Figure \ref{fig:TimeSeries} shows the daily average time series. The daily averages were only calculated from coincident hourly TCCON and in-situ measurements. The plot shows that the TCCON and CFSv2 water vapour columns are in good agreement, as are surface in-situ and the bottom level of the CFSv2 data. The correlation shown in figure \ref{fig:H2oCorrelation}, shows that the TCCON columns are dry relative to the CFSv2 data and had a lower span. The in-situ data is wet compared to CFSv2 data, but with a similar span.

As with all previous comparisons, the TCCON \(\delta\)^2H are too enriched compared to the in-situ measurements, which represent the best case scenario. The mixing models are similar to the in-situ measurements, probably due to how dry the top of the atmosphere is, so has almost no impact on the overall column. The Rayleigh models are on average about 50 per mille depleted compared to the surface.

Figure \ref{fig:dD_correlation} shows the correlation between the TCCON \(\delta\)^2H column and the different modeled columns. The mixing models show the best agreement but this probably depends on the selection of the dry end member \(\delta\)^2H and H_2O mixing ratio. It would be expected that the column \(\delta\)^2H would be more depleted than the surface, so I have probably selected a dry mixing ratio that is too low to apply a mixing model. The \(\delta\)^2H profiles shown in \ref{fig:DryProfiles} show a constant value until high in the atmosphere (>500mbar) even though there is clear boundary layer. All in-situ measurements within the free troposphere when a clear planetary boundary layer has developed show that \(\delta\)^2H is depleted relative to the surface \cite{Herman_2014,Dyroff_2015,Worden_2011}.

The correlations show that both rayleigh models are depleted compared to the TCCON retrievals for all \(\delta\)^2H. Interestingly, the bias between the TCCON data and all models becomes worse at higher \(\delta\)^2H.