Radar Data Assimilation in a Case of Deep Convection in ... · PDF fileRadar Data Assimilation...
20
Introduction Performance of the SCALE–LETKF Sensitivity to IC and BC Sensitivity to filter parameters Summary l Radar Data Assimilation in a Case of Deep Convection in Argentina Paula Maldonado Centro de Investigaciones del Mar y la Atmósfera (CIMA-CONICET/UBA) Departamento de Ciencias de la Atmósfera y los Océanos (DCAO-FCEN-UBA) Supervisor: Takumi Honda DA Seminar June 16th, 2017 P. Maldonado, T. Honda CIMA-CONICET/UBA DA Seminar August 16th, 2017 1 / 17
Transcript of Radar Data Assimilation in a Case of Deep Convection in ... · PDF fileRadar Data Assimilation...
Introduction Performance of the SCALE–LETKF Sensitivity to IC and BC Sensitivity to filter parameters Summary
l
Radar Data Assimilation in a Case of Deep Convection inArgentina
Paula MaldonadoCentro de Investigaciones del Mar y la Atmósfera (CIMA-CONICET/UBA)
Departamento de Ciencias de la Atmósfera y los Océanos (DCAO-FCEN-UBA)
Supervisor: Takumi HondaDA Seminar
June 16th, 2017
P. Maldonado, T. Honda CIMA-CONICET/UBA DA Seminar August 16th, 2017 1 / 17
Introduction Performance of the SCALE–LETKF Sensitivity to IC and BC Sensitivity to filter parameters Summary
l
High-impact Weather Events
Days per Year with Favorable Severe Parameters (Brooks et. al, 2003)Soundings with CAPE > 100 J kg−1 and ∂θ
∂t (2 − 4 km) > 6.5 K km−1
March 9th, 2017 - Buenos Aires May 18th, 2017 - Córdoba
April 3rd, 2013 - La Plata
P. Maldonado, T. Honda CIMA-CONICET/UBA DA Seminar August 16th, 2017 2 / 17
Introduction Performance of the SCALE–LETKF Sensitivity to IC and BC Sensitivity to filter parameters Summary
l
High-impact Weather Events
Days per Year with Favorable Severe Parameters (Brooks et. al, 2003)Soundings with CAPE > 100 J kg−1 and ∂θ
∂t (2 − 4 km) > 6.5 K km−1March 9th, 2017 - Buenos Aires May 18th, 2017 - Córdoba
April 3rd, 2013 - La Plata
P. Maldonado, T. Honda CIMA-CONICET/UBA DA Seminar August 16th, 2017 2 / 17
Introduction Performance of the SCALE–LETKF Sensitivity to IC and BC Sensitivity to filter parameters Summary
l
Convective-scale Forecasts
High-resolution NWP Models (< 4 km)4 Eliminate uncertainty associated with
cumulus parameterization
% Significant errors in location andtiming of convective systems
Remote Sensing Observations4 Describe the state of the atmosphere
in the convective scale
Advantages of Using Radar Data
High-resolution, 3D observations
Temporal frequency necessary toretain the storm’s structure
SINARAME Project
FIGURE – Weather radar network nowadays (blueline) and SINARAME radars (red line).Simple-pol (dash line) and dual-pol (so-lid line) radars.
P. Maldonado, T. Honda CIMA-CONICET/UBA DA Seminar August 16th, 2017 3 / 17
Introduction Performance of the SCALE–LETKF Sensitivity to IC and BC Sensitivity to filter parameters Summary
l
Objectives
MAIN GOALDevelop, implement and evaluate a radar data assimilation system based on the
Local Ensemble Transform Kalman Filter (LETKF) to improve very short-term weatherforecast of high-impact weather events in South America
TALK’S GOALS
1 Evaluate the performance of the SCALE–LETKF system using differentassimilation strategies
2 Assess the sensitivity of the system to :Initial and boundary conditionsFilter parameters
P. Maldonado, T. Honda CIMA-CONICET/UBA DA Seminar August 16th, 2017 4 / 17
Introduction Performance of the SCALE–LETKF Sensitivity to IC and BC Sensitivity to filter parameters Summary
l
Objectives
MAIN GOALDevelop, implement and evaluate a radar data assimilation system based on the
Local Ensemble Transform Kalman Filter (LETKF) to improve very short-term weatherforecast of high-impact weather events in South America
TALK’S GOALS
1 Evaluate the performance of the SCALE–LETKF system using differentassimilation strategies
2 Assess the sensitivity of the system to :Initial and boundary conditionsFilter parameters
P. Maldonado, T. Honda CIMA-CONICET/UBA DA Seminar August 16th, 2017 4 / 17
Introduction Performance of the SCALE–LETKF Sensitivity to IC and BC Sensitivity to filter parameters Summary
