Abstract: Accurate maize yield prediction is often required for regional crop growth monitoring in modern agriculture. However, existing sequential models cannot simultaneously represent continuous local temporal variations and full-season global dependencies. In this study, a deep learning model was developed to accurately predict the maize yield for early warning and spatial generalization at the county scale, according to multi-source agricultural information. A Dual-Branch Global and Local Feature Fusion Network (DGLF-Net) was developed for maize yield prediction in Jilin Province from 2009 to 2022. Moderate. The model was also integrated with the Resolution Imaging Spectroradiometer (MODIS) surface reflectance, vegetation indices, meteorological data, and soil properties. A Local Temporal Feature Extractor (LTFE), composed of a Convolutional Neural Network (CNN) and a Long Short-Term Memory (LSTM) network, was used to extract local continuous temporal features. A Global Temporal Attention Module (GTAM) was used to capture long-range temporal dependencies over the growing season. A gated fusion module (Gating) was then used to combine local temporal and global contextual information. Data from 2009 to 2020 were used for training, and data from 2021 to 2022 were used for testing. The correlation analysis showed that maize yield formation was dominated by both full-season cumulation and key-month local dynamics. Vegetation index (EVI) in July showed the strongest positive correlation with yield, with a correlation coefficient of 0.51, indicating that yield prediction depended on the remote sensing information during key growth stages. Ablation experiments confirmed the contribution rates of the main components of DGLF-Net. Furthermore, the complete model achieved a coefficient of determination (R²) of 0.812 0 in 2022, a root mean squared error (RMSE) of 441.95 kg/hm², and a mean absolute error (MAE) of 325.30 kg/hm². The R² values decreased to 0.711 7 and 0.773 9, respectively, after removing GTAM and Gating. The R² reduced to 0.605 3 in LTFE after removing CNN, while R² reduced to 0.745 2 after removing LSTM, indicating that both local feature extraction and temporal dependency modeling contributed to prediction performance. Model comparison experiments showed that DGLF-Net outperformed random forest regression (RFR), extreme gradient boosting (XGBoost), CNN, Transformer, LSTM, CNN-Transformer, and CNN-LSTM-Attention in 2021 and 2022. DGLF-Net improved R² by 5.52% in 2022, whereas RMSE and MAE were reduced by 9.69% and 17.38%, respectively, compared with CNN-LSTM-Attention. Spatial independent validation showed that DGLF-Net maintained stable performance in counties with different yield levels. The prediction accuracies for Ji’an, Linjiang, and Gongzhuling were 96.10%, 97.47%, and 95.08%, respectively, in 2022. Early prediction analysis showed that the model reached an R² of about 0.77 using data from May to September, indicating a reliable prediction for one month before harvest. The vegetation index analysis showed that the combination of NDVI, CVI, and green normalized difference vegetation index (GNDVI) performed best in 2022, with an R² of 0.826 0, RMSE of 425.30 kg/hm², and MAE of 309.40 kg/hm². SHapley Additive exPlanations (SHAP) analysis indicated that soil organic carbon (SOC), bulk density (Bd), and vapor pressure deficit (VPD) were stable and important features in both years. Spatial prediction maps further showed that DGLF-Net better preserved the county-level yield gradient in Jilin Province for the more balanced spatial distribution of errors. The DGLF-Net effectively improved county-level maize yield prediction using local temporal learning dynamics and global seasonal dependencies, indicating the strong prediction accuracy, early prediction potential, and spatial generalization. Future studies can further incorporate higher-resolution remote sensing data, planting density, field information, cultivar differences, and pest or disease information in more complex agricultural scenarios.