DAWP: A framework for global observation forecasting via Data Assimilation and Weather Prediction in satellite observation space

TL;DR

DAWP addresses biases and delays inherent in reanalysis-driven AI weather prediction by learning directly in the satellite observation space. It combines an observation-space data assimilation stage (AIDA) using a multi-modal masked autoencoder (MMAE) with a CBC-enabled spatiotemporal transformer for forecasting, followed by a mask ViT-VAE-based encoding and precipitation mapping. On a large composite satellite dataset, DAWP improves direct observation predictions and demonstrates potential for global precipitation forecasting, outperforming baselines and benefiting from targeted ablations of AIDA and CBC. This framework offers a path toward faster, observation-space–oriented forecasting with broad applicability to other Earth-system sensing and retrieval tasks.

Abstract

Weather prediction is a critical task for human society, where impressive progress has been made by training artificial intelligence weather prediction (AIWP) methods with reanalysis data. However, reliance on reanalysis data limits the AIWPs with shortcomings, including data assimilation biases and temporal discrepancies. To liberate AIWPs from the reanalysis data, observation forecasting emerges as a transformative paradigm for weather prediction. One of the key challenges in observation forecasting is learning spatiotemporal dynamics across disparate measurement systems with irregular high-resolution observation data, which constrains the design and prediction of AIWPs. To this end, we propose our DAWP as an innovative framework to enable AIWPs to operate in a complete observation space by initialization with an artificial intelligence data assimilation (AIDA) module. Specifically, our AIDA module applies a mask multi-modality autoencoder(MMAE)for assimilating irregular satellite observation tokens encoded by mask ViT-VAEs. For AIWP, we introduce a spatiotemporal decoupling transformer with cross-regional boundary conditioning (CBC), learning the dynamics in observation space, to enable sub-image-based global observation forecasting. Comprehensive experiments demonstrate that AIDA initialization significantly improves the roll out and efficiency of AIWP. Additionally, we show that DAWP holds promising potential to be applied in global precipitation forecasting.

Figures (20)

• Figure 1: A paradigm shift from physical space to observation space. Our DAWP is illustrated in (c).
• Figure 2: The framework of our DAWP. There are two stages in our DAWP: (1) Initialization and (2) Forecasting.
• Figure 3: Curves of MAE for the prediction of different modalities. The max leadtime is 72h with a 1h temporal resolution.
• Figure 4: A visualization of rollout predictions for global satellite observation forecasting.
• Figure 5: Training loss curve.