This paper leveraged a large anemometric dataset from 21 landfalling Atlantic tropical cyclones (TCs) to investigate the suitability of two popular machine learning (ML) approaches, namely artificial neural networks (ANN) and support vector regression (SVR), for accurately predicting vertical wind turbulence in the atmospheric surface layer (ASL). The dataset comprised 3013 10-min wind speed records taken at 5 m and 10 m heights and collected by portable weather stations as part of the Florida Coastal Monitoring Program (FCMP) between 1999 and 2018. Input features to the ML models were limited to longitudinal wind flow velocity statistics, while model outputs consisted of normalized friction velocity and vertical turbulence intensity predictions. A robust nested Monte Carlo cross-validation technique was applied to extract uncertainty measures and assess overall ML performance. ML-based predictions for unseen FCMP data subsets agreed well with field observations, particularly for 10 m measurements. However, ML performance metrics for vertical turbulence intensity predictions were consistently superior to friction velocity estimates, and better ML accuracy was found for extreme wind speed records (>45 m/s). Findings of this work can be applied to infer statistics of TC-induced vertical turbulent fluxes in the ASL from limited (or incomplete) wind speed records.
扫码关注我们
求助内容:
应助结果提醒方式:
