Determining variabilities of non-Gaussian wind-speed distributions using different metrics and timescales

Quantification of long-term wind-speed variability is a critical component in wind resource assessment, and effective wind-farm operations require proper assessment of this variability. Yet, wind-speed variations differ across averaging temporal scales because hourly mean wind speeds fluctuate more than yearly averages. In this study, we quantify the influence of averaging timescale to the resultant variability. We assess three spread metrics (standard deviation, coefficient of variation, and robust coefficient of variation) and two distribution measures (skewness and kurtosis) based on 38 years of wind speeds from the National Aeronautics and Space Administration's MERRA-2 reanalysis data set over the contiguous United States. The spatial distributions of wind-speed variability differ with metrics and timescales: wind speeds of fine temporal resolution generate strong variabilities that dilute spatial contrasts; small sample size becomes a constraint in calculating interannual variabilities via annual means and leads to inaccurate results. Overall, we find that metrics based on monthly data portray the largest spatial differences of wind-speed variability. Although standard deviation yields consistent geographical projections, none of the wind-speed data of any time frame are perfectly Gaussian. Therefore, the robust coefficient of variation, a statistically robust and resistant approach, appears to be the ideal metric for quantifying wind-speed variabilities based on monthly mean data.