Table 1 Overview of the selected weather forecast models
From: Global performance benchmarking of artificial intelligence models in atmospheric river forecasting
Model | Brief description | Input | Output frequency | Output resolution |
|---|---|---|---|---|
Pangu-Weather | 3D Earth-Specific Transformer | z,t,u,v,q,t2m,u10,v10,msp | 6 h | 0.25° |
FCN2 | Vision Transformer with Spherical Fourier Neural Operators | z,t,u,v,r,t2m,u10,v10,msp,sp | 6 h | 0.25° |
GraphCast | Muti-Mesh Graph neural network | z,t,u,v,q,w,t2m,u10,v10,msp,tp | 6 h | 0.25° |
FuXi | U-Transformer | z,t,u,v,r,t2m,u10,v10,msp,tp | 6 h | 0.25° |
NeuralGCM | Dynamical cores combined with neural networks for the physics tendencies | u,v,z,t,q,sciw,sclw,SST,SIC | 1 day | ~0.7° |
FGOALS | Finite volume dynamical core for the atmosphere model coupled with ocean (POP2; Parallel Ocean Program version 2) and sea ice (CICE4; Los Alamos Sea Ice Model version 4) models | full-field initialization strategy (t, u,v, q sp for the atmospheric part) | 1 day | ~1° |
