Hi all, There were still some problems in the code with respect to the logrel option. And looking back on the documentation it seems that I have been unclear about some changes around rel and logrel.
The present strategy is that both retrieval input and output strictly follow the selected unit. And to be clear, there is no conversion for logrel, neither for input Sx or output data. (This is no change regarding the input, but differs on the output side from Qpack1.) I am sorry of if this point not has been expressed clearly enough. In addition, there have been several changes around these units. The rel option has hopefully throughout been OK. For logrel several bugs have been fixed, and likely there have been undetected ones. In short, this originates in bad implementation in Qpack1 and it has taken some time remove the old dust. Anyhow, things should now be OK. But you are still encouraged to keep an eye open if you use logrel. The main issue should be qp2_rel2vmr. Ole Martin and I have together gone through the expressions to apply. Hopefully, they are correctly implemented. I have now tried to improve the documentation around these points. For example, a new section in the manual. Note that qp2_rel2vmr is just converting the fields starting with species (and not e.g. J). And note that conversion between different units can cause "unexpected" effects. For example, inside qpack2_demo you can play around with the retrievals units. Try here to use the rel option (case 2). Convert this L2 with qp2_rel2vmr, and plot the measurement response! It is highly oscillating. But you get the same result with case 3, which is the corresponding case defined directly in VMR. And this is still a quite "nice case" compared to e.g. H2O and CO where the VMR profile has more sharp gradients. It seems that you most easily get a nice measurement response by a fixed a priori uncertainty, independently of the unit. (I don't think this an issue of numerical accuracy) Can anyone explain this? Bye, Patrick _______________________________________________ qpack mailing list [email protected] https://www.sat.ltu.se/mailman/listinfo/qpack
