Changelog

0.4.1 (2023-07-31)

  • !PR33 FIX: the covariance inflation is now applied before the forecast step.

0.4.0 (2023-06-28)

  • !PR30 ENH: Allow to perform the analyse step by batches of gridblocks and avoid the construction on large (N, M) correlation matrices. Three new parameters have been introduced - batch_size: Number of parameters that are assimilated at once. This option is available to overcome memory limitations when the number of updated parameters is large. In that case, the size of the covariance matrices tends to explode and the update step must be performed by chunks of parameters. is_parallel_analyse_step: Whether to use parallel computing for the analyse step if the number of batch is above one. The default is True. n_batches: Number of batches required during the update step. It also introduces the support for sparse matrices for correlation matrices and observation covariance matrix.

  • !PR29 ENH: Add a function to get the ensemble variance. Convinient to compute the uncertainty a posteriori without the full covariance matrix approximation.

  • STYLE: Format with updated version of black.

0.3.3 (2022-12-12)

  • !PR27 STYLE: Add a DOI number from zenodo and correct typos.

0.3.2 (2022-10-07)

  • !PR21 FIX: design - some static methods should be moved to a utils.py file.

0.3.1 (2022-08-12)

  • !PR20 Fix ESMDA-RS documentation and change the cov_m_prior input parameter to its diagonal std_m_prior to be consistent with the implementation and be less memory consuming.

0.3.0 (2022-08-12)

  • !PR15 Implement ESMDA-RS (restricted step) which provides an automatic estimation of the inflation parameter and determines when to stop (number of assimilations) on the fly.

  • !PR14 Add keyword is_forecast_for_last_assimilation to choose whether to compute the predictions for the ensemble obtained at the last assimilation step. The default is True.

  • !PR13 Implementation: Faster analyse step by avoiding matrix inversion.

  • !PR12 Add a seed parameter for the random number generation seed in the prediction perturbation step. To avoid confusion , cov_d has been renamed cov_obs.

  • !PR11 Implement the covariance localization. Introduces the correlation matrices dd_correlation_matrix and md_correlation_matrix. To avoid confusion , cov_d has been renamed cov_obs.

  • !PR10 Implement the parameters auto-covariance inflation. Add the estimation of the parameters auto-covariance matrix. The parameter alpha becomes cov_obs_inflation_factors.

0.2.0 (2022-07-23)

  • !PR6 The parameter stdev_d becomes cov_d.

  • !PR5 The parameter n_assimilation becomes n_assimilations.

  • !PR4 The parameter stdev_m is removed.

  • !PR3 Type hints are now used in the library.

  • !PR2 Add the possibility to save the history of m and d. This introduces a new knew keyword (boolean) for the constructor save_ensembles_history. Note that the m_mean attribute is depreciated and two new attributes are introduced: m_history, d_history respectively to access the successive parameter and predictions ensemble.

0.1.0 (2021-11-28)

  • First release on PyPI.