A comparison of hydrological models for assessing the impact of land use and climate change on discharge in a tropical catchment

Journal of Hydrology 498 (2013) 221–236

Contents lists available at SciVerse ScienceDirect

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

A comparison of hydrological models for assessing the impact of land use and climate change on discharge in a tropical catchment Thomas Cornelissen ⇑, Bernd Diekkrüger, Simone Giertz Department of Geography, University of Bonn, 53115 Bonn, Germany

a r t i c l e

i n f o

Article history: Received 30 March 2012 Received in revised form 24 May 2013 Accepted 10 June 2013 Available online 21 June 2013 This manuscript was handled by Andras Bardossy, Editor-in-Chief, with the assistance of Axel Bronstert, Associate Editor Keywords: Model type intercomparison Model uncertainty Discharge prediction Climate change scenarios Land use scenarios

s u m m a r y This study assesses the suitability of different model types for simulating scenarios of future discharge behaviour in a West African catchment (2344 km2) in the context of climate and land use change. The comparison of models enables the identification of possible sources of uncertainty in hydrological modelling of a tropical catchment. All models were calibrated and validated for the period from 1998 to 2005 with reasonable quality. The simulation of climate and land use change impacts on discharge behaviour results in substantial differences caused by model structure and calibration strategy. The semi-distributed conceptual model UHP-HRU is shown to be the most suitable for the simulation of current discharge dynamics because the simulated runoff components most closely match the current perception of hydrological processes based on field data interpretation. In addition, the model does not introduce new uncertainties into the simulation by imposing high data demands. All models simulate an increase in surface runoff due to land use change. The application of climate change scenarios resulted in considerable variation between the models and points not only to uncertainties in climate change scenarios but also gives an idea of the possible range of future developments. Overall, this study indicates that the major weakness of all hydrological models is their poor representation of the catchment’s soil characteristics and flow processes. Ó 2013 Elsevier B.V. All rights reserved.

1. Introduction West Africa and, in particular, the Republic of Benin suffer from a large discrepancy between the amount of renewable water resources and the availability of water for domestic, agricultural and industrial purposes. This discrepancy is the result of physical, socio-economic and institutional constraints (Hadjer et al., 2010). The physical constraints arise from the seasonality of the discharge, a result of the yearly shift between the dry and wet seasons, whereas the socio-economic and institutional limitations imply an infrastructure that cannot make sufficient use of the amount of available freshwater. Apart from these limitations, climate and land use change already have a large impact on the hydrological cycle in West African countries (Giertz et al., 2010; Kasei, 2010; Götzinger, 2007; Jung, 2006; Busche et al., 2005). In view of these challenges and the importance of reliable data on water availability, a proper estimation of the yearly water balance requires the application of hydrological models. ⇑ Corresponding author. Address: Department of Geography, University of Bonn, 53115 Bonn, Meckenheimer Allee 172, Germany. Tel.: +49 (0)228 732401; fax: +49 (0)228 735393. E-mail addresses: [email protected] (T. Cornelissen), [email protected] (B. Diekkrüger), [email protected] (S. Giertz). 0022-1694/$ - see front matter Ó 2013 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.jhydrol.2013.06.016

For more than a decade, a large number of modelling approaches have been applied to tropical catchments. Examples are the study by Andersen et al. (2001) on the Senegal River Basin, the study by Güntner (2002) on a catchment in north-eastern Brazil and the study by Leemhuis et al. (2007) on a catchment in Indonesia. The number of modelling efforts has increased rapidly since 2004, with studies undertaken in the White Volta Basin (Kasei, 2010; Jung, 2006; Wagner et al., 2006; Ajayi, 2004) and the Ouémé Catchment (Benin) (e.g., Götzinger, 2007) and its subcatchments, including the Donga Catchment (Séguis et al., 2011; Le Lay et al., 2008; Varado et al., 2006) and the Aguima Catchment (Bormann et al., 2005; Giertz et al., 2005), as well as studies that compare simulation results for different catchments (Giertz et al., 2010; Hiepe and Diekkrüger, 2007; Bormann and Diekkrüger, 2004). The simulation results of the previously cited studies show that the calculated fraction of each discharge component depends on the model type applied. For example, Hiepe and Diekkrüger (2007) use a time-continuous but semi-distributed model and find that base flow and surface runoff are the dominant discharge components in the Térou Catchment, a tributary of the Ouémé River. In contrast, Giertz et al. (2010) find that interflow is the dominant discharge component in the Térou Catchment by applying a conceptual model. This assumption is supported by electrical

222

T. Cornelissen et al. / Journal of Hydrology 498 (2013) 221–236

conductivity measurements conducted by Giertz (2004) and Hiepe (2008) and hydrochemical measurements conducted by Fass (2004), who find that fast runoff components are predominant. If hydrological models are applied within a single study, model calibration often results in reliable simulations of the past; however, the influence of model choice and model calibration on the simulation of climate and land use change impacts remains unclear, even if uncertainties (e.g., soil (Bossa et al., 2012) and climate (Varado et al., 2006)) are considered. From a theoretical viewpoint, a physical model represents the underlying hydrologic and land surface processes in greater detail than conceptual or statistical models (Beven, 2001). However, more parameters and greater calibration effort are required as the degree of physical representation of relevant processes in a model increases. Despite the potential of comparing the results of different models and model types, this method has not yet been widely used in hydrology. For example, Bormann et al. (2009), Huisman et al. (2009) and Viney et al. (2009) apply a multi-model modelling approach to assess the impacts of land use change on the hydrology of a catchment in Germany. To date, no comparative study of model types has been performed for the Térou Catchment, although time-continuous, semi-distributed and conceptually or physically based model types have all been applied to this catchment (Giertz et al., 2010; Hiepe and Diekkrüger, 2007; Sintondji, 2005; Busche et al., 2005; Bormann and Diekkrüger, 2004). The aforementioned studies heavily differed in the simulated fraction of interflow, varying from <3% (e.g., Busche et al., 2005) to 60% (Giertz et al., 2010). Thus, the application of these models in impact studies may yield completely different scenario projections. To differentiate between the effects caused by model choice and the effects caused by climate and land use change impacts, different model types, i.e., physically based and conceptual, are used in this study to improve the understanding of hydrological processes and to provide new insights into the influence of land use and climate change on discharge behaviour. This paper addresses the uncertainty in discharge modelling of a tropical catchment by comparing simulation results based on different model types. Furthermore, the paper aims to compare the predictions of future discharge obtained by applying land use and climate change scenarios. Towards these aims, four different models were applied to simulate the past (1998–2005) and future scenarios (2001–2049).

2. Research area 2.1. Geographical overview The Republic of Benin is located in West Africa and extends from the Gulf of Guinea to 12°300 north and from 0°450 to 4° east. With an area of 112,622 km2, Benin is one of the smaller African countries. The Ouémé Catchment drains the major part of Benin. The research area, i.e., the Térou Catchment (2344 km2), is a subcatchment of the Ouémé Catchment (location and major characteristics shown in Fig. 1). Benin has a diurnal climate, with a mean annual temperature of 27.2 °C, ranging between 21.9 °C and 32.6 °C (Ermert and Brücher, 2008). There is a strong precipitation gradient of 300 mm/a, with rainfall increasing from northern Benin, which receives 1008 mm/a (period 1961–1990) in a unimodal distribution, to southern Benin, which receives 1309 mm/a in a bimodal distribution (Ermert and Brücher, 2008). Rainfall is primarily generated by squall lines and, to a minor degree, heavy thunderstorms, which form in hot monsoon air masses (Fink et al., 2010).

Fink et al. (2010) note that the high interannual and decadal variability in rainfall intensities and the number of rainy days in Benin can result in severe droughts, such as the one that occurred in the early 1970s and mid-1980s, during which rainfall decreased to a minimum of 800 mm/a. In the Térou catchment, the precipitation distribution is unimodal, with the rainy season occurring between the beginning of May and the end of October and the dry season occurring between November and the end of April. Mean annual rainfall rates vary around 1152 mm, with a maximum rainfall rate of 260 mm in September and a minimum rate of 0 mm in December (Ermert and Brücher, 2008). Due to millennia-long land use, Benin’s potential natural vegetation, i.e., the tropical dry forest, has been replaced by savannahtype vegetation (Anhuf and Frankenberg, 1991). The commonly used classification of Benin’s vegetation is based on physiognomic characteristics according to the rules of the 1956 Yangambi Conference (Aubréville, 1957). Based on these rules, the Térou Catchment is primarily covered by savannah vegetation types, whereas 20% of the area is covered by light and dense dry forest. Agriculture only represents 11% of the total land coverage (Judex, 2008). The land surface of the study area, termed peneplain, was formed by repeating cycles of the pedimentation process (Runge, 1990; Rohdenburg, 1969) creating a specific slope sequence. The upper parts of a slope are covered by iron crusts embedded in stone lines. These structures are layers with an average thickness of 40 cm (Faust, 1991) consisting of coarse material, including angular and curved blocks of quartz (Runge, 1990). The stone lines cover the saprolith, which is a layer a few metres in depth that contains weathered bedrock. The saprolith substrate is characterised by its high clay content. In the main portion of the slope, the stone lines are overlaid by hillwash sediment. The sediment was formed by bioturbation as a consequence of high termite activity and denudative processes. Due to the translocation of clay, the hillwash sediment is characterised by the soil texture ‘‘loamy sand’’. The lower slope contains a river bed formed by recent fluvial erosion processes (Runge, 1990; Rohdenburg, 1969). Recent pedogenetic processes are dominated by lessivation (Junge and Skowronek, 2007) and the translocation of iron (Faust, 1991), resulting in small-scale differences in hydrologic processes. 2.2. Land use and climate change Judex (2008) compares the land use classification for the years 1991 and 2000 for the Upper Ouémé catchment. During this period, the cultivated area expanded by 70%, and the settlement area expanded by 0.8%. In the northern part of the Térou Catchment, the cultivated area increased by 61–100%, whereas it only increased by 21–40% in the southern part. This expansion is the result of migration, agro-colonisation and the vegetation dynamics induced by cultivation practices. According to Paeth et al. (2008), the future climate of Benin will be characterised by increasing mean annual temperatures (up to 4 °C) and further drying (at least a 25% reduction in mean annual rainfall) until 2050. Christoph et al. (2010) explain the reduction in rainfall by the weakening of the hydrological cycle, especially the recycling of rainfall. They add that the onset of the rainy season will be delayed by more than 10 days by 2025 and that heavy rainfall events will decrease. 2.3. Current perception of hydrological processes The current perception of the dominant hydrological processes is primarily based on the results of Bormann et al. (2005), Giertz (2004) and Fass (2004). The brief description given here does not include the effect of small-scale inland valleys, locally termed

