Impact of dumped sediment structures on hydrological modelling in the artificial Chicken Creek catchment, Germany

Journal of Hydrology 477 (2013) 189–202

Contents lists available at SciVerse ScienceDirect

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

Impact of dumped sediment structures on hydrological modelling in the artificial Chicken Creek catchment, Germany Herwig Hölzel a,⇑, Bernd Diekkrüger b, Detlef Biemelt a, Anne Gädeke a a Chair of Hydrology and Water Resources Management, Faculty of Environmental Sciences and Process Engineering, Brandenburg University of Technology, Konrad Wachsmann Allee 6, 03046 Cottbus, Germany b Hydrological Research Group, Department of Geography, University of Bonn, Meckenheimer Allee 166, 53115 Bonn, Germany

a r t i c l e

i n f o

Article history: Received 5 January 2012 Received in revised form 23 October 2012 Accepted 15 November 2012 Available online 23 November 2012 This manuscript was handled by Andras Bardossy, Editor-in-Chief, with the assistance of Ezio Todini, Associate Editor Keywords: Process-based hydrological modelling Hydrological pattern and processes Structure-process interactions Initial ecosystem development

s u m m a r y Revealing the hydrological impact of sediment structures promises a better understanding of the influence of the spatial variability of sediment properties on the hydrological patterns and processes at the catchment scale. To improve the knowledge of structure-process interactions in initial ecosystems, the 6-ha artificial Chicken Creek Catchment in Germany was investigated by the Transregional Collaborative Research Centre 38 (SFB/TRR 38). Sediment structures called pour-ribs, which are dumped by stackers during the construction process, lead to differently compacted sediment zones, which increase the spatial variability of sediments’ hydraulic properties. Although levelled afterwards by bulldozers, the majority of these structures remain in the subsurface. To analyse the effects of pour-ribs on the hydrological catchment’s behaviour, the process-based spatially distributed Water balance Simulation Model (WaSiM-ETH) was applied. The results show that the consideration of pour-ribs improves the runoff simulation and significantly affects the simulated soil moisture patterns and, thereby, the initial stage of the ecosystem development. Compacted zones act as hydraulic barriers and inhibit subsurface lateral water flow, whereas non-compacted zones constitute areas with increased water storage capacity. Both effects cause reduced catchment runoff. Moreover, disregarding of the pour-ribs was identified as a source of model uncertainty in previous studies. A further outcome of this study is the importance of a global sensitivity analysis as a tool for model improvement. Finally, the results stress the importance of considering the variability of sediment properties for hydrological modelling at the catchment scale. Ó 2012 Elsevier B.V. All rights reserved.

1. Introduction Hydrological models can be used for testing hypotheses of structure-process interactions at the catchment scale and therefore support fundamentally oriented ecological research. As a prerequisite, the spatial variability of hydrological processes and patterns must be considered during hydrological modelling (Grayson and Blöschl, 2000). Additionally, Zehe et al. (2005, 2006) pointed out that a better representation of major catchment structures and processes is crucial for reducing model uncertainty especially for process-based distributed models. This point has already been stressed in several studies. For example, Herbst and Diekkrüger (2006) demonstrated the importance of the spatial variability of soil hydraulic parameters on simulated runoff in a 28-ha catchment, and Sciuto and Diekkrüger (2010b) studied their effects on simulated water budget in a 27-ha catchment. Merz and Bárdossy

⇑ Corresponding author. Tel.: +49 0355 69 5065. E-mail addresses: [email protected] (H. Hölzel), [email protected] (B. Diekkrüger), [email protected] (D. Biemelt), [email protected] (A. Gädeke). 0022-1694/$ - see front matter Ó 2012 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.jhydrol.2012.11.029

(1998) investigated road effects and the spatial distribution of soil parameters and soil moisture on simulated runoff in a 31-ha catchment. The impact of roads on runoff, soil erosion, and sedimentation was simulated by Hölzel and Diekkrüger (2012) in a 21-ha catchment. However, the parameter sensitivity to the available data resolution and spatial distribution of properties is model dependent (Bormann et al., 2009). Understanding the influence of the spatial variability of sediment properties on hydrological patterns and processes at the catchment scale can also help to answer questions about driving forces in initial ecosystems, about which knowledge is still rather sparse (Schaaf et al., 2011). Several projects have continuously investigated mature ecosystems by long-term monitoring programs, such as TERENO (Zacharias et al., 2011), LTER (Long Term Ecological Research Network, 2007), or NEON (Pennisi, 2010). Only a few studies have focused on initial ecosystems that have often not yet reached steady-state conditions (Schaaf et al., 2011). Such studies deal with destroyed regions due to volcanic activities, such as Mount St. Helens (Bishop, 2002), or newly created landscapes, such as Surtsey Island (Fridriksson, 2005). Therefore, knowledge gaps about structure-process interactions in initial ecosystems are still evident (Schaaf et al., 2011).

190

H. Hölzel et al. / Journal of Hydrology 477 (2013) 189–202

To fill these gaps, the Transregional Collaborative Research Centre (SFB/TRR38) conducted an investigation of the structures and processes of the initial ecosystem development in the 6-ha artificial Chicken Creek catchment in the lower lusatian lignite mining district in Germany (Gerwin et al., 2009). To reveal the role of the initial patterns and processes for ecosystem development, comprehensive measurement devices were installed. The catchment remains unplanted and is fenced to guarantee an almost undisturbed development. Because some processes, such as subsurface flow or evapotranspiration, cannot be observed directly in the Chicken Creek catchment, hydrological modelling studies were conducted to supplement the measurements (Bormann, 2011; Hofer et al., 2011; Holländer et al., 2009; Hölzel et al., 2011). The results of the monitoring, published in several studies (Gerwin et al., 2010; Mazur et al., 2011; Schaaf et al., 2010; Schneider et al., 2011; Zaplata et al., 2011), provide a sufficient database for a wide range of hydrological model-based research. Hofer et al. (2011) simulated the impact of subsurface flow-paths on runoff and showed that the development of a subsurface flow-path network is crucial for hydrological connectivity. He also revealed that the observed nonlinear runoff behaviour is caused by threshold processes. Holländer et al. (2009) investigated the influence of different modellers and their assumptions on runoff and water budget simulations by an a priori model comparison and stressed the importance of the human factor. The comparison reveals distinct differences of model simulations, although all modellers used the same, though limited data, information, and boundary conditions. Simulated surface flow was underestimated, whereas total runoff and soil moisture were overestimated by the majority of the models (Holländer et al., 2009). The simulations were improved using additional information on an initial soil moisture deficit and a surface-sealing layer with reduced infiltration capacity. One of the results of the increase in information is that the high variation of the predicted water budget components between the models decreases (Bormann et al., 2011). Nevertheless, the observed runoff was still overestimated (Bormann, 2011; Hölzel et al., 2011). One reason for overestimating the runoff observations could be that artificial sediment structures, namely, spoil-cones dumped by stackers during catchment construction, were disregarded by the models. The existence of spoil-cones in the catchment and their importance for the ecosystem’s development and hydrological processes was mentioned by Gerwin et al. (2009) and Maurer et al. (2011). The impact of spoil-cones on physical and hydraulic sediment properties was investigated by Buczko et al. (2001) in lusatian lignite open-cast mine dumps. One outcome of this study was that the dumping processes lead to the spatial segregation of sediment compaction with higher bulk density in the core and lower parts of the outer sleeves of the spoil-cones. This sediment segregation affects the hydraulic conductivity. Maurer et al. (2011) investigated spoil-cones comparable to the Chicken Creek catchment and also found higher bulk densities at the centres and lower bulk densities at the flanks of the cones. In the Chicken Creek catchment, these cones are strung together and form pourribs running transverse to the flow-paths (Gerwin et al., 2009). Therefore, a hydrological impact can be expected at the catchment scale. The objective of this study is to investigate the impact of pourribs on the hydrological behaviour of the Chicken Creek catchment, whereby driving forces for the initial ecosystem development may be revealed. Moreover, a better understanding of the influence of the spatial variability of sediment properties on the hydrological patterns and processes at the catchment scale can be achieved in general. To accomplish this goal, pour-ribs were implemented in a process-based, spatially distributed hydrological model to study the impact on hydrological processes (e.g., runoff) and patterns (e.g., soil moisture).

