Anda di halaman 1dari 15

Geomorphology 78 (2006) 250 – 264

www.elsevier.com/locate/geomorph

Reconstructing ancient topography through erosion modelling


Iris Peeters a,⁎, Tom Rommens a , Gert Verstraeten a , Gerard Govers a ,
Anton Van Rompaey a , Jean Poesen a , Kristof Van Oost b
a
Katholieke Universiteit Leuven, Physical and Regional Geography Research Group, Geo-Institute, Celestijnenlaan 200E,
B-3001 Heverlee, Belgium
b
University of Exeter, Hydrology and Earth Surface Processes Research Group, Amory Building Rennes Drive, Exeter EX4 4RJ, United Kingdom
Received 28 July 2005; received in revised form 23 January 2006; accepted 24 January 2006
Available online 10 March 2006

Abstract

One of the main aims of geomorphology is to understand how geomorphic processes change topography over long time scales.
Over the last decades several landscape evolution models have been developed in order to study this question. However, evaluation
of such models has often been very limited due to the lack of necessary field data. In this study we present a topography based
hillslope erosion and deposition model that is based on the WATEM/SEDEM model structure and works on a millennial time scale.
Soil erosion, transport and deposition are calculated using slope and unit contributing area. The topography is iteratively
rejuvenated by taking into account modelled erosion and deposition rates, thereby simulating topographic development backwards
in time. A first attempt has been made to spatially evaluate the model, using detailed estimates for historical soil erosion and
sediment deposition volumes, obtained from an augering campaign in a small catchment in the Belgian Loess Belt. The results
show that the model can simulate realistic soil redistribution patterns. However, further research is necessary in order to deal with
artificial flaws that cause routing problems and significantly influence results. Common problems and issues related to this type of
backward modelling are also discussed.
© 2006 Elsevier B.V. All rights reserved.

Keywords: Erosion modelling; Landscape evolution; Model calibration; Late Holocene; DEM

1. Introduction and tectonics over long time scales (e.g. Willgoose et al.,
1991; Tucker and Slingerland, 1994; Braun and Sam-
Several landform evolution models, focusing on bridge, 1997). Some models have been used to
different temporal and spatial scales, already exist (e.g. investigate the relative roles of land cover and climate
Ahnert, 1976; Willgoose et al., 1991; Howard, 1994; on the Holocene evolution of river catchments (e.g.
Tucker and Slingerland, 1994; Moglen and Bras, 1995; Coulthard et al., 1997).
Coulthard et al., 1997; Tucker and Slingerland, 1997; Many authors stressed the importance and need of a
Braun and Sambridge, 1997; Tucker et al., 1999). A critical and accurate model evaluation based on actual
review of landscape evolution models is given by Coult- erosion data (e.g. Addiscott et al., 1995; Coulthard,
hard (2001). These models have often been applied to 2001; Van Rompaey et al., 2001, 2002; Hancock, 2004).
examine relationships of catchment form with hydrology So far, only few attempts have been made to test land-
scape evolution models against observed spatial varia-
⁎ Corresponding author. Tel.: +32 16 326406; fax: +32 16 322980. tion in soil erosion rates, because of the limited amount
E-mail address: Iris.Peeters@geo.kuleuven.be (I. Peeters). of data available on spatial patterns of erosion and
0169-555X/$ - see front matter © 2006 Elsevier B.V. All rights reserved.
doi:10.1016/j.geomorph.2006.01.033
I. Peeters et al. / Geomorphology 78 (2006) 250–264 251

deposition (Govers et al., 1993; Desmet and Govers, rill erosion processes, based on slope gradient S and
1995; Takken et al., 1999; Jetten et al., 2003). This is upslope area A:
especially the case for long-term erosion models: very
often the results of such models can only be compared EPT ¼ C1 ðaS b þ cS d Ae Þ ð1Þ
with sporadic point data on sediment deposition rates where EPT is the potential mean annual water erosion rate
(e.g. Coulthard et al., 1997). This lack of spatial eval- (kg m− 2); S is the slope gradient (m m− 1); A is the
uation is problematic, as models may produce a rea- contributing area or upslope drainage area (m2); and C1,
sonable estimation of sediment output from a basin even a (kg m− 2), b, c (kg m− 4), d, and e are parameters. The
if internal dynamics of the catchment are entirely mis- first term in the parentheses of the right hand side, aSb,
represented (e.g. Takken et al., 1999; Van Oost et al., represents interrill erosion, while the second term, cSdAe,
2005). For example, the spatial variation of erosion and stands for rill erosion. Some important factors such as
deposition simulated by such “good” models may not rainfall erosivity and soil erodibility are expressed
correspond to field observations. through the parameters C1, a and c, which affect the
In this study we investigate to what extent the pattern calculated erosion intensity.
of historical soil erosion and sediment deposition can be Van Oost (2003) found that two-dimensional patterns
simulated using a simple but spatially explicit topography of erosion and deposition were best described if the area
driven erosion and deposition model. The simplicity of exponent (e) is assumed to be a power function of the
the model guarantees a good understanding of the model contributing area (A), which confirms the findings of
responses, and through validation we will make sure that Desmet et al. (1999):
the model can simulate the observed data adequately. The
objectives of this paper are: e ¼ 0:3 þ ðA=Aref Þe1 ð2Þ

