Impact of the spatial and thematic resolution of Holocene anthropogenic land-cover scenarios on modeled soil erosion and sediment delivery rates

Abstract

Recent reconstructions of global Holocene vegetation may provide a powerful means to quantify long-term cumulative sediment fluxes induced by anthropogenic land-cover changes. However, the former’s low spatial resolutions question their potential for use in geomorphic models, since sediment redistribution processes operate at much smaller scales. Furthermore, current land-cover reconstructions often do not differentiate the typology of human impact, although the susceptibility of different anthropogenic land uses towards erosion varies greatly. Hence, the present study investigates the sensitivity of a spatially distributed erosion and sediment redistribution model to the spatial and thematic resolution of input land-cover information. The geomorphic model was applied to the Scheldt Basin (~19,000 km²) with low-resolution and high-resolution land-cover input, and subsequently compared to a field-based reference sediment budget of the Dijle subcatchment. Results indicate that low-resolution land-cover information, expressed as proportions of different land-cover types within each grid cell, leads to largely overestimated sediment fluxes. In contrast, spatial allocation of individual land-cover types to a 100-m grid yields more accurate results. These variations in model outcomes are related to differences in landscape connectivity between high- and low-resolution land cover. Moreover, geomorphic model results are non-linearly related to the area under cropland. This indicates that there is not only a need for land-cover reconstructions at high spatial resolution but also that differentiation of anthropogenic land-cover types is essential for accurate geomorphic modeling.

Keywords

Belgium geomorphic modeling human impact land-cover change sediment dynamics spatial resolution

Introduction

Since the introduction of agriculture in the Neolithic, anthropogenic land-cover changes have impacted the landscape through increased soil erosion and sediment redistribution. The gradual clearance of forested areas for agricultural purposes throughout the Holocene, driven by population growth and increasing food consumption per capita, exposed soils and hence caused erosion of sediments (e.g. Dotterweich, 2008; Notebaert and Verstraeten, 2010).

The relative importance of agricultural activities versus climate as a driving factor in Holocene sediment dynamics has been investigated by many authors. Results indicate that although the sensitivity to climatic fluctuations is increased by changes in land use, the role of climate in sediment dynamics is often of minor importance compared to the anthropogenic impact in regions with a long history of human occupation (e.g. De Moor et al., 2008; Giosan et al., 2012; Hoffmann et al., 2008; Lang et al., 2003; Oldfield and Dearing, 2003). An accurate quantification of these long-term sediment fluxes caused by anthropogenic land-cover changes is essential to better understand how human-induced erosion has impacted the Earth’s biogeochemical cycles. Inspired by Ruddiman’s (2003) early anthropogenic hypothesis, several authors have calculated global cumulative carbon emissions resulting from agricultural expansion and intensification since prehistoric times (Boyle et al., 2011; Kaplan et al., 2011; Lemmen, 2009; Olofsson and Hickler, 2007). However, although soil erosion constitutes an important component of the carbon cycle (Lal, 2005; Van Oost et al., 2007), this indirect effect of anthropogenic land-cover change is not included in current Holocene carbon budgets. The quantification of long-term human-induced sediment redistribution on continental to global scales is critical to achieve this.

Although population growth has been a major driving force of land-cover changes throughout the Holocene, detailed analysis of historic sediment fluxes shows that the response of sediment fluxes along the hillslope–river continuum is not linearly scaled to population, and that colluvial and fluvial systems respond differently (Verstraeten et al., 2009b). As yet, the impact of long-term agriculture on sediment redistribution within a landscape has mainly been quantified through field-based sediment studies in a variety of regions and at different spatial scales (e.g. Bertran, 2004; Fuchs et al., 2011; Lespez, 2003; Macaire et al., 2002; Passmore and Macklin, 2001; Rommens et al., 2005; Trimble, 1999). Major drawbacks of this method, however, are its reliability on datable material with sufficiently high temporal resolution, and its time-consuming nature (Verstraeten et al., 2009a). This hampers the quantification of human impact on erosion and sedimentary systems at scales ranging from individual hillslopes to large fluvial basins.

Geomorphic model approaches can to a large extent address these problems (e.g. Baartman et al., 2012; De Moor and Verstraeten, 2008; Ward et al., 2009), provided that accurate input data, including topography and land cover, are available. Previous studies have indeed shown that geomorphic models respond differently depending on the spatial resolution and accuracy of elevation data (Schoorl et al., 2000; Verstraeten, 2006). Modern digital elevation models from remote sensing are widely accessible on global scales and at high resolution, and Holocene topography has already been reconstructed through iterative landscape evolution modeling for a number of catchments (e.g. Coulthard et al., 2002; Temme et al., 2011). Moreover, global and continental databases of anthropogenic land use during historic and prehistoric times have recently been developed based on population and agricultural land per capita estimates (e.g. Kaplan et al., 2011; Klein Goldewijk et al., 2011; Lemmen, 2009; Pongratz et al., 2008), thus offering opportunities to estimate soil erosion and sediment redistribution on a Holocene timescale and in any region of interest. However, at present, the accuracy of the various land-cover reconstructions remains unknown. Furthermore, the spatial resolution of these land-cover maps is currently limited to 5 arcminutes at best. This questions their usefulness in geomorphic models, as erosion and sediment redistribution processes operate at much smaller spatial scales. Whereas geomorphic model sensitivity to observed and simulated land-cover conversions has been quantified previously (e.g. Collins et al., 2004; Nearing et al., 2005; Van Rompaey et al., 2003), the effect of the spatial resolution of land-cover data on modeled soil erosion and sediment delivery rates was not yet investigated. Therefore, the objective of the present study is to investigate how the spatial and thematic resolution of input land-cover maps affect the output of a geomorphic model that was calibrated to a spatial resolution of 100 m, and hence whether available land-cover reconstructions are suitable to be used for this purpose. Results of the modeling approach, conducted for the Scheldt catchment, are subsequently confronted with a field-based temporally explicit sediment budget.

Study area

The Scheldt River originates in northern France and flows over a course of 335 km through Belgium to the North Sea near Vlissingen, the Netherlands (Figure 1). The river’s catchment area is ~19,000 km², when its estuary downstream of Rupelmonde is not included. Elevation in this area ranges from 0 to 275 m a.s.l., and the topography is characterized by a dense river network, creating a hilly landscape. Soils in the Scheldt River basin consist mainly of podzols in the sandy areas of the northern part, while the central and southern parts are dominated by silty loamy luvisols. Rainfall in the study area ranges between 600 and 900 mm/yr with an average value of approximately 800 mm/yr.

Figure 1.

Elevation and delineation of the Scheldt Basin upstream of Rupelmonde. The hatched area represents the Dijle subcatchment.

