The dimension-reduced projection four-dimensional variational data assimilation （DRP-4DVar） approach utilizes the ensemble of historical forecasts to estimate the background error covariance （BEC） and directly obt...The dimension-reduced projection four-dimensional variational data assimilation （DRP-4DVar） approach utilizes the ensemble of historical forecasts to estimate the background error covariance （BEC） and directly obtains the analysis in the ensemble space. As a result, the quality of ensemble members significantly affects the DRP-4DVar performance. The historical-forecast-based initial perturbation samples are flow-dependent and can describe the error-growth pattern of the atmospheric model and the balanced relationship between different model variables. However, the ensemble spread is not big enough because of the short time interval between adjacent historical samples and the limited ensemble size. In this study, the BEC of the Weather Research and Forecasting Model （WRF） three-dimensional variational data assimilation （3DVar） system is employed to produce initial perturbation samples for the DRP-4DVar. The control variable perturbation method based on the structure characteristics of the 3DVar BEC produces initial perturbation samples that have reasonable background error correlations. Moreover, the estimated BEC also has good dynamic and physical consistency between variables after the initial perturbation samples undergo a development with a 6- or 12-h model forward integration. In terms of computational expense, the historical forecast results can be obtained without any additional computational cost at the operational numerical weather forecast centers, while the integration of samples from the 3DVar-based control variable perturbation method is time-consuming, but this difficulty can be alleviated through parallel computing. Although the assimilation run with the historical-forecast-based ensemble generates slightly better initial analysis field, the forecasts from the assimilation experiment using the 3DVar method performs better during the period from 12 to 30 h. Moreover, precipitation is simulated significantly better when the new ensemble is used. In conclusion, the performance of the DRP-4DVar approach can be improved especially in predicting heavy rains when the initial perturbation samples are derived from the BEC of the WRF 3DVar system.展开更多
基金Supported by the "973" Program of the Ministry of Science and Technology and the Ministry of Finance of China under Grant No.2004CB418304the Meteorological Sector Special Project of China Meteorological Administration under Grant No.GYHY(QX) 200906011
文摘The dimension-reduced projection four-dimensional variational data assimilation （DRP-4DVar） approach utilizes the ensemble of historical forecasts to estimate the background error covariance （BEC） and directly obtains the analysis in the ensemble space. As a result, the quality of ensemble members significantly affects the DRP-4DVar performance. The historical-forecast-based initial perturbation samples are flow-dependent and can describe the error-growth pattern of the atmospheric model and the balanced relationship between different model variables. However, the ensemble spread is not big enough because of the short time interval between adjacent historical samples and the limited ensemble size. In this study, the BEC of the Weather Research and Forecasting Model （WRF） three-dimensional variational data assimilation （3DVar） system is employed to produce initial perturbation samples for the DRP-4DVar. The control variable perturbation method based on the structure characteristics of the 3DVar BEC produces initial perturbation samples that have reasonable background error correlations. Moreover, the estimated BEC also has good dynamic and physical consistency between variables after the initial perturbation samples undergo a development with a 6- or 12-h model forward integration. In terms of computational expense, the historical forecast results can be obtained without any additional computational cost at the operational numerical weather forecast centers, while the integration of samples from the 3DVar-based control variable perturbation method is time-consuming, but this difficulty can be alleviated through parallel computing. Although the assimilation run with the historical-forecast-based ensemble generates slightly better initial analysis field, the forecasts from the assimilation experiment using the 3DVar method performs better during the period from 12 to 30 h. Moreover, precipitation is simulated significantly better when the new ensemble is used. In conclusion, the performance of the DRP-4DVar approach can be improved especially in predicting heavy rains when the initial perturbation samples are derived from the BEC of the WRF 3DVar system.
