Prey availability limits modeled Sumatran tiger habitat in Gunung Leuser

Prey availability was the strongest limiting influence on modeled habitat suitability for Sumatran tigers in Gunung Leuser National Park, Indonesia, according to a two-stage species distribution study that combined records of five prey species with environmental data. The researchers modeled suitable habitat for sambar deer, wild boar, lesser mouse-deer, barking deer and Sumatran serow before using the resulting prey predictions to estimate tiger distribution. Their final ensemble model identified 80.4% of the park as potentially suitable for tigers, but only 40.3% overlapped with the modeled distribution of the three prey species retained for the tiger analysis. The findings indicate that habitat conditions alone may substantially overstate the park’s capacity to support tigers, although the predictions remain subject to sampling and modeling uncertainty.

The study used 6,386 tiger presence records collected through regular monitoring in 2021–2023 and additional surveys in June and August 2024. The records came from SMART patrols, camera traps, direct observations and indirect signs including feces, urine, hair, footprints, food remnants, claw marks, scratches and ground scrapes. They covered Gunung Leuser National Park, an 8,302.69-square-kilometer landscape divided between Aceh and North Sumatra provinces, as well as a 35-kilometer buffer outside the park boundary. The buffer was included to capture nearby settlements, roads, cropland, plantations, rivers and other environmental conditions that could influence habitat selection near the park edge.

Prey records came from the same monitoring and field programs and were supplemented with filtered occurrence data from the Global Biodiversity Information Facility. To improve coordinate reliability, the researchers retained records from 2014–2024, assessed the supporting evidence and spatial accuracy, and applied spatial thinning. The final datasets contained 12 Sumatran serow records, 61 barking deer records, 20 sambar deer records, 18 wild boar records and 17 lesser mouse-deer records after filtering and thinning. These occurrence records were combined with elevation, 19 bioclimatic variables, distance to rivers, roads and settlements, plantation and cropland distance, soil type, land cover, and satellite-derived vegetation and water indices. The predictors were processed on a common 30-arc-second grid, equivalent to approximately 1 square kilometer, to provide a uniform spatial extent and resolution for modeling.

Before constructing the tiger model, the researchers fitted separate distribution models for each prey species using maximum entropy modeling, random forest and an ensemble approach combining the two algorithms. Pseudo-absences were generated with environmental, geographic and k-means constraints within a 10-kilometer radius. The best-performing model was selected for each prey species using the area under the receiver operating characteristic curve, Kappa, the true skill statistic, Jaccard and Sørensen indices. Random forest performed best for four prey species, while maximum entropy was selected for sambar deer. The ensemble did not outperform the best single algorithm for any prey species. Most discrimination scores exceeded 0.8, although the selected Sumatran serow model ranged from 0.667 to 0.876.

The prey models identified different leading environmental predictors. Annual temperature range was the most important variable for Sumatran serow, wild boar and lesser mouse-deer, while seasonal rainfall led for barking deer and land cover ranked first for sambar deer. The response curves showed an inverse association between annual temperature range and the predicted presence of all five prey species: locations with greater annual temperature variation were less favorable. After the researchers tested the prey predictors against other environmental variables for multicollinearity, they retained sambar deer, wild boar and lesser mouse-deer for the tiger model. The other prey species were excluded because their distributions were too highly correlated with other predictors to add independent information.

For the tiger analysis, the three modeled prey distributions became variables alongside the environmental layers. Maximum entropy and random forest each produced discrimination scores above 0.9, yet their spatial estimates differed sharply. The maximum entropy model classified 2,042.83 square kilometers as suitable, while random forest identified 6,754.25 square kilometers. The combined ensemble covered 6,672.11 square kilometers and outperformed both single algorithms across all five discrimination measures. Its sensitivity was 0.986, meaning that the model correctly identified 98.6% of observed tiger presences in the evaluation data. The researchers therefore used the ensemble for the park-wide distribution analysis, while drawing variable-importance and response interpretations from the underlying maximum entropy and random forest models because the software did not provide importance values for the ensemble itself.

