Catena 113 (2014) 24–40

Contents lists available at ScienceDirect

Catena

j ourna l homepage: www.e lsev ie r .com/ locate /catena

Soil erosion in gully catchments affected by land-levelling measures inthe Souss Basin, Morocco, analysed by rainfall simulation and UAVremote sensing data

Klaus Daniel Peter a,⁎, Sebastian d'Oleire-Oltmanns b, Johannes B. Ries a, Irene Marzolff b, Ali Ait Hssaine c

a Physical Geography, University of Trier, D-54296 Trier, Germanyb Remote Sensing & GIS Research Group, Department of Physical Geography, Goethe University Frankfurt am Main, D-60438 Frankfurt a.M., Germanyc Development and Geoenvironment of Arid and Semi-Arid Lands, Department of Geography, University Ibn Zohr Agadir, Morocco

⁎ Corresponding author at: Universität Trier, BehriTel.: +49 6512014545; fax: +49 6512013976.

E-mail addresses: peterk@uni-trier.de (K.D. Peter), do(S. d'Oleire-Oltmanns), riesj@uni-trier.de (J.B. Ries), marz(I. Marzolff), Aithssaine55@gmail.com (A. Ait Hssaine).

0341-8162/$ – see front matter © 2013 Elsevier B.V. All rihttp://dx.doi.org/10.1016/j.catena.2013.09.004

a b s t r a c t

a r t i c l e i n f o

Article history:Received 7 September 2012Received in revised form 13 September 2013Accepted 16 September 2013Available online xxxx

Keywords:Soil erosionLand levellingRainfall simulationUAVSFAPGully

Land-levelling measures are widely used in the Souss valley, SouthMorocco, for the implementation of land usechange. However, their impact on soil erosion in this region is unclear. This paper presents the approach ofcombining punctual process analysis through experimental rainfall simulation and gully mapping as well asvolume quantification analysing on a local scale using unmanned aerial vehicle (UAV) remote sensing data.Thus, the influence of the impacts of land levelling in the catchment area on the linear soil erosion by gulliescan be elucidated. Soil surface characteristics, modified by land levelling, lead to higher runoff generation andsediment production. Mean runoff coefficients from 54% to 58% are found in levelled study sites, and 38% to47% are found in undisturbed areas. Mean sediment loads reach 48.6 g m−2 to 81 g m−2 under the influenceof levelling, but only 9.3 g m−2 to 23.7 g m−2 without it. Quantification of soil erosion by UAV data showedthat a gully in a levelled study site eroded about 720 m3 of soil within only one rain period. The surface of thecatchment area was lowered 0.054 m on average due to land levelling, whereas in study sites without landlevelling, the monitoring did not show significant differences of shape and extent of the gullies at two differentpoints in time. The strong influence of land levelling can be documented with the connection of these twomethods. A clear amplification of soil erosion is induced by land-levelling measures.

© 2013 Elsevier B.V. All rights reserved.

1. Introduction

The Souss valley, South Morocco, is characterized by ongoingdynamic land-use change with transformations from traditional agri-culture to vast agro-industrial plantations of citrus fruits, bananas andvegetables. These plantations, as well as other arable land, are threat-ened by gully and other forms of soil erosion triggered by heavy rainfallevents. In this region, land-levelling measures are used for the imple-mentation of the land-use change. They are usually employed for thereclamation of severely eroded areas, including gullies and badlandsand for the removal of unwanted vegetation (Borselli et al., 2006).However, land-levelling measures also lead to disturbances in theenvironment, thus to the degradation or alteration of soil properties(Borselli et al., 2006; Cots-Folch et al., 2006; Lundekvam et al., 2003).Due to the reduction of vegetation cover, the bare soil is left vulnerablefor intense rainfall events. The progressive loss of soil material, a

ngstr., 54296 Trier, Germany.

leire@em.uni-frankfurt.deolff@em.uni-frankfurt.de

ghts reserved.

reduction in organic matter content and effective soil depth, calciumcarbonate enrichment of arable layers and degradation of soil structureis found in the study by Martínez-Casasnovas and Ramos (2009) thatanalyses the impact of land-levelling measures on vineyards in NorthEast Spain. Only few studies refer to land levelling and terracing inEuropean agriculture, but associated problems and impacts have notbeen widely studied (Cots-Folch et al., 2006). Peter and Ries (2013)investigate the impact of land-levelling measures on infiltration ratesin South Morocco. They found 3.5 times higher infiltration rates onnon-levelled than on levelled study sites. However, there is still littleinformation about the effects of land-levelling measures in arid andsemi-arid Northern Africa. Therefore three main questions arise thatneed to be answered: (1) the impact of land levelling on soil surfacecharacteristics, (2) the influence of land levelling on soil erosion and(3) the dimension of gully erosion in levelled areas.

A recent approach for the investigation of soil erosion on land-levelling influenced areas in southern Morocco is presented in thisarticle: Combining punctual process analysis through experimentalrainfall simulation and aerial surveying using an unmanned aerialvehicle (UAV) for gullymapping andquantification ofminimumerosionrates. Thus, it is possible to clarify the relationship of soil erosion in thegully catchment area to the linear erosion of gullies themselves. Rainfall

25K.D. Peter et al. / Catena 113 (2014) 24–40

simulations are widely used for analysis concerning runoff generationand sediment erosion in different landscape units and elements(Cerdà et al., 1998; Dunkerly, 2012), as well as for different stages ofland abandonment and land-use change (Lasanta et al., 2000; Nadal-Romero et al., 2011; Ries et al., 2003). So they are applicable for thegully catchments in this fast changing region of the Souss. The soil sur-face characteristics and thus the influencing factors for erosion such assoil crusts and vegetation cover can be studied with this method (Huet al., 2012;Mayor et al., 2009; Ries, 2010; Ries andHirt, 2008). Measur-ing the linear erosion of gullies by aerial surveying campaign is aprevalent approach (Aber et al., 2010; d'Oleire-Oltmanns et al., 2012;Marzolff and Poesen, 2009; Ries and Marzolff, 2003). Analysing the

Fig. 1. Souss catchment in South Morocco (left; own illustration; Basemap: USGS HydroSHEDSof Oued Irguitène. The Souss river bed runs along the lower edge of the study area; the cityinterpretation of the references to colour in this figure legend, the reader is referred to the we

linked information gained through both methods is the aim of thefollowing article.

2. Materials and methods

2.1. Study area

2.1.1. Souss basinThe study area is located in the centre of the Souss valley in South

Morocco near the city of Taroudannt (Fig. 1). The Souss basin extendsfrom 30 to 31° northern latitude and from 7 to 9° western longitude. Itis drained by the river Souss. Seven test sites, La Glalcha (GLA), El

). Investigated study sites (red rectangles) within the whole study area on the alluvial fanof Taroudannt is located near the lower right edge (right; Quickbird 2009/10/26). (Forb version of this article.)

Fig. 2. Typical raw soil (here in HAM) with different layers of substrate, freshly cut afterrainfall event.

26 K.D. Peter et al. / Catena 113 (2014) 24–40

Houmer (HOU), Gchechda (GCH), Talaa (TAL), Lastah (LAS), onebetween Lastah and Hamar (LAM) and Hamar (HAM), were chosenon an alluvial fan which is discharging from the High Atlas Mountainsinto the plain (Fig. 1).

The High Atlas Mountains with Palaeozoic, Mesozoic and Cenozoicrocks in the North and the Anti-Atlas Mountains with Precambrianand Palaeozoic rocks in the South build up the frame around the valley.There is a complex interconnection of both Atlasic domains beneath theSouss valleywhichbuilds up the alluvial depressionof the Souss basin. Itis filled by thick Pliocene and Quaternary deposits which overlay aCretaceous–Eocene succession, including significant fluvial–lacustrinesequences (Dijon, 1969). Loamy Quaternary alluvial fans and terracescover the surface. Ait Hssaine and Bridgland (2009) give a precisesequence of Pliocene–Quaternary material deposition. An all-seasonnegative water balance exists due to the precipitation of only around200 mm and a mean annual temperature of 20 °C. The arid climate ischaracterized by a hot dry summer and a mild winter with short rainperiods showing high variability. Soils are mostly immature due torecent accumulation and little pedogenesis, which is caused by hotand dry climatic conditions. High proportions of fine sand, silt and clayare induced by colluvial and alluvial formation. Ghanem (1974)mapped three different types of soils in the Taroudannt area using theFrench soil classification (“référentiel pédologique français”). Hefound “sols à pédogenèse externe”, which are underlying strongmorphodynamic processes. Fluvial erosion and accumulation constrictpedogenetic development. These mostly alluvially accumulated, xericsoils have little content of humus but high contents of silt and sand, aswell as local pebbles. In context of the World Reference Base for SoilResources (WRB), these soils can be classified as Fluvisols. They aremostly found along the wadis, which are dissecting the Taroudanntarea. This is also mostly the location where our test sites are situated.The second soil group classified by Ghanem (1974) is the “solsholocalcaires xériques”. From this group, he mapped crusted Regosols(WRB) in our test area, which are silty–sandy and calcareous. Theseraw soils are also characterized by the dry climatic conditions. Thethick layer of substrate in the initial state of soil formation is prone tosealing and crusting. This type of soil is mostly found in the lessdisturbed areas between the wadis. The last soil type is a mixture ofthe first type and modest Regosols (WRB) which are profound witha mixed proportion of silt, sand and clay. They are called “solspolygéniques, polyphasés et association de sols” and are mostly foundin the southern part of the study area. Similar allocation of soil typeswas done by Tagma (2011), El Ghannouchi (2007) and Watteeuw(1964). Also our own studies confirm the statements of the formerauthors. We found in all study sites mostly profound raw soils withmixed substrate and little pedogenesis. The strong morphodynamicsof this region are continuously reshaping the substrate (Fig. 2).

