A validation technique for assessing regional climate statistics from numerical models.

摘要

Validation studies are needed to determine the quality of model-generated local climate statistics. There is a problem with the direct comparison of model data with raw observations, since a model produces results which are spatially filtered due to the minimum resolution of the grid and spatial derivative schemes used in the model. A validation technique has been developed to compare model simulations and observations in order to determine the accuracy of model-generated climate statistics. This technique allows a direct comparison between model and observations at individual grid points.; This technique involves a comparison of the scale-adjusted station probability distribution functions (PDFs) for various meteorological fields, with those generated by the model at a grid point. Consequently, any deviations will be due to model errors or observational limitations, and are not a result of the scale differences between the model and the observations. This technique relies on the central limit theorem, and while the central limit theorem defines the limiting distribution (i.e., Gaussian) it does not provide the path to the Gaussian distribution. Geostatistical techniques provide a quantifiable method for establishing this evolutionary process by allowing estimates of the intermediate distribution. This technique determines a transform function which expresses a random variable in terms of a standard normal variate using Hermite polynomials. The transformation process assumes permanence of distribution, stating that the point and transformed distributions can be described by the same function. As a result, the observations can be transformed to produce a distribution which has the characteristics of the observations with model resolution.; Results indicate that the technique is capable of accounting for the effects of filtering in model predictions. In addition, this method points out the possible problems with the model which would have otherwise been undetected. Furthermore, this technique can be applied to any variable, scale, and grid point or spectral model.