Vulcan Climate Modeling (a small philanthropically-supported group in Seattle) and NOAA/GFDL are collaborating on a pilot project to use machine learning to develop a skillful corrective parameterization for a full-complexity global atmospheric model that helps it evolve more like a reference data set, which could be a reanalysis or a finer-grid global model. We have applied this approach to climate-oriented versions of FV3GFS with 100-200 km grids. Encouragingly, it significantly improves their 0-7 day weather forecasts and time-mean precipitation distribution, both in present and SST-perturbed climates.
