According to the International Atomic Energy Agency/Global Network of Isotopes in Precipitation (IAEA/GNIP) precipitation sampling guide, a precipitation collector was manually constructed in an open area near the BLH meteorological station. To avoid the contamination of water vapor from the evaporation of shallow soil and surface water and the mixing of windblown snow, this collector was installed 2 m above the ground. We define 1 complete precipitation day as beginning at 20:00 UTC+8 on 1 d and ending at 20:00 on the next day; then, one sample was collected. All the rainfall samples were immediately collected after the end of the precipitation to minimize the effects of evaporation. Hail and snow samples were filled in pre-cleaned plastic bags, the air was pressed out from the bags and sealed to avoid the water vapor exchange, and all samples in sealed bags were melted at room temperature (25 °C). In order to clarify the changes comprehensively and accurately in the precipitation isotopes in the BLH, we tried to collect all samples during every precipitation event, including light rain and short time events (usually with precipitation amounts of less than 5 mm). Accordingly, a wide-mouth stainless-steel plate (400 mm × 600 mm) was used to collect enough samples of light rain and short time rain/snow events for analysis.

Regarding preserving samples, 100 mL high-density polyethylene (HDPE) bottles were used. Before the sampling, the bottles were washed three times with rainwater and then rapidly filled. In total, 554 precipitation samples were collected, including 224 rain samples, 203 snow samples, 85 hail samples, and 42 sleet samples.

Samples of thermokarst lakes/ponds (Fig. 2) were collected by hand using a self-made water sample collector at monthly intervals during ice-free seasons (between May and October) from 2017 to 2022 in the BLH (Fig. 1). During the observation periods, the occurrence numbers of thermokarst lakes dynamically changed among different sampling years (Table 2) due to the interannual variations in the precipitation, active layer thickness, supra-permafrost water, and near-surface ground ice. Some of the sampled lakes disappeared in the next sampling year, and additional new lakes emerged. Accordingly, we obtained sufficient lake water samples to constrain the seasonal changes in the lake water hydrology and try to clarify the influence of permafrost and climate on the water balance of thermokarst lakes in this region. Influenced by the COVID-19 pandemic and lockdown policies between August 2022 and December 2022 in China, only 2 months of sampling work (June and July) were conducted in 2022. Lake water samples were taken at the center of lakes from 20–40 cm below the water surface. The running water samples of stream water samples were collected at each fixed point 20–30 cm beneath the water surface. In addition, the supra-permafrost water and sub-permafrost water were randomly collected using an artificial water ladle at the location where the springs gushing out during each fieldwork. The water ladle was washed using the spring water before sampling.

In total, as many as 2402 thermokarst lake/pond samples, 675 stream water samples (Table 2), 102 supra-permafrost water samples, and 19 sub-permafrost water samples were collected during 6 years of continuous sampling work.

To clarify the characteristics of ground ice and its role in the local hydrological cycles and regional eco-environment, we have designed 17 boreholes (∼20 m depth) in the BLH (Fig. 1). A total of 12 boreholes was drilled near the Qinghai–Tibet Highway (QTH) in 2014, and 5 boreholes were distributed in the center of BLH, which were drilled between 2011 and 2021. In addition, two thaw slumps were dug (Fig. 1). Frozen soil cores were extracted from different depths using a mechanical drilling rig with a drilling diameter of 157 mm (Fig. 3). All visible ground ice samples were collected immediately after the core barrel was pulled out. During the sampling work, disposable PE gloves were used, and the exterior of each sample was removed to avoid contamination from mud and the surplus water in the borehole. In total, 355 ground ice samples were collected from 17 boreholes, and 4 samples were obtained from 2 profiles (Fig. 3; Table 3).

Liquid water storage. All the samples were transferred to 100 mL high-density polyethylene (HDPE) bottles. The sample bottles were filled up without bubbles and sealed with Parafilm. The collection date of the sample types (precipitation, lake water, stream water, and groundwater) was added to the bottle label. For the precipitation samples, the precipitation types (rain, snow, and hail) were recorded. All of the samples were stored at 4 °C and shipped to the State Key Laboratory of Frozen Soil Engineering (SKLFSE) at the Northwest Institute of Eco-Environment and Resources, Chinese Academy of Sciences (CAS), China.

Ground ice storage. All the treated raw frozen soil samples were immediately preserved in HDPE bottles. The massive ice and pure ice layers were sealed in pre-cleaned plastic bags. The depths and drilling site information were recorded. All of the frozen soil and ground ice samples were kept frozen at -4 °C in the field to avoid the sublimation of the ice and the evaporation of the water in the soil.

