Quantitative analysis of sediment source contributions to the different deposits in the Taklamakan Desert, western China, determined using discriminant function analysis

Abstract

Provenance analysis of aeolian sediments is key for understanding its origin, transportation and climatic significance. The ability to identify and quantify source contributions of aeolian sediments remains challenging. This study applies a discriminant function analysis (DFA) method to analyze the geochemistry of five different types of sediments (mobile dunes, vegetated dunes, fluvial sands, sandy loess and loess) in the Taklamakan desert and quantify the relative contribution of sediment sources. The geochemical compositions of sediments are influenced by the source rock compositions, chemical weathering and grain size. We analysed the effects of these three factors on the geochemical characteristics of sediments in detail and identified seven source tracers. The seven tracers of potential source areas and target sediments were compared by DFA, to obtain a series of discriminant functions which could identify the sources and their relative contributions. The results show that the aeolian sediments in the Taklamakan Desert are primarily sourced from the Kunlun Mountains and the Tianshan Mountains, with different grain sizes having different sources. The provenance of different types of sediments within the same grain size fractions were found to remain unchanged. The contribution of sediments sourced from the Kunlun Mountain and Tianshan Mountain to the > 63 μm fractions are 60% and 40% respectively, while the contribution to the < 63 μm fractions are 40% and 60% respectively. The findings are valuable for further understanding the source and transport mechanism of sediment in arid and semi-arid areas.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-025-32135-2.

Introduction

Global desert ecosystems are unique and vital components of the Earth’s environment. They are primarily located between 15°S and 60°N, covering regions of extreme aridity, arid, and semi-arid zones, which together account for approximately 22% of the Earth’s land surface¹. Deserts are also a significant source of dust worldwide². These ecosystems have a profound impact on the global climate, contributing to the global water cycle, influencing atmospheric circulation patterns, and playing a crucial role in the carbon cycle. Within this context, aeolian sediments serve as essential archives for reconstructing past and present environmental dynamics. Consequently, investigating the provenance of its aeolian sediments provides crucial insights into land formation processes and the environmental significance of these valuable sedimentary archives³.

In the Tarim Basin, the Kunlun and Tianshan Mountains are generally considered the primary source areas⁴. However, considerable controversy persists regarding the provenance of aeolian sediments, with significant discrepancies in quantitative estimates of contributions from various potential sources^5,6. Furthermore, several key questions remain unresolved—such as how sediment sources differ between grain-size fractions and across geomorphic units (e.g., mobile dunes, vegetated dunes, and loess). Current provenance research on Taklamakan Desert sediments predominantly focuses on single sediment types (such as mobile dunes)^4,7–9, while studies involving sediments from different geomorphological landscapes (e.g., mobile dunes versus vegetated dunes) remain limited. The differences in provenance and contribution rates of sediments from different land surface types remains unclear and these questions have yet to be effectively addressed. Sediment fingerprinting techniques, integrating geochemical tracers with statistical methods such as discriminant function analysis (DFA), offer promising approaches to resolve these uncertainties.

This study aims improve our understanding of regional sediment sources in the Taklamakan Desert by: (1) quantifying the relative contributions of major potential sources, including the Kunlun and Tianshan Mountains; and (2) comparing provenance characteristics among different sedimentary landscapes, such as loess, sandy loess, fluvial sands, vegetated dunes, and mobile dunes. The results are expected to provide new insights into sediment transport mechanisms and the evolution of aeolian landscapes in the Taklamakan Desert.

Study area

The Taklamakan Desert, located in the center of the Tarim Basin in the Xinjiang Autonomous Region (Fig. 1), covers an area of approximately 324,000 km², of which 85% is covered by mobile sand dunes¹⁰. The desert is encircled by high mountains which provide sediments to the region, including Tianshan to the north and the Kunlun Mountains to the south. Numerous rivers originating from these mountains transport detrital materials into the basin, where they are subsequently reworked and redistributed by wind.

The Kunlun and Tianshan Mountains, as major orogenic belts surrounding the Tarim Basin, exhibit significant differences in their fundamental geological composition. The Kunlun Mountains are characterized by ancient Proterozoic to Archean crystalline basements, consisting mainly of granites and metamorphic rocks such as gneisses, schists, and phyllites, formed in an active continental margin setting^12,13. In contrast, the Tianshan Mountains, forming part of the Central Asian Orogenic Belt, display a younger tectonic history with widespread Late Paleozoic magmatic rocks, particularly alkaline and A-type granites^14–16, associated with intraplate and passive margin environments following the closure of the Paleo-Asian Ocean^17,18. These fundamental differences in rock types, formation ages, and tectonic attributes between the two mountain systems control the geochemical characteristics of their weathering products, thereby influencing the provenance composition and sedimentary patterns of downstream depositional systems such as the Taklamakan Desert.

The paleohydrological network in the Taklamakan Desert has changed significantly since the Han Dynasty¹⁹. Around 1800 years ago (during the Han Dynasty), all the major tributaries from the Tianshan and Kunlun Mountains (nine drainage systems in total) flowed into the Tarim River, while today only four headstreams flow into the Tarim River. The surface processes in the Taklamakan Desert are mainly attributed to the wind and runoff, which play a significant role in the landscape evolution⁷.

