A Space-Time Transformer for Precipitation Forecasting

2025-11-14

AI Summary

Traditional numerical weather prediction (NWP) suffers from high computational cost and poor short-term (0–4 hour) precipitation forecasting accuracy. Meanwhile, existing AI-based approaches lack sufficient capability to model spatiotemporal dynamics in satellite video sequences. To address these limitations, we propose SaTformer—a full spatiotemporal attention-driven video Transformer that frames precipitation forecasting as a multi-level classification task. We introduce a class-weighted loss function to mitigate the long-tailed distribution of rainfall intensities. By deeply integrating spatial and temporal dependencies within infrared satellite radiance time-series videos, SaTformer significantly enhances extreme precipitation detection. Evaluated on the NeurIPS Weather4Cast 2025 Accumulated Precipitation Challenge, SaTformer achieved first place, outperforming baseline methods in short-term precipitation accuracy. Results demonstrate its dual advantages in both computational efficiency and predictive performance.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Computer Vision: Remote Sensing / Geospatial AI
• Planning, Routing, and Scheduling: Optimization of Spatio-temporal Systems

Application Category

• Search and Retrieval-Augmented AI: Large language models for search
• Web Mining and Content Analysis: Large pretrained models with web data
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

Abstract

Meteorological agencies around the world rely on real-time flood guidance to issue live-saving advisories and warnings. For decades traditional numerical weather prediction (NWP) models have been state-of-the-art for precipitation forecasting. However, physically-parameterized models suffer from a few core limitations: first, solving PDEs to resolve atmospheric dynamics is computationally demanding, and second, these methods degrade in performance at nowcasting timescales (i.e., 0-4 hour lead-times). Motivated by these shortcomings, recent work proposes AI-weather prediction (AI-WP) alternatives that learn to emulate analysis data with neural networks. While these data-driven approaches have enjoyed enormous success across diverse spatial and temporal resolutions, applications of video-understanding architectures for weather forecasting remain underexplored. To address these gaps, we propose SaTformer: a video transformer built on full space-time attention that skillfully forecasts extreme precipitation from satellite radiances. Along with our novel architecture, we introduce techniques to tame long-tailed precipitation datasets. Namely, we reformulate precipitation regression into a classification problem, and employ a class-weighted loss to address label imbalances. Our model scored first place on the NeurIPS Weather4Cast 2025 Cumulative Rainfall challenge. Code and model weights are available: https://github.com/leharris3/satformer

Problem

• Improving precipitation forecasting accuracy at nowcasting timescales
• Addressing computational limitations of traditional numerical weather models
• Overcoming data imbalance in extreme precipitation prediction datasets

Innovation

• Transformer model with full space-time attention
• Reformulating precipitation regression as classification
• Using class-weighted loss for label imbalance