Model-based quantification of runoff generation processes at high spatial and temporal resolution
Heavy precipitation-induced flash floods are still a serious hazard and generate high damages. In the context of climate change, an increase of the occurrence of flash floods is very likely. To improve flash-flood predictions and allow measures to reduce damage in vulnerable catchments, the spatial dynamics of runoff generation at a high spatial resolution during extreme rainfall events need to be better predicted. The results of these models can then be included into hydraulic models to predict the surface water level and flow dynamics based on high-resolution topographic data. Long-term discharge data does generally not exist in the small headwaters mostly influenced by flash floods, which would allow to calibrate conventional rainfall-runoff models. But hydrological models predicting runoff generation processes without calibration based on available spatially distributed data sets are still lacking. Such a model [Runoff Generation Research (RoGeR)] was developed for the state of Baden-Württemberg. It is based on an extensive collection of spatial data, including a digital elevation model of 1 × 1 m2 resolution, degree of sealing of the earth surface for the same resolution, and soil properties and geology at the scale of 1:50,000. Within the state of Baden-Württemberg, different regions were selected encompassing distinct environmental characteristics regarding climate, soil properties, land use, topography and geology. RoGeR was tested and validated by simulating 33 observed flood events in 13 mesoscale catchments without calibration and by modelling seven 60-m² artificial rainfall experiments on five different hillslopes in different regions of Switzerland. The results showed that the model was able to reproduce the temporal runoff dynamics as well as the peak discharge and the runoff volume in the mesoscale catchments as well as the 60-m² hillslope plots. The model could reproduce processes and hydrological response under different antecedent soil moisture and precipitation characteristics without any calibration, despite applying it to different regions and different scales. This suggests that RoGeR is predestinated to quantify runoff generation processes during heavy rainfall events at different scales without the typical model calibration procedure allowing to better quantify input and model uncertainty.
KeywordsRunoff generation Uncalibrated model Infiltration Preferential flow Subsurface flow Flash floods
Streamflow gauging has been carried out in Germany and many other countries at many rivers for a long time. These streamflow time series are the basis for estimating return periods of floods and calibrating and applying rainfall-runoff models. Hence, flood hazards can be well predicted along most of the gauged rivers assuming unchanging catchment characteristics and climatic conditions. According to the EU Floods Directive, flood hazard maps along rivers have been generated for catchment areas larger than 10–20 km2.
Extreme rainstorms, especially in the summer, regularly lead to flash floods on hillslopes, in small watersheds, and in urban areas, often generating serious damage. In the context of climate change, it must be expected that the risk of extreme rainstorms will increase even further in central Europe (Berg et al. 2013). Due to the various difficulties in forecasting extreme rainstorms and measuring discharge in small catchments, there are virtually no long-term and continuous measurements of heavy rainfall-induced flash floods available. Hence, return periods for small-scale flash floods cannot be derived as is possible for larger watersheds. Furthermore, rainfall-runoff models which depend heavily on observed data cannot be calibrated (Beven 2011). However, to estimate the extent and magnitude of flash floods, it is crucial to estimate the spatial and temporal distribution of runoff generation during extreme precipitation without parameter calibration in a hydrological model and to represent the idiosyncrasies of individual hillslopes and headwaters.
In Germany rainfall-runoff models such as LARSIM (Bremicker 2000) or WaSIM-ETH (Schulla 1997) are widely applied and calibrated for catchments with measured streamflow data and therefore valid only for the scale of such catchments. For smaller sub-catchments or catchments without any streamflow gauges, the runoff generation processes might not be well represented. Moreover, many rainfall-runoff models do not take relevant infiltration processes through macropores and shrinkage cracks explicitly into account. Regionalization approaches to estimate runoff coefficients (e.g. US-SCS 1972; Lutz 1984) are only valid for mesoscale catchments and should be calibrated as well (DVWK 1984; Merz 2006). Furthermore, those simpler approaches only provide a constant runoff coefficient that does not consider the decrease of infiltration capacity during a rain event or the temporal rainfall characteristics. In addition, the contribution of fast subsurface flow on floods is not considered in such approaches. Markart et al. (2006) developed a method to estimate runoff coefficient classes for small alpine catchments considering a wide range of factors and processes. Since the method was developed for alpine catchments, it cannot easily be applied in other regions. Approaches to determine dominant runoff processes (e.g. LUWG 2006) identify the dominant runoff generation processes in high spatial resolution but do not specifically quantify the individual runoff amount, generated by that process.
