The Quaternary loess and Miocene–Pliocene Red Clay sequences of the Chinese Loess Plateau (CLP) compose one of the most complete terrestrial eolian sediment archives of the late Cenozoic1,2,3. Understanding the spatial and temporal provenance variations of the CLP is key to accurately describing Asian monsoon evolution, aridification, and dust generation in eastern Asia over that period4,5,6. The detrital zircon U-Pb approach, in some instances, provides a tool for deconvolving multiple source areas in highly mixed provenance signals like loess7,8, and is often applied as a complementary tool to bulk geochemical analysis as a provenance indicator9. However, detrital zircon U-Pb geochronology has limitations in tracing provenance. For example, zircons are limited in carbonate and mafic/ultramafic rocks, and they only account for a very small portion of the bulk sediments10. Furthermore, zircon grain size and transport distance should be considered11,12. Despite these uncertainties, it is still useful provenance tool that has been demonstrated effective in many cases. This approach has been widely applied to the Quaternary loess sequences in recent years and fundamentally changed our understanding of the nature of the CLP4,9,13,14,15. Specifically, the proposed origin of the Quaternary CLP has evolved from it being largely composed of sediment deflated from proximal up-wind deserts to it being a combination of sediment from up-wind deserts and sediment sourced from the upper Yellow River5,7,15,16 (Fig. 1). This understanding has become more resolved through large observation (large-n) U-Pb detrital zircon provenance analysis. The added statistical robustness of large-n datasets from the law of large numbers and more widely distributed sampling have demonstrated spatial and temporal provenance differences across the CLP for the Quaternary strata16 —differences which could not be inferred from early U-Pb detrital zircon datasets with fewer observations from each sample7,13,17 (n ≈ 100).

Fig. 1: Map of detrital zircon sample sites in the Chinese Loess Plateau (CLP) and potential sources (modified from Ref. 16).
figure 1

The yellow star indicates the location of the Jiaxian site (JX). White dots indicate the locations of other loess-Red Clay sites. CN Chaona, BD Baode, DW Dongwan, LT Lantian, SL Shilou, WMU Western Mu Us Desert, EMU Eastern Mu Us Desert, XN Xining Basin.

Previous studies aimed at describing the provenance of the Quaternary CLP have demonstrated that the northeastern CLP—used in this study—has a distinct provenance from the central-western CLP4,16 (Fig. 1). Cretaceous sandstone overlying the North China Craton provided more sediments to the Quaternary loess-paleosol strata of the northeastern CLP than the central-western CLP. Zhang et al.16 divided the Quaternary CLP into western-central, eastern, and northeastern areas based on provenance differences. They concluded that the Quaternary strata of the central-western CLP received a higher proportion of detritus from the Huangshui River, a branch of the upper Yellow River, than other sites on the CLP16 (Fig. 1). In contrast, data from sites in the eastern CLP indicate an abundance of detritus supplied by the upper Yellow River and western Mu Us Desert whereas data for the northeastern CLP indicate proportionally more detritus sourced from the middle Yellow River and eastern Mu Us Desert. The formation of the CLP was eventually linked with the evolution of rivers5,8,15, like several loess provinces globally18,19,20,21.

The duration of time over which the late Miocene–Pliocene Red Clay sequences accumulated is roughly twice that of the Quaternary loess-paleosol sequences on the CLP, but much less zircon U-Pb geochronology data have been generated. Nie et al.22 presented the first detrital zircon U-Pb geochronology data in the international literature from the Red Clay sequence on the central-western CLP. Those data show temporal variations in provenance between 8 and 5.5 Ma, and between 4 and 3 Ma. Shang et al.23 presented zircon U-Pb geochronology data from three Red Clay sites (Baode, Lantian, Dongwan; Fig. 1) and compared the results with those from Chaona (CN). The data indicate that the northeastern CLP has a different provenance from the other sites, a pattern like the Quaternary period. Pan et al.24 presented zircon U-Pb geochronology data from the Shilou Red Clay sequences of the eastern CLP (Fig. 1). They found a clear provenance shift in the Shilou strata around 5.7 Ma which they attributed to the uplift of the bounding Lüliang Shan. Sun et al.25 analyzed zircon U-Pb geochronology data from the CLP, and they inferred that the late Miocene uplift of the northeastern Tibetan Plateau and the intracontinental orogenic belt produced large amounts of loose detritus, which served as dust sources for the late Miocene–Pliocene Red Clay sequences.

