were 371 km2 (5.7%; Table 4). If the suitable habitats in both seasons were combined, their total area was approximately 13% of the study area. In addition, the overlapped suitable habitats of the RNH in both breeding and non-breeding seasons covered 390 km2 or 6.0% of the study area. After the preliminary habitat remnants that were smaller than a home range size (Tifong, 2007) were eliminated, the final distribution areas The accuracies of the predictive models were tested by the confusion matrix and the kappa coefficient. The contingency matrix shown in Table 3 indicates that the overall accuracy of prediction in the breeding season was 83.53% and in nonbreeding season was 87.34%. It was noted that the kappa coefficient value in the breeding season was approximately 67% while it was 74% in the non-breeding season. The higher kappa coefficient in the non-breeding season was 15 Jinamoy et al.: Predictive distribution modelling for rufous-necked hornbill in Western Forest Complex Table 3. Contingency matrix of predicted distributions in breeding and non-breeding seasons for rufous-necked hornbill. Breeding Season Observation Predicted Class Presence Absence Total Commission errors KHAT Non-breeding Season Observation Presence Absence Total Omission errors 34 6 40 8 37 45 42 43 85 19.05 13.95 15.00 17.78 83.53 0.67 Predicted Class Presence Absence Total Commission errors KHAT Presence Absence Total Omission errors 34 4 38 6 35 41 40 39 79 15.00 10.26 15.53 14.63 for the RNH in the breeding and non-breeding seasons were 421 and 595 km2 respectively. In addition, the habitat use for both seasons and the shifting habitats were 351 and 313 km2 respectively (Table 4). 87.34 0.74 for RNH male #15 in the non-breeding season was 1.8 times greater than in the breeding season (11.16 km2 versus 6.19 km2). The fact that the male hornbill can feed the female up to 73 times a day during the breeding season (Chimchome et al., 1998) limits the male’s range distance from nest tree. The distribution patterns appeared to be similar in both seasons, predicting the RNH’s absence from the southeastern corner of the study area with the main distribution being in the center of Thung Yai Naresuan West and along the western boundary of Huai Kha Khaeng adjoining Thung Yai Naresuan East (Fig. 3A, 3B). The largest patch was predicted in Thung Yai West, followed by Thung Yai Naresuan East and Huai Kha Khaeng. The predicted habitat sizes and patterns of distribution for RNH in Huai Kha Khaeng between the breeding and non-breeding seasons were quite stable, while in the other two sanctuaries these were quite different. For example, the predicted habitat sizes in the non-breeding season were greater than in the breeding season (30% for Thung Yai East and 35% for Thung Yai West) due to the fact that they are connected with the remaining evergreen forest, while suitable habitat patches in Huai Kha Khaeng are isolated and surrounded by deciduous forest (WEFCOM, 2004). Several previous studies (Poonswad & Kemp, 1993; Pattanavibool & Dearden, 2002; Sitompul et al., 2004; Kinnaird & O’Brien, 2007; Trisurat et al., 2013) indicated that landscape patterns and human disturbance are significant factors affecting hornbill distribution. Rufous-necked hornbill was formerly recorded in northern Thailand, such as in Doi Inthanon and Doi Suthep-Pui national parks, Chiang Mai Province (Poonswad & Kemp, 1993). Unfortunately, it is now locally extinct at these two parks owing to hunting and deforestation by the Hmong and Karen tribes, with a 2005–2007 survey led by the Thailand Hornbill Project finding only Oriental pied hornbills in some protected areas. These patterns support the finding from this research where, even though the study area is large, the distribution of the RNH was limited to specific areas of evergreen forest, particularly large patches of hill evergreen forest, and is also dependent on particular combinations of altitude, temperature, and available foods. In this research, hornbills were found on trails 10–12 in both seasons but not on trails 13 and 14 (Fig. 1) because the topography of trails 13 and 14 is generally flat and easily accessible by people living nearby (WEFCOM, 2004). DISCUSSION Species occurrence data for RNH. The species occurrence data recorded for RNH in the breeding season were more restricted to certain areas and/or less frequent than in the non-breeding season. We detected RNH on three trails in the breeding and four in the non-breeding season in Huai