Palynological and geoarchaeological research in the study area indicates that Neolithic farmers first cultivated the landscape in the Atlantic period (~7800–5000 BP; e.g. Bakels, 1992; Cauwe et al., 2001; De Smedt, 1973; Mullenders and Gullentops, 1957; Mullenders et al., 1966; Van Berg and Hauzeur, 2001), as elsewhere in Europe (Price, 2000). Clearance of the mixed deciduous forest occurred locally, in the vicinity of settlements. Large-scale deforestation took place only since the Iron Age, and peaked during the Roman Period and from the Middle Ages onwards (Broothaerts et al., 2013; De Smedt, 1973; Rommens et al., 2007; Vanwalleghem et al., 2004; Verbruggen and Van Dongen, 1976). At present, land cover in the Scheldt Basin according to the CORINE 2000 land-cover maps (^©European Environment Agency, http://www.eea.eu; Bossard et al., 2000) consists mainly of cropland (54%), urban areas (22%), pasture (16%), and forest (8%).

Methodology

In order to investigate the impact of the accuracy and the native spatial resolution of global land-cover maps on simulated historic erosion rates, we applied the WaTEM/SEDEM geomorphic model (Van Oost et al., 2000; Van Rompaey et al., 2001; Verstraeten et al., 2002). WaTEM/SEDEM was developed and calibrated based on contemporary soil erosion and sediment delivery rates in the Belgian Loess Belt, and has been used to simulate long-term sediment fluxes in the Dijle catchment, a tributary of the Scheldt River (Notebaert et al., 2011b), and several other watersheds varying in size (e.g. De Moor and Verstraeten, 2008; Ward et al., 2009).

Land-cover information in this study is derived from global reconstructions by Kaplan and Krumhardt, hereafter called KK10, which are resolved spatially at 5′ and temporally at 1 year from 8000 to 100 BP (Kaplan et al., 2011). Ages in the KK10 data refer to years before present, that is, before 1950, and are comparable to calibrated radiocarbon ages. These maps indicate the proportion of anthropogenic land cover, without further differentiation of land-use types. Since geomorphic models require high thematic resolution, we introduced five scenarios where cropland constitutes 0%, 25%, 50%, 75%, and 100% of the anthropogenic land cover and the remainder is constituted by pasture. At the same time, these scenarios allow to assess how land-cover accuracy affects geomorphic model results. The dates considered in the present study were chosen to cover the entire period between the beginning of the Neolithic (i.e. ~8000 BP) and 100 BP with a temporal resolution of 1000 years. In addition, some historically important dates regarding human impact on land cover and data availability in the study region were included, representing the Roman Period (1750 BP), early and late Medieval times (650 and 300 BP), and the period in which the oldest available land-cover map of the region was created (180 BP).

Three sets of model runs were performed in WaTEM/SEDEM for the sensitivity analysis, elaborated below in more detail. First, low-resolution KK10 land-cover data were used without allocation, that is, expressed as proportions of different land-cover types within each 5′ grid cell. In a second set of model runs, estimated anthropogenic land cover was spatially allocated to a high-resolution grid (100 m) using a logistic regression model that predicts the pixels’ suitability for different land-cover types, based on various environmental parameters. Last, since previous results have shown that WaTEM/SEDEM calibration parameters are sensitive to the quality of land-cover input data (Verstraeten, 2006), the model was recalibrated based on contemporary sediment yield data for use with low-resolution land-cover data, which were subsequently used as input in a third set of model runs. Results of all sets of model runs were compared mutually to analyze the impact of resolution on model output before and after recalibration, and with a field-based sediment budget of the Dijle subcatchment (Notebaert et al., 2011a; Van Oost et al., 2012) for evaluation.

Spatial allocation of land cover to a high-resolution grid

A relation between the present-day occurrence of land-cover types such as forest, pasture, and cropland in the Scheldt Basin, on the one hand, and a series of environmental factors, on the other hand, was established through logistic regression analyses in SAS^® software in order to create suitability maps for each land-cover type. Logistic regression models were previously used in geographic probability assessments for a variety of purposes, such as habitat selection (e.g. Keating and Cherry, 2004), landslide susceptibility mapping (e.g. Guzzetti et al., 1999; Van Den Eeckhaut et al., 2006), and land-cover allocation (e.g. Temme and Verburg, 2011; Van Dessel et al., 2011). Environmental variables taken into account are the following:

Soil drainage and texture, both from the European Soil Database (ESDB; ^©European Union, http://eusoils.jrc.ec.europa.eu) at 1 km spatial resolution;

Slope and elevation, based on the 90-m Shuttle Radar Topography Mission digital elevation model (SRTM DEM) (^©National Aeronautics and Space Administration (NASA) and National Geospatial-Intelligence Agency (NGA), http://www2.jpl.nasa.gov/srtm), which was bilinearly resampled to 100 m spatial resolution;

Landforms classified in Esri ArcGIS^® based on the 30-m ASTER Global DEM (© NASA, https://lpdaac.usgs.gov), according to the procedure described by Morgan and Lesh (2005), based on Dikau (1989);

The distance to rivers, the latter being defined through the Runoff module in IDRISI^® applied on the SRTM DEM and with an upstream area threshold of 1 km².

All maps were resampled to a spatial resolution of 100 m. Information on the occurrence of forest, pasture, and cropland was derived from CORINE land-cover maps of ad 2000 at 100 m resolution.

The resulting suitability maps of each land-cover type were subsequently used to downscale and allocate the KK10 land-cover data to a 100 m resolution grid in the Scheldt Basin. Anthropogenic land cover within each original KK10 pixel with a 5′ resolution, containing over 5000 high-resolution pixels, was allocated independently using a multi-objective land allocation (MOLA) procedure in IDRISI. MOLA assigns a grid cell to the land-cover class for which it has the highest relative rank, until areal demands of each class are met (Eastman et al., 1995). The result is a set of 65 land-cover maps with a 100 m resolution, that is, five cropland/pasture ratios for each of the 13 considered dates.

Soil erosion and sediment delivery model

In this study, soil erosion and sediment deposition were modeled in WaTEM/SEDEM. WaTEM/SEDEM is a spatially distributed model that predicts sediment fluxes from hillslopes to river channels by calculating for each grid cell mean annual gross soil erosion, E (kg/m²/yr), from the revised universal soil loss equation (Wischmeier and Smith, 1978), and the transport capacity, TC (kg/m/yr; Van Oost et al., 2000; Van Rompaey et al., 2001; Verstraeten et al., 2002):

E = R K L S C P

(1)

T C = K_{T C} R K (L S - 4.1 s^{0.8})

(2)

where R is the rainfall erosivity factor (MJ·mm/m²/h/yr), K is the soil erodibility factor (kg·h/MJ/mm), LS is a two-dimensional slope-length factor as described in Desmet and Govers (1996), C is the crop factor, P is the erosion control practice factor, K_TC is the transport capacity coefficient (m), and s is the slope gradient (m/m). Eroded sediment is routed down to the river network, and gross deposition occurs when the sum of inflowing and in situ produced sediments exceeds the transport capacity. WaTEM/SEDEM predicts net annual hillslope erosion or colluvial deposition as the difference between gross erosion and deposition within each pixel. Modeled hillslope sediment delivery equals the total mass of sediment that is redistributed within the catchment from the hillslopes to the fluvial system. Note that WaTEM/SEDEM calculates sediment delivery to the river network, but not sediment transport within this system. Hence, there is no differentiation between floodplain and channel deposition, riverbank erosion, and total sediment export from the catchment. In this study, simulated hillslope sediment delivery includes all of these components.

