2019-03-28

Agenda Item 1:

Invited Talk: Surface Radiative Transfer Modeling in Support of Satellite Data Assimilation (CSEM)

Discussion:

Talk:

CSEM will supersede the existing CRTM surface model. CSEM is standalone model. CSEM has added value in model software structure and physics. Own interface and own models; Inclusion of model components and improvement of surface models.

CSEM has already been integrated in CRTM and has been tested on Thea; Working towards operational release of CSEM;

New model structure eases addition of new components -> easier to test new model impact <- just modify text register model and turn on new components. Several options in different surface model classes implemented.

Continuous model additions into CSEM after first operational release.

Focus of talk: MW improvements for land & ocean.

Current CRTM surface model: FASTEM6, but no physics-based model. No problem for early clear-sky CRTM cases.

Now: new scheme for multiple stream scattering necessary;

Original two-scale model by Yueh, 1997 is full Stokes model; It captures Stokes components by Geometrical Optics and the small perturbation method. Ocean surface wave spectrum is decomposed by cutoff wavenumber (not intrinsic variable); computation is involved; -> Just one parameter in original model but much more in physical models. Demonstration of effect by animation (doesn’t work).

Two-scale model can still address diel variation (not so much scattering from waves);

Incoherent Bistatic Scattering in Phase Matrix

Ben: Any azimuthal dependence in model? -Yes.

Promising development of physics-based approach towards diel requirement from e.g. MODIS.

Two-Scale model is regression-based. -> Deficiency from regression and two-scale model approach.

FASTEM is to be applied for viewing angles < 60 deg.

In new development a new regression technique is to be applied based on machine learning.

Emily: Is 1997 algorithm based on regression too? -No

E.: FASTEM has 20K difference at large angles and no dependence on wind speed? – Yes, no wind speed dependence for regression. But physically there should be one.

Yianqu: Incidence angles larger than 55 not included in regression? – Yes, highly nonlinear regression necessary. Polynomial fitting not sufficient to fit entire space.

Y.: Improve regression scheme? – Yes, via machine learning to catch all original features. Current FASTEM not applicable for large angles.

Deficiencies also in the original two-scale model. 10-20K difference in horizontal polarization compared to ground measurements.

è Improve two-scale model and then improve regression technique

Ben: Better physics models? – Yes, but two-scale is sufficient. It depends on how much higher-order scattering is included. Cox-Munk model for GO. But difference probably not large here.

Emily: Does the FASTEM6 result have the same conclusion? – Yes.

E.: Does horizontal polarization fit less? -Yes

E.: Is there a physical reason? -Yes, harder to model.

Also difference for azimuth angle results. FASTEM6 doesn’t have all the Stokes components and FASTEM5 is out of phase compared to original radiative transfer model.

Machine learning NFASTEM fits much better to original RT result and no phase difference.

Ben: You trained NFASTEM against what? – Just emissivity results. BRDF even more challenging.

B.: You can use Legendre coefficients for training. – Physically consistent emissivity necessary. Not sure what the best way is.

Many other deficiencies in FASTEM6.

In Europe extension to 700GHz for cubesats in development. (RTTOV-TESSEM). Deficiency at high frequencies. Good agreement of TESSEM and NFASTEM.

Medium permittivity model very important. Constant sea surface salinity is assumed constant in old model. Sea salinity atlas for new model. In the tropics salinity is probably too low and in the arctic too high. Salinity variation as a function of lat and lon in new model; Atlas will be implemented in next release.

Emily: 300 is the constant value? – No 35 ppm.

This is slide from Emily -> Sea surface adjoint becomes negative at high latitudes. Paradox has good physical explanation. Physically reasonable, because emissivity is a strong function of media temperature.

Normally emissivity will decrease as temperature increases. If SSE is used as control variable values are always positive. But negative adjoint for combined wind and temperature (salinity).

For land: emissivity as cv -> very large surface temperature increments.  In this case two implications: retrieval of land temp. is not good + it should not only be based on one term.

Ben: Different dependencies at different frequencies though.

(Ming explains Yianqu the slide with the adjoint problem)

Normally in retrieval sensitive parameters are chosen.

In the current CSEM TL-AD for land are implement. Try to implement analytic TL-AD -> emissivity is most sensitive to soil moisture and vegetation not temp.

Look at two window channels -> compare to adjust input (soil moisture etc.). This technique indicates that even with input parameter uncertainties TL-AD can retrieve values.

Impact of improvements by adjusting input: a lot more assimilation information over Sahara and Amazon region.

Ben: These are observed Tbs? Ok.

Improvement of solid dielectric model. CRTM Default model has large difference towards other models.

Solution: new dielectric model from average of all models.

Roughness correction and polarization in the MW land model in 2015 targets the L-band; Over ocean few adjustable parameters, too many on land;

For next AOP: clear CRTM milestone and coordinate CSEM development. Cooperation with Emily to test all other components. Current RTTOV model already implement in the CSEM CRTM, but not for official release due to copyright issues.

Visible water model difficult due to sunglint. For VIS and IR land atlas from Wisconsin will be implement. Current MW land uses type based parameter.

Ben: What about vegetation? – Leaf area index atlas can also be implemented. Two correlated parameters.

(15:56)

Questions:

Y.: Input table idea very good for users. Cloud coefficient table also for scattering properties? – Yes, great suggestion.

Tong: Difference between RTTOV and NESDIS FASTEM? - Almost the same but different software structure. Advantage in CRTM because not everything is recalculated.

Ben: Steve Albers comments on the phone about VIS? – Not on the phone.

Ming: Ben asked for CSEM use cases. New models can be used easily with CSEM. No need to rewrite anything, just combine different model.

Ben: Concern about physical consistency, but not so relevant for regression.

Ming: I don’t like any current regression model in the CRTM.

Emily: The table says (not understood)

Ben: Any questions from the phone?

Tom Greenwald: I’m writing full Stokes model for atmosphere. Good to combine with new FASTEM model. Is it available?

-       There are several. Very old version is available, but many things are changed. New phase matrix comes from two scale model, difficult to extract.

Result:

First version of CSEM is available for use in the CRTM, but no operational release yet.

Tasks:

clear CRTM milestone and coordinate CSEM development. Cooperation with Emily to test all other components.

Responsible People:

Ming Chen

Deadline:

End of next AOP.

Agenda Item 2:

REL-2.3.1 update

Discussion:

Talk:

Alpha and Beta is already available. ATMS surface emissivity should be improved (discussed in CRTM biweekly meeting).

Questions:

Ming: It’s ok to make conditions at lower level of model. Users should pass down useful values. Condition control should happen as early as possible. Should be part of QC.

Just regression tests will be prepared until code changes from Ming Chen from technical meeting are received.

Ben: Also cloud scattering coefficients from Fuqing’s group received. Patrick is testing it.

Tong: Fuqing is data assimilation guy, not light scattering.

Ben: A postdoc has done this. Patrick will be looking at the pure Mie coefficients.

Y: Fuqing’s group only changed the PSD. – Not clear.

Result:

Issues have been identified in biweekly CRTM technical meeting that need to be solved, but otherwise no major problems.

Tasks:

Just regression tests will be prepared until code changes from Ming Chen from technical meeting are received.

Responsible People:

Tong Zhu (NOAA STAR)

Deadline:

No fixed deadline.