Parallel Multi-level Genetic Ensemble for Numerical Weather Prediction Enhancement

The need for reliable predictions in environmental modelling is well-known. Particularly, the predicted weather and meteorological information about the future atmospheric state is crucial and necessary for almost all other areas of environmental modelling. Additionally, right decisions to prevent damages and save lives could be taken depending on a reliable meteorological prediction process. Lack and uncertainty of input data and parameters constitute the main source of errors for most of these models. In recent years, evolutionary optimization methods have become popular to solve the input parameter problem of environmental models. We propose a new parallel meteorological prediction scheme that uses evolutionary optimization methods based on Multi-Chromosome Genetic Algorithm to enhance the quality of weather forecasts by focusing on the calibration of input parameters. This new scheme is parallelized and executed on a HPC environment in order to reduce the time needed to obtain the final prediction. The new approach is called Multi-Level Genetic Ensemble (M-Level G-Ensemble) and it has been tested using historical data of a wellknown weather catastrophe: Hurricane Katrina that occurred in 2005 in the Gulf of Mexico. Results obtained with our approach provide both significant improvements in weather prediction and a significant reduction in the execution time.