2. Research area The catchment is located approximately 150 km southeast of Berlin, Germany, and was constructed in the post-mining recultivation area of the still active open-cast Welzow-South lignite mine (Fig. 1). The climate conditions are sub-continental, with an average precipitation of 559 mm/a and a yearly average air temperature of 8.9 °C (Gerwin et al., 2009). The 6-ha area consists of two sediment layers, a clay base layer in the underground overlaid by a sandy top layer (Fig. 2). The clay basement is 2–3 m thick and functions as an aquiclude inhibiting seeping water (Kendzia et al., 2008). Two structures were formed at the basement, a subterranean clay dam to stabilise the sandy top layer and a depression to contain a later pond. The sandy top layer is shaped as a slight hillslope with an average inclination of 3.5° and functions as an aquifer (Gerwin et al., 2010). The mean soil texture of the top layer is composed of quaternary sand (84%), silt (9%), and clay (7%) (Maurer et al., 2011). Therefore, the eastern part of the catchment consists of loamy sand whereas the western part is composed of sandy loam. The strip-wise deposition of sediments was conducted consecutively by the stacker (Fig. 3) originates from different source areas. Consequently, the soil texture differs between the eastern and western parts, resulting in two spatial soil clusters (Fig. 4) (Schaaf et al., 2010). The eastern soil cluster consists of 87% sand, 8% silt, and 5% clay; the western soil cluster consists of 82% sand, 11% silt, and 7% clay (Maurer et al., 2011). Despite the sandy substrates, the infiltration capacity is reduced due to a comprehensive surface-sealing layer, which is the result of interactions between abiotic and biotic structures at the early development stage (Fischer et al., 2010; Gerwin et al., 2011) and is responsible for a flashy runoff response to rainfall events. A gully network quickly established itself and drains the catchment because the initially bare, non-vegetated surface as well as the absence of preferential flow-paths caused considerable rill erosion after extreme rainfall events (Gerwin et al., 2011). Due to the higher silt and clay content, the gullies are smaller and deeper in the western part of the catchment. Plant immigration processes led to the spreading of vegetation. In 2006, vegetation coverage was less than 1.5%, and in 2007, it was already 7% and 13% by 2008 (Mazur et al., 2011; Zaplata et al., 2011). In 2009, the vegetation coverage reached nearly 30%. With spreading vegetation, rill erosion is reduced and transpiration processes becomes increasingly important. The catchment has been intensively monitored since the construction was completed in August 2005. Detailed information about the catchment’s construction, initial properties, and development as well as the monitoring program are provided by Gerwin et al. (2010), Elmer et al. (2011), and Schaaf et al. (2010). 3. Artificial sediment structures Due to the machinery used for the catchment’s construction, the sandy top layer is partly composed of artificial sediment structures, the so-called pour-ribs. The pour-ribs consist of individual spoilcones with a width of 3-5 m. These pour-ribs were dumped by a stacker strip-wise in the eastern and the western parts of the catchment above the clay dam (Gerwin et al., 2010) (Fig. 5, left). The orientation of the pour-ribs is transverse to the flow-paths, with a central trench between them (Maurer et al., 2011). The trench and the lower part of the catchment below the clay dam were filled with sediments from both sides by bulldozers. Thereby, the spoil-cones heads were cut away, and the pour-ribs were levelled, but the majority of the structures still remain in the subsurface (Gerwin et al., 2010). Observations in the Chicken Creek catchment show that the soil textures of the spoil-cones differ. However, there is no information on whether differences exist in their bulk densities. Measurements

H. Hölzel et al. / Journal of Hydrology 477 (2013) 189–202

191

Fig. 1. Location of the research area southeast of Berlin, Germany.

Fig. 2. Cross-sectional sketch of the Chicken Creek catchment (SFB/TRR 38, changed).

Fig. 4. Soil cluster and soil thickness.

Fig. 3. Schematic sketch of the sediment dumping process by the stacker (Vattenfall Europe Mining AG, changed).

crease of bulk density up to 1.84 g/cm3 was assumed. Compaction is caused during the dumping of sediment by the stacker (Buczko et al., 2001) and the artificial compaction of sediments by bulldozers (Maurer et al., 2011). Based on assumptions on the bulk density and geometry of the pour-ribs (Fig. 5, left), a map with spatially differentiated bulk densities was created (Fig. 5, right). Because of the dumping process by the stacker, an alternation of compacted and non-compacted sediment conditions was assumed for the pourribs. Compacted sediment conditions were also assumed for the trench and below the clay wall because these regions were mainly affected by bulldozers. However, to avoid interventions inside the catchment, these assumptions cannot be verified by observations. 4. Methods

conducted by Maurer et al. (2011) at comparable test sites suggest that there is a spatial variation of bulk density in the Chicken Creek catchment. Based on this result, they estimated the initial bulk density of the sediments in the Chicken Creek catchment to be 1.45 g/cm3. Additionally, due to compaction processes, a zonal in-

4.1. Hydrological model In this study, the process-based, spatially distributed Water balance Simulation Model (WaSiM-ETH) developed by Schulla and

192

H. Hölzel et al. / Journal of Hydrology 477 (2013) 189–202

Fig. 5. Aerial photography from October 2004 with derivation of pour-rib geometry (Vattenfall Europe Mining AG, 2004, in Gerwin et al. (2010) (left) and the considered pour-ribs for modelling based on aerial photography (right).

Jasper (2007) was used. This model is based on an existing model configuration from a prior model comparison study (Holländer et al., 2009; Hölzel et al., 2011). The major improvement was achieved by replacing the conceptual groundwater model by a process-based 2-dimensional groundwater model. Originally, WaSiMETH was developed to investigate climate change impacts on hydrology in alpine regions (Schulla, 1997). Since then, the model has been substantially improved, successfully applied in various contexts, and transferred to other regions, such as flood forecasting in alpine catchments (e.g., Jasper, 2005), soil moisture simulations in low mountain ranges (e.g., Hölzel and Diekkrüger, 2012), and impact studies in lowland regions (e.g., Bronstert, 2005). Depending on data availability, the model can be used for process-based simulations of hydrological patterns and processes over a wide range of spatiotemporal scales. Potential evapotranspiration is calculated using the Penman– Monteith equation depending on vegetation coverage (Monteith and Unsworth, 1990), whereas for real evapotranspiration, a suction-dependent linear reduction approach is applied. Interception is simulated using a Leaf Area Index (LAI) dependent bucket approach, while snow accumulation and snowmelt is based on a temperature-index approach (Schulla and Jasper, 2007). Infiltration is calculated with a modified Green-Ampt approach (Peschke, 1987), while the Richards’ equation with parameterisation according to van Genuchten (1980) is used for simulating the water dynamics in the unsaturated soil zone. Groundwater flow is described 2-dimensionally using Darcy’s flow equation solved with the Gauss–Seidel-Algorithm (Schulla and Jasper, 2007). Compared with the conceptual linear storage approach often used when applying WaSiM-ETH, the process-based 2-dimensional approach enables the lateral routing of base flow, which is a prerequisite for simulating the effects of subsurface sediment structures on the hydrological catchment behaviour. Finally, flood routing is calculated using a kinematic wave approach (Lighthill and Whitham, 1955) with flow velocity according to the Manning–Stricker equation. 4.2. Database and model parameterisation The modelling is based on data from the monitoring program of the SFB/TRR38:  hourly measured precipitation, air temperature, air humidity, wind velocity, and global radiation at a local climate station from 2006 to 2010,

