Minichamber Results: last updated 6pm 17 Nov. 1998

M12 results

Some general observations are:

• The signal-to-noise is actually rather low, but I think consistent with expectations. If we ignore noise introduced by subtracting off the collimator cell cold-chopper radiance, our computation of expected noise is within a factor of three of the "observed" noise. Once we add the cold-chopper noise into our model, our computed noise should be even closer to what we observe. We need to model the minichamber and full-up flight model gas cell tests to determine what signal-to-noise we need for those tests.
• Given this noise, we will not have great sensitivity for determination of the SRF half-width. Note that the CO2 lines (except for the Q-branch) are only 10% absorbing. The noise level is about 0.005 RMS so this is only a signal-to-noise of 20/1.
• So far, we have done two fits. For each of these fits we "pre-fit" the wavenumber scale, and it appears that the observed dispersion is quite different from what was "expected". However, I don't know how LIMIRS generated the wavenumber scale of the spectrum we were provided, and maybe the changes to the AIRS optical system with the new grating will have some small effect? See our plot of the observed wavenumber scale versus what was provided with the M12 data. We fit the observed data to a wavenumber scale that included an offset, linear, and quadratic terms. Further work may show that the quadratic term is not justified.

Fit #1

We let the SRF width and the optical depth (CO2 absorption coefficient * pathlength) float. This procedure ignores errors in the three different "296K" temperatures (gas, collimator optics, 296K blockbody source) and errors in the 310K blackbody temperature (relative errors in the blackbody temperatures between the near and far measurements). A graph of those results shows significant errors in the Q-branch, and an SRF 25% below the nominal value. Also plotted on this graph are results if you keep the SRF widths at their nominal value, and increase the effective pathlength (optical depth) from 20 inches to 24 inches. Away from the Q-branch, this calculation keeps the relative variation in the obs-calcs quite low, but they have a constant bias. This indicates that the SRF width differences might just be due to errors/variations in the various thermal terms that contribute to the calculation of the observed transmittances. At first the Q-branch errors looked like errors in line-mixing, but we believe that is not the case. (Note that for this 1 atm pressure spectrum, line mixing is very important, but we think we can model it much more accurately than the observed errors.) I don't think some combination of (1) thermal errors, (2) optical depth errors, and (3) small changes in the SRF width will fix the Q-branch problem, but more work is need to verify that. It is true that the stronger Q-branch absorption is more sensitive to thermal errors than the weaker P/R branch lines. We checked to see the effect of a "pedestal" upon a nominal spectra; see effect of pedestal plot. The pedestal (actually a hole) used for this test comes from the EM SRF data. If there is a pedestal in the FM SRFs, it must be well below the 1% level.

Fit #2

In this fit we let the optical depth vary, and introduced a term in the fit equivalent to floating the 310K black body temperature. The SRF widths were held fixed at their nominal values. A graph of this fit shows that the Q-branch problem remains. You can also see larger obs-calcs for the P/R branch lines than when we let the SRF float. This indicates that we are going to have to model how errors in the 310K temperature and the three 296K temperatures affect the fits.

M3 results