Fig. 6
From: Spherical multigrid neural operator for improving autoregressive global weather forecasting
Globally averaged latitude-weighted anomaly correlation coefficient (ACC) of IFS T42 (black lines), FourCastNet (blue lines), SFNO (purple lines), and SMgNO (red lines) for 3 surface variables, 4 upper-air variables at 500 hPa pressure level, and 5 upper-air variables at 850 hPa pressure level with spatial resolution of 5.625° × 5.625° in 7 days forecasts using testing data from 2017 to 2018. In the Figure, the x-axis represents lead times, and the y-axis is the values of globally averaged latitude-weighted ACC. The higher the value of ACC, the better the performance. IFS T42 were not available for 10U and 10 V.
