Bias Correction to GEV Shape Parameters Used to Predict Precipitation Extremes

Abstract

Previous studies have shown that the shape parameter of the generalized extreme-value (GEV) distribution, a distribution commonly applied in precipitation frequency analysis, is difficult to estimate from short records because of sampling bias. This bias is confirmed in the present study using annual precipitation maxima data from NOAA Atlas 14, which covers much of the United States and its territories. The bias depends on record length as well as on the magnitude of the population shape parameter. Correcting sample shape parameter estimates for bias yields extreme precipitation quantiles that are more conservative (i.e., higher) than would be obtained without a correction. For example, in a test case presented here, the bias correction increased the 1- and 24-h 10,000-year quantiles by 8 and 10%, respectively. This paper presents an expression and coefficients that may be used to develop approximately unbiased GEV shape parameter estimates for regions covered by NOAA Atlas 14.