This article appeared in a journal published by Elsevier. The attached copy is furnished to the author for internal non-commercial research and education use, including for instruction at the authors institution and sharing with colleagues. Other uses, including reproduction and distribution, or selling or licensing copies, or posting to personal, institutional or third party websites are prohibited. In most cases authors are permitted to post their version of the article (e.g. in Word or Tex form) to their personal website or institutional repository. Authors requiring further information regarding Elsevier’s archiving and manuscript policies are encouraged to visit: http://www.elsevier.com/copyright

Author's personal copy

Journal of Hydrology 398 (2011) 246–259

Contents lists available at ScienceDirect

Journal of Hydrology journal homepage: www.elsevier.com/locate/jhydrol

Global uncertainty analysis of suspended sediment monitoring using turbidimeter in a small mountainous river catchment O. Navratil a,⇑, M. Esteves a, C. Legout b, N. Gratiot a, J. Nemery c, S. Willmore a, T. Grangeon b a

Laboratoire d’étude des Transferts en Hydrologie et Environnement (LTHE) – Université Grenoble 1/IRD, BP 53, 38041-Grenoble Cedex 9, France Laboratoire d’étude des Transferts en Hydrologie et Environnement (LTHE) – Université Grenoble 1, BP 53, 38041-Grenoble Cedex 9, France c Laboratoire d’étude des Transferts en Hydrologie et Environnement (LTHE) – Université Grenoble 1/G-INP, BP 53, 38041-Grenoble Cedex 9, France b

a r t i c l e

i n f o

Article history: Received 28 June 2010 Received in revised form 11 October 2010 Accepted 22 December 2010 Available online 31 December 2010 This manuscript was handled by A. Bardossy, Editor-in-Chief, with the assistance of Erwin Zehe, Associate Editor Keywords: Monte Carlo simulations Turbidity Turbidimeter calibration Pumping sampler Sediment yield

s u m m a r y A major challenge confronting the scientific community is to understand both patterns of and controls over spatial and temporal variability of suspended sediment dynamics in rivers, as these sediment govern nutriment export, river morphology, siltation of downstream reservoirs and degradation of water quality. High-frequency suspended sediment monitoring programs are required to meet this goal, particularly research in highly erodible mountainous catchments which supply the sediment load of the entire downstream fluvial network. However, in this context, analysis of the data and their interpretation are generally limited by many sources of uncertainty in river monitoring. This paper proposes to estimate the global uncertainty of suspended sediment monitoring using turbidimeter in a small mountainous river catchment (22 km2; Southern French Alps). We first conducted a detailed analysis of the main uncertainty components associated with the turbidity approach, i.e. a widely used method to continuously survey the suspended sediment concentration (SSC). These uncertainty components were then propagated with Monte Carlo simulations. For individual records, SSC uncertainties are found to be on average less than 10%, but they can reach 70%. At the flood scale, the mean and the maximum SSC uncertainties are on average 20% (range, 1–30%), whereas sediment yield uncertainty is a mean 30% (range, 20–50% depending on the flood considered; discharge error, 20%). Annual specific sediment yield (SSY) was then 360 ± 100 t km2 year1. Uncertainty components associated with the automatic pumping procedure, discharge measurement and turbidity fluctuation at the short time scale were found to be the greatest uncertainties. SSC and SSY uncertainties were found highly site- and time-dependent as they vary significantly with the hydro-sedimentary conditions. This study demonstrates that global uncertainty accounts for only a small part of inter-flood SSC and SSY variability. It outlines the controlling factors of land use, relief, geology and rainfall regime on suspended sediment yields. Ó 2011 Elsevier B.V. All rights reserved.

1. Introduction Improving knowledge on suspended sediment yields, dynamics and water quality is one of today’s major environmental challenges addressed to scientists and hydropower managers (Owens et al., 2005). These advances will continue in the future as the acquisition of reliable and long-term suspended sediment concentration (SCC) time series are generalised to many hydrometric stations. In mountainous catchments, major fractions of the annual suspended sediment yields (SSY) are transported over a very short time period generally corresponding to several floods (e.g. Meybeck et al., ⇑ Corresponding author. Address: Cemagref Grenoble, Unité de Recherche ETNA (Erosion Torrentielle, Neige et Avalanches), Domaine Universitaire, 2 rue de la Papeterie, BP76, 38402 Saint-Martin-d’Hères, France. Tel.: +33 4 76 63 55 39; fax: +33 4 76 82 50 01. E-mail address: [email protected] (O. Navratil). 0022-1694/$ - see front matter Ó 2011 Elsevier B.V. All rights reserved. doi:10.1016/j.jhydrol.2010.12.025

2003; Mano et al., 2009). Therefore high-frequency SSC monitoring is required for reliable SSC and SSY estimates. Nevertheless, a reliable and easy method to obtain a direct, continuous SSC measurement is not currently available. Although great progress is expected with, for instance, the backscatter acoustic method (Wren et al., 2000; Gray and Gartner, 2009), their application is still limited to large rivers and canals. However, problems remain in mountainous catchments where river channels generally show significant and rapid erosion–transport–sedimentation processes, with very high specific sediment yields (suspended as well as bed-load transport). Moreover, measuring suspended sediment dynamics in this restrictive context remains of prime importance, all the more so because these river basins determine the fine sediment delivery to the whole downstream lowland river system (Milliman and Syvitski, 1992; Hicks et al., 2000; Lenzi et al., 2003; Mathys et al., 2003; Meybeck et al., 2003; Mano et al., 2009; Navratil et al., 2010) and they govern reservoir siltation

Author's personal copy

247

O. Navratil et al. / Journal of Hydrology 398 (2011) 246–259

dynamics, water resources and aquatic habitat qualities (Wood and Armitage, 1997; House and Warwick, 1999; Rees et al., 1999; Valero-Garcés et al., 1999; Packman and Mackay, 2003; Owens et al., 2005). Using only SSC samples with a predefined sampling frequency most often leads to large errors in suspended sediment flux (SSF) estimates (e.g., Thomas and Lewis, 1993, 1995; Lewis, 1996; Phillips et al., 1999; Coynel et al., 2004; Moatar et al., 2006). Therefore, given the problems of direct high-frequency SSC estimates, turbidity measurement is currently used as a surrogate measurement (Lenzi et al., 2003; Brasington and Richards, 2000; Mathys et al., 2003; Orwin and Smart, 2004; Stott and Mount, 2007; Mano et al., 2009; López-Tarazón et al., 2009). This method is based on the quasi-continuous measurement of turbidity, the sensor being calibrated with suspended sediment collected during many flood events. It remains the easiest and most widely used method for suspended sediment monitoring with high-frequency acquisition (Wren et al., 2000; Downing, 2006; Némery et al., 2010). However, this method shows many limitations and uncertainties within river applications and particularly in mountainous catchments exhibiting a very high SSC and complex spatio-temporal SSY variability. The present paper aims (1) to analyse the major uncertainty components (referred to as UCs) associated with the turbidity approach for suspended sediment monitoring in a highly erodible mountainous catchment; (2) to assess the propagated uncertainty on SSC and SSY with Monte Carlo simulations; and (3) to classify the UCs according to their relative effect on the final results. This work will provide a practical methodological framework that can be applied to other hydrometric stations using similar monitoring techniques. This study was based on a 2-year data set (October 2007– December 2009) collected in a small mountainous river basin, the Galabre River in Southern French Pre-Alps. Nine major uncer-

