Using regression and neural networks to reconstruct winter circulation indices and precipitation in the Southwest

The short time period for which instrumental records exist limits our knowledge of long-term climate variability. The American Southwest is rich in long tree-ring chronologies, many of which have been little affected by interactions between neighbors and contain particularly clean signals. The aim of this research is to link tree-ring variability with atmospheric circulation/precipitation and to use these linkages to reconstruct the winter climate of the southwest in the last 1000 yrs. Ninety years of observed records are used to derive reconstruction models based on linear regression (LR) and nonlinear artificial neural networks (ANN) techniques. Preliminary results indicate that the ANN model performs better than the LR for both the circulation indices and the precipitation reconstructions. In the last 1000 years the precipitation in the southwest shows a statistically significantly linkage to several indices such as, the Pacific Decadal Oscillation (PDO), the North Pacific (NP) pattern and the southern oscillation index (SOI).