 digital elevation model with a 1-m spatial resolution of the clay basement and the sandy top layer from 2008,  soil texture and bulk density according to Maurer et al. (2011),  saturated hydraulic conductivity based on 22 undisturbed surface infiltrometer measurements,  yearly measurements of mean vegetation coverage at approximately 100 plots of 25 m2 each from 2006 to 2010. Additional data from the monitoring program of the SFB/TRR38 were used for parameter and model evaluation:  daily observed runoff as pond inflow (calculated based on measured precipitation and pond water level fluctuations) from 2006 to 2010,  daily measured soil moisture in different depths at two soil pits (Fig. 4) in the eastern and western parts of the catchment from July 2008 to 2010,  laboratory-based measured water retention curves at two soil pits (Fig. 4) in the eastern and western parts of the catchment at different soil depths,  monthly manually measured groundwater levels at 15 plots from 2006 to 2010. The relief data were spatially aggregated to a 5-m raster resolution. Therefore, small erosion gullies cannot be considered by the model. The simulation period depends on the existence of the catchment and spans a time period of 5 years from January 2006 to December 2010 in hourly intervals. The pond in the lower part of the catchment can be neglected because the study focuses on the hydrological impact of the pour-ribs situated upslope the clay wall. To simulate the spread of vegetation starting in 2008, two model configurations with different vegetation parameters were coupled offline. This was necessary, because only the seasonal, not the interannual, development of vegetation can be defined by the user of WaSiM-ETH. Previous to 2008, bare soils and thus only evaporation were assumed, while afterwards, standard parameters for extensive grassland adjusted by the measured average of the vegetation coverage were used. Despite the availability of spatially distributed data concerning vegetation coverage, a homogeneous distribution was assumed, and the clustering of vegetation was therefore disregarded. Model parameters could not sufficiently be determined by the data provided so that standard parameters of WaSiM-ETH were applied. The consideration of a spatial distribution may therefore lead to model uncertainty. Moreover, the spatial variability of vegetation was comparably low at the

H. Hölzel et al. / Journal of Hydrology 477 (2013) 189–202

beginning and does not influence the spatial patterns of soil moisture. Soil hydraulic parameters were calculated with the pedotransfer functions appropriate for sandy soils (Schaap et al., 2001) based on the measured soil texture and bulk density. Differences between the western and the eastern parts of the catchment were considered by two spatial soil clusters (Fig. 4). Soil thickness was derived based on the intersection of the digital elevation model of the sandy top layer and the clay basement. A 1-cm-thick surface soil layer of was implemented to represent the surface sealing with reduced infiltration capacity. The median of 22 infiltration measurements (10.2 cm/d) was used as the saturated hydraulic conductivity of this surface sealing. The initial conditions for WaSiM-ETH must be calculated internally by the model and cannot be provided manually. They were therefore estimated by a spinup run across the entire model period, with internally defined steady-state conditions. To consider the initial soil moisture deficit caused by the dumped sediments (Holländer et al., 2009), the spin-up run was conducted without precipitation. Therefore, the initial soil moisture decreases from 24.5 Vol.% to 13.4 Vol.%. Initially, the standard parameters of WaSiM-ETH were used for conceptual model parameters that are usually adjusted by inverse modelling. To consider the effect of the pour-ribs on soil hydraulic properties, the bulk density was used as key parameter. At first, the soil hydraulic parameters were calculated based on the initial bulk density (1.45 g/cm3) and compacted sediment conditions (1.84 g/ cm3). In addition, the soil hydraulic parameters were distinguished between the eastern and the western soil clusters (Table 1). These effects become visible when analysing the calculated water retention curves (Fig. 6). The selected PTF is suitable for sandy soils (Schaap et al., 2001) and the calculated retention curves mostly conform with the range of the measurements. Then, the non-compacted regions in the pour-rib map (Fig. 5, right) were parameterised based on the initial bulk density. For compacted regions, the bulk density for the compacted sediment conditions was used. Finally, the pour-rib map was intersected with the soil map (Fig. 4). In this way, both the soil clusters and the pour-ribs could be considered. For the pour-ribs, a constant width of 5 m was assumed, limited by the spatial model resolution. Thus, the differences of soil hydraulic properties could only be considered between, but not inside, the pour-ribs. The total catchment runoff is measured by a combined V-profile and a rectangle profile at the outlet of the pond. The pond itself works as storage. This is of major relevance for studies focusing on the fast runoff components. Therefore, the fast inflow into the pond which was directly linked to precipitation events was calculated based on the change in pond storage, the precipitation into the pond, and the pond outflow. The storage capacity of the pond is determined by a water level-volume relation. The pond was surveyed in 2006, 2008 and 2010. The water level in the pond was equipped with a pressure transducer (SEBA, Insider) measuring with a frequency of 10 min.

Table 1 Soil hydraulic parameters of the pour-rib model configuration based on different bulk densities derived with the pedo-transfer function of Schaap et al. (2001). Soil hydraulic parameter

Western soil cluster

Eastern soil cluster

Bulk density [g/cm3] Saturated soil water content [cm3/cm3] Residual soil water content [cm3/cm3] Van Genuchten alpha [1/m] Van Genuchten n [–] Saturated hydraulic conductivity [cm/d]

1.45 0.41 0.05 3.61 1.85 153.19

1.45 0.40 0.05 3.60 2.19 259.06

1.84 0.30 0.04 4.28 1.65 36.22

1.84 0.29 0.04 3.93 1.97 69.87

193

Fig. 6. Calculated range of 20 laboratory based measurements of water retention curves, ten curves for each soil pit.

4.3. Sensitivity analysis A global sensitivity analysis (GSA) according to the GLUE methodology (Beven and Freer, 2001) was used to evaluate the effects of soil, land use, and conceptual parameters on runoff simulation. In contrast to a simple local sensitivity analysis, parameter interactions were taken into account to enable a systematic analysis of the parameters’ influence on the model outcomes (Beven, 2009). The GSA was facilitated by the program GLUEWIN (Ratto and Saltelli, 2001) and was performed for a modelling period of 3 years, from 2008 to 2010. This period was chosen because vegetation was of minor importance before 2008. To guarantee consistent modelling despite the parameter variation, a uniform soil distribution was assumed. The reason for using a uniform soil distribution was simply necessary for a consistent modelling. Using a heterogeneous soil distribution may lead to model instabilities. This simplification limits the validity of the GSA. However, the main outcome (high model equifinality) will be the same. Altogether, 15 model parameters were varied within real boundaries facilitated by the program SIMLAB (IPCS, 2003) (Table 2). The lower and upper limits of the parameter ranges were determined based on the range of default values by WaSiM-ETH. To reduce the sample amount, the stratified Latin Hypercube approach (McKay, 1988), a classified Monte Carlo method that enables a sample reduction based on parameter distribution functions (Melching, 1995) was used. This approach is recommended for models with many parameters (Christiaens and Feyen, 2001). Because the parameter distribution functions were unknown, a homogeneous distribution was assumed for all parameters. According to IPCS (2003), the number of samples (model runs) should at least exceed the number of parameters by 10 times. Overall, 300 samples (20 times the number of model parameters) were generated by SIMLAB, covering the entire parameter space. To evaluate the model performance, the coefficient of determination (r2) between the simulated and observed discharge was calculated as efficiency criteria. For the identification of parameters with a global optimum concerning model performance suited for calibration, the Hornberger–Spear–Young (HSY) (Beven, 2009) approach was used. The HSY approach is based on the comparison between cumulative density functions of model runs with high (behavioural simulations) and low model performance (nonbehavioural simulations). To distinguish between behavioural and non-behavioural simulations, a threshold value of r2 = 0.6 was used.

