VDRAS - Radar Data Assimilation and Explicit Forecasting of Convections

1
VDRAS - Radar Data Assimilation and Explicit
Forecasting of Convections

• Juanzhen Sun
• National Center for Atmospheric Research

2
Outline

• Introduction background and motivation
• Methodologies for storm-scale DA
• VDRAS - a 4D-Var based radar data assimilation
  system
• Case studies and results
• Issues and opportunities
• Summary

3
Cloud-scale modeling since 1960s

• Used as a research tool to study dynamics of
  moist convection
• Initialized by artificial thermal bubbles
  superimposed on a single sounding
• Rarely compared with observations

From Weisman and Klemp (1984)
4
Lillys motivating publication (1990)-- NWP of
thunderstorms - has its time come?

• Yes, it was time because we had
• NEXRAD network
• Increasing computer power
• Advanced DA techniques
• Experience in cloud-scale modeling

Because of the inherent difficulty of
predicting Initial storm development, our main
focus will probably be on predicting the
evolution of existing storms and development of
new ones from outflow Interaction. We are not
sure what will happen if we start a model with
these incomplete data and fill in the rest of the
volume with mean-flow condition, but it is not
likely to be inspiring.
5
China national radar network -- CINRAD
WSR-98D
The Chinese Meteorological Administration is
developing a network of 126 CINRAD radars 66
radars are S band (red dots) 60 radars are C band
(blue squares)
6
Operational NWP poor short-term QPF skill
0.1 mm hourly precipitation skill scores for
Nowcast and NWP averaged over a 21 day period

• Current operational NWP can not beat
  extrapolation-based radar nowcast technique for
  the first few forecast hours.
• One of the main reasons is that NWP is not
  initialized by high-resolution observations, such
  as radar.

From Lin et al. (2005)
7
Example of model spin-up from BAMEX
6h forecast (July 6 2003)
12h forecast

• Without high-resolution
• initialization
• A model can takes a
• number of hours to
• spin up.
• Convections with weak
• synoptic-scale forcing
• can be missed.

Radar observation at 0600 UTC
at 1200 UTC
Graphic source http//www.joss.ucar.edu
8
Comparing radar DA with conventional DA
Conventional DA
Radar DA
Obs. resolution a few 100 km -- much poorer than model resolutions Obs. resolution a few km -- equivalent to model resolutions
Every variable (except for w) is observed Only radial velocity and reflectivity are observed
Optimal Interpolation Retrieval of the unobserved fields
Balance relations Temporal terms essential
observation
model grid
9

• Methodologies for storm-
• scale DA

10

Two general methodologies

• Sequential initialization
• - Model dynamical, thermodynamical, and
  microphysical
• fields are derived separately using
  different methods
• - Is usually simple and efficient
• - Initial conditions may lack consistency
• Simultaneous initialization
• - Model initial fields are obtained
  simultaneously
• - Is usually computational demanding
• - Initial fields satisfy the constraining
  numerical model

11
An examples of sequential initialization
Large-scale background and radial velocity
Step 1
3DVar constrained by simple balance equations
u,v,w
Reflectivity and cloud information
Step 2
Cloud analysis
T, qr,qc ,qv
12
An examples of simultaneous initialization
Input
Large-scale background, radar radial velocity,
and reflectivity
V1
V2
V3
4DVar constrained by a NWP model
u,v,w,T, qr,qc ,qv
u,v,w,T, qr,qc ,qv
u,v,w,T, qr,qc ,qv
The analysis variables are balanced through the
Numerical model
t2
t3
t1
13

Sequential initialization Techniques

• Successive correction cloud analysis
• LAPS (FSL)
• 3DVar cloud analysis
• ARPS (CAPS)
• 3D wind retrieval thermodymical retrieval
  microphysical
• specification
• (Weygandt et al. 2002)
• 3D wind retrieval latent heat nudging
• (Xu et al. 2004)

14
Simultaneous initialization techniques

• 3D-Var
• WRF (NCAR)
• 4D-Var
• VDRAS (NCAR), MM5 4DVar (FSU)
• EnKF
• (Snyder and Zhang 2004, Dowell et al. 2004)