The present-day landscape and sediment distribution are shaped by the interplay of wind and fluvial processes. Rivers originating from the surrounding high mountains transport detrital materials into the basin, where they are subsequently reworked and redistributed by aeolian action. The Keriya River, with a total length of approximately 650 km, is a key fluvial system in the southwestern part of the desert. It is fed by glacier meltwater and precipitation in the West Kunlun Mountains and flows northward into the desert hinterland^20,21. The catchment of the Keriya River lies within a zone of wind convergence, as indicated by sand drift potential analysis¹¹. Here, NW westerlies passing through the Pamir Plateau converge with NE winds passing through the Tianshan Mountains (Fig. 1). This dynamic wind regime, combined with fluvial inputs from the Keriya River, supports diverse dune morphologies—from sand sheets to compound megadunes—and contributes to loess sedimentation on the northern slopes of the Kunlun Mountains²². Due to its representative hydrology, aeolian dynamics, and diverse sediment types, the Keriya River catchment was selected as the focal area for this provenance study.

Overview of the Taklamakan Desert, including sample locations. The green dots represent the samples collected in this study. The red and yellow dots are the samples from references^4,7, respectively. The modern wind data are from the U.S. National Climatic Data Center (https://www.ncdc.noaa.gov/) from 1956 to 2018. The sand drift potential of 145 μm mean grain-size population was calculated following Fryberger and Dean (1979)¹¹. DP = drift potential; RDP = resultant drift potential; VU = vector unit; black lines showing winds capable of transporting sand from various directions and red arrows point downwind.

Materials

Specifically, a total of 44 surface samples were retrieved along the Keriya River from different geomorphological landscapes in 2018 (Fig. 2) to fingerprint the aeolian sediment sources. They include 16 samples from the mobile sand dunes, 19 from vegetated dunes, 3 from fluvial sands, 3 from loess, and 3 from sandy loess (Supplementary Table 1). To eliminate differences due to sediment sorting, the surface samples of dunes were collected only from dune ridges during wind-free periods. Each sample was taken from the top 0–5 cm depth at each site where there is not any sign of human interference. This paper compiles data from previous studies focused on the geochemical composition of the bedrock from sediment source areas including the Kunlun and Tianshan mountains. Compiled data is then compared with results from the 44 surface samples collected in this study to identify the source area of the sediment.

Sampling location map of sites within the Keriya River Basin. The base image is from ALOS_DEM data: https://www.eorc.jaxa.jp/ALOS/en/index_e.htm.

Sediment analysis

In order to eliminate the particle size effect²³, all 44 surface samples were first oven-dried, and then separated into > 63 μm and < 63 μm fractions by dry sieving with a Retsch AS 200 Control Vibratory Size Shaker. The 63 μm was chosen as the boundary because particles of < 63 μm fractions tend to be transported in suspension, which can be carried to higher atmospheric levels. The > 63 μm fractions are found to primarily move by saltation, which is the dominant process of sand transport by near-surface winds²⁴. Approximately 5 g of each sample (include different fractions) were finely ground in an agate mortar-grinder to a size of < 200 mesh before measuring major and trace elements in the Second Institute of Oceanography, Ministry of Natural Resources in Hangzhou, China. A fully automated sequential wavelength-dispersive X-ray fluorescence (XRF) spectrometer and inductively coupled plasma mass spectrometry (ICP-MS) were used to analyze the major and trace element measurements, respectively (Supplementary Table 2). For Samples No. 24 and No. 33, the dust fraction was insufficient to perform trace element analysis, and thus the data are unavailable. Analytical uncertainties are less than ± 1% for all major elements except for SiO₂ (up to ± 4%) and CaO (up to ± 2%). The analytical uncertainties (relative standard deviation) are less than ± 5% and ± 1% for trace elements and rare earth elements (REEs), respectively.

Tracer selection and discriminant function analysis

The purpose of using the multivariate discriminant function analysis (DFA) method is to search for the optimal combination of tracers capable of discriminating between potential sediment sources²⁵.The selection of appropriate geochemical tracers is fundamental for reliable sediment source fingerprinting²⁶. An ideal tracer should be conservative during transport and deposition, and be independent of grain size²⁶. In this study, seven geochemical parameters were selected based on these criteria and include: Y/ƩREE_N, Eu/Eu^*, Y/Tb_N, (La/Yb)_N, Y/La_N, TiO₂/Al₂O₃ and K/Rb.

Rare earth elements (REEs) and their ratios are particularly suitable for provenance studies due to their conservative behavior during sedimentary processes^27,28. While REEs can be subject to fractionation across grain sizes, specific ratios including Y/ƩREE_N, Eu/Eu*, Y/Tb_N, (La/Yb)_N, and Y/La_N are found to be effectively grain-size independent in northern Chinese deserts²⁸. The TiO₂/Al₂O₃ ratio was selected due to the high immobility of both Ti and Al during weathering and its consistent behavior across different silt-sized fractions, as established in loess studies²⁹. Given the minimal clay content in Taklamakan aeolian sediments³⁰, this ratio remains reliable for the < 63 μm fraction analyzed in this study. The K/Rb ratio serves as an effective tracer owing to the geochemical coherence of K and Rb during magmatic processes and their subsequent incorporation into K-feldspars, which are abundant in Taklamakan sediments^31,32.

