Advancing forest fire prediction: a multi-layer stacking ensemble model approach

A reliable forest fire probability map is vital for disaster management and an essential resource in land use planning. This study evaluates the efficacy of the multi-layer stacking ensemble Machine Learning (ML) method for forest fire susceptibility mapping, presenting a comparative case study within the Malakand division of Pakistan. Our extensive literature review shows that the present ML model has never been used in Pakistan’s forest fire scenarios. We employed several benchmark models for comparative evaluation, including Random Forest (RF), Logistic Regression (LR), Support Vector Machine (SVM), Extreme Gradient Boosting (XGBoost), and K-Nearest Neighbor (KNN). A comprehensive fire inventory database was constructed, including satellite and ground hotspot data and relevant influencing factors. The fire probability indices from the six models were analyzed and validated using accuracy, area under the curve (AUC), precision, recall, and F1 score evaluation metrics. According to the Performance Evaluation Outcomes, the multi-layer stacking ensemble model provides the best outcomes in terms of accuracy 96.24%, AUC 99.43%, precision 97.81%, recall 94.59%, and F1 96.17% metrics. These results underscore the model’s potential as an effective new forest fire Probability mapping tool. Given its evidenced effectiveness, local forestry authorities in the Malakand division should consider its application for enhanced forestry conservation management and fire prevention strategies.

Text 123961.pdf - Published Version Restricted to Repository staff only Download (3MB) Official URL or Download Paper: https://link.springer.com/article/10.1007/s12145-0...

Actions (login required)

View Item