Development of Optimized Radar Data Assimilation Capability within - - PowerPoint PPT Presentation

Development of Optimized Radar Data Assimilation Capability within the Fully Coupled EnKF‐EnVar Hybrid System for Convective‐Permitting Ensemble Forecasting and Testing via NOAA Hazardous Weather Testbed Spring Forecasting Experiments

Chengsi Liu1, Ming Xue1, 2, Youngsun Jung1, Lianglv Chen3 , Rong Kong1 and Jingyao Luo3

1Center for Analysis and Prediction of Storms and 2School of Meteorology University of Oklahoma, USA 3 China Meteorological Administration, China

ISDA 2019

Hazardous Weather Testbed (HWT) Ensembles 2007‐2018

Models: ARPS, WRF‐ARW, WRF‐NMM, COAMPS, FV3 2008 4‐km 3DVAR/cloud analysis ensemble with a 2‐km member 2009‐11 1‐km CONUS member 2013‐14 Added 12Z Members 2015 resolution change from 4‐km to 3‐km 2016 Added a cycled EnKF ensemble 2017 convection‐allowing FV3 member 2018 Added ensemble of coupled global‐regional FV3 forecast

• Aims to guide the near‐future
• perational CAM ensemble
• A distinct feature of the CAPS Storm‐

Scale Ensemble Forecasts (SSEFs): assimilation of full‐volume radial velocity and reflectivity data from the WSR‐88D network

• Part of the Community Leveraged

Unified Ensemble (CLUE) since 2016.

(Slide provided by Youngsun J.)

2018 HWT Model and Configuration

• Forecast model: WRF‐ARW v3.9.1 and FV3
• DA systems: CAPS 3DVAR/cloud analysis, GSI‐EnKF, CAPS EnKF
• 3‐km horizontal grid spacing, 51 vertical levels

http://www.caps.ou.edu/~fkong/sub_atm/spring18.html http://www.caps.ou.edu/~fkong/sub_atm/spring18‐enkf.html Forecast domain (1620 X 1120)

(Slide provided by Youngsun J.)

≥ 1.27 mm

Experiment (number of cases): 3DVar/cloud analysis_2016 (18), EnKF_2016 (16), 3DVar/cloud analysis_2017 (22), EnKF_2017 (21)

a) ≥ 1.27 mm b) ≥ 6.35 mm

Neighborhood ETS of 1‐hour rainfall

(R=15km)

(Slide provided by Youngsun J.)

The capability to directly assimilate reflectivity developed by CAPS in the NCEP’s operational DA system (GSI)

• One unique feature of the CAPS ensemble is the assimilation of full‐volume
• perational WSR‐88D radar radial wind and reflectivity data at the native

model grid resolution (4 km initially, 3 km since 2016) with a complex multi‐ moment microphysics scheme.

• Using temperature‐dependent background error profiles for static B of

hydrometeors (Liu et al. 2018)

• some special treatments to deal with issues associated with the non‐linear

reflectivity operator has been implemented.(Liu et al. 2018 ISDA)

• Three options of hydrometeors control variables
• Using mixing ratio as control variable (CVq)
• Using logarithmic mixing ratio as control variable (CVlogq)
• Using power transforming mixing ratio as control variable(CVpq, new development)

Advantage and disadvantage for CVq and CVlogq

（according to previous studies based on 3DVar: Liu et al. ISDA 2018）

• CVq:
• Have a better analysis for small reflectivity than CVlogq (< 30 dBZ)
• Worse radial wind analysis
• Slower convergence for minimization
• CVlogq:
• Have a better analysis for large reflectivity than CVq (> 30 dBZ)
• Better radial wind analysis
• Faster convergence for minimization

Current progress

DA System Model Control Variable in DA Experiment configuration ISDA 2018 ARPS‐3DVar ARPS CVq and CVlogq OSSE ISDA 2019 GSI‐EnKF/3DVar/hybrid En3DVar WRF‐ARW CVq, CVlogq and CVpq Real cases of NOAA

• perational HWT/WoF

Configuration

• A studied case: A Mesoscale convective system(MCS) occurred on 16th

May 2016 in Texas, U.S.

GSI EnKF/3DVar/Hybrid En3DVar

forecast

The 5th cycle background and analysis

background En3DVar_CVq En3DVar_CVlogq Analysis Obs.

• The analysis and forecast of

the leading edges of MCS from Cvlogq is better than that from CVq.

• The analysis of CVq is more

balanced than CVlogq and has less spurious echoes in the forecast.

Comparison between CVq and CVlogq

• The Vr analysis of CVlogq outperforms CVq in terms of smaller RMSIs.
• However, CVlogq has larger RMSIs in the forecasts due to the imbalance in

the analysis.

CVq CVlogq

Obs. En3DVar 3DVar HEn3DVar20 （20% static B） 0000 UTC 5th Cycle ANL BG (15min‐ forecast) EnKF

• The lead‐edge of MCS in 3DVar forecast is less organized than the others, due to lacking of

balance in the static background error covariance B.

