The Eciency of Elemental Geochemistry and Weathering Indices as Tracers in Aeolian Sediment Provenance Fingerprinting

Eight geochemical elements (Al, Ca, Fe, K, Na, Mg, Si, Ti) and 17 associated weathering indices were measured in 34 aeolian source samples and 10 sand dune target sediment samples in three absolute particulate size fractions. For each fraction, three nal composite ngerprints (i.e., geochemical elements only, weathering indices only and a combination of the two) for discerning and ascribing the aeolian sediment sources were selected. The Modied MixSIR Bayesian un-mixing model was used to apportion aeolian source contributions using the nal composite ngerprints. Regardless of the composite ngerprint used, all results across the different size fractions suggested that the south-eastern alluvial fan is the dominant (average contribution 50.6%, SD 19.0%) source of the sand dune samples, with the western alluvial fan being the second most important (average contribution 38.4%, SD 20.4%) source. Comparisons of the posterior distributions for the predicted source proportions generated using the nine composite ngerprints (three kinds of composite ngerprints*three particle size fractions) showed that the composite ngerprints combining the geochemical elements and weathering indices generated the most powerful source material discrimination. Our results demonstrate the use of weathering indices alongside more conventional elemental geochemistry tracers for investigations into sand dune sediment provenance.


Introduction
Wind erosion has three stages comprising abrasion, de ation, and accumulation 1 . Prevailing winds cause the redistribution, accumulation and migration of sand in many regions of the world. The principal controlling factors for sand dune formation include wind direction and velocity, the available supply of sand in source areas subject to wind erosion, source area surface roughness, aggregate stability, presence of crusting, soil texture and moisture, vegetation cover and the presence of residues [2][3][4] .
Effective wind erosion mobilises erodible soil particles, which can be deposited subsequently on agricultural land, roads and railways, and in settlements. Given these on-site and off-site consequences of wind erosion 5,6 , there is a need to determine the key sources of aeolian sediment in order to target and select robust strategies for stabilising the source areas.
Most areas of arid land in central Iran are under threat of aeolian deposits (landforms) resulting from wind erosion. Due to strong prevailing winds and the lack of moisture and vegetation cover, both ne and coarse soil particles are redistributed to depositional zones where sand dunes form. Sand dunes have been estimated to cover 12 million ha in Iran, with approximately 50% being classi ed as active or semiactive 7 . Available solutions for controlling the threat of sand dune formation and migration include stabilizing sand dune areas with mixed petroleum and tar mulching, removing sand dunes, and either biological or mechanical windbreaks 2,8,9 . But, despite application of these interventions, the disbene ts arising from sand dune formation and migration continue to be severe in Iran. Here, for instance, some studies have reported that petroleum and tar mulching whilst an effective method for sand dune stabilization and reducing storm dust 10 can result in environmental pollution resulting from heavy Page 3/28 metals. The alternative approach to management is to focus on identifying the key sources of aeolian sediment to target interventions for decreasing the supply of sand generating off-site aeolian sediment deposits (i.e., sand dunes).
Traditional techniques for quantifying aeolian sediment provenance were based on the shape and texture of dune sands, and an abrasion coe cient (i.e., sphericity of particles). These methods estimate the distance of particle transport and on that basis estimate the sources of dune sand. An alternative approach is to use source ngerprinting which is founded on the direct comparison of the properties, or composite ngerprints, of dune deposits and potential source areas. Early studies based on aeolian sediment properties used light and/or heavy mineral compositions as well as minor element geochemistry [11][12][13][14] . These studies did not, however, use an un-mixing model to estimate the contribution of each source on the basis of the ngerprint properties analysed. More recently, as global uptake of the ngerprinting approach has continued to expand, especially in uvial environments [15][16][17] , applications to aeolian sediment and storm sand tracing have also started to emerge [18][19][20] . Here, since applications to aeolian systems are rudimentary, in comparison with those in uvial environments, there remains a need to explore the most reliable and cost-bene t tracers to discriminate and apportion sources. In this context, weathering indices have proved to be useful for investigating uvial sediment sources [21][22][23] , but their applicability to the aeolian sediment problem warrants further research.
