Analysis of runoff, sediment dynamics and sediment yield of subcatchments in the highly erodible...
-
Upload
jose-andres -
Category
Documents
-
view
215 -
download
2
Transcript of Analysis of runoff, sediment dynamics and sediment yield of subcatchments in the highly erodible...
-
ANALYSIS AND MODELLING OF SEDIMENT TRANSFER IN MEDITERRANEAN RIVER BASINS
Analysis of runoff, sediment dynamics and sediment yieldof subcatchments in the highly erodible Isbena catchment,Central Pyrenees
Till Francke & Sandra Werb & Erik Sommerer &Jos Andrs Lpez-Tarazn
Received: 30 September 2013 /Accepted: 18 September 2014 /Published online: 17 October 2014# Springer-Verlag Berlin Heidelberg 2014
AbstractPurpose The Isbena catchment (445 km2), Spain, featureshighly diverse spatial heterogeneity in land use, lithology andrainfall. Consequently, the relative contribution in terms ofwater and sediment yield varies immensely between itssubcatchments, and also temporally. This study presents thesynthesis of ~2.5 years of monitoring rainfall, discharge andsuspended sediment concentration (SSC) in the five mainsubcatchments of the Isbena and its outlet.Materials and methods Continuous discharge at thesubcatchment outlets, nine tipping bucket rainfall and auto-matic SSC samplers (complemented by manual samples),were collected from June 2011 until November 2013. Thewater stage records were converted to discharge using a ratingcurve derived with Bayesian regression. For reconstructingsediment yields, the data from the intermittent SSC samplingneeded to be interpolated. We employed non-parametric mul-tivariate regression (Quantile Regression Forests, QRF) usingthe discharge and rainfall data plus different aggregationlevels of these as ancillary predictors. The subsequent Monte
Carlo simulations allowed the determination of monthly sed-iment yields and their uncertainty.Results and discussion The stagedischarge rating curvesshowed wide credibility intervals for the higher stages, withgreat uncertainties associated with the discharge rates, espe-cially during floods. The water yield of the subcatchmentsdiffered considerably. The entire catchments output was dom-inated by the northernmost subcatchment (~360 mm year1).The smaller, southern subcatchments featured much highervariability and lower runoff rates (55250 mm year1). TheSSCs exhibited a wide range and can exceed 100 g l1 for thecentral subcatchments, where most of the badlands are locat-ed. For the reconstruction of the sedigraphs, the QRF methodproved suitable with NashSutcliffe indices of 0.50 to 0.84.The specific sediment yield ranges from relatively low(32 t km2 year1) in the highly vegetated north to high values(3,651 t km2 year1) in areas with many badland formations.Conclusions The Isbena catchment shows high erosion dy-namics with great variability in space and time, with starkcontrasts even between adjacent subcatchments. The naturalconditions make water and sediment monitoring and instru-mentation very challenging; the measurement of discharge isparticularly prone to considerable uncertainties. The QRFmethod employed for reconstructing sedigraphs and monthlyyields proved well suited for the task.
Keywords Mediterranean-mountainous . Non-parametricregression . Sediment yield .Water yield . Badlands
1 Introduction
Assessing erosion rates and the resulting sediment yield of acatchment is of great importance for a variety of reasons.Beyond the scientific interest in understanding geomorphicand fluvial processes, there are numerous practical aspects that
Responsible editor: Ramon J. Batalla
T. Francke (*) : S. WerbInstitute of Earth and Environmental Sciences, University ofPotsdam, 14476 Potsdam, Germanye-mail: [email protected]
E. SommererHelmholtz Centre Potsdam, GFZ German Research Centre forGeosciences, 14473 Potsdam, Germany
J. A. Lpez-TaraznSchool of Natural Sciences and Psychology, Liverpool John MooresUniversity, L3 3AF Liverpool, UK
J. A. Lpez-TaraznFluvial Dynamics Research Group, Department of Environment andSoil Sciences, University of Lleida, 25198 Lleida, Catalonia, Spain
J Soils Sediments (2014) 14:19091920DOI 10.1007/s11368-014-0990-5
-
require or benefit from proper understanding of sedimentfluxes; for example, high specific sediment yields can indicateprogressive degradation of potentially valuable land, and theirassessment can help in identifying hotspots and remediationmeasures. Sediment transported in a river has considerableimpact on water quality as it tends to absorb contaminants andnutrients. Additionally, sediment deposition highly affects theriverbed, reducing pore water fluxes and the rate of hyporheicexchange (e.g. Wood and Armitage 1997; Packman andMackay 2003), and directly affects aquatic life by alteringhabitat quality and biodiversity (e.g. due to changes in lightingconditions, abrasion, suitability for spawning). Such impactshave been reported for fish spawning (Acornley and Sear1999), for macroinvertebrates (Quinn et al. 1992; Buendiaet al. 2013) and for macrophyte communities (Clarke andWharton 2001). It also causes important economic issues(e.g. channel navigability, integrity of riverside infrastructure,water usability for cooling and irrigation, attractiveness forrecreational use). Finally, in areas such as reservoirs or riverdeltas, large amounts of sediment, or the lack thereof, impairwater storage reliability, hydro-electric facilities or lead toaccelerated degradation of estuaries, respectively. In Mediter-ranean regions, many of these issues are especially prominent,both because of their physiographic settings (e.g. climate,shallow soils, steep terrain; Woodward 1995) and their vul-nerability (e.g. dense population and related pressure). Withinthese landscapes, catchments hosting badland formationsshow sediment yields elevated by several orders of magnitude(Nadal-Romero et al. 2011) throughout a wide range of catch-ment sizes, making them a special challenge for managementissues.
Despite the apparent need for reliable sediment data tosupport management decisions, monitoring sediment fluxfrom a catchment remains a challenging task in many casesbecause it requires the continuous measurement of both dis-charge and sediment concentration. The former includesmanyuncertainties (e.g. errors associated with equipment, establish-ment of discharge rating curve, changes in cross section; seeMcMillan et al. 2012). The latter is mostly restricted tosuspended sediment concentration (SSC). Its continuous mea-surement (i.e. turbidimetry) is affected by further influences(e.g. sensor range, fouling, effects of colour or grain size of thesediment; see Hatcher et al. 2000). Manual or automaticintermittent sampling with off-site analysis yields accuratevalues of SSC regardless of their range, but with low temporalresolution and high labour costs. Furthermore, reconstructingcontinuous sedigraphs from these non-continuous measure-ments is often not easy and stringent, despite recent method-ological progress (see Section 3.2.2) and proposed frame-works that account for the various error sources included(e.g. Navratil et al. 2011).
