Hybrid ensemble-3DVar radar data assimilation for the short-term prediction of convective storms
This two-part study develops and tests a hybrid ensemble-3DVar radar data assimilation system for the short-term prediction of convective storms. A key component of this work is the use of the operational regional numerical weather prediction infrastructure of the United States National Weather Service (NWS). Recently, the NWS's Gridpoint Statistical Interpolation system (GSI) has been extended to include a hybrid ensemble-3DVar assimilation capability, allowing for the inclusion of flow dependent background error statistics in the 3DVar cost function. A convenient aspect of the hybrid ensemble-3DVar approach is its resource manageability. The initial implementation of the system may only use 3DVar and the hybrid aspect can be implemented gradually where additional ensemble members can be added as computational resources allow. Therefore the hybrid ensemble-3DVar method may be a particularly appealing approach for an operational numerical weather prediction (NWP) center where resources are at a premium. The first part of this study focuses on the development of a storm-scale, hybrid ensemble-3DVar radar data assimilation system. An observation operator for radar reflectivity is introduced, static background errors for additional hydrometeor control variables are obtained, an ensemble prediction system is implemented, and an algorithm is developed to assimilate radar observations. This system is applied to a real-data case which exhibits varying convective modes. It is found that, when compared to 3DVar, the hybrid ensemble-3DVar assimilation approach provides a closer fit to observations, produces cold pools which are much stronger than what was observed in the 3DVar experiment, and all experiments have a vertical velocity field at the final analysis time which exhibits generally weak upward vertical motion fields. The weak vertical motion field is hypothesized to be a result of the lack of vertical velocity control variable and thus there is no coupling amongst the three components of the wind. The second part of this study tests the radar data assimilation experiments through the evaluation of short, 1 hour forecasts initialized from the storm-scale analyses. It is found that the weak upward vertical velocity fields found in the storm-scale analyses did not preclude the development of deep convective storms with upward vertical motion representative of the observed storm types. In all radar data assimilation experiments a general eastward displacement of forecast storms relative to observed storms is observed. This displacement is hypothesized to be a result of storm re-development along cold pools during the first 10 to 20 minutes of the forecast. Furthermore, objective verification indicates that radar data assimilation compared to a case of no data assimilation generally improves the forecasts, and hybrid ensemble-3DVar assimilation yielded initial conditions which provided the best forecasts. A sensitivity was noted to the relative weights given to the static- and ensemble-based background error statistics. Experiments with 50% and 25% of the weight given to the static background error generally yielded the best forecast verification scores overall. It is noted that while these results demonstrate the effectiveness of hybrid ensemble-3DVar radar data assimilation at the convective-scale with the regional operational NWP infrastructure of the NWS, this is only an evaluation with a single case. Additional case studies are recommended before a more general conclusion may be obtained.
Baldwin, Purdue University.
Off-Campus Purdue Users:
To access this dissertation, please log in to our