Remote Sensing of Environment 218 (2018) 44–54

Contents lists available at ScienceDirect

Remote Sensing of Environment

journal homepage: www.elsevier.com/locate/rse

Retrieving river baseflow from SWOT spaceborne mission T

a,* a,* a b
Fulvia Baratelli , Nicolas Flipo , Agnès Rivière , Sylvain Biancamaria
a
MINES ParisTech, PSL Research University, Geosciences Department, Fontainebleau, France
b
LEGOS, Université de Toulouse, CNES, CNRS, IRD, UPS, 14 avenue Edouard Belin, Toulouse 31400, France

A R T I C LE I N FO A B S T R A C T

Keywords: The quantification of aquifer contribution to river discharge is of primary importance to evaluate the impact of
Surface Water and Ocean Topography (SWOT) climatic and anthropogenic stresses on the availability of water resources. Several baseflow estimation methods
Hydrology require river discharge measurements, which can be difficult to obtain at high spatio-temporal resolution for
Baseflow large basins. The future Surface Water and Ocean Topography (SWOT) satellite mission will provide discharge
Hydrograph separation
estimations for large rivers (> 50–100 m wide) even in ungauged basins. The frequency of these estimations
Groundwater
Seine river basin
depends mainly on latitude and ranges from zero to more than ten values in the 21-day satellite cycle. This work
aims at answering the following question: can baseflow be estimated from SWOT observations during the
mission lifetime? An algorithm based on hydrograph separation by Chapman's filter was developed to auto-
matically estimate the baseflow in a river network at regional scale (> 10 000 km2). The algorithm was applied
to the Seine river basin (75 000 km2, France) using the discharge time series simulated at daily time step by a
coupled hydrological-hydrogeological model to obtain the reference baseflow estimations. The same algorithm is
then forced with discharge time series sampled at SWOT observation frequency. The average baseflow is esti-
mated with good accuracy for all the reaches which are observed at least once per cycle (relative bias less than
8%). The time evolution of baseflow is also rather well retrieved, with a Nash-Sutcliffe coefficient above 0.7 for
96% of the network length. An analysis of the effect of SWOT discharge uncertainties on baseflow estimation
shows that bias is the component of discharge error that most contributes to the error on baseflow. Anyway,
when the combined effect of SWOT discharge sampling and SWOT discharge uncertainties is considered, the
error on baseflow estimates is slightly smaller than that on discharge. This work provides new potential for the
SWOT mission in terms of global hydrological analysis and water cycle closure.

1. Introduction
Streamflow can be conceptually separated into two components: a
high frequency component (quickflow) including runoff, interflow and
direct precipitation, and a low frequency component (baseflow) in-
cluding mainly groundwater discharge to the river but also the con-
tribution of other natural storages like snow, glaciers, lakes, wetlands
(Brodie et al., 2007; Brodie and Hostetler, 2005; Smakhtin, 2001). The
quantification of baseflow is of major importance for the management
of water resources and the assessment of river water availability (low-
flow hydrology and flood hydrology) and quality. This is especially true
when streamflow is sustained by baseflow in the dry season (Miller
et al., 2016; Partington et al., 2012; Howard and Merrifield, 2010;
Brodie and Hostetler, 2005). During those periods, most of the bio-
geochemical processes depends on the hydrodynamics in the river-
aquifer interfaces (Flipo et al., 2014; Marmonier et al., 2012). It was for
instance recently shown that even though the bacterial activity is
significant in the hyporheic zone, its contribution to green house gas
emissions depends on the river-aquifer configuration (losing or gaining
reach), which is related to climate (Newcomer et al., 2018). Also
groundwater discharge to rivers contributes to the formation of cold
thermal refuges for some fish species, helping to protect those species
from the negative effects of climatic warming (Wawrzyniak et al., 2016;
Dugdale et al., 2013).
Several baseflow estimation methods, especially filtering separation
techniques, require river discharge measurements, which are difficult to
obtain at high spatio-temporal resolution for large basins, especially in
remote areas (Durand et al., 2016; Sichangi et al., 2016; Tarpanelli
et al., 2013). The Surface Water and Ocean Topography (SWOT) sa-
tellite mission could provide such data. The SWOT mission is a colla-
boration between the National Aeronautics and Space Administration,
the Centre National d’Études Spatiales (the French Spatial Agency), the
Canadian Space Agency and the United Kingdom Space Agency
(Biancamaria et al., 2016), with launch planned in 2021. The objective

*
Corresponding authors.
E-mail addresses: fb2mountains@gmail.com (F. Baratelli), nicolas.flipo@mines-paristech.fr (N. Flipo).

