Abstract: This video discusses PILA, a framework for improving physical models in Earth Observation by augmenting incomplete models with learnable residuals, applied to forest and volcanic data, with a focus on interpretability and accuracy.