Auxiliary Data

Version as of 10:46, 16 Dec 2018

to this version.

Return to Version archive.

View current version

Auxiliary information required to compare TCCON data to other data or model output are posted on the TCCON database auxiliary data page for GGG2012, and at the bottom of this page for the (now obsolete) GGG2009 data.

Comparing TCCON data with profile measurements or models

Note: this is a work-in-progress!

In order to compare our data with, say, model or high-resolution aircraft profile data, you need to use our column averaging kernels and a prioris. This information is also covered in Wunch et al. (2010) and is based on Rodgers and Connor (2003). The main equation is:

where is the quantity of interest: the smoothed column DMF (a scalar), is the TCCON a priori column (a scalar), is like a pressure weighting function (a vector), is the TCCON absorber-weighted column averaging kernel (a vector), is the DMF "truth" (either the model profile or the aircraft profile) and is the TCCON a priori profile (vector). There is one a priori profile per local day of measurements (many TCCON sites measure over two UTC days per local day because of their time zones). In order to compute the smoothed columns, you need to know the pressure weighting function, which is the ratio of the vertical column of the gas in each layer  () to the vertical column of dry air (): 

where is the dry mole fraction of the gas of interest ( or from the first equation)

where is defined by convention:

and the and are in kg/molecule, requiring Avogadro's constant to convert the molar weights. g is the gravitational acceleration and i is the atmospheric layer.

This makes the first equation become:

A future version of our .map files will contain g and all the constants necessary to compute the equations above, but you can find relationships for g as a function of altitude and latitude from, e.g.,



  • The TCCON apriori is treated as if it is a wet mole fraction. For inclusion in your model/profile-measurement intercomparison with TCCON data, it must therefore be converted to a dry-air mole fraction, via:
  • The TCCON values given are level values, NOT layer means. This means they are defined for exactly the altitude given, not an average for the layer above/below/centred at that altitude. When integrating over layers some interpolation/averaging is therefore necessary.
  • The contribution of H2O to the equations above pertaining to the integration CANNOT be ignored. We recommend using your model's own output of H2O for this, but failing that one can use the TCCON apriori. The difference induced by different H2O formulations is small, but ignoring H2O can cause a significant effect.


Sensitivity to assumptions

In order to test the effect of neglecting some of the aspects mentioned above, and highlight there importance, a number of sensitivity studies have been undertaken. These are done using CT2011 simulations, provided as "special output" at 90 minute intervals at TCCON locations. In these studies, we generate one CT "smoothed" point for each FTS measurement, thereby temporally interpolating between the spanning CT model outputs. This has a couple of advantages:

1. We can then use the solar zenith angle of the measurement to interpolate from the generic averaging kernels to provide the averaging kernel for smoothing, along with the daily TCCON apriori profile.

2. Because the TCCON data are not temporally uniform, this ensures that there are no biases between the model and data introduced by non-uniform sampling, even when averaging to daily/weekly/less-frequent periods.

Here we use 3 sites to illustrate the sensitivities - Darwin (wet, tropical), Lamont (mid-lat) and Spitsbergen (dry, polar).


  1. Incorrectly using the TCCON apriori as if it is a dry-air mole fraction.


As the figure below shows, this induces significant differences (on the order of the TCCON precision and accuracy), and could thereby compromise any comparison between TCCON measurements and models. The differences are noticeably smaller at the dry site, Spitsbergen.


  1. Incorrectly treating the as wet


By the formulation above, when integrating it is strictly necessary to consider the H2O correction as being the dry-air mole fraction of H2O (by definition, by considering it as a mixing ratio rather than a mole fraction). This makes a small difference (<< 0.1ppm) to the smoothed xCO2.




  1. Incorrectly treating the values as layer means rather than level values.


TCCON averaging kernels, aprioris and other values are provided at exact levels, not averaged across an altitude range. Here we have compared the formulation integrating across every layer using the mean of the level boundaries of the layer with assuming that the value at the lower altitude (i.e. higher pressure) is representative of the entire layer above. The differences are bordering on significant, at approximately 0.1 ppm.




  1. Incorrectly ignorning the apriori contribution to the smoothing equation


This is a common mistake, and critical. The effect of convolving the profile with the averaging kernel can be large, however it is mostly cancelled out by the effect of also convolving the apriori profile with the kernel. Ignoring the apriori therefore leads to large errors, sometimes larger than 1% depending on the zenith angle of the measurement (because of the zenith angle dependence of the averaging kernels).





  1. Incorrectly ignoring the in the integration to column values


Because the integration deals with dry-air mole fractions, but an integration over pressure is effectively with respect to wet-air (i.e. the pressure includes the contribution of H2O), this dilution due to H2O must be taken into account. This differences from ignoring this are generally small because of cancellation between the VC(gas) and the VC(air), but can exceed 0.1 ppm.




Note: These things are to be added:

  • The finer the vertical grid the better... but cannot go finer than the lowest resolution grid
    • check this (TCCON vs model grid differences)
  • time-resolution of H2O profiles (NCEP 1x vs 4x daily, e.g.)
  • How to treat smoothing of tracer components (e.g. fossil fuel, biosphere, ocean, fires in CT)?
    • subtract components from total and assess variability? fractional variability (of total)?
  • Treatment of SZAs (for AK interpolation)?


A Priori Profiles and Column Averaging Kernels for the (obsolete) GGG2009 Data

Site Averaging Kernels A Priori Profiles
Lamont, Oklahoma lamont_averaging_kernels_co2.dat
lamont_averaging_kernels_ch4.dat lamont_averaging_kernels_co.dat
Park Falls, Wisconsin

parkfalls_averaging_kernels_co2.dat parkfalls_averaging_kernels_ch4.dat

JPL, California (the instrument is now in Lamont, OK) Please use the Lamont averaging kernels jpl_apriori_profiles.tgz
Lauder, New Zealand  Please use the Lamont averaging kernels



Darwin, Australia Please use the Lamont averaging kernels. If necessary extrapolate to lower SZA. AKs binned into 5 degree SZA averages are also available on request.  db_apriori.tar.gz
Wollongong, Australia Please use the Lamont averaging kernels. AKs binned into 5 degree SZA averages are also available on request.  wg_apriori.tar.gz 
Bremen, Germany Please use the Lamont averaging kernels.  br_aprioris.tar.gz
Bialystok, Poland Please use the Lamont averaging kernels.  bi_aprioris.tar.gz 
Orleans, France Please use the Lamont averaging kernels.  or_aprioris.tar.gz 
Tsukuba, Japan  Please use the Lamont averaging kernels.



Sodankyla, Finland Please use the Lamont averaging kernels. so_apriori.tar.gz 
Izana, Spain     iz_aprioris.tar.gz 
Karlsruhe, Germany  Please use the Lamont averaging kernels.


Eureka, Canada
Garmisch, Germany Please use the Lamont averaging kernels. gm_aprioris.tgz


This site has been viewed 25781 times.