194

H. Hölzel et al. / Journal of Hydrology 477 (2013) 189–202

Table 2 Model parameters considered in the global sensitivity analysis. Model parameter

Parameter class

Lower boundary

Upper boundary

Saturated soil water content [cm3/ cm3] Van Genuchten alpha [1/m] Van Genuchten n [–] Saturated hydraulic conductivity [cm/d] Albedo [–] Soil surface resistance [s/m] Vegetation surface resistance [s/m] Leaf area index [m2/m2] Vegetation high [m] Vegetation coverage [–] Root depth [m] Recession constant for surface flow [h] Recession constant for interflow [h] Drainage density [1/m] Reduction of saturated hydraulic conductivity with soil depth [–]

Soil

0.05

0.55

Soil Soil Soil

1 1 1

14 3 300

Vegetation Vegetation Vegetation Vegetation Vegetation Vegetation Vegetation Conceptual Conceptual Conceptual Conceptual

0.1 100 50 1 0.01 0.01 0.01 1 1 1 0.1

0.9 1000 100 12 10 1 1 120 120 35 1

5. Results 5.1. Parameter sensitivity The dotty plots of the GSA reveal that the majority of parameters lack a clear global optimum (Fig. 7, bottom). To represent a parameter without a clear global optimum, the saturated hydraulic conductivity (ksat) is presented as proxy (Fig. 7, bottom right). For the whole range of ksat-values, both high and low model performance is obtained, and no trend can be observed. Only one conceptual parameter, namely, the recession constant for surface flow (Qdir), shows a significant trend, and a real global optimum can be observed. For lower values of Qdir, the model performance increases (Fig. 7, bottom left). It must be kept in mind, however, that the dotty plots only provide a 2-dimensional view of the multidimensional parameter space, which exacerbates the interpretation of the results. Therefore, the cumulative density functions of the behavioural and non-behavioural simulations were also analysed (Fig. 7, top). The comparison between behavioural and non-behavioural simulations shows a clear distinction in the variation of Qdir (Fig. 7, top left). Therefore, this parameter is suitable for calibration. As consequence of the GSA, Qdir was reduced from 24 h (model standard parameter) to 1 h for further modelling because a better model performance could be expected. It must be taken into account that Qdir is a conceptual parameter, and the unit hour (h) does not have direct physical meaning. None of the other parameters showed a clear distinction between the behavioural and non-behavioural simulations, as represented by the ksat (Fig. 7, top right). Due to high equifinality of these parameters, a calibration is not recommended. As an outcome of the GSA, all other parameters were kept uncalibrated. 5.2. Simulation of runoff, soil moisture, groundwater levels, and water budget In contrast to the sensitivity analysis, where homogenous soil conditions were assumed, the following hydrological simulations are based on a heterogeneous soil distribution. By considering heterogeneous soil distribution, the spatial variability also increases which may lead to higher model equifinality compared to the sensitivity analysis conducted. Moreover, the effects of pour-ribs were also considered. Observed and simulated runoff directly react on precipitation and are therefore characterised by high peaks and short concentration

and recession times (Fig. 8, left). High runoff was observed especially after snowmelt and convective extreme rainfall events in summer. The observed runoff between the runoff peaks is relatively low and sometimes even interrupted (no flow conditions). Based on visual evaluation, the observed runoff behaviour is satisfactorily reflected by the model, although not every peak is exactly matched. Some simulated runoff peaks underestimate the observations, whereas others overestimate them. The uncertainty bounds of the 300 model runs of the GSA reveal that the model is suitable for adequately simulating the runoff peaks but has problems reflecting low and no flow conditions (Fig. 8, right). No flow was measured in 67% and runoff was observed in 33% of the days during the time period of the GSA. However, in 71% of the days with observed runoff, the runoff is within the uncertainty bound. It must be taken into account that the observed runoff was calculated based on measured precipitation and pond water level fluctuations and may therefore also be uncertain. ‘‘Observed’’ runoff was computed from the change in water volume of the pond on rainy days. A slight daily runoff, as often simulated, cannot be considered by this method and was therefore set to no flow. Considering the uncertainties in the discharge measurements, in our view the results are satisfying for this study. The model performance of the runoff simulation was evaluated using the coefficient of determination (r2), the coefficient of model efficiency (cme) according to Nash and Sutcliffe (1970), and the mass balance error (mbe [%]). Applying the model configuration for the period 2006–2007 (without vegetation) results in a higher model performance (r2 = 0.75, cme = 0.69, mbe = 1.4%) compared with the time period with vegetation (r2 = 0.59, cme = 0.51, mbe = 6.5%). The model performance obtained can be partly attributed to the reduction of the conceptual parameter Qdir as result of the GSA. Using the model standard value for Qdir as 24 h instead of the reduced value of 1 h, the calculated values of r2 and cme were always less than 0.42. The mbe was barely affected because the reduction of Qdir influences only the runoff concentration. Surface flow is mainly responsible for the simulated runoff dynamic, and with 63% of the total discharge, it is the dominant runoff component. The runoff coefficient, the ratio between runoff and precipitation, amounts to 24%. The interflow and base flow are less important, at 12% and 25%, respectively. However, a shift between the runoff components can be observed. The base flow increases over time, from less than 1% in 2006 up to 33% in 2010, whereas the surface runoff is reduced from 70% to 66% and the interflow from 30% to 1%. In addition to runoff, simulated soil moisture and groundwater levels were also used for the model evaluation. The observed soil moisture dynamic depends on the precipitation (Fig. 9, left). The measured and simulated soil moisture differs quantitatively depending on soil depth. Observed and simulated soil moisture dynamic decrease with increasing soil depth. Therefore, calculated averages of the measurements were used for the comparison of the measured and simulated soil moisture. The simulated soil moisture matches well with the calculated average of the measurements (Fig. 8, left). The coefficient of determination (r2) amounts to 0.7 (Fig. 10, middle), which is higher compared with the cme (0.29) and the root mean square error (rmse = 2.79). This result is an indication that the model reflects the soil moisture dynamic better than the quantity. The gradual increase of the soil moisture as well as the seasonal dynamic with higher values in winter and spring and lower values in summer and autumn is based on a visual plot interpretation satisfactorily reflected by the model. Similar to the soil moisture, the measured groundwater levels also fluctuate seasonally and are characterised by a gradual increase over time (Fig. 9, right). Two measurement plots were located in erosion gullies, which explains the continuously high groundwater levels ob-

H. Hölzel et al. / Journal of Hydrology 477 (2013) 189–202

195

Fig. 7. Cumulative density functions of behavioural (r2> 0.6) and non-behavioural (r2 < 0.6) simulations for a model parameter with (top left) and without (top right) a clear global optimum. Dotty plots of model performance concerning runoff simulation for a model parameter with (bottom left) and without (bottom right) a clear global optimum. Each point represents one model simulation.

Fig. 8. Observed and simulated runoff (left) and uncertainty bounds of runoff simulation based on the global sensitivity analysis (300 model runs, right).

served at the two measurement plots. However, the calculated averages of observed and simulated groundwater levels clearly indicate the formation of a groundwater body. Corresponding to the simulated soil moisture, the average of the simulated groundwater levels reflects well the average of the measurements. Compared with the soil moisture, a higher model agreement with the observed average at r2 = 0.8 (Fig. 10, right), cme = 0.64, and rmse = 0.14 was achieved. Due to deviations between simulated and observed runoff peaks, r2 for the runoff simulation is lower than for simulated soil moisture and groundwater levels (Fig. 10, left). In addition to the good agreement in simulating runoff, the analysis of the soil moisture and groundwater levels proves the good model quality, which is further evidence that the evapotranspiration and water budget (Table 3) are sufficiently simulated. The

annual average of the simulated potential evapotranspiration for 2006–2010 accounts for 662 mm/a, whereas the real evapotranspiration accounts for only approximately half that value (322 mm/a). Higher values of real evapotranspiration were simulated for 2008– 2010 compared with 2006 and 2007. The increase is caused by gradually increasing soil moisture and the consideration of transpiration starting from 2008. Based on the yearly simulated water budgets (Table 3), the average water budget for the entire modelling period can be calculated. The yearly average measured precipitation amounts to 562 mm/a, the simulated runoff to 123 mm/a, the real evapotranspiration to 322 mm/a, and the calculated storage change to 117 mm/a. Based on the measured precipitation and the simulated potential evapotranspiration (662 mm/a), the climatic water budget amounts to 100 mm/a.