(i) to present a simple topographic-based hillslope where Aref is a reference contributing area and e1 is a
model that works backward in time from one to parameter. Van Oost (2003) determined an optimal value
5000 years; of 3000m2 for Aref. The contributing area can be used as
(ii) to investigate the possibilities of the model in a proxy for runoff discharge. The area is calculated by
reconstructing ancient topography and simulating routing the area of each grid cell from the highest to the
long-term erosion and sedimentation patterns, by lowest cell, using the (multiple flow) flux decomposition
comparing the model results with field data ob- algorithm (Desmet and Govers, 1996a) by which flow is
tained in a small agricultural catchment (103ha) in routed to two cardinal downslope neighbours defined
the Belgian Loess Belt (Rommens et al., 2005); from the aspect direction. When there is only one direct
(iii) to discuss the problems with topographic recon- neighbouring cell with a lower height than the con-
struction using the erosion model and the limita- tributing cell, a single flow algorithm is used and all the
tions of the proposed type of numerical modelling. sediments are routed towards this lower cell.
Once the volume of eroded soil for each cell is
2. Erosion model calculated, the sediments are routed through the catch-
ment, also with the flux decomposition algorithm. Sedi-
2.1. Process description ment deposition is modelled by comparing sediment
fluxes with local transport capacity Tc (kg m− 1), which is
The model presented in this paper is based on the a function of the total erosion potential:
spatially distributed model WATEM/SEDEM, which is a Tc ¼ kTC EPT ð3Þ
steady-state model that calculates tillage and water
erosion rates on an annual basis (Van Oost et al., 2000a, where kTC is the transport capacity coefficient (m), which
Van Rompaey et al., 2001). Some adjustments have been corresponds to the distance where a point along a rill on a
made to this model in order to simulate soil redistribution uniform slope has a transport capacity corresponding to
over a long time scale. The model takes into account the the sediment flux. Ephemeral gully erosion is not expli-
two main processes controlling soil redistribution in the citly taken into account in the model although the process
study area, i.e. water erosion and tillage erosion. is approximated using a multiple flow routing algorithm
which simulates flow convergence (Desmet and Govers,
2.1.1. Water erosion 1996a). This is important since gully erosion is res-
The water erosion potential at a given location is ponsible for 40–50% of the sediment production in
calculated using a simple representation of interrill and typical Belgian catchments (Poesen et al., 1996).
252 I. Peeters et al. / Geomorphology 78 (2006) 250–264

2.1.2. Tillage erosion topography exist. On the other hand, backward calcula-
Tillage causes the movement of large volumes of soil tion is fraught with some specific difficulties that will be
on arable land. While water erosion rates are largest on discussed later. In this study, a resolution of 20m was
the steepest slope sections, tillage erosion shows high applied for input and output maps and a bulk density of
values on convex sections (Van Oost et al., 2000b). 1350 kg m− 3 is used to convert erosion values to height
Consequently, at some places tillage translocation coun- differences (Desmet and Govers, 1995).
teracts water erosion and deposition processes. Van Oost
et al. (2003, 2004, 2005) demonstrated a marked shift in 2.2.2. Model output data
the relative contribution of water and tillage erosion over The model uses an annual time step. Water and tillage
recent decades in the Loess Belt of Belgium. Over a few erosion values are added up to give a total erosion or
thousand years, water erosion was more important than deposition mass in kg m− 2 for each grid cell. This value
tillage erosion. However, tillage erosion has become is then transformed into a height difference by dividing it
dominant during the last fifty years. In addition, even for by the dry soil bulk density, and finally this difference is
longer time spans, tillage erosion cannot be ignored since subtracted from the input DEM. Since erosion is
the spatial influence of this process is significantly represented by negative values and deposition by
different from the impact of water erosion. positive values, height is updated backwards in time.
The net soil flux caused by tillage on a hillslope of The resulting DEM thus shows the topography of the
infinitesimal length and unit width is proportional to the previous year which can then be used as an input for the
local slope gradient (Govers et al., 1994): next iteration. This process is repeated for each iteration,
so that finally the topography of thousands of years ago
Qs;t ¼ ktil S ¼ −ktil dh=dx ð4Þ is reconstructed.
Apart from the resulting reconstructed DEM, maps of
where Qs,t represents the net downslope flux due total water erosion, total tillage erosion and grand total
to tillage translocation (kg m− 1 year− 1), ktil is the tillage erosion are generated, as well as a text file with ad-
transport coefficient (kg m− 1 year− 1), S is the local slope ditional information on the mass balance.
gradient (m m− 1), h is the height at a given point of the
hillslope (m) and x is the horizontal distance (m). 2.2.3. Pit removal
The model does not account for soil erosion due to A topographically based model requires a good input
crop harvesting (Poesen et al., 2001; Ruysschaert et al., DEM. Such a DEM has to be sufficiently smooth, so that
2004). Although this process is at present quite the routed sediments are not trapped in small local
important in the study area, its relative importance depressions, causing discontinuities, although the main
over longer time spans is limited due to the relatively topographic features should be well represented. Usually
late introduction of root crops such as sugar beets and this is achieved by a combination of filtering and pit
chicory. For a more detailed description of the tillage removal algorithms. Mean filtering allows removing
erosion module, please refer to Van Oost et al. (2000a). small discontinuities that may affect model results
(Desmet and Govers, 1996b; Desmet, 1997). A pit re-
2.2. Model implementation moval procedure needs to be applied in order to remove
unrealistic closed depressions inherent to DEMs.
2.2.1. Model set-up We assume that all pits found in our DEM are either
At present, the model used is purely based on artificial or due to spatially intermittent data sampling in
topographic attributes derived from a Digital Elevation creating the DEM. In the Loess Belt of central Belgium,
Model (DEM). Long-term landform modelling in prin- where our study area is situated, local closed depressions
ciple requires the original topography at the start of the can be found, but Gillijns et al. (2005) have shown that
period of interest. However, Rommens et al. (2005) most of them have an anthropogenic origin. Therefore,
showed that reconstructing original topography based on we opted for the irrevocable removal of all pits.
soil profile data leads to significantly different results Many pit removal procedures exist, and the most
depending on the interpolation method used. Therefore, widely used one is the method of Jenson and Domingue
a different approach was chosen, i.e. starting from (1988). A disadvantage of this and similar methods is
present-day topography shown by the input DEM and that pits are turned into flat areas. This causes problems
calculating backwards to reconstruct previous topogra- when the sequence of cells through which the sediments
phy. This backward approach will improve the applica- are routed is based on the slope gradient. Planchon and
bility of the model to areas where no data on ancient Darboux (2001) proposed another way of managing pits.
I. Peeters et al. / Geomorphology 78 (2006) 250–264 253

Fig. 1. Recalculating topography through erosion and deposition modelling: calculating forwards vs. backwards in time. Calculating forwards
smoothens the surface, while calculating backwards triggers the occurrence of a local depression.

