Journal of Hydrology 643 (2024) 131914

Contents lists available at ScienceDirect

Journal of Hydrology
journal homepage: www.elsevier.com/locate/jhydrol

Research papers

Electrical conductivity fluctuations as a tracer to determine time-dependent

transport characteristics in hyporheic sediments
Jonas L. Schaper a,* , Olaf A. Cirpka a, Joerg Lewandowski b,c, Christiane Zarfl a
a
Department of Geosciences, Eberhard Karls University of Tübingen, Tübingen 72076, Germany
b
Department Ecohydrology and Biogeochemistry, Leibniz-Institute of Freshwater Ecology and Inland Fisheries, Berlin 12587, Germany
c
Geography Department, Humboldt University Berlin, Berlin 12489, Germany

A R T I C L E I N F O A B S T R A C T

Keywords: Assessing solute transport in riverbed sediments is important for quantifying the effective reactivity of hyporheic
Transient hyporheic exchange flows sediments and the magnitude of exchange flows between rivers and their river beds. A typical approach of
Reactive transport simulation estimating transport in riverbed sediments is by measuring natural tracers such as fluctuations of temperature or
Non-parametric deconvolution
electrical conductivity (EC) and fitting models to them that assume time-independent travel time distributions,
Natural tracers
implying steady-state flow. Here, we use a transport parameterization that is based on the advection–dispersion
equation (ADE) with coefficients that continuously vary in time. The ADE is solved numerically and its solution is
fitted to measured EC time series using Bayesian parameter inference. A continuous function of model param-
eters is constructed by smoothly interpolating between point values with different temporal resolution, and
Tikhonov regularization is used to avoid spurious parameter fluctuations. The approach is tested using EC time
series synchronously measured in river water and hyporheic porewater of two urban rivers in Germany and one
urban river in South Australia. For all datasets the goodness of fit was improved by introducing a time-dependent
EC offset. Estimated porewater velocities were highly transient in three out of the four datasets with values
increasing by up to a factor of six over the course of a day. Flux transients were likely related to both variations of
hydraulic gradients along and spatial shifting of flow paths. Comparison to stationary, non-parametric decon-
volution indicated that transient flow may induce multimodality in stationary travel time distributions. Given the
high temporal dynamics of transport in the streambed sediments of the three investigated urban rivers, we
envision that the presented model is also a valuable tool to improve the assessment of reactive transport in
riverbed sediments.

1. Introduction temperature and biogeochemistry control the structure of the microbial

community (Peralta-Maraver et al., 2018) and the reactivity of anthro-
The hyporheic zone, i.e., the sediments where groundwater and river pogenic compounds, nutrients and dissolved organic carbon in the
water interact, provides habitats for freshwater biota, regulates the ex- sediments (Harvey et al., 2013; Schaper et al., 2019). Knowledge on
change between groundwater and surface water, and plays an important transport characteristics in riverbed sediments furthermore informs in-
role for biogeochemical cycling and pollutant dynamics in lotic systems vestigations on hyporheic exchange flows (Knapp et al., 2017) and can
(Boano et al., 2014; Lewandowski et al., 2019). Quantitative assess- be relevant for drinking-water management when riverbank filtration is
ments of hyporheic exchange, biogeochemical cycling, and the fate of used for drinking-water production (Heberer, 2002; Huntscha et al.,
anthropogenic compounds in riverbed sediments require a thorough 2013).
understanding of the transport characteristics therein. Transport char- Knowledge on the transport characteristics across interfaces of sur-
acteristics control the time available for reactions to occur (Frei and face water and groundwater is commonly based on measurements of
Peiffer, 2016; Pittroff et al., 2017; Ginn, 1999) and influence the tem- artificial or natural tracers in the surface water and the porewater of the
perature distribution and biogeochemical conditions including the riverbed sediment. The tracers can either be measured via in-situ probes
redox zonation in riverbed sediments (Zarnetske et al., 2011). Both, (Fig. 1) or via active sampling (Knapp et al., 2017; Schaper et al., 2019).

* Corresponding author.
E-mail address: [email protected] (J.L. Schaper).

https://doi.org/10.1016/j.jhydrol.2024.131914

Available online 2 September 2024

0022-1694/© 2024 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC license (http://creativecommons.org/licenses/by-
nc/4.0/).
J.L. Schaper et al. Journal of Hydrology 643 (2024) 131914

Subsequently, the distribution of travel times from river water to sam- temperature records (Gossler et al., 2019; Palmer et al., 1992; Shanafield
pling points in riverbed sediments is assessed by fitting parametric (Luo et al., 2011). Also, temperature signals undergo comparably strong
et al., 2006; Maloszewski and Zuber, 1993) or non-parametric transfer dampening by conduction, making diurnal variation vanish at depths
functions (Cirpka et al., 2007; McCallum et al., 2014) to time series of beyond about 1 m (Vogt et al., 2010).
artificial or natural tracers. Fitting parametric transfer functions to EC fluctuations in lotic surface water can be caused by natural in-
measured time series includes adjusting the advection–dispersion stream processes such as diurnal variations of stream metabolism and
equation (Höhne et al., 2021; Schaper et al., 2019), a dual-domain associated biogeochemical cycling (de Montety et al., 2011; Hayashi
model (e.g., Briggs et al., 2018), the fractional advection–dispersion et al., 2012) or time-varying mixing of waters from sources with
equation (e.g., Meerschaert and Tadjeran, 2004), or a Continuous-Time different EC, e.g., effluents from wastewater treatment plants versus
Random Walk model (e.g., Cortis and Berkowitz, 2005) to the data. groundwater discharge (Jaeger et al., 2019). In freshwater bodies, the
Because of mixing along and between hyporheic flow paths, porewater ions contributing to EC hardly sorb and thus, transport of EC signals is
samples collected at points within the hyporheic zone are typically much less affected by retardation than temperature signals. Also, the
characterized by a distribution of travel times rather than a single travel- dispersion of the ions is practically that of an ideal tracer, rendering EC a
time value (Danckwerts, 1953; Engdahl et al., 2016; Leray et al., 2016; more precise tracer to determine characteristics of travel-time distri-
McCallum et al., 2015, 2014; Varni and Carrera, 1998). Artificial tracers butions in the riverbed sediments than heat. Conversely, EC signals in
have the advantage that background concentrations are practically zero riverbed sediments are influenced by precipitation and dissolution of
and that tracer breakthrough curves can unambiguously be related to minerals (particularly calcite), cation exchange, and other reactive
input signals. However, depending on river flow, large amounts of the processes so that EC is not fully conservative (Hoagland et al., 2017;
artificial tracer might have to be injected into the surface water to be Laudon and Slaymaker, 1997), causing an offset in EC values between
detectable in riverbed sediments. Moreover, tracer recoveries and river water and porewater at many sites. A traditional approach to deal
travel-time distributions inferred from non-continuous artificial-tracer with the offset in EC values is to subtract the respective mean values
injections are valid only for the hydrological conditions encountered from the EC time series measured in the river and in the riverbed sedi-
during the tracer test and thus fail to capture temporal variability ments, or to subtract a fitted trend signal. As example, Cirpka et al.
inherent to lotic systems. Consequently, time series of natural or envi- (2007) subtracted seasonal trends before deconvolution of EC time se-
ronmental tracers such as heat and dissolved ion-concentration are often ries for the analysis of riverbank filtration. However, the processes
used to infer transport characteristics in riverbed sediments. Due to causing the offset, such as calcite dissolution and precipitation, may vary
measurement constraints, time series of dissolved ion-concentration are with time in a manner that cannot be addressed by simple detrending.
often approximated by time series of electrical conductivity (EC), which EC time series in riverbed sediments and alluvial aquifers have pre-
can be readily measured via data-loggers. dominately been evaluated by methods that assume transport parame-
The applicability of dissolved-ion concentration as EC and heat as ters to be time-invariant such as cross-correlation (Schmidt et al., 2012;
natural tracers for riverbed sediments relies on the propagation of Sheets et al., 2002; Vieweg et al., 2016), non-parametric deconvolution
fluctuations of these properties from the river into the riverbed. One of (Cirpka et al., 2007), or numerical solutions of the advection dispersion
the most common approaches uses diurnal temperature fluctuations as equation (ADE) with constant coefficients (Schaper et al., 2019). A
signal (Hatch et al., 2006; Stallman, 1965), which can be interpreted general advantage of parameterized transfer functions, such as the
both for gaining and losing conditions. However, in comparison to sol- inverse-Gaussian distribution resulting from the ADE, is that they are
utes, the transport of heat in sediments is retarded, with the corre- fully characterized by a few parameters, which need to be estimated in
sponding retardation factor strongly depending on the porosity and the model calibration. This simplicity comes at the expense that travel-time
mineral composition of the sediments, which introduces a relatively distributions must adhere to pre-defined shapes and thus potential fea-
large uncertainty in the estimate of porewater velocities from tures of the real travel-time distributions such as broad peaks, multi-

Fig. 1. Schematic setup of the measurements. a) Placement of EC probes (CTD Diver, van Essen Instruments) within the riverbed sediments in River Ammer, the Sturt
River, and in River Erpe in 2019 to sample unconstrained flow paths. b) The River Erpe 2016 data set was recorded inside a 32 cm long ring enclosure (PVC pipe) to
minimize the effects of horizontal flow, i.e. to sampled constrained, near straight flow paths. Hypothetical flow paths are depicted as dotted arrows.

