Documente Academic
Documente Profesional
Documente Cultură
Journal of Hydrology
journal homepage: www.elsevier.com/locate/jhydrol
a r t i c l e i n f o s u m m a r y
Article history: Soil erosion models need detailed runoff and infiltration data as input. This information can be deduced
Received 11 April 2011 from empirical relationships or from dynamic models. Here we investigate to what extent it is possible to
Received in revised form 28 October 2012 derive meaningful parameterizations for the Green–Ampt Mein-Larson infiltration model (GAML) using a
Accepted 31 October 2012
dataset consisting of 350 rainfall simulation experiments carried out over a period of 6 years in Central
Available online 9 November 2012
This manuscript was handled by Corrado
Belgium. We applied three different parameterization strategies for the Green–Ampt Mein-Larson model
Corradini, Editor-in-Chief to evaluate the number of degrees of freedom that would result in optimum model performance. We
found that a good description of the evolution of infiltration during an experiment was achieved for most
Keywords:
experiments independent of the number of parameters that was allowed to vary. However, acceptable
Rainfall simulation simulations, based on estimated parameter values, could only be obtained for rainfall simulations per-
Green–Ampt Mein-Larson infiltration model formed in spring or early summer when the soils were relatively dry and infiltration was dominantly con-
Model calibration trolled by soil surface conditions. Under these conditions, major controls in effective hydraulic
Soil cover conductivity were soil cover and crusting stage as well as the clay content of the topsoil. Even under these
Soil crusting circumstances, good simulations were only possible when only a single parameter value needed to be
estimated from the measured data. When more than one parameter value needed to be estimated, pre-
dictions were much poorer, mainly due to the large uncertainty associated with the parameter value esti-
mates.
No good simulations were possible for the winter period: due to the wet state of the soil, infiltration is
then no longer controlled by soil surface conditions so that it became impossible to derive meaningful
parameter values form the measured field data. This may pose severe limitations on runoff prediction
from agricultural land under these conditions. In comparison to a regression-based model based on
the available data, the use of the GAML model resulted in slightly worse predictions but the use of a
dynamic model allows for greater flexibility. However, one should be aware that local calibration and val-
idation remain necessary and that the prediction uncertainty remains relatively large.
Ó 2012 Elsevier B.V. All rights reserved.
0022-1694/$ - see front matter Ó 2012 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.jhydrol.2012.10.051
A. Van den Putte et al. / Journal of Hydrology 476 (2013) 332–344 333
empirical in nature. Dynamic infiltration models are easier to was performed by Nearing et al. (1996) using the approach devel-
incorporate in spatially distributed modelling frameworks than oped by Risse et al. (1994) and including data for cropped plots.
the curve number method, but require more input data as well Both the investigations showed that the GAML model performed
as proper calibration before they can be successfully applied. They better than the curve number method.
are still much easier to implement than the full Richards equation, All the studies discussed above attempted to derive average val-
without a major loss in prediction accuracy if the conditions for ues for Ke, while not accounting explicitly for the temporal varia-
which the model was developed are satisfied (Mein and Larson, tion that might be induced by changing soil surface conditions
1973). due to crusting, crop growth etc. Freebairn et al. (1989) found that
The Green–Ampt infiltration equation is one of the most widely the optimal Ke value to simulate runoff production was positively
used infiltration equations in hydrological/erosion models. It has, related to soil cover and negatively to the amount of rainfall prior
among others, been incorporated in the WEPP (Laflen et al., to the rainfall event. Risse et al. (1995a,b) used a series of fallow
1997), SWAT (Tuppad et al., 2011) and ANSWERS (Bouraoui and plot data to assess the effect of crusting on the temporal variability
Dillaha, 1996) models. In its original form, the model was designed of the effective hydraulic conductivity. They found that crusting
to simulate infiltration under ponding conditions in a homoge- negatively affected Ke, while there was a positive relationship with
neous soil. Over time, several modifications have been proposed. event size and again a negative relationship with initial soil mois-
Mein and Larson (1973) modified the model so that infiltration ture content. Introducing submodels for these factors in the calcu-
during a steady rainfall event could be simulated: this form of lation procedure allowed improving the average ME from 0.56 to
the model is commonly called the Green–Ampt Mein-Larson 0.70. Zhang et al. (1995a,b) used the results of this study to calcu-
(GAML) model. Other studies have added additional flexibility to late temporally variable values for Ke for row crops. They found
the Green–Ampt model and formulations exist to account for the that when crops are present effective hydraulic conductivity de-
development of a soil crust (Rawls et al., 1990), temporally-varying pends on the effective total soil cover and the total event precipi-
effective hydraulic conductivity (Ahuja and Tsuji, 1976) and lay- tation. ME ranged from 0.30 to 0.82 with an average of 0.66,
ered soil profiles (Childs and Bybordi, 1969). Esteves et al. (2000) excluding one site for which predictions were particularly poor.
used a two-layer formulation to account for crust formation on A similar analysis for perennial crops was carried out by Zhang
bare soils in the Sahel. A review of all variants of the Green–Ampt et al. (1995a,b), whereby an extra correction was introduced, sim-
model has recently been published by Kale and Sahoo (2011). ply multiplying the effective hydraulic conductivity as calculated
However, in most model implementations the GAML model as for row crops by 1.81. Resulting ME values were between 0.28
modified by Chu (1978) to account for temporally varying rainfall and 0.68. Esteves et al. (2000) did not calibrate the effective
is used. The assumption that the GAML model can be applied using hydraulic conductivity in their model, but the suction at the wet-
spatially uniform parameter values is acceptable as a first approx- ting front.
imation at the plot scale. At the field scale more complex formula- The overview above shows that calibration of the GAML model
tions of infiltration accounting for the spatial variability of is indeed possible and can be reasonably successful. The ap-
hydraulic soil properties should be used, otherwise predicted run- proaches vary in complexity with respect to the (assumed)
off simulations may differ strongly from observations (Corradini description of the factors controlling infiltration and hence also
et al., 2002). Formulations of this type involving the GAML model with respect to the information that is required. It is, however, still
as basic support were proposed by Govindaraju et al. (2001) for unclear whether the incorporation of additional effects through a
homogeneous soil vertical profiles and Corradini et al. (2011) for more refined model description and/or through more refined
two-layered soil profiles with a more permeable upper layer. parameter estimation methods do indeed result in better predic-
Theoretically, the GAML model describes piston-type infiltra- tive capabilities. Clearly, studies like the ones carried out by e.g.