Verstraeten (2006) calibrated WaTEM/SEDEM based on observed sediment yield data, to model contemporary sediment fluxes to rivers draining the Scheldt Basin with SRTM elevation data at 100 m resolution. Elevation and K factor maps employed in the present study, as well as values of the R, C, P, and K_TC factors, were obtained from Verstraeten (2006). The annual rainfall erosivity R is assumed constant through time to exclude variations in model output other than those induced by varying land cover, and since Notebaert et al. (2011b) have shown that sediment fluxes on Holocene timescales in the Dijle subcatchment are nearly insensitive to climatic variations. C factor values corresponding to each land-cover type are 0.37 for cropland, 0.01 for pasture, and 0.001 for forest. Calibrated K_TC-values equal 7 m for non-erosion-prone land cover (forest and pasture) and 23 m for cropland, implying that the latter can transport more than three times as much sediment as pasture or forest.

Sensitivity of WaTEM/SEDEM to land-cover resolution

Low-resolution land-cover input

In a first set of WaTEM/SEDEM model runs, soil erosion, deposition, and hillslope sediment delivery are calculated based on low-resolution, non-allocated land cover. A mask of the study area at 100 m resolution was used as land-cover input map, since no information on individual land-cover parcels is available. Instead, we accounted for land cover by creating a 100 m resolution C factor map in which each pixel’s C value is calculated as a linear combination of the different land-cover types contained in the original 5′ pixel of the KK10 map:

\begin{array}{l} C = 0.001 * (1 - f_{anthropogenic}) + \\ [f_{anthropogenic} * (0.37 * f_{cropland} + 0.01 * (1 - f_{cropland}))] \end{array}

(3)

where $f_{anthropogenic}$ is the fraction of anthropogenic land cover in the KK10 map, and $f_{cropland}$ is the proportion of cropland within deforested areas. The latter varies according to the five cropland/pasture scenarios described above. A low-resolution 5′ pixel is hence represented in the C factor map by c. 5400 high-resolution pixels that all have the same C value. Note that the absence of spatially distributed land-cover input implies that all pixels within the study area have the same transport capacity coefficient K_TC, that is, 23.

High-resolution land-cover input

To model sediment redistribution based on high-resolution land-cover data, allocated maps were used as land-cover input maps. Consequently, K_TC and C factor values vary according to land-cover type. The presence of individual land-cover parcels also implies that LS factors in this model setup are lower than in the previous. Again, calculations were performed for all scenarios and dates.

Low-resolution land-cover input after model recalibration

Since transport capacity coefficients in WaTEM/SEDEM are calibrated for use with high-resolution (100 m) elevation and land-cover input, we recalibrated the model for use with non-allocated land-cover data. For this purpose, modern CORINE land cover was expressed as percentages of natural vegetation, cropland, and pasture at the level of the original KK10 low-resolution pixels. These proportions were subsequently used to create a C factor map (Eq. 3). The optimal K_TC value was calculated as described in Verstraeten (2006), based on a contemporary sediment yield data set, containing several catchments within the Scheldt Basin.

Evaluation of geomorphic model results

Finally, the results of different model runs are evaluated through comparison with independent sediment redistribution estimates in the study area. A field-based, time-differentiated sediment budget of the Dijle catchment upstream of Leuven (~760 km²) was established by Notebaert et al. (2011a) using accelerator mass spectrometric radiocarbon dating and optically stimulated luminescence dating, and updated by Van Oost et al. (2012). Their cumulative hillslope sediment delivery values within three distinct time periods were compared with sediment delivery to the Dijle simulated with WaTEM/SEDEM. In order to calculate cumulative sediment fluxes for a given time period, annual hillslope sediment delivery values of each simulated date were extrapolated both forward and backward to the midpoint of the period between two consecutive simulated dates. Ratios of cropland to pasture were assumed constant within a single time period to limit the number of possible outcomes. Note that the sediment budget by Notebaert et al. (2011a) dates back to 9000 bc, while KK10 maps are only available until 8000 BP or c. 6000 bc. The last 150 years of the field-based sediment budget are also lacking in the KK10 data set. Hence, annual hillslope sediment delivery at 6000 bc for each scenario and model setup was linearly interpolated towards the value corresponding to an entirely forested catchment in 7000 bc and the preceding millennia, which is independent of the land-cover resolution. With respect to the last 150 years, hillslope sediment delivery values at 100 BP, or ad 1850, were extrapolated to ad 2000.

Results

Spatial allocation of land-cover to a high-resolution grid

Bidirectional stepwise logistic regression analysis at a 0.05 significance level showed that all tested variables except elevation are correlated to the presence of cropland, pasture, and forest. Locations of cropland are positively correlated to well-drained and fine-grained (loamy) soils, weak slopes, and distance to rivers. The opposite applies to pasture and forest, except that the latter is less likely to occur close to rivers. Measures of model performance are provided in Table 1. Receiver operating characteristic (ROC) values of the suitability maps of cropland, pasture, and forest are all larger than 0.50 and thus confirm that these maps perform better than random in predicting land-cover locations (Pontius and Schneider, 2001).

Table 1.

Logistic regression model performance for predicting the land’s suitability for cropland, pasture, and forest. Wald and likelihood ratio statistics are all significant on a <0.0001 level.

Evaluation statistic	Cropland suitability	Pasture suitability	Forest suitability
Wald χ²	713	370	531
Likelihood ratio χ²	824	409	604
ROC-value	0.69	0.65	0.76

ROC: receiver operating characteristic.

Figure 2 provides a detailed view on the low-resolution KK10 land-cover reconstruction of 3000 BP with a 50/50 cropland/pasture ratio before and after the allocation procedure. Note how the soil texture and drainage data, resolved at 1 km and used as explaining variables in the land-cover suitability assessment, lead to rectangular patterns in the allocated land-cover map.

Figure 2.

Fragment of the original KK10 map for 3000 BP at 5′ resolution (above) and the same map after spatial allocation with a 50/50 cropland/pasture ratio to a 100-m grid (below).

Geomorphic model results

The average fraction of anthropogenic land cover according to the KK10 land-cover reconstruction data increases steadily through time in the Scheldt Basin, from 11% in 8000 BP to a maximum of 75% in 100 BP (Figure 3a). An overview of modeled hillslope sediment delivery values since the Neolithic, based on high-resolution data as well as low-resolution land-cover maps before and after recalibration, is presented in Figure 3b–f.

Figure 3.

Temporal evolution of (a) KK10 anthropogenic land cover within the Scheldt catchment and (b–f) the modeled annual hillslope sediment delivery for three sets of model runs and varying cropland/pasture ratios.