Fig. 2. Height distribution in the Nodebais catchment (hatched) and location of the catchment in central Belgium. Contour interval: 2.5 m.
254 I. Peeters et al. / Geomorphology 78 (2006) 250–264

It first inundates the surface with a thick layer of water deposition amount is subtracted from the surface. Pits are
and then removes the excess water. A slightly adapted thus exaggerated through calculations, which is prob-
version of this method has been integrated in our model. lematic for obtaining the final model results. Conse-
A value ε, which corresponds to the minimal height quently, pit removal is performed for each iteration step
difference between neighbouring cells, is defined to of modelling.
create gently sloping instead of flat surfaces. When this
value is set to zero, flat areas will originate. In our study, 3. Study area
ε was set to
The hilly loess landscapes of Western and Central
e ¼ 0:0001⁎i⁎j ð6Þ Europe provide unique opportunities to assess the extent
to which observed landscape evolution can be simulated
where i is the row number and j is the column number. As by geomorphological models, as landscape evolution
this number is unique for every cell, neighbouring cells over the last millennia has been relatively rapid due to
do not end up with equal heights where they have the human impact and rates of erosion and deposition can be
same lowest neighbour. Depressions are thus replaced relatively well quantified. Calibration of the model is
with gently sloping surfaces. therefore done using data from a soil augering campaign
Problems that may occur due to the pit handling in a small catchment (103 ha) in the Belgian Loess Belt, a
procedures are triggered by the fact that local slopes can west–east oriented loess region with a gentle rolling hill
become very small and therefore cause large amounts of topography. The catchment is situated in Nodebais,
sediments to be deposited on the wrong places. The 13km south of Leuven and is part of the Nethen
backward calculation may reinforce small topographic catchment, a tributary of the Dijle (Fig. 2). In the south of
discontinuities instead of smoothing (Fig. 1). This stresses the study area a plateau with slopes less than 3% can be
the importance of pit removal schemes when calculating found, while in the north three dry valleys with slopes up
backwards. The main problem in this case is the ap- to 28% converge to the outlet in the north–east.
pearance of pits on the recalculated versions of the DEM. Soils in the region are mainly loess-derived Luvisols
If pits are not removed correctly they are very likely to with a silt loam texture. These soils, known for their
remain through several iterations, since such a feature in fertility, have been intensively cultivated since pre-his-
the topography causes deposition to occur and this toric times. At present, land use in the Nodebais

Fig. 3. Location of augering points (stars) and a trench (grey bar) used to make quantitative estimates of historical erosion and deposition amounts
based on soil profile truncation (Rommens et al., 2005).
I. Peeters et al. / Geomorphology 78 (2006) 250–264 255

catchment is still mainly arable (N90%), with wheat, characterised by a rather large uncertainty. Nevertheless,
sugar beets, potatoes, chicory and flax as main crops. the experimental error on the delineation of most of the
soil horizons, sediment packages and the decalcification
4. Methods of model calibration and evaluation front is 5cm (Rommens et al., 2005).
Apart from having good information on volumes of
4.1. Use of long-term soil erosion data erosion and deposition, it is crucial to know when intense
water erosion started in order to run the model for the
The soil profile typical to Luvisols was formed after appropriate time period. Rommens et al. (2005) took four
deposition of the loess in the late Pleistocene. Decalci- charcoal samples from the sediments directly overlying
fication occurred and clay became mobile in the upper the palaeosol in an upslope hollow. They returned
soil horizons, forming a Bt horizon with clay-accumula- calibrated 14C-ages of 970–800, 770–400, 770–410 and
tion. As erosion leads to soil profile truncation and sed- 790–410 BC, respectively (with 95% probability). The
iment deposition buries the original soil profile under a calibration was done using OxCal v3.9 (Bronk Ramsey,
colluvial layer, the current depth of this Bt horizon and the 1995, 2001). These results indicate that the colluvial
decalcification front can be used for the spatial deter- deposition of sediments in the catchment started in the
mination of past soil redistribution processes. Rommens late Holocene (late Bronze Age to Iron Age). Our soil
et al. (2005) used this information to reconstruct a survey in the study area revealed no traces of soil
sediment budget for the Nodebais catchment. The depth formation in the colluvial deposits, which indicates con-
of erosion or deposition was estimated by hand augerings tinuous erosion and deposition. This significant water
at 160 locations that were selected to represent different erosion was triggered by anthropogenic factors including
topographic positions in the landscape, mainly along extensive forest clearing, cultivation and grazing, as
transects perpendicular to the dry valley bottom or indicated in literature (e.g. Martens, 1981; Starkel, 1992;
thalweg (Fig. 3). Augering density was increased in zones Vanwalleghem et al., 2003, 2005).
with a larger topographical variation, namely the incised
valleys in the northern part of the catchment. These data 4.2. Comparing simulation results with measured data
were further complemented by a detailed analysis of the
soils and sediments in a field trench (Fig. 3). The comparison between the modelled and measured
From this information three assumptions were made erosion/deposition volumes can be done in different
about the early Holocene soil geography: ways. We tested the following three ways.

