Fig. 4
From: Spherical multigrid neural operator for improving autoregressive global weather forecasting
Weighted mean relative \(\mathcal{L}2\) norm losses for ablation experiments. In the Figure, the x-axis represents autoregressive steps (1 to 144) i.e. lead times (1 h to 144 h), and the y-axis is the values of weighted mean relative \(\mathcal{L}2\) norm losses. The lower the weighted mean relative losses the better the performance. (a) mean relative losses of geopotential height; (b) mean relative losses of vorticity; (c) mean relative losses of divergence. In the Figure “w/o” is short for “without”.