In this study, we use the data collected during ~2.5 years ofmonitoring water and sediment flux in the Isbena catchment,
Spain, and its tributaries to characterise the hydro-sedimentological regime of a highly active mesoscale catch-ment in regard to space and time. The objectives were (1) todemonstrate the applicability and consistency of a data-drivenmethod for sedigraph reconstruction, (2) the analysis of waterand sediment dynamics and yields and their variability inspace and time in comparison to reported values, and (3) theassessment of related uncertainties.
Further motivation arises from the specific local setting.The River Isbena drains into the River sera in the backwa-ters of the Barasona reservoir, which experiences massivesiltation problems (Avendao et al. 1997) due to the amountof fine sediments transported by both rivers, thus endangeringthe provision of water to approximately 100,000 ha of irrigat-ed land in the Ebro depression (Lpez-Tarazn et al. 2009). Inaddition to its socio-economic importance, the Isbena catch-ment has been the focus of numerous studies because its high-magnitude and heterogeneous sediment response in space andtime poses many hydrological, geomorphological and ecolog-ical questions (e.g. Fargas et al. 1997; Valero-Garcs et al.1999; Navas et al. 2004; Verd et al. 2006; Lpez-Taraznet al. 2012; Buendia et al. 2013).
Among the sediment related issues, a specific (but veryimportant) aspect is the in-channel fine-sediment storage,which for the Isbena catchment has been estimated to be upto 5 % of the annual suspended yield (Lpez-Tarazn et al.2011). This storage displays non-linear dynamics in space andtime and shows marked variations between years (Piqu et al.2014), which cannot be fully understood without extensiveand reliable discharge and sediment data of the contributingsubcatchments. In addition, ongoing research using sedimentfingerprinting approaches (Brosinsky et al. 2014) and terres-trial laser scanning require credible hydro-sedimentologicaldata series. In this context, we demonstrate the application of adata-driven model to reconstruct continuous sedigraphs andits validation. The results give insight into the varying role ofsediment source areas during the monitoring period. More-over, we also provide estimates of sediment yield of thesubcatchments and their relative contributions and discussthe relevant uncertainties in the current monitoring scheme.
2 Study area
The River Isbena drains a mesoscale mountainous catchmentlocated in the Central Pyrenees, NE Iberian Peninsula. It is themain tributary of the River sera; both constitute the mostimportant tributaries of the Cinca basin, in turn the secondlargest tributary of the River Ebro. The Isbena drains an areaof 445 km2 (0.48 % of the Ebro basin). Elevation varies from450 m above sea level (a.s.l.) at the catchment outlet to2,720 m a.s.l. in the northern part. The catchment can bedivided into five main sub-basins: Cabecera (146 km2,
1910 J Soils Sediments (2014) 14:19091920
-
representing 33 % of the total catchment area), Villacarli(42 km2, 9 %), Carrasquero (25 km2, 6 %), Ceguera(28 km2, 6 %) and Lascuarre (45 km2, 10 %) (Table 1,Fig. 1). The rest of the catchment is formed by small creeksthat flow only after rainfall events.
The river shows a rainsnow-fed regime, with very highinter-annual irregularity and remarkable discharge variations.Maximum discharge peaks often occur in spring due to thesnowmelt in the headwaters whereas minimum levels occur insummer. Nevertheless, historical maximum discharges arefound in autumn caused by very intense thunderstorm events.Mean annual discharge is 4.1 m3 s1, with a standard deviation() of 2.2 m3 s1 (0.96 % of the Ebro mean discharge). Themean annual water yield is 177 hm3 (=92 hm3), a value thatrepresents ~1.5 % of the Ebro basins total runoff (Lpez-Tarazn et al. 2009).
The climate of the Isbena catchment is typical of Medi-terranean mountainous areas (pure Mediterranean in his con-tinental variant; Lpez-Tarazn 2011). It is enhanced by thealtitudinal variability, varying from a sub-Mediterranean cli-mate in the southern-lower part to a sub-alpine climate>1,600 m a.s.l. The main characteristic is the important ther-mal contrast, with cold and dry winters, and hot and stormysummers and frequent thunderstorms. Mean precipitationvaries from 450 mm in the lower part to 1,600 mm in thehigher ranges, yielding a mean annual precipitation for thewhole basin of 770 mm, with monthly maximum values inMay and June and minimum rainfall values in July. Meanannual temperature in the catchment varies from 11 to 14 C inthe southern part, and from 9 to 11 C in the northern part.
In the upper part of the basin, the river flows throughnarrow valleys excavated on Cretaceous limestones (i.e. cal-careous rocks) forming a pronounced relief in that area. Ero-sion has left the calcareousmaterials, partially karstified, at thehighest levels of the massifs, with the later Eocene marlsshaping run-down reliefs. These Eocene marls usually out-crop, in the middle part of the basin, as badland structures thathave been identified as the most important source of sedimentduring storm periods, despite representing
-
subcatchments. Water stage was measured mainly at quasi-natural cross sections close to the sub-basin outlets whereconditions were favourable (i.e. bridges); note that theLascuarre gauge features a stable artificial cross section. Themeasurements were conducted with capacitive water-stagesensors/loggers (WT-HR; TruTrack Ltd., Christchurch, NZ)and a microwave stage recorder at Villacarli (RQ 24;Sommer GmbH, Koblach, A) at a resolution of 5 min,complemented by manual readings. Repeated dischargemeasurements at a wide range of flow conditions (seeTable 2) were performed using current meters (C2; OttGmbH, Kempten, D), an acoustic device (ADC; idem)and the salt dilution method for derivation of water-stage rating curves (see Section 3.2.1). For high stagesat the Villacarli gauge, surface velocity measurements werealso performed (RQ 24).
