Ensembling machine learning models to identify forest fire-susceptible zones in Northeast India

Abstract

Forest fires pose significant challenges by disrupting ecological balance, impacting socio-economic harmony, and raising global concerns. North-East India (NEI) experiences high incidences of forest fires, making it crucial to implement suitable management measures considering the driving forces influencing fire likelihood. This study aims to identify forest fire susceptibility zones in NEI by using five machine-learning modeling approaches, Boosted Regression Tree (BRT), Random Forest (RF), Support Vector Machine (SVM), Classification and Regression Tree (CART), and Multivariate Adaptive Regression Splines (MARS), and an ensemble method. Forest fire data from the SNPP – VIIRS sensor (2018–2019) were rectified for spatial autocorrelation. Thirty-two responsive predictor variables related to topographic, climatic, biophysical, and anthropogenic factors were used as model inputs and multicollinearity analysis was performed to eliminate highly correlated predictors. Results indicate that the southern and southeastern regions of NEI, characterized by ample solar radiation, enhanced vegetation index, high human population density, and jhum cultivation, contribute significantly to higher susceptibility to forest fires. The Random Forest model performs best among the models employed, achieving an AUC value of 0.87. The ensemble susceptibility map, binarized based on AUC weighting, covers 29.54% of the total geographic area and 44.42% of the forested area of NEI. The vulnerability levels vary among states, with Mizoram showing the highest susceptibility at 89.27% and Sikkim exhibiting the lowest vulnerability at only 0.49% of their respective geographic areas. This map provides valuable insights for implementing effective forest fire management plans in the region. Moreover, the methodology utilized in this study, which incorporates satellite imagery, GIS techniques, and improved modeling techniques, can be replicated in any geographical region worldwide to facilitate effective forest fire management at a regional to large scale.