Comparisons of kinematical retrievals within a simulated supercell:

Dual-Doppler analysis (DDA) vs. EnKF radar data assimilation

Corey Potvin and Lou Wicker, NSSL

Storm-Scale Radar Data Assimilation WorkshopOctober 17-18, 2011

Norman, OK

Motivation• High-resolution radar datasets (e.g., VORTEX-2) enable fine-scale

supercell wind retrievals that are vital to illuminating supercell/tornado dynamics

• Knowledge of error characteristics of different wind retrieval techniques is needed to:– Select the most accurate method for a given dataset– Determine how much confidence to place in the analyses– Design mobile radar deployment and scanning strategies that mitigate

wind retrieval errors

• Maximizing the scientific value of kinematical/dynamical retrievals from mobile radar datasets requires understanding of limitations of different methods under different scenarios

Primary Questions

• When dual-radar data are available, does EnKF produce better wind estimates than DDA? – Or do model/IC/EnKF errors obviate advantage of

constraining analysis with NWP model?• When only single-radar data are available, can

EnKF produce DDA-quality analyses?– Or is the solution too underdetermined?

• Does EnKF produce reliable wind estimates outside radar data region?

Previous Work• DDA vs. EnKF comparisons have been made for

real tornadic supercells:– Dowell et al. (2004) - 88D; Arcadia, OK, 1981– Marquis et al. (2010) – DOW; Goshen Co., 2009– Marquis et al. (2011) – DOW; Argonia, KS, 2001

• Low-level wind/ζ analyses from 1- or 2-radar EnKF qualitatively similar to DDA

• True wind field unknown, so difficult to evaluate EnKF vs. DDA performance

OSS Experiments

• Simulate supercell using NSSL Collaborative Model for Multiscale Atmospheric Simulation (NCOMMAS)

• Synthesize mobile radar Vobs, Zobs

• Input observations to:– NCOMMAS EnKF system– 3D-VAR DDA technique

• Compare wind analyses from the two methods– 2 vs. 1 radar – good vs. poor radar positioning– perfect (2-moment) vs. imperfect (LFO) microphysics– various scanning strategies

Radar Emulation

• Compute Vobs, Zobs on radar grids using realistic beam weighting (Wood et al. 2009)

• ΔR = 150 m, ΔΦ = 1°• Vobs computed only where Zobs > 5 dBZ• Random errors: σV = 2 m/s, σZ = 2 dBZ

VCP ΔT (min) Max θ ΔθDEEP 2.5 33.5° 1° - 3°

SHALLOW 1.5 12.5° 1°

RAPID 0.5 12.5° 1°

Supercell simulation

• NCOMMAS: nonhydrostatic, compressible numerical cloud model

• Roughly 100 × 100 × 20 km domain; Δ = 200 m • ZVDH microphysics (Mansell et al. 2010)

t = 6

0 m

in, z

= 1

km

dBZ, ζ dBZ, w

θ' u, v, w

EnKF configuration

• Ensemble square root filter• 40 members• Δ = 600 m (model error)• Data interpolated to 2-km grid on conical scan

surfaces, then advection-corrected (1-D)• Data assimilated every 2 min from t = 30 min to t = 84

min• Covariance localization: 6 km (horiz), 3 km (vert)• Filter uses σV = 2 m/s, σZ = 5 dBZ

EnKF configuration (cont.)

• Ensemble initialization:– sounding u, v perturbed at top/bottom– randomized thermal bubbles

• Ensemble spread maintenance:– additive noise (Dowell and Wicker 2009) – bubbles added where Zobs - Zens > 30 dBZ

DDA Technique

• 3D-VAR method (Shapiro et al. 2009; Potvin et al. 2011)

• Data, mass conservation, smoothness constraints weakly satisfied

• Impermeability exactly satisfied at surface• Δ = 600 m; domain = 40 × 40 × 13.8 km (matches

verification domain)• Same advection correction as in EnKF• DDA errors for this case examined in Potvin et al.