Low-resolution land-cover input

Annual hillslope sediment delivery modeled in WaTEM/SEDEM with low-resolution, non-allocated land-cover input increases through time and hence with the proportion of deforested land (Figure 3). Depending on the ratio of cropland to pasture, sediment delivery values at 100 BP constitute 300–550% of the respective 8000 BP value, which is less than the sevenfold increase of anthropogenic land cover. The largest increase is observed for the 50/50 scenario, whereas in the scenarios with larger cropland proportions, the growth in sediment delivery levels off from approximately 2000 BP onwards. Absolute hillslope sediment delivery at 100 BP ranges from 308 to 4189 kt/yr, and is positively correlated to the proportion that cropland constitutes within the deforested area.

Modeled annual sediment production and deposition follow a long-term trend similar to hillslope sediment delivery; however, absolute values are, respectively, higher (483–6931 kt/yr at 100 BP) and lower (175–2742 kt/yr) than the latter. Variation between different scenarios indicates that sediment production and deposition also strongly depend on the area under cropland.

The hillslope sediment delivery ratio (HSDR), which is the proportion of net eroded sediment that is delivered to the river network, only slightly varies through time and between scenarios of the cropland/pasture ratio. For a given age, for example, 100 BP, HSDR based on non-allocated land-cover maps ranges from 58% to 64% depending on the scenario (Figure 4). For a given cropland/pasture ratio, HSDR does not change much through time, except between 8000 and 7000 BP. When the proportion of anthropogenic land cover is very low, relatively more sediment (up to 76%) is delivered to the river for the scenario with only pasture.

Figure 4.

Modeled hillslope sediment delivery ratio (HSDR) at 100 BP and for different scenarios of the cropland/pasture ratio, based on allocated and non-allocated land-cover input.

High-resolution land-cover input

When high-resolution, allocated land-cover input is used in WaTEM/SEDEM, modeled annual hillslope sediment delivery is considerably lower compared to the values based on low-resolution maps (Figure 3): absolute values at 100 BP vary between 187 and 2573 kt/yr. Again, these increase with the proportion of cropland within the total anthropogenic land-cover area. However, temporal differences are much higher for allocated than for non-allocated models, with a maximum increase throughout the studied time period of more than 2000%, corresponding to the scenario where all anthropogenic land cover consists of cropland.

Modeled annual sediment production and deposition in the study area also increase through time, that is, with the proportion of deforested land, as well as with the cropland/pasture ratio. At 100 BP, erosion ranges from 261 to 3891 kt/yr, while deposition constitutes 74–1318 kt/yr, depending on the scenario. When the proportion of anthropogenic land cover equals only 11%, which is the case at 8000 BP, annual sediment production and deposition are much lower (respectively, 106–311 kt/yr and 17–183 kt/yr for different cropland/pasture ratios).

Whereas, for a given age, the HSDR of non-allocated models is almost constant with changing cropland/pasture ratios, this is not the case for model runs with allocated maps. At 100 BP, as shown in Figure 4, the proportion of eroded sediment delivered to the river is highest (72%) for the scenario where all anthropogenic land cover is pasture. In contrast, HSDR reaches a minimum of 42% when cropland and pasture are evenly distributed, after which it increases again to 66% when all anthropogenic land cover is cropland. Differences in HSDR through time for a given ratio between cropland and pasture are not as important as the variation between scenarios. Only for the earliest dates, the HSDR is considerably higher (up to 84%) compared to younger ages when there is only pasture, while it is lower (down to 41%) for the scenario where all anthropogenic land cover is cropland.

Low-resolution land-cover input after model recalibration

Recalibration of WaTEM/SEDEM based on contemporary low-resolution land-cover input and observed sediment yield data, resulted in an optimal transport capacity coefficient K_TC of 10 m with a model efficiency (Nash and Sutcliffe, 1970) of 85%. Using low-resolution land-cover maps in the recalibrated model yields annual hillslope sediment delivery values that lie between the allocated and non-allocated model results described in the previous sections, except for the scenario where all anthropogenic land cover is cropland (Figure 3f). Here, modeled sediment delivery for the last 2000 years is lower than in the other model runs. Absolute values vary between 249 and 1784 kt/yr at 100 BP, and they even become constant for high proportions of anthropogenic land cover in combination with high ratios of cropland to pasture. Hence, increases through time for a given scenario are rather low and reach a maximum of 467% for the scenario where cropland and pasture are evenly distributed. The same trend is observed for modeled annual erosion and deposition, with values of, respectively, 438–3046 kt/yr and 189–1262 kt/yr at 100 BP, depending on the scenario.

Due to the leveling of annual hillslope sediment delivery as well as erosion and deposition values at high proportions of cropland, HSDR is approximately constant, ranging only between 57% and 59% (Figure 4). As for non-allocated models without calibration, only for the earliest dates and for the 100% pasture scenario, HSDR is higher (up to 67%).

Evaluation of geomorphic model results

Table 2 provides a comparison of estimated cumulative hillslope sediment delivery values based on allocated and non-allocated land-cover maps, with reference values of the Dijle catchment. Field-based cumulative hillslope sediment delivery for the three time periods corresponds well to the allocated model’s cropland/pasture ratios of, respectively, 25/75, 75/25, and 100/0. When we look at non-allocated model results, the reference value for the first two time periods lies close to the scenario where all anthropogenic land cover is pasture. In contrast, between ad 100 and ad 2000, a cropland/pasture ratio between 25/75 and 50/50 simulates cumulative hillslope sediment delivery best.

Table 2.

Comparison of modeled cumulative hillslope sediment delivery (Mt/yr) within the Dijle catchment with a time-differentiated field-based sediment budget by Notebaert et al. (2011a) and Van Oost et al. (2012).

Time period	Field-based sediment budget	Allocated model					Non-allocated model
Time period	Field-based sediment budget	0/100	25/75	50/50	75/25	100/0	0/100	25/75	50/50	75/25	100/0
9000–2000 bc	94 ± 14	80	88	117	204	410	112	504	1125	1508	1631
2000 bc to ad 1000	166 ± 25	50	60	93	198	464	91	375	1062	1254	1306
ad 1000–2000	227 ± 34	18	22	35	76	192	33	132	389	446	459

For each time period, five cropland/pasture ratios were simulated in both allocated and non-allocated models.

Discussion

Evaluation of geomorphic model results

Comparison of modeled and field-based cumulative hillslope sediment delivery points out that allocated land-cover maps yield good model results, especially between 2000 bc and ad 1000, whereas the use of non-allocated maps seems to overestimate hillslope sediment delivery. Indeed, estimated proportions of cropland and pasture for historic times suggest that a 50/50 to 75/25 ratio is realistic (e.g. Gregg, 1988; Klein Goldewijk et al., 2011; Pongratz et al., 2008). Also modern land-cover observations indicate that about 75% of anthropogenic land cover within the Scheldt catchment is covered by cropland. However, WaTEM/SEDEM results based on high-resolution land-cover input overestimate the amount of sediment that reaches the river system between 9000 bc and 2000 bc, while hillslope sediment delivery within the last 1000 years is underestimated.

