Our experiences with EnKF assimilation of high-res radar observations collected in the 5 June 2009 Vortex 2 tornadic supercell. Jim Marquis, Yvette Richardson, Paul Markowski, David Dowell, Josh Wurman, Karen Kosiba, Paul Robinson Photo by Sean Waugh Goals of EnKF analysis Increase availability of 3-D kinematic and thermodynamic data (dual-Doppler and in situ obs are spatially/temporally limited). Courtesy of Paul Markowski Goals of EnKF analysis We want a set of smoothly evolving EnKF analyses (i.e., We are not using EnKF to initialize a forecast). Analyze roles that mesocyclone-scale processes play in tornadogenesis, maintenance, and decay: - trajectory analysis, - circulation/vorticity budgets, - mid-upper level features. Model Specifics WRF-ARW 3.2.1: -x,y = 500 m, 80m < z < 2km, [120 x 80 x 20] km 3 -LFO microphysics, -open lateral BCs, -no surface fluxes, no radiation, flat terrain. Homogeneous environment : DA Specifics DART: Ensemble adjustment filter, 50 members, Localization: Gaspari-Cohn, W = r = 6 km Ensemble initiation: 10 randomly placed warm bubbles at model t 0 for each member Ensemble spread maintained with: Additive noise added to T, Td, U, V every 5 min where radar reflectivity is > 25 dBZ, (Dowell and Wicker 2009) Perturbations smoothed to 4 km (horiz), 2 km (vert) scales Courtesy 2 obs = (2 m/s) 2 Radar velocities assimilated every 2 minutes OBAN: Cressman weighting 500 m horizontal grid spacing (for 500m-model grid experiment) data along conical slices Experiment timeline synthetic Data Thinning the amount of observations assimilated: Unthinned50% assimed 25% assimed Model grid Radar vel. ob. Y X No thinning done in vertical direction shown today; though, experiments show poor storm structure when low-level elevation angles are neglected. dBZ W > 0 Z = 0.5 km AGL (Note: these experiments conducted with 2-km model grid) (K) X Y 50% assimilated 25% assimilated unthinned Y radars Dual-Doppler EnKF W m/s Z = 400 m AGL Dual-Doppler EnKF (ensemble mean) kinematics comparison (DOW6 & DOW7) X (km) Y (km) Model errors/idealized conditions require DA for a good storm. Storm structure with/without radar assimilation: Top row: Series of EnKF (Ens. Mean) analyses. Bottom row: Single member forecasted forward from 2157 (no DA). Z = 150 m X (km) Y (km) Ensemble mean analyses temperature fields Z = 50 m (Note: these experiments conducted with 1-km model grid) Z = 50 m Mob. Mesonets Z = 50 m Mob. Mesonets MM obs EnKF outflow comparison (K) Low-level w max trace: dBZ MM obs: Not great, particularly in far left FF Better overall, but still disagreement in FF Decent, but still disagreement in FF Warmth (lasts 1-2analysis times) - Seems to disagree With few available MM obs Pretty good decent ok MMs only just getting into storm, but so far - not good My < 1 sentence impression of the agreement of MM obs and each EnKF analysis Z = 50 m Sequence of EnKF analyses of theta and MM obs (valid +/- 1 min from EnKF analysis) overlayed: Z = 50 m Mesocyclone buoyancy (function of height & time) Circulation and radial motion (function of radius & time) Periods of tornado lifecycle Pre-tornadic T-genesis Intensification Maturity Weakening Time (UTC) Inbound 0 m/s Outbound Positively buoyant Negatively buoyant Trajectories (storm-rel.) calculated from ens. mean analyses W (z = 200m) Ring (radius = 1 km) of 20 parcels centered on peak at z = 200 m; integrated forward in time from 4 times (K) Most parcels rising into updraft Some parcels rising into updraft Few/no parcels rising into updraft No parcels rising into updraft (K) Pre-tornadic T-genesis Intensification Maturity Weakening Comparison across model grid resolutions RMSI Total Spread Consistency Ratio (K) Low-level w max trace: MM obs: Summary EnKF Kinematic Analyses: -Compare well with dual-Doppler fields of similar scale. -Storm-scale structure robust with different model scale. More details come with finer model grid res. EnKF Thermo Analyses: -Mixed success with comparisons to MM in situ obs (where available) m model grid seems verify with MM obs best in mesocyclone area (partially due to more details captured?) -Seemingly random, possibly unphysical(?) anomalies appear more prominently with finer model resolution. Questions 1)Different way to maintain ensemble spread (adaptive methods)? 2)How else can we determine plausibility of analyses (kinematic and thermodynamic)? -Forcing terms along parcel trajectories? 3)More rigorous methods for optimizing scales/amounts of observations (compared to model resolution)? Acknowledgements The EnKF experiments were performed using NCAR CISL supercomputing facilities with the Data Assimilation Research Testbed (DART) and WRF-ARW software. Thank you to Glen Romine, Lou Wicker, Dan Dawson, Chris Snyder, and Nancy Collins for advice/help. Thanks to all VORTEX2 crew for their dedication while collecting data on 5 June This research is funded by NSF grants: NSF-AGS , NSF- AGS The DOW radars are NSF Lower Atmospheric Observing Facilities supported by NSF-AGS