The samples were collected to a depth of 385 cm in two parts: at 0–200 cm, with 2 cm sampling intervals (n = 99 samples; No. A1–A100, A20 missing), and at 200–385 cm, with 5 cm sampling intervals (n = 34 samples; No. B1, B2, B7–B38). The texture and color of the ZZH sedimentary faces were described as followed (Ning et al., 2019): 0–40 cm corresponds to a moderate consolidated sand layer with a grayish yellow color; 40–100 cm reflects the limnetic layer with a gray-black color; 100–125 cm represents rust spots in a yellow sand layer; 125–170 cm represents more silt in a gray-black silt limnetic layer; 175–348 cm, represents silt-sand with a yellow-green color. The 175–248 cm upper layer shows grayish-green rust spots in permanent yellowish-brown fine sand; the 248–348 cm under layer reflects grayish-green; at 348–385 cm, there are abundant plant residues mixed in a black peat layer; and, below 385 cm, there is a brown unconsolidated aeolian sand layer.

Seven samples for dating were collected from the profile, three were sent to the AMS dating laboratory of Peking University for determination, and the remaining four samples were pretreated and dated by BETA Analytic Inc. (Miami, FL, United States). The phytolith extraction for all of the 133 sediment samples was conducted at the School of Geographical Sciences, Northeast Normal University, Jilin, China. Ning et al. (2019) previously established a chronological sequence based on the AMS ¹⁴C dating results for the ZZH profile in the Badain Jaran Desert.

Experiment samples were treated using the wet ashing method (Lu et al., 2006; Gao et al., 2018c) according to the following steps: (1) Samples were dried and their weights recorded; (2) For oxidation, 25 g samples were placed in a centrifuge tube and dilute hydrochloric acid was added to remove the carbonate. The samples were washed with distilled water, and concentrated nitric acid was added to completely oxidize organic matter. (3) For heavy liquid flotation, a zinc bromide solution (‘heavy liquid’) with a specific gravity of 2.34 was added at a volume of more than 1.5 times that of the sample and centrifuged at 2000 rpm for 15 min. The upper suspension was removed to another centrifuge tube and distilled water was added for repeated centrifugal washing until the sample pH was neutral; (4) After flotation, the heavy liquids were cleaned, spores of known concentration (one spore tablet is about 27,640 grains) were added, dissolved with dilute HCl, and the samples were repeatedly washed until the pH was around 7; (5) The phytoliths were then identified and statistically categorized using permanent reference slides and an Olympus optical microscope. Through this analysis, 43,773 phytoliths were identified across the 133 samples from the ZZH profile. In general, the phytolith concentration from 0 to 162 cm was relatively high, (average, 473,000 grains·g⁻¹); whereas the concentration from 162 to 375 cm was comparatively low (average, 1,300 grains·g⁻¹). Calculation of the percent phytolith content was done according to the following formula: p = n / s × 100 % (1) where p is the percentage of each type of phytolith, n is the number of each phytolith type, and s is the total number of phytoliths counted. C ≡ ( s × M ) / ( S × m ) (2)

C is the concentration of phytoliths, s is the total number of phytoliths counted, M is the number of spores on each tablet, S is the total number of spores, and m is the sample weight.

Phytolith types were identified (Figures 2, 3) in the experiment. Most of the phytoliths were derived from Poaceae and Cyperaceae, with occasional ferns and Xylophyta also being present.

From the 133 sediment samples, a large number of phytoliths were observed, we selected ten of the most representative phytolith types, namely rondels, saddles, dumbbells, elongated, fans, long points, short points, rectangles, gobbets, and wavy (Lu et al., 2006). The remaining types were excluded due to their small numbers. We combined short saddles and long saddles into saddles, sinuate elongate and smooth elongate into elongate, and wavy-trapezoid and wavy narrow into wavy.

