We investigate the use of loess as a proxy for the concentration and isotopic composition of highly siderophile elements, specifically Os, in the upper continental crust. The 187Os/188Os, platinum group element, and Re concentrations of 16 loess samples from China, Europe, and South America, previously analyzed for major, trace element, and Sr and Nd isotope composition, reveal subtle differences between loess provinces. Despite those differences, the 187Os/188Os of 1.05 ± 0.23 is surprisingly homogenous. Average 187Os/188Os as well as average Os (31 pg/g) and Ir (22 pg/g) concentrations are similar to the lower limit of previous estimates for average upper continental crust, whereas Ru, Pt, and Pd concentrations are intermediate between previous estimates. We argue that hydrogenous enrichment of Os in riverine sediments led Esser and Turekian  to overestimate the Os concentration of upper continental crust (50 pg/g). On the basis of this argument and correlations with major and trace elements we propose that average platinum group element concentrations of loess (i.e., 31 pg Os/g, 22 pg Ir/g, 210 pg Ru/g, 510 pg Pt/g, 520 pg Pd/g) are a proxy for the upper continental crust. We further suggest that the nonchondritic average Os/Ir of 1.4 reflects the combined effects of radiogenic ingrowth of Os from Re decay over the mean lifetime of the upper continental crust and preferential return of Os to the crust during subduction. Rhenium concentrations scatter significantly, with highest values in loess derived from organic-rich sedimentary rocks. Low median Re concentrations most likely reflect depletion of loess in organic matter, an important sink for Re in the upper continental crust. An average 187Re/188Os of 34.5 was calculated on the basis of the measured 187Os/188Os and Nd model ages. This value corresponds to a Re concentration of 198 pg/g. Correcting measured 187Os/188Os = 1.05 and inferred 186Os/188Os = 0.119871 (from 190Pt/188Os = 0.0176) for the older mean age (2.2 Gyr) of upper continental crust compared to loess (1.6 Gyr) yields average upper crustal 187Os/188Os of 1.40 and 186Os/188Os of 0.119885.
 The continental crust is severely depleted in platinum group elements (PGE: Ru, Rh, Pd, Os, Ir, Pt) relative to the Earth's core and mantle and contains <0.01‰ of the terrestrial PGE budget. This depletion reflects the siderophile and chalcophile nature of PGE as well as their compatible behavior during mantle melting [e.g., Hart and Ravizza, 1996]. Owing to the lithological complexity of the continental crust, PGE concentrations, 187Re/188Os and 187Os/188Os in crustal rocks vary by several orders of magnitude. This makes estimating average crustal compositions a challenging task.
Calculated value for bulk upper continental crust using measured values for average global river sediments (TDM = 1.45 Gyr) and correcting the isotopic composition using a Nd crustal residence age of 2.2 Gyr and a 187Re/186Os of 400 (i.e., 187Re/188Os = 48.1). Note that Esser and Turekian  calculate a 187Os/186Os of 15.8 (p. 3100) but cite a value of 16 in the conclusions of their paper (p. 3102). This explains the discrepancy between the 187Os/188Os tabulated above and the 187Os/188Os value of 1.92 tabulated by Saal et al. [1998, Table 1].
Esser and Turekian  consider the commonly cited 187Os/186Os value of 10.5 (i.e., 187Os/188Os = 1.26) as representative of currently eroding upper continental crust (i.e., worldwide river sediment), not bulk upper continental crust.
Esser and Turekian  analyzed Fe and Os concentrations in a North American loess (Roxana silt) and a Mississippi River sediment sample from the TAMU station. They argue that the positive correlation between Fe and Ir in Mississippi River sediments [Fenner and Presley, 1984], and the fact that Os/Ir are close to unity in a wide variety of terrestrial and extraterrestrial materials despite orders of magnitude variations in absolute concentrations, can be used to derive an average Os concentration of 50 pg/g for the upper continental crust (Figure 2b). This estimate is based on an average Fe concentration of 3.5 wt % for the upper continental crust [Taylor and McLennan, 1985]. We argue below that the positive correlation between Ir and Fe in Mississippi River sediments has a steep slope that is caused by scavenging of Ir (and Os) onto Fe-Mn oxyhydroxide coatings of riverine particles in sediments with elevated Fe concentrations. The study by Esser and Turekian  is the only comprehensive attempt to estimate the average 187Os/188Os of the upper continental crust. Their estimates of 187Os/188Os ∼ 1.26 for average global river sediments (i.e., currently eroding continental crust) and 187Os/188Os ∼ 1.9 for average 2.2 Gyr old upper continental crust are widely used in the literature to characterize this important reservoir. The average Os concentration and isotopic composition together with an estimate for the average age of the continental crust from Nd isotope systematics were also used by Esser and Turekian  to estimate the average Re concentration of the upper continental crust at ∼400 pg/g.
Wedepohl  reports average concentrations of Ru, Rh, Pd, Ir, and Pt for 17 European graywackes, analyzed by G. Hartmann using NiS fire assay followed by Te coprecipitation. Unfortunately, no uncertainties are reported for the concentration estimates, and the raw data have not been published (G. Hartmann, personal communication, 2000).
