Weakly-Constrained 4D Var for Downscaling with Uncertainty using Data-Driven Surrogate Models

2025-03-04

AI Summary

Pure data-driven models (e.g., FourCastNet) suffer from long-term forecast instability and systematic error accumulation in dynamic downscaling. Method: This work proposes the first integration of a weak-constraint 4D-Var data assimilation framework with FourCastNet, using the latter as a surrogate model within the variational framework to explicitly model and dynamically correct temporal evolution errors. The approach incorporates ERA5-based prior information and uncertainty-aware modeling to enable rigorous error propagation and quantification. Results: In hurricane tracking experiments, the hybrid system significantly outperforms both the ensemble Kalman filter (EnKF) and the baseline FourCastNet in trajectory and intensity forecasting accuracy. Moreover, it yields narrower, more discriminative uncertainty intervals—demonstrating improved reliability. This work establishes the first variational assimilation paradigm for data-driven meteorological modeling that simultaneously ensures forecast stability, physical interpretability, and rigorous uncertainty quantification.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Reasoning under Uncertainty: Stochastic Optimization
• Intelligent Robots: State Estimation

Application Category

• Web Mining and Content Analysis: Web data integration and cleaning
• Systems and Infrastructure for Web, Mobile and WoT: Data management and stream processing for Web, mobile and wireless applications
• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics

Abstract

Dynamic downscaling typically involves using numerical weather prediction (NWP) solvers to refine coarse data to higher spatial resolutions. Data-driven models such as FourCastNet have emerged as a promising alternative to the traditional NWP models for forecasting. Once these models are trained, they are capable of delivering forecasts in a few seconds, thousands of times faster compared to classical NWP models. However, as the lead times, and, therefore, their forecast window, increase, these models show instability in that they tend to diverge from reality. In this paper, we propose to use data assimilation approaches to stabilize them when used for downscaling tasks. Data assimilation uses information from three different sources, namely an imperfect computational model based on partial differential equations (PDE), from noisy observations, and from an uncertainty-reflecting prior. In this work, when carrying out dynamic downscaling, we replace the computationally expensive PDE-based NWP models with FourCastNet in a ``weak-constrained 4DVar framework"that accounts for the implied model errors. We demonstrate the efficacy of this approach for a hurricane-tracking problem; moreover, the 4DVar framework naturally allows the expression and quantification of uncertainty. We demonstrate, using ERA5 data, that our approach performs better than the ensemble Kalman filter (EnKF) and the unstabilized FourCastNet model, both in terms of forecast accuracy and forecast uncertainty.

Problem

• Stabilize data-driven models for long-term weather forecasting.
• Improve dynamic downscaling accuracy using weak-constrained 4DVar framework.
• Quantify uncertainty in forecasts using data assimilation techniques.

Innovation

• Weak-constrained 4DVar framework for downscaling
• Data-driven surrogate models replace PDE-based NWP
• Uncertainty quantification in dynamic downscaling tasks