Recharge Model Features - Peterson & Western / Flex (Groundwater Uptake)

[martinvonk]

Description

This pull request introduces two new features to the recharge models:

1. Peterson & Western nonlinear recharge model. This model was originally described by Peterson & Western (2014) - Nonlinear time-series modeling of unconfined groundwater head. The recharge model is based on equation 6 from this article and has five parameters. If parameter gamma = 1, the model behaves similarly to the FlexModel which is already implemented in Pastas. Some good time series model results were obtained with this recharge model, compared to other nonlinear recharge models.
2. A root uptake parameter for the gw_uptake=True optionality in the FlexModel is added. Currently the excess evaporation is subtracted fully from the recharge flux when gw_uptake=True. With this pull request a parameter f is introduced which scales the excess evaporation between 0 and 1. This parameter is found to take sensible values, while not changing other parameter values too much.

I have not made tests for these features yet (don't know how :)).
I'll add an example notebook to the repository pastas/test-datasets because we have permission to use the data of some nonlinear groundwater time series.

Edit 04-03-2022

Example notebook can be found on pastas/test-datasets/nonlinear_wrij

Checklist before PR can be merged:

• is documented
• PEP8 compliant code
• tests added / passed
• Example Notebook (for new features)
• API changes documented in Release Notes

[raoulcollenteur]

Thanks, interesting addition. Curious how identifiable the gw evaporation factor is in practice. Couple of things before we can Pull:

• Remove all unnecessary changes in the FlexModel (they cause water balance errors in the final time step).
• Add a test for FlexModel (ps.rch.FlexModel(gw_uptake=True)
• Add tests for Peterson (look at / copy the tests for the FlexModel in the tests folder).
• Add docstrings for the Peterson model. Mention you use explicit Euler where Peterson basically uses implicit.
• Should we import Peterson model by default ?
• We could add a code-block to the recharge notebook with this model (where we compare Berendrecht, Flex and Linear already).
• Edit: Open PR That proposes this changes such that it will show up in the release notes automatically :-)

[raoulcollenteur]

See comments in the code and the GH Conversation

[martinvonk]

Thanks, interesting addition. Curious how identifiable the gw evaporation factor is in practice. Couple of things before we can Pull:

• Remove all unnecessary changes in the FlexModel (they cause water balance errors in the final time step).
• Add a test for FlexModel (ps.rch.FlexModel(gw_uptake=True)
• Add tests for Peterson (look at / copy the tests for the FlexModel in the tests folder).
• Add docstrings for the Peterson model. Mention you use explicit Euler where Peterson basically uses implicit.
• Should we import Peterson model by default ?
• We could add a code-block to the recharge notebook with this model (where we compare Berendrecht, Flex and Linear already).
• Edit: Open PR That proposes this changes such that it will show up in the release notes automatically :-)

All done.
The changes to the FlexModel were due to some merge conflicts. I approved the wrong changes, my bad.

Tests are now added for Peterson (except water balance test). I don't think Peterson should be default (yet). Some significant water balance errors might occur still because of using forward Euler. Currently a more quick & dirty implementation is used but this could/should be improved in the future.

[raoulcollenteur]

Going to check the notebook now and will merge after if all works.