Schmidt et al. [1997a, 1997b] argue that impact melts can be used as regional averages of the impact target areas. Linear trends in plots of Ru, Rh, Pd, Re, Pt, and Au versus Ir concentrations are interpreted as mixing lines between a meteoritic component and a regional crustal component. They estimate average crustal Ru, Rh, Pd, and Pt concentrations by extrapolating the linear trends to an average crustal Ir concentration of 30 pg/g, thereby eliminating contributions from the impactor. The correlation of Pd concentrations with Ir is shown in Figure 2a as an example for these correlations and the uncertainties involved in extrapolating the trends to crustal Ir concentrations. The scatter about individual regression lines as well as the scatter between different impact target areas is significant. This leads to significant uncertainties in the estimated average crustal abundances. Moreover, crustal end-member concentrations for Os cannot be derived because of pervasive Os depletion relative to Ir, presumably caused by volatile loss of Os during impact. The most notable differences between the average crustal estimate of Wedepohl  and Schmidt et al. [1997a, 1997b] are the significantly lower Ru, Pd, and Pt concentrations in European graywacke [Wedepohl, 1995]. We will address this difference in section 4.2.
 Other estimates (see Table 1) for average crustal PGE concentrations include Pd and Ir concentrations for bulk, upper, and lower continental crust by Taylor and McLennan [1985, 1995] that are mainly based on a large number of individual rock analyses from the Canadian shield by Shaw et al. [1967, 1976, 1986] and Ir analyses of Mississippi River sediments by Fenner and Presley . Gao et al.  recently published Pd and Pt concentrations for crustal sections in China, analyzed by emission spectrometry after fire assay preconcentration. Palladium/Pt values are close to unity, with average upper, middle, and lower crustal concentrations (carbonate free) of ∼1.2 ng/g, ∼0.9 ng/g, and ∼2 ng/g, respectively. Average Os and Re concentration for the lower continental crust have been estimated by Saal et al.  at 49 pg Os/g and 184 pg Re/g on the basis of median values for granulite-facies lower crustal xenoliths from Northern Queensland, Australia. These values are similar to median Os (31 pg/g) and Re (161 pg/g) concentrations in lower crustal xenoliths from Arizona [Esperança et al., 1997].
 In this study we investigate whether PGE concentrations in loess are a valid proxy for the average composition of the upper continental crust. The use of fine-grained detrital sediments as a proxy for the composition of the upper continental crust has a long history in geochemistry, dating back to Goldschmidt , and these sediments have been successfully used for estimating upper crustal concentrations of major and some trace elements, such as the rare earth elements (REE) [e.g., Taylor and McLennan, 1985, 1995]. However, the applicability of this concept for trace elements such as PGE that are hosted by trace phases with unusual densities and susceptibility to alteration such as sulfides and PGE alloys is not (a priori) clear. Of all fine-grained detrital sediments, loess, a predominantly eolian deposit typically formed under arid glacial conditions, may constitute an upper crustal analog least affected by hydrogenous alteration.
2. Samples and Analytical Procedure
 The samples used in this study have been previously analyzed for major and trace element concentrations as well as Sr and Nd isotope composition. The results have been published by Gallet et al.  and Jahn et al. , and details of the analytical procedures for those elements/isotopes can be found there. As we are using the same sample names, the analytical results can be easily found in earlier publications and are, with the exception of Nd isotope data and model ages, not tabulated in Table 2.
Table 2. Re-Os-PGE Systematics, 187Os/188Os, 143Nd/144Nd, Nd Model Ages, and Trace Element Concentrations of Loessa
Sample names correspond to those used by Gallet et al.  and Jahn et al. . Complementary major and trace element and 87Sr/86Sr data are reported there as well, unless they are listed above. Neodymium isotope data and Nd mantle extraction ages listed above are from Gallet et al. . Trace element concentrations of Chinese loess were measured by ICP-MS after flux fusion at the Centre de Recherches Pétrographiques et Geochimiques (CNRS) in Nancy, France, with a relative uncertainty of <5% for Cr, V, Ga, and Zn, <10% for Co, and <15% for Cu. No uncertainties are given for Ni concentrations.
Osmium concentrations were duplicated by analyzing a split of the residual sparging solution by ICP-MS using conventional liquid uptake. Values in the first column are reported as ICP-MS sparging/conventional ICP-MS. Values for the WHOI Os standard, measured to monitor accuracy and precision of the sparging method, are 0.17514±1.05% (1σ, n = 5) for 80 pg total Os, 0.17453±0.88% (1σ, n = 10) for 80–400 pg total Os and 0.17404±0.21% (1σ, n = 9) for 1.23 ng total Os. These values are within uncertainty identical to the long-term average of 0.17410±0.14% (1σ, n = 16, 0.8 pg to 100 ng total Os) for the same standard using NIMA-B in negative ion mode (discrete dynode electron multiplier operated in analog mode; for details of the data reduction procedure, see Peucker-Ehrenbrink et al. ). The values are corrected for blank contributions.
Rhenium was determined by isotope dilution on a separate ∼150 mg split by acid dissolution, anion column separation followed by ICP-MS (Finnigan Element). Values are corrected for blank (0.6 pg total Re).