Discriminant Function Analysis (DFA) was used to quantitatively identify sediment sources based on the seven tracers selected. The application of DFA relies on three key assumptions³³: tracer conservativeness, significant discriminatory power between sources, and the ability to define source groups mathematically. Tracer conservativeness was verified using a bracket test³⁴, ensuring that target sediment values fell within the range of potential sources. The discriminatory power of each tracer was statistically validated using the Kruskal-Wallis H test (p < 0.05)^33,35. DFA then identified the optimal linear combination of tracers that best differentiates the potential source areas (both two-source and four-source models). The quantitative apportion of sediment source contributions by the DFA method is based on the distances between the sediment and the center of each source area³⁶. The sediment is closer to the C1 source area, indicating a greater contribution from the C1 source to the sediment (Fig. 3). Liu et al.³⁷ used the sum of the horizontal and vertical distances when calculating the distances, while this study calculates the distance between sediments and the center of each source using the Pythagorean theorem. Source contributions were calculated as follows³⁶:

1

2

3

4

Fig. 3.

An example of inferring sediment source contributions using DFA and calculating distances. Here, D1 (used in Eq. (1)) is the distance from the center of sediment to the source 1.

where m is the number of source areas, n is the number of functions which have the capacity to separate sources, D_m is the distance of sediments from source m, ρ_i is the variance of function i, F_i (sediment) is the center position of the sediments, F_i (source_m) is the center position of the source m, W_m is the weight of source m, W is the total weight of all sources, P_m is the contribution of source m.

Results

Geochemical characteristics of different sediments

Generally, the chondrite-normalized REE patterns are an effective approach for comparing REE compositions and distinguishing provenances among different sediments³⁷. This study found no significant difference in REE distribution patterns among different types of sediments (Fig. 4a-b). The chondrite-normalized REE patterns of all target sediments were uniform with steep LREE, relatively flat HREE profiles and negative Eu anomalies. The REE contents of < 63 μm fraction was higher than that of the > 63 μm fraction in the same type of sediments (Fig. 4a-b). Bivariate diagrams of several tracers demonstrated that there is a very clear boundary between the > 63 μm and < 63 μm fractions (Fig. 4c-f). The > 63 μm fractions are characterized by lower Y/La_N values and higher Eu/Eu^*, (La/Yb)_N values, while the < 63 μm fractions of the same sediment samples exhibit higher Y/La_N values and lower Eu/Eu^*, (La/Yb)_N values (Fig. 4d-f). Although there is little variability in the TiO₂/Al₂O₃ values, there is still a boundary between the > 63 μm and < 63 μm fractions with the TiO₂/Al₂O₃ of > 63 μm fractions higher than that of < 63 μm fractions (Fig. 4c). Since the ratios, including the Y/ƩREE_N, Y/Tb_N, Y/La_N and TiO₂/Al₂O₃, are independent from grain sizes^28,38, the difference between the > 63 μm and < 63 μm fractions likely comes from different sources rather than from grain size sorting. By contrast, different types of sediments within the same grain size fractions show no significant difference in the bivariate diagram (Fig. 4), suggesting indicating the same provenance.

Fig. 4.

(a, b) The Chondrite-normalized REE patterns of sand (> 63 μm) and dust fractions (< 63 μm). The error bars are measurement errors of 2σ. (c-f) The bivariate diagrams of several tracers in sand and dust fractions. Among this, Eu/Eu^∗ =Eu_N/(Sm_N × Gd_N)^[1/238. Different types of sediments are represented by dots of five different colors, sand fractions are represented by solid dots, and dust fractions are represented by hollow dots.

Assessment of sedimentary processes

Prior to provenance analysis, the potential influences of sedimentary recycling, grain size sorting, and chemical weathering on sediment geochemistry were evaluated. The Th/Sc vs. Zr/Sc plot (Fig. 5a) show distinct patterns for different grain-size fractions. The > 63 μm fractions exhibit a synchronous covariation of Th/Sc and Zr/Sc, indicative of first-cycle sediments whose composition is primarily controlled by source rocks. In contrast, the < 63 μm fractions show a significant increase in Zr/Sc with minimal change in Th/Sc, suggesting these finer sediments have undergone degrees of sedimentary recycling. This interpretation is supported by the La/Th–Hf diagram (Fig. 5b), where most < 63 μm samples plot along a trend associated with an increasing component of recycled old sediment.

Fig. 5.

Plots indicating sedimentary recycling trend of surface sediments in this study. (a) Th/Sc versus Zr/Sc diagram³⁹. (b) La/Th versus Hf diagram modified from reference⁴⁰. UCC and PAAS values recommended by reference³⁷. Different size fractions of sediment are represented by the black, red and blue circles.

Results show that, the SiO₂/Al₂O₃ ratio has a positive relationship with grain size and can be employed to reflect grain size sorting^29,41,42. To assess grain size effects, the relationship between seven selected geochemical tracers (Y/ΣREE_N, Eu/Eu*, Y/Tb_N, (La/Yb)_N, Y/La_N, TiO₂/Al₂O₃, K/Rb) and grain size was examined. The results show no obvious correlations between any of these tracer ratios and grain size (Fig. 6).

Fig. 6.

Scatter plots testing the effect of grain size sorting on tracer ratios. Different types of surface sediment are represented by the five circles of five colors. In the legend, the grain size fractions are represented by solid circles (denoting > 63 μm) and hollow circles (denoting < 63 μm), respectively.

The intensity of chemical weathering was determined using the A-CN-K diagram and Chemical Index of Alteration (CIA) values⁴³. The weathering trend of all samples is sub-parallel to the A-CN line (Fig. 7a), with no trend towards the K apex, indicating an absence of significant potash metasomatism³⁹. Furthermore, the CIA values for all samples are below 58, consistent with a low degree of chemical weathering corresponding to the initial stage of Na and Ca removal.

Fig. 7.

The A-CN-K ternary diagram (a) and La-Th-Sc ternary diagram of surface sediments (b). Different size fractions of sediments are represented by the red, green and blue circles. CIA=[(Al₂O₃)/(Al₂O₃ + Na₂O + K₂O + CaO^*)]100, where CaO^* = Na₂O (in moles), following reference 60, represents the CaO incorporated only in the silicate minerals. T = tonalite; Gd = granodiorite; G = granite; Ka = kaolinite; AG = A-type granite; Sm = smectite; IL = illite; Pl = plagioclase; Mu = Muscovite; Ks = K-feldspar³⁸.

Types of main source rocks

The underlying geology of the source material and rocks tend to control the mineralogical and geochemical composition of sediments. The intersection of weathering trend line and the feldspar join (Pl-Ks) (Fig. 7a) lies between the point of granite (G) and granodiorite (Gd), indicating that granite and granodiorite are likely the main source rocks for most samples. Zircon is ubiquitous in virtually all felsic igneous rocks and is an immobile mineral that is usually resistant to chemical and mechanical destruction. Zr and Hf, often enriched in Zircon, are very similar crustal incompatible elements and can be used to distinguish different crustal processes. The values of Zr/Hf ratio for all samples is about 40, further indicating the granitoid source of samples⁴⁴.

The La-Th-Sc diagram can also be used to discriminate sediments from different sources³⁷. Mafic igneous rocks are enriched in Sc, while silicic rocks are enriched in La and Th. Therefore, on a ternary plot, sediments from the mafic sources will plot near the Sc pole, while sediments from the silicic sources will plot near the La and Th pole. As shown in Fig. 7b, our results fall close to the field of silicic rocks and sit between granites and granodiorite compositions. This result provides strong evidence that the source rocks of sediments in this study are a mix of granites, granitoids and granodiorite.

The classification results of DFA

Based on geochemical evidence indicating that the source rocks are primarily composed of granites, granitoids, and granodiorite (Fig. 7), we compiled the granite geochemical data from the Kunlun^45–54 and Tianshan Mountains^55–60, and used a bracket test to test the conservativeness of seven tracers (Table 1). The maximum and minimum tracer values in the source areas and target sediment samples was used to identify outliers. All tracers passed the bracket test and were then subjected to the KW-H test.

Table 1.

Minimum and maximum values of the seven ratios of geochemical tracers in the potential sources (Kunlun + Tianshan Mountains) and target sediments.

Tracer	Source		> 63 μm		Bracket test	< 63 μm		Bracket test
	Min	Max	Min	Max		Min	Max
Eu/Eu*	0.092	1.223	0.446	0.745	Passed	0.423	0.762	Passed
(La/Yb)N	1.056	34.866	7.654	27.758	Passed	7.153	11.099	Passed
Y/TbN	0.297	2.178	0.942	1.383	Passed	1.113	1.399	Passed
Y/N	0.072	1.577	0.146	0.508	Passed	0.354	0.507	Passed
Y/LaN	0.018	1.085	0.071	0.224	Passed	0.152	0.227	Passed
K/Rb	108.634	390	179.758	309.39	Passed	176.905	286.652	Passed
TiO2/Al2O3	0.002	1.197	0.035	0.108	Passed	0.034	0.155	Passed

The Kunlun and Tianshan Mountains can be subdivided into the East Kunlun, West Kunlun, East Tianshan, and West Tianshan Mountains, with a total of four geological sources identified as potential sediment source areas for the Taklamakan Desert. Granite geochemical data were collected from the West Kunlun^45–50, East Kunlun^51–54, West Tianshan^55,56 and East Tianshan Mountains^57–60, and the KW-H test was performed. The results show that the TiO₂/Al₂O₃ failed this test with a p-value exceeding 0.05 (Table 2) and was therefore excluded from further analysis. The remaining six tracers—Y/ΣREEN, Eu/Eu*, Y/TbN, (La/Yb)N, Y/LaN, and K/Rb—were used in the DFA.

Table 2.

The Kruskal-Wallis H-test outputs for tracers among different source areas.

Provenance group	Tracers	H-value (Chi-square)	P-value
Kunlun + Tianshan	K/Rb	21.725	< 0.001
	TiO2/Al2O3	0.88	0.348*
	Eu/Eu*	5.506	0.019
	(La/Yb)N	19.118	< 0.001
	Y/TbN	7.571	0.023
	Y/N	13.173	< 0.001
	Y/LaN	17.072	< 0.001
West Kunlun + West Tianshan +East Kunlun + East Tianshan	K/Rb	32.953	< 0.001
	TiO2/Al2O3	5.256	0.072*
	Eu/Eu*	37.600	< 0.001
	(La/Yb)N	91.570	< 0.001
	Y/TbN	12.235	0.007
	Y/N	89.691	< 0.001
	Y/LaN	94.992	< 0.001

* Not statistically significant at P > 0.05.

Prior to conducting DFA, the source areas were grouped into two scenarios: Scenario One included two broad source areas (the Kunlun Mountains and the Tianshan Mountains), while Scenario Two subdivided these into four areas (Eastern Kunlun, Western Kunlun, Eastern Tianshan, and Western Tianshan). DFA was then applied to the six selected tracer ratios from the source areas. The results revealed that for Scenario One (two sources), only one discriminant function was sufficient to distinguish between the Kunlun and Tianshan Mountains. In contrast, for Scenario Two (four sources), three functions were required to discriminate among the source areas (Table 3).

Table 3.

The unstandardised discriminant function coefficients from the DFA outputs.

Provenance group	Kunlun + Tianshan	West Kunlun + West Tianshan + East Kunlun + East Tianshan
Function	f	f1	f2	f3
K/Rb	0.009	0.006	-0.007	0.009
Eu/Eu*	-2.045	-0.995	2.659	3.579
(La/Yb)N	-0.008	0	0.009	0.003
Y/TbN	-1.708	-1.218	1.801	1.326
Y/N	14.915	7.307	-14.613	-2.565
Y/LaN	-14.175	1.423	15.828	3.119
Constant	-1.677	-3.156	-0.058	-5.631

The performance of the discriminant functions in classifying sediment source types in the Taklamakan Desert was evaluated using stepwise DFA on all tracers accepted by the KW-H test (Table 4). All functions passed the chi-square test (P < 0.001), indicating statistical significance. Wilks’ lambda was used to assess the discriminant power of each function, with values closer to zero indicating stronger discriminatory ability. For function f3, the Wilks’ lambda value was relatively high, suggesting a weaker capacity to differentiate between source areas.

Table 4.

The test results of the discriminant function derived using Stepwise DFA.

Provenance group	Function	Percent of variance	Wilks’lambda	Chi-square	P-value
two source areas	f	100	0.388	112.558	< 0.001
four source areas	f1	75	0.033	404.301	< 0.001
	f2	17.2	0.242	167.413	< 0.001
	f3	7.8	0.599	60.482	< 0.001

For Scenario One, when there are only two source areas, the formula for the discriminant function f that distinguishes between the Kunlun Mountains and the Tian Shan source areas is as follows (Table 3):

5

Based on the formula (5), and the tracer indicator values for each sample, the numerical value of f could be obtained. The f values represent the center positions of target sediments. According to the DFA result, the center positions of Kunlun and Tianshan sources in the one-dimensional space of function f are F_K=-0.859, F_T=1.804, respectively (Fig. 8a). Reclassification of source area samples using function f and these centroids yielded a correct classification rate of 90.3%, confirming the reliability of the discriminant function. The distance between each sediment sample and the two source centroids was calculated using the Pythagorean theorem (Formula 1), to determine which source contributes more to each sediment sample.

Fig. 8.

Classification plots of function F for the Taklamakan sediments with two source areas (a) and classification plots of function F1, F2 and F3 for the Taklamakan sediments with four source areas (b). Black circles and triangles represent the sediments collected from the Keriya river sediments in this study. F_K and F_T are the center positions of Kunlun Mountain and Tianshan Mountains respectively. C1, C2, C3 and C4 are the center positions of West Kunlun, West Tianshan, East Kunlun and East Tianshan Mountains, respectively.

To enhance the robustness of the results, the 44 sediment samples from the Keriya River collected in this study were combined with previously published data. Samples from references 4 and 7, along with the Keriya River samples, were correctly classified using function f (Fig. 8a). It should be noted that for grain-size standardization, reference 4 applied a 63 μm cutoff while reference 7 used a 75 μm threshold. To maintain analytical consistency, particles below both 63 μm and 75 μm thresholds were uniformly classified as the “dust fraction,” while those above these thresholds were designated as the “sand fraction.” This approach established a consistent three-tier grain-size framework comprising bulk sediment, dust fraction, and sand fraction. Following this standardization, the corresponding trace element ratios from all samples were incorporated into the established discriminant functions (DFA) for systematic calculation and comparison. The final results demonstrate that, both bulk samples and the sand fraction (> 63 μm or > 75 μm fraction) were closer to the Kunlun Mountains, indicating that the Kunlun Mountains are the primary source for these fractions. In contrast, the dust fraction (< 63 μm or < 75 μm fraction) was predominantly derived from the Tianshan Mountains (Supplementary Table 3). The overall consistency between our discriminant results and previously inferred provenance assignments supports the validity of the discriminant function as a reliable tool for provenance determination in the Taklamakan Desert with two source areas.

