The QLB lies in the northeastern sector of the QTP, encompassing a total area of 2.97×104 km² (Fig. 1). Among these, permafrost primarily occurs in the upstream and high-altitude areas, covering approximately 1.23×104 km², which represents 41.5 % of the basin's entire area. Seasonally frozen soil is mainly distributed in the midstream, downstream, and areas surrounding the lake, covering 1.74×104 km², accounting for 58.5 % of the total basin area. This basin is located in the transitional zone from the eastern monsoon region to the inland arid region, and belongs to the temperate continental semi-arid climate zone of the QTP, characterized by pronounced dry-wet seasonal variations, strong evaporation, abundant sunshine, and intense solar radiation (Li et al., 2016). Alpine meadows constitute the dominant vegetation type, followed by alpine steppes; together, they cover roughly 85 % of the basin's vegetated area. The basin features diverse soil types, mainly including alpine cold desert soil, alpine meadow soil, and alpine steppe soil, among others. Considering the influences of elevation, topography, and groundwater burial depth, the spatiotemporal variations in this study are classified into the downstream basin and middle and upper reaches of the basin.

The groundwater system exhibits distinct spatial differentiation: mountainous areas primarily serve as groundwater recharge zones, piedmont plains act as the main migration pathways and infiltration areas for subsurface runoff, while the plains surrounding the lake are predominantly characterized by groundwater discharge. In mountainous areas, pore water is stored as unconfined groundwater in the detrital layer, with a water table depth of approximately 15 m. The piedmont plain mainly consists of unconfined groundwater in Quaternary sand and gravel layers, with a water table depth of 5–25 m and an aquifer thickness of 25–98 m. The plain surrounding the lake is characterized mainly by unconfined and confined groundwater in sand and gravel layers, with a water table depth of approximately 6 m and an aquifer thickness of 10–62 m. Overall, the depth to the groundwater table and the thickness of the aquifer in the watershed exhibit an increasing trend with elevation (Fig. 2).

From May to October 2025, we collected precipitation, groundwater, and soil samples on a monthly basis in the QLB, taking into account elevation, soil type, and permafrost distribution. A total of 15 groundwater sampling points, 17 soil sampling points, and 11 precipitation collection points were established (Fig. 1). During the study period, a total of 481 samples were collected, comprising 90 groundwater samples, 306 soil samples, and 85 precipitation samples (Table 1). Groundwater was sampled from the well system used by local herders in the QLB. Before sampling, the well was flushed for 20 s, and after the flow stabilized, water samples were collected into 100 mL polyethylene bottles. Soil samples were collected in layers at depths of 0–30, 30–60, and 60–90 cm using a modified electric soil auger. Sampling depths typically ranged from 60 to 90 cm, with a maximum depth of 110 cm and a minimum depth of 50 cm. Immediately after collection, samples were placed into 12 mL sampling vials and aluminum boxes, sealed with Parafilm, and quickly frozen in a vehicle-mounted refrigerator for preservation. Rainfall samples were collected using standard evaporation-proof rain gauges installed in open areas. Rainfall samples were placed in 100 mL polyethylene bottles and stored frozen. Snow samples were collected from the snowpack using clean, sealed plastic bags. During sampling, the surface layer of snow was first removed, and snow samples were then collected immediately adjacent to the soil interface and allowed to melt naturally at room temperature on the day of sampling. All water samples, including snowmelt, rainfall, and groundwater, were filtered through a 0.45 µm membrane filter before being transferred to 100 mL polyethylene sample bottles.

All experimental analyses of the samples were completed at the State Key Laboratory of Earth Surface Processes and Hazards Risk Governance, Beijing Normal University. Precipitation, groundwater, and soil water were analyzed for δ2H and δ18O using an Isotope Ratio Infrared Spectrometer (DLT-100, Los Gatos Research, Mountain View, USA). The instrument's analytical precision was δ2H: ±0.5 ‰ and δ18O: ±0.1 ‰. The measurement results were calibrated using the LWIA Post Analysis liquid water isotope analysis software. Soil water was extracted using an automatic cryogenic vacuum distillation water extraction system (LI-2100, LICA United Technology Limited, China), followed by δ2H and δ18O measurements. To determine soil water content (SWC), soil samples in aluminum boxes were weighed and dried in an oven at 105 °C for 24 h.