2
J.L. Schaper et al. Journal of Hydrology 643 (2024) 131914

modality and long tails might be missed. Non-parametric deconvolution, to non-stationarity of flow and transport.
on the contrary, can estimate any shape of the travel-time distribution,
but is prone to noise in the time series. Both parameterized and non- 2. Materials & methods
parametric transfer functions assume that transport characteristics are
time invariant. The non-stationary extension of non-parametric decon- 2.1. Data acquisition and study sites
volution by Liao et al. (2014) is very sensitive to noise and does not yield
unique results. In analogy to the approach of Keery et al. (2007), Vogt Time series of specific electrical conductivity (referenced to 25 ◦ C),
et al. (2010) used dynamic harmonic regression (Young et al., 1999) to absolute pressure, and temperature in overlying river water and hypo-
extract the non-stationary harmonic component with the period of one rheic pore water were collected using multi-purpose probes (CTD Diver,
day from time series of EC in a river and its sediment to estimate time- van Essen Instruments, The Netherlands). The measurement cell of the
dependent mean travel times from surface water to groundwater from EC probes installed in the riverbeds was covered in nylon socks to pre-
the phase shift between the two signals. Unfortunately, this approach vent sediment from entering the measurement chamber of the logger
requires fluctuations of river EC to be quasi-sinusoidal and its accuracy which would interfere with EC measurements, before the probes were
is largely determined by the performance of the respective signal- manually pushed into the relatively fine-grained streambed sediments
filtering technique. Model approaches that allow estimating time- encountered at all of our field sites. In the present study, four pairs of EC
varying porewater velocities and metrics of solute spreading from EC time series were recorded in three urban rivers, which all receive
time series are thus widely lacking. In addition, a highly flexible model considerable amounts of treated wastewater effluent.
that allows time-variable transport coefficients, including a time- The River-Ammer site is located upstream of the City of Tübingen
variable offset, is prone to misinterpretation of the data. In the most (Germany, study site coordinates 48.520751◦ N, 9.019334◦ E). It is
extreme case, all differences between two EC time series would be characterized by loamy to sandy sediments. Here, an EC probe was
interpreted as transport-independent offsets, whereas another extreme installed in 8 cm depth according to the scheme depicted in Fig. 1a. Data
case would make the velocities fluctuate at very high frequencies to was recorded for 3 weeks in December 2020 and January 2021. In the
match every noise-related fluctuation in the data. To avoid such effects, River Sturt (Warriparri, Australia), riverbed sediments were also char-
the temporal resolution of the apparent porewater velocity and the EC acterized by loamy to sandy material (Schaper et al. 2018a, b). Here,
offset must be chosen wisely, and smoothing of the time-varying co- data was collected in March and April 2017 at a site roughly 3 km
efficients may need to be enforced by Tikhonov regularization downstream of a municipal WWTP (study site coordinates 35.024957◦ S,
(Tikhonov and Arsenin, 1977). 138.688016◦ E). The EC probe was installed in 9 cm depth in the
Exchange flows between rivers and groundwater can be highly dy- hyporheic zone according to Fig. 1a. Two time series were recorded in
namic as a result of flow variations in the river (Wu et al., 2020, 2018). River Erpe at Heidemühle, Berlin (Germany, study site coordinates
Direct field observations of flux transients resulting from river-flow 52.478700◦ N, 13.635176◦ E). Previous investigations at the site
variations over relatively short, i.e. sub-daily, timescales are relatively included studies on the fate of trace organic compounds in hyporheic
scarce and the ability of EC as tracer to capture flux transients in sediments (Posselt et al., 2018; Schaper et al. 2018a, b) and on river-to-
streambed sediments has hardly been explored. It is furthermore un- groundwater travel times (Schaper et al., 2022). The location is char-
clear, to which extent flux transients influence the shape of travel-time acterized by losing conditions, i.e. surface water is infiltrating into an
distributions in riverbed sediments and thus the accuracy of ADE- alluvial aquifer, and the sediments are sandy. The first eight-day long
based model approaches over relatively short distances, like the depth dataset at River Erpe was collected in April 2016 in 19 cm depth by
of an EC sensor pushed into the riverbed. Over these distances the effects placing the EC probe within a cylindrical tube (PVC pipes with a
of (sub-)diurnal variation in velocity on the tracer signal will most likely diameter of 15 cm and a height of 32 cm, Fig. 1b). Similar to Schaper
not be averaged out. et al. 2018a, b, the rim of the tube was leveled with the sediment–water
The aim of the present study is to develop and test a versatile interface and the tube was installed to create a near straight flow path. A
numerical-modeling approach that allows determining dynamic travel- second dataset was collected in River Erpe in 8 cm depth in January
time distributions, parameterized by an apparent longitudinal dis- 2019 (Fig. 1a).
persivity and time-varying porewater velocities, from EC time series in
riverbed sediments. Because we have observed offsets in the mean EC
values between the river water and the porewater in the sediment, we 2.2. Model
also estimate a time-varying systematic offset that may reflect reactive
processes in the sediment. We choose the time-varying ADE as param- 2.2.1. Governing equation
eterization of the travel-time distribution rather than non-local models In the present study, we parameterize transport of dissolved ions,
because typical EC time series hardly allow determining time-varying expressed as the propagation of the EC signal in the riverbed, via the
coefficients of the travel-time distribution beyond its mean and one-dimensional (1-D) advection–dispersion equation with coefficients
spread. Anomalous features of the travel-time distribution, such as assumed to be constant over the travel distance but variable in time
broad peaks and long tails, that become obvious in experiments with (Genuchten and Alves, 1982). As discussed in the introduction section,
pulse-injections and require non-local transport models for their inter- some ions that contribute to EC such as chloride or sodium hardly un-
pretation, are largely lost with smooth input signals, particularly when dergo any reactions. Others, however, such as (bi)carbonate and calcium
transport conditions vary with time. may be involved in precipitation/dissolution reactions that in turn are
In the present study, we outline a strategy how to determine driven by slight changes in pH (Hayashi et al., 2012). In order to
continuous, time-dependent porewater velocities from tracer time series compensate for drifts in EC measurements and for effects introduced by
measured in-situ. Subsequently, we discuss the applicability of the reactions a time-dependent offset, o(t) (S m− 1), is added:
approach and the effects of parameter regularization schemes in light of ∂σ cons
(
∂σ cons ∂2 σ cons
)
EC time series collected in riverbed sediments (flow paths < 20 cm) of + v(t) − DC (t) = 0, (1)
∂t ∂z ∂z 2
three rivers with different patterns of EC fluctuations in their surface
water. We then compare the estimates of the time-dependent advective σ sim (t) = σ cons (t) + o(t), (2)
travel times from the rivers to points within their hyporheic zones ob-
tained by the numerical model to stationary travel-time distributions DC (t) = |v(t)|α + Dp (3)
inferred by non-parametric deconvolution, testing whether features
obtained by the non-parametric stationary deconvolution can be related in which σcons and σ sim (S m− 1) denote the conservative EC signal and the

3
J.L. Schaper et al. Journal of Hydrology 643 (2024) 131914

EC signal accounting for the offset, respectively, t (s) is time, v (m s− 1) is Inference of model parameters (v, o, α) was achieved using the dif-
the apparent vertical porewater velocity, z (m) denotes the depth below ferential evolution adaptive metropolis algorithm DREAM of Vrugt
the sediment–water interface, Dp is the pore diffusion coefficient (m2 (2016). DREAM uses multiple Markov chains, subspace sampling,
s− 1), and α (m) is the longitudinal dispersivity. In our model, we assume outlier detection and a Metropolis-Hasting sampling to derive posterior
spatially constant coefficients, not because we are sure that the sedi- distributions of the model parameters. Twenty-eight Markov chains
ments are homogeneous, but because we have only one probe in the were run in parallel during each DREAM simulation. The posterior
river water and one in the riverbed sediment so that trends in co- distribution was sampled during 14,000 iterations performed after the
efficients would have to be based on additional, unavailable informa- Markov chains had converged, which was defined by the R-statistic
̂ of
tion, e.g., from sediment cores. Likewise, the EC offset o(t), is likely the Gelman and Rubin (1992) for each model parameter to be smaller than
result of spatially dependent source-sink terms for EC, but lacking any 1.2.
additional information we simply add the offset., We are fully aware that Assuming a constant measurement error SDobs (S m− 1) of EC, the log-
flowpaths in dynamic hyporheic sediments can shift with time, and that likelihood of the measurements can be written as:
the spatial variability of hydraulic properties in conjunction with [( ) ]
transverse mixing can lead to travel-time distributions that cannot fully 1∑ m
σmeas,t − σsim,t (v, o, α) 2
L (v, o, α|SDobs , σ meas ) = − , (4)
be described by the 1-D advection–dispersion equation. However, 2 t=1 SDobs
inferring the parameters for 3-D spatially variable flow fields from the
available data would be impossible. Thus, we use Eq. (1) only for where σ meas denotes the vector of all EC-measurements, and σ sim,t (v, o, α)
parameterization to obtain time-dependent transfer functions, and we is the simulated EC value at the measurement time with index t (− ).
consider the coefficients determined upon parameter inference as To avoid overfitting regularization is needed. Regularization can be
apparent ones. In particular, the time-variable velocity may in reality achieved by penalizing large differences in consecutive parameter
reflect effects of both changes in the true porewater along fixed flow values to gain a relatively smooth continuous function of the apparent
paths velocities and shifts in travel paths. porewater velocity v(t) and the EC offset o(t). Towards this end, the sum
Similar to previous studies (Schaper et al., 2019), Eq. (1) is solved in of squared normalized differences between two subsequent parameter
Python using central differentiation in space and implicit-Euler time estimates, multiplied by regularization weights wv (− ) for porewater
integration using the SciPy (Virtanen et al., 2020) and NumPy (Harris velocity and wo (− ) for the EC offset, were added to the log-likelihood
et al., 2020) packages. EC time series in the surface water were used as function to construct the posterior log-likelihood L a of the parame-
the upper boundary condition, while EC time series measured in the ters v, o, α, given the measurements σ meas and the model:

[( ) ]
1∑ m
σmeas,t − σ sim,t (v, o, α) 2 n∑v− 1 (
vj+1 − vj )2 n∑o− 1 (
oj+1 − oj )2
L a (v, o, α|SDobs , σ meas ) = − − wv − wo , (5)
2 t=1 SDobs j=1
v j=1
o

sediment were used for parameter inference (compare section 2.2.2).

