Influence of missing wind measurements on wind turbine power production using various measure-correlate-predict methods and reanalysis datasets

This study analyzes the impact of missing wind data points on the accuracy of Annual Energy Production (AEP) estimation in wind resource assessment (WRA). Evaluations were made under different scenarios using various Measure-Correlate-Predict (MCP) methods and reanalysis datasets. One year of wind measurements was collected from an inland met mast located in the Gashiri area of Jeju Island, South Korea. Three types of long-term reanalysis datasets- ERA-5, MERRA-2 and WRF (ERA-5)- were obtained, each exhibiting different levels of correlations with the met mast wind measurements. To simulate missing data points scenarios, a yearly percentage sampling method was applied to the one-year met mast wind data with sampling rates ranging from 10% to 90%. To ensure statistical reliability, random sampling was performed 12 times for each sampling rate. The MCP method was applied after pairing each sampled dataset with the reanalysis datasets. Long-term predictions were generated using four MCP approaches- two machine learning techniques (Random Forest and Gradient Boosting Regression) and two traditional methods (Regression and Matrix). AEP was calculted from these predictions and compared to the reference AEP derived from the complete dataset. Results show that accurate AEP estimation remained feasible even when using reanalysis datasets with low correlation to the measured data. Moreover, all four MCP methods demonstrated similar performance, with machine learning–based approaches producing results comparable to those of traditional methods. While conventional WRA practice recommends a data recovery rate above 90% for accurate AEP estimation, this study demonstrated reliable AEP estimates could be achieved with rates as low as 50%.