Under the background of global climate warming， the sustainable development of ice-snow tourism is highly dependent on the stability and superiority of regional climate resources. Altay in Xinjiang， known as the “Snow Capital of China”， possesses outstanding ice-snow resources. However， systematic and refined assessments of its ice-snow tourism climate resources are still lacking. This study aims to： （1） quantitatively reveal the composition and stability of top-level ice-snow climate resources in Altay； （2） explore the long-term evolution characteristics of key resource elements under climate warming； （3） compare the differentiated advantages of three representative ski resorts； and （4） determine the optimal daily operational weather windows for each ski resort through hourly data analysis. A multi-scale assessment framework of “region-ski resort-hour” is constructed to provide a scientific paradigm for resource evaluation， differentiated development， and risk management of ice-snow tourism destinations in China.This study is based on daily observation data from the Altay National Reference Climate Station from 1954 to 2024 （71 years） and refined meteorological data from three major ski resorts （Jiangjunshan， Jikepulin， and Keketuohai） during the 2023/2024 and 2024/2025 snow seasons. A comprehensive evaluation system was constructed following the Xinjiang local standard Specification for Climate Resource Assessment of Ice-Snow Tourism （DB65/T 4875—2024）. Seven core indicators were selected： ice-snow season length， winter snowfall total， number of stable snow cover days （snow depth ≥10 cm）， snow-water ratio （representing snow quality）， number of days with sufficient sunshine （sunshine duration ≥6 h）， number of days with high visibility （minimum visibility ≥5 000 m）， and skiing meteorological suitability. Each indicator was classified into four grades with corresponding scores （1~4）. The total annual score （maximum 28 points） determines the suitability grade. A comprehensive ice-snow tourism climate resource index （I C） was constructed by standardizing the seven indicators and summing them， allowing integrated quantification and comparison of complex climate resources. For the three ski resorts， daily and hourly analyses were conducted， and a dual-threshold method was applied to hourly skiing meteorological index data to identify “highly suitable” time periods.（1） Altay possesses top-grade and stable ice-snow tourism climate resources. From 1954 to 2024， the average ice-snow season length was 140 days， average snowfall was 73.2 mm， the snow-water ratio ranged from 1.3 to 2.0 mm·cm-1， and the combined proportion of days with Grade 1 （highly suitable） and Grade 2 （suitable） skiing meteorological indices reached 89.4%. The long-term averages of all seven core indicators reached the highest “highly suitable” grade， with compliance rates exceeding 90% for key indicators.（2） Against the background of climate warming， Altay’s ice-snow tourism climate resources show a continuous “improvement” trend. Over the past 71 years， winter snowfall increased significantly at a rate of 1.49 mm·（10a）-1， the number of Grade 1 skiing days increased at a rate of 0.9 d·（10a）-1， and the comprehensive climate resource index （I C） rose significantly at a rate of 0.15 （10a）-1 （P<0.05）.（3） The three ski resorts each have distinct climate resource advantages while all attaining the highest “highly suitable” grade. Keketuohai has the highest comprehensive index （I C=2.97）， the longest snow season （average 172.0 d）， the best snow quality， and the most Grade 1 skiing days. Jikepulin’s core advantage lies in its extremely abundant snowfall （two-season average 360.5 mm）. Jiangjunshan is characterized by highly concentrated high-quality weather （annual average proportion of Grade 1 days is 62.8%）， and its meteorological conditions remain consistently “highly suitable” throughout all operating hours （10：00—23：00）.（4） Refined assessment based on hourly data clarified the optimal operational weather windows： Jiangjunshan is suitable all day （10：00—23：00）， Jikepulin from 12：00 to 20：00， and Keketuohai from 13：00 to 19：00， providing precise guidance for operational scheduling and safety management.This study systematically confirms the top-grade and improving ice-snow climate resources of Altay， the “Snow Capital of China”， and reveals the spatial heterogeneity among three major ski resorts. The constructed multi-scale assessment framework （“region-ski resort-hour”） and the comprehensive climate resource index （I C） method enable integrated quantification and comparison of complex climate resources. The findings provide scientific support for regional industrial planning， differentiated ski resort operation， and tourist service improvement. The methodology can serve as a reference for the resource assessment and sustainable development of other ice-snow tourism destinations in China. Future research should extend the observation period to verify interannual variability and long-term stability， explore weighted evaluation methods based on expert or tourist preferences， and couple regional climate models to project future changes under different emission scenarios.
Ice and snow resources， as critical natural assets that supply water， regulate climate change， and maintain ecosystem balance， hold irreplaceable significance and value for the sustainable development of humanity. Under the constraints of resource and ecological sustainability， promoting the high-quality protection and sustainable utilization of ice and snow resources has become an urgent issue to be addressed. This paper focuses on the core topic of realizing the ecological product value of ice and snow resources and systematically constructs a trinity research framework of “logic-mechanism-model”， aiming to resolve the theoretical dilemmas and practical pathways for transforming their value from a singular ecological focus to a composite of ecological， economic， and social values. The fundamental logic of realizing the value of ice and snow resources is analyzed from four dimensions： theoretical foundation， institutional mechanisms， technical support， and practical pathways. At the theoretical level， grounded in Marx’s labor theory of value， which posits the duality of “commodities possessing both value and use value”， and guided by the principle that “ice and snow are also as valuable as gold and silver”， this study reveals the dialectical relationship between the use value and economic value of ice and snow resources. It further emphasizes that realizing this value requires the condensation and monetization of human labor. At the institutional level， focusing on the core role of the natural resource asset property rights system and incorporating policy documents such as the Overall Plan for the Reform of the Ecological Civilization System， this paper clarifies an institutional guarantee system composed of property rights definition， equity distribution， and policy incentives. At the technical level， it integrates ecosystem accounting technologies with digital technologies driven by new quality productive forces， constructing a technical support system that includes satellite remote sensing monitoring， blockchain-based rights confirmation， and big data accounting. At the practical level， the feasibility of value realization is validated through typical cases such as the industrialization of Tianshan glacier meltwater supply， ice and snow tourism in Harbin， the “Ice and Snow Resource Bank” in Zhangjiakou. The study further proposes a transformation mechanism of “resources-assets-capital”. The government-led assetization mechanism addresses the question of “to whom do resources belong” through resource surveys and property rights registration， while achieving value explicitness through ecological pricing and policy supervision. The market-led capitalization mechanism accomplishes value appreciation via supply-demand pricing， technological innovation， financial instruments， and industrial chain integration. The synergistic feedback mechanism between the government and the market ensures that economic benefits are reinvested into ecological protection through institutional designs such as franchising and tax regulation， forming a closed-loop system of “ecology-economy-ecology”. Building on this， the paper distills four typical realization models： the eco-industrial operation model for the ice and snow industry， the green financial support model for the ice and snow economy， the environmental rights trading model for ice and snow， and the ecological compensation model for ice and snow. Among these， the eco-industrial operation model for the ice and snow industry focuses on reshaping the ice and snow economy through green transformation. It strictly adheres to ecological redlines in planning， drives comprehensive industrial chain upgrades through technological empowerment， optimizes operations using big data and artificial intelligence， and overcomes equipment bottlenecks with smart manufacturing and green materials. It innovatively leverages resource recycling and mobile facilities to develop ice and snow tourism， reinvesting profits into ecological engineering and achieving waste resourceization through restoration projects. The green financial model for ice and snow includes green credit， green bonds， and asset securitization related to ice and snow resources. The environmental rights trading model for ice and snow explores two pathways： exclusive ice surface leasing rights trading and non-exclusive ice and snow tourism carrying capacity trading. The ecological compensation model for ice and snow distinguishes between horizontal market-based compensation and vertical fiscal transfer payments based on the accessibility and usability of ice and snow resources， constructing a diversified compensation network. This paper establishes a theoretical framework for realizing the value of ice and snow resources， proposes pathways for value realization， enriches and extends the application of the “two mountains” theory in the field of ice and snow， provides theoretical support and practical approaches for the comprehensive development， protection， and sustainable utilization of ice and snow resources， and offers important reference value for improving the mechanism for realizing ecological product value. The research findings not only hold significant academic importance but also provide practical references for promoting the coordinated advancement of regional economic development and environmental protection.
The Three-River Source Region， known as “China’s Water Tower”， serves as the headwater area for the Yangtze， Yellow， and Lancang rivers. Snow cover changes in this region exert a critical influence on downstream water availability， ecosystem stability， and regional climate feedbacks. Although previous studies have analyzed variations in snow depth or snow cover area using ground-based observations or single-source remote sensing products， a systematic assessment of snow water equivalent （SWE）—a direct indicator of water storage in snowpack—remains scarce， particularly regarding its long-term （over 40 years） spatiotemporal evolution and the quantitative attribution of climatic drivers across different seasons. To fill this gap， this study aims to： （1） characterize the macroscale spatial pattern and seasonal distribution of SWE in the Three-River Source Region； （2） analyze its interannual trends and spatial heterogeneity； and （3） quantitatively identify the relative contributions of temperature and precipitation to SWE variations and the dominant factors in different seasons. This study utilized a daily SWE product over China with a resolution of 25 km from 1980 to 2020， generated by a spatiotemporal XGBoost fusion model integrating multiple existing SWE datasets with topographic and temporal covariates. This product was validated against 647 in situ stations， yielding cross-validation accuracies of r=0.77 （random） and r=0.70 （year-based）， with MAE=7.54 mm （random）， 8.62 mm （year-based） and RMSE=12.29 mm （random）， 13.73 mm （year-based）， respectively. Meteorological forcing data were obtained from the China Meteorological Forcing Dataset （CMFD v2.0， 0.1° resolution）， including near-surface temperature， precipitation， wind speed， and radiation. All data were resampled to a unified 25 km grid. The analytical methods included standardized multivariate linear regression （to derive beta coefficients for comparing the relative importance of temperature and precipitation）， partial R 2 decomposition （to quantify the independent contribution of each predictor）， and linear trend estimation with significance testing （t-test， P<0.05）. The results revealed the following key findings. （1） Intra-annual SWE exhibited a pronounced unimodal distribution， characterized by a four-phase seasonal evolution： “winter accumulation—spring melt—summer depletion—autumn recharge”. The peak occurred in January （average SWE>1.5 mm in high-altitude areas）， while summer values dropped below 0.02 mm， indicating that snowpack was almost completely melted. （2） Spatially， high SWE values were consistently distributed on the northern slopes of the Tanggula Mountains， the western segment of the Bayan Har Mountains， and the Lancang River source area， all above 4 500 m a.s.l.， with local maxima exceeding 3.0 mm. In contrast， low values were confined to low-elevation areas （e.g.， the Tongtian River valley and the Zhaling-Eling Lake basin）. SWE increased markedly with elevation and tended to stabilize above 5 000 m. （3） Over the 41‑year period， the annual average SWE showed a weak and statistically insignificant increasing trend （slope=0.001 mm⋅a-1， P>0.05）， yet with strong interannual variability. The largest interannual amplitude occurred in winter （with a maximum difference of up to 2.5 mm）， with an exceptional peak in 1985—1986 （>2.0 mm）， driven by concurrent above‑average precipitation and below‑average temperature， which favored snowfall over rainfall. Decadally， SWE was relatively low in the 1980s （except for 1985—1986）， followed by higher values from the late 1990s to early 2000s， and exhibited fluctuations after 2010. （4） The spatial trend pattern exhibited marked east-west heterogeneity. Significant increasing trends dominated the eastern Yellow River source area and high-elevation areas， particularly in winter and spring， whereas decreasing trends prevailed in the western valley and adjacent regions， especially in summer and autumn. This indicated that the region was experiencing a spatially differentiated seasonal trend pattern characterized by winter-spring increases in the east and summer-autumn decreases in the west， highlighting asymmetric seasonal responses to ongoing climate warming. （5） Standardized regression coefficients demonstrated that temperature was the dominant controlling factor for SWE in all seasons except spring. The negative temperature effect was strongest in autumn （Beta=-0.63）， followed by winter （-0.31） and summer （-0.26）， indicating that autumn was the most sensitive season for snowpack to warming—where a slight temperature rise can shift precipitation phase from solid to liquid， drastically reducing snow accumulation. In spring， however， precipitation exerted a greater positive influence （Beta=0.35） than temperature （Beta=-0.15）， indicating that spring snowpack was primarily governed by snowfall supply rather than melt. The combined temperature-precipitation model explained 51.7% of SWE variations in autumn （R 2=0.517）， but much less in summer （0.072） and winter （0.092）， suggesting that other factors （e.g.， wind redistribution， sublimation， or radiation） may play non‑negligible roles in those seasons. These findings have several important implications. First， the pronounced seasonal shift in SWE—increasing in cold seasons but decreasing in warm seasons—indicates that the timing and magnitude of snowmelt runoff are undergoing changes， which may intensify spring flood risks and reduce summer baseflow， posing challenges to water resource management in downstream basins. Second， the spatially uneven trend of “increase in the east and decrease in the west” calls for region-specific adaptation strategies. The western decreasing zone requires intensified long-term monitoring， while the eastern increasing zone needs attention to potential eco-hydrological feedbacks， such as vegetation greening or frozen soil degradation. Third， the dominant roles of autumn temperature and spring precipitation offer clear guidance for improving snow hydrological models. Future projections should prioritize accurate representations of autumn phase transition processes and spring snowfall variability. The fusion-based SWE dataset and the quantitative attribution framework established in this study provide a robust scientific basis for operational snowmelt runoff forecasting， ecological protection， and climate change impact assessment in the Three-River Source Region.
