A Comparison of Errors in objectively Analyzed Fields for Uniform and Nonuniform Station Distributions

Abstract

An error analysis of the Purdue Regional Objective Analysis of the Mesoscale (PROAM) scheme is described. PROAM is an interactive, multiple iteration objective analysis package for surface meteorological data. It utilizes a Gaussian weighting function similar to that of Barnes. A problem with applying interpolation routines to surface data is that the surface station network has a highly nonuniform spatial distribution. Smith and Leslie have shown that this nonuniformity can impact the accuracy of the analysis. The objective of this investigation is to determine quantitatively the contribution that this nonuniformity of station locations has on the analysis of data for two different distributions of stations. Experiments were done by varying the values of important analysis parameters in order to generate error curves for each parameter. Analytic functional representations for pressure and temperature were used to generate input data free of observational error, as well as to produce exact analyses. With exact fields available, root-mean-square errors and average grid point errors were calculated for analyses using uniform, actual and edited station distributions. Results show that the PROAM scheme performs a better interpolation of the observations when the stations are more uniformly distributed. Root-mean-square and average grid point errors for analyzed fields decrease as the uniformity of the station distribution increases. Furthermore, elimination of some data stations (when several are clustered within a small area) can result in an analysis with smaller errors than a field with more stations but a less uniform distribution. The error analysis performed in this study suggests methods which can minimize errors in the analyzed fields.