Evolution of the Upper Yellow River as Revealed by Changes in Heavy-Mineral and Geochemical (REE) Signatures of Fluvial Terraces (Lanzhou, China)

Despite decades of study, the factors that controlled the formation and evolution of the upper reaches of the Yellow River, including uplift of the northeastern Tibetan Plateau, Pliocene-Pleistocene climate change, and autogenetic processes are still poorly constrained. The stratigraphic record of such paleogeographic evolution is recorded in the sequence of nine terraces formed during progressive incision of the Yellow River in the last 1.7 Ma. This article investigates in detail for sediment provenance in terraces of the Lanzhou area, based on heavy-mineral and geochemical (REE) signatures. Two main provenance changes are identified, pointing each to a major paleogeographic reorganization coupled with expansion of the upper Yellow River catchment and enhanced sediment fluxes. The first change took place between the deposition of terrace T9 (formed around 1.7 Ma) and terrace T8 (formed around 1.5 Ma), when rapid fluvial incision point to tectonic control and active uplift of northeastern Tibetan Plateau. The second change took place between deposition of terrace T4 (formed around 0.86 Ma) and terrace T3 (formed around 0.14 Ma), during a period of low incision rates and notably enhanced sediment fluxes as a response to enhanced East Asian Summer Monsoon and consequently increased precipitations, pointing instead chiefly to climatic control.


Introduction
The Yellow River is the sixth longest river and carries the highest sediment load in the world [1,2]. The upper course of the Yellow River drains the central and northeastern Tibetan Plateau, and reconstructing its origin and early history is key to understanding the relationship between fluvial incision, climate variation, and basement uplift [3][4][5][6][7][8][9]. In the northeastern Tibetan Plateau, the Yellow River is deeply incised into a series of intermontane basins and tectonic ranges. The valley contains a series of fluvial terraces, which preserve the record of successive stages of plateau uplift and climatic

Yellow River Terraces in the Lanzhou Area
The Lanzhou Basin, located at the northeastern edge of the actively uplifting Tibetan Plateau, is part of the larger Longzhong Basin crossed by the upper Yellow River [32,33] (Figure 1) and including deposits of Late Cretaceous to Cenozoic age [34]. The Lanzhou basin, a graben-like structure 5-10 km wide and extending over a length of 40 km from Hutouya in the west-northwest to Sangyuanxia in the east-southeast, and it is divided into a western sub-basin delimited to the north by the Jinchengguan fault and an eastern sub-basin delimited by the Baitashan and Leitanhe fault in the west and by the Gaolan Mountains in the south [35] (Figure 1b).
Along the northern margin of the eastern Lanzhou sub-basin, the stepwise incision of the Yellow River has formed nine terraces (from the oldest T9 at the top to the youngest T1 at the base) (Figure 1a,b). Terraces T9 to T2 are strath terraces covered by loess-paleosol of different thickness and beveled across either Cretaceous-Cenozoic sandstone or Precambrian basement [10,11] (Figure 1c). The highest terrace T9, studied in the Yaogou section, lies 333 m above the modern river, is based on red Cenozoic sandstone, and consists of 10 m of fluvial gravel overlain by~22 m of fluvial sand and clay [36][37][38]. Terrace T8, studied in the Jiuzhoutai section, lies 100 m below T9, is also based on red Cenozoic sandstone, and consists of 2-3 m of fluvial gravel [11]. Terrace T7, studied in the Dunwashan section, lies 140 m above the modern river and consists of 6 m of fluvial gravel overlain by 12 m of overbank deposits. Terrace T6, exposed along the northern margin of the eastern Lanzhou basin and studied in the Dalanggou section, lies 127 m above the modern river and consists of 3-4 m of fluvial gravel overlain by 18 m of fluvial sand. Terrace T5, exposed in a narrow strip between the Xiaoshagou and Xiaogou sections, lies 108 m above the modern river consists of 3-4 m of fluvial gravel overlain bỹ 11 m of interchannel sand. Terrace T4, exposed almost along the entire length of the Lanzhou basin and studied in the Zaoshugou section, lies 99 m above the modern river and consists of 4-6 m of fluvial gravel overlain by 6-10 m of fluvial sand. The most widely exposed and preserved best terrace T3 lies 60 m above the modern river and comprises of 5-7 m of fluvial gravel overlain by 10 m of fluvial sand in the Baitashan section. The lowest strath terrace T2 lies 23 m above the modern river and consists of 5 m of fluvial gravel and~10 m of fluvial sand in the Luoguogou section. Terrace T1, consisting of fluvial gravel overlain by fluvial silt, is not studied here because it is poorly exposed and mostly occupied by buildings of Lanzhou city. Based on paleomagnetic and optically-stimulated-luminescence (OSL) dating of the base of loess-paleosol sequences, the ages of these terraces have been determined as 0.05 Ma (T2), 0.14 Ma (T3), 0.86 Ma (T4), 0.96 Ma (T5), 1.05 Ma (T6), 1.24 Ma (T7), 1.5 Ma (T8), and 1.7 Ma (T9) [10,11,38].