Accurate daily runoff prediction in alpine， data-scarce basins remains a challenging issue in hydrological modeling due to the strong nonlinear interactions among precipitation， temperature-driven snow and glacier melt， hydrothermal processes in the permafrost active layer， and complex topographic controls. This results in highly nonlinear runoff-generation mechanisms characterized by strong time lags and multiscale coupling. The source region of the Yangtze River， located in the hinterland of the Qinghai-Xizang Plateau， is a typical alpine basin characterized by a fragile climatic environment， complex topography， and limited in-situ observations. In the context of climate change， the uncertainty in runoff process has further increased. These characteristics make it difficult for conventional process-based or empirical hydrological models to achieve stable predictive performance across multiple temporal scales. To address these challenges， this study proposed a hybrid deep learning model based on a serial combination of long short-term memory （LSTM） and gated recurrent unit （GRU） architecture （LSTM-GRU） for daily runoff prediction in alpine basins. To improve model generalization ability and reduce the sensitivity to manually tuned hyperparameters， two optimization strategies were incorporated， namely particle swarm optimization （PSO） and Bayesian optimization algorithm （BOA）. These methods were used to automatically tune key hyperparameters， including learning rate and the number of hidden-layer neurons， thereby improving both convergence efficiency and prediction stability. Three different input configurations were designed to systematically analyze the effects of different hydro-meteorological variables on runoff prediction performance： （1） precipitation-only （P）， （2） precipitation-runoff （PR）， and （3） precipitation-temperature-runoff （PTR）. Historical runoff was incorporated to enhance the model’s representation of basin storage and release processes， while temperature was introduced to characterize energy-driven hydrological processes such as snow accumulation and melt. This improved model’s ability to represent the complex runoff generation and concentration mechanisms in alpine basins. Meanwhile， a sliding-window approach with window lengths ranging from 6 to 35 days was employed to construct time-series samples， allowing the model to capture hydrological response characteristics at different temporal scales. Forecast lead times of 1， 3， 5， and 7 days were considered to evaluate short-， medium-， and relatively long-term prediction performance. The model was trained and validated using daily hydro-meteorological observations from 2006 to 2019 in the source region of Yangtze River. The dataset was divided chronologically into training and testing periods to ensure temporal consistency. Model performance was assessed using four widely adopted indicators： Nash-Sutcliffe efficiency coefficient （NSE）， relative error （RE）， root mean square error （RMSE）， and mean absolute error （MAE）， enabling a comprehensive evaluation of model accuracy and stability. The results indicated that the proposed LSTM-GRU model consistently outperformed the single LSTM baseline， demonstrating the advantage of combining long-term memory representation with gated feature refinement. Among the three input scenarios， PTR achieved the best overall performance， followed by PR and P， indicating that temperature information played a critical role in improving runoff simulation in alpine basins. This improvement could be attributed to the enhanced representation of energy-driven hydrological processes， particularly snow and ice melt contributions， which were not captured when only precipitation was used. In addition， both PSO and BOA significantly improved model performance compared with the non-optimized configuration. However， BOA showed superior generalization ability and more stable performance across all forecast horizons， especially for medium- and long-term predictions. This advantage was mainly due to the probabilistic surrogate modeling mechanism of Bayesian optimization， which efficiently balanced exploration and exploitation in the hyperparameter search space and reduced the risk of suboptimal convergence associated with heuristic search methods such as PSO. Furthermore， model performance decreased with increasing forecast lead time， reflecting increasing uncertainty in hydrological processes at longer temporal scales. Nevertheless， the relative ranking among input scenarios remained consistent across all forecast lead times， indicating that incorporating temperature and historical runoff information provided stable improvements in predictive skill. The results also suggested that precipitation alone was insufficient to represent the complex runoff generation mechanisms in alpine basins， where cryospheric processes played a dominant role. Overall， this study demonstrates that integrating an LSTM-GRU hybrid deep learning architecture with Bayesian optimization provides an effective and robust framework for daily runoff prediction in alpine， data-scarce environments. The proposed approach not only improves prediction accuracy and stability across multiple forecast horizons but also provides a transferable modeling strategy for hydrological forecasting and water resource management in similar alpine basins under changing climate conditions. Future research should consider incorporating additional cryospheric variables and physical constraints to further enhance model interpretability and long-term predictive reliability.
Digital twin technology is profoundly reshaping the disaster management paradigm within the framework of Earth system science. Under sustained global warming， the cryosphere has retreated extensively， characterized by pronounced glacier mass loss， snow cover reduction， and permafrost degradation. As a core region of the global cryosphere， the Qinghai-Xizang Plateau is one of the areas with the most concentrated and prominent cryospheric disasters， where ice-rock avalanches， snow avalanches， glacial lake outburst floods， and glacier debris flows occur frequently. These hazards have pronounced cascading effects and are characterized by multiple causative factors， multi-scale dynamics， and sudden onset， posing severe challenges to disaster monitoring， risk assessment， prevention， and control. However， existing research and disaster prevention practice still face several problems. Multi-source heterogeneous data are scattered across different agencies with inconsistent standards； risk assessment mostly relies on single-hazard analyses and static indicators； and emergency decision support remains dominated by post-hoc analysis and static visualization. More importantly， a digital twin system framework and preliminary implementation designed specifically for multi-hazard， multi-process， and multi-scale cryospheric disasters on the Qinghai-Xizang Plateau are still lacking. To address these gaps， this study employed digital twin technology and conducted systematic research at two levels： conceptual framework construction and preliminary practical implementation. At the theoretical level， targeting the differentiated needs of researchers， practitioners， and the general public， this study proposed a five-layer conceptual framework comprising the data layer， service layer， model layer， application layer， and presentation layer. The data layer achieves unified integration and quality control of multi-source data from remote sensing， unmanned aerial vehicle surveys， and ground observations. The service layer encapsulates data services and model capabilities into standardized interfaces. The model layer integrates physical mechanism models and AI surrogate models as the scientific computing core. The application layer comprises five modules： disaster database management， disaster process simulation， risk assessment， decision support， and system management. The presentation layer provides a WebGIS-based visualization interface. Four design principles were established： layered decoupling， data-driven operation， model extensibility， and intuitive interaction. A perception-synchronization-feedback digital twin closed-loop mechanism was embedded， providing a technical pathway for dynamic mapping between physical and digital spaces. At the practical level， a standardized cryospheric disaster database was constructed， integrating six categories of core datasets： basic geographic information， meteorological forcing data， hydrological station measurements， socio-economic data， multi-hazard exposure data， and historical disaster records. A hybrid storage architecture combining SQLite/SpatiaLite， MySQL， and MongoDB was adopted to support spatial queries， transactional management， and high-frequency writes. Based on this， engineering prototypes of the core modules were then developed， integrating the long short-term memory （LSTM） runoff model and the probabilistic inundation model. These prototypes implemented the complete technical workflow from data input and model configuration to computation execution and result output， and achieved the functional integration of the disaster database management and flood process simulation modules. Taking the Yarkant River Basin on the northwestern margin of the Qinghai-Xizang Plateau as the validation area， this study evaluated an LSTM model using measured daily runoff from the Kaqun hydrological station during 1956—2000， with ERA5-Land reanalysis data as meteorological forcing. The training period was 1956—1990， and the testing period was 1991—2000. The model achieved a Nash-Sutcliffe efficiency coefficient （NSE） of 0.79， a coefficient of determination （R 2） of 0.79， a root mean square error （RMSE） of 6.93 m3⋅s-1， and a percent bias （PBIAS） of 0.07% during training； and an NSE of 0.91， an R 2 of 0.91， an RMSE of 4.28 m3⋅s-1， and a PBIAS of -3.86% during testing， indicating that the model effectively captured the hydrological characteristics of the river basin. The trained model was then applied to the entire Qinghai-Xizang Plateau to reconstruct daily gridded runoff data for 1956—2000. Combined with a probabilistic inundation model based on spatialized percentile threshold fields （the 10th， 50th， 90th， 95th， and 99th percentiles）， the system automatically identified grids where runoff reached or exceeded the selected threshold as potential inundation areas. This approach enabled multi-scenario discrimination ranging from high-frequency inundation to extreme events without relying on stage-discharge relationships， making it suitable for data-scarce plateau regions. The results preliminarily verified the technical feasibility of the proposed framework in terms of core data flow and modeling chain， providing an extensible architecture for the digital management of cryospheric disasters on the Qinghai-Xizang Plateau. However， current validation is limited to a single flood hazard type using historical data for offline testing. Real-time data assimilation， multi-hazard model coupling， and a complete digital twin closed-loop mechanism remain to be implemented. Future studies will focus on improving the real-time closed-loop mechanism， developing and coupling multi-hazard process models for snow cover， glaciers， and frozen ground， establishing real-time data links with operational departments， and quantitatively evaluating system performance. These efforts aim to advance the system from a conceptual prototype to operational applications， thereby providing scientific and technological support for disaster prevention and mitigation， as well as emergency management on the Qinghai-Xizang Plateau.
Against the backdrop of global warming， the frequency and intensity of extreme snowfall events on the Qinghai-Xizang Plateau show an increasing trend， posing a serious threat to livestock production and pastoral livelihoods in the Three-River Source Region. Accurately projecting the spatiotemporal evolution characteristics of snow disaster risk under different future climate scenarios is of great scientific value and practical significance for regional disaster prevention and mitigation， climate change adaptation， and the sustainable development of pastoral areas. Based on CMIP6 climate model data， this study constructed a quantitative snow disaster risk projection model using a weighted comprehensive evaluation method incorporating three dimensions： hazard of disaster-causing factors， sensitivity of the disaster-prone environment， and exposure of disaster-affected bodies， to project snow disaster risk in the Three-River Source Region from 2025 to 2100. The future climate projection data were mainly derived from the RCM_BCC dataset released by the National Climate Center， which was based on the BCC-CSM2-MR model from CMIP6 and downscaled using the RegCM4.4 regional climate model at a spatial resolution of 25 km. Future population and socioeconomic projection data were derived from the gridded population and economic dataset of Shared Socioeconomic Pathways （SSPs） released by Science Data Bank at a spatial resolution of 0.5°. In the risk projection model， the hazard of disaster-causing factors comprised winter minimum temperature and precipitation， which were closely related to the formation and development of snow disasters. Precipitation determined snow depth， while minimum temperature influenced snow duration. Under low-temperature conditions， greater precipitation led to deeper snow cover and more severe grassland burial. The sensitivity of the disaster-prone environment included elevation and slope aspect， affecting the development and persistence of snow disasters. Higher elevations had lower temperatures and longer-lasting snow cover， while shady slopes received less solar radiation than sunny slopes and were more prone to snow disaster development. Exposure comprised population and GDP that were directly affected by snow disasters. Population represented the primary disaster-affected body， while GDP reflected both potential direct economic losses caused by snow disasters and the status of pastoral infrastructure. Indicator weights were determined using the entropy weight method. The results showed that： （1） under both SSP2-4.5 and SSP5-8.5 scenarios， winter minimum temperature and precipitation in the region exhibited increasing trends， with more significant increases in temperature and snowfall under SSP5-8.5. （2） Yushu， Nangqian， and southern Golog were identified as high-risk zones for snow disasters. Under SSP5-8.5， high-risk zones accounted for 14.45% of the total area， an increase of nearly 5 percentage points compared to SSP2-4.5. （3） The GDP and population potentially affected by high-risk zones increased significantly with the increase in emission scenario intensity and over time. By 2100， the affected GDP under SSP5-8.5 could reach 22.686 billion yuan. Although the future climate in the Three-River Source Region shows an overall “warming and wetting” trend， the spatial extent of high-risk snow disaster zones and their potential impact on the regional socio-economic system will still expand significantly， particularly under the SSP5-8.5 high-emission scenario. This study reveals the evolution characteristics of snow disaster risk under different future climate scenarios， thereby providing a scientific basis for snow disaster prevention and decision-making support for addressing climate change risks in the Three-River Source Region.