Kha Khaeng. Although we sighted the species over all five trails in both seasons in Thung Yai Naresuan West, its detection in the breeding season was lesser than the non-breeding season (e.g., two versus 18 records on trail nine, respectively). This phenomenon may be explained by the ecological niche concept, which is defined as a species only being able to establish populations in areas that match the ecological conditions that limit its native distribution (Peterson, 2003), which, in this case, produce seasonal movements in response to variations in availability of fruits and hence, home range size. Tifong (2007) found that the average home range size Species distribution model. We selected the MaxEnt model to predict the RNH distribution because its distribution prediction provides a powerful new tool that uses only presence data. MaxEnt estimates the most uniform distribution (maximum entropy) of the occurrence points of the RNH across the study areas, given the assumption that an expected value for each environmental predictive variable within this estimated distribution matches its empirical average (average values for a set of RNH occurrence data). In addition, it takes into consideration the interactions between environmental variables and seems to perform relatively well with small sample sizes of occurrence data (Phillips et al., 2006). Our research demonstrated that this approach can be applied to 16 243.59 (3.75%) 313.19 (4.83%) many taxa in Thailand that have only presence data, such as plant specimens (Trisurat et al., 2011). Species distributions for hornbills were conducted previously at either regional (Kinnaird & O’ Brien, 2007) or national levels (Trisurat et al., 2013), and analyses were undertaken at a resolution of 1 km. However, previous studies did not consider seasonal movement. This research used a pixel resolution of 100 m for the estimation of the distribution of the RNH—a scale appropriate to study the species’ distribution at the microhabitat level in response to variations in the habitat factors used in the model. In this study, altitude (m), slope, land use, temperature, precipitation, and the proximity to ranger stations, to villages and to roads were used to generate the distribution models for the RNH because they had been defined previously as potential environmental factors (Kinnaird & O’ Brien, 2007; Tifong, 2007; Trisurat et al., 2013). However, the results derived from the MaxEnt model indicated that altitude, land use, and minimum temperature were preferable habitat factors in the breeding season, and altitude and land use were identified as habitat preferences in the non-breeding season (Table 2). The contribution of altitude was very high for both seasons (48% in the non-breeding season and 33% in the breeding season). The current findings indicated that the elevation range of the RNH in the breeding season was 748–1,640 m and 493–1,640 m asl in the non-breeding and breeding seasons, respectively, and was concentrated in an altitudinal range between 1,000–1,400 m asl for both seasons. These findings were quite different from previous research (Chimchome et al., 1998; Ouithavon et al., 2005; Tifong, 2007), except the concentration areas, which indicated that the RNH normally occurs above 1,000 m. In addition, the response curves derived from MaxEnt indicated that evergreen forest was a preferable habitat for the RNH in both seasons. Chimchome et al. (1998) and Ouithavon et al. (2005) revealed that there were 13 and 15 plant species (mainly figs) in the dry evergreen and hill evergreen forests of the Huai Kha Khaeng Wildlife Sanctuary identified as fruit-foods for the RNH in the breeding season and these contributed approximately 80% of the total fruit in the RNH diet. Sitompul et al. (2004) and Kinnaird & Brien (2007) reported that figs have been recognised as “keystone species” for tropical frugivorous vertebrates, especially birds (hornbills) and primates. However, after the RNH female and chick emerge from the nest, they need more food, especially proteins for the growth of the fledged chicks (Poonswad, 1998; Ouithavon et al., 2005; Tifong, 2007), and they range much farther afield. Therefore, the contribution of land use factor on a particular hill evergreen forest was substantially reduced from 29% in the breeding season to 8% in the nonbreeding seasons (Table 4). Note: * Predicted presence derived from the MaxEnt model, ** Predicted presence greater than the home range of RNH#15 (breeding season = 6.19 km2; non-breeding season = 11.16 km2), (1) Number of patches in the breeding season = 990 and in the non-breeding season = 1,092. (2) Number of patches in the breeding season = 9 and in the non-breeding season = 6. 