Predicting the global temperature with the Stochastic Seasonal to Interannual Prediction System (StocSIPS)

Many atmospheric fields-in particular the temperature-respect statistical symmetries that characterize the macroweather regime, i.e. time-scales between the approximate to 10 day lifetime of planetary sized structures and the (currently) 10-20 year scale at which the anthropogenic forcings begin to dominate the natural variability. The scale-invariance and the low intermittency of the fluctuations implies the existence of a huge memory in the system that can be exploited for macroweather forecasts using well-established (Gaussian) techniques. The Stochastic Seasonal to Interannual Prediction System (StocSIPS) is a stochastic model that exploits these symmetries to perform long-term forecasts. StocSIPS includes the previous ScaLIng Macroweather Model (SLIMM) as a core model for the prediction of the natural variability component of the temperature field. Here we present the theory for improving SLIMM using discrete-in-time fractional Gaussian noise processes to obtain an optimal predictor as a linear combination of past data. We apply StocSIPS to the prediction of globally-averaged temperature and confirm the applicability of the model with statistical testing of the hypothesis and a good agreement between the hindcast skill scores and the theoretical predictions. Finally, we compare StocSIPS with the Canadian Seasonal to Interannual Prediction System. From a forecast point of view, GCMs can be seen as an initial value problem for generating many stochastic realizations of the state of the atmosphere, while StocSIPS is effectively a past value problem that estimates the most probable future state from long series of past data. The results validate StocSIPS as a good alternative and a complementary approach to conventional numerical models. Temperature forecasts using StocSIPS are published on a regular basis in the website: http://www.physics.mcgill.ca/StocSIPS/.