Classifying forest cover and mapping forest fire susceptibility in Dak Nong province, Vietnam utilizing remote sensing and machine learning

Abstract

Forest fires can cause significant harm to the biodiversity, air quality, economy, and industries that depend on forests, which is particularly critical given the current climate change scenario. Therefore, it is crucial to predict potential fire risks, especially in Dak Nong, a province located in the highland region of Vietnam, where forest farming is prevalent. In this study, the utilization of remote sensing data from unmanned aerial vehicles (UAV) revealed a decline in forest areas while agricultural rubber and industrial crops witnessed an increase. Furthermore, the integration of SPOT satellite images with UAV and the eXtreme gradient boosting (XGBoost) classification algorithm proved highly efficient in mapping forest types in this province (OA = 81.446%, and Kappa = 0.803). To assess forest fire susceptibility in this study, a hybrid model known as Artificial bee colony-Adaptive neuro fuzzy inference system (ABC-ANFIS) was employed, which indicated that the majority of forests in the area face high to very high of fires, particularly in bamboo and dipterocarp forest areas. The present model is proposed to suggest better prevention and suppression strategies. These facts can help managers stop fires or deal with this disturbance more effectively that may occur in the future. Policymakers and researchers in different areas, such as the limestone forest in the north or the Dipterocarp forests in Vietnam, may consider replicating this study to evaluate specific wildfire risks and develop appropriate fire prevention strategies tailored to their local circumstances.