Hybrid error correction model to enhance long-term wind field forecasting performance of coastal bridges by the WRF model

Abstract

Wind is a dominant load for coastal bridges, yet long-term wind field data are often lacking, making accurate design wind speed evaluation challenging. To efficiently and accurately predict long-term wind fields for bridges, a hybrid error correction model is proposed by integrating machine learning (ML) methods with the inverse distance weighting (IDW) method. First, an error model is established between the measured meteorological data and the parameter-optimized weather research and forecasting (WRF) data. Furthermore, the WRF-based data as model inputs with corresponding error data serving as target outputs are integrated through various ML algorithms. An error correction model for the selected bridge is developed by integrating the optimal ML approach with the IDW method. Finally, the validity of the proposed methodology is successfully verified using short-term field measurements. A 10-year bridge wind field is generated through the proposed hybrid error correction model. Results demonstrated that a set of common WRF physical parameterization schemes was recommended for long-term wind field simulation of coastal bridges. Compared with the raw WRF output, the novel hybrid model achieved improvements surpassing 30% in terms of root mean square error. Unlike general meteorological sites, coastal bridges far above the sea surface featured sharp vertical wind gradients. The proposed framework explicitly accounted for this by incorporating site-specific observations and WRF output into the ML–IDW correction layer.