tion, whereby rain water is drawn into the soil based on the com- Freebairn et al. (1989) and Zhang et al. (1995a, 1995b) show that
bined effects of capillary suction and gravity, resulting in a sharply accounting for more crusting and vegetation effects does allow
defined wetting front. The basic GAML model has two parameters for a better simulation. Thus, the model is better capable of
for which values need to be provided, being the effective hydraulic describing variations in runoff production when these effects are
conductivity (Ke) and the capillary pressure at the wetting front incorporated. However, at least two main reasons can be identified
(w). The use of adequate values for these parameters is key to run- as to why models may not necessarily allow making better inde-
off and erosion prediction. The results of process-based overland pendent predictions when more complex process descriptions,
flow and erosion models have repeatedly been shown to be (very) requiring more parameterization, are introduced. First, more com-
sensitive to the values used, especially with respect to hydraulic plex descriptions generally require more input data which are al-
conductivity (e.g. De Roo and Jetten, 1999; Nearing et al., 1990). ways subject to measurement errors and uncertainty: at a certain
Furthermore, one might expect the model to produce reasonable point the uncertainty associated with additional input data will
predictions only when the field conditions correspond more or less overwhelm the gain in predictive ability due to a better process
to the basic conditions for which the model was designed. description (Van Rompaey and Govers, 2002). Second, having more
Several studies have been carried out to derive parameter val- parameters may lead to overparameterization: parameter values
ues for the GAML model. Brakensiek and Onstad (1977) used pond- will then become highly interdependent, poorly constrained and
ed infiltration, but it is well known that such data cannot be difficult to relate to measured system properties so that it becomes
considered to be representative for infiltration under rainfall. Risse impossible to obtain meaningful parameter values independently
et al. (1994) derived ‘optimal’ values for the effective hydraulic (Beven, 1989).
conductivity over a series of rainfall events (thus not allowing for With respect to the GAML model it is not known whether or not
temporal variability in effective hydraulic conductivity) on fallow more sophisticated model formulations and/or an increased num-
land by minimising the least square error on runoff production ber of parameters to be calibrated indeed lead to better prediction
as measured on a series of USLE plots on seven different sites. Mod- results. In particular, the possible improvement that may result in
el efficiency (ME) values for different sites were between 0.07 and overland flow prediction by allowing the model to account for
0.84. Risse et al. (1995a,b) also compared the runoff predictions ob- changes in soil surface properties during a rainfall event have hith-
tained with optimised values with those obtained when the curve erto not been systematically explored. Furthermore, information
number method was used to derive so-called baseline hydraulic on how the effects of land management techniques should be
conductivity values under fallow conditions. A similar analysis accounted for when the GAML model is used for infiltration
334 A. Van den Putte et al. / Journal of Hydrology 476 (2013) 332–344
simulation is still scarce. Our knowledge of factors controlling run- The GAML model describes the situation when runoff occurs
off generation on arable land has increased significantly because only after some time, called the ponding time tp (h). Two equations
large amounts of data were collected, both under natural and sim- describe the two stages of the GAML model (Mein and Larson,
ulated rainfall (e.g. Auerswald and Haider, 1996; Cerdan et al., 1973):
2002; Freebairn and Gupta, 1990; Le Bissonnais et al., 2005; Leys
et al., 2007). However, these data were mainly used to develop F ¼ rt for t 6 tp ð2Þ
and/or calibrate empirical models. To the best of our knowledge,
only Zhang et al. (1995a) and Zhang et al. (1995b) proposed tested wDh þ F
F ¼ K e ðt t p Þ þ F p þ wDh ln for t P t p ð3Þ
methodologies to account for the presence of a crop and/or residue wDh þ F p
in Green and Ampt parameterization. Such information is vital for
practical applications, as reductions in runoff and soil loss can be where r is the rainfall intensity (mm h1), Fp is the cumulative infil-
most easily reached through adjusted land management strategies. tration until ponding (mm) and t is the time (h).
The availability of a well-parameterised GAML model accounting Ponding time can be calculated by applying Eq. (2) at ponding
for these effects would allow to evaluate to what extent modified time and using it to rearrange Eq. (1). This yields following
practices may alter runoff generation during high-intensity, ero- equation:
sive rainfall events, the effects of which may not be well captured
K e wDh
when using an empirical approach focusing on the calculation of tp ¼ ð4Þ
rðr K e Þ
total runoff amounts at the daily time scale.
Finally, it is not certain that the GAML model is capable of whereby it is assumed that the increase in volumetric moisture con-
describing and predicting infiltration on arable land for all possible tent at the time of ponding is equal to the volumetric moisture con-
conditions: the piston-type infiltration that the model assumes tent increase over the whole experiment.
may be a reasonable assumption when the soil is relatively dry Surface sealing due to rainfall impact may cause a decrease of
prior to a rainfall event, but may break down under more wet con- the hydraulic conductivity with time. To account for such variable
ditions. The identification of such conditions is essential to define Ke, Ahuja and Tsuji (1976) assumed that Ke would decrease expo-
the conditions under which the model may successfully be applied. nentially with time. However, crust development is dependent
Thus, applying the GAML model for runoff prediction is mainly on cumulative rainfall energy, rather than time, so that their equa-
hampered by difficulties in choosing the adequate number of tion may be reformulated as follows (Van Doren and Allmaras,
parameters to be calibrated and selecting appropriate parameter 1978):
values. At the same time it is not clear under which field conditions
the model may be successfully applied. The first objective of this K e ðtÞ ¼ K f þ ðK i K f Þ expðB KEcum Þ ð5Þ
paper is therefore to investigate to what extent it is possible to 1
where Kf is the final effective conductivity (mm h ) (infinite time),
use rainfall simulation data to derive meaningful parameteriza-
Ki is the initial effective conductivity (mm h1), KEcum = the cumula-
tions for the GAML model and to investigate which parameteriza-
tive rainfall energy since the beginning of the event (J m2), and B is
tion strategy yields optimal results in terms of simulation. In order
a rate constant (J1 m2).
to do this, we used a large dataset consisting of 350 rainfall simu-
lations carried out over a period of 6 years in Central Belgium on
arable land under various crops, crop stages and management 2.2. Data
techniques. The second objective is to investigate under which of
the tested conditions the GAML model can be successfully applied 2.2.1. Field measurements
and how model parameters depend on factors related to tillage In total, 350 rainfall simulations were carried out on small plots
technique used and soil conditions. (ca. 0.75 by 0.75 m) on 21 arable fields in the Belgian Loess Belt
from 2004 to 2010. The trial fields, with slopes ranging from 1%
to 21%, are listed in Tables 1a and 1b.