https://doi.org/10.1016/j.rse.2018.09.013
Received 9 February 2018; Received in revised form 10 September 2018; Accepted 16 September 2018
Available online 21 September 2018
0034-4257/ © 2018 Elsevier Inc. All rights reserved.
F. Baratelli et al.
of this mission is to measure ocean and terrestrial surface water (lakes,
reservoirs, rivers and wetlands) topographies at high accuracy and high
spatial resolution at global scale. In particular, the mission is required
to provide water level, slope and width for rivers larger than 100 m.
Moreover, rivers larger than 50 m might also be observed if resources,
instruments and processing methods are able to. The mission will have
a nominal orbit repeat period of 21 days and a duration of 3 years. The
spatial coverage is unprecedented: 96.45% of land area between 78° S
and 78° N will be observed by the satellite at least once per cycle. The
frequency of SWOT observations depends on both latitude and long-
itude and ranges from zero to more than ten observations per cycle.
Given a longitude, the observation frequency increases with latitude: at
the equator the maximum frequency is two observations per cycle,
while above 70° N/S it reaches ten (Biancamaria et al., 2016). The
mission will transform land hydrology, as no data similar to SWOT
observations are currently available at global scale (Altenau et al.,
2017; Pavelsky et al., 2014; Alsdorf et al., 2007). SWOT observations of
water surface elevation can be assimilated in hydrologic/hydraulic
models to improve river depth modeling (Biancamaria et al., 2011).
They can be used to retrieve hydraulic information of ungauged braided
rivers (Garambois et al., 2017) as well as to support flood modeling
(Mason et al., 2016) and the development of global inundation maps
(Pekel et al., 2016; Fluet-Chouinard et al., 2015). River discharge es-
timation from satellite observations is one of the mission goals. Some
algorithms have been developed for this purpose and work is in pro-
gress to improve their performance (Durand et al., 2016, 2014;
Garambois and Monnier, 2015; Paiva et al., 2015; Gleason and Smith,
2014).
The future availability of discharge time series for rivers larger than
50–100 m opens a new perspective on the quantification of various
water cycle components, including aquifer contribution to river dis-
charge. One of the main issues is that the discharge time series esti-
mated for river reaches which are observed only once or twice per cycle
might not be suitable for the application of baseflow separation algo-
rithms, as flood peaks and recession periods might not be detected. An
evaluation of the impact of SWOT irregular time sampling on the esti-
mation of average discharge around the season peak flow (Papa et al.,
2012) showed that the uncertainty is acceptable for all watersheds
analyzed. However, to the authors' knowledge, the possibility to esti-
mate baseflow from SWOT data has not been investigated so far.
In this framework, the objective of the present work is to answer the
following question: can baseflow be estimated at global scale from
SWOT observations during the mission lifetime? To this aim, an algo-
rithm based on Chapman filter is developed to automatically estimate
baseflow in a river network from SWOT-like discharge time series, i.e.,
from discharge time series with the same time resolution as what will
be obtained from SWOT observations. As a proof of concept, the algo-
rithm is tested on the Seine river basin (75 000km 2, France), for which
a coupled hydrological-hydrogeological model simulates river dis-
charge at daily time step (Pryet et al., 2015). The application of the
Chapman filter to these daily time series provides a reference baseflow
distribution which is used to assess the performance of the algorithm.
Guidance is then provided for initializing the filter properly, which will
strengthen baseflow estimates at the beginning of the mission. The
impact of the uncertainties of SWOT discharge estimations on the
performance of the baseflow algorithm is also analyzed.
2. Methods for estimating baseflow from SWOT mission discharge
product
Among the different baseflow estimation methods that have been
developed in the literature (Brodie and Hostetler, 2005; Smakhtin,
2001; Nathan and McMahon, 1990), the filtering separation methods
are widely used due to their minimal input requirements and simple
implementation that can easily be automated. The recursive digital
filters are signal processing techniques that remove the high frequency
45
Remote Sensing of Environment 218 (2018) 44–54
quickflow signal from a streamflow time series in order to obtain the
low frequency baseflow signal. Such filters include the Lyne and Hollick
filter (Lyne and Hollick, 1979), the Chapman one-parameter algorithm
(Chapman, 1991; Chapman and Maxwell, 1996), the Boughton two-
parameter algorithm (Chapman, 1999), the Eckhardt two-parameter
algorithm (Eckhardt, 2005) and the IHACRES three-parameter algo-
rithm (Chapman, 1999). In this work, baseflow is estimated from si-
mulated daily river discharge time series by means of the Chapman
filter, as described in the following section.
2.1. The Chapman filter
The Chapman filter (Chapman, 1999) is a one-parameter algorithm
separating river discharge into a high frequency component, which
includes runoff, interflow and direct precipitation, and a low frequency
component, which corresponds to baseflow. The algorithm was first
developed by Chapman (1991) and subsequently improved by
Chapman and Maxwell (1996). It assumes a linear relationship between
storage in the aquifer and river discharge during the recession period
when the high frequency component ceases. This linear behavior can be
derived from the 1D Darcy equation in a confined aquifer of constant
thickness and is a reasonable approximation for unconfined aquifers
when the underlying impermeable layer is well below the stream bed
(Chapman, 1999). According to this hypothesis, baseflow Qb(t)
[m3 s−1] decreases exponentially with time during the recession period:
Qb (t ) = Q0 e−t / τ = Q0 k t , (1)
3 −1
where Q0 [m s ] is the peak river discharge at the beginning of the
recession period, τ [s] is the residence or turnover time and k = e−1/τ is
the recession parameter. For a general time t, baseflow is given by the
following expression (Chapman, 1999):
k 1−k
Qb (t ) = Qb (t − 1) + Q (t ),
2−k 2−k (2)
3 −1
where Q [m s ] is river discharge, with the constraint that Qb ≤ Q.
The recession parameter k is the only parameter required by the
Chapman algorithm. Eq. (1) shows that the logarithm of k can be es-
timated by linear regression between the logarithm of Q(t)/Q0 and time
during flow recession.
Given a discharge time series, baseflow is estimated in this study
according to the following methodology. First, the discharge peaks are
selected. Then, for each peak, the corresponding recession period of
non-increasing discharge is considered and the associated parameter k
is computed by linear regression. The baseflow time series is then
computed with Eq. (2) using a value of k which is constant between two
discharge peaks and varies between different recession periods. In order
to initialize the algorithm (Eq. (2)), a low-flow period is chosen, so that
Qb(t = t0) = Q(t = t0).
3. Data
3.1. The study site: Seine river basin
The Seine river basin (75 000 km2), located in the north of France
(Fig. 1), is the most urbanized and industrialized basin in France (Billen
et al., 2007). Given its large population and food production, water
resources in the Seine river basin are of high strategic importance as
about 109 m3 of groundwater is extracted every year over the whole
basin (Pryet et al., 2015). The climatic regime is pluvial oceanic,
modulated by seasonal variations in evapotranspiration. The discharge
at the gauging station of Poses (Fig. 1), which is located in the most
downstream part of the river network, at the entrance of the estuary, is
480 m3 s−1 (average over the period 1974–2000). The river network is
28 000 km long. According to the requirements of the mission
(Rodríguez, 2016), SWOT will observe rivers wider than 100 m, that is,
it will observe 594 km in the Seine river network. According to the
F. Baratelli et al.
Fig. 1. a) Distribution of baseflow relative bias over three years (average over the 30 starting dates) in the Seine river basin. The Seine river downstream of Poses will
be observed by SWOT, but it is not considered in this study as it is tidally influenced. b) Linear regression between baseflow from SWOT discharge and reference
baseflow (average over the 30 starting dates).
goals of the mission, SWOT may also observe rivers wider than 50 m. In
this case, SWOT would observe 1060 km in the Seine river network.
3.2. Hydrosystem modeling platform for simulating river discharge
The distributed hydrological-hydrogeological model CaWaQS is
used to generate river discharge in the Seine river basin to test the
baseflow estimation method. CaWaQS was initially developed by Flipo
et al. (2007a,b) and has been recently upgraded by Labarthe et al.
(2015) based on the concepts of the EauDyssée platform, whose de-
scription can be found in Baratelli et al. (2016), Flipo et al. (2012), and
Pryet et al. (2015). CaWaQS simulates the various components of the
water cycle in a hydrosystem at regional scale (103–106 km2). It couples
different modules which simulate the surface water mass balance, the
runoff, the river flow, the flow in the unsaturated zone and the flow in
the saturated zone. The surface water balance module is a conceptual
model that computes runoff, real evapotranspiration, soil storage and
infiltration from the input data of precipitations and potential evapo-
transpiration. Runoff is routed to the stream network, where river
discharge is simulated with the Muskingum method (Cunge, 1969). In-
stream water levels are calculated from discharge with a simplified
Manning-Strickler approach (Baratelli et al., 2016; Chow, 1959). In-
filtrated water flows through the unsaturated zone before reaching the
saturated zone. This is simulated by a conceptual model which re-
presents the unsaturated zone as a succession of reservoirs. Each re-
servoir discharges into the next one depending on a constant drainage
46
Remote Sensing of Environment 218 (2018) 44–54
coefficient that simulates an exponential drainage throughout time
(Flipo et al., 2012). The modeled area is discretized into square cells
with size ranging from 1 to 8 km. Each cell is characterized by three
parameters: the number of reservoirs, the reservoir thickness and the
drainage coefficient. The flow in the saturated zone is simulated with a
physically-based distributed hydrogeological model for multi-layer
aquifer systems initially developed by Ledoux (1975) and de Marsily
et al. (1978). This model solves the diffusivity equation (de Marsily,
1986) with a quasi-3D finite-differences scheme: the flow is assumed to
be horizontal in each layer, whereas the aquitards between the layers
are characterized by vertical flow. The different modules are coupled by
the interactions between surface water and groundwater (stream-
aquifer exchanges, aquifer overflow). River-aquifer exchanges are
computed for each river cell with a conductance model (Flipo et al.,
2014).
CaWaQS was implemented in the Seine river basin by Pryet et al.
