# How to convert units of averaging kernel from DU/DU to ppbv/ppbv

For validation of model data, I want to compare the ozone data from model to the satellite results downloaded from the following link:

The satellite data are in DU(Dobson Unit) while the model data are in ppbv. To make them comparable, I converted the units of satellite data from DU to ppbv from the formula specified in the README file.

But I have problem converting values of average kernel(AKs). The readme file specifies that

Note the AKs for different units (e.g., DU or ppbv) are different, the DU/DU does not cancel out because the conversion from DU to ppbv varies with altitude (e.g. resulting from different T, P).

But, I am not sure what is the right way convert the values of Average Kernel from DU/DU to ppbv/ppbv so that I can use its values.

• potentially related Dobson unit
– uhoh
Commented Feb 15, 2022 at 0:31
• If I understand your question, the averaging kernels aren't part of the conversion which is correct. Instead, the averaging kernel is a unitless multiplier (should be near 1.0) that is applied at every layer. You don't convert units of the AK because by definition it is unitless. You can do the conversion without the AKs, and compare the biases with/without (as discussed in the link you provided).
– f.thorpe
Commented Feb 15, 2022 at 2:32
• Also see Figure 1 for an example of averaging kernels and how they are centered on 1.0. acp.copernicus.org/articles/3/1285/2003/acp-3-1285-2003.pdf
– f.thorpe
Commented Feb 15, 2022 at 2:37
• @f.thorpe So that means I can use the values of AK without converting? I am just confused, as they said in the manual that the AKs for different units (e.g., DU or ppbv) are different. Commented Feb 15, 2022 at 3:32
• The AK would be different for DU vs ppbv since those units are not linearly proportional with altitude. Before you use an AK, you have to convert your model data to the same layers as specified in the AK (discussed in the next page in that README). To properly interpolate your vertical resolution to the satellite retrieval layers, you have to do a thickness-weighted averaging of model layer concentrations. You can't just average the model layers that span the retrieval layers, since they likely have varying thicknesses. Then you can compare model/sat w/ or w/o the AK... your choice.
– f.thorpe
Commented Feb 15, 2022 at 4:28