Thus, so far, there is no rainfall-runoff model that is capable of quantifying runoff generation processes from the plot to the mesoscale in high spatial and temporal resolution without calibration while considering the relevant processes. Such a model should account for infiltration through macropores and shrinkage cracks as well as for fast subsurface flow through lateral preferential flow pathways. To fill this research gap, the model RoGeR (Runoff Generation Research) was developed at the University of Freiburg to quantify runoff generation in high spatial (up to 1 × 1 m2) and temporal (up to 5 min) resolution without the need to calibrate parameters. RoGeR combines the knowledge of runoff generation process research gained over the last decades with spatially distributed data sets describing vegetation and surface as well as subsurface properties. Since the model uses spatially distributed data containing all the information needed to describe the runoff generation processes at any place and any time, it can be applied to a wide range of spatial scales from the plot to mesoscale catchments without the need of a scaling factor. The suitability of RoGeR to properly quantify and predict runoff generation processes was validated on different scales using observed rainfall-runoff events and floods in a wide range of mesoscale catchments as well as with sprinkler experiments on larger hillslopes. This paper describes the considered processes, model architecture, and data requirements of RoGeR ("Runoff generation processes implemented into RoGeR" Section) as well as the model validation at different scales ("Validation of RoGeR" Section).
Runoff generation processes implemented into RoGeR
The occurrence of HOF is required before infiltration through macropores (MP), and shrinkage cracks can occur. This infiltration through preferential flow pathways plays an important role in reducing HOF (Weiler and Flühler 2004). Once field capacity is exceeded, water starts to percolate to the base of the soil. Depending on the permeability of the underlying material, water can continue to percolate vertically (deep percolation) or—if the amount of percolating water exceeds the capacity of the permeability of the material below the soil—a saturated zone develops at the base of the soil. From this saturated zone, subsurface flow (SSF) originates if the lateral permeability and gradient is large enough. SSF occurs as a slow component through the soil matrix and as a fast component through lateral preferential flow paths as, e.g. pipes or root channels (Scherrer 1997). When the soil becomes fully saturated, no more water can infiltrate and saturation overland flow (SOF) starts. This likely occurs especially in locations that have shallow soils with impermeable material below or in areas with a small distance to the groundwater table.
Interception and matrix infiltration
To quantify interception storage for single events, a leaf area index (LAI) approach was adopted including a seasonal dependence of the LAI into a simple interception storage model as described in Bremicker (2000). For sealed or partly sealed areas, surface storage in microdepressions is quantified. Since RoGeR was designed as an event-based model, evaporation from interception storage is not considered.
Database and use in the model currently available for Baden-Württemberg
Use in model
Landesvermessungsamt Baden-Württemberg (LV-BW)
Parameterization of macropores, interception and wetlands
CORINE land cover 2000
Min. polygon size 500 × 500 m2 min. line width 100 m
Parameterization of macropores, interception and wetlands
Degree of sealing of earth surface
1 × 1 m2-grid
Wasser- und Bodenatlas Baden-Württemberg (WaBoA)
Reduction of infiltration
~1 × 1 m2
Digital elevation model (DEM), vegetation height, slope, flow accumulation, depth to groundwater
River network as line file
Landesanstalt für Umwelt, Messung und Naturschutz Baden-Württemberg (LUBW)
Ideal to define a river network from the DEM
Lakes as polygon file
Soil overview map of Baden-Württemberg (BÜK)
Regierungspräsidium Freiburg Landesamt für Geologie, Rohstoffe und Bergbau (LGRB)
Soil storage, soil depth, portion of skeleton, soil texture classes
Soil map 50,000 (BK50) (available since 2015/6)
Soil storage, soil depth, portion of skeleton, soil texture classes, saturated hydraulic conductivity
Transmissibility of under laying material and bedrock
Precipitation radar data for selected events (RADOLAN)
~1 × 1 km2 1 h-sum
Deutscher Wetterdienst (DWD)
Input for modelling selected flood events
Precipitation data from gauging stations
Control and evaluation of precipitation radar data