(2015) and Labarthe et al. (2015). The model was calibrated and va-
lidated over a 17-year period. 17 years is the duration of the cycle as-
sociated with the North Atlantic Oscillation (NAO) (Flipo et al., 2012;
Massei et al., 2010) and represents the minimal period of time for
calculating relevant pluri-annual statistics of hydrological quantities. In
the calibration period the Nash-Sutcliffe efficiency coefficient (Nash
and Sutcliffe, 1970) for river discharge is greater than 0.5 for 60% of
the 74 gauging stations present in the basin and it is 0.81 at the station
of Poses. The global RMSE for the 183 piezometers in the basin is
3.64 m (Pryet et al., 2015).
F. Baratelli et al.
Fig. 2. Discharge at Bazoches-les-Bray. a) Reference discharge (gray line), reference baseflow (black line), baseflow from SWOT discharge (red line: median of the
distribution obtained with 30 starting dates; blue dotted line: 95% percentile, blue dashed line: 5% percentile). b) Baseflow relative residual. c) Difference between
95% and 5% percentile in the case of 3 observations per cycle (black) and 1 observation per cycle (gray). (For interpretation of the references to color in this figure
legend, the reader is referred to the web version of this article.)
3.3. Estimation of reference river discharge and baseflow
The SWOT observable river network (Fig. 1) is discretized in reaches
200 m long. For each reach, the river discharge simulated by CaWaQS
at daily time step in the three years period 2012/09/01–2015/08/31 is
assumed as the reference river discharge. The reference baseflow is
obtained by applying the Chapman filter to the reference discharge as
described in Section 2.1. The filter is initialized from 2011/10/07, that
is, 11 months before the study period and during low-flow conditions.
For example, the discharge measured at that date at the Bazoches
gauging station (Fig. 1), is 16.7 m3/s, which corresponds to the 13%
percentile of the discharge distribution over the 17-year period 1999/
09/01–2016/08/01.
3.4. Estimation of baseflow from SWOT-like river discharge time series
In order to obtain river discharge time series representative of those
that will be estimated from SWOT observations, the reference discharge
time series are sampled at SWOT observation frequency, which is ir-
regular and depends on the position along the river network. In the
Seine river basin, the number of observations ranges from zero to four
for each 21-day cycle. 98.2% of the river reaches larger than 50 m will
be observed by the satellite. 15.7% will be observed once per cycle,
43.4% twice, 30.2% three times and 8.9% four times per cycle. SWOT
observation frequency is calculated for each river reach from the
47
Remote Sensing of Environment 218 (2018) 44–54
ground track estimated by the CNES.
As a first step, only the effect of SWOT irregular sampling on
baseflow estimation is analyzed (Section 4). A simple analysis of the
effect of the uncertainties on discharge time series estimated from
SWOT observations is presented in Section 5, where the combined ef-
fect of SWOT sampling and discharge errors is also considered.
SWOT launch date is planned in April 2021. After the launch there
will be different test phases and a 3-month calibration period.
Therefore, the starting date of SWOT observations is not known yet. In
this study, we consider 30 different values for the starting date of SWOT
observations, namely, between 09/01 and 09/30, and we test the cor-
responding impact on baseflow estimations. A different starting date
implies a different sampling of the discharge time series as well as a
different initialization for the baseflow algorithm. For each of the 30
starting dates, the reference discharge time series are sampled to obtain
SWOT-like discharge time series. The latter are then linearly inter-
polated between the observation dates in order to obtain daily dis-
charge time series to which the Chapman filter is applied to estimate
baseflow.
4. Retrieving distributed baseflow at the basin scale: the effect of
SWOT temporal sampling of discharge
The reference 3-year average baseflow in the Seine River is
46 m3 s−1 at the Bazoches station, 190 m3 s−1 at the Austerlitz station
F. Baratelli et al.
and 260 m3 s−1 at the Poses station. The corresponding values of the
baseflow index, i.e., the ratio between average baseflow and average
discharge, are 0.49 at Bazoches and 0.50 at Austerlitz and Poses. Along
the entire SWOT-observable river network, the baseflow index ranges
between 0.46 and 0.52.
4.1. Performances of the Chapman filter
For each of the 30 potential starting dates of the mission, statistical
criteria are calculated to compare baseflow estimated from SWOT-like
discharge to the reference baseflow. The relative bias, averaged over
the 30 scenarios, ranges from −2.6% to 7.6% and its average over the
network is 1% (Fig. 1a). It amounts to 3.5% at Bazoches and 1.1% at
Poses, which will be observed three times per cycle, and to 0.9% at
Austerlitz, which will be observed once per cycle. The absolute bias
ranges from −5.1 to 12 m3 s−1 and its average is 1.4 m3 s−1. The
average reference baseflow and the average baseflow estimated from
SWOT discharge are very well correlated (Fig. 1b). The slope of the
regression line is 1.01 so that the average SWOT baseflow only slightly
overestimates the average reference baseflow. Moreover, the perfor-
mance does not depend significantly on the number of SWOT ob-
servations per cycle (Fig. 1b) and it is good even for reaches which are
observed only once per 21-day cycle.
In order to assess the performance of the algorithm in terms of
baseflow time variability, the Nash-Sutcliffe coefficient was also cal-
culated for all the starting dates and for all the reaches. The Nash-
Sutcliffe, averaged over the 30 scenarios, ranges from 0.55 to 0.98 and
its average is 0.90. Moreover, it is higher than 0.7 for 96% of the net-
work length. It amounts to 0.95 at Bazoches, to 0.90 at Poses and to
0.85 at Austerlitz. As an example, the baseflow time series obtained
from SWOT-like data at Bazoches is compared to the reference baseflow
time series (Fig. 2). For each time, the distribution of the 30 baseflow
estimations obtained with the different starting date is considered and
the corresponding 5%, 50% and 95% percentiles are calculated. The
reference baseflow is included between the 5% and the 95% percentiles
during 52% of the 3-year period (Fig. 2a). The relative residual between
the median SWOT-baseflow and the reference baseflow ranges between
−16% and 70%, and it is less than 10% for 70% of the time (Fig. 2b).
During the flood periods the difference between the 5% and 95% per-
centiles is larger than during low-flow (Fig. 2c), leading to larger un-
certainties.
Finally, the distributions of the reference baseflow and the SWOT
median baseflow are compared at Bazoches, Austerlitz and Poses
(Fig. 3). The empirical probability density functions and cumulative
density functions (left and mid-panels of Fig. 3) are generally similar
and the largest difference is observed for the extreme values of the
baseflow distributions. The Q-Q plots confirm this observation (right
panels of Fig. 3). In particular, at the station of Bazoches the Q-Q plot is
close to the 45° line for all discharge values. On the other hand, de-
viations are observed at Austerlitz for quantiles lower than 11% and
higher than 80%, and at Poses for quantiles lower than 22% and higher
than 72%. As already mentioned concerning the increase of the spread
of the baseflow estimates during flood periods (Fig. 2c), the filter is
more uncertain during these periods and more sensitive to the starting
date and subsequent initialization problems which is discussed in
Section 6.
4.2. SWOT sampling impact on baseflow retrieval
In order to assess the dependence on the number of SWOT ob-
servations per cycle, the algorithm is also applied at Bazoches assuming
only one observation per cycle. The results show that the difference
between the 95% and the 5% percentiles is generally higher when the
station is observed once per cycle than when it is observed three times
per cycle, especially during flood events, when the spread can double
(Fig. 2b). However, the performance remains good, as with one
48
Remote Sensing of Environment 218 (2018) 44–54
observation per cycle the relative bias is 2.8%, the Nash-Sutcliffe
coefficient is 0.89 and the RMSE is 8.5 m3 s−1, whereas it is 5.6 m3 s−1
in the case of three observations per cycle.
5. Propagation of discharge uncertainties to the baseflow
estimates
The discharge estimates that will be obtained from SWOT will be
affected by an error whose magnitude depends on the algorithm used
and on the river characteristics. Durand et al. (2016) compared five
discharge estimation algorithms and showed that at least one among
them estimates discharge with a relative RMSE less than 35% in 14 of
the 16 nonbraided rivers analyzed. However, in some cases the errors
can be significantly larger. The impact of such errors on baseflow es-
timation must be analyzed in order to really assess the possibility to
retrieve baseflow from SWOT mission. To this aim, the baseflow algo-
rithm developed in this paper is applied to discharge time series
properly corrupted to simulate SWOT derived discharge. As a first step,
such analysis is performed at the two gauging stations of Bazoches and
Poses.
Following Hagemann (2017), the corrupted discharge Q★ is ob-
tained by applying a log-normally distributed error y to the true river
discharge Q:
Q★ (t ) = y (t ) Q (t ). (3)
The mean and standard deviation of y are determined from the
discharge error estimations of Durand et al. (2016). In that paper,
discharge errors are expressed as RRMSE, MMR and SDRR, which are
the root mean square error, the mean and the standard deviation of the
discharge relative residual (RR). The latter is defined as
Q★ (t ) − Q (t )
RR (t ) = .
Q (t ) (4)
★
Definitions (3) and (4) imply that y(t) = 1 + RR(t). If Q is affected
by MRR and SDRR, then the mean of y is 1 + MRR and its standard
deviation is SDRR.
According to Durand et al. (2016), the median of the MRR among
the discharge algorithms varies between −72.9% and 104.7% de-
pending on the river. The median of the SDRR among the discharge
algorithms varies between 3.7% and 83.7%, with an exception of
197.3% for the Platte River. The medians of the MRR and of the SDRR
among all rivers and algorithms are −18.8% and 15% respectively.
5.1. The effect of a bias on the discharge relative residual
First of all, we analyze the effect of a bias on river discharge by
setting SDRR = 0 and MRR ≠ 0. In this case, if α = 1 + MRR, y(t) = α
and the corrupted discharge is given by
Q★ (t ) = αQ (t ). (5)
★
The recession coefficient k for Q is calculated as the slope of the
regression line between ln(Q★ (t )/ Q0★) and t (Section 2.1). Since
Q★(t) = αQ(t), the recession coefficient for Q★ is the same as the one for
Q.
Moreover, if Qb★ is the baseflow obtained by applying the Chapman
filter (Eq. (2)) to the corrupted discharge Q★, the following equation
can be proved by induction:
Qb★ (t ) = αQb (t ). (6)
First, at t = 0 the initial baseflow is a fraction of total discharge
according to Eq. (11): Qb★ (0) = cQ★ (0) . Substituting Eq. (5) leads to
Qb★ (0) = cαQ (0). (7)
By definition of c,
Qb (0) = cQ (0), (8)
F. Baratelli et al. Remote Sensing of Environment 218 (2018) 44–54

