Journal of Hydrology 617 (2023) 128786

Contents lists available at ScienceDirect

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

Research papers

Near surface roughness estimation: A parameterization derived from

artificial rainfall experiments and two-dimensional hydrodynamic modelling
for multiple vegetation coverages
David Feldmann a , Patrick Laux a,b ,∗, Andreas Heckl c , Manfred Schindler c , Harald Kunstmann a,b
a
Karlsruhe Institute of Technology, Campus Alpin, Kreuzeckbahnstraße 19, Garmisch-Partenkirchen, 82467, Germany
b University of Augsburg, Institute of Geography, Alter Postweg 118, Augsburg, 86159, Germany
c Dr. Blasy - Dr. Øverland Ingenieure GmbH, Moosstraße 3, Eching am Ammersee, 82275, Germany

ARTICLE INFO ABSTRACT

This manuscript was handled by Marco Borga, Roughness is the key parameter for surface runoff simulations. This study aims to determine robust Manning
Editor-in-Chief, with the assistance of Francesco resistance coefficients (𝑛) on the basis of consecutive artificial rainfall experiments on natural hillslopes
Comiti, Associate Editor. available in literature, obtained at 22 different sites with different degrees of vegetation cover and type. The
Keywords: Manning resistance coefficient is particularly important in the context of two-dimensional (2D) hydraulic heavy
Depth-dependent roughness coefficients rainfall simulations. Since there is a wide range of possible resistance values available leading to significantly
Flow resistance different results regarding the accumulation of surface runoff, especially for shallow water depths. The planning
Artificial rainfall experiments of flood protection structures is directly affected by these uncertainties. This work also improves the knowledge
Hydraulic rainfall-runoff modelling between roughness and the shape of the hydrograph allowing a better calibration of infiltration models. As flow
Surface runoff velocity, water depth, and infiltration rate were not observed during the rainfall experiments, only the outflow
Flash floods of the test field and rain intensity are known. For this purpose, a framework was developed to parameterize
shallow water depth (< 1 cm) -dependent roughness coefficients. To test the robustness of the framework,
three different formulations of depth-dependent roughness and a constant Manning coefficient are used by
comparing the measured discharge under different rainfall intensities with simulations in a 2D-hydraulic model.
We identified a strong dependency of Manning’s 𝑛 on the degree of vegetation cover and -type as well as an
influence of consecutive rainfall events. This finally leads to a more robust parameterization of near surface
roughness for hydrodynamic modelling, which is particularly important for the simulation of heavy rainfall
events.

1. Introduction use case (Zhang and Cundy, 1989; Fernández-Pato et al., 2016; Savant
et al., 2019; David and Schmalz, 2020).
Flash floods resulting from torrential rainfall events can cause Using hydraulic 2D-models for heavy rainfall simulations, the pre-
great damage and loss of human lives, as numerous events in Cen- cipitation is distributed over the model domain. This means that an
tral Europe in recent years have demonstrated (Kaiser et al., 2020). individual amount of rain is added to each mesh-node or cell depending
While climate change is potentially increasing the intensity and fre- on the infiltration approach. Subsequently, the discharge concentration
quency of such events, maps depicting the expected flooding risk in flow paths caused by the surface topology is determined by solving
caused by heavy rainfall are becoming increasingly important for the shallow water equations. This enables the creation of accurate
communities (Sañudo et al., 2020; Bulti and Abebe, 2020). The use of flood maps showing endangered critical infrastructure or buildings.
detailed two-dimensional (2D) hydraulic models is the most common In contrast to fluvial flood simulations, these simulations can show
and accurate method for risk mapping (Rai et al., 2010). hazards far away from water bodies caused by short and intense rainfall
The original purpose of hydraulic 2D-models is to simulate fluvial events.
flood events, where the river overflows its bank. Meanwhile, the direct Beside the hydrologic parameters, the flow resistance expressed by
generation of surface runoff within a hydraulic model, considering hy- the surface roughness is the key parameter for 2D hydraulic surface
drological processes like infiltration excess, has become also a standard runoff simulations (Rai et al., 2010; Barros and Colello, 2001). The

∗ Corresponding author at: Karlsruhe Institute of Technology, Campus Alpin, Kreuzeckbahnstraße 19, Garmisch-Partenkirchen, 82467, Germany.
E-mail address: patrick.laux@kit.edu (P. Laux).

https://doi.org/10.1016/j.jhydrol.2022.128786
Received 25 May 2022; Received in revised form 1 November 2022; Accepted 4 November 2022
Available online 1 December 2022
0022-1694/© 2022 The Author(s). Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-
nc-nd/4.0/).
D. Feldmann et al. Journal of Hydrology 617 (2023) 128786

definition of the term roughness varies depending on the discipline and the research concerning the Manning formula in two fields: the first
question. It can represent surface or flow properties, or simply serve type tries to develop new or extend existing formulas for the calculation
as a model parameter. The lack of a uniform parameterization and of flow resistance as the formulation of the Manning formula is based
scale dependency complicates the subject. Since no single surface prop- only on empirical assumptions (Jarrett, 1990). The second type tries to
erty can be defined as ‘‘the roughness’’, roughness primarily reflects determine better or more exact coefficients valid for specific use cases
the accuracy of the measurement technique and the motivation for and conditions (Wu et al., 1999; Lawrence, 2000; Wilson and Horritt,
parameterization (Smith, 2014). The obstructive effects of roughness 2002; Ding et al., 2004).
determine the velocity of the runoff on the surface and therefore the In this study, we contribute to the second type of research studies.
formation speed of a flood wave. In general, a smooth surface leads Based on existing artificial rainfall experiments (Ries et al., 2020),
to a more rapidly rising flood wave, while peak discharge can be we derive empirical parameters describing depth-dependent roughness
significantly lower on rough surfaces (Sanz-Ramos et al., 2021). relations valid for the micro topography and different land-use types.
Vegetation coverage has the most significant effect on near surface We use four different formulations describing roughness in order to
runoff processes and surface resistance. Studies differentiate by vegeta- compare their suitability for the proposed task. Combined with further
tion type, flexibility, degree of submergence (Cantisani et al., 2014) or data from literature, this leads to a better, more universal represen-
vegetation stem diameter (Zhang et al., 2021). A distinction is also of- tation of overland flow in 2D-hydraulic models. Finally, this allows a
ten made between submerged and partially submerged vegetation (Kim more accurate assessment of flash flood risks and therefore, a better
et al., 2012; Wu et al., 1999). planning of counter measures.
Roughness implies the effects on flow by crop ridges, rocks, tillage, To achieve this goal, an idealized hydraulic 2D-model is applied
and the frictional drag over the surface (Engman, 1986). Romkens and representing the test sites of the artificial rainfall experiments. By
Wang (1986) distinguish surface roughness into four categories: The examining all possible parameter combinations to maximize the Nash
first category includes microrelief variations in the range of 1–2 mm Sutcliffe Efficiency (NSE) between observed and simulated discharge
as the friction drag of the surface. The second type results from non- and a comparison with results from other experiments, a range of
directional clumping of the soil as it arises, for example, from tillage valid values is determined. In addition, the framework offers a way
in the range of 100 to 200 mm. The third class represents systematic or to determine roughness coefficients knowing only discharge, rainfall
directional differences, such as plough tracks. The fourth type contains rate and slope of the test sites, thus avoiding complex measurements
non-directional variations of the terrain with elevation differences. The of water level or velocity.
increasing availability of high resolution terrain data, often with a
horizontal resolution of less than 1 m and a vertical accuracy of a 2. Materials and methods
few centimetres, allows to depict most of the topographic complexity
directly from the terrain model. The roughness parameterization has 2.1. Artificial rainfall experiments
to represent all processes which lie beyond the terrain resolution as
different flow processes occur at different topographic scales. So rough- Artificial rainfall experiments are one option to examine roughness
ness parameterization has a scale dependency and parameters have to coefficients on real surfaces outside a laboratory. Ries et al. (2019)
be chosen in dependency of the computational mesh resolution (Lane, and Seibert et al. (2011), for example, provide collections of these
2005; Horritt et al., 2006). experiments. In most cases, sprinkling experiments use a constant rain
The roughness of arable land is particularly difficult to standard- intensity (e.g., 70 mm∕h), which corresponds to a statistical return
ize (Huang and Bradford, 1992; Sepaskhah and Bondar, 2002). Huang interval for a local heavy rainfall event (e.g., a 100-year event). The
and Bradford (1992) demonstrated this with laboratory tests experi- water is sprayed via a sprinkling system over a defined period of
menting with different rain intensities and successive rain events. After time (e.g., 1 h) onto the test field. Simultaneously, the outflow is
an initial rainfall event of 63 mm, soil was crusted and roughness measured at the lower boundary of the test field as the volume of water
reduced. After a simulated rain event of 92 mm thereafter, microrills running downstream as surface runoff in a high temporal resolution.
developed and the visible roughness of the surface increased due to its The data set provided by Ries et al. (2019), for instance, includes the
stronger structuring, whereby the roughness relevant to the runoff had entire hydrograph with ascending and descending parts, whereas the
decreased due to a channelling of the runoff. collection of Seibert et al. (2011) does not include the descent of the
The most common representation of roughness is done by the use of hydrograph.
the Manning parameter (𝑛) (Manning, 1891) as a value between 1 and Sprinkling experiments with tracers (Tatard et al., 2008; Mügler
0. There are widely accepted and generally valid roughness coefficients et al., 2011) try to circumvent the problem of lacking flow velocity
for larger flow depths, such as those that occur during fluvial floods data by measuring the propagation rate of a tracer substance. A similar
(Chow, 1959). Apart from this the roughness at shallow water depths approach with a sprinkling setup to simulate rainfall has been con-
is subject to greater uncertainties. Especially surface runoff simulations ducted by Katz et al. (1995) using fluorescent dye and a slow motion
require a reliable choice of roughness parameters due to the high camera system in an artificial channel to estimate flow velocities.
sensitivity of the results (Gaur and Mathur, 2003; Sauer and Ortlepp, However, these approaches focused mainly on one test site and did not
2021) and thus, possibly to errors in risk assessment of flash floods. cover a greater variety of surface properties. This lack of a sufficient
Many studies tried to find a relation with dimensionless parameters quantity of data does not allow a comparison of multiple land cover
between Reynolds number or Froude number and vegetation or slope types and therefore, artificial rainfall experiments with flow velocity
(Chow, 1959; Emmett, 1970; Wu et al., 1999; Díaz, 2005). Meanwhile, measurements are not considered in this study.
the use of water depth-dependent roughness relations rather than a One of the most comprehensive data sets available for Germany is
single value representing the range of possible roughness effects for provided by Ries et al. (2019). The artificial rainfall experiments took
the simulation of surface runoff has become a standard in engineering place at 23 locations in Baden-Wuerttemberg (Germany) with different
(Gaur and Mathur, 2003; Rai et al., 2010; Mügler et al., 2011; Fraga vegetation and soil conditions on a defined field of 10 m length. Six
et al., 2013). In practice, the application of roughness coefficients experiments were conducted on each field, each with different rainfall
is a definition of a water depth-dependent relationship, as it is, for duration, intensity and initial soil moisture. The following information
example, the case for the 2D-model HYDRO_AS-2D (Hydrotec, 2022). was logged in a one minute time interval, which is relevant for our
These models originally developed for river hydraulics make use of study:
the Manning formula which is also the most common and widely used
approach to quantify flow velocity (Díaz, 2005). Díaz (2005) divides • discharge at the lower end of the test field