The domain was chosen considerably longer than the distance from the in which nv (− ) and no (− ) are the numbers of velocity and offset values,
river to the measurement point to minimize the effects of the outflow respectively, subject to a constant measurement error of EC (SDobs =
boundary condition, where we have set the diffusive flux to zero. Setting 0.005 mS cm− 1 ) that does not depend on the parameters. v and o used for
the diffusive flux to zero at the outflow boundary implies that the con- normalizing the parameter differences are the respective mean param-
centration gradient is forced to be zero at the end of the model domain, eter values.
which can lead either to over- or underestimation of the concentration at Determining the optimal number of interpolation knots for the
distances on the order of DC /v. Continuous time-series of v(t) and o(t) apparent porewater velocity v(t) and the EC offset o(t) and the regula-
were obtained by cubic-spline interpolation between a fixed number of rization weights wo and wv is a trade-off. The larger the density of knots,
distinct time points, denoted knots, at regular intervals. and the smaller the weights, the more flexible is the model to address
transient behavior of transport, resulting in a better goodness of fit.
2.2.2. Regularized parameter inference Conversely, this flexibility may lead to physically unreasonable results
The time-invariant longitudinal dispersivity α and the m-tuples of the that essentially map noise in the river signal to noise in the streambed
apparent porewater velocities (v) and the EC offset (o) at distinct time signal. To save computational costs, we assigned the same values to the
points were adjusted during parameter inference. In the context of the weights wo and wv . In the present study, we applied the L-curve method
present study, bold fonts indicate column vectors as used in linear (Hansen and O’Leary, 1993) to optimize these regularization weights.
algebra. In the present study, both the time resolution of the apparent The L-curves were constructed by plotting the decadic logarithm of the
( )
porewater velocity and the model of temporal variability of the EC offset residual-related contribution to L a (log10 ‖erel ‖22 , Equation SI-4) versus
were altered during different model simulations introducing twelve the decadic logarithm of the term stemming from the relative parameter
different model scenarios for each dataset with varying numbers of pa- ( )
differences (log10 ‖xrel ‖22 , Equation SI-5) for different values of wv and
rameters to be inferred: For the EC offset, three different conceptual
wo . Optimal weights were determined by maximization of the L-curve’s
models were used. The first one assumed a constant offset value
curvature c(wv , wo ). Details of the approach are outlined in the Sup-
throughout the simulation time. The second one assumed a linear trend
porting Information (SI). At the point of maximum curvature, additional
model, while the third one assumed a time-variable offset for which the
smoothing would incur relatively large deterioration of the model fit,
time resolution was set to one value per day. For the apparent porewater
whereas less smoothing would lead to unnecessary parameter
velocity, four model variants with different time resolution were
fluctuations.
considered with one, two, four, and eight apparent velocity values per
For all model runs, the root-mean square error (RMSE) was calcu-
day, respectively. The prior distributions for α, v and o were set to be
lated as a measure of the goodness of fit between the measured EC signal
uniform with minimum and maximum values of 0.005 m and 0.1 m,
and the EC signal simulated using each parameter set accepted during
0.0025 m/h and 0.15 m/h and − 0.1 and 0.1 µS cm− 1, respectively.

4
J.L. Schaper et al. Journal of Hydrology 643 (2024) 131914

posterior sampling of the respective model run. Additionally, to monotonic. The advective travel time τ(t) at the observation point at
compare the different model scenarios with respect to the goodness of fit time t is then defined by the following root-finding problem :
and the number of parameters, the Akaike Information Criterion (AIC,
d(t) − d(t − τ(t) ) − Δz = 0, (6)
see SI-section 3 for details) was used.
where Δz is the measurement depth of the tracer signal. In case that v(t)
2.2.3. Calculation of time-dependent advective travel time changes sign, there are several solutions, as the same water parcel may
The time-dependent advective travel time from the surface water to a pass the measurement point several times. In this case, we take the time
location in the streambed sediment was calculated from the time- of first advective arrival and disregard the late arrivals.
dependent apparent porewater velocity in two steps. First, the cumula-
tive distance, d(t), covered by a water parcel along a straight, vertical
2.3. Non-parametric deconvolution
flow line between the surface water and the measurement point in the
streambed sediment was calculated as a function of the time t* , that has
To investigate the consequences of neglecting transient flow in in-
passed since the end of the simulation period (t* ):
vestigations of hyporheic transport, we compare the results provided by
∫t
our non-stationary model to the results yielded by the non-parametric
d(t) = v(t* )dt* . (6)
o deconvolution approach developed by Cirpka et al. (2007), which as-
sumes that travel-time distributions are constant in time but are not
If v(t) is monotonic the cumulative-distance function d(t) is also restricted to a certain functional shape of the distribution. Under the

( )
Fig. 2. L-curve, i.e., the logarithm of the squared Euclidian norm of the model residuals, log10 ‖erel ‖22 , versus the squared Euclidian norm of the consecutive
( )
difference of parameter values, divided by the respective mean of each parameter vector, log10 ‖Δxrel ‖22 , for all model runs conducted using the River Erpe 2016
dataset and the 24 h discretization for the EC offset, i.e., the model scenario where the temporal resolution of the EC offset interpolation points was one knot per day.
For porewater velocity, one, two, four and eight knots per day. Weights are considered optimal along the L-curve where its curvature was the highest (indicated by
black arrows, e.g., wo = wv = 0.5 for eight knots per day, compare Table 1).

5
J.L. Schaper et al. Journal of Hydrology 643 (2024) 131914

Fig. 3. Measured time series and estimated parameters at River Erpe 2016. a) Stream-stage variations relative to the installation depth of the EC probe in the surface
water, b) measured (dots) and modeled (lines) time series of electrical conductivity in River Erpe 2016 dataset c) measured and modeled time series of apparent
porewater velocity v and d) mean advective travel time τ from surface water to the sediment depth of the logger. Shaded areas denote one standard deviation around
the conditional mean calculated from all accepted model realizations during posterior sampling.

