Based on the concept that sedimentary processes average large areas of exposed crust, sediment data have been widely used to estimate the average Pb isotopic composition of the upper continental crust. However, the possible effects of mineral sorting processes on sediment Pb isotopes have never been fully investigated. Here, we report Pb isotopic compositions of Himalayan river sediments as well as those of several grain-size fractions and mineral separates. We demonstrate that Pb isotopes of both bed loads and suspended loads are biased toward more radiogenic values than their source rocks due to a “heavy mineral effect” caused by mineral sorting during fluvial transport on continents. The sparse zircons, monazites and allanites present in all samples (<1 wt%), including suspended loads, generate a Pb isotopic variability as large as that observed in the Earth's mantle. After correction of this effect, we propose an average value for the composition of the upper Himalayan crust together with a new Pb isotopic value for the Earth's upper continental crust. We conclude that mineralogical effects must be evaluated carefully before using Pb isotopes of sediments as provenance and anthropogenic tracers.
 Knowing the average Pb isotopic composition of the continental crust is crucial to constrain the evolution of the Earth since its formation. Current estimates are rare. Some come from Earth evolution models [Kramers and Tolstikhin, 1997; Zartman and Doe, 1981] but most are based on data acquired from river or oceanic sediments [Asmerom and Jacobsen, 1993; Allègre et al., 1996; Millot et al., 2004]. The latter assume that sediments are representative of their continental sources but bias can be introduced if sedimentary mineral sorting is ignored. Patchett et al.  demonstrated that Hf isotopes are fractionated by the so-called “zircon effect,” i.e., preferential concentration of unradiogenic Hf-rich zircons in coarse sediments which produces finer sediments with much more radiogenic Hf isotopes than their continental sources. Because heavy minerals have sometime extremely radiogenic Pb isotopes, here, we evaluate the impact of the “heavy mineral effect” on Pb isotopes of river sediments. We used a comprehensive set of samples including bed loads, bank sediments, and suspended loads sampled in the Ganga and its major tributaries draining the Himalayan mountain range together with several grain-size fractions and 12 mineral separates isolated from some of these sediments. Using all these data, we demonstrate that mineral sorting processes during fluvial transport bias the sedimentary record and generate a large Pb isotopic variability. We quantify this effect and suggest an alternative to properly estimate the sediment source compositions. Finally, we estimate an average Pb isotopic composition for the upper Himalayan crust and propose a new Pb isotopic value for the Earth's upper continental crust.
2. Context and Samples
 The Ganga is one of the largest rivers on Earth, delivering annually ca. 400 million tons of sediments to the Bay of Bengal [Lupker et al., 2011]. Because of its critical role in the world's sedimentation system [Milliman and Meade, 1983], this fluvial network constitutes an excellent laboratory to assess the global isotopic systematics of river sediments. The Ganga fluvial system essentially carries the erosion products of the Himalayan orogenic belt which can be divided into four main geological units from north to south [Le Fort, 1975]: (1) the variably metamorphosed sedimentary rocks of the Tethyan Sedimentary Series, (2) the highly metamorphosed crystalline rocks of the High Himalayan Crystalline, (3) the weakly metamorphosed sedimentary rocks of the Lesser Himalaya, and (4) the Siwaliks comprised of Neogene to Quaternary floodplain deposits (Figure 1).
 In this paper, we report Pb isotopes of (a) sediments carried by Himalayan rivers draining monolithologic catchments in the mountain range and (b) bed load, bank, and suspended load sediments sampled in the floodplain at the Himalayan front and further downstream in the Ganga and its major tributaries. To understand what controls Pb isotopes in river sediments, we also analyzed pure separates of biotite, K-feldspar, muscovite, plagioclase, magnetite, epidote, titanite, vermiculite, amphibole, carbonate and one fraction enriched in zircon, monazite, and allanite separated from a bed load sampled at the outflow of the Ganga, as well as several grain-size fractions separated from a suspended load and a bank sediment.
