Catena 129 (2015) 9–17

Contents lists available at ScienceDirect

Catena

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

Assessing the sources of sediment transported in gully systems using afingerprinting approach: An example from South-east China

Jinshi Lin, Yanhe Huang ⁎, Ming-kuang Wang, Fangshi Jiang, Xubin Zhang, Hongli GeCollege of Resource and Environment, Fujian Agriculture and Forestry University, Fuzhou 350002, PR China

⁎ Corresponding author.E-mail address: yanhehuang@163.com (Y. Huang).

http://dx.doi.org/10.1016/j.catena.2015.02.0120341-8162/© 2015 Elsevier B.V. All rights reserved.

a b s t r a c t

a r t i c l e i n f o

Article history:Received 8 May 2014Received in revised form 18 December 2014Accepted 16 February 2015Available online 28 February 2015

Keywords:Composite fingerprinting techniqueErosionDiscriminate function analysis (DFA)Sediment sourceSediment transportation

Eroded sediment is an important feature of permanent gully's development; however, obtaining the specificcharacteristics of the sediment transport in the gully can be difficult. Thus, a composite fingerprinting technique,incorporating uncertainty analysis, has been employed to investigate themean relative contribution of sedimentsources in permanent gullies. In this study, 31 tracers were measured at 62 different sampling sites from threelayers of a gully wall [surface layer (SL), sandy soil layer (SSL), and semi-weathering rock layer (SWL)] and 36sediment samples from the slumping deposit and alluvial fan of the gullies. The sediment source tracing proce-dure was therefore used to assess the relative contribution of each source. The mean relative contribution fromthe SL to the sediment of the slumping deposit is higher in the active gully (G1), and is relatively lower in thepotentially stable (G2) and stable gullies (G3); thus there was no obvious gully bank retreat in G2 and G3. In dif-ferent places of the G1 alluvial fan, the mean relative contribution of the SL, SSL and SWL ranged from 9–11,36–43 and 46–55%, respectively. However, the sediment contribution of the SL increased in G2 and G3 due tothe changes in the erosion type. The mean relative contributions of the SL were 33 ± 5 (G2) and 30 ± 3%(G3). These findings have important implications for the establishment of a scientific basis for permanent gullysediment transport management and control policies.

© 2015 Elsevier B.V. All rights reserved.

1. Introduction

Gully erosion (especially from a permanent gully) is a widespreadphenomenon in many agricultural landscapes (Poesen et al., 2003)and it is a serious land management issue in many parts of the world(Betts andDeRose, 1999). Granitic regions are some of themost serious-ly affected areas of gully erosion in southern China (Feng et al., 2009).Gullies usually initiate from fluvial transport and rapidly develop intomass movement complexes (Betts et al., 2003). Erosion gullies can beas large as 3.5 ha for a single gully, and the erosion modulus can reach2.3 × 104 t km−2 per year. These gullies cause soil loss, downstream ag-gregation, and increased degradation of agro-ecological systems; thus,more research is focusing on gully erosion.

Long-term studies report that gullies develop in associationwithmassmovement in natural landscape evolution that is linked to thedegradation of vegetation cover. Many studies have described gullyprocesses, such as gully initiation, development and growth(Martinez-Casasnovas et al., 2003; Porto et al., 2014; Sidorchuk et al.,2003; Vandekerckhove et al., 1998). Gully development processes thathave been modeled include initiation, gully bank retreat, sediment de-tachment, and sediment transport (Butler and Memon, 1999; Casaliet al., 2003; Conoscenti et al., 2014; Herzig et al., 2011; Peter et al.,

2014; Prosser and Slade, 1994). Gully erosion occurs randomly with sed-iment from the gully bank being deposited into a slumpormoving into analluvial fan. Thus, simulation of the sediment movement within a gully isdifficult. Despite the existence of models of gully development processes,few studies have been able to define the specific characteristics of the sed-iment transport within a gully, although understanding the sedimentsources and transport is important for developing sediment budgetsand yield models (Wilson et al., 2011). Therefore, it is necessary to finda new way to assess sediment sources and transport within gullies.

The fingerprinting method has been used to identify sources of sedi-ment since the 1970s. This method identifies a single or multiple soilchemical parameters of different sources andusesmixingmodels to delin-eate the relative contribution of each source (Collins et al., 1997a; Collinset al., 2013; Smith and Blake, 2014; Voli et al., 2013; Walling et al., 1999;Wilson et al., 2011). Wilson et al. (2011) used total carbon to identifythe sediment sources in a small-plot runoff experiment. Different soil pa-rameters have been used to identify sediment sources, such as variousminerals, mineral-magnetic properties, chemicals, isotopic and radiomet-ric properties, and physical size parameters (Collins et al., 1997a, 2001;Deboer and Crosby, 1995; Schuller et al., 2013; Walling et al., 2008).Collins et al. (2010a) refined the existing approaches and developed amodified fingerprinting technique incorporating theweighting of proper-ties and prior information. Following its development, the fingerprintingtechnique has been used in various studies. However, to our knowledge,it has never been employed to quantify gully sediment transport.