2. Materials and methods
Simulations were carried out in spring or early summer on
fields where maize, sugar beet or potato were grown and in late
2.1. The GAML model
winter on fields with winter wheat. Each trial field was split up
in at least two parts: one part was ploughed conventionally (CP,
The Green–Ampt infiltration model assumes a homogeneous
mouldboard plough to depth of 25–30 cm) and on the other part(s)
soil and a sharp wetting front, and under these assumptions it
non-inversion tillage was applied, which can be deep (DRT, 20–
can be derived from Buckingham-Darcy’s law for water flow
30 cm) or shallow (SRT, 5–15 cm) reduced tillage or direct drilling
through a soil. The movement of water is assumed to be the result
(DD). On both the non-inversion tilled and conventionally
of both capillary forces that suck the water front into the soil and
ploughed part(s) three locations were selected for a rainfall simu-
gravity. This results in so-called piston-type infiltration. The origi-
lation. Wheel tracks were avoided and care was taken that the
nal Green–Ampt equation, describing infiltration under ponded
three replicates were located on similar slope gradients. At each
conditions, was extended by Mein and Larson (1973) to a two-
location a small runoff plot of ca. 0.75 m 0.75 m was installed
stage model describing infiltration under constant rain: the
and four rain gauges were placed at the plot borders to measure
Green–Ampt Mein-Larson model (GAML). The original Green–
simulated rainfall intensity (I, mm h1). For all experiments, a noz-
Ampt model describes the infiltration rate by the following equa-
zle-type field rainfall simulator was used equipped with a Lechler
tion (Mein and Larson, 1973):
460.788 nozzle (for more details see Poesen et al., 1990) which was
wDh suspended at 3 m height. Under laboratory conditions this setup
f ¼ Ke 1 þ ð1Þ results in a rainfall intensity of ca. 45 mm h1 with a kinetic energy
F
of ca. 15 J m2 mm1, which is about 60% of that of natural
where f is the infiltration rate (mm h1), Ke is the effective hydraulic rainstorms of the same intensity (van Dijk et al., 2002). Due
conductivity (mm h1), F is the cumulative infiltration (mm), w is to variations in weather conditions and water temperature,
the suction at the wetting front (mm) and Dh is the difference be- measured rainfall intensity varied considerably and averaged
tween saturated and initial volumetric moisture content (m3 m3). 55 ± 29 mm h1. It may be assumed that the drop size distribution
A. Van den Putte et al. / Journal of Hydrology 476 (2013) 332–344 335
Table 1a
Main characteristics of the trial fields and number of rainfall simulations performed in (a) 2004, 2005 and 2006, (b).
Year Field Tillage techniquea Cropb Nc Sandd (%) Siltd (%) Clayd (%) Slope (%)
2004 F4 CP, DRT M 6 3.7 ± 0.6 81.5 ± 0.9 14.8 ± 0.5 8.8 ± 0.9
F14 SRT, DD SB 6 3.9 ± 0.3 81.2 ± 0.8 14.9 ± 0.7 4.0 ± 0.3
F1 CP, DRT, SRT M 9 27.0 ± 9.1 60.6 ± 6.8 12.4 ± 2.4 9.7 ± 1.1
F2 CP, SRT M 4 3.2 ± 0.6 82.0 ± 1.2 14.8 ± 1.4 17.5 ± 2.0
F7 CP, SRT M 6 61.3 ± 2.3 30.1 ± 3.0 8.6 ± 1.5 7.7 ± 0.8
F10 CP, SRT SB 5 46.1 ± 4.4 41.9 ± 4.3 12 ± 0.6 11 ± 0.4
F5 CP, SRT, DD SB 7 22.5 ± 5.1 65.4 ± 4.4 12.1 ± 0.9 8.6 ± 2.0
F16 CP, DRT SB 6 3.4 ± 0.9 80.3 ± 0.6 16.3 ± 1.2 4.5 ± 0.4
F15 CP, DRT SB 6 4.7 ± 1.3 78.5 ± 0.9 16.7 ± 0.7 11.3 ± 0.7
F6 CP, DRT P 4 41.0 ± 1.9 47.5 ± 3.4 11.6 ± 1.6 8.4 ± 0.1
2005 F18 CP, 2X DRT SB 9 3.9 ± 0.4 80.5 ± 1.3 15.6 ± 1.4 8.6 ± 0.5
F2 CP, DRT M 6 3.7 ± 0.6 81.2 ± 0.4 15.0 ± 1.0 17.9 ± 1.7
F7 CP, DRT SB 6 58.7 ± 1.0 33.7 ± 1.3 7.6 ± 1.2 7.0 ± 0.9
F17 CP, SRT, DD SB 7 5.2 ± 0.9 79.9 ± 0.6 15.0 ± 0.9 9.5 ± 1.6
F10 CP, SRT WW 6 49.1 ± 4.4 40.3 ± 3.2 10.6 ± 1.3 11.5 ± 1.3
F15 CP, DRT WW 6 5.1 ± 1.0 80.8 ± 1.2 14.1 ± 1.0 10.7 ± 0.5
F3 CP, SRT P 6 5.6 ± 0.6 76.6 ± 3.8 17.8 ± 3.9 5.0 ± 0.7
F5 CP, SRT WW 7 25.5 ± 3.6 62.4 ± 4.1 12.1 ± 1.4 8.1 ± 1.0
F8 CP, SRT WW 6 18.6 ± 1.6 71.6 ± 1.1 9.8 ± 1.2 5.4 ± 0.8
F1 CP, DRT, DD WW 11 26.0 ± 4.9 63.4 ± 3.9 10.6 ± 1.5 8.5 ± 0.9
2006 F20 CP, DRT SB 6 26.3 ± 3.1 61.3 ± 3.2 12.3 ± 1.4 8.9 ± 1.2
F21 DRT SB 6 4.1 ± 0.2 79.9 ± 0.7 16.1 ± 0.9 5.4 ± 0.7
F7 CP, DRT WW 6 58.7 ± 1.0 33.7 ± 1.3 7.6 ± 1.2 7.9 ± 0.8
F1 CP, DRT, SRT SB 12 25.1 ± 4.5 64.3 ± 2.8 10.6 ± 1.9 8.5 ± 0.9
a
CP: conventional mouldboard plough; DRT: deep reduced tillage; SRT: superficial reduced tillage; DD: direct drilling.
b
M: maize; P: potato; SB: sugar beet; WW: winter wheat.
c
N: number of conducted rainfall simulations.
d
Sand (2–0.063 mm), silt (0.002–0.063 mm), clay (<0.002 mm); average based on three topsoil samples for each tillage technique.
Table 1b
Main characteristics of the trial fields and number of performed rainfall simulations in 2007, 2008, 2009 and 2010.
