ALT_06C0: Quasi-continuous, surface measurements from Alert, Nunavut, Canada located at 82.4508° N, 62.5072° W and 185.00 masl. Measurements are made by Environment Canada (EC). Assimilation statistics for this and other sites are included in the Documentation.

Time series of CH4 mole fractions, prior, post-assimilation and observed, at this CarbonTracker observation site. In the top panel, measured mole fractions (**red** circles, "OBS") are plotted along with CarbonTracker post-assimilation values (**dark blue** circles, "POST"), and values produced by using prior emission estimates with no data assimilation (**light blue** circles, "PRIOR"). At some sites, there are observations that CarbonTracker cannot simulate successfully; these are shown as **yellow** circles. The bottom panel shows a time series of residuals (the difference between the post-assimilation and measured mole fractions) shown with **dark blue** circles ("POST-OBS"). These residuals should be uncorrelated in time, unbiased (i.e., have a mean of zero), and distributed normally. Residuals computed using prior flux estimates and no data assimilation are also shown (**light blue** circles), along with the imposed model-data mismatch (**red** lines), which defines how well the site is expected to be simulated by the transport model (TM5) used with CarbonTracker (see documentation). Simulated mole fractions more than three times the MDM from zero are rejected by the optimization system. Note that without adjustment of fluxes by assimilation of observations, the simulated CH4 values diverge significantly from observations over the period spanned by CarbonTracker. Units are 10-9 mol mol-1 of CH4 (ppb).

Histograms of residuals at this site for the selected period (residuals are the difference between the post-assimilation and measured mole fractions). The left panel shows residuals for the CH4 mole fractions with no data assimilation; the right panel shows the residuals after data assimilation. The **red** curves show summaries of the residuals interpreted as normal distributions, with the standard deviation (σ (ppb)), the mean (μ (ppb)), and assumed model-data mismatch (σMDM (ppb)) shown in text at the upper part of each figure. The vertical scales are number of observations. Ideally, the distribution should have a mean of zero and a posterior standard deviation that is smaller than the assumed model-data mismatch error. The result is a more sharply peaked and narrow normal distribution after assimilation of observations.