Inland valleys are seasonally waterlogged headwater wetlands, widespread
across western Africa. Their role in the hydrological cycle in the humid,
hard-rock-dominated Sudanian savanna is not yet well understood. Thus, while
in the region recurrent floods are a major issue, and hydropower has been
recognized as an important development pathway, the scientific community
lacks precise knowledge of streamflow (
It is widely recognized that wetlands exert a strong influence on the hydrological cycle. However, they hardly support generalization in their functioning, according to their different types and settings, and the limited availability of hydrological observations (Bullock and Acreman, 2003). In sub-Saharan western Africa, “Bas-fonds”, or inland valleys, are seasonally waterlogged wetlands located at the headwater of streams and cover 22–52 Mha (Andriesse et al., 1994). They can be found on the whole continent, although they have different names, such as dambos, mbuga, inland valleys, or valley bottoms. Their role in the hydrological cycle is still unclear and lacks significant attention. This is particularly true for the humid Sudanian zone, mostly located on hard-rock lithology with rather shallow water tables and low-capacity aquifers (MacDonald et al., 2012) where all components of the critical zone are tightly linked (e.g. tree root systems are likely to be connected to the water tables; Mamadou et al., 2016; Richard et al., 2013). There, for instance, the role of inland valleys in streamflow generation is still poorly understood, while they are suspected of providing most of the observed streamflow as baseflow through exfiltration of perched seasonal water tables (Séguis et al., 2011).
Streamflow generation mechanisms deserve more attention, as (1) the region
suffers from severe floods, which affect an increasing number of people (Di
Baldassarre et al., 2010), and (2) hydropower development is also
ongoing (Cervigni et al., 2015) and
is recognized to be a fundamental large-scale source of energy for the
development of sub-Saharan Africa
(
Moreover, the strong western African demographic rates (2.6 % on average in the period 2000–2010) imply rapid changes in land use and land cover (LULC) (FAO, 2012). For instance, about 42 % (or 8 %) of forested land (or non-forested vegetation) was converted to agricultural land (which amounts to an increase of more than 11 %) in the Sudanian savanna between 1975 and 2000 (Eva et al., 2006). This is also likely to affect the functioning of inland valleys, although it is still unknown to what extent.
Despite the possible role of inland valleys in the hydrological cycle, they carry an important agronomic potential currently underexploited (Katic et al., 2014; Rodenburg et al., 2014). Due to their high water content, they can be cultivated offseason, and thus they buffer the high climatic variability of the area (Nyamadzawo et al., 2015) as well as human pressure on uphill plots located closer to the villages (Lounang Tchatchouang et al., 2014). However, inland valley exploitation limits should be carefully studied, as overexploitation may have consequences for these specific ecosystems (Böhme et al., 2016; Brouwer et al., 2014; Wood, 2006). Rodenburg et al. (2014) estimate that production derived from less than 10 % of the total inland valley area over Africa could meet the total current demand for rice in Africa. In these studies, little is mentioned about water exploitation sustainability. This is due to the heterogeneity of inland valleys and the difficulty of capturing fundamental hydrological processes to be generalized. This again calls for documenting water redistribution processes within inland valleys, so as to include water use sustainability in the process of inland valley selection and cultivation (Böhme et al., 2016; Schmitter et al., 2015).
Headwater inland valleys are characterized by a rapid decrease in permeability with depth due to the accumulation of clays, which is ultimately associated with a low-permeability layer, allowing the formation of perched water tables in the valley thalweg (Blavet, 1997; Brabant, 1991; Hector et al., 2015; von der Heyden and New, 2003; Von Der Heyden, 2004). The hydrology of inland valleys is also controlled by classical catchment features like vegetation distribution and the functioning of contributive areas (Von Der Heyden, 2004). For instance, concerning vegetation distribution, Balek (2006) reports on increased flood flow when trees were removed from two of their observed catchments. The fast LULC changes occurring today call for the scientific community to clearly understand the role of different vegetation distributions and the impact of their changes on the hydrology of inland valleys. These interactions between flow generation and vegetation require study of the hydrological functioning of inland valleys in an integrated, critical zone (CZ)-like approach. Concerning contributive areas, the dominance of baseflow in the hydrology of inland valleys calls for investigation of the properties of subsurface regions prone to contributing to streamflow through water exfiltration in the valley thalweg.
As a complement of CZ observations, physically based CZ models are appropriate tools to study the functioning of a hydrosystem (e.g. Kollet and Maxwell 2008; Srivastava et al. 2014). These models explicitly link all hydrological compartments of the CZ, from the deep impermeable bedrock to the top of the canopy, and provide a robust framework to conduct virtual experiments (Weiler and McDonnell, 2004).
In this paper, we seek to address the following two questions: (1) what are the main characteristics of the hydrological functioning of inland valleys in the Sudanian savanna of western Africa? (2) What is the impact of LULC changes on such systems? The approach undertaken here is to build a set of virtual experiments in an idealized elementary V-shaped catchment to test the sensitivity of a physically based CZ model to different features: (a) the presence of the low-permeability layer in the valley thalweg, which characterizes inland valleys, (b) the pedology of contributive areas and (c) the vegetation distribution (from natural woodland cover to herbaceous-like fallow/crop cover). This idealized elementary V-shaped catchment is built based on the main features of elementary catchments comprising an inland valley (topography, clay lens, vegetation distribution), but also following a highly instrumented elementary catchment for which we previously built a conceptual lithological/pedological model (Hector et al., 2015).