Vegetation cover is determined by subtropic, desert andMediterranean species. Argania spinosa, Acacia, Euphorbia, Artemisiaherba-alba and Ziziphus are typical for the widespread bush and shrubsteppe. Gramineous species are Dactylus glomerata, Cynodo dactylonand Andropogon hirtus. Tamarix, Salicornia and Salsola are only foundin years with high precipitation. In the Souss valley, a highly dynamicland-use change is going on, accompanied by labour migration. Nowa-days, land use is dominated by citrus fruit and banana plantationswhich are irrigated by deep wells (Fig. 1). There is further cultivationof vegetables, cereals and small remains of the traditional small-areamixed agriculture. The sedimentary fans of the Souss valley are heavilydissected by gully erosion in their distal part, particularly in theTaroudannt area. The linear incision was triggered by the decline ofsugarcane cultivation at the end of the 17th century, when the near-complete degradation of the argan forests coincided with the abandon-ment of the agricultural fields (Ait Hssaine, 1994, 2002). Since this time,the Souss valley has been affected by soil erosion (Dijon, 1969). Theresulting badlands reach deep between the modern fruit tree planta-tions and irrigation areas. In order to increase the productive area,

currently it is attempted to restore them to cultivated land bybulldozing and land levelling. These measures are emerging recently,as heavy machines are easily available since finishing the highwayMarrakesh–Agadir. Hired for a couple of days, they are used to infillgullies with the surrounded material and to level the area in shorttime. However, not much value is set on soil consolidation, stabilisationand smoothing of the surface, and a distinct pattern of levelling furrowsby the bulldozer's grating action often remains. The same methods areused for the generation of new spaces in the periphery for buildingconstructions, as the population is still growing.

2.1.2. Study site TaroudanntAll seven study sites were chosen along a transect from South–West

to North–East through the same alluvial fan, which is named after thewadi Oued Irguitène (Fig. 1). The fan is controlled by similar conditions,accumulated material originates from the same source, a homogeneousgentle slope is found throughout the study area. The texture of soils isdiffering, there is a slightly decreasing of grain sizes from proximal todistal locations. Also, the treatment strategies on the study sites differ.We set the study sites on areas with land levelling, regular ploughing,fallow land and no treatment in recent years (Table 1). Each of thestudy sites is characterized by a gully system which traverses it. Thesegullies are in close vicinity to or even cut into arable land, plantationsof bananas and citrus fruits or housing areas. The fan originates in thefoothills of the High Atlas Mountains in the North of the Souss valleynear the village Bou Lajelat (Fig. 1). It reaches the wadi Oued Souss inthe South of Taroudannt along about 15 km of the riverbed. The alluvialfan is incised by the wadi Oued Ouaar into which many smaller wadis

Table 1Land use, treatment strategies and soil parameters (mean values) of different study sites.

Test site Main land use Treatment strategy pH-value Organic matter[%]

Bulk density[g cm−3]

Inorganic carbon[%]

Mean grain diameter[mm]

GLA Vegetable field (fallow) Levelling 8.61 5.11 – 1.61 0.08HOU Building lot/maize field/fallow land Levelling 8.29 5.88 1.59 1.62 0.08GCH Fallow land Non 7.94 3.77 1.49 1.36 0.11TAL Fallow land/cereal field/banana plantation Partly ploughing 7.72 3.83 1.46 1.62 0.14LAS Orange plantation/fallow land Non 7.98 5.81 1.40 1.94 0.13LAM Fallow land/cereal field Ploughing, partly levelling 7.93 4.16 1.46 2.26 0.16HAM Fallow land/cereal field Partly ploughing 7.63 2.78 1.50 1.98 0.19

27K.D. Peter et al. / Catena 113 (2014) 24–40

discharge. The whole area, but especially the southern part of thealluvial fan, underlies heavy gully erosion. As the different impacts ofland-levellingmeasures on soil erosion concerning runoff and sedimentrates will be investigated in this article, study sites were selected inlevelled and non-levelled areas underlying different land use. GLA andHOU present levelled study sites. They are located southernmost nearthe wadi Oued Ouaar. GLA was levelled once in the second half of2009 and again in spring 2011. In HOU, several levelling measurestook place since 2000, the southern part of the area was also ploughed.Following to north-eastern direction are the study sites GCH, TAL, LAS,LAM and HAM. They are all not levelled except of a small part of LAMwhich was levelled in autumn 2011. Land use and treatment strategiesas well as soil parameters can be found in Table 1.

2.2. Methods

For the investigation of the gully catchments and the gulliesthemselves, rainfall simulations and aerial survey campaigns are used.Rainfall simulations are widely used for the quantification of runoffand erosion processes (Cerdà et al., 1998; Ries, 2010). Comparison ofrunoff and erosion under different conditions such as varying land useor soil surface structures is possible. The reasons for applying aerial sur-vey campaigns are (1) mapping the study sites at a larger areal extentwith very high image resolution, (2) the photogrammetric analysis ofsmall-format aerial photographs (SFAPs) for quantification of actualerosion and (3) the monitoring of the long-term gully development. Inthis article the focus is set on photogrammetric analysis of SFAPs forquantification of actual erosion.

2.2.1. Rainfall simulationIn total, 122 rainfall simulations were conducted on the seven test

sites. The test plotswere selected systematically along the gully systemsof each study site to cover their catchment areas. In order to cover mostfeatures of the landscape, test plots were set on positions representativefor different land use, surface characteristics and treatment strategies inthese areas.

Various characteristics of surface types were analysed by the rainfallsimulations. The features of each test plot were registered descriptively.Some of the features could also be recorded quantitatively, e.g. forvegetation cover, physical crust cover or stone fragment cover (in % ofthe surface of the test plot). A visual estimation of these features wasdone in the entire area of each test plot. The grain size distribution ofthe fine earth was determined by means of the pipette method (Köhn,1929). On each of the seven different study sites, two to four soilsamples were taken of the 0.1 m topsoil. The location of sampling wasnear test plotswhich typify the study site; themean valuewas calculatedfor each site. Analysed particle size fractions are coarse sand = 2–0.63 mm, medium sand = 0.63–0.2 mm, fine sand = 0.2–0.063 mm,coarse silt = 0.063–0.02 mm, medium silt = 0.02–0.0063 mm, finesilt = 0.0063–0.002 mm and clay ≤ 0.002 mm (after BS EN ISO,14688-1: 2002). In total, 21 soil samples were examined.

A small portable nozzle rainfall simulator (Iserloh et al., 2012) basedon the one designed by Calvo et al. (1988) and Lasanta et al. (2000), wasused for the rainfall simulation experiments. The rainfall simulations

were carried out on circular plots with a size of 0.28 m2. Each of themis delimited by a 2 mm thin and0.07 mhigh steel ring,which is insertedinto the soil at least 0.03 m. Great carewas takenwith the installation ofthe ring to avoid disturbances of the soil. The outlet is V-shaped andplaced at the deepest point of the plot at surface level (Fister et al.,2011). A motor pump boosts the water regulated by a flow metre intothe commercial full cone nozzle (Lechler 460.608) at a height of 2 m.The rainfall intensity was maintained at about 40 mm h−1 for thewhole experiment, which has been used as a standard in the studygroup over the last 15 years (Peter and Ries, 2013; Ries and Langer,2002; Ries et al., 2000, 2009). Each rainfall simulation lasts for 30 minand is divided into six measuring intervals of 5 min each. Runoffwater and suspended sediment are collected in 0.5 L wide-openingPE-bottles for each interval separately. The duration from the beginningof the experiment until the start of runoff generation is recorded as wellas the exact times of exchanging the bottles when runoff exceeds thecapacity of the PE bottles (0.5 L) during the measuring intervals. Theamount of runoff water is calculated by subtracting the tare weight ofthe plastic bottles and the sedimentweight from theweight of the filledplastic bottles. Runoff amount in litres is assumed to be the same as thecalculated weight. Each bottle with the collected runoff water and thesuspended sediment is filtrated separately with circular fine-meshedfilter papers (Rotilabo® round filter type 15A, less than 2 μm mesh-width). The filters are dried to constant weight at 105 °C and weightedthereafter for determining suspended sediment load for eachmeasuredinterval (Iserloh et al., 2013).We used common tapwater with an elec-tric conductivity of around 720 μS cm−1 for the simulations.We did notobserve any soil dispersivity problems when using this water. We dried25 filtrates of the rainfall simulations to determine the weight ofdissolved solids. They did not vary significantly from the weight of thedissolved solids in the tap water (mean difference 0.01 g), neither didthe electric conductivity. Therefore dissolved solids were neglected.

2.2.2. Statistical analysisFor 122 rainfall simulations, data of runoff generation and sediment

production were summarised for the whole experiments of 30 min.Furthermore, runoff coefficients and sediment concentrations for eachrainfall simulationwere calculated. For a better understanding of factorson the plot influencing runoff and sediment generation, explorativedata analysis and correlation analysis were performed with the SPSS18.0 statistical package as well as with Sigma Plot 11.0. Spearman's-Rho correlation coefficient was chosen for correlation estimationbetween the quantitative characteristics of the soil surface and runoffand sediment production (Seeger, 2007). This rank-correlation methodis considered robust against outliers and non-normal distribution of thedata. Boxplot diagrams are used to display the various plot characteris-tics of the study sites as well as to clarify the differences of runoff andsediment generation in the different study sites. Each boxplot consistsof the upper and lower quartile, the median, minimum and maximumand outliers if necessary. As the data were not normally distributed, aKruskal–Wallis test was performed to analyse the differences amongthe test sites. An all pairwise multiple comparison procedure (Dunn'sMethod) gives the significance of the difference of one study site toeach of the others.