2
D. Feldmann et al. Journal of Hydrology 617 (2023) 128786

initial value 𝑓0 to its final one 𝑓𝑐 , using the decay coefficient 𝛼 (Eq. (1)).

𝑓 (𝑡) = 𝑓𝑐 + (𝑓0 − 𝑓𝑐 )𝑒−𝛼𝑡 (1)

The variables of the Horton function are calculated with the non-
linear least square method as proposed by Esen (1987) to fit the
function (Eq. (1)) to the observed data. The applied method using
the Levenberg–Marquardt algorithm is implemented in the MINPACK-1
package (Moré et al., 1980). A Horton curve is displayed exemplarily
in Fig. 1 as maximum possible infiltration rate 𝑓𝑚𝑎𝑥 .
The schematic artificial rainfall experiment can be separated in
three time segments: Ascent of the hydrograph, nearly constant dis-
charge and descent of the hydrograph. During the ascent of the hy-
drograph, the infiltration cannot be determined unambiguously (orange
area in Fig. 1), as the physical process of soil saturation is not repre-
sented by the empirical Horton approach. Consequently the impact of
Manning’s 𝑛 on the shape of the hydrograph cannot be separated from
infiltration, since 𝑞 is a function of 𝑓 , Manning’s 𝑛 and precipitation 𝑃
Fig. 1. Water fluxes in a schematic sprinkling experiment.
(Eq. (2)).

d𝑞∕d𝑡 > 0 𝑎𝑛𝑑 𝑞𝑡0 ≤𝑡≤𝑡1 = 𝐹 (𝑓 , 𝑛, 𝑃 ) (2)

• rain intensity on the test field
• soil moisture at three different depths in the test field A constant wetting rate leads to a gradually increasing 𝑞 and con-
verges in many observed cases to a steady rate. The observed soil
Vegetation density [%] (0–100%), -height [m] (0–0.3 m), land-use moisture data (not shown) shows also a plateau leading to the conclu-
(pasture, arable land), and slope [%] (9–32%) of each site have been sion, that the maximum saturation has been reached, the depression
documented. A detailed description of the experimental setup can be storage is filled and infiltration into the soil is constant. Infiltration
found in Ries et al. (2020). before the termination of artificial rainfall 𝑓𝑚𝑖𝑛 is in many cases equiv-
alent to the equilibrium infiltration rate 𝑓𝑐 resulting in a constant
2.2. The partitioning of roughness and infiltration surface runoff. So infiltration is clearly predictable for these cases as
a difference between precipitation 𝑃 and discharge 𝑞. But no inference
The basic idea is to use the shape of the hydrograph to infer the on 𝑛 can be made in this time segment, as roughness has no visible
roughness of the surface. The shape of the hydrograph is generally
effect on the hydrograph, because runoff is nearly constant (Eq. (3)).
dependent on precipitation rate, infiltration rate and roughness. As
precipitation rate 𝑃 [mm∕h] is constant, the partitioning of runoff 𝑞 d𝑞∕d𝑡 ≅ 0 𝑎𝑛𝑑 𝑞𝑡1 ≤𝑡≤𝑡2 = 𝐹 (𝑓𝑐 , 𝑃 ) (3)
[mm∕h], infiltration 𝑓 [mm∕h] and roughness is the key for determining
1
Manning’s 𝑛 [sm− 3 ]. The relevant fluxes can be seen in the schematic Finally, the third time segment begins with the cessation of rainfall
drawing of a fictive sprinkling experiment in Fig. 1. resulting in the descent of the hydrograph. The decreasing 𝑞 is a
The partitioning of runoff and infiltration changes during the rain- function of 𝑓 and 𝑛 (Eq. (4)).
fall experiment. This behaviour is not only caused by changes in soil
d𝑞∕d𝑡 < 0 𝑎𝑛𝑑 𝑞𝑡2 ≤𝑡≤𝑡𝑒 = 𝐹 (𝑓 , 𝑛) (4)
moisture but also from rainfall induced surface sealing (Assouline,
2004). The impact of raindrops causes surface sealing by the destruc-
A nearly constant infiltration rate can be extrapolated by using the
tion of soil aggregates, which reduces the soil porosity by siltation
fitted Horton function for the falling limb segment of the hydrograph.
(McIntyre, 1958; Farres, 1978). Also chemical dispersion depending on
Consequently the impact of 𝑛 on the shape of this segment of the
the chemistry of the soil affects sealing (Agassi et al., 1996). Soil sealing
hydrograph can finally be separated from infiltration, allowing a clear
is a crucial process for surfaces with low vegetation cover, as the soil
becomes less and less permeable caused by rapidly transforming soil determination of roughness.
surface features with persistent or consecutive rain events (Bresson and In addition, high roughness coefficients cause a slower ascending
Boiffin, 1990; Ribolzi et al., 2011). In addition, Ribolzi et al. (2011) hydrograph. This effect can be used to verify the estimated 𝑛, excluding
stated that high intensity rain on surfaces with a slope gradient less roughness coefficients leading to a simulated slower rise of the hydro-
than 30% (which is the case for the data) leads to less permeable erosion graph simultaneously applying the minimum infiltration rate 𝑓min in
crusts, while higher slopes cause more permeable structures. the model (Eq. (5)).
Consequently the change of infiltration rate over time is a non-linear
curve and must be considered for each rainfall experiment individually. 𝑄𝑠𝑖𝑚𝑡0 ≤𝑡≤𝑡1 ≥ 𝐹 (𝑃 , 𝑛, 𝑓min ) (5)
Most equations describing the infiltration process are monotonically
The collection of rainfall experiments provides a large variety of
decreasing functions based on power laws or exponential decays (As-
different hydrographs with maximum discharge ranging from 5 to 150
souline, 2013). To describe the infiltration process of the artificial rain-
fall experiments, a Hortonian depletion curve (Horton, 1933; Beven, mm∕h. No meaningful roughness can be derived from experiments
2004) is used to calculate the maximum possible infiltration rate 𝑓𝑚𝑎𝑥 of with very low discharges. To recalculate the observed hydrograph, a
the rainfall experiment by fitting the equation to the difference of rain- comparatively much higher roughness would be needed, partly even
fall and observed discharge. The minimum possible infiltration rate 𝑓𝑚𝑖𝑛 with a Manning value greater than 1. This is presumably caused by
can be estimated, when the hydrograph is nearly constant. The actual continuous drainage of soil and vegetation, distorting the hydrograph.
infiltration rate 𝑓 lies between the maximum and minimum infiltration For this reason, experiments whose discharge does not exceed 15 mm∕h
rate, which implies that 𝑓 can only be determined when 𝑓𝑚𝑎𝑥 = 𝑓𝑚𝑖𝑛 . are generally excluded and discharge values below 5 mm∕h are not
The empirical Horton equation decreases 𝑓𝑚𝑎𝑥 exponentially from its considered in the comparison of simulated and observed discharge.