Globally, weathering and erosion generate quartz and silicate dune sediments from quartz-rich rocks but, locally, sand deposits can consist of grains weathered from carbonate or by near-surface chemical and physical processes 24 . Dune sands originating from the weathering of bedrock appear to be limited whereas contributions from eroding uvial sediment deposits are more common 25 . Collectively, this means that physical, chemical, and mechanical weathering can all play a role in the earth surfrace processes generating dune sediments 26,27 . Accordingly, weathering indices have the potential to provide useful tracers for aeolian source discrimination and apportionment since they re ect the susceptibility of potential source areas to wind erosion arising from interactions between climate, geology, pedology, tectonic processes, relief, land cover and human activity 27,28 . The main objective of this study, accordingly, was to apportion aeolian sediment source contributions to aeolian target sediment (i.e., sand dune deposits) samples comprising different particle size fractions in a case study located within central Iran, using geochemical properties and weathering indices as tracers and the Bayesian un-mixing model MixSIR.

Study area
The Hoseinabad Mish Mast region (340 km 2 ) is located in the central part of Iran in boundary Qom and Isfahan Provinces between 51° 02' 29" to 51° 15' 29" E longitude and 34° 18' 06" to 34 34′ 44′′ N latitude. The topography of the study area includes plain, piedmont and hills. Maximum elevations are up to 1039 m whereas the minimum elevation is 805 m above sea level. Eastern parts of the study region comprise hilly terrain (maximum slope of 40%). Much of the study area is, however, at. Land use is dominated by rangelands and cultivated land. Based on the 1:100,000 scale lithology map of Qom (Fig. 1), the main lithology comprises: upper red formation (marl with intercalations of conglomerate); marl with intercalations of sandstone, gypsum, and siltstone; association of marl, shale and greyish green caverneous sandstone with gypsiferous marl; association of greyish red to brown sandstone, conglomerate and marl; conglomerate with calcareous pebbles; red marl, shale, siltstone, dark red gypsiferous sandstone, gypsum and dark green sandstone in some localities); lower red formation (alternation of red and dark grey silty shale, sandy marl, green marl with intercalations of sandstone, gypsum and salt); the Qom formation (marl, limestone, sandstone, shale, gypsum and volcanic rocks in some localities), and a Quaternary unit (young alluvial terraces; old alluvial terraces; recent river deposits; scree; sand dunes; salty mud at). Based on data  collected by the Qom synoptic station, the mean annual rainfall is ~ 144 mm (coe cient of variation 43%). Long term (1986-2015) mean monthly minimum / maximum temperatures are 4.6°C and 31.4°C, respectively. According to the Köppen climate classi cation scheme, the climate of the study area is dry desert. The dominant winds in the study area are western and north-western (Fig. 1).
The case study area is recognised as being fragile and susceptible to wind erosion (Fig. 2a). The village of Hoseinabad Mish Mast has been surrounded by mobile sand dunes (Fig. 2b). The Sarajeh gas installation (one of the most important gas installations in the Qom Province; Fig. 2c) and associated roads (Fig. 2d) as well as the main railway (Fig. 2e) in this region are all threatened by mobile sand dunes. In total, 271 km 2 of this region is identi ed as being at risk from the aeolian sediment problem. In response, interventions comprising petroleum and tar mulching (Fig. 2f) and windbreaks (Fig. 2e&g) have been used to control the threat of the aeolian sediment hazard. These methods do not, however, deliver sustained e ciency meaning that the dunes continue to migrate (Fig. 2e&h). Consequently, there is an urgent need to quantify the sources of the aeolian sediments instead of relying on methods to stabilise sand dunes.

Field sampling, geochemical tracer measurements, and calculation of weathering indices
Identi cation of potential aeolian sediment sources: In order to prepare a map of potential aeolian sediment sources the location of aeolian sediment deposits (i.e., sand dunes) was determined (Fig. 1). Next, the wind rose for the study area based on the wind data collected at the Qom synoptic meteorological station was prepared. By merging the dominant wind directions and the mapped locations of sand dunes and considering the mountains (which act as barriers to the winds) surrounding the study area, the key potential aeolian sediment sources were determined (Fig. 1). Using Google Earth images and eld surveys, three potential sources of aeolian sediment were identi ed: crop lands, western alluvial fans, and south-eastern alluvial fans (Fig. 1).
Aeolian sediment source sampling: based on a random-systematic sampling design, a total of 34 samples were collected from the potential aeolian sediment sources: 14 samples from crop lands, 11 samples from the western alluvial fans, and a further 9 samples from the south-eastern alluvial fans ( Fig. 1). Each sample represented a composite combining ve sub-samples to take account of potential micro-spatial (i.e., within ~ 100 m 2 ) variations in geochemical elements at each individual sampling location selected to represent the sources. The sampling depth was ~ 5 cm to represent the typical depth of wind erosion effects.