Permafrost cryostratigraphy embodies the long-term coupling among surface thermal boundary conditions， ground ice phase change， geothermal heat flow， and near-surface hydrothermal processes. Its vertical cryostratigraphic configuration modulates land surface energy partitioning， hydrological cycles， and mass transport in cold regions， thereby governing biogeochemical cycling， landscape evolution， and the stability of engineering infrastructure. Traditionally， regional and global land surface models represent permafrost as a simple binary system consisting of an active layer overlying the permafrost table. Such an oversimplification fails to capture the cryostratigraphic heterogeneity produced by interannual freeze-thaw cycles， nor does it adequately resolve the transient thermal states prevailing in marginal permafrost regions. Crucially， we conceptualize this paradigm shift of permafrost cryostratigraphic configurations from the conventional binary model to a vertically heterogeneous， multi-layered framework comprising the active layer， transition zone， suprapermafrost subaerial talik （SST）， permafrost body， relict permafrost， and two-layer permafrost. This review systematically classifies the cryostratigraphic components based on their genetic and evolutionary mechanisms， focusing on freeze-thaw front dynamics， interannual variations in active layer thickness， the evolutionary trajectory of the transition zone， and the formation and vertical development of the SST. This process-based perspective demonstrates that interfaces within a permafrost profile are highly dynamic rather than static stratigraphic boundaries. Their migration is driven by the local surface energy budget modulated by snow cover and vegetation， along with their feedback interactions with soil moisture， ground ice content， lithology， and deep geothermal heat flux. Furthermore， this review establishes the mapping relationships between these cryostratigraphic configurations and the vertical distribution and seasonal dynamics of ground temperature-depth profiles. The geometry of the annual ground temperature envelope， the depths of seasonal freezing and thawing， and the intersection characteristics of temperature profiles with the 0 ℃ isotherm provide a physically interpretable basis for distinguishing profile configurations. A conventional attached active-layer-permafrost configuration is marked by a seasonally thawed layer directly overlying a persistent frozen body. In contrast， a profile featuring a transition zone， which preserves ice-rich or intermittently frozen material immediately beneath the active layer， reflects the “thermal legacy” of historical climate anomalies buffered by the latent heat of ground ice. When winter cooling fails to completely refreeze the active layer down to the permafrost table， an SST develops， thermally and hydraulically decoupling the seasonally frozen layer from the underlying permafrost. Continued warming or local thermal disturbance expands this unfrozen zone， promoting permafrost fragmentation and leaving isolated relict permafrost bodies， whereas epigenetic refreezing can generate complex， superimposed two-layer configurations. Therefore， these configurations delineate distinct evolutionary stages and nonlinear response pathways of permafrost adjustment rather than a single linear degradation sequence. Because non-equilibrium thermal states and phase-change thermal lag allow similar near-surface ground temperatures to mask highly divergent deep permafrost conditions， accurate identification requires integrating high-resolution， long-term borehole thermometry with cryostratigraphic observations and geophysical imaging， thereby avoiding reliance solely on active-layer thickness. Permafrost cryostratigraphic configuration not only responds to climate change but also actively modulates Earth system feedbacks. During ground ice melting， latent heat consumption acts as a thermal buffer， slowing the increase in ground temperature despite a contraction of the frozen ground volume， thereby significantly prolonging the zero-curtain effect. Consequently， the apparent thermal stability represented by quasi-isothermal conditions near 0 ℃ cannot be inferred solely from ground temperature trends. Progressive thaw in deeper strata mobilizes deep organic carbon pools for microbial decomposition， whereas talik development and coalescence establish hydraulic windows connecting surface water， shallow groundwater， and subpermafrost aquifers. This enhanced hydraulic connectivity shifts the subsurface heat transfer regime from pure conduction to coupled convective-advective transport， which is poorly resolved by conventional homogeneous heat-conduction models. Furthermore， cryogenic structures， excess ice， sedimentary sequences， and stable hydrogen and oxygen isotope （δ 18O and δD） signatures preserve paleoclimatic records of historical permafrost aggradation and degradation. This allows contemporary profile geometries to be interpreted as cryostratigraphic archives rather than mere indicators of transient thermal states. A deeper understanding of these dynamics requires multidisciplinary observations and modeling frameworks that explicitly resolve vertical cryostratigraphic heterogeneity. This review proposes key research priorities， including： （1） establishing an integrated multi-source， multi-scale monitoring system encompassing borehole thermometry， ground-penetrating radar （GPR）， electrical resistivity tomography （ERT）， InSAR-based land surface deformation inversion， and hydrological isotope tracing； （2） developing physics-informed machine learning （PIML） models that explicitly couple moving phase-change boundaries and mass transport； （3） identifying bifurcation thresholds where nonlinear hydrothermal feedbacks trigger abrupt stratigraphic transitions； and （4） quantitatively assessing cascading feedbacks among carbon cycling， hydrological processes， ecological succession， and geomorphic evolution. In cold-region engineering， this framework offers a mechanistic basis for assessing foundation stability through accurate identification of ice-rich horizons， talik development， and differential thaw settlement. In conclusion， reframing permafrost cryostratigraphy from a static classification to an observable， process-based， and predictable sequence of multi-layered configurations establishes a comprehensive cryostratigraphic framework for assessing permafrost evolution， ecological feedbacks， and engineering risks in the context of climate change.
The Qinghai-Xizang Plateau is one of the regions in China where permafrost is highly concentrated and major cold-region engineering projects are intensively developed. Permafrost not only controls hydrothermal processes in the active layer and the spatial patterns of alpine ecosystems but also provides a critical foundation for maintaining the stability of cold-region engineering infrastructure. Under the combined effects of climate warming and engineering-induced thermal disturbance， improperly implemented roads， railways， pipelines， mines， and other engineering projects may alter surface albedo， specific heat capacity， and surface energy balance through vegetation destruction， topsoil stripping， mechanical trampling， excavation and backfilling， and the thermal effects of engineering structures. These disturbances can disrupt active-layer freeze-thaw processes and shallow hydrological connectivity， resulting in a downward shift of the permafrost table， active-layer thickening， ground-ice thawing， thermokarst subsidence， and reduced soil bearing capacity. They can further cause soil desiccation， texture coarsening， compaction， reduced pore connectivity， nutrient loss， and changes in soil biological communities. Focusing on the ecological status and restoration practices in engineering-disturbed permafrost regions of the Qinghai-Xizang Plateau， this study systematically reviews the effects of engineering construction and operation on site conditions in disturbed regions， including soil texture， hydrological processes， nutrient status， and soil biota. It summarizes the characteristics of changes in vegetation community structure， species diversity， and biomass； analyzes the feedback effects of vegetation degradation on the permafrost environment and engineering stability； and compares the applicable conditions， restoration effectiveness， and major limitations of existing ecological restoration techniques. Existing studies collectively indicate that engineering disturbance reshapes hydrothermal， nutrient， and biological processes in disturbed regions， making vegetation recovery constrained by multiple factors， including soil substrate conditions， shallow soil moisture， permafrost thermal state， and the native propagule pool. Vegetation in disturbed regions generally exhibits retrogressive community succession， decreases in hygrophilous and mesophytic plants， increases in cold-， drought-， and infertile-soil-tolerant species， and declines in biomass and diversity. Vegetation destruction weakens the shading and thermal insulation functions of the canopy， turf layer， and litter layer； reduces the shear resistance and soil-reinforcing capacity of the root-soil complex； and intensifies the risks of wind erosion， water erosion， thaw settlement， thermally induced landslides， and slope instability. These processes form a feedback chain of “engineering disturbance-vegetation degradation-permafrost degradation-engineering distress.” Current techniques， including spray seeding with borrowed soil， vegetation bags， turf transplantation， fertilization and soil amendments， biochar application， microbial inoculation， microtopographic reconstruction， permafrost-layer reconstruction， backfilling with artificially frozen soil blocks， grass mulching， and three-dimensional vegetation soil-reinforcement nets， can improve surface substrate conditions， increase vegetation cover， promote local water and nutrient accumulation， and alleviate degradation in disturbed regions to some extent. However， several problems remain， including excessive emphasis on short-term revegetation rather than permafrost thermal stability， reliance on single restoration techniques rather than engineering-ecological coordination， insufficient adaptation to ecological zoning， inadequate configuration of native provenances and functional traits， and weak long-term monitoring and quantitative evaluation of restoration effectiveness. On this basis， this study argues that ecological restoration in engineering-disturbed permafrost regions of the Qinghai-Xizang Plateau should take the maintenance or recovery of permafrost thermal stability as its core objective. Restoration should follow the basic principles of zoned and graded management， low secondary disturbance， and environmental safety. It should be based on native species screening， complementary configuration of functional traits， construction of habitat heterogeneity， and restoration of community diversity. Soil amendment， vegetation reconstruction， active-layer thermal regulation， and engineering protection measures should be integrated to establish a long-term， multi-indicator monitoring and comprehensive evaluation system covering the soil temperature field， active-layer thickness， ground-surface deformation， soil moisture and nutrients， vegetation functional-group composition， community recovery trajectory， and engineering stability. This framework can provide guidance for effective ecological restoration， maintenance of engineering safety， and improvement of cold-region engineering resilience under climate change in engineering-disturbed permafrost regions of the Qinghai-Xizang Plateau.