T. Cornelissen et al. / Journal of Hydrology 498 (2013) 221–236

223

Fig. 1. Overview of the Ouémé and Térou Catchments: (a) location of Benin in Africa; (b) location of the Ouémé Catchment and the Upper Ouémé Catchment, the administrative units of Benin and the location of Benin’s capital Porto Novo; (c) location of the Térou Catchment, its relief and its major cities.

bas-fonds. These depressions influence discharge on a local scale but cannot be considered at the regional scale analysed in this study. The influence of the soil characteristics on the hydrological processes is shown in Fig. 2 (refer to next page). Hortonian runoff occurs on iron crusts and cultivated areas but is highly unlikely in catchments with savannah-type vegetation. This pattern is due to the high infiltration rates of savannah soils, usually more than 1000 mm per hour (Giertz et al., 2010) due to biological activity. Agricultural land use reduces the secondary pore system significantly, resulting in increased surface runoff (Germer et al., 2010). The analysis of hydrological processes by Bormann et al. (2005) depicts a strong scale-dependency of the flow processes. At the local scale, groundwater flow is of minor importance, whereas fast surface and subsurface processes dominate. In contrast, groundwater flow plays a major role in the water cycle at the regional scale. Séguis et al. (2011) confirm this analysis. They found that baseflow was the major source of annual discharge in catchments ranging from 10 to 600 km2 in size. Apart from the dependence on scale, the soil characteristics of the catchment suggest that distinguishing between interflow and baseflow might be difficult in the Upper

Ouémé Catchment. First, field observations by Giertz et al. (2005) revealed a very high mean density of macropores of (219 per m2) for a savannah, indicating that bypass flow plays a major role in the water cycle. Séguis et al. (2011) and Giertz (2004) observe that it is necessary to distinguish between shallow groundwater and deep groundwater. The deep groundwater does not contribute to the local water cycle. Séguis et al. (2011) find that the percolation rate to deeper groundwater is between 9% and 17%. Fass (2004) reports a fraction of fast runoff components of 73%, based on hydrochemical measurements for two different small-scale catchments. The challenge is to distinguish the runoff components on the large scale. The discharge behaviour in the study area is highly seasonal. Discharge begins in June, peaks in September and ends at the beginning of December. Giertz et al. (2010) show mean discharge and runoff coefficient values for different subcatchments of the Ouémé between 1993 and 2004. The mean discharge of the Térou Catchment is 212 mm/a, with a runoff coefficient of 0.16. Fink et al. (2010) show that mean potential evapotranspiration can vary between 1690 and 2420 mm/a, with lower values during the rainy season. An analysis of actual evapotranspiration by Giertz et al.

224

T. Cornelissen et al. / Journal of Hydrology 498 (2013) 221–236

Fig. 2. Hydrological processes on a typical slope in the Ouémé Catchment (Fass, 2004, p. 119).

(2010) shows simulation results from the UHP-HRU model of different subcatchments of the Upper Ouémé Catchment. The results yield an estimate of 843 mm/a for the Térou Catchment.

3. Materials and methods 3.1. Comparison of model structures In this study, four different hydrological models are applied. These models vary in complexity, spatial resolution and process representation. All of the models are time-continuous and are applied at daily time steps. The models’ basic characteristics are described in a tabular summary (Table 1), while the text provides detailed descriptions of the most important differences in process representation between the models. The Water balance-Simulation Model (WaSiM) has been developed by Schulla (1997) to evaluate the influence of climate change on water balance in low and high mountain ranges. This deterministic, spatially distributed model incorporates physical and conceptual approaches to describe relevant hydrological processes. Version 7.9.11 of the model, developed at the end of 2007, is used in this study (Schulla and Jasper, 2007). The Soil Water Assessment Tool (SWAT) is a time-continuous semi-distributed catchment model developed by the US Agricultural Research Service to evaluate the influence of climate, land use and agricultural cultivation techniques on water quality and sediment yield (Arnold et al., 1998). The UHP-HRU model is a semi-distributed conceptual model, originally developed by Bormann and Diekkrüger (2004) and extended by Giertz et al. (2006) using the hydrological response unit concept and by Giertz et al. (2010) using a routing routine. The GR4J model (Génie Rural à 4 Paramètres Journaliers; Perrin et al., 2003) is a daily lumped rainfall-runoff model with only four parameters, all of which must be calibrated. In contrast to WaSiM, which uses a grid based spatial discretisation, SWAT and UHP-HRU divide the catchment into subbasins that are further divided into Hydrological Response Units (HRUs). While SWAT creates fixed HRUs based on superposed land use and soil maps, UHP-HRU allows for yearly changing land use. For the subunits that are created by the superposition of subbasins and soil maps, a share of each

land use type is calculated for each year. This allows a dynamic change in land use and an adaption in the size and number of HRUs according to land use changes. Thus, each HRU is characterised by uniform land coverage, soil properties and land use management, but it is not georeferenced within a subbasin. GR4J does not require spatial discretisation. Whereas SWAT and WaSiM divide the soil into numerical layers, UHP-HRU divides the soil into two storage components, i.e., the root and unsaturated zones, whose recession constants need to be calibrated. Root-zone storage is filled by the amount of rainfall that is not captured by interception and runoff; unsaturatedzone storage is filled by percolation, which depends on the water level in root-zone storage, its maximum water storage capacity (field capacity) and a recession constant. All models differ in their representation of interflow. In WaSiM, lateral flow (interflow) is a function of slope length and inclination, saturated water conductivity, drainable porosity and the amount of drainable water stored in the saturated zone. Interflow is calculated by comparing a maximum possible interflow rate, depending on the actual water content, with a second interflow rate that is a function of river density, the hydraulic gradient and conductivity. The smaller value of both interflow rates is taken as the actual rate. UHP-HRU determines the amount of interflow as a ratio between the actual and maximum water storage of the unsaturated zone. In SWAT, interflow is calculated using a linear function that consists of slope length and angle, saturated conductivity, drainable porosity and the amount of water that is stored in the saturated zone. All models use a linear storage approach to calculate baseflow, but they differ in their representation of the groundwater layer. SWAT and UHP-HRU simulate one deep aquifer, which does not contribute to baseflow, and one shallow aquifer, which is linked to the river system. The shallow aquifer is recharged by percolation and reduced by deep percolation to the deep aquifer and capillary rise to the unsaturated zone. Percolation is only calculated if the water content is above field capacity. GR4J consists of two storage components, the production store and the routing store. Production storage represents soil–water storage. If precipitation is larger than potential evapotranspiration, the net rainfall is calculated by subtracting potential evapotranspiration from rainfall. The amount of water filling the production store from net rainfall is calculated using a parabolic function resembling infiltration. The remaining portion of the water will

Linear storage, soil is divided into root and unsaturated zone SCS curve number (SCS, 1972) Tipping bucket, soil is divided into numerous numerical layers SCS curve number (SCS, 1972)

225

directly feed the discharge. This process is represented by considering a unit hydrograph approach. If precipitation is lower than potential evapotranspiration, the net evapotranspiration rate is calculated by subtracting rainfall from potential evapotranspiration. Actual evapotranspiration is computed from soil water storage using a parabolic equation. Percolation from the production store is calculated using a power function based on water storage. Total discharge, computed from net precipitation and percolation, is divided into two unit hydrographs with a fixed ratio of 9:1. A groundwater exchange rate resembling the fast and slow groundwater components is calculated and applied to the amount of routed water. The total discharge is calculated as the superposition of the two unit hydrographs (Perrin et al., 2003). WaSiM requires twelve parameters for each soil mapping unit to simulate soil water dynamics: four parameters for the incorporation of macropores, the saturated hydraulic conductivity and its recession for each horizon, the van Genuchten parameters describing the soil water retention function of the horizon and the thickness of the horizons. The van Genuchten parameters are calculated from available soil data by applying the pedotransferfunction of Carsel and Parish (1988). The recession of saturated hydraulic conductivity with depth is a calibration parameter. Due to limited data, the macropore model is not applied and bypass flow is not simulated. For the simulation of the evapotranspiration module, WaSiM depends on thirteen different parameters for each land use class. Five parameters (threshold value for the start of dryness stress, two parameters which describe the reduction of transpiration due to oxygen stress, root distribution in the soil and the interception capacity per LAI) do not vary during the year, whereas the other parameters (i.e., evapotranspiration resistance, aerodynamic resistance, leaf area index, vegetation height, albedo and root depth) are variable in time. All of these parameter values were taken from literature, except the evapotranspiration resistances that are associated with agriculture, which had to be calibrated. For example, the parameters for the ‘‘wood savannah’’ land use type were taken from Bronstert et al. (2001), Güntner (2002), Hagemann (2002) and Steyaert and Knox (2008). In SWAT, the soil module is parameterised with hydrologic groups attributed according to soil texture, effective soil depth and shrink-swell potential, soil depth, organic carbon content, soil texture distribution (all available from measurements) and bulk density (estimated by a regression model derived from measured soil data from Giertz (2004) and Sintondji (2005)); available water capacity and saturated hydraulic conductivity were estimated using the pedotransfer function of Rawls and Brakensiek (1995). The original land use classes were attributed to the predefined land use classes of SWAT, but plant heights and the vegetation parameters affecting the seasonal development of LAI had to be adapted according to data from Mulindabigwi (2006). UHP-HRU requires the soil depth, water holding capacity and storage constants for the root and unsaturated zone for each soil unit. While the storage constants were calibrated, the soil depth and water holding capacity were taken from measurements (Giertz (2004), Hiepe (2008), Sintondji (2005)). A Curve Number (CN II) is required for each combination of soil and land use type, which was assessed based on SCS (1972). Other required vegetation parameters are root depth, LAI, albedo and plant height. These parameters were taken from investigations in the Ouémé catchment by Orthmann (2005) and Mulindabigwi (2006). Missing data were taken from Scurlock et al. (2001). 3.2. Data sources