tainty components (UC1–UC9) associated with turbidity measurement were identified and are summarised in Table 1: the choice of a turbidimeter (UC1) and its calibration (UC8); the temporal (UC2, UC3) and spatial (UC4) field sampling strategies; the technical field problems (UC5); the field (UC6) and laboratory (UC7) water sample sampling and procedures; and finally, the discharge estimation (UC9). UC2, UC4 and UC9 were considered in this study in comparison to the literature studies and on-site characteristics. Others (UC1, UC3, UC5–UC8) for which literature data were insufficient or not transposable to this context were quantified with specific analyses. Uncertainty propagation applied to long-term time series is quite difficult to perform since the UCs may be correlated with each other; so analytical computations remain very complex. Therefore, Monte Carlo simulations were undertaken to evaluate the combined effect of UCs and to estimate the propagation of uncertainties (Fig. 1; Coleman and Steele, 1999; Joint Committee for Guides in Metrology, 2008a,b; Lacour et al., 2009).

2. Suspended sediment monitoring 2.1. Study area The Galabre River (lat.: 44°100 2700 N, long.: 06°120 5900 E) is a small tributary of the Bléone River in the Rhône River catchment, France (drainage area, 35 km2; Fig. 2). Its Mediterranean and mountainous climate (with frost in winter and high-intensity rainfall in summer) and its geology result in a badlands topography, gully development and substantial transfer of sediment. Highly erodible areas cover about 2% of the river basin. Mean annual temperature ranges between 12 and 13 °C at 400 m ASL (above sea level), with a high temperature amplitude between summer and winter (approxi-

Table 1 Uncertainty components (UCs) for the suspended sediment monitoring using turbidimeter: notations, descriptions and literature studies. An overview of the scope and limitations of our study is also presented: () UCs assumed negligible; (+) UCs shown to be negligible; (#) UC not determined, but considered by our methodology; (O) UCs quantified and considered. UC Id.

Title

UC1

Choice of a suitable turbidimeter Turbidity and water level acquisition frequency Turbidity signal fluctuations Representativity of the point SSC measurement Technical field problems Field sampling procedure Laboratory procedure

UC2

UC3 UC4

UC5 UC6 UC7

UC8

Turbidimeter calibration

UC9

Discharge estimation

Description

UCs analysed or at least discussed In the literature studies

In this study

Suitability of the turbidimeter used (SSC range, automatic cleaning) in regard to the study site characteristics

Wren et al. (2000), Downing (2006), Lewis and Eads (2008)

+

Data acquisition strategy for suspended sediment dynamics monitoring in the studied basin in regard to the purpose of the study

Phillips et al. (1999), Coynel et al. (2004), Moatar et al. (2006)

+

Turbidity signal fluctuations during each time-step of record and choice of a reliable time-duration of signal integration SSC homogeneity in the water column and at the cross-section scale in regard to the hydraulic and sediment characteristics during the flood events Risk of data gaps: sensor silting, fouling, power supply and data logger failures, automatic water sampling failures Uncertainty associated with water and sediment sampling for the range of SSC and the sediment type observed on the site Uncertainty associated with the laboratory procedure (method, material, operator) for the range of SSCs and sediment type observed on the site Reliability of the T-SSC calibration curve in regard to the data dispersion, the laboratory uncertainties (UC6-7) and the turbidity fluctuations (UC3)

Joannis et al. (2010)

O

Horowitz et al. (1989), Horowitz et al. (1989), Horowitz (2008), Steegen and Govers (2001), Gray and Gartner (2009), Némery et al. (2010) Wren et al. (2000), Downing (2006), Lewis and Eads (2008), Francke et al. (2008), Talebizadeh et al. (2010) Lewis (1996), Horowitz et al. (1989), Horowitz (2008), Gray and Gartner (2009) Gautheron (1994), Gray et al. (2000), Orwin and Smart (2004), Clark and Siu (2008)



Foster et al. (1992), Riley (1998), Brasington and Richards (2000), Lenzi and Marchi (2000), Orwin and Smart (2004), Pfannkuche and Schmidt (2003), Downing (2006), Minella et al. (2008), Lacour et al. (2009), Némery et al. (2010) Sauer and Meyer (1992), European ISO EN Rule, 748 1997, Baldassarre and Montanari (2009), McMillan et al. (2010)

O

Water level and discharge measurement uncertainties (material, operator) and quality of the stage – discharge rating curve: range of flow measured, extrapolation, dataset size and actualisation, stability of the gauging cross-section

 O O

#

Author's personal copy

248

O. Navratil et al. / Journal of Hydrology 398 (2011) 246–259

Step

1:

10

minutes

time-step

T (t )

turbidity measurement

and

signal variability σ T (sample size n=60) considering UC1, UC2 and UC5 2

Step 2: Random selection of p timeseries { T (t ) }p within the distribution ( T (t ); σ

2

T

Step 3: Laboratory measurement of suspended sediment concentration SSCi of sample i collected during the flood events

Step 4: Random selection of p sets {SSCi}p within the distribution ( k ⋅ SSCi ; σ

(t ) / n ) considering UC3

2

SCC

)

considering

UC4, UC6 and UC7

Step 5: Estimation of p² sets { Ti ; SSCi}p², i.e. the association between {SSCi}p and the corresponding turbidity value { T (t ) } p Step 6: Estimation of p² calibration curves and p² hysteretic relationships with

{ Ti

; SSCi} p² considering UC8 Step 7: Estimation of p² {SSC(t)}p² times series and statistics Step 9: 10 minutes time-step discharge measurement Q(t)

Step 8: {SSC(t)}p

Random selection of p time-series from the

distribution ( SSC (t ); σ

2

SSC

(t ) )