(1) the decalcification and soil development rates in (1) The original measurement values at 160 augering
non-erosive circumstances are independent of the points are compared with the simulated values at
slope gradient and aspect; the corresponding cells.
(2) the depth of decalcification in a non-eroded soil in (2) The average of values for the target cell and its
the catchment is 2.3 m, which is the average for 14 neighbouring eight cells is used for comparison.
soil profiles at flat topographic positions where This method eliminates local errors, as outliers
soil erosion processes are considered insignifi- can be smoothened through aggregation (Van
cant, and Rompaey et al., 1999), but has the disadvantage
(3) the depth of the upper boundary of the clay that cells with different behaviours (erosion vs.
illuviation horizon (Bt) of non-eroded soils in the deposition) are treated together.
catchment is 0.4 m and the depth of the lower (3) A comparison is made based on a classification
boundary of the Bt horizon is 1.5 m, also into landscape elements. The study area was
estimated from 14 soil profiles in flat landscape classified into nine classes, defined by a com-
positions. bination of slope and contributing area (Table 1).
For each class the average of the simulated val-
The combination of augering data with a more de- ues was compared with that of the measured
tailed analysis of the sediments in the trench allowed a values.
quantitative estimate of total eroded and deposited
volumes as well as that of the spatial pattern of erosion 4.3. Parameter sets and goodness of fit
and deposition. It should be kept in mind though that
some estimates, especially on eroded landscape posi- Thousand parameter sets were generated by varying
tions or places where deposition also occurred, are parameters d, C1, e1, kTC and ktil of Eqs. (1)–(4) within
256 I. Peeters et al. / Geomorphology 78 (2006) 250–264

Table 1 4.4. Comparison of forward and backward simulations


Division of the landscape into nine landscape element classes, based
on slope gradient and contributing area
In order to investigate the performance of the newly
Class number Slope (°) Upslope area (m2) developed backward procedure, we tested the model on a
1 0–1.5 0–10,000 two-dimensional convex–concave hillslope profile.
2 1.5–3 0–10,000 First, the total erosion amount of the last 2500 years
3 3–6 0–10,000
was simulated based on backward calculation. The re-
4 N6 0–10,000
5 0–1.5 10,000–200,000 sulting profile was then used as an input for the forward
6 1.5–3 10,000–200,000 counting version of the model, and total erosion amounts
7 3–6 10,000–200,000 were compared in order to find out whether the results of
8 0–1.5 N200,000 the backward approach are equal to those of the traditio-
9 1.5–6 N200,000
nal forward calculations. The same exercise was per-
formed for the three-dimensional study area of Nodebais.
ranges inferred by Govers et al. (1993), Desmet and
Govers (1995, 1997), and Van Oost (2003) for water 5. Results
erosion, and by Van Oost et al. (2003) for tillage erosion.
The values of parameter a, b and c are kept constant at 5.1. Comparing simulated data with measured data
0.012 kg m− 2, 0.7 and 0.0015 kg m− 4, respectively. The
value of e is restricted to values smaller than 0.8. A parameter set with the highest model efficiency
For each parameter set the model was run for a (MEF) for each comparison method, determined after one
time period of 2500 iterations. Model performance thousand Monte Carlo simulations, is given in Table 2.
was determined using the model efficiency parameter These best-fit values are comparable to parameter values
(MEF), which measures the absolute agreement be- previously found (Govers et al., 1993; Desmet and
tween observed and simulated data (Nash and Sutcliff, Govers, 1997; Van Oost, 2003) (Table 2)). Our sensitivity
1970): analysis suggests that the parameter d (the slope exponent
for rill erosion) has the greatest influence on the model
MEFi ¼ 1−r2i =r2obs ð5Þ performance.

where MEFi is the model efficiency of the parameter set 5.1.1. Cell-based approach
i, σi2 is the variance of the residuals between measured Optimal values for parameters d, C1, e1, kTC and ktil to
and simulated variables of the parameter set i and σobs
2
is obtain the highest cell-based MEF are 1.35, 4.3, 0.68,
the variance of the observations. The value of MEFi lies 400m and 140kg m− 2, respectively (Table 2). This
between − ∞ and +1, with the model performance optimal parameter set gives a MEF value of 0.29. The
increasing proportionally. Usually the model perfor- scatter plot shown in Fig. 4 offers a better insight in the
mance is said to be satisfactory when MEFi is higher agreement between the modelled and measured data. It is
than 0.5 (Quinton, 1997). obvious that the cell-based model has difficulties

Table 2
Optimal parameter values of Eqs. (1)–(4) after calibration for each comparison method, and previously found parameter value ranges (Govers et al.,
1993; Desmet and Govers, 1997; Van Oost, 2003)
Parameter Optimal values after calibration Suggested Mean
symbol value ranges in values in
Cell-based Cell-aggregation Class-based
literature literature
a 0.012 0.012 0.012 0.0011–0.012 0.006
b 0.7 0.7 0.7 0.5–2.0 0.95
c 0.0015 0.0015 0.0015 0.0003–0.0015 0.0008
d 1.35 1.20 0.83 0.5–2.0 1.09
C1 4.3 10 7.4 – –
e1 0.68 0.80 0.96 0.80 0.80
e 0.3–0.8 0.3–0.8 0.3–0.8 0.500–0.886 0.58
KTC 400 400 200 200–250 225
ktil 140 140 110 140a 140a
a
Long-term value.
I. Peeters et al. / Geomorphology 78 (2006) 250–264 257

Fig. 4. Simulated erosion (b0) and deposition (N0) values against measured erosion and deposition values for 160 augering points. The dotted line
represents the 1 : 1 line, which expresses the ideal situation where simulated and measured values are equal.

predicting rapid erosion as well as all deposition values. sudden change from high erosion rates to deposition at
In general, the intensity of erosion and deposition is the footslopes.
underestimated and several outliers affect the results. A more detailed analysis revealed that the most sig-
Nevertheless, the overall spatial pattern of erosion and nificant problems occur on the hillslope transition zones
deposition (Fig. 5) looks realistic, with the strongest where strong erosion suddenly shifts to deposition, since
erosion on the steepest hills in the north, an overall these zones are very sensitive to local conditions in both
decrease of erosion rates in a downslope direction, and a reality and the model. Fig. 6 shows that a minor change

