In February 2018 a sediment core, 253 cm in length (JL18-02-A), was collected from the southwestern area of Jili Lake using a Piston corer. The core was frozen and transported to Lanzhou University, sampled at a 1-cm intervals, and freeze-dried. The lithology of the c. 120 cm of the core can be divided into three intervals: 0–6 cm - cyan mud with loose sediments; 6–34 cm - the color of sediments is brown; 34–123 cm - pale silty clay with relatively compact sediments.

We focused on the upper 120 cm of the core because there was a hiatus in the sediment below this depth based on 7 radiocarbon dates (Lang, 2020). For the age-depth model of the upper 120 cm, three bulk organic samples from the core were processed for accelerator mass spectrometry (AMS) radiocarbon dating at the Beta Analytic Radiocarbon Dating Laboratory in Miami, Florida, United States (Table 1) (Lang, 2020). The ¹⁴C specific activity values of the lake-water DIC sample and the samples from the upper most layers of cores largely approximate to the pMC (percent modern carbon) values of the modern atmosphere (105.4 ± 1.0 pMC in 2007; Fellner and Rechberger, 2009), indicating that the CO₂ exchange between the lake and atmosphere tends to be balanced, reflecting that the “reservoir effect” of the lake water is low. The age-depth model for Jili Lake (Figure 2) was calculated using “Clam” version 2.2 (Blaauw, 2010) and the IntCal13 calibration curve (Reimer et al., 2013). The core depth-chronology has a linear relationship and it provides a record of sediment deposition over the last 2400 years.

A total of 96 samples were taken from the core for pollen analysis. Pollen grains were extracted from 1–3 g of dried sample, and preparation used HCl (10%) and HF (40%) to remove carbonates and silicates (Fægri and Iversen, 1989). Samples were sieved through a 10-μm mesh to remove small clay-sized particles, and clean samples were mounted in glycerin on glass slides. Pollen grains and charcoal particles were identified and counted using a Nikon ECLIPSE 80i optical microscope at ×400 magnification. More than 500 pollen grains and 300 charcoal particles were counted for each sample, and charcoal particles were grouped by long axis length: >100, 50–100, and <50 μm. To calculate pollen and charcoal concentration, one Lycopodium tablet (27,637 grains) was added to each sample prior to chemical pre-treatment (Maher, 1981). The percentages of pollen were calculated based on the sum of all counted terrestrial pollen grains, and the aquatic pollen percentages were calculated based on the sum of all counted pollen grains. The pollen diagrams were plotted using Tilia software (Grimm, 2011), and the pollen zones were divided by stratigraphically constrained cluster analysis using CONISS (Grimm, 1987).

Based on the investigations of modern surface pollen in northern Xinjiang, it has shown that Artemisia and Amaranthaceae (old name is Chenopodiaceae) (APG, 1998), are dominant species in desert-steppe and desert areas (Yu et al., 1998; Luo et al., 2009; Li et al., 2017). A lot of modern pollen investigations have shown that the Artemisia and Chenopodiaceae ratio (A/C ratio) is a valid indicator/proxy of humidity in desert and desert-steppe areas (e.g., El-Moslimany, 1990; Sun et al., 1994; Huang et al., 2009; Zhao et al., 2009; Li et al., 2010; Zhao et al., 2012; Zhang D. et al., 2018). As Ephedra is also one of the main taxa in desert areas like Amaranthaceae (Huang et al., 2009; Huang et al., 2018), here we use the sum of Amaranthaceae and Ephedra (Am + E) to indicate a relatively dry condition, and use the ratio of Artemisia (Ar) and (Am + E) as a new indicator of climatic humidity like A/C ratio.

The signal of regional human activity can be indicated by some specific pollen types and charcoal abundance. For example, Poaceae pollen grains among the 35–50 μm size range were considered to be cereal-type Poaceae (mostly might be wheat pollen), which can be used as an indicator of agricultural activity (Li et al., 2008; Li et al., 2012). Charcoal is a particularly useful proxy for recording the disturbance of vegetation by humans, in which macro-charcoal (>100 μm) could indicate changes in local fire occurrence in the past (Whitlock and Larsen, 2001; Li et al., 2008).

The major pollen types included Asteraceae (0.2–3.6%, mean 2.3%), Ephedra (0.6–3.09%, mean 2%), Polygonaceae (0.2–3.6%, mean 2.74%), Rosaceae (1.0–5.8%, mean 2.1%), and Typha (1.0–2.8%, mean 1.60%). The sum of tree pollen types is c. 5.1% and including Pinus, Picea, Betula and Salix. The abundance of Poaceae (0.3–2.5%, mean 1.4%) and Cyperaceae (0–0.9%, mean 0.3%) was low, and the Ar/(Am + E) ratio (0.2–0.7, mean 0.4) had the lowest value in the whole core sequence. Overall, this zone had the lowest pollen concentration (c. 18873 grains/g). The pollen assemblages in this zone, can be further divided into two sub-zones. Sub-zone Ⅰ-1 (119–106 cm, c. 380–70 BCE) can be readily distinguished from sub-zone Ⅰ-2 (106–87 cm, c. 70 BCE-400 CE) by the decreasing percentage of Poaceae pollen.