 Sampling techniques for bed loads, bank sediments, and suspended loads can be found in Galy et al.  and Lupker et al. . Grain-size fractions were separated from two samples (PB 60 and BGP 6) by sieving and settling techniques. Mineral separates were obtained from different grain-size fractions of sample BR 717. We used successive centrifugations in sodium metatungstate and methylene iodide to separate the minerals as a function of their densities and recovered them by partial freezing in liquid nitrogen. Each mineral separate was then carefully purified under the binocular microscope. Unfortunately, we could not isolate pure zircon, monazite, and allanite fractions because the grains were too small to be handpicked under the binocular microscope. Instead, we separated all the heavy minerals (density > 3.3 g. cm−3) from the 2–63 μm fraction of BR 717 and analyzed this heavy mineral concentrate, which is highly enriched in zircon, monazite, and allanite as demonstrated by the petrologic and chemical studies conducted by Garzanti et al. [2010, 2011] on samples collected at the same location. A 5–10 mg mineral separates and 50–100 mg powder for both bulk sediments and grain-size fractions were dissolved for 2 days in HNO3 14 N on a hot plate at 130°C prior to evaporation and further digestion in a mixture of HF and HClO4 in Teflon containers placed in steel PARR® bombs for 6 weeks at 140°C or 1 week at 200°C. Following the analytical procedure described by Chauvel et al. , the isolation of Pb was carried out with anion AG1-X8 resin. Pb isotopic ratios were measured on a Nu Plasma Multicollector-Inductively Coupled Plasma-Mass Spectrometer (MC-ICP-MS) at École Normale Supérieure (ENS) Lyon (France). Mass fractionation bias was corrected using a thallium spike [White et al., 2000]. Analytical drift was corrected by the standard bracketing technique using the measured values for the National Bureau of Standards (NBS) 981 standard run every two or three samples and those published by Galer and Abouchami . The reproducibility (1σ) of our measurements for the NBS 981 standard, measured 66 times, is better than 270 ppm for 208Pb/204Pb, 180 ppm for 207Pb/204Pb, and 420 ppm for 206Pb/204Pb. Total procedural blanks average 280 pg and are always negligible relative to the amount of Pb isolated in each sample.
4. Results and Discussion
4.1. What Causes the Pb Isotopic Diversity?
 Pb isotopic compositions of river sediments sampled in the Himalayan mountain range and in the floodplain are provided in supporting information Table S1 and shown in Figure 2.1Lupker et al.  provide major element data on the same samples and Garçon  and Garçon et al.  report trace elements and Nd-Hf isotopes. The most striking feature seen in Figure 2 is that Himalayan sediments define a remarkable linear trend in Pb isotopic spaces, spanning an extremely large range of values compared to that known for terrestrial rocks. Such a linear array in Pb diagrams is puzzling: it could be interpreted as an isochron but because Nd isotopes of Himalayan river sediments and U-Pb ages of individual zircons demonstrate that the sources of sediments are diverse and have different ages [Parrish and Hodges, 1996; DeCelles et al., 2000; Richards et al., 2005], this is very unlikely. Alternatively, the trend could result from mixing of sediments derived from Himalayan sources with different Pb isotopic signatures. We also reject this interpretation because sediments eroded from single geological units have extremely variable Pb compositions and do not define end-member values along the array. For example, sediments eroded from the Lesser Himalaya units, which include much older rocks than the other geological units [Parrish and Hodges, 1996; DeCelles et al., 2000], do not exhibit the expected highly radiogenic Pb signatures (Figure 2). In addition, some of the sediments eroded from the Lesser Himalaya have Pb isotopic compositions similar to those derived from the Tethyan Sedimentary Series and the High Himalayan Crystalline units but different Nd and Hf isotopic compositions [Galy and France-Lanord, 2001; Richards et al., 2005; Singh et al., 2008; Garçon et al., 2013; Garçon, 2012]. These observations lead us to conclude that Pb isotopes of river sediments do not faithfully reflect that of their sources.
 Sediments collected in the floodplain also define a large range of Pb isotopic compositions (Figure 2). This is in sharp contrast with their uniform Nd isotopic compositions [Galy and France-Lanord, 2001; Singh et al., 2008; Garçon et al., 2013; Garçon, 2012] indicating that Himalayan detritus is well homogenized before entering the floodplain and that the tributaries along the river course do not supply new material. It is therefore extremely unlikely that the Pb isotopic variations defined by sediments are due to source variability. The histogram shown in Figure 2c highlights a systematic difference between suspended load and bed load/bank sediments, the latter generally displaying more radiogenic 206Pb/204Pb ratios. The most straightforward explanation is that the observed variability of Pb isotopes results from fluvial processes occurring during sediment transport and deposition.