Pingos， as key periglacial landforms in permafrost regions， profoundly record groundwater thermal dynamics， paleoclimate evolution， and hydrological cycle characteristics， serving as core carriers for deciphering the permafrost-hydrology-climate coupling system in cold regions. Current research on pingos has predominantly focused on macro-scale distribution description and single-point recharge source tracing， inferring their formation mechanisms based on morphology， while lacking empirical support from geochemical methods such as stable isotopes. Particularly， the understanding of how microtopographic differences between sunny and shady slopes—by altering surface radiation and freeze-thaw cycle intensity—regulate pingo system openness， freezing processes， and ground ice accumulation mechanisms remains insufficient. Addressing this research gap， this study targeted a representative pingo in the source region of the Yellow River on the Qinghai-Xizang Plateau （34°55′53.64″ N， 98°31′23.34″ E） to achieve three core objectives： accurately trace the multiple recharge sources of ground ice， reveal the regulatory effect of topographic differences between sunny and shady slopes on pingo formation， and clarify the ground ice formation mechanism. The findings aim to provide a scientific basis for projecting permafrost changes， simulating regional water cycles， and protecting cold-region engineering safety under a warming climate. The study area is located in the core zone of the source region of the Yellow River National Nature Reserve on the Qinghai-Xizang Plateau， with an average elevation of 4 200 m. It features a cold alpine semi-arid climate， with an annual average temperature of -4 ℃ and annual precipitation of 418 mm. The region is a coexistence zone of seasonally frozen ground， isolated permafrost， and discontinuous large-area permafrost， dominated by warm permafrost with a mean annual ground temperature （MAGT） mostly above -1.0 ℃ and a thickness generally less than 40 m， showing a degradation trend. The study site is situated in the Duogerong Basin， a well-developed area of large pingo clusters. The target pingo has a height of approximately 7.6 m， with continuous ice layers exceeding 2 m in thickness and ice-rich sediments with a volumetric water content of over 50%， demonstrating typical research representativeness. To achieve the research goals， three independent boreholes were drilled at the summit， sunny slope， and shady slope of the pingo. A total of 81 ground ice samples were collected through systematic drilling， core logging， and sampling （18 from the summit， 41 from the sunny slope， and 22 from the shady slope）. The first occurrence of ground ice was at 18.8 m （summit）， 4.5 m （sunny slope）， and 1.7 m （shady slope）， with no ice detected below 33.1 m （summit）， 28.4 m （sunny slope）， and 32.8 m （shady slope）. After natural thawing and filtration through a 0.22 μm membrane， the samples were analyzed for stable isotopes （δ¹⁸O， δD） using a high-precision water isotope analyzer （Picarro L2130）， and d-excess values were calculated. A unified elevation datum （taking the surface of the summit borehole as 0 m） was established for cross-borehole comparative analysis. Combined with vertical isotope profile characteristics and statistical models， the hydrothermal processes and formation mechanism of the pingo were systematically deciphered. The results showed that the isotopic composition of the pingo’s ground ice significantly deviates from the Global Meteoric Water Line， characterized by low freezing line slopes （2.69~2.98） and extremely negative d-excess values （mean values of -44.92‰， -27.70‰， and -46.75‰ for the summit， sunny slope， and shady slope， respectively， with a minimum of -91.99‰）， indicating intense evaporation and kinetic fractionation of the water body. Moisture source analysis revealed that the ground ice is primarily derived from confined groundwater （initial water： δ¹⁸O=-13.54‰， δD=-100.93‰， which is highly consistent with the isotopic characteristics of confined groundwater）. Distinct isotopic abrupt layers at specific depths （30.5 m at the summit， 15.2 m at the sunny slope， 21.5 m at the shady slope） recorded signals of recharge from strongly evaporated paleo-lake water （average： δ¹⁸O=5.17‰， δD=-47.27‰， d-excess=-88.71‰）， while shallow ice bodies are mainly contributed by precipitation infiltration， confirming the multiple moisture sources of the pingo as “confined groundwater-dominated + paleo-lake water secondary modification + shallow precipitation recharge”. The topographic difference between sunny and shady slopes exerted a significant regulatory effect on the freezing mechanism： the sunny slope， affected by intense solar radiation， experienced frequent shallow freeze-thaw cycles， resulting in erratic isotopic signals and open-system evaporation enrichment in the deep layer （d-excess decreasing to -86.23‰）； the cool environment of the shady slope inhibited disturbance， leading to a clear Rayleigh fractionation trend in mid-deep （17.5~27 m） ground ice （δ¹⁸O decrease >8.5‰） and a 2~3 m lag in the downward migration of the freezing front compared to the sunny slope. The ground ice formation process in all three boreholes can be divided into four stages： precipitation-dominated surficial modification stage （0~17.5 m）， differential ground ice growth stage （17.5~27 m）， thermal gradient-dominated freezing front differentiation stage （27~32.5 m）， and isotopic fluctuation stage induced by water source disturbance （>32.5 m）. Isotopic trends in different stages reflect ice growth differentiation under the combined action of climate drivers and slope-aspect thermal differences. Notably， both open and closed system characteristics coexist on different slopes of the same pingo， with the lower limit of ground ice on the sunny slope （31.4 m） elevated by 3.4 m compared to the shady slope （34.8 m）， confirming that thermal differentiation between sunny and shady slopes achieves differentiated regulation of ice core growth by altering permafrost thermal stability and water phase transition paths. This study reveals the key controlling role of thermal differentiation between sunny and shady slopes in pingo formation from a multi-location isotope profile perspective， breaking through the single-point limitation and macro-scale perspective of traditional pingo research. It provides empirical support for the pingo formation mechanism at the stable isotope level and deepens the understanding of permafrost-hydrology-climate coupling processes in alpine regions. The identified multiple paths of pingo moisture evolution in extreme environments offer an important scientific basis for projecting permafrost changes and simulating regional water cycles under climate warming. From an engineering application perspective， the shallower lower limit of ground ice and active hydrothermal processes on the sunny slope indicate a higher risk of thermal instability under warming scenarios， providing targeted references for disaster early warning and protection of linear projects in cold regions such as the Qinghai-Xizang highway and railway. It is recommended to establish an automated monitoring system to track the dynamic changes of pingos in real time and ensure long-term engineering stability.
Permafrost degradation induced by climate warming and human engineering activities severely constrains the long-term stability of road infrastructure in high-latitude permafrost regions. The Greater Khingan Mountains represent the only high-latitude and low-altitude permafrost region in China and contain Xing’an-Baikal-type permafrost at the southern margin of the circum-Arctic permafrost. This region is characterized by high mean annual ground temperature and low thermal stability. Continuous permafrost degradation has triggered frequent thaw settlement， frost heave， cracking， and uneven subgrade deformation. Existing studies on permafrost thawing mechanisms and subgrade distress mainly focus on macroscopic parameters such as temperature fields and water content， and most laboratory tests employ remolded soil samples. However， the preparation of remolded samples destroys the original structure， preventing them from accurately preserving the ice occurrence characteristics， cryostructure， and pore structure of undisturbed frozen soil， which may lead to biased test results. Moreover， the internal relationships among ice morphology， pore structure， and the thawing response of frozen soil remain unclear， and systematic research on the thaw drainage， water retention characteristics， and deformation evolution patterns of frozen soil with different cryostructures is still insufficient. To address this research gap， this study investigated undisturbed frozen clay collected from the permafrost region of the Greater Khingan Mountains. Compared with conventional tests using remolded soil， undisturbed samples can completely preserve the original cryostructure and ice occurrence characteristics of frozen soil. Three typical cryostructures of frozen soil—massive， fibrous， and laminated—were quantitatively identified through binary image analysis. Ice morphology， area proportion， and spatial connectivity were adopted as core indicators to establish a standardized classification system for frozen-soil cryostructures. A series of laboratory self-weight thawing tests were conducted at a constant ambient temperature of （20±1） ℃， and a total of 32 groups of undisturbed frozen soil samples with different total water contents were tested. By defining two quantitative indicators， namely the vertical change rate （VCR） and lateral change rate （LCR）， this study systematically investigated the drainage characteristics， water retention behavior， and differential response mechanisms of bidirectional deformation during the thawing process of frozen soil with three types of cryostructures.The test results indicated that total water content was the primary factor governing the thaw drainage rate of frozen soil， whereas cryostructure morphology exerted a critical differential control on water retention performance. The thawing stabilization time and drainage rate exhibited increasing trends with increasing total water content for three types of frozen soil. Under similar water content conditions， the drainage capacity of frozen soil followed the order of laminated>fibrous>massive， and the peak drainage rate of laminated frozen soil was much higher than those of the other two types. Massive frozen soil exhibited the strongest water retention capacity， and low-water-content samples retained all thaw water after thawing within the soil pores. In contrast， the water retention capacity of laminated frozen soil decreased with increasing total water content. The underlying mechanism was that ice-layer freezing under high-ice-content conditions markedly increased the compactness of soil around fissures， compressed soil water storage space， and thereby weakened the soil water retention capacity. Significant differences were observed in the thaw deformation characteristics of frozen soil with different cryostructures. Low-water-content massive frozen soil exhibited an unusual thaw expansion phenomenon in both vertical and lateral directions after thawing， without settlement deformation. Fibrous and laminated frozen soil exhibited only vertical settlement deformation after thawing， without expansion. Massive frozen soil was dominated by lateral deformation under high drainage conditions， whereas fibrous and laminated frozen soil was dominated by vertical thaw settlement deformation. The thawing process of frozen soil could be divided into three stages： the sedimentation stage， the particle contact stage， and the consolidation stage. The deformation of laminated and fibrous frozen soil mainly occurred during the early sedimentation stage， while the deformation of massive frozen soil was concentrated in the middle and late particle contact and consolidation stages. Moreover， after the completion of thaw drainage， massive frozen soil exhibited greater residual consolidation deformation and poorer late-stage deformation stability. This study systematically reveals the control mechanism of cryostructure morphology on the thawing hydro-mechanical response of frozen clay， and clarifies the internal relationships among ice occurrence state， pore structure， thaw drainage， and deformation characteristics of frozen soil. The results explain the fundamental causes of differential subgrade distress in the permafrost region of Greater Khingan Mountains from a microstructural perspective. The findings improve the theoretical framework of the thawing response of undisturbed frozen soil with different cryostructures. They provide important theoretical support for the optimization of drainage systems， subgrade structure design， and distress prevention and control in highway projects in high-latitude permafrost regions， and offer a scientific reference for the long-term stable operation and maintenance of cold-region geotechnical engineering and infrastructure under ongoing climate warming.
The shear behavior of frozen soil-structure interfaces is closely associated with the stability and long-term performance of engineering structures in cold regions. Although previous studies have extensively investigated the macroscopic shear response of such interfaces， the evolution of the interface shear band at the particle scale remains insufficiently understood. Moreover， the effects of temperature， strain rate， and interface roughness have generally been examined separately， and their respective roles in controlling interface strength， particle movement， and shear localization have not been systematically quantified. To address these limitations， this study investigated the evolution of the shear band at frozen soil-structure interfaces under the effects of temperature， strain rate， and interface roughness， with particular emphasis on the relationship among interparticle ice bonding， interface shear strength， particle displacement， and shear band geometry. Laboratory tests were first conducted to characterize the mechanical behavior of pure ice under different temperature and strain rate conditions. Based on the experimental results， quantitative relationships were established between the strength and modulus of ice and the corresponding temperature and strain rate. Ice was subsequently represented as the bonding medium between frozen soil particles， and a discrete element contact model incorporating the interparticle ice-bonding effect was developed. In this model， the mechanical properties of the ice bonds were determined using the experimentally derived temperature- and strain-rate-dependent relationships， thereby establishing a direct connection between the mechanical response of pure ice and the particle-scale bonding behavior of frozen soil. The proposed model was then used to simulate the direct shear process of a frozen soil-structure interface. Its reliability was evaluated by comparing the numerical results with previously reported direct shear test results for frozen soil-structure interfaces. On this basis， a series of numerical direct shear simulations was performed using a controlled-variable approach to investigate the effects of temperature， strain rate， and interface roughness on peak shear strength， void ratio evolution， particle displacement， and shear band development. The results indicated that temperature， strain rate， and interface roughness all affected the shear response of the frozen soil-structure interface， although their relative contributions and underlying mechanisms differed considerably. A decrease in temperature and an increase in strain rate enhanced the strength of the ice bonds between soil particles， leading to an increase in the peak interface shear strength. Within the investigated range， interface roughness exerted a comparatively smaller influence on the peak shear strength. Nevertheless， interface roughness had a pronounced effect on particle movement and the spatial development of the shear band. During the shearing process， the overall void ratio of the specimen initially increased and subsequently decreased， reflecting a transition from shear-induced dilation to particle rearrangement and local compaction. Changes in the void ratio were primarily concentrated within the shear band， whereas the region outside the shear band exhibited relatively limited structural variation. Interface roughness and strain rate had more pronounced effects on the evolution of the void structure within the shear band than temperature. These results demonstrated that the macroscopic shear behavior of the interface was controlled not only by the strength of the interparticle ice bonds but also by particle rearrangement， interface interlocking， and localized volumetric deformation. Quantitative analysis further revealed that temperature， interface roughness， and strain rate governed shear band evolution through distinct mechanisms. When the temperature decreased from -10 ℃ to -40 ℃， the maximum particle displacement ratio increased from 1.04% to 1.99%， whereas the shear band thickness ratio decreased from 11.63% to 5.87%. This result indicated that stronger ice bonding at lower temperatures restricted shear deformation to a narrower localized zone while intensifying particle displacement within that zone. When the interface roughness increased from 0.5 to 1.5， the maximum particle displacement ratio and shear band thickness ratio increased to 2.99% and 30.84%， respectively. Greater interface roughness therefore enhanced mechanical interlocking between the frozen soil particles and the structural surface， mobilized a larger number of particles， and expanded the zone involved in shear deformation. By contrast， when the strain rate increased from 10-5 s-1 to 10-2 s-1， the maximum particle displacement ratio and shear band thickness ratio decreased to 0.95% and 7.64%， respectively. A higher strain rate reduced the time available for particle rearrangement and coordinated movement， thereby suppressing particle sliding and limiting the expansion of the shear band. The results demonstrated that temperature primarily governed the overall interface strength and the degree of shear localization by changing the strength of the interparticle ice bonds. Interface roughness mainly controlled particle interlocking， collective particle movement， and the spatial extent of the shear band， whereas strain rate influenced the time-dependent processes of particle sliding and rearrangement. The principal contribution of this study is the establishment of a quantitative link between the mechanical properties of pure ice and the bonding parameters used in discrete element simulations of frozen soil. The proposed approach provides a particle-scale interpretation of interface strength formation and shear band evolution under coupled thermal， loading-rate， and surface-morphology conditions. The findings deepen the understanding of the shear failure mechanism of frozen soil-structure interfaces and provide a theoretical basis for interface stability assessment and parameter selection in foundations， piles， linings， retaining structures， and other engineering systems constructed in cold regions.
