This is not the latest CarbonTracker update! Link to latest.
CO2 Time Series (CT2007B)

Time series of the CO2 mole fractions used in CarbonTracker. The title reflects the name and location of the site. The red symbols shows the actual observed values, the green symbols is the model calculated value after surface CO2 exchange has been estimated. The grey shaded bars denotes the tolerance we set on the observations and are determined mostly by our ability to model CO2 each site correctly in CarbonTracker. If the modeled CO2 exceeds the tolerance by more than a factor of 3, the observation is not used in our system and colored blue in this figure. Units are parts per million of CO2 [ppm]. Grey bars are the assigned model-data mismatch values found in this table. Note that time series can be discontinuous and do not necessarily span the full time domain.

Time series of the CO2 residuals (final CarbonTracker simulated values minus the actual observations) from the above figure. Non-zero residuals reflect the fact that CarbonTracker is unable to match all the observations perfectly which is to be expected for a data assimilation system. The first guess fluxes and observations covariances matrices as well as the numerical transport model are not perfect. The study of the residuals is a powerful diagnostic tool of the performance of CarbonTracker. The time correlation in the residual exhibited at several sites, for example, shows the inability of CarbonTracker to extract all the information from the observations. The analysis of the residuals is ongoing and will help improve future CarbonTracker updates.
Histogram of innovations for all observations at this site. &chi2 is the average of the squared innovation sums, N rejected is the number of rejected samples, and N total is the number of observations available. Innovations are calculated as (z-H(x))/√((HPHT+R)) with H the observation operator, P the prior covariance matrix, R the model-data-mismatch, x the prior fluxes, and z the observations. It indicates whether the model performance matches the assigned uncertainty in the system and shows if there are systematic problems matching this sites' CO2 record or if the mismatches are Gaussian.