 Samples were ground in an agate mortar prior to analysis. About 10 g each was spiked with a mixed PGE spike enriched in 99Ru, 105Pd, 190Os, 191Ir, and 198Pt. The PGE were preconcentrated using a NiS fire assay technique developed at the Woods Hole Oceanographic Institution [Ravizza and Pyle, 1997]. The 187Os/188Os and Os concentrations were measured by sparging volatile OsO4 directly into a single-collector, magnetic sector inductively coupled plasma-mass spectrometer (ICP-MS) (Finnigan ELEMENT) according to a procedure described by Hassler et al. . The external reproducibility of 187Os/188Os for 80 pg Os (in-house standard) is ∼1% (1σ) for the sparging technique. The average 187Os/188Os of 80 pg in-house standard analyses is, within uncertainty, identical to ICP-MS analyses of the same standard at higher concentrations (0.41 and 1.23 ng) as well as the long-term average of the same standard analyzed by negative thermal ionization mass spectometry (N-TIMS) (see Table 2). Subsequent to 187Os/188Os analyses, the liquid residues after sparging were analyzed for Ru, Pd, Os, Ir, and Pt concentrations using conventional liquid uptake into the ICP-MS. To check the internal consistency of the calculated concentrations, Ru, Pd, Os, and Pt concentrations were calculated using two isotope ratios. In most cases the calculated concentrations agree to within a few percent of each other. Osmium concentrations determined by sparging (first number in first column, Table 2) agree very well with Os concentrations determined by ICP-MS using conventional liquid uptake of the residual sparging solution (second number in first column, Table 2). However, Ru concentrations calculated using 99Ru/101Ru and 99Ru/102Ru corrected for interferences from Pd, sometimes agree less well. These unexplained discrepancies are generally smaller than 20% but reach 30% in sample 24–26. In addition, blank corrections for Ru are relatively large (up to 44% at low concentrations), adding to the overall higher uncertainty of the reported Ru concentrations. Relative blank corrections for the other PGE are Pd, 5–36%; Os, 1–3%; Ir, 2–14%; and Pt, 2–24%. Rhenium concentrations were determined by isotope dilution ICP-MS on separate 0.1–0.2 g splits of the loess powder used for Os isotope and PGE analyses. The samples were dissolved in HF-HNO3-HCl and equilibrated with the 185Re spike in all-Teflon microwave vessels. Subsequently, Re was separated on a 1 mL column of AG1X8 (200–400 mesh) in 0.25N HNO3, eluted with 8N HNO3 and analyzed by ICP-MS. Blank corrections for Re do not exceed 2%. Additional details on the analytical procedures are given in Table 2.
 We have analyzed six loess samples from the central Chinese loess plateau, three loess samples from a drill core in central eastern Argentina, and seven loess samples from northern (Svalbard) and central (United Kingdom, France, Belgium) Europe. The 187Os/188Os values range from 0.470 to 1.564, with an average/median value of 1.05/1.06 and a weighted average of 1.06. Although slightly lower than the 187Os/188Os of 1.26 for global river sediments [Esser and Turekian, 1993], the loess average is indistinguishable from this estimate at the 99.5% confidence level (i.e., 1.05 ± 0.23). Osmium concentrations range from 19 to 100 pg/g, with an average/median value of 31/29 pg/g, significantly lower than the value of 50 pg/g reported by Esser and Turekian . Iridium concentrations are even lower than Os concentrations and range from 6 to 41 pg/g, with an average/median of 22/19 pg/g. This results in suprachondritic Os/Ir of ∼1.4 (1.25, if sample LO94 with an Os/Ir of 4.3 is excluded). Ruthenium concentrations vary between 37 and 567 pg/g, with an average/median concentration of 210/118 pg/g. These concentrations are intermediate between values reported by Wedepohl  and Schmidt et al. [1997a, 1997b]. Loess from Europe appears to be depleted in Ru relative to other loess deposits, consistent with low Ru concentrations reported by Wedepohl  for a central European graywacke composite. Platinum concentrations vary between 103 and 1938 pg/g, with an average/median value of 506/335 pg/g, whereas Pd concentrations range from 137 to 1397 pg/g, with an average/median value of 518/406 pg/g. These concentrations are within the range of concentrations previously estimated for the upper continental crust [e.g., Taylor and McLennan, 1995; Wedepohl, 1995; Schmidt et al., 1997a, 1997b] and are most similar to values reported by Wedepohl . Rhenium concentrations vary by nearly two orders of magnitude from 51 to 3588 pg/g, with an average/median value of 367/110 pg/g. Loess sample LO94 from Svalbard, characterized by high Os concentration, radiogenic 187Os/188Os , and suprachondritic Os/Ir of 4.3, also has the highest Re concentrations and highest 187Re/188Os (201) of all loess samples analyzed. In general, Re concentrations in loess are lower than values previously estimated for the upper continental crust (400 pg/g [Esser and Turekian, 1993]). We discuss the difficulties in estimating reliable Re concentrations for the upper continental crust in more detail in section 4.6. The 190Pt/188Os values range from 0.0059 to 0.0475, significantly lower than the value of ∼0.056 calculated from Os and Pt concentrations of Schmidt et al. [1997a, 1997b] (assuming a 187Os/188Os of 1.4; see below). Using average Os (31 pg/g) and Pt (510 pg/g) concentrations as well as 187Os/188Os of 1.05 of loess, we calculate an average 190Pt/188Os of 0.0176. Neodymium isotope data for the loess samples are consistent with an average mantle-extraction age (TDM) of 1.6 Gyr (see Gallet et al.  for discussion). Taking the present-day 186Os/188Os ([0.1198340 [Brandon et al., 2000]) and 190Pt/188Os of the depleted mantle (0.001734; recalculated from Anders and Grevesse , Walker et al. , and Begemann et al. ) together with our estimate of the average 190Pt/188Os of loess, we estimate the 186Os/188Os of average loess at 0.119871. Osmium isotope values for average 2.2 Gyr old upper continental crust can be calculated using the approach of Esser and Turekian  and average Os and Pt concentrations of 31 and 510 pg/g, respectively. Average upper crustal 187Os/188Os and 186Os/188Os are 1.40 and 0.119885, respectively (see Esser and Turekian  for a discussion on potential pitfalls in equating Nd model ages with Os model ages).