10 J. Lin et al. / Catena 129 (2015) 9–17

On the basis of its proven and successful application in other soil ero-sion studies, the fingerprinting approach was selected as the most ap-propriate method to identify sediment sources in large, permanentgullies. The objectives of this study were (1) to determine the relativecontribution of various sediment sources to slumping deposits and allu-vial fans, (2) to examine the sediment sources within the gully systems,and (3) to provide reliable sediment contribution data to improve theaccuracy of gully model use in southern China.

2. Material and method

2.1. Study area

This study was conducted at the Longmen catchment of Anxi Coun-ty, Fujian Province, in southeast China (Fig. 1). This location has a

Fig. 1. Location of study catchm

subtropical monsoon climate, an annual mean temperature of 18 °Cand a mean precipitation of 1800 mm during the past 30 years (from1980–2010). Approximately 70% of the annual precipitation occursfrom June to September.

The Longmen catchment represents one of the largest andmost typ-ical gully erosion areas of southern China due to the soil properties andintense precipitation periods. The parent materials of the soil is granite,and because tropical weathering often produces a deepweathering pro-file, the depth of the soil is over 10 m in the research area (sometimesexceeding 50 m). The regolith consists of quartz grains and felsic clayminerals (Sheng and Liao, 1997). Soils can be divided into three layersin each gully wall; the soil layer characteristics are summarized inTable 1.

Local researchers describe a permanent large gully as a “collapsinghill” or “collapse gully” (Feng et al., 2009). The gully erosion in the

ents and sampling sites.

11J. Lin et al. / Catena 129 (2015) 9–17

study area is quite different from the Loess Plateau in northwesternChina. Collapse gullies are usually started from rill erosion; then, the re-moval of topsoil and subsoil by fast-flowing surface water createsabrupt, deep and wide gullies. There are three different periods of col-lapse gully in the study area (Fig. 2); the characteristic of the differentperiods are summarized in Table 2.

2.2. Soil sampling and analysis

Soil samples were collected in each layer of each of the differentcatchments of gully wall (Table 3). Five random locations were chosenat the samedepth for each layer and then combined to form a compositesample for each sample point. The number of samples collected for eachsource layer is shown in Table 4. This research was conducted within agully system; therefore, the number of samples was low compared toother studies conducted within a catchment or river system. Soil sam-ples from a slumping deposit and an alluvial fan were also collected ata sampling depth of 1 m. All of the samples were collected using asmall, non-metallic shovel, which was repeatedly cleaned to avoidinter-sample contamination.

All of the soil source material and sediment surface samples weretransported to the laboratory, where the soil samples were air-dried,manually disaggregated using a mortar and passed through a200 mesh nylon sieve for chemical analysis. The particle size distribu-tion of the sieved samples were measured using a laser granulometer.A range of potential fingerprint properties was selected for analysis topermit the identification of effective composite fingerprints. A total of31 soil characteristicswere observed in the soil analyses. After digestionwith nitric and hydrochloric acids, total concentrations of Al, K, Na, Mg,Ti, Zr, Ni, V, Co, Cr, Cd, As, Mo, Fe, Mn, Si, and Zn in the soil samples weredetermined using Inductively Coupled PlasmaMass Spectrometry (ICP-MS) (NexION, 300X, Perkin Elmer Inc., Waltham, MA) (Pansu andGautheyrou, 2006). Soil organic carbon was determined by wetoxidation with H2SO4–K2Cr2O7 (Walkley and Black, 1934). Soil total ni-trogen (N) was determined using the Kjeldahl method (Bremner andMulvaney, 1982). Soil phosphorus concentrationswere colorimetricallydetermined using the ascorbic-blue method (Watanabe and Olsen,1965). Soil iron oxides were extracted with dithionite-citrate-bicarbonate (DCB) solutions (Mehra and Jackson, 1960) for free Fe-oxides (Fed) and with NH4–oxalate solutions of pH 3 (0.2 M) for non-crystalline Fe-oxides (Feo) (McKeague et al., 1971). Available Fe andZn were determined using an atomic absorption spectrophotometer(AAS) (280 ZAA, Agilent, Beijing, China), after extraction by DTPA.After extraction with DCB solutions, soil extractable Mn, Si, and Alwere also determined by AAS. Exchangeable and reducibleMnwere de-termined by AAS after extractionwith ammoniumacetate and ammoni-um acetate–hydroquinone, respectively (Pansu and Gautheyrou, 2006).The relative standard deviation based on the three replicates of eachdeterminant in each sample was less than 2%. The analytical error ofeach laboratory methods is showed in Table 5.

2.3. Sediment source discrimination

Soil samples were collected to characterize the three potential sedi-ment source types that were discriminated on the basis of compositefingerprint method (Collins et al., 1997b, 2010a). These signaturesused a combination of the Kruskal–Wallis H-test and multivariate

Table 1Characteristics of different soil layers from the gully wall.

