Errors in Hailpad Data Reduction

Abstract

A study of the hailfall parameter estimates determined by 11 persons (readers) measuring the hailstone dents on the same aluminum-covered styrofoam hailpads has shown significant differences among many of the readers. The hailpads in the study were collected by the National Hail Research Experiment (NHRE) in 1976. The results apply directly to the pads and the method of data reduction used by the NHRE, but have implications for other projects using hailpad data. The experiment analyzed the estimates of five hail parameters on each of 23 hailpads reduced by the same 11 persons. Point estimates of parameters such as hail mass were found to have relative standard deviations of about 30% when determined by one reader. The effect of using multiple readings is to reduce the error by the factor n-m, where n is the number of readers. The total error of each reader was decomposed into systematic and random components by analysis of variance. The standard deviation of the systematic component was -40% of that of the random component for allfive hail parameters. Analysis of individual size category counts determined by the same readers on the same pads showed that small counts, usually associated with the largest hail sizes, could be in error by as much as 100% while large counts (near 1000) had standard deviation of only 16%. The reader standard deviation exceeded the expected Poisson error for mean counts per category greater than 2. A theoretical statistical model was used to examine the effect of reader variance on areal estimates of hailfall. Areal estimates were found to be much less affected by reader variation than point estimates, notably when the number of hailpads hit by hail exceeded 30. Further, since the majority of the readererror was random rather than systematic, it was not very helpful in reducing the error to divide the pad reading for a hail day among several readers. In evaluating parameters such as mass for an area in a hail suppression experiment it was concluded that multiple readings were not required.