These prior studies improve our understanding of the provenance of the Red Clay sequences, but the number of zircon grains (n) for each stratigraphic interval was normally less than 200. As demonstrated by Pullen et al.26, inferring provenance information from a dataset of this sort has large uncertainties because of the statistical limitations of most of these datasets. To increase n, previous workers often aggregated results from adjacent strata or the entire Red Clay unit. However, that approach sacrifices temporal resolution, which might have prevented a robust understanding of temporal source changes, as we demonstrate here. The only large-n dataset from the late Miocene–Pliocene Red Clay strata is available for the CN site of the central-western CLP8 (Fig. 1). Recognizing the spatial and temporal variations in the Quaternary CLP strata, we present large-n detrital zircon U-Pb data for Miocene–Pliocene samples from Jiaxian (JX) in the northeastern CLP (Figs. 1 and 2). These samples have depositional ages similar to samples from the CN site, allowing a direct evaluation of spatial-temporal provenance evolution between the central and northeastern CLP. We recognize stepwise increased spatial provenance shift on the CLP from the late Miocene to the Pleistocene, which was mainly driven by temporal provenance variations on the northeastern CLP.

Fig. 2: Structure map of the study area (modified from ref. 44).
figure 2

The inferred uplift area near Lanzhou is indicated by the dashed rectangle.


To assess sediment sources of the late Miocene–Pliocene Red Clay at JX, we compared the JX zircon U-Pb age data with published detrital zircon U-Pb ages of potential source regions5,6,7,8,14,15,16,22,27,28,29,30 (see Supplementary Table 1 for detailed data sources). We used probability density plots (PDP), kernel density estimation plots (KDE), and histograms for comparison31. We also calculated the ratio of zircon ages within several diagnostic age ranges and plotted them along with depositional ages (Fig. 3, Supplementary Fig. 1 and Supplementary Data 1)—an approach that is beneficial in discriminating between potential source areas8. To aid comparison, a non-parametric multidimensional scaling (MDS) statistical approach was used32. In the MDS space, dissimilarities are plotted as distances with more dissimilar samples plotted further apart (Fig. 4 and Supplementary Fig. 2). Finally, we performed inverse Monte Carlo modeling experiments based on the Kuiper test to estimate the relative contribution of each potential source33. Ten thousand iterations of fitting were performed to recover a particular detrital age spectrum by varying the contributions from the potential sources and the best 1% fitting is presented (Fig. 5, Supplementary Figs. 3-6 and Supplementary Data 2).

Fig. 3: JX dust accumulation rate, grain size, magnetic susceptibility, and contribution variations of different zircon U-Pb age populations.
figure 3

a Dust accumulation rate (this study). b >20 μm percentage of all clastic noncarbonate grains (data from Ref. 56). c Magnetic susceptibility record (this study). d The number of zircon ages within the 200–300 Ma range over all zircon ages [n(200–300 Ma)/n(all)] (this study). e The number of zircon ages within the 400–500 Ma range over all zircon ages [n(400–500 Ma)/n(all)] (this study). f The number of zircon ages within the 800–1000 Ma range over all zircon ages [n(800–1000 Ma)/n(all)] (this study). g The number of zircon ages within the 1500–2750 Ma range over all zircon ages [n(1500–2750 Ma)/n(all)] (this study). The gray bars indicate stage divisions mentioned in the text.

Fig. 4: Nonparametric multi-dimensional scaling (MDS) plot of zircon U-Pb age data of the loess and Red Clay sequences on the CLP and comparison with potential sources.
figure 4

Blue, pink, and orange dots represent late Miocene Red Clay, Pliocene Red Clay, and loess samples from the JX site, respectively. The pink and blue stars represent all 5.6–4.2 Ma samples and all 7–6.1 Ma samples, respectively. Purple dots are the Red Clay samples from the CN site of different ages (numbers indicate ages in Ma). Black dots are potential source areas. Solid lines mark the closest neighbors and dashed lines link the second closest neighbors. JX Jiaxian, CN Chaona, HS, Huangshui River sediment, XN Xining Basin sediment, QB Qaidam Basin sediment, MYR Middle Yellow River sediment, EMU Eastern Mu Us Desert. Two red arrows show the stepwise increased spatial provenance contrast on the CLP.

Fig. 5: The relative contribution plots of potential sources to loess and Red Clay in JX.
figure 5

Red dots show the best-estimated contributions of different potential sources in the optimized simulation model. MYR Middle Yellow River sediment, EMU Eastern Mu Us Desert. HS Huangshui River sediment, XN Xining Basin sediment, QB Qaidam Basin sediment.

