An improved rainfall disaggregation technique for GCMs

Abstract

Meteorological models represent rainfall as a mean value for a grid square so that when the latter is large, a disaggregation scheme is required to represent the spatial variability of rainfall. In general circulation models (GCMs) this is based on an assumption of exponentiality of rainfall intensities and a fixed value of areal rainfall coverage, dependent on rainfall type. This paper examines these two assumptions on the basis of U.K. and U.S. radar data. Firstly, the coverage of an area is strongly dependent on its size, and this dependence exhibits a scaling law over a range of sizes. Secondly, the coverage is, of course, dependent on the resolution at which it is measured, although this dependence is weak at high resolutions. Thirdly, the time series of rainfall coverages has a long-tailed autocorrelation function which is comparable to that of the mean areal rainfalls. It is therefore possible to reproduce much of the temporal dependence of coverages by using a regression of the log of the mean rainfall on the log of the coverage. The exponential assumption is satisfactory in many cases but not able to reproduce some of the long-tailed dependence of some intensity distributions. Gamma and lognormal distributions provide a better fit in these cases, but they have their shortcomings and require a second parameter. An improved disaggregation scheme for GCMs is proposed which incorporates the previous findings to allow the coverage to be obtained for any area and any mean rainfall intensity. The parameters required are given and some of their seasonal behavior is analyzed. Copyright 1998 by the American Geophysical Union.