Sampling and Methods
Provenance analysis was carried out on eight sand samples collected in the eastern Lanzhou basin from alluvial terraces T9 to T2, and on one modern sand sample (T0) collected from an active bar on the Yellow River bed at Lanzhou (Figure 1a).
From each sample, heavy minerals were separated using the dense liquid tribromomethane (density 2.89 g/cm3). Heavy-mineral separates were analyzed by QEMSCAN at Colorado School of Mines (Golden, CO, USA), following the procedure described in detail in [39][40][41]. From 290 to 398 heavy-mineral grains were counted in each sample. For composite grains including more than one mineral, the QEMSCAN determines the surface area percentage of each (Figure 2), thus providing full information on all components. heavy-mineral grains were counted in each sample. For composite grains including more than one mineral, the QEMSCAN determines the surface area percentage of each (Figure 2), thus providing full information on all components.
The concentration of major and trace elements, including REE (rare earth elements) are widely considered as useful tracers of provenance [42][43][44][45][46]. Geochemical analyses were carried out on a 30-40 mg aliquot of each bulk sample crushed in an agate mortar, weighed into an airtight Teflon decomposer, and moistened with a little extra-pure water. Next, we added 1 mL HNO3 and 1 mL HF in Teflon containers left in ultrasonic bath for over 20 min, and the samples were steamed nearly dry on a heating plate at 150 °C After repeating the operation three times, the residues placed in a stainless-steel pot with 2 mL HNO3 and 3 mL extra-pure water were heated in oven at ∼150 °C for over 24 h. The concentration of chemical elements was determined by LA-ICP-MS using Agilent 7700X and Photon Machine Analyte 193nm at Key Laboratory of Mineral Resources in Western China, Lanzhou University (Lanzhou, China).
Grain-size analyses on bulk samples were carried out by Laser Diffraction Particle Size Analyzer at the Key Laboratory of Western China's Environment Systems (Ministry of Education) of Lanzhou University. All samples were air-dried and analyzed by a Mastersizer 2000 produced by Malvern Instruments Ltd. (Malvern, UK) with Hydro2000G automatic sampler. The measuring range is from 0.02 to 2000 μm, and the median particle size D50 has an accuracy of ±1%.  The concentration of major and trace elements, including REE (rare earth elements) are widely considered as useful tracers of provenance [42][43][44][45][46]. Geochemical analyses were carried out on a 30-40 mg aliquot of each bulk sample crushed in an agate mortar, weighed into an airtight Teflon decomposer, and moistened with a little extra-pure water. Next, we added 1 mL HNO 3 and 1 mL HF in Teflon containers left in ultrasonic bath for over 20 min, and the samples were steamed nearly dry on a heating plate at 150 • C After repeating the operation three times, the residues placed in a stainless-steel pot with 2 mL HNO 3 and 3 mL extra-pure water were heated in oven at~150 • C for over 24 h. The concentration of chemical elements was determined by LA-ICP-MS using Agilent 7700X and Photon Machine Analyte 193nm at Key Laboratory of Mineral Resources in Western China, Lanzhou University (Lanzhou, China).
Grain-size analyses on bulk samples were carried out by Laser Diffraction Particle Size Analyzer at the Key Laboratory of Western China's Environment Systems (Ministry of Education) of Lanzhou University. All samples were air-dried and analyzed by a Mastersizer 2000 produced by Malvern Instruments Ltd. (Malvern, UK) with Hydro2000G automatic sampler. The measuring range is from 0.02 to 2000 µm, and the median particle size D50 has an accuracy of ±1%.

