HyPar_DAFiltering module

class HyPar_DAFiltering.DAFilter(DA_Filter_Configs)

Bases: _filter_base.DAfilterBase

This creates an object that that will take care of the whole DA process. To initialize an object:

>> Filter_Object = DAFilter(Config_dict)

while Config_dict is a configuration dictionary obtained by reading the file da_solver.inp. This can be obtained as follows:

>> Config_dict=_utility.read_filter_configurations([file_name = <file_name='da_solver.inp'>])

Filter_Object is the data assimilation object with the necessary filter configurations set.

DA_Monitor_Add(monitor)
DA_filtering_cycle(Model_Object, Xf, X_k1, ObsVec, nens, update_B=True)
DA_filtering_cycle(Model_Object, Xf, X_k1, ObsVec, nens[, update_B=True])

given observation vector, background state/ensemble, and statistics at a specific time instant, this function produces the analysis ensemble for the given cycle.

Parameters:
  • Model_Object (object) – An object created from the class ModelBase_HyPar.
  • Xf (numpy.ndarray) – the forecast ensemble array. (Nvar x Nens): each memeber is stored as a column.
  • X_k1 (numpy.ndarray) – previous analysis ensemble (at the previous assimilation time instance). (Nvar x Nens): each memeber is stored as a column.
  • ObsVec (numpy.ndarray) – observation vector at the current time instance.
Returns:

Xa, Filter_Statistics.

  • Xa is the analysis ensemble. (Nvar x Nens): each memeber is stored as a column.

  • Filter_Statistics is a dictionary containing filter statistics such as:

    • Effective sample size,
    • Acceptance probabilities,
    • Number of objective function evaluations,
    • Number of gradient evaluations,
    • etc.

    Statistics differ from filter to filter.
DA_filtering_process(Model_Object, in_Ref_IC=None)
DA_filtering_process(Model_Object[, in_Ref_IC=None])

Sweep the filter timespan and do assimilation at points where observations are available as requested in the filter configurations file da_solver.inp.

Parameters:
  • Model_Object (object) – An object created from the class ModelBase_HyPar.
  • in_Ref_IC (numpy.ndarray) – reference initial condition vector. This is needed for RMSE evaluation and to generate observation if it is synthetic. It is created by default as Model_Objec._Ref_IC when the model object is initalized. For example, in HyPar models this is called by compiling ``aux/init.c/C file in the model directory.
Returns:

This function does not return something as a variable. It saves the assimilation results to the assimilation results directory which by default is results_directory=assimilation_results
create_Initil_Ensemble(Model_Object, forecast)
create_Initil_Ensemble(Model_Object, forecast)

Create and save an initial ensemble to be used as analysis at the initial time point. The current scheme is so simple. More advanced schemes will be considered!

Parameters:
  • Model_Object (object) – An object created from the class ModelBase_HyPar.
  • forecast (numpy.ndarray) – background/forecast initial condition vector. This is obtained by perturbing the initial reference solution by adding gaussian noise with covariance matrix B.
Returns:

This function does not return something as a variable. It saves the initial ensemble in the temporary directory out_dir = 'ensemble_out
set_DA_filter(Model_Object, set_uncertainty_parameters=True, create_initial_ensemble=True)
set_DA_filter(Model_Object[, set_uncertainty_parameters=True][, create_initial_ensemble=True])

Sets up the data assimilation model/filter based on information in the configurations file. a shallow copy of the model object is passes just to extract necessary information

Parameters:
  • Model_Object (object) – An object created from the class ModelBase_HyPar.
  • set_uncertainty_parameters (bool) – if, True (default) uncertainty parameters such as model error covariance matrix, background error covariance matrix and observation error covariance matrix, based on settings in the filter configuration file da_solver.inp.
  • create_initial_ensemble (bool) – if True (default) an intial ensemble is created and saved to disk by calling Model_Object.generate_initialForecast().