Fig. 5. Simulated total erosion (b0) and deposition (N0) after cell-based calibration (in m). Deposition is simulated in thalwegs and erosion on steep
slopes.
258 I. Peeters et al. / Geomorphology 78 (2006) 250–264

in the height of a cell in the transition zone gives a major Deposition also characterises the cells downstream of the
change in erosion and deposition at the cell and its fourth cell. The erosion value in the seventh cell remains
neighbouring cells. When the height of the fifth cell is quasi-stable.
close to the height of the downstream (fourth) cell, an A second source of problems is thalwegs, where
almost flat area will originate between these cells. As a alternating high erosion and deposition rates were
consequence, more deposition occurs in these cells while simulated. If data from the transition zone and thalwegs
erosion occurs further downstream. No major changes are omitted, a MEF of 0.8 is obtained.
occur upstream in this case. When the height of the fifth
cell is only 0.1m lower than that of the upstream (sixth) 5.1.2. Cell-aggregation approach
cell, a knick point originates between the fourth and the The model encountered difficulties simulating the
fifth cell. Consequently, erosion occurs in the fourth and aggregated averages properly, and gives a maximum
fifth cells, while local deposition occurs in the sixth cell. MEF of only 0.17 at values of 1.2, 10, 0.8, 400 m and

Fig. 6. Modelling height of cells in a erosion/deposition transition zone along a two-dimensional transect. (a) Location of cells along a transect and
altering the height of the fifth cell. (b) Change in the height of the third, fourth, sixth and seventh cells as a function of the height of the fifth cell. This
is purely a model experiment.
I. Peeters et al. / Geomorphology 78 (2006) 250–264 259

Fig. 7. Comparison of the effectiveness of the cell-based and class-based models. The two arrows indicate the parameter sets with the best model
efficiency after cell-based and class-based calibration.

140kg m− 2 for d, C1, e1, kTC and ktil, respectively (Table 2). 5.1.3. Class-based approach
Again, several outliers affect the result. The poor result is The best parameter set of d, C1, e1, kTC and ktil for the
mainly due to the fact that aggregation goes beyond the class-based approach contains values of 0.83, 7.4, 0.96,
borders of landscape elements so that values of cells with 200 m and 110kg m− 2, respectively (Table 2). These
high erosion and deposition rates are averaged out. values differ significantly from the values obtained by
Especially in the thalweg region, where the transition the cell-based approach. Fig. 7 shows that there is no
between erosion and deposition is very sudden, results clear relationship between the cell-based and the class-
were worse. based MEF values (see the two arrows for the best-fit

Fig. 8. Simulated vs. measured average erosion (b0) and deposition (N0) values for each landscape element class.
260 I. Peeters et al. / Geomorphology 78 (2006) 250–264

models). This is due to the fact that the evaluation disadvantage of this method is the loss of spatial
criteria differ significantly, and might also indicate that detail.
the scale at which the model is useful is an issue.
The maximum class-based MEF is 0.78, and indeed 5.2. Comparing forward and backward simulations
the simulated averages approach the observed averages
for each hillslope/upslope area class (Table 1 and Fig. Our approach of calculating backwards in time is
8). The simple class-based model thus seems to gene- valid for a two-dimensional surface because no signif-
rate overall soil redistribution patterns rather well. A icant differences could be detected between the forward

Fig. 9. Ancient topography (a) and total erosion (b) for a two-dimensional hillslope profile. Calculating backwards gives almost the same result as
calculating forwards.
I. Peeters et al. / Geomorphology 78 (2006) 250–264 261

Fig. 10. The difference in simulated total erosion between backward and forward calculations (in m).

and backward calculations for a hillslope profile (Fig. 9). accurate output, as the introduction of additional details
For the study area, the largest difference in total erosion in the process descriptions will inevitably be accompa-
amounts between forward and backward calculations is nied by an increase in data input error (Loague and
only 0.18m and is found in the thalweg (Fig. 10). The Freeze, 1985; Jetten et al., 1999; Van Rompaey and
difference was mainly caused by the pit removal scheme, Govers, 2002).
which is applied much more often during the backward Our model could reproduce the overall evolution of
calculation, but it lies within an acceptable range. The the landscape over the last 2500 years quite well: erosion
greater part of the study area does not show any and sedimentation patterns and rates can be well
significant differences, which shows that the backward reproduced. It may be concluded that a simple,
approach has major possibilities at the proposed topographically based model has the potential to capture
spatiotemporal scale in three dimensions. the long-term dynamics of sediment movement in a
relatively large area. However, on the level of each raster
6. Discussion cell, it was difficult to accurately reproduce the measured
values of erosion and deposition, as already indicated by
6.1. Evaluation of the model Jetten et al. (2003). Part of the problems may be related to
fundamental issues in geomorphological modelling,
Landscape evolution models that are designed for use such as polygenesis and equifinality. Polygenesis
on a millennial scale tend to be rather complicated and implies that topographic information per se is not
data demanding due to their physically based nature (e.g. sufficient to reconstruct ancient topography as present-
Tucker et al., 1999). Van Oost et al. (2003) stated that day landforms might be the result of the action of various
simple topography based models, which focus on the possible combinations of processes. However, in this
dominant processes, may be more appropriate to study we consider a landscape that has been evolving
simulate erosion and deposition patterns on the decadal over a time period of ca. 2500 years so that changes in
to millennial time scale as they pose fewer problems for topographical attributes (slope, height, contributing area
parameterisation and calibration. Furthermore, the most and curvature) are relatively small with respect to their
complex model does not necessarily result in the most absolute values, and the fundamental character of the
262 I. Peeters et al. / Geomorphology 78 (2006) 250–264

landscape has not been altered. Furthermore, we coupled functioning of the model. A spatial evaluation indeed
topographic information with detailed, spatially distrib- allows us to identify weaknesses in our modelling
uted, field data on erosion and deposition volumes. approaches that may not be discovered in any other way.
Therefore, present-day topography is not the only In our case the main issues were adequate modelling of
constraint on our modelling results. Also, we can safely the transition between erosion and deposition at the foot
assume that water erosion has been the only dominant slopes as well as erosion and deposition in thalweg
process controlling landscape evolution in the study area areas. These problems may be partly related to
over the last 2500–3000 years, albeit that tillage erosion technical–numerical issues but they may also point to
has been dominant over the last 50 years. Evidence has inadequate process description in relation to extreme
been found in the Belgian Loess Belt that historical events as discussed below. Such information helps us to
gullies under cropland were filled up by water erosion consider strategies for further improvement of a
and thus stabilised without tillage practices (Vanwalle- modelling technique.
ghem et al., 2005), and permanent gullies are only
observed under forest. 6.2. Sediment delivery ratio
The issue of equifinality makes it hard to reconstruct
topography, since different initial forms can evolve into Rommens et al. (2005) calculated a sediment de-
the same current topography. In the most extreme case, livery ratio (SDR) of 20% to 42% for the Nodebais
the end topography is flat, which obviously makes it catchment. However, our modelling result for the catch-
impossible to reconstruct the previous slopes. However, ment yielded a SDR of only ca. 1%. Thus, although the
in this study we are investigating a landscape that has WATEM/SEDEM model was calibrated and validated
undergone only minor topographic changes. We there- using catchment sediment export data collected in the
fore believe that neither equifinality nor polygenesis region including the study area, the model is not capable
pose fundamental obstacles for the interpretation of our of predicting sediment export properly in this particular
results. Nevertheless, this issue leads to an important case. Although we tried some routing algorithms other
discussion concerning the validity of backward than flux decomposition, the choice of the algorithm
calculation. did not have a significant impact on the calculated sedi-
In both the forward and backward approaches, ment export.
uncertainties are intrinsic to the process of reconstructing A possible explanation for the extremely low SDR is
ancient landscapes, and both methods have advantages the low gradient of the catchment thalwegs. The
and disadvantages. We believe that there is a potential in WATEM/SEDEM model, on which our model is
backward calculation although it has not been explored conceptually based, calculates the transport capacity as
properly so far. Backward calculations do not need a direct function of the rill erosion potential, as shown in
estimations of past landscapes before a model can be run. Eq. (3). Consequently, when erosion potential is low, for
In addition, only a minor difference is found between example on gentle slopes, transport capacity will be low
backward and forward calculations in our modelling. It so that most of the incoming sediments will be
should be noted, however, that some problems with deposited. In order to overcome this difficulty, a
long-term geomorphological modelling are related to the different equation for transport capacity was tested:
numerical techniques applied: non-linear feedbacks and
discontinuities lead to numerical instability and/or the Tc ¼ fS g Ah ð7Þ
generation of topographical pits. Evidently, such pro- where f, g and h are parameters. Optimal values of g and
blems are much more critical when calculating backward h were taken from Prosser and Rustomji (2000) and are
in time rather than calculating forward. More advanced equal to 1.4 and 1.4. For the parameter f, a range
simulation techniques than ours (e.g. the use of adaptive between 0.003 and 0.03 was chosen. The use of Eq. (7)
grids) may help to remedy part of the problems. led to much higher sediment delivery ratios, up to 50%
However, numerical stability may pose fundamental depending on the value of the value of f. However,
limits on the time range for which backward modelling extreme outliers of deposition disturbed the spatial
can be used, especially when data to validate/constrain pattern of erosion and deposition: the higher the SDR is,
the model are limited. the worse the spatial pattern is.
Despite these difficulties, we believe that the spatial Another most likely explanation of the unrealistic
evaluation of a long-term landscape evolution model SDR value is the steady-state character of the model
based on field data offers important perspectives, since a used. Extreme events were not taken into account, but
quantitative evaluation gives us new insights into the they may be responsible for a significant account of the
I. Peeters et al. / Geomorphology 78 (2006) 250–264 263