In the first section, we briefly discuss the physical environment and how we
model it over a 7-year period using a physically based CZ model. Then, we
present the results of a single simulation, which are compared to
observations from the elementary headwater catchment, available within the
AMMA-CATCH observation service (
In this section, we briefly describe our conceptual representation of the
main soil and vegetation characteristics found in the hard-rock basement
areas of the Sudanian climatic region (yearly precipitation amount between
700 and 1400 mm) dominated by metamorphic settings. This eco-physical model
is built following the literature of the region and on inland valleys and the
work of Hector et al. (2015), who studied and described the functioning of an
elementary catchment in northern Benin, the Nalohou catchment
(0.16 km
Our conceptual model of the physical environment is made up of three layers through depth (H1: 0–0.5, H2: 0.5–2, H3: 2–10.5 m, Fig. 1b–c), to which we prescribed specific hydrodynamic properties in a deterministic way and following the literature, but is also based on numerous field and laboratory measurements by Hector (2014), Richard (2014) and Robert (2012) in the same area (Table 1). In the following text, we refer to sand-like and clay-like hydrodynamic properties in a broad sense: sand-like relates to highly permeable/low-retention media and clay-like relates to low-permeable/high-retention media. The reader may refer to Tables 1 and 3 for actual property ranges and values.
The thicknesses of the two shallowest soil layers (H1 and H2) are well defined from observations in the Nalohou catchment, and seem in accordance with other locations in (sub)tropical humid regions (de Condappa et al., 2008). The thickness of the deeper layer (H3) is that of the regolith, located above the fissured basement and the hard-rock basement, following Lachassagne et al. (2011).
In the tropical-humid Sudanian climate, topsoils are usually permeable, due to microfauna bioturbation or soil work from farmers (Giertz and Diekkrüger, 2003; Richard, 2012; Robert, 2012). As a consequence, the infiltration-excess overland flow is very limited in the area (e.g. Giertz et al., 2006; Masiyandima et al., 2003). The topsoil layer, H1, is therefore permeable, with sand-like soil hydrodynamic properties. Some authors note a significant relationship between saturated hydraulic conductivity and land cover in the area (Giertz et al., 2005; Giertz and Diekkrüger, 2003; Richard, 2014), but in a limited range (less than 1 order of magnitude). The H1 layer is thus considered laterally homogeneous in this study.
The second layer (H2) is usually more variable laterally, and variations in clay content in this potential illuviation layer (a clay-accumulation horizon, as described by de Condappa et al., 2008) may generate perched water tables (e.g. Giertz et al., 2006). In this layer, some macroporous hardpan units with high permeability may also be found (Bonsor et al., 2014; Hector et al., 2015).
It is beyond the scope of this study to analyse the effect of this layer (H2) on the hydrological processes. There is an extensive literature using the same critical zone modelling framework that provides insights into the effect of lateral variability of soil properties through dedicated sensitivity studies (Atchley and Maxwell, 2011; Meyerhoff and Maxwell, 2011; Meyerhoff et al., 2014; Gilbert et al., 2016). We therefore chose to consider a rather sandy, permeable layer to mimic a hardpan unit behaviour (as part of the polycyclic soil profile described by Faure and Volkoff, 1998), and favour quick lateral subsurface transfer, which has been observed in the area (e.g. Hector et al., 2015; Masiyandima et al., 2003 and Séguis et al., 2011).
The composition of the third layer (H3) is variable laterally. On the one hand, there is a low-permeability, clayey horizon, similar to what is observed by Hector et al. (2015) and Wubda et al. (2017) in the western half of the Nalohou catchment, and also described as a major feature from larger-scale observations by Faure and Volkoff (1998). On the other hand, there is a clay-eluviated, saprolite-like, sandy horizon, which is expected to produce a strongly contrasted behaviour. It is also similar to what is observed in the eastern half of the Nalohou catchment, and is the second major feature from larger-scale observations by Faure and Volkoff (1998).
The inland valley is located in the downstream part of the valley thalweg (circles of Fig. 1a). It has a specific H1/H2/H3 sequence. It usually consists of sandy upper horizons. H1 and H2 layers are the same as elsewhere, and are considered to be of a sandy nature and to favour quick vertical and lateral transfers, as observed in the field. At about 2 m depth, in the upper part of the H3 layer just below the H2 horizon, is a low-permeability, clayey layer, which actually characterizes the inland valley. This allows the formation of perched water tables during the rainy season (Masiyandima et al., 2003). Such a shallow drainage system is the main contributor to baseflow, which represents roughly 80 % of total streamflow for such a catchment (Giertz et al., 2006; Hector et al., 2015; Kamagaté et al., 2007; Séguis et al., 2011). This clay-rich accumulation horizon is often found above a poorly weathered horizon (Brabant, 1991), as is the case in the Nalohou catchment. Therefore the H3 layer underneath the inland valley is made of a low-retention, high-permeability fissured basement (in the sense of Lachassagne et al., 2011).
In a thesis work, Hector (2014) summarized the variability of field measurements to build up classes of main soil/lithology units for the three main layers identified in this study (Table 1). Hydrodynamic parameters (following Van Genuchten, 1980) for this study are shown below in Table 3. Saturated conductivities of saprolite and fissured basement in H3 are given by the range of observations derived from both pumping tests and magnetic resonance soundings in the area (Vouillamoz et al., 2014).
Ranges of hydrodynamic properties in the area, following Van Genuchten (1980). From Hector (2014).
