Hi Patrick

It was the Frequency retrieval called using FFIT and not a 'species'. I didn't 
think about the fact that this was also a retrieval quantity... The problem was 
that I had set Q.FFIT.DF to zero. Thank you for your help!


Am 30.09.2011 um 13:05 schrieb Patrick Eriksson:

> Hi Rolf,
> 
> Can you check which retrieval quantity that generates the NaNs? That is, what 
> retrieval variable that correspinds to column 61.
> 
> You can do that checking which row in R.ji that contains 61, and then looking 
> into the corresponding array element in R.jq. jq contains text strings that 
> give the information.
> 
> Bye,
> 
> Patrick
> 
> 
> 
> 
> On 09/30/2011 12:12 PM, Rolf Rüfenacht wrote:
>> Dear qpack users
>> 
>> I get an error because of NaNs when running qpack2.m when attempting to
>> do a retrieval. The forward model with the same settings works fine.
>> The error occurs on line 119 when calling oem. In the first iteration of
>> the while loop between line 561 and 608 in oem.m I get a column of NaN's
>> in column 61 out of 65 of J when it is calculated using calc_y_and_j(
>> comfun, Q, R, O, xnorm, x, iter );. All the other columns look fine.
>> This NaN's then propagate into x when it is updated on line 594. In the
>> next iteration an error is produced because of the NaN's in x.
>> 
>> I don't understand why these NaN entries come into J?
>> 
>> When looking at the apriori it appears that the entries for xa to which
>> x is set for the first iteration are ==0 for the entries 61 to 65 and
>> ~=0 for all other entries. Could that be related to the NaN problem?
>> I only use midlatitude-winter xml files for O3, H20, N2 and O2 a
>> prioris. Only O3 was retrieved in this run.
>> 
>> Thank you for any suggestions
>> 
>> Rolf

_______________________________________________
qpack mailing list
[email protected]
https://www.sat.ltu.se/mailman/listinfo/qpack

Reply via email to