CL-LGFM: Early-Season Winter Wheat Mapping by Integrating Sentinel-2 NDVI and GPM Precipitation Data—A Case Study in the Chaohu Basin, China

Early-season winter wheat mapping is crucial for agricultural management and food security, but reliable identification remains challenging under weak spectral conditions during early growth stages. To address this challenge, this study developed a CNN–LSTM with a lag-aware gated fusion model (CL-LGFM) for winter wheat mapping in the Chaohu Basin, China, using a reconstructed 5-day Sentinel-2 NDVI time series and precipitation data from the Global Precipitation Measurement (GPM) mission. The model employs a dual-branch architecture to jointly learn vegetation and precipitation features and introduces a lag-aware dynamic gated fusion module to capture the delayed response of vegetation to precipitation and enhance multi-source feature fusion. The results show that the proposed method achieved reliable early-season winter wheat mapping (OA ≥ 0.90, Kappa ≥ 0.80) on 26 January, at least 10 days earlier than traditional methods, including SVM, RF, DTW, and TCN, using the same reconstructed 5-day NDVI time series. Optimal performance uses a 7 × 7 patch size and 30-day precipitation window.