Assimilation of GPS Radio-Occultations at DWD Harald Anlauf - - PowerPoint PPT Presentation

▶
assimilation of gps radio occultations at dwd
SMART_READER_LITE
LIVE PREVIEW

Assimilation of GPS Radio-Occultations at DWD Harald Anlauf - - PowerPoint PPT Presentation

Feb 05, 2023 95 likes •319 views

Assimilation of GPS Radio-Occultations at DWD Harald Anlauf Research and Development, Data Assimilation Section Deutscher Wetterdienst, O ff enbach, Germany IROWG 2nd Workshop, Estes Park (Colorado, USA) 28. March 2012 Harald Anlauf (DWD)

slide-1
SLIDE 1

Assimilation of GPS Radio-Occultations at DWD

Harald Anlauf

Research and Development, Data Assimilation Section Deutscher Wetterdienst, Offenbach, Germany

IROWG 2nd Workshop, Estes Park (Colorado, USA)

  • 28. March 2012

Harald Anlauf (DWD) Assimilation of GPSRO at DWD

  • 28. March 2012

1 / 14

slide-2
SLIDE 2

NWP Models at DWD

GME

Global model, hydrostatic Triangular grid, mesh size: 20 km 60 levels (top: 5 hPa) (1474562×60 grid points) Forecast times: 174h from 00Z, 12Z; 48h from 06Z, 18Z

COSMO-EU

Non-hydrostatic Mesh size: 7 km 40 levels Forecast times: 78h from 00Z, 12Z; 48h from 06Z, 18Z

COSMO-DE

Non-hydrostatic, “convection allowing” Mesh size: 2.8 km 50 levels Forecast times: 21h from 00Z, 03Z, . . . , 21Z

COSMO-DE-EPS

(Pre-operational) Ensemble prediction system 20 ensemble members (operationally: 40 members) Forecast times: 21h from 00Z, 03Z, . . . , 21Z

Harald Anlauf (DWD) Assimilation of GPSRO at DWD

  • 28. March 2012

2 / 14

slide-3
SLIDE 3

Global Data Assimilation System at DWD

3D-Var-PSAS, 3-hourly update cycle Available Forward models for GPSRO

I 1d bending angle operator (Implementation by Michael Gorbunov);

fixed/effective tangent point for profile or individual tangent point

I 3d ray tracer (Michael Gorbunov)

Implementations tested and evaluated in collaboration with GFZ using data from CHAMP and GRACE (Pingel and Rhodin, 2009)

I Ray tracer: best in terms of std.dev. of OBS-FG, numerically expensive!

Ray tracer also needs additional data not provided in BUFR messages (satellite positions and velocities)

I 1d-operator (Abel integral) with effective location of occultation

probably good enough for initial operational implementation (still needed major optimization efforts for the NEC SX-9)

Refractivity 3-term expression as recommended by GRAS-SAF “Initialization” MSIS-90 climatology matched at model top (∼ 36 km)

Harald Anlauf (DWD) Assimilation of GPSRO at DWD

  • 28. March 2012

3 / 14

slide-4
SLIDE 4

Assimilation of GPS Radio-Occultations

Observation errors (S. Healy):

I Linear decrease from 10% to 1% for impact height from 0 to 10 km I 1% from 10 km to 30 km

Quality control of observations:

I Consistency checks of profiles I Observation-minus-first guess check: 4σ (should be made stricter) I B.a. < 0.02 rad to avoid ducting (replace by condition on refractivity) I Clip lowest section of GPS-RO profiles when non-monotonous

Vertical thinning to model resolution, exponential smoothing Use impact heights 3 km–30 km Exclude occultations starting above 20 km GPS Radio-Occultations operationally used since 2010-08-03

I COSMIC/FORMOSAT-3 FM 1-2, 4-6 (FM-3 dead since 2010-08-01) I GRACE-A I GRAS on METOP-A I TerraSAR-X (since 2010-12-09) I C/NOFS, SAC-C (since 2012-02-29) Harald Anlauf (DWD) Assimilation of GPSRO at DWD

  • 28. March 2012

4 / 14

slide-5
SLIDE 5

Impact of the Assimilation of GPSRO

(90% c.l.)

