Fig. 8: Comparison of the average latitude-weighted RMSE (first and second rows) and ACC (third and fourth rows) of the 139 sets of 7-day’s medium forecast using ERA5 and outputs of 4DVarFormer (red lines for using 3-day forecasting background filed as 4DVarFormer’s input, blue lines for 4-day forecasting background field as the input, and yellow lines for 5-day forecasting background field as the input). | npj Climate and Atmospheric Science

Fig. 8: Comparison of the average latitude-weighted RMSE (first and second rows) and ACC (third and fourth rows) of the 139 sets of 7-day’s medium forecast using ERA5 and outputs of 4DVarFormer (red lines for using 3-day forecasting background filed as 4DVarFormer’s input, blue lines for 4-day forecasting background field as the input, and yellow lines for 5-day forecasting background field as the input).

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

Fig. 8

Four surface variables, such as MSLP, T2M, U10, and V10, and six upper-air variables are illustrated, including Z500, T850, T500, R500, U500, V500, using testing data from 2021 to 2022. All models are evaluated against the ERA5 reanalysis dataset.

Back to article page