We suggest that the poorer model performance of allocated land-cover maps for the first period is mainly due to overestimated anthropogenic land-cover areas within the Dijle catchment, which covers 24 of the 5′ KK10 pixels partially or fully. This assumption is supported by palynological research within this area, indicating that large-scale deforestation occurred from c. 2500 BP onward (Broothaerts et al., 2013; De Smedt, 1973; Mullenders et al., 1966). In contrast, according to the KK10 data, half of the Dijle catchment was deforested as early as 7000 years ago. Of course, KK10 land-cover information was originally calculated at the national level and subsequently spatially allocated to a 5′ grid, implying that inaccuracies at pixel level do not necessarily reflect inaccuracies at the national scale. Neolithic anthropogenic land-cover estimates for the entire Scheldt Basin are indeed more moderate (Figure 3a) and correspond better to palynological and archaeological records. For the last 1000 years, on the other hand, documentation on historical land-cover areas in Flanders suggests that the reconstructions implemented in our models underestimate anthropogenic land cover: whereas the KK10 land-cover reconstruction sets the proportion of natural vegetation at 180 BP to 34%, only 11% of the landscape was forested according to the Ferraris map of the 1770s (Geografische Data Infrastructuur (GDI)-Vlaanderen; http://www.agiv.be/gis/).

We should note, however, that apart from anthropogenic land-cover estimates, several other model uncertainties might contribute to inaccuracies in simulated hillslope sediment delivery. First, rainfall erosivity is assumed constant through time. However, climatic variations play a minor role in sediment fluxes on Holocene timescales (e.g. Notebaert et al., 2011b). Second, our interpretation of what is considered anthropogenic land cover, that is, a combination of cropland and pasture, is rather simplistic. Depending on the studied region, a more diverse landscape with a variety of intermediate land-cover types such as meadow, fallow fields, villages, woodlots, and grazed forests, is probably more realistic (Gregg, 1988). Also, the C factors implemented in this study are based on contemporary land-cover characteristics and are not necessarily representative of prehistoric extensive land use. Other potential sources of error include topography smoothing and exposure of deeper, more erodible soil horizons through time, resulting in higher LS and lower K factors, respectively, for prehistoric times (Peeters et al., 2008; Rommens et al., 2007). However, when working with a 100-m DEM, changes in topography throughout the Holocene are expected to be small compared to the inaccuracies related to the modern SRTM data (Verstraeten, 2006). Furthermore, in contrast to, for example, Mediterranean environments (Dusar et al., 2012), sediment redistribution rates in our study area are not significantly influenced by Holocene erosion histories due to the thick and homogeneous soil profiles typical to the Loess Belt. Finally, land-cover allocation could be improved. Whereas archaeological research and more recent observations suggest that settlement and cultivation strategies are sensitive to changes in environmental, socioeconomic, and technological characteristics (e.g. Bakker et al., 2011; Kohler and Parker, 1986), land-cover suitability assessment in the present study is solely based on modern landscapes and hence assumes constant drivers of land-cover conversions through time. Moreover, the fragmented and heterogeneous character that results from the MOLA allocation procedure is expected to underestimate hillslope sediment delivery. Comparison of allocated model results with contemporary sediment yield calculations for the Scheldt catchment based on the CORINE land-cover map, in which urban land is replaced with forest, confirms this assumption. With the same areas of cropland, pasture, and forest, our allocated map results in an annual hillslope sediment delivery of 1.04 Mt/yr compared to a value of 1.82 Mt/yr when WaTEM/SEDEM is ceteris paribus applied with CORINE land-cover input. This points to the need for a better algorithm to assign land-cover areas to pixels, taking into account neighborhood and autocorrelation effects. Still, allocated model results yield more realistic results than the non-allocated maps, which simulate a sediment delivery value of 3.98 Mt/yr for the same land-cover areas.

Impact of land-cover resolution on geomorphic model results

Modeled annual hillslope sediment delivery, erosion, and deposition based on non-allocated land-cover maps are for all dates and cropland/pasture ratios much larger than when allocated maps are used in WaTEM/SEDEM. Several factors explain this. Since no information is available on individual land-cover units in the low-resolution maps, there is no differentiation of the transport capacity coefficient K_TC between more and less erosion-prone vegetation types, leading to high transport capacities over the entire study area. Moreover, the lack of information on individual patches of vegetation implies that land cover in the Scheldt catchment is considered homogeneous by WaTEM/SEDEM, resulting in unrealistically high LS factors that lead to overestimation of gross erosion in Eq. 1 (Van Oost et al., 2000). Combination of these elements, together with the averaging of the C factor over each original 5′ pixel of the KK10 data set, causes increased sediment flux connectivity between hillslopes and the river system in non-allocated models, regardless of the considered date and scenario. This also explains why HSDRs based on these maps are relatively high and constant.

When low-resolution land-cover maps are used in a recalibrated model, modeled sediment fluxes are lower due to the K_TC value of 10 m. Moreover, decreased transport capacity leads to a transport-limited system at high proportions of cropland, which results in constant erosion, deposition, and delivery. However, landscape connectivity remains high, and therefore, absolute values are still higher than for allocated models.

High-resolution land-cover maps, on the other hand, are characterized by a patchwork of cropland, pasture, and forest parcels, creating a heterogeneous landscape (Figure 2). Modeled annual erosion for allocated maps is limited due to the presence of parcel boundaries between individual vegetation patches, and the preferential allocation of cropland on weak slopes, which both imply relatively low LS factors (Van Oost et al., 2000). Furthermore, as a result of their low C factor and transport capacity coefficient K_TC, both forest and pasture parcels act as buffers preventing downward transport of sediment eroded upslope, and hence increase deposition at field boundaries (Beuselinck et al., 2000) and at the same time limit the connectivity between the hillslopes and the river network. This buffer effect plays an important role in the modeled sediment dynamics, since cropland is preferentially allocated on well-drained, weak slopes, that is, on the plateaus, while forest and pasture are located on steep slopes and close to the river, respectively. Connectivity, and hence hillslope sediment delivery, increases as vegetation within the study area is further converted to cropland. HSDRs hence take a range of values, depending on the amount of anthropogenic land cover and the cropland/pasture ratio.

Impact of cropland to pasture ratio on geomorphic model results

The sensitivity of modeled erosion, deposition, and hillslope sediment delivery to the cropland/pasture ratio can be even larger than the effect of the land-cover map’s spatial resolution and the proportion of anthropogenic land cover (Figure 3). The large discrepancies between scenarios result from the difference in the C factor, and in the case of allocated models, the K_TC coefficient of cropland and pasture. Furthermore, the location of cropland area also plays an important role when high-resolution land-cover maps are used in WaTEM/SEDEM.

In the non-allocated model, the average C factor within each original 5′ cell increases as the area under cropland expands, thereby inducing a rise in gross erosion according to Eq. 1. Since the transport capacity coefficient K_TC and hence the connectivity is invariably high in non-allocated models, a large part of the produced sediment is transported to the river network, resulting in higher hillslope sediment delivery and net erosion. Deposition occurs when the sediments eroded upslope exceed the transport capacity TC, and is also positively correlated to gross erosion and hence to the C factor. However, as the proportion of cropland within the study area further increases, the catchment becomes increasingly transport-limited as the transport capacity is reached in more and more pixels, higher upslope. Consequently, each additional increase in C factor results in only a marginal increase of modeled hillslope sediment delivery, net erosion, and deposition. This weakening increase of sediment redistribution with cropland area is illustrated in Figure 5b.