Fig. 3. Comparison between reference discharge distribution and SWOT discharge distribution at the gauging stations of Bazoches (a), Austerlitz (b) and Poses (c).
Left panels: empirical probability density function. Center panels: empirical cumulative distribution function. Right panels: Q-Q plots.

so that Qb★ (0) = αQb (0) . Second, assuming that Eq. (6) holds at time t, Q★ is calculated with Eq. (5) and the Chapman's filter is applied to
we compute Qb★ (t + 1) with the Chapman filter (Eq. (2)): obtain Qb★ . The results confirm that the bias of baseflow depends line-
arly on the bias of discharge and that the slope of the regression line
k 1−k ★
Qb★ (t + 1) = Qb★ (t ) + Q (t + 1). equals the baseflow index (Fig. 4). The correlation coefficient is 1 at
2−k 2−k (9)
both stations.
Given the induction hypothesis (Eq. (6)) and definition (5), Eq. (9)
reduces to 5.2. The effect of time-varying errors on the discharge relative residual
k 1−k
Qb★ (t + 1) = α ⎡ Qb (t ) + Q (t + 1)⎤ = αQb (t + 1). As a second step, the effect of time-varying errors on discharge is
⎣2 − k 2−k ⎦ (10)
analyzed (MRR = 0, SDRR ≠ 0). To this aim, we test five values of
In other words, Eq. (6) holds for t + 1. The proof by induction is SDRR, chosen in the range identified by Durand et al. (2016): (5, 10, 20,
then completed. 50, 100) %. For each value of SDRR, 1000 realizations of y are drawn
Given Eqs. (6) and (5), the ratio between the bias of baseflow and from the log-normal distribution and the corresponding Q★ and Qb★ are
the bias of discharge corresponds to the baseflow index. Moreover, if computed. For each realization, MRRb and SDRRb are computed as the
MRRb is the mean relative residual for baseflow, then MRRb=MRR. mean and standard deviation of the relative residual for baseflow. The
This theoretical result is confirmed by the results of the baseflow distributions of MRRb and SDRRb are then analyzed as a function of
algorithm at the stations of Bazoches and Poses. Indeed, 13 values of SDRR. Fig. 5 shows that at the Bazoches station the average of the
MRR are chosen in the range identified by Durand et al. (2016): MRRb distribution is negative and decreases non-linearly with SDRR.
( ± 5, ± 10, ± 20, ± 25, ± 50, ± 75,100) %. For each value of MRR, The absolute value of MRRb is less than 2% for SDRR < 20% and is

49
F. Baratelli et al.
Fig. 4. Bias of baseflow and bias of discharge for the different values of MRR
listed in Section 5.1 at the Bazoches station.
16% for SDRR = 100%. On the other hand, SDRRb is less than 11% for
SDRR < 20% (Fig. 6). For higher values of SDRR, SDRRb increases
significantly and is 53% for SDRR = 100%. Recalling that
RRMSE2 = MRR2 + SDRR2 (Durand et al., 2016), RRMSEb has the
same order of magnitude as RRMSE when the latter is 10% or smaller.
For higher values of RRMSE, RRMSEb is roughly smaller of a factor 1.8
(Table 1).
It is important to notice that high levels of SDRR imply high levels of
noise on the discharge time series, so that the calculation of the re-
cession coefficient is no more possible. Indeed, for SDRR > 50%, the
probability that the recession coefficient calculation fails is more than
70%.
Fig. 5. Distribution of the mean relative residual for baseflow (MRRb) at the Bazoches station for different discharge error levels (SDRR). The inset shows the mean
and standard deviation of the MRRb distribution as a function of SDRR.
50
Remote Sensing of Environment 218 (2018) 44–54
The results at the downstream station of Poses are not displayed as
they are very similar to those presented for Bazoches.
5.3. The cumulative effect of discharge estimation error and of discharge
sampling
Finally, the cumulative effect of three sources of errors is con-
sidered: bias (MRR ≠ 0), time-varying errors (SDRR ≠ 0) and sampling
at SWOT observation frequency. In this case, the values of the MRR and
of the SDRR are set at those found by Durand et al. (2016) for the Seine
River (MRR = −21.2%, SDRR = 6.6%, RRMSE = 22.2%). 1000 reali-
zations of y are drawn from the log-normal distribution. At the Ba-
zoches station the distribution of the MRRb has a mean of −18% and a
standard deviation of 2%, while its minimum and maximum values are
−22% and −14% respectively (Fig. 7). In absolute value, the mean
MRRb is smaller than MRR. The distribution of the SDRRb has a mean
of 11%, which is larger than SDRR. The standard deviation is 3%, the
minimum is 6% and the maximum is 18.5% (Fig. 7). RRMSEb is 21.3%,
i.e., slightly smaller than RRMSE. Similar results are observed at the
Poses station, where the mean MRRb and SDRRb are −18% and 9.8%
respectively, and RRMSEb = 20.9%. Notice that, as a result of the
combined effect of the three sources of error, the absolute value of the
mean MRRb is smaller than the one expected from the analysis in
Section 5.1, according to which MRRb equals MRR.
In summary, the analysis of the impact of SWOT discharge estima-
tions errors shows that
• if MRR ≠ 0 and SDRR = 0, the corresponding baseflow will be af-
fected by MRRb = MRR, that is, RRMSEb = RRMSE
• if MRR = 0 and SDRR ≠ 0, the corresponding baseflow will be af-
fected by both a MRRb and a SDRRb. Their combined effect is such
that RRMSEb has the same order of magnitude than RRMSE for
RRMSE ≤ 10%, whereas for higher values of RRMSE, RRMSEb is
roughly half RRMSE
• if MRR ≠ 0, SDRR ≠ 0 and SWOT observation frequency is con-
sidered, RRMSEb is slightly smaller than RRMSE, which displays
non-linear effects in the propagation of errors in the baseflow
F. Baratelli et al. Remote Sensing of Environment 218 (2018) 44–54

Fig. 6. Distribution of the standard deviation for baseflow relative residual (SDRRb) at the Bazoches station for different discharge error levels (SDRR). The inset
shows the mean and standard deviation of the SDRRb distribution as a function of SDRR.

Table 1 algorithm.
Root mean square error for relative residual of dis-
charge (RRMSE) and baseflow (RRMSEb). 6. Baseflow estimation improvement: initialization strategy
RRMSE (%) RRMSEb (%)
The 30 starting dates considered in this work are all chosen in a low-
5 7
flow period, so that baseflow can be initialized with river discharge.
10 8
20 11 However, depending on the real start of the mission and on the position
50 29 on the earth's surface, the first estimation of discharge from SWOT
100 56 observations could occur in another period than the low-flow one. In
this case, initializing with river discharge is not correct physically and
may lead to significant errors in baseflow estimation. To better illus-
trate this point, the impact of baseflow initialization is analyzed at the
Bazoches station. The baseflow algorithm is applied several times to the

Fig. 7. Distribution of the mean (MRRb) and standard deviation (SDRRb) for baseflow relative residual at the Bazoches station for a discharge estimation char-
acterized by MRR = −21.2% and SDRR = 6.6%.

51
F. Baratelli et al. Remote Sensing of Environment 218 (2018) 44–54

Fig. 8. a) Average number of days for the stabilization of baseflow to its reference value as a function of the quantile class of initial river discharge at the Bazoches
station. b) Distribution of stabilization times for all the elements of the river network.

reference discharge time series, each time with a different initialization 7. Discussion
date. 730 initialization dates are considered in order to span different
hydrological conditions. The time after which the absolute difference Results show that in the Seine river basin (75 000 km2) baseflow can
between baseflow and its reference value is less than 1% of the average be estimated with good accuracy from real discharge time series which
reference baseflow is represented as a function of the initial discharge are sampled at SWOT observation frequency. Performances are good
classed in percentiles (Fig. 8a). When the initial discharge is lower than even in reaches which will be observed by the satellite only once per
the 10% percentile, on average 40 days are necessary for the stabili- 21-day cycle. When the discharge time series are characterized both by
zation of baseflow, whereas around 74 days are necessary when the SWOT observation frequency and SWOT discharge estimation errors,
initial discharge is higher than the 90% percentile. If SWOT observa- the corresponding errors on baseflow estimations are always slightly
tions start in a high-flow period, it would be necessary to wait for the lower than the errors on discharge. These positive results support the
low-flow period in order to correctly initialize the baseflow algorithm. possibility to estimate baseflow at global scale from the SWOT mission.
As a consequence, the length of the estimated baseflow time series To this aim, the methodology should be tested on other SWOT-ob-
would be shorter than the mission duration. servable basins and, especially, to basins characterized by larger size
In order to reduce the stabilization time for high discharges, the and different geological context than the Seine river basin.
following initialization function is proposed: The prerequisite of the methodology used in this paper to estimate
baseflow is the availability of discharge time series. The baseflow al-
⎧ 1 Q (t0) ≤ Q1
gorithm is then strictly dependent on the discharge algorithms that are
⎪ Q (t0) − Q1
⎪ 1 − (1 − β ) Q1 < Q (t0) ≤ Q2 currently being developed to estimate discharge from SWOT observa-
Qb (t0) ⎪ Q2 − Q1
= tions (Garambois and Monnier, 2015; Durand et al., 2014; Gleason and
Q (t0) ⎨ Q (t0) − Q2 Smith, 2014). The applicability to truly ungauged basins depends on
⎪ β − (β − γ ) Q3 − Q2 Q2 < Q (t0) ≤ Q3
⎪ the algorithm, as some require only global limits on velocity and depth,
⎪ γ Q (t0) > Q3, (11) whereas others rely on a priori information like local discharge esti-
⎩
mates or bathymetry (Durand et al., 2016). Bias is often the biggest
where the five parameters β, γ, Q1, Q2 and Q3 are estimated by mini-
component of the error on SWOT discharge estimations (Durand et al.,
mizing the average number of days necessary for baseflow stabilization
2016). The results in this paper show that bias is also the component of
to its reference values (β = 0.6, γ = 0.4, Q1 = 1% percentile, Q2 = 2%
discharge error that most contributes to the error on baseflow. Efforts
percentile, Q3 = 85% percentile). According to Eq. (11), baseflow is
are then needed to reduce uncertainties of discharge algorithms, espe-
initialized with the total discharge only when the latter is lower than
cially in terms of bias.
Q1, otherwise it is initialized with a fraction of the total discharge.
Baseflow is estimated here by means of the Chapman filter, which is
Using the initialization function (Eq. (11)), the average stabilization
chosen because of its simplicity. Indeed, it depends on one parameter
time decreases significantly (Fig. 8a): the stabilization time decreases to
only, which can be estimated from the discharge time series itself with a
15 days when the initial discharge is lower than the 10% percentile and
recession analysis. We are aware of the discussions on the validity of the
to 45 days when the initial discharge is higher than the 90% percentile.
different kinds of filters (Zhang et al., 2017; Su et al., 2016; Stewart,
When the initialization function (Eq. (11)) with the 5 parameters op-
2015; Eckhardt, 2008). In particular, Eckhardt (2005) showed that the
timized at the Bazoches station is applied to the whole river network, a
Chapman filter implies a maximum baseflow index - the long term ratio
general decrease of the stabilization time is observed (Fig. 8b). When
between baseflow and total discharge - of 0.5, whereas for catchments
averaged over the river network, the stabilization time decreases from
characterized by perennial streams and porous aquifers, like the Seine
62 days to 43 days. An initialization function with three parameters
basin, a maximum baseflow index of 0.8 should be applied. However,
instead of five was also tested, but in this case the decrease in the
this value was derived from the analysis of a few basins only and more
stabilization time was not significant.
data need to be acquired to validate it (Eckhardt, 2008). Besides, the
For an ungauged basin, no reference discharge is available to opti-
choice of the maximum baseflow index impacts the baseflow estimates
mize the parameters of the initialization function. In this case, it is
but not the assessment of the SWOT mission potential for retrieving
necessary to estimate the convergence time of the filter. To do so, we
baseflow from its discharge estimates, which is the objective of the
recommend to start the filter with Qb = Q at various dates (for instance,
present work. The choice of the proper baseflow separation method to
every day during a month as for the Seine basin) and to estimate the
be used with SWOT data remains a challenge and requires the avail-
time needed to converge to a unique solution. This unique solution can
ability of independent baseflow estimates or measurements for valida-
then be used as the reference baseflow to refine the initialization of the
tion.
filter as mentioned above for the Seine river basin.