3.1.2 Sediments
The (quasi-)ephemeral runoff dynamics and the large range ofSSC precluded the use of turbidimeters for all subcatchmentsdue to the high maintenance effort. Instead, water sampleswere taken manually and automatically (ISCO sampler;Teledyne Tech. Inc., Thousand Oaks, USA) on an event basis.The sampler was triggered by water stage, initiating the ac-quisition of up to 24 samples with intervals of 15120 min.The predominant fine particle size distribution and the gener-ally shallow and well-mixed flow ensured representative sam-pling, which was verified by comparing samples from differ-ent positions of the cross section. The samples were decantedwhen concentrations were >2 g l1 or vacuum filtered(0.045 mm pore size), oven dried (105 C) and weighed todetermine SSC. For the catchment outlet (Capella),
Fig. 1 The Isbena catchment,Spain, with locations ofsubcatchments and rain and rivergauges (colours correspond toscheme used in Fig. 5) (left);location on the Iberian Peninsula(top right); typical quasi-naturalcross section in Villacarlisubcatchment (bottom right)
Table 2 Summary of monitoring data
Rainfall (mm year1) Discharge (m3 s1) Suspended sediment concentration (SSC) (g l1)
Mean Mean Range Median Range
Cabecera 820 1.58 0.219.0 0.06
-
turbidimeter records of 15 min resolution (Turbimax WCUS41; Endress+Hauser AG, Reinach, CH) calibrated withISCO samples were used.
3.2 Data processing
3.2.1 Discharge
The logger stage records were corrected using the manualstage readings to account for sensor drift. For the Villacarlisubcatchment, the change in zero-flow level wasaccounted for by a time-variant correction based onobservations. Power-law rating curves were fitted tothe measurements using the HYDRASUB software(Reitan and Petersen-verleir 2007), which employs Bayesregression. Applying the rating curves to the stage recordsresulted in time series of discharge, which were subjected tolinear interpolation to fill any gaps and removal of sensorartefacts (Reusser et al. 2012).
3.2.2 Sediment flux
Classically, intermittent SSC samples are used to construct asediment rating curve (SRC; Walling 1977; Asselman 1999;Bhutiyani 2000; Rondeau et al. 2000) which relates SSC todischarge using a power law. Thus, a SRC can be considered aunivariate (i.e. using a single predictor) statistical model forSSC. However, when other drivers apart from discharge gov-ern SSC (e.g. hysteresis, seasonality, exhaustion of sedimentsupply), the corresponding proxies for these processes need tobe included. For that purpose, various advanced statistical andmachine-learning approaches have been applied in recentyears, such as Fuzzy Logic (Kisi et al. 2006), Artificial NeuralNetworks (Nagy et al. 2002) and other multivariate regressionmethods (Schnabel and Maneta 2005; Francke et al. 2008b).Among these, the non-parametric, tree-based techniques ofRandom Forests (RF; Breiman 2001) and the related QuantileRegression Forests (QRF; Meinshausen 2006) haveproven to be efficient for sediment prediction (Franckeet al. 2008a; Zimmermann et al. 2012; Mohr et al.2013) under various hydro-sedimentological settings.This favourable performance can be attributed to theproperties of RF and QRF (e.g. no assumptions on datadistributions, handling of non-linearities, interactions andnon-additive behaviour), and the inclusion of predictors withhigh predictive power. Moreover, the use of QRF, in particu-lar, allows the computation of uncertainties by using thequantiles of the predictions.
Pre-processing of predictors and model building Previousstudies (e.g. Lpez-Tarazn et al. 2012) and field observationshave identified processes such as weathering dynamics, sed-iment exhaustion on the slopes, seasonality and temporary
sediment storage in the riverbed. Therefore, the predictordischarge (of the respective subcatchment) was supplement-ed by the rate of change in discharge (delta_Q), time series ofrainfall (of nearby stations) and the day of the year (DOY).Additionally, ancillary predictors were computed from theserainfall and discharge time series by aggregating theirvalues in windows of increasing size in the past, whichthen serve as a descriptor for the past conditions for asubcatchment (for details, see Zimmermann et al. 2012) andthus act as proxies for the aforementioned processes. Formal-ly, this leads to a multivariate statistical model for SSC:
SSC QRFmodel pi t ; aggr pi ttback ;winsize j 1
which is not only driven by the instantaneous state of apredictor pi (with i=1number of predictors included, in thiscase fourrainfall, discharge DOY and delta_Q) but also byits values aggregated over a period of winsizej shifted backinto the past (with j=1number of aggregation levels, in thiscase 9, 7, 0 and 0, respectively).
The measured SSC concentrations served as training datato build a QRF model for each subcatchment, resulting in fiveindependent QRF models.
Model validation We validated the model based on threemetrics. We used the out-of-bag estimates of the QRFmodel, which provide an unbiased estimate of the errorrate (Breiman 2001), to compute the NashSutcliffeindex (NSOOB) and the root mean squared error(RMSEOOB). This corresponds to a leave-one-outcross-validation, where each single observation is with-held from the training data before making a predictionfor it.
Additionally, we employed an even more rigorous schemeby splitting the dataset into five contiguous parts andperforming a cross-validation, with the RMSE value beingcomputed for each validation period. Both approaches differsignificantly from conventional measures of fit (e.g. R2) thatare much more optimistic and imply unrealistically goodperformance in prediction mode.
Model application Using the continuous time series of pre-dictors as input for the model allowed the reconstruction ofcontinuous hydrographs. Additionally, suspended sedimentyields (SSY) were computed on a monthly basis using aMonte Carlo approach with 250 replicates, as QRFproduces distributions of predictions (instead of singlevalues), which can be randomly drawn from for eachtimestep. Aggregating the 250 realisations of sedigraphsto monthly aggregated SSY values consequently alsoresults in distributions of values for SSY. These distri-butions are used to assess the uncertainty of the esti-mates (for details, see Francke et al. 2008b).
