Abstract:The refinement level of wind forecast by numerical prediction models cannot meet the needs of inland waterway transportation, and its adaptability to different regions varies. This article takes most of western hilly area and central plain area in Hubei Province, including parts of the Yangtze River waterway, as the research areas. Referring to the 10 m wind real-time product of ART_1 km, this article analyzes the adaptability of 10 m wind forecast of the European Centre for Medium-Range Weather Forecasts high-resolution atmospheric model deterministic forecasting product (EC-HRES) and the China Meteorological Administration mesoscale model forecasting product (CMA-MESO) in the research areas. A U-Net++ deep convolutional network model is constructed to achieve downscaling correction of wind speed forecast. The correction model improves the sampling module and adds waterway item and terrain item into the loss function, enhancing expressive ability and robustness of the model and improving the correction effect on the waterway. The verification result shows that this method can effectively reduce the prediction error of wind speed made by numerical prediction models in the waterway areas.