An uncertainty-aware spatial overlay framework for prioritizing conservation planning

This study introduces a groundbreaking framework for conservation planning in data-scarce landscapes, focusing on spatial uncertainty and ensemble inter-model disagreement. The framework, developed for the endangered Rheum ribes L. plant species in northeastern Iran, integrates multiple machine learning models to identify priority conservation zones.

Four distinct algorithms (RF, GBM, ANN, DNN) were calibrated using various environmental predictors and combined through a TSS-weighted ensemble, achieving a high predictive accuracy (AUC = 0.94, TSS = 0.75). The researchers discovered that tree-based algorithms demonstrated more spatial alignment with the ensemble compared to neural network models, indicating differences in niche representation.

By quantifying spatial uncertainty, the framework revealed that high-suitability habitats are consistently associated with low inter-model disagreement. This dual-filter approach delineated 19.2% of the study area as priority habitat zones, offering a reproducible and transferable method for biodiversity-informed conservation planning across ecosystems and taxa.

The authors express gratitude to Dr. Hassan Yeganeh, the Department of Environment of North Khorasan Province, and the local rangers for their support.