4.1. Loess as a Proxy for the Upper Continental Crust: Major and Trace Elements
 As discussed previously in the literature, loess predominantly is an eolian sediment formed during glacial stages and often has experienced some fluvial transport [e.g., von Richthofen, 1877; Marosi, 1975; Taylor et al., 1983; Liu et al., 1985; Zhang et al., 1991]. It is critical for the interpretation of geochemical data of loess as a crustal analog to understand the nature and timing of element fractionation processes. For instance, Nesbitt and Young's  chemical index of alteration (CIA) (as defined by McLennan ) could reflect the sum of alteration processes detrital particles have experienced on their way from their unaltered source to the deposition as loess. Alternatively, the CIA could be a measure of the relative contribution of altered rocks (i.e., cannibalistic recycling of sediments) to loess. Figure 3 illustrates that loess plots between estimates of upper/bulk continental crust and average shale/sandstones [Condie, 1992] on a trajectory pointing away from primary minerals such as plagioclase, hornblende, and clinopyroxene toward secondary minerals such as illite, kaolinite, and chlorite. The temporal evolution of the CIA can be estimated with radiogenic isotope systems that are sensitive to surficial fractionation, e.g., the Rb/Sr system. Recent alteration during loess formation leads to changes in 87Rb/86Sr that are decoupled from changes in 87Sr/86Sr. Conversely, ancient Rb/Sr fractionation affects the 87Sr/86Sr, as this ratio is the time-integrated measure of the 87Rb/86Sr. Loess analyzed in this study, to first order, is characterized by a positive correlation between 87Sr/86Sr and 87Rb/86Sr (r2 = 0.884, excluding three samples from Argentina with young TDM(Nd) and the loess from Belgium, for which we do not have complementary chemical analyses), consistent with significant contributions of previously altered rocks to loess. This interpretation is also consistent with two other, internally consistent facts: (1) ∼66% of the land surface is covered by sedimentary rocks [Blatt and Jones, 1975], and (2) the average contribution of recycled (i.e., previously weathered) material in typical sediments, although highly variable, has been estimated at ∼65% [e.g., Veizer and Jansen, 1979; Taylor et al., 1983]. Much of the altered signature observed in loess thus seems to be inherited from previous weathering cycles, rather than being produced during loess formation.
4.2. Platinum Group Element Systematics
 Major and trace element concentrations as well as Nd and Sr isotope systematics of loess analyzed in this study have been discussed elsewhere [e.g., Gallet et al., 1998; Jahn et al., 2001]. We avoid duplicating these discussions by summarizing that many elemental (e.g., REE pattern, La/Th, Th/U) and isotope tracers (e.g., Nd, Sr) point to a close similarity between loess and the upper continental crust, with a considerable sedimentary contribution. This is demonstrated in three diagrams for Chinese loess (Figure 4, top), Argentine loess (Figure 4, middle), and European loess (Figure 4, bottom) that show element concentrations normalized to upper crust [McLennan, 2001]. For comparison, the range of literature estimates of upper crustal composition is also shown [Wedepohl, 1981, 1995; Weaver and Tarney, 1984; Taylor and McLennan, 1985, 1995; Shaw et al., 1986; Condie, 1992] (vertical black lines in Figure 4). Figure 5 summarizes primitive mantle-normalized [McDonough and Sun, 1995] PGE pattern for loess from China (top), Argentina (middle), and Europe (bottom).
 Most major and trace element concentrations in Chinese loess fall within the range of previous estimates of upper crustal composition. The exceptions are depletions in Ba and Na, known to be mobile during alteration. Cesium concentrations are generally elevated, reflecting absorption onto clay minerals (see discussions by Gallet et al.  and Jahn et al. ). Formation of soil carbonates during loess deposition/soil formation generally leads to very variable Ca concentrations in loess. Most important for our discussion is the fact that concentrations of compatible elements such as Mg, Cr, and Ni fall within the range of previously reported concentrations for upper continental crust. This increases our confidence that loess formation does not systematically discriminate against host phases of compatible trace elements such as the PGE but does not exclude the possibility that PGE concentrations are biased. Primitive mantle-normalized PGE concentrations of Chinese loess show a pronounced step pattern (low Os and Ir, high Ru, Pt, Pd, and Re), typical of continental crust.
