new data quality of aeolus wind measurements · 2020. 5. 4. · data quality of aeolus wind...
TRANSCRIPT
Data quality of Aeolus wind measurements
Isabell Krisch1
Christian Lemmerz1, Oliver Lux1, Uwe Marksteiner1, Nafiseh Masoumzadeh1, Oliver Reitebuch1, Fabian Weiler1, Benjamin Witschas1, Fabio Bracci2, Markus Meringer2, Karsten Schmidt2, Alexander Geiß3, Dorit Huber4, Ines Nikolaus5, Michael Vaughan6, Alain Dabas7, Thomas Flament7, Dimitri Trapon7, Saleh Abdalla8, Lars Isaksen8, Michael Rennie8, Dave Donovan9, Jos de Kloe9, Gert-Jan Marseille9, Ad Stoffelen9, Thomas Kanitz10, Anne-Grete Straume10, Denny Wernham10, Jonas von Bismarck11, Sebastian Bley11, Peggy Fischer11, Tommaso Parrinello11
1DLR, Institute of Atmospheric Physics 2DLR, Remote Sensing Technology Institute 3Ludwig-Maximilians-Universität München 4DoRIT 5Munich University of Applied Sciences 6OLA
7Météo-France 8ECMWF 9KNMI 10ESA-ESTEC 11ESA-ESRIN
The Aeolus Data Innovation and Science Cluster (DISC)
DLR.de Chart 2 EGU General Assembly 4-8 May 2020 © Aeolus DISC. All rights reserved.
Evolution of Aeolus random and systematic errors ECMWF operational monitoring of Aeolus Rayleigh and Mie winds
DLR.de Chart 3 EGU General Assembly 4-8 May 2020
Random error of ALADIN Rayleigh winds is in the order of 5-7 m/s and 3-4 m/s for Mie winds (mostly clouds): random errors in both channels were increased since launch and did not improve.
Systematic errors (bias) for both Mie and Rayleigh winds were enhanced since launch (several m/s), and show strong temporal variations (slow drifts), orbital variations, differences for ascending and descending orbits, and occurrence in some range-gates combination of 4 unexpected sources for the bias identified up to now
Rayleigh clear global, daily O-B statistics Mie cloudy global, daily O-B statistics
O-B: Difference between Aeolus observation and ECMWF forecasted HLOS wind
FM-B FM-B FM-A FM-A
FM-B with FM-A calibration
FM-B with FM-A calibration
© M. Rennie (ECMWF). All rights reserved.
Evolution of Aeolus random and systematic errors Comparison of Aeolus Rayleigh and Mie winds with 4 German radar wind-profilers
DLR.de Chart 4 EGU General Assembly 4-8 May 2020
Radar wind profilers support error estimates obtained by ECMWF O-B.
© A. Geiß (University LMU Munich). All rights reserved.
1. Laser emit energy
Lower than expected (60mJ instead of 80mJ)
Negative trend
2. Optical signal throughput in receive path for atmospheric signal
Lower than expected (factor 2-3)
Negative trend
What drives the random errors?
DLR.de Chart 5 EGU General Assembly 4-8 May 2020
Rayleigh atmospheric path signal and laser energy for FM-B
Discrepancy between these lines indicates that laser energy is not representative for instrument performance. This hints to a signal
loss in optical emit and/or receive path.
© K. Schmidt (DLR). All rights reserved.
1. Laser emit energy
Lower than expected (factor 1-2)
Negative trend
2. Optical signal throughput in receive path for atmospheric signal
Lower than expected (factor 2-3)
Negative trend
3. Solar background noise
Impact higher than expected due to lower atmospheric signal
Seasonal variation of solar background by factor 18: Rayleigh random errors of 7-8 m/s were obtained in summer months for polar regions
What drives the random errors?