6
J.L. Schaper et al. Journal of Hydrology 643 (2024) 131914

assumption of linear and time-invariant transport, any output signal y(t) Table 1
can be calculated by convolving the transfer function g(τ) and the input Overview of mean (±one standard deviation) Akaike Information criterion (AIC,
signal x(t) (Jury, 1982): − ) and root mean square error (RMSE, µS cm− 1) values calculated for different
∫T temporal resolutions of porewater velocity interpolation points (velocity-vari-
ability) and different conceptual models of EC offset variability (constant EC
y(t) = g(τ)x(t − τ)dτ. (8)
0 offset, linear trend, one and two interpolation points d-1, respectively) from all
model realizations accepted during posterior sampling. The AIC and RMSE
In hydrological terms, the transfer function g(τ), also denoted as Green’s values are shown for model runs, for which the weighting factors were deter-
function or impulse-response function, describes the linear response of mined to be ideal according to the L-curve method. The number in brackets
( )
the system to a pulse-like input function under stationary conditions. For denotes the decadal logarithm of the ideal weighting factor wv,o located at the
solute transport, it can be interpreted as a travel-time distribution times point of maximum curvature along each L-curve. Bold values highlight model
runs with lowest AIC values that were deemed to be optimal and chosen for
the signal recovery. In deconvolution, g(τ) is inferred from measured x(t)
further analysis.
and y(t) signals. In the present study, transfer functions were estimated
from the measured data without predefining the shape of g(τ) under the Velocity- EC offset variability
variability
restriction that g(τ) is smooth and non-negative, using the non-
parametric approach of Cirpka et al. (2007). Details of the method Temporal Conceptual model
resolution
have been described by Cirpka et al. (2007) and are outlined in the SI.
1/d Constant Linear 1/d
3. Results and discussion River 8 AIC 1245 ± 10 1144 ± 10 595 ± 12
Erpe, (10) (10) (5.0)
3.1. Variations of electrical conductivity and groundwater-surface water 2016 RMSE 3.1 ± 0.0 2.9 ± 0.0 1.5 ± 0.0
4 AIC 1922 ± 8 1865 ± 7 852 ± 10
interactions (10) (10) (0.1)
RMSE 4.3 ± 0.0 4.2 ± 0.0 2.9 ± 0.0
Time series of electrical conductivity (EC) measured in the river 2 AIC 3686 ± 6 3299 ± 7 1595 ± 7
water at the three study sites show pronounced and quasi-periodic (50) (10) (0.1)
RMSE 6.0 ± 0.0 5.8 ± 0.0 4.1 ± 0.0
fluctuations that, together with diurnal temperature signals, propa-
1 AIC 6126 ± 5 5774 ± 5 5264 ± 7
gated into the streambed sediments (Fig. 3b, Figures SI-07, SI-09, SI-11 (1 0 0) (0.1) (10.0)
and SI-13). In River Erpe and in Sturt River, EC fluctuations are char- RMSE 8.0 ± 0.0 7.9 ± 0.0 7.4 ± 0.0
acterized by a periodicity of approximately 24 h. EC time series in River River 8 AIC 3755 ± 12 1455 ± 12 582 ± 13
Erpe are in line with diurnal temperature and stream-stage variations Erpe, (10) (10) (10)
2019 RMSE 5.8 ± 0.0 3.5 ± 0.0 1.9 ± 0.0
(Figure SI-07 & SI-09), which are caused by time-varying discharge of 4 AIC 5486 ± 9(1) 1873 ± 8 682 ± 10
the large municipal wastewater treatment plant Münchehofe located (10) (10)
approximately 0.8 km upstream of the study site (Jaeger et al., 2019). At RMSE 7.6 ± 0.0 4.3 ± 0.0 2.3 ± 0.0
the site, further research on river-to-groundwater travel times con- 2 AIC 8067 ± 6 2452 ± 6 883 ± 8
(10) (0.1) (0.01)
ducted in June 2019 (Schaper et al., 2022) showed that stream-stage
RMSE 9.3 ± 0.0 5.1 ± 0.0 3.0 ± 0.0
variations resulted in diurnal hydraulic-head fluctuations in the adja- 1 AIC 9663 ± 5 3946 ± 5 1075 ± 6
cent riparian aquifer (recorded approximately 3 m away from River (10) (1000) (10)
Erpe, Figure SI-05). RMSE 10.2 ± 0.0 6.3 ± 0.0 3.2 ± 0.0
In River Ammer, EC fluctuations occur with a slightly irregular River 8 AIC 27619 ± 11 17138 ± 9818 ± 12
Ammer (5.0) 16(1) (10)
periodicity of approximately 8 h. They are caused by the quasi-regular RMSE 10.7 ± 0.0 8.5 ± 0.0 6.2 ± 0.0
release of effluents from a drinking-water softening plant into River 4 AIC 46499 ± 9 38472 ± 9 13404 ± 9
Ammer upstream of the sampling site (Schwientek et al., 2013). In the (1000) (1000) (0.1)
Sturt River, diurnal EC fluctuations were already present approximately RMSE 13.3 ± 0.0 12.2 ± 0.0 7.6 ± 0.0
2 AIC 49156 ± 6 40620 ± 7 22411 ± 7
100 m downstream of the WWTP effluent discharge point (Figure SI-05).
(1 0 0) (1 0 0) (1 0 0)
They were likely amplified along the 3 km river reach upstream of the RMSE 14.4 ± 0.0 13.1 ± 0.0 9.5 ± 0.0
study site by mixing with tributaries and in-stream processes such as 1 AIC 56848 ± 5 43958 ± 5 27297 ± 6
stream metabolism and hyporheic exchange flows. At the River-Ammer (1 0 0) (1 0 0) (1 0 0)
and Sturt-River field sites, levels of riparian groundwater were not RMSE 15.5 ± 0.0 13.7 ± 0.0 10.7 ± 0.0
River 8 AIC 7953 ± 12 961 ± 12 496 ± 11
recorded and thus information on differences between stream stages and
Sturt (10) (1) (1)
riparian groundwater levels is not available. Due to the propagation of RMSE 5.5 ± 0.0 1.7 ± 0.0 1.1 ± 0.1
EC signals from the surface water into the hyporheic sediments at all 4 AIC 7907 ± 9 1355 ± 8 521 ± 10
field sites, however, it is reasonable to assume that the hyporheic flow (0.1) (0.1) (0.5)
RMSE 5.8 ± 0.0 2.3 ± 0.0 1.3 ± 0.0
paths sampled as part of the present study predominantly originated in
2 AIC 13090 ± 7 2435 ± 7 1158 ± 8
the rivers and that mixing with old groundwater can be neglected at all (10) (0.01) (0.1)
sites. RMSE 7.3 ± 0.0 3.2 ± 0.0 2.2 ± 0.0
1 AIC 14396 ± 6 3191 ± 6 2784 ± 6
(10) (1) (5 0 0)
3.2. Model performance and effects of parameter regularization RMSE 7.8 ± 0.0 3.7 ± 0.0 3.2 ± 0.0

As seen in Fig. 3b, Figure SI-09c, Figure SI-11c and Figure SI-13c an
excellent agreement between the measured EC timeseries in the hypo- larger effective degree of freedom. Introducing a time-variable EC offset
rheic sediments and the best-fitting model could be achieved. For all for which the time resolution was set to one value per day, further
four datasets and all conceptual models of EC offset variation, measured improved model fits compared to model scenarios in which EC offsets
EC time series were better met with low to medium regularization were set to be constant or prescribed to vary according to a linear trend
weights and with a higher temporal resolution of the apparent pore- model. For the River-Ammer and River-Erpe-2016 datasets an increase
water velocity (Fig. 2, Figures SI 01–04, Table 1). This is expected, as a in temporal resolution from 4 to 8 velocity values per day resulted in a
higher temporal resolution with lower regularization weights implies a considerable improvement of the model performance and fit, as judged

7
J.L. Schaper et al. Journal of Hydrology 643 (2024) 131914

by a lower mean AIC and mean RMSE (±one standard deviation), 3.3. Estimated model parameters and temporal variations of porewater
respectively, calculated from all model realizations computed during velocities
posterior sampling (ΔRMSE=1.4 & 1.4 µS cm− 1 & ΔAIC=3586 & 257,
respectively). For the River-Erpe-2019 and River-Sturt datasets, how- For the highest temporal resolution of porewater velocities and
ever, only slight improvements of the fit (ΔRMSE=0.4 & 0.2 µS cm− 1, optimal weighting factors, the median (±1 interquartile range IQR)
respectively) were achieved by increasing temporal resolution from 4 to apparent porewater velocity calculated from the posterior probability
8 velocity values per day while the performance of the model as judged density distributions (posteriors) of all but the first estimated value
by the AIC was better for 8 velocity values per day (ΔAIC=100 & 25, ranged from 1.3 ± 1.7 cm/h for River Ammer to 2.7 ± 1.4 for River Erpe
respectively). Based on the mean AIC values, optimal temporal resolu- 2016, respectively, while the median (±1 IQR) of the EC offset ranged
tion of the apparent porewater velocity was thus 8 knots per day for from − 21 ± 12 µS cm− 1 for River Ammer to 58 ± 13 µS cm− 1 for River
River-Erpe-2019, River-Erpe-2016, River-Sturt and the River-Ammer Erpe 2019 (Table 2). For the same model scenarios, mean (±2 SD)
dataset, with corresponding optimal weighting factors for the EC longitudinal-dispersivity values ranged between 0.9 ± 0.0 cm in the
offset (wo ) and for porewater velocity (wv ), determined by the point of River-Erpe-2016 dataset to 7.6 ± 0.1 cm in the River-Ammer dataset
maximum curvature along the L-curves, of 10, 5.0, 1 and 0.1, (Table 2). Different temporal resolutions of apparent porewater veloc-
respectively. ities had only little effect on their overall temporal mean velocity, lon-
The temporal resolution of apparent porewater velocity is restricted gitudinal dispersivity, and EC offset values (Table SI-01). The estimated
by the mean travel time from the surface water to the measurement apparent porewater velocities and EC offsets, calculated for the highest
location and the amplitude and the frequency of traceable features in the temporal resolution of porewater velocity and optimal weighting fac-
input and output signals. An improvement of model fit with an increase tors, showed pronounced temporal variability in all datasets (Fig. 3,
in temporal resolution is likely related to higher-frequency components Figures SI-09, SI-11, SI-13 and Figure SI-19). Mean ± 1 SD (maximum)
of EC fluctuations in the river water. However, it remains unclear diurnal relative porewater velocity changes (in % vmin /vmax ) ranged from
whether the differences in improvement of model fit with an increase in 2 ± 0 (3) % in the Sturt River time series to 26 ± 8(42) % in the River-
temporal resolution from 4 to 8 velocity values per day are related to Erpe-2019 time series, respectively (Table 2).
characteristics of the input time series (e.g., higher amplitudes of higher Longitudinal dispersivity (αL ) is a function of sediment heterogeneity
order harmonics) or to flow-path characteristics. The observation, that and grain size distribution and differences observed between the data-
the constant EC offset model and the linear trend perform worse than the sets analyzed in the present study are likely related to sediment prop-
time-variable EC offset model for which the time resolution was set to erties of the respective sampling sites. The mean values estimated from
one value per day suggests that the EC offset is not only caused by drift the four datasets evaluated in the present study are in line with values
but, most likely also due to the fact, that EC is not entirely conservative. estimated by a laboratory (Chou and Wyseure, 2009) and a field study
In the subsequent sections, model parameters are reported and discussed (Liu et al., 2019) where α was found to be on the order of several cen-
for the model scenarios with the lowest AIC value and optimal weighting timeters. However, it should be noted that the EC signal measured in the
factors only. For comparison, the model results obtained by the other hyporheic zone is treated as a point measurement, although the EC
model scenarios are shown in the SI. measurement in the EC probes used in the present study is performed in
a measurement cell with an inner diameter of approximately 1 cm. As
the estimated dispersivities in the riverbed sediments of River Erpe and
Table 2 Sturt are within the same order of magnitude as the length of the mea-
Overview of dataset parameters (measurement depth and interval), model set- surement cell, they should not be considered properties of the sediments
tings, scenarios, and results of both the advection–dispersion-equation-based alone.
transport model and the non-parametric deconvolution approach. τ ADE : mean Reactive processes in the hyporheic zone alter concentrations of
advective travel time. dissolved ions and thus EC of the porewater. First of all, precipitation
River River River Sturt and dissolution reactions, particularly involving carbonates, can cause
Erpe, Erpe, Ammer River changes of the EC offset. These reactions can be altered by variations of
2016 2019 the CO2 partial pressure in the infiltrating water (caused by photosyn-
Measurement depth cm 8 19 9 8 thesis and respiration) and by temperature fluctuations. In addition, EC
Measurement interval min 5 5 2 2 measurement devices including the EC probes used here can be subject
to drift. The estimated best-fitting time series of the EC offset did not
ADE correlate to the measured temperature in the streambed sediment (i.e.,
Mean αL±1 SD α cm 1.6 ± 0.9 ± 7.6 ± 1.4 ± Spearman ρ < 0.3), neither to the estimated time series of apparent
0.0 0.0 0.1 0.1
porewater velocity nor to the measured EC values in both the surface
Median v ± IQR v cm 2.7 ± 2.8 ± 1.3 ± 2.3 ±
h− 1 1.4 1.0 1.7 3.5
and porewater (Figures SI-15 to Figures SI-18). An exception was found
Median o ± IQR o µS − 7±7 − 4 ± 14 − 21 ± 58 ± 13 for the Sturt River dataset in which the EC offset correlated positively
cm− 1
12 with the measured EC in both porewater and surface water (i.e.,
Mean ± 1 SD % 10 ± 2 26 ± 8 5 ± 0(6) 2± Spearman rank correlation coefficient > 0.8, p-value < 0.01, Figure SI-
1
vmin v−max (max.) (18) (42) 0 (3)
18). Without detailed knowledge on the ion compositions of the surface
Median τADE ± IQR H 7.1 ± 2.9 ± 7.1 ± 3.7 ±
2.3 1.0 2.3 4.2
water and porewater, we cannot make qualified statements on the most
Spearman ρ: rel. (− ) 0.72 − 0.56 0.12 − 0.35 likely processes having caused the variations in the estimated EC offsets.
Stream stage – cond. Estimates of apparent porewater velocity in the present study
mean v generally fall within the range of values found in streambed sediments in
a variety of rivers which can range from a few cm per day up to a meter
Deconvolution per hour (Boano et al., 2014). For the two datasets collected in River
Epistemic error mS 0.01 0.013 0.025 0.0045 Erpe, the estimated porewater velocities compared well to previous
cm− 1
Slope of variogram θ d-3 1.9 10.96 11.55 2.00
studies at the same site, in which vertical velocities were estimated from
Length of transfer n 500 500 600 1300 EC and temperature time series resulting in values on the order of
function meas. several cm per hour (Schaper et al. 2018a, b; Schaper et al., 2019).
Median τ ± IQR h 7.3 ± 2.6 ± 3.8 ± 18.9 ± Water exchange flows across the sediment–water interface are driven
13.6 2.9 9.3 24.6
by hydraulic gradients, which typically occur over geomorphological

