assimilating infra-red sounder data over land

© Crown copyright Met Office

Assimilating infra-red sounder data over land

John Eyre for Ed PavelinMet Office, UK

Acknowledgements: Brett Candy

DAOS-WG, 4th meeting, Exeter, 27-28 June 2011


Assimilating infra-red sounder data over land

Outline

• The problem• Met Office implementation: 1D-var + 4D-Var• Retrieval assimilation studies:

• on simulated data• on real data

• Conclusions


Pre

ssur

e (h

Pa)

IASI temperature Jacobians

Surface-sensitivechannels

Large proportion oftropospheric channels!

The problem


Land surface emissivity (8.5m)from NASA AIRS product, June-August 2008


Use of IR radiances over land

• Current operations:

• Assume infrared emissivity = 0.98 over land

• Not good enough – don’t use channels sensitive below ~ 400hPa

• Options to increase data use over land• Use fixed emissivity “atlas”• Use land surface model / surface type atlas• Retrieve surface emissivity from observations


Met Office assimilation of AIRS and IASI radiances

• 1D-Var analysis for each radiance vector:• … to estimate atmospheric profile + surface variables• … to provide:

• QC on radiances passed to 4D-Var• estimate of variables not in 4D-Var control variable:

• skin temperature• [temperature profile above model top]• cloud top height, cloud fraction• ***NEW*** – surface emissivity variables

• Radiances and 1D-Var-retrieved variables passed to 4D-Var


1D-Var fit to observations

Window channel, assuming = 0.98

Over sea:use ISEM emissivity model(quite accurate)

Over land:assume emissivity=0.98Large mis-fit


0.98


Simulation experiment: retrievals from simulated AIRS radiances (1)

Simulated radiances:

• simulated using RTTOV radiative transfer model• 13495 atmospheric profiles from ECMWF ERA40 dataset• surface emissivity from UWisc/CIMSS IR emissivity atlas

(2006 data used)• derived from MODIS observations, fitted to laboratory spectra• independent of training dataset

• simulated observation errors


Simulation experiment: retrievals from simulated AIRS radiances (2)

1D-Var retrievals:

• 12 emissivity PCs in retrieval vector

• first guess emissivity = 0.98

• emissivity retrieved simultaneously with T, q and cloud

• idealised case: • using all AIRS channels• clear sky only


Atlas: 9.32 m (mineral signal)

True Emissivity


Retrieval: 9.32 m (mineral signal)

Retrieved Emissivity


920 hPa temperature bias:without emissivity retrieval


920 hPa temperature bias:with emissivity retrieval


Promising results from simulations…

Test with real data

…but first: what about clouds?


Need good and

Tskin to detect cloud

Need good and


Need knowledge of cloud to

analyse sfc!


analyse sfc!


Need good and


Need good and



analyse sfc!


analyse sfc!

Use a piori emissivity

atlas?

Use a piori emissivity

atlas?

Use a priori

emissivity atlas

Use a priori

emissivity atlas


First-guess emissivity:a global IR emissivity atlas

• CIMSS/Univ. Wisconsin IR emissivity atlas• based on MODIS products• emissivities fitted to high-spectral-resolution

laboratory measurements• monthly variability, high spatial resolution• becoming widely used

• Other atlases available from other sources• Products from AIRS, IASI, …


Global NWP trials

• Bottom line: impact on Met Office “global NWP index”• Winter: +0.0 vs obs, -0.1 vs anal• Summer: +0.1 vs obs, +0.3 vs anal

• Sufficiently +ve to include in next operational change (July 2011)

• However:

• Problems with model Tskin biases during daytime

• At present data included nighttime only, and only IASI• First step - improvements expected


Conclusions

• Currently, low-peaking IR sounding channels are rejected over land - uncertainties in emissivity and Tskin

• Improvements implemented via 1D-Var:• analysis of spectrally-varying surface emissivity• “first-guess” global emissivity atlas

• Poor fit to observations over daytime deserts• possible systematic biases in model surface temperature


Questions ?

assimilating infra-red sounder data over land

Documents