DLR.de Chart 6 EGU General Assembly 4-8 May 2020
Seasonal variation of Rayleigh solar background noise
Orbital variation of Rayleigh solar background noise
March 1, 2019
© K. Schmidt (DLR). All rights reserved.
What causes systematic errors? Combination of 4 unexpected error sources with different temporal characteristics
Systematic dark signal offsets with 10-3 to 10-4 of signal or 1% -10% of noise
1. Higher dark current rates for some “hot pixels”
affects specific range gates; currently 13 pixels on Mie ACCD and 14 pixels on Rayleigh ACCD
DLR.de Chart 7 EGU General Assembly 4-8 May 2020
April 2020, Hot pixel map Hot pixel evolution
Oct Jan Apr Jul Oct Jan Apr Jul Oct 2018 | 2019 | 2020
30
25
20
15
10
5
0
Nu
mb
er o
f h
ot
pix
els
Hot pixels (mean per DCMZ/DUDE measurement)
Mie ACCD
Rayleigh ACCD
© F. Weiler (DLR). All rights reserved.
Correction of hot pixels implemented in June 2019
Enhanced dark signals (hot pixels) cause
systematic errors of ±(1-3 m/s) for Rayleigh
winds for some range gates
=> horizontal stripes
New instrument modes introduced, algorithms developed and implemented in operational processors on June, 14, 2019 for correction of hot pixels by measuring dark signals 4 times per day for real-time datasets
Hot pixels appearing in between the dark signal measurements still cause biases in NRT dataset (are flagged in L2B winds) !
DLR.de Chart 8 EGU General Assembly 4-8 May 2020 R
ayleigh H
LOS w
ind
in m
/s
June 14, 2019
Alt
itu
de
in m
© Fig: VirES for Aeolus; correction developed by
F. Weiler (DLR) and D. Huber (DoRIT) All rights reserved.
What causes systematic errors? Combination of 4 unexpected error sources with different temporal characteristics
1. Higher dark current rates for some “hot pixels”
2. Error in the on-board software in calculation of residual projection of the satellite ground speed on the line-of-sight LOS:
harmonic variation of bias along the orbit
correction with on-board calculated vSAT in L1B and L2B processors de-activated in summer 2019
correction for on-ground L1A processing identified and implementation envisaged for autumn 2020
DLR.de Chart 9 EGU General Assembly 4-8 May 2020
vSAT during June 29 – Jul 05, 2019
Amplitude of vLOS is zero at equator and maximum at poles with around 0.6 m/s, but opposite phase as in December
vSAT during Dec 18 – 24, 2018
© I. Nikolaus (Psol). All rights reserved.
© B. Witschas (DLR). All rights reserved.
What causes systematic errors? Combination of 4 unexpected error sources with different temporal characteristics
1. Higher dark current rates for some “hot pixels”
2. Error in the on-board software in calculation of residual projection of the satellite ground speed on the line-of-sight LOS
3. Slow drifts in the illumination of the Rayleigh/Mie spectrometers causing a slowly, linear drifting constant bias
DLR.de Chart 10 EGU General Assembly 4-8 May 2020
09.2018 11.2018 01.2019 03.2019 05.2019 07.2019 09.2019 11.2019 01.2020 03.2020