8
J.L. Schaper et al. Journal of Hydrology 643 (2024) 131914

features of various spatial scales ranging from streambed ripples and positively with relative stream stage (Spearman ρ = 0.72, p < 0.01,
dunes to pool-riffle sequences and meanders and can be strongly influ- Table 2), while in the River Erpe 2019 dataset no such correlation was
enced by riparian and regional groundwater levels (Krause et al., 2022). found (Spearman ρ = 0.13, p < 0.01), although mean (±1 SD) diurnal
In the River Erpe 2016 dataset, apparent porewater velocity correlated stream stage ranges were only slightly smaller in the River Erpe 2019

Fig. 4. Time series of simulated mean advective travel time τ(t) from surface water to the sediment depth of the EC probe calculated from estimated time-dependent
velocities (left panels) and residence time distributions g(τ), estimated via non-parametric deconvolution, and a histogram of τ(t) (right panels), for the River Erpe,
River Ammer and Sturt River datasets. Note that the y-axis ticks and tick labels of the left panels also apply to the right panels.

9
J.L. Schaper et al. Journal of Hydrology 643 (2024) 131914

(17 ± 4 cm) compared to the River Erpe 2016 modeling period (23 ± 1 deconvolution to correctly align the phase of the two semi-sinusoidal
cm). Because stream-stage variations at the field site translate to varia- EC signals with a period of 24 h.
tions in hydraulic gradients between the river water stage and the ri-
parian groundwater (section 3.1, Figure SI-05), it is likely that flux 3.5. Limitations and practical recommendations
transients in the River-Erpe-2016 dataset are induced by diurnal varia-
tions of hydraulic gradients. However, at the River Erpe site, diurnal In the approach outlined in the present study, spatial parameters
hydraulic gradients seem only to result in diurnal porewater velocity such as porosity and dispersivity are assumed to be constant across the
fluctuations if the flow field is constrained to near one-dimensional flow model domain which is of course not the case in environmental systems.
and uniform conditions by a cylindrical tube. Time series of stream stage Spatial gradients of the dispersion coefficient and porosity would have
and apparent porewater velocity in the Sturt River and River Ammer the effect of an additional minor velocity component (compare section
datasets were also characterized by diurnal fluctuations, but in both SI-01 for a detailed discussion). However, due to lacking depth profiles
datasets, the time series did not correlate with each other. In River of porosity or dispersion coefficients over the distance to the EC probe,
Ammer, the Sturt River and the River Erpe 2019 datasets, it is thus likely we have opted to use spatially constant coefficients.
that the flow field in the hyporheic zone was non-uniform or multidi- Moreover, we assume that flow is predominantly vertical and that
mensional or both and that flux transients therein were related to spatial the curvature of the flow path is small. In the center of a stream,
shifts of flow paths over time rather than variations of hydraulic hyporheic flow paths often have strong vertical components (Cuthbert &
gradient along a fixed flow path. The fitted time series of porewater MacKay, 2013) while in regions with intense hyporheic exchange flows
velocities should thus be seen as apparent velocities, characterizing and near streambanks, hyporheic flow fields are characterized by larger
mean travel times related to a hypothetical path length, whereas the true non-vertical flow components and curved flow paths (Fox et al., 2014;
path lengths may vary in time due to shifts in the spatial arrangement of Reeves & Hatch, 2016; Roshan et al., 2014). Nevertheless, assuming
flow paths (see also the schematic flow paths in Fig. 1). one-dimensional flow is common in methods that attempt to determine
vertical exchange flows across sediment–water interfaces, including
3.4. Dynamic travel times in comparison to stationary travel-time VFLUX (Gordon et al., 2012), FLUXBOT (Munz & Schmidt, 2017), and
distributions LPMLEn (van Kampen et al., 2022). Besides the increased computational
cost of spatially multi-dimensional models, this is mainly due to exper-
The median advective travel time (median τADE ( ± IQR)), estimated imental difficulties one faces when attempting to measure time series of
using the non-stationary ADE model (sections 2.2 and 2.2.2) and state variables such as temperature, or in the case of the present study,
calculated from all posterior model realizations (n = 1200) over all EC, in the field in 3-D. In addition, an increased presence of instru-
model timesteps ranged from 2.9 ± 1.0 h in the River-Erpe-2019 dataset mentation in the sediment would lead to greater perturbation of natural
to 7.1 ± 2.3 h in the River-Erpe-2016 dataset (Table 2). conditions and flow paths, especially along flow paths that are relatively
The median (±IQR) travel time estimated by stationary non- short compared to the spatial dimensions of the measurement devices.
parametric deconvolution ranged from 2.6 ± 2.9 h for the River-Erpe- As long as the curvature of the flow field is small and the flow is uniform,
2019 dataset to 18.9 ± 24.6 h for the River-Sturt dataset (Table 1). 1-D estimates are reliable (Cuthbert & MacKay, 2013). However, the
For the River-Erpe-2019 and the River-Ammer datasets, median and IQR results of the approach outlined in the present study should be treated
τADE thus compare well to the estimated stationary residence time dis- with caution when the flow field is likely to be characterized by non-
tribution. For the River-Erpe-2016 and the Sturt-River datasets, how- uniform and curved flow components.
ever, the mean travel time and its standard deviation of the stationary In its current form, the model described here, is intended to be used
approach exceed the ADE-derived median and IQR values. The travel- only under strict downwelling conditions. If more than two EC time
time distribution estimated from the River-Erpe-2019 dataset by sta- series were measured along a vertical profile, our approach could
tionary non-parametric deconvolution was found to be unimodal, theoretically be used to estimate v(t) also under gaining conditions, i.e.,
whereas those derived from the Erpe-2016 and River-Ammer datasets in situations where a stream is gaining water from groundwater. How-
have a pronounced primary peak, followed by minor secondary peaks, ever, its applicability under such conditions remains unclear because EC
and the non-parametric travel-time distribution for the River-Sturt in groundwater is relatively constant over time and the effects of pore
datasets is bimodal (Fig. 4). For the datasets of River Erpe and River diffusion are typically negligible.
Ammer, the estimated non-parametric travel-time distributions in To minimize disturbance of streambed sediments and effects of probe
principle resemble inverse Gaussian distributions, suggesting that in installation on model results, we recommend not walking within 1 m of
these cases ADE-based approaches may be used to parameterize trans- the deployment site and discarding the first week of recorded data.
port along the sampled flow path (Simmons, 1982). Similar to heat-based models that use temperature depth profiles
Multimodality and broad peaks in travel-time distributions in satu- measured with multi-level temperature sticks (e.g., Munz et al., 2011),
rated porous media have been considered to be the result of mixing we note that preferential flow along the EC probes cannot be completely
between distinct flow paths which may occur in highly heterogenous or excluded in our test cases.
layered aquifers (Leray et al., 2016; McCallum et al., 2014). The Finally, our model approach is also limited by computational re-
approach presented in section 2.2 demonstrates that for three out of four quirements. The simulation of one of the eight-day time series of the
datasets analyzed in the present study, solute transport in streambed present study takes less than one second. However, the estimation of
sediments is highly time-dependent (section 3.3). Moreover, the 2016 ideal weighting factors using the L-curve approach and the subsequent
dataset was collected within a cylindrical tube and thus flow conditions sampling of a posterior distribution of model parameters given an ideal
around the EC probe were most likely one-dimensional. Multi- and weighting factor takes up to several days, depending on the number of
bimodality in deconvolution-derived travel-time distributions are most parameters, because parameter inference was performed using DREAM.
pronounced in data sets with high temporal variability of the porewater The approach outlined in the present study is thus an efficient tool for
velocity. Assuming that heterogeneity of the riverbed sediment along analyzing time series spanning up to a few weeks.
the sampled flow paths had little effects on the shape of the estimated Computational expenses can be reduced by reducing the number of
transfer functions, it is therefore likely that the multimodality, partic- parameters to be fitted. For instance, reducing the temporal resolution of
ularly in case of the River-Erpe-2016 dataset, is caused by flux tran- porewater velocity knots to 6 per day, still results in reasonable agree-
sients. It should be noted, however, that the bimodal stationary travel- ment between modelled and measured EC time series (Table 1,
time distribution estimated from the Sturt dataset, could also be a ΔRMSE=0.1–1.4 µS cm− 1) and estimated porewater velocity time series
modeling artefact, which arose due to the inability of non-parametric show similar patterns for all datasets investigated in the present study