 Figure 3 shows the Pb isotopic compositions of the mineral separates together with the grain-size fractions. Data for these samples are provided in supporting information Table S2. Except for titanite, all pure mineral separates lie at the lower end of the trend defined by the Himalayan sediments (Figure 3). The same is true for the clay fractions (<2 μm). In contrast, the 2–50 and 50–63 μm fractions from BGP 6 and, more importantly, the heavy mineral fraction from BR 717 have extremely radiogenic Pb values and lie on the extension of the array defined by all river sediments (Figure 3). We could not measure the trace element concentrations of the heavy mineral fraction from BR 717 but petrologic observations under the binocular microscope indicate that it is highly enriched in monazite, allanite, and zircon. Additionally, we note that the 2–50 and 50–63 μm fractions from BGP 6 have very high rare earth element (REE), Zr, and Hf contents (see supporting information Table S2), a feature indicating the presence of high amount of monazite, allanite, and zircon. We thus suggest that the elevated Pb isotopic ratios of the heavy mineral separate from BR 717 and the 2–50 and 50–63 μm fractions from BGP 6 result from the presence of high proportions of zircon, monazite, and allanite. We can estimate the proportion of heavy mineral present in these samples using the Hf content of zircon, the REE content of monazite and allanite (see supporting information Table S2), and the Hf and REE contents of the <2 μm fraction from BGP 6 which, we assume, contains negligible zircon, monazite, and allanite. Such a calculation shows that the two grain-size fractions from BGP 6 contain up to 2.4 wt% zircon and 0.8 wt% monazite and allanite. Due to their chemical stability and their very high U/Pb and Th/Pb ratios (See supporting information Table S2), old zircon, monazite, and allanite accumulate considerable radiogenic Pb through time and their isotopic ratios become extremely radiogenic compared to most terrestrial rocks and minerals. The linear trend displayed by Himalayan river sediments in Figure 2 is thus likely due to a “heavy mineral effect” caused by mineralogical sorting during sediment transport and deposition. In other words, the array seen in Figure 2 results from mixing between extremely radiogenic minerals (zircon, monazite, and allanite) and a more subdued end member corresponding to the heavy mineral-free component of the Himalayan crust.
 We defined the composition of the heavy mineral-free component (pink star in Figure 3) by averaging the Pb isotopic composition of all heavy mineral-poor samples (clay fractions < 2 μm) together with that of the pure K-feldspar separate because K-feldspar hosts most of the Pb present in crustal rocks together with very little U and Th [Patterson and Tatsumoto, 1964; Hemming and McLennan, 2001]. The Pb isotopic composition of the radiogenic heavy mineral end member cannot be precisely constrained using the values measured in our heavy mineral fraction because this fraction contains large amounts of monazite, allanite, and zircon but it also includes garnet and epidote that significantly lower its bulk Pb isotopic composition. The Pb isotopic composition of the pure radiogenic heavy mineral end member was thus estimated using the Pb contents, the isotopic compositions, and the estimated heavy mineral content (3.2 wt%) of samples BGP 6 B and BGP 6 C (see supporting information Table S2). We expect its composition to slightly vary depending on the relative proportions of zircon, monazite, and allanite in the assemblage but, given the very radiogenic composition of these minerals, such variations should not create much scatter at the lower end of the mixing line. The resulting mixing lines are shown in Figures 3 and 4.
 Observations supporting our “heavy mineral effect” interpretation are:
 1. Our calculation demonstrates that less than 1 wt% of extremely radiogenic minerals (zircon, monazite, and allanite) added to the nonradiogenic component is enough to explain the entire range of Pb isotopic values measured for Himalayan river sediments (Figures 4a and 4c). Such proportions are consistent with the proportion determined independently by Garzanti et al. [2010, 2011] in sediments from the Ganga in Bangladesh.
 2. Radiogenic Pb in zircon, monazite, and allanite explains why bed loads have more radiogenic Pb isotopes than suspended loads (Figure 2c). Mineral sorting during sediment transport enriches bed loads in fast-settling coarse and dense minerals, such as zircons, monazites, and allanites, whereas suspended loads are preferentially enriched in platy phyllosilicates and other slow-setting minerals.
