Strategies of using radar/conventional data for improving QPF at cloud-resolving scale
by the ensemble Kalman filter
Kao-Shen Chung1 , Weiguang Chang1, Seung-Jong Baek2 and Luc Fillion2
Collaborators: Isztar Zawadzki1, M.K Yau1
1. Dept of Atmospheric and Oceanic Sciences, McGill University2. Meteorological Research Division, Environment Canada
Kao-Shen Chung( 鍾高陞 )National Central University
May 6th, 2014
Outline
1. Introduction of the Canadian High Resolution Ensemble Kalman Filter (HREnKF) system
2. Examine the impact of position error with radar data assimilation
3. Strategies of improving QPF at convective scale a) regional assimilation system b) Adaptive Radar observation
4. Summary and future works
1. High Resolution Ensemble Kalman Filter System ( HREnKF )
Initial guess
Ensemble members
Add random perturbations
Data assimilation
Observations
Perturbed observations
GEM-LAM forecast for all the members
Add random perturbations(model error) Analysis step
Forecast step
GEnKF(2005 operational system)
LAM 1-km (300x300)
HREnKF
for radar data assimilation
Features of the system
Sequential processing of batches of observations
60 10~y
Houtekamer and Mitchell 2001
Sub-ensemble 1
Sub-ensemble 2
Sub-ensemble 3
Ensemble members (80)
Sub-ensemble 4
Gain matrix K1K1 Gain matrix K2K2 Gain matrix K3K3 Gain matrix K4K4
Partitioning the ensemble (to deal with the underestimation of the error structure) )HK( ffa xyxx
• Control variables: U, V, W, T, HU (specific humidity)
• Observations are perturbed according to its variance (no correlation).
• Simplified random perturbations to consider the model errors
• Localization: 10-km in horizontal; 2 * ln( Pressure levels ) in vertical
• 80 members
Some features of the current set up:
For the HREnKF
For the GEM_LAM model at 1-km resolution
• Cycling hydrometeor variables
• Microphysical scheme: double moment scheme (Milbrandt and Yau, 2005)
• Fixed lateral boundary conditions Ensemble lateral boundary conditions for all ensemble members
Summer cases Features
June 12, 2011 Scattered and localized convection. (up to 90-min)
June 23, 2011 Wide spread stratiform system (up to 90-min)
June 29, 2011 Squall line system
QPF Improvement only up to 1-h
Analysis 30-min
90-min60-min
Position error of precipitation
Poor background fields
Is it important? (impact of assimilating radar data)
Is there any way to improve it?
sin)(cos)cossin( Tr VWVUV
In the low elevation angle:
Directly update U and V
Indirect update W through flow-dependent background error cross-covariance
)(),(
22VrInnov
WVrCovW j
oVrj Increment:
22
22
22
22
1
),(),(
VrO
W
OVrW
WVrWVrCov
Error reduction:
2. Examine the impact of position error with radar data assimilation
Experiment designed:
Global EnKF system(GEnKF)
Regional EnKF system(REnKF)
High resolution EnKF(HREnKF)
傳統氣象資料衛星資料
Obtain ensemble members
傳統氣象資料衛星資料
氣象雷達資料
3. Strategies of improving QPF at convective scale
a) conventional observations + regional assimilation system
Impact of Regional EnKF with conventional observations
Vertical velocity W
Global EnKF (ensemble mean) Regional EnKF (ensemble mean)
REnKF_15km 1-day cycling
0000 1200 0000 0600
6-h short-term forecast
Control run (from DF) Precipitation (mean of REnKF)
Improvement of the background field
Control run: No radar assimilation, from ensemble mean of the REnKF
0000
HREnKF: cycling for 60-min and launch the short-term forecast
0000 1830 UTC
0230
Radar radial wind (assimilating every 5-min)
0100
short-term ensemble forecasts 1.5 hr
2.5 h model integration
Cycling assimilation procedure
Experiment of the HREnKF with radial wind assimilation
Impact of using the ensemble set from REnKF system From random perturbations
From REnKF
Converge toward observations
underestimate
Covered the rms of Pf
Verification
Analysis 30-min
90-min60-min
Adaptive radar observations How to optimize using radar observations?
4. Summary and future works
• At cloud-resolving scale, if there is any position error of precipitation, it is important to correct it before assimilate radar observations.
• Initial ensemble set from the REnKF is better than random perturbations 1. Capture mesoscale circulation better 2. Ensemble spread is able to cover forecast errors assimilate more radar observations
• The verification of the radial component shows that the improvement of short-term forecast is up to 1.5-hr. (Both bias and root-mean-square errors)
• How to use radar observation properly? Adaptive observation strategy is able to improve the effectiveness of assimilating radar data
Data assimilation(bridge)
observationsNumerical
model
About future work
The solid line represents the theoretical limit of predictability, the dashed line indicates NWP models, and the dotted line represents nowcasting methods (Austin et al., 1987).
Berenguer et al. 2012
Forecast skill ( nowcasting versus NWP )precipitation
0 - 6 hr QPF
• Resolution of NWP• Extra Observations
Final goal
Reflectivity
0000 UTC 0000 UTC
Simulated reflectivity
0000 UTC
• Construct regional and mesoscale analysis fields
Simulated reflectivity
observations Poor background field Good background field
Large scale forcing Data assimilation
How many cycling of regional EnKF & conventional data?Optimal assimilation window ( 6-h or 3-h ) ?
a. Assimilate both radial wind and reflectivity observations
• Extra Observations (other than conventional data)
(Feng et al. 2009)
(Humidity)(Precipitation)
b. Refractivity Apply to a EnKF system
C. Dual-Polarization observations:
( Putnam et al 2013 )
Microphysics versus Dual-Polarization parameters
( Putnam et al 2013 )
d. Cloud radars
• Use more complicated observation operator
i
i
beam
M
jj
N
iii
e
beam
M
jj
N
iii
er
binr
drrrWddGr
rZ
drrrWddGr
rZrV
V
11
22
11
22
'cos,'cos2exp,,
'cos,'cos2exp,,
,,
• Identify error structure in a) observations
sin)(cos)cossin( Tr VWVUV
Consider: proper geometry, accurate propagation Include: the sampling volume, signal and its processing
< simplified operator >
Fabry and Kilambi (2011)
• Identify error structure in B) numerical model
Error correlation of TT profile V.S. Vertical correlation of TT tendency ( Ensemble Forecasts) (stochastic perturbation of SCM)
Single column model (SCM) Represent the error structure
Microphysics
移動雷達車移動雷達車移動雷達車移動雷達車
JWDJWD雨滴譜儀雨滴譜儀JWDJWD雨滴譜儀雨滴譜儀
2DVD2DVD雨滴譜儀雨滴譜儀2DVD2DVD雨滴譜儀雨滴譜儀
剖風儀剖風儀剖風儀剖風儀
MRRMRR微波雨量微波雨量雷達雷達
MRRMRR微波雨量微波雨量雷達雷達
雨量計雨量計雨量計雨量計
• Verify QPF (quantitative precipitation forecast) over Taiwan region
• Identify error structure in B) numerical model
Global Precipitation Mission (GPM)
報告完畢
歡迎指教
謝謝!
3. Impact of assimilating radial wind componentIs it able to propagate information to other control variables?
Temperature
HumidityV-wind
Obs Z
McGill Algorithm for Precipitation Nowcasting by Lagrangian Extrapolation(MAPLE)
Variational Echo Tracking technique (Laroche and Zawadzki 1995 ):
Estimate the motion field of precipitation and a modified semi-Lagrangian backward scheme for advection.(capable of stretching and rotation)