At the Upper Ouémé catchment scale (14 300 km
Non-disturbed natural vegetation in this area is mostly woodland, composed of
trees with a herbaceous layer underneath. Under human influences, trees are
usually cut in favour of agricultural plots. However, trees in the riparian
areas and the inland valley fringes are generally kept alive, as well as
specific species on the hillslopes for consumption purposes (agroforestry
system). This is why we consider in this study two vegetation classes: trees
and herbaceous. While the trees class refers to a fully tree-covered state
with some herbaceous layer underneath, the herbaceous cover refers to a mixed
fallow-field landscape with sparse trees, composed of either cultivated plots
or young fallows under the influence of farmers or with some isolated trees
(with height
Tree phenology mostly consists in defoliation around January, which occurs asynchronously between different species and different individuals of the same species (Seghieri et al., 2012). On the other hand, in the so-called herbaceous cover, the ground is commonly covered by scattered shrubs, while the actual herbaceous layer usually starts growing in April after the first rains and reaches its maximum height (up to 2.5 m) in October. At this time its height exceeds that of shrubs. Then, the herbaceous layer stops transpiring and dries up from October to January. Controlled fire is a traditional practice in Benin; the vegetation (mainly the herbaceous layer) is burnt between November and February. Hence, during the dry season, the soil is bare for about 5 months. Leaf area index (LAI) and stem area index of trees and herbaceous classes are later shown in Fig. 6a.
The CZ model ParFlow-CLM (PFCLM) is composed of ParFlow, a three-dimensional
integrated physical model that simultaneously solves variably saturated flow
and surface outflow coupled with CLM, a land surface model which solves both
the water and energy budget at the surface. ParFlow uses a free-surface
overland flow boundary condition which routes ponded water as overland flow
using the kinematic wave equation and Manning equation (Kollet and Maxwell,
2006). In this study, the topographic effect is taken into account by a
terrain-following grid (Maxwell, 2013) and only requires two-dimensional
surface slopes to be provided. ParFlow solves the Richards equation using
cell-centered finite differences in space and an implicit backwards Euler
scheme in time. The Newton–Krylov method is used together with a multigrid
preconditioner and shows excellent parallel scaling, which allows the
simulation of large-scale domains (i.e. continental US) at high resolution
(i.e.
We model an idealized elementary headwater catchment, comprising an inland
valley, of 720 m
Model set-up
The vertical resolution varies with depth following a geometric progression of common ratio 1.2 for the first 13 cells (which are coupled to CLM), ranging between 0.134 and 1.195 m, and 4 more 1 m thick cells, uncoupled to CLM, are added at the bottom. The domain is therefore 10.5 m thick. Selected hydrodynamic parameters based on expertise-driven sensitivity studies within the range of observed values (Table 1) are shown in Table 3.
Hydrodynamic properties for the virtual V-shaped catchment, following Van Genuchten (1980).
No-flow boundary conditions are applied to the catchment edges (symmetric assumption) and bottom. The only way for the water to leave the domain, apart from evapotranspiration, is through overland flow at the catchment outlet.
Forcing data are obtained from the Benin site of the AMMA-CATCH observatory
(AMMA-CATCH, 1996) and are available on the data portal
(
Western African rainfall follows the monsoon dynamics and is characterized by a strong variability at interannual scales (see e.g. Lawin (2007) and Le Lay and Galle (2005) for detailed analyses of the Sudanian rainfall regime). In the simulated years, yearly precipitation amounts ranged from 913 to 1572 mm (Fig. 2).
Figure 2 shows the meteorological variables used to force PFCLM. They have
been described in Mamadou et al. (2014) for the year 2008 and recently
compared to data from a site 13 km away over a forested cover for the period
2008–2010 (Mamadou et al., 2016). The daily meteorological conditions of the
two sites were expectedly close. Minimal values of incoming short-wave
radiations (SW
PFCLM is forced at a 30 mn time step, over the 7-year period 2006–2012, with spatially homogeneous forcings. Initial conditions are set as a water table depth of 3 m, in accordance with several water table measurements. The model is spun up for 1 year, which corresponds to the year 2007, a rather average year.
Model forcing. Upper panel: daily precipitation. SW
In this study, we forced the model with observation-based time series of LAI, SAI (stem area index), displacement heights and roughness lengths for the two vegetation classes of our domain (trees and herbaceous).
Vegetation height was monitored over a herbaceous-covered site (Nalohou) and
a forest site 13 km away and further used to calibrate a simple API
(Antecedent Precipitation Index)-like model which takes precipitation as
input (Kohler and Linsley, 1951), and with the assumption that vegetation
growth follows soil moisture availability. This allowed derivation of
vegetation heights during the growing period at a daily time step for
herbaceous cover for years where only precipitation data were available.
Roughness lengths and displacement heights have been derived from these time
series following the Brutsaert approach (Brutsaert, 1982) calibrated on in
situ eddy covariance
measurements. The LAI time series is a combination of satellite LAI products
(CYCLOPE, MODIS, SEVIRI), constrained by in situ measurements derived from
hemispherical photographs based on the method proposed by Weiss et
al. (2004), on both the forested and herbaceous-covered sites. Herbaceous SAI
is also calculated from LAI using the equation
(LAI
For the herbaceous cover, the root distribution in the model follows that of
Zeng (2001), which is an exponential shape for the cumulative root fraction
(
As a first step, we evaluate the relevance of the model with respect to real-world data (water table depth measurements, soil moisture, streamflow records, evapotranspiration and gravity data). This section describes the evaluation data obtained from the Nalohou catchment, whose size and structure are comparable to that of our virtual V-shaped catchment (Fig. 1a). The Nalohou catchment comprises a small inland valley and is considered to be representative of a cultivated site. It has been the focus of previous studies (Descloitres et al., 2011; Guyot et al., 2012; Hector et al., 2015; Richard et al., 2013). Land surface elevation ranges from 436 to 460 m.a.s.l., and topographic slopes are gentle (mean N–S and E–W slopes correspond to those imposed on the virtual V-shaped catchment).
Observations in P1 and P2 (Fig. 1a) were used to evaluate the simulated groundwater dynamics. At the inland valley location in P1 there is a shallow, 1 m deep piezometer which aims at documenting the seasonal perched water table and a deeper (10 m) piezometer monitoring the permanent water table.
