Use of a Genetic Algorithm to Optimize a Numerical Weather Prediction System
- 1 April 2021
- journal article
- research article
- Published by American Meteorological Society in Monthly Weather Review
- Vol. 149 (4), 1089-1104
- https://doi.org/10.1175/mwr-d-20-0238.1
Abstract
An important step in an ensemble Kalman filter (EnKF) algorithm is the integration of an ensemble of short-range forecasts with a numerical weather prediction (NWP) model. A multiphysics approach is used in the Canadian global EnKF system. This paper explores whether the many integrations with different versions of the model physics can be used to obtain more accurate and more reliable probability distributions for the model parameters. Some model parameters have a continuous range of possible values. Other parameters are categorical and act as switches between different parameterizations. In an evolutionary algorithm, the member configurations that contribute most to the quality of the ensemble are duplicated, while adding a small perturbation, at the expense of configurations that perform poorly. The evolutionary algorithm is being used in the migration of the EnKF to a new version of the Canadian NWP model with upgraded physics. The quality of configurations is measured with both a deterministic and an ensemble score, using the observations assimilated in the EnKF system. When using the ensemble score in the evaluation, the algorithm is shown to be able to converge to non-Gaussian distributions. However, for several model parameters, there is not enough information to arrive at improved distributions. The optimized system features slight reductions in biases for radiance measurements that are sensitive to humidity. Modest improvements are also seen in medium-range ensemble forecasts. An important step in an ensemble Kalman filter (EnKF) algorithm is the integration of an ensemble of short-range forecasts with a numerical weather prediction (NWP) model. A multiphysics approach is used in the Canadian global EnKF system. This paper explores whether the many integrations with different versions of the model physics can be used to obtain more accurate and more reliable probability distributions for the model parameters. Some model parameters have a continuous range of possible values. Other parameters are categorical and act as switches between different parameterizations. In an evolutionary algorithm, the member configurations that contribute most to the quality of the ensemble are duplicated, while adding a small perturbation, at the expense of configurations that perform poorly. The evolutionary algorithm is being used in the migration of the EnKF to a new version of the Canadian NWP model with upgraded physics. The quality of configurations is measured with both a deterministic and an ensemble score, using the observations assimilated in the EnKF system. When using the ensemble score in the evaluation, the algorithm is shown to be able to converge to non-Gaussian distributions. However, for several model parameters, there is not enough information to arrive at improved distributions. The optimized system features slight reductions in biases for radiance measurements that are sensitive to humidity. Modest improvements are also seen in medium-range ensemble forecasts.Keywords
This publication has 21 references indexed in Scilit:
- Parameter estimation using data assimilation in an atmospheric general circulation model: From a perfect toward the real worldJournal of Advances in Modeling Earth Systems, 2013
- Estimating Model Parameters with Ensemble-Based Data Assimilation: A ReviewJournal of the Meteorological Society of Japan. Ser. II, 2013
- NWP model forecast skill optimization via closure parameter variationsQuarterly Journal of the Royal Meteorological Society, 2012
- Ensemble prediction and parameter estimation system: the conceptQuarterly Journal of the Royal Meteorological Society, 2011
- Ensemble prediction and parameter estimation system: the methodQuarterly Journal of the Royal Meteorological Society, 2011
- Toward Random Sampling of Model Error in the Canadian Ensemble Prediction SystemMonthly Weather Review, 2010
- Model Error Representation in an Operational Ensemble Kalman FilterMonthly Weather Review, 2009
- Strictly Proper Scoring Rules, Prediction, and EstimationJournal of the American Statistical Association, 2007
- A generalized approach to parameterizing convection combining ensemble and data assimilation techniquesGeophysical Research Letters, 2002
- Novel approach to nonlinear/non-Gaussian Bayesian state estimationIEE Proceedings F Radar and Signal Processing, 1993