This zone was typified by an increase in Poaceae (0.5–13.2%, mean 6.5%) and Cyperaceae (0.2–1.5%, mean 0.8%) at the expense of Artemisia (mean 34%), Asteraceae (0.5–2.6%, mean 1.4%), and Rosaceae (0–2.4%, mean 0.5%) whose percentages decreased. This zone also sees the first appearance of cereal-type Poaceae (0–0.6%, mean 0.3%).

This zone can be further divided into two sub-zones. Sub-zone Ⅱ-2 (52–40 cm, c. 1160–1350 CE) can be readily distinguished from Sub-zone Ⅱ-1 (86–52 cm, c. 400–1160 CE) by the following features: further increases in Artemisia (32.6–41.7%, mean 36.8%), Poaceae (5.6–13.2%, mean 9%) percentages and Ar/(Am + E) ratio (0.6–1.1, mean 0.8) and the presence of cereal-type Poaceae/Poaceae, Amaranthaceae (37.3–52.3%, mean 43%), and a decrease in pollen concentration (now with a mean 82166 grains/g).

There were no obvious changes in the percentage of Poaceae pollen (3.1–9.8%, mean 6.7%) but there was a higher abundance of Cyperaceae pollen (0.4–2.2%, mean 1.1%) compared to the previous zone. In Zone Ⅲ, there was an obvious increase in aquatic pollen (Typha and Sporangium) percentages (0.4–6.5%, mean 2.3%), tree pollen percentages (0.8–9.6%, mean 3.9%) and Cereal-type Poaceae/Poaceae (0–33.3%, mean 10.6%). The Ar/(Am + E) ratio was relatively higher (0.6–1.1, mean 0.8), but was interrupted by a brief period (c. 1550–1700 CE) with lower values (0.7–0.8, mean 0.7).

Sub-zone Ⅲ-2 (23–0 cm, c. 1650 CE to present) is distinguished from sub-zone Ⅲ-1 (40–23 cm, c. 1350–1650 CE) by an apparent increase in Artemisia (30.7–42.3%, mean 32.3%), Salix (0.2–2.7%, mean 0.9%), Sparganium (1.8–6.5%, mean 2.6%), Rosaceae (0–2.3%, mean 1.0%), and Cereal-type Poaceae/Poaceae (0–33.3%, mean 12.1%) and a slight decrease in Poaceae (3.1–9.8%, mean 6.1%).

The size of pollen source area depends on the size of the depositional basin, characteristics of pollen types and spatial distribution of the plant species (Jackson, 1990; Sugita, 1994). Sugita (1994) suggests that pollen assemblages of sedimentary basins with a radius larger than 750 m is mainly influenced by regional pollen rather than local pollen, and hence we considered the pollen assemblages in Jili Lake (∼174 km²) can reflect regional vegetation variations. Because of strong aeolian activity in Northern Xinjiang, we suggested pollen assemblages are a mixed signal to the regional vegetation in the basin. As modern vegetation types show there are less forest around Jili Lake except some specific planted poplar trees to protect farmlands (Xinjiang Comprehensive Investigation Team, Chinese Academy of Sciences, 1978), the tree pollen types in the lake could have been transported by wind and/or river from the southern slopes of the Altai Mountains.

The Sparganium is adapted to growing in wet conditions, and the high content of Sparganium might indicate a low lake level (Davis, 1999). The aquatic plants in JL18-02-A core from Jili Lake were mainly Typha and Sparganium, both of which are emergent plants. A previous study showed the higher content of aquatic pollen and lower total pollen concentration may indicate a shallower lake environment (Huang et al., 2010).

To verify the reliability of Ar/(Am + E) ratio as an indicator of climatic humidity, we compared it to the relative humidity records from meteorological stations (Figure 4). It was demonstrated that the Ar/(Am + E) ratio fluctuations of JL18-02-A generally follow the relative humidity records from the Fuhai station, both showing an increasing trend in humidity during the 1950–2018 CE (Figure 4).Therefore, the Ar/(Am + E) ratio was a reliable indicator of humidity in this region, which is similar to A/C ratio used by many other previous studies (e.g., Huang et al., 2009; Zhang J. et al., 2018).