Streamflow measurements have been described in AMMA-CATCH (2008). A Parshall
flume was built in 2004 at the outlet of the inland valley (Q2 in Fig. 1a)
and recorded any level change above 10
Two eddy covariance flux towers have been installed, respectively, on the Nalohou site (where the flux tower footprint is representative of cultivated, herbaceous-like areas) and on the woodland site (Mamadou et al., 2016). Furthermore, an infrared large-aperture scintillometer (LAS) was deployed from 2006 to 2008 on the Nalohou site (Guyot et al., 2009, 2012). LAS measurements give access to an average sensible heat flux over the footprint area and, by closing the energy budget, this allows us to derive the average latent heat flux (i.e. evapotranspiration). On the Nalohou site, the beam path was 2.4 km long, and comprised many more trees than in the flux tower footprint area.
Time-domain reflectometry (TDR) probes have been set up on the eastern bank
of the Nalohou catchment at about 150 m eastward of latitude point
Gravity instrumentation originates from the GHYRAF (Gravity and Hydrology in
Africa) project, which sought to study gravity to water storage change (WSC)
relationships (Hinderer et al., 2009, 2012). Gravity instruments were an
absolute (AG hereafter) FG5 gravimeter (Hector et al., 2013), a relative
superconducting gravimeter (Hector et al., 2014, SG hereafter) and a CG5
relative portable spring gravimeter (Hector et al., 2015). Relative spring
microgravimeters measure differences in gravity between stations, while other
instruments such as SG or AG measure changes at a specific site. Hybrid
gravimetry is the joint use of gravity surveys using relative
microgravimeters and monitoring of temporal changes at a base station, using
an SG and/or AG (Hinderer et al., 2016). From the hybrid gravity data set of
Nalohou, Hector et al. (2015) produced a map of the seasonal amplitude of
gravity changes (their Fig. 11). In this study, we convert gravity changes to
water storage changes using a simple scalar admittance (gravity changes to
WSC ratio) of 0.04
WSC analysis is conducted using empirical orthogonal function (EOF)
decomposition. EOF decompositions can be seen as principal component analysis
applied to the time series of a spatial field. The method produces spatial
patterns of variability (in terms of variance), their associated time
variation together with a measure of the weight (importance) of each pattern
(in terms of explained variance). In other words, the whole data set of
coordinates (
To measure the goodness-of-fit between simulations and observations, we use
the Kling–Gupta efficiency (KGE) criterion (Gupta et al., 2009):
We model an elementary headwater catchment and test its sensitivity to (1) the presence of a low-permeability layer underneath an inland valley, (2) the vegetation distribution, and (3) the hydrodynamic properties of contributive areas (H3 layer). To do so, we build a set of virtual experiments that are compared to a benchmark simulation (named the “reference” case hereinafter, Table 4). This reference simulation includes an inland valley characterized by an impervious layer made up of a clay lens underlying an upper sandy layer in the valley thalweg. Vegetation distribution and H3 layer are both chosen arbitrarily in this simulation. An anthropized vegetation distribution (mostly herbaceous but also with trees in the riparian areas) allows comparison of model outputs with observed hydrological variables at the Nalohou catchment. A clay-like contributive area (H3 outside the inland valley cells) is chosen between the two main pedological features (clay and saprolite). The three virtual experiments are as follows (Table 4).
The first experiment explores the impacts of the low-permeability layer in the valley thalweg on the hydrological functioning of the catchment. This experiment consists in removing the impervious layer below the H2 horizon (upper part of H3 as shown in Fig. 1b) in the valley thalweg with respect to the reference case. Other parameters (hydrodynamic, vegetation) are kept identical.
The second experiment explores the impact of the vegetation cover with three cases: the “natural” state where the catchment is fully covered with trees, and the “anthropized” state with two different vegetation covers: (1) either the catchment is covered with grass, except in the riparian and inland valley fringe areas where trees remain (as in Fig. 1a – this is the reference case), or (2) the catchment is fully covered with grass (no trees).
Virtual experiments. “Trees and herbaceous” refers to trees only in riparian areas and inland valley fringes, as shown in Fig. 1.
In the third experiment, we test three different configurations for the H3 layer outside the valley thalweg, with H1 and H2 left unchanged, as the contributive area within H3 drives the subsurface and groundwater transfer toward the inland valley. The reference case considers clay material only in the H3 layer, and the two other cases consider, respectively, (1) a sandy to loamy saprolite in the H3 layer and (2) a clay zone on the western side of the virtual catchment and a sandy zone on the eastern side (asymmetric case), both above the hardrock considered to be impervious. These configurations are based on two distinct soil profiles commonly found in the Sudanian landscape (Faure and Volkoff, 1998).
Water storage changes (WSCs) are shown in Fig. 3 as the spatial and temporal patterns of the first mode of the EOF decomposition. While the left panel shows the simulated WSCs, the right panel shows WSCs obtained from the hybrid gravity data by interpolation from point-based values (blue dots). The first mode of the simulated WSCs represents more than 90 % of the variance. This means that more than 90 % of the total signal variance is contained in the first EOF spatial pattern multiplied by its associated time series (Fig. 3c, red line). The time series shows that this mode corresponds to the seasonal cycle (the common part of the signal which explains most of the total variance). In that sense, we can directly interpret the spatial patterns (i.e. the EOFs) as maps of seasonal water storage amplitudes. The first mode of gravity-derived WSCs also explains a significant part (79 %) of the total variance. As this map derives from an interpolated field, and although gravity data are spatially integrative, observed geometries should not be overinterpreted. Mode 2 (not shown) accounts for about 8 % of the total variance, and contains interannual storage variations together with some residuals at the seasonal level.