5320
5340
5360
5380
5400
5420
5440
5460
5480
5500
5520
-(0.07 ± 0.02) MHz/d
Switch
to FMB
-(0.32 ± 0.02) MHz/d
-(0.87 ± 0.06) MHz/d
Spacin
g/(
MH
z)
Date
-(1.52 ± 0.03) MHz/d
-(0.12 ± 0.04) MHz/d
Rayleigh spectrometer filter spacing
Rayleigh filter A and B spectral difference nominal 2.3 pm = 5460 MHz
=> change of 0.4 fm/d (2∙10-4 /d) => wind speed dependent error
filter A
Response
𝑹 =𝑰𝑨 − 𝑰𝑩𝑰𝑨 + 𝑰𝑩
filter
B
What causes systematic errors? Combination of 4 unexpected error sources with different temporal characteristics
1. Higher dark current rates for some “hot pixels”
2. Error in the on-board software in calculation of residual projection of the satellite ground speed on the line-of-sight LOS
3. Slow drifts in the illumination of the Rayleigh/Mie spectrometers causing a slowly, linear drifting constant bias
DLR.de Chart 11 EGU General Assembly 4-8 May 2020
Rayleigh spectrometer center frequencies
07.2019 08.2019 09.2019 10.2019 11.2019 12.2019 01.2020 02.2020 03.2020 04.2020
844.95526
844.95528
844.95530
844.95532
844.95534
844.95536
844.95538
844.95540
(0.29 ± 0.02) MHz/day
(0.11 ± 0.1) MHz/day
(0.26 ± 0.03) MHz/day
f0/(
TH
z)
- d
irect
cha
nn
el
Date
150 MHz
Modell Line
Gleichung y = A + B*x
Chi-Quadr Reduziert
4.8999998E-12
Kor. R-Quadrat -0.5
Wert Standardfehler
f0/(THz)A 844.6743259 0.3477182
Modell Line
Gleichung y = A + B*x
Chi-Quadr Reduziert
2.6666664E-13
Kor. R-Quadrat -0.3333333
Wert Standardfehler
f0/(THz)A 844.2528445 0.0573589
Laser energy increase by 0.5 mJ
positive T-change CP
filter A
07.2019 08.2019 09.2019 10.2019 11.2019 12.2019 01.2020 02.2020 03.2020 04.2020
844.96060
844.96062
844.96064
844.96066
844.96068
844.96070
844.96072
844.96074
(0.14 ± 0.05) MHz/day.
(0.11 ± 0.06) MHz/day.
(0.27 ± 0.03) MHz/day.
f0/(
TH
z)
- re
fle
cte
d c
han
ne
l
Date
150 MHz
Modell Line
Gleichung y = A + B*x
Chi-Quadr Reduziert
8.7689859E-13
Kor. R-Quadrat 1.0065909
Wert Standardfehler
f0/(THz) A 844.2868598 0.0677506
filter B
filter A
Response
𝑹 =𝑰𝑨 − 𝑰𝑩𝑰𝑨 + 𝑰𝑩
filter
B
© B. Witschas (DLR). All rights reserved.
What causes systematic errors? Combination of 4 unexpected error sources with different temporal characteristics
1. Higher dark current rates for some “hot pixels”
2. Error in the on-board software in calculation of residual projection of the satellite ground speed on the line-of-sight LOS
3. Slow drifts in the illumination of the Rayleigh/Mie spectrometers causing a slowly, linear drifting constant bias
Was in the past corrected manually by tuning of input parameters for L2B processing
Now corrected in real-time products as part of M1-correction (see next slide)
DLR.de Chart 12 EGU General Assembly 4-8 May 2020
filter A
Response
𝑹 =𝑰𝑨 − 𝑰𝑩𝑰𝑨 + 𝑰𝑩
filter
B
Temporal evolution of internal Response 𝑹𝑰𝒏𝒕
Aug 2019 Jan 2020
0.01 = 20 MHz = 5.8 m/s HLOS
slowly, linear drifting constant bias
© F. Weiler (DLR). All rights reserved.