The probability density plots (PDP), kernel density estimation plots (KDE), and histogram plots show that the late Miocene–Pliocene Red Clay samples from the northeastern CLP are different from the Red Clay samples of the central CLP with similar depositional ages (Fig. 6 and Supplementary Data 1). In general, the Red Clay samples from the northeastern CLP are distinguishable by lower relative proportions of 800–1000 Ma ages and 400–500 Ma ages, and higher proportions of 200–300 Ma ages and 1500–2750 Ma ages than the central-western CLP. However, the provenance contrast between the two sites seems to become stepwise larger up-section (Fig. 6). It is clear from the MDS plot that the increased provenance contrast between the two sites is mainly due to large temporal provenance shifts in the JX site (because the CN site data are consistently more similar).

Fig. 6: A comparison of detrital zircon U-Pb ages from the northeastern and central-western CLP.
figure 6

Black and blue lines are normalized probability density function (PDP) and kernel density estimation plots (KDE), respectively, and open rectangles are age histograms. From left to right, the blue, purple, green and orange shadows indicate the ages in the range of 200–300 Ma, 400–500 Ma, 800–1000 Ma, and 1500–2750 Ma, respectively. JX-Jiaxian from the northeastern CLP; CN-Chaona from the central-western CLP.

The JX samples cluster in two areas: 7–6.1 Ma and 5.6–4.2 Ma samples (Fig. 4). The samples deposited at 7–6.1 Ma are statistically closer to CN samples than the samples deposited at 5.6–4.2 Ma. This indicates that the zircon-based provenance difference between the two sites is larger in the Pliocene. The provenance contrast between the two sites is larger for Pleistocene samples as indicated by the MDS location of the JX sample deposited at 2.4 Ma, which is closer to the middle Yellow River and eastern Mu Us Desert and further from the CN data (Fig. 4).

We note that the sample deposited at 7.6 Ma (JX-7.6) and 2.96 Ma (JX-2.96) plot close to the clusters of samples of 5.6–4.2 Ma, and 7–6.1 Ma, respectively. However, the proportion of zircons ages in the 200–300 Ma range for JX-7.6 and JX-2.96 are respectively lower and higher than that of samples in the 5.6–4.2 Ma and the 7–6.1 Ma range. The proportion of 200–300 Ma ages in JX-2.96 is similar to that for samples deposited between 5.6 Ma and 4.2 Ma. These differences are not notable in the MDS plot. By considering both the MDS plot and the major zircon age proportion ratios (Figs. 3 and 4), we divided temporal provenance variations in JX into five stages. From stage I ( > 7 Ma) to II (7–5.6 Ma), the clearest changes are the increased proportion of 200–300 Ma ages and decreased proportion of 1500–2750 Ma ages (Fig. 3). From stage II (7-5.6 Ma) to III (5.6–3.5 Ma), proportions of the 200–300 Ma and the 1500–2750 Ma ages increased, and the proportion of the 400–500 Ma ages decreased. The proportion of the 1500–2750 Ma ages in stage IV (3.5–2.7 Ma) is lower, back to about the same proportions as stage II (7–5.6 Ma); however, proportions of the other age groups are like stage III (5.6–3.5 Ma). The proportion of the 200–300 Ma ages in stage V ( < 2.7 Ma) is about the same as in stage III. However, proportions of the 400–500 Ma and the 800–1000 Ma ages are at the lowest observed level in stage V, and the proportion of the 1500–2750 Ma ages are at the highest level of the dataset (Fig. 3). These five-stages roughly align with dust accumulation rate (DAR) variations and with grain size variations to a lesser degree. The DAR and grain size are highly variable in stages I and II, are more consistent during stage III, and are higher and more variable for stages IV and V.

The JX-7–6.1 and the JX-5.6–4.2 samples do not have close source sample neighbors in MDS space suggesting the mixing of multiple sources. However, we note that samples JX-7–6.1 are closer to the Huangshui River and Xining basin sediments, which are primarily derived from the Qilian Shan16. The JX-5.6–4.2 samples plot closer to the middle Yellow River and eastern Mu Us Desert samples.

