Researchers have used deep learning to feed daily satellite measurements of sun-induced chlorophyll fluorescence into a land surface model, allowing the model to detect irrigation patterns it was never designed to represent. The work uses data from the TROPOMI instrument on Copernicus Sentinel-5P and was tested in Spain's Ebro basin, where the analysis picked up summer regrowth on irrigated cropland while surrounding dryland areas stayed flat. The approach improves vegetation state estimates and supports better tracking of the land carbon cycle, a key uncertainty in national carbon budgets. Co-assimilating the fluorescence signal with standard leaf area index data gave the best results. The authors note that the method still cannot fully fix gross primary production estimates when the underlying model lacks a realistic representation of soil water supply from irrigation, so the water side of land surface models remains the limiting factor. For carbon market and climate policy users, this points to a practical monitoring tool for land-based emissions and water use. It also shows why model physics matter: satellite data alone cannot fix missing processes.
