Kilometre-Scale Generative Downscaling of Weather
Generative models that turn coarse global forecasts into kilometre-scale local fields for hazards such as heavy rain.
Tier Grades
Tier A: Frontier AI labs
feasible
NVIDIA built CorrDiff on 2 km Taiwan data.
Tier B: Indonesian AI labs
feasible
CorrDiff needs paired km-scale regional model output, which a national weather agency holds for its own area.
Tier C: Institution units such as ITB
stretch
Training needs km-scale regional data that institution units may not hold.
Tier D: Organisations such as SEMAI (PoT, PoC, PoV)
stretch
A downscaled rainfall field for one region fits a proof of technology, but needs regional training data.
Where this comes from
Tiers: a) frontier AI labs, b) Indonesian AI labs, c) institution units such as ITB, d) organisations such as SEMAI delivering proof of technology, concept and value. Each rationale cites the anchors below.
Anchors
- ComputeCorrDiff: a UNet plus a corrector diffusion model; training compute not stated in the abstract.
- DataCorrDiff: 2 km regional model data over Taiwan conditioned on 25 km global reanalysis.
- Team size (proxy)Proxy only (author count of an exemplar paper, not a headcount): 13 authors on the CorrDiff paper (arXiv 2309.15214).
- Time horizonYears.
- Shelf-life signalsNVIDIA followed CorrDiff (arXiv 2309) with StormCast (arXiv 2408) and cBottle (arXiv 2505).
- Pipeline roleTier A publishes downscaling models; Tier B adapts them with national km-scale data; Tier D could use them for local hazard proofs.
Where this comes from
Team size is a proxy (author counts of exemplar papers), not a headcount.
Who and What
Links
R&D TopicKilometre-Scale Generative Downscaling of Weather
Models
CorrDiffNVIDIA, weatherStormCastNVIDIA, weather
Varda-single-1.0MeteoSwiss, weatherClimate in a Bottle (cBottle)NVIDIA, weather
CorrDiffNVIDIA, weatherStormCastNVIDIA, weatherVarda-single-1.0MeteoSwiss, weatherClimate in a Bottle (cBottle)NVIDIA, weatherUse Casesnone recorded
Who Runs Itnone recorded
Papers
- DeepSD: Generating High Resolution Climate Change Projections through Single Image Super-ResolutionThomas Vandal et al., arXiv preprint, 2017 founding
- Residual Corrective Diffusion Modeling for Km-scale Atmospheric DownscalingMorteza Mardani et al., arXiv preprint, 2023 state of the art
- ClimaX: A foundation model for weather and climateTung Nguyen et al., arXiv preprint, 2023 state of the art
Where this comes from
The papers to read first: where the direction began, its current best, and a survey when one exists.
Sources
- arxiv.org/abs/2309.15214 https://arxiv.org/abs/2309.15214
To cite this page: Project AURION, AI Use-Cases Research & Innovation, "Kilometre-Scale Generative Downscaling of Weather", https://aurion.semai.id/topics/km-scale-generative-downscaling/, records of October 2026.