This week's seminar features Camilo Matus-Olivares, PhD in Agrifood and Environmental Sciences from the University of La Frontera, Chile, who, along with Keyla Alencar (UTEM), Horacio Gilabert, Andres Weintraub and Miguel Castillo; as well as Jaime Carrasco-Barra, Felipe de la Barra, Erico Kutchartt, Diego Terán and Núria Aquilué, developed an advanced methodology for mapping surface fuels of the KITRAL classification.

To achieve this, a Data Cube was constructed by integrating seasonal time series from the Copernicus program, combining optical indices from Sentinel-2 and radar backscatter from Sentinel-1, fused with topographic and vegetation structure variables. To model the high spatial complexity and imbalance of 33 classes, a Stacking Ensemble architecture was implemented. Baseline Gradient Boosting Models (XGBoost, LightGBM, and CatBoost) were trained, and their predictions were combined using a Meta-Learner. Evaluated using Spatial Block Cross-Validation, the Meta-Learner model successfully learns complex geographic contexts, significantly outperforming individual algorithms in accuracy for land management.