 3. The slope of the array defined by all sediments in 207Pb/204Pb versus 206Pb/204Pb isotopic space (Figure 2a) is controlled by the very radiogenic isotopic signature of the heavy mineral fraction, and it provides an age of 1.6 ± 0.1 Ga. This age is consistent with the average U-Pb age of 1.7 Ga that we calculated using data published by DeCelles et al.  on detrital zircons from modern Himalayan Rivers. We are therefore confident that the array defined by Himalayan river sediments in Pb isotopic spaces represents the varying contribution of heavy minerals in the sedimentary materials.
 We can take the argument further and use our data to estimate the average Pb isotopic composition of the upper Himalayan crust provided that the lead present in sediments is representative of its crustal source. In rivers, most Pb is transported in solid phases (i.e., suspended and bed loads) but some is also present as dissolved phase. Considering the average Pb concentration of river sediments given by Viers et al.  (i.e., 61 ppm) and the average worldwide sediment flux of Hay et al.  (i.e., 20 × 1012 kg yr−1), we estimate that about 1200 kt of sediment Pb reach the world ocean each year. By contrast, Gaillardet et al.  calculated that only 3 kt yr−1 of dissolved Pb are delivered into the world ocean. This means that the amount of Pb transferred from continent to ocean in a dissolved form is very small compared to the amount of Pb initially present in crustal rocks submitted to weathering. Lead present in suspended and bed loads thus likely reflects the composition of the upper crust provided that a correction for the “heavy mineral effect” is made. We are therefore confident that the average Pb isotopic composition of the upper Himalayan crust can be evaluated using our Pb data.
 Assuming that all Zr in the upper continental crust (i.e., 193 ppm after Rudnick and Gao ) resides in zircon, and all its REE (i.e., 148 ppm after Rudnick and Gao ) in monazite and allanite, crustal rocks contain a maximum of 0.03 wt% of zircon and 0.04 wt% of monazite and allanite. With these proportions, mixing calculations indicate that the average Pb isotopic composition of the upper Himalayan crust is, within error, similar to our unradiogenic end member (pink star in Figures 3 and 4). This value is less radiogenic than the average Pb composition of the Himalayan-derived suspended loads, bed loads, and bank sediments (Figure 2c), indicating that river sediments are more radiogenic than their sources due to the “heavy mineral effect.” In addition, it suggests that zircon, monazite and allanite are not restricted to the bed load but are also present in fine suspended load, either as small individual grains or as inclusions in other minerals (e.g. phyllosilicates). Consequently, Pb isotopic variability of sediments cannot be used as a proxy for source and/or anthropogenic changes unless mineralogical effects are carefully evaluated and corrected for. Finally, we emphasize that the “heavy mineral effect” is particularly visible in the Himalayan sediments because the detrital heavy minerals being relatively old, their extremely radiogenic Pb isotopic compositions can easily influence the bulk sediment Pb isotopic budget. Of course, a smaller effect is expected if the mean age of the eroded sources is younger.
4.2. New Value for the Average Upper Continental Crust Pb Isotopic Composition
 At a more global scale, we evaluate the Pb isotopic composition of the upper continental crust following the two-stage approach of Asmerom and Jacobsen  using our Pb isotopic values and an average extraction age for the Himalayan crust from the mantle. The average extraction age of the Himalayan crust is constrained using the average Nd model age of Himalayan river sediments, i.e., 2.2 ± 0.1 (1σ) Ga calculated using the Nd data published by Singh et al. , Garçon et al. , and Garçon . Using this Nd model age and its associated uncertainties, we obtain a mantle-stage μ1 value (238U/204Pb) of 8.40 ± 0.02 (1σ) followed by a crustal-stage μ2 value of 11.6 ± 0.2 (1σ) (see Figure 4), values that are in complete agreement with those commonly suggested for mantle and upper continental crust [Asmerom and Jacobsen, 1993; Kramers and Tolstikhin, 1997]. A similar calculation for κ1 and κ2 values (232Th/238U) using 208Pb/204Pb ratios is unconstrained because the two κ values are interdependent, but κ is, on average, around 4.0. Using our μ1 and μ2 values and κ values of 4.0, we can calculate the Pb isotopic composition of upper continental crust as a function of its average extraction time from the mantle (Table 1). Nd model ages suggested for average upper continental crust range between 1.5 and 2.2 Ga [Goldstein and Jacobsen, 1988; Jacobsen, 1988] and can be used to estimate the average crustal Pb isotopic composition. For an upper crust with an average age of 1.5 Ga, the 206Pb/204Pb ratio is 18.76 and it increases to 19.24 for an average age of 2.2 Ga as shown by the black line in Figures 4b and 4d. Using the commonly accepted average value of 1.8 Ga for the upper crust, we suggest the following Pb values: 206Pb/204Pb = 18.96 ± 0.06(1σ), 207Pb/204Pb = 15.73 ± 0.02(1σ), and 208Pb/204Pb = 39.16 ± 0.06 (1σ) (Table 1). If, in the future, a different average age for crust differentiation is established using independent methods, another value can be easily calculated.