Relative contribution estimates based on the inverse Monte Carlo simulations support the above observations based on visual data comparison in MDS space (Fig. 5, Supplementary Figs. 3 and 4). For the JX-7.6 sample (stage I), the Huangshui+Xining+Qilian dust—combined because of the same derivation— and the middle Yellow River and eastern Mu Us Desert dust have similar contributions (close to 50%). The Huangshui+Xining+Qilian contributed ~62% of the detrital zircon crystals to the JX-7–6.1 sample (stage II) based on the results of the Monte Carlo simulations, followed by the middle Yellow River and eastern Mu Us Desert dust (~29%). By contrast, the middle Yellow River and eastern Mu Us Desert dust is estimated to contribute ~58% to the JX-5.6–4.2 samples (stage III), whereas the Huangshui+Xining+Qilian dust contributed only ~22%. The Huangshui+Xining+Qilian dust is estimated to contribute ~36% to the JX-2.96 sample (stage IV), whereas the middle Yellow River and eastern Mu Us Desert dust contributed ~33%. Sample JX-2.4 is unique in this suite of samples; the middle Yellow River and eastern Mu Us Desert dust contributed ~90% of the detrital zircon crystals and the Lüliang Shan provided the additional ~10% contribution (Fig. 5, Supplementary Figs. 3 and 4). The Gobi Desert (stony desert where gravels cover most of the surface, located in the north and northwest of the major deserts in the upwind of the CLP (Fig. 1), leaving only about 10% of the surface covered by other compositions34,35) provided <15% of the detrital zircon crystals in stages III and IV, and Liupan Shan provided another ~15% contribution in stage IV. The Qaidam basin was also considered a potential source area for the JX samples. However, the Monte Carlo modeling suggests that the Qaidam basin provided minimal detrital zircon to the JX samples (e.g., ≤10%) over any of the studied intervals. Although our unmixing modeling results are consistent with those based on visual comparison and proportion ratios, the quantitative ratios are better treated with caution because there are uncertainties associated with potential source regions. Adding or removing some potential sources may alter these ratios. For example, our simulations still yield reasonable fits to the measured zircon U-Pb age spectra even without the contribution of the Gobi Desert (Supplementary Figs. 5 and 6), suggesting that it is permissible, but not necessary for the Gobi Desert to have been a major source to arrive at the observed zircon age distributions.

Increased spatial provenance contrast of the CLP from the late Miocene to the Pleistocene

Loess accumulation is a function of dust supply, availability, and wind strength in which dust supply is a function of fluvial-eolian interaction, and dust availability is determined by climate and vegetation8,36. We explore the possible source regions for JX dust and surface processes responsible for provenance shifts from stage I to stage V below, within this framework.

The Pleistocene JX sample (stage V) is similar to the modern northeastern CLP samples37 having few 400–500 Ma zircon ages but a high proportion of 1500–2750 Ma ages. This was derived by erosion of the middle reaches Yellow River (and branch rivers) from the Cretaceous sandstone overlying the North China Craton16. Compared with the Pleistocene JX sample, 400–500 Ma ages compose ~18% of all zircon ages for the late Miocene JX samples (stages I and II), a percentage similar to that of CN from the central CLP (Fig. 3 and Supplementary Fig. 1). This similarity in the proportion of 400–500 Ma age zircons in CN and JX can be explained by common sediment sourcing from the Qilian Shan where detrital zircon crystals are known to have higher relative proportions of U-Pb ages in the 400–500 Ma range than other potential (dust) source regions16. The Qilian Shan experienced rapid rock and surface uplift during the late Miocene38,39,40,41,42. This would have provided ample unconsolidated detritus to be carried downwind to the CLP. In stages I and II, in addition to sediment eroded from the Qilian Shan, erosion of local basement rocks is interpreted because of the high proportion of 1500–2750 Ma ages15 (Figs. 4 and 5). In comparison with stage I, the proportion of the 400–500 Ma ages is slightly higher in stage II, accompanied by higher DAR. East Asian summer monsoon and winter monsoon had larger variations during stage II than I, as can be inferred from the magnetic susceptibility and grain size results (Fig. 3). Under these conditions, more sediments would have been made available from the Qilian Shan source (Figs. 5 and 7a, b), explaining the higher proportion of 400–500 Ma ages and higher DAR in JX. The middle Yellow River had probably not yet developed by the late Miocene: the zircon-based provenance indicators for the late Miocene JX samples would be more similar to, and possibility indistinguishable from, the Pleistocene (stage V) provenance signal if the middle Yellow River had been active at that time.