Interflow Baseflow Routing

Overland flow Percolation

Infiltration

Based on Peschke (1977) and Green and Ampt (1911) Horton overland flow SCS curve number (SCS, 1972) Function based on soil saturation and saturated conductivity Storage routing; water content must be above field capacity Storage approach; comparing maximum and actual rate Kinematic storage model Linear storage approach Linear storage approach Kinematic wave approach considering retention and Continuity equation using Manning’s translation equation

SCS curve number (SCS, 1972) Storage routing; water content must be above field capacity Linear storage approach No distinction Linear storage approach No distinction Continuity equation using a simplified storage approach Water is staggered into a number of inputs for the two unit hydrographs

Calculation of net precipitation with parabolic function No distinction Power function

If PET- Precipitation > 0, actual evapotranspiration is taken from soil water storage using parabolic equation Production and routing store Depends on PET and water availability in root storage zone Calculated separately for evaporation and transpiration, reduction of PET by soil water content

Separate calculation of evaporation from vegetated soils considering all soil layers and from bare soil for the first soil layer; both reduced by soil water content of first soil layer Richards equation Soil module

Not considered Penman (1956)

GR4J (Perrin et al., 2003)

Storage approach; function of LAI Penman (1956)

UHP-HRU (Giertz et al., 2010) SWAT (Arnold et al., 1998) WaSiM (Schulla and Jasper, 2007)

Storage approach; function of LAI Penman–Monteith (Monteith, 1975)

Interception Potential Evapotranspiration Actual Evapotranspiration

Table 1 Hydrological processes and process representation of the four models used in this study (LAI = leaf area index; PET = potential evapotranspiration).

Storage approach; function of LAI Penman–Monteith (Monteith, 1975)

T. Cornelissen et al. / Journal of Hydrology 498 (2013) 221–236

Table 2 summarises the data used and the data sources. All models were run with a temporal resolution of one day. As WaSiM is the only spatially distributed model considered, only this model

226

T. Cornelissen et al. / Journal of Hydrology 498 (2013) 221–236

Table 2 Data availability for the four models used (DMN = Direction de la Météorologie Nationale; CATCH = Couplage de l’Atmosphère Tropicale et du Cycle Hydrologique; IRD = Institut de Recherche pour le Développement; DGE = Direction Générale de l’Eau; PET = potential evapotranspiration, IMPETUS = An Integrated Approach to the Efficient Management of Scarce Water Resources in West Africa). WaSiMa Climate data Precipitation Discharge data Elevation Data Soil data Land use classification a b

SWATb

UHP-HRUb

1 Station (DMN) 2 Stations (DMN, IMPETUS) Same as SWAT 4 Stations (DMN) 13 Stations (CATCH, IRD, IMPETUS) 18 Stations (IRD, IMPETUS) DGE; daily data, same for all four models SRTM-Data; resolution: 90  90 m SRTM-Data; resolution: 90  90 m SRTM-Data; resolution: 90  90 m Soil map 1:200,000. Soil properties were provided by Sintondji (2005) and Hiepe (2008): one representative profile for each soil mapping unit (sheet Djougou; Faurè, 1977) Judex (2008); classified from LANDSAT images with a resolution of 28.5  28.5 m

GR4Jb PET is taken from UHP-HRU; Same as UHP-HRU – – –

Data are valid for the Térou Catchment. Data are valid for the Upper Ouémé Catchment; see figure 1 for locations.

was run with a spatial resolution of 1 km2. SWAT and UHP-HRU use the Hydrologic Response Unit concept for spatial discretisation, and GR4J does not require spatial data. For the spatial discretisation of the Térou Catchment, SWAT uses 177 HRUs, whereas UHP-HRU uses 281 HRUs. The distributed catchment precipitation is calculated in different ways for the four models. The UHP-HRU and the SWAT model assign one precipitation value for each day to each subbasin. This method may create sharp boundaries between precipitation regions. This spatial distribution of precipitation values follows the Thiessen polygon interpolation method. For WaSiM, the inverse distance weighting method was assigned, thus resulting in a distinct precipitation value for each grid cell. This approach yields smoother patterns. Data for the UHP-HRU model were taken as inputs for GR4J. Nevertheless, the convective character of precipitation can neither be accurately represented by the continuous precipitation field of the inverse distance weighting method nor by the sharp boundaries of the Thiessen polygons. In addition to uncertainty resulting from the density of rain gauges, the interpolation method may influence model results. Bormann (2005) notes that the choice of interpolation method has a weaker influence on simulation results than errors in the precipitation data. 3.3. Calibration and validation procedures As calibration and validation were done by different authors, the available data were split into calibration and validation periods inconsistently. Calibration for the UHP-HRU and GR4J model was performed for the years 2003 and 2004, while validation was performed for 1997 through 2002. The years 2002 through 2005 were chosen as the calibration period for the WaSiM simulation, while the years 1998 through 2001 were chosen as the validation period for the WASIM simulation. A spin-up time of two years was used to define proper initial conditions for both the validation and calibration periods. The SWAT model was calibrated for the years 1998 through 2001, and validated for 2002 through 2005. Due to the substantial number of parameters, equifinality may occur (Beven, 2001). To reduce the possible parameter space, additional information on the fraction of discharge components may be useful. According to Séguis et al. (2011), baseflow is the dominant runoff generation process in the savannah-dominated, 586 km2 large Donga catchment. Moreover, the measurements made by Giertz (2004) show a fraction of surface runoff of 30%. This value is the mean value of the conductivity measurements performed in the upper Aguima Catchment, which has a size of 3.2 km2, is predominantly savannah vegetation. The IMPETUS project, which focuses on investigating the effect of global environmental change on the water cycle in Benin and Morocco and the development of a decision support system, also chose the Aguima Catchment as a representative catchment in terms of hydrological processes (Giertz,

2004). Thus, the calibration aim of WaSiM is to achieve a tradeoff between the simulation of a fraction of surface runoff of approximately 30%, the simulated discharge amount and the quality of the simulated discharge dynamics. The calibration aims were achieved by adjusting the following parameters, which Kasei (2010), Jung (2006), Schulla and Jasper (2007) and Cullmann et al. (2006) found to be sensitive: the reduction of saturated conductivity with depth (influencing the partitioning between the interflow and base flow), the drainage density (influencing interflow), the scaling factor and the recession constants for baseflow and inter- and overland flow. The validation and calibration of the SWAT model was performed by Hiepe (2008) for the Térou Catchment. The calibration aim is comparable to that of the WASIM simulation, with the exceptions that a fraction of surface runoff of 45% was assumed to be correct and the Nash–Sutcliffe Coefficient (Nash and Sutcliffe, 1970) and Coefficient of Determination reached at least 0.7. The fraction of surface runoff is different from that of WaSiM because the fraction for SWAT is a result of baseflow filtering, which cannot differentiate between fast and slow baseflow components. Calibration was performed manually and with the automatic calibration implemented in SWAT (Shuffled complex evolution algorithm) at yearly and weekly time steps for the SCS curve number parameters, the saturated hydraulic conductivity (values smaller than 5 mm per hour were set to 5 mm per hour) and the parameters affecting the recession of baseflow and surface runoff, as well as for the percolation to the shallow and deep aquifers and evaporation. For each land use, four different SCS curve numbers were calibrated according to their assigned hydrologic group, which were separated according to different infiltration rates. The calibration of the UHP-HRU model was performed manually by Giertz et al. (2010) for root depth, the curve number parameters and the recession constants for interflow and groundwater. The calibration aims to optimise the simulation of total discharge and its seasonality and maximise the Nash–Sutcliffe Coefficient (Nash and Sutcliffe, 1970) and the Coefficient of Determination. GR4J was calibrated automatically for the four model parameters: the maximum capacity of the routing store and production store, the groundwater exchange coefficient and the time base of the first unit hydrograph. The aim of the calibration is to maximise the Nash–Sutcliffe coefficient (Nash and Sutcliffe, 1970). 3.4. Climate and land use scenarios The WaSiM, SWAT and UHP-HRU models were run in the first simulation step with three land use scenarios; in the second simulation step, all four models were run with two climate scenarios consisting of three ensemble runs each. For the climate change simulation, the model results of Paeth and Diederich (2011) for the REMO model (Regional Model, Jacob, 2001) were used. The results are based on IPCC scenarios A1B and

227

T. Cornelissen et al. / Journal of Hydrology 498 (2013) 221–236

B1 (IPCC, 2007). For each scenario, three ensemble runs were available and were applied to the four hydrological models. The REMO model is able to consider the change in the concentration of atmospheric greenhouse gases, a change in land coverage and soil degradation. The simulation period ranged from 2001 through 2049. Because REMO is a mesoscale model, the simulated precipitation pattern has a resolution of 0.5°, which does not match the local precipitation data. For this reason, a downscaling method was applied. The method includes a physical downscaling component to incorporate the effects of local topography, a stochastic component with random rainfall distributions and a statistical component that adjusts the statistical characteristics of the simulated precipitation pattern to those of the observed daily rainfall data. The method maintains the climate signal from REMO and matches the simulated pattern to the observed spatial and temporal rainfall distribution (Christoph et al., 2010). Hiepe (2008) compared the results of the REMO simulation for the period 1960–2000 with measured data at Parakou (see Fig. 1). She concluded that REMO reproduces monthly rainfall and potential evapotranspiration as well as the frequency distribution of daily rainfall very well. Land use was simulated on a yearly basis for the years 2000 through 2024 by Judex (2008). Based on socio-economic scenarios developed by the IMPETUS project (Reichert and Jaeger, 2010), Judex (2008) computed annual land use change at a high spatial resolution. In the first step, he identified local driving forces by analysing land use changes between 1990 and 2000. He then calculated the demand for arable land, depending in part on population growth and technological development, with a yearly time step. In the third step, the location where land use change is expected is calculated based on soil suitability, protected areas and access to infrastructure or markets. Table 3 gives an overview of socio-economic development and the resulting change in arable land. The values mentioned in Table 3 for the expansion of cultivated land refer to the Upper Ouémé Catchment but are also valid for the Térou Catchment. 4. Results 4.1. Calibration and validation To avoid confusion resulting from the different calibration and validation periods, the results are presented for 2 periods: a first period between 1998 and the end of 2001 (calibration period for SWAT and validation period for the three other models) and a second period between 2003 and 2004 (validation period for SWAT and calibration period for the three other models). Table 4 summarises the components of the simulated water budget and the fractions of discharge components for each model and gives the Nash–Sutcliffe-coefficient (Nash and Sutcliffe, 1970) and the Coefficient of Determination for both periods. A correct interpretation of the results requires knowledge about the climate variability during the two periods. During both periods, the first years (1998, 2003) experienced 1459 mm and 1323 mm precipitation and were wetter than the other years. The years 1999 and 2004 had comparable precipitation amounts (1189 mm and 1115 mm, respectively), but during the first period precipitation decreased, declining from 1076 mm in 2000–1000 mm in

