Evaluating the role of prior information in atmospheric inverse modeling frameworks through comparison with geostatistical inverse modeling techniques

Atmospheric inverse modeling techniques make it possible to constrain terrestrial CO2 fluxes on various scales based on information extracted from atmospheric CO2 observation networks. Improvements in modeling frameworks in recent years have significantly increased the spatial and temporal resolution of simulated flux fields, and improved the ability to assess the uncertainties in model output. These trends have highlighted the potential to use atmospheric inverse modeling to analyze mechanisms driving biosphere-climate feedbacks at the disturbance scale, and to monitor compliance with regional greenhouse gas reduction policies. Because atmospheric inverse modeling generally constrains a large number of fluxes with comparatively few atmospheric observations, the problem is usually severely underconstrained. Accordingly, modeling approaches are designed to assign meaningful flux estimates under these conditions. The most common approach to better constrain atmospheric inverse modeling is to start with initial estimates of flux fields and their associated uncertainties. This technique makes it possible to assimilate large additional databases into the framework, such as fluxes from eddy-covariance networks or remote sensing datasets. At the same time, however, each piece of prior information limits the ability of the inverse model to characterize the carbon cycle from the perspective of the atmospheric observations themselves. The use of geostatistical inverse modeling (GIM) holds the potential to overcome these limitations, replacing rigid prior patterns with information on how flux fields are correlated across time and space, as well as ancillary environmental data related to the carbon fluxes. A challenge is that this technique relies on the ability to derive representative covariance matrices. We present results from a regional atmospheric inversion study that focuses on generating terrestrial CO2 fluxes at high spatial and temporal resolution in the Pacific Northwest U.S. We compare two simulation setups, one that heavily relies on prior information on flux fields and their uncertainties, the second avoiding the use of prior flux estimates as far as possible through the use of GIM. For the former, we optimize the prior information through assimilation of observations from multiple eddy-covariance sites. For the latter, we test different options of integrating ancillary information from remote sensing sources and reanalysis meteorology. Results are compared against independent flux estimates for the region from inventory studies and existing modeling results. Strategies are suggested to synthesize the strengths of both approaches to maximize assimilation of existing databases without imposing too many constraints on the inverse model.