The moisture variations inferred from the Ar/(Am + E) ratio from Jili Lake (Figure 5A) over the last 2400 years is consistent with other regional records (e.g., Sayram Lake (Figure 5C) and Ebinur Lake (Figure 5D), and suggested that an increase in regional moisture, which is consistent with an increase in the average moisture index of Xinjiang towards the present (Wang et al., 2013). The nearby Lake Ulungur (Figure 5B) also showed an increase in moisture prior to 1450 CE, but then decreased sharply (Liu et al., 2008).

From 400 BCE-400 CE, there is lower water level and lower effective moisture recorded in Jili Lake, which is also seen in other records. For instance, climate conditions were drier at Sayram Lake (Figure 5C) and Ebinur Lake (Figure 5D) during this period (as interpreted from a low A/C ratio). This period of drying is coincident with strong solar activity (Figure 5J), and the known warming associated with the Roman Warm Period (RWP, 0–400 CE) (Biintgen et al., 2011). In low-elevation areas in northern Xinjiang, there are many sites that suggest a dry climate during the RWP (Feng et al., 2017). In the Aral Sea area, there was a period of low lake level from 1 CE to 425 CE (Sorrel et al., 2006), and at Bosten Lake, a period of high salinity and lower lake level occurred from 280 to 480 CE (Fontana et al., 2019; Li et al., 2021). Historical documents show that in the 4th Century, the city of the Loulan Kingdom experienced a severe drought (Li et al., 1991), which might also have been recorded in Jili Lake, as indicated by a lower pollen humidity index (PHI) at 300–400 CE. The results of lithological and grain-size analyses from Jili Lake suggest the lake was shallower before 400 CE, hence the hydrodynamics caused by wave action on the lake was relatively strong (Lang, 2020). We interpret the low pollen concentration as evidence of strong hydrodynamics and/or drier conditions. There are also differences in the records of humidity at different altitudes in the Xinjiang region. From 400 BCE to 400 CE, a decrease in moisture occurred in high altitude areas [e.g., Tielishahan Peat (Zhang et al., 2016) and Narenxia Peat (Zhang D. et al., 2018), Yushenkule Peat (Yang et al., 2019), and Sayram Lake (Lan et al., 2020)], while a slight increase in moisture occurred in low altitude areas [e.g., Jili Lake (Figure 5A), Ulungur Lake (Figure 5B) and Ebinur Lake (Figure 5D)]. In addition, during this period moisture records from Bosten Lake in the southern Tienshan Mountains (Figure 5H; Huang et al., 2009) showed a slight decreasing in moisture during this period, which is inconsistent with low latitude lakes in northern Xinjiang.

From 400 CE to 850 CE [Dark Ages Cold Period (DACP)], the PHI indicates an increase in regional moisture, overlain by fluctuations in wet and dry phases at a centennial scale. The water level in Jili Lake began to increase, likely reaching its present-day level after c. 400 CE (Lang, 2020). Similarly, the moisture record from Ebinur Lake (Figure 5D) shows an increase in regional moisture, and pollen records from Kanas Lake suggests an increase in effective humidity from 550–1050 CE (Huang et al., 2018).

During the MWP (850–1200 CE), the climate was warm and dry. Although warm climatic conditions prevailed in the Northern Hemisphere (Figures 5I, J), the PHI suggests relatively dry conditions in the Jili Lake region. These drier conditions may be related to a period of higher evaporation at Bosten Lake, and also coincided with a glacial retreat event recorded at Karakuli Lake (Liu et al., 2014). During this period, there is a higher content of coarse grain sizes (40–200 μm) from Jili Lake suggests an intensification of storm activity at Jili Lake (Lang, 2020). Likewise, the increased coarse fraction in the record from Harnur Lake (Figure 5E) and decreased A/C ratios at Ebinur Lake (Figure 5D) corroborate this interpretation.

During the Little Ice Age (LIA: 1200–1850 CE), there was an increase in moisture from 1200 CE (Figure 5A), consistent with the A/C ratios of Sayram Lake (Figure 5C), Ebinur Lake (Figure 5D) and Harnur Lake (Figure 5E). Cold climate also recorded in Manas Lake (Song et al., 2015) and Bosten Lake (Figure 5G) during this period. This period is not characterized by uniform wet conditions. For example, the period 1560–1700 CE was relatively dry with a lower PHI, though the regional climate was still wetter than the period prior to 1200 CE (Figure 5A). At Jili Lake the abundance of aquatic pollen (Typha and Sparganium) peaked at 1665 CE, with a brief interval of lower water level at 1600–1855 CE. Other regional records also attest to relatively dry conditions during 1560–1700 CE: Sichanghu peatland, Hutubi River Basin, Manas Lake, Kesang Cave and Dalong Pond (Chen F. et al., 2015; Song et al., 2015; Cai et al., 2017; Lan et al., 2019; Ren et al., 2019). At Bosten Lake, the carbonate content (Figure 5G) suggests an interval of high evaporation during 1550–1700 CE; higher temperatures are recorded at the same times [e.g., at Manas Lake (Song et al., 2015) and Belukha glacier (Eichler et al., 2011)], suggesting that temperature was the main driving force of decreased moisture during this period. After 1850 CE, the PHI derived from the Jili Lake record suggests a drier climate, which is consistent with the records of other lakes in northern Xinjiang (Huang et al., 2009; Feng et al., 2016; Li et al., 2017; Yang et al., 2019; Yang et al., 2020).

