Satellite Data-Based Logistic Regression and Neural Network Modeling for Wildfire Prediction

As the frequency and severity of wildfires continue to increase worldwide, the need for rapid and reliable wildfire detection and prediction technologies has become increasingly crucial. In this study, satellite data were utilized to predict wildfire occurrences, and the predictive performances of logistic regression and artificial neural network models were compared and analyzed. Land surface temperature (LST), normalized difference vegetation index (NDVI), and thermal anomalies (TA) were selected as input variables for wildfire prediction. The logistic regression model achieved a recall of approximately 82% and an F1 score of 0.80, demonstrating its applicability to satellite-based wildfire prediction. The artificial neural network model exhibited an overall improved predictive performance compared with the logistic regression model. Furthermore, to facilitate practical field implementation, the trained model was expressed in the form of explicit parametric equations, enabling the direct calculation of probability of wildfire occurrence.