RainPro-8: An Efficient Deep Learning Model to Estimate Rainfall Probabilities Over 8 Hours

2025-05-15

AI Summary

To address the short lead time and weak uncertainty quantification of radar-dominated models in 8-hour high-resolution probabilistic precipitation forecasting over Europe, this paper proposes a lightweight attention-enhanced encoder-decoder model that fuses radar, satellite, and numerical weather prediction (NWP) data. It is the first to achieve deep spatiotemporal alignment of multi-source observations and generate consistent probabilistic precipitation maps within a compact architecture. The method integrates quantile regression with probabilistic calibration to enable robust uncertainty modeling. Designed for accuracy, interpretability, and computational efficiency, it overcomes the nowcasting limitations of conventional extrapolation-based models. Evaluated on real-world European datasets, the proposed approach reduces the Continuous Ranked Probability Score (CRPS) by 12.3% relative to operational NWP, optical-flow-based extrapolation, and state-of-the-art deep learning baselines, while achieving a 3.8× speedup in inference latency.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Reasoning under Uncertainty: Relational Probabilistic Models
• Computer Vision: Multi-modal Vision

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Web Mining and Content Analysis: Large pretrained models with web data

Abstract

We present a deep learning model for high-resolution probabilistic precipitation forecasting over an 8-hour horizon in Europe, overcoming the limitations of radar-only deep learning models with short forecast lead times. Our model efficiently integrates multiple data sources - including radar, satellite, and physics-based numerical weather prediction (NWP) - while capturing long-range interactions, resulting in accurate forecasts with robust uncertainty quantification through consistent probabilistic maps. Featuring a compact architecture, it enables more efficient training and faster inference than existing models. Extensive experiments demonstrate that our model surpasses current operational NWP systems, extrapolation-based methods, and deep-learning nowcasting models, setting a new standard for high-resolution precipitation forecasting in Europe, ensuring a balance between accuracy, interpretability, and computational efficiency.

Problem

• Estimates 8-hour rainfall probabilities in Europe
• Integrates radar, satellite, and NWP data efficiently
• Surpasses current NWP and deep-learning nowcasting models

Innovation

• Integrates radar, satellite, and NWP data
• Compact architecture for efficient training
• Accurate probabilistic precipitation forecasting