Results
Different heavy-mineral assemblages in superposed Yellow River terraces of the Lanzhou area, together with varying REE patterns, indicate major provenance changes through time.

Heavy-Mineral Assemblages
Transparent heavy-mineral suites in modern Yellow River sand and in Pleistocene to Holocene terraces are invariably dominated by amphibole (46-58%) and epidote (22-29%) (Figure 3a, Table 1). The abundance of amphibole indicates prominent contributions from basement rocks [24,47], with the amphibole series forming in a wide range of temperature and pressure conditions in igneous and metamorphic rocks [48]. A ratio of amphibole versus total heavy minerals has been used to reflect past changes in the origin of loess from the central Chinese Loess Plateau [49]. Because more than 73% of the heavy-mineral fraction is composed of amphibole plus epidote in Yellow River terraces, we here use the ratio between these two dominant minerals as a tracer of provenance change. Amphibole accounts for less than 50% of the assemblage in terraces T8 and T3-T2 as well as in T0, whereas epidote displays peak abundance in terraces T3 and T2 where the amphibole/epidote ratio is close to 1 (1.01-1.08 versus 1.5-2.7 in older terraces).

Results
Different heavy-mineral assemblages in superposed Yellow River terraces of the Lanzhou area, together with varying REE patterns, indicate major provenance changes through time.

Heavy-Mineral Assemblages
Transparent heavy-mineral suites in modern Yellow River sand and in Pleistocene to Holocene terraces are invariably dominated by amphibole (46-58%) and epidote (22-29%) (Figure 3a, Table 1). The abundance of amphibole indicates prominent contributions from basement rocks [24,47], with the amphibole series forming in a wide range of temperature and pressure conditions in igneous and metamorphic rocks [48]. A ratio of amphibole versus total heavy minerals has been used to reflect past changes in the origin of loess from the central Chinese Loess Plateau [49]. Because more than 73% of the heavy-mineral fraction is composed of amphibole plus epidote in Yellow River terraces, we here use the ratio between these two dominant minerals as a tracer of provenance change. Amphibole accounts for less than 50% of the assemblage in terraces T8 and T3-T2 as well as in T0, whereas epidote displays peak abundance in terraces T3 and T2 where the amphibole/epidote ratio is close to 1 (1.01-1.08 versus 1.5-2.7 in older terraces). Correlation between formation age of Yellow River terraces and height above the river (mod. from [10,38]). The less steep slope between T7 and T3 indicates lower rates of rock uplift and river incision at that time. (c) Variations of the East Asian summer monsoon during the past 1.8 Ma, based on the stacked magnetic susceptibility (χ) records on the Chinese Loess Plateau (CLP) [50]. (d) Variations of global ice volume during the past 1.8 Ma, based on the benthic oxygen-isotope record [51]. Mineral color-code as in Figure 2. Comparison between heavy-mineral spectra from Yellow River terraces and the paleoclimatic record. (a) Heavy-mineral data from Yellow River terraces in the Lanzhou area. (b) Correlation between formation age of Yellow River terraces and height above the river (mod. from [10,38]). The less steep slope between T7 and T3 indicates lower rates of rock uplift and river incision at that time. (c) Variations of the East Asian summer monsoon during the past 1.8 Ma, based on the stacked magnetic susceptibility (χ) records on the Chinese Loess Plateau (CLP) [50]. (d) Variations of global ice volume during the past 1.8 Ma, based on the benthic oxygen-isotope record [51]. Mineral color-code as in Figure 2.

