Radar Data Assimilation Using VDRAS and WRF-VAR
Juanzhen SunNCAR, Boulder, Colorado
Oct 17, 2011
AcknowledgmentHongli WangQingnong XiaoYing ZhangZhuming Ying
Outline
• Historical Background• Key findings with VDRAS• Experiences with WRF-VAR
- Relative impact of VR and RF
• Thoughts on the future
Oct 17, 2011
Historical Background (VDRAS)
• Early works focused on proof of concept - Single Doppler retrieval for the boundary layer (Sun et al 1991, 1994)
- Initializing cloud-scale model (Sun and Crook 1997, 1998)
• Real-time applications of VDRAS - Real-time nowcasting for NWS, ATEC, DOD,…(Sun & Crook 2001)
- Demonstrations in two summer Olympics (Crook and Sun 2004, Sun et al 2010)
• VDRAS as a tool for
- Understanding convective dynamics and developing conceptual models for nowcasting
- Providing predictor fields for Automated nowcasting systems
- Initialization of mesoscale models (Liou et al 2011)
Historical Background (WRF-VAR)
• WRF 3DVAR radar data assimilation - Convective rainfall (Xiao et al 2005, Xiao and Sun 2007)
- Tropical cyclone (Xiao et al 2007, Pu et al 2009)
- Statistical evaluation over consecutive periods (Xiao et al 2008,
Sun et al 2011)
- Recent improvement on reflectivity assimilation
> Cloud analysis
> Assimilate rainwater instead of reflectivity
> Use of saturation water vapor as data in the cost function
• WRF 4DVAR radar data assimilation - Adjoint of a warm rain microphysics
- Control variables of microphysics
- Being tested with convective cases
What is the adequate resolution to resolvesome small-scale features?
VDRAS continuous analyses of divergence and windFrame interval: 15 min
3KM 1KM
Inserting VDRAS analysis into WRF inner domain
• Interpolated fields of VDRAS to WRF inner domain– U-wind at the 1st level– Without (left) and with (right) blending of VDRAS & WRF near boundaries
RTFDDA_d02
VDRAS
19 UTC 15 June 2002
0
0.15
0.3
0.45
6 10 14 18Thresholds (mm)
ET
S
OBS_EC
WRF
OBS_EC+WRF
ETS score for accumulated 2-hr rainfallOBS_EC: Using VDRAS alone.WRF : Using WRF alone.OBS_EC+WRF: Combining VDRAS and WRF.
Lessons learned by running VDRAS
• 10-15 min 4DVAR window seems to be optimal for analyzing the convective-scale dynamical and thermodynamical structures • Continuous 4DVAR cycling reveals dynamically consistent evolution of convective features• The rapid updating and use of derived rainwater avoids displacement error of storms, a common problem in microphysical initialization• Radial velocity plays a more important role than reflectivity• VDRAS analysis can be used to initialize mesoscale models• A two-step procedure (non-radar data are used to provide an improved storm environment before radar DA) enables a closer fit to radar observations• 1 km resolution resolves much more convective details
Study of a supercell storm using a 4DVAR system VDRASSun (2004)
Observation Forecast
Color contour: qr
w
w
qv
qv
Radial velocity only
Reflectivity only
Observation
Z only
vr only
vr and z
• Without radial velocity, the rain falls out quickly.• Radial velocity assimilation results in slantwise updraft and moisture, but not the reflectivity assimilation
Rainwater correlation
Recent WRF 3DVAR Experiments
• IHOP June 10 – June 16 one week continuous run
- Active convective period - 3 hourly update cycle - 3 km horizontal resolution - Assimilate 25 radars - Cloud analysis option
• Beijing 4 cases evaluation
- 3 hourly update cycle - 3 km horizontal resolution - Assimilate 6 radars - Assimilate in-cloud saturated water vapor
Beijing
IHOP Results
• NORD: Control with no radar DA • RV: Assimilate radial velocity• RF: Assimilate reflectivity• RVRF: Assimilate both
One-week FSS skill (5mm)
RFRV
6-h Forecasts after four 3DVAR cycles
Beijing Results
• NORD: Control with no radar DA• RV: Assimilate radial velocity• RF: Assimilate reflectivity• RVRF: Assimilate both
FSS skill for four 2009 summer cases
FSS skill for July 22, 2009
RV RF
FSS skill for July 23, 2009
WRF 4DVAR Radar Data Assimilation4-hour forecasts from a case study (13 June 2002)
OBS 3DVAR
4D_RV 4D_RF
Thoughts on the future
• Technical improvement of DA systems
- Observation error statistics – based on information content - Background error statistics – evolving with system improvement - Rapid update cycle less than 1 hour for 3DVAR - Choice of control variables - Add terrain effect (VDRAS)
• Polarimetric radar data assimilation
- Observation operator - Use estimated microphysics - Quantify the impact on forecast
Thoughts on the future…Continued
• Further study of Radar DA impact on convective forecasting
- QPF: dependence on convection type, diurnal cycle, scale, etc… - Wind, temperature, humidity
• Improving accuracy of storm environment
- Operational models that are used as background do not have required accuracy for convective initiation forecast, especially in the low-level - Radar clear-air returns do not have adequate coverage - Make better use of other observations, e.g., surface obs.
Diurnal variation of Radar DA impact
00Z
12Z
• Radar DA has longerpositive impact for late Evening initializations
• The positive impactonly lasted 4 hours formorning initializations
• It seems to indicate that the radar DA worksmore effectively for growing storms thandissipation storms