The Influence of North American Carbon Flux Spatial Distribution on the Temporal Variability of Atmospheric Carbon Dioxide

A small subset of biospheric model estimates of net ecosystem exchange (NEE) are used to assess the ability of the North American CO2 sampling network to detect regional spatial variability (i.e., 1° x 1°) in land-atmospheric carbon flux. The atmospheric signal at continuous observation locations operating in 2004 resulting from biospheric model derived NEE is quantified using the WRF-STILT atmospheric transport model. The resulting CO2 concentration time series are then compared to determine if changes in the spatial distribution and degree of grid-scale variability in surface fluxes translate into detectable differences in their corresponding atmospheric CO2 signal. If no significant differences are observed, then inverse modeling approaches may be unable to infer a unique grid-scale (i.e., 1° x 1°) surface flux distribution (given the current sampling network). Tower-specific model-data mismatch error derived from real data and estimated using Restricted Maximum Likelihood (RML) is used to assess the significance of observed differences among transported CO2 signals. In general, distinct atmospheric CO2 signals of transported fluxes resulting from the different biospheric models are seen more frequently during the growing versus the dormant season and at towers located in areas of greater flux magnitude (e.g., forested regions). In addition, the greatest difference in CO2 time series is observed at towers with larger measurement footprints, such as Park Falls, Wisconsin (WLEF). Although the magnitude of carbon flux variability varies significantly among the biospheric models examined here, in many cases, these differences do not necessarily translate into significant differences in atmospheric CO2 signal at measurement locations. Thus, while inversions using atmospheric measurements may be able to detect ecoregion-scale or coarse resolution differences among surface flux distributions, they may be unable to detect unique distributions of grid-scale flux variability. Overall, this work provides information about the uniqueness of NEE estimations from inversions, or rather the information content of the atmospheric data with regards to the estimated fluxes. Such information will help inform inverse modeling, and improve our understanding of land-atmosphere carbon exchange.