According to Lu et al. (2006), there were 243 sites with calibration sets for the modern surface sediment based on the wide ecological and climatic gradients in China. Therefore, we chose the 27 sites within the 243 surface sediments calibration set from Lu et al. (2006). The mean annual precipitation for these 27 sites was <200 mm which included the northwestern arid region in China. We then extracted the temperature data for these 27 sites in concert with the China meteorological forcing data (CMFD) during the time span of 1979–2018. This study applied the transfer function to the phytolith types from the ZZH profile in the Badain Jaran Desert; the R (4.2.0) package of “rioja” was selected for the modern training set of 27 sites from the data set of 243 surface sediments (Lu et al., 2006). First, the phytolith-inferred MAT_{pre. 1} was the result of the weighted-averaging partial least squares regression (WA-PLS) model (Ter Break and Juggins, 1993), which extracted the most efficient component. The modern predicted temperature reconstruction (MAT_{pre. 1}) was then calculated in R [R_-boot ² = 0.26, root-mean-square-error (RMSE) = 2.62]. Second, the MAT_{pre. 1} was linearly fitted with CMFD (MAT_{obs. 1}) (known from 1979 to 2018) and a linear fitted equation was obtained. The inference data with the cross-validation are reflected by the MAT_{obs. 1} and the MAT_{pre. 1} had a statistical correlation (Figure 4). Third, the phytolith data in this study were substituted into the above model, and the Holocene paleotemperature reconstruction was obtained (MAT_pre.) [R_-boot ² = 0.28, RMSE = 4.00]. Due to the value of R ² (MAT_pre.) being relatively low, we used the linear fitted equation to obtain the results for the observed Holocene paleotemperature (MAT_obs.). M A T o b s . = 1.02 × M A T p r e . + 0.03 (3) where MAT_pre. is the predicted Holocene paleotemperature, and MAT_obs. is the observed Holocene paleotemperature.

Many other Holocene climate records have been obtained from some arid regions in China. However, the timing and mechanisms of climate change during the Holocene is still being debated (Long et al., 2014). The phytolith-based paleotemperature reconstruction for the Badain Jaran Desert may provide more reliable information for the paleoclimate record. We compared the reconstructed paleotemperatures for different regions in China using various proxies. Based on the reconstructed temperature sequence, Fang and Hou (2011) divided the Holocene paleotemperature into three stages in China, corresponding to the early, middle, and late Holocene. The paleotemperature during the early Holocene (11.5–8.9 ka BP) was fluctuating and increased and was approximately 1°C higher than in modern times. The paleotemperature in the middle Holocene (8.9–4.0 ka BP) reached its warmest peak during this period (8.0–6.4 ka BP). In the late Holocene (4.0 ka BP to the present), the paleotemperature decreased and the climate became cold (Figure 6B). Based on the geochemistry record of δ¹³C in Dajiuhu peat by Ma et al. (2008), the early Holocene was cold, but the paleotemperature gradually increased. The middle Holocene was warm overall, but the representative cold event at 8.2 ka BP was strongly reflected, and the paleotemperature changed from warm to cold during the late Holocene (Figure 6A). We also compared the δ¹³C record from the Hani peat (Hong et al., 2005) in northeastern China (Figure 6C) and the Hongyuan peat on the Tibetan Plateau (Hong et al., 2003) (Figure 6D), which was found to be similar to the Dajiuhu peat record. The pollen record in Daihai Lake (Xiao et al., 2004) also revealed the history of paleotemperature change during the Holocene in this region. From 10,250 to 7,900 ka BP, an increasing proportion of tree pollen implied that the environment had become warmer. Between 7,900 and 4,450 ka BP, the percentage of tree pollen reached its maximum value and the paleotemperature fluctuated but was relatively warm overall. During the period of 4,450 to 2,900 ka BP, tree pollen appeared and there was a sharp cooling with the paleotemperature dropping lower than in the previous period. In addition, the paleotemperature variation trend for the Qinghai Lake (Hou et al., 2016) (Figure 6E) showed results similar to our reconstructed MAT (Figure 6F). Hafsten (1970) also put forward the term ‘megathermal’ to explain the warmest period in the middle Holocene (8.2–3 ka BP). The data of the sediment core indicated that the Holocene megathermal was an arid period in Inner Mongolia (Chen et al., 2003). The above research suggests that the middle Holocene was the warmest period during the epoch.

