Fig. 4: Visualization of examples when only assimilating 10-m wind observations. | npj Climate and Atmospheric Science

Fig. 4: Visualization of examples when only assimilating 10-m wind observations.

From: Accurate initial field estimation for weather forecasting with a variational constrained neural network

Fig. 4

a Results of the 7-day medium forecast using ERA5 and state estimations only assimilating 10-m wind observations as initial fields, separately. The illustrated results are the forecasting RMSEs and ACCs of two upper-air variables (Z500 for the first column and Z850 for the third column) and two surface variables (U10 for the third column and V10 for the fourth column). b A typical example of background error (xb − xt). Two levels of geopotential variables (Z500 for the first column and Z850 for the second column) and two components of 10-m wind (U10 for the third column and V10 for the fourth column) are illustrated. c The visualization of the analysis increment (xa − xb) of the compared models is represented. For each case, 4DVarNet (first row), ViT (second row), and 4DVarFormer (third row) are shown. For all illustrated cases, the time is 00:00 UTC on 19 July 2021, for there is a tropical cyclone at this time in the East China region.

Back to article page