  • H. Anlauf et al.,
  • Atmos. Meas. Tech.,

4, 1105–1113

Better fit to radiosondes in upper troposphere/lower stratosphere,

  • esp. southern hemisphere (but mixed results in Antarctic)

Significant forecast improvements with assimilation of GPSRO

0.4 0.5 0.6 0.7 0.8 0.9 1 24 48 72 96 120 144 168 Anomaly correlation forecast time (h) ANOC geopotential 500 hPa SH 2010070100 till 2010073100 (31 forecasts) GME routine GMEP GPSRO

Anomaly Correlation of Geopotential 500 hPa, Southern Hemisphere for July 2010 ⇒ gain of several hours vs. operational system

Harald Anlauf (DWD) Assimilation of GPSRO at DWD

  • 28. March 2012

5 / 14

slide-6
SLIDE 6

Comparison to Radiosondes

  • H. Anlauf et al.,
  • Atmos. Meas. Tech.,

4, 1105–1113

Mean departures of temperature and rel. humidity observations from radiosondes to 3-h forecasts, Southern Hemisphere

  • 2
  • 1

1 2 Bias of Temperature [K] 925 850 700 500 400 300 250 200 150 100 70 50 30 20 10

Area=-20/-60 0/360

0.0 0.5 1.0 1.5 2.0 2.5 3.0 RMS error of Temperature [K] 925 850 700 500 400 300 250 200 150 100 70 50 30 20 10 Pressure (hPa) Pressure (hPa)

1641 2347 2530 2561 2480 2460 2479 2407 2384 2398 2139 2197 1964 1580 732 211

  • 1

11 13 53 44 50 40 73 85 73 266 93 232

  • 18

114 11

  • 10
  • 5

5 10 Bias of Relative Humidity [%] 925 850 700 500 400 300 250 200

Area=-20/-60 0/360

5 10 15 20 RMS error of Relative Humidity [%] 925 850 700 500 400 300 250 200 Pressure (hPa) Pressure (hPa)

1666 2273 2158 2127 2262 2216 2183

  • 16

1

  • 20

24 56 53 17

(blue: ctrl (operational), red: GPSRO experiment)

Harald Anlauf (DWD) Assimilation of GPSRO at DWD

  • 28. March 2012

6 / 14

slide-7
SLIDE 7

Problems with the Assimilation of GPSRO over Antarctic

Large temperature bias, got even worse with assimilation of GPSRO! Partially understood: poor representation of vertical correlations in

  • perational assimilation system, revised in December 2010

Some issues with the forecast model, but investigations ongoing

Harald Anlauf (DWD) Assimilation of GPSRO at DWD

  • 28. March 2012

7 / 14

slide-8
SLIDE 8

Issues with Assimilation of GPSRO Data (I)

Bias, RMS differences for different satellites (processing, model, . . . ) Lower troposphere: largest bias in the tropics, smaller in extratropics

Tropics

  • 0.04 -0.02 0.00 0.02 0.04

Bias of (O-B)/O

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30

Area=20/-20 0/360

Solid=OBS-FG, Dashed=OBS-AN

0.00 0.02 0.04 0.06 0.08 0.10 0.12 RMS error of (O-B)/O

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30

Impact height (km) Impact height (km)

488 2607 8181 10805 25943 26022 46611 36484 40312 43359 44738 46080 47819 25177 30679 40141 22432 25797 78 24258 62 23340 46 21601 9915 17931 9704 3451 12438 30277 20876 32138 21818 32627 22187 23454 24019 23941 23550 18774 16320 11084 21239 10233 11291 88 10308 17 9725 22 9308 26 12087 18

Statistics for Bending Angles from METOP / GPS RO EXP=rou OBS minus FG/AN for: Surface=all Flag=used SatId= 4 Time period = 20110101 00UTC - 20110831 21UTC, STEP=3h Statistics for Bending Angles from COSMIC / GPS RO EXP=rou OBS minus FG/AN for: Surface=all Flag=used SatId= 740 Time period = 20110101 00UTC - 20110831 21UTC, STEP=3h Harald Anlauf (DWD) Assimilation of GPSRO at DWD

  • 28. March 2012

8 / 14

slide-9
SLIDE 9

Issues with Assimilation of GPSRO Data (I)

Bias, RMS differences for different satellites (processing, model, . . . ) Lower troposphere: largest bias in the tropics, smaller in extratropics

Northern Extratropics

  • 0.04 -0.02 0.00 0.02 0.04

Bias of (O-B)/O

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30

Area=60/20 0/360

Solid=OBS-FG, Dashed=OBS-AN

0.00 0.02 0.04 0.06 0.08 0.10 0.12 RMS error of (O-B)/O

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30

Impact height (km) Impact height (km)