As critical infrastructure in water conveyance and distribution systems， canals play essential roles in irrigation district development， inter-basin water transfer projects， and national water network engineering. However， canal engineering in seasonally frozen soil regions commonly faces severe challenges from frost heave damage. Slope toe anchorage is a widely used slope reinforcement measure that can effectively mitigate scouring effects and enhance the stability of canal slope lining plates. The alteration of boundary conditions at the slope toe results in frost heave characteristics for this type of canal that differ from those of three-plate spliced canals. However， corresponding frost heave analysis models are currently lacking. To address this， a mechanical model for frost heave in trapezoidal canals with slope toe anchorage was established based on the theory of elastic beams on frozen soil foundations， yielding analytical expressions for the frost heave deformation and internal forces of the slope plate. The frost heave deformations of the slope plate were calculated using both the proposed model and a finite element model， and comparisons with measured values demonstrated the applicability of the proposed method. Unlike in three-plate spliced trapezoidal canals， the frost heave deformation curve of the slope plate in a slope toe-anchored canal exhibited an inflection point near the slope toe with zero cross-sectional rotation， which was consistent with the deformation characteristics of a fixed end in engineering mechanics. Additionally， under typical working conditions without top cover plate restraint， uplift displacement （2.4 mm） occurred at the top of the slope plate， which was consistent with theoretical expectations. This indicated that debonding and void formation were likely to occur near the slope top due to the failure of freezing constraints. In engineering practice， the proposed method can be used to calculate uplift displacement for the preliminary verification of such frost damage. Finally， the influence of groundwater depth on the frost heave mechanical behavior of the canal was analyzed in detail. The results indicated that as groundwater depth decreased， uplift displacement at the top of the slope plate gradually increased. When the groundwater depth decreased from 3.6 m to 3.2 m， the uplift displacement at the slope top increased from 2.4 mm to 3.8 mm. Therefore， a reduction in groundwater depth increased the risk of debonding and void formation between the frozen soil and the lining plate. Furthermore， due to the constraint provided by slope toe anchorage， an inflection region existed in the deformation curve near the toe of the slope plate， which aligned with engineering practice. Both bending moment and shear force exhibited positive and negative extreme values. The areas near the slope toe and the middle-lower part of the slope plate were the most susceptible locations for cracking. As groundwater depth decreased， the maximum values （absolute values） of both bending moment and shear force increased. When the groundwater depth decreased from 3.6 m to 3.2 m， the absolute value of the negative bending moment at the slope toe increased by approximately 25 kN⋅m， while the maximum positive bending moment in the middle-lower section of the slope plate increased by about 7.16 kN⋅m. Bending moments in cross-sections near the slope top remained relatively small and gradually approached zero. The relationship between cross-sectional bending moment and groundwater depth followed a linear function， while the relationship between cross-sectional shear force and groundwater depth followed an exponential function. This study provides a reference for the frost heave resistance design of trapezoidal canals with flexible bottom plates and slope toe anchorage.
The Three-River Source Region， as the headwaters of the Yangtze River， Yellow River， and Lancang River， is a critical water conservation area in China. In recent decades， the cryospheric and hydrological systems of this region have undergone major changes due to significant climate warming. To systematically quantify the spatiotemporal characteristics of glacier changes and elucidate their driving mechanisms affecting runoff components， this study integrated multi-source datasets， including long-term glacier observations， remote sensing retrievals， and hydrological model outputs. The study covered the period from 1981 to 2018， with a focus on revealing the hydrological effects of glacier changes. Furthermore， the geodetector was employed to quantify the influence of key climatic and hydrological factors on the spatial heterogeneity of river runoff. The results revealed several significant trends and underlying mechanisms. （1） Significant glacier retreat and thinning： Over the past five decades， glaciers in the Three-River Source Region experienced continuous retreat and thinning. The total glacier area decreased by 433.44 km²， with a loss rate of 26.26%. The rate of retreat has accelerated notably since 2000. During the same period， the average glacier elevation dropped by 8.82 m. The regional glacier equilibrium line altitude （ELA） remained at a relatively high level， with an average elevation of 5 658 m. Additionally， the glacier mass balance remained persistently negative， with an average value of -0.46 m w.e.， indicating sustained ice mass loss. Notably， glacier retreat and ablation were particularly significant in the southwestern part of the region. （2） Profound impact on regional hydrological dynamics： From 1981 to 2018， both the total natural runoff and the glacier meltwater runoff components in the Three-River Source Region showed statistically significant increasing trends （P<0.05）. Runoff composition analysis revealed changes in contributions from different sources. Glacier runoff and rainfall runoff increased significantly， whereas snowmelt runoff exhibited a decreasing trend. （3） Driving factors identified by correlation and spatial analysis： Correlation analysis showed that rising air temperature was the primary direct driving factor of the increase in glacier meltwater runoff （mean correlation coefficient r = 0.81）. To further clarify the factors influencing runoff changes， the geodetector model was applied. This analysis revealed that precipitation was the dominant factor controlling the spatial distribution of natural runoff， with a maximum average q value of 0.78 （P<0.01）. The q statistic represented the explanatory power of a factor， and a higher value indicated a stronger influence on the observed spatial heterogeneity. Glacier runoff was identified as the secondary influencing factor， followed by air temperature and downward shortwave radiation flux. This influence hierarchy highlighted the combined mechanism of climatic drivers （precipitation and air temperature） and cryospheric response （glacier ablation） in shaping the region’s water yield. （4） Quantified runoff increases and their impact： Under continuous climate warming， accelerated glacier ablation significantly increased river runoff， particularly during the summer melt season. During the ablation period （May to September）， the incremental contributions of glacier melt were 0.18×10⁸ m³， 1.33×10⁸ m³， and 0.32×10⁸ m³ for the source regions of the Yellow River， Yangtze River， and Lancang River， respectively. The corresponding total increases in natural runoff were more significant， reaching 13.62×10⁸ m³， 55.12×10⁸ m³， and 37.14 × 10⁸ m³ for the Three-River Source Region， Yangtze River Basin， and Lancang River Basin， respectively， highlighting the combined effects of all runoff components. This study demonstrates that accelerated glacier ablation leads to a transient increase in water resource availability for the Three-River Source Region and its middle and lower reaches in the short term. However， from a long-term perspective， the continuous decrease in glacier storage， which functions as a natural water reservoir， will increase the unpredictability of future water security. The persistent negative mass balance indicates a limited contribution from glacial sources. These findings provide a scientific basis for water resource assessment， risk evaluation， and the formulation of adaptive management strategies in this ecologically sensitive and hydrologically vital region.
Rain-on-snow （ROS）， in which liquid precipitation falls onto an existing snow cover， is a compound hydrological process that links atmospheric forcing， snowpack energy exchange， solid-liquid phase change， water release， and basin runoff. Under a warming climate， shifts from snowfall to rainfall may alter ROS occurrence and the timing of snow-water release， with implications for short-term runoff and flood formation in cold， high-elevation basins. However， conventional hydrological models often simplify the rapid albedo decay and wet-snow densification induced by ROS， and the propagation of these changes through snowmelt and snowpack outflow to runoff remains insufficiently quantified. This study aims to improve the representation of key ROS processes in the VIC-CAS model and analyze their effects on snowpack energy balance， snowmelt， water release， and runoff in the Upper Heihe River Basin， a cold and semi-arid alpine basin in northwestern China. The VIC-CAS model was driven by the China Meteorological Forcing Dataset version 2.0 （CMFD2.0） at 3 h and 0.1° resolutions and evaluated using daily discharge data at the Yingluoxia hydrological station. Local ROS was identified at the grid-elevation band-3 h scales when liquid precipitation was at least 1.0 mm and initial snow water equivalent （SWE） at the beginning of the time step was at least 2.0 mm. Daily events were retained when the ROS-affected area covered at least 20% of the basin area and basin-equivalent rainfall on snow was at least 0.5 mm. This procedure identified 91 typical events during 1979—2018. In the VIC-CAS model， an equivalent snow-age increment proportional to ROS rainfall was incorporated to accelerate snow-albedo decay and preserve its post-event memory， while a liquid-water-induced densification term was introduced as a function of liquid-water fraction and snow density. Single-factor sensitivity tests were conducted to adjust the two newly introduced parameters by ±5% and ±10%， respectively， to examine the sensitivity of the simulation results to parameter perturbations. The baseline and ROS-enhanced models used identical meteorological forcing， initial conditions， and calibrated hydrological parameters， and their responses were compared from the ROS event day through the following seven days. Kling-Gupta efficiency （KGE）， root mean square error （RMSE）， and relative peak error （RPE） were used to evaluate long-term and event-window discharge simulations. The baseline model achieved KGE values of 0.88 during calibration and 0.85 during validation， thereby providing a basis for comparative analysis of ROS processes. Of the 91 typical events， 69 occurred from June to September， and the mean elevation of the affected areas was approximately 3 900 m， indicating that typical ROS events in the Upper Heihe River Basin were mainly concentrated in high-elevation snow-covered areas during the warm season. Within the ±10% perturbation range， both new parameters showed weak sensitivity in snowmelt and runoff. Variations in the snow-age conversion parameter changed these variables by less than 0.5%， whereas the liquid-water-induced densification parameter primarily affected snow density and snow depth. On the ROS event day， mean snow albedo decreased by 0.15 and net shortwave radiation flux increased by 30.34 W⋅m-2. Although changes in net longwave radiation flux， sensible-heat flux， and latent-heat flux partly offset the increase in shortwave energy， net snowpack energy flux still rose by 25.08 W⋅m-2. Within seven days after the ROS event， the cumulative increases in net shortwave radiation flux and net energy flux were 5.17 and 2.43 MJ⋅m-2， respectively， indicating that the added energy was redistributed among snow warming， melting， and heat exchange between snowpack and atmosphere. The enhanced energy input shifted the snowmelt peak from the first day after an event to the event day and advanced snowpack outflow. The differences in snowmelt and snowpack outflow were closely related， with a Pearson correlation coefficient of 0.86， although water storage and retention within the snowpack caused differences in response magnitude and timing. Mean ΔSWE， defined as the SWE difference between the ROS-enhanced and baseline models， was approximately -1.6 mm on the event day， indicating accelerated depletion of snow storage. Event-day snowmelt and total runoff increased by 37.9% and 15.6%， respectively， and the snowmelt contribution to runoff rose from 26.1% to 34.2%. However， rainfall runoff remained the dominant water source， contributing 65.8% of total runoff. Within three days after an event， the increase in cumulative total runoff decreased to 2.6%. By the end of the seven-day window， the difference in cumulative total runoff between the two models was nearly zero. This indicated that ROS primarily redistributed snowmelt and water release toward the event onset rather than persistently increasing event-window water yield. The long-term and event-window KGE differences between the two models were no greater than 0.01. Among 46 events underestimated by the baseline simulation， RPE improved in 39 events and RMSE declined in 28 events， whereas the added early runoff response increased errors for the 45 overestimated events. These contrasting results indicated that improved ROS representation did not necessarily produce consistent improvements in outlet-discharge accuracy because precipitation forcing， soil runoff generation， and routing also affect model performance. Overall， the study establishes a continuous process chain from ROS-induced albedo reduction and energy enhancement to earlier snowmelt and snowpack outflow and enhanced short-term runoff response. The principal hydrological effect of ROS in the Upper Heihe River Basin is therefore a shift in the timing and intensity of snow-water release. Although its influence on cumulative runoff gradually weakens over longer event windows， concentrated liquid water input and release during the early stage of events may still enhance runoff and affect flood peak formation. The findings provide a process-based foundation for assessing how intensifying ROS may affect alpine runoff and flood risk under climate warming. Because both event identification and snow-process analysis rely on simulated snow variables， future studies should incorporate independent snow observations and event-scale hydro-meteorological measurements to further constrain the magnitude and timing of ROS responses.