Step 10: random selection of p {Q(t)}p time-series within ( (1 − θ ) ⋅ Q(t ); (1 + θ ) ⋅ Q(t ) ) considering UC9

Step 11: Estimation of p² suspended sediment flux time-series {SSF(t)}p² with {SSF(t)}p = {SSC(t)}p · {Q(t)}p

Step 12: Estimation of {SSY}p² and the flood indicator {ssy}p² with the distribution ( SSF (t ); σ

2

SSF

(t ) )

Fig. 1. Monte Carlo simulation flow chart with corresponding uncertainty components (UCs).

mately 18 °C). Mean annual rainfall in the catchment varies between 600 and 1200 mm at 400 m ASL. Rainfall is characterised by considerable seasonal variations, with a maximum in spring and autumn (Mano et al., 2009). The catchment is composed of marly calcareous (54%), molasses (31%), black marl (9%), gypsum (4%) and conglomerate (2%). It is overlain by thin soils under grasslands (67%), sparse vegetation (19%) or forests (11%) and it is characterised by a very low human impact (no dams, very low urbanisation). A hydrometric station was installed in October 2007 to survey the suspended sediment dynamics (drainage area, 22 km2). The monitoring station continuously records the turbidity and the water level in order to estimate SSC and discharge with high-frequency acquisition. The site is easy to access and is located in the vicinity of a bridge (Fig. 2). The station studied is part of an international monitoring program assessing the spatio-temporal variability of suspended sediment dynamics in mountainous catchments from the flood scale to the pluri-annual scale, and from 1 to

about 1000 km2 (e.g. Navratil et al., 2010; Gratiot et al., 2010). The program is based on continuous measurement of suspended sediment fluxes with high temporal resolution at six hydrometric stations (from 20 to 860 km2) in the Bléone River basin (Navratil et al., 2008, 2010) and at five stations (from 3 to 600 km2) in the Mexican Central Highlands (Gratiot et al., 2010, Duvert et al., 2010). 2.2. Field monitoring A 24-GHz radar (Paratronic CrusoeÒ) was installed to continuously survey the water level (Fig. 2). Flow discharges were measured every 3 weeks with the salt (NaCl) dilution method and a current flow meter: 28 flows discharge were measured from 2007 to 2009 ranging from 0.005 m3 s1 to 1.18 m3 s1. The surveyed cross-section was assumed to be very stable (i.e. bedrock cross-section without sedimentation or erosion). The water-level discharge rating curve was then built to estimate the discharge time series (Fig. 4a).

Author's personal copy

O. Navratil et al. / Journal of Hydrology 398 (2011) 246–259

249

Fig. 2. The Galabre River at La Robine in the Southern French Pre-Alps. (a) Study area located in the Bléone River basin. (b) Picture of the instrumentation during a flood event (May 5th, 2008), with a flow discharge of about 3 m3 s1.

A nephelometric turbidimeter (WTW VisolidÒ 700-IQ) and a IQ182 sensor net transmitter measured the turbidity in the water flow by infra-red light (860 nm) 90°-diffusion (for low concentrations) and retro-diffusion (for high concentrations). A nephelometric sensor was preferred over a transmission sensor for two main reasons: first it measures a greater SCC range; second, these sensors are less sensitive to fouling. The sensor was built to measure high suspended sediment concentrations in rivers and wastewater-treatment plants; so it was calibrated by the manufacturer with silica (SiO2) in the [0–300 g l1 SiO2] range, corresponding to a [0–10,000 NTU] range. The [g l1 SiO2] unit for turbidity is used herein. The coefficient of variation (CV), i.e. the ratio of the standard deviation to the mean, of the signal provided by the manufacturer is less than 4%; resolution is 0.01 g l1 in the [0–25 g l1 SiO2] range, 0.1 g l1 in the [25–300 g l1 SiO2] range. Ultrasonic cleaning of the sensor during the measurement prevents organic and fine sediment fouling. To avoid any bias with natural solar light, the sensor was oriented downward (angle, approximately 30°) in the upstream direction of the river flow and 20 cm above the river bottom. Every 10 min, a data logger (CampbellÒ CR800) recorded the water level and the turbidity for 1 min (frequency, 1 Hz) and calculated the basic signal statistics: the mean turbidity T and water level H, the standard deviations (respectively, rT and rH) and the minimum and maximum values. Data integration during a 1-min time step was found to be a good compromise between the short time scale fluctuations of the signals and their variations at the flood event scale (a few hours in this case study). Statistics were then downloaded daily via GSM transmission and stored at the laboratory. The turbidimeter was calibrated with SSC measurements of water samples collected during several floods. An automatic water sampler (Teledyne ISCOÒ 3700) containing 24 bottles of 1 l capacity operated approximately 3 m above the intake of the pumping pipe according to the method described by Lewis (1996): when turbidity critical thresholds T1 = 5 g l1 SiO2 and T2 = 20 g l1 SiO2 were exceeded, the data logger triggered water sampling at

regular intervals: every 60 and 30 min, respectively. Turbidity thresholds and sample frequency were adjusted according to SSC dynamics and seasonal variability. The vinyl pumping pipe was 5 m long and 9.5 mm in diameter; its intake was placed at a fixed height close to the turbidimeter, i.e. 20 cm from the river bottom. The pumping line was purged before and after each sample in order to limit auto-contamination between successive samples. The ideal orientation of the intake would be facing upstream (Gautheron, 1994), but debris fouling and the inability of the sampler to purge the intake line against the flow current makes this orientation impossible (Lewis and Eads, 2008). The turbidimeter was also subjected to calcification resulting in calcareous precipitation on the optic cell. It was then regularly washed with 5% HCl acid. 2.3. Laboratory analysis and data processing One-litre samples were processed in the laboratory using two different methods to estimate SSC (AFNOR T90-105, 1994). The method based on the whole water sample was preferred to the method that uses sub-samples (or aliquot) in order to limit measurement bias associated with operator handling (Gray et al., 2000). At low SSCs (estimated by expert assessment, 125 lm) on average accounted for less than 5% of the sediment sample volume; however, for specific flood events, the fine sand fraction can increase significantly (up to 66%) with a simultaneous increase in discharge and SSC (Fig. 6), which could then refute the hypothesis of SSC homogeneity. However, as the flow remains very turbulent during most of the flood events and the gauge station is distant from sediment sources, the sediment and water are assumed to be well mixed. We therefore hypothesise that for most flood events, there is a little or no bias at the point sample (Table 1), as suggested by Lewis and Eads (2008). 4.1.5. Technical field problems (UC5) In this study, the uncertainties or missing data associated with sensor fouling and equipment problems are at least corrected with

Fig. 6. Relation between discharge and SSC for suspended sediment samples collected during four contrasting flood events (associated with different circles greyscale). Circle size is proportional to the sand fraction in the sample (>125 lm; in percent).

