Answers are grounded in this paper and cite their sources.
Accurately representing surface precipitation is crucial for the operational use of weather and climate models. Presently, global numerical weather prediction (NWP) models struggle to accurately generate precipitation due to their parametrization of unresolved deep convective clouds and, in regions of grid-resolved ascent, inadequate parameterizations of cloud microphysics. Here we bypass these parameterizations with a machine learning model that diagnoses precipitation from 13 ERA5 fields that are easily observed and assimilated. Only three of these fields are moisture-related and all three are better observationally constrained than hydrometeor-class variables like rain or cloud liquid water. We train a pair of models; ML ERA5 using ERA5 precipitation as the target, and ML IMERG using a satellite based precipitation product. ML ERA5 closely reproduces the ERA5 precipitation at all intensities. When evaluated against the satellite dataset, ML IMERG closely matches observations, notably reproducing the diurnal cycle of the satellite product. ML IMERG generally captures extremes better than ERA5 while also avoiding the overproduction of light precipitation present in ERA5. When evaluated against a third ground-and-radar-based dataset, ML IMERG inherits the strengths of the satellite dataset which is superior to ERA5 in the summer months. Both ML IMERG and ML ERA5 are computationally inexpensive to run at inference, taking less than a second per output time to compute precipitation globally at 1°×1° resolution.
No takes yet. Share an insight, caveat, or question.
Moreno et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: