Introduction
Interactions between surface and subsurface flow processes are key
components of the continental hydrological cycle (Winter et al., 1998; Sophocleous,
2002), which have received particular attention in the last decades partly
because of their substantial impact on the overall response of hydrologic
systems (Boano et al., 2014; Brunner et al., 2017, and citations herein).
Several studies have recently highlighted the hydrological interactions
between surface and subsurface that have a major impact on the
biogeochemical and ecological responses of hydrosystems (e.g., Stegen et
al., 2016, 2018; Danczak et al., 2016; Partington et al., 2017). These
interactions, which are partly driven by the geomorphological
structure and the channel dynamics (Namour et al., 2015), influence flow
pathways, water mixing, residence time in the hyporheic zone along
streambeds, and the overall ecological functioning (Schmitt et al., 2011).
They are complex for several reasons, including (a) the nonlinearity of the
processes involved, (b) the strong heterogeneity of the hydrological
systems, and (c) the incidence of small-scale features on large-scale
behavior (Hester et al., 2017). Although these surface–subsurface
interactions have been extensively investigated in the last decades, several
issues relating to them require a deeper understanding to address
contemporary challenges associated with water quality and water resources
management (Brunner et al., 2017). Among these issues, monitoring and
modeling the evolution of these interactions over space and time is
fundamental (Krause et al., 2014), especially in the context of river restoration.
River restoration has been applied worldwide to counteract the undesired
effects of anthropogenic actions on river ecosystems and ecosystem services
(e.g., Wohl et al., 2015, and citations herein). From a general perspective,
the goal of restoration projects is to enhance the hydrological,
biogeochemical, and ecological functioning of large rivers and stream
hydrosystems through the reactivation of lost geophysical, geochemical, or
biological processes. Due to their firm control on biogeochemical and
ecological signatures in the so-called hyporheic zone (e.g.,
Peralta-Maraver et al., 2018), the interactions between surface and
subsurface hydrological processes may become a focus of restoration projects
(e.g., Boulton et al., 2010; Friberg et al., 2017). As examples,
surface–subsurface water exchanges generate oxygen–carbon transfers (e.g.,
Stegen et al., 2016; Danczak et al., 2016) and thermal refuges for various
aquatic species (e.g., Kurylyk et al., 2015); they also revive ponding and
renewal of water in wetlands that could otherwise turn to perishing swamps
partly disconnected from stream flow. Many projects try to improve the water
quality and/or ecological processes of the hydrosystem through engineering
works that target hyporheic exchange enhancements. Maintaining or amplifying
these interactions could prove crucial regarding climate change effects to
preserve aquatic species. Nevertheless, it is still very difficult to assess
the efficiency of such restoration projects as this requires a refined
characterization of the location and amplitude of surface–subsurface
interactions (e.g., Morandi et al., 2014).
Several advances in measurement techniques and modeling approaches appear
very promising to improve our current understanding and our forecasting
capabilities regarding surface–subsurface interactions (Krause et al., 2014;
Brunner et al., 2017). Many experimental/field projects are related to the
use of temperature as a tracer of hydrological connectivity and locations
where groundwater discharges into surface water bodies (e.g., Pfister et
al., 2010; Daniluk et al., 2013). Two different thermal techniques –
fiber-optic distributed temperature sensing (FO-DTS) and thermal infrared (TIR)
survey – have been used for their potential to inform on spatial and
temporal patterns of water fluxes in large areas of the hyporheic zone
through the determination of thermal anomalies. FO-DTS provides
one-dimensional profiles of these anomalies with a fine spatial resolution
by submerging fiber-optic cables along a streambed. TIR surveys can be
performed from air and satellite and informs on surface temperature with
two-dimensional images of various resolutions (e.g., Hare et al., 2015).
For their part, integrated hydrologic models emerged in the late 1990s, and
they are now recognized as suitable tools to investigate streamflow
generation processes at the catchment scale (e.g., Paniconi and Putti, 2015;
Fatichi et al., 2016). Although most integrated models rely on the solution
to the 3-D Richards equation to describe subsurface flow (e.g., Maxwell et
al., 2014), alternative low-dimensional approaches that simplify the
description of the subsurface compartment (still with some physical meaning)
have recently appeared (e.g., Hazenberg et al., 2015, 2016; Jeannot et al.,
2018). Solving the 3-D Richards equation with a proper discretization to
capture the complex and small-scale physics of flow in the vadose zone over
large areas may require substantial computational resources. Low-dimensional
integrated approaches that are efficient regarding computation time could
also prove beneficial to tackle practical water management issues.
Integrated models, irrespective of their level of complexity, explicitly
account for the interaction between surface and subsurface hydrological
processes. Thus, their application to hydrosystems renders insights on the
evolution over time and space of surface–subsurface interactions (e.g.,
Partington et al., 2013; Camporese et al., 2014).
Hydrologic modeling has already been used to assess the potential effects of
restoration works on the hydrologic response of a given system. The studies
reported in the ongoing literature are mainly geared towards the effect of
restoration on subsurface water table dynamics (e.g., Ohara et al., 2014),
floodplain responses (e.g., Martinez-Martinez et al., 2014; Clilverd et al., 2016),
and vegetation dynamics (e.g., Hammersmark et al., 2010). To our knowledge,
the prediction with models of hyporheic exchanges has not yet been
considered. No integrated hydrologic model has been applied to a restored
fluvial hydrosystem even though the application could reveal noteworthy data
in rendering quantitative indicators of restoration efficiency. In addition,
the combined use of thermal information with integrated hydrological models
is not yet common even though comparing and discussing both seems fruitful.
Ala-aho et al. (2015) used thermal imaging and integrated modeling to study
the exchanges between groundwater and lakes in Finland. Glaser et al. (2016)
used integrated modeling and TIR surveys to improve the calibration procedure
and investigate the dynamics of the saturated area in a small catchment in
Luxembourg. Munz et al. (2017) combined thermal measurement along the banks
of a stream and integrated modeling at the reach scale to improve the
determination of residence times in the hyporheic zone.
(a) Location of the studied area (France), (b) aerial
view of Rohrschollen Island, and (c) network of hydrologic response
measurements (mainly hydraulic heads and water fluxes).
![]()
In this paper, the low-dimensional integrated hydrologic model NIHM (for
Normally Integrated Hydrologic Model) is applied to the restored hydrosystem
of Rohrschollen Island, which is an artificial island located 8 km south of
Strasbourg (Upper Rhine, France; see Fig. 1a). Previous studies have shown
that the hydrological, sedimentological, and geomorphological dynamics of the
island were very active due to intense hyporheic exchanges and surface
processes (Eschbach et al., 2017, 2018). These dynamics were tightly linked
to the flood dynamics of the Rhine river that were progressively lost
because of territorial developments along the Rhine fluvial corridor. A
restoration project started in 2012 with the idea of improving the overall
functioning of the ecosystem through artificial injections. The restoration
actions specifically target short-term enhancement of hyporheic exchanges
over the whole island and the reactivation of sediment transport in the main
channel of the island. Even though short-term horizon effects are the main
target of the restoration, it is expected that duplicating
flooding episodes over time in the island could result in beneficial impacts on
the long-term ecological and biological health of the island.
The proposed study addresses and models a couple of these flooding episodes
with the four main objectives that are (i) to test the performance of NIHM
regarding the description of highly transient hydrologic behavior over short
periods of time; (ii) to check on the correspondences and discrepancies
between model results and TIR imaging in the delineation of exfiltration
patterns; (iii) to investigate the efficiency of restoration actions
undertaken at Rohrschollen Island, especially regarding surface–subsurface
water exchanges, and (iv) to propose optimal short-term management
procedures regarding the enhancement of surface–subsurface exchanges.
It could be argued that short-term analysis of a restored system does not
fit the general understanding stating that restoration processes are
intended to render benefits over long-term horizons. In the present case
(but also in many other cases), restoration works are recent and the system
is still evolving. This means that long-term simulations on the basis of the
actual settings of the system would probably miss its further evolution. It
makes sense to assess the behavior of a recently restored hydrosystem in
response to short-terms events. Duplicating calculations for various short
stress periods is also a way to foresee how the system could behave, even
though uncertainty and model robustness associated with the evolution of the
system over time persist. This study is limited to the analysis of the
short-term response (to flood events that are also pulse stresses) of a
transient hydrosystem via a highly resolved model in time and space.
Results and discussion
Model outputs
Figure 5 displays the evolution over time of simulated and observed
piezometric heads at two locations (transects) in the island. It also plots
simulated versus observed heads for all locations and sampling times used
during the calibration period. Heads at transects in Fig. 5 were selected to
show the best and worst match concerning RMSE between simulation and
observation. It is worth noting that, before injections peaks, the simulated
heads are mainly influenced by the Dirichlet-type boundary conditions on the
east and west sides of the island. Few data (one measure each 15 days)
were available to set up these Dirichlet boundary conditions, and the
almost-constant-over-time simulated heads before peak injections do not fully match
up to the head transients observed along the BGW. That being said, in general the
model adequately reproduces the system dynamics, capturing the two peaks of
head response associated with the injection patterns at the new channel
inlet. The recession part of the response is also captured well with a
slight overestimation of the final head value for transect T8 (Fig. 5a).
The plot of simulated versus observed heads (Fig. 5c)
confirms that the model tends to overestimate the piezometric heads as more
points are located above the 1:1 straight line. This feature is associated
with one of the founding assumptions of the model regarding the vadose zone,
which is integrated with the saturated zone and can be excessively or not
sufficiently capacitive, depending on the mean soil moisture (see Weill et
al., 2017). The values of the two performance indicators that are the RMSE
and the KGE are satisfying, at 17 cm and 0.93, respectively. Regarding the
KGE value of all measured versus simulated heads, the Pearson correlation
coefficient is 0.97, the bias ratio is 1, and the variance ratio is 1.07.
Figure 6 depicts the same information as Fig. 5 but for the validation period.
The agreement between simulated and measured heads remains good with an RMSE
of 24 cm and a KGE of 0.75, associated with a Pearson correlation
coefficient of 0.94, a bias ratio of 1, and a variance ratio of 1.24. The
decrease in the KGE values from calibration to validation steps does not
generate bias between observed and simulated head values. Nevertheless, the
variance ratio slightly increases, showing that errors between observed and
simulated heads also increase from calibration to validation. That being
said, both exercises show that the NIHM and its calibrated set of parameters
render convincing simulations of the highly transient hydrologic behavior of the system.
Comparison between simulated and measured hydraulic heads in the
subsurface during the validation period. (a, b) Evolution over time
at the two transects, that is, the worst (a) and best (b)
transects regarding RMSE. (c) Local in space and time values of
simulated hydraulic heads as a function of observed ones. RMSE is the root
mean square error, and KGE is the Kling–Gupta efficiency.
![]()
Interactions between surface and subsurface in Rohrschollen Island
Once calibration and validation were completed, the ability to capture the
interactions between surface and subsurface was checked by comparing the
modeled exfiltration patterns simulated on 22 January 2015 with the
thermal anomalies identified via airborne TIR imaging performed the same day
(see Sect. 2). In Fig. 7, the thermal anomalies are represented as pink
spots, and the simulated exfiltration patterns are represented as colored
patches ranging from blue to red as a function of the exfiltration rate.
Figure 7 focuses on the area of the island where a vast majority of the
thermal anomalies were identified. The simulated exfiltration patterns
usually coincide with the thermal anomalies from the TIR, even though their
spatial extension may be wider than thermal anomalies. This feature can be
the consequence of multiple factors, such as (a) the substantial sedimentary
heterogeneity of the streambed not sufficiently represented in the model,
(b) a spatial resolution of the computation mesh not fine enough to capture
the very small-scale surface–subsurface interactions, and (c) the
measurement uncertainty plaguing the TIR analysis. Keeping these
approximations in mind, the hydrologic model correctly locates the
surface–subsurface interactions in the island and provides flux values that
are not accessible via TIR surveys.
Comparison between simulated exfiltration patterns and thermal
anomalies identified via thermal infrared imaging close to the junction between
the new channel (southeast corner) and the BGW (Bauerngrundwasser; center of
figure). Red transects a and b are locations where surface water and groundwater
head are followed to exemplify surface–subsurface interactions in Fig. 9.
![]()
Given that a rigorous sensitivity analysis to model parameters was not
undertaken, it could be stated that flawed model parameter values are at the
origin of mismatches between TIR images and the exfiltration zones modeled
by NIHM. Nevertheless, the macroscopic hydraulic diffusion (the ratio of
conductivity to specific storage) is correctly fitted as shown by the good
match of observed heads both in time and amplitude. The point is that
thermal anomalies are visible at a scale on the order of less than 10 m,
which is also the scale of local heterogeneity of clay, sand, gravel, and
pebble deposits in alluvial systems. A numerical model handling local
heterogeneity at that scale should employ a mesh of 1–2 m resolution. In
view of the available data, building this model is unfeasible, except by
conjecturing the distribution of hydraulic parameters (as can be done for
example in stochastic approaches to the inverse problem). The lack of data
suggests that perfect accuracy cannot be expected, and the mismatch between
the measurement and model resolutions is the main reason for discrepancies
between TIR and model delineation of exfiltration zones. In addition and
under the present modeling constraints, we suggest that the quality of model
results does not relate to the fact that the model accurately represents
data over a single scenario, but rather to the fact that it roughly
represents data over multiple different scenarios (events). Unfortunately,
we only had one single set of TIR imagery at our river reach.
Groundwater head, surface water thickness, and exfiltration rate over
the whole of Rohrschollen Island for three different periods (in hours after
the beginning of injection) of the calibration period. Notably, the last period
is also the date of the airborne thermal infrared imaging.
![]()
Figures 8 and 9 picture the transient interactions between surface and
subsurface and tell us why the banana-shaped exfiltration zone reported in
Fig. 7 is close to the junction of the new artificial channel and the BGW.
Figure 8 displays maps of the groundwater head, the surface water thickness,
and the exfiltration rates over the whole island at three different times of
the calibration period that are t=50 h (i.e., after the first injection
peak), t=59 h (i.e., at the second injection peak), and t=1072 h
(i.e., the date of the airborne TIR flight). As evidenced by the snapshots
of groundwater head and surface water thickness, the water injected upstream
of the island, flowing into the BGW, its dead ends, and the associated
floodplain, rapidly infiltrates, producing an important increase in
groundwater levels alongside the new artificial channel (and also the BGW),
which had been excavated but was still not clogged with fine sediments. When
the maximum injection rate is reached (t=59 h), surface ponding occurs
on a significant portion of the island and the groundwater mounding invades
all the upstream part of the BGW. Note that the exfiltration rates (Fig. 8,
right panels) are localized in small topographic depressions during the injection
period, and the banana-shaped exfiltration pattern (Fig. 7) is still
inactive. The latter pattern only appears during the recession period (t=1072 h)
when the strong injection rates have stopped. It appears alongside
the BGW in the vicinity of the area where the groundwater level previously
increased the most. Figure 9 represents cross sections along locations a
and b in Fig. 7 for t=59 and t=1072 h and reports on the subsurface
water head, the surface water elevation (set to the topography elevation
when surface water thickness is zero), and infiltration–exfiltration rates.
It shows that (a) the topography mainly controls the banana-shaped
infiltration–exfiltration zone (depressions in Fig. 9) and (b) the temporal
dynamics and amplitude of exfiltration are the combined effect of surface
water rapidly flowing toward the system outlet (i.e., surface water
thickness diminishes) and a slow recession of the groundwater heads after
the main peaks of injected flow rates have vanished.
Evolution of surface water elevation (blue), groundwater head (red),
and exchange fluxes (arrows) along transects a and b (located in Fig. 7) at
two periods (hours after the beginning of injection) of the calibration period.
A thick grey line represents the topographic profile. The grey scale indicates
values of the saturated hydraulic conductivity at the interface between surface
and subsurface.