Figure 5.

Modeled annual erosion, hillslope sediment delivery and deposition based on (a) allocated and (b) non-allocated land-cover maps, as a function of the proportion of cropland within the study area. The remaining proportion is considered forest.

Modeling results with non-allocated land-cover input after recalibration are similar to the modeled erosion, delivery and deposition shown in Figure 5b; however, they have lower absolute values and become fully transport-limited at cropland proportions above 60%. HSDRs after recalibration are slightly lower than in the low-resolution model, due to the lower transport capacity coefficient.

Also in allocated models, sediment fluxes are mainly controlled by the proportion of cropland within the study area, and less by pasture area. In the allocation procedure, expansion of cropland takes place on increasingly unsuitable, that is, steeper, and hence, more erodible land. Also the size, contained in the LS factor, of individual cropland patches increases. As a result, erosion rises more than linearly with cropland proportion (Figure 5a). Moreover, as forest and pasture on the hillslopes and near the river are converted to cropland, connectivity within the catchment drastically increases due to the closer distance to the river network as well as the disappearance of buffer patches with low transport capacity where sediment is deposited. This explains the progressive increase of hillslope sediment delivery, while deposition increases approximately linearly. In our study area, the cultivation of the erodible hillslopes and the disappearance of buffers are marked by a threshold in the relation when the cropland proportion exceeds 50–60% (Figure 5a).

The presence of such a geomorphologic threshold has also been observed in field-based studies, and is reflected in the HSDR (e.g. Houben, 2012; Notebaert et al., 2011a; Van Rompaey et al., 2002; Verstraeten et al., 2009b). Indeed, in the case of allocated models, HSDR is highly sensitive to the cropland area and its location (Figure 4). When land cover in the catchment consists entirely of forest and pasture, sediment supply is limited and the transport capacity, although low, is not exceeded. Moreover, pasture is preferentially allocated near the river. Hence, most of the eroded sediment reaches the river system, resulting in a high HSDR. When cropland area and erosion increase, transport capacity becomes the limiting factor and part of the eroded sediment is deposited as colluvium. Further expansion of cropland until the threshold is exceeded leads to increased erosion and delivery, resulting again in a high HSDR.

Implications

Whereas application of low-resolution land-cover input in WaTEM/SEDEM leads to overestimated soil erosion and sediment delivery rates, spatial allocation of land-cover information to a high-resolution grid results in more heterogeneous landscapes with decreased connectivity that are better suited to simulate sediment redistribution processes. Recalibration of WaTEM/SEDEM for use with low-resolution land-cover input mitigates overestimation of sediment fluxes but also fails to simulate the observed non-linearity in the relation between cropland area and hillslope sediment delivery – and even aggravates it. However, spatial allocation of KK10 land-cover information to a 100-m grid for application in a geomorphic model is probably not computationally feasible at a global scale and at annual temporal resolution for the period between 8000 and 100 BP.

The varying response of allocated and non-allocated models to changing cropland proportions implies that the difference in modeled soil erosion, deposition, and sediment delivery between both types of model runs depends on the cropland/pasture ratio and on the total amount of anthropogenic land cover, or the considered date. This dependence on cropland proportion complicates the application of a correction on non-allocated model output. Especially for the purpose of long-term sediment redistribution modeling, accurate information on land-cover proportions is essential to minimize errors when annual sediment fluxes are integrated over thousands of years. Indeed, modeled cumulative hillslope sediment delivery within the Scheldt catchment since 8000 BP varies with an order of magnitude depending on the choice of scenario (Figure 6). Archaeological, palynological, and macrobotanical studies may provide valuable information to help increase the thematic resolution of Holocene land-cover reconstructions.

Figure 6.

Modeled cumulative hillslope sediment delivery between 8000 and 100 BP based on (a) allocated and (b) non-allocated land-cover input, for five scenarios of the cropland/pasture ratio.

However, even if detailed information on anthropogenic land-cover types and their respective proportions is available, modeled sediment redistribution still depends on the spatial distribution of land-cover parcels at subcatchment scale, as well as on other catchment characteristics. Consequently, scaling relations between non-allocated and allocated model results would need empirical calibration and are not universally applicable. Further research on the application of low-resolution land cover in spatially distributed geomorphic models should hence focus on the incorporation of a land-cover heterogeneity or connectivity measure in erosion and transport capacity calculations. Previous studies that attempted to model sediment delivery at regional scales, with varying degrees of success, suggested to introduce an empirical, spatially variable HSDR function at pixel or subcatchment scale instead of a transport capacity (e.g. Van Dijk and Kwaad, 1999; Vigiak et al., 2012), while others used a spatially lumped approach (Walling, 1983). A major drawback of the latter method, whether or not it includes land-cover information, is that it fails to model spatial variability in sediment delivery processes.

Conclusion

Global reconstructions of Holocene land cover may constitute a powerful means for the quantification of long-term anthropogenic impact on soil erosion and sediment redistribution through a geomorphic modeling approach. Model runs with land-cover input at different spatial resolutions and subsequent evaluation demonstrated that global low-resolution, non-allocated maps lack spatial detail and heterogeneity to simulate sediment fluxes within the Scheldt catchment. Overestimation of connectivity between the hillslopes and the fluvial system leads to unrealistically high sediment redistribution rates and complete transport-limitedness as deforestation increases, which cannot be solved through model recalibration. In contrast, spatially allocated land-cover maps at 100 m resolution do better capture the scale at which geomorphic processes operate, and yield good estimates of annual soil erosion and hillslope sediment delivery provided that land-cover proportions are accurate. Particularly, the cropland area and its location within the landscape control sediment fluxes from the hillslopes to the river. Hence, to accurately model Holocene soil erosion and sediment redistribution based on land-cover reconstructions, precise allocated and differentiated land-cover information at spatial scales characteristic for geomorphic processes is indispensable. For medium- to large-scale catchments with thick loess soils situated in a temperate climate, other model uncertainties, for example, resulting from temporal variability in climatic, topographic, and soil factors, affect model results at a spatial resolution of 100 m minimally compared to the large range of cumulative hillslope sediment delivery estimates caused by differing spatial resolutions and land-cover proportions.

Footnotes

Acknowledgements

The authors would like to thank Dr Jed Kaplan and Dr Kristen Krumhardt for providing the KK10 land-cover reconstruction data set used in this study.

Funding

This work was supported by the KU Leuven Center for Archaeological Sciences, and the Interuniversity Attraction Poles program initiated by the Belgian Science Policy Office (grant number P7/09).

References

Baartman

Temme

Schoorl

. (2012) Did tillage erosion play a role in millennial scale landscape development? Earth Surface Processes and Landforms 37: 1615–1626.

Bakels

(1992) The botanical shadow of two early Neolithic settlements in Belgium: Carbonized seeds and disturbances in a pollen record. Review of Palaeobotany and Palynology 73: 1–19.

