A Spatiotemporal Framework for Wildfire Risk Prediction and Management in Nigeria’s Savanna Ecosystems

Wildfires are a pervasive ecological process and recurring hazard in Nigeria's savanna ecosystems, yet predictive risk modelling remains methodologically immature relative to the scale of the problem. This study develops a spatiotemporal framework for wildfire risk prediction and management in Nigerian savanna ecosystems across the North Central and North East geopolitical zones. The framework integrates remote sensing-derived fire weather indices, fuel load estimates from MODIS-based vegetation biomass products, historical fire occurrence data from MODIS Active Fire and Burned Area products spanning 2005 to 2023, and infrastructure exposure data in a machine learning prediction model. A Gradient Boosted Trees model trained on the integrated dataset produces probabilistic fire risk maps at 500-metre spatial resolution and 16-day temporal frequency. The model achieves an area under the ROC curve of 0.91 for fire occurrence prediction in the validation period. Temporal trend analysis reveals a statistically significant increase in fire frequency and burned area in the Guinea Savanna zone over the 18-year record, inconsistent with rainfall trends but correlated with increasing dry season biomass accumulation. Keywords: wildfire risk, savanna, remote sensing, spatiotemporal modelling, Nigeria