Deep Learning-Based Prediction of Black Sea Nearshore Wind

Machine learning technologies are increasingly being used in data analysis as potential tools for assessing natural resources. The goal of this study is to determine how successfully the recently released Time Series Modeller (MATLAB R2026a) predicts wind conditions in the Black Sea basin’s nearshore zones. Ten years of hourly ERA5 data (2016–2025) were considered, with the predicted parameters being related to wind speed (at 100 m height) and the corresponding wind direction. A number of deep learning models were evaluated, and the findings were presented using statistical metrics such as RMSE (Root Mean Squared Error), relative error, and the Weibull distribution. The ERA5 and anticipated wind speed showed fair agreement, with the exception of extremely low (<3 m/s) and high wind speeds (>18 m/s), which may have been inflated by the statistical indicator used. The wind dispersion was examined on a monthly and hourly basis. In terms of wind direction, it was discovered that deep learning models are unable to replicate wind conditions from the north sector; nevertheless, this appears to be an issue with models based on RMSE prediction.