Author's personal copy

254

O. Navratil et al. / Journal of Hydrology 398 (2011) 246–259

linear data interpolation, or at best avoided given the real-time survey of the measurements (e.g. GSM or phone network) and the frequents visit to the study sites. Finally, from October 2007 to December 2009, the missing turbidity or water level data totalled less than 3%, which could be a lot, given that 90% of SSY is generated during 2% of the time. But, in our study, technical problems only occurred during small flood events which did not generated significant sediment yields at the other gauging stations in the catchment (Fig. 2). Therefore, UC5 was assumed negligible (Table 1). 4.1.6. Field sampling procedure (UC6) and laboratory analysis (UC7) Aggregated automatic sampling (UC6) and laboratory (UC7) uncertainties combined systematic and random errors (Fig. 7a). Systematic error was on average 17%, i.e. the mean ratio between SSC_true and SSC_estim was 1.17 for SSC, ranging from 0.5 to 150 g l1. The random error was a mean CV = 17%, but standard deviation (Std_SSC) increased with SSC (Fig. 7b):

rSSC estim ¼ 0:23  SSC estim0:79 ðr2 ¼ 0:96Þ

ð3Þ

The normality of the distribution of each SSC range was positively tested with the Shapiro–Wilk test (p-value = 0.05). Then  SSC_true can be defined as N 1:17  SSC estim; r2SCC estim , and can be implemented in the Monte Carlo simulation flowchart (Fig. 1). Laboratory uncertainty estimated alone (UC7), i.e. corresponding to the repeatability of the laboratory measurements, corresponded to only a small part of the previous aggregated UC6 and UC7. Indeed, systematic error attributed to laboratory manipulations was on average 3% vs. 17% for the aggregated UC6 and UC7.

Moreover, random error associated with UC7 alone remained small: CV = 5% on average for the entire range of SSC_estim (range, 1.5–16%). Systematic errors may correspond to a loss of sediment during operator handling in the laboratory, whereas random error may correspond to uncertainties associated with the water volume measurement and the dried sediment weighing. Therefore, the uncertainties associated with the automatic pumping procedure predominated. During the automatic sampling tests carried out in a controlled environment, no significant deviation in SSC was evidenced from the 1 st to the 11th collected sample for the 5 g l1, 50 g l1 and 150 g l1 SSC ranges (Fig. 7c). Only for the 10 g l1 and 100 g l1 SSC ranges was a slight reduction in the SSC_true/SSC_estim ratio observed for the first four samples, whereas an increase was observed for the 0.5 g l1 range. While the mixing conditions achieved in the bucket with the paint mixer handle drill seem to be the same for most of the samples, we have no idea whether sufficient homogeneity was achieved from the bucket (i.e. all sediment in suspension with no stratification) or whether significant particle siltation occurred at the bottom of the bucket that could introduce systematic and random errors. These problems are inherent to this type of experimentation and can at best be limited, but cannot be totally prevented. Nevertheless, they may be quite similar to field sampling bias, as referred to UC4. The sediment size before the sampling (D50% = 63 lm; D10% = 12 lm; D90% = 158 lm) remained very stable among the different ranges of SCC sampled in the bucket (CV, approximately 1%). Fig. 7d shows a systematic and well-marked reduction in the sandsized fraction volume, but an increase of the clay and silt-sized fractions during the pumping phase for all SSC ranges.

Fig. 7. Automatic sampling test. (a) Aggregated uncertainties of both the automatic water sampling (ISCO) and the laboratory SSC analysis. SSC_true corresponds to the controlled SSC prepared from dry suspended sediment and SSC_estim to the estimated SSC after pumping and laboratory analysis (circles). Mean SSC_estim corresponds to the mean of suspended sediment replicates (black points). (b) Relation between the standard deviation and mean SSC_estim for the different ranges of SSC. (c) Evolution of the SSC_true/SSC_estim ratio during each test. (d) Variation of sediment grain size (% volume) before and after the automatic sampling for each class of SSC (g l1).

Author's personal copy

255

O. Navratil et al. / Journal of Hydrology 398 (2011) 246–259

Fig. 8. T-SSC calibration curve analysis. (a) Ratio between the field/laboratory uncertainties (UC6 and UC7) and fit residuals, i.e. the absolute difference between the SSC estimated with the calibration curve Eq. (1) and the SSC measured in laboratory. (b) SSC-discharge for samples of dataset#1: (black plot) and dataset#2 (grey plot).

The experimentation provides a quantification of UC6 and UC7 aggregated uncertainty. UC6 remains the most significant uncertainty. However, it is difficult to asses whether the systematic and random errors during the sampling process can be attributed to the siltation processes of the sand-sized fraction at the bottom of the 10-l bucket or to the pumping process itself. The intake location and orientation may play an important role depending on the hydrodynamic conditions, the power of the automatic sampler and its elevation (4 m in our test, 3 m in the field), the length of the pipe (5 m) and the purge efficiency before and after the sampling may also be very important. These problems, evidenced here in a controlled environment, are of course quite similar in a natural river environment (e.g. Thomas, 1985).

4.1.7. Turbidimeter calibration (UC8) Here we asses the turbidimeter calibration relevance by comparing the field and laboratory uncertainties combined (UC6 and UC7) with the residuals associated with Eq. (1) (Fig. 3). For dataset#1, i.e. the data set used to fit the global calibration curve, the residuals for turbidity less than 1 g l1 SiO2 were very high (>600%). But, residuals significantly decrease for higher SSC: on average 24% in the [1–20 g l1 SiO2] range and 10% in the [20– 200 g l1 SiO2] range. For dataset#2, i.e. the data set used to build hysteretic relations, residuals were found logically higher as the flood samples that composed the hysteresis relations were initially chosen with respect to their large scatter from the calibration curve: on average 35% in the [1–20 g l1 SiO2] range and 50% for greater turbidity values. We cannot conclude if these hysteresis

Fig. 9. SSC(t) and flood indicators uncertainties (95% confidence level). (a) SSC uncertainty (DSSC(t); in %) vs. SSCðtÞ. (b) Distribution of the SSCmean, SSCmax and ssy uncertainties (median, 25% and 75% quartiles, and outliers) for n = 66 flood events (with DQ(t) = 5, 10 and 20%).