Table 4 Mean water balance, mean discharge components, coefficient of determination and Nash–Sutcliffe-coefficients for the periods 1998–2001 and 2003–2004 for each model.

a

WaSiM

SWAT

UHP-HRU

GR4J

1998–2001 Measured discharge (mm) Simulated discharge (mm) Fraction of surface runoff (%) Fraction of interflow (%) Fraction of baseflow (%) Potential evapotranspiration (mm) Actual evapotranspiration (mm) Precipitation (mm) Nash–Sutcliffe coefficient Coefficient of determination

219 220 50 26 24 2258 861 1089 0.81 0.83

219 229 45 0 55 1635 599 1087 0.91 0.92

219 210 21 65 14 1626 760 1181 0.69 0.71

219 289 – – – 1626 664 1181 0.75 0.81

2003–2004 Measured discharge (mm) Simulated discharge (mm) Fraction of surface runoff (%) Fraction of interflow (%) Fraction of baseflow (%) Potential evapotranspiration (mm) Actual evapotranspiration (mm) Precipitation (mm) Nash–Sutcliffe coefficient Coefficient of determination

222 287 50 25 25 1931 993 1328 0.77 0.89

222 212 43 0 57 1479a 767 1224 0.84 0.85

222 175 26 62 12 1659 872 1219 0.81 0.85

222 225 – – – 1659 803 1219 0.86 0.90

Values are the mean values for the period 2002–2005.

2001. This resulted in a sharp decline in precipitation and discharge amount during both periods. Considering the precipitation data of the full calibration and validation periods, SWAT has been calibrated for drier conditions and all other models have been calibrated for wetter than mean conditions. According to Table 4, both statistical measures reach at least 0.75 for all models except for the UHP-HRU, which only reaches 0.69 and 0.71. For WaSiM, UHP-HRU and GR4J, this outcome indicates good agreement between the measured and simulated discharge data during the validation period. SWAT reaches statistical values higher than 0.9, which indicates excellent agreement between the measured and simulated discharge. The analysis of the water balance components (Table 4) shows that UHP-HRU simulates the lowest fraction of surface runoff and that SWAT and WaSiM calculate the highest values. For the WaSiM and SWAT simulations, surface runoff represents 50% and 45%, respectively, and is thus higher than the mean measured fraction of 30% surface runoff (Giertz, 2004). The calibration aim for the WaSiM simulation, i.e., the limitation of surface runoff to approximately 30%, is only achieved for three individual years. Fig. 3 compares the difference between the observed and measured discharge for the applied models during the first period. The months from December to June are not shown because no discharge was measured and only the UHP-HRU model calculated low discharge peaks during the dry period caused by single rainfall events. In the dry period, the surface runoff that occurs will infiltrate in the river bed. This process is not included in UHP-HRU. Moreover, the UHP-HRU simulation overestimates the discharge at the beginning of the discharge period, but underestimates the discharge towards the end of the year (Fig. 3). WaSiM underestimates the discharge at the beginning of the discharge period, but

Table 3 Main assumptions of the land use scenarios LU1, LU2 and LU3 and the resulting change in the fraction of arable land between 2000 and 2024 (Judex, 2008). LU1 scenario

LU2 scenario

LU3 scenario

 Economic growth  Improved political and socio-economic situation  Innovations in the agricultural sector spread due to good administrative structures  Expansion of cultivated areas by 15%

 Economic stagnation  Protected forests are not controlled  No new technologies that enhance production  Expansion of cultivated area by 30%

 ‘‘Business as usual’’  Productivity does not increase  Expansion of cultivated area by 20%

T. Cornelissen et al. / Journal of Hydrology 498 (2013) 221–236 4.00

0

3.00

20

2.00

40

1.00

60

0.00

80

-1.00

100

-2.00

120

-3.00

140

-4.00

Precipitaon (mm/d)

Observed - Simulated Discharge (mm/d)

228

160 07/98

09/98

11/98

07/99

09/99

Precipitaon (GR4J Input)

11/99 GR4J

07/00

09/00 WaSiM

07/01

11/00 UHP-HRU

09/01

11/01

SWAT

Fig. 3. Difference between observed and measured daily discharge for the WaSiM, SWAT, UHP-HRU and GR4J models between July 1998 and the end of November 2001.

Observed - Simulated Discharge (mm/d)

correctly simulates the decrease in discharge at the end of the year. In addition, it is visible from the lower range of differences between the observed and simulated discharge that the models’ performance is much better during the drier years (1999, 2000 and 2001). This behaviour can be explained by a delayed response of the discharge components and the groundwater level to the rainfall events between the beginning of July and mid-August. The comparison of the results between the two periods shows that the performances of WaSiM, UHP-HRU and GR4J, evaluated in terms of the discharge simulation and statistical measurements, are better during the second period. The large difference between the Nash–Sutcliffe-coefficient and the Coefficient of Determination

for WaSiM is due to the overestimation of peak discharge rates between July and the end of September, 2003 (Fig. 4). SWAT and WaSiM overestimate the discharge rates until the end of August 2003, but after that date, WaSiM shows a slight underestimation. Both UHP-HRU and GR4 J underestimate the discharge during the whole period. In 2004, overestimations by UHP-HRU at the beginning of the year and by WaSiM during August are visible. The two figures also indicate that the model performance is better during drier years than during wetter years. According to the data in Table 4, the SWAT and GR4J models simulate nearly the same amount of discharge for the period 2003–2004, whereas WaSiM computes the highest amounts and UHP-HRU simulates the lowest amounts of discharge. The overestimation of the total discharge during the

4.00

0

3.00

20

2.00

40

1.00

60

0.00

80

-1.00

100

-2.00

120

-3.00

140

-4.00

160 07/03

09/03

Precipitaon (GR4J Input)

11/03

GR4J

07/04

09/04

WaSiM

UHP-HRU

11/04

SWAT

Fig. 4. Difference between simulated and observed daily discharge for the WaSiM, SWAT, UHP-HRU and GR4J models between July 2003 and the end of November 2004.

229

T. Cornelissen et al. / Journal of Hydrology 498 (2013) 221–236

calibration period of WaSiM results from the calibration procedure. To achieve the goal of 30% surface runoff, parameters influencing the amount of inter- and baseflow had to be increased. These changes in the parameter values caused the total discharge amount to increase. This observation also explains the substantial overestimation of discharge peaks in 2003. The GR4J simulation shows only a small overestimation of total discharge during the calibration period, but an overestimation of 70 mm during the validation period. This might be a result of the calibration process, as this process will maximise the Nash–Sutcliffe-coefficient. The simulated fractions of the discharge components are very similar for all models during both periods, except for the UHP-HRU model, which simulates at least a 5% increase in surface runoff during the second period corresponding to the calibration period. As shown in Table 4, WaSiM calculates the highest amount of potential evapotranspiration of all models evaluated with 2258 mm/a during the period 1998–2001 but only 1931 mm/a during the period 2003–2004. Both values are consistent with Class A pan evapotranspiration measurements in Benin reported in Fink et al. (2010).The data come from the Direction de la Météorologie Nationale (DMN) of

Benin and consider the oasis effect. According to Fink et al. (2010), potential evapotranspiration can vary from 10–12 mm per day during the dry period and from 2 to 4 mm per day in the rainy season. These values correspond to a yearly mean value of between 1690 and 2420 mm. Furthermore, the results of Kasei (2010) show that the potential evapotranspiration calculated by WaSiM for the White Volta Basin, which has a drier climate than the Térou Catchment, can vary between 1759 mm (calibration period) and 2272 (validation period). 4.2. Land use scenarios The climate data from 1998 to 2004 were used to simulate the land use scenarios. To exclude the effects of varying climate, a mean value of the model results over the seven simulation years was calculated. The three land use scenarios are differentiated by their increase in agricultural area, as well as by other factors. The computed discharge is shown as a seven-year mean monthly value for all scenarios for total and surface runoff in Figs. 5–7; the values for all water balance components are given in Table 5.

100 95 90 85 80 75 Discharge (mm/month)

70 65 60 55 50 45 40 35 30

c

25

d

b

20

a

15 10

Land use 2000 (a)

2024 Scenario B1 (b)

2024 Scenario B2 (c)

December

November

September

August

July

June

May

April

March

February

January

0

October

5

2024 Scenario B3 (d)

Fig. 5. Total discharge (solid line) and surface runoff (dashed line) simulated by UHP-HRU for the land use scenarios. Characters a-d show the ordering of the curves. The ordering is the same for total discharge and surface runoff. 100 95 c

90

d

85 80

Discharge (mm/month)

75

b

70 65 60 a

55 50 45 40 35 30 25 20 15 10 5

Land use 2000 (a)

2024 Scenario B1 (b)

2024 Scenario B2 (c)

December

November

October

September

August

July

June

May

April

March

February

January

0

2024 Scenario B3 (d)

Fig. 6. Total discharge (solid line) and surface runoff (dashed line) simulated by WaSiM for the land use scenarios. Characters a-d show the ordering of the curves. The ordering is the same for total discharge and surface runoff.