196

H. Hölzel et al. / Journal of Hydrology 477 (2013) 189–202

Fig. 9. Observed and simulated soil moisture (left) and groundwater levels (right).

Fig. 10. Scatterplots of simulated and observed runoff from March 2008 to 2010 (left), average soil moisture from July 2008 to 2010 (middle), and average groundwater levels from 2008 to 2010 (right).

Table 3 Simulated water budget of the pour-rib model configuration. Hydrological year

Precipitation [mm/a]

Pot. Evapotranspiration [mm/a]

Real evapotranspiration [mm/a]

Runoff [mm/a]

Storage change [mm/a]

2006 2007 2008 2009 2010

373 565 605 549 716

691 682 686 640 613

207 280 362 381 380

57 124 96 111 225

109 161 147 57 111

5.3. Impact of pour-ribs on the simulated hydrological processes and patterns To demonstrate the effects of artificial sediment structures, the model configuration with the pour-ribs is compared with two configurations where the pour-ribs are neglected. For both configurations without pour-ribs, a spatially homogeneous bulk density based on the initial (1.45 g/cm3) and compacted sediment conditions (1.84 g/cm3) were assumed and are hereafter called noncompacted and compacted configurations, respectively. The consideration of the pour-ribs improved runoff simulation, as shown by comparing the model configurations with and without the pour-ribs (Fig. 11). The non-compacted and compacted configurations both overestimated the observed runoff. Therefore, the model performance of the configuration without the pour-ribs is lower than with the pour-ribs (Table 4). Only the mass balance er-

ror decreased (by 1%) for the time period after 2008 using the configuration with compacted sediment conditions. The compacted configuration clearly obtains a better model performance compared with the non-compacted configuration for the runoff simulation. This result is an indication of the importance of the artificial compaction processes of sediments for the hydrological behaviour of the Chicken Creek catchment, and it affects the yearly average water budget (Table 5). Comparing the non-compacted and compacted configurations, the simulated runoff is reduced when considering the pour-ribs. On average, 55 mm/a (45%) more runoff was simulated for the non-compacted configuration and 17 mm/ a (14%) for the compacted configuration compared with the pour-rib configuration. As expected, the potential evapotranspiration is not affected by the pour-ribs. The real evapotranspiration is reduced by 32 mm/a (10%) for the non-compacted and increased by 2 mm/a (0.5%) for the compacted configurations compared with

197

H. Hölzel et al. / Journal of Hydrology 477 (2013) 189–202

real evapotranspiration, the pour-ribs effects are hardly visible in the simulated patterns of the groundwater levels. To show the range of values in the simulated patterns, the first (2006) and last (2010) model years are displayed in addition to the average (Fig. 12). Generally, lower mean values were simulated in 2006 compared with 2010 and the average. Moreover, the effects of pour-ribs on patterns of real evapotranspiration are less visible in 2010 and on average compared with 2006. Because of the lack of vegetation, only real evaporation is simulated in 2006. Effects of pour-ribs are lacking in spatial patterns of the compacted and non-compacted model configuration (Fig. 13) but general structures are already reflected. As a result, for the non-compacted model configuration 9% higher soil moisture, 10% lower real evapotranspiration, and 36% deeper groundwater levels are simulated compared to the compacted model configuration.

the pour-rib configuration. Compared with the non-compacted and the compacted configurations, the storage change is increased by 23 mm/a (20%) and 19 mm/a (16%), respectively, for the pour-rib configuration. The average soil moisture of the pour-rib configuration (684 mm) is higher compared with the non-compacted (682 mm) and compacted (623 mm) configurations. On average, the same groundwater level (1.4 m below surface) is simulated for the pour-ribs and the compacted configuration, whereas for the noncompacted configuration, deeper groundwater levels (1.9 m below surface) were simulated. As expected, the consideration of pour-ribs also affected the simulated hydrological patterns, such as soil moisture, real evapotranspiration, and groundwater levels (Fig. 12). The pattern of soil moisture revealed the strongest effects, followed by real evapotranspiration. A higher soil moisture was simulated for non-compacted compared with compacted pour-ribs. In addition to the differences due to pour-ribs, the effects on soils moisture caused by soil thickness are also visible. In contrast to the soil moisture, non-compacted pour-ribs are often characterised by a lower real evapotranspiration compared with compacted pour-ribs. Without the consideration of pour-ribs, no spatial distribution of the real evapotranspiration patterns was simulated, and the soil moisture patterns were only affected by soil depths and texture differences between the soil clusters. Compared with the soil moisture and

6. Discussion 6.1. Simulation of crucial hydrological processes in respect to parameter sensitivity and model uncertainty The global sensitivity analysis (GSA) reveals a high model equifinality, which implies that various parameter interactions lead to multiple model solutions. In addition, a number of parameters ex-

Fig. 11. Root squared error of simulated runoff of the model configurations from the measurements (left) and the cumulative runoff simulations and measurements (right). Pour-ribs are neglected for the non-compacted and compacted model configurations.

Table 4 Model performance evaluated based on the coefficient of determination (r2), the coefficient of model efficiency (cme), and the mass balance error (mbe [%]) considering runoff of the pour-rib, non-compacted, and compacted model configuration. Model configuration

Pour-ribs configuration

Efficiency criteria March 2006–December 2010 March 2006–December 2007 January 2008–December 2010

r2 0.62 0.75 0.59

cme 0.55 0.69 0.51

Non-compacted configuration mbe 5.1 1.4 6.5

r2 0.38 0.72 0.31

cme 0.35 0.71 0.27

Compacted configuration mbe 31.33 33.84 30.41

r2 0.61 0.75 0.58

cme 0.51 0.66 0.47

mbe 6.73 10.08 5.49

Table 5 Simulated yearly average water budget of the pour-rib, non-compacted, and compacted model configurations for the hydrological years 2006–2010. Model configuration

Precipitation [mm/a]

Pot. evapotranspiration [mm/a]

Real evapotranspiration [mm/a]

Runoff [mm/a]

Storage change [mm/a]

Pour-ribs configuration Non-compacted configuration Compacted configuration

562 562 562

662 662 662

322 290 324

123 178 140

117 94 98

198

H. Hölzel et al. / Journal of Hydrology 477 (2013) 189–202

Fig. 12. Simulated patterns of soil moisture (left), real evapotranspiration (middle), and groundwater levels (right) of the pour-rib model configuration in the first (2006, top) and last (2010, middle) model years as well as the average (2006–2010, bottom).

ist that do not have a global optimum and, therefore, are not suited for calibration, as represented by saturated hydraulic conductivity (ksat) (Fig. 7, right). Moreover, parameters with a physical meaning usually have limited calibration potential and measured values should be used if available. Therefore, we waive the calibration and adjusted only the recession constant for surface flow (Qdir), a conceptual parameter, which is the only parameter for which a global optimum could be identified (Fig. 7, left). The effect of Qdir on runoff is so dominant that it could not be superimposed by the interactions of other parameters. Therefore, Qdir is characterised by a significant trend and a clear distinction between the behavioural and non-behavioural simulations. By reducing Qdir, a better model performance concerning runoff simulation was achieved that proves that GSA is a suitable tool for model improvement. Reduced values for Qdir lead to higher peaks and a faster recession of

surface runoff, whereby the runoff simulations are closer to the observations. In addition, the sensitivity of Qdir reflects the importance of surface runoff in the Chicken Creek catchment. As mentioned previously, surface runoff is the dominant runoff component and is responsible for the creation of a gully network that drains the catchment (Gerwin et al., 2009). The high surface runoff amount is mainly caused by the high amount of non-vegetated bare soil and by the sealed surface layer with reduced infiltration. Previous studies revealed that the consideration of this sealed layer, simulated by reducing the ksat, significantly improved the runoff simulations (Bormann, 2011; Hölzel et al., 2011). Contrary to previous studies in which the ksat of the sealed layer was based on assumptions, the ksat in this study is based on measurements that may reduce model uncertainty. Due to the spatial model resolution, small gullies cannot be considered by the model,