![]()
Figure 10 reports on the evolution over time of the total infiltration and
exfiltration fluxes calculated over the whole surface area of the island
during the two-peak calibration period. While the injection rate is kept at
2 m3 s-1, both infiltration and exfiltration fluxes are stable
with much more infiltration than exfiltration. When the injected flow rate
increases, the infiltrated flux follows a slightly delayed evolution over
time, which is very similar to the injection hydrograph (with a two-peak
shape; see Fig. 3). Meanwhile, as the hydraulic gradient between surface and
subsurface changes at some locations, the exfiltration decreases in areas
that turn from an exfiltration to an infiltration regime due to excess of
surface water associated with injection peaks. Once the injection of water
into the new artificial channel stops, the infiltration flux sharply
decreases while the exfiltration flux increases. An exfiltration peak can be
observed just at the end of the recession period. It is noteworthy that,
during the recession period, the exfiltration flux is almost constant over
time and kept at a value twice that observed before injection (Fig. 10). In
the end, forced water injections at the new channel inlet foster water
exfiltration from the subsurface that maintains ponds and wetlands on the
surface over long periods (say, approximately 15 days for each injection, as
simulated by the model but not reported in Fig. 10).
Evolution of the infiltration and exfiltration volumetric fluxes during
the first steps of the calibration period (where evolutions are essential).
![]()
Efficiency of the restoration actions
One of the issues targeted in this study is the assessment of the efficiency of
hydrological restoration projects. The previous results indicate that water
injections in the new channel enhance the interactions between surface and
subsurface compartments of the island, noting that it was observed during
the excavation that the new channel had been dug in highly conductive
sedimentary formations. It may be interesting to check via a modeling
approach what causes differences between the current restored circumstances
and a pre-restoration situation. As the pre-restored island is not well
documented in terms of hydraulic data, we considered a scenario where the
pre-restored island is similar to the current situation (including, e.g.,
geometry and boundary conditions) with the exception that the newly
excavated channel connecting Rohrschollen Island's BGW and the Rhine river
is absent. Therefore, no forced injection may occur at the southern boundary
of the pre-restored island. The hydrological behavior of the pre-restored
situation has been simulated and compared with an actual case where the
injection rate in the new channel is at the usual year-round configuration
of 2 m3 s-1.
Patterns of exfiltration for the pre-restored and the restored
situations. The focus is on the most active zone of Rohrschollen Island
regarding surface–subsurface interactions.
![]()
Figure 11 displays snapshots of exfiltration rates in a subarea of the island
for the pre-restored and the restored scenarios. Even with an injected flow
rate of 2 m3 s-1, both the exfiltration surfaces and exfiltration
rates are much higher in the restored situation. In other words, the base
flow regime of the restored situation is sufficient to positively impact the
interactions between surface and subsurface compartments of the island. When
forced injections enhance the development of wetlands and maintain high
rates of exfiltration over long periods, from the mere hydrological
standpoint, restoration works are successful.
Suggestions for management practices
The injection scenarios tested in the hydrological model with maximum peaks
reaching 80 m3 s-1 are designed as a routine inlet for feeding
Rohrschollen Island with water, but some other inlet procedures can also be
considered to improve the functioning of the island. We analyzed with the
hydrological model how these routine injections could be designed to
maximize either the spatial extension of exfiltration areas maintaining
wetlands in surface or the time over which exfiltration occurs. Two
hypothetical injections superimposed to a base flow of 2 m3 s-1 in
the new channel were proposed, with the first one being of short duration (24 h)
with an injection rate of 15 m3 s-1 and the second one being of longer
duration (120 h) but with a weaker injection rate of 5 m3 s-1 (see
Fig. 12a). As the total injected water volume differs between both
scenarios (the weaker injection flushes almost twice the volume of the
stronger injection), it can also be determined which of the two
configurations – high rate–small volume or small rate–high
volume – maximizes the extension and/or duration of exfiltration.
(a) Injection rates of two scenarios seeking optimal
exfiltration surface areas and durations at Rohrschollen Island.
(b) Evolution over time of excess or lack of exfiltration surface
area compared with exfiltration surface produced by a routine injection rate
of 2 m3 s-1 at the inlet of the system.
![]()
Figure 12b plots the excess or lack of exfiltration surface areas during
injections compared with surface areas sustained by base flow
(2 m3 s-1) in the new channel. The evolution over time of these
excess exfiltration areas (or lack thereof) occurs for both injection
scenarios with a lack of exfiltration areas occurring during the injection
periods when infiltration from the surface dominates. After the injection
peak is completed, the recession period – starting at t=52 h for the
high injection rate and t=162 h for the small injection rate (Fig. 12) – always
shows an excess of exfiltration areas. The interesting point is
that the high injection rate delivers a smaller volume of water in the
system but maintains increased areas of exfiltration over extensive periods.
