Kalman filter based techniques for assimilation of radar data.

摘要

The Ensemble Square Root Filter (EnSRF) data assimilation technique is applied to examine the impact of assimilating high temporal frequency radar observations over a shorter assimilation period. To reduce the heavy computation cost of assimilating large number of radar observations using EnSRF technique, synthetic radar observations are generated at coarser spatial resolution. Two sets of experiment are conducted with identical settings based on perfect model framework where model error does not play a role. One experiment assimilates radar observations, in which a volume scan is conducted every 5 min, while the other experiment assimilates observations, in which a volume scan is conducted every 1 min. Results indicate that assimilating observations at 1-min intervals over short 15-min period yields significantly better analyses and forecasts than those produced using observations at 5-min intervals. However, the very good performance obtained from perfect model experiments is not expected in real-world experiments where models unavoidably have errors. Therefore to account for model error, another two sets of experiments are conducted using both a perfect and an imperfect model framework and the EnSRF data assimilation technique. In addition, the value of using a range of intercept and density parameters for hydrometeor categories in different ensemble members within the same microphysics scheme also is examined. Results show that the EnSRF system performs reasonably well with the imperfect model assumption. Results also indicate that in the presence of model error, a combination of different hydrometeor density and intercept parameters leads to improved forecasts over experiments that use a constant, hydrometeor intercept and density parameter.;While the EnSRF data assimilation technique shows promise in radar data assimilation, one limitation of EnSRF technique is that it assimilates observations serially, making it computationally very expensive when the number of observations is very large. Thus in an effort to explore efficient data assimilation method, the feasibility of the information filter as an alternate to the EnSRF data assimilation technique when the number of observations is very large is examined. The extended information filter (EIF) is implemented using the Lorenz 96 model and the performance of EIF in assimilating both low and high spatial resolution observations are compared with the benchmark extended Kalman filter (EKF) assimilation technique. Results indicate that both EKF and EIF produce similar results for different spatial resolution observation assimilation. The computational time for the EIF is larger than that of the EKF filter as expected due to the higher computational cost of matrix inversion in EIF technique. However, the increment in computational cost for EIF technique is much smaller than that of EKF technique for increased number of observation assimilation.