Year Field Tillage techniquea Cropb Nc Sandd (%) Siltd (%) Clayd (%) Slope (%)
2007 F10 CP, DRT SB 6 34.3 ± 1.7 57.2 ± 1.9 8.5 ± 1.7 10.3 ± 0.9
F12 CP, DRT SB 4 25.3 ± 1.4 67.0 ± 0.5 7.7 ± 1.4 5.5 ± 0.6
F8 CP, DRT M 6 23.3 ± 1.2 67.4 ± 1.5 9.3 ± 1.3 6.7 ± 1.0
F1 CP, DRT, DD WW 20 38.2 ± 6.2 56.1 ± 5.7 5.7 ± 0.7 6.3 ± 1.5
2008 F10 CP, DRT M 6 35.4 ± 3.9 55.5 ± 2.8 9.1 ± 1.3 11.7 ± 1.0
F2 CP, DRT M 6 NA NA NA 14.6 ± 2.7
F6 CP, DRT SB 6 34.3 ± 3.0 56.6 ± 2.3 9.0 ± 0.8 8.4 ± 0.7
F7 CP, DRT SB 6 47.6 ± 2.2 45.0 ± 2.6 7.3 ± 0.8 9.0 ± 1.3
F8 CP, DRT M 4 NA NA NA 7.9 ± 1.5
IWT1 CP, 4X DRT SB 15 11.3 ± 3.5 78.6 ± 1.6 10.1 ± 2.2 3.1 ± 0.5
IWT2 CP, 4X DRT P 15 11.4 ± 1.5 79.2 ± 1.0 9.4 ± 1.0 2.7 ± 0.6
IWT3 CP, 2X DRT, SRT P 12 10.0 ± 1.6 79.9 ± 0.8 10.1 ± 0.9 3.4 ± 1.2
F1 CP, DRT, SRT WW 27 34.1 ± 5.1 59.6 ± 4.5 6.3 ± 1.4 7.3 ± 1.6
2009 F10 CP, DRT M 5 37.2 ± 2.1 54.2 ± 1.6 8.6 ± 1.0 10.3 ± 1.0
F12 CP, DRT WW 6 23.7 ± 1.0 66.4 ± 1.1 9.9 ± 0.4 5.4 ± 1.7
F17 CP, DRT WW 6 17.3 ± 0.5 74.0 ± 0.7 8.7 ± 0.4 6.6 ± 0.8
F6 CP, DRT M 6 39.3 ± 2.1 52.4 ± 2.4 8.3 ± 1.4 8.7 ± 0.8
IWT2 CP, DRT WW 5 10.9 ± 0.3 79.6 ± 0.8 9.5 ± 0.9 3.0 ± 1.2
IWT5 CP, 3X DRT P 12 24.5 ± 8.9 65.8 ± 8.4 9.7 ± 0.7 3.8 ± 1.7
F1 CP, DRT, SRT M 9 34.3 ± 4.5 59.7 ± 4.4 5.9 ± 0.6 7.1 ± 0.8
2010 F1 CP, DRT, SRT WW 9 17.8 ± 1.8 72.9 ± 1.4 9.3 ± 1.4 9.0 ± 1.2
a
CP: conventional mouldboard plough; DRT: deep reduced tillage; SRT: superficial reduced tillage; DD: direct drilling.
b
M: maize; P: potato; SB: sugar beet; WW: winter wheat.
c
N: number of conducted rainfall simulations.
d
Sand (2–0.063 mm), silt (0.002–0.063 mm), clay (<0.002 mm); average based on three topsoil samples for each tillage technique.
and hence the rainfall kinetic energy were not strongly affected by ment of the soil surface state and cover (Cv, %). Soil surface
these variations in intensity as the same nozzle and water pressure crusting (Cr, %) was assessed by determining the fraction of the to-
were used throughout all experiments. tal surface covered by either a depositional or a structural crust,
Before the start of each experiment, a soil sample was collected based on a series of reference pictures. Pictures for 10%, 30%, 70%
with a Kopecky ring (100 cm3, 0–0.05 m topsoil) next to every plot and 100% crusting are shown in Fig. 1.
for the determination of initial dry volume weight (qbi, g cm3), Next, simulated rainfall was applied until steady-state runoff
initial volumetric moisture content (hini, %), soil organic matter was reached: the duration of the experiments varied between 30
(OM, %) and texture. Before the start of each experiment an orthog- and 60 min. For all rainfall simulation experiments the runoff-time
onal top view picture of every plot was taken to make an assess- relationship was measured and consequently the cumulative
336 A. Van den Putte et al. / Journal of Hydrology 476 (2013) 332–344
Fig. 1. Pictures illustrating crusting percentages of 10% (a), 30% (b), 70% (c) and 100% (d).
infiltration (F, mm) was calculated and used to evaluate the appli- 2.2.2. Laboratory analysis
cability of the GAML model. Volumetric and gravimetric moisture content of the topsoil be-
After the experiment another Kopecky ring sample was taken to fore and after the experiment were determined by weighing the
determine the final moisture content. The crop planting date was Kopecky samples before and after oven-drying (24 h at 105 °C).
obtained from the farmers. Rainfall depth (Px, mm, with x being Grain size analysis was carried out using laser diffractometry
the number of days) over different periods preceding the rainfall (Coulter LS 13 320) after boiling a small subsample for 15 min
simulation and maximum daily rainfall amount (IMAX, mm day1) and maintaining dispersion during the analysis by ultrasonic stir-
in the period between planting and the experiment were obtained ring. For the samples collected between 2004 and 2008, the OM
from rain gauges placed at the field site or from a weather station content was determined with the Walkley and Black method
located nearby (<6 km) the field. All variables that were considered (Walkley and Black, 1934), while for those taken in 2009 and
are presented in Table 2. 2010 OM contents were measured with a vario MAX CN Macro
Table 2
Range of values of measured soil/surface and rainfall characteristics during the spring/summer and winter experiments.
Elemental Analyzer (Elementar Analysensysteme GmbH, Germany). which is valid for 0 P x P exp(1) and has a maximum relative
A linear regression was therefore conducted using 21 samples that error of 0.03%. In our analysis
were analysed using both methods (OCCN = 1.25 OCWB + 0.114,
R2 = 0.96, where OCCN = the organic carbon content as measured x ¼ ð1 þ F p Þ expð1 t F p Þ ð14Þ
with the CN analyzer and OCWB is the organic carbon content as
measured with the Walkley–Black method) so that the Walkley F p and t are needed to obtain x. To calculate F p and t four values
and Black data could be made comparable to those obtained from are required: Ke, w, tp and Dh. For the calibration datasets, the latter
total combustion. was always directly calculated as the difference between the mea-
sured volumetric soil water at the end and the beginning of the
2.3. Model calibration and validation rainfall simulation. Three different fitting procedures were used to
obtain optimised values for the model parameters Ke and w and
For some of the experiments that were performed a quality for tp for each rainfall event in the calibration datasets.
check revealed that not all necessary data were collected with suf- In the first procedure (fit1), both Ke and w were obtained by
ficient accuracy. Therefore the results of only 298 experiments model optimisation and tp was then obtained from Eq. (4). The
were used for model calibration and validation. As it was immedi- model optimisation was done by comparing the model fit for a
ately clear that the runoff response in spring and early summer (on wide range of different values of Ke and w in all possible combina-
summer crops) was different from that in late winter (on winter tions. Values for Ke ranged from zero to the average rainfall inten-
crops), these datasets were analysed separately. Both datasets sity of the rainfall simulation (mm h1), with steps of 0.25 mm h1.
were further split in a calibration and validation set by ranking Evaluated values for w ranged from zero to 500 mm, with steps of
all observations based on the value of the runoff coefficient. Subse- 2.5 mm.