3
D. Feldmann et al. Journal of Hydrology 617 (2023) 128786

2.3. Numerical hydraulic model

3
For one-dimensional and stationary conditions, discharge 𝑞 [ 𝑚
𝑠𝑚
] can
be calculated under consideration of roughness as followed according
to the converted Manning equation (Eq. (6)) (Manning, 1891).
√
1 5
𝑞 = × ℎ 3 × 𝑆𝑓 (6)
𝑛
During the artificial rainfall experiments only 𝑞 has been measured
as integral value and the energy slope 𝑆𝑓 is known as a property of the
respective site. So Manning’s equation does not allow an unambiguous
calculation of 𝑞, as it can be determined with different roughness
coefficients 𝑛 and their respective flow depth ℎ. In addition, the insta-
tionary distribution of flow depth (Fig. 2) cannot be represented by the Fig. 2. Illustration of the model domain displayed as rectangular cells in combination
Manning equation, as well as the ascent and descent of the hydrograph, with the simulated water fluxes.
due to water accumulation to the end of the experimental site. So the
flow processes must be calculated by a two-dimensional approach.
Calculation of instationary flow on a surface is based on the two- applied both for the simulation of fluvial flood events and for surface
dimensional depth-averaged flow equations (shallow water equations). runoff caused by heavy rainfall. The discharge calculation is carried
They integrate the three-dimensional continuity equation and the out via the finite volume method, based on a mesh consisting of
Reynolds-averaged-Navier–Stokes equations for incompressible fluids triangles and quadrilaterals. The use of a combined mesh of triangles
over the water depth by assuming hydrostatic pressure distribution and quadrilaterals allows an easier adaptation to the topographical and
(Pironneau, 1988). The shallow water equations in compact vector hydrodynamic conditions of the respective task. In a standard use case,
format are defined as followed: the structure of the hydraulic model is based on laserscan data. The
𝛿𝒘 𝛿𝒇 𝛿𝒈 model representing the test sites consists of an inclined plane, whose
+ + +𝒔=0 (7)
𝛿𝑡 𝛿𝑥 𝛿𝑦 slope is specified in the database (Fig. 2).
with: As the test sites are homogeneous areas, microtopography is not
considered as height variation in the computational mesh. The surface
⎡𝐻 ⎤ roughness is modelled with water depth-dependent or constant Man-
𝒘 = ⎢𝑢ℎ⎥ (8)
⎢ ⎥ ning coefficients on each node. So all flow affecting parameters like
⎣𝑣ℎ⎦
microtopography, vegetation resistance, splash effects of raindrops or
⎡ 𝑢ℎ ⎤ rill formation due to erosion must be represented by the roughness
⎢ 𝛿𝑢 ⎥
𝒇 = ⎢𝑢2 ℎ + 0.5𝑔ℎ2 − 𝑣ℎ 𝛿𝑥 ⎥ (9) coefficient.
⎢ 𝛿𝑣
𝑢𝑣ℎ − 𝑣ℎ 𝛿𝑥 ⎥ Corresponding to the real experiments, the surface of the model has
⎣ ⎦
a length of 10 m. The width of the site is not relevant, because flow is
⎡ 0 ⎤ measured in discharge per metre. The selection of a sufficient number
𝒔 = ⎢𝑔ℎ(𝑆𝑓 𝑥 − 𝑆𝑏𝑥 )⎥ (10) of nodes has been done by a sensitivity analysis leading to an optimal
⎢ ⎥
⎣ 𝑔ℎ(𝑆𝑓 𝑦 − 𝑆𝑏𝑦 ) ⎦ node spacing of 25 cm. This is the key to exclude an influence of model
⎡ 𝑣ℎ ⎤ discretization on the results. The rain is modelled via so-called source-
⎢ 𝛿𝑢
⎥ nodes as a spatial homogeneous addition of water. The infiltration is
𝒈=⎢ 𝑢𝑣ℎ − 𝑣ℎ 𝛿𝑦 ⎥ (11)
⎢ ⎥ modelled similarly to the rain as negative source-nodes, if necessary,
⎢𝑣 ℎ + 0.5ℎ − 𝑣ℎ ⎥
2 2 𝛿𝑣
in a time dependent manner. The advantage of this procedure in
⎣ 𝛿𝑦 ⎦
comparison to a simple reduction of the amount of rainfall is that the
𝐻 = ℎ+𝑧 which represents the water level as the sum of water depth
infiltration can be active while or after the rainfall. At the lower end
(h) and elevation of the surface (z). 𝑢 and 𝑣 are the velocity components
of the surface, the discharge and the water level is logged for further
in x- and 𝑦-direction. The source term 𝑠 involves the friction slope 𝑺 𝑓
analysis.
(with the components 𝑆𝑓 𝑥 and 𝑆𝑓 𝑦 ) and the bed slope (𝑆𝑏𝑥 , 𝑆𝑏𝑦 ). The
slope of the surface (bed slope) is defined by the gradient of bed level
2.4. Reducing the solution space
𝑧 in 𝑥 and 𝑦 direction:
−𝛿𝑧 −𝛿𝑧 Each experiment leads to a solution space representing several
𝑆𝑏𝑥 = , 𝑆𝑏𝑦 = (12)
𝛿𝑥 𝛿𝑦 acceptable roughness functions with minimal difference in quality of
The friction slope 𝑆𝑓 is calculated with the Darcy-Weisbach equa- the results. Comparing the solution spaces of experiments with similar
tion: properties, overlapping areas can be obtained leading to the assump-
𝜆𝒗 |𝒗| tion, that there must be a smaller general valid solution space.
𝑆𝑓 = (13)
2𝑔𝐷 Three steps are now taken to reduce the solution space. The first
step is to exclude smooth (low 𝑛) roughness relations based on the
With the Manning equation the friction factor 𝜆 is determined:
ascending hydrograph as described in Section 2.2. In the second step,
2𝑔𝑛2 the individual experiments of one site are compared with each other to
𝜆 = 6.34 (14)
𝐷1∕3 find similarities, that are valid for all. Finally, the remaining solution
g is defined as earth gravity and 𝐷 = 4𝑟ℎ𝑦 is representing the space can be compared to sites with similar properties to further reduce
hydraulic diameter. Applying the 2D shallow water equations the hy- or verify the results. Fig. 3 shows the described process schematically.
draulic radius 𝑟ℎ𝑦 is equivalent to the water depth ℎ. Using the finite The method followed in this study for determining the roughness
volume method the shown equations provide depth and flow velocity coefficient is the iterative testing of different roughness values and the
at each node of the computational mesh in a temporal resolution. For subsequent comparison of the hydrographs using a statistical perfor-
our study, we used the hydrodynamic model HYDRO_AS-2D (Hydrotec, mance measure like the Nash Sutcliffe Efficiency (NSE). The identified
2022). The model HYDRO_AS-2D is widely used in Germany and is roughness coefficients with a high NSE score can be compared to the

4
D. Feldmann et al. Journal of Hydrology 617 (2023) 128786

Mathur, 2003). On the basis of these findings Jain et al. (2004) sim-
plified Kadlec’s Power Law for the application in a distributed rainfall
model (Eq. (16)).
( )−𝜖
ℎ
𝑛 = 𝑛0 (16)
ℎ0
with the limits: ℎ < ℎ0 and 𝑛 = 𝑛0 for ℎ ≥ ℎ0 ℎ0 defines minimum flow
depth, beyond which the roughness coefficient 𝑛0 is assumed constant.
The exponent 𝜖 represents the influence of vegetation drag. This for-
mulation of a depth-dependent roughness function is well established
and has been used in several studies (Mügler et al., 2011; Simons
et al., 2014; Özgen et al., 2015). The formulation is opposite to the
exponential function, since it emphasizes low roughness more strongly
for a linear iteration of 𝑛0 . The simplified Power Law can result in a
decreasing or increasing function and has therefore, a wider range of
possible solutions.
Fig. 3. Framework for reducing the solution space.