15

• VDRAS - a 4D-Var based radar data assimilation
  system

16
History of VDRAS/VLAS
1987 First study of 4DVar for 3-D wind retrieval
(Wolfberg 1987) 1991 First version of VDRAS
developed and successfully applied to simulated
radar data (Sun et al 1991) 1994 Applied to
real single-Doppler observations (Sun and Crook
1994) 1997 Extended to a full troposphere cloud
model (Sun and Crook 1997,1998) 1998
Implemented in real-time at Sterling, NWS (Sun
and Crook 2001) 2000 Installed at Sydney,
Australia for the Olympics (Crook and Sun, 2002)
17
History of VDRAS/VLAS Cont
2000-2004 Field Demonstrations every summer for
the FAA convective weather program 2001 Applied
to simulated lidar data for convective boundary
layer (Lin et al. 2001) 2000-now Research on
forecasting gust front and storm evolution (Crook
and Sun 2004, Warner et al. 2002, Sun 2005,
) 2003-now VLAS and VDRAS applications for
homeland security projects 2005-2006 Real-time
demonstration using multiple WSR-88Ds for
high-resolution analysis and QPF
18
Flow chart showing major processes of VDRAS

• Data Preprocessing
• Quality control
• Interpolation
• Background analysis
• First Guess

• Data Ingest
• Rawinsondes
• Profilers
• Mesonet
• Doppler radars

• 4DVAR Assimilation
• Cloud-scale model
• Adjoint model
• Cost function
• Weighting specification
• Minimization

• Display (CIDD)
• Plots and images
• Animations
• Diagnostics and statistics

19
The Numerical Model

• Anelastic approximation
• Adams-Bashforth time differencing
• Arakawa C-grid spatial differencing
• Liquid water potential temperature is used as the
  
• thermodynamical variable. Cloud water and
  temperature are diagnosed.
• Bulk warm-rain parameterization
• Constant diffusion coefficients
• No surface fluxes

20
Cost Function
vr - (u,v,w) Relation
Z-qr Relation
21
What is an adjoint model?
Model state at time 0
Model state at time k
Forecast model
Tangent linear model
Adjoint model

• The adjoint operator is the transpose of the
  tangent linear
• model operator.
• Integration of the adjoint model from the time
  step k to 0
• gives the gradient of J with respect to x0

22
Continuous 4DVar analysis cycles
V1
V2
V5
V4
V3
90
Time (min)
0
6
12
18
24
4DVAR Cycle 1
4DVAR Cycle 2

• First guess
• Mesoscale analysis
• VAD analysis
• Mesoscale analysis
• Barnes mesonet analysis

• First guess
• Cycle 1 analysis
• VAD analysis
• Mesoscale analysis

• First guess
• Cycle 2 analysis
• VAD analysis
• Mesoscale analysis

• First guess
• Mesoscale analysis
• VAD analysis
• Mesoscale analysis
• Barnes mesonet analysis

23
Procedures of the mesoscale background analysis
RUC first-pass Barnes analysis with a radius of
influence of 200km
Surface data Barnes analysis
VAD second-pass Barnes analysis with a radius of
influence of 50 km
Combine surface and upper-air analyses via
vertical least-squares fitting
Mesoscale background
24
4D-Var cycles
Atmospheric State
5
10
20
25
30
15
TIME (Min)
25

• Case Studies and Results

26
Low-level wind analysis

• Apply VDRAS to the low-level (below 5 km)
• Focus on low-level convergence and gust front
• Has been run in real time for a number of years
  in several locations

27

Sydney 2000 Tornadic hailstorm
November 3rd tornadic hailstorm event,
left-moving supercell, clockwise rotating tornado.
gust front
sea breeze
28
November 3rd, VDRAS-Dual Doppler comparison
¼ of analysis domain
rms(udual uvdras) 1.4 m/s rms(vdual vvdras)
0.8 m/s
29
Sydney 2000
Verification of VDRAS winds using aircraft data
(AMDARs)
Date Mean vector difference Mean vector
9/18/2000 2.1 m/s 6.2 m/s
10/3/2000 3.5 m/s 9.4 m/s
10/8/2000 2.6 m/s 5.0 m/s
11/03/00 2.2 m/s 5.0 m/s
30
Fort Worth NWS office real time VDRAS
Severe storm of May 4-5, 2006
VDRAS analyses every 20 min
Perturbation temperature
Convergence White contour 30 dBZ
reflectivity
31
Initialization and forecasting of a supercell
storm

• Occurred near Bird City, Kansas, on June 29,
  2000
• 5 hour life time 2200 300 UTC
• Formed ahead of an advancing surface boundary
• Produced large hail and a F1 tornado
• VDRAS assimilates data from one radar (KGLD)

32
Storm evolution (40 dBZ contour)
33
Vertical profile of radial wind RMS error
Height (km)
RMS error (m/s)
34
Comparison of forecast with observation(40 dBZ
contours every 20 min for two hours)
Observation
Forecast
35
Storm Track for three experiments(40 dBZ
contours every 20 min for two hours)
No evaporative cooling in both analysis and
forecast
Evaporative cooling in analysis but not in
forecast
Control
36
Initialization and forecasting of an IHOP squall
line

