Remote Sensing of CO2: Geostatistical Tools for Assessing Spatial Variability, Quantifying Representation Errors, and Gap-Filling.

Abstract

Currently, approximately half of the anthropogenic emissions of CO2 are absorbed by oceans and the terrestrial biosphere, thus greatly reducing the rate of atmospheric CO2 increase and related climate change. The current understanding of the global carbon cycle, and of the sustainability of natural carbon sinks, is limited, however. To enhance this knowledge, scientists use process-based biospheric models and atmospheric transport models, together with the limited global ground-based CO2 measurement network to infer global CO2 fluxes. Current estimates of carbon budgets at regional to continental scales vary significantly, however, in large part due to limited atmospheric observations of CO2.

Satellite-based observations provide the possibility of global coverage of column-averaged CO2 (XCO2), which could improve the precision of estimated CO2 fluxes. XCO2 observations will have large data gaps, however, which will limit the use of XCO2 observations for evaluating CO2 flux estimates. In addition, remote sensing soundings will often be representative of fine scales relative to the resolution of typical atmospheric transport models, causing representation errors that should be quantified for accurate CO2 flux estimation.

In this dissertation, the spatial variability of the XCO2 signal is quantified using geostatistical analysis. Geostatistical methods that depend on the knowledge of this spatial variability are then presented for evaluating representation errors. Unlike previous estimates of representation errors, the proposed method accounts for the regionally-variable XCO2 spatial variability, and the spatial distribution of retrievals. Further, a spatial mixed-effects statistical model that best represents the quantified XCO2 variability is presented for gap-filling XCO2 retrievals. The presented geostatistical gap-filling method, which is based on a multi-resolution model of the spatial trend and variability of XCO2, is tested using eight realistic scenarios of expected spatial distributions of XCO2 retrievals. The method yields XCO2 estimates over regions with data gaps, together with an estimate of the associated gap-filling uncertainties.

The presented methods provide flexible tools that can be applied to estimate representation errors and gap-fill XCO2 or other remotely sensed data. As such, they provide the potential for improving and evaluating estimated CO2 fluxes, process-based models, and atmospheric transport models.