During the thawing period, precipitation exhibited the most depleted δ2H and δ18O values, along with the highest degree of variability. The isotopic signature of groundwater remained comparatively constant, with δ18O and δ2H ranging from -8.65 ‰ to -6.25 ‰ and from -52.48 ‰ to -36.55 ‰, respectively. In the 0–30 cm soil layer, the δ18O and δ2H of soil water were relatively enriched, with the lowest lc-excess, whereas the δ18O and δ2H of soil water in the 30–60 and 60–90 cm layers were closer to those of groundwater (Fig. 3a). During the thawed period, the δ18O values of precipitation shifted overall from depleted to enriched (δ18O: from -11.57 ‰ to -8.93 ‰). The δ18O and δ2H values of soil water in all soil layers increased. Among these, the 0–30 cm soil water had the highest δ2H and δ18O values, and its lc-excess remained significantly lower than that of the 30–90 cm soil water and groundwater (p<0.05). In contrast, the variations in the 30–90 cm soil water and groundwater were relatively small. During the freezing period, the δ2H and δ18O values of soil water in the 0–30 cm depth interval were more depleted than those during the thawed period, while lc-excess showed a slight increase (Fig. 3b). Meanwhile, the δ18O and δ2H values of the 60–90 cm soil water and groundwater remained close to each other. Overall, precipitation exhibited the largest isotopic fluctuation during the FT periods, followed by soil water in the 0–30 cm layer, while the 60–90 cm soil water and groundwater were the most stable. Meanwhile, soil water δ2H and δ18O gradually decreased with increasing depth, whereas lc-excess gradually increased with depth (Fig. 3c).

In the QLB, the δ2H and δ18O values for precipitation, soil water, and groundwater all cluster around the LMWL, but the soil water line (SWL) and groundwater line (GWL) were generally below the LMWL (Fig. 4). During the thawing period, the difference between GWL and SWL was most pronounced. Among them, GWL was closest to the LMWL, whereas the 0–30 cm SWL exhibited the lowest slope and the poorest fit (Fig. 4a). During the thawed period, the slopes and intercepts of the soil water lines for all layers approached the LMWL, with the 30–60 cm SWL being the closest to the LMWL. However, compared to the thawing period, the slope of the GWL remained lower than those of all SWLs, and its R2 decreased (Fig. 4b). During the freezing period, the GWL slope and intercept rose markedly, approaching the LMWL closely. Meanwhile, the SWLs again diverged, exhibiting slopes of 5.28, 4.70, and 4.16 for the 0–30, 30–60, and 60–90 cm layers, respectively (Fig. 4c). During the FT periods, the slope of the SWL generally approached that of the LMWL as depth increased. The slope of the groundwater was 5.81, and the slope for 60–90 cm SWL was 5.90. These two values were the closest to each other (Fig. 4d).

The SWC in the QLB exhibited pronounced spatiotemporal variation characteristics (Fig. 5). In the middle and upper reaches, SWC in the 0–30, 30–60, and 60–90 cm layers was 26.9 %, 28.4 %, and 31.1 %, respectively, showing an increasing trend with depth during the thawing period (Fig. 5a). During the thawed period, the SWC in the 0–30 cm layer of the middle and upper reaches of the basin was the highest (27.8 %), while that in the 30–60 and 60–90 cm layers decreased to 26.2 % and 24.5 %, respectively. During the freezing period, the differences among the layers diminished, with values of 26.9 %, 25.5 %, and 25.9 %, respectively. Throughout the QLB, SWC was lowest in the 0–30 cm layer during the thawing period, while it was relatively higher in the 30–90 cm layer. During the thawed period, the differences among the layers decreased. The SWC in the 0–30 and 30–60 cm layers rose to 26.9 % and 26.3 %, respectively, while that in the 60–90 cm layer dropped to 24.7 %. During the freezing period, SWC generally increased across all soil layers, with the 60–90 cm layer showing the most significant increase, rising to 26.7 %. In the downstream basin, SWC in the 0–30 cm layer during the thawing period was only 15.7 %, significantly lower than the 23.8 % and 23.5 % observed in the 30–60 and 60–90 cm layers, respectively (p<0.05, Fig. 5b). During the thawed period, the SWC in all layers decreased to approximately 16.0 %, and the differences among the soil layers became significantly smaller. During the freezing period, SWC in all layers increased overall, with the most pronounced increase observed in the 60–90 cm layer (28.3 %), which was significantly higher than the 22.3 % and 22.8 % in the 0–30 and 30–60 cm layers, respectively (p<0.05). Overall, SWC in the middle and upper reaches of the basin was generally higher than in the downstream basin.