28 K.D. Peter et al. / Catena 113 (2014) 24–40

2.2.3. Aerial survey campaignsThe acquisition of SFAPs by survey campaigns using a fixed-wing

UAV enables photogrammetric analysis of aerial photographs finallyleading to quantification of actual erosion rates. The battery-poweredfixed-wing aircraft type Sirius I (MAVinci, Germany) is used for the sur-vey flights. The UAV is made from Elapor, has a wingspan of 1.63 m anda length of 1.20 m. In total the UAV weighs approx. 2.7 kg includingpayload. The UAV is hand-launched and controlled autonomously dur-ing take-off and flight by the autopilot software from MAVinci. Duringsurvey flights it follows predefined flight plans. After completing asurvey flight, the pilot is landing the UAV using a half-autonomousmode, the so-called assisted mode. The pilot is permanently supportedby the autopilot software during UAV control. This assisted flyingmode enables thepilot to safely steer and land the planewhile confiningthe navigation area to a predefined range. For difficult terrain, wherefully-autonomous landing may not be possible, this is highly useful(d'Oleire-Oltmanns et al., 2012).

As optical onboard-sensor a digital interchangeable lens systemcamera is installed (Panasonic Lumix GF1). The complete descriptionincluding all technical details of the applied hardware is given byd'Oleire-Oltmanns et al. (2012).

Aerial survey campaigns took place in autumn 2010 and in autumn2011. A total number of more than 17,500 SFAPs was acquired. In thisarticle the study site GLA (Fig. 1) was selected for detailed analysis. Itis an example of a dynamic location where land-levelling measuresare continuously taking place.

Ground control points (GCPs) were installed before image acquisi-tion took place. Each GCP was precisely measured with a total station(approx. 0.01–0.02 m precision) using a local coordinate system. Thisis required for georeferencing during the photogrammetric analysis.

Different flying heightswere applied for the surveying flights. Lowerflying heights result in higher image resolution values and smallerimage extents whereas higher flying heights lead to proportionaldecrease of image resolution while the image extents grow simulta-neously. For the analysis of aerial photographswith very high resolutionthe flying height was set at 70 m above ground. In order to achievesufficient resolution and simultaneously arrive at a large image extentfor the overview photographs, the flying height during the survey ofthe surrounding areas was set at 400 m above ground. The matter ofdifferent scales for image acquisition using a UAV has recently beendiscussed by d'Oleire-Oltmanns et al. (2012). The calculated resolutionof the images at a flying height of 70 m above ground is at 1.5 cm ×1.5 cm per pixel. This resolution at a flying height of 400 m aboveground still is at 8.6 cm × 8.6 cm per pixel.

2.2.4. Acquisition and analysis of the SFAPsDuring data acquisition the aerial photographs were saved on a

memory card within the camera. The original RAW file format wasconverted to TIFF format, and modest image enhancement regardingcontrast, saturation and brightness was applied. Leica PhotogrammetrySuite (LPS) was used for all photogrammetric analysis. In addition, GISsoftware was used for the quantification of erosion rates.

For the creation of photogrammetric image blocks of selected aerialphotographs acquired at a flying height of 70 m above ground, themeasured coordinateswere assigned to the GCPs on each image, addingadditional tie points manually where necessary and finally using anautomated tie point generation routine within LPS software. Theexterior orientation of each image was then established by bundle-block triangulation. From the resulting stereo models, a very detailedDigital Terrain Model (DTM) with a resolution of 0.05 m × 0.05 mwas created for the gully site (d'Oleire-Oltmanns et al., 2012; cf. Aberet al., 2010). The x/y/z errors of position of this precise DTM werebelow 0.01 m.

Using a similar workflow on aerial photographs acquired at a flyingheight of 400 m above ground (Fig. 11e), a second DTM was createdwith a resolution of 1.0 m × 1.0 m, covering the farther surroundings

of the gully site. For this workflow, GCPs were not measured in thefield but assigned after manually identifying coordinate values fromthe satellite image provided. The selected aerial photographs cover thelarger area neighbouring the main gully environment and thereforeallow analysis of environment the study site GLA is located within.The errors of position of the second DTM (resolution of 1.0 m ×1.0 m) amount to x/y/z values of 1.7/1.04/1.1 m.

Based on the secondDTM, hydrological analysis took place in the GISsoftware environment. After determining the flow directions for eachpixel, the existing sinks were identified and filled to assure that thereare no drainless areas left. The intermediate result was a depressionlessDTM. The pour point was set at the exit point of themain gully enteringthe wadi. Next, a flow accumulation raster was calculated and thisfinally led to the derivation of a synthetic hydrological stream network.A second product derived from the depressionless DTM is the water-shed containing the hydrological network. These two data products –

the watershed and the modelled hydrological network – were used inorder to relate the erosion rates to the runoff contributing area asdescribed in Section 3.2.

The secondDTMwas created from aerial photographswith a smallerimage scale. Due to this finer scale, the image resolution is lower and thedegree of details is reduced compared to the detailed gully site DEM.Therefore, the micro-relief of rills and furrows created by the levellingprocess is not captured by the DTM and the modelled hydrologicalnetwork which follows the actual downslope direction of the terrainrather than being diverted by the levelling pattern.

In Fig. 3 the three main working steps for quantifying the gullyvolume are schematically illustrated. The generated gully-site DTMwas required as input data for the quantification of the minimum gullyvolume. Moreover, the digital stereo models were used in a stereo-digitizing environment (3D graphics card and shutter glasses) formanually digitizing the edge of the gully system (Fig. 3, Step 1). Thegully edge is well identifiable for the presented study site GLA: Occur-ring regressive erosion created a sharp cut in the natural continuoussurface. This breaklinewhere the vertical gully sidewalls and the terrainsurface meet was set as gully edge.

On the basis of the raster DTM and the 3D polyline of the gully edge,the calculation of the erosion volume took place in the GIS softwareenvironment: A 3D-polygon was created using the digitized 3D edgeline. This 3D-polygon was taken as the minimum terrain level beforeerosion occurred. The polygon was then transformed to raster formatand the DTM was subtracted (Fig. 3, Step 2). This step led to a rastercontaining the height difference from before erosion minus the currentstate for every raster cell. Multiplying each value with the surface ofeach cell and finally summing up all single cell values result in the quan-tified volume of the gully system (Fig. 3, Step 3).

3. Results

3.1. Rainfall simulation analysis

On the gently tilted alluvial fan, we found mostly rather flat areas inthe study sites. These sites were tested by the rainfall simulationswith amedian inclination of only 3° for all test plots. In total, the slope of theplots ranged from 0.5° to 11°, but only seven test plots were installedon slopes steeper than 7° on the flanks of gully- or wadi-systems. Thehighest median of slope angles was found in GLA with 4.5°. We foundthe lowest slope inclination in GCH with a median of only 1° (Fig. 6a).The slope varied here significantly (P b 0.05) compared to all othertest sites. Vegetation cover (Fig. 6b) was also rather low. Nevertheless,vegetation cover could vary from 0 to 100% on the entire test plots,but the median reached only 20%. The lowest vegetation cover wasfound in GLA and HOU with medians of only 5% and 0% respectively.We measured the highest result in LAS with a median of 40%. Signifi-cantly different (P b 0.05) was only the study site HOU from TAL andLAS. The rock fragment cover for the test plots ranged from 0 to 95%,

Fig. 3. Quantification of the gully volume. The digitized 3D gully edge is set as minimum elevation level for the creation of the pre-erosion surface.

29K.D. Peter et al. / Catena 113 (2014) 24–40

the median however was only 15%. All study sites had similar medianrock fragment covers of around 15%. Only HAM with a median of 35%had a slightly highermedian (Fig. 6c). GCHwith amedian of 5% differedsignificantly (P b 0.05) from HOU, TAL, LAM and HAM. Soils weredominantly covered by physical crusts, with a median of 50% of thetotal area. The highest crust cover with medians of 90% was found inGLA and HOU, where the bare soils after land levelling are highly sealedand crusted after the first rainfall events. These two test sites were alsosignificantly (P b 0.05) different from all other test sites, except thatGLA and LAM did not vary significantly. We found the lowest amountsof physical crusts in LAS with only 20% (Fig. 6d).

In LAS, as well as in GCH, there were high proportions of biologicalcrust with medians of 40%. For all other study sites except for TAL(25%) biological crusts were rare. Visual differentiation of physical andbiological crusting was easily possible in the field and is illustrated bypolished sections in Fig. 4. The biological crust shows a roughmicromor-phological surface. Single grain structure, cavities and vertical connec-tivity are recognizable. The physical crust is characterised by a flatsurface, vesicles and a dense, platy structure which develops under

Fig. 4. Two examples of biological (left) and physical (right) soil crusts with their respective pstructure.

multiple pressure impacts by heavy machines. In general, soils werevery dry. On the upper 0.05 m of the soil at the beginning of each exper-iment, the soil moisture content ranged from 0.48 to 13%, with a verylow median of only 1.80%. High soil moisture contents of over 5% wereonly found on 19 test plots. These contents were due to natural rainfallevents in short advance of the rainfall experiment. However, there wasno relationship found between the initial soil moisture content and therainfall simulation outcomes.

The pH-values in the study sites are high around8. The highestmeanvalue of 8.61 is found in GLA, the lowest with 7.63 in HAM. The bulkdensity of soils varies from 1.40 g cm−3 in LAS to 1.59 g cm−3 inHOU, where also the absolute maximum of 1.82 g cm−3 is found.Here, the influence of levelling can be distinguished. Soil organic matterranges between 2 and 6%. Highest contents over 5% are found in HOU,LAS and GLA. The particle size of the fine fraction decreases from biggerfractions in the North to smaller fractions in the South (Fig. 5). Thesmallest proportion of coarse and medium sand is measured in thelevelled test sites HOU and GLA which are also the southernmoststudy sites in the distal part of the alluvial fan. The highest proportions

olished section. a) biological crust, b) single grain structure, c) cavity, d) vesicles, e) platy

Fig. 5. Particle size distribution of fine earth (≤2 mm) of study sites.

30 K.D. Peter et al. / Catena 113 (2014) 24–40

are found in HAM and LAM, which are the northernmost test sites. Allother results can be seen in Fig. 5.