421.17 (6.49%)(2) 6,487.52 equal training sensitivity and specificity, and generalised using home range** 595.16 (9.17%)(2) 351.57 (5.42%) 69.60 (1.07%) 371.66 (5.73%) 285.20 (4.40%) 86.46 (1.33%) 675.37 (10.41%)(1) 476.63 (7.35%)(1) equal training sensitivity and specificity* Study areas Huai Kha Khaeng, Thung Yai Naresuan East and and West Wildlife Sanctuary Site 390.17 (6.01%) Habitat Loss Habitat Gain Uses in Both Seasons Non-Breeding Season Distribution Threshold (cut off value) Total Area (km2) Table 4. Comparison of data from presence-absence maps in the breeding and non-breeding seasons. Breeding Season Habitat Suitability Area in km2 (percent per study area) Seasonal Shifts RAFFLES BULLETIN OF ZOOLOGY 2014 Habitat suitability and seasonal shifts. The results of the MaxEnt models indicated that suitable habitats for the RNH after generalisation were located in three forest patches (Fig. 3). The first patch was consistent with Tifong (2007), who depicted that suitable habitats were located in intact and degraded evergreen forests, at altitudes 800–1,400 m asl, 17 Jinamoy et al.: Predictive distribution modelling for rufous-necked hornbill in Western Forest Complex mostly near streams. The second patch was situated along the junction west of Huai Kha Khaeng and east of Thung Yai Naresuan East. Even though it was located in a large patch of evergreen forest (238 km2), not all of this area was defined as suitable RNH habitat (Figure 3B), though other hornbill species that are less threatened than the RNH were found, such as the great hornbills, Tickell’s brown hornbills, and Oriental pied hornbills (BirdLife International, 2001). The third patch in the central zone of Thung Yai Naresuan West was the largest extent of suitable RNH habitat found in this study and thus vital in maintaining local RNH populations in this area (Bailey, 2007). These preliminary suitable habitats consisted of 990 sub-patches in the breeding season and 1,092 in the nonbreeding season. Nevertheless, the predicted areas may be overestimated since the RNH did not appear in many of the small patches of suitable habitats estimated as used from the MaxEnt model. This was especially so on mountain tops where patch sizes were smaller than the home range and distance to other suitable patches was greater than the average daily movement. In addition, Akçakaya (2005) suggested that the estimation of metapopulations for endangered species with a small population size and limited distribution should employ the smallest home-range size in the analyses. Therefore, this study used the average home range size for RNH male #15 in Huai Kha Khaeng to generalise the preliminary map. Hence, the number of sub-patches was substantially reduced in both seasons (Table 4). When suitable habitats in the breeding season were subtracted from habitats in the non-breeding season, the predicted suitable habitats for RNH before and after generalisation differed by 29%. This estimate conforms to the studies by Chimchome et al. (1998) and Tifong (2007), which found that during the breeding season a male RNH utilised a habitat that is 45% smaller than that in non-breeding season because of sufficient availability of fruits around the nest tree. In addition, this finding is consistent with Poonswad et al. (1987), who noticed that differences in habitats used by different hornbill species potentially correlated with differences and variations in the availability of fruit foods and the stages of the reproductive period. Conservation and protection measures. Poonswad et al. (1999) indicated that nest trees, home range, food source, and roosting sites during flocking are limiting factors for hornbills. The results of this research clearly indicated that in the non-breeding season, RNH used extended contiguous patches in the study area. On the other hand, in the breeding season, patches were clearly segregated, which resulted in three sub-populations (Fig. 3A) that can be best explained by the concept of “habitat fragmentation” (Akcakaya, 2005). The predicted habitat patches situated along the junction west of Huai Kha Khaeng and east of Thung Yai Naresuan East can be used by RNH as stepping stones for dispersion to adjacent and larger suitable habitats (Sitompul et al. 2004) located in