The MixSIAR model results indicate that groundwater recharge sources differ markedly during the FT periods (Fig. 6). During the thawing period, the contributions of soil water from the 0–30, 30–60, and 60–90 cm layers to groundwater were 25.61 %, 23.18 %, and 23.05 %, respectively, all higher than those of rainfall (16.29 %) and snowmelt water (11.88 %) (Fig. 6a, b). During the thawed period, the differences in contributions among sources (except snowmelt) diminished, with rainfall and soil water from different layers contributing 21.06 %, 21.50 %, 22.00 %, and 22.42 % to groundwater, respectively, and snowmelt water contributing 13.01 % (Fig. 6c, d). During the freezing period, the contributions from the 30–60 and 60–90 cm soil water layers rose to 25.30 % and 25.57 %, respectively. The contributions from 0–30 cm soil water, rainfall, and snowmelt water were 21.43 %, 18.34 %, and 9.33 %, respectively (Fig. 6e, f). Overall, during the FT periods, groundwater recharge was primarily sourced from soil water and precipitation, with snowmelt contributing the least. Specifically, soil water in the 60–90 cm layer contributed the most (23.15 %), followed by soil water in the 30–60 and 0–30 cm layers, which had similar contributions (22.94 % and 22.85 %, respectively). Rainfall contributed 19.02 %, while snowmelt contributed the least (12.02 %).

The sources and composition of groundwater recharge are influenced by a variety of factors, including meteorological conditions, topography, geological structures, and permafrost (Pavlovskii et al., 2017; Li et al., 2020; Hyman-Rabeler and Loheide, 2023). During the thawing period, groundwater in the QLB was recharged not only by precipitation but also by antecedent soil moisture storage released during the thawing process (Zhang et al., 2025), and the contribution ratios of these two sources differed significantly (Fig. 6). In the middle and upper reaches of the basin, the contributions of soil water across the three layers were relatively balanced, with rainfall and snowmelt contributing 16.17 % and 11.82 %, respectively (Fig. 7a). Although the contribution from snowmelt is relatively small, it cannot be ignored. In contrast, in the downstream basin, the contribution of soil water in the 0–30 cm layer was 27.93 %, higher than the 21.27 % in the 30–60 cm layer and the 22.36 % in the 60–90 cm layer (Fig. 7b). Overall, the direct recharge of groundwater from snowmelt is relatively weak. It is worth noting that the contribution of snowmelt water is significantly lower than that of rainfall. There are two main reasons for this. On the one hand, during the thawing period, the air temperature is low, leading to slow snowmelt (Fig. S1). On the other hand, under conditions of strong radiation and high winds, part of the snow water in the basin is lost through runoff or sublimation (Imran et al., 2025). Consequently, the proportion that is actually converted into groundwater is relatively small.

During the thawed period, air temperature, precipitation, and active layer thaw depth in the QLB all reached their annual peaks. In the middle and upper reaches of the basin, the differences in contributions from various water sources to groundwater during the thawed period were relatively small (Fig. 7a). It is worth noting that the contribution of soil water from the 0–30 cm layer in the middle and upper reaches of the basin was slightly higher than that of the other two layers and also higher than that of the same layer in the downstream basin. This may be related to the fact that during the thawed period, SWC in each layer of the middle and upper reaches of the basin remained at 26.90 %–31.05 %, significantly higher than that in the downstream basin. In the downstream basin, the contribution of rainfall to groundwater rose to 21.18 %, an increase compared to the thawing period (Fig. 7b). The contribution of soil water in the 0–30 cm layer decreased from 27.93 % during the thawing period to 22.45 %, while the contributions in the 30–60 and 60–90 cm layers remained at 25.60 % and 22.50 %, respectively. However, during the thawed period, the SWC of all three layers in the downstream basin was significantly lower than that during the thawing period (Fig. 5b), indicating that antecedent soil water storage had been doubly depleted by both continuous discharge to groundwater and strong summer evaporation. In addition, with the widespread melting of snow cover in summer, the contribution rate of snowmelt to groundwater dropped to 10.60 %.

During the freezing period, the continuous decline in air temperature in the QLB led to bidirectional freezing of the active layer from both the top down and the bottom up (Hu et al., 2023), causing soil water to transition from liquid to solid state. In the middle and upper reaches of the basin, the contribution rates of rainfall and snowmelt to groundwater decreased to 18.40 % and 9.33 %, respectively, while the contribution rate of soil water from the 60–90 cm layer increased to 26.87 %, which was significantly higher than the 19.33 % from the 0–30 cm layer (Fig. 7a). The downstream basin also exhibited similar characteristics. The contribution of snowmelt was the lowest, indicating that snowfall during this period participated in subsequent hydrological cycles more as temporary storage rather than being immediately converted into groundwater recharge.

