Unraveling scale-dependent flood responses to changing climate extremes over the Tibetan Plateau

Unraveling scale-dependent flood responses to changing climate extremes over the Tibetan Plateau | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Unraveling scale-dependent flood responses to changing climate extremes over the Tibetan Plateau Xueqin Zhang, Xiang Li, Peng Cui, Pengke Shen This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7153596/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 18 Mar, 2026 Read the published version in Communications Earth & Environment → Version 1 posted You are reading this latest preprint version Abstract Understanding of the dynamic interplay between plateau floods and climate extremes has long been constrained by scale fragmentation in existing studies. Our multi-scale analysis unravels and interprets the scale-dependent responses of floods to changes in climate extremes across the Tibetan Plateau (TP). Extreme precipitation, temperature, and snowmelt drive the average flood day increase (0.7 d/10a) at the plateau scale, while the rise in annual maximum daily discharge (Q max ) (2.1 m 3 /s/10a) is modulated by extreme precipitation and drought indices. Watershed-scale analysis uncovers a distinct east-west partitioning of flood drivers, whereas river order-scale analysis reveals patterned shifts in flood drivers from main streams to tributaries. Cross-watershed analysis shows that upstream temperature changes contribute 5.3% to downstream flood frequency and 4.8% to magnitude variability via hydrological connectivity. The scale-specific disparities, shaped by the synergistic effects of watershed hydrological processes, underlying surface heterogeneity, climate factor sensitivities, and climate-cryosphere interactions, establish a framework for alpine flood attribution and predictive models. Earth and environmental sciences/Natural hazards Earth and environmental sciences/Climate sciences/Climate change/Climate-change impacts Flood Climate extreme Scale-dependent response Hydrological connectivity Tibetan Plateau Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Figure 9 Introduction Plateau regions have emerged as critical hotspots for extreme flood events under accelerated global warming. The IPCC Sixth Assessment Report indicates that flood risk in high-altitude areas has increased by 2–3 times compared to low-altitude regions (IPCC, 2023 ), with the "high-altitude amplification effect" being particularly prominent in high mountain areas (Bai et al., 2025 ; Wang et al., 2025 ). Diverging from traditional flood patterns, plateau floods driven by climate extremes exhibited three emerging characteristics: enhanced flash flood occurrences in small watersheds, increased frequency of precipitation-snowmelt compound floods, and a growing proportion of glacial lake outburst floods (GLOFs) (Kundzewicz et al., 2014 ; Wu et al., 2022 ; Wang et al., 2022 ). The complex shifts in climate extremes on the plateau and its cascading impacts on the cryosphere precipitated a new flood regime. Take TP as an illustrative example, it has experienced significant alterations in extreme precipitation patterns over recent decades, characterized by concurrent increases in total precipitation, frequency and intensity of extreme precipitation (Yao et al., 2025 ). These pronounced changes in extreme precipitation directly drive more frequent flash floods in small watersheds. Concurrently, the frequency and intensity of extreme high-temperature events has increased, while extreme cold events declined sharply (Yang et al., 2022 ; Yin et al., 2017 , 2019 ). The divergence triggered a chain reaction in the cryosphere system: although the overall snow depth, snow cover, and total snow accumulation decreased (Bormann et al., 2018 ; Gao et al., 2023 ), the snowmelt and the frequency of extreme snowmelt increased locally, leading to more precipitation-snowmelt compound floods (Li et al., 2024 ). Additionally, accelerated warming exacerbated glacial retreat across TP, causing an increase in the area of glacial lakes by 22%/10a (1990–2020) and a consequent rise in the risk of GLOFs (Zhang et al., 2024 ). While a clear connection existed between evolving climate extremes and flood regimes on the plateau, the quantitative characterization of their coupling mechanisms remained a persistent scientific challenge (Harrison et al., 2025 ; Otto, 2023 ; Scussolini et al., 2024 ). Current understandings of flood change attribution faced limitations, particularly regarding the scale-dependent disparities in flood responses to climate extremes. At the global scale, warming amplified atmospheric water-holding capacity, leading to a robust increase in the frequency and intensity of extreme precipitation (Min et al., 2011 ; Park and Min, 2017 ). However, no pronounced long-term trend in river flood magnitude was observed globally due to offsetting effects between different flood types. For instance, increasing precipitation-driven floods were counterbalanced by decreasing trend in snowmelt-induced floods (Zhang et al., 2022 ). At the basin scale, flood responses to climate change were governed by basin size and underlying surface conditions (Brunner et al., 2021 ). In small watersheds, short flow concentration paths and limited regulatory capacities resulted in a tight coupling between precipitation and runoff, enhancing sensitivity to extreme precipitation (McGuire et al., 2005 ; Sivapalan, 2003 ). Large basins, with their extensive catchment areas, exhibited flood processes more regulated by antecedent basin wetness. Additionally, surface characteristics further shaped flood response patterns (Torre Zaffaroni et al., 2023 ): basins with abundant vegetation exhibited stronger adaptability to climate extremes through canopy interception and soil infiltration (Peel et al., 2002 ), while highly urbanized areas amplified flood risks from extreme precipitation due to expanded impervious surfaces (Balaian et al., 2024 ). Our study systematically explored the scale-dependent response characteristics of plateau floods to changes in climate extremes, focusing on four hierarchical scales: plateau-wide, watershed, river hierarchy, and cross-watershed dynamics. To address divergent linear and nonlinear response patterns of floods to climate extremes across scales, we devised targeted research strategies: classical regression methods for quasi-linear flood responses at plateau and watershed scales, and a machine learning-based framework for complex nonlinear responses at cross-watershed scales. Unlike existing machine learning-based attribution approaches (Jiang et al., 2024 ), our method featured two key innovations. Firstly, we incorporated spatially explicit variables enabling a more accurate representation of heterogeneous distribution of climate extreme indices across watersheds. Secondly, we established a hydrological connectivity transmission mechanism that can quantify cascading effects of upstream climate extremes on downstream floods through river network system. The proposed multi-scale analytical framework established a generalizable methodological paradigm for flood attribution and adaptation strategy development in global alpine regions. Results Spatio-temporal changes in floods Annual flood days were 6.99 d from 1980 to 2019 in TP, with maximum (13.45 d) and minimum (4.72 d) occurring in 2019 and 1997 respectively (Fig. 1 a). A significant increasing trend ( p < 0.05) of 0.7 d/10a was observed over the past 40 years. M-K mutation analysis identified 2016 as a turning point for flood days: flood days fluctuated moderately without significant trends during 1980–2016, followed by rapid increases post-2016. Spatially, most watersheds experienced increased annual flood days across the plateau in the past 40 years. In particular, flood days increased significantly in northern TP (Fig. 1 b); while regions with reduced flood days were concentrated in southwest, northwest, and southeastern regions. Notably, the most substantial reductions in flood frequency occurred near the Himalayan range and northern Pamir Plateau. Besides, the flood day change patterns of rivers at different orders revealed significant hierarchical differentiation: the average flood days of first-order rivers showed no considerable change trend over the past 40 years, whereas the increases in flood days presented an obvious stepwise enhancement with the rise in river order. Average flood days for second- to seventh-order rivers rose significantly ( p < 0.05) at rates of 0.5, 1.8, 2.7, 3.5, 3.8, and 5.0 d/10a, respectively (Fig. 1 e). Additionally, Q max served as a key indicator of the intensity of the strongest flood event in each river annually, averaged 142.15 m³/s during 1980–2019, peaking in 2019 (189.55 m³/s) and troughing in 1997 (126.04 m 3 /s). Though the overall Q max was non-significant trend (2.1 m 3 /s/10a), its temporal variation closely mirrored average flood days (Fig. 1 c): stable during 1980–2016 followed by post-2016 acceleration. This consistency between Q max and flood day changes indicated that floods became not only more frequent but also more intense. Spatially, Q max increased across most plateau watersheds from 1980–2019, with particularly significant rises in the Hindu Kush Mountains region and the central/northwestern Kunlun Mountains (Fig. 1 d). In contrast, significant declines in Q max were observed in watersheds of the northern Pamir Plateau, south of the western Kunlun Mountains, and near the eastern Himalayas. Additionally, the average Q max of first- to seventh-order rivers showed no significant change trends (− 0.2, − 0.4, − 1.5, and − 1.2 m³/s/10a; p > 0.05), whereas fifth- to seventh-order rivers exhibited increases of 35.8, 62.9, and 152.8 m³/s/10a respectively (Fig. 1 f). Multi-scale attribution of flood change Plateau-average scale We compared model performance in analyzing the impacts of changes in climate extremes on average flood days and Q max across the TP. The analysis incorporates 33 climatic parameters (25 climate extreme indices in Tab. S1, plus average/maximum/minimum temperature, annual precipitation, annual snowfall, annual snowmelt, extreme snowmelt days, and continuous extreme snowmelt occurrences). For these 33 parameters, the average and maximum values of each watershed are calculated, generating a total of 66 independent variables, with the average flood days and Q max of plateau rivers as dependent variables. Results demonstrated that linear models achieved significantly higher explanatory power than machine learning models given identical factor selections. Stepwise regression performed best among linear models (flood days: R² >0.9; Q max : R 2 > 0.8). By contrast, the bootstrap forest method was optimal performance among nonlinear machine learning models (flood days: R² = 0.80; Q max : R 2 = 0.61). Consequently, we quantified linear climate-flood linkages via stepwise regression. For flood day variability, we identified 17 key drivers which influence it. The resulting model demonstrated excellent predictive performance, yielding an exceptionally high coefficient of determination (R² = 0.96), explaining 96.3% of observed variance in flood days. Model robustness was further confirmed by an adjusted R² of 0.93 and a low root mean square error (RMSE) of 0.45, with information criteria values (AICc = 101.96; BIC = 96.05) demonstrating an optimal balance between model parsimony and explanatory power. The ANOVA analysis revealed distinct contributions: extreme precipitation indices accounted for the majority (59.9%) of observed changes in flood days, while extreme temperature and snowmelt contributed 18.2% and 17.9% respectively (Fig. 2 a). The RX5day, annual precipitation, snowmelt, and snowfall emerged as the most influential drivers according to LogWorth significance values (Fig. 2 b). These factors were closely associated to precipitation and snowmelt in the TP, corresponding to the two main sources of river discharge in the region. Specifically, the contribution of RX5day to flood day changes was 20.8% (10.9% for the average RX5day and 9.9% for the maximum RX5day). RX5day captured extreme precipitation intensity, which directly drove rapid increases in river discharge and flood events. Annual precipitation modulated 18.5%, with average (8.5%) and maximum values (9.9%) showing comparable impacts. Average annual precipitation primarily modulated fundamental hydrological conditions by elevating key factors exacerbating flood risk, such like baseflow, water levels, and soil moisture saturation. Simultaneously, both maximum RX5day and maximum annual precipitation exhibited identical contributions, highlighting the critical role of spatial precipitation concentration in flood dynamics. Snowmelt processes contributed significantly to the average flood day variability in TP, with annual maximum snowmelt and snowfall respectively explaining 8.1% and 8.0% of the changes. These snow-related indices influence flooding through two primary mechanisms. On the one hand, sustained snowmelt, as a major source of river baseflow, regulated long-term hydrological conditions and flood susceptibility. On the other hand, the concentrated spring meltwater pulse (typically March–June) interacted synergistically with precipitation. This compound effect accelerated watershed response through simultaneous liquid water inputs, reduced soil infiltration capacity, and shortened runoff concentration times, collectively amplifying flood risks beyond what either factor would produce independently. Additionally, indices such as EP, R20, FD0, TNx, and CLT also played critical roles in flood day changes on the plateau. For Q max variability, 14 significant climate drivers were identified across TP using stepwise regression (R² = 0.82, adjusted R² = 0.72). The model demonstrated robust predictive performance, with an RMSE of 6.66, AICc of 302.01, and BIC of 305.38. The ANOVA analysis revealed distinct contributions of different climate extremes: extreme precipitation dominated for 51.2% of Q max variability, followed by drought (22.7%) and extreme temperature (8.1%) (Fig. 3 a). Critically, precipitation extremes emerged as the primary controls: RX1day exhibited the strongest influence (average contribution: 16.7%; maximum contribution: 9.1%), followed by R99p (maximum contribution: 9.6%) (Fig. 3 b). These indices characterized the intensity and magnitude of extreme precipitation events that directly governed flood peak discharges. Drought-related indices dominated the remaining top factors, with maximum PDSI contributing 7.8%. These indices represented long-term moisture conditions that established critical baseline hydrological states by regulating annual patterns of soil moisture, groundwater storage, and baseflow. Upon occurrence of extreme precipitation, hydrological factors rapidly transitioned from baseline to flood peaks. Thus, flood magnitude was ultimately controlled by synergistic coupling between pre-existing hydrological baselines and superimposed extreme precipitation events, explaining 73.9% of Q max variability across TP. Watershed-scale Extreme precipitation was recognized as the most critical driver of flood days, exerting the broadest spatial influence, particularly in the southeastern, southern, and northeastern TP (Fig. 4 a). Extreme drought indices further exhibited considerable influence on flood day variations, primarily impacting the western plateau, the lower reaches of the Yarlung Tsangbo River, and select watersheds near the source of the Yellow River. Extreme temperatures dominated flood day changes in the western, central-northern, and scattered central-southern watersheds of the plateau. In addition, snowmelt processes dominated in high-elevation (> 4000 m asl) western watersheds, reflecting distinct altitudinal controls on flood generation mechanisms. Simultaneously, extreme precipitation emerged as the primary driver of Q max changes across most plateau watersheds, notably in the Altun Mountains, Qilian Mountains, eastern Kunlun Mountains, and the Qiangtang Plateau (Fig. 4 b). Extreme high temperatures predominantly modulated discharge changes in the central, northwestern, and southern regions of the plateau. In contrast, extreme drought indices ranked as the second most important factor, mainly affecting the southwest, northwest, and eastern parts of the plateau. Meanwhile, extreme snowmelt played a more localized role, primarily impacting specific western plateau watersheds. Based on spatial differences in the dominant climate extreme indices for watershed flood days and Q max changes, the study area was divided into three parts from west to east. First, flood changes in the western plateau (west of 82 °E) were dominated by extreme temperature, snowmelt, and drought indices. Extreme temperature and drought indices drove most of the significantly increased flood days in this part. The significant decrease in flood days in the northern Pamir Plateau was caused by extreme temperature changes. Besides, the increase in Q max around the Hindu Kush Mountains were dominated by extreme snowmelt indices, whereas those near the Pamir Plateau were jointly driven by extreme temperature and drought indices. Conversely, Q max increased around the Western Himalayas, which was attributed to extreme drought indices. Second, flood changes in most watersheds of the central plateau (between 82 °E and 95 °E) were mainly affected by extreme temperature and precipitation indices. Extreme temperature indices dominated the majority of watersheds with significantly increased flood days and Q max . Notably, Q max around the Central Himalayas decreased significantly, driven by extreme