quently every observation with an odd rank number was assigned In a second procedure (fit2), only Ke was optimised. The value of
to the calibration dataset and every observation with an even rank w was fixed at the average value that was obtained during fit1 for
number was assigned to the validation dataset. The final number of all experiments considered, i.e. 182 mm. Evaluated Ke values ran-
observations in each subdataset is given in Table 5. ged again from zero to the rainfall intensity of the rainfall simula-
tion (mm h1), with steps of 0.25 mm h1. The time to ponding was
2.3.1. Calibration procedures again derived from Eq. (4).
The GAML infiltration formula for predicting cumulative infil- In the first and second procedure, Ke was assumed to remain
tration to dry or uniformly wet soil (Eq. (3)) can be expressed in constant over the rainfall experiment. In a third procedure (fit3),
dimensionless form as (Barry et al., 2005): the effective hydraulic conductivity was assumed to vary with
" # cumulative rainfall kinetic energy as described by Eq. (5): this re-
F þ 1
F ¼t þ
F p þ ln for t P 0 ð6Þ quired four parameters (w, Ki, Kf and B) to be estimated. The time
Fp þ 1 to ponding (tp) was assumed to be equal to the time to runoff
where F is the dimensionless cumulative infiltration, t is the which was obtained by extrapolating the measured runoff–time
dimensionless time and F p is the dimensionless cumulative infiltra- curve to zero runoff. Ki values were evaluated in a range from zero
tion at time of ponding, using the following substitutions: to the rainfall intensity of the rainfall simulation (mm h1), with
steps of 0.25 mm h1 while Kf was allowed to vary between zero
2K e F to Ki, again in steps of 0.25 mm h1. The rate constant (B) was
F ¼ ð7Þ
S2 tested in a range from 0.006 to 0.03 m2 J1, with steps of
0.0012 m2 J1. Evaluated values for w ranged from zero to
2K e F p 500 mm, with steps of 10 mm.
F p ¼ ð8Þ
S2 The parameter value set having the lowest sum of squared er-
rors for the differences between the observed and the computed
2K 2e values of the cumulative infiltration curve was retained as the
t ¼ ðt t p Þ for t P t p ð9Þ
S2 ‘best’ parameter set. The fitting procedures were performed using
Matlab (Mathworks, 2009).
where S is the sorptivity for a Green–Ampt soil, defined as
S2 ¼ 2K e wDh ð10Þ
2.3.2. Validation of the GAML model
The LambertW function is the function W(x) such that We examined if the GAML model parameters obtained with the
WðxÞ exp½WðxÞ ¼ x ð11Þ three procedures can be related to measured system properties. To
this end, the GAML parameter values obtained for the calibration
A solution of the dimensionless form of the GAML model in terms of datasets were related to the measured soil/surface and rainfall
the LambertW function (Barry et al., 2005) is characteristics using multiple regression. Next, parameter values
F ¼ 1 W 1 ½ð1 þ F p Þ expð1 t F p Þ for F P F p for the validation dataset were predicted using these regression
equations, and these parameter values were used to predict the
P 0; t P 0 ð12Þ infiltration for comparison with the measured infiltration. Apart
in which W1 is the lower branch of the LambertW function. from the GAML parameter values, the increase of volumetric mois-
Approximations of the LambertW function can thus be used to find ture content during the experiment also needed to be predicted:
F. To evaluate the GAML model and to fit the F versus t curve the this was estimated from the measured initial soil moisture content.
following approximation of the LambertW function, discussed by As the distribution of model parameter values was strongly
Parlange et al. (2002), was used: skewed, Ke and Dh values were transformed to their square root
while w values were transformed to their base-10 logarithm before
8 9 regression analysis was applied. In case of the 3rd procedure (fit3)
< ½2 2 lnðxÞ2 =
1
Kf, Ki, w and B were estimated by regression. Ki and Kf values were
W 1 ðxÞ lnðxÞ ln lnðxÞ þ 1
ð13Þ not transformed while w and B values were transformed to their
: 1þ½22 lnðxÞ2 ;
6 base-10 logarithm before regression analysis was applied. The
338 A. Van den Putte et al. / Journal of Hydrology 476 (2013) 332–344
Statistics of parameter values and model efficiency for the different GAML parameterization procedures (see text for symbols). The average value (AVG), standard deviation (STD), median (Q50), minimum (MIN) and maximum (MAX)
soil/surface and rainfall characteristics used are described in
0.23
0.43
0.73
0.64
52.75
1.63
1.88
4.19
1.00
1.00
Table 2.
68.8
63.3
67.0
25.0
1.0
MAX
500
500
For each experiment in the calibration and validation datasets
the cumulative infiltration curve was then calculated using the
parameter values predicted by regression from the corresponding
calibration set. Similar to the approach used by Risse et al.
(1994), model performance was then quantified by calculating root
0.0073
mean square error (RMSE) and model efficiency (ME) by comparing
0.048
0.001
3.75
0.35
0.22
0.81
0.01
0.01
0.02
0.04
MIN
2.5
5.0
10.0
measured and estimated infiltration at each measured data point.
0
For our study, these measures give an indication of the goodness
of fit and can be compared to evaluate the relative performance
of different models but they may not be used to derive significance
levels as measurements of cumulative infiltration are obviously
0.999
0.996
autocorrelated. An average value of ME and RMSE was then calcu-
0.18
0.15
0.41
0.38
0.06
0.10
0.10
28.5
87.5
21.5
30.5
45.0
5.0
10.0
1.0
Q50
lated per dataset for every GAML model type.
We also simulated the total runoff coefficient as well as the
runoff coefficient after 10, 20 and 30 min. When a negative runoff
coefficient was obtained, it was set to zero. The predicted total
runoff coefficient for each experiment was then compared to the
0.35
0.13
0.11
0.38
0.11
0.64
0.06
0.09
0.09
0.03
observed total runoff coefficient using RMSE and ME statistics.
6.2
17.9
205.8
13.5
14.9
147.0
13.0
STD
Simulated runoff coefficients at intermediate times could not
directly be compared to observations given that measurement
Winter (n = 101)
times often did not correspond exactly to 10, 20 or 30 min. There-
fore, values for the RC at these moments as interpolated from the
cumulative infiltration curve were used.
0.12
0.98
0.32
0.17
0.97
0.54
0.13
0.99
0.59
0.09
28.9
23.9
200.5
8.4
80.3
28.0
42.0
AVG
2.4. Multiple regression
value of the optimal parameter values as well as of the model fit statistics (ME and RMSE) are given for spring/summer and winter data.
0.59
0.79
0.47
1.76
2.02
5.03
1.00
1.00
1.00
53.3
31.3
62.5
50.5
25.0
MAX
500
accurately as possible given the information collected with respect
to soil/surface and rainfall characteristics. A regression model may
therefore outperform a dynamic model such as GAML both in the
calibration and validation stage, as there are no a priori constraints
on how the available information can be assimilated in the model.
0.014
0.043
3.75
0.75
0.56
0.01
0.01
0.05
0.02
0.01
0.01
0.05
2.5
1.5
5.0
10.0
MIN
0
GAML provides an indication as to how well the information mea-
sured in the field is transferred towards the GAML model through
the parameter estimation techniques used here.
The performance of the regression model was evaluated for the
validation datasets by comparing its predictions of the total runoff
0.999
0.997
0.18
0.13
0.16
0.19
0.36
0.14
0.99
0.44
10.3
40.5
18.5
62.5
7.0
16.0
Q50
3. Results
0.13
0.27
0.18
0.35
0.11
0.73
0.07
0.08
0.09
0.04
199.1
13.4
11.6
204.4
5.4
6.7
13.0
The fit1 procedure resulted in a GAML model that fitted the ob-
Summer (n = 197)
0.17
0.98
0.26
0.24
0.98
0.46
0.16
0.98
0.69
172.7
181.4
19.2
38.6
10.9
10.3
18.0
the fact that only one degree of freedom was allowed (Table 3).