Fig. 7: Schematic diagram showing dust provenance changes in JX and corresponding surface processes.
figure 7

The basemap was produced by an open-source software GMT63. The regional topography was produced by the SRTM datasets64. The yellow star indicates the location of the Jiaxian site (JX). a Stage I: Similar contributions from erosion of local basement (Cretaceous sandstone overlying the North China Craton) and distal source (Qilian Shan of northeastern Tibetan Plateau). b Stage II: Increased Qilian Shan exhumation and global cooling increased distal contribution. c Stage III: Decreased Qilian Shan exhumation and climatic wetting decreased erosion and distal contribution. Following uplift in the Lanzhou area and possible rerouting of the upper Yellow River course44, more northeastern Tibetan Plateau-derived sediments were routed towards the Yinchuan-Hetao graben. However, climate warming and wetting promoted lacustrine environments to prevail45. Thus, northeastern Tibetan Plateau-derived sediment was temporarily stored in the paleolake(s) in the Yinchuan-Hetao graben, resulting in decreased distal contribution and dust accumulation rate. d Stage IV: Climatic dry-wet fluctuations amplified, resulting in increased erosion and release of trapped sediments in the Yinchuan-Hetao graben, increasing the distal contribution. e Stage V: Following the onset of intensive Northern Hemisphere glaciations, increased wet-dry contrast in comparison with the Pliocene increased river erosion, and released more unconsolidated sediments from the local basement. Climatic fluctuation amplitude increase promoted middle Yellow River incision, resulting in a connection between the middle and upper Yellow River. Increased erosion of the Lüliang Shan by frost wedging and other weathering processes enhanced during glacial intervals resulted in more production of unconsolidated sediment.

In stage III, the DAR decreased to the lowest level of the entire sequence, and it increased back to the level of stages I and II at stage IV (Fig. 3). The distal Qilian Shan source’s contribution decreased in proportion in stages III and IV, as indicated by the decreased proportion of the 400–500 Ma ages (Fig. 3). Thermochronology data suggest that rapid denudation of the Qilian Shan slowed or ceased by the Pliocene41,42. Therefore, less detritus was being produced during the early Pliocene, which reduced sediment supply (Figs. 5 and 7c). Decreased dust supply due to the tectonic factor may be strengthened by the intensification of Asian monsoon precipitation (as indicated by the high and less variable magnetic susceptibility) driven by high atmospheric CO2 concentration and the warm climate in the early Pliocene43. We reckon that under these conditions, vegetation in the Qilian Shan source areas was dense so that dust availability diminished, explaining the low DAR and decreased distal dust contribution to JX (Fig. 3). Lin et al.44 argued that the upper Yellow River was flowed along the course of the modern Weihe River in the late Miocene but switched to flow along its current river course due to uplift of the Lanzhou area (Fig. 2). If valid, this would have promoted movement of Qilian Shan-derived sediments to JX. However, the warm and wet early Pliocene climate may have promoted lake expansion in the grabens around the Ordos block (e.g. Yinchuan-Hetao graben; Fig. 2)45, decreasing sediment availability to the wind. We infer that decreased dust supply and availability in the distal source due to the above factors made the dust contribution from proximal sources proportionately more prominent in stage III.

Climatic dry-wet variations increased in amplitude in the late Pliocene (stage IV) at JX. This is inferred from the JX magnetic susceptibility record (Fig. 3). Increased climatic fluctuation amplitudes can promote dust accumulation in several ways. First, increased dry-wet fluctuation amplitudes tend to promote erosion and increase sediment supply. More river incision occurs during glacial-interglacial transitions46,47,48. Second, sediment availability tends to increase in conditions of increased dry-wet fluctuations49. It is reasonable to expect that the detritus sourced from the northeastern Tibetan Plateau and stored in the Yinchuan-Hetao graben were periodically exposed over dry intervals after ~3.5 Ma, increasing sediment availability for the wind to deflate (Figs. 5 and 7d). Furthermore, pollen records from the CLP suggest that tree percentage decreased from 30% to 10% and vegetation gradually changed to typical grassland and even to desert steppe after ~3.5 Ma50. This may also have increased sediment availability due to decreased vegetation stabilization. Increased wet-dry climatic fluctuations in stage IV may have also increased dust availability in the Gobi Desert51, resulting in the highest contribution of dust derived from the Gobi source to JX of all intervals (Fig. 5).