precipitation, temperature, and drought indices. Third, changes in extreme precipitation and drought indices, especially changes in extreme precipitation, dominated flood day and Q max changes in most watersheds of the eastern plateau (east of 95°E). Specifically, extreme precipitation changes induced significant increase in flood days in watersheds east of the Qaidam Basin, while changes in both extreme precipitation and drought indices jointly contributed to the significant increase in annual Q max in this region. River order-scale We applied LASSO regression method to identify the key climate extreme drivers governing the flood changes across different river orders in TP (Fig. 5 ). Representativeness differences among rivers of varying orders were explicitly considered in the analysis: Order 1 rivers exhibited a slight discharge, resulting in a low frequency of flood events, while Order 7 rivers existed in only a few sections in TP, making it challenging to comprehensively reflect the flood change characteristics of the entire plateau. Consequently, analysis primarily discussed the impacts of climate extremes on flood variations in rivers of Orders 2–6. Extreme precipitation indices emerged as the primary drivers of flood day variations across different river orders. RX1day was particularly influential for flood day changes in Order 2–5 rivers, contributing 47.0%, 43.6%, 27.8%, 39.9%, and 34.6% of the flood day variations, respectively. In contrast, annual precipitation, R20, and EP collectively dominated flood changes in Order 6 rivers, with a total contribution exceeding 49%. In contrast, the impacts of snowmelt indices on flood day changes exhibited significant differentiation by river order. The contribution of extreme snowmelt indices was generally low (typically < 5%) for Order 2–4 rivers but increased obviously with rising river orders. In detail, CESM contributed over 10% to flood day changes in Order 5 rivers and emerged as the most critical climate extreme index for Order 6 rivers, indicating that snowmelt became an important auxiliary driving factor in larger basins. This shift was primarily attributed to the cumulative effect of expanding basin scale: as river orders increased, snowmelt runoff from multiple upstream sub-basins continuously accumulated during the confluence process, making snowmelt runoff a key water source for high-order rivers. Besides, the dominant climate drivers of Q max varied systematically with river order. R95p and CWD contributed 74.8% and 67.8% to Q max variations in Order 1 and 2 rivers, respectively, indicating that low-order rivers were predominantly regulated by precipitation (both short-duration heavy rainfall and prolonged wet periods). In Order 3 rivers, CWD had the largest contribution (29.2%), and the influence of PDSI increased (20.0%), suggesting that Q max in medium-scale basins began to be modulated by antecedent soil moisture conditions. EP and PDSI collectively dominated Q max variations in Order 5 and 6 rivers (54.1% and 50.7% total contributions, respectively), revealing that large-scale basins were influenced not only by extreme precipitation but also by antecedent basin moisture status. This progression underscored the cumulative impact of climate factors on hydrological processes in high-order rivers, where both acute precipitation events and longer-term soil moisture dynamics in shaping flood magnitude within complex basin systems. Therefore, we discovered that a gradient evolved in a gradient pattern. As river order increased, the variation pattern of flood days transitioned from being dominated by extreme precipitation to one controlled by the synergy of extreme precipitation and snowmelt. Concurrently, the variation pattern of Q max evolved from an extreme precipitation-driven regime to one regulated by the combined effects of extreme precipitation and soil moisture conditions. Cross-watershed scale We conducted a cross-watershed flood change attribution study by treating annual flood days and Q max values from discrete watersheds as independent yet network-connected observational units. Z-score standardization was applied to both flood indicators and climate extreme variables within each watershed to facilitate cross-watershed comparison while preserving intra-watershed dynamics. Methodologically, we systematically integrated three representative spatial feature metrics (watershed-averaged, maximum, and upstream-averaged values) for each of the 33 parameters, generating 99 affecting variables to capture both local and network-scale drivers of flood variability. We evaluated eight distinct modeling approaches, including five machine learning methods (Random Forest, Gradient Boosting, Support Vector Machine, K-Nearest Neighbors, and Artificial Neural Network) and three linear statistical techniques (Ordinary Least Squares, LASSO Regression, and Stepwise Regression), to identify the optimal method for attributing flood changes. Through rigorous comparison of model fitting performance across multiple evaluation metrics, we selected the most appropriate algorithm that best captured the complex climate-flood relationships while maintaining practical interpretability. Firstly, the Bootstrap Forest algorithm demonstrated superior performance among the machine learning models developed for flood day fitting, achieving the highest predictive accuracy with an R² of 0.78 and the lowest Root Average Square Error (RASE) of 0.12 (Fig. 6 a, 7 b). Boosted Trees followed closely with an R² of 0.70 and RASE of 0.14. In contrast, Support Vector Machine and K-Nearest Neighbors exhibited poorer performance, with R² values of 0.68 and 0.46 and RASE values of 0.13 and 0.18, respectively. All five machine learning methods surpassed the three linear statistical approaches. The Bootstrap Forest model, identified as the optimal performer, was subsequently employed for factor importance analysis. Climate extreme indices collectively explained over 78% of variation in flood days. Specifically, extreme precipitation, extreme temperature, extreme drought, and extreme snowmelt indices accounted for 42.9%, 22.8%, 7.3%, and 5.0% of variations, respectively (Fig. 6 c). Extreme precipitation and temperature were the dominant drivers, jointly contributing 65.7%. Additionally, changes in watershed-average, watershed-maximum, and upstream-average climate extreme indices contributed 41.0%, 28.5%, and 8.5% of variations, respectively (Fig. 6 d). Hence, although climate extreme index changes within the watershed were the primary determinants of flood frequency, upstream climate extremes conditions also played a non-negligible role. Notably, upstream changes in extreme temperature indices contributed 5.3% of the flood day variation, highlighting the downstream propagation of snow and glacier melt processes through hydrological connectivity and confirming the hydrological coupling effect in high-altitude glacier-river systems. At the individual index level, annual precipitation, R10, PDSI, CWD, and SPEI were the most influential indices, contributing 22.1%, 7.4%, 4.3%, 3.2%, and 3.0% of the flood day variation, respectively. These results diverged significantly from the attribution analysis of flood days across the entire plateau (Fig. 6 e). Extreme precipitation and drought indices dominated among the top contributors. The persistently high importance of annual precipitation further supported the critical role of precipitation background in flood frequency. Meanwhile, the substantial contribution of R10 underscored the key influence of extreme precipitation frequency on flood day variations in most watersheds of TP. Although extreme drought indices made a relatively modest overall contribution (7.3%), individual drought indices, such as PDSI (4.3%), ranked among the most influential, demonstrating their significant impact on flood frequency. Prolonged drought can reduce the soil moisture retention capacity, affecting not only the basin's baseline hydrological state but also decreasing precipitation infiltration. If rainfall cannot effectively penetrate overly dry soils, more water will flow into river systems as surface runoff, thereby elevating flood risks. At the same time, the Bootstrap Forest method again demonstrated superior performance for models with Q max as the dependent variable (Fig. 7 a). The Bootstrap Forest model achieved an R² of 0.71 with a relatively low RASE (0.13) (Fig. 7 b), explaining over 71% of the variation in Q max . Specifically, extreme precipitation, extreme temperature, extreme drought, and extreme snowmelt indices contributed 38.6%, 19.1%, 8.0%, and 5.3% of variations, respectively (Fig. 7 c). Extreme precipitation and temperature remained the dominant drivers, jointly contributing 57.7% to the changes in Q max . Changes in the watershed-average, watershed-maximum, and upstream-average climate extreme indices contributed 37.9%, 25.5%, and 7.5% of variations, respectively (Fig. 7 d). Thus, upstream changes in climate extremes also exerted a non-negligible influence on peak discharge. Notably, upstream changes in extreme temperature indices alone explained 4.8% of the variation. Examining specific indices, annual precipitation, PDSI, RX5day, FD0, and CWD emerged as key factors affecting Q max , with contribution rates of 7.9%, 5.4%, 5.0%, 4.8%, and 4.7%, respectively (Fig. 7 e). Specifically, annual precipitation and PDSI showed the highest contributions—changes in these indices directly modulated the long-term hydrological state of the watershed, corresponding to the "baseline hydrological state" discussed earlier. RX5day, to some extent, reflected the intensity of the most extreme precipitation events in the watershed, representing the "magnitude of extreme precipitation events". The significant contribution of FD0 was noteworthy, as the index was demonstrated to be a key driver of snow and ice melt processes in TP. Remarkably, the upstream-average FD0 contributed 1.3% to the Q max variation, further demonstrating the regulatory effect of upstream snow and ice melt processes on downstream extreme flows through modified water supply conditions. In summary, extreme precipitation and temperature emerged as the dominant factors influencing both flood frequency and magnitude, while drought indices (e.g., PDSI) exerted critical controls by modifying soil hydrological properties and the watershed's baseline hydrological state. Notably, upstream climate extremes significantly regulated downstream flood processes through their effects on snow and ice melt. Discussion Scale-dependent flood responses to changing climate extremes In light of the scale-dependent heterogeneity in the dominant climate extreme indices driving interannual flood variations, our discussion aimed to comprehensively elucidate the distinctive flood response patterns observed across spatial scales. Specifically, we addressed three key questions: the differences in linearity of flood responses between the plateau average scale and other scales; the pronounced east-west differentiation of climate extreme indices driving flood variations across the TP watersheds; and the hierarchical differentiation of flood responses to climate extreme changes across rivers of varying stream orders. First, linear trends in climate extreme indices demonstrated markedly superior explanatory power for average flood changes at the plateau scale compared to other scales, reflecting a quasi-linear flood-climate response pattern at the aggregate level. This phenomenon arose from the statistical aggregation effects of spatial heterogeneity. Watershed-specific forcing fields triggered differentiated runoff responses through complex surface hydrological processes via the spatial heterogeneity of underlying surface parameters such as soil moisture and vegetation cover. For example, soil moisture deficit in arid regions caused most precipitation to be absorbed by the soil, leading to low runoff production efficiency and nonlinear responses, while near-saturated soil conditions in humid regions promoted nearly complete and rapid conversion of precipitation into runoff, generating a significantly linear response pattern (Ran et al., 2022 ; Yang et al., 2025 ). Nevertheless, these differences in nonlinear responses between watersheds tended to cancel each other out during large-scale spatial averaging, thereby ultimately establishing a quasi-linear relationship between precipitation and runoff at the plateau scale (Li and Sivapalan, 2011 ; Liu et al., 2019 ; Yu et al., 2023 ). Second, the climate extreme indices driving flood variations in TP exhibited pronounced east-west differentiation, rooted in the climate-cryosphere coupling effect. The eastern TP, situated within the monsoon zone, was characterized by abundant precipitation and frequent extreme precipitation, coupled with relatively limited glacier and snow cover. Here, extreme precipitation events directly triggered surface runoff, while soil conditions (represented by drought indices) influenced surface water storage capacity—these two factors jointly modulated flood dynamics. Conversely, western TP constituted a cold-arid zone under westerly control. Despite arid climate conditions, the western TP harbored abundant glacial and snow resources, with snow and ice melt serving as the primary source of runoff. Marked warming accelerated snow and ice ablation, while drought indices reflected surface water retention capacity, collectively regulating the conversion efficiency of meltwater to runoff and establishing a flood-driving mechanism dominated by extreme temperatures, snowmelt, and drought. The central TP served as a transitional zone between the monsoon and westerly domains, characterized by moderate precipitation and modest glacier and snow cover. Summer extreme precipitation directly induced flooding here, while spring warming accelerated glaciers and snowmelt, converging with extreme precipitation to generate floods. Consequently, flood variations in this area were jointly driven by extreme precipitation and temperature indices. Meanwhile, the hierarchical differentiation of flood responses to the changes in climate extremes across river systems was fundamentally governed by the interplay of scale-dependent hydrological processes, underlying surface complexity, and climate factor sensitivities. In small tributary watersheds, precipitation events exhibited near-instantaneous conversion to surface runoff, constrained by limited catchment areas and rapid concentration times (McGuire et al., 2005 ; Sivapalan, 2003 ). This process was further amplified by minimal soil water storage capacity, accentuating the immediacy of precipitation-runoff relationships (Kraaijenbrink et al., 2021 ; Veatch et al., 2009 ). Conversely, large mainstem basins demonstrated pronounced spatial heterogeneity, integrating both cryospheric and precipitation-driven processes. Glacier/snow-covered areas sustained baseflow through meltwater contributions, while precipitation interacted with these melt signals across temporal and spatial gradients. Additionally, soil moisture exerted stronger regulatory control on runoff generation as basin size increased. Crucially, climate sensitivity diverged markedly with scale—small watersheds responded predominantly to localized convective precipitation, whereas large basins synchronized with expansive, system-scale precipitation patterns (Dai, 2006 ; Pfahl et al., 2017 ). These scale-dependent mechanisms collectively underscore the pivotal role of basin dimensions in modulating hydrological behavior under climatic extremes. In summary, the pronounced multi-scale differentiation in the response of flood changes to climate extremes arose from differences in watershed hydrological processes, underlying surface complexity, climate factor sensitivities, and climate-cryosphere coupling (Fig. 8 ). This insight not only deepened our understanding of the mechanisms underlying hydrological cycle responses to climate change in alpine regions but also provided critical theoretical foundations and practical guidance for developing cross-scale coupled hydrological disaster models and formulating targeted flood risk management strategies tailored to regional hydrological and climatic characteristics. Conclusion Our study systematically elucidated the multi-scale response characteristics of flood changes to climatic extremes across TP, integrating linear and nonlinear attribution approaches. Quantitative analysis at the plateau-average scale determined the long-term driving contributions of extreme precipitation, temperature, and snowmelt to flood variations in TP, and the watershed-scale analysis revealed a distinct east-west divide in the climate drivers of flood changes. River order-scale investigation uncovered a stepwise evolution of flood-driving mechanisms, and cross-watershed scale analysis confirmed the critical regulatory role of upstream changes in climate extremes on downstream floods. These multi-scale differentiations originated from watershed hydrological processes, underlying surface complexity, climate factor sensitivities, and climate-cryosphere coupling. This work challenges the conventional "one-size-fits-all" analytical paradigm in flood research, offering a novel perspective on the relationship between climate extremes and flood responses. Collectively, these insights provide a robust solid theoretical foundation for constructing multi-scale, dynamic flood risk assessment models and formulating regionally differentiated flood resilience strategies. Materials and methods Study area The Tibetan Plateau, often referred to as the "Roof of the World", is located in south-central Asia. It is a global biodiversity hotspot, a natural habitat for rare