Ke (mm h1)
Ke (mm h1)
Kf (mm h1)
Ki (mm h1)
The RMSE never exceeded 2.02 mm and the mean RMSE was
3
w (mm)
w (mm)
B (h1)
ME (F)
ME (F)
ME (F)
tp (h)
tp (h)
Table 3
fit1
fit2
fit3
Again, minimum ME was very low, especially for spring/summer under which the datasets were collected: In winter, more than
data (0.01). For winter data, minimum ME was 0.22. 45% of the plots had a crust cover exceeding 90% whereas in
The fit3 procedure, using a variable effective hydraulic conduc- spring/summer, this was only the case for 16% of the observations.
tivity, resulted in equal fits for the spring/summer data and better In spring/summer, 80% of the observations had a vegetation cover
fits for the winter data (Table 3). The RMSE never exceeded below 20%, whereas in winter, the vegetation cover was higher
5.03 mm and the mean RMSE was 0.69 mm for spring/summer than 40% for 72% of the observations (Fig. 3).
experiments and 0.59 mm for winter experiments. The mean ME For fit3, Kf could not be related to any measured variable for
was 0.98 for all spring/summer experiments and 0.99 for all winter spring/summer data, and could be related poorly to P4 for winter
experiments. For spring/summer data, the minimum ME was 0.56 data (r2 = 0.20) (Table 4). The variation in Kf predicted for the
and for winter data the minimum ME was 0.81, so all experiments calibration dataset (17 mm h1 for spring/summer data and
were modelled well using this fitting procedure. 24.3–41.9 mm h1 for winter data) was again much lower than
the observed variations in final infiltration rates (0–70 mm h1).
Log10(w) was reasonably well related to rainfall intensity for the
3.2. Relating GAML model parameters to measured data
spring/summer data (r2 = 0.43) and for the winter data (r2 = 0.32).
Rate constant B was well related to four different variables for
The Ke values obtained from the fit1 procedure were poorly re-
the spring/summer data (r2 = 0.52), but could not be related to
lated to crusting stage for the spring/summer data (R2 = 0.19) and
any measured variable for winter data. Ki was poorly related to
to P4 for the winter data (R2 = 0.18) (Table 4). Also, the resulting
some of the measured variables for the spring/summer data
predicted variation in Ke for the calibration data (10–32 mm h1
(r2 = 0.18) and for the winter data (r2 = 0.22).
for spring/summer data and 20–46 mm h1 for winter data) was
much lower than the observed variations (0–70 mm h1) in final
infiltration rates. Log10(w) was only very poorly related to soil 3.3. Multiple regression model
crusting for spring/summer data (R2 = 0.08) and could not be re-
lated to any measured variable for winter data. For the spring/summer calibration dataset, the variation in the
When only Ke was determined by optimisation (fit2), a much total runoff coefficient could be described by the following multi-
better correlation relation with measured variables was found (R2 ple regression equation:
of 0.49 for spring/summer data and 0.34 for winter data) (Fig. 2).
Controls on Ke appeared to be different in winter compared to sqrtðRCÞ ¼ 0:39 þ 0:004C r 0:02Cl 0:0063C v þ 0:0041I
spring/summer. In spring/summer, Ke was mainly determined by ðn ¼ 66; r 2 ¼ 0:59; ME ¼ 0:59; RMSE ¼ 0:12Þ ð15Þ
soil crusting, soil cover and soil clay content. For winter experi-
ments, Ke was positively related to initial soil moisture content A dummy variable for non-inversion tillage was also included in the
and the amount of rain during the 2 days preceding the simulation. regression analysis, but it did not appear to be a significant factor.
Part of this difference may be related to the different conditions For the winter calibration dataset (n = 42), no acceptable regression
Table 4
Regression equations relating GAML parameter values to measured data for the different fitting procedures. For an explanation of the variables, see Table 2. CALs is the calibration
dataset for spring/summer data and CALw is the calibration dataset for winter data.
Table 5
Average ME and RMSE for the prediction of cumulative infiltration with the GAML model using parameter values estimated from measured data through regression (see text for
symbols). Values are given for the different fitting procedures and for spring/summer and winter datasets (CALs and VALs are calibration and validation datasets for spring/
summer data and CALw and VALw are calibration and validation datasets for winter data).
Dataset N GAML model (fit1) GAML model (fit2) GAML model (fit3)
Fig. 2. Simulated versus observed values for Ke (fit2 procedure). (Ke is the effective hydraulic conductivity; CALs and VALs are calibration and validation datasets for spring/
summer data and CALw and VALw are calibration and validation datasets for winter data).
Fig. 3. Histograms of crust (Cr) and residue and crop (Cv) cover percentages for spring/summer and winter field plots.
model could be found to describe the variation in runoff one-parameter fit2 procedure resulted in a clearly higher ME as
coefficients. compared to the fit1 and fit3 procedures. However, the highest
ME (0.44) was obtained for the regression model (Table 6). Predic-
3.4. Model validation tions obtained using the GAML model fit2 are shown in Fig. 4 and
predictions obtained using the MR model are shown in Fig. 5.
3.4.1. Cumulative infiltration curves For the spring/summer data, the simulation of the runoff coeffi-
It appeared that the GAML model with parameters estimated by cients improved with rainfall duration (Table 7). Model efficiency
regression could not be meaningfully used to predict cumulative after 10 min of rainfall was still very low and meaningful simula-
infiltration during the winter period: the GAML model performed tions for the validation dataset were only possible for rainfall dura-
very poorly, regardless of the fitting procedure used (Table 5). tions of 20 and 30 min. Again, the results from the one-parameter
For spring/summer experiments, model performance for the cali- model calibration (fit2) were best. The results from the four-
bration dataset improved when one parameter was used (fit2) in parameter model calibration (fit3) were poor for all rainfall
comparison to a two-parameter and four-parameter model (fit1 durations. For the winter data, model performance was again very
and fit3). The one-parameter model (fit2) also clearly performed poor, regardless of the calibration procedure used.
best for the validation data (Table 5).
4. Discussion
3.4.2. Runoff coefficients
As for the cumulative infiltration curves, the parameters for The first part of our analysis shows that, in most cases, the
none of the GAML model calibration procedures could be success- GAML model is indeed capable of describing correctly the time-
fully estimated by regression to predict runoff coefficients for dependent evolution of cumulative infiltration using a single
winter data (Table 6). For spring/summer validation data, the parameter (Ke) to optimise the model fit. This is surprising as some
A. Van den Putte et al. / Journal of Hydrology 476 (2013) 332–344 341
Fig. 4. Simulated versus observed total runoff coefficient. Simulated values obtained using the GAML model with parameter estimates obtained from the fit2 procedure. (CALs
and VALs are calibration and validation datasets for spring/summer crops and CALw and VALw are calibration and validation datasets for winter crops). Full lines are
regression lines for calibration datasets, while dotted lines are for validation datasets.
