Evaluating the model with TM5 or local AMFs
Hello Arjo,
Sorry for the potentially silly question but there is one point not entirely clear to me. The output of CSO gridded includes the observed and simulated profiles with the TM5 air mass factor (ys
and yr
, respectively), as well as the observed and simulated profiles with the local (in our case MONARCH) air mass factor (ys_m
and yr_m
, respectively). I understand that in all these profiles, the averaging kernels are correctly applied. According to the documentation, "as long as the averaging kernels are used to simulated a retrieved product, a comparison between the simulation and the retrieval is independent of the air mass factor". Therefore, I would expect that evaluating the MONARCH model considering ys
versus yr
would give similar results than evaluating it considering ys_m
versus yr_m
.
However, considering for instance the CSO gridded output of 01/01/2019 at 13h, I am calculating a few statistics (I express the tropospheric columns in Pmolec/cm2) :
MB nMB RMSE nRMSE slope PCC N Mo Mm
0.05 2.4 1.20 61.2 0.59 0.56 15680 1.96 2.00 (ys versus yr)
0.07 3.5 1.38 66.9 0.70 0.63 15680 2.06 2.13 (ys_m versus yr_m)
(NB: Mean Bias, nMB: normalized Mean Bias, RMSE: Root Mean Square Error, nRMSE: normalized Root Mean Square Error, slope: slope of the scatterplot, PCC: Pearson Correlation Coefficient, N: number of points, Mo: Mean observation, Mm: Mean model)
As you can see, there are some substantial differences in the statistics obtained here.
The ratios ys/ys_m
and yr/yr_m
are quite similar but not identifical:
MB nMB RMSE nRMSE slope PCC N Mo Mm
0.00 0.0 0.05 4.8 0.96 0.97 15680 1.02 1.02 (ys/ys_m versus yr/yr_m)
Are such differences expected? Thanks in advance!