230

T. Cornelissen et al. / Journal of Hydrology 498 (2013) 221–236 100 95 90 85 80 Discharge (mm/month)

75 70 65 60 55 50

c

45

d b

40 35 30

a

25 20 15 10 5

Land use 2000 (a)

2024 Scenario B1 (b)

2024 Scenario B2 (c)

December

November

October

September

August

July

June

May

April

March

February

January

0

2024 Scenario B3 (d)

Fig. 7. Total discharge (solid line) and surface runoff (dashed line) simulated by SWAT for the land use scenarios. Characters a-d show the ordering of the curves. The ordering is the same for total discharge and surface runoff.

Table 5 Seven-year (1998–2004) mean values of water balance components for the three land use scenarios and the land use for the year 2000 for all models.

a b c

2000

2024 LU1

2024 LU2

2024 LU3

UHP-HRU Total discharge (mm) Fraction of surface runoff (%) Fraction of interflow (%) Fraction of baseflow (%) Potential evapotranspiration (mm) Actual evapotranspiration (mm) Precipitation (mm)

188 22 65 13 1640b 804 1189a

203 30 60 10 1634b 786 1186a

219 35 56 9 1624b 778 1195a

210 32 58 10 1629b 783 1190a

WaSiM Total discharge (mm) Fraction of surface runoff (%) Fraction of interflow (%) Fraction of baseflow (%) Potential evapotranspiration (mm) Actual evapotranspiration (mm) Precipitation (mm)

186 42 38 20 2207 921 1165

226 48 30 22 2137 880 1165

249 51 26 23 2095 855 1165

231 49 29 22 2127 874 1165

SWAT Total discharge (mm) Fraction of surface runoff (%) Fraction of interflow (%) Fraction of baseflow (%) Potential evapotranspiration (mm) Actual evapotranspiration (mm) Precipitation (mm)

217 48 0 52 1564c 724 1152

224 51 0 49 1564c 721 1152

232 55 0 45 1564c 714 1152

227 53 0 47 1564c 719 1152

The differences in the mean values are due to the use of monthly output data; the maximum difference in monthly precipitation between the scenarios is 2.1 mm; The Penman model uses albedo data, which changes with changing land use. No change despite Penman–Monteith because the FAO approach for short-grass is used.

The GR4J model is not used for land use change projections because it is not able to capture the major impacts of land use changes due to the missing differentiation between discharge components. In SWAT, it is not possible to change land use dynamically because it determines HRU delineation. Therefore, land use change scenarios can only be performed by independent model runs, which makes comparing land use change effects difficult. In UHP-HRU and SWAT, the land use and soil parameters for each land use type were not modified, but the share and spatial distribution of the different land use types were changed according to the land use scenario maps of Judex (2008). In SWAT, a new land use map changes the delineation of HRUs and thus directly influences the simulation of the discharge fractions.

In WaSiM, changes in land use are described by vegetation parameters. Changes in the distribution of land use parameters result from the usage of different land use maps for the scenarios and cause a change in potential and actual evapotranspiration and, therefore, in total discharge. Table 5 shows that the changes in potential and actual evapotranspiration are pronounced in the WaSiM simulation. Comparable to the results for the validation and calibration periods, WaSiM calculates a significantly higher amount of evapotranspiration than UHP-HRU. The UHP-HRU model simulates an increase in total discharge of 12% and an increase in surface runoff of 47% on average for all scenarios compared to the reference period. WaSiM simulates a mean discharge increase of 27% and a mean increase in surface runoff of 46%. SWAT simulates a mean discharge increase of

231

T. Cornelissen et al. / Journal of Hydrology 498 (2013) 221–236

5% and a mean increase in surface runoff of 17% (refer to Table 5). All models agree that the LU2 scenario has the largest impact on discharge sum and the increase in surface runoff. The simulated discharge amounts and surface runoff fractions are very similar for WaSiM and SWAT, except for the base run in 2000 where SWAT simulates a higher discharge amount than the two other models. Differences in the shape of the surface runoff curve between UHPHRU and WaSiM are apparent from the comparison of Figs. 5 and 6. In the UHP-HRU simulation, surface runoff starts at the end of March, increases rapidly between June and July and decreases steadily until October. In contrast, surface runoff in the WaSiM simulation starts in June, increases rapidly until September and decreases to 0 mm in November. In SWAT, the surface runoff starts in May, but the peak in surface runoff is simulated for August, while the peak in total discharge amount is simulated for September (Fig. 7). 4.3. Climate scenarios The potential impact of climate change on the discharge dynamics is analysed by comparing the yearly mean sums of (1) total runoff, (2) precipitation, (3) potential and actual evapotranspiration and (4) runoff coefficient per decade. Figs. 8 and 9 show the simulated discharge amount per decade between 2001 and 2049 for each model as a mean value for all three ensemble runs of the A1B and B1 scenarios. It is clearly visible from both figures that both climate scenarios lead to a reduction of the total discharge amount, regardless of model type (Table 6). The decrease in total runoff varies between 28% (UHPHRU, WaSiM) and 34% (SWAT, GR4J) for the A1B scenario and between 28% (SWAT, WaSiM, UHP-HRU) and 33% (GR4 J) for the B1 scenario. UHP-HRU calculates the lowest values (160–88 mm per decade). In contrast, WaSiM calculates the highest values (480–200 mm; Table 6). The development of the discharge amount for the A1B-scenario (Fig. 8) is generally comparable among the models evaluated, which is not the case for the B1 scenario (Fig. 9). The UHP-HRU

model simulates a continuous decrease in total discharge between the first and last decades. In contrast, SWAT, GR4J and WaSiM calculate a large decrease in discharge amount between 2001–2010 and 2011–2020, with a maximum decrease of 132 mm (37%) calculated by the SWAT model. The differences between the three ensemble runs of the A1B and B1 scenarios are shown in Figs. 8 and 9, which depict the maximum and minimum values of the mean discharge amount for each decade. Two features of these figures are apparent. First, the difference between the maximum and minimum values decreases from the first to the last decade. Second, the differences between the minimum and maximum values are comparable for the GR4J and the UHPHRU models. Those models also produce the smallest differences, whereas WaSiM computes the largest differences. It has to be noted that the variability between the ensembles is higher than the decrease in discharge for all models. In addition, the variability is also higher than the differences between the models except for the GR4J and the UHP-HRU models in the first and last decade. The variability of the simulated decade-mean precipitation values is much lower than the variability of the discharge (Table 6). In general, the precipitation amount decreases by 138 mm for the A1B scenario and by 164 mm for the B1 scenario between 2001– 2010 and 2040–2049. The value cited is a mean value of all ensembles and all models. Potential evapotranspiration increases in all four models between the first and last decades (Table 6) from 5% to 14% in scenario A1B and from 5% to 12% in scenario B1. A comparison of the simulated potential evapotranspiration of the first and last decades shows an increase of 105 mm for both scenarios in the UHPHRU model. WaSiM and SWAT calculate higher increases than the UHP-HRU for both scenarios, whereas WaSiM calculates the highest increase (267 mm) for the A1B scenario. The development of actual evapotranspiration shows large differences between the models. In contrast to WaSiM, which simulates higher amounts in the last decade than in the first, SWAT simulates a small decrease and GR4 J and UHP-HRU calculate the highest decrease in actual evapotranspiration.

800 750 700 650 600 550

Discharge (mm)

500 450

WASIM A1B SWAT A1B

400

UHP-HRU A1B

350

GR4J A1B

300 250 200 150 100 50 0 Reference period

2001-2010

2011-2020

2021-2030

2031-2040

2041-2049

Fig. 8. Simulated yearly mean discharge per decade of each model in the A1B-scenario and for the reference period 1989–2004. The values shown are the mean values of the three ensembles used. Error bars indicate the maximum and minimum values of the three ensembles for the corresponding decade.

232

T. Cornelissen et al. / Journal of Hydrology 498 (2013) 221–236 800 750 700 650 600 550

Discharge (mm)

500 450

WAS SWA

400

UHP 350

GR4J

300 250 200 150 100 50 0 Reference period

2001-2010

2011-2020

2021-2030

2031-2040

2041-2049

Fig. 9. Simulated yearly mean discharge per decade of each model in the B1-scenario and for the reference period 1989–2004. The values shown are the mean values of the three ensembles used. Error bars indicate the maximum and minimum values of the three ensembles for the corresponding decade. Table 6 Decade mean sums of total discharge, precipitation, actual and potential evapotranspiration and runoff coefficient for both climate scenarios and all four models. WaSiM B1

SWAT A1B

SWAT B1

UHP-HRU A1B

UHP-HRU B1

GR4J A1B

GR4J B1

Total discharge (mm) 2001–2010 435 2011–2020 407 2021–2030 367 2031–2040 375 2041–2049 325

WaSiM A1B

496 390 369 401 373

263 221 219 216 174

357 225 227 263 227

124 121 109 112 88

160 115 114 120 116

182 176 156 156 121

222 166 167 173 148

Precipitation (mm) 2001–2010 1234 2011–2020 1164 2021–2030 1129 2031–2040 1138 2041–2049 1095

1282 1152 1130 1155 1123

1221 1138 1132 1131 1067

1351 1148 1165 1206 1152

1168 1134 1095 1112 1046

1231 1113 1117 1122 1098

1168 1134 1095 1112 1046

1231 1113 1117 1122 1098

Potential evapotranspiration (mm) 2001–2010 1849 2011–2020 1931 2021–2030 1965 2031–2040 2005 2041–2049 2116

1803 1915 1932 1948 2017

1835 1905 1950 1988 2082

1795 1893 1912 1945 1990

1992 2018 2032 2059 2096

1959 2016 2017 2039 2064

1992 2018 2032 2059 2096

1959 2016 2017 2039 2064

Actual evapotranspiration (mm) 2001–2010 755 2011–2020 758 2021–2030 761 2031–2040 764 2041–2049 770

