Reproduce “Machine learning emulation of precipitation from km-scale regional climate simulations using a diffusion model” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Henry Addison, Elizabeth Kendon, Suman Ravuri, Laurence Aitchison, Peter AG Watson
CPMGEM emulates daily-mean precipitation over England and Wales at 8.8-km resolution by generating stochastic rainfall maps conditioned on coarse atmospheric fields rather than interpolating coarse precipitation. On held-out CPM data, its spatial RMS relative standard-deviation bias is 3.6%, versus 13.3% for deterministic U-Net. It better reproduces small-scale variability, but emulates CPM behavior, not observed rainfall; future change magnitudes remain imperfect.
Key insights
- Diffusion models effectively emulate the complex, non-deterministic relationship between coarse-scale atmospheric states and fine-scale precipitation, including stochasticity and realistic spatial structures.
- ML emulators can successfully transfer from a 'perfect model' training setup (coarsened CPM inputs) to practical application using real Global Climate Model inputs, while learning the added value of CPMs.
- The method can achieve good climatological reproduction even with substantially smaller training datasets, increasing its applicability in data-limited contexts.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.