 Compared to Chinese loess, loess from Argentina is slightly more fractionated relative to average upper continental crust. Values for K, Na, Ga, Re, Cu, and Mg are lower than crustal averages. Calcium concentrations are highly variable. Gallet et al.  emphasize that Argentine loess is considered most representative of eolian sediments in South America. Volcanoclastic input (e.g., ɛNd (0) of −6 to −1.5, relatively young TDM(Nd) of 1.0–1.4 Gyr) as well as considerable riverine transport contribute to the slightly more fractionated element pattern relative to average upper continental crust. Normalized platinum group element patterns are, on average, lower than those of Chinese loess and in between the estimates of Wedepohl  and Schmidt et al. [1997a, 1997b].
 European loess is characterized by the most fractionated element pattern of all loess provinces studied. Concentrations of Rb, Ba, K, Sr, Na, Ga, Al, Re (except LO94), Cu, and Mg are lower than all upper crustal estimates. In contrast, Zr and Si concentrations are systematically high, indicative of enrichment of zircon and quartz [e.g., Taylor et al., 1983; McLennan, 2001]. Gallet et al.  speculate that the more strongly fractionated pattern of European loess reflects greater lithologic varibility between the small, local loess deposits in Europe, compared to continent-scale deposits in Argentina and China. Normalized platinum group element patterns mimic those determined by Wedepohl  for a European graywacke composite. The agreement between our analytical data for European loess and Wedepohl's  data for European graywacke, determined by very different analytical techniques, indicates that PGE patterns vary regionally depending on predominant lithologies. Sample LO94 from Svalbard is the clearest example of such a regional, lithologically biased pattern (gray line in Figure 5, bottom). This has implications for calculating average PGE concentrations for the upper continental crust. While our initial intent was to use PGE patterns only from the most upper-crust-like loess provinces (e.g., Chinese loess plateau), the regional variability of PGE patterns prompted us to use a global mean for calculating a new set of PGE concentrations of global loess. These values are Os = 31 pg/g, Ir = 22 pg/g, Ru = 210 pg/g, Pt = 510 pg/g, and Pd = 520 pg/g.
4.3. Effects of Eolian Transport on Element Budgets of Loess
 Although incorporation of previously altered material in loess contributes to elevated CIA values, fractionation processes during loess formation also contribute to the chemical signature of loess. For example, using major element systematics, Gallet et al.  and Jahn et al.  show that worldwide, loess trends toward the chemical composition of sandstones. This trend is most convincingly seen in loess from the central Chinese loess plateau, where a grain-size gradient from the NNW (sandy loess) to the SSE (clayey loess) reflects sorting of eolian particles during transport from NNW to SSE. This granulometric gradient is reflected in chemical gradients, such as the observed decrease in SiO2 content of the Malan loess from ∼66 wt % in the north to ∼57 wt % in the south and the increase in Al from ∼5.5 wt % in the north to ∼7 wt % in the south [Eden et al., 1994]. Other elements such as Ti, Mn, Fe, Co, Ni, Cu, Zn, and Mo follow trends similar to Al [e.g., Wen et al., 1987; Eden et al., 1994]. Loess on the central Chinese loess plateau, derived from a remarkably homogenous source [Jahn et al., 2001], can thus be considered the residue of a natural wind-tunnel experiment with atmospherically transported particles. We currently do not know to what extent this eolian fractionation affects the platinum group element budget of loess. The fact that Ni, an element that correlates well with Os in most mantle-derived rocks, varies systematically from NNW to SSE may indicate that PGE are also affected by eolian fractionation. The magnitude of this effect critically depends on the nature (size, density) of the host phases of PGE in loess. However, the Chinese loess samples analyzed in this study exclusively come from the central, silty loess zone and have average Ni concentrations within the range of estimates for the upper continental crust. The generally poor correlation of PGE with elements enriched in heavy minerals such as zircon (e.g., Zr), known to be enriched in loess [e.g., Taylor et al., 1983; McLennan, 2001], lends further support to our contention that PGE concentrations in loess analyzed in this study are valid proxies for PGE concentrations in the upper continental crust. We therefore consider concentrations given above for global loess to be representative for the upper continental crust.
4.4. Osmium and Iridium Concentrations
 Osmium, and particularly Ir, concentrations in loess are significantly lower than in previous estimates of upper (and bulk) continental crust. Using Os and Ir as an example, we argue below that previous estimates of these elements are biased toward high values. The most widely used estimate of Os in the continental crust is the estimate of 50 pg/g by Esser and Turekian . This estimate is based on Os analyses of two samples: the Roxana silt, an eolian deposit in the Mississippi River valley, and Mississippi River sediment from the TAMU Station near the river delta. Together with Ir analyses of Mississippi River sediments [Fenner and Presley, 1984], Os and Ir concentrations define a linear trend when plotted against the Fe concentrations in the same samples (Figure 6, stippled line). Using an Fe concentration of 3.5 wt % for the upper continental crust [Taylor and McLennan, 1985], Esser and Turekian  use the observed linear correlation to calculate an average Os and Ir concentration in the upper continental crust of 50 pg/g (Figure 6).