741 763 760 755 751

761 743 745 755 749

753 744 754 751 747

894 886 874 889 867

892 882 879 879 864

828 806 800 818 788

825 802 804 800 790

Runoff coefficient (–) 2001–2010 0.35 2011–2020 0.35 2021–2030 0.33 2031–2040 0.33 2041–2049 0.30

0.39 0.34 0.33 0.35 0.33

0.22 0.19 0.19 0.19 0.16

0.26 0.20 0.19 0.22 0.20

0.11 0.11 0.10 0.10 0.08

0.13 0.10 0.10 0.11 0.11

0.16 0.16 0.14 0.14 0.12

0.18 0.15 0.15 0.15 0.13

5. Discussion Chapter 3.1 revealed that discharge simulations with comparable quality measures can be generated by different fractions of discharge components. This finding in the analysis of discharge

modelling is known as ‘‘equifinality’’ (Beven, 2001). However, the performance of the models evaluated for scenario simulations differs substantially. In the case of the first result presented above, all simulations conducted with the SWAT model in the Térou Catchment agree on

T. Cornelissen et al. / Journal of Hydrology 498 (2013) 221–236

a very low fraction of interflow (<5%) (Busche et al., 2005; Hiepe and Diekkrüger, 2007; Sintondji, 2005). This simulation result is in contrast to the results of the WaSiM and the UHP-HRU models, which simulate a fraction of interflow of 25% and 63%, respectively. The low fraction of interflow calculated by SWAT can be explained by the model concept. In this model, the lateral flow depends on the local slope. As the Térou Catchment is relatively flat, the interflow cannot be calculated correctly by the SWAT model (Sintondji, 2005). This weakness leads to a fraction of surface runoff of nearly 50% in Sintondji’s study because the surface runoff and the baseflow compensate for the missing interflow. The simulations of the UHP-HRU model indicate a dominance of interflow because the mean fraction for all seven simulation years is 66%. This finding is supported by Giertz (2004) using a discharge simulation performed with the physical-based model SIMULAT-H, by Fass (2004) using geochemical methods for the small Aguima Catchment (3.2 km2) and by the calibration results in the study by Kasei (2010) undertaken in the White Volta Basin. Kasei (2010) found a dominance of lateral flow during the calibration period and a dominance of surface runoff during the validation period using the WaSiM model. The WaSiM and SWAT models calculate the same fraction of surface runoff (Table 4). In WaSiM, the infiltration process calculation is physically based and is thus dependent on soil parameters. Despite the physical calculation of the infiltration process in WaSiM, which is limited in our study due to the usage of daily time steps, none of the models applied in this study can simulate the saturation-excess runoff mechanism. This property of the models implies that the Hortonian-type surface runoff fully accounts for the high fraction of surface runoff simulated by WaSiM. The dominance of Hortonian runoff is not consistent with the observation by Giertz et al. (2010) that Hortonian runoff is unlikely for savannah vegetation types because of their very high infiltration rates (at least 1000 mm per hour). On agricultural fields, the secondary pore system is reduced significantly, which results in increased surface runoff (Germer et al., 2010). It was previously stated that the infiltration process calculation in WaSiM requires the definition of certain soil parameters, among them the saturated conductivity, which is also used in the SWAT model. Güntner (2002) and Le Lay et al. (2008) have noted that soil parameters, especially hydraulic conductivity values, are a source of uncertainty that produces problems in the correct representation of infiltration characteristics (e.g., shown by Niehoff (2001) for the WaSiM model). The saturated conductivity values used in the present study were measured by Sintondji (2005) and are at least a factor of 10 smaller than the values reported by Giertz et al. (2005). These variations can be explained by different incorporations of the macropore system during laboratory (performed by Sintondji) and field measurements (done by Giertz). One possible solution to the problem would be to calibrate the saturated conductivity, as Varado et al. (2006) did in their study and as Hiepe (2008) did for the calibration of the SWAT model. Their solution was not included because it would limit the physical representativeness of the WaSiM model, which is already limited by the use of the conceptual groundwater module and the calculation of infiltration at daily time steps. The choice of different calibration and validation periods with different durations and different climates clearly limits the comparability between the results. This is especially true for the comparison between WaSiM or SWAT and UHP-HRU or GR4J because the latter only use a calibration period of two years, whereas the former use a period of four years. Despite these differences, it is important to note that SWAT and WaSiM agree on the fraction of surface runoff during both periods. Both models are calibrated and validated for different periods. WaSiM is able to simulate interflow as a fast subsurface discharge component, and both mod-

233

els have been calibrated to match different fractions of surface runoff (WaSiM: 30%; SWAT: 45%). As SWAT generally simulates the discharge curve in a better way, it can be stated either that the calibration of SWAT was more successful or that the distinction between interflow and baseflow does not adequately represents local conditions. The UHP-HRU and SWAT models do not need saturated conductivity data as inputs for calculating infiltration because the SCS curve number method is used for calculating runoff (SCS, 1972). Uncertainties in the simulation of and assumptions for partitioning the discharge components and the differences in calibration periods inevitably raise a question. Of the four applied models, which, if any, can correctly simulate the discharge, even with an optimal database? The models’ weaknesses are (1) the distinction between interflow and baseflow, (2) the exclusive consideration of Hortonian runoff and (3) the calibration affecting the formation of the discharge processes. Instead of distinguishing between interflow and baseflow, a soil module suitable for the Térou Catchment should consist of three zones: one zone on top of the loamy-sand layers or iron crusts, one zone underneath these structures and one zone for the parameterisation of the nearly impermeable soil layers. If this concept is applied, it should be possible to differentiate the subsurface runoff according to the reaction time to rainfall events in terms of three recession constants, one for each soil zone. To avoid the parameterisation of preferential flow, which has a large impact on the discharge components, the recession constant for the first layer should allow variability in time and should depend on land use. The recession constants influence the amount of discharge. With a time-varying recession constant, it should be possible to mimic the abrupt replacement of old water by water infiltration through macropores (Anderson and Burt, 1990). It is necessary to ask whether models whose calibrations affect runoff formation processes are suitable for analysing and understanding discharge processes. This concern is especially valid if a calibration appears to match statistical measures rather than the perception of the processes. Chapter 4.2 has outlined that a change in land use has a substantial influence on the discharge dynamics. The idea that an increase in the fraction of agricultural land causes an increase in discharge is generally confirmed by Bormann (2005) and has been verified for Benin by Giertz et al. (2005) and Götzinger (2007). For example, Giertz et al. (2005) found a higher amount of total discharge (62%) for a catchment dominated by agriculture compared to a catchment dominated by savannah vegetation, although the precipitation only differs by 12%. This model comparison study indicates that the increase in the amount of discharge is due to an increase in surface runoff caused by the lower infiltration rates on agricultural land. This finding is supported by measurements (Giertz et al., 2005) and other model-based land use change impact studies of Giertz et al. (2010), using the UHP-HRU model, and Busche et al. (2005) and Hiepe and Diekkrüger (2007), using SWAT. Busche et al. (2005) compute an increase in surface runoff of 17% for the same land use scenario LU2 (Table 3). The low increase of 17% in surface runoff for the LU2 scenario is in contrast to the results of the UHP-HRU and WaSiM models, which calculate the highest increases in surface runoff for the LU2 scenario (83%) corresponding to the highest increase in arable land (Table 3). These results indicate that SWAT is basically suitable for assessing land use change; however, Hiepe and Diekkrüger (2007) indicate that simulating the effects of land use change scenarios on discharge dynamics with the SWAT model is time consuming because changing land use leads to a change in the delineation of HRUs. Although the UHP-HRU model used in this study also applies

234

T. Cornelissen et al. / Journal of Hydrology 498 (2013) 221–236

the concept of HRUs, changing land use is easier to implement, as UHP-HRU has been programmed to dynamically consider annual land use, whereas SWAT is only able to consider one land use map per simulation run. Contradictory results concerning the applicability of a physically based model to simulate land use change effects can be found in the literature. The results of Bormann et al. (2009) and Pöhler (2006) indicate that WaSiM appears to be unsuitable for analysing changes in discharge due to changing land use. Bormann et al. (2009) explain this finding in terms of the groundwater-dominated parameterisation of the WaSiM model, which would not allow surface runoff and interflow to react to changes in vegetation coverage. It would be necessary to define the saturated conductivity values in relation to land use and not only to soil types. This requirement is especially applicable to Benin because Giertz et al. (2005) find that saturated conductivity values vary more between land use types than between soil types. In contrast to the results of Pöhler (2006) and Bormann et al. (2009), the simulation results of Niehoff (2001) and Leemhuis et al. (2007) show that the WaSiM model can simulate changes in discharge resulting from changes in land use. In the study by Niehoff (2001), the fraction of land cover types is changed by implementing different land use maps. This approach causes a change in discharge during flood events. This method is also used in the present study. Leemhuis et al. (2007) implement land use change in WaSiM by altering the vegetation parameterisation relative to the cover of the land use classes. This method resulted in a significant change in total runoff for a scenario in which all forest at an altitude less than 1200 m above sea level was converted into agricultural land. However, no such change in total runoff occurred for a simulation using the observed land use changes. It can be concluded that, in addition to model structure and approaches, the sensitivity of simulation models to land use change depends on model calibration strategies. The climate during the calibration period, as well as the selection of the objective function, determines the parameters and the applicability of the calibrated parameters for scenario quantification. All models show a decrease in precipitation and total discharge until 2050 for the Térou Catchment under both climate scenarios. The mean decrease in precipitation over all models ranges from 134 mm, or 11% (A1B), to 156 mm, or 12% (B1). As all models used the same input data for the climate scenarios, the differences in calculated precipitation can be explained by the different regionalisation methods in the applied models. The mean decrease in total discharge ranges from 74 mm (A1B), or 29%, to 93 mm, or 30% (B1). Basically, all models compared in this study are suitable for a simulation of climate scenarios. Pöhler (2006) remarks that the results of scenarios are influenced more by the quality of the data than by the model structure of WaSiM. However, the results of climate scenarios must be interpreted with caution because the simulated climate data depend on certain assumptions about the possible development of climate which manifest in the fact that the variability between ensembles is larger than the decrease for all models and larger than the difference between the models. This concern is especially applicable to the precipitation data simulated for Benin because this region is marked by high interannual and decadal variability in precipitation (Fink et al., 2010). During the simulation period from 1998 to 2004, the precipitation rates interpolated by WaSiM for the Térou Catchment range from 952 mm (2000) to 1475 mm (2003), a difference of 55%. In addition, Holländer et al. (2009) remark that the modeller itself is an intrinsic part of the modelling process and thus a source of uncertainty. This uncertainty might explain a large part of the variation in simulated future developments.