Table 2 Notations, definitions and units of the variables used in this study. Variable T(t) SSC(t) SSCmean SSCmax Q(t) qmax SSF (t) SSY ssy a

Definition Turbidity time series Suspended sediment concentration time series Mean SSC at the flood event scale Peak SSC at the flood event scale Discharge time series Peak discharge at the flood event scale Suspended sediment flux time series Suspended sediment yield during the October 2007 to December 2009 period Suspended sediment yield at the flood event scale

Units 1

g l SiO2 g l1 g l1 g l1 m3 s1 m3 s1 t s1 t t

Uncertainty (%) at a 95% confidence levela

DT(t) DSSC(t) DSSCmean DSSCmax DQ(t) Dqmax DSSF(t) DSSY Dssy

Confidence level of 95% for all uncertainties except for DQ(t) and Dqmax as they refer to uniform distributions with 100% confidence level.

Author's personal copy

256

O. Navratil et al. / Journal of Hydrology 398 (2011) 246–259

Fig. 10. Suspended sediment yields uncertainty (95% confidence level). (a) Uncertainty estimation on cumulated SSF(t) (in tons) with time for DQ(t) = 5%, 10% and 20%. (b) Flood event yields (ssy) vs. flood peak discharge (qmax) with uncertainties on both variables (DQ(t) = 20%).

only occur during particular floods, or if they are present for all large events with very high SSCs, but only visible (or sampled) for some large events. We confirm that a relevant and univocal relation exists between turbidity and SSC for most of the collected samples (Fig. 8a) as uncertainties were found greater than residuals for 96% of the dataset#1 samples (>1 g l1). Uncertainties were greater than residuals for only 67% of dataset#2 samples (>1 g l1; Fig. 8a). We must point out that turbidity measurements associated with dataset#2 show very low variability during the 1-min measurement (A1 area; Fig. 5b); consequently, residuals could not be explained by turbidity measurement fluctuations (turbidimeter fouling, etc.). For dataset#2, other factors than SSC uncertainties were responsible for T-SSC hysteretic relations: the sediment size variations, their mineralogy and colour (Foster et al., 1992). Another reason could be the heterogeneity of SSCs at the survey cross-section for specific flood events (UC4). Indeed, hydraulic conditions can discriminate these two data sets for discharges less than 0.1 m3 s1 (Fig. 8b): contrary to dataset#1, many samples of dataset#2 are associated with very high SSCs. So for highly concentrated flow with low discharge, SSC stratification in the water column could occur and explain these hysteresis patterns. As evidenced previously during the automatic sampler test, the SSC estimation would be highly dependent on (1) the hydrodynamic conditions at the vicinity of the ISCO pipe intake (Thomas, 1985), (2) but also its type, orientation and placement within a cross-section and (3) its deviation from the turbidimeter. Furthermore, the intake usually cannot be positioned in one place so that it will sample correctly for all hydro-sedimentary conditions. In the present study, the intake was located at a fixed height above the bed (20 cm), so that it would sample a different relative position in the vertical concentration gradient for a different water level. This problem probably influenced the relationship between the sampled and actual concentration for specific flood events and as a result the T-SSC relationship. 4.1.8. Discharge estimation (UC9) We estimated the uncertainty of individual gauging with repeated measurements at about 5%; it corroborates well with past studies (e.g. Sauer and Meyer, 1992). Water level recording error with radar was less than 1 cm. Contrary to other submerged water level probes (e.g. pressure probes), suspended sediment cannot af-

fect the water level measurement because the radar altimeter is non-submerged. A discharge uncertainty of 10% is expected by the international hydrometric requirements (European ISO EN Rule 748, 1997; Baldassarre and Montanari, 2009). In highly turbulent rivers or torrents, i.e. this study, this uncertainty is probably under-estimated during base flows and flood peaks; it may reach more than 20% (e.g. McMillan et al., 2010). In this study, we considered different ranges of discharge uncertainty: 5%, 10% and 20%. The statistical distribution of Q(t) was expressed as U((1  h)  Q(t); (1 + h)  Q(t)), with h values of 0.05, 0.1 and 0.2.

4.2. Uncertainty propagation analysis 4.2.1. Global SSC uncertainty Since the analytical computation remains very complex for global uncertainty analysis, Monte Carlo procedure appears to be the most-adapted method. However, such method assumes complete intra- and interdependence of the considered variables. The strong dependency between hydrological and sedimentary processes involves that the variables and/or errors are strongly correlated between each other and time-autocorrelated. Consequently, the current approach did not account for these last effects. We made the assumption that turbidity was normally distributed, so statistical  distribution  of mean turbidity TðtÞ can be expressed as N TðtÞ; r2T ðtÞ=n ; with n = 60. This assumption is likely plausible for the measurements with low fluctuations (A1 area at Fig. 5b), but it is probably not verified for turbidity measurements with obstacles that may introduce extreme and transitory turbidity peaks (A2 area; Fig. 5a). For these specific records (i.e. less than 10% of the total records), a log-normal distribution would have been more suited. Monte Carlo simulations provide the mean and standard deviation of SSC every 10 min (Fig. 9a). We defined the uncertainty of variable SSC(t) as:

DSSCðtÞ ¼ 2 

rSSCðtÞ SSCðtÞ

 100

ð4Þ

A coverage factor of 2 was used in this study, i.e. a numerical factor used as a multiplier of the combined standard deviation, in order to define the 95% confidence interval associated SSC(t). The same definition was applied to all the other variables used (Table

Author's personal copy

O. Navratil et al. / Journal of Hydrology 398 (2011) 246–259

2), except Q(t) and qmax. The uncertainty for these last variables, with a 100% confidence interval (uniform distribution), can be written as:

