2.   Detailed Description
This study uses all analyzed air samples taken at the surface from the NOAA ESRL Cooperative Air Sampling Network available for each year studied, except those flagged for analysis or sampling problems, or those thought to be influenced by local sources. The composition of the network thus varies per week depending on successful sampling and analysis, and each site's sampling frequency. In addition, we use in situ quasi-continuous CO2 time series from seven towers: (1) the 396m level of the WLEF tower in Wisconsin; (2) the 107m level of the AMT tower in Argyle, Maine; (3) the 251m level of the KWKT tower in Texas; (4) the 40m level of the tower in Fraserdale, Canada operated by Environment Canada (EC); (5) the 23m level of the tower at Candle Lake (formerly Old Black Spruce), Canada operated by EC; (6) the 9m level of the tower at Storm Peak Laboratory operated by NCAR; and (7) the 5m level of the tower at Niwot Ridge operated by NCAR. Other in situ quasi-continuous CO2 time series used are from the NOAA ESRL observatories at Barrow, Mauna Loa, Samoa, and South Pole, and the EC programs at Alert, Canada and Sable Island, Canada. Note that all of these observations are calibrated against the same world CO2 standard (WMO-2005). Also, note that aircraft observations from the NOAA ESRL program were NOT assimilated, but used for independent assessment of the CarbonTracker results.

At the tower-based quasi-continuous sampling sites, we construct one daytime average (12:00-16:00 Local Standard Time) mole fraction for each day from the time series, recognizing that our atmospheric transport model does not always capture the continental nighttime stability regime while daytime well-mixed conditions are better matched. At mountain-top sites (MLO, NWR, and SPL), we use midnight-4:00 LST averages as this tends to be the most stable time period and avoids periods of upslope flows that contain local vegetative and/or anthropogenic influence. Moreover, observations at sub-daily time scales are likely to be strongly correlated and therefore add relatively little independent information to our results. Also based on Transcom continuous simulations, we decided to move a set of coastal sites by one degree into the ocean to force the model sample to be more representative of the actual site conditions. These sites are labeled for reference in the complete table of sites used in CarbonTracker.

We apply a further selection criterion during the assimilation to exclude non-Marine Boundary Layer (MBL) observations that are very poorly forecasted in our framework. We use the so-called model-data mismatch in this process, which is the random error ascribed to each observation to account for measurement errors as well as modeling errors of that observation. We interpret an observed-minus-forecasted (OmF) mole fraction that exceeds 3 times the prescribed model-data mismatch as an indicator that our modeling framework fails. This can happen for instance when an air sample is representative of local exchange not captured well by our 1x1 degree fluxes, when local meteorological conditions are not captured by our offline transport fields, but also when large-scale CO2 exchange is suddenly changed (e.g. fires, pests, droughts) to an extent that can not be accommodated by our flux modules. This last situation would imply an important change in the carbon cycle and has to be recognized by the researchers when analyzing the results. In accordance with the 3-sigma rejection criterion, ~2% of the observations are discarded through this mechanism in our assimilations. Table 1 gives a summary of the observing sites used in CarbonTracker and the assimilation performance.