H. Hölzel et al. / Journal of Hydrology 477 (2013) 189–202

199

Fig. 13. Simulated patterns of average (2006–2010) soil moisture (left), real evapotranspiration (middle), and groundwater levels (right) of the compacted (top) and noncompacted model configuration (bottom). In contrast to Fig. 12, pour-ribs are neglected in both model configurations.

which is why the observed and simulated flow paths differ. These differences could be one reason for the model uncertainty concerning runoff simulation. Another source of model uncertainty could be the parameterisation of vegetation, as the model configuration with vegetation starting in 2008 showed a lower model performance compared with the configuration without vegetation. On the one hand, the interannual increase of vegetation cannot be simulated with WaSiM-ETH; on the other hand, the vegetation model parameters, except vegetation cover, were unknown. Therefore, except for vegetation coverage, the standard model parameters for extensive grassland were used. Disregarding the spatial variability of the vegetation to reduce the parameterisation effort could also be responsible for the model’s uncertainty. The unknown initial conditions are another source of uncertainty, especially in systems that have not yet reached steady-state conditions. Bormann (2011) also noted that ‘‘In the case of the Chicken Creek catchment, the hydrological systems needed much more time to forget ‘wrong’ initial conditions’’. However, the runoff (Fig. 8, left), soil moisture (Fig. 9, left), and groundwater levels (Fig. 9, right) could be satisfactorily simulated. The changing hydrological conditions, including the gradual filling of the groundwater body and the soil water storage, are well reflected by the model. The steadily observed increase of groundwater levels and soil moisture shows that the hydrological change occurring in the catchment seems to have been incomplete at the end of the modelling period. As a consequence of the filling, the water availability for vegetation increases. Confidence in the model is enhanced by the evaluation of the simulated soil moisture and groundwater levels in addition to runoff. Thus, the simulated real evapotranspiration may also be repre-

sentative, although no validation by measurements is possible (Table 3). In the beginning, the real evapotranspiration was mainly caused by evaporation from bare soil because the vegetation was less developed. As a consequence of the increased vegetation and soil moisture availability, the relevance of transpiration increases, whereas evaporation becomes less important. Therefore, the real evapotranspiration increases over time, while the potential is slightly reduced. Due to the establishment of vegetation coverage, potential evapotranspiration should also increase. However, this effect is superimposed by climate conditions. In 2009 and 2010, the mean air temperature and global radiation, which mainly affect potential evaporation, are lower than in 2006–2008. 6.2. Impact of pour-ribs on simulated hydrological processes, patterns, and model uncertainty The improvement of the runoff simulations due to the consideration of pour-ribs indicates the importance of these artificial sediment structures for the hydrology of the Chicken Creek catchment (Fig. 11, Table 4). Disregarding the pour-ribs appears in dependence to the model type to be one source of uncertainty for runoff simulation in previous studies (Bormann, 2011; Holländer et al., 2009; Hölzel et al., 2011). The consideration of the pour-ribs enabled a reduction of model uncertainty concerning runoff simulation although only assumptions of hydraulic sediment properties were used for the modelling exercise. The dumped pour-ribs, transverse to the flow-paths with altered hydraulic behaviour, created strip-wise zones of different sediment conditions (Fig. 5, Table 1). On the one hand, the compacted pour-ribs act as hydraulic barriers, resulting in reduced subsurface lateral water flow and high surface runoff. On the other hand, the non-com-

200

H. Hölzel et al. / Journal of Hydrology 477 (2013) 189–202

pacted pour-ribs constitute zones with increased water storage capacity. Subsurface lateral runoff accumulates above the compacted zones, therefore inducing an up-filling of the non-compacted zones with a higher water storage capacity. Both the inhibited lateral subsurface flow and the increased water storage capacity explain the reduced runoff and the increased soil moisture compared with the model configurations where pour-ribs are neglected. Non-compacted pour-ribs are able to store more water than compacted pour-ribs and are therefore characterised by higher soil moisture. Additionally, the water drainage from the noncompacted pour-ribs is inhibited due to the adjacent compacted pour-ribs downstream, which became visible by the simulated patterns of soil moisture (Fig. 12). Compared with the patterns of the real evapotranspiration and groundwater levels, the soil moisture patterns show the clearest effects when considering the pour-ribs. Concerning the model structure, the soil moisture is directly linked to the hydraulic sediment properties. The differences concerning real evapotranspiration can also be explained by saturated hydraulic conductivity (ksat). As mentioned previously, in contrast to the runoff and soil moisture, the real evapotranspiration is lower for the non-compacted and higher for the compacted model configurations compared with the pour-rib configuration. On average, the non-compacted configuration is characterised by higher ksat-values, whereas the compacted configuration has lower values. Higher ksat-values indicate faster percolation and reduced water availability in the root zone for evapotranspiration, while lower values indicate the opposite. Thus, the lowest evapotranspiration and the highest cumulated runoff were simulated for the non-compacted configuration. The simulated patterns of real evapotranspiration are, therefore, generally characterised by a spatial alternation with lower evapotranspiration rates for non-compacted and higher for compacted pour-ribs (Fig. 12). Compared with the soil moisture, the patterns of the real evapotranspiration are less visible because the sediment effects are superimposed by other factors, such as potential evapotranspiration, vegetation, and soil moisture conditions. There is, however, no direct linkage between hydraulic sediment properties and evapotranspiration in the process description of WaSiM-ETH. Real evapotranspiration is limited by soil moisture which is determined by soil texture (sediment properties) in the model. Therefore, the sediment structures influence evapotranspiration only indirect via the state variable soil moisture. This is one reason why the effects of sediment properties on patterns of evapotranspiration are less visible compared to soil moisture patterns. The low groundwater levels for the non-compacted configuration can be explained by the differences of runoff and water storage capacity. Both the runoff and sediment porosity of the noncompacted configuration are higher compared with the compacted and pour-rib configurations, and therefore, lower groundwater levels were simulated. Differences in the simulated patterns of the groundwater levels are hardly visible because they are superimposed by multiple factors, such as soil moisture, percolation, uprise, and lateral in- and outflows (Fig. 12). Therefore, it is not surprising that no evidence of the pour-rib effects can be perceived based on the spatially distributed measured groundwater levels. Finally, the comparison of the simulated patterns between the first (2006) and last (2010) model years shows the development of the gradual up-filling of the groundwater and the soil moisture storage as well the effect of vegetation. Therefore, higher values were simulated in 2010 compared with 2006. Compared with the patterns of evapotranspiration in 2006, the effects of the pour-ribs are superimposed by transpiration due to vegetation and are therefore less visible in 2010 on average. Differences between spatial patterns of the compacted and the non-compacted model configuration (Fig. 13) are mainly caused by different assumptions concerning bulk density. The lower bulk

density of the non-compacted model configuration results in higher soil moisture compared to the compacted model configuration. In contrast to this, the impact on the patterns of real evapotranspiration and groundwater levels is less pronounced. This may be caused by higher saturated hydraulic conductivity of the non-compacted model configuration resulting in faster percolation and therefore less time for vegetation to transpire. Without considering pour-ribs, soil thickness and relief conditions as well as the clay dam become more important for the spatial distribution of hydrological fluxes and state variables. Considering soil moisture, spatial patterns are particularly affected by soil thickness whereas soil texture affects specifically patterns of real evapotranspiration. Patterns of groundwater levels are mainly influenced by relief conditions. Moreover, concerning patterns of real evapotranspiration and groundwater levels, the clay dam becomes visible because the clay dam acts as hydraulic barrier. Upslope of the clay dam, groundwater levels are therefore increased. This results in higher soil water availability for vegetation resulting in increased real evapotranspiration. Unfortunately, the simulated impact of the pour-ribs on the hydrological patterns cannot be validated by measurements at this time. Therefore, no evaluation of the uncertainty of the simulated patterns is possible. However, some degree of uncertainty can be expected because the hydraulic properties of the pour-ribs are not based on measurements but on assumptions about bulk-density and its spatial distribution. Nevertheless, the runoff simulations were improved, and the simulated hydrologic patterns seem to be plausible. Therefore, we stress the need for additional measurements to evaluate our model simulations because it promises to further the understanding of structure-process interactions in the initial and later stages of the ecosystem development in the Chicken Creek catchment. Moreover, a better understanding of the hydrological effects of the spatial variability of sediment conditions at the catchment scale in general can thereby be achieved.

