Oxford MIPAS meeting#67
07 Dec 04

 
with reference to the problems on the pT error propagation matrix
there are two different issues.
 
1) many times the pT error propagation matrix is found to be zero.
 
This because the pT error propagation matrix is calculated only for
the nominal OM, when the retrieval uses a non-nominal OM the propagation
matrix is automatically put to zero.
The choice to calculate the pT error propagation matrix only for the
nominal OM was made because the calculation of this matrix is much
time consuming, and the number of non-nominal OM is large, so the
calculation for all the OMs is very expensive.
Besides the estimation of the error in VMR due to the error in p and T
for nominal OM is a reasonable estimation of the order of magnitude
of this error also in the case when a non-nominal OM is used.
 

2) for NO2 the second row of the pT error propagation matrix is zero.
 
The second row of the matrix defines the error at 60 km, where generally
the VMR of NO2 is much small.
In the calculation of pT error propagation matrix came out that the
retrieval at 60 of NO2 is much instable, and the values obtained for
the pT error propagation matrix were not reliable, so it was decided
to put to zero those values, in a way that the user could not use
a non-reliable estimation of the error.
 
The fact that the matrix is not invertible should not be a problem,
because the variance-covariance matrix obtained by the propagation
of the errors of p and T has to be added to the retrieval
variance-covariance
matrix in order to obtain the total variance-covariance matrix of the
measurement,
and this matrix (the total) is invertible.
 

The pT error propagation matrices calculated so far seem to be correct and we
don't see the need to modify the OM auxiliary files.  It is important that the
user knows that zero in the pt error propagation matrix doesn't mean that the
error is zero, but it means that a reliable estimation of the error was not
possible.