Improved Estimates of Regional-scale Land-Atmosphere CO2 Exchange Using Geostatistical Atmospheric Inverse Models.

Abstract

In order to devise strategies to reduce atmospheric CO2 concentrations and predict their future trajectories for climate change mitigation and prediction, it is important to accurately quantify and understand the drivers of regional-scale (~500 x 500 km2) land-atmosphere carbon exchange from biospheric processes and fossil fuel emissions. While CO2 fluxes at this scale cannot be directly measured, inverse models can potentially provide estimates with reasonable uncertainties by tracing back variability in atmospheric CO2 measurements to the most likely distribution of surface CO2 exchange. This dissertation applies a geostatistical approach to inversions, which relies on an estimated spatiotemporal covariance structure to infer fluxes directly at fine scales in both space and time. In addition, process-based datasets can be incorporated into the inversion in a manner analogous to multi-linear regression, improving flux estimates and providing inference regarding significant flux drivers. In the first dissertation component, environmental datasets are incorporated into a global inversion, with results showing that Leaf Area Index and the Fraction of Photosynthetically Active Radiation explain a significant portion of biospheric flux variability, while Gross Domestic Product and Population Density are associated with the fossil fuel emission signal. However, at the continental scale, flux estimates were found to be constrained primarily by the atmospheric measurements, with the grid-scale environmental datasets having minimal impact. The second component investigates the optimal use of continuous, continental CO2 measurements influenced by the biospheric diurnal cycle, heterogeneous land-cover, and point-source fossil fuel emissions. In a series of synthetic data inversions over North America during the growing season, explicitly estimating the diurnal variability of fluxes was found to be critical for inferring unbiased fluxes at the aggregated monthly, ecoregion-scale. In the third component, a North American regional inversion is implemented using real data available from the continuous monitoring network in 2004. The biospheric portion of estimated total CO2 flux is compared to a collection of bottom-up, process-based model output. Results show some convergence in the spatial patterns, seasonal cycle and net annual CO2 flux between the inversion and bottom-up models, although inversion results at robust scales also help to provide insight into the forward model spread.