Rare Earth Elements
The total REE (∑REE) content in Yellow River terraces varies widely from 46 to 116 ppm. In chondrite-normalized REE diagrams ( Figure 4, Table 2), all samples show enrichment in light rare  REE curves are all similar, with minimum values for the oldest terrace T9 and maximum values for terrace T8. REE contents are relatively high for both terraces T3 and T2 (Figure 4). For the interpretation of REE concentration; however, it must be kept in mind that REE are preferentially hosted in ultra-dense minerals such as monazite, and their abundance in sediments is consequently strongly affected by hydraulic-sorting processes [53].

Discussion
The mineralogical and geochemical fingerprints of siliciclastic sediments are mainly controlled by the lithology of source rocks, with superposed effects associated with climatic conditions in the catchment and hydraulic sorting during erosion, transport, and deposition [9,54].
Most of the river sediments have similar and uniform REE patterns displaying enrichment in light REE [55,56] which is inherited from source rocks, confirming that the sediments are not anthropogenically polluted [57,58]. Heavy-mineral assemblages in Yellow River terraces and geochemical data suggest that changes in sediment composition took place between terraces T9 and T8, and between terraces T4 and T3. The provenance changes cannot be discounted as a grain-size effect because T3 and T2 have similar composition despite their markedly different grain-size distributions ( Figure 5). This is confirmed by the principal-component diagram shown in Figure 6, where terraces T3 and T2 plot on the right side of the PC1 component, accounting for ~82% of data  REE curves are all similar, with minimum values for the oldest terrace T9 and maximum values for terrace T8. REE contents are relatively high for both terraces T3 and T2 (Figure 4). For the interpretation of REE concentration; however, it must be kept in mind that REE are preferentially hosted in ultra-dense minerals such as monazite, and their abundance in sediments is consequently strongly affected by hydraulic-sorting processes [53].