2.5.4. Fu’s equation

A newer formulation describing depth-dependent roughness based
other experiments with different rain and infiltration rates of the same on the Manning equation has been provided by Fu et al. (2019). Later
test site. If all experiments lead to a similar roughness value, the result the equation has been adapted to be valid for litter cover (Ding et al.,
can be considered as robust solution. 2021). Fu conducted several laboratory experiments for a vegetation
cover from 0 to 30% with artificial plants with a height of 12 cm. Fu’s
2.5. Roughness functions equation leads to a nearly constant roughness for low vegetation cover
and an increasing roughness with higher water depths for a higher
Three depth dependent formulations of near surface roughness and vegetation cover. The equation calculates Manning’s roughness on stem
one depth independent approach with different advantages are used covered slopes and shows that the influence of plant stem resistance,
to parameterize the roughness function to prove the robustness of the flow resistance from vegetation and boundary resistance depends on
presented framework. water depth and plant basal cover (𝐶𝑣 ). 𝐶𝑣 is defined as the ratio
between the area covered by stems (𝐴𝑠 ) and flume bed (𝐴𝑓 ) (Eq. (17)).
2.5.1. Constant roughness coefficients
The use of constant (depth independent) roughness coefficients is 𝐶𝑣 = 100𝐴𝑠 ∕𝐴𝑓 (17)
the most straightforward method. For this use case, no parameteriza-
tion is needed, but the Manning values can be iterated directly within 𝐶𝑣 is regarded as equivalent to the plant coverage documented
the framework. by Ries et al. (2020) as no further information about the methodology
is provided. The coefficients 𝑎 and 𝑏 vary with vegetation type (Eq. (18)
as provided by Fu et al. (2019)).
2.5.2. Exponential function
−0.219𝐶𝑣
Thus, better fitting shapes of the hydrographs and a universally 𝑛 = (𝑎 + 𝑏(1 − 𝑒−0.061𝐶𝑣 )1.668 )ℎ0.604−0.710𝑒 (18)
valid roughness can only be modelled with depth-dependent roughness
relationships. This assumption is confirmed by many authors (Wu, The solution space of Fu’s equation is limited, because the shape of
2008; Fraga et al., 2013; Rai et al., 2010; Mügler et al., 2011; Fu et al., the function is mainly defined by the vegetation coverage. The advan-
2019). To describe depth-dependent roughness we parameterize 𝑛 as tage is, that this reduces the necessary number of possible iterations
a reciprocal exponential function of ℎ defined by the variables 𝑐 and 𝑑 to find a solution, but this could also result in a lack of accuracy.
(Eq. (15)). Fu’s equation has been determined with relatively low stem covers
(<30%) and flow depths larger than 0.5 cm. So the validity for higher
1 vegetation covers, the transferability of Fu’s stem cover to vegetation
𝑛= (15)
𝑐 + 𝑒𝑑ℎ cover, as used in the data of Ries et al. (2019), and lower flow depth
The possible representation of either a constant roughness (𝑑 = 0) needs to be verified.
or a depth-dependent (𝑑 > 0) decreasing roughness with increasing
water depth is the advantage of this function. This allows to test all 3. Results
possible combinations of the variables 𝑐 and 𝑑 in a reasonable range of
values and to compare the observed with the simulated falling limb of The calibration of the three equations leads to multiple water depth-
the hydrograph. The reciprocal is used to achieve a better distribution dependent roughness functions and therefore to differing hydrographs.
of roughness functions, as functions with a low 𝑛 would otherwise Similarity of the resulting falling limb of the simulated hydrograph with
be underrepresented, if 𝑐 and 𝑑 are iterated linearly. Preliminary test the measured one is represented by the NSE as a value between 0 and
simulation showed that the differences in the simulated hydrograph of 1.
a linear varied 𝑛 with high roughness values (0.6 to 1.0) are less pro-
nounced than with a smaller 𝑛 (0.2 to 0.4). The exponential function has 3.1. The most robust result
no direct link to Manning’s equation, but covers all possible variations
of decreasing roughness with increasing water depth. The result is a heat map where each raster cell stands for one depth-
dependent roughness relation and the quality of its comparison to the
2.5.3. Kadlec’s Power Law descending part of the hydrograph. 22 sites out of 23 are useable for
Experimental investigations studying the shallow overland flow simulation (Table 1), since no runoff could be generated at site 10
found a relationship between vegetative drag and roughness coeffi- with any artificial rainfall experiment due to high soil porosity. For
cient (Kadlec, 1990; Wu et al., 1999; Tsihrintzis et al., 2001; Gaur and each site, up to six matrices, depending on the number of useable

5
D. Feldmann et al. Journal of Hydrology 617 (2023) 128786

Table 1
Overview of the most robust parameter combinations together with the relevant properties of their respective experimental sites.
Site Cva Sa Veg. typea const. 𝑛 exp. function Kadlec’s Power Law Fu’s equation
[%] [%] 𝑛 𝑐 𝑑 𝑛b ℎc 𝑛0 𝜖 𝑛b ℎc 𝑎 𝑏 𝑛b ℎc
[mm] [mm] [mm]
1 100 12 Pasture 0.58 0.70 0.59 0.59 5 0.250 0.45 0.47 5 0.3 0.50 0.57 5
2 100 18 Pasture 0.60 0.50 0.63 0.63 5 0.825 0.05 0.88 7 0.8 0.00 0.57 5
3 90 16 Pasture 0.70 0.40 0.69 0.69 6 0.575 0.10 0.66 6 0.7 0.25 0.68 6
4 40 16 Mustard 0.45 0.40 0.53 0.53 5 0.600 −0.15 0.49 5 0.6 0.00 0.43 4
5 0 14 Triticale (seeded) 0.18 3.90 0.20 0.20 4 0.100 0.40 0.17 3 0.0 0.75 0.11 3
6 100 21 Pasture 0.65 0.40 0.67 0.67 5 0.975 0.00 0.97 6 0.6 0.25 0.61 4
7 80 14 Winter barley 0.38 0.20 0.44 0.44 3 0.550 −0.20 0.42 3 0.5 0.00 0.36 3
8 15 16 Corn (seeded) 0.13 7.00 0.12 0.12 3 0.125 −0.20 0.09 2 0.0 0.25 0.09 2
9 100 21 Pasture 0.58 0.40 0.64 0.64 5 0.650 −0.05 0.61 5 0.8 0.00 0.57 5
10 100 32 Pasture – – – – – – – – – – – – –
11 80 18 Pasture 0.68 0.00 0.82 0.82 5 0.750 −0.05 0.70 4 0.4 0.50 0.64 4
12 100 19 Pasture 0.65 0.40 0.67 0.67 4 0.500 0.15 0.62 4 0.6 0.25 0.61 4
13 40 11 Alfalfa 0.28 1.40 0.28 0.28 2 0.475 −0.30 0.31 2 0.4 0.00 0.29 2
14 100 27 Pasture 0.70 0.00 0.82 0.82 5 0.800 −0.05 0.75 5 – – – –
15 0 14 Winter barley 0.50 0.00 0.62 0.62 3 0.225 0.40 0.39 3 0.0 0.75 0.11 2
16 100 12 Pasture 0.40 1.60 0.38 0.38 4 0.550 −0.25 0.39 4 0.5 0.00 0.36 3
17 100 14 Pasture 0.48 0.50 0.53 0.53 5 0.575 −0.10 0.50 5 0.4 0.25 0.46 5
18 60 12 Alfalfa and clover 0.38 1.50 0.39 0.39 4 0.250 0.20 0.33 4 0.5 0.00 0.36 4
19 100 21 Pasture 0.98 0.00 1.00 1.00 6 0.975 0.00 0.97 6 1.0 0.25 0.89 6
20 0 9 Corn (harvested) 0.05 19.50 0.01 0.01 1 0.050 0.00 0.05 2 0.0 0.25 0.04 2
21 50 14 Green manure 0.35 1.30 0.37 0.37 3 0.525 −0.20 0.40 3 0.0 0.50 0.33 3
22 100 12 Pasture 0.38 1.60 0.38 0.38 4 0.175 0.50 0.35 3 0.3 0.25 0.39 4
23 0 14 Corn (harvested) 0.08 7.00 0.10 0.10 2 0.125 −0.10 0.11 2 0.0 0.50 0.07 2
a
Data taken from Ries et al. (2019).
1
b
Manning’s 𝑛 [sm− 3 ] at a water depth of 4 mm.
c
Maximum water depth during simulation.

Fig. 4. Matrices of NSE values of the experiments 3, 4 and 5 on site 9 conducted with a constant 𝑛, the exponential function, Kadlec’s Power Law and Fu’s equation. The crosses
highlight the roughness relations with the smallest deviation from the optimum. The grey marked cells can be excluded as simulated q is smaller than the rising hydrograph due
to too high roughness as described in Section 2.2.

