Statistical Modeling of Wildfire Occurrence Based on Geomorphometry in La Amistad International Park’s Buffer Zone

Resumen

This study investigates the spatial distribution and environmental drivers of wildfire occurrence in a humid tropical protected area in the Southern Pacific Costa Rica using geomorphometric and statistical analyses. We applied multiple statistical models, including Poisson, Negative Binomial, Zero-Inflated Poisson, and Zero-Inflated Negative Binomial, to evaluate their performance in modeling wildfire count data. Model diagnostics, including Akaike Information Criterion, Bayesian Information Criterion, and Validated Global Deviance, indicated that the Zero-inflated Poisson model provided the best fit. Covariates such as land surface temperature, slope, topographic wetness index, and land use (savanna areas) were significant predictors of wildfire counts, while vegetation exposure showed limited significance. Spatial predictions revealed higher wildfire probabilities in specific areas, enabling the identification of regions with more than 50% probability of wildfire occurrence. These findings enhance the understanding of wildfire dynamics in tropical ecosystems and provide insights for targeted wildfire prevention and management strategies in protected areas.