For the rainfall simulations, the produced runoff varied from 0.29 Lto 4.67 L (median 2.62 L), which is equivalent to runoff coefficients(RCs) from 5% to 79% (median: 46%). The start of runoff ranged from0:30 min to 10:05 minwith amedian of 3:10 min. Suspended sedimentloads of 0.1 g up to 53.1 g (median 4.5 g) were measured on thetest plots, which result in amounts of 0.4 g m−2 to 188.0 g m−2

(median 15.9 g m−2). The average sediment concentration rangedfrom 0.1 g L−1 to 12.8 g L−1 (median 2.0 g L−1) (Table 2).

The highest RCs were found on the levelled study sites in HOU andGLA as well as in GCH (non-levelled) with median values of 57%, 53%

Fig. 6. Boxplots of the soil surface characteristics a) slope; b) vegetation cover; c) rock fragmdifferences in themedian values among the treatment groups are greater than would be expecte

and 51%. We measured less runoff in the other non-levelled studysites with 47.5% to 40% (Fig. 7a). The boxplots also show that there isa wide dispersion of the data in almost all study sites, especially in TALand LAS. Moreover, there are outliers in the minimum and maximumsectors. The highest suspended sediment loads were also collected inthe levelled test sites GLA and HOU with 77.1 g m−2 and 36.6 g m−2

respectively (median values). Very distinctive lower amounts ofsediment were eroded on the non-levelled study sites. The valuesrange from 4.4 g m−2 to 20 g m−2 (Fig. 7b). The dispersion of thedata is not as high as for RCs. However, there are also outliers found,especially in the maximum sectors. The all pairwise multiple compari-son procedure (Dunn's method) indicates that for RCs, only HOU

ent cover; and d) physical crust) in % of the test plot area for the seven study sites. Thed by chance; there is a statistically significant difference (a, c, d: P = b0.001, b: P = 0.003).

Table 2Statistics of the rainfall simulation dataset of 122 rainfall simulations.

Variable Min Max Median Mean

Runoff start [min] 0:30 10:05 3:10 3:48Runoff [L] 0.29 4.67 2.62 2.51Runoff coefficient [%] 5 79 46 44Sediments [g] 0.1 53.2 4.5 7.3Sediment concentration [g L−1] 0.1 12.8 2.0 2.7Vegetation cover [%] 0 100 20 28Rock fragment cover [%] 0 95 15 24Physical crust cover [%] 0 100 50 48Biological crust cover [%] 0 100 0 17Slope [°] 0.5 11 3 3Moisture [%] 0.5 13 1.8 2.9

31K.D. Peter et al. / Catena 113 (2014) 24–40

varies significantly (P b 0.05) from LAM and HAM. However, for thesuspended sediment load, GLA and HOU differ significantly (P b 0.05)from all other test sites except from LAM.

Fig. 8 shows the progress of the ongoing rainfall simulation experi-ments in mean values for the different study sites in six intervals eachwith 5 min. After low overland flow of around 0.1 l for all study sitesin the first interval, there is an immediate rise in the next interval(min. 0.25 L, max. 0.49 L). From the third and fourth step onwards analmost constant level is reached with the highest values in the lastinterval (e.g. 0.69 L in HOU or 0.50 L in HAM). A steady state betweeninfiltration and overland flow is attained. Only the course of LAMshows a further stronger rise in the last intervals. The highest valuesin all intervals are found for the levelled study sites HOU and GLA. Thebasically uniform rise of the RCs in all study sites describes the sameprocess during the rainfall simulation.When the applied water amountexceeds the infiltration after a fewminutes, runoff rises to a certain levelreaching a steady state, while infiltration and runoff remain constant.

With regard to the sediment load, the courses of the rainfall simula-tion experiments in the different study sites are more differentiated. Onthe non-levelled study sites, there is a slight rise of sediment load in thefirst three intervals up to 1.3 g in LAM. This low level is maintainedconstantly to the end of the experiment. The lowest value of 0.51 g inthe last interval is reached in GCH, the highest with 1.65 g in LAM.The curves of the levelled study sites GLA and HOU reach 2 to 4 timeshigher levels. There is a strong increase of sediment load from the firstto the second interval; the sediment load rises in GLA from 1.49 g inthe first interval to 4.51 g in the third. It continues to rise to amaximumof 4.69 g per interval. In HOU, the sediment load remains on a constantlevel of around 2.5 g from the second interval. Themaximum is reachedin the fifth interval with 2.72 g.

By the use of a Spearman-Rho correlation (Table 3), it is discoveredthat there aremostly low correlations between the results of the rainfall

Fig. 7.Boxplots of (a) Runoff in % of total simulated rainfall amount on plot (RC) and (b) sedimentreatment groups are greater than would be expected by chance; there is a statistically signific

simulations and the plot characteristics. The highest correlations arefound between physical crusts and sediment yield (ρ = 0.78) andrunoff (ρ = 0.64). There is only a slight but still significant (at the0.01 level) negative correlation between vegetation cover and sedimentyield of about ρ = 0.47. Further higher negative correlations can befound between runoff and water infiltration depth (ρ = 0.72) andrunoff and runoff start (ρ = 0.71). All other statistically significantcorrelations are below ρ = 0.5 or correspond to autocorrelationphenomena.

For a better understanding of the coherencies between the influenceof plot characteristics and runoff generation and sediment production,scatter plots of different features combined with runoff and sedimentload are used (Fig. 9). The scatters of rock fragment cover and runoffand sediment load do not show a coherency, neither does slope. Bettercorrelations can be seen with vegetation cover. In combination withrunoff, there is a tendency of lower runoff with higher vegetationcover. However, there is still a high dispersion and high RCs are alsofound under high vegetation cover conditions. The sediment load alsodeclines reasonably with more vegetation cover. The combination ofrunoff and crust cover shows a clearer output. Higher RCs are achievedwith higher crust cover. Even more explicit is the correlation of crustcover and sediment load.With higher sealing crust covers, the sedimentload increases continuously.

Furthermore, it is possible to compare the mean values of RCs andsediment loads and the mean values of the surface characteristics ofthe different study sites (Fig. 10). There is a statistically significant(P = 0.004 and P = 0.008) correlation between the RCs and coarseand medium sand as well as with crust cover with coefficients ofdetermination R2 of 0.83 and 0.78. A little lower relation is found withvegetation cover (R2 = 0.48). For sediment load, the correlations arenot very clear. The combination to coarse and medium sand has only acoefficient of determination R2 of 0.27. Hardly better is the relation tovegetation cover with R2 = 0.48. A good coherency is only found forsediment load and crust cover with a coefficient of determination of0.71 (P = 0.017). It can be seen that the levelled study sites GLA andHOU have the highest crust cover which results in the highest runoffand sediments. Same accounts for the lowest vegetation in thesestudy sites. For coarse and medium sand fractions, the runoff followsexactly the order through the alluvial fan from South (distal part)with the highest runoff to North (proximal part) with the lowest runoff.

3.2. Analysis of SFAP

The La Glalcha gully (GLA) investigated in this study is a dendriticsystem developed by regressive erosion, incising the north-westernbank of a small wadi draining into Oued Ouaar. Fig. 11 shows the devel-opment of the GLA gully in a time series from 1968 to 2012. The Corona

t loads in g m−2 for thedifferent test sites. The differences in themedian values among theant difference (a: P = 0.002, b: P = 0.001).

Fig. 8.Mean development of (a) runoff and (b) sediment load for each study site in six 5 min lasting intervals.

32 K.D. Peter et al. / Catena 113 (2014) 24–40

satellite image shows the state of GLA in January 1968 (Fig. 11a).Comparing the gully system of 1968 (Fig. 11a) with the SPOT satelliteimage from December 2006 (Fig. 11b), the existing gully system hasonly slowly developed over this almost 40 years long time period. Itappears slightly deeper and wider incised especially in the easternbranches. On the Quickbird satellite image from October 2009(Fig. 11c), the whole gully system has been filled up and levelled. Thenorthern part is already under cultivation, so it must have been levelledalready earlier. Fresh traces of heavymachines (such as bulldozers leaveafter land levelling) may be recognized in the southern part indicatingrecent levelling measures. The southern area is still not vegetated. InSeptember 2010, the gully has deeply cut the cultivated field againnearly to its former extent (Fig. 11d); brown vegetation litter fromweeds trace the original branches of the former gully that still remainas thalweg depressions. This implies that most of or all infilled soilmaterial within the gully was eroded within only one rain period(2009/2010) with total precipitation of around 430 mm (measured ata nearby fruit plantation). This is an extremely fast gully developmentcompared to the long time period of the initial gully growth. Afterfurther levelling measures in spring 2011, incision processes of thegully are starting already again. Fresh levelling traces and cracks ofsagging are recognizable in the most recent SFAPs from March 2012(Fig. 11e). The material used for infilling the gully is not transportedto the site from other regions, but has obviously been scraped off thesurrounding land by the bulldozer, thus lowering the field surface inthe gully vicinity.

Table 3Spearman-Rho correlation of soil surface characteristics with rainfall simulation results of the

Runoff Runoffcoefficient

Runoffstart

Sediment Sedimentconc.

Slop

Runoff 1Runoff Coefficient 0.989⁎⁎ 1Runoff start −0.700⁎⁎ −0.712⁎⁎ 1Sediments 0.554⁎⁎ 0.543⁎⁎ −0.331⁎⁎ 1Sediment conc. 0.198⁎ 0.190⁎ −0.052 0.884⁎⁎ 1Slope 0.088 0.072 −0.082 0.400⁎⁎ 0.421⁎⁎ 1Vegetation cover −0.398⁎⁎ −0.397⁎⁎ 0.364⁎⁎ −0.472⁎⁎ −0.300⁎⁎ −0.Rock fragment cover 0.119 0.115 −0.131 0.306⁎⁎ 0.221⁎ 0.Crust cover 0.634⁎⁎ 0.643⁎⁎ −0.432⁎⁎ 0.784⁎⁎ 0.573⁎⁎ 0.Biological crusting −0.026 −0.022 −0.123 −0.550⁎⁎ −0.595⁎⁎ −0.Moisture 0.128 0.130 −0.094 −0.084 −0.188⁎ −0.Infiltration depth −0.692⁎⁎ −0.715⁎⁎ 0.523⁎⁎ −0.344⁎⁎ −0.092 0.