 Loess analyzed in this study defines a linear trend with a significantly shallower slope in the same diagram, with, on average, higher Os than Ir concentrations (Figure 6, Os: r2 = 0.3, solid line; Ir: r2 = 0.27, linear correlation not shown). Interestingly, the Roxana silt (triplicate analyses by Esser and Turekian ) plots within uncertainty on the same shallower trend. This indicates that only riverine sediments with elevated Fe concentrations are particularly rich in Os (TAMU Station) and Ir. We suspect that scavenging of Os and Ir onto Fe-oxyhydroxide coatings of particulate riverine matter has led to hydrogenous Os and Ir enrichment. The high Os concentration of 50 pg/g calculated by Esser and Turekian  thus most likely reflects hydrogenous enrichment of Os and does not reflect a primary upper crustal value. Although two studies have shown that labile Os may be isotopically different than weathering residues [Peucker-Ehrenbrink and Blum, 1998; L. A. Jaffe et al., Effects of weathering of black shales on the mobility of rhenium, platinum group elements and organic carbon, submitted to Earth and Planetary Science Letters, 2001 (hereinafter referred to as Jaffe et al., submitted manuscript, 2001)], such processes affect the solid residue much less than the fluids and thus can be neglected here. Following the approach of Esser and Turekian  and using the significantly shallower trend defined by our loess data, we calculate a value of 28 pg Os/g for the upper continental crust. Owing to the shallow slope of the correlation, this value is nearly identical to the average/median Os concentration in loess of 31/29 pg/g. Among the major elements, MgO and SiO2 exhibit tighter correlations with Os than Fe2O3 used above. An independent confirmation of the estimate derived by using Fe is obtained by interpolating the positive linear correlation (r2 = 0.65, without sample LO94 and K-14) between Os and MgO to an average upper crustal MgO concentration (Figure 7, blue circles). A similar, although less well constrained, value is calculated on the basis of interpolating the negative correlation (r2 = 0.38) between Os and SiO2 to an average upper crustal SiO2 content (Figure 7, red squares). Both correlations yield upper crustal Os concentrations of 28–32 pg/g, similar to average Os concentrations in loess. Iridium concentrations scatter significantly but are generally lower than Os concentrations. The average Ir concentration in loess of 22 pg/g may thus be a good estimate of the average Ir concentration in the upper continental crust. This is confirmed by the positive correlation of Ir with Co (r2 = 0.48) that yields an Ir concentration between 17 and 25 pg/g at upper crustal Co concentrations of 10–18 μg/g.
 Suprachondritic Os/Ir (i.e., Os/Ir > 1.01 in CI chondrite [Anders and Grevesse, 1989]; Os/Ir > 1.07 in Bulk Earth [Allègre et al., 2001]), observed in the majority of loess samples analyzed in this study, pose a problem, because Os and Ir are rarely fractionated in geologic materials despite orders of magnitude variations in absolute concentrations. Such fractionations have been observed in the sedimentary environment [e.g., Ravizza and Pyle, 1997; Peucker-Ehrenbrink et al., 1995; Peucker-Ehrenbrink, 1996; Peucker-Ehrenbrink and Hannigan, 2000] and in subduction zone settings [McInnes et al., 1999]. Generally, slowly accumulating oxic sediments are characterized by preferential enrichment of Ir, whereas more rapidly accumulating reducing sediments are preferentially enriched in Os. As these fractionation processes are restricted to sedimentary environments, they should not affect bulk upper crustal estimates, unless oxic sediments are preferentially recycled back into the mantle relative to anoxic sediments. As there is no indication for such differential subduction, disproportionate contribution of organic-rich sediments to loess needs to be evaluated as a reason for suprachondritic Os/Ir in loess before other processes are invoked. Loess from Svalbard (Spitzbergen) is characterized by the highest Os/Ir of ∼4.3, consistent with significant contributions from organic-rich sediments. This is consistent with the regional geology of loess in the lower Adventdalen, a glacial valley on Svalbard that is characterized by large exposures of Jurassic black shales [e.g., Bryant, 1982; Hjelle, 1993]. Platinum group element patterns of loess from Svalbard are very similar to those of black shales [e.g., Peucker-Ehrenbrink and Hannigan, 2000; Jaffe et al., submitted manuscript, 2001]. In addition, loess LO94 is relatively radiogenic (187Os/188Os of 1.236), consistent with radiogenic ingrowth at high 187Re/188Os since deposition in Jurassic seawater. Excluding sample LO94, average Os/Ir in loess decrease from ∼1.4 to ∼1.25. This indicates that Os/Ir higher than unity may be characteristic for the upper continental crust in general. Although nonchondritic Os/Ir estimates for the upper (and bulk) continental crust have been published before [e.g., Taylor and McLennan, 1995], potential fractionation mechanisms have never been discussed. We explore potential reasons for a nonchondritic Os/Ir below.
