Calibration of loading efficiency values

[MattBrst]

Today I tried to calibrate on the loading efficiency with:

 cal.set_aquifer_parameter(
            "leffaq",
            layers=[1],
            initial=leffaq1_river,
            pmin=0,
            pmax=1,
            log_scale=True,
            inhoms=("river"),
            model="transient",
        )

But this results in: ValueError: Parameter 'leffaq' not recognised. Supported: ['kaq', 'c', 'Saq', 'Sll', 'kzoverkh']

So I guess we would need to add leffaq and leffll in this block:

timflow/timflow/calibrate.py

Lines 1381 to 1396 in 882bc67

	def _get_aquifer_parameter_array(model, aq, name: str) -> np.ndarray:
	"""Return reference to the appropriate parameter array in the aquifer object."""
	lookup = {"kaq": aq.kaq, "c": aq.c}
	if hasattr(aq, "Saq"):
	lookup["Saq"] = aq.Saq
	if hasattr(aq, "Sll"):
	lookup["Sll"] = aq.Sll
	if hasattr(aq, "kzoverkh"):
	lookup["kzoverkh"] = aq.kzoverkh
	assert aq.model == model, "Aquifer does not belong to the given model"
	for prefix, arr in lookup.items():
	if name.startswith(prefix):
	return arr
	raise ValueError(
	f"Parameter '{name}' not recognised. Supported: {list(lookup.keys())}"
	)

Or would you guys suggest something else?

[mbakker7]

Neither leffaq nor leffll are currently implemented as parameters that can be calibrated by specifying their string name.
The reason we have not done it (yet) is that these problems already often have correlated parameters that cannot be calibrated independently before the loading efficiency parameters were added. Luckily for the loading efficiency, the values don't vary over a wide range, and reasonable estimates can be made. For example leffaq=0.8 in sand layers or leffll=1 in a clay layer. My suggestion is to do a calibration w/o loading efficiency (leffaq=lefll=0) and then one by setting them to reasonable values. If the estimated aquifer parameters differ significantly, then the loading efficiency apparently plays an important role and you may investigate further.

Regarding estimating them, we can consider building them in so they can be estimated by name, like kaq or Saq. Right now, you can already estimate them by specifying them by reference, see this notebook: https://timflow.readthedocs.io/en/latest/transient/05benchmarks/synthetic_test_calibrate.html

[mbakker7]

I will move this Issue to Discussions