• Occurred in IHOP domain, on June 12-13, 2002
• 12 hour life time 2000 800 UTC
• Formed near a triple point of a dry line and a
  stationary outflow boundary

37
Model and DA set-up

• Domain size
• 480kmx440km
• Resolution 4km
• 4 NEXRAD radars
• 30 METAR surface
• stations
• Cold start first guess
• radiosonde VAD
• surface obs.
• 50 min assimilation
• period which includes
• three 10 min 4DVar
• cycles

015400 UTC
Observation
38
5-hour forecast of IHOP June 12 squall line
Frame interval 20 min. White contour
observation
5-hour forecast
39
Detailed look of the analysis wind
Reflectivity and wind at z3.25 km
Radial velocity and wind at Z 0.25 km
Radar
40
t 0
Evolution of cold pool
The initial cold pool of -8oc played a key role
in the development of the storm.
t 1.5 hr
t 3 hr
41
Forecast verification
Rainwater correlation
Extrapolation
Model
Persistence
42
Issues and Opportunities

• Further improvement of data assimilation
  techniques
• New observations
• - Radar refractivity, polarimetric obs.,
• CASA, TAMDAR
• Accuracy of large-scale analysis
• Model error/physical parameterization
• Computation/limited area implementation

43
Sensitivity with respect to first guess
Humidity first guess Background saturation
within convection
Humidity first guess background
44
Issues and Opportunities

• Further improvement of data assimilation
  techniques
• New observations
• - Radar refractivity, polarimetric obs.,
• CASA, TAMDAR
• Accuracy of large-scale analysis
• Model error/physical parameterization
• Computation/limited area implementation

45
Impact of TAMDAR data
Relative humidity without TAMDAR
Relative humidity with TAMDAR
1-hour qr forecast without TAMDAR
1-hour qr forecast with TAMDAR
White contour Observed reflectivity Greater
than 30 dBZ
46
Issues and Opportunities

• Further improvement of data assimilation
  techniques
• New observations
• - Radar refractivity, polarimetric obs.,
• CASA, TAMDAR
• Accuracy of large-scale analysis
• Model error/physical parameterization
• Computation/limited area implementation

47
Sensitivity of the simulation with respect to
environmental condition
48
Issues and Opportunities

• Further improvement of data assimilation
  techniques
• New observations
• - Radar refractivity, polarimetric obs.,
• CASA, TAMDAR
• Accuracy of large-scale analysis
• Model error/physical parameterization
• Computation/limited area implementation

49
Microphysical parameter retrieval
Adjusting model microphysical parameters along
with initial condition by fitting the model to
radar observations
Change of the parameter with respect to iteration
number
Terminal Velocity
Evaporation rate
First Guess
5 m/s - Value in control simulation
Iteration
Iteration
50
Issues and Opportunities

• Further improvement of data assimilation
  techniques
• New observations
• - Radar refractivity, polarimetric obs.,
• CASA, TAMDAR
• Accuracy of large-scale analysis
• Model error/physical parameterization
• Computation/limited area implementation

51
Summary

• Radar data assimilation is one of the critical
  aspects for
• improvement of QPF.
• VDRAS was developed and applied to study the
  high-
• resolution analysis and initialization using
  radar
• observations.
• Case studies and real time implementations have
• demonstrated that the 4DVar-based technique has
  
• potentials in improving short range QPF
• Improving DA techniques, adding new
  high-resolution
• observations, dealing with scale interaction
  and model
• errors, computational efficiency are among a
  series of future
• challenges.

52
WRF 3DVar radar data assimilation Of the IHOP
June 12-13 squall line
Column maximum reflectivity (dBZ)
1-hour forecast
6-hour forecast
No radar DA

• WRF 3DVar assimilates
• Radial velocity and
• Reflectivity simultaneously.
• A warm rain process is used to balance the
• hydrometeors and
• Thermodynamics.

With Radar DA
Observation
Courtesy of Q. Xiao
53
QPF verification
Red No radar Gray one radar Purple 11 radars
Courtesy of Xiao
54
Use of VDRAS Vertical Velocities in Thunderstorm
Nowcasting
60 min extrapolation
Contours of Vertical velocity

0.1 m/s
0.3 m/s
0.5 m/s
55
Use of VDRAS Vertical Velocities in Thunderstorm
Nowcasting
Verification
56
4D-Var vs. EnKF

• Both are constrained by a numerical model
  (dynamical assimilation).
• 4D-Var finds an analysis trajectory using
  several time levels of observations
  (variational), while EnKF produces an analysis at
  a single time level using observations at that
  time level (sequential).
• EnKF is more dependent on background covariance,
  while 4D-Var relies more on observations.
• 4D-Var has a longer history in atmospheric data
  assimilation than EnKF and have had more
  real-time and operational implementations.