Table 6
ME and RMSE for the total runoff coefficient predicted using the GAML model and different parameterization methods (see text for symbols). VALs is the validation dataset for
spring/summer data and VALw is the validation dataset for winter data.
Dataset N GAML model (fit1) GAML model (fit2) GAML model (fit3) MR model
ME RMSE ME RMSE ME RMSE ME RMSE
VALs 65 0.20 0.17 0.32 0.15 0.38a 0.26a 0.44 0.14
VALw 41 0.37 0.24 0.14 0.22 0.27b 0.35b 0 0.20
a
N = N 1, F-curve could not be fitted for 1 observation.
b
N = N 1, F-curve could not be fitted for 2 observations.
342 A. Van den Putte et al. / Journal of Hydrology 476 (2013) 332–344
outweighs the relatively minor loss in predictive capacity. While also to all the farmers for the maintenance of the split fields and
an empirical regression model predicting the runoff coefficient will the practical assistance.
only be applicable for experiments of approximately the same
intensity and duration, a calibrated dynamic model may be used References
as a basis for predicting runoff amounts produced for events with
varying duration and/or varying intensity, provided that the effect Ahuja, L.R., Tsuji, G.Y., 1976. Use of Green–Ampt equation with variable
conductivity. Soil Sci. Soc. Am. J. 40 (4), 619–622.
of varying rainfall intensities and water depths on effective infiltra- Auerswald, K., Haider, J., 1996. Runoff curve numbers for small grain under German
tion are effectively accounted for (Langhans et al., 2011). However, cropping conditions. J. Environ. Manage. 47 (3), 223–228.
one should be aware that local validation will remain necessary. Barry, D.A., Parlange, J.Y., Li, L., Jeng, D.S., Crapper, M., 2005. Green–Ampt
approximations. Adv. Water Resour. 28 (10), 1003–1009.
Earlier work (Risse et al., 1994) already showed that pedotransfer
Beven, K., 1989. Changing ideas in hydrology – the case of physically-based models.
functions may only be used locally and for the scale for which they J. Hydrol. 105 (1–2), 157–172.
were developed. Bouraoui, F., Dillaha, T.A., 1996. Answers-2000: runoff and sediment transport
model. J. Environ. Eng. – ASCE 122 (6), 493–502.
Brakensiek, D.L., Onstad, C.A., 1977. Parameter-estimation of Green and Ampt
infiltration equation. Water Resour. Res. 13 (6), 1009–1012.
5. Conclusions Brooks, R.H., Corey, A.T., 1964. Hydraulic Properties of Porous Media. Hydrology
Paper. Colorado State University, Fort Collins, Colorado.
Predicting runoff generation on arable land is inherently diffi- Cerdan, O., Le Bissonnais, Y., Couturier, A., Saby, N., 2002. Modelling interrill erosion
in small cultivated catchments. Hydrol. Process. 16 (16), 3215–3226.
cult due to the rapidity with which soil and crop conditions
Childs, E.C., Bybordi, M., 1969. Vertical movement of water in stratified porous
change. While a relatively simple, dynamic model such as the material. I: Infiltration. Water Resour. Res. 5 (2), 446–459.
GAML model allows to describe very well variations in cumulative Chu, S.T., 1978. Infiltration during an unsteady rain. Water Resour. Res. 14 (3), 461–
infiltration (and hence runoff generation) under simulated rainfall, 466.
Corradini, C., Melone, F., Smith, R.E., 1997. A unified model for infiltration and
prediction is not always successful. redistribution during complex rainfall patterns. J. Hydrol. 192 (1–4), 104–124.
The failure of the GAML model to predict the variation in runoff Corradini, C., Govindaraju, R.S., Morbidelli, R., 2002. Simplified modelling of areal
generation on winter wheat fields is mainly due to the fact that the average infiltration at the hillslope scale. Hydrol. Process. 16 (9), 1757–1770.
Corradini, C., Flammini, A., Morbidelli, R., Govindaraju, R.S., 2011. A conceptual
variability in runoff generation under winter conditions is not only model for infiltration in two-layered soils with a more permeable upper layer:
controlled by soil surface characteristics, but also by subsoil prop- from local to field scale. J. Hydrol. 410 (1–2), 62–72.
erties. The latter are almost impossible to measure on a routine ba- De Roo, A.P.J., Jetten, V.G., 1999. Calibrating and validating the LISEM model for two
data sets from the Netherlands and South Africa. Catena 37 (3–4), 477–493.
sis, which may pose important limitations on our capability to Esteves, M., Faucher, X., Galle, S., Vauclin, M., 2000. Overland flow and infiltration
predict runoff generation correctly under these conditions. modelling for small plots during unsteady rain: numerical results versus
Model performance was much better for spring/summer condi- observed values. J. Hydrol. 228 (3–4), 265–282.
Fox, D.M., Le Bissonnais, Y., Bruand, A., 1998. The effect of ponding depth on
tions. The main factors controlling infiltration were soil surface infiltration in a crusted surface depression. Catena 32 (2), 87–100.
crusting state, soil vegetation cover and soil clay content: these Freebairn, D.M., Gupta, S.C., 1990. Microrelief, rainfall and cover effects on
findings are consistent with those of other studies although rela- infiltration. Soil Tillage Res. 16 (3), 307–327.
Freebairn, D.M., Gupta, S.C., Onstad, C.A., Rawls, W.J., 1989. Antecedent rainfall and
tively large discrepancies may exist with respect to the absolute
tillage effects upon infiltration. Soil Sci. Soc. Am. J. 53 (4), 1183–1189.
values that are estimated. Tillage technique as such did not have Gilley, J.E., Finkner, S.C., Varvel, G.E., 1986. Runoff and erosion as affected by
a significant effect: thus, the impact of conservation tillage is ade- sorghum and soybean residue. Trans. ASAE 29 (6), 1605–1610.
quately captured by assessing its impacts on soil surface character- Govindaraju, R.S., Morbidelli, R., Corradini, C., 2001. Areal infiltration modeling over
soils with spatially correlated hydraulic conductivities. J. Hydrol. Eng. 6 (2),
istics (cover and crusting stage). 150–158.
Despite the fact that the infiltration characteristics of a tilled Green, W.H., Ampt, G., 1911. Studies in soil physics. I: The flow of air and water
soil surface clearly evolve throughout a high-intensity rainfall through soils. J. Agric. Sci. 4, 1–24.
Horton, R.E., 1933. The role of infiltration in the hydrological cycle. Trans. Am.
event, the GAML model performed best when only a single param- Geophys. Union 14, 446–460.
eter was allowed to vary and temporal variation is not explicitly ta- Kale, R., Sahoo, B., 2011. Green–Ampt infiltration models for varied field conditions:
ken into account. When more than one parameter needs a revisit. Water Resour. Manage 25 (14), 3505–3536.
