Fig. 1: Comparison of the average latitude RMSE (first and second rows) and ACC (third and fourth rows) of the 139 sets of 30-day assimilation cycle using 4DVarNet (blue lines), ViT (yellow lines), and 4DVarFormer (red lines) for four surface variables, such as MSLP, T2M, U10, and V10, and six upper-air variables, including Z500, T850, T500, R500, U500, V500, using testing data from 2021 to 2022. | npj Climate and Atmospheric Science

Fig. 1: Comparison of the average latitude RMSE (first and second rows) and ACC (third and fourth rows) of the 139 sets of 30-day assimilation cycle using 4DVarNet (blue lines), ViT (yellow lines), and 4DVarFormer (red lines) for four surface variables, such as MSLP, T2M, U10, and V10, and six upper-air variables, including Z500, T850, T500, R500, U500, V500, using testing data from 2021 to 2022.

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

Fig. 1

All models are evaluated against the ERA5 reanalysis dataset.

Back to article page