sediment export. In the 250 ha Kinderveld catchment in Acknowledgements


Belgium, Steegen et al. (2000) calculated a SDR of ca.
75% for an extreme event of 16mm rainfall in ca. We would like to thank James Dixon for linguistic
15 min, which corresponds to a recurrence interval of ca. advice and corrections. The study is financed through the
10 year (Demarée, 1985). Also, in a 290 ha catchment OT-program of the K.U. Leuven, grant no: 3E020001.
near Tongeren, Takken et al. (1999) obtained a SDR of
ca. 60% for an extreme event of ca. 60 mm rainfall in References
60 min, thus with a recurrence interval well exceeding
200 year (Demarée, 1985). These values are clearly Addiscott, T., Smith, J., Bradbury, N., 1995. Critical evaluation of
models and their parameters. Journal of Environmental Quality 24,
higher than the long-term averages. 803–807.
The fact that we nevertheless satisfactorily repro- Ahnert, F., 1976. Brief description of a comprehensive three-
duced the observed erosion and sedimentation pattern dimensional process–response model of landform development.
indicates that the pattern of sediment redistribution Zeitschrift fur Geomorphologie 25, 29–49.
within the catchment during an extreme event is not Braun, J., Sambridge, M., 1997. Modelling landscape evolution on
geological time scales: a new method based on irregular spatial
fundamentally different from that of normal events. discretization. Basin Research 9, 27–52.
Bronk Ramsey, C., 1995. Radiocarbon calibration and analysis of
7. Conclusions stratigraphy: the OxCal program. Radiocarbon 37, 425–430.
Bronk Ramsey, C., 2001. Development of the radiocarbon program
Long-term erosion modelling is an important tool in OxCal. Radiocarbon 43, 355–363.
Coulthard, T.J., 2001. Landscape evolution models: a software review.
geomorphology and can be used to reconstruct ancient Hydrological Processes 15, 165–173.
topography. However, existing landscape evolution Coulthard, T.J., Kirkby, M.J., Macklin, M.G., 1997. Modelling
models have rarely been evaluated spatially, based on hydraulic, sediment transport and slope processes, at a catchment
comparisons between simulated results and measured scale, using a cellular automaton approach. In: Pascoe, R.T. (Ed.),
Proceedings of the second annual conference GeoComputation 97,
data. In this paper, a simple topographic hillslope
Incorporating the Spatial Information Research Centre's 9th
erosion model is presented that works backwards on Annual Colloquium. GeoComputation CD-ROM. University of
long time scales. The model was evaluated using three Otago, Dunedin, New Zealand, pp. 309–318 (ISBN 0-9533477-0-
different methods: (1) on a cell-to-cell level, (2) after XXXX, http://www.geocomputation.org/1997/).
raster cell aggregation, and (3) after classification into Demarée, G., 1985. Intensity–duration–frequency relationship of
landscape elements. The third approach based on point precipitation at Uccle. Reference period 1934–1983.
Koninklijk Meteorologisch Institut van België A-116, 3–52.
upslope area/slope classes shows that the model is Desmet, P.J.J., 1997. Effects of interpolation errors on the analysis of
capable of reproducing basic patterns of erosion and DEMs. Earth Surface Processes and Landforms 22, 563–580.
deposition. Furthermore, the obtained optimal parameter Desmet, P.J.J., Govers, G., 1995. GIS-based simulation of erosion and
values are similar to previously found values. Our study deposition patterns in an agricultural landscape: a comparison of
has shown that long-term erosion/deposition modelling model results with soil map information. Catena 25, 389–401.
Desmet, P.J.J., Govers, G., 1996a. Comparison of routing algorithms
may significantly benefit from spatial evaluation and for digital elevation models and their implications for the
calibration, and that topographically driven erosion/ prediction of the location of ephemeral gullies. First Joint
deposition models have a potential to simulate Holocene European Conference and Exhibition on Geographical Informa-
landform evolution. tion, The Hague, The Netherlands, pp. 324–329.
Desmet, P.J.J., Govers, G., 1996b. A GIS procedure for automatically
However, our results also show some discrepancies
calculating the USLE LS factor on topographically complex land-
between the model results and the field observations. scape units. Journal of Soil and Water Conservation 51, 427–433.
The most important problems are lower model perfor- Desmet, P.J.J., Govers, G., 1997. Two-dimensional modelling of the
mances in the transition zones between erosional and within-field variation in rill and gully geometry and location
depositional areas on hillslopes and in thalwegs, as well related to topography. Catena 29, 283–306.
as the underestimation of the sediment delivery ratio. Desmet, P.J.J., Poesen, J., Govers, G., Vandaele, K., 1999. Importance
of slope gradient and contributing area for optimal prediction of the
These problems may be related to technical issues as initiation and trajectory of ephemeral gullies. Catena 37, 377–392.
well as inadequacies within the model which could only Gillijns, K., Poesen, J., Deckers, J., 2005. On the characteristics and
be identified by comparing modelling results with origin of closed depressions in loess-derived soils in Europe — a
observed spatial data. Besides technical improvements, case study from central Belgium. Catena 60, 43–58.
future model applications may benefit from a better Govers, G., Quine, T.A., Walling, D.E., 1993. The effect of water
erosion and tillage movement on hillslope profile development: a
description of extreme events caused by storms and a comparison of field observations and model results. In: Wicherek,
better simulation for the transition between erosion and S. (Ed.), Farm Land Erosion in Temperate Plains Environment and
deposition, both on hillslopes and in thalwegs. Hills. Elsevier Science Publishers B.V., Paris, pp. 285–300.
264 I. Peeters et al. / Geomorphology 78 (2006) 250–264

