Fig. 3: Comparison of EXP_CTRL, EXP_CORR and EXP_ASSI analysis. | npj Climate and Atmospheric Science

Fig. 3: Comparison of EXP_CTRL, EXP_CORR and EXP_ASSI analysis.

From: FuXi-DA: a generalized deep learning data assimilation framework for assimilating satellite observations

Fig. 3

The time series of the globally-averaged latitude-weighted RMSE for R300 (a), T300 (b), U300 (c), and V300 (d) at 00 and 12 UTC. The time series of the normalized RMSE difference for R300 (e), T300 (f), U300 (g), and V300 (h) at 00 and 12 UTC. The time series of the globally-averaged latitude-weighted RMSE for R300 (i), T300 (j), U300 (k), and V300 (l) at 06 and 18 UTC. The time series of the normalized RMSE difference for R300 (m), T300 (n), U300 (o), and V300 (p) at 00 and 12 UTC. The spatial distribution of time-averaged normalized RMSE difference for EXP_ASSI relative to EXP_CORR for R300 (q), T300 (r), U300 (s), and V300 (t). The solid lines with high transparency in (a–p) represent the actual statistical metrics, while the solid lines with low transparency indicate the values after applying a moving average with a window size of 5. In (q–t), the global domain is divided into 36 × 72 smaller regions, each with a 5° × 5° range, within which the regionally-averaged latitude-weighted RMSE and normalized RMSE difference are computed. The areas filled with plus signs are places in which the normalized RMSE difference passed the 95%-confidence-level t-test of significance.

Back to article page