Evaluating Sensitivity of Simulated Atmospheric CO2 to Varied Transport Model Versus Prior Surface Flux

Atmospheric carbon dioxide (CO2) significantly influences the radiation budget of Earth and therefore contributes to the greenhouse gas climate forcing (Ciais et al., 2013). Because the North American terrestrial biosphere acts as a large sink for atmospheric CO2 (Tans et al., 1990), it is important to determine its spatial distribution and quantify its magnitude (Peters et al., 2007; Schuh et al., 2010), as well as evaluate how contributing carbon sources and sinks may change with time. The carbon cycle includes fluxes from biogenic, fossil fuel, oceanic and fire sources (Ciais et al., 2013). We run the Weather Research and Forecasting (WRF) mesoscale model at 30 km horizontal spatial resolution, using prior carbon fluxes from the CarbonTracker inversion system (version CT2010), and incorporating extensive recent additions in observational data from the tower network to optimize biological carbon fluxes over North America during 2008. We assume fossil fuel emissions, oceanic fluxes, and emissions from biomass burning are well-quantified and known. Because atmospheric transport and prior fluxes play a significant role in the development of such an inverse carbon flux estimate, it is important to evaluate the sensitivity of the inversion to transport and flux variability. Varying prior fluxes, boundary conditions and atmospheric transport enable assessment of posterior flux uncertainty. We develope a mini-ensemble of forward-run simulations of atmospheric CO2 mixing ratios over North America during 2008 using three different transport schemes and two prior fluxes (Figures 1 and 2). Atmospheric transport sensitivity is evaluated by implementing different planetary boundary layer (PBL), moist convection, and land surface parameterizations. The first model configuration uses the Rapid Update Cycle (RUC) land surface scheme (Smirnova et al., 2000), the Yonsei University (YSU) scheme PBL scheme (Hong et al., 2006), and the Betts cumulus parameterization for moist convection. This configuration is prone to a deep PBL bias relative to meteorological observations. The second model configuration employs the Noah land surface scheme, the Mellor-Yamada- Janjić (MYJ) (Janjić et al., 2001) PBL scheme, and the Grell 3D (Grell and Devenyi, 2002) cumulus parameterization. This model configuration is prone to a cold bias and a shallow PBL relative to observations. Model-model differences are computed by subtracting the second configuration results from the first configuration results. The original surface flux prior is the CT2010 product and is modified by increasing it uniformly over the domain by 20%. Flux-flux differences are calculated by subtracting the mixing ratio results of the original flux prior from mixing ratio results developed using the enhanced flux prior. [Figure 1] Figure 1. January through March seasonal mean difference of noon to 4 pm LST PBL atmospheric CO2 mixing ratios from (a) all sources/sinks for two transport models; (b) all sources and sinks for two flux priors; (c) biological sources and sinks only for two transport models; (d) biological sources and sinks only for two flux priors. Figures 1 and 2 show that differences in PBL CO2 mixing ratios vary significantly by season and region. During January – March, transport differences are enhanced throughout the Midwest and the east, likely related to consistent passage of synoptic systems during this time of year. The CO2 that is swept by these systems is largely from fossil fuel sources biological sources and sinks alone reduces the transport model difference. Because mixing ratios simulated in a shallow PBL are subtracted from those developed in a deep PBL, transport differences are largely negative. Flux differences are of a greater magnitude than transport differences during this time of year. Large negative flux-flux differences in CO2 mixing ratios occur in the southern U.S. where vegetation growth continues because the enhanced flux prior option amplifies the uptake of CO2, yielding a negative difference throughout the southeast. [Figure 2] Figure 2. July through September seasonal mean difference of noon to 4 pm LST PBL atmospheric CO2 mixing ratios from (a) all sources/sinks for two transport models; (b) all sources/sinks for two flux priors; (c) biological sources and sinks only for two transport models; (d) biological sources and sinks only for two flux priors. Figure 2 shows the same overview of model-model and flux-flux differences as Figure 1, but for July through September seasonal means. During the winter months, a shallow PBL will result in greater CO2 mixing ratios due to fossil fuel emissions and vegetation respiration. Thus a deep PBL simulation minus a shallow PBL simulation will result in a negative difference as shown in Figure 1. The opposite occurs during the late summer when vegetation growth dominates regional signals. A shallow PBL during this time enhances the CO2 drawdown signal, yielding more negative CO2 mixing ratios. Thus, a deeper PBL simulation yields higher concentrations and subtracting the shallow PBL simulation yields a positive model-model difference. For a surface flux prior that is enhanced uniformly by 20%, this results in more negative mixing ratios during the growing season. Simulations conducted using the unmodified CT2010 flux prior yield a less negative mixing ratio signal over the Midwest corn belt during July through September. NASA's OCO-2 is making global measurements of CO2, providing atmospheric CO2 dry air column mole fractions. These data will require evaluation (Wunch et al., 2011; O'Dell et al., 2012; Crisp et al., 2012) and so we further examine the sensitivity of the simulated atmospheric column CO2 to varied transport model and surface flux, comparing the magnitude, timing, and distribution of column sensitivity to that of the PBL.