6
D. Feldmann et al. Journal of Hydrology 617 (2023) 128786

Fig. 5. left: Minimum NSE of all six simulatable experiments of site 9. right: Resulting depth dependent roughness functions as indicated by the crosses.

experiments, are generated. The number of useable experiments de- indicate robust solution spaces. The more crosses appear, the more
pends on the successful generation of surface runoff. In many cases, possible parameter combinations lead to a good result. To reduce the
the first experiment or experiments with a short rainfall duration did number of solutions, an overlay of the results of all experiments of
not produce any runoff, because soil moisture could not sufficiently one site is performed. By calculating the minimum NSE values of
build up. Fig. 4 exemplarily shows the matrices of three out of six each parameter combination (Fig. 5), the lowest common denominator
experiments simulated for site 9 (100% veg. cover). representing a solution, which is most acceptable for all experiments,
The crosses in the heat maps demonstrate the deviation from the can be found. In order to verify the selection of the minimum as best
highest NSE value and are an indicator for the ambiguousness of the possible solution, the optimum solution space is compared with the
results of one experiment. Thus, coherent and similar areas with crosses optimum of other statistical moments, such as mean, median, standard

7
D. Feldmann et al. Journal of Hydrology 617 (2023) 128786

deviation, or coefficient of variation (not shown). All the statistical Table 2

Clusters obtained from the parameters calculated with Kadlec’s Power Law.
moments of NSE lead to similar results, as well as the simulated hydro-
graphs resemble to the observed ones. But the minimum is considered Nr Cluster (sites) Common characteristics NSE 𝑛0 𝜖

as the most robust solution between the different experiments. Taking 1 8, 20, 23 Low veg. cov., corn 0.87 0.075 −0.30
2 5, 15 Low veg. cov., grain 0.83 0.100 0.50
the mean values, for instance, would not assure that all the experiments
3 1, 22 Longer grass 0.89 0.200 0.45
are contained in the optimum solution space. 4 3, 9, 12, 17 Shorter grass 0.91 0.650 0.05
On site 9 basically all four approaches lead to a range of 𝑛 values 5 4, 7, 13, 16, 21 Medium density field crops 0.88 0.500 −0.25
between 0.5 and 0.7 (Fig. 5). The exponential function and Kadlec’s 6 2, 6, 11, 14, 19 High density pasture 0.74 0.975 0.00
Power Law result in similar optimum roughness functions (black lines)
with 𝑛-values decreasing with ℎ. But the solution space (grey func-
tions) of Kadlec’s Power Law covers also the optimum solution derived
The exponential function also achieves a high level of consistency,
with Fu’s equation. The formulation of Fu’s equation determines an
unless on sites where the roughness is strongly increasing with flow
increasing 𝑛 with flow depth for higher vegetation coverage due to
depth (e.g. site 8 and 16).
its predefined properties and does not allow much variation for full
Fu’s equation tends to result in lower NSE values, since the adapt-
vegetation coverage. Accordingly all 𝑎−𝑏 combinations generate similar
ability of the formula is limited. Significantly lower results compared
results, as the functions in Fig. 5 overlap each other. The optimum
to Kadlec’s Power Law are achieved on sites where roughness clearly
depth dependent functions differ to a minor degree from the constant
decreases with water depth (e.g. sites 22 and 23). At the same time, less
roughness as the flow process takes in this case completely place in
iterations are necessary to obtain reasonable results, since vegetation
partially submerged vegetation. Therefore also the constant approach
coverage is directly considered in the formulation. Executing a higher
results in a high NSE value, while the overall best fitting function leads number of iterations with a finer parameter sampling interval could not
to a slightly decreasing 𝑛 with ℎ. improve the results significantly.
The resulting parameters together with the most important proper- The quality of the results of constant 𝑛 varies strongly with the
ties of their sites are listed in Table 1. properties of the respective site. Sites with a dense vegetation tend to
To enable a straightforward comparison of the results, the 𝑛-value of have a less pronounced depth dependent roughness and therefore a low
each function at a depth of 4 mm is listed. To indicate the limitation of difference to the other approaches is achieved (e.g. sites 12, 14 and 22).
the respective function, the highest flow depth (ℎ) achieved during the In contrast to sites with low or no vegetation cover, the NSE values
numeric simulation is also presented. Since the maximum flow depth of achieved with the constant 𝑛 decrease significantly compared to the
some functions (e.g. sites 13, 20 and 23) lies significantly below 4 mm, depth dependent approaches (e.g. sites 5, 8 and 23).
they can nonetheless be seen as valid, as they show a nearly constant The variation of roughness is shown in Fig. 8 by comparing rough-
𝑛, which can be extrapolated to 4 mm. ness at a depth of 4 mm with the results derived with Kadlec’s Power
Law.
3.2. Numerical and experimental hydrographs Larger deviations occur mainly at the upper end of the scale. In
particular, the comparison with Fu’s equation shows a larger variance
The comparison between the observed and the simulated falling due to the lower adaptability of the formula. The variation compared
limb of the hydrographs, obtained from applying the four roughness to the exponential function is lower as both approaches lead in many
approaches in the hydraulic model in Fig. 6, lead to nearly identical cases to similar solutions. It is noticeable that the constant roughness
results (NSE values from 0.94 to 0.99). The respective minimum NSE has the smallest deviations. This can be explained by the considered
is equivalent to the optimum value derived from the overlay in Fig. 5. the water depth of 4 mm, since the depth independent approach tends
The NSE values of the single experiments correspond to the values to result in a mean roughness value.
presented in Fig. 4 and the position determined by the optimum value.
The simulation has been conducted with the in Section 2.2 described 3.4. Correlation of parameters and site characteristics
minimum infiltration rate 𝑓𝑚𝑖𝑛 .
The criteria for an acceptable shape of a simulated hydrograph is In comparison to the other approaches, the strongest correlation
defined as followed: First, the rise of the simulated hydrograph must between the parameters and vegetation cover could be ascertained with
be higher or equal to the observed 𝑞, since infiltration is modelled Kadlec’s Power Law. Fig. 9 shows the identified 𝑛0 − 𝜖 values calculated
as the minimum obtained from fitting the difference of precipitation with Kadlec’s Power Law with a maximum deviation of 1% of their
and discharge to the Horton function. Second, the simulated descent optimum in dependency of the degree of vegetation coverage. Low
of the hydrograph must fit to the observed data. Some deviation at vegetation cover corresponds clearly with a low 𝑛0 and an 𝜖-value
the lower end is tolerable, as this is assumed to be caused by soil or approaching zero resulting in a nearly constant roughness function.
vegetation draining effects, which cannot be mapped by the roughness A more differentiated result is obtained focusing on medium vegeta-
formulation. The example of site 9 shows, that these criteria can be tion cover (30–60%). Overall, the roughness function tends to become
met, as well as high NSE coefficients can be achieved independently of smoother with increasing flow depth, corresponding to a negative 𝜖.
the applied roughness formulation. Exceptions can be caused by different plant types, such as at site 18
with alfalfa and clover. High vegetation cover (>60%) results in many
3.3. Quality and variability of the results cases in a high 𝑛0 . Lower 𝑛0 values are associated with a significantly
positive 𝜖 corresponding to a rapidly increasing roughness with flow
The quality of the results is shown in Fig. 7 comparing the achieved depth.
NSE values of the four approaches. High NSE values are achieved if Going further into detail, different clusters with common vegetation
the conducted rainfall experiments resulted in similar hydrographs, characteristics can be obtained from the data (Table 2). To verify the
while low NSE values are caused by diverging discharge measurements. coincidence of the cluster, the minimum NSE coefficient of the sites is
Site 19 is a negative exception because of extremely dense vegetation, calculated to find the best possible concordance, that is acceptable for
so that even with a Manning value of 1, the hydrograph cannot be all sites. For all clusters, an adequate parameter combination can be
modelled correctly. obtained, as the NSE values indicate. The high variability (low NSE)
Considering all sites, it can be stated that Kadlec’s Power Law in high density pasture (cluster 6) can be substantiated with larger
gives the best results on average, since this formula reflects best the deviations in vegetation texture, since there are still major differences,
variability of the vegetation with increasing or decreasing roughness. even if vegetation coverage is assumed to be 100%.

8
D. Feldmann et al. Journal of Hydrology 617 (2023) 128786

Fig. 6. Comparison of observed and simulated hydrographs of all six experiments of site 9. The NSE values assess only the falling limb of the hydrograph and discharge higher
than 5 mm∕h.

