Concerns about an Warning Message with groundwater signatures module

[mathiastidjani]

I have a question regarding the groundwater signatures module. Is it fully operational?

While using it, I get this error message that concerns me (WARNING: The estimated recession constant (1000.00) is close to the boundary. This may lead to incorrect results)

[dbrakenhoff]

Hi @mathiastidjani,

The module is operational, but one of the methods is generating a warning. If you look at the code for the recovery constant, you will see that it is attempting to fit an exponential function, and the parameters are bounded. The upper bound is 1000. If the parameter reaches the bound in the optimization, the result might not be correct. In that situation it might worthwhile to explore for yourself whether the result you're seeing is reasonable or not.

pastas/pastas/stats/signatures.py

Lines 1288 to 1356 in eeb8e2b

	def recovery_constant(
	series: Series,
	bins: int = 300,
	normalize: bool = False,
	min_event_length: int = 10,
	min_n_events: int = 2,
	) -> float:
	"""Recovery constant after :cite:t:`kirchner_catchments_2009`.
	Parameters
	----------
	series: pandas.Series
	Pandas Series with DatetimeIndex and head values.
	bins: int, optional
	Number of bins to bin the data to.
	normalize: bool, optional
	normalize the time series to values between zero and one.
	min_event_length: int, optional
	Minimum length of an event in days. Events shorter than this are discarded.
	min_n_events: int, optional
	Minimum number of events in a bin. Bins with less events are discarded.
	Returns
	-------
	float:
	Recovery constant.
	Notes
	-----
	Time constant of the exponential function fitted to percentile-wise binned means
	of the recovery segments. The higher the recovery constant, the slower the head
	recovers, and vice versa. The following function is fitted to the binned data
	(similar to the Exponential response function in Pastas):
	..math::
	h(t) = h_0 * (1 - exp(-t / c))
	where h(t) is the head at time t, h_0 is the final head as t goes to infinity, and
	c is the recovery constant.
	"""
	binned = _get_events_binned(
	series,
	normalize=normalize,
	up=True,
	bins=bins,
	min_event_length=min_event_length,
	)
	# Deal with the case that binned is empty
	if binned.empty:
	return nan
	# Fit an exponential model to the binned data and return the recovery constant
	f = lambda t, *p: p[0] * (1 - exp(-t / p[1]))
	popt, _ = curve_fit(
	f, binned.index, binned.values, p0=[1, 100], bounds=(0, [100, 1e3])
	)
	# Return nan and raise warning if the recovery constant is close to the boundary
	if isclose(popt[1], 0.0) or isclose(popt[1], 1e3):
	msg = (
	"The estimated recovery constant (%s) is close to the boundary. "
	"This may lead to incorrect results."
	)
	logger.warning(msg, round(popt[1], 2))
	return nan
	else:
	return popt[1]

Hope that helps somewhat!

[mathiastidjani]

Dear David, Thank you for your reply. Is it possible to have references regarding thresholds or interpretation intervals for the various signatures? Best regards

…

________________________________ De : Davíd Brakenhoff ***@***.***> Envoyé : lundi 22 avril 2024 14:54 À : pastas/pastas ***@***.***> Cc : Akominon Tidjani ***@***.***>; Mention ***@***.***> Objet : Re: [pastas/pastas] Concerns about an Error Message with groundwater signatures module (Discussion #747) Vous ne recevez pas souvent de courriers de la part de ***@***.*** Découvrez pourquoi cela est important<https://aka.ms/LearnAboutSenderIdentification> Hi @mathiastidjani<https://github.com/mathiastidjani>, The module is operational, but one of the methods is generating a warning. If you look at the code for the recovery constant, you will see that it is attempting to fit an exponential function, and the parameters are bounded. The upper bound is 1000. If the parameter reaches the bound in the optimization, the result might not be correct. In that situation it might worthwhile to explore for yourself whether the result you're seeing is reasonable or not. https://github.com/pastas/pastas/blob/eeb8e2b33783ddb2031112625b4552a4c3232428/pastas/stats/signatures.py#L1288-L1356 Hope that helps somewhat! — Reply to this email directly, view it on GitHub<#747 (comment)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/A7UDSIK2A5EXC6GCBX5C773Y6UB6XAVCNFSM6AAAAABGSSQGBSVHI2DSMVQWIX3LMV43SRDJONRXK43TNFXW4Q3PNVWWK3TUHM4TCOBYGY4DQ>. You are receiving this because you were mentioned.Message ID: ***@***.***>

[dbrakenhoff]

Check the references section in this notebook in the examples section of the documentation: https://pastas.readthedocs.io/latest/examples/signatures.html

Or check the docstrings of the signature functions themselves, these also contain references to literature where appropriate.