The precipitation samples were transferred to HDPE bottles immediately. If multiple rain/snow events occurred during 1 sampling day, the water sample from every single precipitation event was collected separately. At the end of 1 complete sampling day, all the samples collected from a single event were mixed. If the precipitation types changed during 1 sampling day, then different samples were collected separately. The final complete samples were kept cool at 4 °C. All we have done is to avoid the influence of evaporation on the enrichment of δD and 18O and ensure the originality of the samples.

During the sampling work of thermokarst lakes/ponds and streams, we do our best to control the sampling time at the same period during every month (controlling the sampling time within 1 week, i.e., between the 17th and 22th of every month) to make sure that all the samples can represent the average level of the whole month. The sampling HDPE bottles were precleaned three times using raw water. Lake water was taken at the center of lakes from 20–40 cm beneath the water surface. The running water samples of the stream were collected at each fixed point 20–30 cm beneath the water surface.

Before we started to analyze the samples, we first prepared 14 distilled water samples with the same stable isotopes to test the stability of our analyzer. The precision of the δ18O and δD values was calculated by calculating the 1σ standard deviation of groups of 12 injections and then calculating the average of these standard deviations. In order to ensure the data quality, the “high-precision” mode was employed during analysis. Under this mode, the analyzing time for each injection is about 8.75 min. The drift of the analyzer was determined by taking the mean of these same 12 groups of measurements and calculating the difference between the maximum and minimum means. All of these measured precision and drift values were smaller than those of the guaranteed precision (0.025 ‰ and 0.1 ‰ for δ18O and δD, respectively) and drift values (0.2 ‰ and 0.8 ‰ for δ18O and δD, respectively), indicating that the analyzer achieves both a good repeatability and a good reproducibility. If the measured precision and drift values did not pass the guaranteed values, then a comprehensive inspection of the analyzer was conducted, i.e., the instrument analyzing system, the vaporizer module, and the quality of the dry nitrogen. After completing all processes of checking, we repeated the analysis of 14 distilled/tap water samples and calculated the drift values until they passed the guaranteed values. The results were normalized to the Vienna Standard Mean Ocean Water–Standard Light Antarctic Precipitation (VSMOW–SLAP) scale by analyzing the internal standards before and after each set of 10 samples. Five laboratory standards (provided by LICA United Technology Limited, Beijing, China) with given isotopic values were inserted before 10 samples, which were used for instrument calibration with δ18O values of -21.28 ‰, -16.71 ‰, -11.04 ‰, -7.81 ‰, and -2.99 ‰ and δD values of -165.7 ‰, -123.8 ‰, -79.6 ‰, -49.2 ‰, and -9.9 ‰. The best-fit linear relationship between the five known calibration values and the analyzer's reported values was determined. The slope and intercept of the best-fit line through these points are used to calibrate the results of our samples.

To avoid memory effects, the first three results of the measurements were discarded, and arithmetic mean values were calculated from the last three injections. During the analyzing process, the real-time data of the water concentration of all injections were controlled within a range between 19 000 and 20 000 ppm and with a standard deviation smaller than 200 ppm. Once the water concentration values appeared to decrease, the work was stopped, and the syringe was detached for washing with deionized water. All measurements were post-processed with the Picarro ChemCorrect™ software to monitor the organic contamination and correct the data.

The stable isotopes of precipitation exhibit a remarkable seasonal variability during 6 years of observations (Fig. 4). The δ18O and δD of the local precipitation in the BLH ranged from -30.4 ‰ to 6.2 ‰ and from -238.0 ‰ to 65.4 ‰, respectively. The d-excess ranged between -37.5 ‰ and 44.5 ‰. The amounts of the weighted-average values of annual precipitation are -10.9 ‰, -72.1 ‰, and 15.4 ‰ for δ18O, δD, and d-excess, respectively. As shown, the δ18O and δD display distinct seasonal patterns with high values in summer and low values in winter (Figs. 2, S2); this is due to the changes in moisture sources and the influence of local climate conditions (Guo et al., 2022; Tian et al., 2005; Guan et al., 2013; Bershaw et al., 2012).

For comparison, the δ18O and δD of thermokarst lakes/ponds are more positive than those of precipitation due to strong evaporation and the resultant enrichments of heavier isotopes in lake water (Yang et al., 2016; Narancic et al., 2017; Ala-aho et al., 2018). δ18O ranged from -14.4 ‰ to 5.7 ‰ (mean: -6.0 ‰), δD is between -104.1 ‰ and 22.6 ‰ (mean: -48.0 ‰), and the d-excess ranged from -35.8 ‰ to 21.8 ‰ (mean: -0.1 ‰), respectively. Similarly, the isotopic patterns of thermokarst lakes/ponds exhibited strong seasonal variations (Figs. 4, S3), which are due to the changes in source waters (i.e., precipitation, meltwater of thawing permafrost/ground ice, and groundwater) and alternations of evaporation degrees due to air temperature fluctuations (Narancic et al., 2017; Yang et al., 2021; Aichner et al., 2022; Zhu et al., 2022). Generally, the heavy isotope contents of lakes/ponds are lower in August and September (Figs. 4, S3) and are is attributed to the recharges of monsoonal precipitation and water with more negative isotopes fed by melting ground ice (Gibson et al., 2015; Yang et al., 2021). In comparison, the majority of the isotope values of lakes/ponds are positive in May, June, July, and October (Figs. 4, S3) due to evaporation and recharge of isotope-enriched precipitation water.

