Title:Prediction of regional seasonal fluctuations in precipitation based on chaos theory

Abstract:In the past decade, the combined effect of flood and drought resulted in the loss of thousands of lives and billions of dollars. Multi season ahead prediction of regional precipitation extremes could significantly reduce losses. However, the evolution of climate is highly sensitive to initial conditions, or chaotic, so practical long term prediction of precipitation in time is impossible. Adding to the difficulty, the climate system is non-stationary; with the energy available to move water and air as tracked by global average surface temperature (GAST) increasing over the last several decades2. Neither purely empirical autoregression, nor global circulation models (GCM) are sufficiently accurate. Here I use statistical methods motivated by chaos theory to predict seasonal fluctuations in regional and local precipitation with high correlation. The change in GAST is accommodated using special runs of a global circulation model to build an initial set of predictive models, while ground data is used to train, combine, and calibrate them. In examples I show seasonal prediction of precipitation in a few geographical regions with high statistical significance. In one region were precipitation rose near the 4 sigma level above the mean, the correlation was above 0.8. Also, I examine one region over longer time tentatively identifying a tight coupling between GAST and patterns of climate anomalies, with implications for attribution. In addition, being able to invert climate patterns to identify parameters of the dynamic process holds promise for allowing statistical evaluation of parameterizations of climate models. I expect these methods may be applicable both to a number of other measures of climate and weather as well as other high dimensional chaotic systems.