David Aristoff's homepage Title: Superparameterization and Data Assimilation

Abstract: Superparameterization (SP) is a multiscale computational approach wherein a large scale atmosphere or ocean model is coupled to an array of simulations of small scale dynamics on periodic domains embedded into the computational grid of the large scale model. SP has been successfully developed in global atmosphere and ocean models. It has been particularly successful in enabling realistic simulations of the Madden-Julian Oscillation (MJO), among other things. The MJO is an intraseasonal aperiodic cycle of variability in precipitation over the tropical Indian and Pacific oceans, with teleconnections to weather across the globe, from rainfall in West Africa to tornadoes over the continental US. Accurate forecasts with SP models could be very useful, but SP models are not currently used for forecasting because of the difficulty in initializing the multiscale model. This talk develops a 3D-Var variational data assimilation framework for use with SP. The relatively low cost and simplicity of 3D-Var in comparison with ensemble approaches makes it a natural fit for relatively expensive multiscale SP models. The atmospheric modeling community also has extensive experience with 3D-Var assimilation, which sets a relatively low bar for implementing the framework in an SP model. A new, simple multiscale model is developed to demonstrate the assimilation framework. The new system consists of a system of ordinary differential equations similar to the Lorenz-`96 model. The system has one set of variables denoted Y_i; large and small scale parts are defined using the discrete Fourier basis; and the SP approximation to the system is straightforward. The system exhibits dynamics reminiscent of the tropical atmosphere, with small-scale fast `convection' and large-scale wave-like propagation. With the new assimilation framework the SP model approximates the large scale dynamics of the true system even more accurately than an ensemble Kalman filter using the true (non-SP) dynamics.