Bakker

Hatna

Kuhlman

. (2011) Changing environmental characteristics of European cropland. Agricultural Systems 104: 522–532.

Bertran

(2004) Soil erosion in small catchments of the Quercy region (southwestern France) during the Holocene. The Holocene 14: 597–606.

Beuselinck

Steegen

Govers

. (2000) Characteristics of sediment deposits formed by intense rainfall events in small catchments in the Belgian Loam Belt. Geomorphology 32: 69–82.

Bossard

Feranec

Otahel

(2000) CORINE Land Cover Technical Guide – Addendum 2000. Technical Report No. 40, 29 May. Copenhagen: European Environment Agency.

Boyle

Gaillard

Kaplan

. (2011) Modelling prehistoric land use and carbon budgets: A critical review. The Holocene 21: 715–722.

Broothaerts

Verstraeten

Notebaert

. (2013) Sensitivity of floodplain geoecology to human impact – A Holocene perspective for the headwaters of the Dijle catchment, central Belgium. The Holocene 23: 1403–1414.

Cauwe

Vander Linden

Vanmontfort

(2001) The middle and late Neolithic. In: Cauwe

Hauzeur

Van Berg

P-L

(eds) Prehistory in Belgium. Anthropologica et Praehistorica 112: 77–89.

10.

Collins

DBG

Bras

Tucker

(2004) Modeling the effects of vegetation-erosion coupling on landscape evolution. Journal of Geophysical Research 109.

11.

Coulthard

Macklin

Kirkby

(2002) A cellular model of Holocene upland river basin and alluvial fan evolution. Earth Surface Processes and Landforms 27: 269–288.

12.

De Moor

JJW

Verstraeten

(2008) Alluvial and colluvial sediment storage in the Geul River catchment (The Netherlands) – Combining field and modelling data to construct a late Holocene sediment budget. Geomorphology 95: 487–503.

13.

De Moor

JJW

Kasse

van Balen

. (2008) Human and climate impact on catchment development during the Holocene – Geul River, The Netherlands. Geomorphology 98: 316–339.

14.

De Smedt

(1973) Paleogeografie en kwartair-geologie van het confluentiegebied Dijle-Demer. Acta Geographica Lovaniensia 11: 1–141.

15.

Desmet

PJJ

Govers

(1996) A GIS procedure for automatically calculating the USLE LS factor on topographically complex landscape units. Journal of Soil and Water Conservation 51: 427–433.

16.

Dikau

(1989) The application of a digital relief model to landform analysis. In: Raper

(ed.) Three Dimensional Applications in Geographical Information Systems. London: Taylor & Francis, pp. 51–77.

17.

Dotterweich

(2008) The history of soil erosion and fluvial deposits in small catchments of central Europe: Deciphering the long-term interaction between humans and the environment – A review. Geomorphology 101: 192–208.

18.

Dusar

Verstraeten

D’Haen

. (2012) Sensitivity of the eastern Mediterranean geomorphic system towards environmental change during the late Holocene: A chronological perspective. Journal of Quaternary Science 27: 371–382.

19.

Eastman

Jin

Kyem

PAK

. (1995) Raster procedures for multi-criteria/multi-objective decisions. Photogrammetric Engineering and Remote Sensing 61: 539–547.

20.

Fuchs

Will

Kunert

. (2011) The temporal and spatial quantification of Holocene sediment dynamics in a meso-scale catchment in northern Bavaria, Germany. The Holocene 21: 1093–1104.

21.

Giosan

Coolen

Kaplan

. (2012) Early anthropogenic transformation of the Danube-Black Sea system. Scientific Reports 2: 582.

22.

Gregg

(1988) Foragers and Farmers: Population Interaction and Agricultural Expansion in Prehistoric Europe (Prehistoric Archaeology and Ecology Series). Chicago, IL: The University of Chicago Press.

23.

Guzzetti

Carrara

Cardinali

. (1999) Landslide hazard evaluation: A review of current techniques and their application in a multi-scale study, central Italy. Geomorphology 31: 181–216.

24.

Hoffmann

Lang

Dikau

(2008) Holocene river activity: Analysing ¹⁴C-dated fluvial and colluvial sediments from Germany. Quaternary Science Reviews 27: 2031–2040.

25.

Houben

(2012) Sediment budget for five millennia of tillage in the Rockenberg catchment (Wetterau loess basin, Germany). Quaternary Science Reviews 52: 12–23.

26.

Kaplan

Krumhardt

Ellis

. (2011) Holocene carbon emissions as a result of anthropogenic land cover change. The Holocene 21: 775–791.

27.

Keating

Cherry

(2004) Use and interpretation of logistic regression in habitat-selection studies. Journal of Wildlife Management 68: 774–789.

28.

Klein Goldewijk

Beusen

Van Drecht

. (2011) The HYDE 3.1 spatially explicit database of human-induced global land-use change over the past 12,000 years. Global Ecology and Biogeography 20: 73–86.

29.

Kohler

Parker

(1986) Predictive models for archaeological resource location. Advances in Archaeological Method and Theory 9: 397–452.

30.

Lal

(2005) Soil erosion and carbon dynamics. Soil and Tillage Research 81: 137–142.

31.

Lang

Bork

H-R

Mäckel

. (2003) Changes in sediment flux and storage within a fluvial system: Some examples from the Rhine catchment. Hydrological Processes 17: 3321–3334.

32.

Lemmen

(2009) World distribution of land cover changes during Pre- and Protohistoric Times and estimation of induced carbon releases. Géomorphologie: Relief, Processus, Environnement 4: 303–312.

33.

Lespez

(2003) Geomorphic responses to long-term land use changes in Eastern Macedonia (Greece). Catena 51: 181–208.

34.

Macaire

J-J

Bellemlih

Di-Giovanni

. (2002) Sediment yield and storage variations in the Négron River catchment (south western Parisian Basin, France) during the Holocene period. Earth Surface Processes and Landforms 27: 991–1009.

35.

Morgan

Lesh

(2005) Developing landform maps using Esri’s ModelBuilder. In: 25th Esri International User Conference Proceedings, San Diego, CA, 25–29 July. Available at: http://proceedings.esriyom/library/userconf/proc05/papers/pap2206.pdf.

36.

Mullenders

Gullentops

(1957) Palynologisch en geologisch onderzoek in de alluviale vlakte van de Dijle, te Heverlee-Leuven. Agricultura 5: 57–64.

37.

Mullenders

Gullentops

Lorent

. (1966) Le remblement de la vallée de la Nethen. Acta Geographica Lovaniensia 4: 169–181.

38.

Nash

Sutcliffe

(1970) River flow forecasting through conceptual models part I – A discussion of principles. Journal of Hydrology 10: 282–290.

39.

Nearing

Jetten

Baffaut

. (2005) Modeling response of soil erosion and runoff to changes in precipitation and cover. Catena 61: 131–154.

40.

Notebaert

Verstraeten

(2010) Sensitivity of West and Central European river systems to environmental changes during the Holocene: A review. Earth-Science Reviews 103: 163–182.

41.