wildlife, a gene pool for plateau species, a critical ecological security barrier for China and Asia, a vital water source for the continent, and a region of unique cultural heritage. The study adopted the extent of TP proposed by Zhang et al. (2021) (Fig. 9 a), with its boundaries defined by the following coordinates: northernmost (40°1′6″N, 96°50′5″E), southernmost (25°59′26″N, 98°40′33″E), westernmost (34°58′8″N, 67°40′37″E), and easternmost (33°13′41″N, 104°40′43″E). The region covered an area of approximately 3×10⁶ km². The region exhibits a characteristic layered geomorphic pattern, with distinct topographic units developing sequentially from the marginal mountains toward the interior: alpine valleys, periglacial platforms, and wide lake basins. The plateau's major geomorphic units primarily include the Himalayas, Karakoram Mountains, Gangdise Mountains, Kunlun Mountains, Altun Mountains, Qilian Mountains, Hengduan Mountains, Pamir Plateau, Changtang Plateau, and Qaidam Basin. The plateau's radial river system gives rise to major Asian rivers, such as the Yangtze, Yellow, Mekong, Salween, and Brahmaputra. TP also hosts the world's highest-altitude large lake groups, such as Nam Co and Siling Co. These hydrological factors, combined with widespread permafrost and modern glaciers, form a unique hydrogeological system, whose dynamics critically regulate regional hydrological processes and flood formation mechanisms. TP exhibits diverse climate types shaped by altitude differences and latitude-longitude variations, such as mountain humid climate, subtropical humid climate, and plateau monsoon climate. The TP's climate has shown a distinct warming and wetting trend over the past 40 years. Specifically, average temperature and annual precipitation have increased significantly at rates of 0.25°C/10a and 11.6 mm/10a, respectively (Fig. 9 b, 9 c). Data source Meteorological data The study utilized the high-resolution near-surface meteorological forcing dataset for the Third Pole region (TPMFD) and ERA5-land climate reanalysis data. The TPMFD dataset provides seven key meteorological variables, including precipitation, 2 m air temperature, 2 m specific humidity, 10 m wind speed, surface pressure, downward longwave radiation, and downward shortwave radiation. We primarily employed precipitation, temperature, and wind speed data from TPMFD, which integrated short-term high-resolution Weather Research and Forecasting (WRF) simulations, long-term ERA5 reanalysis, and ground station observations. TPMFD dataset integrates multiple data sources to achieve enhanced accuracy and higher spatial resolution than conventional reanalysis products, establishing an optimal resource for hydro-meteorological studies across the Third Pole region. The ERA5 dataset, as the latest climate reanalysis product from the European Centre for Medium-Range Weather Forecasts (ECMWF), provides detailed records of global atmospheric, land surface, and ocean wave conditions since 1950. The study utilized the ERA5-Land dataset, which focused on the land surface component of ERA5 and offered higher precision for long-term climate element monitoring, with a spatial resolution of 0.1° × 0.1°. Snowmelt data from this dataset were employed in the analysis. Hydrological data Watershed data Our study utilized the MERIT Hydro global high-resolution hydrological dataset (Yamazaki et al., 2019) to extract river network and watershed boundaries. The dataset is derived from the 90-meter resolution MERIT Digital Elevation Model (DEM) and incorporates flow direction correction. River networks are hierarchically classified using the Strahler system, ranging from first-order branches (headwater streams) to seventh-order main rivers within TP, comprehensively covering the entire hydrological network from alpine rivulets to major trunk streams of large basins. The dataset further employs the Pfafstetter system to establish a complete global hierarchical watershed framework with 12 nested levels (L1–L12), enabling continuous spatial coverage from continental-scale basins (L1) to local sub-watersheds (L12). Flood records We employed the most comprehensive flood inventory compiled by the Second Tibetan Plateau Scientific Expedition, integrating multi-source flood records across TP from 1961 to 2020. This dataset constitutes the region's most complete flood event compilation to date, encompassing > 3,000 documented cases derived from systematic literature reviews of academic publications, technical reports, historical archives, hydrometric station observations, and official disaster registries (Fig. S1 ). The data producer implemented a rigorous quality control protocol that involved cross-validation of different data sources, complemented by extensive field verification campaigns, to ensure maximum data reliability. Each flood event in the final curated database is characterized by key attributes, including precise geolocation and temporal occurrence. The multi-method integration of documentary evidence, instrumental measurements, and ground-truthing significantly enhances data accuracy, establishing an unprecedented foundation for robust flood characterization and modeling in this critical region. Discharge data Our research utilized simulated discharge data from the Global 3-Hour River Flood Reanalysis (GRFR) and the Global Flood Awareness System (GLOFAS) historical dataset. The GRFR dataset employed the distributed hydrological model VIC and the river routing model RAPID to construct high-resolution, high-accuracy global natural river discharge simulation data. It covers land surface runoff data at a 0.05° resolution and natural discharge simulations for 2.94 million river segments globally from 1980 to 2019. Accuracy assessments based on daily discharge observations from > 14,000 global stations demonstrate that GRFR effectively reproduces daily-scale runoff processes and performs well in capturing flood events. GLOFAS, developed by the ECMWF, is a global flood awareness and monitoring system designed to provide flood warnings and risk analysis worldwide (Alfieri et al., 2013). The GLOFAS discharge dataset includes daily-scale discharge grid data (unit: m³/s) at 0.1° grid resolution simulated globally using the Lisflood hydrological model. The input data are derived from at least four years of observed discharge data and ERA5 meteorological data, with model outcomes calibrated at 1,226 river sections across 66 countries. Here, we assessed both reanalysis datasets' flood detection across the TP (Fig. S2). The results demonstrated superior flood monitoring capabilities in the GRFR data, with over 80% of historical flood records aligned with to flood days identified by GRFR. Over 70% of flood records documented with daily precision matched flood days detected in adjacent river reaches, while 88% of monthly flood records corresponded to GRFR-derived flood periods. Consequently, GRFR data were selected for subsequent analysis. Methods Climate extreme and flood indices Climate extreme indices This study employed 25 standardized climate extreme indices based on those proposed by the Expert Team on Climate Change Detection and Indices (ETCCDI) (Zhang et al., 2011). These indices, derived from daily temperature and precipitation data, are broadly categorized into two classes: extreme value indices (quantifying the intensity, magnitude, or amplitude of extreme climate events) and day-count indices (reflecting the frequency, duration, or occurrence probability of extreme climate events). The selected indices comprise 12 extreme temperature indices, 10 extreme precipitation indices, and three extreme drought indices (Tab. S1). Extreme snowmelt indices We introduced the concept of extreme snowmelt events and developed corresponding indices, building upon the framework of extreme precipitation and temperature indices. The extreme snowmelt days index was characterized by individual days with snowmelt exceeding the 90th percentile threshold of historical daily values during the 1991–2020 reference period. Continuous extreme snowmelt occurrences were identified when daily snowmelt exceeded the 90th percentile threshold for at least three consecutive days. This threshold was determined using a seasonally varying percentile calculation, where the 90th percentile was computed separately for 5-day moving windows to account for the strong seasonal cycle in snowmelt processes. Flood indices Besides, the Peak Over Threshold (POT) method, grounded in Extreme Value Theory (EVT), provides a robust statistical framework for identifying extreme events that exceed predetermined thresholds (Bezak et al., 2014). This method operates by setting an appropriate threshold and considering only values exceeding this threshold as extreme events. In our study, we selected the 90th percentile of daily discharge as the flood identification threshold after consideration of the region's complex terrain and highly variable climate conditions. Historical flood records confirmed that this threshold effectively captured most documented flood events while successfully filtering out low-intensity, non-flood occurrences. Additionally, a minimum discharge threshold of 10 m³/s was established to exclude low-flow periods from flood identification. To characterize historical flood patterns, we employed two complementary indices: flood days and annual maximum daily discharge (Q max ). Flood days quantified flood frequency by counting the number of days exceeding flood thresholds within a given period, while Q max captures flood magnitude by identifying the peak discharge event each year (Jiang et al., 2024 ). These established hydrological indicators represent both the occurrence rate and intensity of flood events across watersheds. Machine learning methods Our research employed multiple machine learning methods—including Bootstrap Forest, Boosted Trees, Support Vector Machines (SVM), K-Nearest Neighbors (KNN), and Neural Networks—for attribution analysis of water bodies and flood changes. These methods effectively captured nonlinear and deep-level relationships between climate extreme indices and flood changes when dealing with complex nonlinear relationships and high-dimensional data. Each algorithm provided distinct analytical advantages, as detailed below. Bootstrap Forest Bootstrap Forest is an ensemble learning method, a variant of Random Forest (Breiman, 2001). It enhances model robustness and accuracy by training multiple decision trees using bootstrap sampling (i.e., random sampling with replacement) of training data. Compared to traditional decision trees, Bootstrap Forest reduces the risk of overfitting and improves prediction accuracy by integrating multiple sub-models. In our study, Bootstrap Forest effectively revealed the relative importance of different climate extreme indices on flood changes and provided more reliable conclusions by aggregating predictions from multiple models through a voting mechanism. Simultaneously, we conducted a feature importance analysis to reveal the relative importance of different climate indices to the model. Specifically, we calculated both the frequency of each predictor variable being selected as splitting nodes across all decision trees and their corresponding sum of squared reductions. These values were then normalized to determine the relative contribution rate of each index. The method effectively captured nonlinear relationships and interaction effects among variables but also improved the stability of contribution assessment through ensemble learning. Boosted Trees Boosted Trees is an algorithm that builds a strong learner by integrating multiple weak learners (decision trees) (Friedman, 2001). The core idea is to iteratively strengthen samples that are incorrectly predicted in previous rounds through weighted adjustments. Compared to traditional decision trees, the Boosted Trees method excels in handling complex nonlinear relationships and continuously improves model prediction accuracy by optimizing loss functions. In our study, Boosted Trees method was employed to identify key climate extreme indices influencing flood changes and to handle complex data with nonlinear characteristics, thereby enhancing the accuracy and robustness of attribution analysis. Support Vector Machines (SVM) SVM is a widely used supervised learning method, particularly suitable for classification and regression tasks (Cortes & Vapnik, 1995). SVM constructs a hyperplane that maximizes the margin to classify data points or maps data to a higher-dimensional space using kernel tricks to solve nonlinear problems. The advantage of SVM is its ability to handle high-dimensional data and achieve good performance even with small datasets. SVM effectively adapts to the different characteristics of datasets by adjusting kernel functions and penalty parameters, providing reliable results in flood change attribution analysis. In our study, SVM was used for regression tasks to fit flood days and Q max . K-Nearest Neighbors (KNN) KNN is a simple yet effective non-parametric supervised learning method suitable for classification and regression tasks (Cover & Hart, 1967). The core idea of KNN is to determine the predicted value of an input sample based on its distance to other samples in the training data, commonly using Euclidean or Manhattan distance metrics. In the study, KNN was used for regression analysis to fit hydrological indicators such as flood days. Neural Networks The neural network trains several neural network models and optimizes them using boosting algorithms to create an integrated strong prediction model (Rumelhart et al., 1986). This method is suitable for prediction tasks involving complex, nonlinear relationships, especially those involving high-dimensional data or problems that are difficult to model with traditional statistical methods. In the study, the method was used to model the complex relationships between climate extreme indices and flood changes. Regression methods and supplementary variables for attribution analysis Our study employed a suite of regression models to systematically analyze the associations between plateau floods and climate extremes, uncovering response characteristics across multiple scales through variable selection, trend quantification, and attribution diagnostics. First, linear regression analysis was applied to multi-decadal records of flood frequency and magnitude to quantify temporal trends (Shi & Neng, 1995). The regression coefficients, along with their 95% confidence intervals (95% CI) and statistical significance levels ( p -values) were systematically examined to assess the direction (increasing or decreasing) and magnitude of trends in key variables and whether these trends reached statistical significance (p < 0.05). This analysis helped reveal the characteristics of flood changes in TP. Second, we quantitatively evaluated the linear impact of climate extreme indices on regional flood changes using a stepwise regression method. Stepwise regression is a variable selection method used in regression analysis to optimize model performance by iteratively adding or removing explanatory variables. The method automatically selects the most significant factors for modeling by evaluating each variable's contribution to the model's explanatory power. We implemented a dual-criterion approach using both the corrected Akaike Information Criterion (AICc) and Bayesian Information Criterion (BIC) for variable selection. The combined strategy ensures an optimal balance between model fit and complexity. The AICc, which adjusts for small sample sizes, minimizes prediction error while preventing oversimplification by accounting for both model accuracy and parameter count. However, AICc's relatively mild penalty for complexity may occasionally permit marginally significant variables to enter the model. In contrast, the BIC imposes stronger penalties on model complexity, serving as a robust safeguard against overfitting. Third, the Least Absolute Shrinkage and Selection Operator (LASSO) regression was used to detect the main climate extreme indices governing changes in flood days and Qmax in different watersheds in TP. LASSO regression is particularly useful for handling high-dimensional data, as it maintains model predictive performance while reducing complexity and avoiding overfitting. The method performs variable selection and model simplification by applying L1 regularization (i.e., adding a penalty term for the sum of absolute coefficients) (Tibs hirani, 2011). The core idea of LASSO is to shrink some regression coefficients to zero through regularization, thereby retaining key variables while automatically eliminating less important ones. Our study further reduced the number of variables in the LASSO model based on the 1SE rule (one standard error rule), making the final model more interpretable and robust. The 1SE rule is a common strategy for selecting regularization strength (i.e., the λ value in LASSO). It suggests choosing the largest λ value where the mean squared error (MSE) falls within one standard error of the minimum MSE. This approach achieves a better balance between model accuracy and complexity, resulting in a simpler yet sufficiently predictive model. Notably, existing attribution studies of regional hydrological changes have predominantly relied on spatially averaged climate variables, neglecting the critical influence of intra-regional spatial heterogeneity. However, spatial distribution patterns of climate extreme indices play a significant role in modulating hydrological processes, particularly flood generation and propagation. To address the gap, we proposed incorporating spatial statistical indicators (e.g., maximum, minimum, and variance) of independent variables as supplementary variables in attribution analysis. Comparative experiments demonstrated that watershed-maximum values of key meteorological parameters (e.g., annual precipitation, R10, and PDSI) significantly enhances the performance of flood attribution models. Equally importantly, upstream changes in climate extremes can significantly influence downstream flood events due to the spatial propagation of flood-generating processes. Hydrological connectivity ensures that extreme precipitation, snowmelt, or drought conditions in upstream areas alter downstream runoff patterns and flood frequency. We thus incorporated spatially averaged climate extreme indices from upstream watersheds as additional variables to account for this mechanism in our cross-watershed flood attribution analysis. This approach explicitly captured the spatial transmission of upstream climate extremes on flood dynamics, offering a framework for assessing the interplay of regional climate variability and hydrological processes. Declarations Competing intersts All authors declare no financial or non-financial competing interests. Author contributions C.P., L.X., and, Z.X. initiated the research. L.X. and Z.X. developed the methods. L.X. wrote the manuscript. Z.X. and S.P. revised the manuscript. C.P. and Z.X. acquired the funding and supervised the study. All authors read and approved the final manuscript. Acknowledgments This work was supported by the Science and Technology Projects of Xizang Autonomous Region, China (XZ202402ZD0001, XZ202401ZY0108), and the Special Project for the Construction of Nyingchi National Sustainable Development Pilot Zone, Xizang Autonomous Region (2023-SYQ-006). Data availability The reanalysis discharge, meteorological, and watershed data sets used in this study are publicly available. The reanalysis discharge data sets are from the GRFR ( https://doi.org/10.11888/Terre.tpdc.272901 ) and GLOFAS ( https://data.jrc.ec.europa.eu/dataset/73305ca5-c002-4124-b1d5-6451cc93af3f ). TPMFD dataset is from National Tibetan Plateau ( https://doi.org/10.11888/Atmos.tpdc.300398 ). The ERA5-Land dataset is from the European Centre for Medium-Range Weather Forecasts (ECMWF) ( https://cds.climate.copernicus.eu/cdsapp#!