Gully wall soil layer Thickness(m)

Soil texture (%) Soil resistanceability of erosion

Sand Silt Clay

Surface soil layer 2 ± 0.4 28.05 36.75 35.20 HighSandy soil layer 3 ± 1 32.55 60.00 7.45 LowSemi-weathering rock soil layer N9 44.97 48.99 6.04 Middle

Discriminate Function Analysis (DFA), as proposed by Collins et al.(Collins et al., 1997b; Collins and Walling, 2002). A number of otherstudies have used this method in sediment tracing (Chong et al., 2009;Collins et al., 2010b; Hatfield and Maher, 2009; Walling et al., 2006;Walling et al., 2008; Mukundan et al., 2010; Samani et al., 2011).Therewere two steps for discriminated sediment sources, and a numberof sediment source tracing studies have employed this method (Collinset al., 1997b; Collins andWalling, 2007; Mukundan et al., 2010; Samaniet al., 2011). In the first step, the Kruskal–Wallis H-test was used as aninitial screening of 31 soil properties collected from three gullies(Tables 3, 4). The results of the test are presented in Table 5. In the de-veloping gully (G1), a total of 21 individual properties passed the test.The weakest of the properties failing the test was Cd, yielding a P-value of 0.127 and 0.938. For the G2 (G3) gully, 20 (15) individual prop-erties failed the selection process.

In the second step, multivariate DFA was used to identify the opti-mum combination of properties for discriminating the source samplescollected from different soil layers. The results of applying DFA to thesource material sample datasets collected for the gully are shown inTable 6. Statistical analysis clearly demonstrates that no single propertyis capable of classifying 100% of the source material samples into thecorrect source categories for different stage gullies. However, several in-dividual soil properties can be used as potential indication of the likeli-hood of correctly classifying all of the source samples using compositesignatures. In the G1 gully, nine soil properties (Zn, K, P, Co, exchange-able Mn, Ni, Si, acid extractable Fe, and reducible Mn) were selectedfor the optimum composite fingerprint that discriminated 99.9% of thesource type samples correctly (Table 6). The classification of the sourcesamples provided by the optimum composite fingerprint is illustratedgraphically in Fig. 3. In the initial stages of the stepwise selection, thescatter plots showed considerable overlap between SSL and SWL. How-ever, at the end of the stepwise selection of the composite fingerprint,the samples collected from each source were heavily clustered. For theG2 gully, five soil properties (Mn, Available Zn, Cd, V,Mg)were selected,correctly distinguishing 97.7% of the source samples. The optimumcomposite fingerprints selected by the DFA for the remaining gully(G3) was 99.7%. Scatter plots illustrating the discrimination of thesource samples collected in the G2 and G3 are shown in Fig. 4.Weighting to reflect thewithin-source variation of individual propertieswas also performed, asweighting has been shown to aid in constrainingthe uncertainty ranges in predicted source contributions (Collins et al.,2010a, 2010b).

2.4. Sediment source apportionment

The sediment source apportionmentwas obtained using a multivar-iate mixingmodel. The mixingmodel optimizes estimates of the sourcecontributions byminimizing the sumof squares of theweighted relativeerrors (Collins et al., 2010a), using Eq. (1):

Xni¼1

Ci−Xns¼1

PsSsiZsOsSVsi

!!=Ci

)2

Wi

(ð1Þ

where Ci = the concentration of the fingerprint property (i) in the sed-iment sample; Ps = the optimized percentage contribution from thesource category (s); Ssi = the median concentration of the fingerprintproperty (i) in the source soil layer (s); Z = the particle size correctionfactor for the source category (s); O = the organic matter content cor-rection factor for the source category (s); SVsi=weighting representingthe within-source variation of the fingerprint property (i) in the sourcecategory (s); Wi= tracer discriminatory weighting; n= the number offingerprint properties comprising the optimum composite fingerprint;and m = the number of sediment source categories.

Particle size and organic matter content correction factors were in-cluded in the model (Collins et al., 1997a). Weighting to reflect the

Fig. 2. Different periods of gully erosion in the study area (A) active gully, (B) potentially stable gully, and (C) stable gully.

12 J. Lin et al. / Catena 129 (2015) 9–17

within-source variation of the fingerprint property has also been con-sidered. Two linear boundary conditions were imposed on the mixingmodel iterations to ensure that the relative contributions (Ps) of the in-dividual sediment sources were non-negative (Eq. (2)) and that thesecontributions sum to unity (Eq. (3)). The efficiency of this approachfor determining sediment source contributions has already been dem-onstrated by several studies (Collins et al., 2011; Samani et al., 2011).

0≤Ps≤1 ð2Þ

Xns¼1

Ps ¼ 1 ð3Þ

The goodness-of-fit provided by the mixing model was assessed bycomparing the actual fingerprint property concentrations measured insediment samples with the corresponding values predicted by theoptimization procedure (Collins et al., 2010a; Smith and Blake, 2014).

Table 2Characteristics of different types of gully erosion in the study area.

Type Characteristic Erosion type

Active gully In obvious headward erosionthe gully extends upstream asa result of waterfallundercutting and thegravitational slumping of thegully head, removal of topsoiland subsoil creates abruptdeep and wide gully. Slumpingsoil deposited against the gullywall.

Splash erosion; headwarderosion; gravity erosion;rill erosion; lateral erosion;

Potential stablegully

Gully is widening throughlateral erosion, and soils in thegully wall, deposited soil, andalluvial fan are run off bysplash or rill erosion; but theheadward erosion slowsdown.

