Modelado espacio-temporal con redes neuronales para la predicción de fenómenos ambientales

The thesis investigates neural approaches to environmental forecasting when observations are distributed across space and evolve through time. Its central applications include air-quality networks, atmospheric aerosols, greenhouse-gas concentrations and urban noise.

Graph representations provide a natural way to connect monitoring stations, cities or global observation points. Temporal models then learn how the corresponding signals change at different horizons and scales.

The resulting research programme links model architecture with the quality of observational data and the reliability of the decisions made after training.