Sand dune sampling: 10 samples were collected from sand dunes based on a random-systematic approach (Fig. 1). Each sample was composited from three sub-samples that were collected from the windward, crest, and leeward sides of the sampled sand dunes.
Determining the dominant particle fractions: in order to assess the particle size distribution of the target sand dune samples for con rming the most appropriate fraction for source ngerprinting , dry-sieving was performed using 1000, 500, 425, 250, 180, 150, 125, 75, and 63 µm meshes to determine the dry mass in each fraction. Based on the cumulative frequency curves (Fig. 3) three absolute particle size fractions comprising <125 µm, 125-250 µm, and 250-425 µm, were selected as the dominant fractions.
Measurements for geochemical elements: Acid digestion was used for elemental extraction for all samples of the three fractions. 1 g of sample was weighed; 3 ml of Nitric acid and 9 ml of Hydrochloric acid added and after 24 h, all samples were heated for 2 h on a hotplate at 90 ˚C. After the digestion, the samples were ltered and diluted to a volume of 50 ml. Al, Ca, Fe, K, Na, Mg, Si, and Ti were measured in the solutions using a Varian SpectrAA-20 Plus. An element standard solution (Merck KGaA, Frankfurt, Germany) and standard curve was used to calculate the element concentrations based on the dilution factor.
Evaluation of elemental tracer content in the three particle size fractions: As the absolute particle size analysis suggested a broad size range was representative of the target sediment samples, it was informative to analyse the elemental tracers in the size fractions to con rm it was appropriate to use all fractions for determining sediment sources with the tracers in question. Here, one-way ANOVA was used to evaluate the presence of signi cant contrasts between the content of individual tracers in the different size fractions.
Calculation of weathering indices: 17 weathering indices (Table 1) were calculated. All elements were transformed to oxide percent and then the molecular weights of elemental oxides were calculated using molar mass. Finally, based on the formulas presented in Table 1, the different weathering indices were calculated.
Summary statistics for all geochemical elements and weathering indices are presented in Table 2. Table 2 Tracer concentration (%) and weathering indices data for the different sand dune sources and target sand dune samples, plusthe Kruskal-Wallis H-test results for discriminating the sources using the individual tracers in the three dominant absolute particle size fractions. For tracer abbreviations see Table 1.

Sand dune source discrimination and apportionment
In total, 25 potential tracers comprising eight geochemical elements and 17 weathering indices were tested for discriminating and apportioning sand dune sources in the three particle size fractions based on nal composite ngerprints comprising: 1) using only geochemical elements; 2) using only weathering indices, and; 3) combining the geochemical elements and weathering indices. This yielded a total of nine nal composite ngerprints. In all three approaches, conservation tests were undertaken. Here, in rst step, the tracer values for the target sand dune samples were compared to the range based on the tracer values for the individual sources (i.e., the minimum average source value minus its standard deviation and the maximum source average value plus its standard deviation). Any tracer exhibiting a target aeolian sediment sample value outside of the corresponding source range was excluded from further statistical analysis. In the second step of the conservation tests, the mean tracer values for all target sand dune samples were compared to the minimum and maximum mean values for the sources. Again, violation of this conservation test was used as a basis for identifying redundant tracers. In order to be sure that tracer concentrations had not been affected by sand dune transportation and sedimentation processes, the nal sets of tracers for discriminating the sand dune sources were assessed using biplots of tracer concentrations in source and target sand dune samples for all three particle size fractions. The use of biplots as an additional test of tracer conservation is now recognised as a useful step in state-ofthe-art source ngerprinting decision trees 15 .
For selecting three nal composite ngerprints for each particle size fraction (i.e., the nine nal signatures) a two step statistical procedure comprising the Kruskal-Wallis H-test (KW-H) as step one and discriminant function analysis (DFA) as step two was used in conjunction with the tracers passing the conservative tests. This con rmed the composite ngerprints for discriminating sand dune sources. DFA statistics comprised Wilks' Lambdas, partial Lambdas, F to remove, the p-levels, the tolerance and 1tolerance values, and a full canonical analysis (eigenvalues, canonical correlation, Wilks' lambda, Chi-Square, p-level) for all functions, as well as classi cation and associated scatter plots. DFA was performed using STATISTICA V.8.0 47 .
The Modi ed MixSIR Bayesian model was used to apportion the aeolian sediment source contributions. This model has been mainly used in sediment source tracing studies in uvial systems [48][49][50][51] , but more recently in aeolian and storm dust source tracing 19 . The model was run for 10 6 iterations with the nine different nal composite ngerprints generated based on the different combinations of tracers. The maximum importance ratio was used to specify that the un-mixing model was effective in estimating the true posterior density distributions. Further details on this un-mixing model are provided in the aforementioned references.
