Checking your initial assimilation¶
You may require several attempts to get your assimilation configured correctly. The next section, Computing filter increments, describes how to take the difference between two assimilation stages to determine whether your initial assimilation worked as intented.
If your assimilation does not change anything in the model state, you may need
filter multiple times to understand what is wrong.
Thus you should make
filter very fast to run. You can do this by:
Making an observation sequence file containing a single observation.
Configuring your run so that filter does a single assimilation and exits without having to advance the ensemble of models or do other work.
Making an observation sequence file containing a single observation¶
You can use one of these methods to make an
obs_seq with just a single
create_obs_sequenceto make a new, short, observation sequence file.
obs_sequence_toolto cut an existing
obs_seq.outfile down to just a few obs by selecting only a subset of the types and setting a very short time window, such as a second or two when you know there are observations available.
These programs are described in the Programs directory.
Configuring your run so that filter does a single assimilation and exits¶
filter to only do a single assimilation:
input.nmlto set the
init_time_secondsto match the observation time in your truncated observation sequence file. This overrides any times in the input files and ensures that
filterwill only assimilate and not try to advance the model.
Make sure the truncated observation sequence file contains only a single observation or observations close enough together in time to fit into a single assimilation window.