After 2.7 Ma, the 400–500 age mode decreases and the 1500–2750 Ma mode increases in the JX samples and DAR increased dramatically (Fig. 3). The mixture modeling simulations suggest mixing between the Cretaceous sandstone overlying the North China Craton (represented by the middle Yellow River and eastern Mu Us Desert) and Lüliang Shan explain the data, and that distal contribution is limited (Figs. 5 and 7e). We note that this corresponds to the amplification of Northern Hemisphere glaciations52 and enhanced glacial dust accumulation in the North Pacific Ocean53. Past research suggests that the East Asian summer monsoon was controlled by ice volume during the early Pleistocene54, as a result, the East Asian summer monsoon exhibited 41-kyr wet-band periodicity after 2.7 Ma and before 1 Ma55. This amplified the contrast between relatively wet and dry intervals. Compared with the Pliocene, this may have increased the productivity of fluvial erosion48, and may have liberated more weathered sediments from the Cretaceous strata overlying North China Craton (Fig. 7e). More pronounced wet-dry intervals may have also promoted the connection between the upper and middle Yellow River by increasing river downcutting48 (e.g., greater stream power from higher, albeit more variable, discharges).

In addition to increased dust availability from the middle Yellow River area, the mixture modeling performed here suggests increased detrital zircon contribution from the neighboring Lüliang Shan. Although the Lüliang Shan is too low in elevation to have been heavily glaciated after 2.7 Ma, frost weathering, more prevalent over glacial periods, may have increased erosion of the Lüliang Shan, increasing the 1500–2750 Ma zircon age component. Ages in the range of 1500–2750 Ma are the only notable ages in the Lüliang Shan27,28. Vandenberghe et al.56 defined the southern limit of permafrost during the Last Permafrost Maximum in northern China; they found that the Lüliang Shan was probably a permafrost zone during 19–23 ka.

In summary, our work demonstrates that understanding of the eolian system relies on a complete understanding of sediment supply, sediment availability, and transporting winds. Failure to pay attention to any factor would potentially result in erroneous interpretations. In contrast to suggestions that dust on the CLP has homogeneous composition according to element composition data, the detrital zircon data demonstrate that the Red Clay-loess sequence on the CLP experienced several shifts in provenance. These results support a genetic link between dust accumulation on the CLP and reorganization of the Yellow River system as well as tectonics, and global and regional climate over the late Miocene–Pleistocene.



We analyzed N = 13 late Miocene–Pliocene samples from the JX site of the northeastern CLP57 (Figs. 1, 2, Supplementary Fig. 7 and Supplementary Data 1). These samples have depositional ages of 2.96 Ma, 4.2 Ma, 4.8 Ma, 4.9 Ma, 5.1 Ma, 5.6 Ma, 6.1 Ma, 6.2 Ma, 6.7 Ma, 6.9 Ma, 6.95 Ma, 7 Ma, and 7.6 Ma. The age model of the JX site is based on the magnetostratigraphy of Qiang et al.57 with interpolation between Chron boundaries based on measured sediment accumulation rates (Supplementary Fig. 7).

Zircon U-Pb dating

The zircon U-Pb ages were measured at the Arizona LaserChron Center at the University of Arizona, using the U-Th-Pb large-n methods described in Sundell et al.58 and Pullen et al.59. The laser beam diameter was set to 20 μm. Zircon FC-160 and R3361 were used as reference materials. The U-Pb age used for interpretations and plotting (known as the ‘best age’) was from 206Pb/238U for <1000 Ma and from 206Pb/207Pb for >1000 Ma ages. For dates >600 Ma, we include the ages with <10% reverse discordance to <30% normal discordance. Following the convention for laser-ablation inductively-coupled-plasma mass-spectrometry U-Pb, a discordance filter was not applied to data <600 Ma thus this dataset is directly comparable to previously published datasets for the CLP. We note that we also tried applying a uniform discordance filter to all ages (i.e., not only those >600 Ma), and the main features of the dataset still hold (Supplementary Fig. 8). The number of ages reported for samples is in the range of n = 398–1112 (Fig. 6).

Quantitative estimation of the provenance contributions

Quantitative comparison is based on the Kolmogorov-Smirnov (KS) test D statistic and Kuiper test V statistic for cumulative distribution functions, and the Cross-correlation coefficient for finite mixture distributions. We primarily used Kuiper’s V value to perform the modeling because this value was used in the MDS statistics. By using the Kuiper’s V value, we can directly compare results between the two techniques. Other than this advantage, the Kuiper test is more sensitive than the KS-statistic to the tails of distributions62 and guarantees equal sensitivity across the entire range of ages in the samples33. We also provided the unmixing results based on the other two methods and associated uncertainty in the Supplementary Data 2, Supplementary Figs. 3, 9, 10, and all three methods show similar results.