No commit message

No commit message

Changed expand_mixingratios to allow R and HPHR to be either matrix or vector depending on algorithm choice (serial or bulk).

Changed R and HPHR matrices to vectors for serial minimum least squares algorithm in optimizer to save memory (the routines needed only vectors). For the bulk algorithm this is not possible, so nothing was changed for that option.

Changed the localization routine and settings, so that the localization is actually used. Also added some extra logging messages related to localization.

The code now tests for the presence of the optimizer file, which is False for the purely forward runs. If not present, some steps related to the forecast variables are skipped.

Minor changes to flow of sample_step and sample_state. Caught a possible problem with the get_initialdata for the ObsOperator, which also needs to happen for an advance step.

Implemented the first outlines of a ForwardPipeline that simply runs TM5 with optimized parameters from another run. To be completed...

reverted to r1186 (the last one from 2012), with changes in observation files from r1192-1194 preserved

Corrected reading in of may_reject and may_localize keys from sites_weights.rc, and option added to include only certain hours of the day (for obspack files containing hourly observations)

called from pipeline as call parameters
*warning* take care about get_covariance, it is called from baseclass's "propagate", whereas this function exists only in the derived class.

a lot of my helper comments, but they will disappear soon.