Previous studies from northern Xinjiang showed that the region alternated between “cold-wet” and “warm-dry” during the late Holocene (Feng et al., 2006). Combining the Northern Hemisphere temperature (Figure 5I) with the PHI derived from the Jili Lake record, there is support for the climate model of “warm-dry” and “cold-wet” (Chen et al., 2019), overlain by a millennial timescale wetting trend from 1 CE to 1550 CE.

By comparison of regional moisture changes with records in monsoon marginal region (Figure 6), it was found that there was an increasing trend in moisture at Jili Lake (Figure 6A) during 400 BCE-2000 CE, which is in contrast to records from Hurleg Lake (Figure 6E) and Genggahai Lake (Figure 6F). The reason for these differences may be the climate of Hurleg Lake and Genggahai Lake being mainly driven by the Asia Summer Monsoon (Liu et al., 2016; Zhao et al., 2010). Records from Tian’E Lake (in the Qilian Mountains) and BYBS profile (in the western part of the Badain Jaran Desert) show a decrease in moisture during 400 BCE-1100 CE and an increase in moisture from 1100 BCE to 1800 CE, which was in antiphase to the moisture record at Sugan Lake (Figures 6B–D). An increase in moisture during 1100–1800 CE at Tian’E Lake and BYBS profile may be due to the weakening of the Asian Summer Monsoon caused by the reducing solar radiation (Steinhilber et al., 2012; Gao et al., 2020). In general, a wetting record from Jili Lake over 2400 years (Figure 6A) is obviously different from that in the monsoon marginal region.

The water vapor in ACA originates from the North Atlantic Ocean, the Mediterranean Sea, Caspian Sea, and regional water recycling within inland Asia, and is transported from west to east by the westerlies (Aizen et al., 2006; Chen et al., 2008). The southward shift of the westerly jet stream facilitates the infiltration of water vapor from the Indian Ocean to ACA and leads to more rainfall over northern Xinjiang (Yang and Zhang., 2008; Zhao et al., 2014). Solar activity plays an important role in atmospheric circulation (Reid, 1991; Steinhilber et al., 2012). Yan et al. (2019) suggesting that changes in solar activity and the intensity and location of the westerly jet stream are the dominant control on hydroclimatic variations in ACA. The stronger westerly and the southern migration of the westerly jet stream, which corresponds to lower total solar irradiance (TSI) and colder conditions, could favor more water vapor transport to ACA, and vice versa (Yan et al., 2019). During 400 BCE-400 CE and the MWP, the Jili Lake region had a warm-dry climate when the solar activity was higher (Figures 5A, J). During the LIA, there was a period of higher evaporation (1560–1700 CE) and an associated low moisture index.

Since 1700 CE, there were higher concentrations of macro-charcoal (>100 μm) (Figure 7C), accompanied by an increase in the Ce/Po ratio (Figure 7E). These indicators suggest that a period of intensified human activity played an important role in influencing the vegetation in the region, and this can be matched with local and state historical documents. The population of Xinjiang gradually increased during 1626–1760 CE (Figure 7F). Irrigation for agriculture in northern Xinjiang began in the early Qing Dynasty (from 1716 CE) (Zhao and Xie, 1988) with a rapid phase of development during 1749–1840 CE (Fang, 1989). After 1755 CE, there was an upsurge in urban construction in northern Xinjiang, driven by political, economic and military factors, as well as expended cultivation and trade development (Yang, 2018). Likewise, during the 1600–1850 CE, Jili Lake had a lower lake level as indicated by the higher percentage of aquatic pollen (Typha and Sparganium) (Figure 7B) and the lower pollen concentration (Figure 7D). Notably, the PHI derived from the Jili Lake sediments was relatively higher during 1700–1850 CE and indicated relatively wet conditions (Figure 7A), which is in contrast to the lower lake level. Therefore, the wetter climate during 1700–1850 can’t explain the lower lake level and we infer that the lower water level in Jili Lake during 1600–1850 CE may have been caused by human activities or the more extensive exploitation of water resources. Similarly, the decline of water level in Ulungur Lake was likely caused by the greatly increased regional population and the associated development of oasis irrigation agriculture along the rivers in the lower basin during the Ming and Qing Dynasties (Liu et al., 2008; Tuerhong, 2011; Ni et al., 2021). Consequently, lower lake level may be mainly influenced by stronger irrigation agricultural activities during 1700–1850 CE.