10
J.L. Schaper et al. Journal of Hydrology 643 (2024) 131914

(Fig. S1-06, SI-08, SI-10 & SI-12). Assuming a linear trend model while series (Schaper et al., 2024, under review) which helped reducing am-
keeping 8 knots per day also reduces the numbers of parameters and biguity. Including 222Rn and other age-tracer measurements may further
results in reasonable model fits (ΔRMSE=0.6–2.3 µS cm− 1) but results in constrain the estimated parameters.
slightly different magnitudes of porewater fluctuations. If one is inter-
ested in general flux magnitudes rather than sub-daily fluctuations, a 5. Plain language summary
linear trend model or a constant offset in combination with 2 to 4 knots
per day for the porewater velocity might also be sufficient as ΔRMSE We use natural fluctuations of electrical conductivity in river water
between these scenarios and the best performing scenarios are generally as a tracer to determine time-dependent travel times between rivers and
less than 10 µS cm− 1. Depending on the required accuracy and resolu- observation points in their riverbeds. Our model assumes that vertical
tion, users of the model approach may thus choose EC offset models and velocities in the riverbed continuously vary with time. We show that the
temporal resolution of porewater velocity knots according to their travel times from the rivers to their riverbeds can be highly dynamic,
needs. which may result from variations of velocity along fixed flow paths due
to variations of hydraulic gradients or by a shift of the spatial arrange-
4. Conclusions ment of the flow paths and their lengths.

Time series of electrical conductivity are an easy-to-measure, cost-

CRediT authorship contribution statement
effective natural tracer to investigate transport in shallow (<20 cm)
streambed sediments under losing conditions. In the present study, we
Jonas L. Schaper: Writing – original draft, Investigation, Formal
estimated temporal dynamics of the apparent porewater velocity in
analysis, Visualization, Data curation, Software. Olaf A. Cirpka: Writing
streambed sediments using a user-defined temporal resolution. How-
– review & editing, Methodology, Conceptualization. Joerg Lew-
ever, a high temporal resolution of the apparent porewater velocity re-
andowski: Writing – review & editing, Funding acquisition, Data
quires regularization, which increases the computational costs and adds
curation, Conceptualization. Christiane Zarfl: Writing – review &
another step in the data analysis. Our results demonstrate that diurnal
editing, Funding acquisition.
variations in stream stage can translate to substantial transients of mean
travel time in riverbed sediments, with the apparent porewater velocity
changing considerably within a day (in our case by a factor of 6 within
Declaration of competing interest
24 h). The fitted variations of apparent velocities may either be due to
actual variations of porewater velocity along fixed flow paths induced
The authors declare that they have no known competing financial
by fluctuations of hydraulic gradients or due to a shift of the spatial
interests or personal relationships that could have appeared to influence
arrangement of flow paths and their lengths. This ambiguity has con-
the work reported in this paper.
sequences for future studies on quantitative analysis of biogeochemical
processes in hyporheic sediments. If variations of apparent velocities
Data availability
relate to variations of porewater velocity along fixed flow paths,
reactive-transport simulations can be performed along these flow paths
The data collected as part the present study are published as Schaper
accounting for the fluctuations of the velocity as one-dimensional
et al. (2024) and can be found at https://zenodo.org/records/10882771
reactive-transport problem. If, by contrast, shifts in the spatial
and https://zenodo.org/records/7997796. The code of the model used
arrangement of flow paths dominate, the reactive-transport simulations
in the present paper is on github (https://github.
may be affected by water parcels being exposed to different reactive
com/jonasschaper/VTraFlux).
zones within the streambed sediments at different times, and a full three-
dimensional resolution of reactive riverbed properties is needed to
reproduce the data. Acknowledgements
In general, the transients observed in the present study are particu-
larly relevant because we considered small travel distances so that The authors acknowledge support by the High Performance and
transient velocities could not be averaged out over the overall travel Cloud Computing Group at the Zentrum für Datenverarbeitung of the
time of the water parcels from the surface water to the sampling loca- University of Tübingen, the state of Baden-Württemberg through
tions. This could be different for riverbank-filtration applications in bwHPC and the German Research Foundation (DFG) through Grant No.
which the mean travel times from the river to an abstraction well may be INST 37/935-1 FUGG. In addition, the authors are grateful to Anja
on the order of several days to weeks so that quasi diurnal or even Hoehne and Lea Antesz, who assisted during field work. In the present
higher-frequent velocity fluctuations, as observed in the present study, study, generative AI was not used to produce text. However, ChatGPT,
have only a small impact on the overall transport behavior. an AI language model developed by OpenAI, was used for assistance
The present study further shows that flux transients can cause mul- with coding.
timodality in non-parametric deconvolution-derived travel-time distri-
butions. Thus, transport characteristics (i.e., statistical moments) Appendix A. Supplementary data
estimated from travel-time distributions using methods that assume
stationarity should be treated with caution. Pronounced tailing and Supplementary data to this article can be found online at https://doi.
multimodality of stationary travel-time distributions may simply reflect org/10.1016/j.jhydrol.2024.131914.
time averaging.
All four datasets evaluated in the present study were collected in References
rivers receiving water from urban water management facilities at time-
variable rates, which was a major driver for the more or less periodic EC Boano, F., Harvey, J.W., Marion, A., Packman, A.I., Revelli, R., Ridolfi, L., Wörman, A.,
2014. Hyporheic flow and transport processes: Mechanisms, models, and
fluctuations encountered in their surface waters. Future studies should
bioghemical implications. Rev. Geophys. 52, 603–1379. https://doi.org/10.1002/
test the applicability of EC as a tracer in streambed sediments and the 2012RG000417.
modeling approach presented in the present study also in rivers where Briggs, M.A., Day-Lewis, F.D., Dehkordy, F.M.P., Hampton, T., Zarnetske, J.P.,
EC fluctuations have less periodic causes (e.g., rainfall). In a parallel Scruggs, C.R., Singha, K., Harvey, J.W., Lane, J.W., 2018. Direct Observations of
Hydrologic Exchange Occurring With Less-Mobile Porosity and the Development of
study, we have extended the framework by inferring time-depending Anoxic Microzones in Sandy Lakebed Sediments. Water Resour. Res. 54, 4714–4729.
transport parameters using joint inversion of EC and temperature time https://doi.org/10.1029/2018WR022823.

