Improving the Statistical Reliability of Data Analysis from Atmospheric Measurements and Modeling

Abstract

Statistical issues of atmospheric data analysis are discussed that address the problem of stationarity and homogeneity of data and the problem of inadequate record lengths. Bandpass filtering of observational data is proposed to make possible the reliable comparisons with model output statistics. Another suggestion is box area measurements that offer considerable advantages in terms of the accuracy of estimation over linear flight path data. Records of stationary data of adequate lengths are unavailable for higher-order statistics, but sufficient amounts of box area data can be obtained from limited domains of 20?40 km. The findings are illustrated by the analyses of data from Project LESS (Lake-Effect Snow Studies) and from large eddy simulation (LES) of Project LESS events.