For the streams, the isotope values varied from -13.7 ‰ to -7.2 ‰ (δ18O; mean: -11.1 ‰) and from -83.8 ‰ to -53.3 ‰ (δD; mean: -73.6 ‰), and the d-excess ranged from -0.6 ‰ to 25.6 ‰ (mean: 15.0 ‰), respectively. The mean values are equivalent to the average values of annual precipitation in the BLH. Compared with thermokarst lakes/ponds, the δ18O values of stream water exhibited relatively stable patterns (Fig. 4) due to the short residence time (Yang et al., 2021; S. Wang et al., 2023; Song et al., 2017), which indicates weak evaporation. However, the stream isotopes also represented seasonal variations during the 6 years of observation (Figs. 4, S4) as lower values prevailed in August and September. The temporal changes in the stream isotopes are mainly influenced by the seasonal variability in the evaporation (Yang et al., 2017) and differences in the source water, i.e., alternative replenishment of precipitation, melting ground ice, and groundwater (Streletskiy et al., 2015; Yang et al., 2019; Ala-aho et al., 2018).

The two kinds of supra-permafrost water (i.e., GSHQ and GSYTQ) exhibited a similar seasonal trend (Fig. 4). For comparison, the GSHQ displayed relatively more positive isotopic peaks during whole sampling periods (Fig. 4), with δ18O ranging from -13.3 ‰ to -5.8 ‰ (mean: -11.2 ‰), the δD ranging between -86.7 ‰ and -39.0 ‰ (mean: -74.2 ‰), and the d-excess varying from 6.5 to 22.4 ‰ (mean: 15.1 ‰), respectively. The isotopes of GSYTQ varied from -13.5 ‰ to -8.4 ‰ (mean: -11.4 ‰), the δD ranged between -83.2 ‰ and -50.6 ‰ (mean: -73.8 ‰), and the d-excess varied from 4.6 to 25.1 ‰ (mean: 16.9 ‰). The isotopic peaks of the two types of springs lagged behind those of precipitation (Fig. 4), indicating the replenishment of precipitation via infiltration. In contrast, the stable isotopes of sub-permafrost water are more negative than those of supra-permafrost water which range between -12.7 ‰ and -11.1 ‰ (mean: -11.8 ‰) for δ18O, from -83.7 ‰ to -77.7 ‰ (mean: -80.7 ‰) for δD, and from 10.9 ‰ to 17.7 ‰ for d-excess (mean: 13.5 ‰). In addition, they stayed nearly stable over a long time series (Fig. 4), suggesting unchanged sources of isotopically light water (monsoonal precipitation, meltwater from thawing permafrost, etc.) and an insignificant influence by precipitation.

The distributions of stable isotope dots of all cores are scattered along various depths (Fig. 5). Generally, the δ18O ranges from -15.0 ‰ to -8.3 ‰ (mean: -12.2 ‰), from -113.7 ‰ to -66.4 ‰ (mean: -94.4 ‰) for δD, and between -13.4 ‰ and 15.5 ‰ (mean: 3.1 ‰) for d-excess, respectively. Comparing with the precipitation, the majority of the δ18O values of ground ice is lower than the precipitation, indicating multiple sources of initial water during ice formation under variable climatic conditions and complex geological contexts on the QTP (Michel, 2011; Yang et al., 2017, 2023; Murton, 2013).

Specifically, the stable isotopes of ground ice varied between different boreholes (Fig. 5; Table 4). This is attributed to the influences of the initial source water and the complex ice formation mechanism. In addition, the isotopic patterns along various depths showed marked differences between boreholes (Fig. 5), suggesting influence of lithology on the water migration and freezing fractionation of stable isotopes (Yang et al., 2020; Lacelle et al., 2014; Fisher et al., 2021). Remarkably, the thaw slump ice was isotopically lighter than that of drilling ground ice (Fig. 5; Table 4); this is due to the considerable differences in the initial source water and freezing processes. The thaw slump ice is considered replenished by winter snowmelt via cracks and quick freezing (Fritz et al., 2011; Porter and Opel, 2020). However, the pore ice with lower isotopic values in these boreholes suffered isotope fractionation due to freeze–thaw conditions under climate transitions (Wetterich et al., 2014; Yang et al., 2023).