11
J.L. Schaper et al. Journal of Hydrology 643 (2024) 131914

Chou, P.Y., Wyseure, G., 2009. Hydrodynamic dispersion characteristics of lateral inflow method using temperature time series. J. Hydrol. 336, 1–16. https://doi.org/
into a river tested by a laboratory model. Hydrol. Earth Syst. Sci. 12, 217–228. 10.1016/j.jhydrol.2006.12.003.
https://doi.org/10.5194/hess-13-217-2009. Knapp, J.L.A., González-Pinzón, R., Drummond, J.D., Larsen, L.G., Cirpka, O.A.,
Cirpka, O.A., Fienen, M.N., Hofer, M., Hoehn, E., Tessarini, A., Kipfer, R., Kitanidis, P.K., Harvey, J.W., 2017. Tracer-based characterization of hyporheic exchange and
2007. Analyzing bank filtration by deconvoluting time series of electric conductivity. benthic biolayers in streams. Water Resour. Res. 1575–1594. https://doi.org/
Ground Water 45, 318–328. https://doi.org/10.1111/j.1745-6584.2006.00293.x. 10.1002/2016WR019393.
Cortis, A., Berkowitz, B., 2005. Computing “anomalous” contaminant transport in porous Krause, S., Abbott, B.W., Baranov, V., Bernal, S., Blaen, P., Datry, T., Drummond, J.,
media: The CTRW MATLAB toolbox. Ground Water 43, 947–950. https://doi.org/ Fleckenstein, J.H., Velez, J.G., Hannah, D.M., Knapp, J.L.A., Kurz, M.,
10.1111/j.1745-6584.2005.00045.x. Lewandowski, J., Martí, E., Mendoza-Lera, C., Milner, A., Packman, A., Pinay, G.,
Cuthbert, M.O., MacKay, R., 2013. Impacts of nonuniform flow on estimates of vertical Ward, A.S., Zarnetzke, J.P., 2022. Organizational Principles of Hyporheic Exchange
streambed flux. Water Resour. Res. 49 (1), 19–28. https://doi.org/10.1029/ Flow and Biogeochemical Cycling in River Networks Across Scales. Water Resour.
2011WR011587. Res. 58 https://doi.org/10.1029/2021WR029771.
Danckwerts, P.V., 1953. Continuous flow systems. Distribution of residence times. Chem. Laudon, H., Slaymaker, O., 1997. Hydrograph separation using stable isotopes, silica and
Eng. Sci. 2, 1–13. https://doi.org/10.1016/0009-2509(96)81811-2. electrical conductivity: an alpine example. J. Hydrol. 201, 82–101. https://doi.org/
de Montety, V., Martin, J.B., Cohen, M.J., Foster, C., Kurz, M.J., 2011. Influence of diel 10.1016/S0022-1694(97)00030-9.
biogeochemical cycles on carbonate equilibrium in a karst river. Chem. Geol. 283, Leray, S., Engdahl, N.B., Massoudieh, A., Bresciani, E., McCallum, J., 2016. Residence
31–43. https://doi.org/10.1016/j.chemgeo.2010.12.025. time distributions for hydrologic systems: Mechanistic foundations and steady-state
Engdahl, N.B., McCallum, J.L., Massoudieh, A., 2016. Transient age distributions in analytical solutions. J. Hydrol. 543, 67–87. https://doi.org/10.1016/j.
subsurface hydrologic systems. J. Hydrol. 543, 88–100. https://doi.org/10.1016/j. jhydrol.2016.01.068.
jhydrol.2016.04.066. Lewandowski, J., Arnon, S., Banks, E., Batelaan, O., Betterle, A., Broecker, T., Coll, C.,
Fox, A., Boano, F., Arnon, S., 2014. Impact of losing and gaining streamflow conditions Drummond, J.D., Garcia, J.G., Galloway, J., Gomez-Velez, J., Grabowski, R.C.,
on hyporheic exchange fluxes induced by dune-shaped bed forms. Water Resour. Herzog, S.P., Hinkelmann, R., Höhne, A., Hollender, J., Horn, M.A., Jaeger, A.,
Res. 50, 1895–1907. https://doi.org/10.1002/2013WR014668. Krause, S., Wu, L., 2019. Is the hyporheic zone relevant beyond the scientific
Frei, S., Peiffer, S., 2016. Exposure times rather than residence times control redox community? Water (Switzerland) 11 (11). https://doi.org/10.3390/w11112230.
transformation efficiencies in riparian wetlands. Journal of Hydrology 543, Liao, Z., Osenbrück, K., Cirpka, O.A., 2014. Non-stationary nonparametric inference of
182–196. https://doi.org/10.1016/j.jhydrol.2016.02.001. river-to-groundwater travel-time distributions. J. Hydrol. 519, 3386–3399. https://
Gelman, A., Rubin, D.B., 1992. Inference from Iterative Simulation Using Multiple doi.org/10.1016/j.jhydrol.2014.09.084.
Sequences. Stat. Sci. 7, 428–456. https://doi.org/10.2307/2246134. Liu, D., Zhao, J., Jeon, W.H., Lee, J.Y., 2019. Solute dynamics across the stream-to-
Genuchten, V., Alves, 1982. Analytical solutions of the one-dimensional convenctive- riparian continuum under different flood waves. Hydrol. Process. 33, 2627–2641.
dispersive solute transport equation. U.S. Department of Agriculture, Technical https://doi.org/10.1002/hyp.13515.
Bulletin 1661, 151. Luo, J., Cirpka, O.A., Fienen, M.N., Wu, W.M., Mehlhorn, T.L., Carley, J., Jardine, P.M.,
Ginn, T.R., 1999. On the distribution of multicomponent mixtures over generalized Criddle, C.S., Kitanidis, P.K., 2006. A parametric transfer function methodology for
exposure time in subsurface flow and reactive transport: Foundations, and analyzing reactive transport in nonuniform flow. J. Contam. Hydrol. 83, 27–41.
formulations for groundwater age, chemical heterogeneity, and biodegradation. https://doi.org/10.1016/j.jconhyd.2005.11.001.
Water Resour. Res. 35, 1395–1407. https://doi.org/10.1029/1999WR900013. Maloszewski, P., Zuber, A., 1993. Principles and practice of calibration and validation of
Gordon, R.P., Lautz, L.K., Briggs, M.A., McKenzie, J.M., 2012. Automated calculation of mathematical models for the interpretation of environmental tracer data in aquifers.
vertical pore-water flux from field temperature time series using the VFLUX method Adv. Water Resour. 16, 173–190. https://doi.org/10.1016/0309-1708(93)90036-F.
and computer program. J. Hydrol. 420–421, 142–158. https://doi.org/10.1016/j. McCallum, J.L., Engdahl, N.B., Ginn, T.R., Cook, P.G., 2014. Nonparametric estimation
jhydrol.2011.11.053. of groundwater residence time distributions: What can environmental tracer data tell
Harris, C. R., Millman, K. J., van der Walt, S. J., Gommers, R., Virtanen, P., Cournapeau, us about groundwater residence time? Water Resour. Res. 50, 2022–2038. https://
D., Wieser, E., Taylor, J., Berg, S., Smith, N. J., Kern, R., Picus, M., Hoyer, S., van doi.org/10.1002/2013WR014974.
Kerkwijk, M. H., Brett, M., Haldane, A., del Río, J. F., Wiebe, M., Peterson, P., … McCallum, J.L., Cook, P.G., Simmons, C.T., 2015. Limitations of the Use of
Oliphant, T. E. (2020). Array programming with NumPy. In Nature (Vol. 585, Issue Environmental Tracers to Infer Groundwater Age. Groundwater 53, 56–70. https://
7825, pp. 357–362). Nature Research. https://doi.org/10.1038/s41586-020-2649-2. doi.org/10.1111/gwat.12237.
Gossler, M.A., Bayer, P., Zosseder, K., 2019. Experimental investigation of thermal Meerschaert, M.M., Tadjeran, C., 2004. Finite difference approximations for fractional
retardation and local thermal non-equilibrium effects on heat transport in highly advection-dispersion flow equations. J. Comput. Appl. Math. 172, 65–77. https://
permeable, porous aquifers. J. Hydrol. 578 https://doi.org/10.1016/j. doi.org/10.1016/j.cam.2004.01.033.
jhydrol.2019.124097. Munz, M., Oswald, S.E., Schmidt, C., 2011. Sand box experiments to evaluate the
Hansen, P.C., O’Leary, D.P., 1993. The use of the L-curve in the regularization of discrete influence of subsurface temperature probe design on temperature based water flux
ill-posed problems. SIAM J. Sci. Comput. 14 (6), 148–1503. calculation. Hydrol. Earth Syst. Sci. 15 (11), 3495–3510. https://doi.org/10.5194/
Harvey, J.W., Böhlke, J.K., Voytek, M.A., Scott, D., Tobias, C.R., 2013. Hyporheic zone hess-15-3495-2011.
denitrification: Controls on effective reaction depth and contribution to whole- Munz, M., Schmidt, C., 2017. Estimation of vertical water fluxes from temperature time
stream mass balance. Water Resour. Res. 49, 6298–6316. https://doi.org/10.1002/ series by the inverse numerical computer program FLUX-BOT. Hydrol. Process. 31
wrcr.20492. (15), 2713–2724. https://doi.org/10.1002/hyp.11198.
Hatch, C.E., Fisher, A.T., Revenaugh, J.S., Constantz, J., Ruehl, C., 2006. Quantifying Palmer, D., Blowes, D.W., Frind, E.O., Molson, J.W., 1992. Thermal energy storage in an
surface water-groundwater interactions using time series analysis of streambed unconfined aquifer: 1. Field injection experiment. Water Resour. Res. 28 (10),
thermal records: Method development. Water Resour. Res. 42, 1–14. https://doi. 2845–2856.
org/10.1029/2005WR004787. Peralta-Maraver, I., Galloway, J., Posselt, M., Arnon, S., Reiss, J., Lewandowski, J.,
Hayashi, M., Vogt, T., Mächler, L., Schirmer, M., 2012. Diurnal fluctuations of electrical Robertson, A.L., 2018. Environmental filtering and community delineation in the
conductivity in a pre-alpine river: Effects of photosynthesis and groundwater streambed ecotone. Sci. Rep. 8, 1–11. https://doi.org/10.1038/s41598-018-34206-
exchange. J. Hydrol. 450–451, 93–104. https://doi.org/10.1016/j. z.
jhydrol.2012.05.020. Pittroff, M., Frei, S., Gilfedder, B.S., 2017. Quantifying nitrate and oxygen reduction rates
Heberer, T., 2002. Tracking persistent pharmaceutical residues from municipal sewage in the hyporheic zone using 222Rn to upscale biogeochemical turnover in rivers.
to drinking water. J. Hydrol. 266, 175–189. https://doi.org/10.1016/S0022-1694 Water Resour. Res. 53, 563–579. https://doi.org/10.1111/j.1752-1688.1969.
(02)00165-8. tb04897.x.
Hoagland, B., Russo, T.A., Gu, X., Hill, L., Kaye, J., Forsythe, B., Brantley, S.L., 2017. Posselt, M., Jaeger, A., Schaper, J.L., Radke, M., Benskin, J.P., 2018. Determination of
Hyporheic zone influences on concentration-discharge relationships in a headwater polar organic micropollutants in surface and pore water by high-resolution
sandstone stream. Water Resour. Res. 53, 4643–4667. https://doi.org/10.1002/ sampling-direct injection-ultra high performance liquid chromatography-tandem
2016WR019717. mass spectrometry. Environ. Sci. Process Impacts 20, 1716–1727. https://doi.org/
Höhne, A., Lewandowski, J., Schaper, J.L., McCallum, J.L., 2021. Determining hyporheic 10.1039/c8em00390d.
removal rates of trace organic compounds using non-parametric conservative Reeves, J., Hatch, C.E., 2016. Impacts of three-dimensional nonuniform flow on
transport with multiple sorption models. Water Res. 206 https://doi.org/10.1016/j. quantification of groundwater-surface water interactions using heat as a tracer.
watres.2021.117750. Water Resour. Res. 52 (9), 6851–6866. https://doi.org/10.1002/2016WR018841.
Huntscha, S., Rodriguez Velosa, D.M., Schroth, M.H., Hollender, J., 2013. Degradation of Roshan, H., Cuthbert, M.O., Andersen, M.S., Acworth, R.I., 2014. Local thermal non-
polar organic micropollutants during riverbank filtration: Complementary results equilibrium in sediments: Implications for temperature dynamics and the use of heat
from spatiotemporal sampling and push-pull tests. Environ. Sci. Tech. 47, as a tracer. Adv. Water Resour. 73, 176–184. https://doi.org/10.1016/j.
11512–11521. https://doi.org/10.1021/es401802z. advwatres.2014.08.002.
Jaeger, A., Posselt, M., Betterle, A., Schaper, J., Mechelke, J., Coll, C., Lewandowski, J., Schaper, J.L., Posselt, M., Mccallum, J.L., Banks, E., Hoehne, A., Meinikmann, K.,
2019. Spatial and Temporal Variability in Attenuation of Polar Organic Shanafield, M., Batelaan, O., Lewandowski, J., 2018a. Hyporheic exchange controls
Micropollutants in an Urban Lowland Stream. Environ. Sci. Tech. 53, 2383–2395. fate of trace organic compounds in an urban stream. Environ. Sci. Tech. 52,
https://doi.org/10.1021/acs.est.8b05488. 12285–12294. https://doi.org/10.1021/acs.est.8b03117.
Jury, W.A., 1982. Simulation of solute transport using a transfer function model. Water Schaper, J.L., Seher, W., Nützmann, G., Putschew, A., Jekel, M., Lewandowski, J., 2018b.
Resour. Res. 18, 363–368. https://doi.org/10.1029/WR018i002p00363. The fate of polar trace organic compounds in the hyporheic zone. Water Res. 140,
Keery, J., Binley, A., Crook, N., Smith, J.W.N., 2007. Temporal and spatial variability of 158–166. https://doi.org/10.1016/j.watres.2018.04.040.
groundwater-surface water fluxes: Development and application of an analytical Schaper, J.L., Posselt, M., Bouchez, C., Jaeger, A., Nuetzmann, G., Putschew, A.,
Singer, G., Lewandowski, J., 2019. Fate of Trace Organic Compounds in the

