Transcript The Impact of Assimilating Radar Observations on Ensemble

The EnKF Analyses and Forecasts of the
8 May 2003 Oklahoma City Tornadic
Supercell Storm
By
Nusrat Yussouf1,2
Edward Mansell2, Louis Wicker2 ,
Dustan Wheatley1,2, David Dowell3,
Michael Coniglio2 and David Stensrud2
1. Cooperative Institute for Mesoscale Meteorological Studies, University of Oklahoma, Norman, OK.
2. NOAA/National Severe Storms Laboratory , Norman, OK.
3. NOAA/ESRL/Global Systems Laboratory, Boulder, CO.
Motivation

Most storm-scale NWP modeling studies assume horizontally
homogenous environmental conditions

Much easier to obtain a high-quality analysis of supercell storm than a
accurate forecast

Stensrud and Gao (2010): Substantial improvement in storm
forecast accuracy when using realistic inhomogeneous
mesoscale environment
 This work focuses on ensemble data assimilation experiments
of tornadic supercell within full mesoscale complexity
A very short-range probabilistic forecasts of
tornadic supercell storms
 In support of Warn on Forecast - a numerical model-based
probabilistic convective-scale analysis and forecast system to
support warning operations within
NOAA
The 8 May 2003 Oklahoma City Tornadic
Supercell
NWS Damage Path of OKC Tornado
Hu and Xue (2007)
HPC Synoptic Scale Surface
Analyses at 18:00 UTC
KOUN Radar Observations at 22:10 UTC
Mesoscale Ensemble
• WRF-ARW v3.2.1
 Mesoscale data assimilation on CONUS domain
 18-km horizontal grid spacing; 51 vertical levels
 Mean initial and boundary conditions from GFS final analysis
• 45 member mesoscale ensemble
 IC/BC perturbations from WRF-Var (Torn et al. 2006)
 Physics Options:
- Cumulus: Kain-Fritsch
- Microphysics: Thompson
- Longwave Radiation:RRTM
•
- PBL: MYJ
- Shortwave Radiation: Dudhia
- Land Surface: Noah
Ensemble Adjusted Kalman Filter (EAKF) approach from the
Data Assimilation Research Testbed (DART)
Mesoscale Data Assimilation
 Observations assimilated:
- Altimeter setting (p)
- Temperature (T)
- Dewpoint (Td)
- Horizontal winds (u and v)
 Observation platforms:
- METAR, Radiosonde, Maritime and Automated Aircraft from MADIS
 Adaptive prior inflation & localization (1600 obs)
 Localization half width: 287/4 km for horizontal/vertical
 Filter configuration adapted from Glen Romine’s system at NCAR
Timeline of mesoscale data assimilation experiment:
- Continous cycling for 3 days
- Every 6 hour DA: 18 UTC May 5 – 12 UTC May 8
- Every 1 hour DA: 13 UTC May 8 - 0 UTC May 9
Storm-Scale Data Assimilation
• A 45 member storm-scale ensemble