/dataset/reanalysis-era5-land?tab=overview ). The flood record data is available from the Second Tibetan Plateau Scientific Expedition and Research Program (STEP) but restrictions apply to the availability of these data, which were used under licence for the current study, and so are not publicly available. Data are however available from the authors upon reasonable request and with permission of STEP. References Bai, Y., Li, D., Wangchuk, S., Kettner, A., Zhao, Y., Deng, R., Liu, Y., Xiao, C., Ni, J., Cui, P., 2025. Flood complexity and rising exposure risk in High Mountain Asia under climate change. Science Bulletin S2095927325001355. https://doi.org/10.1016/j.scib.2025.01.055 Balaian, S.K., Sanders, B.F., Abdolhosseini Qomi, M.J., 2024. How urban form impacts flooding. Nat Commun 15, 6911. https://doi.org/10.1038/s41467-024-50347-4 Bormann, K.J., Brown, R.D., Derksen, C., Painter, T.H., 2018. Estimating snow-cover trends from space. Nature Clim Change 8, 924–928. https://doi.org/10.1038/s41558-018-0318-3 Brunner, M.I., Swain, D.L., Wood, R.R., Willkofer, F., Done, J.M., Gilleland, E., Ludwig, R., 2021. An extremeness threshold determines the regional response of floods to changes in rainfall extremes. Commun Earth Environ 2, 173. https://doi.org/10.1038/s43247-021-00248-x Dai, A., 2006. Precipitation Characteristics in Eighteen Coupled Climate Models. Journal of Climate 19, 4605–4630. https://doi.org/10.1175/JCLI3884.1 Gao, Y., Lu, N., Dai, Y., Yao, T., 2023. Reversal in snow mass trends on the Tibetan Plateau and their climatic causes. Journal of Hydrology 620, 129438. https://doi.org/10.1016/j.jhydrol.2023.129438 Harrison, S., Macklin, M.G., Toonen, W.H.J., Benito, G., Cohen, K.M., 2025. Robust climate attribution of modern floods needs palaeoflood science. Climatic Change 178, 71. https://doi.org/10.1007/s10584-025-03904-9 IPCC, 2023. Climate Change 2022 – Impacts, Adaptation and Vulnerability: Working Group II Contribution to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change, 1st ed. Cambridge University Press. https://doi.org/10.1017/9781009325844 Jiang, S., Tarasova, L., Yu, G., Zscheischler, J., 2024. Compounding effects in flood drivers challenge estimates of extreme river floods. Sci. Adv. 10, eadl4005. https://doi.org/10.1126/sciadv.adl4005 Kraaijenbrink, P.D.A., Stigter, E.E., Yao, T., Immerzeel, W.W., 2021. Climate change decisive for Asia’s snow meltwater supply. Nat. Clim. Chang. 11, 591–597. https://doi.org/10.1038/s41558-021-01074-x Kundzewicz, Z.W., Kanae, S., Seneviratne, S.I., Handmer, J., Nicholls, N., Peduzzi, P., Mechler, R., Bouwer, L.M., Arnell, N., Mach, K., Muir-Wood, R., Brakenridge, G.R., Kron, W., Benito, G., Honda, Y., Takahashi, K., Sherstyukov, B., 2014. Flood risk and climate change: global and regional perspectives. Hydrological Sciences Journal 59, 1–28. https://doi.org/10.1080/02626667.2013.857411 Li, H., Sivapalan, M., 2011. Effect of spatial heterogeneity of runoff generation mechanisms on the scaling behavior of event runoff responses in a natural river basin. Water Resources Research 47, 2010WR009712. https://doi.org/10.1029/2010WR009712 Li, X., Cui, P., Zhang, X.-Q., Zhang, F., 2024. Intensified warming suppressed the snowmelt in the Tibetan Plateau. Advances in Climate Change Research 15, 452–463. https://doi.org/10.1016/j.accre.2024.06.005 Liu, J., Engel, B.A., Wang, Y., Wu, Y., Zhang, Z., Zhang, M., 2019. Runoff Response to Soil Moisture and Micro-topographic Structure on the Plot Scale. Sci Rep 9, 2532. https://doi.org/10.1038/s41598-019-39409-6 McGuire, K.J., McDonnell, J.J., Weiler, M., Kendall, C., McGlynn, B.L., Welker, J.M., Seibert, J., 2005. The role of topography on catchment‐scale water residence time. Water Resources Research 41, 2004WR003657. https://doi.org/10.1029/2004WR003657 Min, S.-K., Zhang, X., Zwiers, F.W., Hegerl, G.C., 2011. Human contribution to more-intense precipitation extremes. Nature 470, 378–381. https://doi.org/10.1038/nature09763 Otto, F.E.L., 2023. Attribution of Extreme Events to Climate Change. Annu. Rev. Environ. Resour. 48, 813–828. https://doi.org/10.1146/annurev-environ-112621-083538 Park, I.-H., Min, S.-K., 2017. Role of convective precipitation in the relationship between subdaily extreme precipitation and temperature. J. Climate 30, 9527–9537. https://doi.org/10.1175/JCLI-D-17-0075.1 Peel, M.C., McMahon, T.A., Finlayson, B.L., Watson, F.G.R., 2002. Implications of the relationship between catchment vegetation type and the variability of annual runoff. Hydrological Processes 16, 2995–3002. https://doi.org/10.1002/hyp.1084 Pfahl, S., O’Gorman, P.A., Fischer, E.M., 2017. Understanding the regional pattern of projected future changes in extreme precipitation. Nature Clim Change 7, 423–427. https://doi.org/10.1038/nclimate3287 Ran, Q., Wang, J., Chen, X., Liu, L., Li, J., Ye, S., 2022. The relative importance of antecedent soil moisture and precipitation in flood generation in the middle and lower Yangtze River basin. Hydrol. Earth Syst. Sci. 26, 4919–4931. https://doi.org/10.5194/hess-26-4919-2022 Scussolini, P., Luu, L.N., Philip, S., Berghuijs, W.R., Eilander, D., Aerts, J.C.J.H., Kew, S.F., Van Oldenborgh, G.J., Toonen, W.H.J., Volkholz, J., Coumou, D., 2024. Challenges in the attribution of river flood events. WIREs Climate Change 15, e874. https://doi.org/10.1002/wcc.874 Sivapalan, M., 2003. Process complexity at hillslope scale, process simplicity at the watershed scale: is there a connection? Hydrological Processes 17, 1037–1041. https://doi.org/10.1002/hyp.5109 Torre Zaffaroni, P., Baldi, G., Texeira, M., Di Bella, C.M., Jobbágy, E.G., 2023. The Timing of Global Floods and Its Association With Climate and Topography. Water Resources Research 59, e2022WR032968. https://doi.org/10.1029/2022WR032968 Veatch, W., Brooks, P.D., Gustafson, J.R., Molotch, N.P., 2009. Quantifying the effects of forest canopy cover on net snow accumulation at a continental, mid‐latitude site. Ecohydrology 2, 115–128. https://doi.org/10.1002/eco.45 Wang, X., Chen, R., Li, H., Li, K., Liu, J., Liu, G., 2022. Detection and attribution of trends in flood frequency under climate change in the Qilian Mountains, Northwest China. Journal of Hydrology: Regional Studies 42, 101153. https://doi.org/10.1016/j.ejrh.2022.101153 Wang, Y., Zheng, D., Zhang, G., Carrivick, J.L., Bolch, T., Ren, W., Guo, L., Su, J., Yuan, S., Li, X., 2025. Patterns and change rates of glacial lake water levels across High Mountain Asia. National Science Review 12, nwaf041. https://doi.org/10.1093/nsr/nwaf041 Wu, B., Zhang, Z., Guo, X., Tan, C., Huang, C., Tao, J., 2022. Spatial and Temporal Analysis of Quantitative Risk of Flood due to Climate Change in a China’s Plateau Province. Front. Earth Sci. 10, 931505. https://doi.org/10.3389/feart.2022.931505 Yang, H., Zhang, Xiaoqi, Yuan, Z., Hong, X., Yao, L., Zhang, Xiuping, 2025. Investigating the effects of spatial heterogeneity of multi-source profile soil moisture on spatial–temporal processes of high-resolution floods. Journal of Hydrology 652, 132672. https://doi.org/10.1016/j.jhydrol.2025.132672 Yang, Keke, Guo, D., Hua, W., Pepin, N., Yang, Kun, Li, D., 2022. Tibetan Plateau temperature extreme changes and their elevation dependency from ground‐based observations. JGR Atmospheres 127, e2021JD035734. https://doi.org/10.1029/2021JD035734 Yao, Y., Wang, Y., Chen, Q., Liao, Y., 2025. Temporal and spatial variation characteristics of extreme precipitation in central and eastern Tibetan Plateau. Journal of Arid Meteorology 41, 714–722. https://doi.org/10.11755/j.issn.1006-7639（2023）-05-0714 Yin, H., Sun, Y., Donat, M.G., 2019. Changes in temperature extremes on the Tibetan Plateau and their attribution. Environ. Res. Lett. 14, 124015. https://doi.org/10.1088/1748-9326/ab503c Yin, H., Sun, Y., Wan, H., Zhang, X., Lu, C., 2017. Detection of anthropogenic influence on the intensity of extreme temperatures in China. Int. J. Climatol. 37, 1229–1237. https://doi.org/10.1002/joc.4771 Yu, T., Ran, Q., Pan, H., Li, J., Pan, J., Ye, S., 2023. The impacts of rainfall and soil moisture to flood hazards in a humid mountainous catchment: a modeling investigation. Front. Earth Sci. 11, 1285766. https://doi.org/10.3389/feart.2023.1285766 Zhang, G., Carrivick, J.L., Emmer, A., Shugar, D.H., Veh, G., Wang, X., Labedz, C., Mergili, M., Mölg, N., Huss, M., Allen, S., Sugiyama, S., Lützow, N., 2024. Characteristics and changes of glacial lakes and outburst floods. Nat Rev Earth Environ 5, 447–462. https://doi.org/10.1038/s43017-024-00554-w Zhang, S., Zhou, L., Zhang, L., Yang, Y., Wei, Z., Zhou, S., Yang, D., Yang, X., Wu, X., Zhang, Y., Li, X., Dai, Y., 2022. Reconciling disagreement on global river flood changes in a warming climate. Nat. Clim. Chang. 12, 1160–1167. https://doi.org/10.1038/s41558-022-01539-7 Additional Declarations There is NO Competing Interest. Supplementary Files SIn.pdf Supporting information MLreportingsummary14.pdf Reporting Summary ML nrreportingsummary12.pdf Reporting Summary Cite Share Download PDF Status: Published Journal Publication published 18 Mar, 2026 Read the published version in Communications Earth & Environment → Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7153596","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":495936670,"identity":"12898447-057b-4097-8ee2-8c29497106e6","order_by":0,"name":"Xueqin Zhang","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA3klEQVRIiWNgGAWjYDACCQglx0ayFmPStSQ2EK1Dfnbzs4df26zT+/iXP3tcuYdBnr+B+dkDfFoY5xwzN5Y5k57bJvHG3PDMMwbDGQfYzA3waWGWSDCTlqg4DNRyhk2y4QAD4wYGHjYJfFrYJNK/SUsYHE5nkzj+DKTFnqAWHokcM8kPFYcT2PgbzEBaEglqkZDIKZNmOJNu2CbBA9IikTzjMJsZXi3yM9K3Sf5ss5aX7wc7zMa2v735GV4tIMDMw8AMtC8BbCsDiE0QMP4AKeM/QITSUTAKRsEoGJEAAMjlPrDQ7quQAAAAAElFTkSuQmCC","orcid":"https://orcid.org/0000-0003-3049-6492","institution":"Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences","correspondingAuthor":true,"prefix":"","firstName":"Xueqin","middleName":"","lastName":"Zhang","suffix":""},{"id":495936671,"identity":"408f6565-0ea7-43a9-a868-8978759e0441","order_by":1,"name":"Xiang Li","email":"","orcid":"","institution":"Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences","correspondingAuthor":false,"prefix":"","firstName":"Xiang","middleName":"","lastName":"Li","suffix":""},{"id":495936672,"identity":"3cd72648-f897-41e6-b070-2dea17dc3359","order_by":2,"name":"Peng Cui","email":"","orcid":"","institution":"Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences","correspondingAuthor":false,"prefix":"","firstName":"Peng","middleName":"","lastName":"Cui","suffix":""},{"id":495936673,"identity":"4098bab1-cd1c-40c7-9726-b01543f7f6f0","order_by":3,"name":"Pengke Shen","email":"","orcid":"","institution":"China Meteorological Administration","correspondingAuthor":false,"prefix":"","firstName":"Pengke","middleName":"","lastName":"Shen","suffix":""}],"badges":[],"createdAt":"2025-07-18 03:55:37","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7153596/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7153596/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1038/s43247-026-03413-2","type":"published","date":"2026-03-18T04:00:00+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":90046179,"identity":"825913ae-1709-4571-af86-f38358503571","added_by":"auto","created_at":"2025-08-27 18:21:46","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":252313,"visible":true,"origin":"","legend":"\u003cp\u003eSpatiotemporal changes of flood days and Q\u003csub\u003emax\u003c/sub\u003e in TP from 1980 to 2019: average flood days (a) and Q\u003csub\u003emax\u003c/sub\u003e (c) change in TP; flood days (a) and Q\u003csub\u003emax\u003c/sub\u003e (c) change in different watersheds; average flood days and Q\u003csub\u003emax\u003c/sub\u003e changes of different river orders in TP.\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-7153596/v1/d31e147f632f7e2717d99f3b.png"},{"id":90046186,"identity":"7fde1f61-9e3c-4a28-b6dc-5eb6b7fe35da","added_by":"auto","created_at":"2025-08-27 18:21:47","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":86636,"visible":true,"origin":"","legend":"\u003cp\u003eThe driving effects of climate extreme indices on changes in annual flood day in TP from 1980 to 2019: contribution of different types of climate extreme indices (a) and key climate extreme indices (b).\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-7153596/v1/7d040e2516b7472dd39b7f92.png"},{"id":90046183,"identity":"7b15065f-9ee6-4600-875b-c4be7294527a","added_by":"auto","created_at":"2025-08-27 18:21:47","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":83866,"visible":true,"origin":"","legend":"\u003cp\u003eThe driving effects of climate extreme indices on changes in Q\u003csub\u003emax\u003c/sub\u003e in TP from 1980 to 2019: contribution of different types of climate extreme indices (a) and key climate extreme indices (b).\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-7153596/v1/443299f638e46973432a110c.png"},{"id":90046903,"identity":"8eec1c1e-c503-4215-9cc3-14ed9124c8ac","added_by":"auto","created_at":"2025-08-27 18:37:47","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":373531,"visible":true,"origin":"","legend":"\u003cp\u003eThe main drivers of changes in flood day (a) and Q\u003csub\u003emax\u003c/sub\u003e (b) across different watersheds in the TP from 1980 to 2019\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-7153596/v1/d3447ed274ba2b10f118718f.png"},{"id":90046196,"identity":"6b41a601-0d88-4717-bf6d-a9d161efcd5b","added_by":"auto","created_at":"2025-08-27 18:21:47","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":189335,"visible":true,"origin":"","legend":"\u003cp\u003eThe driving effects of climate extreme indices on flood changes in different river orders\u003c/p\u003e","description":"","filename":"5.png","url":"https://assets-eu.researchsquare.com/files/rs-7153596/v1/b8abb71b06d1837fc378c20f.png"},{"id":90046902,"identity":"382d83cd-663a-4b69-ba7a-02679aac44ba","added_by":"auto","created_at":"2025-08-27 18:37:47","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":127187,"visible":true,"origin":"","legend":"\u003cp\u003eCross-watershed results for the attribution of flood changes: model comparison (a); fitting results of Bootstrap Forest model (b); contribution of different kinds of climate extreme indices (c), spatial statistical indicators (d), and key climate extreme indices (e).\u003c/p\u003e","description":"","filename":"6.png","url":"https://assets-eu.researchsquare.com/files/rs-7153596/v1/497ae72be8907f610890f3e4.png"},{"id":90046182,"identity":"ccc56366-5232-47a6-8399-b04618318212","added_by":"auto","created_at":"2025-08-27 18:21:47","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":131427,"visible":true,"origin":"","legend":"\u003cp\u003eCross-watershed attribution results for Q\u003csub\u003emax\u003c/sub\u003e changes: model comparison (a); fitting results of Bootstrap Forest model (b); contribution of different kinds of climate extreme indices (c), spatial statistical indicators (d), and key climate extreme indices (e).