Advanced modeling of forest fire susceptibility and sensitivity analysis using hyperparameter-tuned deep learning techniques in the Rajouri district, Jammu and Kashmir

Abstract

Forest resources are crucial for sustaining the global population, regulating climate services, and maintaining overall ecological balance. However, forest fires are causing a significant loss of forest cover worldwide. In this context, advanced deep learning techniques, which are novel to date, have been utilized to prepare forest fire susceptibility mapping. The present study aimed to predict forest fire susceptibility using three hyper-tuned techniques: deep neural network (DNN), elman neural network (ENN), and convolutional neural network (CNN). To identify the importance of influencing factors, sensitivity analysis was conducted using the DNN. The forest fire susceptibility map (FFSM) was categorized into five susceptibility zones: very high, high, moderate, low, and very low. Results indicated that the southern and southeastern parts of the study area are most prone to forest fires. The proportion of high susceptibility zone in the study area was found to be 34% for DNN, 37% for ENN, and 30% for CNN. Among all the models, DNN outperformed the others, achieving the highest accuracy of 0.8925, compared to ENN (0.8825) and CNN (0.87). Sensitivity analysis further revealed that evapotranspiration, temperature, land surface temperature (LST), distance to roads, aridity, and elevation were the most influential factors contributing to forest fires in the region. This study demonstrates an advanced and globally relevant approach to forest fire susceptibility analysis. The findings may be crucial for stakeholders and policymakers to make informed decisions regarding effective forest fire management and to protect vulnerable communities from unexpected losses.