Lateral erosion; splasherosion; headward erosion;rill erosion;

Stable gully No headward erosion. Splasherosion in gully wall, depositsoil and alluvial fan.

Splash erosion; rill erosion;

The overall mean goodness-of-fit for the sediment samples collectedat each gully was estimated at 0.91 (G1), 0.88 (G2), and 0.90 (G3).The mean relative errors for the mixing model calculations rangedfrom 6–15%, indicating that the mixing model provided acceptableestimates of the individual fingerprint property concentrations in thesediment samples (Collins et al., 1997a). Table 7 presents a summaryof the ranges of the fingerprint properties selected for the optimumcomposite fingerprints. These data demonstrate that the sedimenttracer properties were within the mixing polygon defined by the corre-sponding ranges measured for the source types.

AMonte Carlo approachwas used to quantify the uncertainties asso-ciated with the optimized sediment source contributions predicted bythe mixing model (Collins and Walling, 2007). Cumulative normaldistributions were constructed on the basis of the mean and standarddeviations of the measurements for each fingerprint property for eachgully using a random number generator. The mixing model was solved500 times. The confidence limits (95%) for the relative contribution ofeach soil layer to each sediment sample were estimated using the stan-dard error of the mean associated with the results of the repeat itera-tions (Collins and Walling, 2007).

The weighted mean relative contribution of each potential sourcetype to the sediment samples was calculated using Eq. (4), which wasused by Collins and Walling (2007).

Table 3Summary information of gully wall sampling.

Samplingcatchment

Longitude Latitude Type Height(m)

Thickness of eachsoil layer (m)

G1 118°3′48.4″ 24°57′6.50″ Developing 12.8 1.7a3.8b7.3c

G2 118°3′26.26″ 24°57′9.29″ Stable 10.0 2.0a3.8b4.2c

G3 118°3′04.2″ 24°58′36.6″ Potentiallystable

9.3 1.6a2.0b5.7c

a, b, and c represent surface layer, sandy soil layer and semi-weathering rock layer,respectively.

Table 4Summary information on the number of samples collected for the source layers andsediment.

Catchment Source layer Erosion sediment

Surfacelayer

Sandy soillayer

Semi-weatheringrock layer

Slumpingdeposit

Alluvial fan⁎

5 m 10 m 15 m

G1 6 8 8 3 3 3 3G2 6 6 8 3 3 3 3G3 6 6 8 3 3 3 3

⁎ Soil samples of an alluvial fan collected from5, 10 and 15mdistances fromgullywalls.

Table 6The optimum composite fingerprints for discriminating individual sediment source typesin each layer of a collapsing wall.

Gullyerosioncatchment

Step Fingerprintpropertyselected

Wilks'lambda

Cumulativesourcetype samplesclassifiedcorrectly(%)

Tracerdiscriminatoryweighting

G1 1 Zn 0.0693 44.44 2.12 K 0.0233 77.13 1.53 P 0.0078 88.45 1.04 Co 0.0023 93.76 2.05 Exchangeable Mn 0.0007 97.69 1.36 Ni 0.0002 98.73 2.27 Si 0.0001 99.43 1.18 Acid extractable

Fe0.0000 99.71 1.3

9 Reducible Mn 0.0000 99.90 1.5G2 1 Mn 0.2290 80.86 2.4

2 Available Zn 0.0390 94.63 1.03 Cd 0.0079 97.42 1.14 V 0.0007 97.66 1.35 Mg 0.0001 97.70 2.6

G3 1 N 0.3185 73.56 2.12 K 0.1027 97.28 1.83 Zn 0.0078 99.08 1.24 Acid extracted Fe 0.0018 99.49 1.65 Mo 0.0007 99.65 1.06 Al 0.0002 99.70 2.2

13J. Lin et al. / Catena 129 (2015) 9–17

Psw ¼Xns¼1

PsxLxLt

� �ð4Þ

where Psw=weighted relative contribution fromdifferent soil layer (s);Psx=relative contribution of a source type for a given sediment sample(x); Lx = the instantaneous estimate of sediment storage in associationwith a fine sediment sample; and Lt = the sum of the instantaneous es-timates of sediment storage associated with a sediment sample collect-ed from the slumping deposit and alluvial fan. This approach canprovide a more meaningful estimate of the mean source of sedimentsamples collected from slumping deposits and alluvial fans (Collinsand Walling, 2007).