J Soils Sediments (2014) 14:19091920 1913
-
4 Results and discussion
4.1 Monitoring
The monitoring resulted in time series of precipitation andwater stage of >30 months length (see Table 1). Figure 2compares the hydrological activity of the entire catchment interms of monthly peak discharge (Qmax) during the studyperiod with a longer dataset (19982013). The years 2011 and2012 are close to the long-term mean with somewhat lowerflood activity in DecemberFebruary (2011, 2012) and Au-gustSeptember (2012). The year 2013 had values of Qmaxabove average until September, with outstanding peak valuesin January and August. The values for mean monthly dis-charge display a similar pattern (not shown).
Rating curves were established for all gauges, althoughtheir robustness differed markedly; in particular, for thegauges with high dynamics in the riverbed (e.g. Villacarli),Bayes credibility intervals of the coefficients of the ratingcurve were wide. This effect was especially influential forthe estimation of discharge at high stages, where the credibil-ity interval of the estimated discharge can become very wide[e.g. Villacarli, medium flood (stage 0.5 m), credibility inter-val from 60 to 190 % of median Bayes estimate]. On the otherhand, scour and deposition at cross sections also affected low-flow estimates considerably by shifting the zero-flow stage.
Except for the Villacarli and Capella catchments, we expe-rienced some technical difficulties in the period January toMarch 2013, which compromised the quality of the dischargedata. The resulting implications are discussed in Section 4.
Table 2 illustrates the considerable spatial heterogeneity ofrainfall throughout the catchment, which roughly follows theorographic gradient. The resulting discharge displays a widerange, with all but the Cabecera catchment experiencing periodsof flow below the detection limit, being a result of both higherrainfall and cooler climate with less evapotranspiration, and agreater catchment area.
At all gauges, high (>1 g l1) or very high SSCs (>100 g l1)were observed. Range and median values of SSC stronglycorrelate with the areal percentage of badlands (Table 1), whichare the major primary source of sediments (Fargas et al. 1997).The catchments of Villacarli and Carrasquero, in particular,show very high median and maximum values, which testifyto the challenges encountered during monitoring.
Hysteresis between discharge and SSCwas pronounced forall gauges. However, its strength and direction can vary im-mensely, thus requiring advanced regression techniques forsedigraph reconstruction. Possible explanations for this be-haviour have been discussed in Lpez-Tarazn et al. (2009).
4.2 Sedigraph reconstruction and yield computation
4.2.1 Model building and validation
For all subcatchments, the generated QRF models yield a goodor decent fit to the observations, as attested by the values of theNashSutcliffe index (Table 3). The best fit is achieved forCeguera and Lascuarre (see Fig. 3). The uncertainty of themodelprediction generally increases with SSC, as indicated by thewhiskers in Fig. 3, while lowSSCs are predictedwith less scatter.
The RMSE values for the validation periods cover a range of~50335 % of the values of RMSEOOB. This means that themodels can predict the values of some of the periods well, evenif their data were withheld frommodel training. Conversely, forother periods, the model performance deteriorates considerablywhen their data are not included in the training. This suggeststhat these periods contain special hydro-sedimentological con-ditions, which have not been observed in the other periods.Thus, by excluding this information, the model is unable toperformwell under these conditions. Consequently, high valuesin RMSEvalid period of a period attest to increased informationcontent in the respective data, whereas low values indicate datawhich are (partially) redundant for the model.
4.2.2 Model application
Temporal dynamics (duration curves) The temporal distribu-tion of both water and sediment flux is summarised in Fig. 4
2 4 6 8 10 12
020
4060
8012
0
month
Qm
ax [m
/s]
meanrange201120122013
Fig. 2 Flood activity at the catchment outlet (expressed as monthly peakdischarge, Qmax) of the monitored years (20112013) compared torecords of 19982013
Table 3 Performance measures for QRF models
Gauge NSOOB RMSEOOB Range (RMSEvalid period)a
Cabecera 0.75 0.17 0.090.33
Carrasquero 0.50 6.58 3.4614.28
Ceguera 0.84 2.32 3.97.28
Lascuarre 0.77 2.43 1.538.13
Villacarli 0.72 44.1 24.9147.74
a Range of RMSE values computed for the five periods employed in thecross-validation
1914 J Soils Sediments (2014) 14:19091920
-
using duration curves (cumulated time versus the cumulatedordered flux, both expressed as fractions of the total). The morea curve deviates from the 1:1 line, the more episodic (i.e. vari-able) is the flux. The catchment outlet at Capella and thesubcatchments of Villacarli and Cabecera display a steadierhydrological regime, while Ceguera and Carrasquero showmoredynamical conditions. Likewise, sediment flux shows compara-ble variability for Cabecera and Villacarli (steady), while for the
other subcatchments it is considerablymore concentrated in time.This is especially remarkable for the catchment outlet at Capella,featuring a relatively balanced flow regime, while sediment fluxis more concentrated in time than at all other subcatchments;90 % of the sediment is transported in
-
catchments have different orders of magnitude. Low monthlyyields, mainly 5 times of the mean.Monthswith relatively low or high yield did not always correlateamong catchments (e.g. September 2011high yield forCarrasquero, low yield for Lascuarre). However, Januaryand March 2013 had high yields for all catchments. Thegeneral tendency of higher sediment yields in 2013 is alsoapparent for most catchments.
The uncertainty in SSYas computed with the QRF methodis illustrated by the whiskers plots, which increase with thevalues of monthly yields not only in absolute but also in
relative terms; they are an indicator of model-related uncer-tainty in the yield estimates. This shows that during times oflow dynamics, the QRF model is especially successful inproducing precise estimates. During periods of high sedimenttransport, there is considerably more uncertainty due to themore complex interaction of different drivers that may havecaused high concentrations.