Furthermore, the DFA results also show that there is no obvious difference in the provenance of different types of sediments within the same grain sizes (Fig. 8a). The > 63 μm fractions are mainly transported by rivers, and the < 63 μm fractions are primarily transported by wind. Therefore, sediments with different grain sizes might have different provenances. Although there are differences in the landforms of mobile dunes, vegetable dunes, loess, sandy loess and fluvial sands, the main source areas within the same grain sizes remain the same, although the relative contribution of the source areas may be different.

For Scenario Two, when there are four sediment source areas, the formulas for the three discriminant functions are as follows (Table 3):

6

7

8

The value of f1, f2 and f3 of sediment determines its center position in the three-dimensional space of functions (f1, f2, f3). According to the DFA result, the center positions of the sources areas of West Kunlun, West Tianshan, East Kunlun and East Tianshan Mountains in the three-dimensional space of functions (f1, f2, f3) are (-1.963, 0.563, -0.084), (1.138, -2.25, -1.901), (5.263, 1.601, -0.192) and (1.014, -1.355, 1.132), respectively. A plot of the three functions (function 6–8) gives effective separation between West Kunlun and East Kunlun Mountains, but significant overlap occurs between the East Tianshan and West Tianshan categories (Fig. 8b). When the territorial plot was used to reclassify samples from the four predefined source areas, only 79.84% were assigned to the correct groups.

Similarly, application of the three functions to the 44 Keriya River samples and previously published data from the Taklamakan Desert showed that both the > 63 μm and < 63 μm fractions source were classified as closer to the West Kunlun Mountains (Fig. 8b), with no clear provenance differentiation between grain sizes. This result contradicts known sediment transport patterns and may be attributed to the high Wilks’ lambda value of function f3 (Table 4), which implies limited discriminative power and may result in unreliable provenance discrimination in the four-source scenario.

The contribution of sediment sources

Discriminant Function Analysis (DFA) was applied to quantify the contributions of potential sediment sources to bulk, > 63 μm, and < 63 μm fractions across five sediment types in the Taklamakan Desert. The calculations were based on the multivariate distance between sediment samples and source group centroids (formula 1–4).

Under the two-source scenario (Kunlun vs. Tianshan Mountains), the Kunlun Mountains were identified as the dominant source for the bulk and coarser (> 63 μm) fractions (~ 60%), while the Tianshan Mountains contributed more significantly to the finer (< 63 μm) fraction (~ 60%) (Fig. 9a-c). This grain-size-dependent provenance pattern is consistent with earlier studies⁴ suggesting different transport pathways for sand and dust fractions. When the provenance was refined to a four-source model (West Kunlun, East Kunlun, West Tianshan, East Tianshan), the discrimination accuracy decreased, as expected with an increasing number of sources. Re-aggregating the results from the four-source model back into the broader Kunlun and Tianshan categories yielded a consistent pattern: the Kunlun Mountains contributed approximately 55% to bulk and > 63 μm fractions, whereas the Tianshan Mountains contributed about 57% to the < 63 μm fraction (Fig. 9d-f). This reaffirms the robustness of the primary finding that the Tianshan Mountains are a major supplier of fine sediments to the desert.

Fig. 9.

The sediment source contributions to different grain size fractions. The Taklamakan sediments with two (a-c) or four source areas (d-f).

Provenance differences among the five sediment types (loess, sandy loess, vegetated dunes, mobile dunes, and fluvial sands) were primarily governed by grain size rather than geomorphic setting. For bulk and > 63 μm fractions, the Kunlun contribution was higher for mobile dunes and fluvial sands (66–77%) than for loess, sandy loess, and vegetated dunes (50–60%) (Fig. 10a). This is attributed to the coarser nature of fluvial and mobile dune sediments, which are predominantly supplied by Kunlun-sourced rivers⁴. In the four-source model (Fig. 10b), the bulk and coarse fractions of loess, sandy loess, and vegetated dunes received similar contributions from the West Kunlun (37–45%), West Tianshan (25–30%), and East Tianshan (25–30%), with a minor input from the East Kunlun (5–8%). In contrast, mobile dunes and fluvial sands were predominantly sourced from the West Kunlun (52–65%), with less contributions from the Tianshan ranges (15–20% each). For the fine (< 63 μm) fraction, provenance was uniform across all sediment types, with the East Kunlun contributing minimally (~ 8%) and the other three sources each contributing 28–34%. This further underscores that sediment provenance in the Taklamakan is strongly partitioned by grain size, with the Kunlun Mountains, particularly the West Kunlun, being the principal source of coarse fluvial and dune sediments.

Fig. 10.

The sediment source contributions to different types of sediments in different grain size fractions. The Taklamakan sediments with assumption of two source (a) and four source areas (b), respectively.

Discussion

Implications of sedimentary processes for provenance fingerprinting

The different sedimentary history between grain-size fractions provides important context for interpreting provenance results. As clearly demonstrated by the Th/Sc-Zr/Sc and La/Th-Hf results (Fig. 5), the < 63 μm and > 63 μm fractions have undergone distinctly different sedimentary recycling processes. The < 63 μm fractions have experienced some degree of sedimentary recycling, while the > 63 μm fractions are composed of first cycle sediments. This fundamental difference in transport history has resulted in contrasting geochemical characteristics between the two fractions, further supporting their potentially different source provenances.