\u003c/p\u003e","description":"","filename":"7.png","url":"https://assets-eu.researchsquare.com/files/rs-7153596/v1/6b94b84b6c00c6a8d9dfddba.png"},{"id":90046200,"identity":"dc454aad-d80d-4833-9f1e-6c1124559ba8","added_by":"auto","created_at":"2025-08-27 18:21:47","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":425546,"visible":true,"origin":"","legend":"\u003cp\u003eDiagram of scale-different impacts of climatic extremes on floods in TP, with a background inspired by the Hainei Huayi Tu (Map of China and Foreign Lands)—one of the China’s earliest images depicting the distribution of hydrological factors\u003c/p\u003e","description":"","filename":"8.png","url":"https://assets-eu.researchsquare.com/files/rs-7153596/v1/31ff6064f7c39eeb5228b558.png"},{"id":90046202,"identity":"6e283d47-87df-426b-9577-f083339800e7","added_by":"auto","created_at":"2025-08-27 18:21:47","extension":"png","order_by":9,"title":"Figure 9","display":"","copyAsset":false,"role":"figure","size":454751,"visible":true,"origin":"","legend":"\u003cp\u003eGeographical overview and climate change of the TP: the boundaries, elevation, major rivers, and key topographic units (a); the temperature (b) and precipitation changes(c).\u003c/p\u003e","description":"","filename":"9.png","url":"https://assets-eu.researchsquare.com/files/rs-7153596/v1/6c9e762636f6c60bc565bad3.png"},{"id":104952650,"identity":"9e18a2e9-c793-445c-bae0-6f5cc67608d9","added_by":"auto","created_at":"2026-03-19 07:13:39","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2628691,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7153596/v1/6e8b6b66-5c25-49cd-9dd9-bfc2b6c95466.pdf"},{"id":90046180,"identity":"2df551cc-827a-41d4-8055-511978a01aed","added_by":"auto","created_at":"2025-08-27 18:21:46","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":425213,"visible":true,"origin":"","legend":"Supporting information","description":"","filename":"SIn.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7153596/v1/8ab69cdde028fa294d7c6e74.pdf"},{"id":90046686,"identity":"69683552-76a9-45c9-97d1-176811a1c7f6","added_by":"auto","created_at":"2025-08-27 18:29:47","extension":"pdf","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":362673,"visible":true,"origin":"","legend":"Reporting Summary ML","description":"","filename":"MLreportingsummary14.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7153596/v1/8f817cc1d04908b7e102122e.pdf"},{"id":90046194,"identity":"388b6708-6b97-45a1-a2bf-c1ff77afe8a0","added_by":"auto","created_at":"2025-08-27 18:21:47","extension":"pdf","order_by":3,"title":"","display":"","copyAsset":false,"role":"supplement","size":1666965,"visible":true,"origin":"","legend":"Reporting Summary","description":"","filename":"nrreportingsummary12.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7153596/v1/b4dc616e2ded551be232ddd4.pdf"}],"financialInterests":"There is \u003cb\u003eNO\u003c/b\u003e Competing Interest.","formattedTitle":"Unraveling scale-dependent flood responses to changing climate extremes over the Tibetan Plateau","fulltext":[{"header":"Introduction","content":"\u003cp\u003ePlateau regions have emerged as critical hotspots for extreme flood events under accelerated global warming. The IPCC Sixth Assessment Report indicates that flood risk in high-altitude areas has increased by 2\u0026ndash;3 times compared to low-altitude regions (IPCC, \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2023\u003c/span\u003e), with the \"high-altitude amplification effect\" being particularly prominent in high mountain areas (Bai et al., \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Wang et al., \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). Diverging from traditional flood patterns, plateau floods driven by climate extremes exhibited three emerging characteristics: enhanced flash flood occurrences in small watersheds, increased frequency of precipitation-snowmelt compound floods, and a growing proportion of glacial lake outburst floods (GLOFs) (Kundzewicz et al., \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2014\u003c/span\u003e; Wu et al., \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Wang et al., \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2022\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eThe complex shifts in climate extremes on the plateau and its cascading impacts on the cryosphere precipitated a new flood regime. Take TP as an illustrative example, it has experienced significant alterations in extreme precipitation patterns over recent decades, characterized by concurrent increases in total precipitation, frequency and intensity of extreme precipitation (Yao et al., \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). These pronounced changes in extreme precipitation directly drive more frequent flash floods in small watersheds. Concurrently, the frequency and intensity of extreme high-temperature events has increased, while extreme cold events declined sharply (Yang et al., \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Yin et al., \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e2017\u003c/span\u003e, \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). The divergence triggered a chain reaction in the cryosphere system: although the overall snow depth, snow cover, and total snow accumulation decreased (Bormann et al., \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2018\u003c/span\u003e; Gao et al., \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2023\u003c/span\u003e), the snowmelt and the frequency of extreme snowmelt increased locally, leading to more precipitation-snowmelt compound floods (Li et al., \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). Additionally, accelerated warming exacerbated glacial retreat across TP, causing an increase in the area of glacial lakes by 22%/10a (1990\u0026ndash;2020) and a consequent rise in the risk of GLOFs (Zhang et al., \u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e2024\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eWhile a clear connection existed between evolving climate extremes and flood regimes on the plateau, the quantitative characterization of their coupling mechanisms remained a persistent scientific challenge (Harrison et al., \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Otto, \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Scussolini et al., \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). Current understandings of flood change attribution faced limitations, particularly regarding the scale-dependent disparities in flood responses to climate extremes. At the global scale, warming amplified atmospheric water-holding capacity, leading to a robust increase in the frequency and intensity of extreme precipitation (Min et al., \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2011\u003c/span\u003e; Park and Min, \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2017\u003c/span\u003e). However, no pronounced long-term trend in river flood magnitude was observed globally due to offsetting effects between different flood types. For instance, increasing precipitation-driven floods were counterbalanced by decreasing trend in snowmelt-induced floods (Zhang et al., \u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). At the basin scale, flood responses to climate change were governed by basin size and underlying surface conditions (Brunner et al., \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). In small watersheds, short flow concentration paths and limited regulatory capacities resulted in a tight coupling between precipitation and runoff, enhancing sensitivity to extreme precipitation (McGuire et al., \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2005\u003c/span\u003e; Sivapalan, \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2003\u003c/span\u003e). Large basins, with their extensive catchment areas, exhibited flood processes more regulated by antecedent basin wetness. Additionally, surface characteristics further shaped flood response patterns (Torre Zaffaroni et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2023\u003c/span\u003e): basins with abundant vegetation exhibited stronger adaptability to climate extremes through canopy interception and soil infiltration (Peel et al., \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2002\u003c/span\u003e), while highly urbanized areas amplified flood risks from extreme precipitation due to expanded impervious surfaces (Balaian et al., \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2024\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eOur study systematically explored the scale-dependent response characteristics of plateau floods to changes in climate extremes, focusing on four hierarchical scales: plateau-wide, watershed, river hierarchy, and cross-watershed dynamics. To address divergent linear and nonlinear response patterns of floods to climate extremes across scales, we devised targeted research strategies: classical regression methods for quasi-linear flood responses at plateau and watershed scales, and a machine learning-based framework for complex nonlinear responses at cross-watershed scales. Unlike existing machine learning-based attribution approaches (Jiang et al., \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2024\u003c/span\u003e), our method featured two key innovations. Firstly, we incorporated spatially explicit variables enabling a more accurate representation of heterogeneous distribution of climate extreme indices across watersheds. Secondly, we established a hydrological connectivity transmission mechanism that can quantify cascading effects of upstream climate extremes on downstream floods through river network system. The proposed multi-scale analytical framework established a generalizable methodological paradigm for flood attribution and adaptation strategy development in global alpine regions.\u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003e\u003cb\u003eSpatio-temporal changes in floods\u003c/b\u003e\u003c/p\u003e\u003cp\u003eAnnual flood days were 6.99 d from 1980 to 2019 in TP, with maximum (13.45 d) and minimum (4.72 d) occurring in 2019 and 1997 respectively (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003ea). A significant increasing trend (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.05) of 0.7 d/10a was observed over the past 40 years. M-K mutation analysis identified 2016 as a turning point for flood days: flood days fluctuated moderately without significant trends during 1980\u0026ndash;2016, followed by rapid increases post-2016. Spatially, most watersheds experienced increased annual flood days across the plateau in the past 40 years. In particular, flood days increased significantly in northern TP (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eb); while regions with reduced flood days were concentrated in southwest, northwest, and southeastern regions. Notably, the most substantial reductions in flood frequency occurred near the Himalayan range and northern Pamir Plateau. Besides, the flood day change patterns of rivers at different orders revealed significant hierarchical differentiation: the average flood days of first-order rivers showed no considerable change trend over the past 40 years, whereas the increases in flood days presented an obvious stepwise enhancement with the rise in river order. Average flood days for second- to seventh-order rivers rose significantly (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.05) at rates of 0.5, 1.8, 2.7, 3.5, 3.8, and 5.0 d/10a, respectively (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003ee).\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eAdditionally, Q\u003csub\u003emax\u003c/sub\u003e served as a key indicator of the intensity of the strongest flood event in each river annually, averaged 142.15 m\u0026sup3;/s during 1980\u0026ndash;2019, peaking in 2019 (189.55 m\u0026sup3;/s) and troughing in 1997 (126.04 m\u003csup\u003e3\u003c/sup\u003e/s). Though the overall Q\u003csub\u003emax\u003c/sub\u003e was non-significant trend (2.1 m\u003csup\u003e3\u003c/sup\u003e/s/10a), its temporal variation closely mirrored average flood days (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003ec): stable during 1980\u0026ndash;2016 followed by post-2016 acceleration. This consistency between Q\u003csub\u003emax\u003c/sub\u003e and flood day changes indicated that floods became not only more frequent but also more intense. Spatially, Q\u003csub\u003emax\u003c/sub\u003e increased across most plateau watersheds from 1980\u0026ndash;2019, with particularly significant rises in the Hindu Kush Mountains region and the central/northwestern Kunlun Mountains (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003ed). In contrast, significant declines in Q\u003csub\u003emax\u003c/sub\u003e were observed in watersheds of the northern Pamir Plateau, south of the western Kunlun Mountains, and near the eastern Himalayas. Additionally, the average Q\u003csub\u003emax\u003c/sub\u003e of first- to seventh-order rivers showed no significant change trends (\u0026minus;\u0026thinsp;0.2, \u0026minus;\u0026thinsp;0.4, \u0026minus;\u0026thinsp;1.5, and \u0026minus;\u0026thinsp;1.2 m\u0026sup3;/s/10a; \u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026gt;\u0026thinsp;0.05), whereas fifth- to seventh-order rivers exhibited increases of 35.8, 62.9, and 152.8 m\u0026sup3;/s/10a respectively (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003ef).\u003c/p\u003e\u003cp\u003e\u003cb\u003eMulti-scale attribution of flood change\u003c/b\u003e\u003c/p\u003e\u003cp\u003e\u003cb\u003ePlateau-average scale\u003c/b\u003e\u003c/p\u003e\u003cp\u003eWe compared model performance in analyzing the impacts of changes in climate extremes on average flood days and Q\u003csub\u003emax\u003c/sub\u003e across the TP. The analysis incorporates 33 climatic parameters (25 climate extreme indices in Tab. S1, plus average/maximum/minimum temperature, annual precipitation, annual snowfall, annual snowmelt, extreme snowmelt days, and continuous extreme snowmelt occurrences). For these 33 parameters, the average and maximum values of each watershed are calculated, generating a total of 66 independent variables, with the average flood days and Q\u003csub\u003emax\u003c/sub\u003e of plateau rivers as dependent variables. Results demonstrated that linear models achieved significantly higher explanatory power than machine learning models given identical factor selections. Stepwise regression performed best among linear models (flood days: R\u0026sup2; \u0026gt;0.9; Q\u003csub\u003emax\u003c/sub\u003e: R\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;\u0026gt;\u0026thinsp;0.8). By contrast, the bootstrap forest method was optimal performance among nonlinear machine learning models (flood days: R\u0026sup2; = 0.80; Q\u003csub\u003emax\u003c/sub\u003e: R\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;=\u0026thinsp;0.61). Consequently, we quantified linear climate-flood linkages via stepwise regression.\u003c/p\u003e\u003cp\u003eFor flood day variability, we identified 17 key drivers which influence it. The resulting model demonstrated excellent predictive performance, yielding an exceptionally high coefficient of determination (R\u0026sup2; = 0.96), explaining 96.3% of observed variance in flood days. Model robustness was further confirmed by an adjusted R\u0026sup2; of 0.93 and a low root mean square error (RMSE) of 0.45, with information criteria values (AICc\u0026thinsp;=\u0026thinsp;101.96; BIC\u0026thinsp;=\u0026thinsp;96.05) demonstrating an optimal balance between model parsimony and explanatory power. The ANOVA analysis revealed distinct contributions: extreme precipitation indices accounted for the majority (59.9%) of observed changes in flood days, while extreme temperature and snowmelt contributed 18.2% and 17.9% respectively (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003ea). The RX5day, annual precipitation, snowmelt, and snowfall emerged as the most influential drivers according to LogWorth significance values (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eb). These factors were closely associated to precipitation and snowmelt in the TP, corresponding to the two main sources of river discharge in the region.\u003c/p\u003e\u003cp\u003eSpecifically, the contribution of RX5day to flood day changes was 20.8% (10.9% for the average RX5day and 9.9% for the maximum RX5day). RX5day captured extreme precipitation intensity, which directly drove rapid increases in river discharge and flood events. Annual precipitation modulated 18.5%, with average (8.5%) and maximum values (9.9%) showing comparable impacts. Average annual precipitation primarily modulated fundamental hydrological conditions by elevating key factors exacerbating flood risk, such like baseflow, water levels, and soil moisture saturation. Simultaneously, both maximum RX5day and maximum annual precipitation exhibited identical contributions, highlighting the critical role of spatial precipitation concentration in flood dynamics.\u003c/p\u003e\u003cp\u003eSnowmelt processes contributed significantly to the average flood day variability in TP, with annual maximum snowmelt and snowfall respectively explaining 8.1% and 8.0% of the changes. These snow-related indices influence flooding through two primary mechanisms. On the one hand, sustained snowmelt, as a major source of river baseflow, regulated long-term hydrological conditions and flood susceptibility. On the other hand, the concentrated spring meltwater pulse (typically March\u0026ndash;June) interacted synergistically with precipitation. This compound effect accelerated watershed response through simultaneous liquid water inputs, reduced soil infiltration capacity, and shortened runoff concentration times, collectively amplifying flood risks beyond what either factor would produce independently. Additionally, indices such as EP, R20, FD0, TNx, and CLT also played critical roles in flood day changes on the plateau.