Notebaert

Verstraeten

Vandenberghe

. (2011a) Changing hillslope and fluvial Holocene sediment dynamics in a Belgian loess catchment. Journal of Quaternary Science 26: 44–58.

42.

Notebaert

Verstraeten

Ward

. (2011b) Modeling the sensitivity of sediment and water runoff dynamics to Holocene climate and land use changes at the catchment scale. Geomorphology 126: 18–31.

43.

Oldfield

Dearing

(2003) The role of human activities in past environmental change. In: Alverson

Bradley

Pedersen

(eds) Paleoclimate, Global Change and the Future. Berlin: Springer Verlag, pp. 143–162.

44.

Olofsson

Hickler

(2007) Effects of human land-use on the global carbon cycle during the last 6,000 years. Vegetation History and Archaeobotany 17: 605–615.

45.

Passmore

Macklin

(2001) Late Holocene alluvial sediment budgets in an upland gravel bed river: The River South Tyne, northern England. In: Maddy

Macklin

Woodward

(eds) River Basin Sediment Systems: Archives of Environmental Change. Rotterdam: Balkema, pp. 423–444.

46.

Peeters

Van Oost

Govers

. (2008) The compatibility of erosion data at different temporal scales. Earth and Planetary Science Letters 265: 138–152.

47.

Pongratz

Reick

Raddatz

. (2008) A reconstruction of global agricultural areas and land cover for the last millennium. Global Biogeochemical Cycles 22: GB3018.

48.

Pontius

Schneider

(2001) Land-cover change model validation by an ROC method for the Ipswich watershed, Massachusetts, USA. Agriculture, Ecosystems & Environment 85: 239–248.

49.

Price

(2000) Europe’s First Farmers. Cambridge: Cambridge University Press.

50.

Rommens

Verstraeten

Peeters

. (2007) Reconstruction of late-Holocene slope and dry valley sediment dynamics in a Belgian loess environment. The Holocene 17: 777–788.

51.

Rommens

Verstraeten

Poesen

. (2005) Soil erosion and sediment deposition in the Belgian loess belt during the Holocene: Establishing a sediment budget for a small agricultural catchment. The Holocene 15: 1032–1043.

52.

Ruddiman

(2003) The anthropogenic greenhouse era began thousands of years ago. Climatic Change 61: 261–293.

53.

Schoorl

Sonneveld

Veldkamp

(2000) Three-dimensional landscape process modelling: The effect of DEM resolution. Earth Surface Processes and Landforms 25: 1025–1034.

54.

Temme

AJAM

Verburg

(2011) Mapping and modelling of changes in agricultural intensity in Europe. Agriculture, Ecosystems & Environment 140: 46–56.

55.

Temme

AJAM

Peeters

Buis

. (2011) Comparing landscape evolution models with quantitative field data at the millennial time scale in the Belgian loess belt. Earth Surface Processes and Landforms 36: 1300–1312.

56.

Trimble

(1999) Decreased rates of alluvial sediment storage in the Coon Creek Basin, Wisconsin, 1975–93. Science 285: 1244–1246.

57.

Van Berg

P-L

Hauzeur

(2001) Le Néolithique ancien. In: Cauwe

Hauzeur

Van Berg

P-L

(eds) Prehistory in Belgium. Anthropologica et Praehistorica 112: 63–76.

58.

Van Den Eeckhaut

Vanwalleghem

Poesen

. (2006) Prediction of landslide susceptibility using rare events logistic regression: A case-study in the Flemish Ardennes (Belgium). Geomorphology 76: 392–410.

59.

Van Dessel

Van Rompaey

Szilassi

(2011) Sensitivity analysis of logistic regression parameterization for land use and land cover probability estimation. International Journal of Geographical Information Science 25: 489–508.

60.

Van Dijk

Kwaad

(1999) Estimation of Suspended Sediment Supply to the Stream Network of the River Rhine under Present-Day Climate and Land Use. Global Change NOP-NRP Report 410200027, 1 December. Amsterdam: Interuniversitair Centrum voor Geo-ecologisch Onderzoek.

61.

Van Oost

Govers

Desmet

(2000) Evaluating the effects of changes in landscape structure on soil erosion by water and tillage. Landscape Ecology 15: 577–589.

62.

Van Oost

Quine

Govers

. (2007) The impact of agricultural soil erosion on the global carbon cycle. Science 318: 626–629.

63.

Van Oost

Verstraeten

Doetterl

. (2012) Legacy of human-induced C erosion and burial on soil-atmosphere C exchange. Proceedings of the National Academy of Sciences of the United States of America 109: 19492–19497.

64.

Van Rompaey

Govers

Puttemans

(2002) Modelling land use changes and their impact on soil erosion and sediment supply to rivers. Earth Surface Processes and Landforms 27: 481–494.

65.

Van Rompaey

Govers

Verstraeten

. (2003) Modelling the geomorphic response to land use changes. In: Lang

Hennrich

Dikau

(eds) Long Term Hillslope and Fluvial System Modelling: Concepts and Case Studies from the Rhine River Catchment (Lecture Notes in Earth Sciences). Berlin: Springer Verlag, pp. 73–100.

66.

Van Rompaey

Verstraeten

Van Oost

. (2001) Modelling mean annual sediment yield using a distributed approach. Earth Surface Processes and Landforms 26: 1221–1236.

67.

Vanwalleghem

Verheyen

Hermy

. (2004) Legacies of Roman land-use in the present-day vegetation in Meerdaal Forest (Belgium)? Belgian Journal of Botany 137: 181–187.

68.

Verbruggen

Van Dongen

(1976) De geokronologie van het Post-pleniglaciaal in Zandig-Vlaanderen op basis van pollenanalyse en 14C-onderzoek. Natuurwetenschappelijk Tijdschrift 58: 233–256.

69.

Verstraeten

(2006) Regional scale modelling of hillslope sediment delivery with SRTM elevation data. Geomorphology 81: 128–140.

70.

Verstraeten

Lang

Houben

(2009a) Human impact on sediment dynamics – Quantification and timing. Catena 77: 77–80.

71.

Verstraeten

Rommens

Peeters

. (2009b) A temporarily changing Holocene sediment budget for a loess-covered catchment (central Belgium). Geomorphology 108: 24–34.

72.

Verstraeten

Van Oost

Van Rompaey

. (2002) Evaluating an integrated approach to catchment management to reduce soil loss and sediment pollution through modelling. Soil Use and Management 18: 386–394.

73.

Vigiak

Borselli

Newham

. (2012) Comparison of conceptual landscape metrics to define hillslope-scale sediment delivery ratio. Geomorphology 138: 74–88.

74.

Walling

(1983) The sediment delivery problem. Journal of Hydrology 65: 209–237.

75.

Ward

van Balen

Verstraeten

. (2009) The impact of land use and climate change on late Holocene and future suspended sediment yield of the Meuse catchment. Geomorphology 103: 389–400.

76.

Wischmeier

Smith

(1978) Predicting Rainfall Erosion Losses – A Guide to Conservation Planning (Agriculture Handbook No. 537). Washington, DC: US Department of Agriculture.