l
Radar Observations
Radar CharacteristicsC-band, dual-pol, Doppler radar
Resolution : 500 m (range), 1◦ (azimuth)Scanning mode :
1 120 km scan : 10 min (Ref and DV)2 240 km scan : 10 min (Ref)
0 20 40 60 80 100 120 140 160 180 200 220 240Range (km)
02468
101214161820
Height (k
m)
0.5°
1.3°
2.3°
3.5°
5.0°6.9°9.1°11.8°15.1°19.2°
Case Study : Anguil, La Pampa (Argentina) - January 11th, 2010
Strong deep convective clouds includingsupercells were observed
FIGURE – Photographs taken by locals
P. Maldonado, T. Honda CIMA-CONICET/UBA DA Seminar August 16th, 2017 5 / 17
Introduction Performance of the SCALE–LETKF Sensitivity to IC and BC Sensitivity to filter parameters Summary
l
Quality Control and SuperobbingR
eflec
tivity
Raw Data QC Data Super Obs
Dop
pler
Velo
city
P. Maldonado, T. Honda CIMA-CONICET/UBA DA Seminar August 16th, 2017 6 / 17
Introduction Performance of the SCALE–LETKF Sensitivity to IC and BC Sensitivity to filter parameters Summary
l
Experimental Design
Model Domains
D01 D02
Resolution 10 km 2 kmDomain Size (2000 km)2 (500 km)2
IC-BC GFS 0.5◦ D01Observations PREPBUFR RadarCycle Length 1 hr ***Assimilation Period 10 days 3 hs
Ensemble size : 60
Assimilation Strategy Experiments
5min_3D : Assume whole volume ismeasured at same time
5min_4D : Assimilation window with1 min slots
1min_3D : Divide radar volume in1 min files
P. Maldonado, T. Honda CIMA-CONICET/UBA DA Seminar August 16th, 2017 7 / 17
Introduction Performance of the SCALE–LETKF Sensitivity to IC and BC Sensitivity to filter parameters Summary
l
Qualitative Evaluation
Analysis mean after 135 minutes at 1.3◦ elevation
Observation No DA
1min_3D 5min_4D 5min_3D
P. Maldonado, T. Honda CIMA-CONICET/UBA DA Seminar August 16th, 2017 8 / 17
Introduction Performance of the SCALE–LETKF Sensitivity to IC and BC Sensitivity to filter parameters Summary
l
Quantitative Evaluation
Analysis mean RMSE (solid) and BIAS (dot)
0 10 20 30 40 50 60 70 80 90 100110120130140150160170Assimilation time (min)
0
5
10
15
Refle
ctivity
(dBZ
)
1min_3D5min_4D5min_3D
0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 160 170Assimilation time (min)
−2
0
2
4
6
Dopp
ler V
eloc
ity (m
/s) 1min_3D
5min_4D5min_3D
RMSE increase between60 min and 90 min
Systematic error for bothvariables
Small error differencebetween 5min_4D and5min_3D
Overall, assimilating thewhole radar volume at onetime (green line) seems toshow the best results
P. Maldonado, T. Honda CIMA-CONICET/UBA DA Seminar August 16th, 2017 9 / 17
Introduction Performance of the SCALE–LETKF Sensitivity to IC and BC Sensitivity to filter parameters Summary
l
Doppler Velocity
Analysis mean after 10 minutes at 0.5◦ elevation
Observation No DA 5min_3D
Poor representation of Doppler velocity in low levels near the radar site
Synoptic-scale initial condition is not "good" enough
P. Maldonado, T. Honda CIMA-CONICET/UBA DA Seminar August 16th, 2017 10 / 17
Introduction Performance of the SCALE–LETKF Sensitivity to IC and BC Sensitivity to filter parameters Summary
l
Experimental Design
IC and BC Experiments
5min_3D_D01 : IC and BC from D01
5min_3D_GFS : IC and BC from GFSanalysis
Wind difference at 15Z at 1000 m
D01 (black vector), GFS (green vector),D01-GFS (shaded)
P. Maldonado, T. Honda CIMA-CONICET/UBA DA Seminar August 16th, 2017 11 / 17
Introduction Performance of the SCALE–LETKF Sensitivity to IC and BC Sensitivity to filter parameters Summary
l
Results
Analysis mean RMSE (solid) and BIAS (dot)
0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 160 170Assimilation time (min)
0
5
10
15
Refle
ctivity
(dBZ
)
5min_3D_D015min_3D_GFS
0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 160 170Assimilation ime (min)
−2
0
2
4
6
8
Dopp
ler V
eloc
i y (m
/s) 5min_3D_D01
5min_3D_GFS
Similar errors in reflectivity
Systematic errors are stillpresent
Fewer DV observations arerejected by LETKF qualitycontrol in 5min_3D_GFSthan in 5min_3D_D01
P. Maldonado, T. Honda CIMA-CONICET/UBA DA Seminar August 16th, 2017 12 / 17
Introduction Performance of the SCALE–LETKF Sensitivity to IC and BC Sensitivity to filter parameters Summary