Table 1. Our Pb Estimates for Average Upper Continental Crust
Methods Used to Estimate Pb Isotopic Values
Average Model Age (Ga)
Our Pb estimates were calculated using μ1 = 8.40, μ2 = 11.6, and κ1 = κ2 = 4.0 from the isotope ratios of primeval Pb of Canyon Diablo as published by Tatsumoto et al. . Our preferred values are indicated in bold and correspond to an average age of 1.8 Ga for the upper crust. Average Pb isotopic values for the upper continental crust estimated by other studies [Zartman and Doe, 1981; Asmerom and Jacobsen, 1993; Kramers and Tolstikhin, 1997; Hemming and McLennan, 2001; Millot et al., 2004] are also reported for comparison.
 Our suggested 206Pb/204Pb ratio for the upper crust is always displaced to the left of former estimates in Figure 4b notwithstanding whether they are based on Pb evolution models [Kramers and Tolstikhin, 1997; Zartman and Doe, 1981] or on suspended river sediments [Asmerom and Jacobsen, 1993; Millot et al., 2004] (Table 1). We believe that the literature values derived from river sediments are too high because they do not take into account the “heavy mineral effect” on Pb isotopes. Our Pb estimate is also less radiogenic than that of Hemming and McLennan  which is based on deep-sea turbidite data (Figure 4). The difference might be attributed to addition of labile lead in the marine environment due to the very short residence time of dissolved lead in seawater. However, marine hydrogenous sediments such as ferromanganese crusts have crustal-like Pb isotopic compositions (i.e., 18.6–19.3 for 206Pb/204Pb ratios, 15.6–15.7 for 207Pb/204Pb ratios, and 38.6–39.5 for 208Pb/204Pb ratios) [von Blanckenburg et al., 1996] and addition of such lead cannot explain the observed difference between the two estimates. Instead, we think that since turbidites are coarse-grained sediments containing relatively high amount of heavy minerals, their Pb values are also shifted toward too radiogenic values due to a “heavy mineral effect.”
 Our estimate strongly resembles the average composition of global subducting sediments [Plank, 2013] (Figure 4) demonstrating that on average material recycled back into the mantle through subduction zones is not affected by the “heavy mineral effect” and has the same Pb isotopic composition as the upper continental crust. Finally, our Pb estimate for the upper crust still lie on the right on the 4.55 Ga geochron, i.e., the meteorite isochron on which the Earth as a whole is expected to plot (Figure 4), a shift called the first Pb paradox [Allègre, 1969]. Our new Pb values do not resolve this Pb paradox, confirming that an unradiogenic lead reservoir is still needed to balance the radiogenic composition of the upper continental crust.
 Our results clearly show that Pb isotopes of river sediments are biased by a “heavy mineral effect” caused by mineral sorting processes during fluvial transport. Both bed loads and suspended loads have Pb isotopic compositions more radiogenic than their source rocks due to the presence of a few but extremely radiogenic zircons, monazites and allanites. Consequently, Pb isotopic variability of sediments cannot be used as a proxy for source and/or anthropogenic changes unless mineralogical effects are carefully evaluated and corrected for. We suggest that estimates of the average Pb isotopic composition of the eroded crust should be preferentially evaluated using Pb-rich K-feldspars separates and/or clay fractions rather than bulk sediment. After correction of the “heavy mineral effect,” we establish a new Pb isotopic value for the upper continental crust.
 We thank A. Galy, V. Galy, and M. Lupker for sediment sampling, S. Andò for his help during mineral separations, S. Bureau for her help in the clean lab, P. Telouk for assistance during MC-ICP-MS measurements at Lyon, and N. T. Arndt for constructive discussions that helped improving the style and content of the manuscript. We are also grateful to B. S. Kamber, an anonymous reviewer and the editor L. Derry for their constructive comments that greatly improved the content of the discussion. This study was supported by funding from CNRS and INSU programs.