For its part, the small injection rate has no effect beyond t=250 h with
a system coming back to its initial state with 2 m3 s-1 of routine
injection at the inlet. Finally, injecting less volume but with high
injection rates over short periods is better suited to maintaining
exfiltration over long periods as the process feeding wetlands on the island
(Fig. 12). It is also likely (though not studied in this work) that intense
injections favor the unclogging of the BGW, which are the primary surface
water routes contributing to water renewal on the island.
As already mentioned, the short-term behavior of the hydrosystem in response
to flood events motivated this study. In a context where long-term horizons
of the restoration benefits are the principal objective, performing
short-term simulations does not depart from this prescribed objective. The
exploration of injection scenarios discussed above with a model highly
resolved in time and space deciphers how the system currently behaves.
Duplicating that kind of simulations could for example inform on the number
and intensity of flood events needed to maintain a prescribed number of
exfiltration days (and mean flow rates) in a year. In that sense, modeling
short-term events in not necessarily in complete opposition to long-term
considerations on the modeled system.
Conclusions
Restoration projects to counterbalance the undesired effects of
anthropogenic actions on the hydrological, geomorphological, and ecological
status of riverine ecosystems have recently spread worldwide. As the
interactions between surface and subsurface compartments of the hydrosystem
have a strong impact on hydrological, biogeochemical, and ecological
processes, it makes sense to rely upon integrated hydrological modeling when
addressing the question of restoration efficiency. When feasible (i.e., with
tractable problems and models), hydrological modeling with high resolution
in time and space can accurately delineate infiltration–exfiltration areas
and their evolution over time as key factors for maintaining active surface
river networks
Relying upon simplified models, not in their physics but rather on their
dimensionality (as done in the present study), renders many problems
tractable and calculable. This is the case with Rohrschollen Island, which
shows smooth variations of topography that do not help to locate ground
water outcrops. This comment also extends to the very transient hydraulic
behaviors requiring refined time steps to accurately capture temporal
evolutions of the system.
If the focus is placed on infiltration–exfiltration patterns as a reliable
indicator of the effects of restoration in riverine systems, any spatially
distributed modeling exercise needs conditioning regarding both model inputs
and outputs. Concerning the conditioning (or control) of model outputs
associated with the delineation of exfiltration areas, the recent technique
of airborne, low-altitude, and high-resolution thermal infrared imaging is
very promising. The technique is not free of measurement errors and
artifacts, but it has been shown reliable enough to highlight interactions
between surface and subsurface compartments of the hydrosystem that coincide
with simulations. Further investigations should duplicate thermal imaging
over time with the aim of grasping the transient behavior of
surface–subsurface interactions and discussing the best versus the worst
environmental conditions where imaging is applicable.
Rohrschollen Island (and many other fluvial hydrosystems) is very specific
regarding surface–subsurface interactions, meaning that water heads in the
aquifer are often close to surface water levels. This means that slight
variations in both compartments may invert the direction of exchanged fluxes
between compartments. In that case, injecting significant volumes of water
in a system to store them over large periods may be counterproductive, even
though these volumes may contribute to flooding over large areas. Large
volumes are diverted into the rapidly flowing surface water and exit the
system. Intense injections of smaller volumes over short periods foster
intense local infiltration into the subsurface. The subsequent water
mounding in the aquifer then results in long-term storage and smooth release
of water via exfiltration. This behavior, hardly foreseeable, was that
simulated for Rohrschollen Island and could also apply to many other
configurations of fluvial corridors. These results show that management
rules for a restored system may be developed from modeling exercises
handling various forcing scenarios applied to the system. If it is accepted
that exfiltration (sustaining ponding and wetlands) is a valuable indicator
of riverine restoration, additional works should envision various settings
to improve this process. For example, it is not clear if several smaller
inlets could replace a single inlet in the system for higher efficiency. Is
water extraction from the surface and reinjection in the subsurface a
valuable process that can generate slow exfiltration over broad areas?
Physically based integrated modeling of hydrosystems might propose some answers.