GitLab

No commit message

Added additional options for South America fwd runs, including option to run forward when no observation is found in that cycle.

renamed class names Ct- to CO2-
changed class name Observation to Observations 
small cleanup changes to template.py

changed the function calls in order to remove need of sth.DaCycle = DaCycle in start_job. achieved by adding DaCycle to particular function calls and initializing tm5 with DaCycle as call parameter

added the get_covariance generic function to the sv baseclass

1) 1x1 fluxes

Cycle fluxes are now written in cycle-by-cycle files instead of all into one file. This makes file handling easier. Also, I now 
write the full ensemble of 1x1 fluxes to each file, resulting in much more data. The advantage is that this data can be used to post-
aggregate to any mapping needed, including mapping of the full covariance through the ensemble.

2) State fluxes

State fluxes are also written per cycle for the same reasons.

3) Olson fluxes

mean fluxes as well as the ensemble are aggregated to 240 regions (11*19 Olson ecosystems and 30 Ocean regions)

4) TransCom fluxes

mean fluxes as well as the ensemble are aggregated to 23 regions. This is similar to the dedicated routine to make TransCom
fluxes. The disadvantage is that the dedicated routine uses a staevector-to-transcom mapping which is required to be part of the 
Statevector, whereas this is not needed for the new routine. From the ensemble, aggregated fluxes and TransCom covariance can be constructed.

5) Country fluxes

mean and ensemble fluxes are written for a number of large countries (USA, China, Russia, EU27, Brazil,...). This routine can in principle handle all countries of the World, but small countries get unreliable fluxes because of aggregation errors, and the uncertainty on the parameters on smaller scales.

For the aggregation, several dictionary are used to map from the grid to the aggregate regions. These are pickled to make the routine faster, and provided also through SVN.

Fixed an error in the calculation of flux uncertainties for the output arrays. Covariances are now only calculated for TC regions (23,23) and not for the full state vector.

Had to replace a few lines to ensure proper writing of prior fluxes to the files. Until now, prior fluxes 
were retrieved from a savestate.nc file (and erroneously, the poterior was read by default) while the "old"
method to provide priors was as unscaled fluxes (all multipliers equal to 1.0). This behavior is now back.

routine can now be run as part of a cycle so that analysis finishes directly instead of post-processing

TransCom files are now created based on a 1x1 to TC conversion, note that there is no covaraince yet