• Including static B to En3DVar can improve the analysis in the area where ensemble spread is

not enough.

• Both the convective and stratiform precipitation are underestimated in EnKF.

30min‐forecast

En3DVar EnKF 3DVar HEn3DVar20 Obs.

Using power transforming mixing ratio as control variable(CVpq)

• When p=1.0, CVpq = CVq
• When p ‐> 0, CVpq ‐> Cvlogq
• Xue et al. (2010) used a similar power

transform in updating total number concentration by radar DA.

• Yang et al. (2018) also used this power

transform in the analyses of surface visibility and cloud ceiling height.

0.2 0.4 0.6 0.8 1 1.2 1.4 p=0.2 p=0.4 p=0.6 p=0.8 p=1.0 p=10e‐6

ˆ ( 1) /

p

q q p  

another test case

• Discrete tornadic supercells that occurred on 16th May 2017 in Elk City .

RMSI of 30min and 60min forecast

4 5 6 7 p10‐6 p0.2 p0.4 p0.6 p0.8 p1.0

dbz

Z

30mim forecast 60min forecast

4 5 6 7 8 p10‐6 p0.2 p0.4 p0.6 p0.8 p1.0

dbz

Vr

30min forecast 60min forecast

• Optimal power parameter is obtained based on RMSIs of the forecasts.

CVlogq CVq CVlogq CVq

Inner‐loop iteration

• uter‐loop iteration
• CVq has the slowest convergence speed

and worst Vr analysis

0.0 4.0 8.0 12.0 16.0 20.0 24.0 28.0 32.0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 100

cost function

x 100000

iteration steps

Cvloq CVpq_p=0.4 CVq

0.0 1.0 2.0 3.0 4.0 5.0 1 2 3

RMSI(dBZ)

iteration steps

CVlogq CVpq_p=0.4 CVq

Z

0.0 1.0 2.0 3.0 4.0 5.0 6.0 7.0 1 2 3

RMSI(dbz)

iteration steps

CVlogq CVpq_p=0.4 CVq

Vr

10.0 15.0 20.0 25.0 30.0 1 2 3

RMSI(dBZ)

iteration steps

Cvlogq CVpq_p=0.4 CVq

Z(>=40.0)

0.0 1.0 2.0 3.0 4.0 5.0 1 2 3

RMSI(dBZ)

iteration steps

Cvlogq CVpq_p=0.4 CVq

Z(<=25.0)

• For large reflectivity(e.g. >40 dBZ),

more outer‐loop steps are needed due to the high non‐linearity of the reflectivity operator.

10 20 30 40 50

reflectivity (dBZ)

0.5 1 1.5 2 2.5

rain mixing ratio (k kg-1)

Obs No DA

3DVar EnKF HEn3DVar

Background (15 min forecast) Analysis

3 4 5 6 p10‐6 p0.2 p0.4 p0.6 p0.8 p1.0

dbz

Z

30mim forecast 60min forecast

6 7 8 9 p10‐6 p0.2 p0.4 p0.6 p0.8 p1.0

m/s

Vr

30min forecast 60min forecast

5 6 7 8 9 10 11 p10‐6 p0.2 p0.4 p0.6 p0.8 p1.0

dbz

Z

30mim forecast 60min forecast

6 7 8 9 10 p10‐6 p0.2 p0.4 p0.6 p0.8 p1.0

m/s

Vr

30min forecast 60min forecast

4 5 6 7 p10‐6 p0.2 p0.4 p0.6 p0.8 p1.0

dbz

Z

30mim forecast 60min forecast

4 5 6 7 8 p10‐6 p0.2 p0.4 p0.6 p0.8 p1.0

dbz

Vr

30min forecast 60min forecast

6 7 8 9 10 p10‐6 p0.2 p0.4 p0.6 p0.8 p1.0

dbz

Z

30mim forecast 60min forecast

4 5 6 7 8 p10‐6 p0.2 p0.4 p0.6 p0.8 p1.0

m/s

Vr

30min forecast 60min forecast

RMSI of 30min and 60min forecast (4 convective storm cases)

20170509 20170516 20170518 20170527

Summary

• An EnKF coupled with hybrid En3DVar DA system that can directly

assimilate full volume radar data has been developed within the GSI framework.

• Three sets of control variables (CVq, CVloq and CVpq) were tested in

reflectivity DA. CVpq with 0.4‐0.6 power parameter has the best performance in terms of smallest RMSI in the forecasts.

• Hybrid En3DVar performs best in better capturing the intensity and

structure of the storms.

Future plan

• Run the GSI‐based coupled EnKF‐EnVar Hybrid System in real‐time for

2019 NOAA Hazardous Weather Testbed Spring Forecasting Experiments (forecast model will be switched to FV3 in 2020).

• Develop the adjoint for Thompson microphysics(two‐moment)

reflectivity operator within the GSI variational framework.