The present study aims to develop and optimize a spatial risk scenarios model for forest fires in Ecuador through the integration of machine learning and satellite remote sensing. To this end, 103,224 historical records of thermal anomalies captured by the VIIRS sensor were processed, along with five continuous meteorological variables from the ERA5-Land global reanalysis, spatial coordinates (latitude and longitude), and temporal indicators (month of the year), spanning the period between 2015 and 2025. The methodology was based on the construction of a 5-kilometer resolution spatial grid coupled with the XGBoost classification algorithm, designed to efficiently manage class imbalance in the face of extreme climate events. The computational system achieved an accuracy level of 81% on the test set, validating its robustness with a true positive rate of 0.86 and an area under the ROC curve of 0.8382. Furthermore, the feature importance analysis determined that precipitation deficits and seasonal timing act as the primary drivers of ignition risk, exerting a significantly greater influence than surface temperature, concluding that this scalable and open-access tool allows for high-precision mapping of vulnerable areas in the Andean and coastal regions through scenario-based experiments, aiming to optimize the distribution of state preventive resources against the constant threat of global warming.