GWN-BW (groundwater recharge in Baden-Württemberg)
Daily average soil moisture, average area 0.8 km2
Estimation of spatial antecedent soil moisture conditions
Discharge data from gauging stations
Evaluation of modelled discharge at gauging station
Infiltration through macropores and shrinkage cracks
The soil matrix properties are not sufficient to predict runoff response to heavy rainfall. For example, Scherrer (1997) was able to show in numerous sprinkler experiments with high rainfall intensities on larger hillslopes that, frequently, more water can infiltrate than expected because of the soil matrix properties. Infiltration through preferential flow paths like macropores (i.e. earthworm channels) or soil cracks is cited as the reason of this phenomenon, which is playing an important role in runoff generation during high rainfall intensities (e.g. Beven and Clarke 1986; Weiler 2001, 2005; Weiler and Naef 2003; Weiler and Flühler 2004). Consequently, the macropore infiltration process is implemented in RoGeR. Ponding water, described as potential Hortonian overland flow, is a prerequisite for infiltration through preferential flow paths. However, only a certain proportion of overland flow is connected to macropores. The magnitude of this proportion is related to the macropore density (Weiler and Flühler 2004). The other part of the overland flow cannot infiltrate via macropores and hence will run off. How much water actually infiltrates via macropores depends on the interaction between macropores and soil matrix. This horizontal infiltration from the macropores into the soil matrix is estimated using the Green & Ampt method, which needs to be modified for horizontal direction with a radial wetting front for this purpose (Beven and Clarke 1986; Weiler 2005; Fig. 1, top right). During the infiltration process, the progressing wetting front of vertical matrix infiltration from the soil surface successively shortens the active part of the macropore (Fig. 1, top right).
Estimation of macropore density and length for land cover classes and k values for overland flow (vertical macropores for infiltration process, slope parallel macropores for subsurface flow)
Land cover class
Density of vertical macropores (MP/m2)
Length of vertical macropores (cm)
Density of slope parallel macropores (MP/m2)
Strickler k values for overland flow
Settlement (1–100% sealed)
Cmin is set to 3% and F to 7 mm so that the depth of the cracks at the shrinkage limit (maximum depth) equals 500 mm for the soil with the highest clay content (75%). This value lies within the range found in literature (e.g. Baram et al. 2012) and own field studies. Shrinkage cracks typically build a more or less polygon-shaped network (e.g. Baram et al. 2012). According to different studies (Bouma and Dekker 1978; Konrad and Ayad 1997; Li and Zhang 2010; Baram et al. 2012) and own field observations on pelosols around Freiburg (Baden-Württemberg), an average distance between cracks of 20 cm was set resulting in a crack density of 10 m/m2.
Soil storage, deep percolation, subsurface flow and saturation overland flow
Potential soil storage at the beginning of a rainfall event is given by the free proportion of the available field capacity plus the free drainable pore volume. Infiltration through matrix and preferential flow paths fills this soil storage. If the available field capacity is exceeded, water is percolating deeper (field capacity excess). The potential deep percolation flux into the underlying geological substrate is derived from the hydrogeological map (Table 1). If more water percolates out of the soil than the bedrock can absorb, this water (deep percolation excess) starts to fill up drainable pore volume from the base of the soil. Consequently, a saturated layer develops, which represents the active zone for subsurface flow through the soil matrix and lateral macropores. The deep percolation excess is also the amount of water available for subsurface flow.
Water, available for subsurface flow (deep percolation excess), is allocated to (slow) matrix subsurface flow and (fast) macropore subsurface flow according to the quantitative relation between the potential maximum SSF flow of matrix and MP. If the amount of infiltrated water still exceeds the soil porosity after losses by deep percolation and SSF are considered, saturation overland flow is generated by saturation excess. In areas with a high groundwater table, the total available storage of the soil is reduced by the groundwater. An average initial groundwater table is estimated applying the GIS method “vertical distance to groundwater”, available with the GIS software SAGA (Olaya 2004).
