Forest fire risk assessment model optimized by stochastic average gradient descent

• BPNN is used and optimized by the stochastic average gradient descent (SAGD) algorithm. • High-precision datasets are obtained and preprocessed by ArcGIS for accuracy. • Eleven influencing factors were selected to construct forest fire risk assessment index system. • Historical data was used for training and accuracy verification. Forest fire is a serious global natural disaster that occurs frequently and is characterized by its suddenness, destructiveness, and difficulty in emergency response. Therefore, it’s of great importance to research forest fire risk assessment and prediction to protect the ecological environment of forests, respond to disaster damage in time and mitigate the effect of disasters. Most of the current related research methods rely on a stable operating environment and high raw data accuracy, the number of influencing factors considered is always different from the actual in many ways, and there is no systematic validation, etc., which makes the feasibility insufficient. In this study, a comprehensive method for forest fire risk assessment and prediction is proposed using a back-propagation neural network (BPNN) optimized by the stochastic average gradient descent (SAGD) algorithm. The model is based on the Regional Disaster System Theory (RDST), and incorporates 11 indicators of meteorological, vegetation, and human activity factors from the aspects of hazard-formative factor, hazard-formative environment, and hazard-affected body, and achieves a prediction accuracy of 94.38% and a coefficient of determination of 0.9581 when compared with historical data and the Global Fire Weather Index (FWI). The results demonstrate that the SAGD optimization improves the performance of the BPNN, offers a pragmatic solution for forest fire risk assessment in the Guangxi Zhuang Autonomous Region, and facilitates the enhancement of its disaster preparedness and response capabilities, thereby mitigating the adverse effects of forest fires.