Retrieval of O Brien (CSIRO, Australia) Spectra

Cover Slide:

This talk is on the retrieval of oxygen A-band spectra using airborne measurements. The motivation for this work was provided by the OCO Mission, so I’ll start off with a brief overview of the mission.

OCO Mission:

The OCO or Orbiting Carbon Observatory mission was set up to make the first global, space-based measurements of the column integrated dry air mole fraction, which we call

XCO2, with the precision, resolution and coverage needed to characterise sources and sinks of this important greenhouse gas. This accuracy of 1 part per million, or equivalently

0.3%, is unprecedented. These measurements will improve our ability to forecast CO2 induced climate change.

The maximum contribution to the column integrated CO2 abundance comes from near the surface, where most sources and sinks are located. As the figure shows, thermal IR techniques have less sensitivity near the surface because the thermal contrast between the surface and near-surface atmosphere is often small. In contrast, NIR absorption is characterised by averaging kernels that peak near the surface and are thus suitable for our application. So, we measure high resolution spectroscopic observations of near infrared

CO2 and O2 absorption bands in reflected sunlight.

Several factors besides the atmospheric CO2 mole fraction contribute to the CO2 absorption intensity measured from space. These factors include the surface pressure, the atmospheric temperature and humidity profiles, clouds and aerosols. The “weak” CO2 band gives us the column CO2. The O2 A-band provides surface pressure estimates with accuracies of 0.1%, as O’Brien and Mitchell showed in 1992. It is also sensitive to clouds and aerosols. The “strong” CO2 band provides independent constraints on the wavelength dependent optical properties of clouds and aerosols. Measurements in this band also provide information about the atmosphere temperature and humidity along the optical path, minimising systematic errors due to uncertainties in these parameters.

Retrieval of O’Brien (CSIRO, Australia) Spectra:

To test the retrieval scheme, we used high-resolution O2 A-band spectra of sunlight reflected from the sea surface. These measurements were made by Denis O'Brien et al. at CSIRO, Australia, and published in the Journal of Atmospheric and Oceanic Technology in a couple of papers in 1997 and 1998. The objective is to retrieve column O2 with the precision required for OCO.

Our retrieval strategy has a forward model, an instrument model and an inverse model. The forward model is a multistream, multiple scattering radiative transfer model which incorporates a bidirectional reflection function at the surface. The BDRF is based on the Cox and Munk glint model. The instrument line shape was provided by O’Brien and is essentially the convolution of a boxcar entrance slit transfer function with a detector pixel transfer function. The inverse method is based on the optical estimation theory explained by Clive Rodgers in his book. We use the Levenberg-Marquardt method for the iteration.

Retrieval: First Cut

As a first cut at the retrieval, we tried to simulate the spectra using our model. The model atmosphere has 11 layers, 8 of which are in the troposphere. The transmittance here is the normalised upwelling radiance. The green line refers to the computed spectrum and the black line is the observed spectrum. As you can see, the root mean square residual is 8.8%, which is way too bad for our purpose. On blowing the figure up, we noticed that a scaling of the wavelength grid might solve part of the problem. Retrieval: Second Cut

This brings us to the second cut, where O’Brien’s wavelength grid has been scaled to match the calculated spectrum. Here we also take into account absorption by isotopomers of oxygen and trace species like water. Remarkably, just this simple modification gives us a much better residual of 2.3%.

Retrieval: Third Cut

To further improve the residuals, we fit the continuum level, the tilt and the zero offset. We also noticed that O’Brien’s line shape assumed zero response beyond 0.5 cm-1. So, we fit a wide lorentzian to the line shape. The best fit had a width of 10 cm-1 and amplitude 0.02, as compared to O’Brien’s ILS which has a peak of about 1.75. We obtained a residual of 1.4% and were able to retrieve the column O2 to 1%. However, line mixing effects are apparent and also there is a solar feature at 0.772 µm.

Retrieval: Fourth Cut

If we could remove these effects, we would obtain a further reduction in the residuals.

Conclusion:

In conclusion, we were able to retrieve column O2 with a precision of around 1%. This is from just one sounding. Clearly, by averaging sufficient soundings, we should be able to retrieve column O2 to 0.1%. End-to-end retrieval simulation tests have shown that this precision in column O2 is sufficient to retrieve XCO2 to precisions better than 0.3%, which is what the OCO mission requires.