A stochastic approach for assessing the effect of changes in synoptic circulation patterns on gauge precipitation

A stochastic model is described that allows transfer of information from general circulation models to precipitation gauge locations using a weather state classification scheme. The weather states, which are based on present and previous day's sea level pressure, are related stochastically to gauge daily precipitation and temperature. Weather states are defined to give maximal separation of test station precipitation distributions using the Classification and Regression Trees procedure. Precipitation amounts are resampled from historical observations, conditional on the weather state and the previous day's rain state. Daily temperature maxima and minima are simulated by conditioning on the present and previous day's rain state, with the residual modeled as a first-order autoregressive process. The model parameters are estimated using 9 years (1965–1973) of concurrent U.S. National Meteorological Center (NMC) gridded observations and four precipitation stations in the Columbia River Basin. The model is illustrated using both historical NMC gridded observations of sea level pressure and lower atmosphere temperature, as well as the same variables from the General Fluid Dynamics Laboratory general circulation model for present climate and CO2 doubling. A 40-year sequence of simulated precipitation and temperature is used to estimate seasonal streamflows and flood frequency distributions under present and doubled CO2 climates.