7985 14349 28964 27397 51549 43452 74728 56549 59608 62685 65466 68884 71664 37512 47034 64050 36634 38644 210 38273 155 38936 159 38088 12300 27010 12320 24545 32491 53316 36761 57987 39207 57598 37123 36955 36944 37768 38454 31290 27575 19592 38854 19191 19704 75 19516 23 19777 26 19294 22 19829 15

Statistics for Bending Angles from METOP / GPS RO EXP=rou OBS minus FG/AN for: Surface=all Flag=used SatId= 4 Time period = 20110101 00UTC - 20110831 21UTC, STEP=3h Statistics for Bending Angles from COSMIC / GPS RO EXP=rou OBS minus FG/AN for: Surface=all Flag=used SatId= 740 Time period = 20110101 00UTC - 20110831 21UTC, STEP=3h Harald Anlauf (DWD) Assimilation of GPSRO at DWD

  • 28. March 2012

8 / 14

slide-10
SLIDE 10

Issues with Assimilation of GPSRO Data (II)

Apparent positive bias for non-GRAS data due to bugs in first-guess check implementation (non-symmetric w.r.t. OBS and FG!) Current GRAS data are (known to be) biased below ∼ 8 km

I Rising occultations (globally) I Setting occultations (notably tropics, lower troposphere) Harald Anlauf (DWD) Assimilation of GPSRO at DWD

  • 28. March 2012

9 / 14

slide-11
SLIDE 11

Issues with Assimilation of GPSRO Data (II)

GRAS, 30S–30N, 3-4 km

Apparent positive bias for non-GRAS data due to bugs in first-guess check implementation (non-symmetric w.r.t. OBS and FG!) Current GRAS data are (known to be) biased below ∼ 8 km

I Rising occultations (globally) I Setting occultations (notably tropics, lower troposphere)

0.012 0.014 0.016 0.018 0.020 Observation [radian] 0.012 0.014 0.016 0.018 0.020 First Guess [radian] * : mean of values per column

3dvar monitoring Bending Angle [radian] Level Max/Min: 3944.40 / 3058.10 Date: 2011070100 - 2011073121 Satellite: 4 Region: TR30

Lat = -30.0 ... 30.0 min = 0.0116 max = 0.0216 mean = 0.0154 median = 0.0151 min = 0.0121 max = 0.0200 mean = 0.0161 median = 0.0159 bias =-0.0008 rms = 0.0019 corr = 0.4908 total number = 276 max number in bin = 9 slope = 1.1471 const =-0.0019

1 3 5 7 10 15 20 25 30 50 75 100

Harald Anlauf (DWD) Assimilation of GPSRO at DWD

  • 28. March 2012

9 / 14

slide-12
SLIDE 12

Issues with Assimilation of GPSRO Data (II)

GRAS, 30S–30N, 6-7 km

Apparent positive bias for non-GRAS data due to bugs in first-guess check implementation (non-symmetric w.r.t. OBS and FG!) Current GRAS data are (known to be) biased below ∼ 8 km

I Rising occultations (globally) I Setting occultations (notably tropics, lower troposphere)

0.010 0.012 0.014 0.016 Observation [radian] 0.010 0.012 0.014 0.016 First Guess [radian] * : mean of values per column

3dvar monitoring Bending Angle [radian] Level Max/Min: 6938.00 / 6090.30 Date: 2011070100 - 2011073121 Satellite: 4 Region: TR30

Lat = -30.0 ... 30.0 min = 0.0088 max = 0.0159 mean = 0.0110 median = 0.0107 min = 0.0093 max = 0.0164 mean = 0.0111 median = 0.0107 bias =-0.0001 rms = 0.0005 corr = 0.9005 total number = 3055 max number in bin = 192 slope = 0.9884 const = 0.0002

1 3 5 7 10 15 20 25 30 50 75 100

Harald Anlauf (DWD) Assimilation of GPSRO at DWD

  • 28. March 2012

9 / 14

slide-13
SLIDE 13

Issues with Assimilation of GPSRO Data (II)

GRAS, 30S–30N, 6-7 km

Apparent positive bias for non-GRAS data due to bugs in first-guess check implementation (non-symmetric w.r.t. OBS and FG!) Current GRAS data are (known to be) biased below ∼ 8 km

I Rising occultations (globally) I Setting occultations (notably tropics, lower troposphere)

0.010 0.012 0.014 0.016 Observation [radian] 0.010 0.012 0.014 0.016 First Guess [radian] * : mean of values per column

