A Multi-Source Spatiotemporal Large-Scale Model for Typhoon Prediction

Le résumé fourni par la source

Typhoon forecasting is a critical task for meteorological warning and disaster mitigation, aiming to predict key characteristics such as future track, intensity, and affected regions from historical observations. Existing physical and statistical models, however, often struggle to achieve high accuracy under complex atmospheric conditions. Inspired by advances in large language models (LLMs) for time-series analysis, we propose a Multi-source Typhoon Predictive Model with Temporal-Spatial Awareness (MTPM). MTPM encodes multi-source meteorological data into unified spatiotemporal representations and employs specialized embedding modules to capture both spatial distribution and temporal evolution patterns. The model jointly predicts typhoon trajectories and intensities, and experimental results on the Northwest Pacific typhoon dataset demonstrate that MTPM achieves high predictive accuracy while maintaining stability over long forecast horizons.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.