Model architectures we could consider including

[leifdenby]

I am just typing on this github issue to keep of work I come across so that I don't loose track. Feel free to add this list :)

• NowAlpha from KMA (Korea Meteorological Administration) presented at the 22/1/2026 inaugural WMO AI webinar: https://wmo.int/media/news/wmo-ai-webinars

Evolution on LDCast:

Classification	Main Focus	Example
Baseline generative nowcasting	Latent diffusion model for short nowcasting	LDCast (MeteoSwiss) (arxiv.org)
Structural enhancements	Two-stage forecasting / end-to-end training	STLDM (arxiv.org)
Physics-aware generative modeling	Domain constraint integration	PreDiff (openreview.net)
Multi-modal latent fusion	Combining radar + other data sources	CastDiffuser (sciencedirect.com)
Extended domain adaptation	Medium-range / global forecasting	LaDCast (arxiv.org)
Benchmarking & derivative use cases	Alternative domains / training tweaks	Solar irradiance, bias correction, ShadeCast (sciencedirect.com)
Methodological innovations	Better attention / preference-training	RainDiff, SynCast (arxiv.org)