3dvar monitoring Bending Angle [radian] Level Max/Min: 6938.00 / 6090.30 Date: 2011070100 - 2011073121 Satellite: 4 Region: TR30

Lat = -30.0 ... 30.0 min = 0.0088 max = 0.0159 mean = 0.0110 median = 0.0107 min = 0.0093 max = 0.0164 mean = 0.0111 median = 0.0107 bias =-0.0001 rms = 0.0005 corr = 0.9005 total number = 3055 max number in bin = 192 slope = 0.9884 const = 0.0002

1 3 5 7 10 15 20 25 30 50 75 100

Bias might be tolerable for small bending angles (. 15 mrad)

Harald Anlauf (DWD) Assimilation of GPSRO at DWD

  • 28. March 2012

9 / 14

slide-14
SLIDE 14

Issues with Assimilation of GPSRO Data (III)

Occasionally poor convergence of the 3D-Var

I Forward operator was evaluated outside domain of validity, e.g. F d(r · n(r))/dr < 0 for some r, or F rays were extrapolated below model orography (mostly Antarctic) in

line-search during minimization

I Extend forward operator and first-guess checks (not yet operational) F minimum geometric height of rays above orography (1 km) F require d(r · n(r))/dr > 0.5 I Enhance optimization algorithm to enable detection and removal of

bad rays during minimization

I Variational Quality Control (VQC) scheme initially used for surface

pressure observations (Gaussian+Flat) while using an approximate (modified) Huber-function for the other observations didn’t work well

Harald Anlauf (DWD) Assimilation of GPSRO at DWD

  • 28. March 2012

10 / 14

slide-15
SLIDE 15

Variational Quality Control

In variational assimilation schemes, VQC enables dealing with bad

  • bservations during minimization

I Observational cost function for Gaussian error distribution (p)

Jo(y − H(x)) = − log(p(y − H(x))) = 1 2(y − H(x))TR−1(y − H(x)) ⇒ Pull of outliers same as for good observations

I Gaussian+Flat: large outliers have zero impact, but strong

non-linearities, possible multiple minima, slow convergence

I ‘Huber norm’ (ECMWF): quadratic/linear for small/large departures;

  • utliers have small impact, but better convergence, no multiple minima

I Approximate (modified) Huber-function

Jqc ∼ α · s x2 β + 1 − 1 ! , with parameters α, β = β(R, . . .) ⇒ Similar to Huber norm, but smooth gradient, 2nd derivative We now use the approximate (modified) Huber-function approach for all observations

Harald Anlauf (DWD) Assimilation of GPSRO at DWD

  • 28. March 2012

11 / 14

slide-16
SLIDE 16

Results from revised first-guess checks and QC

Global

Bias for lower troposphere reduced, more rays used (except GRAS) Example: TerraSAR-X (blue: control, red: experiment)

  • 0.04 -0.02 0.00 0.02 0.04

Bias of (O-B)/O 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30

Area=90/-90 0/360

Solid=OBS-FG, Dashed=OBS-AN

0.00 0.02 0.04 0.06 0.08 0.10 0.12 RMS error of (O-B)/O 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 Impact height (km) Impact height (km)

5856 5705 8893 6184 10398 7597 11061 6999 7168 7425 7606 7741 7824 3914 3882 7817 3792 3950 3887 3833 3665 14 3850 13 2561 329 52 4

  • 12
  • 6
  • 52
  • 92
  • 26

35 4

  • 15
  • 6
  • 2
  • 9
  • 24
  • 4
  • 3

4

  • 16
  • 11
  • 2
  • 1

Statistics for Bending Angles from TerraSar / GPS RO EXP=8565 OBS minus FG/AN for: Surface=all Flag=used SatId= 42 Time period = 20110701 00UTC - 20110731 21UTC, STEP=3h Statistics for Bending Angles from TerraSar / GPS RO EXP=8670 OBS minus FG/AN for: Surface=all Flag=used SatId= 42 Time period = 20110701 00UTC - 20110731 21UTC, STEP=3h

(blue: ctrl, red: experiment)

Harald Anlauf (DWD) Assimilation of GPSRO at DWD

  • 28. March 2012

12 / 14

slide-17
SLIDE 17

Results from revised first-guess checks and QC

Global

Bias for lower troposphere reduced, more rays used (except GRAS) Example: GRAS (blue: control, red: experiment)

