Assimilating Reflectivity Observations of Convective Storms into Convection-Permitting NWP Models David Dowell 1, Chris Snyder 2, Bill Skamarock 2 1 Cooperative Institute for Mesoscale Meteorological Studies, Norman, Oklahoma, USA 2 National Center for Atmospheric Research, Boulder, Colorado, USA
Ensemble Kalman Filter (EnKF) Assimilation of Radar Data for Convective Storm Analysis Assimilation of Doppler velocity observations –Snyder and Zhang 2003, Zhang et al. 2004, Dowell et al. 2004a, Caya et al. 2005, Tong and Xue 2005 Recently, assimilation of reflectivity observations (together with Doppler velocity observations) –Dowell et al. 2004b –Tong and Xue 2005
Research Questions 1.For retrieving storm characteristics, what is the value of assimilating reflectivity observations? –Reflectivity observations only, and together with Doppler velocity observations 2.Is it necessary/advantageous to process in a special way observations that indicate low values of reflectivity? 3.Do conclusions drawn from “perfect-model” experiments change when model errors are significant? –Uncertainty in storm environmental conditions –Model-physics errors
OSSE Design (part 1) NCOMMAS forecast model –Similar results obtained with WRF –Idealized base state (moderate CAPE, low CIN) –4 hydrometeor types (rain, ice particles, snow, hail/graupel) – x= y=2000 m, z=500 m –Same model used to produce reference simulation and to assimilate synthetic observations
Reference Simulation: Supercell 30 min60 min90 min reflectivity and wind at 2.25 km AGL 100 km
Reference Simulation: Squall Line 1 hour2 hours3 hours reflectivity and wind at 2.75 km AGL 200 km
OSSE Design (continued) Synthetic radar observations –Volumetric observations produced from reference simulation every 5 min –Smith et al equations used to compute reflectivity corresponding to model state variables –“Radar” observes u component of wind in reference simulation, only where reflectivity > 15 dBZ –Random errors added to truth error = 2.0 dBZ for reflectivity error = 2.0 m s -1 for radial velocity
OSSE Design (continued) Initialization of 50-member ensemble –“First echoes” initiated by warm bubbles in random locations within a subdomain containing observed storms Ensemble Kalman filter (square-root filter; Whitaker and Hamill 2002) data-assimilation scheme
Supercell Assimilation Experiment Temperature (K) Vertical velocity (m/s) RMS errors in rain region, EnKF analysis
Squall Line Assimilation Experiment Temperature (K) Vertical velocity (m/s) RMS errors in rain region, EnKF analysis
Suppressing Spurious Cells by Assimilating Reflectivity Observations Velocity observations are available in precipitation regions only. Reflectivity observations are available everywhere. Reflectivity and wind at 1.25 km AGL at 60 min (supercell) Truth Ensemble Member 4 (assimilation of radial velocity only) Ensemble Member 4 (assimilation of reflectivity and radial velocity)
Low-Reflectivity Observations Consider a simpler method for suppressing spurious cells by assimilating low-reflectivity observations: If both the observation and the ensemble mean indicate low reflectivity, then only force outliers (ensemble members with anomalously high reflectivity) back toward the ensemble mean locally: Otherwise, assimilate the reflectivity observations (and velocity observations) with the standard EnKF method, as before.
Standard EnKF vs. EnKF with Simpler Spurious Cell Suppression Temperature (K) Vertical velocity (m/s) RMS errors in rain region, EnKF analysis
Issues Concerning Low-Reflectivity Observations Variations in reflectivity values below ~15 dBZ often represent phenomena not in the model (e.g., insects), so it’s probably best to ignore these variations and treat all low values in the same way. Forecast distributions in low-reflectivity regions are often non-Gaussian (spurious storms in a few members, no storm in most members), so Kalman analysis equation isn’t necessarily valid. In experiments to date, the simpler algorithm for suppressing spurious cells gives comparable results to the standard EnKF, and computation is faster.
Additional Information from Reflectivity: More Efficient Retrieval of Main Supercell Vertical velocity (m/s) RMS errors in rain region, EnKF analysis
Experiment with Sounding Errors ( error = 2 m/s for u and v, 1 K for T and T d ) Temperature (K) Vertical velocity (m/s) RMS errors in rain region, EnKF analysis
Other Data-Assimilation Research Severe Weather Analysis and Prediction Group at CIMMS (1) and NSSL (2) Radar-data quality control, error-covariance estimation, and “3.5DVar” assimilation (Xu) EnKF radar-data assimilation and prediction for real-data cases: supercells and squall lines (Dowell, Coniglio, Wicker) Mesoscale ensemble forecasting (initial-condition, boundary- condition, and model-physics diversity) and EnKF assimilation of surface data (Fujita, Stensrud, Dowell) Precipitation microphysics-scheme diversity for storm-scale data assimilation (Mansell, Wicker) (1) Cooperative Institute for Mesoscale Meteorological Studies, Norman, Oklahoma, USA (2) National Severe Storms Laboratory, Norman, Oklahoma, USA
Conclusions Results of similar quality are obtained for multiple convective modes: supercell and squall line. Both Doppler velocity and reflectivity observations are useful for retrieving the atmospheric state on the scale of convective storms: –Velocity observations provide information about inner storm structure. –Reflectivity observations mainly provide information about where storms are and are not, but also lead to more efficient retrievals of inner storm structure.
Conclusions (continued) When assimilating low-reflectivity observations, it seems to be more efficient to simply force outlier members toward the ensemble mean locally rather than to use the standard EnKF. Typical errors in estimated environmental conditions do not prevent good storm-scale analyses produced by assimilating radar data. However, precipitation microphysics errors (not shown today) do pose a serious challenge.