A Data-Driven Multi-Scale Statistical Investigation of Regional Sources and Sinks to Improve Knowledge of Terrestrial Carbon Cycling.

The current understanding of carbon cycle processes associated with large resolutions (e.g. 1km to 1000km) limits our abilities to forecast climate change and effectively manage the carbon cycle to protect natural sinks. One of the major indications of our lack of knowledge of carbon cycling is the wide discrepancy between carbon budgets from different models at regional scales. The overarching goal of this research is to apply geostatistical methods, which have relatively fewer assumptions known to influence estimates than other widely used approaches, to infer carbon cycling dynamics and surface fluxes that are independent of process-based models. The methods include a geostatistical inversion (GIM) and a geostatistical regression (GR) applied at several spatial and temporal resolutions. GIM uses the variability in atmospheric CO2 concentrations in conjunction with an atmospheric transport model to infer the most likely distribution of surface CO2 fluxes. GR employs estimates of net ecosystem exchange to infer relationships between carbon flux and environmental variables. In addition, GR applies a new adaptation of the Bayes Information Criterion to identify the optimal set of environmental variables that are able to explain the observed variability in carbon flux. The work shows that the existing atmospheric measurements can be used in GIM to obtain largely independent continental monthly and annual CO2 budgets along with associated uncertainties with limited assumptions. The research also shows that GIM holds promise of providing independent estimates of CO2 flux and associated uncertainties at regional scales. Finally, GR was shown to highlight problems with key assumptions employed to in process-based models (such as the linear temporal scaling of a relationship between a variable and flux) that may contribute to the spread of flux estimates from existing models. In general, the work demonstrates that geostatistical methods can provide a means of isolating the information content of the atmospheric measurements while highlighting potential problems associated with model assumptions that otherwise could go unnoticed. As such, geostatistical methods are important tools that can be used in the overall process of reducing the spread of regional and continental flux estimates from different methods.Ph.D.Environmental EngineeringUniversity of Michigan, Horace H. Rackham School of Graduate Studieshttp://deepblue.lib.umich.edu/bitstream/2027.42/84576/1/kimlm_1.pd