Purely data-driven deep learning models have demonstrated good performance in rainfall-runoff simulation because of their strong ability to capture complex and nonlinear relationships between meteorological driving factors and hydrological responses. However， such models generally lack explicit representations of hydrological processes and have weak physical interpretability， which may limit their reliability and spatial transferability when applied to basins with hydroclimatic conditions significantly different from those represented in the training region. Differentiable parameter learning-based hydrological simulation provides a new approach to addressing the above problems. By integrating the strong data learning capability of deep neural networks with the physically interpretable structure of process-based hydrological models， this method is expected to improve physical consistency， interpretability， and generalization ability of models while ensuring simulation accuracy. Nevertheless， whether this framework can be effectively applied to cold-region basins in Northwest China， particularly under the combined effects of snow accumulation and melting， seasonal temperature variations， evapotranspiration， soil moisture dynamics， and complex water balance processes， remains insufficiently investigated. In this study， three representative cold-region basins in the Shiyang River Basin of Northwest China were selected as the study areas. A differentiable parameter learning hydrological simulation framework based on the differentiable HBV （Hydrologiska Byråns Vattenbalansavdelning） hydrological model， referred to as dPL_HBV， was developed by coupling a neural-network-based parameterization module with the differentiable HBV model. Additionally， the applicability of dPL_HBV to runoff simulation in cold-region basins of Northwest China was systematically evaluated. The parameterization network estimated HBV model parameters from basin static attributes and dynamic meteorological driving factors， while the differentiable HBV model explicitly represented major hydrological processes and generated simulated runoff. During model training， the discrepancy between simulated and observed runoff was used to define the objective function， and the gradients of the loss function were propagated backward through the differentiable HBV model to the parameterization network. Consequently， observed runoff data directly constrained the neural network and optimized its weights through gradient-based learning， enabling data-driven learning of hydrological model parameters while maintaining the original physical process structure of the HBV model. The performance of dPL_HBV was systematically compared with that of the conventional HBV model and a purely data-driven long short-term memory （LSTM） model. The results showed that the Nash-Sutcliffe efficiency coefficients （NSE） of dPL_HBV exceeded 0.63 in all three study basins， with its overall performance slightly better than that of both the traditional HBV and LSTM models. To further examine the spatial transferability of the models， spatial extrapolation experiments were conducted by training the models using data from two basins and transferring them to the remaining basin that was not included in the training. Under these conditions， dPL_HBV generally outperformed both HBV and LSTM models， indicating stronger cross-basin transferability and generalization ability. In addition to runoff simulation， the Global Land Evaporation Amsterdam Model （GLEAM） evapotranspiration product was used to assess the physical consistency of internal hydrological variables simulated by the model. The correlation coefficients between the simulated evapotranspiration from dPL_HBV and the GLEAM product exceeded 0.80， indicating that the model can reproduce not only runoff dynamics but also important components of the basin water balance. Overall， the findings demonstrate that differentiable parameter learning can effectively combine the predictive capability of deep learning models with the process interpretability of conceptual hydrological models. It demonstrates considerable applicability in cold-region basins of Northwest China and has great potential for applications in cold-region hydrological simulation， runoff prediction in ungauged basins， and regionalization of hydrological models across basins.
Accurate and reliable runoff prediction is a core technical challenge in the optimal allocation of water resources in cold and arid regions. Under the influence of factors such as global warming and intensified human activities， hydrological time-series data increasingly exhibit nonlinear and non-stationary characteristics， significantly increasing the difficulty of precise runoff prediction. Particularly in the arid regions of Northwest China， issues such as limited data sequence length， strong noise interference， and scarcity of extreme runoff event samples have hindered the training process of traditional data-driven models， leading to insufficient generalization ability and limited prediction accuracy. As a result， the problem of accurate runoff prediction has become a critical bottleneck restricting the safe utilization of regional water resources. Although deep learning models have advantages in fitting complex nonlinear relationships， their “black box” nature weakens the physical interpretability of model results to some extent. Therefore， under the conditions of limited quality observational data， how to enhance model interpretability while improving prediction accuracy has become a key scientific issue that needs to be urgently addressed in current research on hydrological time-series prediction.To address these issues， this study took the upper reaches of the Danghe River in the Qilian Mountains as the study area and constructed a TimeGAN-BiLSTM runoff prediction model coupling the time-series generative adversarial network （TimeGAN） with the bidirectional long short-term memory network （BiLSTM）. To tackle the problem of limited data quality， TimeGAN was introduced to enhance the original runoff and meteorological sequences by generating high-quality synthetic data through learning the latent distribution features of the original data， thereby effectively alleviating the adverse impact of limited data quality on model training. Subsequently， BiLSTM was employed to establish the prediction model， utilizing both historical and future time-step information to improve the capture of long-term dependency features in the sequence and predict daily-scale runoff for 1， 3， 5， and 7 days ahead. On this basis， the SHAP interpretability analysis method was further introduced to overcome the interpretability deficiency of deep learning models， quantitatively analyze the contribution of each factor， and conduct a control analysis using the BiLSTM model trained on the original data to verify the impact of the data augmentation process on the interpretation of variable contribution mechanisms， thereby achieving a balance between prediction performance improvement and physical mechanism interpretability.The results showed that the data generated by TimeGAN were consistent with the original data in terms of statistical characteristics such as central tendency， dispersion， and distribution shape， as well as multi-variable correlation structures and physical consistency， effectively improving the quality of training data. In the runoff prediction experiments， the TimeGAN-BiLSTM model outperformed BiLSTM， TimeGAN-LSTM， and SMOTE-BiLSTM models in terms of various evaluation indicators， demonstrating the best prediction performance in all prediction periods， with significant improvements in overall fitting accuracy and error control. In the prediction of extreme runoff， the analysis of peak flow prediction error （TPE） and extreme runoff Nash-Sutcliffe efficiency coefficient （NSEext） indicated that the TimeGAN-BiLSTM model outperformed the comparison models in terms of peak flow characterization accuracy and prediction stability， and could more accurately reflect the changes in extreme runoff. The SHAP density scatter plot and feature importance plot further revealed the differences in runoff influencing factors under different prediction periods. In short-term predictions， runoff was mainly affected by the autocorrelation characteristics of historical runoff； while in medium- and long-term predictions， the influence of meteorological factors such as minimum temperature and potential evapotranspiration gradually increased. Finally， SHAP analysis was conducted on the BiLSTM runoff prediction model without data augmentation to verify the robustness and consistency of this method in identifying influencing factors. In summary， the interpretable TimeGAN-BiLSTM model addresses three aspects： data augmentation， bidirectional time-series modeling， and model interpretability analysis. It verifies the effectiveness of the time-series generative adversarial network in hydrological data augmentation and constructs a runoff prediction method that combines high-precision prediction ability with model interpretability. This model can provide technical support and theoretical reference for runoff forecasting in the arid regions of Northwest China， and has important scientific significance and practical application value.
Flash floods pose a critical threat to transportation infrastructure and the safety of socioeconomic development in the alpine and arid mountainous areas of Northwest China. Against the backdrop of global warming and atmospheric circulation anomalies， short-duration heavy precipitation has increased in Northwest China in recent years， thereby triggering a frequent occurrence of rainfall-triggered flash floods. However， the formation mechanisms and critical thresholds of small watersheds in the alpine and arid regions of the Qilian Mountains remain poorly investigated. On July 11—12， 2025， heavy precipitation occurred in the Bailang River Basin， located in the upper reaches of the Heihe River Basin in the central Qilian Mountains. This event triggered flash floods and caused local landslides， rockfalls， and roadbed washout along Provincial Highway 313 and National Highway 213， which severely affected transportation routes. Based on minute-scale meteorological and hydrological in situ observations and ERA5 reanalysis data， this study analyzed extreme hydrological processes， flash flood identification and attribution， precipitation-flood peak lag time， and synoptic-scale mechanisms using statistical methods. The results showed： （1） this extreme hydrological event exhibited pronounced flood behavior characteristics. The maximum rainfall intensity in the river basin reached 13.8 mm⋅h-1， and the flood peak discharge at the confluence increased sharply to 33.9 m3⋅s-1. The flood behavior at both the confluence and Xicha hydrological stations was characterized by low-intensity discharge， early peak time， hydrographs exhibiting a typical pattern of “rapid rise and slow retreat”， concentrated flood volume， and sharp and narrow hydrographs， indicating a typical flash-flood event. （2） Short-duration intense rainfall was the dominant driving factor of this flash-flood event. The relatively high antecedent soil water content facilitated rapid runoff generation， while enhanced glacier melt amplified the formation of the flood peak. （3） The hydrological response of the river basin was extremely rapid， and the lag time from the precipitation centroid to the emergence of the flood peak was only 3.5~4.5 h， reflecting the high sensitivity of small watersheds to extreme precipitation in alpine mountainous areas. （4） The short-duration heavy precipitation process was caused by a synoptic circulation configuration involving the eastward-propagating westerly trough and the westward-extending and northward-shifting subtropical high. Warm and moist air from the East China Sea was transported to the northern foothills of the Qilian Mountains and triggered the short-duration heavy precipitation event under the combined effects of topographic forced lifting and dynamic suction characterized by “low-level convergence and upper-level divergence”. This study preliminarily identifies a flood-generation pattern at the event scale in small alpine watersheds of the Qilian Mountains characterized by short-duration intense precipitation dominance amplified by cryospheric background conditions， and proposes precipitation triggering thresholds and an early-warning time window based on this representative event. These findings provide a scientific basis for future flash-flood early warning and risk management in alpine and arid mountainous areas.
Gansu Province， located in northwestern China， experiences frequent extreme floods due to its arid climate and complex topography. The region is arid with limited rainfall， exhibiting extreme spatiotemporal variability in precipitation. Notably， annual evaporation in the inland river basin significantly exceeds precipitation. Furthermore， due to the complex and diverse topography， the Yellow River Basin and Yangtze River Basin are predominantly mountainous， featuring numerous flashy mountain rivers and localized torrential rains that frequently trigger flash floods and debris flows in small watersheds. To elucidate the regional distribution patterns and evolutionary trends of maximum floods in Gansu， this study utilized data from 2 575 sets of measured and surveyed flood events across the inland river basin， Yellow River Basin， and Yangtze River Basin within the province. Known maximum floods corresponding to different catchment areas were selected based on catchment area. Hierarchical cluster analysis and the time-series cumulative value correlation curve method were employed to analyze the inflection points of flood peak modulus across the three major basins. The world flood envelope fitting method was adopted to derive the known maximum flood envelope curves for each basin in Gansu Province. Furthermore， the distribution patterns of maximum floods across the province， along with the spatial heterogeneity and scale inflection characteristics of flood peak modulus in the region， were analyzed. The results demonstrated that the maximum flood envelope curves for the inland river basin， Yellow River Basin， and Yangtze River Basin exhibited inflection points at catchment areas of 2 667 km2， 14.27 km2， and 64.7 km2， respectively. Beyond these thresholds， the peak flood modulus decreased， indicating a reduction in flood intensity. For the same catchment area， the maximum flood envelope curve for the inland river basin showed a significant deviation from the known global maximum flood envelope curve， with a flood peak modulus lower than the global level. In contrast， the maximum flood envelope curves for the Yellow River Basin and Yangtze River Basin were relatively close to the global envelope curve. Notably， individual flood events in the Yellow River Basin exceeded the global envelope curve. In terms of scale effects， for rivers with catchment areas<100 km2， the envelope curves were relatively close to the known global envelope curve， whereas for those exceeding 100 km2， the deviation became pronounced. Additionally， high flood risk areas in Gansu Province were primarily distributed in the eastern part of the Shiyang River system within the inland river basin， the eastern Jing River system and the middle and upper reaches of the Wei River system within the Yellow River Basin， and the main stem of the Jialing River within the Yangtze River Basin. Over the past few decades， for rivers with catchment areas<500 km2， no significant trend was observed in the frequency of maximum flood occurrences. However， for rivers with catchment areas exceeding 500 km2， the frequency markedly decreased. This indicated that ecological restoration projects and water conservancy projects in the river basins played a significant role in peak reduction and flood mitigation. Small watersheds became key areas and weak links for future flood prevention and management. This study provides the first systematic quantification of the regional differentiation and scale inflection characteristics of flood envelope curves for three major river basins in arid and semi-arid regions， providing a theoretical reference for extreme flood risk assessment and regional flood control planning.