Govers, G., Vandaele, K., Desmet, P.J.J., Poesen, J., Bunte, K., 1994. catchment in the Loam Belt of central Belgium. Geomorphology
The role of soil tillage in soil redistribution on hillslopes. European 33, 25–36.
Journal of Soil Science 45, 469–478. Takken, I., Beuselinck, L., Nachtergaele, J., Govers, G., Poesen, J.,
Hancock, G.R., 2004. Modelling soil erosion on the catchment and Degraer, G., 1999. Spatial evaluation of a physically-based
landscape scale using landscape evolution models — a probabi- distributed erosion model (LISEM). Catena 37, 431–447.
listic approach using digital elevation error. SuperSoil 2004: 3rd Tucker, G.E., Slingerland, R.L., 1994. Erosional dynamics, flexural
Australian New Zealand Soils Conference. University of Sydney, isostasy, and long-lived escarpments: a numerical modeling study.
Australia. 9 p. Journal of Geophysical Research 99 (12), 229–243.
Howard, A.D., 1994. A detachment-limited model of drainage basin Tucker, G.E., Slingerland, R.L., 1997. Drainage basin responses to
evolution. Water Resources Research 30, 2261–2285. climate change. Water Resources Research 33, 2031–2047.
Jenson, S.K., Domingue, J.O., 1988. Extracting topographic structure Tucker, G.E., Gasparini, N.M., Bras, R.L., Lancaster, S.T., 1999.
from digital elevation data for geographic information system Overview of the CHILD Model, Version 2.0. Department of Civil
analysis. Photogrammetric Engineering and Remote Sensing 54, and Environmental Engineering, Massachusetts Institute of
1593–1600. Technology, Cambridge. 17 p.
Jetten, V., de Roo, A., Favis-Mortlock, D., 1999. Evaluation of field- Van Oost, K., 2003. Spatial modelling of soil redistribution processes
scale and catchment-scale soil erosion models. Catena 37, in agricultural landscapes. Unpublished Phd-thesis, K.U.Leuven,
521–541. Belgium.
Jetten, V., Govers, G., Hessel, R., 2003. Erosion models: quality of Van Oost, K., Govers, G., Desmet, P., 2000a. Evaluating the effects of
spatial predictions. Hydrological Processes 17, 887–900. changes in landscape structure on soil erosion by water and tillage.
Loague, K.M., Freeze, R.A., 1985. A comparison of rainfall-runoff Landscape Ecology 15, 577–589.
modelling techniques on small upland catchments. Water Van Oost, K., Govers, G., Van Muysen, W., Quine, T.A., 2000b.
Resources Research 21, 229–248. Modeling translocation and dispersion of soil constituents by
Martens, E., 1981. Uit het verleden van de gemeente Oud-Heverlee. tillage on sloping land. Soil Science Society of America Journal
Gemeentebestuur Oud-Heverlee, Stroobants, Neerijse, Belgium. 64, 1733–1739.
253 p. Van Oost, K., Govers, G., Van Muysen, W., Nachtergaele, J., 2003.
Moglen, G.E., Bras, R.L., 1995. The effect of spatial heterogeneities Modelling water and tillage erosion using spatially distributed
on geomorphic expression in a model of basin evolution. Water models. In: Lang, A., Hennrich, K., Dikau, R. (Eds.), Long Term
Resources Research 31, 2613–2623. Hillslope and Fluvial System Modelling. Lecture Notes in Earth
Nash, J.E., Sutcliff, J.V., 1970. River flow forecasting through Sciences, vol. 101. Springer, Berlin, pp. 101–121.
conceptual models, 1, a discussion of principles. Journal of Van Oost, K., Beuselinck, L., Hairsine, P.B., Govers, G., 2004. Spatial
Hydrology 10, 282–290. evaluation of a multi-class sediment transport and deposition
Planchon, O., Darboux, F., 2001. A fast, simple and versatile algorithm model. Earth Surface Processes and Landforms 29, 1027–1044.
to fill the depressions of digital elevation models. Catena 46, Van Oost, K., Van Muysen, W., Govers, G., Deckers, J., Quine, T.A.,
159–176. 2005. From water to tillage erosion dominated landform evolution.
Poesen, J., Vandaele, K., van Wesemael, B., 1996. Contribution of Geomorphology 72, 193–203.
gully erosion to sediment production in cultivated lands and Van Rompaey, A., Govers, G., 2002. Data quality and model
rangelands. IAHS Publication 236, 251–266. complexity for regional scale soil erosion prediction. International
Poesen, J., Verstraeten, G., Soenens, R., Seynaeve, L., 2001. Soil Journal of Geographic Information Science 16, 663–680.
losses due to harvesting of chicory roots and sugar beet: an Van Rompaey, A., Govers, G., Baudet, M., 1999. A strategy for
underrated geomorphic process? Catena 43, 35–47. controlling error of distributed environmental models by aggrega-
Prosser, I.P., Rustomji, P., 2000. Sediment transport capacity relations tion. International Journal of Geographic Information Science 13,
for overland flow. Progress in Physical Geography 24, 179–193. 577–590.
Quinton, J.N., 1997. Reducing predictive uncertainty in model Van Rompaey, A., Verstraeten, G., Van Oost, K., Govers, G., Poesen,
simulations: a comparison of two methods using the European J., 2001. Modelling mean annual sediment yield using a distributed
Soil Erosion Model (EUROSEM). Catena 30, 101–117. approach. Earth Surface Processes and Landforms 26, 1221–1236.
Rommens, T., Verstraeten, G., Poesen, J., Govers, G., Van Rompaey, Van Rompaey, A., Govers, G., Puttemans, C., 2002. Modelling land
A., Peeters, I., Lang, A., 2005. Soil erosion and sediment use changes and their impact on soil erosion and sediment supply
deposition in the Belgian loess belt during the Holocene: to rivers. Earth Surface Processes and Landforms 27, 481–494.
establishing a sediment budget for a small agricultural catchment. Vanwalleghem, T., Van Den Eeckhaut, M., Poesen, J., Deckers, J.,
The Holocene 15, 1032–1043. Nachtergaele, J., Van Oost, K., Slenters, C., 2003. Characteristics
Ruysschaert, G., Poesen, J., Verstraeten, G., Govers, G., 2004. Soil and controlling factors of old gullies under forest in a temperate
loss due to crop harvesting: significance and determining factors. humid climate: a case study from the Meerdaal Forest (central
Progress in Physical Geography 28, 467–501. Belgium). Geomorphology 56, 15–29.
Starkel, L., 1992. The feedback mechanisms between the environ- Vanwalleghem, T., Bork, H.R., Poesen, J., Schmidtchen, G.,
mental systems and the behaviour of prehistoric man. In: Frenzel, Dotterweich, M., Nachtergaele, J., Bork, H., Deckers, J., Brüsch,
B., Reisch, L. (Eds.), Evaluation of Land Surfaces Cleared from B., Bungeneers, J., De Bie, M., 2005. Rapid development and
Forests by Prehistoric Man in early Neolithic Times and the Time infilling of a historical gully under cropland, central Belgium.
of Migrating Germanic Tribes. ESF Project: European Climate and Catena 63, 221–243.
Man, vol. 3. Gustav Fischer Verlag, Stuttgart, pp. 25–37. Willgoose, G.R., Bras, R.L., Rodriguez-Iturbe, 1991. A physically
Steegen, A., Govers, G., Nachtergaele, J., Takken, I., Beuselinck, L., based coupled network growth and hillslope evolution model, 1,
Poesen, J., 2000. Sediment export by water from an agricultural theory. Water Resources Research 27, 1671–1684.

Anda mungkin juga menyukai