Although the geochemical composition of clastic sediments can be affected by post-depositional alterations, our research confirms that neither grain size sorting nor chemical weathering has significantly altered the original geochemical signatures. The absence of clear correlations between the seven selected geochemical tracers (Y/ΣREE_N, Eu/Eu^*, Y/Tb_N, (La/Yb)_N, Y/La_N, K/Rb, TiO₂/Al₂O₃) and grain size indicates that sedimentary sorting has minimal influence on these parameters (Fig. 6). Similarly, the low Chemical Index of Alteration (CIA) values and A-CN-K weathering patterns (Fig. 7a) demonstrate that chemical weathering has not been strong enough to significantly modify key elemental ratio.

Therefore, the geochemical characteristics of Y/ΣREE_N, Eu/Eu^*, Y/Tb_N, (La/Yb)_N, Y/La_N, K/Rb and TiO₂/Al₂O₃ in the Taklamakan Desert sediments primarily reflect source rock properties. This validation is crucial for the application of Discriminant Function Analysis (DFA), as it ensures that quantitative provenance identification based on these tracers can accurately represent actual source contributions without significant bias introduced by sedimentary processes.

The transport processes of different grain size fractions

The grain size composition and transport pathway of sediments should not be ignored when we quantitatively calculate the sediment source contributions to different grain size fractions^61,62. Previous studies have shown that in the Keriya river sediments³⁰, the < 20 μm, 20–63 μm, 63–90 μm and > 90 μm fractions account for about 4%, 26%, 19% and 51%, respectively. Among the > 63 μm fractions, the 63–90 μm fractions accounts for 27% of the total sand-sized fractions and is primarily transported by short-distance suspension, whereas the > 90 μm fraction constitutes 73% and is mainly carried by saltation⁶³. Therefore, the sand fractions should be dominated by proximal sedimentary deposits.

The DFA results indicate that the Kunlun Mountains contribute 60% of the > 63 μm fraction, largely transported by fluvial processes, while the Tianshan Mountains contribute 40%, primarily delivered by wind (Figs. 9a and 10a). This distribution suggests that the Kunlun Mountains are likely the main source of the > 90 μm component, whereas the Tianshan Mountains appear to be a major supplier of the 63–90 μm component. However, the 63–90 μm fraction constitutes only 27% of the total > 63 μm material, yet the Tianshan contribution to the > 63 μm fraction reaches 40%. This discrepancy indicates that the Tianshan Mountains must also contribute significantly to the > 90 μm component. The Aksu River, with headwaters in the Tianshan Mountains, provides more than 70% of the total runoff for the modern Tarim River⁶⁴. Although the Aksu River flows into the Tarim River and towards the eastern margin, the zircon U-Pb age and heavy minerals composition⁶⁵ of the Aksu River sediments are similar to those of the Taklamakan sediments. This suggests that some Tianshan sediments carried by the Aksu River can be transported to the interior of the sand sea.

For the < 63 μm fraction, the DFA results reveal significant contributions from both mountain ranges, with the Kunlun Mountains accounting for 40% and the Tianshan Mountains for 60% (Figs. 9a and 10a). Given that the 20–63 μm component which is transported mainly by short-distance suspension comprises 87% of the < 63 μm fraction, and considering the modern near-surface winds in the region are predominantly north to south (Fig. 1), we propose that the Kunlun-derived materials are likely transported into this grain size fraction primarily through short-distance suspension processes. While the Kunlun-derived rivers supply < 63 μm particles to the Taklamakan Desert, the glacial, fluvial and aeolian abrasion of surrounding high-altitude mountains are also important dust-generating mechanisms processes⁶⁶.

The < 20 μm fractions are mainly found to be transported over long distances in suspension²⁴. The < 20 μm fraction has multiple sources³⁸, including dust from Central Asia, the adjoining arid/semi-arid areas to the east, and the abrasion of coarser particles from multiple regional sources. In addition, materials from the underlying bedrock in the Taklamakan Desert also substantially contribute to this fraction³⁸. The sediments are homogenized by the westerlies and local wind systems prior to reposition. The results from this study show that Tianshan Mountain contributes 40–45% to the bulk sediments, primarily in the 20–90 μm fraction, while Kunlun Mountain contributes 55–60% to the bulk sediments, mainly supplying the > 90 μm fraction, while the < 20 μm fraction is a multi-source component.

