Table 3 Brief descriptions of five MLWP models
From: Evaluation of five global AI models for predicting weather in Eastern Asia and Western Pacific
Model | Algorithm | Vertical pressure levels (excluding surface) | Frequency Options (h) | Time scale (days) | Variables (short name of ECMWF grib2 Table) | |
|---|---|---|---|---|---|---|
|  |  |  |  |  | On pressure Levels (in addition to u, v, t, z) | At surface or specific height (in addition to msl, 10 u, 10 v, 2t) |
Pangu-Weather | 3D Earth-specific transformer (Swin Transformer v1 with Earth-specific positional bias) | 13 | a) 1 b) 3 c) 6+ d) 24 | ----- | q | ------ |
FourCastNet v2 | Vision Transformer with Spherical Fourier Neural Operators | 13 | 6 | ----- | r | 100 u, 100 v, sp, tcwv |
GraphCast | Muti-Mesh Graph neural network | a) 13+ b) 37 | 6 | ----- | q, w | tp* |
FuXi | U-Transformer (Swin Transformer v2 with cube embedding) | 13 | 6 | 0 ~ 5+ 6 ~ 10+ 11 ~ 15 | r | tp* |
FengWu | Transformer based on Cross model fusion | a) 13+ b) 37 | 6 | ----- | q | ----- |