Empirical orthogonal function decomposition of water storage changes
(WSC).
The reference simulation includes an inland valley (a low-permeability layer in the upper part of the H3 horizon) and the clayey material in the H3 layer outside the inland valley corresponds to what is found on the western bank of the Nalohou catchment. In this regard, the simulated spatial pattern matches two important features of the observations: a high amplitude in the inland valley and a low amplitude in the clayey area. The high simulated amplitude in the downstream section (500 to 700 m) is due to tree withdrawals during the dry season. Gravity data do not cover this part of the catchment, which prevents the comparison with the simulated data. The low amplitude on the sides of the catchment is due to the impermeable clay in the H3 horizon, which prevents water from percolating downward and favours lateral transfer to the valley thalweg where water accumulates during the wet season, and drains down toward the riparian areas during the dry season. High amplitude is simulated along the whole valley thalweg and not only in the inland valley, as the observation shows.
Figure 4a shows simulated time series of saturation profiles at P1, in the inland valley for the reference simulation. Red lines show the simulated water tables, while white lines show the observed water tables. The model is able to reproduce seasonal fluctuations of both the permanent water table and the temporary perched water table. Although numerous discrepancies exist (e.g. the seasonal amplitudes of the permanent water table), mainly because of the horizontal and vertical variability of hydrodynamic parameters which have been considered constant in layer H3, the first order of this complex behaviour is reproduced. In particular, the interannual behaviour of the permanent water table is rather satisfying, despite a slight phase shift, but higher-frequency responses of the shallow perched water table at the core of the rainy season are also well reproduced.
Figure 4b shows simulated time series of saturation profiles at P2, in the upstream area for the saprolite simulation (sandy H3 layer everywhere but in the inland valley). This sandy H3 layer is a common feature of the environment, particularly on the eastern bank of the Nalohou catchment. The simulated permanent water table matches satisfyingly the observation representative of this sandy H3 unit.
Simulated saturation and observed and simulated water table.
Figure 5 shows simulated and observed streamflow expressed in mm h
Observed and simulated streamflow for the wet seasons of the period 2008 to 2012 at station Q2.
Figure 6a shows the LAI and SAI forcing time series used in this study for the two vegetation covers. Figure 6b (or Fig. 6d) shows the simulated daily evapotranspiration spatially averaged over herbaceous-covered cells out of the thalweg in light blue line for the reference simulation (or the saprolite simulation). The shaded area gives the whole spatial range of daily values between minimal (lower black line) and maximal (upper black line) spatial evapotranspiration. Figure 6c (or Fig. 6e) is similar, but for tree-covered cells. Reference (clay in the H3 unit everywhere but below the inland valley) and saprolite simulations are compared because the Nalohou flux tower is located on a saprolite-like material. The saprolite simulation actually matches more closely the observed evapotranspiration in the dry season.
During the dry season, trees renew their leaves and show a low LAI and limited evapotranspiration. The herbaceous cover have been burnt, and only some scattered shrubs and trees imply a non-zero LAI and little evapotranspiration. During the transition period between dry and wet seasons, tree LAI increases, herbaceous grows and evapotranspiration increases with both soil moisture (not shown) and plant ability to transpire with a higher LAI. During the wet season, the evapotranspiration is never water limited (Mamadou et al., 2016). During the transition period between wet and dry seasons, the herbaceous layer dries up, and LAI is replaced by SAI, which diminishes drastically the evapotranspiration. Herbaceous are eventually burnt at the end of this transition period (SAI drop in Fig. 6a), while trees start renewing their leaves.
Overall, the spatially averaged simulated evapotranspiration matches the observations, over both herbaceous and tree covers. At the end of the wet season, the simulated evapotranspiration (dominated by transpiration) is a little higher than the EC observations, but not higher than the LAS observations in 2008. EC is known to underestimate evapotranspiration (Foken, 2008).
Because soil moisture probes are located on the eastern side of the Nalohou catchment, observed soil moisture is here compared to the “saprolite” simulation where H3 outside the inland valley is composed of sandy-like saprolite (Fig. 7). Simulations match the observed temporal patterns in a satisfying way, including recession behaviours.
Observed and simulated soil moisture for the saprolite simulation at location P2.
The water budget simulated for the reference case is shown in Table 5. As
expected, evapotranspiration dominates the water budget (from 58 % to
83 % of the yearly precipitation amount) but shows a marked interannual
variability as simulated in the area at basin scale (Cornelissen et al.,
2013; Séguis et al., 2011) and from a review of different other spatial
scales (Hector, 2014). For dry years, the ET part (or the
Water budgets for the reference case. Mean and SDs are calculated on the period 2007–2012.
The most important variations in yearly water budgets occur when comparing a tree-covered catchment and a herbaceous-covered catchment (Table 6). Mean interannual evapotranspiration is higher (or lower) and streamflow is lower (or higher) for the trees (or herbaceous)-covered catchment, while the reference case shows a somewhat intermediate behaviour. Interannual variability (i.e. as shown by the SD) is lower for the trees than for the herbaceous cover. Unlike other cases, interannual water storage changes may become significant in the tree-covered experiment, with a net decrease in water storage during the dry years. This results in a higher interannual SD of the storage term.
The presence of the impermeable layer (in the reference case) has very little effect on the water budget, lowering the streamflow by 2 % (9 mm) on average, in favour of evapotranspiration, with respect to the case without the impermeable layer.
The reference case (clayey) evapotranspirates slightly more than the saprolite-like case, 5 % on average, while the mixed simulation (clayey left bank, saprolite right bank) shows an intermediate behaviour.
Interannual mean and SD of simulated evapotranspiration (ET),
streamflow (
Figure 8 shows water storage changes as the first modes of the EOF decompositions for all tested cases. In all cases, the first mode represents more than 90 % of the variance.