⁎⁎ Significant at the 0.01 level.⁎ Significant at the 0.05 level.

Several products result from photogrammetric analysis of theacquired SFAPs for the study site GLA. A gully-site DTMwith a resolutionof 0.05 m × 0.05 m was created for the gully state of 2010/09/28 asshown in Fig. 11d. The heights given in Fig. 12a are relative values tothe zero level that was generated while creating the local coordinatesystem (see 2.2.3 Aerial survey campaigns). The relative height valuesrange from 2.49 m to −6.93 m. The actual maximum incision of thegully is 4.25 m at the wadi bottom. The pink line indicates the digitizedgully edge and defines the gully area of 869 m2 thatwas used for furtherprocessing. The quantification of the actual erosion volume produces atotal of around 720 m3 of material that was carried out due to theheavy rainfall events between October 2009 and September 2010.Precipitation values that were measured at a fruit plantation duringthis time period reached a maximum of 73 mm per day (measured2010/02/16). The catchment, which extends from the mouth of thegully at the wadi bend towards North–North-East, has a total area of3.53 ha, with a maximum slope length of 310 m (Fig. 13). The southernpart of the overall catchment area that is affected by land-levellingmeasures has a slope length of 160 m and covers an area of13,690 m2. It is located NW from the wadi, ranges up to the ditch thatvertically crosses the main gully north and ends in the west pointwhere the ploughing rills disappear (see Figs. 11e and 13).

The outlines of this re-activated gully system show a strikinglyparallel course of side branches rather than the typical dendritic pattern.After the levelling took place, the main gully developed at the samelocation as before, re-excavating the original main drainage line of the

whole dataset, 122 rainfall simulations.

e Vegetationcover

Rock fragmentcover

Crustcover

Biologicalcrusting

Moisture Infiltrationdepths

145 1365⁎⁎ −0.274⁎⁎ 1119 −0.576⁎⁎ 0.151 1333⁎⁎ 0.140 −0.311⁎⁎ −0.506⁎⁎ 1064 0.077 −0.137 −0.014 0.088 1131 0.291⁎⁎ 0.080 −0.471⁎⁎ −0.158 0.198⁎ 1

Fig. 9. Scatter plots of RC and sediment load against soil surface characteristics (crust cover, vegetation cover, rock fragment cover and slope).

33K.D. Peter et al. / Catena 113 (2014) 24–40

old system. The lateral rills, however, do not follow the original gullyside-branches (as seen in Fig. 11b and traced by the weeds in Figs. 11dand 12b): Their course in most places deviates from the old systemand follows the levelling furrows instead. In order to explain and discussthis, the gully system was divided into the main gully and the lateralrills. The two different values for gully volume were calculated andamount to 693 m3 for the main gully and 27 m3 for the lateral rills.These are minimum values as the main gully is at maximum as deepincised as before refilling.

The volume of the main gully also equals the minimum amount ofsoil that was scraped from the hillslope into the channel of the old gullysystem during infilling and land levelling. Subtracting the gully area(869 m2) from the levelled part of the catchment area (13,690 m2)results in the area of 12,750 m2 from which the eroded soil material hasbeen ablated which equals the main gully volume of 693 m3. Dividingthe gully volume by this area results in the minimum average height of0.054 m by which the area was lowered due to the mechanized scrapingand thereby amplified erosion events.

Fig. 10. Comparing mean RC and mean sediment load to (a) mean crust cover, (b) mean vegetation cover and (c) mean coarse and medium sand of the study sites.

34 K.D. Peter et al. / Catena 113 (2014) 24–40

In Fig. 12b the orthorectified image mosaic for the study site GLA isillustrated and allows a more realistic impression. The gully may beidentified as well as further smaller rills that are located right from the

Fig. 11. Time series showing the development of the gully system at study site GLA: a) Januasatellite imagery, resolution: 10 m × 10 m), c) October 2009 (Quickbird satellite imagery, r0.086 m × 0.086 m), e) March 2012 (SFAP survey, resolution: 0.086 m × 0.086 m).

investigated main gully which started incising from the wadi bottomtowards the agricultural area (regressive erosion process). In additionto the direction of the main gully also smaller rills may be recognized.

ry 1968 (Corona satellite imagery, resolution: 7.5 m × 7.5 m), b) December 2006 (Spotesolution (pan-sharped): 0.6 m × 0.6 m), d) September 2010 (SFAP survey, resolution:

Fig. 12. a) DTM of study site GLA (resolution 0.05 m × 0.05 m), 2010/09/28. The light blue areas are the lowest (see legend). b) Orthorectified image mosaic of study site GLA (imagesacquired September 2010). Darker areas bordering gully rills are vegetation. These areas show former extent of the gully system. (For interpretation of the references to colour in thisfigurelegend, the reader is referred to the web version of this article.)

35K.D. Peter et al. / Catena 113 (2014) 24–40

These rills cut in orthogonal direction to themain gully. This emphasizesthe assumption that land-levelling measures increase gully growthalong the bulldozer tracks.

4. Discussion

Comparing the study sites with respect to runoff coefficients andsediment production confirms a clear difference between the studysites under the influence of land-levelling measures and the non-levelled study sites. For the RCs, there is a difference of 15.1% betweenthe levelled study sites with a mean RC of 56.8% and the non-levelledstudy sites with 41.7%. Only GCH has a similar high RC, too. This is dueto biological crusting, which can repelwater, limit infiltration and there-fore promote runoff (Eldridge et al., 2000). However, the suspendedsediment load in this study site is the lowest also due to biologicalcrusting. Polysaccharides extruded by cyanobacteria and microfungientrap and bind soil particles together, creating larger soil aggregates(Belnap, 2001b). The RCs of the other study sites are on a distinctlower level. However, runoff is still relatively high in all test sites and

Fig. 13. Part of study site LAM: initially small rill in fallow cereal field in autumn 2010 (left), afrainfall event in autumn 2012 (right).

able to erode soil and create gullies all over the entire study area. Resultsare comparable to the study of Seeger (2007), who analysed rainfallsimulations in semi-arid regions of North-East Spain. Evenmore consid-erable is the difference for the sediment erosion. The mean sedimentproduction for the levelled study sites is 59.2 g m−2, 3.5 times higherthan the sediment erosion in non-levelled study sites (17.1 g m−2).This agrees to the observations of Bazzoffi et al. (2006), who point outthat land levelling determines severe disturbance of soil profile andthat in the high vulnerable conditions of soil after land levelling soilerosion by water resulted unacceptable. A similar impact could directlybemeasured by rainfall simulations on a cereal field in the test site LAM.During observation time, a small rill was infilled and levelled. Bulkdensitywas increased from1.37 up to 1.62 g cm−3. Rainfall simulationsin this area prior to levelling showed only relatively low results in runofffrom 20 to 30% and suspended sediments from 16 to 20 g m−2, where-as after land levelling up to 40% of runoff and even 4 times higheramounts of suspended sediments with 79 g m−2 were measured.After further ploughing for field preparation, a subsequent heavy rain-fall event caused high soil losses by a deeply, about 3 m, incising gully

ter land-levelling measure in autumn 2011 (middle) and deeply incised gully after heavy

36 K.D. Peter et al. / Catena 113 (2014) 24–40

(Fig. 13). This sequence of treatment processes and their erosionalimpacts corroborates the negative influence of land-levelling measureson soil erosion. Due to compaction and sealing of the bare soil, overlandflow generation is accelerated and more sediment can be transported.Thus, gully development is triggered. The same processes occurred inLa Glalcha (Fig. 11a–e) and El Houmer.

Soils of the non-levelled study sites are consolidated and despite therelatively high runoff amounts, they are not very prone to soil erosion.Thus, sediment erosion in these areas is much lower.

Furthermore and similar to Ramos andMartínez-Casasnovas (2007),who stated that sealing is higher in high disturbed areas than in lowdisturbed areas after land-levelling measures, we found that studysites with levelling measures (GLA, HOU) have the highest amounts ofsealing and physical crust cover. After the heavy machines have flat-tened the terrain, bare soils tend to seal and crust in the first rainfallevents due to their texture with high amounts of fine sand and silt.These physical crusts reduce infiltration and also inhibit consequentvegetation growth (Belnap, 2001a), which leaves soils unprotectedagainst splash and soil erosion. Crust cover of non-levelled study sitesis explicitly lower (2.3 times lower). In contrast, GLA and HOU havethe lowest fractions of vegetation cover; the other study sites are stilllow, but have 2.2 times higher amounts of vegetation cover. Land level-ling may also be used for the removal of unwanted vegetation (Borselliet al., 2006). Thus, soils may be left bare and vulnerable for heavy rain-fall events.

These results are also highlighted in the course of the rainfall simu-lations as well as in the boxplots. The curves of the levelled study sitesHOU and GLA start off with the highest runoff due to the compactedand sealed soil surfaces. Infiltration is hardly occurring on these areas(Peter and Ries, 2013). However, at the end of the experiment bothrunoff rates are only a little higher than in GCH and LAM. In GCH,most of the test plots are characterized by biological crusting whichreduces infiltration (Maestre et al., 2002) and can lead to high overlandflow. The high runoff of LAM is due to high sealing crust values (Fig. 6d)on the fallow arable land. The rise of sediment load in the first intervalsis caused by rising runoff water which washes loose material from thetest plots. Only on the levelled study sites GLA and HOU, very high sed-iment loads are reached. Due to the levellingmeasures and consecutivephysical crusting, these study sites have very even surfaces with littleroughness. Physical crusts cause smoothing of the soil surface(Rodríguez-Caballero et al., 2012). Soil crusts may still be broken upby splash during the experiments; runoff is not constrained due tolow roughness (FAO, 1993) and thus, runoff water is able to erode alot of sediments. Additionally, due to low infiltration and fast ponding,high overland flow on the crusts is able to transport all the detachedsediment (Singer and Shainberg, 2004).

