Data Assimilation

earth2studio.models.da

Data assimilation models are a class of models that integrate observational data into model states or grids. These models can ingest both sparse observations (via DataFrames) and dense fields (via DataArrays) to produce output suitable for downstream tasks such as driving a prognostic model or generating a guided forecast. Data assimilation models support both stateless and stateful operation, allowing them to process observations independently or maintain internal state across time steps.

Warning

Data Assimilation models are a new addition to Earth2Studio and APIs might be subject to change without warning while the implementation is hardened.

RegionGlobalNAEUASAUAFSA

ClassNWCDSMRFS2SDACM

Product

Release202120222023202420252026

Rec VRAM96 GB80 GB48 GB40 GB

HealDAGlobalDA202640 GB	HealDA data assimilation model for global weather analysis from sparse observations on a HEALPix grid.
InterpEquirectangularGlobalNADA	Interpolation assimilation model that interpolates sparse observations to a lat-lon grid.
StormCastSDANADANWC202440 GB	StormCast with score-based data assimilation (SDA) using diffusion posterior sampling for convection-allowing regional forecasts.