The most striking discrepancies arise when comparing the results related to the contributive area experiments (Fig. 8a, e, f). While the reference case (clayey H3 layer, Fig. 8a) shows low seasonal amplitude in the valley banks and higher amplitude along the valley axis including the inland valley, the saprolite case (Fig. 8e) shows a rather opposite behaviour with higher amplitude on the banks and lower amplitude in the valley axis. In the reference case, the southern, upstream edge of the inland valley exhibits a break in the seasonal amplitude, while this is not visible in the saprolite case (in this case there is no specific imprint of the inland valley). In the mixed case (Fig. 8f), the southern, upstream edge of the inland valley also exhibits a different seasonal amplitude, but of opposite behaviour, as in the reference case.
In the vegetation experiment (Fig. 8a, c, d), the tree-covered catchment (Fig. 8c) shows higher seasonal amplitude than the herbaceous-covered catchment (Fig. 8d), while the reference case shows patterns closer to the herbaceous-covered catchment. In the time series, seasonal WSC increase occurs earlier on herbaceous cover than on trees.
The absence of the impermeable layer in the inland valley increases the seasonal water storage amplitude along the thalweg (Fig. 8b).
In all cases, there is low seasonal amplitude at the catchment outlet (the northern blue patch). This is the imprint of the topography (all cells converge to that point) and the model boundary condition: the only way for the water to exit the domain, apart from evapotranspiration, is through this outlet.
The mixed simulation characteristics are actually the closest to the conceptual model presented by Hector et al. (2015). And the observed spatial pattern (Fig. 2) matches several features of the mixed simulation: a high amplitude in the inland valley, a low amplitude on the clayey, left bank, area, and some intermediate amplitude in the saprolite-like right bank area. While this is satisfying, it has to be noted that the conceptual model proposed by Hector et al. (2015) lacks some details. In the mixed simulation (which follow their descriptive model), while streamflow, evapotranspiration and spatial pattern of WSC are well reproduced, the simulated permanent water table completely fills up the inland valley due to high permeability in the H3 layer on the right bank (see further), which is not observed. Hence there must be some barrier to the lateral flow feeding the inland valley. However, it would be interesting to investigate the existence of other inland valleys which do not show such disconnexion between the upper perched water table and lower permanent water table. Our study would anticipate that, all scaling issues discarded, such an inland valley would produce a similar water budget.
Simulated water storage changes: first mode of the EOF for all cases: upper panel: spatial patterns. Lower panel: temporal patterns.
Yearly simulated streamflow time series (sampled on cell Q1, Fig. 1) are shown in Fig. 9 for the year 2011 and as semilog plots, with the reference time series repeated in all plots in black. Overall, streamflow simulations are rather similar. Many of the differences among the cases are related to the first part of the rainy season, in the timing of the streamflow onset and the streamflow amplitude during this period (this is also true for other years, not shown). Differences in the dry season are very low in terms of water amount (cf. the log scale), but exhibit clearly different behaviours: some cases are able to sustain a low baseflow during the dry season, while some others, including the reference case, prevent water flow during the dry season.
The presence of the impermeable layer in the inland valley (reference case) induces a higher streamflow in August (Fig. 9a). On the other hand, the absence of the impermeable layer favours infiltration and maintains a low baseflow longer in the dry season, fed by the permanent water table which is then connected all across the catchment.
The tree-covered catchment (Fig. 9b) tends to exhibit lower
streamflow in August. The storm peaks are much attenuated, as compared to
the reference case. The herbaceous-covered catchment shows a higher
streamflow in the dry season and the early wet season, with slightly higher
storm peaks than in the reference case. One may note how the reference
The simulation with a saprolite-like contributive area (Fig. 9c) shows a higher dry season and early wet season streamflow as compared to the reference case and the mixed case. At the core of the rainy season, in August–September, the saprolite simulation shows a lower streamflow.
Simulated streamflow of year 2011 for the three virtual experiments.
The reference run is always displayed as a black line. In the lower panel,
the clay
Figure 10 shows the simulated water tables at two locations, in the inland valley (P1, Fig. 1), and in the upstream contributive area (P2, Fig. 1). When present, perched water tables are shown too.
As expected, the presence of the impermeable layer (Fig. 10a, as in the reference case) creates two water tables: a deep permanent, low-amplitude, water table and a shallow seasonal water table within the sandy layer of the inland valley (around P1; see Fig. 4). This is not seen when the impervious layer is removed (no I. valley experiment), and where only a strong-amplitude permanent water table is present. In the upstream contributive area (around P2, Fig. 10d), no difference is observed.
All vegetation experiments (Fig. 10b, P1) simulate a perched water table in the inland valley. The tree-covered catchment shows a deeper water table than the herbaceous-covered catchment, while the reference simulation shows an intermediate behaviour. In the upstream contributive area (Fig. 10e, P2), the herbaceous cover matches the reference simulation. The strong water table drawdown induced by tree water uptake lowers the water table down within the clay layer, and no more seasonal variations are observed. In fact, the simulation did not even reach an equilibrium state at that point.
When the whole catchment or half of the catchment deeper layer (H3) is saprolite (Fig. 10c), the fissured basement below the inland valley (in H3) fills up and no perched water table is observed. In the upstream contributive area (Fig. 10f, P2), however, there is a strong seasonal water table amplitude (as also shown in Fig. 4) not observed in the clayey, reference simulation.