 Although a disproportionate contribution of organic-rich sediments to loess analyzed in this study cannot be excluded without additional Corg data, two additional processes preferentially enrich Os in the continental crust. First, ∼14% of the present-day Os inventory (i.e., ∼4.3 pg/g) was generated over the past 2.2 Gyr through decay of Re (OsRe, i.e., radiogenic Os from Re decay) in the continental crust, leading to an increase in 187Os/188Os from primitive mantle values 2.2 Gyr ago to a present-day value of 1.40 ± 0.30 (99.5% confidence level). This effect has led to a secular increase in Os/Ir of the bulk continental crust. Second, several studies have inferred preferential return of subducting Os to the overlying crust in some subduction zones [e.g., Brandon et al., 1996, 1999; McInnes et al., 1999]. We can estimate, to first order, the efficiency of the subduction zone recycling process for Os (OsSF, i.e., subduction flux of Os) over the average age of continental crust using average Os (Osm = 31 pg/g) and Ir (Irm = 22 pg/g) concentrations for the upper continental crust. We define the Os excess (Osx) above chondritic (CI) [Anders and Grevesse, 1989] or Bulk Earth (BE) [Allègre et al., 2001] values as
with OsRe = 4.3 pg/g. Given an Osx of 7.5 pg/g (relative to BE) to 8.8 pg/g (relative to CI chondrite), OsZF ranges from 3.2 to 4.5 pg/g. Using the mass of the bulk continental crust (2.1 × 1022 kg [Reymer and Schubert, 1984]) as an upper limit for the Os excess (i.e., Os/Ir upper crust = Os/Ir lower crust), the mass of Os added to the continental crust by preferential recycling in subduction zones is 6.7 − 9.4 × 1010 kg. This is equivalent to a steady state addition over the average age (2.2 Gyr) of continental crust of ∼30–43 kg Os/yr. The maximum loss of Os per gram of subducting slab (igneous part only) can be calculated by assuming that the average production rate of oceanic crust (25 km3/yr [Reymer and Schubert, 1984]; ρ = 3.0 g/cm3) during the Phanerozoic is representative for the past 2.2 Gyr. With this the maximum loss of Os from the subducting slab (OsSF) amounts to 0.4–0.57 pg Os per gram of subducting slab, if steady state between oceanic crust production and subduction is assumed at Phanerozoic rates for the past 2.2 Gyr. As Proterozoic crust production rates were most likely higher than Phanerozoic rates, average Os loss per gram of subducting slab is likely to be smaller. Given an average Os concentration of ∼12 pg Os/g in altered oceanic crust [Peucker-Ehrenbrink et al., 2000] and neglecting the minor portion of Os associated with subducting sediments, a loss of 0.4–0.57 pg Os/g crust amounts to 3–5% of the Os budget of subducting oceanic crust. To first order this maximum recycling efficiency does not seem unreasonable and may help explain the suprachondritic Os/Ir value of the upper continental crust. Preferential return of Os hosted in the sedimentary section of the subducting slab to continental crust during subduction, not explicitly considered in the calculation above, would only lower the fraction of Os lost from the bulk subducting slab.
4.5. Ruthenium, Platinum, and Palladium
 Independent confirmation of average Ru, Pt, and Pd concentrations in loess as upper crustal proxies is difficult, because interelement correlations are generally weak (even if we exclude sample LO94 from the statistical analysis). The positive correlation of Ru contents with MgO concentrations (r2 = 0.5) yields a Ru concentration of ∼340 pg/g at 2.20 wt % MgO [Taylor and McLennan, 1985, 1995; McLennan, 2001]. A similar value (330 pg Ru/g) is obtained if the positive correlation between Ru and Cu (r2 = 0.43) is interpolated to an upper crustal Cu concentration of 25 μg/g [Taylor and McLennan, 1995; McLennan, 2001]. Both estimates are ∼50% higher than the average and three times higher than median Ru concentration in loess. Interelement correlations with Pt are too weak to allow independent confirmation of average Pt concentrations in loess. In contrast, Pd exhibits a number of reasonably strong (r2 > 0.5) positive correlations, most notably with Pb (r2 > 0.67), Co (r2 > 0.65), Ce (r2 > 0.63), and MnO (r2 > 0.6). These correlations yield Pd concentrations of 350–750 pg/g (15–20 μg/g Pb), 350–730 pg/g (10–18 μg/g Co), 550–740 pg/g (57.5–64 μg/g Ce), and 450–500 pg/g (0.08–0.10 wt % MnO). These estimates agree well with the average Pd concentration of 520 pg/g in loess. As correlations with major and trace elements generally agree better with average than with median PGE concentrations, we feel justified in using average rather than median concentrations in loess as a proxy for the upper continental crust.