7. Conclusion and outlook The process-based, spatially distributed hydrological model WaSiM-ETH was applied to analyse the impact of pour-ribs (dumped sediment structures) on the hydrological processes in the artificial Chicken Creek catchment. Knowledge on the importance of the spatial variability of hydrological sediment properties for the initial ecosystem development as well as a better understanding of hydrological catchment behaviour in general was thereby achieved. The study revealed the importance of pour-ribs for the simulation of runoff and soil moisture patterns. Neglecting these artificial sediment structures was identified as one of the reasons for the overestimation of simulated runoff in previous model studies (Bormann, 2011; Holländer et al., 2009; Hölzel et al., 2011). The implementation of pour-ribs enabled a runoff simulation that was significantly improved both quantitatively and dynamically, and it affected simulated hydrological patterns, such as soil moisture and real evapotranspiration. It can therefore be concluded that the consideration of artificial sediment structures, such as pourribs, is meaningful for revealing the driving forces of the initial ecosystem development and later stages in the artificial Chicken Creek catchment. Moreover, the significance of considering the spatial variability of sediment conditions when modelling hydrological processes at the catchment scale is demonstrated. The simulations further indicate an impact sediment structures on the real evapotranspiration. However, this assumption cannot be proven in our case even though affects due to sediments structures concerning the initial development of vegetation can be expected. Further lessons learned from this study are the importance of global sensitivity analysis (GSA) as a tool for model improvements

H. Hölzel et al. / Journal of Hydrology 477 (2013) 189–202

as stressed by e.g. Sivapalan (2009). The identification of sensitive model parameters enabled a significantly improved runoff simulation. Therefore, we stress the importance of conducting GSA before the model exercise is performed, not afterwards, as often happens, because the results can support identifying model parameters that are suitable for calibration. Nevertheless, the simulation results should be interpreted carefully, especially in catchments that have not yet reached hydrological steady-state conditions. For progress in hydrological modelling, further attention should be paid to the parameterisation and consideration of the spatial distribution of vegetation. The use of standard model parameters instead of measurements seems to be one significant source of model uncertainty. Due to the spread of vegetation in the Chicken Creek catchment, increasing relevance for the hydrological processes can be expected in the future. Therefore, further vegetation measurements may focus on the demand of hydrological models. Assumptions about the spatial distribution of the bulk density of the pour-ribs should be investigated inside the catchment by field measurements. Because only data from the surroundings were available (Maurer et al., 2011), assumptions of bulk density distribution as a key factor for varying hydrologic sediment properties had to be used for modelling. Under this condition, a propagation of uncertainty may occur. The simulated soil moisture patterns could not be evaluated because of the absence of spatially distributed measurements at present. Measurements regarding this can be provided by a continuous monitoring network of soil moisture, as used in TERENO (Bogena et al., 2006) which was successfully considered in simulating spatio-temporal patterns of the soil moisture dynamic in a 27 ha headwater catchment in West Germany (Sciuto and Diekkrüger, 2010a). However, the simulated patterns show that the spatially distributed modelling approach is suitable for simulating the interactions between hydrological patterns and processes and will therefore benefit fundamentally oriented ecological research. Acknowledgements The Transregional Collaboration Research Centre 38 (SFB/TRR 38) is gratefully acknowledged for providing fundamental data for this catchment study. We would also like to thank Prof. H. Bormann, Dr. H. Holländer, Dr. O. Rössler, Mr. R. Enders, Mr. K. Mazur, Prof. H. Neumeister, and Prof. U. Grünewald for fruitful discussions and helpful advice. The SFB/TRR 38 is financially supported by the Deutsche Forschungsgemeinschaft (DFG, Bonn) and the Brandenburg Ministry of Science, Research and Culture (MWFK, Potsdam). Finally, we acknowledge useful comments from anonymous referees on an earlier version of this paper. References Beven, K., 2009. Environmental Modelling: An uncertain future? An Introduction to Techniques for Uncertainty Estimation in Environmental Prediction. Routledge, 310 pp. Beven, K., Freer, J., 2001. Equifinality, data assimilation, and uncertainty estimation in mechanistic modelling of complex environmental systems using the GLUE methodology. J. Hydrol. 249 (1–4), 11–29. Bishop, J.G., 2002. Early primary succession on Mount St. Helens: impact of insect herbivores on colonizing lupines. Ecology 83 (1), 191–202. Bogena, H., Schulz, K., Vereecken, H., 2006. Towards a network of observatories in terrestrial environmental. Adv. Geosci. 9, 1–6. Bormann, H., 2011. Treating an artificial catchment as ungauged: Increasing the plausibility of an uncalibrated, process-based SVAT scheme by using additional soft and hard data. Phys. Chem. Earth A/B/C 36 (13), 615–629. Bormann, H., Holländer, H., Blume, T., Buytaert, W., Chirico, G.B., Exbrayat, J.-F., Gustafsson, D., Hölzel, H., Kraft, P., Krause, T., Nazemi, A., Stamm, C., Stoll, S., Blöschl, G., Flühler, H., 2011. Comparative discharge prediction from a small artificial catchment without model calibration: representation of initial hydrological catchment development. Die Bodenkultur 62 (1–4), 23–29.

