Estimation of non-uniform wind field over a meandering reservoir

This study evaluated a mass-conserving model (WindNinja) and deep learning model to produce a non-uniform wind field over a meandering reservoir using wind data from AMeDAS and the in-situ wind data at certain locations over the surface. Results showed that WindNinja can estimate wind magnitude and direction for the stations near the input AMeDAS station. On the other hand, stations that are far from the input data location yielded less accurate results. Moreover, the wind estimate using various cases of deep learning models showed better results than the WindNinja simulation. The prediction of wind magnitude using deep learning model is less affected by the type of input and output parameters. On the contrary, the prediction accuracy for wind direction significantly changes for each case. Finally, deep learning models that utilize the results from the WindNinja simulation were also considered and yielded the most accurate wind prediction. The overall results of this study proved that an acceptably accurate non-uniform wind field can be generated using a mass-conserving wind model and deep learning model despite the limitation in the amount of available data.