Mae Wong National Park and Umphang Wildlife Sanctuary even though these patches are small. Poonswad (1993) indicated that the availability of nesting cavities of appropriate size may be the most important population-limiting factor for hornbills as hornbills are large birds and require large nesting cavities that exist naturally only in large trees. Most hornbills’ nesting holes occur in trees of the genus Dipterocarpus. Before Thung Yai Naresuan was declared as a wildlife sanctuary in 1957, indigenous hill tribes had settled in the area and converted the forest into agricultural lands. The GIS database (WEFCOM, 2004) and Fig. 3. Presence-absence binary models for RNH distributions after introducing equal sensitivity-specific threshold values to the continuous MaxEnt model based on the smallest home range size of the male RNH #15, as derived from Tifong (2007), during: a, the breeding season; b, the non-breeding season; and c, showing the combined habitat classification. 18 RAFFLES BULLETIN OF ZOOLOGY 2014 vegetation assessment (WEFCOM, 2003) reported that 30% of Thung Yai East and 19% of Thung Yai West are classified as degraded evergreen forest. Hence, big trees that are both potential nest trees and sites with suitable cavities were removed (Duengkae & Chimchome, 2007) and this became a critical factor for the RNH. Recent attempts to install artificial nests to assist hornbill conservation at the BudoSu-Ngai Padi National Park, Southern Thailand showed that the number of nests visited by hornbills steadily increased after installation (Pasuwan et al., 2011). This conservation effort and a proper artificial nesting design can be applied to the study area in order to increase the carrying capacity and enhance degraded habitats. LITERATURE CITED Agresti A (1996) An Introduction to Categorical Data Analysis. John Wiley and Sons Inc, New Jersey, 357 pp. Akçakaya HR (2005) RAMAS GIS: Linking Spatial Data with Population Viability Analysis. Version 5. Applied Biomathematics, New York, 201 pp. Allouche O, Tsoar A & Kadmon R (2006) Assessing the accuracy of species distribution models: prevalence, kappa and the true skill statistic (TSS). Journal of Applied Ecology, 43: 1223– 1232. Bailey S (2007) Increasing connectivity in fragmented landscapes: An investigation of evidence for biodiversity gain in woodlands. Forest Ecology and Management, 238: 7–23. Beck PSA, Kalmbach E, Joly D, Stien A & Nilsen L (2005) Modelling local distribution of an Arctic dwarf shrub indicates an important role for remote sensing of snow cover. Remote Sensing of Environment, 98: 110–121. BirdLife International (2001) Threatened Bird of Asia: The BirdLife International Red Data Book. BirdLife International, Cambridge, 3,038 pp. Broton L, Thuiller W, Araújo MB & Hirzel AH (2004) Presenceabsence versus presence-only modelling methods for predicting bird habitat suitability. Ecography, 27: 437–448. Chefaoui RM, Hortal J & Lobo JM (2005) Potential distribution modelling, niche characterization and conservation status assessment using GIS tools: a case study of Iberian Copris species. Biological Conservation, 122: 327–338. Chimchome V, Vidhidharn A, Simchareon S, Bumrungsri S & Poonswad P (1998) Comparative study of the breeding biology and ecology of two endangered hornbill, species in Huai Kha Khaeng Wildlife Sanctuary, Thailand. In: Poonswad P (ed.) The Asia Hornbill: Ecology and Conservation. Thai Studies in Biodiversity No. 2, Biodiversity Research and Training Program, Bangkok, pp. 111–136. Corsi F, Skidmore A & de Leeuw J (2000) Modelling species distribution with GIS. In: Boitani L & Fuller TK (eds.) Research Techniques in Animal Ecology. Columbia University Press, New York, pp. 389–434. Duengkae P & Chimchome V (2007) Recovery of avian diversity in an abandoned human settlement in Western Forests of Thailand. Kasetsart Journal (Natural Science), 41: 371–376. Engler R, Guisan A & Rechsteiner L (2004) An improved approach for predicting the distribution of rare and endangered species from occurrence and pseudo-absence data. Journal of Applied Ecology, 41: 263–274. Guisan A, Edwards TC Jr & Hastie T (2002) Generalized linear and generalized additive models in studies of species distributions: setting the scene. Ecological Modelling, 157: 89–100. Guisan A & Zimmermann NE (2000) Predictive habitat distribution models in ecology. Ecological Modelling, 135: 147–186. Kemp AC (1995) The Hornbill: Bucerotiformes. Oxford University Press, Oxford, 352 pp. Kinnaird MF & O’Brien TG (2007) The Ecology and Conservation of Asian Hornbills: Farmers of the Forest. The University of Chicago Press, Chicago, 352 pp. Lekagul B & Round PD (1991) A Guide to the Birds of Thailand. Saha Karn Bhaet Co Ltd, Bangkok, 467 pp. Lobo JM, Jiménez-Valverde A & Real R (2008) AUC: a misleading measure of the performance of predictive distribution models. Global Ecology and Biogeography, 17: 145–151. Ouithavon K, Poonswad P, Bhumbhakpan N & Laohajinda V (2005) Some characteristics of two sympatric hornbill species (Aves: Bucerotidae) and fruit availability during their breeding season in Huai Kha Khaeng Wildlife Sanctuaty, Thailand. In: Lum S & Poonswad P (eds.) Proceedings of the Third International Hornbill Workshop on The Ecology of Hornbill: In addition to enhancing the habitat connectivity between three suitable large patches and degraded habitats located within ranging distance, and the installation of artificial nests, protection measures are also essential to maintain RNH and other hornbill populations in the WEFCOM. Pattanavibool & Dearden (2002) and Poonswad & Kemp (1993) revealed that hornbill species were extinct from most protected areas in northern Thailand owing to hunting pressures and habitat destruction. Currently, seven and nine villages are located inside Thung Yai Naresuan East and Thung Yai Naresuan West respectively, with populations of 2,900 and 1,200 individuals (WFCOM, 2004). Field surveys also showed the evidence of poaching near these communities. The Department of National Parks, Wildlife and Plant Conservation, with the technical and financial support from the Wildlife Conservation Society-Thailand Program, has implemented a smart patrol system in the Thung Yai-Huai Kha Khaeng World Heritage Site since 2005. It aims to equip park rangers and managers with information technology to better protect the areas. It is expected that the conservation and protection measures will increase the carrying capacity of suitable habitats, enhance the connectivity between suitable habitats outside the current distribution ranges, and subsequently help to maintain the population viability of the RNH population in the WEFCOM. ACKNOWLEDGEMENTS For the success of this research, we thank the Department of National Parks, Wildlife and Plant Conservation of Thailand for granting permission to conduct this research in the Western Forest Complex. We greatly appreciate the support received from the staff at the Huai Kha Khaeng and Thung Yai (East and West) wildlife sanctuaries for providing information and accommodation and from the park rangers during the field work. We are also overwhelmingly grateful to the Thailand Hornbill Project for scholarship support through their project on the genetics of hornbills in fragmented-forest landscape and their population and habitat status in Thailand, funded by the National Center for Genetic Engineering and Biotechnology (BIOTEC), Thailand. We thank the Wildlife Conservation Society (WCS) Thailand for assistance with field surveys and materials. Special thanks are given to the Associate Editor, Frank Rheindt, and to two anonymous reviewers for raising important points and for providing constructive comments to improve the manuscript. 19 Jinamoy et al.: Predictive distribution modelling for rufous-necked hornbill in Western Forest Complex Reproduction and Population. Pimdee Karnpim Co., Ltd, Bangkok, pp. 131–140. Pasuwan C, Pattanakiat S, Navanugraha C, Chimchome V, Rattanarungsikul P, Thiensongrusamee P, Boonsriroj T & Poonswad P (2011) An assessment on artificial nest construction for hornbills in Budo Su-Ngai Padi National Park, Thailand. Raffles Bulletin of Zoology, Supplement 24: 85–93. Pattanavibool A & Dearden P (2002) Fragmentation and wildlife in montane evergreen forests, northern Thailand. Biological Conservation, 107: 155–164. Pearce J & Ferrier S (2000) Evaluating the predictive performance of habitat models developed using logistic regression. Ecological Modelling, 133: 225–245. Pearson RG, Dawson TP, Berry PM & Harrison PA (2002) SPECIES: a spatial evaluation of climate impact on the envelope of species Ecological Modelling, 154: 289–300. Peterson AT (2003) Predicting the geography of species invasions via ecological niche modeling. The Quarterly Review of Biology, 78: 419–433. Peterson AT (2006) Uses and requirements of ecological niche models and related distributional models. Biodiversity Informatics, 3: 59–72. Phillips SJ, Anderson RP & Schapired RE (2006) Maximum entropy modelling