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eFor Q\u003csub\u003emax\u003c/sub\u003e variability, 14 significant climate drivers were identified across TP using stepwise regression (R\u0026sup2; = 0.82, adjusted R\u0026sup2; = 0.72). The model demonstrated robust predictive performance, with an RMSE of 6.66, AICc of 302.01, and BIC of 305.38. The ANOVA analysis revealed distinct contributions of different climate extremes: extreme precipitation dominated for 51.2% of Q\u003csub\u003emax\u003c/sub\u003e variability, followed by drought (22.7%) and extreme temperature (8.1%) (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003ea). Critically, precipitation extremes emerged as the primary controls: RX1day exhibited the strongest influence (average contribution: 16.7%; maximum contribution: 9.1%), followed by R99p (maximum contribution: 9.6%) (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eb). These indices characterized the intensity and magnitude of extreme precipitation events that directly governed flood peak discharges.\u003c/p\u003e\u003cp\u003eDrought-related indices dominated the remaining top factors, with maximum PDSI contributing 7.8%. These indices represented long-term moisture conditions that established critical baseline hydrological states by regulating annual patterns of soil moisture, groundwater storage, and baseflow. Upon occurrence of extreme precipitation, hydrological factors rapidly transitioned from baseline to flood peaks. Thus, flood magnitude was ultimately controlled by synergistic coupling between pre-existing hydrological baselines and superimposed extreme precipitation events, explaining 73.9% of Q\u003csub\u003emax\u003c/sub\u003e variability across TP.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003cb\u003eWatershed-scale\u003c/b\u003e\u003c/p\u003e\u003cp\u003eExtreme precipitation was recognized as the most critical driver of flood days, exerting the broadest spatial influence, particularly in the southeastern, southern, and northeastern TP (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003ea). Extreme drought indices further exhibited considerable influence on flood day variations, primarily impacting the western plateau, the lower reaches of the Yarlung Tsangbo River, and select watersheds near the source of the Yellow River. Extreme temperatures dominated flood day changes in the western, central-northern, and scattered central-southern watersheds of the plateau. In addition, snowmelt processes dominated in high-elevation (\u0026gt;\u0026thinsp;4000 m asl) western watersheds, reflecting distinct altitudinal controls on flood generation mechanisms.\u003c/p\u003e\u003cp\u003eSimultaneously, extreme precipitation emerged as the primary driver of Q\u003csub\u003emax\u003c/sub\u003e changes across most plateau watersheds, notably in the Altun Mountains, Qilian Mountains, eastern Kunlun Mountains, and the Qiangtang Plateau (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eb). Extreme high temperatures predominantly modulated discharge changes in the central, northwestern, and southern regions of the plateau. In contrast, extreme drought indices ranked as the second most important factor, mainly affecting the southwest, northwest, and eastern parts of the plateau. Meanwhile, extreme snowmelt played a more localized role, primarily impacting specific western plateau watersheds.\u003c/p\u003e\u003cp\u003eBased on spatial differences in the dominant climate extreme indices for watershed flood days and Q\u003csub\u003emax\u003c/sub\u003e changes, the study area was divided into three parts from west to east. First, flood changes in the western plateau (west of 82 \u0026deg;E) were dominated by extreme temperature, snowmelt, and drought indices. Extreme temperature and drought indices drove most of the significantly increased flood days in this part. The significant decrease in flood days in the northern Pamir Plateau was caused by extreme temperature changes. Besides, the increase in Q\u003csub\u003emax\u003c/sub\u003e around the Hindu Kush Mountains were dominated by extreme snowmelt indices, whereas those near the Pamir Plateau were jointly driven by extreme temperature and drought indices. Conversely, Q\u003csub\u003emax\u003c/sub\u003e increased around the Western Himalayas, which was attributed to extreme drought indices.\u003c/p\u003e\u003cp\u003eSecond, flood changes in most watersheds of the central plateau (between 82 \u0026deg;E and 95 \u0026deg;E) were mainly affected by extreme temperature and precipitation indices. Extreme temperature indices dominated the majority of watersheds with significantly increased flood days and Q\u003csub\u003emax\u003c/sub\u003e. Notably, Q\u003csub\u003emax\u003c/sub\u003e around the Central Himalayas decreased significantly, driven by extreme precipitation, temperature, and drought indices.\u003c/p\u003e\u003cp\u003eThird, changes in extreme precipitation and drought indices, especially changes in extreme precipitation, dominated flood day and Q\u003csub\u003emax\u003c/sub\u003e changes in most watersheds of the eastern plateau (east of 95\u0026deg;E). Specifically, extreme precipitation changes induced significant increase in flood days in watersheds east of the Qaidam Basin, while changes in both extreme precipitation and drought indices jointly contributed to the significant increase in annual Q\u003csub\u003emax\u003c/sub\u003e in this region.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003cb\u003eRiver order-scale\u003c/b\u003e\u003c/p\u003e\u003cp\u003eWe applied LASSO regression method to identify the key climate extreme drivers governing the flood changes across different river orders in TP (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e). Representativeness differences among rivers of varying orders were explicitly considered in the analysis: Order 1 rivers exhibited a slight discharge, resulting in a low frequency of flood events, while Order 7 rivers existed in only a few sections in TP, making it challenging to comprehensively reflect the flood change characteristics of the entire plateau. Consequently, analysis primarily discussed the impacts of climate extremes on flood variations in rivers of Orders 2\u0026ndash;6.\u003c/p\u003e\u003cp\u003eExtreme precipitation indices emerged as the primary drivers of flood day variations across different river orders. RX1day was particularly influential for flood day changes in Order 2\u0026ndash;5 rivers, contributing 47.0%, 43.6%, 27.8%, 39.9%, and 34.6% of the flood day variations, respectively. In contrast, annual precipitation, R20, and EP collectively dominated flood changes in Order 6 rivers, with a total contribution exceeding 49%. In contrast, the impacts of snowmelt indices on flood day changes exhibited significant differentiation by river order. The contribution of extreme snowmelt indices was generally low (typically\u0026thinsp;\u0026lt;\u0026thinsp;5%) for Order 2\u0026ndash;4 rivers but increased obviously with rising river orders. In detail, CESM contributed over 10% to flood day changes in Order 5 rivers and emerged as the most critical climate extreme index for Order 6 rivers, indicating that snowmelt became an important auxiliary driving factor in larger basins. This shift was primarily attributed to the cumulative effect of expanding basin scale: as river orders increased, snowmelt runoff from multiple upstream sub-basins continuously accumulated during the confluence process, making snowmelt runoff a key water source for high-order rivers.\u003c/p\u003e\u003cp\u003eBesides, the dominant climate drivers of Q\u003csub\u003emax\u003c/sub\u003e varied systematically with river order. R95p and CWD contributed 74.8% and 67.8% to Q\u003csub\u003emax\u003c/sub\u003e variations in Order 1 and 2 rivers, respectively, indicating that low-order rivers were predominantly regulated by precipitation (both short-duration heavy rainfall and prolonged wet periods). In Order 3 rivers, CWD had the largest contribution (29.2%), and the influence of PDSI increased (20.0%), suggesting that Q\u003csub\u003emax\u003c/sub\u003e in medium-scale basins began to be modulated by antecedent soil moisture conditions. EP and PDSI collectively dominated Q\u003csub\u003emax\u003c/sub\u003e variations in Order 5 and 6 rivers (54.1% and 50.7% total contributions, respectively), revealing that large-scale basins were influenced not only by extreme precipitation but also by antecedent basin moisture status. This progression underscored the cumulative impact of climate factors on hydrological processes in high-order rivers, where both acute precipitation events and longer-term soil moisture dynamics in shaping flood magnitude within complex basin systems.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eTherefore, we discovered that a gradient evolved in a gradient pattern. As river order increased, the variation pattern of flood days transitioned from being dominated by extreme precipitation to one controlled by the synergy of extreme precipitation and snowmelt. Concurrently, the variation pattern of Q\u003csub\u003emax\u003c/sub\u003e evolved from an extreme precipitation-driven regime to one regulated by the combined effects of extreme precipitation and soil moisture conditions.\u003c/p\u003e\u003cp\u003e\u003cb\u003eCross-watershed scale\u003c/b\u003e\u003c/p\u003e\u003cp\u003eWe conducted a cross-watershed flood change attribution study by treating annual flood days and Q\u003csub\u003emax\u003c/sub\u003e values from discrete watersheds as independent yet network-connected observational units. Z-score standardization was applied to both flood indicators and climate extreme variables within each watershed to facilitate cross-watershed comparison while preserving intra-watershed dynamics. Methodologically, we systematically integrated three representative spatial feature metrics (watershed-averaged, maximum, and upstream-averaged values) for each of the 33 parameters, generating 99 affecting variables to capture both local and network-scale drivers of flood variability.\u003c/p\u003e\u003cp\u003eWe evaluated eight distinct modeling approaches, including five machine learning methods (Random Forest, Gradient Boosting, Support Vector Machine, K-Nearest Neighbors, and Artificial Neural Network) and three linear statistical techniques (Ordinary Least Squares, LASSO Regression, and Stepwise Regression), to identify the optimal method for attributing flood changes. Through rigorous comparison of model fitting performance across multiple evaluation metrics, we selected the most appropriate algorithm that best captured the complex climate-flood relationships while maintaining practical interpretability.\u003c/p\u003e\u003cp\u003eFirstly, the Bootstrap Forest algorithm demonstrated superior performance among the machine learning models developed for flood day fitting, achieving the highest predictive accuracy with an R\u0026sup2; of 0.78 and the lowest Root Average Square Error (RASE) of 0.12 (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003ea, \u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003eb). Boosted Trees followed closely with an R\u0026sup2; of 0.70 and RASE of 0.14. In contrast, Support Vector Machine and K-Nearest Neighbors exhibited poorer performance, with R\u0026sup2; values of 0.68 and 0.46 and RASE values of 0.13 and 0.18, respectively. All five machine learning methods surpassed the three linear statistical approaches.\u003c/p\u003e\u003cp\u003eThe Bootstrap Forest model, identified as the optimal performer, was subsequently employed for factor importance analysis. Climate extreme indices collectively explained over 78% of variation in flood days. Specifically, extreme precipitation, extreme temperature, extreme drought, and extreme snowmelt indices accounted for 42.9%, 22.8%, 7.3%, and 5.0% of variations, respectively (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003ec). Extreme precipitation and temperature were the dominant drivers, jointly contributing 65.7%. Additionally, changes in watershed-average, watershed-maximum, and upstream-average climate extreme indices contributed 41.0%, 28.5%, and 8.5% of variations, respectively (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003ed). Hence, although climate extreme index changes within the watershed were the primary determinants of flood frequency, upstream climate extremes conditions also played a non-negligible role. Notably, upstream changes in extreme temperature indices contributed 5.3% of the flood day variation, highlighting the downstream propagation of snow and glacier melt processes through hydrological connectivity and confirming the hydrological coupling effect in high-altitude glacier-river systems.\u003c/p\u003e\u003cp\u003eAt the individual index level, annual precipitation, R10, PDSI, CWD, and SPEI were the most influential indices, contributing 22.1%, 7.4%, 4.3%, 3.2%, and 3.0% of the flood day variation, respectively. These results diverged significantly from the attribution analysis of flood days across the entire plateau (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003ee). Extreme precipitation and drought indices dominated among the top contributors. The persistently high importance of annual precipitation further supported the critical role of precipitation background in flood frequency. Meanwhile, the substantial contribution of R10 underscored the key influence of extreme precipitation frequency on flood day variations in most watersheds of TP.\u003c/p\u003e\u003cp\u003eAlthough extreme drought indices made a relatively modest overall contribution (7.3%), individual drought indices, such as PDSI (4.3%), ranked among the most influential, demonstrating their significant impact on flood frequency. Prolonged drought can reduce the soil moisture retention capacity, affecting not only the basin's baseline hydrological state but also decreasing precipitation infiltration. If rainfall cannot effectively penetrate overly dry soils, more water will flow into river systems as surface runoff, thereby elevating flood risks.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eAt the same time, the Bootstrap Forest method again demonstrated superior performance for models with Q\u003csub\u003emax\u003c/sub\u003e as the dependent variable (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003ea). The Bootstrap Forest model achieved an R\u0026sup2; of 0.71 with a relatively low RASE (0.13) (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003eb), explaining over 71% of the variation in Q\u003csub\u003emax\u003c/sub\u003e. Specifically, extreme precipitation, extreme temperature, extreme drought, and extreme snowmelt indices contributed 38.6%, 19.1%, 8.0%, and 5.3% of variations, respectively (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003ec). Extreme precipitation and temperature remained the dominant drivers, jointly contributing 57.7% to the changes in Q\u003csub\u003emax\u003c/sub\u003e. Changes in the watershed-average, watershed-maximum, and upstream-average climate extreme indices contributed 37.9%, 25.5%, and 7.5% of variations, respectively (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003ed). Thus, upstream changes in climate extremes also exerted a non-negligible influence on peak discharge. Notably, upstream changes in extreme temperature indices alone explained 4.8% of the variation.\u003c/p\u003e\u003cp\u003eExamining specific indices, annual precipitation, PDSI, RX5day, FD0, and CWD emerged as key factors affecting Q\u003csub\u003emax\u003c/sub\u003e, with contribution rates of 7.9%, 5.4%, 5.0%, 4.8%, and 4.7%, respectively (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003ee). Specifically, annual precipitation and PDSI showed the highest contributions\u0026mdash;changes in these indices directly modulated the long-term hydrological state of the watershed, corresponding to the \"baseline hydrological state\" discussed earlier. RX5day, to some extent, reflected the intensity of the most extreme precipitation events in the watershed, representing the \"magnitude of extreme precipitation events\". The significant contribution of FD0 was noteworthy, as the index was demonstrated to be a key driver of snow and ice melt processes in TP. Remarkably, the upstream-average FD0 contributed 1.3% to the Q\u003csub\u003emax\u003c/sub\u003e variation, further demonstrating the regulatory effect of upstream snow and ice melt processes on downstream extreme flows through modified water supply conditions.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eIn summary, extreme precipitation and temperature emerged as the dominant factors influencing both flood frequency and magnitude, while drought indices (e.g., PDSI) exerted critical controls by modifying soil hydrological properties and the watershed's baseline hydrological state. Notably, upstream climate extremes significantly regulated downstream flood processes through their effects on snow and ice melt.\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003e\u003cb\u003eScale-dependent flood responses to changing climate extremes\u003c/b\u003e\u003c/p\u003e\u003cp\u003eIn light of the scale-dependent heterogeneity in the dominant climate extreme indices driving interannual flood variations, our discussion aimed to comprehensively elucidate the distinctive flood response patterns observed across spatial scales. Specifically, we addressed three key questions: the differences in linearity of flood responses between the plateau average scale and other scales; the pronounced east-west differentiation of climate extreme indices driving flood variations across the TP watersheds; and the hierarchical differentiation of flood responses to climate extreme changes across rivers of varying stream orders.