Simulated water table in the valley thalweg
The presence of the impermeable layer in the inland valley does not influence
the yearly water budget much (1 % – or 2 % – change in the
interannual mean for ET – or
In a similar context, Giertz and Diekkrüger (2003) observed a
significantly higher total runoff amount and peak flow and a longer runoff
period in a catchment containing a larger inland valley area (catchment size:
3.5 km
The vegetation distribution does influence the yearly water budget. For evapotranspiration, for instance, average yearly amount varies between 74 % and 62 % of the precipitation for, respectively, a fully tree-covered and a herbaceous-covered catchment (68 % for the reference case). The relative difference between trees and herbaceous simulations in average yearly amount is 16 % (with respect to the tree simulation), with a larger difference for dry (24 % for 2011) than for wet (12 % for 2010) years. More importantly, the interannual variability (estimated by the SD) is strongly reduced (26 %) for a tree-covered catchment (53 mm) as compared to a herbaceous-covered catchment (72 mm).
Evapotranspiration is higher for the tree-covered catchment, mostly because trees are able to get water from the ground for a longer period than herbaceous cover during the dry season. This is because trees have a longer non-zero LAI period, as they take a short time to renew their leaves and are not impacted by the provoked fires as the herbaceous cover is. They also have access to deeper water thanks to the deeper root system. Finally, trees undergo higher interception and subsequent evaporation thanks to their low-angle leaf orientation and higher LAI and SAI, specified in CLM. Between a herbaceous-covered catchment and a tree-covered catchment, yearly average interception losses increase from 3.2 % to 6.5 % of annual precipitation. Herbaceous cover dries up at the end of the rainy season, and is further burnt by local farmers. When transpiration ceases, evaporation is also limited due to the lack of shallow soil water.
The consequence of a higher evapotranspiration for the tree-covered catchment is a low yearly streamflow budget of 28 % on average as compared to a herbaceous-covered catchment, ranging from 20 % for the wettest year (2010) to 47 % for a dry year (2011). This occurs because during the dry season, when there is no or negligible streamflow, the trees lower the total water storage of the catchment, as shown by the lower simulated water tables (Fig. 10b, e) and the higher seasonal water storage amplitudes (Fig. 8a, c). In the early rainy season, in the tree-covered simulation, rainwater takes a longer time to infiltrate and sufficiently recharge the water table until streamflow finally reaches the streamflow level of the herbaceous-covered simulation (Fig. 9b). This is particularly visible in dry years (i.e. 2011), where highly intermittent rainfall and long dry periods during the rainy season reduce percolation and further delay the “saturated” condition of the catchment.
These results must be tempered by the fact that we used a highly idealized
“herbaceous cover”, which is, in the actual catchment, a mixture of
different vegetation classes (perennial shrubs, isolated trees and perennial
and annual grass species, mainly
As a consequence, tree covers buffer the hydrological cycle by (1) lowering
streamflow, (2) reducing interannual variability of evapotranspiration,
(3) increasing interannual water storage change by decreasing the water
storage during the dry season and (4) increasing interception losses. In the
Sudanian savanna, streamflow has decreased since the 80s, following
precipitation decreases, sometimes by twice as much (Descroix et al., 2009).
LULC changes occurred simultaneously, mostly by conversion of savanna or
woodlands to crops (e.g. Eva et al., 2006). The question of the attribution
of the hydrological changes to either climate forcings or LULC changes has
not yet been solved. While it is likely that LULC changes have played a minor
role since the 80s (because of the observed streamflow decrease), if the
still ongoing deforestation maintains its pace, there is a chance that the
hydrological cycle could be significantly altered in the future due to LULC
changes. For instance, Yira et al. (2016) used a process-based hydrological
model to show that converting savanna to crops in the last 25 years in a
Sudanian catchment in Burkina Faso has led to
When sand-like transmissive hydrodynamic properties are set for lateral and upstream areas of the inland valley (the saprolite case, as opposed to the clay-like reference case), the simulation allows the permanent water table to fill up the area below the inland valley (Fig. 10c). This means that simulated seasonal water storage amplitudes are lower in the valley thalweg than in the valley sides (Fig. 8e), showing the opposite pattern to the reference case, where water storage amplitudes are stronger in the valley thalweg (Fig. 8a). Despite these strong spatial differences, the annual water budget exhibits little differences.
In terms of streamflow generation, the subsurface behaviour of the inland valley is rather similar in the saprolite case and the reference case, as whether the shallow perched water table is connected or not to the deeper permanent water table does not matter in its ability to supply the streamflow. In both cases, there is a sufficiently high permeability layer (H2 hardpan layer in the reference simulation, and H3 in the saprolite-like simulation) which allows lateral transfer to feed the inland valley.
However, although the annual water budgets show little differences, there may still be significant differences in the temporal distribution of fluxes. For instance, the saprolite case sustains a higher baseflow (Fig. 9c). Also, the simulated evaporation differences during the dry season, over herbaceous cover (see Fig. 6d), imply different surface feedbacks on the atmosphere, which may control larger-scale atmospheric features. Evapotranspiration is known to be linked to soil properties through soil moisture availability. Descloitres et al. (2011) showed in the area that evapotranspiration as measured by a large-aperture scintillometer was controlled by soil properties. To check the consistency of these simulated evaporation differences during the dry season, evapotranspiration should be monitored on a hardpan–clay soil profile, as they are widely present in the region. Furthermore, one should note that Fig. 4a and b show results from different simulations (reference and saprolite, respectively), although both associated water table levels are real-world data of the same catchment. This means that some features are still missing in the simplified V-shaped model in order to comprehensively reproduce the behaviours of this specific catchment. If a high permeability is set to the valley banks (as in the saprolite simulation), it will result in the filling of most of the valley thalweg by the permanent water table and the absence of the perched water table, which is not observed. On the contrary, in the low-permeability scenario (reference simulation), the simulated water table on the valley banks mismatches the observed data. To successfully match all the observations, lateral variations in the permeability are needed, but this was not the target of this study, which was instead the extraction of the main features and behaviours of inland valleys.
