Ensemble Kalman Filter Data Assimilation of VORTEX-SE P3 Tail Doppler Radar and Compact Raman Lidar Data for a 13 April 2018 Tornadic Supercell

Abstract

An integral part of any numerical weather prediction (NWP) system is data assimilation (DA): the process of optimally combining observations and model state backgrounds to minimize the overall error in the resulting analysis. Over the past two decades, data from several different observing systems have been assimilated for real convective scale cases. The assimilation of reflectivity (Z) and radial velocity (Vr) data collected by Weather Surveillance Radar-1988 Doppler (WSR-88D) radars has been well studied and their positive impact in NWP is well known. Unfortunately, due to their stationary nature, the distance between WSR-88D radars and storms of interest can result in low-resolution data or incomplete data coverage. Consequently, DA of specialized weather radar observations has also been studied, including observations collected by truck-based mobile radars. Results in these studies have shown that analyses produced using both WSR-88D DA and specialized radar DA are improved versus analyses generated from WSR-88D DA alone. The assimilation of thermodynamic profiles, retrieved from thermodynamic remote sensors, has also been conducted with the goal of improving analyzed storm environmental conditions, particularly in the boundary layer. Building upon experiments conducted in existing literature, this study performs ensemble Kalman filter (EnKF) DA of observations obtained during the 2018 Verification of the Origins of Rotation in Tornadoes Experiment Southeast (VSE18) field campaign for a cyclic, multi-tornado producing supercell that occurred during the late evening of 13 April 2018 near Monroe, Louisiana. In addition to surface observations and WSR-88D Z and Vr data, other unique specialized VSE18 data are also assimilated. These data include dual tail Doppler radar (TDR) Vr data, downward facing compact Raman lidar (CRL) retrieved water vapor mixing ratio (qv) profiles, Vr data collected by an S-band WSR-88D equivalent radar operated by University of Louisiana Monroe (KULM), and special radiosondes launched from Minden and Monroe Lousiana. The DA was performed in a series of experiments on nested 2500 m and 500 m grids using the Weather Research and Forecasting (WRF) model and the Advanced Regional Prediction System (ARPS) EnKF DA system. Deterministic forecasts were then performed using ensemble mean analyses for each DA experiment, on two-way nested 2500 m, 500 m, and 100 m WRF model grids. The impact of the data on the analyses was evaluated through a series of data denial experiments. CRL DA results in much larger values of qv in the inflow region of the Monroe supercell than the other analyses. TDR and KULM DA results in higher values of low-level vertical vorticity, and thereby a stronger tornadic like vortex (TLV), in the analysis. The TLV is absent in other analyses. TDR and KULM DA also increases the strength of low-level and mid-level updrafts in the analysis. The impact of assimilating special soundings in earlier DA cycles on the analyses of later DA cycles is relatively small, and the sign of the impact on analyzed updraft intensity and low-level rotation is not always consistent. This limited impact is likely attributable to the advection of the sounding-modified environment away from the supercell. Ultimately, experiments assimilating both CRL, TDR, and KULM data produced superior analyses, since they feature strong TLVs, deep intense updrafts, and improved boundary layer conditions in the inflow region of the supercell. The impact of the DA on storm forecasts was also examined. In CRL DA experiments, the forecasted Monroe supercell retains supercellular characteristics, including strong mid-level updrafts and associated mesocyclone, longer than the forecasts from other experiments. Generally, CRL DA experiments better forecast reflectivity than other experiments. CRL DA experiments have longer, more intense UH tracks than most other experiments. TDR and KULM Vr DA experiments better predict the intensity, track, and overall structure of the TLV. In TDR and KULM Vr DA experiments, the predicted TLV intensifies significantly within the first 5 to 10 minutes of the forecast. Unfortunately, the TDR and KULM Vr DA seems to degrade reflectivity forecasts, as indicated by poor statistical skill scores, and the qualitatively observed disorganization of the forecasted supercell 15 to 20 minutes after forecast initialization. Special sounding DA does not consistently improve forecasts. Additional work is required to determine why TDR and KULM DA degraded predicted reflectivity forecasts. If possible, CRL, TDR, and KULM DA should be conducted for additional VSE18 cases. Ensemble forecasts should also be conducted using ensemble analyses, as better predictions of the TLV may be obtained in the forecasts for individual ensemble members.