3. Results and discussion

3.1. Contributions to the slumping deposit sediment

Fig. 5 presents the overall mean relative contribution of each sourcetype to the sediment samples collected from the slumping deposit in

Table 5The output of Kruskai–Wallis H-test to the source type fingerprint property datasets for each l

Property Analytical errors of each property Gully erosion c

G1

H-value

C 0.05 g/kg 1.769N 0.0125 g/kg 1.769P 0.25 mg/kg 13.752K 4.5 μg/kg 17.397Na 6.36 μg/kg 13.779Mg 1.97 μg/kg 3.672Ti 0.46 μg/kg 3.163Zr 0.04 μg/kg 4.343Ni 0.06 μg/kg 13.350V 0.08 μg/kg 14.315Co 0.03 μg/kg 16.969Cr 0.11 μg/kg 13.752Cd 0.05 μg/kg 0.127As 0.12 μg/kg 9.034Mo 0.06 μg/kg 14.518Fe 0.82 μg/kg 15.038Fed 0.01 mg/kg 15.896Feo 0.01 mg/kg 0.885Available Fe 0.01 mg/kg 18.041Acid-extractable Fe 0.01 mg/kg 8.390Mn 0.12 μg/kg 5.308Available Mn 0.03 mg/kg 13.028Exchangeable Mn 0.03 mg/kg 11.822Reducible Mn 0.03 mg/kg 14.181Acid extracted Mn 0.12 μg/kg 0.885Zn 0.67 μg/kg 13.108Available Zn 0.02 mg/kg 2.052Si 0.10 μg/kg 13.108Al 1.15 μg/kg 7.533Extracted Si 0.10 μg/kg 6.675Extracted Al 0.3 mg/kg 12.679

⁎ Means statistically significant at P b 0.05.

each gully. Fig. 5 also shows important contrasts in the relative contri-butions of each soil layer. The sediment originating from the surfacelayer ranged between 69 ± 5% (G1) and 26 ± 7% (G2). The mean rela-tive contribution of the sandy soil layer in the active gully's slumping

ayer of collapsing wall.

atchment

G2 G3

P-value H-value P-value H-value P-value

0.413 3.063 0.216 6.905 0.032⁎

0.413 8.928 0.012⁎ 6.190 0.045⁎

0.001⁎ 4.445 0.108 11.012 0.004⁎

0.000⁎ 3.646 0.162 13.393 0.001⁎

0.001⁎ 0.576 0.750 3.869 0.1440.159 5.374 0.068 2.381 0.3040.206 0.952 0.621 3.869 0.1440.114⁎ 7.208 0.027⁎ 0.536 0.7650.001⁎ 12.214 0.002⁎ 1.012 0.6030.001⁎ 7.608 0.022⁎ 4.821 0.0900.000⁎ 9.980 0.007⁎ 3.869 0.1440.001⁎ 1.620 0.445 6.190 0.045⁎

0.938⁎ 10.433 0.005⁎ 4.048 0.1320.011⁎ 2.411 0.300 1.190 0.5510.001⁎ 5.124 0.077 11.905 0.003⁎

0.001⁎ 0.699 0.705 2.381 0.3040.000⁎ 16.121 0.000⁎ 2.440 0.2950.643⁎ 2.426 0.297 10.714 0.005⁎

0.000⁎ 2.794 0.247 0.536 0.7650.015⁎ 16.121 0.000⁎ 8.631 0.013⁎

0.070 2.303 0.316 6.905 0.032⁎

0.001⁎ 2.794 0.247 8.095 0.017⁎

0.003⁎ 1.988 0.370 10.714 0.005⁎

0.001⁎ 2.426 0.297 8.095 0.017⁎

0.643 0.576 0.750 8.095 0.017⁎

0.001⁎ 4.998 0.082 6.726 0.035⁎

0.359 6.287 0.043⁎ 4.821 0.0900.001⁎ 7.608 0.022⁎ 1.190 0.5510.023⁎ 0.768 0.681 8.155 0.017⁎

0.036⁎ 2.794 0.247 1.488 0.4750.002⁎ 9.243 0.010⁎ 2.143 0.343

Fig. 3. Scatter plots constructed from the first and second discriminant functions calculated using DFA in association with the stepwise selection of the optimum composite fingerprint forthe active gully (G1) sampling site. (Mn1 means exchangeable Mn, Fe2 means acid extractable Fe, Mn3 means reducible Mn).

14 J. Lin et al. / Catena 129 (2015) 9–17

deposit was 13 ± 4%. The gully bank usually retreat as a result of threeprocesses; falling soil, transport of soil material by splashes, or waterrunning along gully banks (Chaplot et al., 2011). The active gully ex-tended upstream as a result of waterfall undercutting and gravitationalslumping of the gully head (Derose et al., 1998). Surface layer soils havehigh erosion resistance, but contributemost of the sediment in an activegully slumping deposit (approximately 70%). Therefore, this estimate islikely to reflect, at least in part, the obvious gravitational slumping in adeveloping gully and the falling soil deposited against the gully wall.However, in the stable and potentially stable gullies (G2 and G3). themean relative contributions of the surface layer soil were only 26 ± 4and 31 ± 5%, respectively. The contributions of the other two layers inG2 were 43 ± 5 and 30 ± 4%, and 49 ± 5 and 19 ± 4% in G3. Thismeans that in the development of this gully, there is no obvious gravita-tional slumping in the stable gully (or no obvious gully bank retreat),and the sediment from the surface layer soil was eroded by rain or run-off in the stable gully after slumping from the gully head.