12
J.L. Schaper et al. Journal of Hydrology 643 (2024) 131914

Hyporheic Zone: Influence of Retardation, the Benthic Biolayer, and Organic Carbon. Varni, M., Carrera, J., 1998. Simulation of groundwater age distributions. Water Resour.
Environ. Sci. Tech. 53, 4224–4234. https://doi.org/10.1021/acs.est.8b06231. Res. 34, 3271–3281. https://doi.org/10.1029/98WR02536.
Schaper, J.L., Zarfl, C., Meinikmann, K., Banks, E.W., Baron, S., Cirpka, O.A., Vieweg, M., Kurz, M.J., Trauth, N., Fleckenstein, J.H., Musolff, A., Schmidt, C., 2016.
Lewandowski, J., 2022. Spatial variability of radon production rates in an alluvial Estimatingtime-variable aerobic respiration in the streambed by combining
aquifer affects travel time estimates of groundwater originating from a losing stream. electrical conductivity and dissolved oxygen time series. J. Geophys. Res. Biogeosci.
Water Resour. Res. 1–22 https://doi.org/10.1029/2021wr030635. 121, 2199–2215. https://doi.org/10.1002/2016JG003345.Received.
Schaper, J.L., Zarfl, C., Lewandowski, J., Cirpka, O.A., 2024. VTraFlux - A model toolbox Virtanen, P., Gommers, R., Oliphant, T.E., Haberland, M., Reddy, T., Cournapeau, D.,
to determine transient vertical exchange fluxes in hyporheic sediments from time Burovski, E., Peterson, P., Weckesser, W., Bright, J., van der Walt, S.J., Brett, M.,
series of natural tracers. J. Hydrol. in Review. Wilson, J., Millman, K.J., Mayorov, N., Nelson, A.R.J., Jones, E., Kern, R., Larson, E.,
Schmidt, C., Musolff, A., Trauth, N., Vieweg, M., Fleckenstein, J.H., 2012. Transient Vázquez-Baeza, Y., 2020. SciPy 1.0: fundamental algorithms for scientific computing
analysis of fluctuations of electrical conductivity as tracer in the stream bed. Hydrol. in Python. Nat. Methods 17 (3), 261–272. https://doi.org/10.1038/s41592-019-
Earth Syst. Sci. 16, 3689–3697. https://doi.org/10.5194/hess-16-3689-2012. 0686-2.
Schwientek, M., Osenbrück, K., Fleischer, M., 2013. Investigating hydrological drivers of Vogt, T., Hoehn, E., Schneider, P., Freund, A., Schirmer, M., Cirpka, O.A., 2010.
nitrate export dynamics in two agricultural catchments in Germany using high- Fluctuations of electrical conductivity as a natural tracer for bank filtration in a
frequency data series. Environ. Earth Sci. 69, 381–393. https://doi.org/10.1007/ losing stream. Adv. Water Resour. 33, 1296–1308. https://doi.org/10.1016/j.
s12665-013-2322-2. advwatres.2010.02.007.
Shanafield, M., Hatch, C., Pohll, G., 2011. Uncertainty in thermal time series analysis Vrugt, J.A., 2016. Markov chain Monte Carlo simulation using the DREAM software
estimates of streambed water flux. Water Resour. Res. 47 (3) https://doi.org/ package: Theory, concepts, and MATLAB implementation. Environ. Model. Softw.
10.1029/2010WR009574. 75, 273–316. https://doi.org/10.1016/j.envsoft.2015.08.013.
Sheets, R.A., Darner, R.A., Whitteberry, B.L., 2002. Lag times of bank filtration at a well Wu, L., Singh, T., Gomez-Velez, J., Nützmann, G., Wörman, A., Krause, S.,
field, Cincinnati, Ohio, USA. J. Hydrol. 266, 162–174. https://doi.org/10.1016/ Lewandowski, J., 2018. Impact of Dynamically Changing Discharge on Hyporheic
S0022-1694(02)00164-6. Exchange Processes Under Gaining and Losing Groundwater Conditions. Water
Simmons, C.S., 1982. A stochastic-convective transport representation of dispersion in Resour. Res. 54 (12), 10076–10093. https://doi.org/10.1029/2018WR023185.
one-dimensional porous media systems. Water Resour. Res. 18, 1193–1214. https:// Wu, L., Gomez-Velez, J.D., Krause, S., Singh, T., Wörman, A., Lewandowski, J., 2020.
doi.org/10.1029/WR018i004p01193. Impact of Flow Alteration and Temperature Variability on Hyporheic Exchange.
Stallman, R.W., 1965. Steady one-dimensional fluid flow in a semi-infinite porous Water Resour. Res. 56 (3) https://doi.org/10.1029/2019WR026225.
medium with sinusoidal surface temperature. J. Geophys. Res. 70, 2821–2827. Young, P.C., Pedregal, D.J., Tych, W., 1999. Dynamic harmonic regression. J. Forecast.
https://doi.org/10.1029/jz070i012p02821. 18, 369–394. https://doi.org/10.1002/(SICI)1099-131X(199911)18:6<369::AID-
Tikhonov, A., Arsenin, V.J., 1977. Solutions of ill-posed problems. Halsted, New York. FOR748>3.0.CO;2-K.
van Kampen, R., Schneidewind, U., Anibas, C., Bertagnoli, A., Tonina, D., Zarnetske, J.P., Haggerty, R., Wondzell, S.M., Baker, M.A., 2011. Dynamics of nitrate
Vandersteen, G., Luce, C., Krause, S., van Berkel, M., 2022. LPMLEn–A Frequency production and removal as a function of residence time in the hyporheic zone.
Domain Method to Estimate Vertical Streambed Fluxes and Sediment Thermal J. Geophys. Res. Biogeosci. 116, 1–12. https://doi.org/10.1029/2010JG001356.
Properties in Semi-Infinite and Bounded Domains. Water Resour. Res. 58 (3) https://
doi.org/10.1029/2021WR030886.

13