  • 0.04 -0.02 0.00 0.02 0.04

Bias of (O-B)/O 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30

Area=90/-90 0/360

Solid=OBS-FG, Dashed=OBS-AN

0.00 0.02 0.04 0.06 0.08 0.10 0.12 RMS error of (O-B)/O 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 Impact height (km) Impact height (km)

1401 8029 10913 21192 18995 33243 25431 27397 28987 30093 31366 32674 16733 21224 28937 16632 17498 3 17241 9 16948 12 16200 5483 12018 5264

  • 3502
  • 4956
  • 3796
  • 131
  • 723
  • 80
  • 208
  • 160
  • 91
  • 1
  • 61
  • 64
  • 51
  • 26
  • 44
  • 46
  • 36
  • 4
  • 13
  • 70

1

  • 55
  • 31
  • 26
  • 4

Statistics for Bending Angles from METOP / GPS RO EXP=8565 OBS minus FG/AN for: Surface=all Flag=used SatId= 4 Time period = 20110701 00UTC - 20110731 21UTC, STEP=3h Statistics for Bending Angles from METOP / GPS RO EXP=8670 OBS minus FG/AN for: Surface=all Flag=used SatId= 4 Time period = 20110701 00UTC - 20110731 21UTC, STEP=3h

(blue: ctrl, red: experiment)

Harald Anlauf (DWD) Assimilation of GPSRO at DWD

  • 28. March 2012

12 / 14

slide-18
SLIDE 18

Comparison to Radiosondes

Mean departures of T, rH obs. to 3-h forecasts, Southern Hemisphere

  • 2
  • 1

1 2 Bias of Temperature [K] 925 850 700 500 400 300 250 200 150 100 70 50 30 20 10

Area=-20/-60 0/360

0.0 0.5 1.0 1.5 2.0 2.5 3.0 RMS error of Temperature [K] 925 850 700 500 400 300 250 200 150 100 70 50 30 20 10 Pressure (hPa) Pressure (hPa)

1567 2349 2721 2773 2545 2513 2498 2523 2466 2487 2211 2241 2149 1567 957 217 1 7

  • 7

12

  • 2
  • 8

2

  • 5
  • 1

5 4

  • 10

7

  • 4
  • 7
  • 4
  • 10
  • 5

5 10 Bias of Relative Humidity [%] 925 850 700 500 400 300 250 200

Area=-20/-60 0/360

5 10 15 20 RMS error of Relative Humidity [%] 925 850 700 500 400 300 250 200 Pressure (hPa) Pressure (hPa)

1584 2274 2275 2223 2309 2298 2303

  • 7

2 12 7 15

  • 7

1

(blue: ctrl, red: experiment)

Only small improvements w.r.t. radiosondes, neutral on forecasts /

Harald Anlauf (DWD) Assimilation of GPSRO at DWD

  • 28. March 2012

13 / 14

slide-19
SLIDE 19

Summary and Outlook

GPS Radio-Occultations are a useful component of the global

  • bserving system for Numerical Weather Prediction at DWD

I Improved analyses and forecasts in particular in data-sparse regions I Improved stability of (static) bias correction for satellite radiances I Exhibit deficiencies in the data assimilation (e.g. background error

model)

I Help locating forecast model deficiencies I Strong non-linearity of forward operator poses challenges for quality

control in data assimilation

Future developments

I Optimize and test impact of forward operator with tangent point drift I Implement ROPP 2d forward operator I Evaluate options to re-activate 3d ray tracer

(needs satellite positions and velocities missing in BUFR!)

Harald Anlauf (DWD) Assimilation of GPSRO at DWD

  • 28. March 2012

14 / 14

slide-20
SLIDE 20

Summary and Outlook

GPS Radio-Occultations are a useful component of the global

  • bserving system for Numerical Weather Prediction at DWD

I Improved analyses and forecasts in particular in data-sparse regions I Improved stability of (static) bias correction for satellite radiances I Exhibit deficiencies in the data assimilation (e.g. background error

model)

I Help locating forecast model deficiencies I Strong non-linearity of forward operator poses challenges for quality

control in data assimilation

Future developments

I Optimize and test impact of forward operator with tangent point drift I Implement ROPP 2d forward operator I Evaluate options to re-activate 3d ray tracer

(needs satellite positions and velocities missing in BUFR!)

A big Thank You to all involved in making data available in Near Real-Time!

Harald Anlauf (DWD) Assimilation of GPSRO at DWD

  • 28. March 2012

14 / 14