This study focuses on the Yakou catchment， a representative permafrost catchment in the northeastern Qinghai-Xizang Plateau. It aims to elucidate the characteristics of summer water movement and runoff sources in the catchment， clarify the hydrological connections among precipitation， shallow soil water （0~30 cm）， deep soil water （30~90 cm）， suprapermafrost water， and stream runoff， and integrate conventional hydrological components with those unique to permafrost regions， thereby increasing attention to permafrost-specific hydrological components such as suprapermafrost water. Based on meteorological and hydrological variables during June-August in 2021 and 2022， including precipitation， air temperature， soil temperature， and water content at different depths， thaw depth， the thickness of the saturated zone， and suprapermafrost water level， as well as stable isotopes （δ 2H， δ 18O， lc-excess， and d-excess） and total dissolved solids （TDS） data of different water bodies （precipitation， shallow soil water， deep soil water， suprapermafrost water， and stream runoff）， the characteristics of summer water movement and runoff sources in the permafrost catchment were analyzed using simple linear regression， boxplot analysis， Pearson correlation analysis， and partial least squares path modeling （PLS-PM）. The results showed that： （1） during rainfall events， precipitation infiltrated and recharged shallow soil water. Shallow soil water migrated toward deep soil water， and part of the deep soil water subsequently recharged suprapermafrost water， while part remained within the saturated zone. Under the combined effects of gravity and the frozen layer acting as an aquitard， suprapermafrost water tended to migrate laterally downslope and recharge the stream. The movement of precipitation through the soil profile was dominated by matrix flow rather than preferential flow. Stream runoff increased as suprapermafrost water level， the thickness of the saturated zone， and deep soil water content increased. The stable isotope indices （lc-excess and d-excess） of soil water decreased progressively with increasing soil depth， and the lc-excess and d-excess values of soil water at 60~90 cm were closer to those at 30~60 cm than to those of precipitation. （2） Stream runoff was mainly derived from suprapermafrost water and deep soil water. The TDS of stream runoff was significantly correlated with both suprapermafrost water and deep soil water （P<0.05）. On the same sampling dates， the δ 18O values of stream water were relatively close to those of suprapermafrost water and deep soil water， and the mean absolute differences in δ 18O between stream water and suprapermafrost water and between stream water and deep soil water were the lowest. The mean δ 2H， d-excess， lc-excess， and TDS values of stream runoff were also relatively close to those of suprapermafrost water and deep soil water. The δ 2H values of different water bodies showed weak correlations with individual local meteorological and hydrological variables， such as precipitation and temperature， and none of these correlations were statistically significant. （3） Partial least squares path modeling and two-tailed Pearson correlation analysis indicated that suprapermafrost water and deep soil water directly and significantly recharged stream runoff. These findings deepen the understanding of water movement within the active layer and runoff generation mechanisms in permafrost regions. They provide new insights into the close coupling between the hydrological effects of soil water and suprapermafrost water， further demonstrating the hydrological connections among soil water， suprapermafrost water， and stream runoff in permafrost catchments. Conventional hydrological models describe catchment runoff generation processes using variables such as soil temperature and soil water content， which may make it difficult to fully represent the hydrological processes of permafrost catchments. Future studies should further consider suprapermafrost water and its dynamic coupling with active-layer soil water to improve the simulation and prediction capabilities of catchment hydrological responses and runoff changes under permafrost degradation， thereby providing a reference for understanding watershed water-cycle processes in cold regions under climate change.
The Qinghai-Xizang Plateau is a unique geographical region， and its soil moisture is affected not only by climate change but also by freeze-thaw cycles. Therefore， the soil moisture sequence in this region has obvious seasonal and nonlinear characteristics， making it difficult to conduct detailed simulations. To improve the accuracy of soil moisture prediction while maintaining model interpretability， this study proposes an integrated model based on the seasonal and trend decomposition using LOESS （STL） method. Monthly FLDAS soil moisture data （0~10 cm） from 2001 to 2020 were used to decompose the original sequence into trend， seasonal， and residual components using STL. These subsequences were then input into three models—gradient boosting machine （GBM）， Prophet， and random forest （RF）—for prediction. The obtained results were integrated through an additive model to obtain the final predicted value of soil moisture content in the region. To verify the superiority and generalization ability of the integrated model， it was compared with six basic models， and ablation experiments were conducted. Subsequently， the model was extended from the site scale to the regional scale. The results showed that： （1） the soil moisture content at observation stations in the southern Qinghai-Xizang Plateau showed a downward trend， while that at observation stations in the central and western Qinghai-Xizang Plateau showed an upward trend. The soil moisture content at ten observation stations on the Qinghai-Xizang Plateau showed stable annual periodic fluctuations， with peak values concentrated from July to September， and then falling back to valley values in winter. Some observation stations also experienced a small peak in spring. The results of the ablation experiment also indicated that trends and seasonal components constituted the main predictable information. （2） Across 10 stations during the testing period， the integrated model achieved mean RMSE， MAE， NSE， and KGE values of 0.018 m3⋅m-3， 0.012 m3⋅m-3， 0.898， and 0.888， respectively， and overall outperformed the undecomposed GBM， Prophet， and RF baselines， as well as the homogeneous STL-GBM， STL-Prophet， and STL-RF models. These results indicated that STL decomposition was an important source of performance improvement， while the heterogeneous model configuration provided additional gains after decomposition. （3） Incorporating meteorological variables further improved prediction performance at relatively difficult-to-predict stations such as Maqu and Ali， whereas the improvement was limited at stations with more stable seasonal cycles. In addition， the seasonal soil moisture peak generally coincided with， or lagged by approximately one month， the precipitation peak， while precipitation anomalies showed the most consistent statistical association with soil moisture anomalies in the residual component. （4） When the same STL-based heterogeneous ensemble framework was further applied to ERA5-Land gridded soil moisture series， the spatially averaged MAE during the test period was approximately 0.037 m3⋅m-3 and the coefficient of determination （R 2） was approximately 0.827， indicating that the framework remained applicable to a certain extent when extended to continuous gridded data. The integrated model based on the STL method proposed in this study has high accuracy and generalization ability. This model framework provides an extensible framework for hydrological research on the Qinghai-Xizang Plateau with similar temporal characteristics， and has important application value for strengthening water resource management on the Qinghai-Xizang Plateau.
The cold-region water environment is an important component of the global water cycle and cold-region ecosystems， playing a critical role in regional water resource regulation and ecological stability. Under the influence of global warming， glacier retreat， permafrost degradation， and changes in freeze-thaw processes are reshaping hydrological cycles and ecological characteristics in cold regions， resulting in increasing attention to related research. However， existing studies have mainly focused on local processes or individual environmental factors， and a systematic understanding of the overall development trends， research hotspots， and knowledge structure of cold-region water environment research remains limited. In this study， publications related to cold-region water environments indexed in the Web of Science core collection from 2009 to 2026 are selected as the data source. A total of 13 351 publications are retrieved and analyzed using bibliometric methods， the VOSviewer visualization tool， and a bibliometric analysis platform. The development trends of global cold-region water environment research are investigated from the perspectives of publication trends， contributions of countries and institutions， disciplinary distribution， and keyword evolution. The results show that： （1） global research on cold-region water environments has generally exhibited a continuously increasing trend with distinct developmental stages. The period from 2009 to 2019 represented a stage of gradual accumulation， characterized by steady growth in annual publication counts. Since 2020， the field has entered a rapid development stage， with publications from 2020 to 2025 accounting for 45.35% of the total during the study period， indicating that cold-region water environments have become an important research focus under global climate change. （2） Global cold-region water environment research shows clear regional clustering characteristics. A total of 136 countries have contributed to this field， among which the United States， China， and Canada are the major research contributors and occupy central positions in publication output and international collaboration networks. Institutional analysis indicates that the Chinese Academy of Sciences ranks first worldwide， with 1 133 publications and 30 300 total citations， highlighting its important contribution to cold-region environmental and cryosphere research. However， some European and North American institutions still have greater influence in terms of research impact. （3） Cold-region water environment research is characterized by significant interdisciplinary integration， involving mainly earth sciences， environmental sciences， hydrology， ecology， and climate change studies. Keyword clustering analysis reveals four major research directions in this field： hydrological processes and cryospheric hazards in cold-region basins， aquatic ecological evolution and biogeochemical cycles in cold regions， water environmental quality and pollutant migration and transformation， and cryospheric hydrological responses driven by climate change. （4） Research hotspots have undergone significant temporal evolution. Early studies mainly focused on fundamental issues such as glacier changes， cryosphere evolution， and paleoenvironmental reconstruction. With increasing impacts of global warming， research gradually shifted toward permafrost degradation， hydrological responses， and ecological effect analysis. In recent years， freeze–thaw mechanisms， emerging pollutant transport processes， and intelligent simulations based on machine learning and multi-source remote sensing data have become new research frontiers， promoting a transition of cold-region water environment research from describing environmental changes to analyzing underlying process mechanisms. This study systematically reveals the development trajectory， scientific contribution distribution， and hotspot evolution patterns of global cold-region water environment research. Future studies should strengthen investigations into the coupling mechanisms among cryospheric components， including glaciers， snow cover， and permafrost； deepen the understanding of interactions among climate change， hydrological processes， and ecological responses； and integrate remote sensing technologies， artificial intelligence， and multi-source data fusion methods to improve the simulation and prediction capabilities of cold-region water environment changes. The findings provide scientific references for cold-region water resource management， ecological conservation， and climate change adaptation.
Cultivated land serves as the fundamental carrier for ensuring national food security， and black soil regions constitute the core area for stable and high-yield grain production in China. As the main distribution area of black soil resources and a national-level commodity grain production base， the three provinces of Northeast China undertake the strategic mission of stabilizing national grain supply and maintaining national food security. Long-term intensive agricultural cultivation， unreasonable cultivated land use patterns， and regional industrial transformation have triggered prominent conflicts between intensive cultivated land use and ecological environment protection in the region. Problems such as black soil layer degradation and agricultural non-point source pollution have become increasingly prominent， severely restricting the high-quality and sustainable development of agriculture in black soil regions. Existing studies mostly focus on the temporal variation analysis of a single system， i.e.， cultivated land use efficiency （CLUE） or ecological environment quality （EEQ） in individual provinces， while systematic research on the spatiotemporal evolution， hierarchical transfer characteristics， and core influencing factors of the coupling coordination degree （CCD） between the two systems at the scale of the entire three provinces of Northeast China over a long-term period remains insufficient， which can hardly meet the practical demands for cultivated land protection and coordinated governance of the ecological environment in black soil regions.To address this research gap， this study conducted an empirical investigation in the three provinces of Northeast China based on panel statistical data from 2014 to 2023. First， an evaluation indicator system for CLUE was constructed following the “input-output” framework， and an EEQ assessment system was established based on the classic pressure-state-response （PSR） model. The entropy weight method was employed to calculate the comprehensive indices of both systems. On this basis， the CCD model was applied to analyze the CCD level and coordinated development status between the two systems. The Markov chain model was used to reveal the dynamic evolution characteristics of CCD levels. Finally， gray relational analysis combined with Pearson correlation analysis was adopted to identify the intensity， direction， and inter-provincial heterogeneity of the effects of various influencing factors.The research results indicated that （1） during the study period， CLUE in the study area showed a steady upward trend， with the comprehensive score increasing from 0.2313 in 2014 to 0.2979 in 2023， an average annual growth rate of 2.87%. Spatially， it presented a distribution pattern of “higher in the north and lower in the south”， with Heilongjiang Province consistently ranking first in the region. EEQ increased steadily before 2018， dropped significantly from 2018 to 2019， then showed a recovering fluctuating trend， with all three provinces experiencing a synchronous downward trend from 2022 to 2023. Spatially， it displayed a differentiation characteristic of “higher in Jilin， lower in Liaoning， and medium in Heilongjiang”. （2） Throughout the study period， the coupling degree between the two systems remained stable within a high range of 0.936~0.941， and the overall CCD showed a steady upward trend， rising from 0.399 in 2014 to 0.472 in 2023， presenting a gradient pattern of higher in the north and lower in the south spatially. Heilongjiang Province was the only one that advanced to the primary coordination stage， while Jilin and Liaoning provinces had not yet entered the coordinated development stage by the end of the study period. （3） CCD levels demonstrated strong overall stability， with level transfers mainly occurring between adjacent levels. The probabilities of upward and downward transfers were basically balanced， showing a slight tendency toward downward transfer， and a sustained positive benign evolution pattern had not yet been formed. （4） All influencing factors exerted significant effects on CCD with prominent inter-provincial heterogeneity. Agricultural input capacity and green innovation level served as core positive driving factors across all three provinces， while traditional labor input showed a significant negative restrictive effect throughout the entire region.