As Fu’s equation implies vegetation cover directly, the parameters 𝑎 hydraulic modelling is, that pasture, as a frequently applied land-use,
and 𝑏 should represent the plant type according to Fu et al. (2019). So can be classified by the use of the data of twelve sites resulting in a
an overlay summarizing plant types in order to calculate the minimum solution space around the indicated position of 𝑎 = 0.4 and 𝑏 = 0.5.
NSE value is conducted (Fig. 10). For the shown vegetation types, a best
corresponding parameter combination could be determined. But it must 3.5. Roughness and consecutive rainfall events
be admitted that, for most of the plant types, experiments have been
conducted only on one site. But as alfalfa, mustard and winter barley Ribolzi et al. (2011) state, that consecutive rainfall may cause severe
depict a cluster as well as the recently seeded or harvested types, the changes in the surface structure, especially for higher rain intensities on
results can be seen as plausible. The most important information for bare soil. This possibly affects the roughness coefficient. To visualize

9
D. Feldmann et al. Journal of Hydrology 617 (2023) 128786

Fig. 10. Optimum 𝑎 − 𝑏 combinations for Fu’s equation classified according to

Fig. 7. Comparison of optimum NSE values, ordered by value and labelled with
vegetation type.
the respective site number. Green bars indicate the vegetation coverage [%] of the
respective site.

the effect of the consecutive rainfall experiments, the deviation from

mean 𝑛 of the experiments conducted at a site at a flow depth of 4 mm
is calculated.
The consideration of a single site does not yet show a clear shift of
roughness, but the comparison of all sites implies an influence of rain
intensity and duration on roughness (Fig. 11).
The first and second experiment feature equal rainfall intensity and
duration. The difference in the distribution of 𝑛 is probably caused
by the size of the sample. For the experiments 3 and especially 4,
the rainfall duration was shorter while intensity has been increased.
Here a shift towards a higher 𝑛 can be recognized. In contrast to
experiment 5 with a long duration and significant smaller rain intensity,
the roughness decreases. The last experiment alters again towards a
higher roughness. In conclusion, a general trend in shifting roughness
towards one direction caused by consecutive rain experiments, as stated
by Huang and Bradford (1992), cannot be recognized. Probably, the
high variability in intensity and duration overlays this effect. But it can
be concluded, that a high intensity and especially a short duration leads
to significantly higher roughness values (exp. 4 and 6), while a long
Fig. 8. Comparison of optimum 𝑛 calculated with Kadlec’s Power Law compared with duration has the opposite effect (exp. 5). During longer rain events,
the other three approaches at a flow depth of 4 mm. more pronounced flow paths and grooves may develop, while the
raindrop impact may increase roughness during shorter events (Fraga
et al., 2013). The effect is slightly stronger on bare soil, but the sample
size is not sufficient to make robust conclusions.

4. Discussion

Our study shows that the presented framework robustly determines

roughness values from artificial rainfall experiments without data of
flow velocity or water depth. Independently of the used method, all
four applied approaches result in similar roughness functions, while
they offer different advantages.
A constant roughness is the most straightforward formulation, but
results only in valid values as long as the depth dependence is negligi-
ble, which is mainly the case for surface runoff in partially submerged
vegetation.
The exponential function covers all possible decreasing or constant
water depth functions and leads, in these cases, to accurate roughness
relations. The disadvantage is, that there is no clear correlation between
the calculated parameters and vegetation cover.
Kadlec’s Power Law covers both, decreasing and increasing func-
tions and therefore, delivers a wider range of functions. The results
Fig. 9. 𝑛0 − 𝜖 combinations calculated with Kadlec’s Power Law with maximum 1%
deviation from optimum point of the respective site.
indicate, that increasing or decreasing functions are necessary to de-
scribe the near surface roughness as different vegetation types may

10
D. Feldmann et al. Journal of Hydrology 617 (2023) 128786

Fig. 11. Roughness variability caused by rainfall intensity and -duration. Top row: duration 𝑡 of the rain experiments (orange bars). Middle row: precipitation intensity 𝑃 (blue
bars) combined with its variability (light blue bars). Bottom row: Deviation from mean 𝑛 (at a flow depth of 4 mm) of the respective site. The distribution is shown by violin plots,
while blue dots presents the number of samples. The red line shows the median of the distribution.

cause different shapes of the function. In addition a clear link between vegetation on surface runoff. Water-depth-dependent Manning curves
vegetation properties and the parameters could be recognized (Fig. 9 were determined for different vegetation heights. With fully submerged
and Table 2) vegetation, the roughness coefficient tends to increase at low depths,
Fu’s equation, based on the Manning equation and empirical ob- but then decreases to an asymptotic constant, as the water level con-
servations, needs much less iterations to find a suitable roughness tinues to rise. Since the mattresses are equivalent to a dense vegetation
function, as vegetation cover is directly implemented in the formula- coverage, they correspond well to the results of the present study.
tion. Fu et al. (2019) states that the parameters are dependent on plant Díaz (2005) combined laboratory and field experiments, investi-
type, which is confirmed by our results and valid parameters for the gating the influence of different vegetation types on roughness. For
near surface can be provided (Fig. 10). grassland, similar values were determined as in Wu et al. (1999). Fraga
The result of the study are water depth-dependent or constant et al. (2013) conducted rainfall runoff experiments on concrete and
roughness relationships for water depths under 1 cm. Further research grass surfaces and tried to predict the rising and falling limb of the
should generate higher water depths in field experiments, which can observed hydrograph with a numerical model solving the unsteady
be achieved using longer test fields (in the range of 10 m or higher) shallow water equations. Compared to the other studies shown, a very
or higher rain intensities. However, the relevance of this water depth low vegetation height (5 mm) was used, which leads to a more rapidly
range for hydraulic heavy rainfall simulations should not be under- decreasing roughness. The results showed also a significant rise of
estimated, as very shallow water depths are becoming increasingly roughness at very low water depths and an increasing surface resistance
important in 2D-hydraulic modelling with high resolution terrain mod- with higher rain intensities due to the raindrop impact. This correlates
els (Rai et al., 2010). The choice of roughness leads to diverging water with our findings as we detected also a coherence of rain intensity and
depth, flow velocities and thus different concentration times resulting roughness (Fig. 11).
in a temporal and quantitative diverging peak discharge (Sauer and Several authors compared the roughness functions in dependency of
Ortlepp, 2021; Sanz-Ramos et al., 2021). Our results show clear de- the degree of submergence with a strong agreement between different
pendencies of the roughness coefficient with vegetation cover, plant experimental setups (Graf and Chhun, 1976; Wilson and Horritt, 2002;
type and rain intensity, which should be taken into account in hydraulic Fraga et al., 2013). But this comparison is not appropriate for our data,
simulations to improve the quality of the results. This study therefore because flow depth in the used experimental setup did not submerge
contributes to more robust calibrations in infiltration models, since the vegetation.
influence of roughness on the shape of the hydrograph can now be Fu et al. (2019) conducted laboratory experiments on an inclined
estimated with higher accuracy. flume and varied artificial vegetation cover from almost 0 to 30 %.
Overall, our study is in agreement with previous studies demonstrat- Their research stated an increasing roughness with flow depth which
ing the increasing or decreasing roughness with increasing water depth is contrary to the other studies, but can be explained by the used
in laboratory and field experiments (Díaz, 2005; Fraga et al., 2013; Wu vegetation as the density increases with flow depth. The great achieve-
et al., 1999; LUBW, 2016; Graf and Chhun, 1976; Fu et al., 2019). A ment of their work is the derived empirical formula (Eq. (18)) directly
summary of these values is shown in Fig. 12. considering the influence of vegetation coverage.
For example, Wu et al. (1999) investigated the impact of rough- The LUBW (2016), an environmental authority of the state of Baden-
ness in dependence to water level using an artificial channel and Wuerttemberg in Germany, published a range of minimum and maxi-
horsehair mattresses representing the influence of a ground-near dense mum plausible depth-dependent roughness values for the application in

11
D. Feldmann et al. Journal of Hydrology 617 (2023) 128786

Fig. 12. Depth-dependent roughness relations from literature compared to the results achieved with Kadlec’s Power Law. The label in the legend gives information about examined
vegetation type and its height.

