13.2 Impacts of Assumed Observation Errors in EnKF Analyses and Ensemble Forecasts of a Tornadic Mesoscale Convective System

Thursday, 10 January 2013: 3:45 PM
Room 9C (Austin Convention Center)
Nathan Snook, CAPS/Univ. of Oklahoma, Norman, OK; and M. Xue and Y. Jung

In recent years, the ensemble Kalman filter (EnKF) has become a widely-used technique for atmospheric data assimilation, and has been successfully applied to produce analyses and initialize forecasts from global to convective scale. When using an EnKF, the error statistics of the observations must be estimated; their values are sometimes chosen based on the typical measurement errors of the instruments used to collect the observations, as is often the case with radar measurements. While measurement error is important, it is only one of many sources of error that affect atmospheric data assimilation. Other sources of error include representativeness error, observation operator error, and gross error. Proper estimation of the observational error is required for optimal state estimation.

Recently, the authors have successfully demonstrated the ability of an EnKF in assimilating real radar observations to produce accurate analyses and forecasts for a tornadic mesoscale convective system (MCS) that occurred over Oklahoma and Texas on 9 May 2007. Though skillful ensemble forecasts of radar reflectivity and mesovortex locations were obtained, significant biases were noted in the forecast ensemble. In the current study, a set of experiments are performed using various assumed observation errors for radar reflectivity and radial velocity observations in the EnKF system for the same case, and the resulting impacts on the ensemble analyses and forecasts are examined. Results indicate that the assumed 1 m s-1 and 2 dB errors for radial velocity and reflectivity, respectively, used in our earlier study (and similar to values used in a number of other real radar data assimilation studies), were under-estimated. Increasing the errors to 3-6 m s-1 and 6-9 dB, respectively, results in reduced under-dispersion of the analysis ensemble and an increase in the skill and reliability of the ensemble probabilistic forecasts of reflectivity (a proxy for precipitation). The increased forecast skill appears to be, at least in part, due to a reduction in the forecast bias.

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner