Extended Abstract (744K)19A.2
The Application of an Evolutionary Algorithm to the Optimization of a Mesoscale Meteorological Model
David W. Werth, Savannah River National Laboratory, Aiken, SC; and L. O'Steen
Improvement in mesoscale atmospheric model simulations is often thought to be primarily a matter of finer spatial resolution. While this is generally true, there is a limit to the improvement one can obtain by simply decreasing the grid size of a numerical model. Further improvements in forecasts may be achieved with better model parameterizations, but this leaves the mesoscale modeler with the task of determining which parameterizations to use for a specific problem and what values to use for individual model parameters. The accuracy of a given numerical simulation is often a matter of a judicious choice of these values.
In this presentation, we show how a simple evolutionary programming (EP) algorithm can optimize a given set of parameters in a mesoscale atmospheric model with respect to agreement between simulation and observations. This is illustrated using the Regional Atmospheric Modeling System (RAMS). As an initial test case, data from a RAMS simulation with a default set of parameters, rather than actual data, is used to create a single sounding as the 'observation'. The process begins with the random assignment of parameters to the model. The model is run, and the results are used to calculate an objective function based on the weighted rms error between the simulated and 'observed' soundings. Multiple runs are done with different parameter sets, and the sets with the lowest objective functions are selected to become 'parents' of the next set. Ideally, the model parameters will evolve toward those of the default simulation as the objective function is minimized. This type of experiment also tests the ability of EP to find the global minimum of the objective function since the optimum is known.
Our primary goal was to demonstrate that an EP algorithm can provide a systematic and objective method for optimizing a mesoscale atmospheric model. We are now working towards exploring the path the model follows as it is pushed towards its target, and how such a scheme could be applied operationally.
Session 19A, Modeling Part II
Friday, 5 June 2009, 10:30 AM-12:00 PM, Grand Ballroom East
Previous paper Next paper
Recorded presentation