To evaluate the MixSIAR predictions, known and modelled source contributions using each of the nine nal composite ngerprints were compared on the basis of the root mean squared error (RMSE), mean absolute error (MAE), and Willmott's index of agreement (d) 23,51,52 . Comparison of known and modelled source proportions using this virtual mixture approach is now seen as an informative step in source ngerprinting procedures 15 . The model predictions of source proportions were evaluated using 7 sets of virtual sand dune mixtures for each composite signature for the three particle size fractions.

Particle size fraction effects on elemental concentrations
The result of one-way ANOVA (Table 3) indicated that in the crop land source samples, concentrations of Al, Fe, Na, Si, and Ti were signi cantly affected by the absolute particle size fractions. Ti concentrations were higher in the < 125 µm fraction ( ner size), whereas concentrations of Al, Fe, Na, and Si were higher in the medium (125-250 µm) size fraction of the crop land samples. In the case of the samples collected to represent both the western and south-eastern alluvial fan sources, concentrations of Fe, Na, and Si differed signi cantly by size fraction. Additionally, in both of these sources, Fe, Na, and Si concentrations were higher in the medium (125-250 µm fraction) size fraction. Critically, however, all tracers were present across all size fractions. Table 3 The results of one-way ANOVA for the effect of particle size fractions on the elemental concentrations. Degree of freedom (df) = 2, SS, sum of square; MS, mean of square. *Critical pvalue = 0.05. 3) Final composite ngerprint using combing the geochemical elements and weathering indices: Using the DFA, Fe, WIP, PI, RI, AKN, SOC were selected in the nal signature (94.1% correct classi cation; Table SI 1&2).
The DFA scatter plots showing the source group discrimination visually for all nine nal composite ngerprints are presented in Fig. 4. Figure 5 shows biplots of the nal tracers. These biplots demonstrated that the tracers selected in the nal composite signatures had not been subjected to major transformation during redistribution from the sampled sources to the target sand dune sampling locations.
For the 125-250 µm fraction and using the geochemical elements composite signature, the mean relative contributions (5%-95% uncertainty ranges) from the crop lands, western alluvial fans, and south-eastern alluvial fans were estimated at 13.9% For the 250-425 µm fraction and using the geochemical elements composite signature, the relative contributions (5%-95% uncertainty ranges) from the crop lands, western alluvial fans, and south-eastern alluvial fans were predicted to be 5.6% (0.  (Fig. 6).

Discussion
Using the < 125 µm fraction, the geochemical elements and weathering indices only composite signatures suggested that the south-eastern alluvial fan is the main source of the sand dunes (mean contributions of 57.8% and 63.6%, respectively), whereas the signature combining geochemical elements and weathering indices suggested that the western alluvial fan is the main source (52.9%) of the sampled sand dune sediment. Averaging the mean contributions predicted by using all three composite ngerprints, suggested that the south-eastern alluvial fan is the main source (55.4%) and the western alluvial fan the second most important (overall average 41.7%) source. The RMSE between the Modi ed MixSIR apportionment estimates for the contributions from crop lands, western alluvial fans, and southeastern alluvial fans using the three nal composite ngerprints were 1.6%, 14.8%, and 13.6%, respectively. The corresponding MAE was 1.3%, 13.9%, and 12.5%, whereas the d values were 0.2%, 0.33%, and 0.38% (Table 4).
For the 125-250 µm fraction, the geochemical elements composite signature indicated that the western alluvial fan is the main source (76.4%) of the sampled sand dune sediment, whereas both composite signatures comprising either weathering indices only or a combination of weathering indices and geochemical elements suggested that the south-eastern alluvial fan is the main source of the sampled sand dune sediment (62.7%, and 75.9%, respectively). By averaging the contributions of all three composite signatures, the results showed that the south-eastern alluvial fan is the main source (49.1%) and the western alluvial fan (overall average 34.2%) the second most important source. The RMSE between the apportionment estimates for the contributions from crop lands, western alluvial fans, and south-eastern alluvial fans using the three nal composite ngerprints was 9.0%, 54.6%, and 50.4% respectively. The corresponding MAE was 8.0%, 50.5%, and 44.9%. The corresponding d values were 0.42%, 0.33%, and 0.22% (Table 4).