From the TIC and TOC sediment records from Daihai lake, Xiao et al. (2006) found that the climate of the early Holocene (11,500–8,100 ka BP) was warm and dry, whereas the middle Holocene (8,100–3,300 ka BP) was warm and wet due to the high TIC and TOC values, and during the late Holocene the climate became cold and dry (Figures 7D,E). Chen et al. (2008) hold the opinion that the Holocene climate in arid central Asia was driven by the surface sea temperature in the North Atlantic, the waning ice-sheets in the northern Hemisphere, the high air temperature at mid-latitudes, and the high degree of summer insolation. Our research found that the Holocene paleotemperature in the Badain Jaran Desert was similar to the sea surface temperature (SST) in the western tropical pacific (WTP) which has been reconstructed using Mg/Ca data and oxygen isotope levels from foraminifers extracted from sediment cores (Stott et al., 2004) (Figure 7G). From the δ¹⁸O record in the Dongge Cave, China (Figure 7F), Dykoski et al. (2005) found that the Asian monsoon intensity generally correlated with the insolation changes, and our results were similar to the solar forcing mechanism in the Northern Hemisphere (Laskar et al., 2004) (Figure 7A). Peng et al. (2016) considered that the peak in Northern Hemisphere summer insolation was from 12 to 10 ka, correlating with the dune mobility in the Tengger Desert and other sand fields. This phenomenon might correlate with the strengthening of the Asian summer monsoon. This evidence could support our results that there were high paleotemperatures during the early Holocene.

The high paleotemperature during the early Holocene may have caused the high lake level and relative humid environment in the Badain Jaran Desert. In the early Holocene, the global temperature began to rise as solar radiation increased in the Northern Hemisphere, and the Badain Jaran Desert responded accordingly. Some researchers suggested that groundwater recharge was an important factor in lake-level changes in the Badain Jaran Desert. This is because groundwater recharge contributes 90% of the water balance, while precipitation contributes only 10% to the lake replenishment (Ma et al., 2014; Dong et al., 2016; Sun et al., 2018; Dong et al., 2022). The pan-lake period of Badain Jaran began in the early Holocene at 10 ka BP (Wang et al., 2016). This phenomenon can be attributed to increasing meltwater from the Qilian Mountains and the Tibetan Plateau, which is controlled by solar radiation in the Northern Hemisphere, and the high paleotemperature during the early Holocene. Thus, a considerable amount of meltwater flowed into the ground and became groundwater (Wu et al., 2004; Dong et al., 2022), while the glacier meltwater and precipitation in the northeastern Tibetan Plateau provided a large volume of groundwater to the Badain Jaran Desert (Ning et al., 2019). According to Dong et al. (2022), the lake level frequently fluctuated between 3.82 and 9.21 m during the early Holocene (10.6–8.6 ka BP) and slightly fluctuated between 3.41 and 5.26 m during the middle Holocene (8.6–4.7 ka BP) (Figure 7B,C). The literature examined in this discussion could further explain the hypothesis that the high temperature in the early Holocene led to lake expansion in the Badain Jaran Desert.

Li et al. (2015b) indicated that calcareous root tubes predominantly formed in environments with high relative humidity and could not form in an arid environment. Therefore, they found that the frequency of calcareous tubes was mainly concentrated in the middle Holocene in the Badain Jaran Desert, and their presence suggested that the middle Holocene was relatively humid. The strengthened summer monsoon also resulted in increased moisture in the middle Holocene (Yang et al., 2011). Ning et al. (2019) found that the concentration of n-alkanes was higher in the humid environment from 7.8 to 5.8 ka in the Badain Jaran Desert, which indicated the presence of a relatively wet environment during the middle Holocene. Moreover, when we referred to the pollen of Artemisia and Chenopodiaceae in arid areas, their A/C ratio proved that they afforded a reliable proxy for relative humidity (Zhao et al., 2012). Thus Ning et al. (2021) posited that the highest A/C ratio, between 7.2 and 5.4 ka, explained the high humidity conditions in the Badain Jaran Desert during the middle Holocene. During 7.7–5.3 ka BP, precipitation reached 200 mm yr⁻¹ in the Badain Jaran Desert (Wang et al., 2016). Combined with our MAT data and the above evidence, we have deduced that the middle Holocene period was warm and wet and that this was due to extensive precipitation from the summer monsoon and the low evaporation in the desert (Li et al., 2015b; Wang et al., 2016; Ning et al., 2019; Dong et al., 2022), which had a strong impact on the high lake levels during this period.

During the late Holocene, many lakes retreated or dried up in the Badain Jaran Desert (Wang et al., 2016). This may have been caused by weak summer monsoons (Yang et al., 2010) along with human activities. From our MAT results, at approximately 3,200 ka BP, there was a sharp decrease in temperature, which may be a result of the neoglacial period (Cheng et al., 2020) or the arrival of the late Neolithic age.