The goal of the paper is to derive the controlling factors of inland valley
critical zone systems – through virtual experiments – among factors that
are either susceptible to evolving (land cover), that characterize inland
valleys (thalweg clay lens), or that are still largely unknown (subsurface
lithology). In this respect, slopes and topographic effects are not unknowns
or susceptible to evolving much (as land cover which has been observed to
change significantly, and which is also projected to do so) and are not
targets of this study. However, as topography drives lateral flow, catchment
slopes may significantly impact the water budgets. We ran two extra
experiments, using the reference case and multiplying the slopes by a factor
of 2 and a factor of 0.5 to extract the effect of extreme topography ranges
found in this environment (Runge, 1991). Table 7 compares the results from
these experiments to the reference case and shows that there is weak impact
on the interannual variability (SD) but a marked impact on yearly averages.
The impact of slopes is comparable to the impact of vegetation distribution
(in Table 6: trees case
Interannual mean and SD of simulated evapotranspiration (ET),
streamflow (
In this paper, we studied the hydrological functioning of Sudanian inland valleys and their sensitivity to land cover and their main features (pedology of contributive areas, clay lens), through deterministic sensitivity experiments using a physically based critical zone (CZ) model applied on a virtual generic catchment which comprises an inland valley. This is a first approach to try to understand what controls and explains the behaviour of an inland valley.
We first showed that a CZ model could be used to reproduce the complexity of
a small hydrosystem (0.16 km
Using numerical experiments comprising different virtual configurations of the catchment, we found that yearly water budgets were almost insensitive to the presence (or absence) of the low-permeability layer commonly found below the inland valley, although intra-seasonal variations in water storage and water table showed marked differences. Early season streamflow was slightly delayed when the low-permeability layer of the inland valley was virtually removed, which corresponds to the time required for water to percolate below the inland valley. Higher water table levels were obtained when infiltration below the inland valley was enhanced, resulting in a low but permanent baseflow during the dry season, which could have a significant impact downstream. This virtual experiment confirms that the observed streamflow dynamics, particularly in the dry season, is partly due to the low-permeability layer, a typical inland valley feature due to the accumulation of clays (Blavet, 1997; Brabant, 1991; Von Der Heyden, 2004).
On the other hand, yearly water budgets were much more sensitive to the vegetation distribution: average yearly ET for a tree-covered catchment (944 mm) exceeds that of herbaceous cover (791 mm). ET differences between the two covers vary between 12 % and 24 % of the annual precipitation for the wettest and driest years, respectively. Therefore, the tree-covered catchment produces a yearly streamflow 28 % lower on average than that of a herbaceous-covered catchment, ranging from 20 % for the wettest year to 47 % for a dry year. Trees also buffer interannual variability in ET by 26 %. Extensive LULC changes in the last 30 years, resulting in forest and woodland conversion to agricultural lands, still undergoing today, call for closer attention to the distinct roles of climate changes and land cover changes in the resulting hydrological functioning of Sudanian systems.
We also found little sensitivity of the yearly water budget to the hydrodynamic properties of upstream and lateral areas, although they implied completely opposite behaviours in the intraseasonal space–time patterns of water storages and marked changes in water table depths together with different dynamics and amplitudes of fluxes (e.g. streamflow, soil evaporation). Higher water table levels in the valley thalweg, induced by more permeable materials, foster a low but permanent dry season baseflow.
To summarize, it is shown that while vegetation cover significantly changes water budgets, pedological features have much more impacts on intraseasonal hydrological processes (sustained/unsustained baseflow in the dry season, catchment water storage redistribution) than on yearly water budget amounts. Both factors may therefore potentially impact water-dependent ecosystems but also water use as hydropower generation or irrigation from surface water bodies.
This study was limited to an elementary headwater catchment, and results derived here may not apply at larger scales like basins with larger inland valley / catchment area ratios, or different soil profile distributions. Larger-scale models should be run to investigate these effects, and thus large-scale spatial features should be mapped, which poses significant issues regarding subsurface structure. Indeed, the in situ measurements of the underground structure are missing and the structure is not directly available through remote sensing products for instance.
Forcing data (Meteorological: AMMA-CATCH, 2005a, and precipitation: AMMA-CATCH, 1999) are available in the AMMA CATCH Database, and simulation configuration files as well as simulation outputs are available upon request.
The critical zone is by essence a complex and interdisciplinary object: all co-authors brought their field and disciplinary expertise to this work (data collection, analyses discussions). Particularly, JMC, LS and BH designed the experiments that BH performed.
The authors declare that they have no conflict of interest.
The authors would like to thank many people who helped in acquiring, maintaining and processing data and instruments in Djougou: Sarè Tahirou, Frederic Littel, Jean-Daniel Bernard, Bernard Luck, Nicolas LeMoigne, Idrissou Imorou, Emile Pagou, Théo Ouani, Simon Afouda, Maxime Wubda. We would also like to thank the project partners who consented to the use of their infrastructures and provided valuable information and advices: Arnaud Zannou, Jean-Claude Gbodogbé (Direction Générale de l'Eau, Cotonou, Bénin) and Nicaise Yalo (University of Abomey-Calavi, Bénin).
The AMMA-CATCH regional observing system (which is part of the OZCAR research
infrastructure in critical zones) was set up thanks to an incentive funding
of the French Ministry of Research that allowed pooling together of various
pre-existing small-scale observing set-ups. The continuity and long-term
perennity of the measurements are made possible by continuous IRD funding
since 1990 and by continuous CNRS-INSU funding since 2005. This work has been
supported by a grant from Labex OSUG@2020 (Investissements d'avenir – ANR10
LABX56). Most of the computations presented in this paper were performed
using the Froggy platform of the CIMENT infrastructure
(