In the 250-425 µm fraction, the geochemical elements and the composite signatures with either geochemical elements only or elemental geochemistry combined with weathering indices suggested that the south-eastern alluvial fans are the main source of the sampled sand dunes (51.4% and 50%, respectively). The weathering indices only composite signature suggested that the western alluvial fan is the main source of the sampled sand dunes (41.8%). By averaging the contributions of all three composite signatures, the results indicated that the south-eastern alluvial fan is the main source (47.6%) and the western alluvial fan (overall average 39.4%) the second most important source. The RMSE between the predicted contributions for for the crop lands, western alluvial fans, and south-eastern alluvial fans sources using the three nal composite ngerprints was 8.4%, 6.3%, and 7.7%, respectively.
The MAE was estimated to 7.1%, 5.3%, and 6.7%, whereas the d values were 0.04%, 0.47%, and 0.22% (Table 4). estimates based on the nine composite tracers for the three particle size fractions ranged between 1.6% − 54.6% (with a mean value of 18.5%), 1.3% − 50.5% (with a mean value of 16.7%), and 0.04% − 0.5% (with a mean value of 0.3%), respectively (Table 4). Except in the 125-250 µm fraction for both the weathering indices only and a combination of the geochemical elements and weathering indices, the errors were acceptable for all other fractions. Similar ndings have been reported by studies applying sediment source ngerprinting in both uvial and aeolian systems 23,51,52 .
In order to explore the e ciency of weathering indices there is a need to investigate these type of tracers during all fundamental steps of sediment source ngerprinting comprising conservation tests, discrimination of potential sources and source apportionment using an un-mixing model. Although conventional geochemical tracers are the most widely used in source ngerprinting studies 17 , it remains informative to test other kinds of tracers. Regardless of the absolute particle size fraction in question, 25% of the conventional geochemical tracers were non-conservative, while 33% of the weathering indices were non-conservative. conventional geochemical tracers only, weathering indices only, and both combined for the 250-425 µm fraction were 85.3%, 88.2%, and 94.1%; respectively.The discriminatory power was therefore higher using the weathering indices across all size fractions. Except in the case of the nest fraction, source sample discrimination was highest when combining the conventional geochemical tracers and weathering indices. This underscores a key bene t of using weathering indices which provide a physico-chemical basis for helping to discriminate sediment sources 17 . Table 5 clearly shows that the estimated source contributions were sensitive to the kind of composite signature used, because all pairwise comparisons were signi cantly different (p ≤ 0.05). Sediment sourcing studies should therefore apply more than one nal composite signature to evaluate consistency in source predictions 15 . For the < 125 µm fraction (Table 6)    Overall, for the ner fraction, the RMSE, MAE, and d values from the virtual mixture tests using the different composite signatures showed that the signatures combining the geochemical elements and weathering indices returned the greatest accuracy. For the medium particle size fraction, the composite ngerprints based on the weathering indices alone returned the best accuracy. For the coarsest fraction, the conventional geochemical composite ngerprints returned the most accurate predictions of source contributions.
There are inevitably some limitations associated with the different steps of this source ngerprinting study. Determining the sand dune sources map is very di cult and complex because the main erosive agent for this study is wind erosion. In this study we only selected the sources that are surrounded by mountains to decrease the potential impacts of regional winds on patterns of sand mobilisation and redistribution. Accordingly, our source sampling only needed to consider the prevailing local wind directions based on the wind rose. Sampling design and sample collection are also important stages of source ngerprinting. In the case of environments dominated by wind erosion processes, the terrain is frequently at meaning that the conventional approach to targeting source sampling using visual appraisals of connectivity pathways along steeper terrain is not useful. In addition, seasonal changes in prevailing wind directions need to be given due consideration. The temporal and spatial scales of sampling for aeolian environments are therefore likely to differ from those associated with applications of source ngerprinting in environmental settings dominated by hydro-uvial processes. Depending on the tracers being used, it could be informative to assess the presence of seasonal variations in source tracer properties. Equally, in some settings, both uvial and aeolian processes will be interacting in the erosion and redistribution of sediments, including, for instance, in the case of abandoned stream channels. Here, it will be necessary to design sampling strategies encompassing the sources impacted by water and wind agents.

Conclusions
Our results demonstrate that the use the weathering indices in addition to conventional elemental geochemistry can be useful for investigations into sand dune sediment provenance. Assembling reliable information on the provenance of aeolian sediment deposits supports the targeting of source-control mitigation measures instead of sand dune stabilization interventions including chemical mulching which can suffer from unintended consequences. Future research should expand assessment of the application of weathering indices in combination with more conventional ngerprint properties to additional aeolian and storm sand systems since there is an ongoing need to assist management of the aeolian sediment problem globally.