The FT processes significantly influence groundwater recharge pathways in alpine regions by controlling the active layer's thaw depth, SWC, and soil water infiltration patterns (Wang et al., 2017, 2023a). The differences in soil water recharge pathways to groundwater are mainly attributed to their distinct hydrogeological and topographical conditions. The gentle topography, low elevation, and shallow groundwater table promote preferential flow of shallow soil water along large pores and fissures, leading to rapid infiltration and groundwater recharge (Li et al., 2025b; Rowland et al., 2011) (Fig. 8).

During the thawing period, as temperatures rise, the active layer thaws gradually from top to bottom, and the water in the permafrost gradually transitions from a solid to a liquid state. Especially in the middle and upper reaches of the basin, the depth of thaw is limited at this time and the lower soil layers have not yet thawed, creating a “near-impermeable layer” that impedes vertical infiltration (Hinzman et al., 2022; Vonk et al., 2023). Consequently, snowmelt and rainfall more readily form a transient saturated zone in the shallow subsurface and migrate as interflow or shallow lateral flow (Wu et al., 2024; Xie et al., 2024), thereby delaying the recharge of groundwater by precipitation and soil water (Du et al., 2024). Subsequently, this interflow or shallow runoff may indirectly recharge groundwater through pathways such as riparian seepage and soil layer redistribution (Xu et al., 2024). Meanwhile, SWC increased sharply during the thawing period (Li et al., 2020). This is consistent with results from modeling analyses indicating that spring hydrological processes in the alpine regions of the QTP are mainly governed by a combination of active layer temperature, thaw depth, and SWC (Wang et al., 2023b). The direct contribution of snowmelt in the middle and upper reaches of the basin is relatively weak, indicating that mountain slopes respond more rapidly to warming (Jay et al., 2023). The surface thawing process, together with the large slope gradients, promotes the generation of interflow (Evans et al., 2018), thereby recharging groundwater indirectly.

During the thawed period, air temperature, precipitation, and the thaw depth of the active layer in the QLB all reached their annual peaks, while the “near-impermeable layer” that limits vertical infiltration largely disappeared (Chang et al., 2015; Starkloff et al., 2017). In particular, in the downstream seasonal frozen ground basin, the vertical connectivity of the soil profile increases substantially after complete thawing (Li et al., 2020). Furthermore, a study in the Zuomaokong River basin on the QTP shows that surface runoff also increases with rising active layer temperature (Wang et al., 2009). Under these conditions, precipitation and soil water from the 0–30 cm layer can more easily move downward through the subsoil and eventually recharge groundwater (Ji et al., 2021). However, during the thawed period, the SWC in all three soil layers in the downstream basin was markedly lower than that during the thawing period (Fig. 5), indicating that the antecedent soil water storage has been depleted by both continuous groundwater discharge and strong summer evaporation (Fig. 3c). Low SWC makes infiltrated precipitation more likely to be retained by the soil rather than directly generating lateral flow (Chowdhury et al., 2011), thereby weakening the recharge of groundwater from soil water in the 0–30 cm layer. This also indicates that, despite increased vertical connectivity, soil water in the 0–30 cm layer has a short residence time (Wallach and Shabtai, 1992; Zuo et al., 2023b) and is more prone to downward transport (Lu et al., 2025). Generally, the higher the SWC, the greater the unsaturated hydraulic conductivity. After precipitation infiltrates, excess water is more easily discharged as interflow and replenishes groundwater (Hu et al., 2022), thereby maintaining a high contribution of shallow soil water to groundwater (Li et al., 2025a).

During the freezing period, the continuous decline in air temperature in the QLB led to bidirectional freezing within the active layer from both the surface downward and from the base upward (Hu et al., 2023), causing soil water to transition from a liquid to solid state. Notably, SWC in the 0–30 cm layer across the entire basin increased compared to the thawed period (Fig. 5c). This is primarily due to the formation of a new freezing front at the surface, which inhibits the continued downward percolation of water in the shallow layers, causing water to accumulate above the freezing front (Zhang and Sun, 2011; Bao et al., 2016). At the same time, the vertical connectivity of the soil profile has not yet been completely sealed (Jiang et al., 2024), and some shallow stored water continues to recharge groundwater downward via preferential flow pathways (Pittman et al., 2020; Cheng et al., 2024), thereby explaining why soil moisture in the 30–60 cm layer continues to make a significant contribution during the freezing period. Hydrological modeling of permafrost basins has also shown that when the active layer has not yet fully frozen at freezing onset, shallow soil water can still effectively recharge groundwater (Lu et al., 2023; Wang et al., 2025c) (Fig. 8).

