Laxmi

AI weather model from Schmitt, Jain and co-authors that retrains ECMWF's AIFS on satellite precipitation observations to forecast rain more accurately at the local scale.

Laxmi is a global precipitation forecasting model described in a September 2026 arXiv paper by Julian F. Schmitt, Bertrand Delorme, Robert C. King, Yashica Patodia, Tapio Schneider, Aditi Sheshadri and Ravi Jain. It starts from AIFS, ECMWF's open-source 0.25-degree probabilistic weather model, and retrains it on satellite-based precipitation observations instead of the ERA5 reanalysis most AI weather models are trained against.

The paper's argument is that gains in global-average skill do not automatically help with floods or drought, and that ERA5 carries known biases in rainfall. Against that baseline the authors report a 19% improvement in global probabilistic accuracy, less over-prediction of drizzle, better capture of extreme rainfall, and the best 150 mm event-total forecasts in seven of ten Indian tropical storms studied.

How the model is distributed, and under what licence, is not stated on the abstract page. Status: research, as of the paper's third revision on 23 September 2026.