ATM-9310959 Tsonis, A.A. University of Wisconsin-Milwaukee Title: Climate Diagnostics and Prediction Using Nonlinear Empirical Models ABSTRACT This project's main objectives are to develop and test nonlinear statistical models for climate diagnostics and for climate predictions. The methodologies to be applied, based on artificial neural networks and local-approximation models, have been developed recently in the emerging fields of deterministic chaos and parallel distributed processing. The goal is to achieve skillful models of sea surface temperature (SST) anomalies on seasonal to interannual time scales for both the Pacific and Atlantic basins and, through prediction, to enhance understanding of the processes involved. Climate variability on the seasonal to interannual time scale is strongly related to ENSO. Recent research suggests that ENSO, primarily a Pacific signal, can be considered as a low-dimensional chaotic system. In addition, the Atlantic region also has major climate variations shown in atmospheric and oceanic data. Predictions of both ENSO and Atlantic oscillations will be attempted by applying the nonlinear statistical modeling techniques developed for the prediction of chaotic system behavior. The performance of several nonlinear statistical models will be validated and optimized through hindcast experiments with data sets from as early as 1950. The absolute verification experiments will be hindcasts of ENSO SST events in the period of 1985-1990, as established by the Working Group of the TOGA Program on Seasonal to Interannual Prediction (T-POP). The algorithms developed, particularly those for the Connection Machine, are based on new modeling techniques and will be made readily available to the meteorological and oceanographic communities.