Discussion
The mineralogical and geochemical fingerprints of siliciclastic sediments are mainly controlled by the lithology of source rocks, with superposed effects associated with climatic conditions in the catchment and hydraulic sorting during erosion, transport, and deposition [9,54].
Most of the river sediments have similar and uniform REE patterns displaying enrichment in light REE [55,56] which is inherited from source rocks, confirming that the sediments are not anthropogenically polluted [57,58]. Heavy-mineral assemblages in Yellow River terraces and geochemical data suggest that changes in sediment composition took place between terraces T9 and T8, and between terraces T4 and T3. The provenance changes cannot be discounted as a grain-size effect because T3 and T2 have similar composition despite their markedly different grain-size distributions ( Figure 5). This is confirmed by the principal-component diagram shown in Figure 6, where terraces T3 and T2 plot on the right side of the PC1 component, accounting for~82% of data variance. Principal-component analysis also suggest that terrace T8 and modern fluvial sand T0 are distinguished from other samples on PC2, accounting for~15% of data variance. This reflects at least in part the slightly lower content in hornblende relative to samples T9 and T7-T4. Some studies in Europe have shown that higher uplift rates may decrease the preservation of climate-controlled fluvial terraces, whereas moderate to low uplift rates will enable their preservation [59][60][61]. Accurate field investigations have failed to find any Yellow River terrace in the Lanzhou area formed between 0.86 Ma and 0.14 Ma [10]. However, it is noteworthy that this provenance change took place between 0.86 and 0.14 Ma, a long period of decreased tectonic activity when river terraces were not formed-or were formed but were not preserved-in the Lanzhou basin [10] (Figure 3b). For this reason, we could not more precisely constrain the age of such a compositional change. variance. Principal-component analysis also suggest that terrace T8 and modern fluvial sand T0 are distinguished from other samples on PC2, accounting for ~15% of data variance. This reflects at least in part the slightly lower content in hornblende relative to samples T9 and T7-T4. Some studies in Europe have shown that higher uplift rates may decrease the preservation of climate-controlled fluvial terraces, whereas moderate to low uplift rates will enable their preservation [59][60][61]. Accurate field investigations have failed to find any Yellow River terrace in the Lanzhou area formed between 0.86 Ma and 0.14 Ma [10]. However, it is noteworthy that this provenance change took place between 0.86 and 0.14 Ma, a long period of decreased tectonic activity when river terraces were not formedor were formed but were not preserved-in the Lanzhou basin [10] (Figure 3b). For this reason, we could not more precisely constrain the age of such a compositional change.   variance. Principal-component analysis also suggest that terrace T8 and modern fluvial sand T0 are distinguished from other samples on PC2, accounting for ~15% of data variance. This reflects at least in part the slightly lower content in hornblende relative to samples T9 and T7-T4. Some studies in Europe have shown that higher uplift rates may decrease the preservation of climate-controlled fluvial terraces, whereas moderate to low uplift rates will enable their preservation [59][60][61]. Accurate field investigations have failed to find any Yellow River terrace in the Lanzhou area formed between 0.86 Ma and 0.14 Ma [10]. However, it is noteworthy that this provenance change took place between 0.86 and 0.14 Ma, a long period of decreased tectonic activity when river terraces were not formedor were formed but were not preserved-in the Lanzhou basin [10] (Figure 3b). For this reason, we could not more precisely constrain the age of such a compositional change.    (terraces T8 and T9), the East Asian summer monsoon was relatively stable, as documented by magnetic-susceptibility data on Chinese loess [50] (Figure 3c), and there is no evidence of large variations in ice volume [51] (Figure 3d). This speaks against climate change as a main cause for the change in heavy-mineral assemblages and geochemical signatures observed between terraces T9 and T8. In this time window-i.e., between 1.7 and 1.5 Ma-the Yellow River rapidly incised its talweg by as much as 100 m. If this was not triggered by climatic change, then active tectonic uplift remains as the most plausible cause [62]. Between 1.24 Ma (terrace T7) and 0.86 Ma (terrace T4), the Yellow River course was shorter, its drainage basin smaller, and its sediment load notably less than today [12,13,63]. The timing of major paleogeographic change took place during, and in the final part of the so-called "mid-Pleistocene climatic revolution" (or early-middle Pleistocene transition; 1.4-0.424 Ma), characterized by a progressive increase in the amplitude of climatic oscillations [64,65]. At that time, the East Asian summer monsoon intensified and the amplitude of ice-volume variation between glacial and interglacial stages increased, as documented by magnetic-susceptibility data on Chinese loess [50,51] (Figure 3c,d). Increased glacial erosion may have enhanced sediment fluxes [66][67][68] at that time, when a phase of decreased rock uplift and deformation in the Lanzhou area is suggested by the decrease in incision rates as highlighted in Figure 3b. Tectonic control is unlikely for this time period [62], during which heavy-mineral and geochemical signatures have remained remarkably constant.
After 0.86 Ma, the change in heavy-mineral assemblages and geochemical fingerprints observed between terraces T4 (0.86 Ma) and T3 (0.14 Ma) may be explained with the stepwise addition of large areas to the upper reaches of the Yellow River by headward erosion [12,63]. This scenario is consistent with the increase in precipitations as a consequence of enhanced East Asian Summer Monsoon [50] and with the increased amplitude in ice-volume variation [51]. Between~0.5 and 0.1 Ma, Yellow River sediment fluxes markedly increased, while the Gonghe basin, the Tongde basin, and the Jungong basin were successively incorporated in the Yellow River drainage system [4,12,13].

Conclusions
The study of heavy-mineral assemblages and REE patterns characterizing Yellow River terraces in the Lanzhou area reveals two notable provenance changes, from~1.7 Ma (terrace T9) to~1.5 Ma (terrace T8) and from~0.86 Ma (terrace T4) to~0.14 Ma (terrace T3). Two paleogeographic reorganizations took place during those time intervals, both associated with phases of expansion of the upper Yellow River catchment and consequently increased sediment fluxes.
The height difference of~100 m between terraces T9 and T8 indicates rapid incision during the early Pleistocene (early Calabrian), suggesting that the first shift was triggered tectonically by a phase of rapid uplift of the northeastern Tibetan plateau. The second reorganization took place during the final part of the middle Pleistocene transition (late Calabrian), when the East Asian summer monsoon intensified and the amplitude of ice-volume variation between glacial and interglacial stages increased. River incision was much slower at this time, indicating that this second, major reorganization was chiefly climatically controlled.