A climate-informed framework for farm-level benchmarking of offshore wind power using ERA5 and generation data

Farm-level benchmarking of offshore wind power from public generation records is challenged by changes in available capacity, operational anomalies, and meteorological variability. This study develops a climate-informed framework that combines ENTSO-E generation records with ERA5 reanalysis across 57 hourly generation traces from 40 European offshore wind farms. A rolling-capacity preprocessing procedure harmonizes records across sites, while two independently implemented nonparametric estimators, k -nearest neighbors (KNN) and Additive Multiplicative Kernel (AMK) regression, quantify the incremental predictive information contained in 13 atmospheric descriptors under nested, time-aware blocked validation. Wind speed provides the dominant baseline, wind direction is the most reproducible additional predictor, and wind gust adds a smaller complementary contribution; no fourth variable satisfies the prespecified selection criterion. On held-out observations, the speed–direction–gust model reduces hourly NRMSE by 4.86% for KNN and 5.15% for AMK relative to wind speed alone, with corresponding daily reductions of 10.57% and 11.04%. The directional contribution is strongly site dependent yet highly reproducible across estimators. The predictor hierarchy remains stable across temporal validation designs, preprocessing choices, directional representations, and hub-height wind-speed extrapolation, supporting a scalable and interpretable basis for comparative offshore wind-farm benchmarking using public data.