Evaluación de un modelo de susceptibilidad a incendios forestales en el Cantón Loja para la planificación territorial

Abstract

Abstract: Forest fires are a global phenomenon, the impact they generate on the territory is well known, while the identification of areas susceptible to this phenomenon is not so well known, therefore, it requires more research; the current susceptibility model of Canton Loja has a deficiency, it is affected by the presence of clouds. The research focused on generating predictive models based on three machine learning techniques: logistic regression (LR), multivariate adaptive regression splines (MARS), logistic decision tree (LMT) and implementing more accurate satellite information that is not affected by cloud cover. The machine learning models were trained from a database with 1436 points, which consisted of ten variables related to topography, proximity to human activities and soil moisture. They were compared with models generated in preliminary studies to identify significant differences. After testing the performance of the generated models, they were compared with the results of models generated in preliminary studies, and a good predictive capacity was observed in MARS to identify the probability of fire occurrence.