Slopes in all study sites are rather flat due to their position on theplane alluvial fan. Rock fragment cover is very equal on a low level inall study sites. Only HAM in theNorth andGCH in the South–West differwith high and low values respectively. The influence of stone cover ishighly variable due to their ambivalent effect on infiltration dependingon position, size and cover (Poesen et al., 1998). As stone occurrence onthe test plots is mixed with embedded in and positioned on the soilsurface, there is no correlation recognizable. Moreover, the influenceof the other soil parameters is rather unclear. Solely the bulk densityof HOU is clearly high due to the levelling measures. The compactionof the soil reduces infiltration and thus increases runoff and sedimenterosion. The bulk densities of the non-levelled study sites are slightlylower. It is assumed that aggregate stability is generally strongly corre-lated to organic matter content (Chaney and Swift, 1984). However, forour study sites two of the highest organic matter contents were mea-sured in GLA and HOU, which have instable soils and the highest soillosses. This is contradictory to the thesis. Only LAS with similar highresults in organic matter shows very low soil erosion. Prasad andPower (1997) stated that under similar climate conditions, the organicmatter content in fine textured (clayey) soils is two to four times that

of coarse textured (sandy) soils. At least GLA also has the highestamount of clay. The highest pH-values are also found in the study sitesGLA and HOU. An alkaline influence of sodium carbonate is assumedin the study site GLA which is intensified by irrigation water. Inducedfast clay dispersion might be a further factor for high erosion results(Mamedov et al., 2002).

When using the Spearman-Rho correlation (Table 3), mostly lowcorrelations for the results of the rainfall simulations and the plotcharacteristics are found. Similar are the correlation results of Seeger(2007). He found only a weak to none correlation between the soilsurface characteristics on the plot and runoff or erosion for rainfallsimulations throughout different landscapes. In our results, only soilcrusting tends to higher correlationswith runoff generation and erosionquantity. Soil surface sealing sharply reduces infiltration, decreaseswater storage in the soil and triggers runoff, and hence soil erosion(Singer and Shainberg, 2004; Valentin and Bresson, 1992). Clearertrends can be seen in the scatter plots (Fig. 9). Here, also crust covershows the best correlations. There are also tendencies recognizable forvegetation cover. However, they are not as clear as for crusting andhigh runoff generation and sediment erosion are also possible withhigh vegetation cover. Ries (2010) stated that dense vegetation coverdoes not always correlate with stability in erosion and the increase incover does not generally lead to a stabilisation of erosion processes.Even under increasing vegetation cover of more than 60%, high erosionand runoff rates are possible, particularly in connected rills and on sheeptrails.

In contrast, the correlations of the mean values of the study sites(Fig. 10) indicate a clearer influence of plot characteristics on runoffgeneration and sediment production and combine the abovementionedcorrelation approaches. There are rather good correlations betweenrunoff generation and crust cover, vegetation cover and coarse andmedium sand. Again, the levelled study sites GLA and HOU have thehighest crust cover and lowest vegetation cover which result in thehighest RCs and sediment losses. For the coarse particle size fractions,there is a clear decrease of this fraction on the alluvial fan from thedistal to the proximal part, which is caused by the fluvial deposition(Stock et al., 2008). This leads to a decrease of infiltration and a higherrunoff. However, this development does not account for sedimentload. Higher runoff does not automatically lead to higher sedimentproduction.

All these correlations illustrate the influence of the soil surface char-acteristics on runoff generation and sediment erosion. Not one specificbut the combination of all characteristics is responsible for the scale ofsoil erosion. The lowest variations in these conditions are also found inGLA and HOU due to the equalising effects of the levelling measures.This also leads to low variations in the RCs with standard deviationsof only 9.7 and 13.2. In the study sites TAL and LAS, there are highvariations in vegetation, rock fragment and crust cover (Fig. 6). Thus,there are also high variations in runoff results with standard deviationsof 20 and 19.1.

Overall seen, it is shown that the strong disturbances to which landlevelling leads, especially themodification of the surface characteristics,high soil crusting and low vegetation cover, result in the highestamounts of water runoff and sediment erosion. Other soil conditionslike organic matter or sodicity might influence erosion, but no moredominant effect than levelling was determined. Besides, on non-levelled study sites which are less disturbed, less crusting and morevegetation is found, thus lower RCs and clearly lower sediment loadsare measured. The study of Ramos and Martínez-Casasnovas (2007)shows similar results, when the most disturbed plot showed a highersediment concentration in runoff, which together with higher runoffvolumes gave higher erosion rates and soil losses than the lowdisturbedone.

Results from the SFAP analysis also prove the negative impact ofland-levelling measures on gully erosion development and thereforecomplement the results from the rainfall simulation analysis. Mapping

37K.D. Peter et al. / Catena 113 (2014) 24–40

the gully as well as generating the catchment results in a quantifiedstatement on soil erosion: Within the catchment of 3.53 ha an amountof 720 m3 of soil erosion occurred on a gully area of 869 m2.

Considering several aspects, this value of 720 m3must be consideredthe minimum value for the soil material removed by gully erosion.Especially within the gully, the automatic terrain extraction process issubject to errors and omissions of height values due to shadowing andsight-shadowing (Giménez et al., 2009; Marzolff and Poesen, 2009).Steep walls, narrow and deep rills and subsurface concavities createdby piping processes may not well or not at all be modelled in theDTM. Nevertheless, the derivation of high-resolution DTMs from SFAPis considered a valuable method for erosion quantification purposes(Betts et al., 2003).

In addition, the manual digitizing of the gully edge only delineatesthe main gully with the most prominent lateral rills, not includingevery single small and deep rill developing in the furrows created bythe bulldozer during land levelling. Furthermore, the digitized gullyedge is based on individual human perception of the person to digitizethe gully edge. In the presented case of GLA the gully edge was rathereasy to be defined whereas other gully systems may be more challeng-ing and would therefore stronger influence the accuracy of the overallresults. Defining the location of the gully edge therefore still remains achallenge which requires further discussion as for mapping purposesgully edges may be considered as the critical features (Evans andLindsay, 2010).

Land-levelling measures clearly influence the runoff characteristicswithin the catchment: The orthogonal orientation of ploughing rillstowards the main gully (see Fig. 12a and b) leads to the accumulationand concentration of runoff water during rainfall events. This leads toan increase in the runoff speed while simultaneously the amount ofwater per area unit also increases. Shortly, more water may run offfaster. Subsequently, the erosive impact is constantly increasing alongploughing rills and gullies, though rainfall events may remain constantconcerning the water amount.

The channelling role of the bulldozer furrows is also well illustratedwhen the alignment of the gully's side rills (Fig. 12b) is compared tothe original hydrological network of the gully: One of the biggestcontributing rills – in the upper left of Fig. 11b at around x10/y95, orjust left of the centre of Fig. 11d – running orthogonal to the maingully clearly does not follow the old direction of the gully branchestraced by the dried weeds (and visible in Fig. 11a), but cuts across thefield following the pattern of the furrows created during land levelling.In Fig. 14 the orthorectified image mosaic showing study site GLA isoverlaid with the modelled hydrological network in order to illustratethe described mismatch between ploughing rills and modelled hydro-logical network.

In the case of GLA, nearly all of the gully's side-branching lateral rills –a minimum of 27 m3 eroded soil material – have developed along newcourses pre-defined by the levelling pattern, rather than re-excavatingthe infilled old gully system. This important role of ploughing andland-levelling patterns for the enhancement of runoff and the develop-ment of linear erosion forms has also been discussed by Ries and Hirt(2008) and Marzolff (1999) for abandoned fields in semi-arid Spainand by Kalisch (2009) for the study site HOU in this Moroccan studyarea.

Furthermore visual interpretation of SFAPs from study sites withoutland-levelling measures such as HAM and LAS also took place. Thesenon-levelled study sites do not show significant differences of shapeand extent comparing SFAPs from two different points in time. It maybe concluded that the impact of erosion on these study sites is amultiplelower than on study sites with anthropogenic land levelling takingplace. Betts et al. (2003) confirm this assumption with their findingthat there is a more direct response to rainfall events by incipientgullies.

It is shown that the two presented methods, rainfall simulations aswell as aerial survey campaigns, identify the levelled study sites to be

more prone to soil erosion than the undisturbed study sites. Higher run-off generation and sediment production on the test plots of the levelledstudy sites also lead to more linear runoff which finally initiates gullyerosion and increases gully growth.

A comparison of the dimension of the erosion in the catchment areaand of the erosion of the gully itself may be attained by an attempt toupscale the results of the rainfall simulations of GLA to the catchmentarea. We clearly consider the importance of various spatial and tempo-ral scales in geomorphic and hydrological processes (García-Ruiz et al.,2010). However, as it is only a small catchment with relatively uniformsurface characteristics, we operate a rough calculation using roundedvalues. The median soil erosion rate of around 150 g m−2 h−1

(~75 g m−2 in 30 min) in GLA, measured by the rainfall simulations,is extrapolated to the gully catchment area of 3.5 ha. Assuming that allthe precipitation of the rain period 2009/2010 (around 430 mmmeasured at a nearby fruit plantation) would fall in the same rainfallintensity as the simulated rainfall (40 mm h−1), the soil loss could becalculated as follow:

SC ¼ ES � C� PPS

ð1Þ

with SC = soil loss in catchment area [g], ES = simulated erosion rate[g m−2 h−1], C = catchment area [m2], P = natural precipitation[mm] and PS = simulated precipitation rate [mm h−1]. In total, around56,500,000 g (56.5 t) of soil material would be eroded in the catchmentarea. The gully volume eroded in the rain period of 2009/2010 amountsto a volume of 693 m3 for the main gully, resulting in a weight of1039.5 t soil material, and the lateral rill volume of 27 m3 in 40.5 t.This assumes a bulk density of 1.5 g cm−3, which is the mean bulkdensity of 38 measurements in all study sites. This big amount of soilloss has to be seen with care as it is measured in one specific yearwith high precipitation compared to the mean annual precipitation ofaround 200 mm. Seeger et al. (2009) found a very high temporal vari-ability of gully growth for two gullies in Spain. The variability of processintensity is attributed to the high variability of the rainfall pattern in theMediterranean, similar to our region. Nevertheless, the results of thisyear indicate that the main gully erosion accounts for 91% of the totalsoil loss, lateral rill erosion for 4% and the interrill erosion only for 5%.These proportions fit to the results of Poesen et al. (1996), who foundthat 80% and 83% of the total soil loss were due to gully erosion in twoMediterranean catchments in Portugal and Spain. In the context ofPoesen et al. (2003), who also compared soil loss rates by gully erosionfrom different parts of the world which ranged from 10 up to 94%, ourresults would be allocated in the maximum group. The interrill area isonly important as runoff source. The presence of a highly active gullyin the catchment may imply the soil erosion of the interrill area ascomparatively negligible, considering overestimation of the rainfallsimulations due to assumed maximum erosion on the test plots andmaximum rainfall application, but disregard of connectivity. In thisspecial case, human activity by bulldozing the gully is the main reasonfor the high erosion rates as the supply of soil material, which can beeroded, is sustained by scraping it from the catchment area into thegully. However, comparing the erosion results of both methods, rainfallsimulation and SFAP analysis, have to be seen with caution. It is impor-tant to consider spatial and temporal scales. It is well known that geo-morphic and hydrological processes are scale-dependent (De Venteand Poesen, 2005; Lesschen et al., 2009; Skien et al., 2003), with eachscale relying on certain processes (Cammeraat, 2002; Cerdan et al.,2004; Yair and Raz-Yassif, 2004). Studies focused only on experimentalplots or rainfall simulation emphasize processes such as infiltration,splash or runoff generation, but do not consider connectivity with thefluvial channel and the consequences on sediment outputs from catch-ments and on temporal sediment stores (García-Ruiz et al., 2010).

Fig. 14.Modelled hydrological network for study site GLA. The blue stream lines are illustrated on the orthorectified imagemosaic (Images acquired September 2012). (For interpretationof the references to colour in this figure legend, the reader is referred to the web version of this article.)

38 K.D. Peter et al. / Catena 113 (2014) 24–40

5. Conclusions

In the Souss valley, South Morocco, the influence of land-levellingmeasures on soil erosion was investigated. The results of 122 rainfall

simulations showed a clear impact of land levelling on runoff generationand sediment production. On levelled study sites, runoff was 1.4 timeshigher than in undisturbed areas. Sediment production was even 3.5times higher under the influence of land levelling. It was also illustrated

39K.D. Peter et al. / Catena 113 (2014) 24–40

that the soil surface characteristics which are most important for soilerosion were influenced by land levelling. Vegetation cover is 2.2times higher in undisturbed areas than in smoothed areas. In contrast,crust cover is 2.3 times higher in levelled study sites.

Similar results were found using the UAV remote sensing data. In acatchment of 3.53 ha, a gully of 869 m2 eroded about 720 m3 of soil inonly one rain period. Due to levelling measures, the surface of the areasurrounding the gully was lowered 0.054 m on average. This implies avast anthropogenic amplification of the erosional event. Other studysites in this area without land levelling did not show significant differ-ences of shape and extent comparing SFAPs from two different pointsin time.

The connection of both methods showed that in the catchment arealand levelling leads to higher runoff and soil erosion rates which waselucidated by rainfall simulations. Thus, due to higher accumulation ofwater in depression lines development of gullies increased and theirgrowth was amplified, measured by the UAV data. Altogether the soilerosion in the catchment area accounts only for 5% of the total soilloss, whereas the gully erosion (including side-branches) provides95% of the total soil loss.

Acknowledgements

All work presented is taking place within the ongoing internationalresearch project AGASouss (Assessment of gully erosion in agro-industrial landscapes of the Souss Basin (Morocco)), a bi-national coop-eration of Frankfurt, Trier and Agadir University (for more informationon the project constellation, see d'Oleire-Oltmanns et al., 2011).Funding for the research project “AGASouss” is granted by the GermanResearch Foundation (Deutsche Forschungsgemeinschaft/DFG) underresearch contracts MA 2549/3 and Ri 835/5 and gratefully acknowl-edged. The acquisition of the UAV system was financially supportedby the Vereinigung von Freunden und Förderern der Goethe-Universitätand the University of Trier. The authors also want to thank allcolleagues, students and friends who have participated in the fieldcampaigns and supported data acquisition: W. d'Oleire-Oltmanns,M. Düspohl, H. Ghafrani, C. Giudici, A. Hanna, A. Kaiser, P. Kluter,J. König, R. Kusdian, A. Mellali, C. Müller, R. Nagel, M. Nägle, J.Schimpchen, J. Tumbrink and T. Wilms. Further thanks go to A. Kaiserfor his work in the lab.

Constructive and valuable feedback from the two anonymousreviewers helped to increase the quality of the paper.

References

Aber, J., Marzolff, I., Ries, J.B., 2010. Small Format Aerial Photography: Principles,Techniques and Geoscience Applications. Elsevier, Amsterdam (256 pp.).

Ait Hssaine, A., 1994. Géomorphologie et Quaternaire du piémont de Taroudant-OuladTeima, vallée du Souss, Maroc. (Doctorat d'Etat, PhD) Université de Montréal,Montréal.

Ait Hssaine, A., 2002. Le cadre physique de la Dépression du Souss et la dégradation del'environnement sédimentaire. In: Bouchelkha, M. (Ed.), L'espace rural dans leSouss: Héritage et changements. Méditerranée, Aix-en-Provence, pp. 22–27.

Ait Hssaine, A., Bridgland, D., 2009. Pliocene–Quaternary fluvial and aeolian records in theSouss Basin, southwest Morocco: a geomorphological model. Global Planet. Change68 (4), 288–296.

Bazzoffi, P., Abbattista, F., Vanino, S., Pellegrini, S., 2006. Impact of land levelling forvineyard plantation on soil degradation in Italy. Boll. Soc. Geol. Ital. 191–199 (volumespecial 2006).

Belnap, J., 2001a. Comparative structure of physical and biological soil crusts. In: Belnap, J.,Lange, O.L. (Eds.), Biological Soil Crusts: Structure, Function, and Management.Springer-Verlag, Berlin, pp. 177–191.

Belnap, J., 2001b. Biological soil crusts and wind erosion. In: Belnap, J., Lange, O.L. (Eds.),Biological Soil Crusts: Structure, Function, and Management. Springer-Verlag, Berlin,pp. 339–347.

Betts, H.D., Trustrum, N.A., Rose, R.C.D., 2003. Geomorphic changes in a complex gullysystem measured from sequential digital elevation models, and implications formanagement. Earth Surf. Proc. Land. 28, 1043–1058.

Borselli, L., Torri, D., Øygarden, L., de Alba, S., Martínez-Casasnovas, J.A., Bazzoffi, P., Jakab,G., 2006. Soil erosion by land levelling. In: Boardman, J., Poesen, J. (Eds.), Soil Erosionin Europe. John Wiley and Sons Inc., Chichester, pp. 643–658.

BS EN ISO 14688-1, 2002. Geotechnical Investigation and Testing. Identification andClassification of Soil — Part I: Identification and Description.British StandardsInstitution, London.

Calvo, A., Gisbert, B., Palau, E., Romero, M., 1988. Un simulador de lluvia portátil de fácilconstrucción. In: Sala, M., Gallart, F. (Eds.), Métodos y técnicas para la medición deprocesos geomorfológicos. Sociedad Espanola de Geomorfologia, Monografia, 1,pp. 6–15 (Zaragoza).

Cammeraat, L.H., 2002. A review of two strongly contrasting geomorphological systemswithin the context of scale. Earth Surf. Proc. Land. 27, 1201–1222.

Cerdà, A., Schnabel, S., Ceballos, A., Gomez-Amelia, D., 1998. Soil hydrological responseunder simulated rainfall in the Dehesa land system (Extremadura, SW Spain) underdrought conditions. Earth Surf. Proc. Land. 23, 195–209.

Chaney, K., Swift, R.S., 1984. The influence of organic matter on aggregate stability insome British soils. J. Soil Sci. 35, 223–230.

Cots-Folch, R., Martínez-Casasnovas, J.A., Ramos, M.C., 2006. Land terracing for newvineyard plantations in the north-eastern Spanish Mediterranean region: landscapeeffects of the EU Council Regulation policy for vineyards' restructuring. Agric. Ecosyst.Environ. 115, 88–96.

d'Oleire-Oltmanns, S., Marzolff, I., Peter, K.D., Ries, J.B., 2012. Unmanned aerial vehicle(UAV) for monitoring soil erosion in Morocco. Remote Sens. 4, 3390–3416.

d'Oleire-Oltmanns, S., Marzolff, I., Peter, K.D., Ries, J.B., Ait Hssaine, A., 2011. Monitoringsoil erosion in the Souss Basin, Morocco, with a multiscale object-based remote sens-ing approach using UAV and satellite data. Proceedings of the 1st World Sustainabil-ity Forum, 1-30 November 2011; Sciforum Electronic Conferences Series. www.sciforum.net/presentation/562/pdf. (last accessed 16.08.2012).

De Vente, J., Poesen, J., 2005. Predicting soil erosion and sediment yield at the basin scale:scale issues and semiquantitative models. Earth Sci. Rev. 71, 95–125.

Dijon, R., 1969. Etude hydrogéologique et inventaire des ressources en eau de la vallée duSouss. Notes et Memoires du Service Géologique du Maroc, Rabat.

Dunkerley, D., 2012. Effects of rainfall intensity fluctuations on infiltration and runoff:rainfall simulation on dryland soils, Fowlers Gap, Australia. Hydrol. Process. 26 (15),2211–2224.

El Ghannouchi, A., 2007. Dynamique eolienne dans la plaine de Souss: Approchemodelsatrice de la lutte contre l'ensablement. (Thèse de Doctorat) UniversitéMohammed V, Rabat.

