Forest Fire Susceptibility Zonation using dNBR and Machine Learning models: A case study at the Similipal Biosphere Reserve, Odisha, India

Abstract Forests play a pivotal role in maintaining environmental equilibrium, chiefly due to their biodiversity. This biodiversity is instrumental in atmospheric purification and oxygen production. Nowadays forest fires are an exciting phenomenon, identification of forest fire susceptible (FFS) areas is necessary for forest fire mitigation and management. This study delves into forest fire trends and susceptibility in the Similipal Biosphere Reserve (SBR) over the period of 2012–2023. Utilizing four machine learning models such as Extreme Gradient Boosting Tree (XGBTree), AdaBag, Random Forest (RF), and Gradient Boosting Machine (GBM). Forest fire inventory was prepared using the Delta Normalized Burn Ratio (dNBR) index. Incorporating 19 conditioning factors and rigorous testing for collinearity, FFS maps were generated, and finally, model performance was evaluated using ROC-AUC, MAE, MSE, and RMSE methods. From the results, it was observed that, overall, about 33.62% of the study area exhibited high to very high susceptibility to forest fires. RF exhibiting the highest accuracy (AUC = 0.85). Analysis of temporal patterns highlighted a peak in fire incidents in 2021, particularly notable in the Buffer Zone. Furthermore, a significant majority (94.72%) of fire incidents occurred during March and April. These findings serve as valuable insights for policymakers and organizations involved in forest fire management, underscoring the importance of targeted strategies for high-risk areas.