Open Paszka1 opened 5 months ago
You are correct - the standard_deviation
column in the external observation data is just a simple way to separate the weights and the noise. So if you dont provide a cov matrix (or noise realizations), both pest++ and pyemu will look for non-zero weighted observations that have a non-null standard_deviation
values and use those in the noise draw. Here is where pyemu checks for it:
I would like to get a confirmation about my understanding of how the method
ObservationEnsemble.from_gaussian_draw()
behaves. The description of the method in the online pyEMU documentation doesn't mention about the optionalstandard_deviation
column in the 'observation data external' file. It only states that the draw will be based on weights if cov is missing. However, I understand based on section 9.1.6 of the PEST++ manual, on demo exercises and videos, that if a covariance matrix is not provided for this method and if thestandard_deviation
column is present, then weights will only be used for history matching purposes, and thestandard_deviation
column is used for all other purposes involving uncertainties (including drawing ensembles). Is this correct?