Conditioning stochastic properties of daily precipitation on indices of atmospheric circulation

The relationship between precipitation and large-scale measures of atmospheric circulation has received much attention in recent years (e.g. Bardossy & Caspary, 1990; Bardossy & Plate, 1992; Hughes et al., 1993; Woolhiser et al., 1993; Katz & Parlange, 1993). Some of the interest relates to quantifying the possible change in atmospheric circulation within the past two decades, with associated changes to precipitation occurrence and intensity (Bardossy & Caspary, 1990; Kiely et al., 1998). Other interest is due to the fact that General Circulation Models (GCMs) can provide sea level pressure fields for future climate conditions (e.g. under increased greenhouse gas concentrations). However, because of their coarse resolution (e.g. there might be only one grid point for Ireland), they cannot adequately represent local precipitation. A stochastic precipitation model, conditioned on a suitable atmospheric circulation parameter/index, could be used in simulating monthly precipitation distributions. A comparison of unconditioned chain-dependent models for daily precipitation with conditioned models showed the latter to be an important improvement for the modelled stochastic precipitation process (e.g. Katz & Parlange, 1993). It is common in hydrology to fit stochastic models to the full time series of precipitation amounts. However, precipitation amount lends itself to splitting (or conditioning) the time series into two (e.g. the part of the series of lower precipitation amounts associated with high sea level pressure and the other part of the time series of higher precipitation amounts associated with low sea level pressure). A suitable atmospheric circulation index on which to condition precipitation could include among others: sea level pressure (SLP), geostrophic wind direction (GWD), sea surface temperature (SST) and a Lamb (1950) type wind direction classification index. Much use has been made of the ‘El Niño’ or Southern Oscillation phenomenon as a conditioning index in regions known to be sensitive. The North Atlantic Oscillation has also been used as a circulation index (Hurrell, 1995). Katz & Parlange (1993), using precipitation data for California, developed two conditioned daily models, which differ in terms of the parameters of the occurrence process and the intensity process. Each of these daily effects contributes to changes in the distribution of monthly total precipitation associated with the SLP circulation index. They showed how a daily conditioned model could be used to induce an improved monthly model. The induced model is a combination of both conditioned models Meteorol. Appl. 5, 75–87 (1998)