Multi-method validation of source contributions

The results of this study show that the aeolian sediments in the Taklamakan Desert exhibit distinct provenance characteristics across different grain-size fractions, with varying contributions from potential source regions. This grain-size-dependent provenance pattern aligns well with previous investigations^5,6, though methodological differences warrant careful consideration. For example, the Sr-Nd isotope geochemical study found that 25% to 70% of the detrital material found in the < 75 μm fraction of the Taklamakan Desert is from the South Tianshan and Central Tianshan⁵. Another study based on the Frequentist Model suggested that the Kunlun-Altun sediment is a dominant source (an average of 65%) of dune sand in the Taklamakan Desert, followed by the Pamirs-Kunlun source (33%) and the sediments of Tianshan areas (11%)⁶. Our quantitative results using Discriminant function analysis generally concur with these established patterns, though certain variations exist. These discrepancies likely stem from fundamental methodological differences: (1) variations in sampling strategies for source materials that may capture different spatial heterogeneities; (2) the application of distinct mathematical approaches for source apportionment, each with inherent assumptions and sensitivities. Particularly noteworthy is that the DFA approach proposed in this study, while confirming the general provenance framework, provides more refined discrimination between source contributions across specific grain-size fractions. The consistency across different methods reinforces the robustness of the identified provenance patterns, yet highlights the necessity of methodological transparency. We emphasize that when using sophisticated computational models—especially those with limited traceability—comprehensive validation is crucial to minimize uncertainties in quantitative source attribution and increase the reliability of sediment provenance interpretations.

Limitations of the methods used in this study

Identification of aeolian sediment provenance is a challenge due to methodological limitations, including method selection, sampling strategy, sample representativeness and grain-size effect^67,68. In this study the DFA method is applied to the sediments in the Taklamakan Desert to determine the relative contributions of each source. However, there are several limitations to consider when the DFA method is used for provenance tracing in the arid and semi-arid areas.

First, the potential source areas need to be fully understood before tracing the provenance. When using DFA for provenance discrimination, the number of source areas how the sources are categorized can significantly influence the outcomes. Therefore, a correct understanding of the potential source areas is essential for the reliability of the DFA methods. Currently, the East Tianshan and West Tianshan Mountains cannot be effectively distinguished, and other tracing indicators are needed in the future.

Second, the sampling strategy was necessarily constrained by logistical accessibility, resulting in a primary focus on the Keriya River catchment. Consequently, the geochemical signatures and resulting provenance apportionments are most representative of the mid-western Taklamakan Desert. Although we supplemented our dataset with published data to broaden the geographical context, the conclusions’ direct extrapolation to the entire desert should be made with caution. The findings are thus presented as a robust case study for this significant sector of the desert, and the methodological framework is proposed for testing in other catchments.

Third, through the provenance analysis of five sediment types in different landscapes, we found that the differences in provenances are mainly caused by grain size composition and transport pathway. However, the limited number of samples for the five sediment types, particularly for fluvial sands, loess, and sandy loess, may have constrained the quantitative provenance identification, potentially explaining the lack of distinct differences observed among various surface sediment types. In this study, only the dune crest samples were analyzed, but the grain size distribution of sediments across a dune (windward side, leeward side, crest) is different⁶³. Therefore, further work is needed to evaluate the sensitivity of source contributions to grain size variability in the Taklamakan Desert.

Finally, the target sediments in this study are collected in the same year, however, the geochemical and sedimentological properties of sediments will be changed with the level of in situ desertification in the Taklamakan Desert³⁰. Therefore, the results of this paper provide a snapshot of source contributions, and further work is needed to study the temporal dynamics of these contributions.

Conclusions

In this study, the DFA method was first used to quantitatively determine the contribution of the sediment sources to sediments of different grain sizes in the Taklamakan Desert. Geochemical analysis of bulk, > 63 μm, and < 63 μm fractions from five sediment types—mobile dunes, vegetated dunes, fluvial sands, loess, and sandy loess—revealed that sediment provenance varies with grain size rather than with surface sediment type.

When the Kunlun Mountains and Tianshan Mountains were treated as two broad source areas, DFA effectively discriminated between them and yielded robust results. In contrast, subdividing sources into eastern and western sections of each mountain range significantly reduced the discriminative power of the model. Quantitative contributions show that the Kunlun Mountains account for approximately 55–60% of the bulk sediments, compared to 40–45% from the Tianshan Mountains. The Tianshan Mountains primarily contribute the 20–90 μm fraction, while the Kunlun Mountains supply most of the > 90 μm material. The < 20 μm component appears to have multiple sources. For the > 63 μm fraction, about 60% originates from the Kunlun Mountains, mainly the West Kunlun. In contrast, the Tianshan Mountains contribute roughly 60% of the < 63 μm fraction, though their eastern and western sectors could not be clearly distinguished geochemically.

Among sediments of the same grain size, different surface sediments have similar sources, though relative contributions vary slightly. The Kunlun Mountains contribute more than 60% of the bulk and > 63 μm fractions in mobile dunes and fluvial sands—higher than in other sediment types. Contributions to the < 63 μm fraction are generally consistent across sources, except for the East Kunlun Mountains.

The application of the DFA method offers an intuitive method for characterizing sediment source contributions in the Keriya River catchment, and may provides new insights into the study of sediment provenances in arid and semi-arid areas.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Ye Zhou: Writing – original draft, review and editing, conceptualization, data curation. Bo Chen: Conceptualization, review and editing. Wei Huang: Conceptualization, data curation, review and editing. All authors reviewed and approved the manuscript before submission.

Data availability

Data is provided within the manuscript or supplementary information files.

Declarations

Competing interests

The authors declare no competing interests.