6. Conclusions In this study, four different model types were used to simulate current and future discharge of a tropical catchment. The simulation quality for the current total discharge and its components was shown to vary between model types. This variation was attributed to serious uncertainties in input data, particularly in precipitation and saturated conductivity data, calibration strategy, parameterisation and differences in model structure. The results of the study suggest the question whether hydrological models whose calibration allows for a direct influence on the discharge amount are a good choice for analysing hydrological processes. Following the dictum by Box and Draper, 1987 that ‘‘essentially, all models are wrong but some are useful,’’ it is not possible to validate models in the strict meaning of this term. The aim of model comparison studies is therefore to analyse which models may be useful for certain research questions. Comparing the results of land use scenarios revealed that SWAT, UHP-HRU and WASIM are suitable for assessing land use change because they provide similar results. In WaSiM, land use change is simulated by a change in evapotranspiration. It is doubtful that the change in evapotranspiration fluxes represents the important processes linked with the effects of land use change. For example, the reduction in the saturated conductivity values following the expansion of arable land is not considered. Although WaSiM applies the Penman–Monteith approach considering plant properties (e.g., bulk-stomata resistances, root depths), these properties are predominantly unknown for most of the vegetation types in West Africa and therefore introduce uncertainty into the model. Thus, this study concurs with the opinion of Bormann et al. (2009) and Pöhler (2006), who doubt that WaSiM is suitable for assessing land use change. All of the models evaluated (with the exception of the WaSiM model) are suitable for the simulation of future discharge because they produce comparable results for discharge development. The WaSiM model produces considerably higher total amounts of discharge. Despite the choice of different calibration and validation periods and methods, chapter 4.1 showed that the fractions of discharge components do not vary between the two periods we have chosen for the comparison of discharge results. Additionally, only the UHP-HRU model is able to simulate the dominance of fast subsurface flow components that we assume to be the dominant discharge component, based on the assumption that the measurements by Giertz (2004) are representative of processes at larger scales. Thus, according to this study, the UHP-HRU model is the most suitable for discharge simulation in Benin because it exhibits the best tradeoffs between parameterisation and calibration efforts and physical representativeness. In contrast, the GR4J model reproduces the discharge curve very well but does not contribute to the understanding of the underlying processes. This finding supports the opinion that any statistically driven calibration, even of a physical model, would lead to a poor representation of processes, especially in data-sparse regions such as Benin.

Acknowledgements The authors would like to thank the Federal German Ministry of Education and Research (BMBF, Grant No. 01 LW 06001B) as well as the Ministry of Innovation, Science, Research and Technology (MIWFT) of the federal state of North Rhine-Westfalia (Grant No. 313-21200200) for the funding of the IMPETUS project in the framework of the GLOWA-program. Many thanks to our partners in Benin and all colleagues of the IMPETUS project, who provided data and assistance. We further thank Dr. Constanze Leemhuis

T. Cornelissen et al. / Journal of Hydrology 498 (2013) 221–236

and Dr. Susannah Moore for their support and Dr. Claudia Hiepe for providing SWAT model results. Finally, we acknowledge useful comments from anonymous referees on an earlier version of this paper. References Ajayi, A.E., 2004. Surface Runoff and Infiltration Processes in the Volta Basin, West Africa: Observation and Modeling. Ecology and Development Series 18. Culliver, Göttingen, 151pp. . Andersen, J., Refsgaard, C., Jensen, K., 2001. Distributed hydrological modelling of the Senegal river basin – model construction and validation. J. Hydrol. 247 (3– 4), 200–214. Anderson, M., Burt, T., 1990. Subsurface runoff. In: Anderson, M., Burt, T. (Eds.), Process Studies in Hillslope Hydrology. Wiley, Chichester, pp. 365–400. Anhuf, D., Frankenberg, P., 1991. Die Naturnahen Vegetationszonen Westafrikas. Die Erde 122 (1), 243–265. Arnold, J.G., Srinivasan, R., Muttiah, R.S., Williams, J.R., 1998. Large area hydrologic modelling and assessment. Part I: Model development. J. Am. Water Resour. As. 34 (1), 73–89. Aubréville, A., 1957. Accord a Yangambi sur la nomenclature es types africaines de végétation. In: Bois et Fôrets des Tropiques, vol. 51, pp. 23–27. Beven, K., 2001. Rainfall-Runoff Modelling – The Primer. Wiley, Chichester. Bormann, H., 2005. Regional hydrological modelling in Benin (West Africa): uncertainty issues versus scenarios of expected future environmental change. Phys. Chem. Earth. 30 (8–10), 472–484. Bormann, H., Diekkrüger, B., 2004. A conceptual, regional hydrological model for Benin (West Africa): validation, uncertainty assessment and assessment of applicability for environmental change analyses. Phys. Chem. Earth. 29 (11–12), 759–786. Bormann, H., Fass, T., Giertz, S., Junge, B., Diekkrüger, B., Reichert, B., Skowronek, A., 2005. From local hydrological process analysis to regional hydrological model application in Benin: concept, results and perspectives. Phys. Chem. Earth. 30 (6–7), 347–356. Bormann, H., Breuer, L., Gräff, T., Huisman, J.A., Croke, B., 2009. Assessing the impact of land use change on hydrology by ensemble modelling (LUCHEM) IV: model sensitivity to data aggregation and spatial (re-) distribution. Adv. Water Resour. 32 (2), 171–192. Bossa, A., Diekkrüger, B., Igué, A.M., Gaiser, T., 2012. Analyzing the effects of different soil databases on modeling of hydrological processes and sediment yield in Benin (West Africa). Geoderma 173–174, 61–74. Box, G., Draper, N., 1987. Empirical Model-Building and Response Surfaces. Wiley, Chichester. Bronstert, A., Fritsch, U., Katzenmaier, D., 2001. Quantifizierung des Einflusses der Landnutzung und -bedeckung auf den Hochwasserabfluss in Flussgebieten. Potsdam-Institut für Klimafolgenforschung, Potsdam, 243pp. . Busche, H., Hiepe, C., Diekkrüger, B., 2005. Modelling the effects of land use and climate change on hydrology and soil erosion. In: 3rd International SWAT Conference, pp. 434–443. . Carsel, R.F., Parish, R.S., 1988. Developing joint probability distributions of soil water retention characteristics. Water Resour. Res. 24 (5), 755–769. Christoph, M., Fink, A.H., Paeth, H., Born, K., Kerschgens, M., Piecha, K., 2010. Climate scenarios. In: Speth, P., Christoph, M., Diekkrüger, B. (Eds.), Impacts of Global Change on the Hydrological Cycle in West and Northwest Africa. Springer, Berlin, pp. 402–425. Cullmann, J., Mishra, V., Peters, R., 2006. Flow analysis with WaSiM-ETH – model parameter sensitivity at different scales. Adv. Geosci. 9 (73), 73–77. Ermert, V., Brücher, T., 2008. The climate of Benin (1961–1990). In: Judex, M., Thamm, H.-P. (Eds.), IMPETUS Atlas Benin. Research Results 2000–2007. University of Bonn, Bonn, pp.17–18. . Fass, T., 2004. Hydrogeologie im Aguima Einzugsgebiet in Benin/Westafrika. University of Bonn, 161pp. . Faurè, P., 1977. Carte pédologique de reconnaissance de la République Populaire du Bénin à 1:200,000: Feuille Djougou. Notice Explicative 66. ORSTOM, Paris, 51pp. Faust, D., 1991. Die Böden der Monts Kabyè (N-Togo) – Eigenschaften, Genese und Aspekte ihrer agrarischen Nutzung. Frankfurter Geowissenschaftliche Arbeiten D 13. Institute for Physical Geography of the Johann-Wolfgang-GoetheUniversity, Frankfurt am Main, 174pp. Fink, A.H., Paeth, H., Ermert, V., Pohle, S., Diederich, M., 2010. Meteorological processes influencing the weather and climate of Benin. In: Speth, P., Christoph, M., Diekkrüger, B. (Eds.), Impacts of Global Change on the Hydrological Cycle in West and Northwest Africa. Springer, Berlin, pp. 135–149. Germer, S., Neill, C., Krusche, A., Elsenbeer, H., 2010. Influence of land-use change on near-surface hydrological processes: undisturbed forest to pasture. J. Hydrol. 380 (3–4), 473–480. Giertz, S., 2004. Analyse der hydrologischen Prozesse in den sub-humiden Tropen Westafrikas unter besonderer Berücksichtigung der Landnutzung am Beispiel des Aguima-Einzugsgebietes in Benin. University of Bonn, Bonn, 267pp. .