One-way nested down from mesoscale ensemble analyses at 21Z, May 8
2-km horizontal grid spacing , 450 x 360 km wide, 50 vertical levels
KTLX WSR-88D radar doppler velocity (Vr) and reflectivity (dBZ)
Radar data objectively analyzed to 4-km grid using OPAWS
Both adaptive inflation and additive noise to maintain spread
Adaptive localization (2000 obs)
Observation errors: Vr = 2 m s-1, Z = 5 dBZ
Localization half-width: 12/6 km for horizontal/vertical
• Timeline of storm-scale data assimilation experiment:
- One hour DA every 3 minutes:
21 UTC – 22 UTC, May 8
- One hour ensemble forecast:
22 UTC - 23 UTC, May 8
Storm-Scale Data Assimilation
Experimental Design
 Three ensemble DA experiments using different bulk microphysics
schemes:
- Thompson 1.5 moment (Thompson et al. 2004, 2008)
Mixing ratio: Qc, Qi, Qs, Qr and Qg
Number Concentrations: ice (Ni) and rain(Nr)
- NSSL Variable Density Double Moment (NVD-DM; Mansell et al. 2010)
Mixing ratio: Qc, Qi, Qs, Qr, Qg and Qh
Number Concentrations: Nc, Nr, Ni, Ns, Ng and Nh
- NSSL Fixed Density Single Moment (NFD-SM; Gilmore et al. 2004)
Mixing ratio: Qc, Qi, Qs, Qr and Qg
 Remaining physics options are identical to mesoscale ensemble
Observation-Space Diagnostics: rmsi and total ensemble spread
Ensemble spread for reflectivity is
consistently smaller than the rmsi
Reflectivity rmsi from Thompson is
relatively smaller during the later
assimilation period
Z statistics are calculated where observed Z > 10 dBZ
Radial velocity ensemble spread is
comparable to rmsi
rmsi and spread are similar in
magnitude for the 3 microphysics
Scheme experiments for Vr
Vr statics are calculated at all observed values over the entire domain
Observation-Space Diagnostics: Consistency ratio
Reflectivity consistency ratio
is well below 1.0
Consistency ratio =
(ens. variance + obs-error variance)
/ (mean-squared innovation)
Analyses at 2200 UTC at 1 km AGL
Thompson
Member 31
mesocyclone
NFD-SM
NVD-DM
Member 14
Member 12
mesocyclone
mesocyclone
Vorticity contours from 0.001 to 0.01 at 0.001 s-1
U-V Winds vector (m/s)
KTLX Reflectivity Obs.
The areal extent and the reflectivity
distribution in the forward flank region is
closer to the observation in Thompson
and NVD-DM scheme compared to
NFD-SM.
Reflectivity Forecast at 1 km AGL
NVD-DM
15 min Fcst at 2215 UTC
NFD-SM
Member 31
Member 12
Member 14
45 min Fcst at 2245 UTC
Thompson
Member 31
Member 12
Member 14
KTLX Reflectivity Obs
Ensemble Mean Coldpool Analyses and Forecast
≥ 0.003 s-1 at 1 km AGL
≥ 0.0015 s-1 at 150m AGL
1-hr Forecast Probability of Vorticity
(2145-2245 UTC) after 45-min assimilation
Thompson
NFD-SM
NVD-DM
~22:38
~22:38
~22:06
~22:06
Observed damage
track and times
Observed damage
track and times
~22:06
Observed damage
track and times
Observed damage
track and times
~22:38
~22:38
~22:38
~22:06
~22:38
~22:06
~22:06
Observed damage
track and times
Observed damage
track and times
% Probability
≥ 0.003 s-1 at 1 km AGL
≥ 0.0015 s-1 at 150m AGL
45-min Forecast Probability of Vorticity
(2200-2245 UTC) after 1-hr assimilation
Thompson
NFD-SM
~22:38
~22:38
~22:06
~22:06
Observed damage
track and times
Observed damage
track and times
~22:38
~22:06
Observed damage
track and times
Observed damage
track and times
~22:38
~22:38
~22:06
NVD-DM
~22:06
Observed damage
track and times
~22:38
~22:06
Observed damage
track and times
% Probability
Summary and Future work

The results show promise for short-range, ensemble-based, stormscale tornadic supercell forecasts initialized from EnKF analyses

The reflectivity structure of the supercell storm using a DM scheme
compare better to the observations than that using a SM scheme

Storm-scale ensemble system can predict the track of the strongest
rotation with some accuracy in 0-1 hour time frame

Future work:
Vary the microphysical parameters across the ensemble to
improve spread
Use of higher resolution grid of 1 km or less
Acknowledgement
Glen Romine and Nancy Collins for help with DART
Kevin Manross for providing the edited radar data
Additional Slides
Reflectivity Analyses
at 2200 UTC
Reflectivity Forecasts
at 2230 UTC