Integrating earth observation and machine learning for forest fire risk assessment in Rasuwa district, Nepal

The complex terrain and increasing forest fire incidents in recent years make high-mountain environments vulnerable to forest fires, posing a threat to ecological and economic well-being. However, research related to forest fires remains limited in the mountain region. The study aimed to assess the forest fire risk by integrating eleven predictive layers (biophysical, topographical, climatic, and anthropogenic factors) and a Random Forest (RF) model in Rasuwa, a mountainous district of Nepal. However, the precipitation factor was excluded due to multicollinearity concerns. The model was validated using Area Under the Curve (AUC) and True Skill Statistics (TSS). With an AUC of 0.9 and TSS of 0.67, the model shows strong predictive performance and high reliability of the generated fire risk map. The most significant factors in the model’s ability to predict the forest fire risk were elevation, temperature, Normalized Difference Vegetation Index (NDVI), and proximity to settlements. The study shows that 29.66% of Rasuwa district’s forest is at fire risk, with the highest risk in Kalika Rural Municipality (53.7%). In the study area, approximately 30% of fire incidents occurred at elevations above 3000 m. The results can assist the concerned stakeholders and authorities in developing effective forest management plans by prioritizing identified high-risk areas for early warning systems and fuel management.