College of Engineering | Apply to CSU | Disclaimer | Equal Opportunity Statement | Privacy | Search CSU

The Potential of Clear Sky Carbon Dioxide Satellite Retrievals

You must be on the CSU network—either physically or using VPN—to watch this or any of the videos on this site.

December 12, 2014
Rob Nelson
Hosted by Chris O’Dell (advisor), Chris Kummerow, Scott Denning, Michael Lefsky (Ecosystem Science and Sustainability)

Abstract

This work describes an investigation into the potential use of a simplified non-scattering, or "clear sky", retrieval designed to measure the column-averaged dry-air mole fraction of carbon dioxide (XCO2). The initial motivation for this study was primarily a comparison of retrieved aerosol optical depths from the Greenhouse Gases Observing Satellite (GOSAT) measurements to aerosol optical depths from the AErosol RObotic NETwork (AERONET), which revealed a lack of significant correlation and suggested that the XCO2 retrievals were unable to gain much information about the aerosol content of a given scene. It was thought that clear sky retrievals, which neglect scattering and absorption by clouds and aerosols, could potentially avoid errors and biases brought about by attempting to measure properties of clouds and aerosols when there are none present. Clear sky retrievals have the benefit of being orders of magnitude faster and potentially as accurate as retrievals that attempt to gain information about clouds and aerosols. Real data from GOSAT and simulated data from the Orbiting Carbon Observatory-2 (OCO-2) were analyzed to find conditions under which a clear sky retrieval might perform as well as a "full physics" retrieval, which includes scattering and absorption by clouds and aerosols. We tested retrievals over a wide geographic range that included both land and ocean data. The data were filtered both manually and with the assistance of a genetic algorithm, and bias corrections were applied when necessary.

Regarding the retrieved XCO2, we found that for real GOSAT data the clear sky retrieval performed relatively well over land but not as well over ocean. The opposite conclusion was found for simulated OCO-2 data: it performed well over ocean but poorly over land. For both real GOSAT data and simulated OCO-2 data, high levels of filtering were needed for the clear sky retrieval to be able to perform nearly as well as or better than the full physics retrieval for both land and ocean surfaces. The use of a genetic algorithm to select filters revealed that certain parameters have the potential to be used in pre-filtering data.
Spectral residuals were then examined to determine if the clear sky algorithm's performance was tied to errors in the spectral fitting. It was found that the clear sky retrievals had larger residuals than the full physics retrievals but that reducing the clear sky residuals by allowing them to fit for a customized residual pattern did little to reduce the XCO2 errors. It was also shown that even very clear scenes can result in small but detectable clear sky residual patterns.

Finally, early work in this study prompted investigation into how sensitive the XCO2 algorithm is to the first guess of aerosol properties. Keeping the prior constant, and thus fixing the cost function, χ2 space was explored by varying the first guess of various aerosol parameters. It was revealed that the retrieved aerosol information and XCO2 values can be highly sensitive to the first guess, indicating significant non-linearity in the retrieval's forward model.
Two main conclusions were derived from this work. The first is that an analysis of real GOSAT clear sky XCO2 retrievals and simulated OCO-2 clear sky XCO2 retrievals revealed that the clear sky algorithm is generally inferior to the full physics algorithm, except for when high levels of filtering are applied to remove low quality scenes. The second conclusion is that the current aerosol parameterization leads to unacceptable levels of non-linearity in the XCO2 retrieval. These conclusions motivate further study to improve the retrieval algorithm's aerosol parameterization, either directly or by including additional information, which may result in an improvement of the retrieval algorithm's ability to accurately measure XCO2.