2D-hydraulic surface runoff modelling. For a water depth higher than Finally, the study shows clearly the possibilities and limitations
5 cm, they fit quite well to the other values presented, although the veg- of estimating roughness coefficients with data from artificial sprin-
etation height was not considered. Comparing the given range of short kling experiments. The results of the study can be directly applied
grass and grassland to the present study, they highly underestimate the in hydraulic 2D modelling, as they give modellers an orientation for
roughness for the near surface runoff. the selection of parameters, particularly, when low water depths are
We now provide robust values describing roughness for near surface crucial for the generation of surface runoff, which is important for the
runoff (<1 cm) for different types of vegetation and degree of vegetation simulation of heavy rainfall events.
coverage. The results give the hydraulic modeller an orientation, what
range of roughness to choose for near surface runoff (Fig. 12). Fur-
thermore, the knowledge gained about successive rain events can now CRediT authorship contribution statement
help to better assess the influence of rainfall duration and intensity. For
example, the roughness of very short and intense rain events would be David Feldmann: Conceptualization, Methodology, Software, Vi-
underestimated with averaged values (Fig. 11). sualization, Writing – original draft. Patrick Laux: Conceptualiza-
tion, Writing – original draft, Supervision, Project administration, Fund
5. Conclusions
acquisition. Andreas Heckl: Writing – review & editing. Manfred
Schindler: Writing – review & editing. Harald Kunstmann: Writing
Manning’s 𝑛 or a depth-dependent function of 𝑛 is suitable to calcu-
– review & editing, Supervision.
late the behaviour of surface runoff. A general valid roughness function
should include the degree of vegetation coverage and the vegetation
type. Declaration of competing interest
Based on the main findings of our study, roughness values can be
determined by measuring the outflow based on a comparably simple
The authors declare that they have no known competing finan-
framework as. This renders the conduction of complex measurements
of flow velocity and water level unnecessary. cial interests or personal relationships that could have appeared to
The derived roughness values extend the current knowledge for influence the work reported in this paper.
shallow water depths. It is shown, that roughness has tended to be
underestimated in some cases. For all examined sites, robust values Data availability
could be derived, while surfaces with no vegetation or an extremely
high vegetation density showed largest variability. For bare soil, this
can be explained by rapid changes in the surface structure, potentially The authors do not have permission to share data.
caused by crust formation and erosion. The variability of high density
vegetation is assumed to be caused by plant type, as high vegetation Acknowledgements
coverage alone is not able to describe the differences between the sites.
This study contributes also to a more accurate calibration of infiltra-
tion models. Roughness has been estimated using the falling limb of the We highly acknowledge the data from artificial rainfall experiments
hydrograph when infiltration is nearly constant. Based on the identified provided by Ries et al. (2019). The raw data can be retrieved from
roughness the impact of infiltration rate and roughness on the increas- https://doi.org/10.6094/UNIFR/151460. This work was funded by the
ing segment of the hydrograph can be clearly separated. Consequently, BMBF, Germany research project KARE (grant number 01LR2006D).
it is possible to precisely calculate the change in infiltration rate for the We thank the two anonymous reviewers for their fruitful comments,
entire artificial rainfall experiment. which helped to improve the quality of the manuscript.