Knisel, W., 1980. CREAMS – A Field-Scale Model for Chemicals, Runoff and Erosion
calibration, uncertainties in parameter prediction overwhelm pos- from Agricultural Management Systems. USDA Conservation Research Report
sible gains in predictive capability. Thus, under the conditions No. 26, 640 pp.
tested, simpler is indeed better: the use of a single parameter ver- Laflen, J.M., Elliot, W.J., Flanagan, D.C., Meyer, C.R., Nearing, M.A., 1997. WEPP-
predicting water erosion using a process-based model. J. Soil Water Conserv. 52
sion of the model yields to better predictions. It is also feasible as (2), 96–102.
the suction at the wetting front can indeed be estimated from Langhans, C., Govers, G., Diels, J., Clymans, W., Van den Putte, A., 2010. Dependence
existing pedotransfer functions. of effective hydraulic conductivity on rainfall intensity: loamy agricultural soils.
Hydrol. Process. 24 (16), 2257–2268.
As expected, the use of the GAML model resulted in a loss of
Langhans, C. et al., 2011. Experimental rainfall–runoff data: reconsidering the
predictive capability when compared to an empirical, regression- concept of infiltration capacity. J. Hydrol. 399 (3–4), 255–262.
based model. However, the loss of predictive capability is relatively Le Bissonnais, Y., Singer, M.J., 1993. Seal formation, runoff, and interrill erosion from
limited and is outweighed by the gains in flexibility and applicabil- 17 California soils. Soil Sci. Soc. Am. J. 57 (1), 224–229.
Le Bissonnais, Y. et al., 2005. Variability of soil surface characteristics influencing
ity that are obtained by using a conceptual model. runoff and interrill erosion. Catena 62 (2–3), 111–124.
Leys, A., Govers, G., Gillijns, K., Poesen, J., 2007. Conservation tillage on loamy soils:
explaining the variability in interrill runoff and erosion reduction. Eur. J. Soil Sci.
Acknowledgements 58 (6), 1425–1436.
Martin, P., Papy, F., Capillon, A., 2001. Agricultural field state and runoff risk:
proposal of a simple relation for the silty-loam soil context of the Pays de Caux
This study was supported by the IWT-project ‘From demonstra- (France). In: Stott, D.E., Mohtar, R.H., Steinhardt, G.C. (Eds.), Sustaining the
tion to optimalisation: development of a decision-support instru- Global Farm. Selected papers from the 10th International Soil Conservation
ment for the optimalisation of reduced tillage techniques starting Organization Meeting held May 24–29, 1999 at Purdue University and the
USDA-ARS National Soil Erosion Research Laboratory, pp. 293–299.
from a multi-disciplinary approach’ and by the EU-LIFE project SO-
Mathworks, 2009. Matlab. Mathworks, Natick, MA, USA.
WAP (www.sowap.org) as well as Syngenta Inc. The authors wish Mein, R.G., Larson, C.L., 1973. Modeling infiltration during a steady rain. Water
to thank K. Auerswald and M. Nearing for their valuable comments Resour. Res. 9 (2), 384–394.
and Jan Valckx, Katleen Gillijns, Greet Dielis, Lore Fondu, Sabrina Migliaccio, K.W., Srivastava, P., 2007. Hydrologic components of watershed-scale
models. Trans. ASABE 50 (5), 1695–1703.
Vandevelde, Jos Meersmans, Mustafa Ayyad and many other col- Nearing, M.A., Deerascough, L., Laflen, J.M., 1990. Sensitivity analysis of the WEPP
leagues for their help with the rainfall simulations. Many thanks hillslope profile erosion model. Trans. ASAE 33 (3), 839–849.
344 A. Van den Putte et al. / Journal of Hydrology 476 (2013) 332–344
Nearing, M.A., Liu, B.Y., Risse, L.M., Zhang, X., 1996. Curve numbers and Green– Schroeder, R., Auerswald, K., 2000. Modelling the seasonal variation of a seal-
Ampt effective hydraulic conductivities. Water Resour. Bull. 32 (1), 125– induced runoff generation in small agricultural catchments. Zeitschrift fuer
136. Kulturtechnik und Landentwicklung 41 (4), 167–172.
Parlange, J.-Y., Barry, D.A., Haverkamp, R., 2002. Explicit infiltration equations and Takken, I. et al., 1999. Spatial evaluation of a physically-based distributed erosion
the Lambert W-function. Adv. Water Resour. 25, 1119–1124. model (LISEM). Catena 37 (3–4), 431–447.
Poesen, J., Ingelmo-Sanchez, F., Mücher, H., 1990. The hydrological response of soil Tuppad, P., Douglas-Mankin, K.R., Lee, T., Srinivasan, R., Arnold, J.G., 2011. Soil and
surface to rainfall as affected by cover and position of rock fragments in the top water assessment tool (SWAT) hydrologic/water quality model: extended
layer. Earth Surf. Process. Landforms 15, 653–671. capability and wider adoption. Trans. ASABE 54 (5), 1677–1684.
Ponce, V.M., Hawkins, R.H., 1996. Runoff curve number: has it reached maturity? J. van Dijk, A., Bruijnzeel, L.A., Rosewell, C.J., 2002. Rainfall intensity-kinetic energy
Hydrol. Eng. 1 (1), 11–19. relationships: a critical literature appraisal. J. Hydrol. 261 (1–4), 1–23.
Rawls, W.J., Brakensiek, D.L., Miller, N., 1983. Green–Ampt infiltration parameters Van Doren Jr., D.M., Allmaras, R., 1978. Effect of residue management practices
from soils data. J. Hydraulic Eng.-ASCE 109, 62–70. on the soil physical environment, microclimate and plant growth, Crop
Rawls, W.J., Brakensiek, D.L., Simanton, J.R., Kohl, K.D., 1990. Development of a crust Residue Management Systems. Spec. Pub. 31. ASA, Madison, Wis., pp. 49–83.
factor for a Green Ampt model. Trans. ASAE 33 (4), 1224–1228. Van Rompaey, A.J.J., Govers, G., 2002. Data quality and model complexity for regional
Risse, L.M., Nearing, M.A., Savabi, M.R., 1994. Determining the Green–Ampt effective scale soil erosion prediction. Int. J. Geogr. Inform. Sci. 16 (7), 663–680.
hydraulic conductivity from rainfall–runoff data for the WEPP model. Trans. Walkley, A., Black, I.A., 1934. An examination of the Degtjareff method for
ASAE 37 (2), 411–418. determining soil organic matter and a proposed modification of the chromic
Risse, L.M., Liu, B.Y., Nearing, M.A., 1995a. Using curve numbers to determine base- acid titration method. Soil Sci. 27, 29–38.
line values of Green–Ampt effective hydraulic conductivities. Water Resour. Zhang, X.C., Nearing, M.A., Risse, L.M., 1995a. Estimation of Green–Ampt
Bull. 31 (1), 147–158. conductivity parameters. I: Row crops. Trans. ASAE 38 (4), 1069–1077.
Risse, L.M., Nearing, M.A., Zhang, X.C., 1995b. Variability in Green–Ampt effective Zhang, X.C., Nearing, M.A., Risse, L.M., 1995b. Estimation of Green–Ampt
hydraulic conductivity under fallow conditions. J. Hydrol. 169 (1–4), 1–24. conductivity parameters. II: Perennial crops. Trans. ASAE 38 (4), 1079–1087.