Model parameter (per grid cell)
Degree of sealing
Density of vertical macropores
Depth of vertical macropores
Density of slope parallel macropores
Mean distance of shrinkage cracks
Depth of groundwater level
Plant available field capacity of soil
Air capacity of soil
Saturated hydraulic conductivity of soil
Wetting front suction
Saturated hydraulic conductivity below the soil
Velocity of overland flow
Velocity of fast (preferential) subsurface flow
Velocity of slow (matrix) subsurface flow
Velocity of groundwater
Free plant available field capacity
Free air capacity
Depth of shrinkage cracks
Time step dependent
Validation of RoGeR
RoGeR’s capability to correctly reproduce runoff generation process and the resulting discharge reaction was validated for two different spatial scales. At the hillslope scale, sprinkler experiments with measurements of overland flow and subsurface flow were used to directly compare the simulated runoff generation processes. At the mesoscale, observed flood events in catchments with gauging stations were modelled with RoGeR and the model simulations were compared to the observed streamflow hydrographs.
Runoff generation in space and time is usually not measured during natural rain events. The only information available is the integrative streamflow response of the catchment at a gauging station. The shape of the hydrograph depends considerably on the temporal and spatial distributions of the runoff generation processes (e.g. McGlynn et al. 2004; Zillgens et al. 2007). Events dominated by overland flow processes generally show a fast increase and fast decrease of the hydrograph as well as relatively high peak discharge but small runoff yields. In contrast, events with dominating subsurface flow generate hydrographs of moderate peak flow but long-lasting recession and large runoff yield. Thus, besides the peak discharge and runoff coefficient, the shape of the hydrograph is a suitable indicator to verify the ability of a model to correctly reproduce the processes of runoff generation.
The long-lasting advective events are characterized by moderate peak discharges and long-lasting recessions. Both were well reproduced by RoGeR. Fast subsurface flow (Fig. 7, brown line) is the major process responsible for the long recession in all catchments. In catchment no. 9 (Starzel), the outflow from Karst springs contributes to those events as well, which is reflected by the predicted deep percolation runoff (grey line) within RoGeR. Usually, the contribution of this flow component to the flood hydrograph predicted by RoGeR is marginal. But based on typical flow velocity in Karst systems, it is assumed that this runoff component will contribute significantly to the event runoff for catchments located in limestone areas.
The agreement between observed and modelled dynamics of the hydrographs was evaluated and grouped subjectively into three classes: “good”, “moderate” and “not satisfying” (Fig. 8, right column). A good agreement means that shape and quantitative proportion of the simulated runoff components as well as the total flood hydrograph were well suited to reproduce the observed flood hydrographs shape. A moderate agreement means that the shape of the observed hydrograph was well represented by the shape of the simulated runoff components, but the quantity was not well represented. A not satisfying agreement indicates that shape and quantity were not well reproduced. This subjective approach was selected since all objective functions (e.g. coefficient of correlation, efficiency) are unable to reflect the agreement of the hydrograph shape properly. For example, in Fig. 7 for the long-lasting event in catchment no. 6, the coefficient of correlation is only at 0.56, but the similarity of the modelled versus the observed dynamic is quite satisfying considering the small time lag between simulated and observed peak flow and the shape of the recession. Overall, 21 of 33 events could be modelled with a good agreement of the dynamics between the observed and simulated hydrographs. Five events show a moderate agreement and 7 were not satisfying. Five of the not satisfying results were likely caused by underestimation of subsurface flow. One was caused by an overestimation of SOF, which is quite sensitive to the estimation of the groundwater level (see also discussion above regarding Fig. 7, left bottom). Finally, one event (in catchment 13) was influenced by lake retention, which is not well represented by RoGeR.
One of the main problems of the event-based simulation is the high uncertainty of precipitation and antecedent moisture conditions. The rain radar data are calibrated to precipitation gauging stations and is therefore, for example, affected by shading effects, station density, topographic effects and precipitation type. The antecedent moisture data are itself a model output, which depends on the model structure and input data uncertainty. Taking these input uncertainties into account, the agreement between RoGeR output and observed discharge is satisfying, in particular since the model was not calibrated at all.