DQ ðtÞ ¼ Dqmax ¼ h  100

ð5Þ

SSC uncertainty (DSSC(t)) were very high for SSC < 1 g l1 (Fig. 9a) because of the high turbidity fluctuations that generally occur (see Section 5). In the SSC [1–10 g l–1] range, the median DSSC(t) was 10% (range, 4–67%) and in the [10–100 g l1] range, the median DSSC(t) was 5% (range, 3–25%). The solid lines observed on Fig. 9a are some autocorrelation of uncertainties during flood events with very high concentrations. At the flood event scale (66 events), the median uncertainties on peak SSC (DSSCmax) and mean SSC (DSSCmean) were 18% and 17%, respectively (Fig. 9b). These uncertainties varied greatly from one flood to another, from 1% to more than 30% for both indicators. As mentioned above, we knew that SSC uncertainty was sitedependent; these results indicate that SSC uncertainty is also highly time-dependent and varies significantly depending on the hydro-sedimentary conditions. Therefore, although a mean uncertainty value is always useful for operational studies, such information remains incomplete and sometimes unreliable. 4.2.2. Global SSY uncertainty Uncertainties on the time-cumulated SSF(t) (DSSF(t)) are relatively stable over time, with a mean value of 15%, 19% and 29% (with DQ(t) = 5%, 10% and 20% respectively) for the period studied (Fig. 10a). Specific sediment yield SSY is 360 ± 100 t km2 year1 for the site studied (considering DQ(t) = 20%). Sediment yield uncertainties estimated at each flood event (referred to as Dssy) were in the same range (median = 27%; with DQ(t) = 20%) but showed high variations (range, 22–47%) from one flood to another, depending on turbidity fluctuations, discharge magnitude and the T-SSC calibration relation used (Fig. 9b). Classical relations found in the literature (e.g. Lenzi et al., 2003) can now be expressed with their uncertainties. For instance, for the relation between SSY and the flood peak discharge (qmax; Fig. 10b), cumulated uncertainties only explain a small part of the fit residuals (on average 10% of residuals for all the flood events). This evidences and confirms that physical factors (e.g. land use, soil erodibility, rainfall regime) play an essential role in the sediment transport dynamics and variability. Global uncertainty on SSC and SSY will thus help define and discriminate the main factors controlling the suspended sediment dynamics. 4.2.3. Calibration curve sensitivity analysis To estimate the significance of the use of T-SSC hysteretic relations for the turbidimeter calibration (UC8), we estimated SSY following two approaches: (1) in the first step, we used both the TSSC calibration curve and the hysteretic relations following our initial methodology, and (2) second, only the T-SSC calibration curve was used (Eq. (1)). SSY during the study period was slightly underestimated (about 5%) if T-SSC hysteretic relations were not used. This difference was much greater at the flood event scale, as on average 23% of difference was observed (range, 3–60%). Therefore, in most of cases, hysteretic relations are required when studying the inter-event variability of sediment yields. If the studies aimed only at surveying the annual or inter-annual sediment loads, a calibration of the turbidimeter with a T-SSC calibration curve is acceptable, with occasional verification during the following flood events. 4.2.4. Turbidity measurement sensitivity analysis We considered two turbidity record strategies, with n = 60 and n = 1 (step 1, Fig. 1), and we studied how this choice affected the final estimation of sediment yields. If only a single turbidity value

257

is recorded (n = 1) every 10-min time step, median Dssy were 30%, vs. 18% with n = 60 (with DQ(t) = 10%; Fig. 9b). This analysis shows that averaging turbidity for 1 min makes it possible to significantly decrease measurement uncertainties. 4.2.5. Uncertainty classification and recommendations This study helps to quantify the uncertainties and proposes monitoring recommendations designed to minimise uncertainties. UC1 and UC2 can be negligible with good knowledge and experience of river monitoring. UC3 can also be very low if mean and standard deviation of turbidity are recorded. This method will first minimise the uncertainties and second provide the signal variability as a useful quality indicator of the turbidity signal. In the present study, SSC homogeneity in the water column of the cross-section was reached for the major flood events; however, doubt remains for very specific flood events (low discharge and high SSCs). UC5 can be avoided with a real-time monitoring and data validation with other monitoring stations located in the same catchment. However, this experiment showed that in a mountainous environment system failures are frequent and very difficult to prevent (e.g. during storm events or flash floods). When these occur during high intensity floods, SSY estimations may be severely affected despite relatively short data gaps. Uncertainty in the laboratory (UC7) remains low when using the entire collected sample (Gray et al., 2000). UC8 can also be very low if considering T-SSC hysteretic relations. UC8 will be largely dependent on UC4 and the heterogeneity of the sediment sources. Turbidimeter calibration needs to be checked regularly, particularly if the study focuses on the SSY at the flood event scale. Finally, SSY uncertainties are very dependent on method of automated water sampling (UC6) and discharge estimation (UC9). It is difficult to suggest a solution to minimise the field sampling uncertainties because the main factor governing this uncertainty was not identified; however, we can suggest reducing the length of the pipe line of the automatic sampler and its height above the water surface. The orientation and depth of the pipe intake remain problematic because in many cases, it is placed according to on-site constraints. It also seems important to evaluate the field and laboratory uncertainties for each site because they probably depend on suspended sediment size, i.e. sand-sized fraction errors may be greater. For UC9, we naturally suggest respecting the discharge monitoring basics in rivers, with a focus on the choice of a stable gauging cross-section removed from any backwater effect, the long-term survey of the whole range of flow discharge, the choice of a non-submerged water level recorder in the mountainous context to avoid silting and measurement corrections. 5. Conclusions Suspended sediment monitoring cumulates many uncertainties from field monitoring to the SSC and SSY computation procedure. Assessing uncertainties is therefore required because it would help distinguish the spatio-temporal variability of SSY attributed to natural (e.g. rainfall, erodibility, relief) or anthropogenic factors (e.g. land use change) from the variability attributed to monitoring uncertainties. Furthermore, the quantification of these uncertainties is of prime importance to a relevant calibration and validation of physical-based hydro-sedimentary models as well as for pollutant transport estimations in rivers. In this paper, we have estimated the uncertainties on SSC and SSY monitored with the turbidity approach at a hydro-sedimentary station located in a mountainous context. SSC uncertainties are on average less than 10% for individual records (with a 95% confidence level), but it is highly variable, reaching approximately 70%. Flood indicator value uncertainties (SSCmean, SSCmax; Table 2) are about 20% on average. Annual specific sediment yield SSY is