3.2. Contributions of alluvial fan sediment

Fig. 6 shows that different positions of the alluvial fan have differentsediment inputs from each layer. Only 10 ± 3% of the sediment in thealluvial fan originated from the surface layer soil in the active gully.This is completely different from the situation in the slumping deposit.Sediment generated from gullies usually contained a large proportionof small particle, 60% of which are finer than 2 mm in diameter (Bettset al., 2003; Derose et al., 1998). Furthermore, more confluent streamsare in an active gully watershed than in a stable gully. Run-off waterconcentrated in rills or depressions, removes soil particles from thegully through sluicing. Approximately 30% of soils in the surface layerare clay and 35% are silt; the particle size for more than 80% of the soilis less than 1 mm in diameter. Therefore, surface layer soils are suscep-tible to runoff in active gully alluvial fans, even though these soils arehighly erosion resistant. Comparatively, in the potentially stable andstable gullies, the surface layer of soil contributes overall mean relative

Fig. 4. Scatter plots constructed from the first and second discriminant functions calculated using DFA in association with the stepwise selection of the optimum composite fingerprint forthe potentially stable gully (G2) and the stable gully (G3) sampling site. (Zn1 means available Zn, Fe1 means acid extractable Fe).

15J. Lin et al. / Catena 129 (2015) 9–17

inputs of 33 ± 5% (stable gully, G2) and 30 ± 3% (potentially stablegully, G3) to the alluvial fan sediment samples. The semi-weatheringrock layer contributed 26± 4% of the input to the sediment in the allu-vial fan, and approximately 70 ± 8% of the overall mean relative inputsof sediment originated from the top two layers in the stable gully. Gully

Table 7Comparison of the composite fingerprint property.

Gullyerosioncatchment

Property Sourcemin

Sourcemax

Sedimentmin

Sedimentmax

G1 Zn (mg/kg) 36.93 622.13 38.93 324.88K (g/kg) 4.23 22.12 7.67 18.65P (g/kg) 0.09 0.23 0.10 0.21Co (μg/kg) 0.61 14.24 3.09 5.88Exchangeable Mn(mg/kg)

10.01 42.36 10.50 21.75

Ni (μg/kg) 0.00 3.45 0.50 1.57Si (g/kg) 27.67 38.89 28.54 36.97Acid extractable Fe(g/kg)

0.41 1.32 0.50 1.12

Reducible Mn (mg/kg) 6.83 36.05 8.02 27.34G2 Mn (mg/kg) 255.12 1147.38 583.33 1082.46

Available Zn (mg/kg) 0.53 3.71 0.84 3.60Cd (μg/g) 0.00 0.13 0.00 0.11V (μg/g) 8.59 19.42 9.52 13.09Mg (mg/kg) 71.98 533.97 174.02 284.76

G3 N (g/kg) 0.07 0.15 0.08 0.13K (g/kg) 12.00 19.18 12.99 17.91Zn (mg/kg) 38.50 197.19 62.32 159.29Acid extracted Fe (g/kg) 6.65 18.46 6.93 12.74Mo (μg/g) 0.11 4.79 0.71 2.46Al (g/kg) 97.16 130.34 98.55 125.62

erosion is usually stabilized by watershed foresting or fluvial incision(Herzig et al., 2011). Stable and potentially stable gullies do not haveupstream confluent streams (Betts et al., 2003). The erosion types aremainly splash and rill erosion, and eroded soils are easily depositedin the alluvial fan rather than running out of the gully. The semi-weathering rock layer contribution to the alluvial fan sediment slowlydecreased from 5–15 m in the distance from the gully wall.

The above results can be usefully in predicting gully wall retreat.However, compared to other studies (such as Collins et al., 1997a,1997b; Walling et al., 2008), the transportation distances between thesediments and their sources are too short. Additionally, the number ofsamples was low compared to other studies conducted withincatchment or river systems. The lack of mixing distance and the limitednumber of samples could affect the source apportionment data and,subsequently, the results.

4. Conclusions

Sediment tracing has been conducted in watershed studies to iden-tify sediment sources. In this study, the same method was employed toidentify transport sediment in different areas of a large, permanentgully. Sediment transportation in active and inactive gullies are differentin terms of soil slumping or runoff from a gully bank. The surface layersoil contributed approximately 70% of the slumping deposit sedimentin the active gully. In potentially stable and stable gullies, becausethere is no gully bank retreat, sediments in the slumping deposit mainlyoriginate from the sandy soil layer and the semi-weathering rock layer.For the alluvial fan, the soil surface layer is more easily washed away inan active gully than in a potentially stable or stable gully. Only 10% of thealluvial fan's sediment originated from the surface layer soils in the ac-tive gully. Through simple analysis of gully sediment sources, the results

Fig. 5.Overall mean relative contribution of each soil layer to the slumping deposit sediment samples collected at different gully stages. (SLmeans surface layer, SSLmeans sand soil layer,SWL means semi-weathering rock layer).

Fig. 6. The overallmean relative contribution of each soil layer to the alluvial fan sediment samples collected at different place of in each gully (5, 10, 15mmeans soil samples of alluvial fancollected from 5, 10 and 15 m distances from gully wall).

16 J. Lin et al. / Catena 129 (2015) 9–17

17J. Lin et al. / Catena 129 (2015) 9–17

of this investigation provide valuable information and initial guidanceon the use of the fingerprinting technique in gully erosion research.Furthermore, this study provides data for a gully erosion model that isdifficult to test directly. Resolving the uncertainties of sediment sourcesin gullies will improve the accuracy of gully models.