Discussion and conclusions
While extensive parameter calibration is still the status quo in rainfall-runoff modelling, RoGeR proved to be a suitable model framework to predict event-based runoff generation processes at the hillslope scale, in headwaters and in mesoscale catchments without any parameter calibration. Taking into account the input uncertainty and the still unresolved problems such as parameterizing macropore properties for a heterogeneous landscape, the results are very satisfying and promising. The application of RoGeR to hillslope sprinkler experiments reproduced the measured overland and subsurface discharge very well. Considering the efforts to simulate these hillslope experiments with numerical 2D Richards-based models (e.g. Faeh et al. 1997), the results show a great potential even at this smaller spatial scale. Larger sets of data from more sprinkler experiments might lead to an even better validation of the model on the one hand. On the other hand, they may also provide crucial information to evaluate if important processes are missing in the model. For example, the development of the model algorithms describing infiltration by shrinkage cracks resulted from poor model prediction in catchments with clayey soils during dry summer months.
There are some other potentially relevant processes like water repellency, soil sealing on silty soils, lake retention, fluctuation of groundwater table or the influence of tile drains. These processes have not yet been implemented into RoGeR because of the lack of data or process understanding. Further progress in deriving geodata can further enhance the ability of RoGeR and reduce uncertainty. The ability of RoGeR to quantify the spatial and temporal distribution and dynamics of runoff generation without parameter calibration makes it a suitable tool for the prediction of flash floods induced by heavy rainfall events at scales up to 10–20 km2 where discharge gauging data are lacking. Currently, RoGeR is applied for the whole area of Baden-Württemberg to derive planning criteria for flash flood management based on scenarios of extreme precipitation (LUBW 2016).
A dynamic hydraulic approach for RoGeR is currently still in the developmental stage. It accounts for time variable flow velocities and re-infiltration of overland flow into areas with higher infiltration rates. It might improve the prediction of flow accumulation for small areas and highly intensive rain events. A description of this approach will be subject of a separate publication.
We thank Simon Scherrer for providing the data of the six hillslope sprinkling experiments. The work was partly funded by the State Office for the Environment, Measurement and Nature Conservation of the Federal State of Baden-Württemberg.
- Ad-Hoc-Arbeitsgruppe Boden (2005) Bodenkundliche Kartieranleitung. 5. Auflage, Bundesanstalt für Geowissenschaften und Rohstoffe (Hrsg.), Hannover. Schweizerbart, StuttgartGoogle Scholar
- Beven KJ (2011) Rainfall-runoff modelling: the primer. Wiley, LondonGoogle Scholar
- Bremicker M (2000) Das Wasserhaushaltsmodell LARSIM—Modellgrundlagen und Anwendungsbeispiele Freiburger Schriften zur Hydrologie, 11, Institute of Hydrology (Hrsg.)Google Scholar
- Casper M (2002) Die Identifikation hydrologischer Prozesse im Einzugsgebiet des Dürreychbaches (Nordschwarzwald). Mitteilungen des Instituts für Wasserwirtschaft und Kulturtechnik der Universität Karlsruhe (TH), Heft 210Google Scholar
- DVWK (1984) Arbeitsanleitung zur Anwendung von Niederschlag-Abfluß-Modellen in kleinen Einzugsgebieten, Teil II: Synthese, DVWK-Regeln zur Wasserwirtschaft 113, Deutscher Verband für Wasserwirtschaft und Kulturbau e.V. (DVWK), BonnGoogle Scholar
- Geitner C, Mätzler A, Bou-Vinals A, Meyer E, Tusch M (2014) Bodeneigenschaften und die Besiedlung durch Regenwürmer (Lumbricidae) auf Grünlandstandorten im Brixenbachtal (Kitzbüheler Alpen, Tirol). Die Bodenkultur 65(1):39–51Google Scholar
