We analyzed the behavior of forest fires for daily prediction purposes in one of the regions with the highest fire incidence in Mexico. The main objective was to build logistic regression models (LRMs) for daily prediction of forest fire ignition based on meteorological drought indices. We built 252 LRMs for seven types of vegetation cover of greater representativeness and interest for the study area. Three dynamic variables were considered to estimate daily dryness in combustible fuels based on the effective drought index and the standardized precipitation-evapotranspiration index. Additionally, two weather data sources were included in drought indices: conventional weather stations (CWS) and automatic weather stations (AWS). Prediction efficiency assessment for LRMs was done through the relative operating characteristic (ROC) and model precision efficiency (MPE). The results show that LRMs using data from CWS performed relatively better than those based on data from AWS, as the former data sources have higher spatial density and thus generate predictions with higher accuracy (ROC ≥ 0.700, MPE ≥ 0.934). For both data sources, the use of standardized precipitation-evapotranspiration index as a fuel dryness estimator is recommended, as it reflects an atmospheric moisture balance between precipitation and reference evapotranspiration (ROC ≥ 0.734, MPE = 1). Such predictive models can be used as inputs in early warning systems for forest fire prevention or mitigation.
Full Text Finder