The research team, working within the Central Visayas region of the Philippines, faced a familiar dilemma in tropical conservation: the species most in need of protection are often the ones about which least is known. Many of the threatened endemics in Rajah Sikatuna have sparse, scattered occurrence records, making it difficult to decide which forest patches deserve the strictest protection. To overcome this, the researchers turned to Species Distribution Modelling, a technique that links known locations of a species with environmental variables to predict where suitable habitat may exist across the wider landscape. Using the MaxEnt algorithm, they modelled habitat suitability for thirteen threatened species spanning birds, mammals, amphibians, reptiles and trees.

The environmental backbone of the models came from two complementary sources. Nineteen bioclimatic variables, describing temperature and precipitation patterns, were interpolated from observations provided by the Philippine Atmospheric, Geophysical and Atmospheric Services Administration using inverse distance weighting, a deterministic method chosen deliberately because only three local sampling points were available, too few to fit the semi-variogram models required by geostatistical approaches such as kriging. Terrain variables, including elevation, slope, aspect and the Topographic Ruggedness Index, were derived from a 12.5-metre resolution digital elevation model supplied by the ALOS satellite mission, fine enough to capture the rugged karst topography that shapes microclimates and species movements across the reserve.

Species occurrence records were compiled from the Global Biodiversity Information Facility, supplemented by local biodiversity monitoring reports from the Department of Environment and Natural Resources and field surveys conducted by the Central Visayas Agriculture, Aquatic and Natural Resources Research and Development Consortium. After cleaning for duplicate entries and erroneous coordinates, and applying spatial thinning to reduce sampling bias, the final dataset contained just eighty-three spatially independent records across all thirteen species. To guard against overfitting with such limited data, the researchers reduced an initial set of twenty-three environmental predictors to five statistically independent ones, all with variance inflation factors below four, well under the conservative threshold of five: Annual Mean Temperature, Annual Precipitation, Elevation, Slope and Terrain Ruggedness Index.

The resulting models performed remarkably well for a data-poor tropical landscape. The mean Area Under the Curve, a standard measure of a model’s ability to discriminate between suitable and unsuitable habitat, reached 0.905, with individual species scores ranging from 0.860 for the Mindanao cuckooshrike Edolisoma mindanense to 0.950 for the critically endangered Mindanao bleeding-heart Gallicolumba crinigera. The mean True Skill Statistic, which measures accuracy independently of how common a species is, stood at 0.717, comfortably above the 0.60 benchmark generally regarded as indicating substantial model performance. Validation used five-fold cross-validation for species with fewer than ten records and a leave-one-out jackknife procedure for the most data-limited taxa, ensuring that performance estimates remained reliable across varying sample sizes.

One of the study’s most technically careful decisions concerned how continuous probability surfaces were converted into binary presence-absence maps. The researchers initially considered the minimum training presence threshold, which minimises omission errors, but a sensitivity analysis revealed that this approach inflated predicted presence by up to seventy-four percent in certain taxa, effectively encoding spatial noise into the conservation maps. They instead adopted the maximum sum of sensitivity and specificity threshold, widely recognised as one of the most robust options for conservation planning because it optimises the balance between failing to detect suitable habitat and wrongly classifying unsuitable areas as suitable.

The maps that emerged revealed a landscape of stark ecological contrasts. Some species, including the Mindanao bleeding-heart, the dipterocarp tree Dipterocarpus grandiflorus and the hardwood Shorea guiso, showed highly restricted and fragmented suitability patterns, concentrated in the western and northern sectors where intact forest interiors persist. Others, such as the sailfin lizard Hydrosaurus pustulatus, the long-tailed macaque Macaca fascicularis and the winchell’s kingfisher Todiramphus winchelli, occupied narrow localised hotspots vulnerable to site-specific disturbance. In contrast, the amphibian Platymantis guentheri, the cuckooshrike and the tree Voacanga megacarpa displayed broader, more continuous suitability gradients, suggesting greater ecological plasticity and resilience to moderate landscape-level change.

The environmental drivers behind these patterns were equally species-specific. Annual Mean Temperature contributed 29.2 percent and Elevation 28.2 percent across the ensemble, but individual models told sharper stories: temperature dominated the model for the Mindanao bleeding-heart at 62.3 percent, while Annual Precipitation drove sixty percent of the model for the karst-dwelling frog Platymantis guentheri, reflecting the physiological dependence of amphibians on moisture regimes. Elevation was the primary predictor for dipterocarp trees, contributing 54.9 percent for Pitta steerii and 45.2 percent for Dipterocarpus grandiflorus, while terrain ruggedness and slope shaped the distributions of the macaque and the bent-winged bat Miniopterus schreibersii respectively. The researchers are careful to note that these climatic drivers, interpolated from only three sampling points, should be read as proxy gradients reflecting broad landscape transitions rather than precise mechanistic limits on where species can live.

Aggregating the individual binary maps produced a species richness map that identified the northern and north-western sections of the reserve, with richness classes of ten to twelve species, as the primary biodiversity hotspots, corresponding to areas of relatively intact forest cover and topographic complexity. The southern and eastern portions, classified in the lowest richness classes of zero to three species, likely reflect degraded or fragmented habitats offering fewer niches and fewer resources. The frequency distribution of richness across the landscape was multimodal, with a primary peak at three species per pixel and a secondary peak at five, confirming that the landscape is dominated by moderate biodiversity while high-richness areas remain spatially rare and restricted.

The final and arguably most consequential step was deciding where to draw the line between protected core zones and the wider landscape. The researchers tested five statistical thresholds against the richness distribution, measuring how many known species occurrences each captured. The most conservative options, isolating only the top five percent of the landscape, captured just twenty-nine percent of occurrences, while the ninetieth and ninety-fifth percentiles pinpointed rare hotspots but excluded moderately rich areas that may function as corridors or buffers. The winning candidate was the inflection point of the Cumulative Richness Frequency Distribution, at approximately six species per cell, which captured ninety percent of all species occurrences while maintaining a spatially inclusive footprint. Because Rajah Sikatuna already holds formal protected status, the authors argue that thresholds must balance ecological significance with management feasibility, and the inflection-point approach achieves exactly that, protecting both primary biodiversity nodes and the functional buffer habitats required for long-term species persistence. The team frames its outputs not as immutable boundaries for legal zoning but as spatial hypotheses and decision-support tools, intended to prioritise future ground-truthing, guide denser sensor deployment and steer restoration efforts toward the degraded southern slopes where recovery could reconnect fragmented habitat islands.

Subject of Research: Species distribution modelling and conservation prioritisation of threatened species in a Philippine protected landscape

Article Title: Identification of potential suitable habitats and high conservation value areas for threatened species in Rajah Sikatuna Protected Landscape, Philippines

Article References: Aureo, W., Tiburan Jr., C., Calzeta, E., Lit Jr., I., Yan, Y., Taylor, D. M., & Ancog, R. (2026). Identification of potential suitable habitats and high conservation value areas for threatened species in Rajah Sikatuna Protected Landscape, Philippines. One Ecosystem, 11, Article e184595. https://doi.org/10.3897/oneeco.11.e184595

Image Credits: AI Generated

DOI: 10.3897/oneeco.11.e184595

Keywords: species distribution modelling, MaxEnt, Rajah Sikatuna Protected Landscape, Philippines, threatened species, High Conservation Value Areas, species richness, karst forest, habitat suitability, conservation planning, biodiversity hotspot, endemic species