52
F. Baratelli et al.
Baseflow measurements could be obtained through isotope-based
hydrograph separation (Klaus and McDonnell, 2013). This method se-
parates streamflow in two or more components using a mass balance
approach and measurements of stables isotopes of water, possibly
combined with geochemical tracers (Wels et al., 1991; Hooper and
Shoemaker, 1986), radioisotopes (Martinez et al., 2015) or electrical
conductivity (Longobardi et al., 2016; Penna et al., 2014). Jasechko
et al. (2016) compiled a database of oxygen isotope ratios in rain and
streamflow for 254 watersheds around the world. These values could be
used to validate baseflow algorithms. More data could be provided by
laser spectrometers (Tweed et al., 2016; Berman et al., 2009) but al-
together a worldwide database of multiple tracers would be crucial for
the validation of hydrograph separation algorithms.
The initialization function improves the performance of the algo-
rithm significantly in estimating baseflow time fluctuations at the be-
ginning of the calculation period. In the case of the Seine basin, the five
parameters of the initialization function (Eq. (11)) are optimized at a
single river reach and then applied to the whole network, so that the
computational cost remains low. However, the proposed initialization
function needs to be tested on larger basins around the world. It is
important to notice that the availability of a reference discharge time
series can be useful to improve the initialization of the baseflow algo-
rithm. However, as discussed in Section 6, it is not necessary. Moreover,
the initial baseflow value has a negligible influence on the baseflow
index for long time series (Eckhardt, 2012). The methodology proposed
in this paper to estimate baseflow only requires SWOT discharge esti-
mations and can potentially be applied globally.
Global baseflow estimates from SWOT mission could improve the
retrieval of the different components of the global water cycle from
satellite data (Munier and Aires, 2018; Sahoo et al., 2011), leading to
the water budget closure. Moreover, global models of the water cycle,
climate models and land surface models could take into account the
groundwater component more precisely (Flipo et al., 2014; Vergnes and
Decharme, 2012) and improve their predictions using SWOT data. Fan
et al. (2013) estimated a global groundwater table depth map from
literature data and from a global groundwater model which does not
take into account water exchanges between streams and aquifers. de
Graaf et al. (2017) developed a transient global-scale groundwater
model considering stream-aquifer interactions by means of a con-
ductance model. In the calibration of groundwater models, baseflow is
the most important target (Hunt et al., 2006; Yager, 1998). This is
especially true for baseflow at the basin outlet, which depends on the
overall basin behavior. River baseflow estimated from SWOT mission
will therefore be fundamental to calibrate these models, refine global
predictions and significantly improve global hydrological analysis.
8. Conclusions
Baseflow is estimated with good accuracy from discharge time series
sampled at SWOT observation frequency in all the river reaches of the
Seine river basin which will be observed at least once per 21-day cycle.
Indeed, the relative bias with respect to the reference baseflow - ob-
tained from daily discharge - is less than 8% and its average over the
network is 1%. The temporal dynamics of baseflow is also well re-
trieved with Nash-Sutcliffe coefficients above 0.7 for 96% of the SWOT-
observable river network. These results are obtained by assuming that
discharge is not affected by estimation errors.
The impact of SWOT discharge uncertainties is assessed through
three steps. First, a daily discharge time series affected by a certain bias
determines a baseflow which is affected by the same bias. Second, when
a daily discharge time series is affected by time-varying errors and no
bias, the corresponding baseflow is affected by both bias and time-
varying errors. If the discharge error is smaller than 10%, the baseflow
error has the same order of magnitude than that on discharge. For
higher values of the discharge error, the baseflow error is roughly half
the discharge error. Finally, when the combined effect of SWOT
53
Remote Sensing of Environment 218 (2018) 44–54
discharge sampling and SWOT discharge uncertainties is considered,
the error on baseflow estimates is slightly smaller than that on dis-
charge.
This work demonstrates the feasibility of retrieving baseflow from
SWOT-derived discharge time series. The good performances obtained
even in river reaches with only one observation per cycle indicate that
the SWOT mission will be able to provide baseflow estimates with an
unprecedented global coverage.
As a perspective of this work, the methodology should be tested on
gauged basins characterized by larger size and different geological
context. Finally, it is of paramount importance to improve and develop
methodologies to measure baseflow along the river network of a basin
in order to select and validate the proper baseflow separation algo-
rithm.
Acknowledgments
This work was supported by the CNES. It is based on simulated
observation times from the future SWOT mission. The authors ac-
knowledge particularly Selma Cherchali and Nicolas Picot.
References
Alsdorf, D., Rodríguez, E., Lettermaier, D., 2007. Measuring surface water from space.
Rev. Geophys. 45, RG2002. https://doi.org/10.1029/2006RG000197.
Altenau, E.H., Pavelsky, T.M., Moller, D., Lion, C., Pitcher, L.H., Allen, G.H., Bates, P.D.,
Calmant, S., Durand, M., Smith, L.C., 2017. AirSWOT measurements of river water
surface elevation and slope: Tanana River, AK. Geophys. Res. Lett. 44, 181–189.
https://doi.org/10.1002/2016GL071577.
Baratelli, F., Flipo, N., Moatar, F., 2016. Estimation of distributed stream-aquifer ex-
changes at the regional scale using a distributed model: sensitivity to in-stream water
level fluctuations, riverbed elevation and roughness. J. Hydrol. 542, 686–703.
https://doi.org/10.1016/j.jhydrol.2016.09.041.
Berman, E.S.F., Gupta, M., Gabrielli, C., Garland, T., McDonnell, J., 2009. High-frequency
field-deployable isotope analyzer for hydrological applications. Water Resour. Res.
45, W10201. https://doi.org/10.1029/2009WR008265.
Biancamaria, S., Durand, M., Andreadis, K., Bates, P., Boone, A., Mognard, N., Rodríguez,
E., Alsdorf, D., Lettenmaier, D., Clark, E., 2011. Assimilation of virtual wide swath
altimetry to improve Arctic river modeling. Remote Sens. Environ. 115, 373–381.
Biancamaria, S., Lettenmaier, D., Pavelsky, T., 2016. The SWOT mission and its cap-
abilities for land hydrology. Surv. Geophys. 37, 307–337. https://doi.org/10.1007/
s10712-015-9346-y.
Billen, G., Garnier, J., Mouchel, J.-M., Silvestre, M., 2007. The Seine system: introduction
to a multidisciplinary approach of the functioning of a regional river system. Sci.
Total Environ. 375, 1–12.