Eldridge, D.J., Zaady, E., Shachak, M., 2000. Infiltration through three contrasting biologicalsoil crusts in patterned landscapes in the Negev, Israel. Catena 40, 23–36.

Evans, M., Lindsay, J., 2010. High resolution quantification of gully erosion in uplandpeatlands at the landscape scale. Earth Surf. Proc. Land. 35, 876–886.

FAO Soils Bulletin 69, 1993. Soil Tillage in Africa: Needs and Challenges. Editorial Group,FAO Information Division (http://www.fao.org/docrep/T1696E/T1696E00.htm (lastaccessed 16.08.2012)).

Fister, W., Iserloh, T., Ries, J.B., Schmidt, R.G., 2011. Comparison of rainfall characteristicsof a small portable rainfall simulator and a combined portable wind and rainfallsimulator. Z. Geomorphol. 55 (Suppl. 3), 109–126.

García-Ruiz, J.M., Lana-Renault, N., Beguería, S., Lasanta, T., Regüés, D., Nadal-Romero, E.,Serrano-Muela, P., López-Moreno, J.I., Alvera, B., Martí-Bono, C., Alatorre, L.C., 2010.From plot to regional scales: interactions of slope and catchment hydrological andgeomorphic processes in the Spanish Pyrenees. Geomorphology 120, 248–257.

Ghanem, 1974. Monographie pédologique de la plaine ducbe Souss. Serv. Rech. Ecol.Direction de la recherche agronomique. (101 pp.).

Giménez, R., Marzolff, I., Campo, M.A., Seeger, M., Ries, J.B., Casalí, J., Álvarez-Mozos, J.,2009. High-resolution photogrammetric and field measurements of gullies withcontrasting morphology. Earth Surf. Proc. Land. 34, 1915–1926.

Hu, X., Liu, L.-Y., Li, S.-J., Cai, Q.-G., Lü, Y.-L., Guo, J.R., 2012. Development of soil crustsunder simulated rainfall and crust formation on a loess soil as influenced bypolyacrylamide. Pedosphere 22 (3), 415–424.

Iserloh, T., Fister, W., Seeger, M., Willger, H., Ries, J.B., 2012. A small portable rainfallsimulator for reproducible experiments on soil erosion. Soil Tillage Res. 124, 131–137.

Iserloh, T., Ries, J.B., Cerdà, A., Echeverría, M.T., Fister, W., Geißler, C., Kuhn, N.J., León, F.J.,Peters, P., Schindewolf, M., Schmidt, J., Scholten, T., Seeger, M., 2013. Comparativemeasurements with seven rainfall simulators on uniform bare fallow land. Z.Geomorphol. 57 (1), 11–26.

Kalisch, A., 2009. Ableitung und Analyse von Erosionsrinnen-Netzwerken aus digitalenGeländemodellen mittels großmaßstäbiger Photogrammetrie und GIS — Südwest-Marokko. (Unpublished diploma thesis) Department of Physical Geography,Goethe-University Frankfurt am Main (http://www2.uni-frankfurt.de/45217488/Kalisch_Photogrammetrische-Ableitung-Erosionsrinnen-Netzwerke_Diplomarbeit2009.pdf (last accessed 14.08.2012)).

Köhn, M., 1929. Korngrößenanalyse vermittels Pipettanalyse. Tonindustrie-Zeitung 53,729–731.

Lasanta, T., García-Ruiz, J.M., Pérez-Rontomé, C., Sancho-Marcén, C., 2000. Runoff andsediment yield in a semi-arid environment: the effect of land management afterfarmland abandonment. Catena 38, 265–278.

Lesschen, J.P., Schoorl, J.M., Cammeraat, L.H., 2009.Modelling runoff anderosion for a semi-arid catchment using a multi-scale approach based on hydrological connectivity.Geomorphology 109, 174–183.

Lundekvam, H.E., Romstad, E., Øygarden, L., 2003. Agricultural policies in Norway andeffects on soil erosion. Environ. Sci. Policy 6, 57–67.

Maestre, F.T., Huesca, M., Zaady, E., Bautista, S., Cortina, J., 2002. Infiltration, penetrationresistance and microphytic crust composition in contrasted microsites within aMediterranean semi-arid steppe. Soil Biol. Biochem. 34, 895–898.

Mamedov, A.I., Shainberg, I., Levy, G.J., 2002. Wetting rate and sodicity effects on interrillerosion from semi-arid Israeli soils. Soil Tillage Res. 68, 121–132.

Martínez-Casasnovas, J.A., Ramos, M.C., 2009. Soil alteration due to erosion, ploughingand levelling of vineyards in north east Spain. Soil Use Manage. 25, 183–192.

40 K.D. Peter et al. / Catena 113 (2014) 24–40

Marzolff, I., 1999. Großmaßstäbige Fernerkundung mit einem unbemanntenHeißluftzeppelin für GIS-gestütztes Monitoring von Vegetationsentwicklung undGeomorphodynamik in Aragón (Spanien). Freibg. Geogr. Hefte. 57, Freiburg i.Br.(226 pp.).

Marzolff, I., Poesen, J., 2009. The potential of 3D gully monitoring with GIS using high-resolution aerial photography and a digital photogrammetry system. Geomorphology111 (1–2), 48–60.

Mayor, Á.G., Bautista, S., Bellot, J., 2009. Factors and interactions controlling infiltration,runoff, and soil loss at the microscale in a patchy Mediterranean semiarid landscape.Earth Surf. Proc. Land. 34, 1702–1711.

Nadal-Romero, E., Lasanta, T., Regüés, D., Lana-Renault, N., Cerdà, A., 2011. Hydrologicalresponse and sediment production under different land cover in abandoned farmlandfields in a Mediterranean mountain environment. Bol. Asoc. Geógr. Esp. 55, 303–323.

Peter, K.D., Ries, J.B., 2013. Infiltration rates affected by land levelling measures in theSouss valley, South Morocco. Z. Geomorphol. 57 (1), 59–72.

Poesen, J., Vandaele, K., vanWesemael, B., 1996. Contribution of gully erosion to sedimentproduction in cultivated lands and rangelands. IAHS Publ. 236, 251–266.

Poesen, J., van Wesemael, B., Bunte, K., Solé-Benet, A., 1998. Variation of rock fragmentcover and size along semiarid hillslopes: a case-study from southeast Spain.Geomorphology 23 (2–4), 323–335.

Poesen, J., Nachtergaele, J., Verstraeten, G., Valentin, C., 2003. Gully erosion and environ-mental change: importance and research needs. Catena 50, 91–133.

Prasad, R., Power, J.F., 1997. Soil Fertility Management and Sustainable Agriculture. LewisPublishers, New York.

Ramos, M.C., Martínez-Casasnovas, J.A., 2007. Soil loss and soil water content affected byland levelling in Penedès vineyards, NE Spain. Catena 71 (2), 210–217.

Ries, J.B., 2010. Methodologies for soil erosion and land degradation assessment inMediterranean-type ecosystems. Land Degrad. Dev. 21, 171–187.

Ries, J.B., Hirt, U., 2008. Permanence of soil surface crusts on abandoned farmland in theCentral Ebro Basin/Spain. Catena 72, 282–296.

Ries, J.B., Langer, M., 2002. Runoff generation of abandoned fields in the Central EbroBasin. Results from rainfall simulation experiments. In: García-Ruiz, J.M., Jones,

J.A.A., Arnaez, J. (Eds.), Environmental Change and Water Sustainability. InstitutoPirenaico de Ecología, Zaragoza, pp. 65–81.

Ries, J.B., Marzolff, I., 2003. Monitoring of gully erosion in the central Ebro basin by largescale aerial photography taken from a remotely controlled blimp. Catena 50,309–328.

Ries, J.B., Marzolff, I., Seeger, M., 2003. Einfluss der Beweidung aufVegetationsbedeckungund Geomorphodynamik zwischen Ebrobecken und Pyrenäen. Geogr. Rundsch. 55(5), 52–59.

Ries, J.B., Seeger, M., Iserloh, T., Wistorf, S., Fister, W., 2009. Calibration of simulatedrainfall characteristics for the study of soil erosion on agricultural land. Soil TillageRes. 106, 109–116.

Rodríguez-Caballero, E., Cantón, Y., Chamizo, S., Afana, A., Solé-Benet, A., 2012. Effects ofbiological soil crusts on surface roughness and implications for runoff and erosion.Geomorphology 145–146, 81–89.

Seeger, M., 2007. Uncertainty of factors determining runoff and erosion processes asquantified by rainfall simulations. Catena 71, 56–67.

Seeger, M., Marzolff, I., Ries, J.B., 2009. Identification of gully-development processes insemi-arid NE-Spain. Z. Geomorphol. 53 (4), 417–431.

Singer, M.J., Shainberg, I., 2004. Mineral soil surface crusts and wind and water erosion.Earth Surf. Proc. Land. 29, 1065–1075.

Skien, J.O., Blosch, G., Western, A.W., 2003. Characteristic space scales and timescales inhydrology. Water Resour. Res. 39, 111–119.

Stock, J.D., Schmidt, K.M., Miller, D.M., 2008. Controls on alluvial fan long-profiles. Geol.Soc. Am. Bull. 120 (5–6), 619–640.

Tagma, T., 2011. Ressources en eau souterraine de l'aquifère du Souss-Massa: étudede laqualité et de la vulnérabilité. (Thèse de Doctorat) Universität Ibn Zohr, Agadir.

Valentin, C., Bresson, L.M., 1992. Morphology, genesis and classification of surface crustsin loamy and sandy soils. Geoderma 55, 225–245.

Watteeuw, R., 1964. Les sols de la plaine du Souss et leur réparation schématique au1/500.000. Tire à part, Al Awamia, Rabat, Maroc, 10, pp. 151–185.

Yair, A., Raz-Yassif, N., 2004. Hydrological processes in a small arid catchment: scaleeffects of rainfall and slope length. Geomorphology 61, 55–69.