RustMAE: A Spatiotemporal Transformer for Short- to Medium-Term Warning of Wheat Stripe Rust Spring Spread

Wheat stripe rust poses a significant threat to food security in China. This study proposes RustMAE, a spatiotemporal Transformer-based early warning method utilizing masked autoencoders for 14-day disease spread forecasting during the spring epidemic season. RustMAE employs a dual-branch MAE pre-training framework that integrates spectral reconstruction and spatial structure recovery to learn disease-relevant features from large-scale unlabeled remote sensing imagery. A multi-scale spatiotemporal attention mechanism captures disease evolution patterns across 3-, 7-, and 14-day windows. By fusing optical imagery, synthetic aperture radar, and meteorological factors, the model constructs a multi-source driven disease risk prediction system. Validation on 1,847 samples from Shaanxi, Gansu, and Henan provinces (2019-2024) demonstrates that RustMAE achieves 87.23% overall accuracy and 0.924 AUC, significantly outperforming benchmark methods. This approach provides an innovative pathway for precise wheat stripe rust early warning and control.