Figure 6 provides a comparison of the relative contributionof water and sediment to the total output of the catchment. Interms of water yield, Cabecera (being the largest and mosthumid subcatchment) is clearly reflected in its dominatingcontribution to the total water yield (68 %). Villacarli contrib-utes the second largest share, while the other subcatchmentsgenerate only low total runoff. This pattern is basically con-sistent for all months observed. Concerning sediment yield,the proportions are almost inverse; the yield of Cabecera ranksamong the lowest, while the other catchments generate rough-ly the same or more sediment. In total, >80 % of all sediment
0200
400
600
800
SS
Y [
t]
Ju
l 11
Sep 1
1
Nov 1
1
Ja
n 1
2
Mar 1
2
May 1
2
Ju
l 12
Sep 1
2
Nov 1
2
Ja
n 1
3
Mar 1
3
May 1
3
Ju
l 13
Sep 1
3
150
50
0
rain
fall [
mm
]
Cabecera
dis
ch
arg
e [
m/s
]
05
10
15
05000
10000
15000
SS
Y [
t]
Ju
l 11
Sep 1
1
Nov 1
1
Ja
n 1
2
Mar 1
2
May 1
2
Ju
l 12
Sep 1
2
Nov 1
2
Ja
n 1
3
Mar 1
3
May 1
3
Ju
l 13
Sep 1
3
150
100
50
0
rain
fall [
mm
]
Carrasquero
dis
ch
arg
e [
m/s
]
05
10
01000
3000
5000
SS
Y [
t]
Ju
l 11
Sep 1
1
Nov 1
1
Ja
n 1
2
Mar 1
2
May 1
2
Ju
l 12
Sep 1
2
Nov 1
2
Ja
n 1
3
Mar 1
3
May 1
3
Ju
l 13
Sep 1
3
100
50
0
rain
fall [
mm
]
Ceguera
dis
ch
arg
e [
m/s
]
010
20
0500
1500
2500
SS
Y [
t]
Ju
l 11
Sep 1
1
Nov 1
1
Ja
n 1
2
Mar 1
2
May 1
2
Ju
l 12
Sep 1
2
Nov 1
2
Ja
n 1
3
Mar 1
3
May 1
3
Ju
l 13
Sep 1
3
100
50
0
rain
fall [
mm
]
Lascuarre
dis
ch
arg
e [
m/s
]
05
10
05000
15000
25000
SS
Y [
t]
Ju
l 11
Sep 1
1
Nov 1
1
Ja
n 1
2
Mar 1
2
May 1
2
Ju
l 12
Sep 1
2
Nov 1
2
Ja
n 1
3
Mar 1
3
May 1
3
Ju
l 13
Sep 1
3
150
100
50
0
rain
fall [
mm
]
Villacarli
dis
ch
arg
e [
m/s
]
05
10
15
020000
60000
SS
Y [
t]
Ju
l 11
Sep 1
1
Nov 1
1
Ja
n 1
2
Mar 1
2
May 1
2
Ju
l 12
Sep 1
2
Nov 1
2
Ja
n 1
3
Mar 1
3
May 1
3
Ju
l 13
Sep 1
3
150
100
50
0
rain
fall [
mm
]
Capella
dis
ch
arg
e [
m/s
]
050
10
0
Fig. 5 Discharge, monthly rainfall and reconstructed monthly sediment yield of subcatchments. Thewhiskers denote the mean1 standard deviations ofmodel predictions. For Capella, turbidimeter data were used
1916 J Soils Sediments (2014) 14:19091920
-
comes from Villacarli; it also dominates all monthly yields.The unusually high contribution of Carrasquero in January 13may partly be caused by the aforementioned data loss duringthat period.
The monitored subcatchments cover approximately twothirds of the entire catchment. Likewise, their combined wateryield is close to that fraction in relation to the total catchmentoutput, i.e. approximately one third deficit at the outlet.However, for some months this deficit is negative. On amonthly scale, it is closely related to the total water yield;R(Deficitwater, Yieldwater)=0.91. This means that the outletsubcatchment itself contributes no or little water under dry toregular conditions. Instead, transmission losses occur in thefloodplain. This fact is consistent with the relatively low annualrunoff of Lascuarre (Table 4), which shows comparable land-scape attributes. Conversely, during wet periods the outletsubcatchments contribution increases disproportionally, whiletransmission losses become negligible.
An analogous effect can be observed for sediment in therelation between monthly SSYand deficit [R(Deficitsediment,Yieldsediment=0.96)], although the yields and deficits are only
loosely correlated [R(Yieldwater, Yieldsediment)=0.54, R-(Deficitwater, Deficitsediment)=0.58]. On one hand, this maybe caused by sediment generation in the outlet subcatchmentitself. On the other hand, especially the effect of partialdecoupling of water and sediment transport can partly beexplained by the role of sediment storage (Lpez-Taraznet al. 2011); the sediment accumulates mainly in the monthsof lower sediment flux (e.g. DecemberFebruary 2011, AprilSeptember 2012). It is then preferentially re-mobilised duringtimes of high sediment flux (e.g. MarchJuly 2013). In thatregard, the temporary sediment storage in the river would notdampen, but increase variability in sediment flux, as alreadyseen in the duration curves (Fig. 4).
Apart from these systematic discrepancies, the beginningof 2013 was affected by occasional logger failures, which mayhave caused the loss of data for some floods. Total riverfreezing of some tributaries during winter 20122013 is an-other error factor for this period. As these uncertainties in thedischarge data also propagate to the sedigraph reconstruction,both water and sediment yields for this period should betreated with caution.
Specific yields For better comparability, the total yields ofwater and sediments have been normalised to area and timein Table 4. Again, the high spatial differences become appar-ent; while water yield generally follows the orographic/climatic gradient, sediment yield behaves differently andmay differ by the factor of 100 (i.e. the adjacentsubcatchments of Villacarli and Cabecera). Thus, when com-pared to other Mediterranean mesoscale catchments (Table 5),both subcatchments represent ends of the observed range. Inthat context, the SSY of Carrasquero and that of the entireIsbena catchment also rank as high, while the othersubcatchments show low tomoderate values.When comparedto catchments with badlands, Villacarlis SSYexceeds that forall other catchments of comparable size collected in the studyof Nadal-Romero et al. (2011). However, at the scale of theentire Isbena catchment, the high yield of Villacarli is aver-aged out, putting the resulting yield into the range observedfor other badland catchments.
Figure 7 illustrates the spatial distribution of specific sedi-ment yield. The sediment yield value corresponds closely to theareal fraction of badlands within the various subcatchments (cf.Table 1, R=0.98), which supports the notion of badlands beingthe primary source of sediments. There are no apparent corre-lations (R>0.5) to any other land use class. However, the denservegetation cover in Cabecera (woodland, pastures) seems tooffset its stronger forcing by rainfall, when compared to thelower lying catchments with less annual rainfall but a higherareal fraction of bare and arable lands.