l
Analysis mean after 10 minutes at 0.5◦ elevation
Observation 5min_3D_D01 5min_3D_GFS
Analysis mean after 65 minutes at 1.3◦ elevation
Observation 5min_3D_D01 5min_3D_GFS
P. Maldonado, T. Honda CIMA-CONICET/UBA DA Seminar August 16th, 2017 13 / 17
Introduction Performance of the SCALE–LETKF Sensitivity to IC and BC Sensitivity to filter parameters Summary
l
Experimental Design
Based on the 5min_3D_GFS experiment :
1 Assimilation of clear–sky observations to eliminate spurious convectionClear–sky observation : ref < 10 dBZAssimilation : XX number of members of the background ensemble have ref > 10 dBZ
2 Covariance inflation to increase ensemble spreadMultiplicative inflation factorRelaxation to prior spread (RTPS)
Filter Parameters Experiments
Cntl (5min_3D_GFS) : RainMem = 10 ; Multiplicative inflation = 1.0 ; RTPS = 0.9
RainMem05 : At least 5 members of the first guess ensemble must have rain
RainMem01 : At least 1 members of the first guess ensemble must have rain
Inflation : Multiplicative inflation = 1.1 ; RTPS = 0.95
P. Maldonado, T. Honda CIMA-CONICET/UBA DA Seminar August 16th, 2017 14 / 17
Introduction Performance of the SCALE–LETKF Sensitivity to IC and BC Sensitivity to filter parameters Summary
l
Results
Analysis mean RMSE (solid) and BIAS (dot)
0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 160 170Assimilation time (min)
0
5
10
15
Refle
ctivity
(dBZ
)
CntlRainMem05RainMem01Inflation
0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 160 170Assimilation time (min)
−2
0
2
4
6
8
Dopp
ler V
eloc
ity (m
/ ) Cntl
RainMem05RainMem01Inflation
Assimilation of a largeramount of clear-skyobservations improves thereflectivity RMSE
Increasing inflation has aclear impact on Dopplervelocity
P. Maldonado, T. Honda CIMA-CONICET/UBA DA Seminar August 16th, 2017 15 / 17
Introduction Performance of the SCALE–LETKF Sensitivity to IC and BC Sensitivity to filter parameters Summary
l
Short-term deterministic forecast10
min
fcst
Observation at 1 km No DA RainMem01
30m
infc
st
P. Maldonado, T. Honda CIMA-CONICET/UBA DA Seminar August 16th, 2017 16 / 17
Introduction Performance of the SCALE–LETKF Sensitivity to IC and BC Sensitivity to filter parameters Summary
l
Summary
Assimilation of radar observations using the SCALE–LETKF system has a positiveimpact on both, analysis mean and very short-term weather forecasts
Initialization of the ensemble from the GFS analysis shows better results possiblybecause of the lack of conventional observations in the Southern Hemisphere
Assimilation of clear-sky observations improves the reflectivity pattern byeliminating spurious convective zones
Bigger ensemble spread helps reduce errors in the Doppler velocity field
Future Work
Improve quality control of observations
Test the assimilation system for other cases (i.e. for different types of convectionorganization)
Assimilation of satellite observations to improve the mesoscale ensemble
Assimilation of polarimetric variables
Thank you very much!
P. Maldonado, T. Honda CIMA-CONICET/UBA DA Seminar August 16th, 2017 17 / 17
Introduction Performance of the SCALE–LETKF Sensitivity to IC and BC Sensitivity to filter parameters Summary
l
Summary
Assimilation of radar observations using the SCALE–LETKF system has a positiveimpact on both, analysis mean and very short-term weather forecasts
Initialization of the ensemble from the GFS analysis shows better results possiblybecause of the lack of conventional observations in the Southern Hemisphere
Assimilation of clear-sky observations improves the reflectivity pattern byeliminating spurious convective zones
Bigger ensemble spread helps reduce errors in the Doppler velocity field
Future Work
Improve quality control of observations
Test the assimilation system for other cases (i.e. for different types of convectionorganization)
Assimilation of satellite observations to improve the mesoscale ensemble
Assimilation of polarimetric variables
Thank you very much!
P. Maldonado, T. Honda CIMA-CONICET/UBA DA Seminar August 16th, 2017 17 / 17
