A monitoring network design tool for atmospheric carbon dioxide:Validation over North America

Inverse methods to estimate fluxes of carbon dioxide (CO2) rely on networks of atmospheric CO2 measurements. The current network of continuous observations in North America has grown from 9 continuous CO2 monitoring towers in 2004 to approximately 35 in 2008. Despite its growth, however, the sparse network of monitoring stations has been a limiting factor in the effort to constrain CO2 fluxes at sub-continental scales. Current methods to assess the number and optimal locations for additional monitoring sites either rely heavily on expert opinion, which can be subjective, or on observational system simulation experiments(OSSEs), which are computationally expensive and are sensitive to specific model assumptions. We instead propose a flexible and computationally inexpensive tool to examine current and future atmospheric CO2 monitoring network configurations. This tool has been developed to inform the expansion of continuous CO2 monitoring stations based upon the variability in the atmospheric CO2 signal seen through modeled CO2 fields. The spatial variability is quantified through a geostatistical analysis that yields information about the spatial scales over which the CO2 concentrations are correlated. This information is then used to define a correlation length criterion (CLC) which is used to assess the coverage provided by current networks, more importantly, to inform the optimal placement of towers in future network designs. The CLC approach places towers using knowledge of both the location of pre-existing towers as well as the local variability in the atmospheric CO2 signal. Thus, the density and placement of the new towers are not uniform in space but such that the correlation length criterion is fulfilled over the entire domain. Two sample hypothetical networks are created to cover North America with varying degrees of coverage. The less strict coverage regime of one correlation length (CL1) requires an additional 8 towers relative to the current North American network, while the more stringent regime of one half of a correlation length (CLp5) requires an additional 43 towers. The additional benefit provided by the two hypothetical networks is evaluated relative to the existing 2008 network using a synthetic data inversion where the "true" fluxes are prescribed and used to create measurements at the hypothetical tower locations. Flux estimates from the two expanded hypothetical networks are analyzed at various spatial and temporal scales and assessed by their correspondence with the "true" fluxes. The hypothetical networks show marked improvement over the pre-existing 2008 network, and offer insights into addressing the weaknesses of the current network. Overall, CLC approach offers a quantitative design tool to aid in the expansion of atmospheric CO2 measurement towers for constraining surface fluxes of CO2.