What causes systematic errors? Combination of 4 unexpected error sources with different temporal characteristics
1. Higher dark current rates for some “hot pixels”
2. Error in the on-board software in calculation of residual projection of the satellite ground speed on the line-of-sight LOS
3. Slow drifts in the illumination of the Rayleigh/Mie spectrometers causing a slowly, linear drifting constant bias
4. Thermal variations of the M1 telescope mirror
Rayleigh bias with orbital phase (argument of latitude) and longitude
Use correlation between M1 temperatures and mean model bias for correction
DLR.de Chart 13 EGU General Assembly 4-8 May 2020 Mean telescope M1 mirror temperatures
M1
Te
mp
. /
°C
11.95
11.90
11.85
11.80
11.75
11.70
180° 150°W 120°W 90°W 60°W 30°W 0° 30°E 60°E 90°E 120°E 150°E 180°
90°N
75°N
60°N
45°N
30°N
15°N
0°
15°S
30°S
45°S
60°S
75°S
90°S July 1 – Oct 25, 2019
Rayleigh bias against ECMWF model (O-B)
Aug 13 – Sep 14, 2019 400hPa
ascending orbits
© F. Weiler (DLR) and M. Rennie (ECMWF). All rights reserved.
Correction of bias caused by thermal variations of the M1 mirror
DLR.de Chart 14 EGU General Assembly 4-8 May 2020
Multiple linear regression (1d of data)
Effects of M1 correction: 1. „flattens out“ orbital variation (M1 telescope mirror) – reduces std. deviation of O-B (here: 2.62 m/s to 0.76 m/s) 2. Corrects for bias drifts (e.g. illumination effects)
stdev(<O-B>): • 2.62 m/s • 1.05 m/s • 0.76 m/s
Rayleigh bias versus time on 09/08/2019
Hour of day
HLO
S w
ind
me
an(O
-B)
/ m
/s
6
4
2
0
-2
-4
-6
© F. Weiler (DLR) and M. Rennie (ECMWF). All rights reserved.
Correction of bias caused by thermal variations of the M1 mirror Activated on April 20, 2020
DLR.de Chart 15 EGU General Assembly 4-8 May 2020
New instrument modes introduced, algorithms developed and implemented in operational processors on April, 20, 2020 for correction of M1 temperature biases and slowly, linear drifting constant biases by using daily mean correlation between ECMWF model bias (O-B) and M1 temperature measurements.
Additional investigations are on-going to use ground measurements instead of ECMWF O-B to regain model independence.
Mie cloudy global, daily O-B statistics Rayleigh clear global, daily O-B statistics
© M. Rennie (ECMWF). All rights reserved.
Re-processing of Aeolus data
• Re-processing activities started beginning of this year
• It involves multiple manual and only semi-automated steps, e.g. Manual production of calibration files and
processing up to L1B Correction of hot pixels also in time periods
between dark signal measurements Semi-automated processing up to L2B Manual estimation of M1 temperature correction Verification of bias correction and delivery of all
calibration files to ESA Re-processing at ESA Validation and quality control of re-processed
dataset
• First re-processed dataset is FM-B data from 2019. It will be available in Sept. 2020.
DLR.de Chart 16 EGU General Assembly 4-8 May 2020
Sept 2020
early 2021
2022
Re-processed FM-B data of 2019 including M1 correction will become
available
Re-processed FM-A data with hot-pixel and M1 correction or re-processed FM-B
data of 2020 will become available
Re-processing of complete Aeolus dataset since launch with optimized processor settings and all available corrections.
© Aeolus DISC. All rights reserved.
Summary and Conclusion
• The Aeolus DISC consortium is responsible for calibration, processor evolution, product quality and impact studies.
• Both systematic and random errors after launch were higher than expected
• Precise instrument characterization and use of O-B statistics enabled a drastic reduction of systematic errors
• Hot pixel correction was implemented in June 2019 for NRT data stream
• M1 temp. correction was implemented in April 2020 for NRT data stream and is currently examined by experts.
Public data release on May 12, 2020
• Re-processing started recently and first re-processed data (June – December 2019) will be available in Sept. 2020.
DLR.de Chart 17 EGU General Assembly 4-8 May 2020
Sept. 2020 early 2021 2022
FM-B 2019 data FM-A data Complete
Aeolus dataset
Reprocessing
Hot pixel correction implemented
M1 bias correction implemented
Public NRT data release 12 May
2020
© Aeolus DISC. All rights reserved.