201

Bormann, H., Breuer, L., Croke, B., Gräff, T., Huisman, J.A., 2009. Assessing the impact of land use change on hydrology by ensemble modelling (LUCHEM) IV: model sensitivity on data aggregation and spatial (re-) distribution. Adv. Water Resour. 32, 171–192. Bronstert, A., 2005. Rainfall–runoff modeling for assessing impacts of climate and land use change. In: Anderson, M.G., McDonnell, J.J. (Eds.), Encyclopaedia of Hydrological Sciences. John Willey & Sons Lds., Chichester, pp. 2033– 2059. Buczko, U., Gerke, H.H., Hüttl, R.F., 2001. Spatial distributions of lignite mine spoil properties for simulating 2-D variably saturated flow and transport. Ecol. Eng. 17 (2–3), 103–114. Christiaens, K., Feyen, J., 2001. Analysis of uncertainties associated with different methods to determine soil hydraulic properties and their propagation in the distributed hydrological MIKE SHE model. J. Hydrol. 246, 63–81. Elmer, M., Schaaf, W., Biemelt, D., Gerwin, W., Hüttl, R.F. (Eds.), 2011. The Artificial Catchment ‘‘Chicken Creek’’ Ecosystem Developement 2005–2010 Ecosystem Development, 3. Brandenburg University of Technology, Cottbus, 162 pp. Fischer, T., Veste, M., Schaaf, W., Dümig, A., Kögel-Knabner, I., Wiehe, W., Bens, O., Hüttl, R.F., 2010. Initial pedogenesis in a topsoil crust 3 years after construction of an artificial catchment in Brandenburg, NE Germany. Biogeochemistry 101, 165–176. Fridriksson, S., 2005. Surtsey Ecosystem Formed. Surtsey Research Centre, Reykjavik, 112 pp. Gerwin, W., Schaaf, W., Biemelt, D., Elmer, M., Maurer, T., Schneider, A., 2010. The Artificial Catchment ‘‘Hühnerwasser’’ (Chicken Creek): Construction and Initial Properties. Ecosystem Development, 1. Brandenburg University of Technology, Cottbus. Gerwin, W., Schaaf, W., Biemelt, D., Fischer, A., Winter, S., Hüttl, R.F., 2009. The artificial catchment ‘‘Chicken Creek’’ (Lusatia, Germany) – a landscape laboratory for interdisciplinary studies of initial ecosystem development. Ecol. Eng. 35, 1786–1796. Gerwin, W., Schaaf, W., Biemelt, D., Winter, S., Fischer, A., Veste, M., Hüttl, R.F., 2011. Overview and first results of ecological monitoring at the artificial watershed Chicken Creek (Germany). Phys. Chem. Earth A/B/C 36 (1–4), 61–73. Grayson, R.B., Blöschl, G., 2000. Spatial Patterns in Catchment Hydrology – Observation and Modelling. University Press, Cambridge, 406 pp. Herbst, M., Diekkrüger, B., 2006. Numerical experiments on the sensitivity of runoff generation to the spatial variation of soil hydraulic properties. J. Hydrol. 326 (1– 4), 43–58. Hofer, M., Lehmann, P., Biemelt, D., Stähli, M., Krafczyk, M., 2011. Modelling subsurface drainage pathways in an artificial catchment. Phys. Chem. Earth A/B/ C 36 (1–4), 101–112. 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, 2069–2094. Hölzel, H., Diekkrüger, B., 2012. Predicting the impact of linear landscape elements on surface runoff, soil erosion, and sedimentation in the Wahnbach Catchment, Germany. Hydrol. Process. 26 (11), 1642–1654. Hölzel, H., Rössler, O., Diekkrüger, B., 2011. Grope in the dark – hydrological modelling of the artificial Chicken Creek catchment without validation possibilities. Phys. Chem. Earth A/B/C 36 (4), 113–122. IPCS, 2003. SimLab 2.2 – Simulation Environment for Uncertainty and Sensitivity Analyses – Reference Manual, 158 pp. Jasper, K., 2005. Hydrological Modelling of Alpine River Catchments Using Output Variables from Atmospheric Models. Eidgenössische Technische Hochschule, Zürich, 138 pp. Kendzia, G., Reißmann, R., Neumann, T., 2008. Targeted development of wetland habitats for nature conservation fed by natural inflow in the post-mining landscape of Lusatia. World Min. 60, 88–95. Lighthill, M.H., Whitham, G.B., 1955. On kinematic waves. 1. Flood movement in long rivers. Proc. R. Soc. Lond. A 229, 281–316. Long Term Ecological Research Network, 2007. The Decadal Plan for LTER: Integrative Science for Society and the Environment. LTER Network Official Publication Series, 24, Albuquerque, New Mexico, 154 pp. Maurer, T., Schneider, A., Gerke, H.H., 2011. A structure generator for modelling the initial sediment distribution of an artificial hydrologic catchment. Hydrol. Earth Syst. Sci. 15, 3617–3638. Mazur, K., Schoenheinz, D., Biemelt, D., Schaaf, W., Grünewald, U., 2011. Observation of hydrological processes and structures in the artificial Chicken Creek catchment. Phys. Chem. Earth A/B/C 36 (1–4), 74–86. McKay, M.D., 1988. Sensitivity and uncertainty analysis using a statistical sample of input values. In: Ronen, Y. (Ed.), Uncertainty Analysis. CRC Press Inc., Boca Raton, Florida, pp. 45–186. Melching, C.S., 1995. Reliability estimation. In: Singh, V.P. (Ed.), Computer Models of Watershed Hydrology. Water Resources Publications, Colorado, pp. 69–118. Merz, B., Bárdossy, A., 1998. Effects of spatial variability on the rainfall runoff process in a small loess catchment. J. Hydrol. 112–113, 304–317. Monteith, J.L., Unsworth, M.H., 1990. Principles of Environmental Physics. Second Edward Arnold, London. Nash, J.E., Sutcliffe, J.V., 1970. River flow forecasting through conceptual models – part 1: a discussion of principles. J. Hydrol. 10, 282–290. Pennisi, E., 2010. A groundbreaking observatory to monitor the environment. Science 328, 418–420. Peschke, G., 1987. Soil moisture and runoff components from a physically founded approach. Acta hydrophys. 31 (3–4), 191–205.

202

H. Hölzel et al. / Journal of Hydrology 477 (2013) 189–202

Ratto, M., Saltelli, A., 2001. Model Assessment in Integrated Procedures for Environmental Impact Evaluation: Software Prototypes. IMPACT/JRC/WP6/ D18, 18. Joint Research Centre of European Commission, Institute for the Protection and Security of the Citizen, Ispra, 62 pp. Schaaf, W., Bens, O., Fischer, A., Gerke, H.H., Gerwin, W., Grünewald, U., Holländer, H., Kögel-Knabner, I., Mutz, M., Schloter, M., Schulin, R., Veste, M., Winter, S., Hüttl, R.F., 2011. Patterns and processes of initial terrestrial-ecosystem development. J. Plant Nutr. Soil Sci. 174 (2), 229–239. Schaaf, W., Biemelt, D. and Hüttl, R.F., 2010. Initial Development of the Artificial Catchment ‘‘Chicken Creek’’ – Monitoring Program and Survey 2005-2008. Ecosystem Development, 2. Brandenburg University of Technology, Cottbus, 187 pp. Schaap, M.G., Leij, F.J., van Genuchten, M.T., 2001. A computer program for estimating soil hydraulic parameters with hierarchical pedotransfer functions. J. Hydrol. 251 (3–4), 163–176. Schneider, A., Gerke, H.H., Maurer, T., 2011. 3D initial sediment distribution and quantification of mass balances of an artificially-created hydrological catchment based on DEMs from aerial photographs using GOCAD. Phys. Chem. Earth A/B/C 36 (1–4), 87–100. Schulla, J., 1997. Hydrologische Modellierung von Flussgebieten zur Abschätzung von Folgen von Klimaänderungen. Eidgenössische Technische Hochschule, Zürich, 163 pp. Schulla, J., Jasper, K., 2007. Model description WaSiM-ETH. Eidgenössische Technische Hochschule, Zürich, 181 pp.

Sciuto, G., Diekkrüger, B., 2010a. Influence of soil heterogeneity and spatial discretization on water balance modeling in a headwater forest catchment. Vadose Zone J. 9, 955–969. Sciuto, G., Diekkrüger, B., 2010b. Patterns in soil-vegetation-atmosphere systems: monitoring, modeling, and data assimilation. Vadose Zone J. 9, 821–827. Sivapalan, M., 2009. The secret to ‘doing better hydrological sciences’: change the question! Hydrol. Process. 23, 1391–1396. van Genuchten, M.T., 1980. A closed-form equation for predicting the hydraulic conductivity of unsaturated soils. Soil Sci. Soc. Am. 44 (5), 892–898. Zacharias, S., Bogena, H., Samaniego, L., Mauder, M., Fuß, R., Pütz, T., Frenzel, M., Schwank, M., Baessler, C., Butterbach-Bahl, K., Bens, O., Borg, E., Brauer, A., Dietrich, P., Hajnsek, I., Helle, G., Kiese, R., Kunstmann, H., Klotz, S., Munch, J.C., Papen, H., Priesack, E., Schmid, H.P., Steinbrecher, R., Rosenbaum, U., Teutsch, G., Vereecken, H., 2011. A network of terrestrial environmental observatories in Germany. Vadose Zone J. 10 (3), 955–973. Zaplata, M., Winter, S., Biemelt, D., Fischer, A., 2011. Immediate shift towards source dynamics: the pioneer Conyza canadensis in an initial ecosystem. Flora – Morphol. Distrib. Funct. Ecol. Plants 206 (11), 928–934. Zehe, E., Becker, R., Bárdossy, A., Plate, E., 2005. Uncertainty of simulated catchment runoff response in the presence of threshold processes: Role of initial soil moisture and precipitation. J. Hydrol. 315 (1–4), 183–202. Zehe, E., Lee, H., Sivapalan, M., 2006. Dynamical process upscaling for deriving catchment scale state variables and constitutive relations for meso-scale process models. Hydrol. Earth Syst. Sci. 10, 981–996.