Diffusion Model Emulates High-Resolution Climate Precipitation Patterns
Researchers present CPMGEM, a diffusion-model emulator that downscales 60 km global climate model inputs to 8.8 km daily-mean precipitation, trained on 2.2 km convection-permitting simulations over England and Wales covering 1980–2080 under a high-emissions scenario. The stochastic emulator reproduces spatial structure, intensity distributions, and extreme-event statistics up to ~100-year return periods, and transfers to GCM inputs, offering lower-cost high-resolution precipitation for large ensembles and uncertainty sampling.
Key Points
- 1Emulates 2.2 km convection-permitting precipitation using a diffusion model downscaling 60 km inputs to 8.8 km
- 2Produces stochastic outputs that match spatial structure, intensity distribution, and extremes up to ~100-year return periods
- 3Enables lower-cost high-resolution precipitation for large ensembles and cross-GCM uncertainty sampling in climate impact studies
Scoring Rationale
High methodological novelty and practical applicability, limited by arXiv preprint status, regional training domain, and magnitude errors noted.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