- Gudera T, Morhard A (2015) Hoch aufgelöste Modellierung des Bodenwasserhaushalts und der Grundwasserneubildung mit GWN-BW. Hydrologie und Wasserbewirtschaftung 59.JahrgangHeft 5:205–216Google Scholar
- Kohl B, Stepanek L, Pirkl H, Kirnabauer R, Meyer V, Markart G (2012) Verbesserungspotentiale bei der prozessorientierten Ereignismodellierung. Forum für Hydrologie und Wasserbewirtschaftung Heft 33(13):147–156Google Scholar
- LUBW Landeanstalt für Umweltschutz Baden-Württemberg (Hrsg.) (1997) Boden als Lebensraum für Bodenorganismen—bodenbiologische Standortklassifikation—Literaturstudie, Handbuch BodenGoogle Scholar
- LUBW Landeanstalt für Umweltschutz Baden-Württemberg (Hrsg.) (2016) Leitfaden Kommunales Starkregenrisikomanagement in Baden-WürttembergGoogle Scholar
- Lutz W (1984) Berechnung von Hochwasserabflüssen unter Anwendung von Gebietskenngrößen. Mitteilung Nr. 24 des Instituts für Hydrologie und Wasserwirtschaft, Universität KarlsruheGoogle Scholar
- LUWG, Landesamt für Umwelt, Wasserwirtschaft und Gewerbeaufsicht Rheinland-Pfalz (Hrsg.) (2006) Bestimmungsschlüssel von hochwasserrelevanten Flächen. Bericht Nr. 18Google Scholar
- Markart G, Kohl B, Sotier B, Schauer T, Bunza G und Stern R (2006) Geländeanleitung zur Abschätzung des Oberflächenabflussbeiwertes bei Starkregen—Grundzüge und erste Erfahrungen. Wiener Mitteilungen Band 197: Methoden der hydrologischen Regionalisierung. 159–178Google Scholar
- Merz R (2006) Regionalisierung von Ereigniskenngrößen. Wiener Mitteilungen Band 197: Methoden der hydrologischen Regionalisierung. 179–194Google Scholar
- Moldenhauer K-M, Heller K, Chifflard P, Hübner R, Kleber A (2013) Influence of Cover Beds on Slope Hydrology. Dev Sedimentol 66:140–152Google Scholar
- Olaya V (2004) A gentle introduction to SAGA GIS. Edition 1.1. The SAGA User Group e.V. (Hrsg.), Gottingen, GermanyGoogle Scholar
- Peschke G (1985) Zur Bildung und Berechnung von Regenabfluss. Wissensch. Zeitschrift der TU Dresden Nr. 34(4):195–200Google Scholar
- Pohl B (2000) Charakterisierung der hydrologischen Systemparameter des Einzugsgebietes einer Trinkwasserquelle im Karst am Beispiel Röschenz (Nordwestschweiz), zur differenzierten Ausscheidung von Grundwasserschutzzonen. Diploma thesis at the Institute of Hydrology Freiburg (not published)Google Scholar
- Raissi F, Müller U, Meesenburg H (2009) Ermittlung der effektiven Durchwurzelungstiefe von Forststandorten. GeoFakten 9:1–7Google Scholar
- Scherrer S (1997) Abflussbildung bei Starkniederschlägen. Identifikation von Abflussprozessen mittels künstlicher Niederschläge. Mitteilungen der Versuchsanstalt für Wasserbau, Hydrologie und Glaziologie, ETH Zürich, Nr. 147Google Scholar
- Schmocker-Fackel P (2004) A method to delineate runoff processes in a catchment and its implications for runoff simulations. ETH Zürich, DissGoogle Scholar
- Schulla J (1997) Hydrologische Modellierung von Flussgebieten zur Abschätzung der Folgen von Klimaänderungen. Diss. 12018, ETH Zürich, 161 SGoogle Scholar
- US-SCS (1972) National engineering handbook section 4 hydrology. US Department of AgricultureGoogle Scholar
- Weiler M (2001) Mechanisms controlling macropore flow during infiltration. Dye tracer experiments and simulations. Schriftenreihe des Instituts für Hydromechanik und Wasserwirtschaft der ETH Zürich, Band 7Google Scholar
- Wessolek G, Kaupenjohann M, Renger M (2009) Bodenphysikalische Kennwerte und Berechnungsverfahren für die Praxis. Bodenökologie und Bodengenese Heft 40Google Scholar
- Winter F (2013) Prozessorientierte Modellierung der Abflussbildung und -konzentration auf verschlämmungsgefährdeten landwirtschaftlichen Nutzflächen. Mitteilungen der Universität der Bundeswehr München, Institut für Wasserwesen Wasserwirtschaft und Ressourcenschutz, Heft 118Google Scholar