Error sources Navratil et al. (2011) identified nine sources ofuncertainty in sediment monitoring, of which four are
Fig. 6 Comparison of monthly water and sediment yields (November2013 omitted because of incomplete data). Suspended sediment yield(SSY) for Capella for January 13 is 98,920 t
Table 4 Specific annual yield of water and sediment (based on data fromJuly 2011 to October 2013)
Catchment Suspended sediment yield(SSY) (t km2 year1)
Water yield(mm year1)
Cabecera 32 357
Carrasquero 627 246
Ceguera 361 170
Lascuarre 82 55
Villacarli 3651 253
Capella 517 269
J Soils Sediments (2014) 14:19091920 1917
-
exclusively related to the use of turbidimeters. Of the remain-ing five sources, the one related to the closing of data gaps (i.e.interpolation) has been dealt with in our study, explicitly withour approach. Navratil et al. (2011) proved the source of errorassociated with the representativity of the point SSC mea-surement to be negligible, which we also assume, as sedi-ments were mostly of fine grained and flow was turbulent andwell mixed in our case. As for the uncertainty arising from thefield sampling procedure and laboratory procedure, theformer is reportedly the more significant source of error; thiscomprises effects caused by time-delayed or selective sam-pling of automatic samplers in the context of purging intakehoses or associated contamination. We acknowledge that allthe automatically taken samples in our study may be subject tothis issue.
Finally, errors in discharge estimation are especially dif-ficult to assess (Navratil et al. 2011). They arise from uncer-tainty in metering readings, issues associated with the water-stage sensor [clogging, effects of temperature, salinity oncapacitive measurements (e.g. Larson and Runyan 2009), datagaps] and uncertainties in the stagedischarge rating curve(e.g. hysteresis, unstable cross section). While we were unableto include them in the final computation of errors, their as-sessment within the Bayesian framework suggested a hugepotential influence. When varying the stagedischarge curveof Villacarli within the 95 % confidence interval, the resulting
Table 5 Specific sediment yield (SSY) of (circum-)Mediterranean mesoscale catchments (100104 km2)
Catchment Country Dominant land cover Area (km2) SSY (t km2 year1) Source
Dronne upstream catchment France Agricultural 1,100 813 1
Ribera Salada Spain Woodland 224 1235 2
Save catchment France Agricultural 1,110 1570 3
Gers catchment France Agricultural 970 41 4
Tordera River Spain Woodland 894 50 5
Arige upstream catchment France Agricultural 1,172 5759 1
Bas catchment France Agricultural 1,330 63 4
Bou Namoussa Algeria Dry desert 575 270 5
Bs catchment France Woodland 165 402 6
Upper Llobregat basin Spain 532 430 7
Nahal Eshtemoa catchment Israel Dry desert 119 433 8
Leham river Algeria Dry desert 470 2,028 5
Sra river Morocco Dry desert 493 3,500 5
Mediterranean basins Iberian Peninsula 100200 9
44 Italian catchments Italy 133,067 11661 10
Western humid areas of Jordan valley Israel 100103 16840 11
Mediterranean basins Chile, USA (CA) 100103 2501,500 11
61 Spanish catchments Spain 3013,000 102,600 13
Badland catchments Mediterranean With badlands 101706 2110,090 14
1Veyssy 1998 in Oeurng et al. 2010, 2Vericat and Batalla 2010, 3Oeurng et al. 2010, 4Maneux et al. 2001 in Oeurng et al. 2010, 5 Rovira and Batalla2006, 6Milliman and Syvitski 1992 inWoodward 1995, 7Navratil et al. 2010, 8 CEDEX 2002 in Gallart et al. 2013, 9 Reid et al. 1998 in Thornes et al.2009, 10 Oeurng et al. 2010, 11 de Vente et al. 2006, 12 Vericat and Batalla 2010, 13 de Vente et al. 2008, 14 Nadal-Romero et al. 2011
Fig. 7 Specific water and sediment yield. Note that the outlet representsthe entire Isbena catchment, while Lower Isabena refers to the outletsubcatchment
1918 J Soils Sediments (2014) 14:19091920
-
error ranges for monthly SSY increased by up to 360 % (30 %on average). This increase is a direct result of computing thesediment loads as the product of altered discharge and SSC.That said, uncertain discharge data as a predictor in the QRFmodel (both in its training and prediction phase) is not yetincluded. It would be desirable to account for these uncer-tainties in a consistent framework, including the errors indischarge, SSC and their temporal and mutual structure.
5 Summary and concluding remarks
This study presents the results of ~2.5 years of monitoringwater and sediment fluxes in a mesoscale mountainous catch-ment and its subcatchments. The monitoring of both runoffand sediment flux involved considerable challenges becauseof issues associated with natural cross sections, violent floodregimes and very high sediment concentrations.
With regard to the stated objectives:
(1) The aforementioned challenges were addressed with arating curve based on Bayesian regression and a non-parametric multivariate regression technique to recon-struct the sedigraph. The subsequent computation ofmonthly and total sediment yields and their uncertaintyproved robust and plausible when compared to the com-bined yields at the outlet. For the training data, a widerange of hydro-sedimentological situations need to becovered to allow for adequate training of the model.
(2) The results showed that sediment flux is extremely vari-able in space and time. No apparent relation between theyields (water and sediment) and their variability wasdetected. The main sediment sources are highly concen-trated and correspond to parts of the catchment with a highfraction of badlands. When compared to literature data,the Villacarli subcatchment generates very large amountsof sediments, well in excess of 3,500 t km2 year1, whilethe adjacent Cabecera subcatchment displays low valuesof 32 t km2 year1. The sediment yields of the othersubcatchments rank as moderate to high.
(3) Due to their highly dynamic behaviour, the periods ofhigh SSY are associated with the largest model uncer-tainty. This uncertainty is an optimistic estimate of thetotal uncertainty of sediment yield calculation. In thatcontext, the largest portion of uncertainty originates inthe discharge data.
