Workflow

This package is based around file input and output, with experiment configurations defined as function arguments using the NamedTuple data type. A basic workflow to run a data assimilation twin experiment is to first generate a time series for observations using a choice of tuneable parameters using the GenerateTimeSeries submodule. Once the time series data is generated from one of the benchmark models, one can use this data as a truth twin to generate pseudo-observations. This time series can thus be re-used over multiple configurations of filters and smoothers, holding the pseudo-data fixed while varying other hyper-parameters. Test cases in this package model this workflow, to first generate test data and then to implement a particular experiment based on a parameter configuration to exhibit known behavior of the estimator, typically in terms of forecast and analysis root mean square error (RMSE).

Standard configurations of hyper-parameters for the truth twin and the data assimilation method are included in the SingleExperimentDriver submodule, and constructors for generating maps of parallel experiments over parameter grids are defined in the ParallelExperimentDriver submodule. It is assumed that one will Install a dev package this package in order to define new parameter tuples and constructors for parallel experiments in order to test the behavior of estimators in new configurations. It is also assumed that one will write new experiments using the FilterExps and SmootherExps submodules as templates.