Acknowledgments

We gratefully acknowledge the funding provided by the NationalNatural Science Foundation of China (No: 41001169, 40671113), andthe National Science Technology Support Projects (No: 2014BAD15B03).Thanks are also extended to Jiping Chen, Xiaofan Xie andHen Fu for assis-tance with sample analysis and to Qijun Chen for soil sampling.

References

Betts, H.D., DeRose, R.C., 1999. Digital elevation models as a tool for monitoring and mea-suring gully erosion. Int. J. Appl. Earth Obs. Geoinf. 1, 91–101.

Betts, H.D., Trustrum, N.A., De Rose, R.C., 2003. Geomorphic changes in a complex gullysystemmeasured from sequential digital elevationmodels, and implications forman-agement. Earth Surf. Process. Landf. 28, 1043–1058.

Bremner, J.M., Mulvaney, C.S., 1982. Total nitrogen. In: Page, A.L., et al. (Eds.), Methods ofSoil Analysis, Part 2. Chemical and Microbiological Properties, 2nd ed. American Soci-ety of Agronomy, Madison, WI, pp. 595–624.

Butler, D., Memon, F.A., 1999. Dynamic modeling of roadside gully pots during wetweather. Water Res. 33, 3364–3372.

Casali, J., Lopez, J.J., Giraldez, J.V., 2003. A process-based model for channel degradation:application to ephemeral gully erosion. Catena 50, 435–447.

Chaplot, V., Brown, J., Dlamini, P., Eustice, T., Janeau, J.L., Jewitt, G., Lorentz, S., Martin, L.,Nontokozo-Mchunu, C., Oakes, E., Podwojewski, P., Revil, S., Rumpel, C., Zondi, N.,2011. Rainfall simulation to identify the storm-scale mechanisms of gully bank re-treat. Agric. Water Manag. 98, 1704–1710.

Chong, C.W., Tan, G.Y.A., Wong, R.C.S., Riddle, M.J., Tan, I.K.P., 2009. DGGE fingerprinting ofbacteria in soils from eight ecologically different sites around Casey Station. Antarctica.Polar Biology 32, 853–860.

Collins, A.L., Walling, D.E., 2002. Selecting fingerprint properties for discriminating poten-tial suspended sediment sources in river basins. J. Hydrol. 261, 218–244.

Collins, A.L., Walling, D.E., 2007. Sources of fine sediment recovered from the channel bedof lowland groundwater-fed catchments in the UK. Geomorphology 88, 120–138.

Collins, A.L., Walling, D.E., Leeks, G.J.L., 1997a. Source type ascription for fluvial suspendedsediment based on a quantitative composite fingerprinting technique. Catena 29,1–27.

Collins, A.L., Walling, D.E., Leeks, G.J.L., 1997b. Sediment sources in the Upper Severncatchment: a fingerprinting approach. Hydrol. Earth Syst. Sci. 1, 509–521.

Collins, A.L., Walling, D.E., Sichingabula, H.M., Leeks, G.J.L., 2001. Suspended sedimentsource fingerprinting in a small tropical catchment and some management implica-tions. Appl. Geogr. 21, 387–412.

Collins, A.L., Walling, D.E., Webb, L., King, P., 2010a. Apportioning catchment scale sedi-ment sources using a modified composite fingerprinting technique incorporatingproperty weightings and prior information. Geoderma 155, 249–261.

Collins, A.L., Zhang, Y., Walling, D.E., Grenfell, S.E., Smith, P., 2010b. Tracing sediment lossfrom eroding farm tracks using a geochemical fingerprinting procedure combininglocal and genetic algorithm optimization. Sci. Total Environ. 408, 5461–5471.

Collins, A.L., Naden, P.S., Sear, D.A., Jones, J.I., Foster, I.D.L., Morrow, K., 2011. Sediment tar-gets for informing river catchment management: international experience and pros-pects. Hydrol. Process. 25, 2112–2129.

Collins, A.L., Zhang, Y.S., Duethmann, D., Walling, D.E., Black, K.S., 2013. Using a novel trac-ing–tracking framework to source fine-grained sediment loss to watercourses at sub-catchment scale. Hydrol. Process. 27, 959–974.

Conoscenti, C., Angileri, S., Cappadonia, C., Rotigliano, E., Agnesi, V., Maerker, M., 2014.Gully erosion susceptibility assessment by means of GIS-based logistic regression: acase of Sicily (Italy). Geomorphology 204, 399–411.

Deboer, D.H., Crosby, G., 1995. Evaluating the potential of SEM/EDS analysis for finger-printing suspended sediment derived form 2 contrasting topsoils. Catena 24,243–258.

Derose, R.C., Gomez, B., Marden, M., Trustrum, N.A., 1998. Gully erosion inMangatu forest,New Zealand, estimated from digital elevation models. Earth Surf. Process. Landf. 23,1045–1053.

Feng, M.H., Liao, C.Y., Li, S.X., Lu, S.L., 2009. Investigation on status of hill collapsing andsoil erosion in southern China (in Chinese with English abstract). Yangtze River 40,66–75.