Acknowledgements This research was carried out within the projectGeneration, transport and retention of water and suspended sediments inlarge dryland catchments: Monitoring and integrated modelling of fluxesand connectivity phenomena (BR 1731/4-1-2) funded by the DeutscheForschungsgemeinschaft (DFG). We gratefully acknowledge the field-work conducted by numerous students. The fourth author was supported
by the SCARCE-CONSOLIDER project (ref. CSD2009-00065) fundedby the Spanish Ministry of Economy and Competitiveness. The authorsalso wish to thank the EbroWater Authorities for permission to install themeasuring equipment at the Capella gauging station and for providinghydrological data.
References
Acornley RM, Sear DA (1999) Sediment transport and siltation of browntrout (Salmo trutta L.) spawning gravels in chalk streams. HydrolProcess 13:447458
Alczar J, Ferrn I (1998) La vegetacin de ribera de los ros sera yCinca en el tramo afectado por el vaciado del embalse de JoaqunCosta. Limntica 14:7382
Asselman NEM (1999) Suspended sediment dynamics in a large basin:the River Rhine. Hydrol Process 13:14371450
Avendao C, Cobo R, Sanz ME, Gmez JL (1997) Capacity situation inSpanish reservoirs. Transactions of the International Congress onLarge Dams. ICOLD, Paris, pp 849862
Avendao C, Sanz ME, Cobo R (2000) State of the art of reservoirsedimentation management in Spain. In: Proceedings of theInternational Workshop and Symposium on ReservoirSedimentation Management, Water Resources EnvironmentTechnology CenterJapan (WEC), Tokyo, Japan, pp 2735
Bhutiyani MR (2000) Sediment load characteristics of a proglacial streamSiachen Glacier and the erosion rate in Nubra valley in KarakoramHimalayas, India. J Hydrol 227:8492
Breiman L (2001) Random Forests. Mach Learn 45:532Brosinsky A, Foerster S, Segl K, Lpez-Tarazn J, Piqu G, Bronstert
(2014) A spectral fingerprinting: characterizing suspended sedimentsources by the use of VNIR-SWIR spectral information. J SoilsSediments. doi:10.1007/s11368-014-0927-z (this issue)
Buendia C, Gibbins CN, Vericat D, Batalla RJ, Douglas A (2013)Detecting the structural and functional impacts of fine sediment onstream invertebrates. Ecol Indic 25:184196
Clarke SJ, Wharton G (2001) Sediment nutrient characteristics andaquatic macrophytes in lowland English rivers. Sci Tot Environ266:103112
de Vente J, Poesen J, Azzoffi P, Van Rompaey A, Verstraeten G, Oesen J(2006) Predicting catchment sediment yield in Mediterranean envi-ronments: the importance of sediment sources and connectivity inItalian drainage. Earth Surf Process Landf 31:10171034
de Vente J, Poesen J, Verstraeten G, Van Rompaey A, Govers G (2008)Spatially distributed modelling of soil erosion and sediment yield atregional scales in Spain. Glob Planet Chang 60:393415
Fargas D, Martnez-Casasnovas JA, Poch R (1997) Identification ofcritical sediment source areas at regional level. Phys Chem Earth22:355359
Francke T (2009) Measurement and Modelling of Water and SedimentFluxes in Meso-Scale Dryland Catchments. 145 pp., PhD thesis,University of Potsdam, Potsdam, Germany, http://nbn-resolving.de/urn:nbn:de:kobv:517-opus-31525
Francke T, Lpez-Tarazn JA, Vericat D, Bronstert A, Batalla RJ (2008a)Flood-based analysis of high-magnitude sediment transport using anon-parametric method. Earth Surf Process Landf 33:20642077
Francke T, Lpez-Tarazn JA, Schrder B (2008b) Estimation ofsuspended sediment concentration and yield using linear models,random forests and quantile regression forests. Hydrol Process 22:48924904
Gallart F, Prez-Gallego N, Latron J, Catari G, Martnez-Carreras N,Nord G (2013) Short- and long-term studies of sediment dynamicsin a small humid mountain Mediterranean basin with badlands.Geomorphol 196:242251
J Soils Sediments (2014) 14:19091920 1919
http://dx.doi.org/10.1007/s11368-014-0927-zhttp://nbn-resolving.de/urn:nbn:de:kobv:517-opus-31525http://nbn-resolving.de/urn:nbn:de:kobv:517-opus-31525
-
Hatcher A, Hill P, Grant J, Macpherson P (2000) Spectral optical back-scatter of sand in suspension: effects of particle size, compositionand colour. Mar Geol 168:115128
Kisi O, Emin Karahan M, Sen Z (2006) River suspended sedimentmodelling using a fuzzy logic approach. Hydrol Process 20:43514362
Larson P, Runyan C (2009) Evaluation of a capacitance water levelrecorder and calibration methods in an urban environment. 36 pp.Reporthttp://www.umbc.edu/cuere/BaltimoreWTB/pdf/TM_2009_003.pdf
Lpez-Tarazn JA (2011) The sediment budget of a highly erodiblecatchment. The River Isbena (Ebro Basin, Central Pyrenees).319 pp, Dissertation, University of Lleida, Spain
Lpez-Tarazn JA, Batalla RJ, Vericat D, Francke T (2009) Suspendedsediment transport in a highly erodible catchment: the River Isbena(Southern Pyrenees). Geomorphol 109:210221
Lpez-Tarazn JA, Batalla RJ, Vericat D (2011) In-channel sedimentstorage in a highly erodible catchment: the River Isbena (EbroBasin, Southern Pyrenees). Zeitschrift fr Geomorphol 55:365382
Lpez-Tarazn JA, Batalla RJ, Vericat D, Francke T (2012) The sedimentbudget of a highly dynamic mesoscale catchment: the River Isbena.Geomorphol 138:1528
Mamede G (2008) Reservoir sedimentation in dryland catchments:modelling and management. PhD thesis, University of Potsdam,Potsdam, Germany, http://opus.kobv.de/ubp/volltexte/2008/1704/
Martnez-Casasnovas JA, Poch RM (1998) Estado de conservacin de lossuelos de la cuenca del embalse Joaqun Costa. Limntica 14:8391
