Abstract:Based on forecast products of the European Center for Medium-Range Weather Forecasts - Integrated Forecasting System(ECMWF-IFS) and hourly temperature observation data from the China Meteorological Administration Land Data Assimilation System(CLDAS), an enhanced model named ED-LSTM-FCNN is constructed, incorporating an embedding layer module to handle high-dimensional spatial and temporal features. A fully connected neural network was utilized to integrate various features types and achieve regression prediction of temperature, generating gridded hourly temperature forecast products with a resolution of 0.05°×0.05°. Verification for the 2022 forecast in Hunan Province revealed that the model exhibits a notable capacity to mitigate forecast errors inherent in the numerical model, thereby enhancing the overall forecast stability. The root mean square errors (RMSE) for forecast lead times ranging from 1 to 24 hours exhibit a reduction of 25.4% to 37.7% when compared to ECMWF-IFS and a decrease of 15.8% to 40.0% in comparison to the SCMOC. The model significantly enhances the forecast performance of ECMWF-IFS in spatial prediction, particularly in regions characterized by intricate terrain features. The RMSEs across most areas vary within the range of 1.2 ℃ to 1.6 ℃. The forecast accuracy of the model, with an error margin of ±2 ℃, surpasses 85.0% across various seasons, demonstrating a significant improvement compared to both ECMWF-IFS and SCMOC. The forecasting performance is notably superior, particularly in stable extreme high-temperature weather conditions, when compared to alternative products. In conclusion, this method proved to be effective for high-resolution temperature grid forecasting operations.