On what depends the robustness of multi-source models to missing data in Earth observation?

This work investigates the factors determining the robustness of multi-source models to missing data in Earth Observation. Six state-of-the-art models are evaluated under scenarios where either a single data source is missing or only one source is available at inference time. The analysis shows that model robustness depends strongly on the downstream task, the complementarity between data sources, and the model architecture. Interestingly, removing certain sources can sometimes improve predictive performance, challenging the assumption that incorporating all available data is always beneficial. These findings highlight the importance of understanding data-source interactions and model complexity when designing robust and efficient multi-source EO models.

Recommended citation: Mena, Francisco, et al. "On what depends the robustness of multi-source models to missing data in Earth observation?." IGARSS 2025-2025 IEEE International Geoscience and Remote Sensing Symposium. IEEE, 2025.
Download Paper | Download Slides