The local meteoric water line (LMWL), determined by three different methods, i.e., ordinary least squares regression using the daily isotopic data, the arithmetic mean isotopic values, and the amounts of weighted multi-monthly mean isotopic values during 6 years (2017–2022). They are expressed as LMWLevent, with δD = 7.97δ18O + 15.26 (r2=0.96); LMWLmonthly, with δD = 8.06δ18O +12.58 (r2=0.93); and LMWLPWA, with δD = 7.78δ18O +8.78 (r2=0.92). The slope is nearly identical to that of the global meteoric water line (GMWL; Craig, 1961). However, the intercepts are quite different (Fig. 6) due to the influences of precipitation amounts and the exceptional meteorological conditions (Barešić et al., 2006; Hughes and Crawford, 2012; Kern et al., 2016).

The δ18O–δD diagrams of lakes, streams, and groundwater were built using the monthly stable isotopic values and defined as the local evaporation line (LEL). The LELs observed during 6 years are calculated as δD =5.88δ18O -12.80 (r2=0.95), δD = 4.89δ18O -19.41 (r2=0.83), δD = 5.69δ18O -10.50 (r2=0.85) (supra-permafrost water), and δD = 3.54δ18O -39.06 (r2=0.92) (sub-permafrost water), respectively. The slopes of the three LELs are all lower than those of the LMWL (Fig. 6) and range between 4 and 6, indicating strong evaporation (Cui et al., 2017; Yang et al., 2019) due to the lower relative humidity (Clark and Fritz, 1997). Interestingly, the correlation coefficients of streams and supra-permafrost water are much lower (smaller than 0.9), and the slopes are smaller than those of precipitation and lakes/ponds (Fig. 6), which may be affected by the transitions of source water during warm seasons and the evaporative concentration of isotopes.

The δ18O–δD relationship for ground ice was established using the stable isotopic values of the ice samples, and the correlation is defined as the freezing line (FL; Souchez et al., 2000). In this study, the freezing line of the ground ice at 16 borehole sites was calculated as δD =5.36δ18O -29.15 (r2=0.73), which is significantly different from the LMWL (Fig. 6). The difference reflects the freezing characteristics of liquid water under different conditions (Lacelle, 2011). Our freezing slope between 6.2 and 7.3 was usually obtained during equilibrium freezing, Rayleigh-type fractionation (Lacelle, 2011). The lower correlation coefficient (Fig. 6) suggests variable freezing rates (Souchez et al., 2000), kinetic isotopic fractionation during ice formation (Souchez et al., 2000), and the influence of the initial source water of the ground ice at different sites (Lacelle, 2011; Yang et al., 2017).

Majority of the stable isotopes of the stream lie on the LMWL (Fig. 6) and embrace in the range of supra-permafrost water (Fig. 7). In addition, the mean value is close to the amounts of the weighted-average value of annual/summer precipitation, indicating the direct recharge of precipitation and supra-permafrost waters. However, some of the isotopic points do not lie on the LMWL, exhibiting a clear evaporative effect. The supra-permafrost water and sub-permafrost water display concentrated isotopic patterns compared with precipitation, reflecting relatively stable recharge sources. In addition, the scattered isotopic dots of supra-permafrost water rather than sub-permafrost water indicated changeable sources and climate conditions. For comparison, some of the isotope points of the supra-permafrost water are overlapping with those of precipitation and stream water, suggesting an important replenishment of precipitation and streams. However, the isotopic cluster of sub-permafrost water deviated significantly from the LMWL, and all the isotope values are lower than the annual average value of modern precipitation, suggesting the recharge signal of past water with negative δ18O (δD) values under cold climate conditions. The LEL of thermokarst lakes/ponds significantly deviated from LMWL (Figs. 6, 7), and some of the isotopic dots overlapped with precipitation, groundwater, and ground ice, indicating the hydrological connections between them (Yang et al., 2016, 2017).

The cluster of ground ice is partly overlapped with precipitation, groundwater, lakes, and streams (Fig. 7). It is indicative of the mutual replenishment relations between them. Some of the isotope dots are more positive than the summer precipitation, implying the recharge from the evaporative active layer water. A clear freezing slope is shown, indicating typical freezing of liquid water (Jouzel and Souchez, 1982; Souchez and Jouzel, 1984; Lacelle, 2011; Perşoiu and Pazdur, 2011). However, the d-excess values of ground ice are lower than those of river water and the amount of the weighted-average value of annual/summer precipitation (Fig. 7), suggesting the important recharge of the active layer of water (subjected to evaporation) by the near-surface ground ice (Yang et al., 2013; Throckmorton et al., 2016). In addition, the thaw slump ice exhibited more negative isotopes which are even lower than the amount of the weighted average, indicating the main recharge of the snowmelt water (Yang et al., 2020; Opel et al., 2018).