Bayesian optimization of modeled CO2 fluxes in Oregon using a dense tower network, aircraft campaigns, and the community land model 4.5

Modeled estimates of net ecosystem exchange (NEE) calculated with CLM4.5 at 4 km horizontal resolution were optimized using a classical Bayesian inversion approach with atmospheric mixing ratio observations from a dense tower network in Oregon. We optimized NEE in monthly batches for the years 2012 through 2014, and determined the associated reduction in flux uncertainties broken up by sub-domains. The WRF-STILT transport model was deployed to link modelled fluxes of CO2 to the concentrations from 5 high precision and accuracy CO2 observation towers equipped with CRDS analyzers. To find the best compromise between aggregation errors and the degrees of freedom in the system, we developed an approach for the spatial structuring of our domain that was informed by an unsupervised clustering approach based on flux values of the prior state vector and information about the land surface, soil, and vegetation distribution that was used in the model. To assess the uncertainty of the transport modeling component within our inverse optimization framework we used the data of 7 airborne measurement campaigns over the Oregon domain during the study period providing detailed information about the errors in the model boundary-layer height and wind field of the transport model. The optimized model was then used to estimate future CO2 budgets for Oregon, including potential effects of LULC changes from conventional agriculture towards energy crops.