research and development on satellite data assimilation at the canadian meteorological center l....

Research and development on satellite data assimilation at the Canadian

Meteorological Center

L. Garand, S. K. Dutta, S. Heilliette, M. Buehner, and S. MacPhersonWorld Weather Open Science Conference 2014Montreal, QcAugust 16-21, 2014

DRAFT – – April 20, 2023

Outline

• Current data assimilation context

• Progress on the assimilation of surface sensitive infrared radiances over land

• Progress on the use of inter-channel error correlations

• Observing System Simulation Experiments capability

• Conclusion


New data assimilation context

EC moving to ensemble-variational (EnVar) system with:

•Flow-dependent background errors including surface skin temperature correlations with other variables

•Analysis grid at 50 km, increments interpolated to model 25 km grid, 80 levels

•~140 AIRS and IASI channels assimilated: many sensitive to low level T, q, and Ts.

•Assimilation of North America ground-based GPS (impacts humidity analysis)


Ensemble spread of Ts (Jan-Feb 2011) 00 UTC 06 UTC

12 UTC 18 UTC

Maximum ~15h localIn SH (summer)

Maximum at nightTibetan area (winter)

Ts background error over land in 4Dvar isConstant: 3 K.

In EnVar backgroundError is 50% ENKF,50% static (NMC)


Assimilation of surface-sensitive IR radiances over land

Key challenges:

•Reliable cloud mask •Spectral emissivity definition•Highly variable topography•Radiance bias correction•Background and observation error definition


Approach guided by prudence

Assimilate surface sensitive radiances under restrictive conditionsover land and sea ice:

•Estimate of cloud fraction < 0.01•High surface emissivity (> 0.97)•Relatively flat terrain (std of height < 100 m at 50 km scale)•Diff between background Ts and rough retrieval < 4 K

Radiance bias correction approach:

•For channels flagged as being surface sensitive, use only ocean data to update bias coefficients.


Limitation linked to topography

Criterion used: local STD of topography < 50 m (on 3X3 ~50 km areas)

RED: accepted, white std > 100 m, blue 100>std>50 m


Limitation linked to surface emissivity

Surface emissivityAIRS ch 787

Emissivity based onCERES land types(~15 km) + weighted averagefor water/ice/snowfraction.

Accept only emissivity > 0.97 ; Bare soil, open shrub regions excluded.


Assigned observation error to radiances

15.0m 4.13m 15.3m 4.50m

Assigned = f std(O-P) Desroziers = <(O-P)(O-A)>

f typically in range1.6-2.0


First attempt: negative impact in regions 60-90 N/S

00 hr 72hr

Possible cause: cloud contaminationRisk reduction: no assimilation at latitudes > 60 deg.


Second attempt with added constraints

• Exclude latitudes > 60 deg

• Limit gradient of topography to 50 m (previously 100m) Strong sensitivity of this parameter to std(O-B) of surface channels


Temp std difference vs lead time NH-Extra_trop

vs ERA Interim vs own analysis

Consistent positive impact vs ERA Interim and own


T std diff (CNTL-EXP) vs ERA-Interim

Tropics SH Extratropics


Time series of T std diff at 925 hPa

NH extratropics Tropics CNTLEXP

EXP superior to CNTL 55-65% of cases at days 3-5


850 hPa Temp anomaly corr.

NH extratropics TropicsCNTLEXP


120-h N-Extr vs ERA-Interim

CNTLEXP


Validation vs raobs 120-hFEB-Mar 2011 (59 days)

North America EuropeCNTLEXP

U V U V

GZ T GZ T


Added yield: about 15%(for surface sensitive channels)

Number of radiances assimilated for surface channel AIRS 787CNTL: ~1400/6h EXP: ~1600/6h

CNTLEXP

Region: world, EXP excludes surface-sensitive channels at latitudes > 60Radiance thinning is at 150 km


R&D on Inter-channel error correlations

• Approach based on Desroziers et al. 2005. It allows for a simple evaluation of the R matrix using assimilation by-products (O-A) and (O-B) for channels i,j :

R = <(O-A)i(O-P)j>

• Bormann et al. demonstrated that this method gives similar results as other methods

• No adjustable parameter


Known limitations

• The approach makes use of several assumptions (as other methods such as Hollingsworth-Longberg):

– Unbiased observations – Uncorrelated Background and Observation errors– Well specified B matrix


Implementing the approach

tsym RRR~~

2

1~

• Desroziers diagnostic was applied using 21 days of quality controlled and bias corrected radiances assimilated in a reference cycle.

• The resulting Desroziers matrix was symmetrized:

• Estimates done for all radiances: AIRS, IASI, AMSU-A, AMSU-B, and SSM/I.

• Variances on diagonal still inflated (1.6 std(O-P))2

• Added modest 3% to analysis time

• Did not affect convergence of analysis cost function, using standard pre-conditioning


Sample diagnosed correlations

Correlation structureof the symmetrized R for the 142 AIRS channels selected for assimilation

High correlations forwater vapor and surface channels


Forecast scores against ECMWF analyses(std difference CNTL – EXP)

24-h RELATIVE HUMIDITY


72-hr scores against ECMWF

T

U V

GZ


Temp. anomaly correlation against ECMWF analyses (500 hPa, world)

ControlExperiment

General conclusion:

Overall positive impact,Worth implementing


Planned Polar Communication and Weather Mission (PCW)

A possible 2-sat HEO system

Goal:Fill communication and meteorologicalobservation gaps in the Arctic. 15 min Imagery above 55N.

Most likely configuration:2 satellites in highly elliptical orbit

Status:Awaiting approval from governmentCould operate by 2021

Meteo instrument:Similar to ABI or FCI


PCW would fill AMV gaps at high latitudes: what impact to expect?

Example of current AMV coverage per 6-h from 10 sources


Impact on NWP based on OSSE(ECMWF Nature Run used as “truth”)

OSSE showing significant impact of PCW AMVs filling gap in polar areas

Garand et al., JAMC, 2013

Positive impact (blue) at 120-hIn both polar regions from 4 satellites


Conclusion

• R&D assimilation now benefiting from En-Var system.

• Encouraging results on assimilation of surface sensitive IR radiances over land. Next step is to relax limitation on surface emissivity.

• Overall positive impact considering inter-channel error correlation for all radiances. Plan is to implement along with Cris and ATMS in 2015.

• OSSE capability developed at EC, and used to support PCW

research and development on satellite data assimilation at the canadian meteorological center l....

Documents

owndraft page

surface channelsdraft

static nmcdraft page

high surface emissivity

satellite data assimilation

white std

sensitive ir radiances

tropvs era interim vs