Author's personal copy

258

O. Navratil et al. / Journal of Hydrology 398 (2011) 246–259

360 ± 100 t km–2 year–1 (with DQ(t) = 20%). Uncertainty of sediment yields at the flood event scale (Dssy) is about 15%, 19% and 29% depending on the level of uncertainty on discharge estimates (respectively 5%, 10% or 20%). This study provides a practical methodological framework applicable to other hydro-sedimentary stations in order to propagate the uncertainties from the turbidity measurement to the final SSY estimates. These results are rigorously relevant at the station studied, as we have shown that uncertainties mainly depend on factors varying with time and site features. However, we project that these results could provide a rough but useful order of magnitude of uncertainties valuable at other monitoring stations built on the same model, with similar SSC range, sediment type and equipment. Acknowledgments This study was funded by the French ANR-STREAMS project (ANR Blanc 06-1_139157). The authors are grateful to Fred Malinur, Lucas Muller, Vincent Thiabault, Jean-Marie Miscioscia, and Michel Ricard for the field campaigns and laboratory analysis and technical assistance, Linda Northrup (English Solution) for editing the manuscript, and two anonymous reviewers who helped improve the earlier versions of this manuscript. References Baldassarre, G.B., Montanari, A., 2009. Uncertainty in river discharge observations: a quantitative analysis. Hydrology and Earth System Sciences Discussions 6, 39– 61. Brasington, J., Richards, K., 2000. Turbidity and suspended sediment dynamics in small catchments in the Nepal Middle Hills. Hydrological Processes 14, 2559– 2574. Clark, S.E., Siu, C.Y.S., 2008. Measuring solids concentration in stormwater runoff: comparison of analytical methods. Environmental Science and Technology 42, 511–516. Coleman, H.W., Steele, G.W., 1999. Experimentation and Uncertainty Analysis for Engineers, second ed. Willey Ed.. 296 p. Coynel, A., Schafer, J., Hurtrez, J.E., Dumas, J., Etcheber, H., Blanc, G., 2004. Sampling frequency and accuracy of SPM flux estimates in two contrasted drainage basins. Science of the Total Environment 330, 233–247. Downing, J., 2006. Twenty-five years with OBS sensors: the good, the bad, and the ugly. Continental Shelf Research 26, 2299–2318. Duvert, C., Gratiot, N., Evrard, O., Navratil, O., Némery, J., Prat, C., Esteves, M., 2010. Drivers of erosion and suspended sediment transport in three headwater catchments of the Mexican Central Highlands. Geomorphology 123, 243–256. European ISO EN Rule 748, 1997. Measurement of Liquid Flow in Open Channels – Velocity-Area Methods. Reference Number ISO 748:1997 (E). International Standard, 41, 42, 43, 44, 45, 46. Foster, I.D.L., Millington, R., Grew, R.G., 1992. The impact of particle size controls on stream turbidity measurements; some implications for suspended sediment yield estimation. In: Bogen, J., Walling, D.E., Day, T.J. (Eds.), Erosion and Sediment Transport Monitoring Programmes in River Basins, vol. 210. IAHS Publication, pp. 51–62. Francke, T., López-Tarazón, J., Schröder, B., 2008. Estimation of suspended sediment concentration and yield using linear models, random forests and quantile regression forests. Hydrological Processes 22, 4892–4904. doi:10.1002/ hyp.7110. Gautheron, A., 1994. Incertitudes sur la mesure des Matières en Suspension en Basse Durance. EDF DTG, Technical Report, Grenoble, 50 p. Gratiot, N., Duvert, C., Collet, L., Vinson, D., Némery, J., Saenz-Romero, C., 2010. Increase in surface runoff in the central mountains of Mexico: lessons from the past and predictive scenario for the next century. Hydrology and Earth System Science 2 (14), 291–310. Gray, J.R., Glysson, G.D., Turcios, L.M., Schwarz, G.E., 2000. Comparability of Suspended-Sediment Concentration and Total Suspended Solids Data, U.S. Geological Survey Water-Resources Investigations Report 00-4191, 14 p. Gray, J.R., Gartner, J.W., 2009. Technological advances in suspended-sediment surrogate monitoring. Water Resources Research 45, W00D29. doi:10.1029/ 2008WR007063. Hicks, D.M., Gomez, B., Trustrum, N.A., 2000. Erosion thresholds and suspended sediment yields, Waipaoa River Basin, New Zealand. Water Resources Research 36 (4), 1129–1142. Horowitz, A.J., Rinella, F.A., Lamothe, P., Miller, T.L., Edwards, T.K., Roche, R.L., Rickert, D.A., 1989. Cross-sectional variability in suspended sediment and associated trace element concentrations in selected rivers in the US. In: Hadley, R.F., Ongley, E.D. (Eds.), Sediment and the Environment, vol. 184. IAHS Publication, pp. 57–66.