(JTECH, in review)

Experiment Domains

t = 84 min

“poor radar placement”experiments

t = 30 min

default experiments

verification domain (moves

with storm)

Reflectivity Assimilation

• Only helped single-radar EnKF; only Vobs

assimilated in remaining experiments

• Improvement diminished by– model errors– non-Gaussian-

distributed errors

Considerations

• Representativeness of OSSE model/observational errors imprecisely known

• Wind analysis errors sensitive to tunable parameters and methodological choices

• Each EnKF mean analysis is only a single realization of the distribution of possible analyses

• Need to treat small error differences with caution

Deep VCPverified where data available from both radars

t = 36 min t = 84 min

u

w

t = 60 min

w at t = 60 min, z = 1.2 km TRUTH DDA

1-radar EnKF

2-radar EnKF-LFO

2-radar EnKF

1-radar EnKF

1-radar EnKF-LFO

Max updraft

• Except for 1-radar EnKF-LFO, EnKF generally damps main updraft less than DDA at middle/upper levels

t = 36 min t = 84 mint = 60 min

Non-simultaneity errors• EnKF displacement errors and (at later times) pattern errors

are much smaller than DDA errors aloft

v at t = 60 min, z = 9.6 km

TRUTH DDA 2-radar EnKF

1-radar EnKF-LFO

Deep VCP

t = 36 min t = 84 min

u

w

VCP ΔT (min) Max θ ΔθDEEP 2.5 33.5° 1° - 3°

SHALLOW 1.5 12.5° 1°

RAPID 0.5 12.5° 1°

t = 36 min t = 84 min

u

w

VCP ΔT (min) Max θ ΔθDEEP 2.5 33.5° 1° - 3°

SHALLOW 1.5 12.5° 1°

RAPID 0.5 12.5° 1°

t = 36 min t = 84 min

u

w

VCP ΔT (min) Max θ ΔθDEEP 2.5 33.5° 1° - 3°

SHALLOW 1.5 12.5° 1°

RAPID 0.5 12.5° 1°

Impact of dual-Doppler ceiling

TRUTH

DDA SHALLOW

DDA DEEP

2-radar EnKF SHALLOW

2-radar EnKF DEEP

t = 36 min, z = 5.4 km

Impact of data edge (w)t = 36 min, z = 0.6 km

Impact of data edge (v)t = 36 min, z = 0.6 km

Impact of poor cross-beam angle

T=30 min

T=84 min

30° lobe

Impact of poor cross-beam angle

• 3-D winds much less constrained by Vobs alone

• EnKF far less degraded than DDA

• EnKF better than DDA even at low levels later in DA period

36 min

84 min

Good placement

w w

w w

Poor placement

Impact of sounding error

1 km

2 km

3 km

Impact of sounding errort = 36 min t = 84 min

u

w

NO sounding errort = 36 min t = 84 min

u

w

Conclusions: 2-radar case

• When might EnKF help?– Poor radar positioning– Rapid flow advection/evolution– Outside dual-Doppler region

• When might EnKF (mildly) hurt?– At lowest levels– Very early in DA period (e.g., storm maturation)

• Errors in microphysics and/or low-level wind profile had small impact

Conclusions: 1-radar case• 1-radar EnKF analyses much more sensitive to

microphysics/sounding errors• Thus, conclusions more tentative• 1-radar EnKF may produce DDA-quality (or better)

analyses:– At upper levels if flow rapidly translating/evolving– At middle & upper levels later in DA period– At all levels later in DA period if cross-beam angle is poor and

model errors small• Analyses outside data region probably not good enough for

some applications

Future Work

• Compare trajectories, vorticity & vorticity tendency analyses

• 29 May 2004 Geary, OK case• VORTEX-2 case(s)• Longer term: OSSEs to compare storm-scale

analyses and forecasts (WoF) obtained from different schemes (e.g., 3DVAR, EnKF, hybrid)