Prey distributions and forest cover were the most influential predictors in both underlying tiger models. As the predicted probability of sambar deer, wild boar and lesser mouse-deer presence increased, predicted tiger presence also increased. Primary and secondary forests were the most favorable land-cover classes, while areas farther from settlements were also associated with greater tiger suitability. The researchers interpreted these associations as evidence that prey availability may constrain tiger distribution even where temperature, topography, vegetation and land cover otherwise appear favorable. They also proposed that climatic variables may influence tigers indirectly by shaping the distribution and abundance of their prey rather than acting as the dominant direct control. The study therefore links modeled prey availability to habitat suitability; it does not directly measure prey abundance or establish the causes of every distribution pattern.

The overlap analysis exposed a substantial mismatch between potential tiger habitat and the areas in which modeled prey were available. The selected prey species collectively occupied 48.6% of Gunung Leuser National Park in the suitability maps, while the tiger ensemble covered 80.4% of the park, approximately twice that area. Only 40.3% was classified as suitable for both tigers and the retained prey community. The authors cautioned that these percentages represent model predictions rather than direct population-density estimates. Nevertheless, the spatial mismatch suggests that the park’s effective carrying capacity for tigers could be lower than a tiger-only suitability analysis implies, particularly if prey populations are compressed into a smaller part of the landscape by hunting or other human pressures. The park’s tiger population was previously estimated at around 140 individuals, but the study did not estimate how many tigers its prey base can support.

Several limitations constrain the precision of the findings. Prey and tiger records remained unevenly distributed even after spatial thinning, and under-surveyed areas may therefore contain unrecorded occurrences. The prey models added another layer of uncertainty because each generated an estimated distribution rather than a measured prey abundance, and static suitability maps cannot fully reproduce predator-prey dynamics. Coarsening the environmental predictors to approximately 1 square kilometer was necessary to match the resolution of the WorldClim bioclimatic data, but it may conceal fine-scale habitat features relevant to a wide-ranging predator. The study also used climatic and land-cover data spanning different periods, including 1990–2000 bioclimatic averages and more recent satellite and land-cover layers. The researchers consequently advised cautious use of the maps, particularly in areas where prey data are limited.

For conservation planning, the results point toward maintaining both forest quality and prey populations rather than treating protected forest cover as sufficient on its own. The authors proposed prey restoration, stronger anti-poaching patrol coverage and systematic prey-density surveys as management priorities. Systematic surveys would help refine estimates of prey-mediated carrying capacity, especially in areas modeled as suitable for tigers but lacking modeled prey. The study also highlighted the need for long-term camera-trap and remote-sensing monitoring to track predators and their prey through time and identify changes associated with temperature, rainfall and land cover. These are management implications drawn from the study rather than direct evidence that any intervention increased tiger abundance. As the authors concluded, even apparently extensive habitat cannot be expected to sustain a viable tiger population if its prey base is depleted.

Subject of Research: Modeling prey and environmental influences on Sumatran tiger habitat suitability in Gunung Leuser National Park

Article Title: Tracking Sumatran Tiger ( Panthera tigris sumatrae Pocock, 1929) distribution in Gunung Leuser National Park: The influence of prey presence and environmental variables on habitat selection

Article References: Saputra, M. H., Widyastuti, S., Kuswanda, W., Nugraha, R. T. P., Helmanto, H., Anugrah, N., Rahmat, U. M., Napitu, R. S., Adnan, A., Iskandarrudin, Purwoko, A., & Patana, P. (2025). Tracking Sumatran Tiger (Panthera tigris sumatrae Pocock, 1929) distribution in Gunung Leuser National Park: The influence of prey presence and environmental variables on habitat selection. Nature Conservation, 59, 43-64. https://doi.org/10.3897/natureconservation.59.144262

Image Credits: AI Generated

DOI: 10.3897/natureconservation.59.144262

Keywords: Sumatran tiger, Gunung Leuser National Park, prey availability, habitat suitability, species distribution modeling, random forest, maximum entropy, ensemble model, wildlife conservation, Indonesia, Tracking, Sumatran