of species geographic distributions. Ecological Modelling, 190: 231–259. Poonswad P (1993) Comparative Ecology of Sympatric Hornbills (Bucerotidae) in Thailand. Unpublished DSc thesis. Osaka City University, Osaka, 325 pp. Poonswad P (1998) The Asian Hornbills: Ecology and Conservation. Biodiversity Research and Training Program, National Center for Genetic Engineering and Biotechnology with support from the Thailand Research Fund in collaboration with Science Society of Thailand, Bangkok, 336 pp. Poonswad P & Kemp AC (1993) Manual to the Conservation of Asian Hornbills. Sirvatana Interprint Co., Ltd., Bangkok, 544 pp. Poonswad P, Tsuji A & Ngampongsai C (1987) A comparative study on breeding biology of four sympatric hornbill species (Family Bucerotidae) in Thailand with implications for breeding in captivity. In: International Foundation for the Conservation of Birds (ed.) Proceedings of the Jean Delacour/IFCB Symposium on Breeding Birds in Captivity. International Foundation for the Conservation of Birds, Los Angeles, pp. 250–315. Poonswad P, Chimchom V, Plongmai K & Chuilua P (1999) Ecological factors influencing the reproduction of Asian hornbills. In: Adams N & Slotow R (eds.) Proceedings of the 22nd International Ornithology Congress. University of Natal, Durban, pp. 1740–1755. Round PD (2000) Resident Forest Birds in Thailand: Their Status and Conservation. International Council for Bird Preservation, Cambridge, 211 pp. Sanguansombat W (2005) Thailand Red Data: Birds. Office of Natural Resources and Environmental Policy and Planning, Bangkok, 158 pp. Santos X, Brito JC, Sillero N, Pleguezuelos JM, Llorente GA, Fahd S & Parellada X (2006) Inferring habitat-suitability areas with ecological modelling techniques and GIS: A contribution to assess the conservation status of Vipera latastei. Biological Conservation, 130: 416–425. Sérgio C, Figueira R, Draper D, Menezes R & Sousa AJ (2007) Modelling bryophyte distribution based on ecological information for extent of occurrence assessment. Biological Conservation, 135: 341–351. Sibley CS & Monroe BE (1990) Distribution and Taxonomy of Birds of the World. Yale University Press, New Haven, 1360 pp. Sim J &Wright CC (2005) The kappa statistic on reality studies: uses, interpretation and sample size requirements. Journal of the American Physical Therapy, 85: 257–268. Sitompul AF, Kinnaird MF & O’Brien TG (2004) Size matters: the effects of forest fragmentation and resource availability on the endemic Sumba hornbill Aceros everetti. Bird Conservation International, 14: 23–37. Stockwell D & Peters D (1999) The GARP modelling system: problems and solutions to automated spatial prediction. International Journal of Geographical Information Science, 13: 143–158. Tifong J, Chimchome V, Poonswad P & Pattnavibool A (2007) Home range and habitat use of rufous-necked hornbill (Aceros nipalensis) by radio tracking in Huai Kha Khaeng Wildlife Sanctuary, Uthai Thani Province. Thai Journal of Forestry, 26: 28–39. Trisurat Y, Pattanavibool A, Gale GA & Reed D (2010) Improving the viability of large mammal populations using landscape indices for conservation planning. Wildlife Research, 36: 401–412. Trisurat Y, Shrestha R & Kjelgren R (2011) Plant species vulnerability to climate change in Peninsular Thailand. Applied Geography, 31: 1106–1114. Trisurat Y, Chimchome V, Pattanavibool A, Jinamoy S, Thongaree S, Kanchanasaka B, Simcharoen S, Sribuarod S, Mahannop N & Poonswad P (2013) Assessing distribution and status of hornbills in Thailand. Oryx, 47: 441–450. Vidhidharm A, Nuttaro S, Chimchome P, Wangdilok S, Chimchome V, Suraparbmaitri S, Eiadthong W, Kanjanavanit R, Nakkuntod S, Jirawatkari N, Tunhikarn S & Poonswad P (1995) Current distribution and status of hornbill in Thailand. Wildlife Journal of Thailand, 4: 1–11. Western Forest Complex Project (WEFCOM) (2003) The Vegetation and Flora of the Western Forest Complex: Using Rapid Ecological Assessment and Vegetation Description in the WEFCOM area. The Western Forest Complex Ecosystem Management Project, National Park, Wildlife and Plant Conservation Department, Bangkok. Western Forest Complex Project (WEFCOM) (2004) GIS Database and its Applications for Ecosystem Management. The Western Forest Complex Ecosystem Management Project, National Park, Wildlife and Plant Conservation Department, Bangkok. Zweig MH & Campbell G (1993) Receiver-Operating Characteristic (ROC) plots: a fundamental evaluation tool clinical medicine. Clinical Chemistry, 39: 561–577. 20