A Machine Learning Approach to Correct Turbulence Intensity measured by Floating Lidars

Abstract

Abstract In this work we introduce a supervised Machine Learning (ML) model to correct the Turbulence Intensity (TI) measured by Floating LiDAR Systems (FLS) in offshore environment. The model was developed using data from 46 EOLOS-FLS200 validation campaigns (≈ 4.6 years) carried out at three reference sites in the North Sea. It is based on Gradient Boosting Decision Tree (GBDT) and accounts for wind characteristics, atmospheric conditions, buoy motion, and wave features. Numerical analyses pronounced a consistent improvement in both coefficient of determination (R 2) and Mean Bias Error yielded by the ML-corrected TI. In addition, TI estimates in accordance with the state-of-the-art best practices were successfully obtained, even when evaluating the ML model in a site out of the training dataset, which demonstrates the model’s robustness.