Brodie, R., Sundaram, B., Tottenham, R., Hostetler, S., 2007. An Overview of Tools for
Assessing Groundwater-Surface Water Connectivity. Bureau of Rural Sciences,
Canberra, pp. 131.
Brodie, R.S., Hostetler, S., 2005. A review of techniques for analysing baseflow from
stream hydrographs. In: Proceedings of the NZHS-IAH-NZSSS 2005 conference. New
Zealand Hydrological Society, Auckland New Zealand.
Chapman, T., 1991. Comment on “Evaluation of automated techniques for base flow and
recession analyses” by R.J. Nathan and T.A. McMahon. Water Resour. Res. 27,
1783–1784.
Chapman, T., 1999. A comparison of algorithms for stream flow recession and baseflow
separation. Hydrol. Process. 13, 701–714.
Chapman, T., Maxwell, A.I., 1996. Baseflow separation - comparison of numerical
methods with tracer experiments. In: Hydrology and Water Resources Symposium.
Institution of Engineers Australia, Hobart, pp. 539–545.
Chow, V.T., 1959. Open Channel Hydraulics. McGraw Hill Company Inc., New York.
Cunge, J., 1969. Au sujet d’une méthode de calcul de propagation de crues (méthode
muskingum). J. Res. in Hydrol. 7, 205–230 (In French).
de Graaf, I.E.M., van Beek, L.P.H., Gleeson, T., Moosdorf, N., Schmitz, O., Sutanudjaja,
E.H., Bierkens, M.F.P., 2017. A global-scale two-layer transient groundwater model:
development and application to groundwater depletion. Adv. Water Resour. 102,
53–67.
de Marsily, G., 1986. Quantitative Hydrogeology - Groundwater Hydrology for Engineers.
Academic Press, London.
de Marsily, G., Ledoux, E., Levassor, A., Poitrinal, D., Salem, A., 1978. Modelling of large
multilayered aquifer systems: theory and applications. J. Hydrol. 36, 1–34.
Dugdale, S., Bergeron, N., St-Hilaire, A., 2013. Temporal variability of thermal refuges
and water temperature patterns in an Atlantic salmon river. Remote Sens. Environ.
136, 358–373. https://doi.org/10.1016/j.rse.2013.05.018.
Durand, M., Gleason, C., Garambois, P., Bjerklie, D., Smith, L., Roux, H., Rodriguez, E.,
Bates, P., Pavelsky, T., Monnier, J., Chen, X., Di Baldassarre, G., Fiset, J.M., Flipo, N.,
Frasson, R., Fulton, J., Goutal, N., Hossain, F., Humphries, E., Minear, J., Mukolwe,
M., Neal, J., Ricci, S., Sanders, B., Schumann, G., Schubert, J., Vilmin, L., 2016. An
intercomparison of remote sensing river discharge estimation algorithms from mea-
surements of river height, width, and slope. Water Resour. Res. 52, 4527–4549.
F. Baratelli et al.
https://doi.org/10.1002/2015WR018434.
Durand, M., Neal, J., Rodrìguez, E., Andreadis, K.M., Smith, L.C., Yoon, Y., 2014.
Estimating reach-averaged discharge for the River Severn from measurements of river
water surface elevation and slope. J. Hydrol. 511, 92–104.
Eckhardt, K., 2005. How to construct recursive digital filters for baseflow separation.
Hydrol. Process. 19, 507–515. https://doi.org/10.1002/hyp.5675.
Eckhardt, K., 2012. Technical note: analytical sensitivity analysis of a two parameter
recursive digital baseflow separation filter. Hydrol. Earth Syst. Sci. 16, 451–455.
Eckhardt, N., 2008. A comparison of baseflow indices, which were calculated with seven
different baseflow separation methods. J. Hydrol. 352, 168–173.
Fan, Y., Li, H., Miguez-Macho, G., 2013. Global patterns of groundwater table depth.
Science 339, 940–943. https://doi.org/10.1126/science.1229881.
Flipo, N., Even, S., Poulin, M., Théry, S., Ledoux, E., 2007a. Modelling nitrate fluxes at the
catchment scale using the integrated tool CAWAQS. Sci. Total Environ. 375, 69–79.
https://doi.org/10.1016/j.scitotenv.2006.12.016.
Flipo, N., Jeannée, N., Poulin, M., Even, S., Ledoux, E., 2007b. Assessment of nitrate
pollution in the Grand Morin aquifers (France): combined use of geostatistics and
physically-based modeling. Environ. Pollut. 146, 241–256. https://doi.org/10.1016/
j.envpol.2006.03.056.
Flipo, N., Monteil, C., Poulin, M., de Fouquet, C., Krimissa, M., 2012. Hybrid fitting of a
hydrosystem model: long term insight into the Beauce aquifer functioning (France).
Water Resour. Res. 48, W05509. https://doi.org/10.1029/2011WR011092.
Flipo, N., Mouhri, A., Labarthe, B., Biancamaria, S., Rivière, A., Weill, P., 2014.
Continental hydrosystem modelling: the concept of nested stream-aquifer interfaces.
Hydrol. Earth Syst. Sci. 18, 3121–3149 10-5194/hess-18-3121-2014.
Fluet-Chouinard, E., Lehner, B., Rebelo, L.M., Papa, F., Hamilton, S.K., 2015.
Development of a global inundation map at high spatial resolution from topographic
downscaling of a coarse-scale remote sensing data. Remote Sens. Environ. 158,
348–361.
Garambois, P., Calmant, S., Roux, H., Paris, A., Monnier, J., Finaud-Guyot, P., Samine
Montazem, A., Santos da Silva, J., 2017. Hydraulic visibility: using satellite altimetry
to parameterize a hydraulic model of an ungauged reach of a braided river. Hydrol.
Process. 31, 756–767.
Garambois, P., Monnier, J., 2015. Inference of effective river properties from remotely
sensed observations of water surface. Adv. Water Resour. 79, 103–120.
Gleason, C.J., Smith, L.C., 2014. Toward global mapping of river discharge using satellite
images and at-many-stations hydraulic geometry. Proc. Natl. Acad. Sci. 111,
4788–4791.
Hagemann, M., 2017. BAM: Bayesian AMHG-Manning inference of discharge using re-
motely sensed stream width, slope, and height. Water Resour. Res. 53, 9692–9707.
Hooper, R., Shoemaker, C., 1986. A comparison of chemical and isotopic hydrograph
separation. Water Resour. Res. 22, 1444–1454 0043-1397/86/005W-4289505.
Howard, J., Merrifield, M., 2010. Mapping groundwater dependent ecosystems in
California. PLoS ONE 5. https://doi.org/10.1371/journal.pone.0011249.
Hunt, R., Strand, M., Walker, J., 2006. Measuring groundwater-surface water interaction
and its effect on wetland stream benthic productivity, Trout Lake watershed,
Northern Wisconsin, USA. J. Hydrol. 320, 370–384. https://doi.org/10.1016/j.
jhydrol.2005.07.029.
Jasechko, S., Kirchner, J., Welker, J., McDonnel, J., 2016. Substantial proportion of global
streamflow less than three months old. Nat. Geosci. 9, 126–129. https://doi.org/10.
1038/ngeo2636.
Klaus, J., McDonnell, J., 2013. Hydrograph separation using stable isotopes: review and
evaluation. J. Hydrol. 505, 47–64.
Labarthe, B., Pryet, A., Saleh, F., Akopian, M., Flipo, N., 2015. Distributed simulation of
daily stream-aquifer exchanged fluxes in the Seine river basin at regional scale. In:
Lollino, G., Arrattano, M., Rinaldi, M., Giustolisi, O., Marechal, J.C., Grant, G.E.
(Eds.), Engineering Geology for Society and Territory. vol. 3. Springer, pp. 261–265.
https://doi.org/10.1007/978-3-319-09054-2_54.
Ledoux, E., 1975. Programme NEWSAM: principe et notice d’emploi. In: Technical
Report. Centre d’Informatique Géologique, Ecole Nationale Supérieure des Mines de
Paris (In Frence).
Longobardi, A., Villani, P., Guida, D., Cuomo, A., 2016. Hydro-geo-chemical streamflow
analysis as a support for digital hydrograph filtering in a small, rainfall dominated,
sandstone watershed. J. Hydrol. 539, 177–187.