In the QLB, the relationship between δ18O and δ2H indicates that the GWL lies between the LMWL and the SWL, suggesting that the groundwater recharge process is not a single pattern (Fig. 4). By further comparing the lc-excess indices of groundwater and soil water in the 60–90 cm layer during the FT periods, it is found that the two are not completely consistent, indicating that the recharge process is not controlled solely by piston flow. At the same time, the lc-excess value of groundwater is similar to that of precipitation, indicating that focused recharge via fast pathways still exists (Fig. 3c). Notably, during the thawed period, the lc-excess indices of soil water and precipitation are closest (Δlc-excess from 13.43 to 9.51), indicating that the role of preferential flow is gradually increasing. A study in the source region of the Yangtze River has also found that preferential flow accounted for the largest proportion of soil water infiltration during the thawing period, which is consistent with our findings (Li et al., 2025b). The QLB landscape is mainly defined by highland mountains and lakeside plains. The area surrounding the downstream basin mainly consists of plains, which feature gentle topography and relatively shallow groundwater levels (Li et al., 2022). Consequently, as the thawing process progresses, piston flow is more likely to occur. In contrast, the middle and upper reaches of the basin have a higher elevation and is mainly composed of highland mountains. Under the combined influence of FT cycles and topographic differences (Rooney et al., 2022), surface water is more likely to generate preferential flow along fractures (Wang et al., 2018). Observations in an alpine meadow further confirm that by regulating active layer thickness, FT processes significantly alter the proportions of piston flow and preferential flow (Musa et al., 2016; Li et al., 2026). Similarly, studies at the small watershed scale on the QTP indicate that there are significant differences in the contribution of soil moisture from various slope aspects toward groundwater recharge, making the mountain recharge pattern more complex (Zhang et al., 2024). Overall, during the process of groundwater recharge from soil water, preferential flow and piston flow coexist. As thawing proceeds, the dominance of piston flow gradually weakens, while the role of preferential flow in recharging groundwater progressively strengthens.

From an integrated perspective encompassing recharge sources, transport pathways and hydrological connectivity across the soil profile, this study systematically elucidates how FT processes regulate shallow groundwater recharge in the QLB. These findings highlight the critical role of soil water in sustaining shallow groundwater recharge in alpine regions. Against the backdrop of permafrost degradation and changing precipitation patterns, the spatiotemporal dynamics of groundwater recharge may be reorganized. Meanwhile, permafrost and seasonally frozen ground exhibit distinct responses to climate warming, resulting in differing effects on groundwater recharge. Moreover, focusing solely on rainfall sources may lead to an underestimation of the contributions of snowmelt and soil water to groundwater recharge. Although snowmelt accounts for a relatively small proportion of the total contribution throughout the FT period, its role should not be overlooked. Nevertheless, this study has three main limitations. First, the snowmelt samples were collected from the snowpack, and processes such as sublimation before snowmelt and evaporation from the soil surface after snowmelt may alter the isotopic composition of snowmelt, thereby leading to an overestimation or underestimation of its contribution to groundwater recharge (Hyman-Rabeler and Loheide, 2023; Gottlieb and Mankin, 2025). Second, during freezing, soil water transitions from the liquid phase to the solid phase. As the freezing front advances, light isotopes may be preferentially retained in the unfrozen liquid water, whereas heavy isotopes become preferentially enriched in the ice phase, resulting in isotopic fractionation and altering lc-excess, thereby complicating the identification of recharge pathways (Zhang et al., 2025). Finally, because piston flow and preferential flow coexist, water may migrate between different soil layers. Given the vertical stacking of soil layers and the continuous mixing of water from different sources, isotopic characteristics from the surface layer may be transported downward and influence the isotopic composition of deeper layers, thereby affecting the accuracy of groundwater recharge source identification. Consequently, this may introduce some error in MixSIAR modeling. Future research should integrate in situ high-frequency monitoring with a multi-tracer approach combining stable isotopes and dye tracing to accurately quantify the pathways through which soil water recharges groundwater.