As the ecological lifeline of China’s arid northwest and a critical node along the Silk Road Economic Belt， the Heihe River Basin faces severe challenges in water security. The stability of water resources in the basin is highly dependent on the runoff generation in the upper Qilian Mountains， which is currently undergoing profound changes due to global warming. The current water management framework， specifically the 97 Water Diversion Scheme， relies heavily on historical hydrological statistics and stationary assumptions to allocate water between the middle and lower reaches. However， whether this static allocation scheme remains valid under future non-stationary climate conditions is still an unresolved and critical issue. To address this， this study systematically evaluates the future evolution of runoff at the Yingluoxia， investigates the controlling station for the upper basin， and assesses the adaptability of the current water diversion policy through the end of the 21st century. The physically based mesoscale Hydrologic Model （mHM） and the Multiscale Parameter Regionalization （MPR） technique were employed to effectively capture the spatial heterogeneity of the complex mountainous terrain. The model was rigorously calibrated and validated using historical daily runoff data （1990—2009）， achieving high simulation accuracy with a Nash-Sutcliffe Efficiency （NSE） exceeding 0.8 and a Kling-Gupta Efficiency （KGE） greater than 0.9， demonstrating its robustness in reproducing regional hydrological processes. To project future scenarios， the calibrated model was driven by a bias-corrected ensemble of 17 Global Climate Models （GCMs） from the Coupled Model Intercomparison Project Phase 6 （CMIP6） under two Shared Socioeconomic Pathways： the intermediate emission scenario （SSP2-4.5） and the high emission scenario （SSP5-8.5）. The multi-model ensemble projections revealed several critical findings that challenged prevailing assumptions. First， contrary to the expectation of a sustained “wetting” trend in Northwest China， future runoff in the upper reaches of the Heihe River was not projected to increase continuously. Any potential increases in runoff were likely to be limited to the period before the mid-21st century （around 2040）. Subsequently， the runoff trajectory tended to stabilize or even decrease. Attribution analysis indicated that while precipitation was projected to increase ［2.87 mm·（10a）-1 under SSP5-8.5 and 1.83 mm·（10a）-1 under SSP2-4.5］， the concurrent rise in evapotranspiration ［e.g.， 2.92 mm·（10a）-1 under SSP5-8.5］， driven by significant warming， largely offset the runoff-generation potential of this additional precipitation. This “cancellation effect” was particularly pronounced under the SSP5-8.5 scenario， where the elasticity of runoff to precipitation declined due to intensified evaporative demand. Furthermore， this study highlighted a transition from stability to volatility. The core challenge facing the Heihe River would not merely be a directional trend in water volume， but a significant loss of stability and a surge in uncertainty. By mapping the projected runoff against the guarantee rates of the 97 Water Diversion Scheme， a critical shift was identified in the risk structure of water resources. Under the SSP5-8.5 scenario， the combined probability of “dry” and “extremely dry” years was projected to increase substantially in the late 21st century. Specifically， the frequency of “extremely dry” years could rise to 25%， indicating that the basin may experience severe water scarcity once every four years. Additionally， analysis of transition probability matrices suggested an increased risk of “abrupt alternation between droughts and floods，” making inter-annual regulation increasingly difficult. In conclusion， these findings suggest that the rigid thresholds of the 97 Water Diversion Scheme may become frequently unachievable in the future， potentially intensifying water conflicts between the agricultural middle reaches and the ecological lower reaches. The increasing frequency of extreme hydrological events indicates that the current management paradigm， which relies on historical averages， is no longer sufficient. Consequently， this study underscores the necessity of a paradigm shift towards “managing uncertainty”. This necessitates the development of adaptive water resource management strategies， including integrating short-term climate forecasting into allocation decisions， strengthening demand-side management to reduce rigidity in water consumption， and establishing flexible emergency response mechanisms to cope with the normalized volatility of future water resources.
Glaciers and snow cover， as the main components of the cryosphere， provide a fundamental water source for the ecological environment and human socioeconomic activities in cold and arid regions. Under the background of global warming， the cryosphere is experiencing widespread retreat， and changes in glaciers and snow cover have significantly affected water resource utilization in alpine river basins replenished by glacier and snowmelt， as well as the socioeconomic development of downstream areas. As a critical component of cryospheric services， freshwater provisioning service is closely linked to the Sustainable Development Goals （SDGs） and exerts the most direct and significant impact on human well-being among cryospheric services. Assessing the impact of freshwater provisioning service in cryosphere-affected areas on economic development not only provides a reference for the sustainable utilization and management of water resources in the economic belt on the northern slope of the Tianshan Mountains， but also offers decision support for advancing high-quality construction of the core area of the Silk Road Economic Belt and for the optimization and upgrading of regional industrial structure. Based on the present value assessment method， this study constructed an economic loss model of freshwater provisioning service in the cryosphere-affected area of the Urumqi River Basin on the northern slope of the Tianshan Mountains， estimated the future economic losses of glacier and snowmelt provisioning service in the river basin， and analyzed the relationships between provisioning service and economic development under different shared socioeconomic pathways （SSP） scenarios. The results showed that the economic losses caused by snowmelt provisioning service of the Urumqi River across various sectors were generally greater than those caused by glacial meltwater， with the industrial water sector experiencing the largest loss， followed by the domestic water， service， and agricultural water sectors. Under the SSP5-8.5 scenario， the loss value of glacier and snowmelt provisioning service in the Urumqi River was higher than under the medium- and low-emission scenarios. The cumulative economic losses were projected to increase from 0.58 billion yuan in the middle of this century to 1.82 billion yuan by the end of this century， and the value of freshwater provisioning service was projected to decrease by approximately 76.0% by the 2090s compared with the reference year. Under the SSP1-2.6 scenario， the provisioning service and economic development of the agricultural， industrial， and service sectors all showed a coordinated development relationship， with the coordination degree shifting from low-level coordination to moderate coordination， and the corresponding coordination level ranging from 3 to 4. Under the SSP2-4.5 scenario， the provisioning service and economic development of different sectors either showed a coordinated development relationship （with the coordination level ranging from 3 to 4） or a conflicting relationship （with the corresponding conflict level ranging from 5 to 6）. Under the SSP5-8.5 scenario， all sectors exhibited a conflicting relationship between provisioning service and economic development， with the conflict level ranging from 5 to 8， and the conflict level of the service sector was higher than that of the agricultural and industrial sectors. It is recommended that active and effective measures be taken to reduce greenhouse gas emissions before the middle of this century， as this will help mitigate the economic losses of glacier and snowmelt provisioning service. Through targeted transformation initiatives in industry， agriculture， and service industry—such as the construction of new energy bases and digital-intelligent integration in the industrial sector， the urbanized and intensive transformation of agriculture， and the quality improvement and upgrading of the service sector empowered by green electricity—the deep integration of industries with water conservation and ecological protection can be realized. Coordinated upgrading and comprehensive adjustment of the three sectors should be promoted in a holistic way， thereby providing support for the high-quality socioeconomic development of Xinjiang.
During the boreal winter， the Siberian-Mongolian High over Eurasia generates strong， cold， and dry northwesterly winds across East Asia， collectively constituting the East Asian winter monsoon （EAWM）. As the most powerful and active atmospheric circulation system in the Northern Hemisphere during winter， the EAWM exhibits significant intensity fluctuations on intraseasonal， interannual， and interdecadal timescales. It not only profoundly influences winter climate variations across Eurasia， but also directly affects the natural environment and socio-economic development of approximately one-third of the global population. Therefore， investigating its evolution on （sub-）orbital timescales and assessing the relative contributions of various internal driving factors to its intensity changes are crucial for understanding and predicting the future dynamics of the EAWM. This study focused on a typical permafrost-peat bog core （WLR17） retrieved from the Wanlongwama River terrace in the Headwater Area of the Yellow River （HAYR）， northeastern Qinghai-Xizang Plateau. Based on eight accelerator mass spectrometry （AMS） 14C dating results， a reliable chronological framework was established using the Bacon model. Environmentally sensitive grain-size components responsive to changes in EAWM intensity were extracted using the grain size-standard deviation method， with the S3+S4 component （>79.43 μm） selected as a proxy. By integrating multiple grain-size and geochemical elemental proxies—including the environmentally sensitive grain-size component S3+S4， the content of the >63 μm sand fraction， mean grain size， SiO2 and TiO2 contents， and the SiO2/TiO2 and Rb/Sr ratios—the evolution of the EAWM in the study area since 13.7 cal ka BP was reconstructed. The results revealed a relatively strong EAWM in the study area during 13.7~8.8 cal ka BP. After rapidly weakening from approximately 8.8~7.9 cal ka BP， the EAWM remained relatively weak throughout the mid- to late Holocene. Superimposed upon this overall trend was a series of pronounced short-term intensification events of the winter monsoon. In addition to capturing the Older Dryas and Younger Dryas events during the Last Deglaciation， the WLR17 core also recorded nine distinct strong EAWM events throughout the Holocene. These strong winter monsoon events generally corresponded to cold events recorded in the North Atlantic， with only minor chronological discrepancies observed in a few events. On this basis， the reconstructed EAWM variability was compared with other well-dated， high-resolution EAWM records. By considering historical changes in driving factors such as solar radiation， ice-sheet volume， atmospheric CO₂ concentration， and the Atlantic Meridional Overturning Circulation （AMOC）， this study assessed the relative importance of these drivers in modulating the intensity of the EAWM. The results indicated that the evolution of EAWM intensity in the HAYR since 13.7 cal ka BP was not dominated by winter insolation. Instead， the EAWM was jointly controlled by variations in Laurentide Ice Sheet volume and atmospheric CO2 concentration from approximately 13.7 to 7.0 cal ka BP. After 7.0 cal ka BP， atmospheric CO2 concentration emerged as the primary driver. In addition， since 13.7 cal ka BP， other factors， including solar radiation， AMOC fluctuations， and Eurasian snow-albedo feedback， influenced EAWM intensity to some extent throughout this period. In summary， this study provides reliable multi-proxy records of the evolution of the EAWM since the Last Deglaciation and reveals its potential driving mechanisms. These findings not only deepen the understanding of the evolution and driving mechanisms of the EAWM on （sub-）orbital timescales but also provide an important basis for predicting its future evolution.
The cryosphere， defined as the part of the Earth system where water exists in solid form， together with the atmosphere， hydrosphere， biosphere and lithosphere， constitutes the Earth’s climate system. In the context of global warming， cryosphere elements are highly sensitive to climate change and face risks of degradation and abrupt change. Therefore， in-depth exploration of the evolutionary patterns of the cryosphere and their comprehensive effects on the sustainable development of human habitats has gradually become a key issue in frontier geographical research. The geographical elements of the cryosphere， composed of solid water， act as a link in the Earth system and are coupled with the four major spheres mentioned in senior high school geography textbooks， directly influencing the cryosphere geographical environment and the sustainable development of cold regions. However， in the current versions of senior high school geography textbooks， the content related to the Earth’s spheres is mostly limited to the four major spheres of the Earth’s surface， and the introduction of the cryosphere is relatively scarce. Moreover， by reviewing recent high school examination questions， it is found that current geography of College Entrance Examination tends to test knowledge transfer and comprehensive thinking abilities， and the proportion of knowledge examination in the field of cryospheric science has increased. The test situation materials present a trend of frontier and interdisciplinary integration， and the constituent elements， environmental characteristics， and variation characteristics of these elements have become important sources of new examination propositions. Introducing cryospheric science knowledge into senior high school class can not only compensate for the current deficiencies in cryosphere geographical knowledge in senior high school textbooks， but also stimulate students’ interest in exploring natural sciences， promote students’ systematic mastery of the knowledge system of Earth’s spheres， and cultivate their scientific literacy and social responsibility. By analyzing actual cases of cryosphere changes， guiding students to dialectically view the interaction between cryosphere changes and human activities is an effective way to implement the education of human-Earth coordination. This paper is based on the multi-sphere knowledge system of cryospheric science and high school geography， combined with the current teaching situation， College Entrance Examination requirements， and implementation feasibility， to explore effective pathways for integrating the frontiers of cryospheric science into senior high school geography class. By connecting the comprehensive influence of the four major spheres and the cryosphere on humans， students are guided to think about the integrity of the geographical environment， and to view the interrelationships of geographical elements from a relational perspective， thereby expanding their analytical perspective. In addition， students are guided to use comprehensive thinking to analyze the geographical characteristics of the cryosphere， deepen their regional cognition of the cold-region environment， enhance their geographical practical ability for problem-solving， establish a correct human-land coordination， and enhance environmental protection awareness. In conclusion， introducing courses and knowledge related to cryospheric science on the basis of the four major spheres of senior high school geography knowledge can help promote the cultivation of key competencies such as regional cognition， comprehensive thinking， and human-Earth coordination view. It can not only deepen senior high school students’ understanding of the Earth’s multi-sphere system and the overall nature of the natural geographical environment， improve their knowledge system， but also enhance their key problem-solving abilities and demonstrate the unique educational value of the geography subject in the review and preparation for exams.