4.6. The Rhenium Problem
 Average Re concentrations (367 pg/g) are similar to previous estimates of Re in the upper continental crust (400 pg/g [Esser and Turekian, 1993]). However, this average is biased toward a high value by sample LO94 from Svalbard. Median Re concentrations of loess (110 pg/g) are significantly lower than previous estimates for Re in the upper continental crust, indicative of significant Re depletion of loess. It has been previously demonstrated that Corg contents of loess are low (∼0.1 wt % [Schnetger, 1992]). Furthermore, various studies [e.g., Ravizza et al., 1991; Ravizza and Esser, 1993; Colodner et al., 1993; Crusius et al., 1996; Morford and Emerson, 1999] have shown that organic-rich sediments are a significant sink for crustal Re. Several studies have also demonstrated that Re and organic carbon are efficiently lost in the early stages of weathering of organic-rich rocks [e.g., Peucker-Ehrenbrink and Hannigan, 2000; Jaffe et al., submitted manuscript, 2001]. It is therefore likely that Re and Corg are not incorporated into loess according to their crustal abundances. As we do not have Corg data for the loess samples analyzed in this study, we cannot attempt to correct for the missing Re using well-established Corg/Re relationships for organic-rich sediments. We can, however, calculate whether complementary PGE need to be corrected for this effect, if we assume that organic matter is the missing ingredient. The average Re/Os concentration ratio of organic-rich sediments is ∼160. Maximum corrections are calculated by correcting the median Re concentration (110 pg/g) to previous estimates of Re in the upper continental crust (400 pg/g). This requires addition of 290 pg/g Re, equivalent to an addition of <2 pg/g Os. This is consistent with the fact that only a small fraction of the crustal Os budget, but a large fraction of the crustal Re budget, is associated with organic matter [Ravizza et al., 1998; Peucker-Ehrenbrink and Ravizza, 2000]. Iridium corrections will be even smaller than those for Os, because Os/Ir of organic-rich sediments are typically much higher than unity [e.g., Ravizza and Pyle, 1997; Peucker-Ehrenbrink and Hannigan, 2000; Jaffe et al., submitted manuscript, 2001]. For similar reasons, corrections for Pt and Pd are negligible. The large scatter in measured Re concentrations and the lack of Corg data for our loess samples prevent us from deriving meaningful average Re concentrations for the upper continental crust. We therefore adopt the approach of Esser and Turekian  and use an average 187Os/188Os of 1.05 and an average Nd model age of 1.6 Gyr to calculate an average 187Re/188Os of 34.5. Given an average Os concentration of 31 pg/g, this ratio corresponds to a Re concentration of 198 pg/g, half the value of 400 pg Re/g estimated by Esser and Turekian . Improving our understanding of the average Re concentration in the continental crust is a challenging task because it requires properly accounting for the volumetric proportions of a trace lithology in the crust: organic-rich sediments.
 On the basis of PGE data for loess as well as element correlations, we propose revised average concentrations for Os (31 pg/g), Ir (22 pg/g), Ru (210 pg/g), Pt (510 pg/g), and Pd (520 pg/g) in the upper continental crust. To illustrate the insignificance of the upper continental crustal PGE inventories relative to the Bulk Earth budgets, we calculate the fraction (in ppb by weight) of the Bulk Earth PGE budget that currently resides in the upper continental crust. The upper crustal inventories account for only 41 ppb (Os), 31 ppb (Ir), 195 ppb (Ru), 355 ppb (Pt), and 641 ppb (Pd) of the Bulk Earth budgets. The average 187Os/188Os of loess with an average Nd model age of 1.6 Gyr is 1.05 ± 0.23 (99.5% confidence interval). Using loess as a proxy for the upper continental crust and correcting for the younger Nd model age of loess relative to average upper crust (2.2 Gyr), we calculate a 187Os/188Os of 1.40 ± 0.30 (99.5% confidence interval; note that this does not include uncertainties in 187Os/188Os) for the upper continental crust. We suggest that nonchondritic Os/Ir of 1.4 is a signature of the upper continental crust in general and propose that ingrowth of 87Os from decay of 187Re plus preferential return of subducting Os to the crust above subduction zones contribute equally to this bias. The 187Re/188Os values cannot be reliably estimated from measured Re concentrations, because they are biased toward low values, presumably because loess formation discriminates against organic matter, a major crustal Re reservoir. For lack of better information, we use a 187Re/188Os of 34.5 that is based on measured 187Os/188Os values and Nd model ages. The average 190Pt/188Os of the upper continental crust is estimated at 0.0176. Together with an average Nd isotope model age of 1.6 Gyr, we estimate the average 186Os/188Os of loess at 0.119871. Correcting this ratio for the mean age of the upper continental crust of 2.2 Gyr yields an average 186Os/188Os of 0.119885 for the upper continental crust.
 Eolian fractionation needs to be further investigated to demonstrate the degree to which PGE are affected by this process. This, as well as the identification of host phases of PGE and their susceptibility to alteration, will be the focus of future studies.
 J. Han kindly collected loess samples used in this study. We thank Lary Ball and the WHOI ICP Facility for the use of their Finnigan MAT sector field ICP-MS in making the Os isotope, Re, and PGE measurements. Diane McDaniel, Scott McLennan, and Alberto Saal provided insightful reviews that helped to improve the manuscript. We thank Hubert Staudigel for his invitation to contribute to the new GERM theme and for giving us the final push to write this paper by inviting one of us (B.P.E.) to attend the 2001 GERM workshop in La Jolla. B.P.E. also wishes to acknowledge financial support from NSF OCE 9617448 and 9819296. This is WHOI contribution 10,495.