Evaluation and bias correction of reanalysis wind data for resource assessment and long-term power prediction

Reliable long-term wind-power prediction requires linking short-term mast measurements with long-term reanalysis data. This study developed integrated methodology using measurements from five coastal and near-coastal sites, four reanalysis datasets, seven Power Law exponent methods, and seven bias-correction techniques. The proposed Day–Hour Lookup-Table Power Law produced lowest average mean absolute error, root mean square error (RMSE) and standard deviation of error of 0.23, 0.35 and 0.33 m/s, respectively, and maintained relative RMSE of 4.31–6.30% during chronological hold-out validation. ERA5 and MERRA-2 outperformed their predecessors, while ERA5 showed best performance and lower validation errors after correction. Linear Least Squares reduced relative RMSE for all ten site–dataset combinations and decreased the mean absolute difference between short-term and 20-year capacity factors from 5.6 to 1.4 percentage points. The corrected 20-year capacity factor of best-performing 4.5 MW turbine ranged from 49.8 to 65.6%. The methodology supports wind-resource and power assessment in data-scarce regions.