Fuxi-DA: A Generalized Deep Learning Data Assimilation Framework for Assimilating Satellite Observations

TL;DR

FuXi-DA tackles the data assimilation bottleneck in numerical weather prediction by introducing a DL-based, generalized framework that unifies background and satellite observations in a latent fusion space. By assimilating Fengyun-4B AGRI data, it achieves measurable reductions in analysis error and improvements in medium-range forecasts, and it demonstrates consistency with atmospheric physics through targeted single-observation tests. The approach eliminates the need for hand-crafted observation operators and background-error covariances, reduces pre-processing, and delivers rapid assimilation, enabling potential end-to-end DL-based weather prediction. This work suggests DL-based data assimilation can leverage multi-source satellite data for all-sky assimilation and scalable, real-time forecasting.

Abstract

Data assimilation (DA), as an indispensable component within contemporary Numerical Weather Prediction (NWP) systems, plays a crucial role in generating the analysis that significantly impacts forecast performance. Nevertheless, the development of an efficient DA system poses significant challenges, particularly in establishing intricate relationships between the background data and the vast amount of multi-source observation data within limited time windows in operational settings. To address these challenges, researchers design complex pre-processing methods for each observation type, leveraging approximate modeling and the power of super-computing clusters to expedite solutions. The emergence of deep learning (DL) models has been a game-changer, offering unified multi-modal modeling, enhanced nonlinear representation capabilities, and superior parallelization. These advantages have spurred efforts to integrate DL models into various domains of weather modeling. Remarkably, DL models have shown promise in matching, even surpassing, the forecast accuracy of leading operational NWP models worldwide. This success motivates the exploration of DL-based DA frameworks tailored for weather forecasting models. In this study, we introduces FuxiDA, a generalized DL-based DA framework for assimilating satellite observations. By assimilating data from Advanced Geosynchronous Radiation Imager (AGRI) aboard Fengyun-4B, FuXi-DA consistently mitigates analysis errors and significantly improves forecast performance. Furthermore, through a series of single-observation experiments, Fuxi-DA has been validated against established atmospheric physics, demonstrating its consistency and reliability.

Figures (12)

• Figure 1: The comparison between the Fuxi-DA and the variational data assimilation.
• Figure 2: Comparison of analysis after a single assimilation. The time series of the globally-averaged latitude-weighted RMSE (a, b, c, d) and the time series of the normalized RMSE difference (e, f, g, h) of EXP_CTRL (black lines), EXP_CORR (blue lines) and EXP_ASSI (red lines), and the spatial distribution of normalized RMSE difference (i, j, k, l) for EXP_ASSI relative to EXP_CORR for R300, T300, U300 and V300. Sub-figure in (e) displays the original normalized RMSE difference at certain moments. The original data is represented by solid lines with high transparency, while solid lines with low transparency indicate the smoothed values.
• Figure 3: Visualization of spatial distributions, error distributions and analysis increments of R300 and Z500, as well as visualization of observations. Spatial maps of R300 for ERA5 (a), EXP_CORR (b), and EXP_ASSI (c); spatial maps of Z500 for ERA5 (d), EXP_CORR (e), and EXP_ASSI (f); spatial maps of error distribution for R300 (g) and Z500 (j); spatial maps of analysis increments for R300 (h) and Z500 (k); and a spatial map of AGRI channel 9 brightness temperature (i).
• Figure 4: Comparison of global 10-day forecasts after a single assimilation. The normalized RMSE difference of EXP_CTRL (black lines), EXP_CORR (blue lines) and EXP_ASSI (red lines) for R300, R500 and R850 (a, b, c), T300, T500 and T850 (d, e, f), as well as Z300, Z500 and Z850 (g, h, i). The shaded areas represent 95% confidence intervals.
• Figure 5: Comparison of global 10-day forecasts after a single assimilation. The normalized RMSE difference of CTRL (black lines), EXP_ASSI using a single-time-step loss (blue lines) and EXP_ASSI using a multi-time-step loss (red lines) for R300, R500 and R850 (a, b, c), T300, T500 and T850 (d, e, f), as well as Z300, Z500 and Z850 (g, h, i). The shaded areas indicate a 95% statistical confidence interval. The shaded areas represent 95% confidence intervals.