What question did this study set out to answer?

The central aim is to develop an effective forecasting model that integrates spatiotemporal data for better weather predictions.

March 25, 2026Open Access

SIGMAformer: a spatiotemporal Gaussian mixture correlation transformer for global weather forecasting

Puntos clave

The central aim is to develop an effective forecasting model that integrates spatiotemporal data for better weather predictions.
Developed a spatiotemporal forecasting architecture using GMPE and DSTC modules.
Calculated temporal correlations within each station and spatial correlations across stations using dynamic weights.
Conducted experiments on global weather datasets to evaluate performance.
Compared performance with state-of-the-art forecasting models.
SIGMAformer showed consistent improvement over existing models in weather forecasting.
Significantly enhanced wind speed predictions compared to traditional methods.
Eliminating the DSTC module resulted in mean squared error increases of 7.18% for wind speed and 7.22% for temperature.

Resumen

Abstract SpatIotemporal Gaussian Mixture correlAtion transformer (SIGMAformer) is a spatiotemporal forecasting architecture that integrates a Gaussian mixture pattern extractor (GMPE) with a dynamic spatiotemporal correlation (DSTC) mechanism. The DSTC module leverages GMPE to automatically compute spatiotemporal pattern-specific weights from the data. These weights are first used to calculate temporal correlations within each station and then integrated with global pattern weights to evaluate spatial correlations across stations. This nonlinear, dynamically adaptive modeling approach emphasizes critical spatiotemporal patterns while suppressing less relevant ones. Experiments on global weather datasets reveal that SIGMAformer consistently outperforms state-of-the-art forecasting models and significantly improves wind speed prediction. Removing DSTC increased the mean squared error values by up to 7.18% and 7.22% for wind speed and temperature predictions, respectively. These findings underscore SIGMAformer’s capacity to capture essential spatiotemporal patterns and establish a scalable methodology for intelligent sensor-network fusion in environmental forecasting.

Me gusta

Guardar

Ver artículo completo