Hatfield, R.G., Maher, B.A., 2009. Fingerprinting upland sediment sources: particle size-specific magnetic linkages between soils, lake sediments and suspended sediments.Earth Surf. Process. Landf. 34, 1359–1373.

Herzig, A., Dymond, J.R., Marden, M., 2011. A gully-complex model for assessing gully sta-bilization strategies. Geomorphology 133, 23–33.

Martinez-Casasnovas, J.A., Anton-Fernandez, C., Ramos, M.C., 2003. Sediment productionin large gullies of the Mediterranean area (NE Spain) from high-resolution digital el-evation models and geographical information systems analysis. Earth Surf. Process.Landf. 28, 443–456.

McKeague, J.A., Brydon, J.E., Miles, N.M., 1971. Differentiation of form of extractable ironand aluminum in soils. Soil Sci. Soc. Am. Proc. 35, 33–38.

Mehra, O.P., Jackson, M.L., 1960. Iron oxides removed from soils and clays by a dithionite-citrate system buffered with sodium bicarbonate. Clay Clay Miner. 7, 317–327.

Mukundan, R., Radcliffe, D.E., Ritchie, J.C., Risse, L.M., McKinley, R.A., 2010. Sediment fin-gerprinting to determine the source of suspended sediment in a southern piedmontstream. J. Environ. Qual. 39, 1328–1337.

Pansu, M., Gautheyrou, J., 2006. Handbook of Soil Analysis: Mineralogical, Organic and In-organic Methods. Springer-Verlag, Berlin Heidelberg, Germany.

Peter, K.D., d'Oleire-Oltmanns, S., Ries, J.B., Marzolff, I., Hssaine, A.A., 2014. Soil erosion ingully catchments affected by land-levelling measures in the Souss Basin, Morocco,analyzed by rainfall simulation and UAV remote sensing data. Catena 113, 24–40.

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

Porto, P., Walling, D.E., Capra, A., 2014. Using Cs-137 and Pb-120(ex) measurements andconventional surveys to investigate the relative contributions of inter-rill/rill andgully erosion to soil loss from a small cultivated catchment in Sicily. Soil Tillage Res.135, 18–27.

Prosser, I.P., Slade, C.J., 1994. Gully formation and the role of valley-floor vegetation,southeastern Australia. Geology 22, 1127–1130.

Samani, A.N., Wasson, R.J., Malekian, A., 2011. Application of multiple sediment finger-printing techniques to determine the sediment source contribution of gully erosion:review and case study from Boushehr Province, southwestern Iran. Prog. Phys. Geogr.35, 375–391.

Schuller, P., Walling, D.E., Iroume, A., Quilodran, C., Castillo, A., Navas, A., 2013. Using Cs-137 and Pb-210(ex) and other sediment source fingerprints to document suspendedsediment sources in small forested catchments in south-central Chile. J. Environ.Radioact. 124, 147–159.

Sheng, J.A., Liao, A.Z., 1997. Erosion control in South China. Catena 29, 211–221.Sidorchuk, A., Marker, M., Moretti, S., Rodolfi, G., 2003. Gully erosion modelling and land-

scape response in the Mbuluzi River catchment of Swaziland. Catena 50, 507–525.Smith, H.G., Blake, W.H., 2014. Sediment fingerprinting in agricultural catchments: a crit-

ical re-examination of source discrimination and data corrections. Geomorphology204, 177–191.

Vandekerckhove, L., Poesen, J., Wijdenes, D.O., de Figueiredo, T., 1998. Topographicalthresholds for ephemeral gully initiation in intensively cultivated areas of the Medi-terranean. Catena 33, 271–292.

Voli, M.T., Wegmann, K.W., Bohnenstiehl, D.R., Leithold, E., Osburn, C.L., Polyakov, V.,2013. Fingerprinting the sources of suspended sediment delivery to a large municipaldrinking water reservoir: Falls Lake, Neuse River, North Carolina, USA. J. Soils Sedi-ments 13, 1692–1707.

Walkley, A., Black, A., 1934. An examination of the Detjareff method for determining soilorganicmatter and a proposedmodification of the chromic acid titrationmethod. SoilSci. 63, 251–263.

Walling, D.E., Collins, A.L., Jones, P.A., Leeks, G.J.L., Old, G., 2006. Establishing fine-grainedsediment budgets for the Pang and Lambourn LOCAR catchments, UK. J. Hydrol. 330,126–141.

Walling, D.E., Collins, A.L., Stroud, R.W., 2008. Tracing suspended sediment and particulatephosphorus sources in catchments. J. Hydrol. 350, 274–289.

Walling, D.E., Owens, P.N., Leeks, G.J.L., 1999. Fingerprinting suspended sediment sourcesin the catchment of the River Ouse, Yorkshire, UK. Hydrological Processes 13,955–975.

Watanabe, F.S., Olsen, S.R., 1965. Test of an ascorbic acid method for determining phos-phorus in water and NaHCO3 extracts from soils. Soil Sci. Soc. Am. J. 29, 677–678.

Wilson, H., Cruse, R., Burras, L., 2011. Amethod to adapt watershed-scale sediment finger-printing techniques to small-plot runoff experiments. J. Soil Water Conserv. 66,323–328.