Fig. 5
From: Spherical multigrid neural operator for improving autoregressive global weather forecasting
Visualization of forecast results for geopotential (m2/s2) at the 500 hPa pressure level. Columns from left to right correspond to 1 day, 3 days, 5 days, and 7 days of lead time, i.e., 4, 12, 20, and 28 autoregressive steps, respectively. Rows from top to bottom represent the ERA5 (ground truth), IFS T42, Fourcastnet, SFNO, and SMgNO models. In the Figure, RMSE is the abbreviation of root mean square error. For all cases, the input time is 00:00 UTC on 12 March 2017, and the spatial resolution is 5.625° × 5.625°. This Figure was created using Matplotlib library version 3.8.4 (https://matplotlib.org/) and Cartopy library version 0.23.0 (https://scitools.org.uk/cartopy) on Python 3.10.13 (https://www.python.org), with coastline data from Natural Earth public domain datasets (https://www.naturalearthdata.com/).