\u003c/p\u003e\u003cp\u003eFirst, linear trends in climate extreme indices demonstrated markedly superior explanatory power for average flood changes at the plateau scale compared to other scales, reflecting a quasi-linear flood-climate response pattern at the aggregate level. This phenomenon arose from the statistical aggregation effects of spatial heterogeneity. Watershed-specific forcing fields triggered differentiated runoff responses through complex surface hydrological processes via the spatial heterogeneity of underlying surface parameters such as soil moisture and vegetation cover. For example, soil moisture deficit in arid regions caused most precipitation to be absorbed by the soil, leading to low runoff production efficiency and nonlinear responses, while near-saturated soil conditions in humid regions promoted nearly complete and rapid conversion of precipitation into runoff, generating a significantly linear response pattern (Ran et al., \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Yang et al., \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). Nevertheless, these differences in nonlinear responses between watersheds tended to cancel each other out during large-scale spatial averaging, thereby ultimately establishing a quasi-linear relationship between precipitation and runoff at the plateau scale (Li and Sivapalan, \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2011\u003c/span\u003e; Liu et al., \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Yu et al., \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2023\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eSecond, the climate extreme indices driving flood variations in TP exhibited pronounced east-west differentiation, rooted in the climate-cryosphere coupling effect. The eastern TP, situated within the monsoon zone, was characterized by abundant precipitation and frequent extreme precipitation, coupled with relatively limited glacier and snow cover. Here, extreme precipitation events directly triggered surface runoff, while soil conditions (represented by drought indices) influenced surface water storage capacity\u0026mdash;these two factors jointly modulated flood dynamics. Conversely, western TP constituted a cold-arid zone under westerly control. Despite arid climate conditions, the western TP harbored abundant glacial and snow resources, with snow and ice melt serving as the primary source of runoff. Marked warming accelerated snow and ice ablation, while drought indices reflected surface water retention capacity, collectively regulating the conversion efficiency of meltwater to runoff and establishing a flood-driving mechanism dominated by extreme temperatures, snowmelt, and drought. The central TP served as a transitional zone between the monsoon and westerly domains, characterized by moderate precipitation and modest glacier and snow cover. Summer extreme precipitation directly induced flooding here, while spring warming accelerated glaciers and snowmelt, converging with extreme precipitation to generate floods. Consequently, flood variations in this area were jointly driven by extreme precipitation and temperature indices.\u003c/p\u003e\u003cp\u003eMeanwhile, the hierarchical differentiation of flood responses to the changes in climate extremes across river systems was fundamentally governed by the interplay of scale-dependent hydrological processes, underlying surface complexity, and climate factor sensitivities. In small tributary watersheds, precipitation events exhibited near-instantaneous conversion to surface runoff, constrained by limited catchment areas and rapid concentration times (McGuire et al., \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2005\u003c/span\u003e; Sivapalan, \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2003\u003c/span\u003e). This process was further amplified by minimal soil water storage capacity, accentuating the immediacy of precipitation-runoff relationships (Kraaijenbrink et al., \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Veatch et al., \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2009\u003c/span\u003e). Conversely, large mainstem basins demonstrated pronounced spatial heterogeneity, integrating both cryospheric and precipitation-driven processes. Glacier/snow-covered areas sustained baseflow through meltwater contributions, while precipitation interacted with these melt signals across temporal and spatial gradients. Additionally, soil moisture exerted stronger regulatory control on runoff generation as basin size increased. Crucially, climate sensitivity diverged markedly with scale\u0026mdash;small watersheds responded predominantly to localized convective precipitation, whereas large basins synchronized with expansive, system-scale precipitation patterns (Dai, \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2006\u003c/span\u003e; Pfahl et al., \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2017\u003c/span\u003e). These scale-dependent mechanisms collectively underscore the pivotal role of basin dimensions in modulating hydrological behavior under climatic extremes.\u003c/p\u003e\u003cp\u003eIn summary, the pronounced multi-scale differentiation in the response of flood changes to climate extremes arose from differences in watershed hydrological processes, underlying surface complexity, climate factor sensitivities, and climate-cryosphere coupling (Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003e). This insight not only deepened our understanding of the mechanisms underlying hydrological cycle responses to climate change in alpine regions but also provided critical theoretical foundations and practical guidance for developing cross-scale coupled hydrological disaster models and formulating targeted flood risk management strategies tailored to regional hydrological and climatic characteristics.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eOur study systematically elucidated the multi-scale response characteristics of flood changes to climatic extremes across TP, integrating linear and nonlinear attribution approaches. Quantitative analysis at the plateau-average scale determined the long-term driving contributions of extreme precipitation, temperature, and snowmelt to flood variations in TP, and the watershed-scale analysis revealed a distinct east-west divide in the climate drivers of flood changes. River order-scale investigation uncovered a stepwise evolution of flood-driving mechanisms, and cross-watershed scale analysis confirmed the critical regulatory role of upstream changes in climate extremes on downstream floods. These multi-scale differentiations originated from watershed hydrological processes, underlying surface complexity, climate factor sensitivities, and climate-cryosphere coupling. This work challenges the conventional \"one-size-fits-all\" analytical paradigm in flood research, offering a novel perspective on the relationship between climate extremes and flood responses. Collectively, these insights provide a robust solid theoretical foundation for constructing multi-scale, dynamic flood risk assessment models and formulating regionally differentiated flood resilience strategies.\u003c/p\u003e"},{"header":"Materials and methods","content":"\u003cp\u003e\u003cb\u003eStudy area\u003c/b\u003e\u003c/p\u003e\u003cp\u003eThe Tibetan Plateau, often referred to as the \"Roof of the World\", is located in south-central Asia. It is a global biodiversity hotspot, a natural habitat for rare wildlife, a gene pool for plateau species, a critical ecological security barrier for China and Asia, a vital water source for the continent, and a region of unique cultural heritage. The study adopted the extent of TP proposed by Zhang et al. (2021) (Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e9\u003c/span\u003ea), with its boundaries defined by the following coordinates: northernmost (40\u0026deg;1\u0026prime;6\u0026Prime;N, 96\u0026deg;50\u0026prime;5\u0026Prime;E), southernmost (25\u0026deg;59\u0026prime;26\u0026Prime;N, 98\u0026deg;40\u0026prime;33\u0026Prime;E), westernmost (34\u0026deg;58\u0026prime;8\u0026Prime;N, 67\u0026deg;40\u0026prime;37\u0026Prime;E), and easternmost (33\u0026deg;13\u0026prime;41\u0026Prime;N, 104\u0026deg;40\u0026prime;43\u0026Prime;E). The region covered an area of approximately 3\u0026times;10⁶ km\u0026sup2;.\u003c/p\u003e\u003cp\u003eThe region exhibits a characteristic layered geomorphic pattern, with distinct topographic units developing sequentially from the marginal mountains toward the interior: alpine valleys, periglacial platforms, and wide lake basins. The plateau's major geomorphic units primarily include the Himalayas, Karakoram Mountains, Gangdise Mountains, Kunlun Mountains, Altun Mountains, Qilian Mountains, Hengduan Mountains, Pamir Plateau, Changtang Plateau, and Qaidam Basin. The plateau's radial river system gives rise to major Asian rivers, such as the Yangtze, Yellow, Mekong, Salween, and Brahmaputra. TP also hosts the world's highest-altitude large lake groups, such as Nam Co and Siling Co. These hydrological factors, combined with widespread permafrost and modern glaciers, form a unique hydrogeological system, whose dynamics critically regulate regional hydrological processes and flood formation mechanisms.\u003c/p\u003e\u003cp\u003eTP exhibits diverse climate types shaped by altitude differences and latitude-longitude variations, such as mountain humid climate, subtropical humid climate, and plateau monsoon climate. The TP's climate has shown a distinct warming and wetting trend over the past 40 years. Specifically, average temperature and annual precipitation have increased significantly at rates of 0.25\u0026deg;C/10a and 11.6 mm/10a, respectively (Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e9\u003c/span\u003eb, \u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e9\u003c/span\u003ec).\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003cb\u003eData source\u003c/b\u003e\u003c/p\u003e\u003cp\u003e\u003cb\u003eMeteorological data\u003c/b\u003e\u003c/p\u003e\u003cp\u003eThe study utilized the high-resolution near-surface meteorological forcing dataset for the Third Pole region (TPMFD) and ERA5-land climate reanalysis data. The TPMFD dataset provides seven key meteorological variables, including precipitation, 2 m air temperature, 2 m specific humidity, 10 m wind speed, surface pressure, downward longwave radiation, and downward shortwave radiation. We primarily employed precipitation, temperature, and wind speed data from TPMFD, which integrated short-term high-resolution Weather Research and Forecasting (WRF) simulations, long-term ERA5 reanalysis, and ground station observations. TPMFD dataset integrates multiple data sources to achieve enhanced accuracy and higher spatial resolution than conventional reanalysis products, establishing an optimal resource for hydro-meteorological studies across the Third Pole region.\u003c/p\u003e\u003cp\u003eThe ERA5 dataset, as the latest climate reanalysis product from the European Centre for Medium-Range Weather Forecasts (ECMWF), provides detailed records of global atmospheric, land surface, and ocean wave conditions since 1950. The study utilized the ERA5-Land dataset, which focused on the land surface component of ERA5 and offered higher precision for long-term climate element monitoring, with a spatial resolution of 0.1\u0026deg; \u0026times; 0.1\u0026deg;. Snowmelt data from this dataset were employed in the analysis.\u003c/p\u003e\u003cp\u003e\u003cb\u003eHydrological data\u003c/b\u003e\u003c/p\u003e\u003cp\u003e\u003cem\u003eWatershed data\u003c/em\u003e\u003c/p\u003e\u003cp\u003eOur study utilized the MERIT Hydro global high-resolution hydrological dataset (Yamazaki et al., 2019) to extract river network and watershed boundaries. The dataset is derived from the 90-meter resolution MERIT Digital Elevation Model (DEM) and incorporates flow direction correction. River networks are hierarchically classified using the Strahler system, ranging from first-order branches (headwater streams) to seventh-order main rivers within TP, comprehensively covering the entire hydrological network from alpine rivulets to major trunk streams of large basins. The dataset further employs the Pfafstetter system to establish a complete global hierarchical watershed framework with 12 nested levels (L1\u0026ndash;L12), enabling continuous spatial coverage from continental-scale basins (L1) to local sub-watersheds (L12).\u003c/p\u003e\u003cp\u003e\u003cem\u003eFlood records\u003c/em\u003e\u003c/p\u003e\u003cp\u003eWe employed the most comprehensive flood inventory compiled by the Second Tibetan Plateau Scientific Expedition, integrating multi-source flood records across TP from 1961 to 2020. This dataset constitutes the region's most complete flood event compilation to date, encompassing\u0026thinsp;\u0026gt;\u0026thinsp;3,000 documented cases derived from systematic literature reviews of academic publications, technical reports, historical archives, hydrometric station observations, and official disaster registries (Fig. \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e). The data producer implemented a rigorous quality control protocol that involved cross-validation of different data sources, complemented by extensive field verification campaigns, to ensure maximum data reliability. Each flood event in the final curated database is characterized by key attributes, including precise geolocation and temporal occurrence. The multi-method integration of documentary evidence, instrumental measurements, and ground-truthing significantly enhances data accuracy, establishing an unprecedented foundation for robust flood characterization and modeling in this critical region.\u003c/p\u003e\u003cp\u003e\u003cem\u003eDischarge data\u003c/em\u003e\u003c/p\u003e\u003cp\u003eOur research utilized simulated discharge data from the Global 3-Hour River Flood Reanalysis (GRFR) and the Global Flood Awareness System (GLOFAS) historical dataset. The GRFR dataset employed the distributed hydrological model VIC and the river routing model RAPID to construct high-resolution, high-accuracy global natural river discharge simulation data. It covers land surface runoff data at a 0.05\u0026deg; resolution and natural discharge simulations for 2.94\u0026nbsp;million river segments globally from 1980 to 2019. Accuracy assessments based on daily discharge observations from \u0026gt;\u0026thinsp;14,000 global stations demonstrate that GRFR effectively reproduces daily-scale runoff processes and performs well in capturing flood events.\u003c/p\u003e\u003cp\u003eGLOFAS, developed by the ECMWF, is a global flood awareness and monitoring system designed to provide flood warnings and risk analysis worldwide (Alfieri et al., 2013). The GLOFAS discharge dataset includes daily-scale discharge grid data (unit: m\u0026sup3;/s) at 0.1\u0026deg; grid resolution simulated globally using the Lisflood hydrological model. The input data are derived from at least four years of observed discharge data and ERA5 meteorological data, with model outcomes calibrated at 1,226 river sections across 66 countries.\u003c/p\u003e\u003cp\u003eHere, we assessed both reanalysis datasets' flood detection across the TP (Fig. S2). The results demonstrated superior flood monitoring capabilities in the GRFR data, with over 80% of historical flood records aligned with to flood days identified by GRFR. Over 70% of flood records documented with daily precision matched flood days detected in adjacent river reaches, while 88% of monthly flood records corresponded to GRFR-derived flood periods. Consequently, GRFR data were selected for subsequent analysis.\u003c/p\u003e\u003cp\u003e\u003cb\u003eMethods\u003c/b\u003e\u003c/p\u003e\u003cp\u003e\u003cb\u003eClimate extreme and flood indices\u003c/b\u003e\u003c/p\u003e\u003cp\u003e\u003cem\u003eClimate extreme indices\u003c/em\u003e\u003c/p\u003e\u003cp\u003eThis study employed 25 standardized climate extreme indices based on those proposed by the Expert Team on Climate Change Detection and Indices (ETCCDI) (Zhang et al., 2011). These indices, derived from daily temperature and precipitation data, are broadly categorized into two classes: extreme value indices (quantifying the intensity, magnitude, or amplitude of extreme climate events) and day-count indices (reflecting the frequency, duration, or occurrence probability of extreme climate events). The selected indices comprise 12 extreme temperature indices, 10 extreme precipitation indices, and three extreme drought indices (Tab. S1).\u003c/p\u003e\u003cp\u003e\u003cem\u003eExtreme snowmelt indices\u003c/em\u003e\u003c/p\u003e\u003cp\u003eWe introduced the concept of extreme snowmelt events and developed corresponding indices, building upon the framework of extreme precipitation and temperature indices. The extreme snowmelt days index was characterized by individual days with snowmelt exceeding the 90th percentile threshold of historical daily values during the 1991\u0026ndash;2020 reference period. Continuous extreme snowmelt occurrences were identified when daily snowmelt exceeded the 90th percentile threshold for at least three consecutive days. This threshold was determined using a seasonally varying percentile calculation, where the 90th percentile was computed separately for 5-day moving windows to account for the strong seasonal cycle in snowmelt processes.