Rapid Assessment Model for Fire-Induced Forest Damage: A Case Study in Pelalawan Regency, Riau

Abstract

This study examines and develops a model for detecting forest damage due to fires in Riau Province, particularly in Pelalawan Regency, during the 2020-2021 period. This study aims to develop a machine learning decision tree model that can rapidly assess forest and land damage caused by fires, based on multi-spectral satellite imagery, geophysical variables, and forest and land fire incident points. The machine learning decision tree algorithm used in this study is based on spectral variables constructed using the concept of "change vector analysis." This study demonstrates that the decision tree algorithm achieves an overall accuracy (OA) of 96.18% and a Kappa (KA) of 0.949, surpassing the logistic regression approach with an OA of 93.49% and a kappa of 0.863. This study concludes that the decision tree approach produces better estimates than the logistic regression approach, with the main variables being the "magnitude" and "direction" variables generated by CVA. In the logistic regression approach, the most significant variables are identified by the education level variable and social variables, which exhibit higher uncertainty compared to the machine learning method. The development of a rapid assessment model using machine learning can assist decision-makers in determining post-forest fire restoration and rehabilitation steps