Horowitz, A.J., Elrick, K.A., von Guerard, P.B., Young, N.O., Buell, G.R., Miller, T.L., 1992. The use of automatically collected point samples to estimate suspended sediment and associated trace element concentrations for determining annual mass transport. In: Bogen, J., Walling, D., Day, T., (Eds.), Erosion and Sediment Transport Monitoring Programmes in River Basins, Proceedings of the IAHS Symposium, Oslo, Norway, 1992, vol. 210. IAHS Publication, pp. 209–218. Horowitz, A.J., 2008. Determining annual suspended sediment and sedimentassociated trace element and nutrient fluxes. The Science of the Total Environment 400, 315–343. House, W.A., Warwick, M.S., 1999. Interactions of phosphorus with sediments in the River Swale, Yorkshire, UK. Hydrological Processes 13, 1103–1115. Joannis, C., Ruban, G., Aumond, G., Bertrand-Krajewski, J.L., Battaglia, P., Lacour, C., Saad, M., Chebbo, G., 2010. Implementation of turbidity sensors in sewer systems. Techniques–Sciences–Méthodes 1–2, 21–35. Joint Committee for Guides in Metrology (JCGM), 2008a. Evaluation of Measurement Data — Guide to the Expression of Uncertainty in Measurement. JCGM 100:2008. BIPM Ed. 120 p. Joint Committee for Guides in Metrology (JCGM), 2008b. Evaluation of Measurement Data: Propagation of Distributions using a Monte Carlo Method. JCGM 101:2008. BIPM Ed. 82 p. Lacour, C., Joannis, C., Chebbo, G., 2009. Assessment of annual pollutant loads in combined sewers from continuous turbidity measurements: sensitivity to calibration data. Water Research 43, 2179–2190. Lapointe, M., 1992. Burst-like sediment suspension events in a sand bed river. Earth Surface Processes and Landforms 17, 253–270. Lenzi, M.A., Marchi, L., 2000. Suspended sediment load during floods in a small stream of the Dolomites (Northeastern, Italy). Catena 39 (4), 267–282. Lenzi, M.A., Mao, L., Comiti, F., 2003. Interannual variation of suspended sediment load and sediment yield in an alpine catchment. Hydrological Sciences Journal– Journal Des Sciences Hydrologiques 48 (6), 899–915. Lewis, J., 1996. Turbidity-controlled suspended sediment sampling for runoff-event load estimation. Water Resources Research 32 (7), 2299–2310. Lewis, J., Eads, R., 2008. Implementation Guide for Turbidity Threshold Sampling: Principles, Procedures, and Analysis. Gen. Tech. Rep. PSW-GTR-212. Albany, CA: US Department of Agriculture, Forest Service, Pacific Southwest Research Station. 86 p. López-Tarazón, J.A., Batalla, R.J., Vericat, D., Francke, T., 2009. Suspended sediment transport in a highly erodible catchment: The River Isábena (Southern Pyrenees). Geomorphology 109 (3–4), 210–221. Mano, V., Némery, J., Belleudy, P., Poirel, A., 2009. Assessment of suspended sediment transport in four alpine watersheds (France): influence of the climatic regime. Hydrological Processes 22. doi:10.1002/hyp.7178. Martínez-Carreras, N., Udelhoven, T., Krein, A., Gallart, F., Iffly, J.F., Ziebel, J., Hoffmann, L., Pfister, L., Walling, D.E., 2010. The use of sediment colour measured by diffuse reflectance spectrometry to determine sediment sources: application to the Attert River catchment (Luxembourg). Journal of Hydrology 382, 49–63. Mathys, N., Brochot, S., Meunier, M., Richard, D., 2003. Erosion quantification in the small marly experimental catchments of Draix (Alpes de Haute Provence, France). Calibration of the ETC rainfall–runoff–erosion model. Catena 50, 527–548. McMillan, H., Freer, J., Pappenberger, F., Krueger, T., Clark, M., 2010. Impacts of uncertain river flow data on rainfall–runoff model calibration and discharge predictions. Hydrological Processes 24, 1270–1284. doi:10.1002/hyp.7587. Meybeck, M., Laroche, L., Dürr, H.H., Syvitski, J.P., 2003. Global variability of daily total suspended solids and their fluxes. Global Planetary Changes 39, 65–93. Milliman, J.D., Syvitski, J.P.M., 1992. Geomorphic/tectonic control of sediment discharge to the ocean: the importance of small mountainous rivers. The Journal of Geology 100, 525–544. Minella, J.P.G., Merten, G.H., Reichert, J.M., Clarke, R.T., 2008. Estimating suspended sediment concentrations from turbidity measurements and the calibration problem. Hydrological Processes 22, 1819–1830. doi:10.1002/hyp.6763. Moatar, F., Person, G., Meybeck, M., Coynel, A., Etcheber, H., Crouzet, P., 2006. The influence of contrasting suspended particulate matter transport regimes on the bias and precision of flux estimates. Science of the Total Environment 370 (2–3), 515–531. Navratil, O., Esteves, M., Nemery, J., Legout, C., Poirel, A., Gratiot, N., Belleudy, P., 2008. Multi-scale survey of suspended sediment concentration in the Bléone river basin (Southern French Alps), vol. 10. European Geophysical Union Conference, Vienna, March, 2008. Navratil, O., Legout, C., Gateuille, D., Esteves, M., Liebault, F., 2010. Assessment of intermediate fine sediment storage in a braided river reach (southern French Prealps). Hydrological Processes 24 (10), 1318–1332. Némery, J., Mano, V., Navratil, O., Gratiot, N., Duvert, C., Legout, C., Belleudy, P., Poirel, A., Esteves, M., 2010. Feedback on the use of turbidity in mountainous rivers. Retour d’expérience sur l’utilisation de la turbidité en rivière de montagne. Techniques–Sciences–Méthodes 1/2, 61–67. Orwin, J.F., Smart, C.C., 2004. Short-term spatial and temporal patterns of suspended sediment transfer in pro-glacial channels, Small River Glacier, Canada. Hydrological Processes 18, 1521–1542. Owens, P.N., Batalla, R.J., Collins, A.J., Gomez, B., Hicks, D.M., Horowitz, A.J., Kondolf, G.M., Marden, M., Page, M.J., Peacock, D.H., Petticrew, E.L., Salomons, W., Trustrum, N.A., 2005. Fine-grained sediment in river systems: environmental significance and management issues. River Research and Applications 21, 693– 717. Packman, A.I., Mackay, J.S., 2003. Interplay of stream–subsurface exchange, clay deposition and stream bed evolution. Water Resources Research 39, 41–4-9.

Author's personal copy

O. Navratil et al. / Journal of Hydrology 398 (2011) 246–259 Pfannkuche, J., Schmidt, A., 2003. Determination of suspended particulate matter concentration from turbidity measurements: particle size effects and calibration procedures. Hydrological Processes 17, 1951–1963. Phillips, J.M., Webb, B.W., Walling, D.E., Leeks, G.J.L., 1999. Estimating the suspended sediment loads of rivers in the LOIS study area using infrequent samples. Hydrological Processes 13, 1035–1050. Rees, J.G., Ridgeway, J., Knox, R.W.O.B., Wiggans, G., Breward, N., 1999. Sedimentborne contaminants in rivers discharging into the Humber estuary, UK. Marine Pollution Bulletin 37, 316–329. Riley, S.J., 1998. The sediment concentration–turbidity relation: its value in monitoring at Ranger Uranium Mine, Northern Territory, Australia. Catena 32, 1–14. Sauer, V.B., Meyer, R.W., 1992. Determination of Error Individual Discharge Measurements. US Geological Survey Open-File Report, 92-14. Steegen, A., Govers, G., 2001. Correction factors for estimating suspended sediment export from loess catchments. Earth Surface Processes and Landforms 26, 441– 449. Stott, T., Mount, N., 2007. Alpine proglacial suspended sediment dynamics in warm and cool ablation seasons: implications for global warming. Journal of Hydrology 332 (3–4), 259–270.

259

Talebizadeh, M., Morid, S., Ayyoubzadeh, S.A., Ghasemzadeh, M., 2010. Uncertainty analysis in sediment load modeling using ANN and SWAT model. Water Resources Management 24 (9), 1747–1761. doi:10.1007/s11269-009-9522-2. Thomas, R.B., 1985. Measuring Suspended Sediment in Small Mountain Streams. Gen. Tech. Rep. PSW-83. Berkeley, Calif.: USDA, Forest Service, Pacific Southwest Forest and Range Exp. Stn. 9 p. Thomas, R.B., Lewis, J., 1993. A comparison of selection-at-list time and timestratified sampling for estimating suspended sediment loads. Water Resources Research 29 (4), 1247–1256. Thomas, R.B., Lewis, J., 1995. An evaluation of flow-stratified sampling for estimating suspended sediment loads. Journal of Hydrology 170 (1), 27–45. Valero-Garcés, B.L., Navas, A., Machín, J., Walling, D., 1999. Sediment sources and siltation in mountain reservoirs: a case study from the central Spanish Pyrenees. Geomorphology 28, 23–41. Wood, P.J., Armitage, P.D., 1997. Biological effects of fine sediment in the lotic environment. Environmental Management 21, 203–217. Wren, D.G., Barkdoll, B.D., Kuhnle, R.A., Derrow, R.W., 2000. Field techniques for suspended sediment measurement. Journal of Hydraulic Engineering 126 (2), 97–104.