12
D. Feldmann et al. Journal of Hydrology 617 (2023) 128786

References Kaiser, M., Borga, M., Disse, M., 2020. Occurrence and characteristics of flash floods
in bavaria (Germany). In: Climate Change Management. pp. 293–310. http://dx.
Agassi, M., Benyamini, Y., Morin, J., Marish, S., Henkin, E., 1996. Runoff and erosion doi.org/10.1007/978-3-030-37425-9{_}16.
control in Israel. In: Runoff, Infiltration and Subsurface Flow of Water in Arid and Katz, D.M., Watts, F.J., Burroughs, E.R., 1995. Effects of surface roughness and rainfall
Semi-Arid Regions, Vol. 21. Springer, Dordrecht, pp. 63–120. http://dx.doi.org/10. impact on overland flow. J. Hydraul. Eng. 121 (7), 546–553. http://dx.doi.org/10.
1007/978-94-017-2929-1_2. 1061/(ASCE)0733-9429(1995)121:7(546).
Assouline, S., 2004. Rainfall-induced soil surface sealing: A critical review of ob- Kim, J., Ivanov, V.Y., Katopodes, N.D., 2012. Hydraulic resistance to overland flow on
servations, conceptual models, and solutions. Vadose Zone J. 3 (2), 570–591. surfaces with partially submerged vegetation. Water Resour. Res. 48 (10), W10540.
http://dx.doi.org/10.2136/VZJ2004.0570. http://dx.doi.org/10.1029/2012WR012047.
Assouline, S., 2013. Infiltration into soils: Conceptual approaches and solutions. Water Lane, S.N., 2005. Roughness – time for a re-evaluation? Earth Surf. Process. Landf. 30
Resour. Res. 49 (4), 1755–1772. http://dx.doi.org/10.1002/wrcr.20155. (2), 251–253. http://dx.doi.org/10.1002/ESP.1208.
Barros, A.P., Colello, J.D., 2001. Surface roughness for shallow overland flow on Lawrence, D.S., 2000. Hydraulic resistance in overland flow during partial and marginal
crushed stone surfaces. J. Hydraul. Eng. 127 (1), 38–52. http://dx.doi.org/10.1061/ surface inundation: Experimental observations and modeling. Water Resour. Res.
(asce)0733-9429(2001)127:1(38). 36 (8), 2381–2393. http://dx.doi.org/10.1029/2000WR900095.
Beven, K., 2004. Robert E. Horton’s perceptual model of infiltration processes. Hydrol. LUBW, 2016. Leitfaden Kommunales Starkregenrisikomanagement in Baden-
Process. 18 (17), 3447–3460. http://dx.doi.org/10.1002/hyp.5740. Württemberg. LUBW Landesanstalt für Umwelt Messungen und Naturschutz
Bresson, L.M., Boiffin, J., 1990. Morphological characterization of soil crust devel- Baden-Württemberg, Karlsruhe.
opment stages on an experimental field. Geoderma (Neth.) 47 (3), 301–325. Manning, R., 1891. On the flow of water in open channels and pipes. Trans. Inst. Civ.
http://dx.doi.org/10.1016/0016-7061(90)90035-8. Eng. Irel. 20, 161–207.
Bulti, D.T., Abebe, B.G., 2020. A review of flood modeling methods for urban pluvial McIntyre, D.S., 1958. Permeability measurements of soil crusts formed by raindrop
flood application. Model. Earth Syst. Environ. 6 (3), 1293–1302. http://dx.doi.org/ impact. Soil Sci. 85 (4), 185–189.
10.1007/s40808-020-00803-z. Moré, J., Garbow, B., Hillstrom, K., 1980. User guide for MINPACK-1. [In FORTRAN].
Cantisani, A., Giosa, L., Mancusi, L., Sole, A., 2014. FLORA-2D : A new model to Technical Report, Argonne National Laboratory (ANL), Argonne, IL (United States),
simulate the inundation in areas covered by flexible and rigid vegetation. Int. J. http://dx.doi.org/10.2172/6997568.
Eng. Innov. Technol. (IJEIT) 3 (8), 179–186. Mügler, C., Planchon, O., Patin, J., Weill, S., Silvera, N., Richard, P., Mouche, E., 2011.
Chow, V.T., 1959. Open Channel Hydraulics. McGraw-Hill, New-York. Comparison of roughness models to simulate overland flow and tracer transport
David, A., Schmalz, B., 2020. Flood hazard analysis in small catchments: Comparison experiments under simulated rainfall at plot scale. J. Hydrol. 402 (1–2), 25–40.
of hydrological and hydrodynamic approaches by the use of direct rainfall. J. Flood http://dx.doi.org/10.1016/j.jhydrol.2011.02.032.
Risk Manag. 13:e12639 (4), http://dx.doi.org/10.1111/JFR3.12639. Özgen, I., Teuber, K., Simons, F., Liang, D., Hinkelmann, R., 2015. Upscaling the
Díaz, R.G., 2005. Analysis of Manning coefficient for small-depth flows on vegetated shallow water model with a novel roughness formulation. Environ. Earth Sci. 74
beds. Hydrol. Process. 19 (16), 3221–3233. http://dx.doi.org/10.1002/hyp.5820. (11), 7371–7386. http://dx.doi.org/10.1007/S12665-015-4726-7.
Ding, L., Fu, S., Zhao, H., 2021. Hydraulic properties affected by litter and stem Pironneau, O., 1988. Finite Element Methods for Fluids. Wiley, Paris.
cover under overland flow. Hydrol. Process. 35 (3), http://dx.doi.org/10.1002/
Rai, R.K., Upadhyay, A., Singh, V.P., 2010. Effect of variable roughness on runoff. J.
hyp.14088.
Hydrol. 382 (1–4), 115–127. http://dx.doi.org/10.1016/j.jhydrol.2009.12.022.
Ding, Y., Jia, Y., Wang, S.S.Y., 2004. Identification of manning’s roughness coefficients
Ribolzi, O., Patin, J., Bresson, L.M., Latsachack, K.O., Mouche, E., Sengta-
in shallow water flows. J. Hydraul. Eng. 130 (6), 501–510. http://dx.doi.org/10.
heuanghoung, O., Silvera, N., Thiébaux, J.P., Valentin, C., 2011. Impact of slope
1061/(ASCE)0733-9429(2004)130:6(501).
gradient on soil surface features and infiltration on steep slopes in northern laos.
Emmett, W.W., 1970. The hydraulics of overland flow on hillslopes, Vol. 662-A. U.S.
Geomorphology 127 (1–2), 53–63. http://dx.doi.org/10.1016/J.GEOMORPH.2010.
Geological Survey Professional Paper, http://dx.doi.org/10.3133/PP662A.
12.004.
Engman, E.T., 1986. Roughness coefficients for routing surface runoff. J. Irrig. Drain.
Ries, F., Kirn, L., Weiler, M., 2019. Runoff response from extreme rainfall events
Eng. 112 (1), 39–53. http://dx.doi.org/10.1061/(asce)0733-9437(1986)112:1(39).
on natural hillslopes: A data set from 132 large-scale sprinkling experiments in
Esen, I., 1987. Least-squares estimates of the Horton infiltration parameters. Soil Sci.
south-western Germany. 2. Version. http://dx.doi.org/10.6094/UNIFR/151460.
144 (1), 6–10.
Ries, F., Kirn, L., Weiler, M., 2020. Runoff reaction from extreme rainfall events on
Farres, P., 1978. The role of time and aggregate size in the crusting process. Earth
natural hillslopes: A data set from 132 large-scale sprinkling experiments in south-
Surf. Process. 3 (3), 243–254. http://dx.doi.org/10.1002/ESP.3290030304.
western Germany. Earth Syst. Sci. Data 12 (1), 245–255. http://dx.doi.org/10.
Fernández-Pato, J., Caviedes-Voullième, D., García-Navarro, P., 2016. Rainfall/runoff
5194/essd-12-245-2020.
simulation with 2D full shallow water equations: Sensitivity analysis and calibration
Romkens, M.J., Wang, J.Y., 1986. Effect of tillage on surface runoff. Trans. Am. Soc.
of infiltration parameters. J. Hydrol. 536, 496–513. http://dx.doi.org/10.1016/j.
Agric. Eng. 29 (2), 429–433. http://dx.doi.org/10.13031/2013.30167.
jhydrol.2016.03.021.
Sañudo, E., Cea, L., Puertas, J., 2020. Modelling pluvial flooding in urban areas
Fraga, I., Cea, L., Puertas, J., 2013. Experimental study of the water depth and rainfall
coupling the models iber and SWMM. Water 12 (9), 2647. http://dx.doi.org/10.
intensity effects on the bed roughness coefficient used in distributed urban drainage
3390/w12092647.
models. J. Hydrol. 505, 266–275. http://dx.doi.org/10.1016/j.jhydrol.2013.10.005.
Fu, S., Mu, H., Liu, B., Yu, X., Liu, Y., 2019. Effect of plant basal cover on velocity Sanz-Ramos, M., Bladé, E., González-Escalona, F., Olivares, G., Aragón-Hernández, J.L.,
of shallow overland flow. J. Hydrol. 577, 123947. http://dx.doi.org/10.1016/J. 2021. Interpreting the manning roughness coefficient in overland flow simulations
JHYDROL.2019.123947. with coupled hydrological-hydraulic distributed models. Water (Switzerland) 13
Gaur, M.L., Mathur, B.S., 2003. Modeling event-based temporal variability of flow (23), 3433. http://dx.doi.org/10.3390/w13233433.
resistance coefficient. J. Hydrol. Eng. 8 (5), 266–277. http://dx.doi.org/10.1061/ Sauer, A., Ortlepp, R., 2021. Parameter uncertainties in flood hazard analysis of heavy
(asce)1084-0699(2003)8:5(266). rain events. ASCE-ASME J. Risk Uncertain. Eng. Syst. A 7 (2), http://dx.doi.org/
Graf, W.H., Chhun, V.H., 1976. Manning’s roughness for artificial grasses. J. Irrigation 10.1061/AJRUA6.0001125.
Drain. Div. 102 (4), 413–423. http://dx.doi.org/10.1061/JRCEA4.0001116. Savant, G., Trahan, C.J., Pettey, L., McAlpin, T.O., Bell, G.L., McKnight, C.J., 2019.
Horritt, M.S., Bates, P.D., Mattinson, M.J., 2006. Effects of mesh resolution and Urban and overland flow modeling with dynamic adaptive mesh and implicit
topographic representation in 2D finite volume models of shallow water fluvial diffusive wave equation solver. J. Hydrol. 573, 13–30. http://dx.doi.org/10.1016/
flow. J. Hydrol. 329 (1–2), 306–314. http://dx.doi.org/10.1016/J.JHYDROL.2006. J.JHYDROL.2019.03.061.
02.016. Seibert, S.P., Auerswald, K., Fiener, P., Disse, M., Martin, W., Haider, J., Michael, A.,
Horton, R.E., 1933. The Role of infiltration in the hydrologic cycle. Trans. Am. Geophys. Gerlinger, K., 2011. Surface runoff from arable land – a homogenized data base
Union 14 (1), 446–460. http://dx.doi.org/10.1029/TR014i001p00446. of 726 rainfall simulation experiments. Hydrol. Earth Syst. Sci. (Special issue),
Huang, C.-h., Bradford, J., 1992. Applications of a laser scanner to quantify soil http://dx.doi.org/10.1594/GFZ.TR32.2.
microtopography. Soil Sci. Am. J. 56 (1), 14–21. http://dx.doi.org/10.2136/ Sepaskhah, A.R., Bondar, H., 2002. Estimation of manning roughness coefficient for
sssaj1992.03615995005600010002x. bare and vegetated furrow irrigation. Biosyst. Eng. 82 (3), 351–357. http://dx.doi.
Hydrotec, 2022. HYDRO_AS-2D - 2D-flow model for water management applications - org/10.1006/bioe.2002.0076.
Reference Manual. Simons, F., Busse, T., Hou, J., Özgen, I., Hinkelmann, R., 2014. A model for overland
Jain, M.K., Kothyari, U.C., Ranga Raju, K.G., 2004. A GIS based distributed flow and associated processes within the Hydroinformatics Modelling System. J.
rainfall–runoff model. J. Hydrol. 299 (1–2), 107–135. http://dx.doi.org/10.1016/ Hydroinform. 16 (2), 375–391. http://dx.doi.org/10.2166/HYDRO.2013.173.
J.JHYDROL.2004.04.024. Smith, M.W., 2014. Roughness in the earth sciences. Earth-Sci. Rev. 136, 202–225.
Jarrett, R.D., 1990. Hydrologic and hydraulic research in mountain rivers. J. Am. http://dx.doi.org/10.1016/j.earscirev.2014.05.016.
Water Resour. Assoc. 26 (3), 419–429. http://dx.doi.org/10.1111/J.1752-1688. Tatard, L., Planchon, O., Wainwright, J., Nord, G., Favis-Mortlock, D., Silvera, N.,
1990.TB01381.X. Ribolzi, O., Esteves, M., Huang, C.H., 2008. Measurement and modelling of high-
Kadlec, R.H., 1990. Overland flow in wetlands: Vegetation resistance. J. Hydraul. Eng. resolution flow-velocity data under simulated rainfall on a low-slope sandy soil. J.
116 (5), 691–706. http://dx.doi.org/10.1061/(ASCE)0733-9429(1990)116:5(691). Hydrol. 348 (1–2), 1–12. http://dx.doi.org/10.1016/j.jhydrol.2007.07.016.

13
D. Feldmann et al. Journal of Hydrology 617 (2023) 128786

Tsihrintzis, V.A., Wu, F.-C., Shen, H.W., Chou, Y.-J., 2001. Discussion and closure: Wu, F.-C., Shen, H.W., Chou, Y.-J., 1999. Variation of roughness coefficients for
Variation of roughness coefficients for unsubmerged and submerged vegeta- unsubmerged and submerged vegetation. J. Hydraul. Eng. 125 (9), 934–942. http:
tion. J. Hydraul. Eng. 127 (3), 241–245. http://dx.doi.org/10.1061/(ASCE)0733- //dx.doi.org/10.1061/(ASCE)0733-9429(1999)125:9(934).
9429(2001)127:3(241). Zhang, W., Cundy, T.W., 1989. Modeling of two-dimensional overland flow. Water
Wilson, C.A., Horritt, M.S., 2002. Measuring the flow resistance of submerged grass. Resour. Res. 25 (9), 2019–2035. http://dx.doi.org/10.1029/WR025I009P02019.
Hydrol. Process. 16 (13), 2589–2598. http://dx.doi.org/10.1002/HYP.1049. Zhang, J., Zhang, S., Chen, S., Liu, M., Xu, X., Zhou, J., Wang, W., Ma, L., Wang, C.,
Wu, F.-s., 2008. Characteristics of flow resistance in open channels with non-submerged 2021. Overland flow resistance law under sparse stem vegetation coverage. Water
rigid vegetation. J. Hydrodyn. 20 (2), 239–245. http://dx.doi.org/10.1016/S1001- (Switzerland) 13 (12), 1657. http://dx.doi.org/10.3390/w13121657.
6058(08)60052-9.

14