235

Giertz, S., Junge, B., Diekkrüger, B., 2005. Assessing the effects of land use change on soil physical properties and hydrological processes in the sub-humid tropical environment of West Africa. Phys. Chem. Earth Pt. A/B/C 30 (8–10), 485–496. Giertz, S., Diekkrüger, B., Jaeger, A., Schopp, M., 2006. An interdisciplinary scenario analysis to assess water availability and water consumption in the upper Ouémé catchment in Benin. Adv. Geosci. 9, 3–13. Giertz, S., Hiepe, C., Steup, G., Sintondji, L., Diekkrüger, B., 2010. Hydrological processes and soil degradation in Benin. In: Speth, P., Christoph, M., Diekkrüger, B. (Eds.), Impacts of Global Change on the Hydrological Cycle in West and Northwest Africa. Springer, Berlin, pp. 168–197. Götzinger, J., 2007. Distributed Conceptual Hydrological Modelling – Simulation of Climate, Land Use Change Impact and Uncertainty Analysis. Mitteilungen des Instituts für Wasserbau 146. University of Stuttgart, Stuttgart, 144pp. . Green, W., Ampt, G., 1911. Studies on soil physics: I. The flow of air and water trough soils. J. Agric. Sci. 4 (1), 1–24. Güntner, A., 2002. Large-Scale Hydrological Modelling in the Semi-Arid North-East of Brazil. Potsdam Institute for Climate Impact Research and University of Potsdam, Germany, 148pp. . Hadjer, K., Höllermann, B., Bollig, M., 2010. Social organization, livelihoods, and politics of water management in Benin. In: Speth, P., Christoph, M., Diekkrüger, B. (Eds.), Impacts of Global Change on the Hydrological Cycle in West and Northwest Africa. Springer, Berlin, pp. 286–304. Hagemann, S., 2002. An Improved Land Surface Parameter Dataset for Global and Regional Climate Models. Report 336. Max Planck Institute for Meteorology, Hamburg, 28pp. . Hiepe, C., 2008. Soil Degradation by Water Erosion in a Subhumid West-African Catchment: A Modelling Approach Considering Land Use and Climate Change in Benin. University of Bonn, Germany, 335pp. . Hiepe, C., Diekkrüger, B., 2007. Modelling soil erosion in a sub-humid tropical environment at the regional scale considering land use and climate change. In: 4th International SWAT Conference, pp. 73–80. . Holländer, H.M., Blume, T., Bormann, H., Buytaert, W., Chirico, G.B., Exbrayat, J.-F., Gustafsson, D., Hölzel, H., Kraft, P., Stamm, C., Stoll, S., Blöschl, G., Flühler, H., 2009. Comparative predictions of discharge from an artificial catchment (Chicken Creek) using sparse data. Hydrol. Earth Syst. Sci. 13 (11), 2069–2094. Huisman, J.A., Breuer, L., Bormann, H., Bronstert, A., Croke, B.F.W., Frede, H.-G., Gräff, T., Hubrechts, L., Jakeman, A.J., Kite, G., Leavesley, G., Lanini, J., Lettenmaier, D.P., Lindström, G., Seibert, J., Sivapalan, M.G., Viney, N.R., Willems, P., 2009. Assessing the impact of land use change on hydrology by ensemble modelling (LUCHEM) III: scenario analysis. Adv. Water Resour. 32 (2), 159–170. IPCC, 2007. Climate Change 2007: Synthesis Report. Cambridge University Press, Cambridge. Jacob, D., 2001. A note to the simulation of the annual and interannual variability of the water budget over the Baltic Sea drainage basin. Meteorol. Atmos. Phys. 77 (1–4), 61–73. Judex, M., 2008. Modellierung der Landnutzungsdynamik in Zentralbenin mit dem XULU-Framework. University of Bonn, Bonn, 184pp. . Jung, G., 2006. Regional Climate Change and the Impact on Hydrology in the Volta Basin of West Africa. University of Augsburg, Augsburg, 160pp. . Junge, B., Skowronek, A., 2007. Genesis, properties, classification and assessment of soils in Central Benin, West Africa. Geoderma 139 (3–4), 357–370. Kasei, R., 2010. Modelling Impacts Of Climate Change On Water Resources in the Volta Basin, West Africa. Ecology and Development Series 69. Cuvillier, Göttingen, 195pp. . Le Lay, M., Saulnier, G.-M., Galle, S., Séguis, L., Mètadier, M., Peugeot, C., 2008. Model represenation of the sudanian hydrological processes: application on the Donga catchment (Benin). J. Hydrol. 363 (1–4), 31–41. Leemhuis, C., Erasmi, S., Twele, A., Kreilein, H., Oltchev, A., Gerold, G., 2007. Rainforest conversion in central Sulawesi, Indonesia: recent development and consequences for river discharge and water resources. Erdkunde 61 (3), 284–293. Monteith, J. (Ed.), 1975. Vegetation and the Atmosphere 1. Prinicples. Academic Press, London. Mulindabigwi, V., 2006. Influence des systèmes agraires sur l’utilisation des terroirs, la séquestration du carbone et la sécurité alimentaire dans le bassin versant de l’Ouémé supérieur au Bénin. University of Bonn, Bonn, 253pp. . Nash, J.E., Sutcliffe, J.V., 1970. River flow forecasting through conceptual models: Part I – a discussion of principles. J. Hydrol. 10 (3), 282–290. Niehoff, D., 2001. Modellierung des Einflusses der Landnutzung auf die Hochwasserentstehung in der Mesoskala. University of Potsdam, Potsdam, 165pp. . Orthmann, B., 2005. Vegetation Ecology of a Woodland-Savanna Mosaic in Central Benin (West Africa): Ecosystem Analysis with a Focus on the Impact of Selective Logging. University of Rostock, Rostock, 148pp. . Paeth, H., Diederich, M., 2011. Postprocessing of simulated precipitation for impact research in West Africa. Part II: a weather generator for daily data. Clim. Dyn. 36 (7–8), 1337–1348.

236

T. Cornelissen et al. / Journal of Hydrology 498 (2013) 221–236

Paeth, H., Born, K.V., Heuer, K.O., 2008. Human activity and future climate change. In: Judex, M., Thamm, H.-P. (Eds.), IMPETUS Atlas Benin. Research Results 2000– 2007. University of Bonn, Bonn, pp. 13–14. . Penman, H.L., 1956. Evaporation: an introductory survey. Neth. J. Agric. Sci. 19–29 (87–97), 151–153. Perrin, C., Michel, C., Andréassian, V., 2003. Improvement of a parsimonious model for streamflow simulation. J. Hydrol. 279 (1–4), 275–289. Peschke, G., 1977. Ein zweistufiges Modell der Infiltration von Regen in geschichtete Böden. Acta Hydrophys. 22 (1), 39–48. Pöhler, H.A., 2006. Anpassung von WaSiM-ETH und die Erstellung und Berechnung von Landnutzungs- und Klimaszenarien für die Niederschlag-AbflussModellierung am Beispiel des Osterzgebirges. University of Freiberg, Freiberg, 148pp. . Rawls, W.J., Brakensiek, D.L., 1995. Prediction of soil water properties for hydrologic modeling. In: Jones, E., Ward, T.J. (Eds.), Proceedings of the Symposium Watershed Management in the Eighties, Denver, pp. 293–399. Reichert, B., Jaeger, A., 2010. Socio-econmoic scenarios. In: Speth, P., Christoph, M., Diekkrüger, B. (Eds.), Impacts of Global Change on the Hydrological Cycle in West and Northwest Africa. Springer, Berlin, pp. 426–441. Rohdenburg, H., 1969. Hangpedimentation und Klimawechsel als wichtigste Faktoren der Flächen- und Stufenbildung in den wechselfeuchten Tropen an Beispielen aus Westafrika, besonders aus dem Schichtstufenland SüdostNigerias. In: Fölster, H., Rohdenburg, H. (Eds.), Beiträge zur Geomorphologie der wechselfeuchten Tropen. Giessener Geographische Schriften 20. University of Giessen, Giessen, pp. 7–152. Runge, J., 1990. Morphogenese und Morphodynamik in Nord-Togo (9°–11°N) unter dem Einfluss spätquartären Klimawandels. Göttinger Geographische Abhandlungen 90. Goltze, Göttingen, 115pp. Schulla, J., 1997. Flussgebietsmodellierung und Wasserhaushalts-Simulation mit WaSiM-Modellbeschreibung. Eidgenössische technische Hochschule Zürich, Zürich, 189pp. .

Schulla, J., Jasper, K., 2007. Model Description WaSiM-ETH. . SCS, 1972. Estimation of Direct Runoff from Storm Rainfall. National Engineering Handbook, Section 4 – Hydrology. USDA, 10.1–10.24pp. ftp:// ftp1.co.mecklenburg.nc.us/luesa/stormwater/Mapping_Briar_LSC_Preliminary/ References/NRCS-II_Engineering_Handbook.pdf. Scurlock, J.M.O., Asner, G.P., Gower, S.T., 2001. Global Leaf Area Index Data from Field Measurement, 1932–2000. The Oak Ridge National Laboratory Distributed Active Archive Center, Oak Ridge, Tennessee, USA. Séguis, L., Kamagaté, B., Favreau, G., Descloitres, M., Seidel, J.-L., Galle, S., Peugeot, C., Gosset, M., Le Barbè, L., Malinur, F., Vav Exter, S., Arjounin, M., Boubkraoui, S., Wubda, M., 2011. Origins of streamflow in a crystalline basement catchment in a sub-humid Sudanian zone: the Donga Basin (Benin, West Africa) inter-annual variability of water budget. J. Hydrol. 402 (1–2), 1–13. Sintondji, L., 2005. Modelling the Rainfall-Runoff Process in the Upper Ouémé Catchment (Térou in Benin Republic) in a Context of Global Change: Extrapolation from the Local to the Regional Scale. Shaker, Aachen, 226pp. Steyaert, L., Knox, R., 2008. Reconstructed historical land cover and biophysical parameters for studies of land-atmosphere interactions within the eastern United States. J. Geophys. Res., 113(D2). Varado, N., Braud, I., Galle, S., Le Lay, M., Séguis, L., Kamagate, B., Depratere, C., 2006. Multi-criteria assessment of the representative elementary watershed approach on the Donga catchment (Benin) using a downward approach of model complexity. Hydrol. Earth Syst. Sci. 10 (3), 421–442. Viney, N.R., Bormann, H., Breuer, L., Bronstert, A., Croke, B.F.W., Frede, H., Gräff, T., Hubrechts, L., Huisman, J.A., Jakeman, A.J., Kite, G.W., Lanini, J., Leavesley, G., Lettenmaier, D.P., Lindström, G., Seibert, J., Sivapalan, M., Willems, P., 2009. Assessing the impact of land use change on hydrology by ensemble modelling (LUCHEM) II: ensemble combinations and predictions. Adv. Water Resour. 32 (2), 147–158. Wagner, S., Kunstmann, H., Bárdossy, A., 2006. Model distributed water balance monitoring of the White Volta catchment in West Africa through coupled meteorological, hydrological simulations. Adv. Geosci. 9, 39–44.