McMillan H, Krueger T, Freer J (2012) Benchmarking observationaluncertainties for hydrology: rainfall, river discharge and water qual-ity. Hydrol Process 26:40784111
Meinshausen N (2006) Quantile Regression Forests. J Mach Learn Res 7:983999
Mohr CH, Zimmermann A, Korup O, Iroum A, Francke T, Bronstert A(2013) Seasonal logging, process response, and geomorphic work.Earth Surf Dyn Discuss 1:311335
Nadal-Romero E, Martinez-Murillo JF, Vanmaercke M, Poesen J (2011)Scale-dependency of sediment yield from badland areas inMediterranean environments. Prog Phys Geogr 35:297332
Nagy HM, Watanabe K, Hirano M (2002) Prediction of sediment loadconcentration in rivers using artificial neural network model. JHydraul Eng 128:588595
Nash JE, Sutcliffe JV (1970) River flow forecasting through conceptualmodels. Part I: a discussion of principles. J Hydrol 10:282290
Navas A, Valero-Garcs BL, Machn J (2004) Research Note: An ap-proach to integrated assessement of reservoir siltation: the JoaqunCosta reservoir as a case study. Hydrol Earth Syst Sci 8:11931199
Navratil O, Legout C, Gateuille D, Esteves M, Liebault F (2010)Assessment of intermediate fine sediment storage in a braided riverreach (southern French Prealps). Hydrol Process 24:13181332
Navratil O, Esteves M, Legout C, Gratiot N, Nemery J, Willmore S,Grangeon T (2011) Global uncertainty analysis of suspended sedi-ment monitoring using turbidimeter in a small mountainous rivercatchment. J Hydrol 398:246259
Oeurng C, Sauvage S, Snchez-Prez J-M (2010) Dynamics ofsuspended sediment transport and yield in a large agricultural catch-ment, southwest France. Earth Surf Process Landf 35:12891301
PackmanAI,Mackay JS (2003) Interplay of stream-subsurface exchange,clay deposition and stream bed evolution.Wat Resour Res 39:4149
Palau A (1998) Estudio limnolgico del ecosistema fluvial afectado porlos vaciados del embalse de Barasona. Limntica 5:115
Piqu G, Lpez-Tarazn J, Batalla R (2014) Variability of in-channelsediment storage in a river draining highly erodible areas theIsbena, Ebro Basin. J Soils Sediments. doi:10.1007/s11368-014-0957-6 (this issue)
Poch RM, Martnez-Casasnovas JA (1997) Prevention of reservoir silta-tion in large watersheds: from sediment source identification to thedesign of soil conservation measures. J Soil Wat Conserv 52:285286
Quinn JM, Davies-Colley RJ, Hickey CW, Vickers ML, Ryan PA (1992)Effects of clay discharge on streams, 2: benthic invertebrates.Hydrobiologia 248:235247
Reitan T, Petersen-verleir A (2007) Bayesian power-law regressionwith a location parameter, with applications for construction ofdischarge rating curves. Stoch Environ Res Risk Assess 22:351365
Reusser DE, Buytaert W, Vitolo C (2012) RHydroHydrologicalmodels and tools to represent and analyze hydrological data in R.Geophys Res Abstr 14,EGU2012:4166
Rondeau B, Cossa D, Gagnon P, Bilodeau L (2000) Budget and sourcesof suspended sediment transported in the St. Lawrence River,Canada. Hydrol Process 14:2136
Rovira A, Batalla RJ (2006) Temporal distribution of suspended sedimenttransport in aMediterranean basin: The Lower Tordera (NE SPAIN).Geomorphol 79:5871. doi:10.1016/j.geomorph.2005.09.016
Schnabel S, Maneta M (2005) Comparison of a neural network and aregression model to estimate suspended sediment in a semiaridbasin. In: Batalla RJ, Garcia C (eds) Geomorphological processesand human impacts in river basins, IAHS Publ 299. IAHS,Wallingford, pp 91100
Thornes JB, Lpez-Bermdez F, Woodward JC (2009) Hydrology, riverregimes, and sediment yield. In: Woodward JC (ed) The physicalgeography of the Mediterranean. Oxford University Press, Oxford,pp 229253
Valero-Garcs BL, Navas A, Machn J, Walling D (1999) Sedimentsources and siltation in mountain reservoirs: a case study from theCentral Spanish Pyrenees. Geomorphol 28:2341
Verd JM, Batalla RJ, Martnez-Casasnovas JA (2006) Estudiohidrolgico de la cuenca del ro Isbena (Cuenca del Ebro). I:Variabilidad de la precipitacin. Ing del Agua 13:321330
Vericat D, Batalla RJ (2010) Sediment transport from continuous moni-toring in a perennial Mediterranean stream. Catena 82:7786
Walling DE (1977) Limitations of the rating curve technique for estimat-ing sediment loads, with particular reference to British rivers. In:Erosion and SolidMatter Transport in InlandWater, IAHS Publ 122,IAHS, Wallingford, UK, pp 3438
Wood PJ, Armitage PD (1997) Biological effects of fine sediment in thelotic environment. Env Manag 21:203217
Woodward JC (1995) Patterns of erosion and suspended sediment yield inMediterranean river basins. In: Foster IDL, Gurnell AM, Webb B(eds) Sediment and water quality in river catchments. Wiley,Chichester, pp 365389
Zimmermann A, Francke T, Elsenbeer H (2012) Forests and erosion:insights from a study of suspended-sediment dynamics in an over-land flow-prone rainforest catchment. J Hydrol 428429:170181
1920 J Soils Sediments (2014) 14:19091920
http://www.umbc.edu/cuere/BaltimoreWTB/pdf/TM_2009_003.pdfhttp://www.umbc.edu/cuere/BaltimoreWTB/pdf/TM_2009_003.pdfhttp://opus.kobv.de/ubp/volltexte/2008/1704/http://dx.doi.org/10.1007/s11368-014-0957-6http://dx.doi.org/10.1007/s11368-014-0957-6http://dx.doi.org/10.1016/j.geomorph.2005.09.016
Analysis...AbstractAbstractAbstractAbstractAbstractIntroductionStudy areaMethodsMonitoringWaterSediments
Data processingDischargeSediment flux
Results and discussionMonitoringSedigraph reconstruction and yield computationModel building and validationModel application
Summary and concluding remarksReferences