Lyne, V., Hollick, M., 1979. Stochastic time variable rainfall-runoff modelling. In:
Proceedings of the Hydrology and Water Resources Symposium, Perth, 10–12
September,1979. Institution of Engineers National Conference Publication, pp.
89–92.
Marmonier, P., Archambaud, G., Belaidi, N., Bougon, N., Breil, P., Chauvet, E., Claret, C.,
Cornut, J., Datry, T., Dole-Olivier, M., Dumont, B., Flipo, N., Foulquier, A., Gérino,
M., Guilpart, A., Julien, F., Maazouzi, C., Martin, D., Mermillod-Blondin, F.,
Montuelle, B., Namour, P., Navel, S., Ombredane, D., Pelte, T., Piscart, C., Pusch, M.,
Stroffek, S., Robertson, A., Sanchez-Pérez, J., Sauvage, S., Taleb, A., Wantz, M.,
Vervier, P., 2012. The role of organisms in hyporheic processes: gaps in current
knowledge, needs for future research and applications. Ann. Limnol. Int. J. Limnol.
48, 253–266.
Martinez, J.L., Raiber, M., Cox, M.E., 2015. Assessment of groundwater-surface water
interaction using long-term hydrochemical data and isotope hydrology: headwaters
of the Condamine River, Southeast Queensland, Australia. Sci. Total Environ. 536,
499–516.
54
Remote Sensing of Environment 218 (2018) 44–54
Mason, D.C., Trigg, M., Garcia-Pintado, J., Cloke, H.L., Neal, J.C., Bates, P.D., 2016.
Improving the TanDEM-X digital elevation model for flood modelling using flood
extents from synthetic aperture radar images. Remote Sens. Environ. 173, 15–28.
Massei, N., Laignel, B., Deloffre, J., Mesquita, J., Motelay, A., Lafite, R., Durand, A., 2010.
Long-term hydrological changes of the Seine river flow (France) and their relation to
the North Atlantic Oscillation over the period 1950–2008. Int. J. Climatol. 30,
2146–2154. https://doi.org/10.1002/joc.2022.
Miller, M.P., Buto, S.G., Susong, D.D., Rumsey, C.A., 2016. The importance of base flow in
sustaining surface water flow in the Upper Colorado River Basin. Water Resour. Res.
52, 3547–3562. https://doi.org/10.1002/2015WR017963.
Munier, S., Aires, F., 2018. A new global method of satellite dataset merging and quality
characterization constrained by the terrestrial water budget. Remote Sens. Environ.
205, 119–130.
Nash, J., Sutcliffe, J., 1970. River flow forecasting through conceptual models. Part I, a
discussion of principles. J. Hydrol. 10, 282–290.
Nathan, R., McMahon, T., 1990. Evaluation of automated techniques for base flow and
recession analyses. Water Resour. Res. 26, 1465–1473.
Newcomer, M., Hubbard, S., Fleckenstein, J., Maier, U., Schmidt, C., Thullner, M., Ulrich,
C., Flipo, N., Rubin, Y., 2018. Influence of hydrological perturbations and riverbed
sediment characteristics on hyporheic zone respiration of CO2 and N2. J. Geophys.
Res. Biogeosci. 123, 1–21. https://doi.org/10.1002/2017JG004090.
Paiva, R.C.D., Durand, M.T., Hossain, F., 2015. Spatiotemporal interpolation of discharge
across a river network by using synthetic SWOT satellite data. Water Resour. Res. 51,
430–449. https://doi.org/10.1002/2014WR015618.
Papa, F., Biancamaria, S., Lion, C., Rossow, W.B., 2012. Uncertainties in mean river
discharge estimates associated with satellite altimeter temporal sampling intervals: a
case study for the annual peak flow in the context of the future SWOT hydrology
mission. IEEE Geosci. Remote Sens. Lett. 9, 569–573.
Partington, D., Brunner, P., Simmons, C., Werner, A., Therrien, R., Maier, H., Dandy, G.,
2012. Evaluation of outputs from automated baseflow separation methods against
simulated baseflow from a physically based, surface water-groundwater flow model.
J. Hydrol. 459, 28–39.
Pavelsky, T.M., Durand, M.T., Andreadis, K.M., Beighley, R.E., Paiva, R.C.D., Allen, G.H.,
Miller, Z.F., 2014. Assessing the potential global extent of SWOT river discharge
observations. J. Hydrol. 519, 1516–1525.
Pekel, J.F., Cottam, A., Gorelick, N., Belward, A.S., 2016. High-resolution mapping of
global surface water and its long-term changes. Nature 540, 418–422.
Penna, D., Engel, M., Mao, L., Dell’Agnese, A., Bertoldi, G., Comiti, F., 2014. Tracer-based
analysis of spatial and temporal variations of water sources in a glacierized catch-
ment. J. N. Am. Benthol. Soc. 18, 5271–5288. https://doi.org/10.5194/hess-18-
5271-2014.
Pryet, A., Labarthe, B., Saleh, F., Akopian, M., Flipo, N., 2015. Reporting of stream-
aquifer flow distribution at the regional scale with a distributed process-based model.
Water Resour. Manag. 29, 139–159. https://doi.org/10.1007/s11269-014-0832-7.
Rodríguez, E., 2016. Surface water and ocean topography mission (SWOT) project - sci-
ence requirements document. In: Technical Report D-61923. NASA/JPL.
Sahoo, A., Pan, M., Troy, T., Vinukollu, R., Sheffield, J., Wood, E., 2011. Reconciling the
global terrestrial water budget using satellite remote sensing. Remote Sens. Environ.
115, 1850–1865. https://doi.org/10.1016/j.rse.2011.03.009.
Sichangi, A.W., Wang, L., Yang, K., Chen, D., Wang, Z., Li, X., Zhou, J., Liu, W., Kuria, D.,
2016. Estimating continental river basin discharges using multiple remote sensing
data sets. Remote Sens. Environ. 179, 36–53.
Smakhtin, V.U., 2001. Low flow hydrology: a review. J. Hydrol. 240, 147–186.
Stewart, M.K., 2015. Promising new baseflow separation and recession analysis methods
applied to streamflow at Glendhu Catchment, New Zealand. Hydrol. Earth Syst. Sci.
19, 2587–2603.
Su, C.H., Costelloe, J.F., Paterson, T.J., Western, A.W., 2016. On the structural limitations
of recursive digital filters for base flow estimation. Water Resour. Res. 52,
4745–4764.
Tarpanelli, A., Brocca, L., Lacava, T., Melone, F., Moramarco, T., Faruolo, M., Pergola, N.,
Tramutoli, V., 2013. Toward the estimation of river discharge variations using
MODIS data in ungauged basins. Remote Sens. Environ. 136, 47–55.
Tweed, S., Munksgaard, N., Marc, V., Rockett, N., Brass, A., Forsythe, A., Bird, M.,
Leblanc, M., 2016. Continuous monitoring of stream stable isotope and stormflow
hydrograph separation using laser spectrometry in an agricultural catchment. Hydrol.
Process. 30, 648–660. https://doi.org/10.1002/hyp10689.
Vergnes, J.P., Decharme, B., 2012. A simple groundwater scheme in the TRIP river
routing model: global off-line evaluation against GRACE terrestrial water storage
estimates and observed river discharges. Hydrol. Earth Syst. Sci. 16, 3889–3908.
https://doi.org/10.5194/hess-16-3889-2012.
Wawrzyniak, V., Piégay, H., Allemand, P., Vaudor, L., Goma, R., Grandjean, P., 2016.
Effects of geomorphology and groundwater level on the spatio-temporal variability of
riverine cold water patches assessed using thermal infrared (TIR) remote sensing.
Remote Sens. Environ. 175, 337–348.
Wels, C., Cornett, R.J., Lazerte, B.D., 1991. Hydrograph separation: a comparison of
geochemical and isotopic tracers. J. Hydrol. 122, 253–274.
Yager, R.M., 1998. Deftecting influential observations in nonlinear regression modelling
of groundwater flow. Water Resour. Res. 34, 1623–1633.
Zhang, J., Zhang, Y., Song, J., Cheng, L., 2017. Evaluating relative merits of four baseflow
separation methods in Eastern Australia. J. Hydrol. 549, 252–263.