Stochastic Characterization of Regional Circulation Patterns for Climate Model Diagnosis and Estimation of Local Precipitation

Abstract

Two statistical approaches for linking large-scale atmospheric circulation patterns and daily local rainfall are applied to GCM (general circulation model) climate simulations. The ultimate objective is to simulate local precipitation associated with altered climate regimes. Two regions, one in the Pacific-American sector (western region) and one in the American-Mid-Atlantic sector (eastern region), are explored. The first method is based on Classification and Regression Trees (CART) analysis. The CART method classifies observed daily sea level pressure (SLP) fields into weather types that are most strongly associated with the presence/absence of rainfall at selected index stations. After applying this method to historical SLP observations, precipitation simulations associated with GCM SLP output were validated in terms of probability of occurrence and survival time of the weather states identified by the CART analysis. Daily rainfall time series were then generated from weather classes derived by application of CART to both daily SLP fields derived from historical observation and from GCM simulations. While the mean rainfall and probability distributions were rather well replicated, the precipitation generator based on this version of the CART technique had two important deficiencies: the generated dry periods were too short, on average, and the identification of weather states may be not invariant under coordinate rotations. The second rainfall generator is based on the analog method and uses information about the evolution of the SLP field from several previous days. It considers a pool of past observations for the circulation patterns closest to the target circulation. It is similar to the CART method and in certain aspects it performs better, although some downward bias in the simulated rainfall persistence was still present. Applying both methods to the output of a 2 ? CO2 GCM simulation produced only small changes in simulated precipitation, which is due to the small sensitivity of this variable to greenhouse forcing. The selection characteristics of the analogs are similar for observations, a control run, and a 2 ? CO2 run, indicating that analogs for possible altered climates can be found in the historical record.