Skillful joint probabilistic weather forecasting from marginals

2025-06-12

AI Summary

This study addresses the challenge of modeling spatially correlated multivariate joint distributions in probabilistic weather forecasting using only univariate (marginal) point-wise observations. We propose the Forecasting Generative Network (FGN), the first framework enabling high-fidelity multivariate probabilistic modeling under purely marginal supervision. FGN employs learnable constraint-aware perturbations to generate physically consistent ensemble forecasts and adopts a CRPS-optimized deep ensemble architecture that implicitly learns spatial dependencies without requiring joint-label supervision or numerical model priors. Our method achieves state-of-the-art performance across both deterministic (e.g., RMSE) and probabilistic (e.g., CRPS, reliability) metrics. It significantly improves tropical cyclone track forecasting skill and, for the first time under marginal supervision, accurately captures the spatial structure of real atmospheric fields—achieving an optimal balance among forecast accuracy, physical consistency, and computational efficiency.

Technology Category

• Reasoning under Uncertainty: Relational Probabilistic Models
• Machine Learning: Probabilistic Circuits and Graphical Models
• Natural Language Processing: Generation

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Machine learning (ML)-based weather models have rapidly risen to prominence due to their greater accuracy and speed than traditional forecasts based on numerical weather prediction (NWP), recently outperforming traditional ensembles in global probabilistic weather forecasting. This paper presents FGN, a simple, scalable and flexible modeling approach which significantly outperforms the current state-of-the-art models. FGN generates ensembles via learned model-perturbations with an ensemble of appropriately constrained models. It is trained directly to minimize the continuous rank probability score (CRPS) of per-location forecasts. It produces state-of-the-art ensemble forecasts as measured by a range of deterministic and probabilistic metrics, makes skillful ensemble tropical cyclone track predictions, and captures joint spatial structure despite being trained only on marginals.

Problem

• Improving joint probabilistic weather forecasting accuracy
• Outperforming traditional numerical weather prediction models
• Generating skillful ensemble tropical cyclone track predictions

Innovation

• FGN uses learned model-perturbations for ensembles
• FGN minimizes CRPS for per-location forecasts
• FGN captures joint spatial structure from marginals