DEVELOPMENT OF A RELATIONSHIP BETWEEN STATION AND GRID-BOX RAINDAY FREQUENCIES FOR CLIMATE MODEL EVALUATION

The validation of climate model simulations creates substantial demand s for comprehensive observed climate datasets. These datasets need not only to be historically and geographically extensive, but need also t o be describing areally averaged climate, akin to that generated by cl imate models. This paper addresses one particular difficulty found whe n attempting to evaluate the daily precipitation characteristics of a global climate model, namely the problem of aggregating daily precipit ation characteristics from station to area. Methodologies are develope d for estimating the standard deviation and rainday frequency of grid- box mean daily precipitation time series from relatively few individua l station time series. Temporal statistics of such areal-mean time ser ies depend on the number of stations used to construct the areal means and are shown to be biased (standard deviations too high, too few rai ndays) if insufficient stations are available. It is shown that these biases can be largely removed by using parameters that describe the sp atial characteristics of daily precipitation anomalies. These spatial parameters (the mean interstation correlation between station time ser ies and the mean interstation probability of coincident dry days) are calculated from a relatively small number of available station time se ries for Europe, China, and Zimbabwe. The relationships that use these parameters are able to successfully reproduce the statistics of grid- box means from the statistics of individual stations. They are then us ed to estimate the statistics of grid-box means as if constructed from an infinite number of stations (for standard deviations) or 15 statio ns (for rainday frequencies), even if substantially fewer stations are actually available. These estimated statistics can be used for the ev aluation of daily precipitation characteristics in climate model simul ations, and an example is given using a simulation by the Commonwealth Scientific and Industrial Research Organisation atmosphere general ci rculation model. Applying the authors' aggregation methodology to obse rved station data is a more faithful form of model validation than usi ng unadjusted station time series.