\u003c/p\u003e\u003cp\u003e\u003cem\u003eFlood indices\u003c/em\u003e\u003c/p\u003e\u003cp\u003eBesides, the Peak Over Threshold (POT) method, grounded in Extreme Value Theory (EVT), provides a robust statistical framework for identifying extreme events that exceed predetermined thresholds (Bezak et al., 2014). This method operates by setting an appropriate threshold and considering only values exceeding this threshold as extreme events. In our study, we selected the 90th percentile of daily discharge as the flood identification threshold after consideration of the region's complex terrain and highly variable climate conditions. Historical flood records confirmed that this threshold effectively captured most documented flood events while successfully filtering out low-intensity, non-flood occurrences. Additionally, a minimum discharge threshold of 10 m\u0026sup3;/s was established to exclude low-flow periods from flood identification.\u003c/p\u003e\u003cp\u003eTo characterize historical flood patterns, we employed two complementary indices: flood days and annual maximum daily discharge (Q\u003csub\u003emax\u003c/sub\u003e). Flood days quantified flood frequency by counting the number of days exceeding flood thresholds within a given period, while Q\u003csub\u003emax\u003c/sub\u003e captures flood magnitude by identifying the peak discharge event each year (Jiang et al., \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). These established hydrological indicators represent both the occurrence rate and intensity of flood events across watersheds.\u003c/p\u003e\u003cp\u003e\u003cb\u003eMachine learning methods\u003c/b\u003e\u003c/p\u003e\u003cp\u003eOur research employed multiple machine learning methods\u0026mdash;including Bootstrap Forest, Boosted Trees, Support Vector Machines (SVM), K-Nearest Neighbors (KNN), and Neural Networks\u0026mdash;for attribution analysis of water bodies and flood changes. These methods effectively captured nonlinear and deep-level relationships between climate extreme indices and flood changes when dealing with complex nonlinear relationships and high-dimensional data. Each algorithm provided distinct analytical advantages, as detailed below.\u003c/p\u003e\u003cp\u003e\u003cem\u003eBootstrap Forest\u003c/em\u003e\u003c/p\u003e\u003cp\u003eBootstrap Forest is an ensemble learning method, a variant of Random Forest (Breiman, 2001). It enhances model robustness and accuracy by training multiple decision trees using bootstrap sampling (i.e., random sampling with replacement) of training data. Compared to traditional decision trees, Bootstrap Forest reduces the risk of overfitting and improves prediction accuracy by integrating multiple sub-models. In our study, Bootstrap Forest effectively revealed the relative importance of different climate extreme indices on flood changes and provided more reliable conclusions by aggregating predictions from multiple models through a voting mechanism. Simultaneously, we conducted a feature importance analysis to reveal the relative importance of different climate indices to the model. Specifically, we calculated both the frequency of each predictor variable being selected as splitting nodes across all decision trees and their corresponding sum of squared reductions. These values were then normalized to determine the relative contribution rate of each index. The method effectively captured nonlinear relationships and interaction effects among variables but also improved the stability of contribution assessment through ensemble learning.\u003c/p\u003e\u003cp\u003e\u003cem\u003eBoosted Trees\u003c/em\u003e\u003c/p\u003e\u003cp\u003eBoosted Trees is an algorithm that builds a strong learner by integrating multiple weak learners (decision trees) (Friedman, 2001). The core idea is to iteratively strengthen samples that are incorrectly predicted in previous rounds through weighted adjustments. Compared to traditional decision trees, the Boosted Trees method excels in handling complex nonlinear relationships and continuously improves model prediction accuracy by optimizing loss functions. In our study, Boosted Trees method was employed to identify key climate extreme indices influencing flood changes and to handle complex data with nonlinear characteristics, thereby enhancing the accuracy and robustness of attribution analysis.\u003c/p\u003e\u003cp\u003e\u003cem\u003eSupport Vector Machines (SVM)\u003c/em\u003e\u003c/p\u003e\u003cp\u003eSVM is a widely used supervised learning method, particularly suitable for classification and regression tasks (Cortes \u0026amp; Vapnik, 1995). SVM constructs a hyperplane that maximizes the margin to classify data points or maps data to a higher-dimensional space using kernel tricks to solve nonlinear problems. The advantage of SVM is its ability to handle high-dimensional data and achieve good performance even with small datasets. SVM effectively adapts to the different characteristics of datasets by adjusting kernel functions and penalty parameters, providing reliable results in flood change attribution analysis. In our study, SVM was used for regression tasks to fit flood days and Q\u003csub\u003emax\u003c/sub\u003e.\u003c/p\u003e\u003cp\u003e\u003cem\u003eK-Nearest Neighbors (KNN)\u003c/em\u003e\u003c/p\u003e\u003cp\u003eKNN is a simple yet effective non-parametric supervised learning method suitable for classification and regression tasks (Cover \u0026amp; Hart, 1967). The core idea of KNN is to determine the predicted value of an input sample based on its distance to other samples in the training data, commonly using Euclidean or Manhattan distance metrics. In the study, KNN was used for regression analysis to fit hydrological indicators such as flood days.\u003c/p\u003e\u003cp\u003e\u003cem\u003eNeural Networks\u003c/em\u003e\u003c/p\u003e\u003cp\u003eThe neural network trains several neural network models and optimizes them using boosting algorithms to create an integrated strong prediction model (Rumelhart et al., 1986). This method is suitable for prediction tasks involving complex, nonlinear relationships, especially those involving high-dimensional data or problems that are difficult to model with traditional statistical methods. In the study, the method was used to model the complex relationships between climate extreme indices and flood changes.\u003c/p\u003e\u003cp\u003e\u003cb\u003eRegression methods and supplementary variables for attribution analysis\u003c/b\u003e\u003c/p\u003e\u003cp\u003eOur study employed a suite of regression models to systematically analyze the associations between plateau floods and climate extremes, uncovering response characteristics across multiple scales through variable selection, trend quantification, and attribution diagnostics.\u003c/p\u003e\u003cp\u003eFirst, linear regression analysis was applied to multi-decadal records of flood frequency and magnitude to quantify temporal trends (Shi \u0026amp; Neng, 1995). The regression coefficients, along with their 95% confidence intervals (95% CI) and statistical significance levels (\u003cem\u003ep\u003c/em\u003e-values) were systematically examined to assess the direction (increasing or decreasing) and magnitude of trends in key variables and whether these trends reached statistical significance (p\u0026thinsp;\u0026lt;\u0026thinsp;0.05). This analysis helped reveal the characteristics of flood changes in TP.\u003c/p\u003e\u003cp\u003eSecond, we quantitatively evaluated the linear impact of climate extreme indices on regional flood changes using a stepwise regression method. Stepwise regression is a variable selection method used in regression analysis to optimize model performance by iteratively adding or removing explanatory variables. The method automatically selects the most significant factors for modeling by evaluating each variable's contribution to the model's explanatory power. We implemented a dual-criterion approach using both the corrected Akaike Information Criterion (AICc) and Bayesian Information Criterion (BIC) for variable selection. The combined strategy ensures an optimal balance between model fit and complexity. The AICc, which adjusts for small sample sizes, minimizes prediction error while preventing oversimplification by accounting for both model accuracy and parameter count. However, AICc's relatively mild penalty for complexity may occasionally permit marginally significant variables to enter the model. In contrast, the BIC imposes stronger penalties on model complexity, serving as a robust safeguard against overfitting.\u003c/p\u003e\u003cp\u003eThird, the Least Absolute Shrinkage and Selection Operator (LASSO) regression was used to detect the main climate extreme indices governing changes in flood days and Qmax in different watersheds in TP. LASSO regression is particularly useful for handling high-dimensional data, as it maintains model predictive performance while reducing complexity and avoiding overfitting. The method performs variable selection and model simplification by applying L1 regularization (i.e., adding a penalty term for the sum of absolute coefficients) (Tibs hirani, 2011). The core idea of LASSO is to shrink some regression coefficients to zero through regularization, thereby retaining key variables while automatically eliminating less important ones. Our study further reduced the number of variables in the LASSO model based on the 1SE rule (one standard error rule), making the final model more interpretable and robust. The 1SE rule is a common strategy for selecting regularization strength (i.e., the λ value in LASSO). It suggests choosing the largest λ value where the mean squared error (MSE) falls within one standard error of the minimum MSE. This approach achieves a better balance between model accuracy and complexity, resulting in a simpler yet sufficiently predictive model.\u003c/p\u003e\u003cp\u003eNotably, existing attribution studies of regional hydrological changes have predominantly relied on spatially averaged climate variables, neglecting the critical influence of intra-regional spatial heterogeneity. However, spatial distribution patterns of climate extreme indices play a significant role in modulating hydrological processes, particularly flood generation and propagation. To address the gap, we proposed incorporating spatial statistical indicators (e.g., maximum, minimum, and variance) of independent variables as supplementary variables in attribution analysis. Comparative experiments demonstrated that watershed-maximum values of key meteorological parameters (e.g., annual precipitation, R10, and PDSI) significantly enhances the performance of flood attribution models.\u003c/p\u003e\u003cp\u003eEqually importantly, upstream changes in climate extremes can significantly influence downstream flood events due to the spatial propagation of flood-generating processes. Hydrological connectivity ensures that extreme precipitation, snowmelt, or drought conditions in upstream areas alter downstream runoff patterns and flood frequency. We thus incorporated spatially averaged climate extreme indices from upstream watersheds as additional variables to account for this mechanism in our cross-watershed flood attribution analysis. This approach explicitly captured the spatial transmission of upstream climate extremes on flood dynamics, offering a framework for assessing the interplay of regional climate variability and hydrological processes.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003ch2\u003eCompeting intersts\u003c/h2\u003e\u003cp\u003eAll authors declare no financial or non-financial competing interests.\u003c/p\u003e\u003c/p\u003e\u003ch2\u003eAuthor contributions\u003c/h2\u003e\u003cp\u003eC.P., L.X., and, Z.X. initiated the research. L.X. and Z.X. developed the methods. L.X. wrote the manuscript. Z.X. and S.P. revised the manuscript. C.P. and Z.X. acquired the funding and supervised the study. All authors read and approved the final manuscript.\u003c/p\u003e\u003ch2\u003eAcknowledgments\u003c/h2\u003e\u003cp\u003eThis work was supported by the Science and Technology Projects of Xizang Autonomous Region, China (XZ202402ZD0001, XZ202401ZY0108), and the Special Project for the Construction of Nyingchi National Sustainable Development Pilot Zone, Xizang Autonomous Region (2023-SYQ-006).\u003c/p\u003e\u003ch2\u003eData availability\u003c/h2\u003e\u003cp\u003eThe reanalysis discharge, meteorological, and watershed data sets used in this study are publicly available. The reanalysis discharge data sets are from the GRFR (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.11888/Terre.tpdc.272901\u003c/span\u003e\u003cspan address=\"10.11888/Terre.tpdc.272901\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e) and GLOFAS (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://data.jrc.ec.europa.eu/dataset/73305ca5-c002-4124-b1d5-6451cc93af3f\u003c/span\u003e\u003cspan address=\"https://data.jrc.ec.europa.eu/dataset/73305ca5-c002-4124-b1d5-6451cc93af3f\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e). TPMFD dataset is from National Tibetan Plateau (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.11888/Atmos.tpdc.300398\u003c/span\u003e\u003cspan address=\"10.11888/Atmos.tpdc.300398\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e). The ERA5-Land dataset is from the European Centre for Medium-Range Weather Forecasts (ECMWF) (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://cds.climate.copernicus.eu/cdsapp#!/dataset/reanalysis-era5-land?tab=overview\u003c/span\u003e\u003cspan address=\"https://cds.climate.copernicus.eu/cdsapp#!/dataset/reanalysis-era5-land?tab=overview\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e). The flood record data is available from the Second Tibetan Plateau Scientific Expedition and Research Program (STEP) but restrictions apply to the availability of these data, which were used under licence for the current study, and so are not publicly available. Data are however available from the authors upon reasonable request and with permission of STEP.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n \u003cli\u003eBai, Y., Li, D., Wangchuk, S., Kettner, A., Zhao, Y., Deng, R., Liu, Y., Xiao, C., Ni, J., Cui, P., 2025. Flood complexity and rising exposure risk in High Mountain Asia under climate change. Science Bulletin S2095927325001355. https://doi.org/10.1016/j.scib.2025.01.055\u003c/li\u003e\n \u003cli\u003eBalaian, S.K., Sanders, B.F., Abdolhosseini Qomi, M.J., 2024. How urban form impacts flooding. Nat Commun 15, 6911. https://doi.org/10.1038/s41467-024-50347-4\u003c/li\u003e\n \u003cli\u003eBormann, K.J., Brown, R.D., Derksen, C., Painter, T.H., 2018. Estimating snow-cover trends from space. Nature Clim Change 8, 924\u0026ndash;928. https://doi.org/10.1038/s41558-018-0318-3\u003c/li\u003e\n \u003cli\u003eBrunner, M.I., Swain, D.L., Wood, R.R., Willkofer, F., Done, J.M., Gilleland, E., Ludwig, R., 2021. An extremeness threshold determines the regional response of floods to changes in rainfall extremes. Commun Earth Environ 2, 173. https://doi.org/10.1038/s43247-021-00248-x\u003c/li\u003e\n \u003cli\u003eDai, A., 2006. Precipitation Characteristics in Eighteen Coupled Climate Models. Journal of Climate 19, 4605\u0026ndash;4630. https://doi.org/10.1175/JCLI3884.1\u003c/li\u003e\n \u003cli\u003eGao, Y., Lu, N., Dai, Y., Yao, T., 2023. Reversal in snow mass trends on the Tibetan Plateau and their climatic causes. Journal of Hydrology 620, 129438. https://doi.org/10.1016/j.jhydrol.2023.129438\u003c/li\u003e\n \u003cli\u003eHarrison, S., Macklin, M.G., Toonen, W.H.J., Benito, G., Cohen, K.M., 2025. Robust climate attribution of modern floods needs palaeoflood science. Climatic Change 178, 71. https://doi.org/10.1007/s10584-025-03904-9\u003c/li\u003e\n \u003cli\u003eIPCC, 2023. Climate Change 2022 \u0026ndash; Impacts, Adaptation and Vulnerability: Working Group II Contribution to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change, 1st ed. Cambridge University Press. https://doi.org/10.1017/9781009325844\u003c/li\u003e\n \u003cli\u003eJiang, S., Tarasova, L., Yu, G., Zscheischler, J., 2024. Compounding effects in flood drivers challenge estimates of extreme river floods. Sci. Adv. 10, eadl4005. https://doi.org/10.1126/sciadv.adl4005\u003c/li\u003e\n \u003cli\u003eKraaijenbrink, P.D.A., Stigter, E.E., Yao, T., Immerzeel, W.W., 2021. Climate change decisive for Asia\u0026rsquo;s snow meltwater supply. Nat. Clim. Chang. 11, 591\u0026ndash;597. https://doi.org/10.1038/s41558-021-01074-x\u003c/li\u003e\n \u003cli\u003eKundzewicz, Z.W., Kanae, S., Seneviratne, S.I., Handmer, J., Nicholls, N., Peduzzi, P., Mechler, R., Bouwer, L.M., Arnell, N., Mach, K., Muir-Wood, R., Brakenridge, G.R., Kron, W., Benito, G., Honda, Y., Takahashi, K., Sherstyukov, B., 2014. Flood risk and climate change: global and regional perspectives. Hydrological Sciences Journal 59, 1\u0026ndash;28. https://doi.org/10.1080/02626667.2013.857411\u003c/li\u003e\n \u003cli\u003eLi, H., Sivapalan, M., 2011. Effect of spatial heterogeneity of runoff generation mechanisms on the scaling behavior of event runoff responses in a natural river basin. Water Resources Research 47, 2010WR009712. https://doi.org/10.1029/2010WR009712\u003c/li\u003e\n \u003cli\u003eLi, X., Cui, P., Zhang, X.-Q., Zhang, F., 2024. Intensified warming suppressed the snowmelt in the Tibetan Plateau. Advances in Climate Change Research 15, 452\u0026ndash;463. https://doi.org/10.1016/j.accre.2024.06.005\u003c/li\u003e\n \u003cli\u003eLiu, J., Engel, B.A., Wang, Y., Wu, Y., Zhang, Z., Zhang, M., 2019. Runoff Response to Soil Moisture and Micro-topographic Structure on the Plot Scale. Sci Rep 9, 2532. https://doi.org/10.1038/s41598-019-39409-6\u003c/li\u003e\n \u003cli\u003eMcGuire, K.J., McDonnell, J.J., Weiler, M., Kendall, C., McGlynn, B.L., Welker, J.M., Seibert, J., 2005. The role of topography on catchment‐scale water residence time. Water Resources Research 41, 2004WR003657. https://doi.org/10.1029/2004WR003657\u003c/li\u003e\n \u003cli\u003eMin, S.-K., Zhang, X., Zwiers, F.W., Hegerl, G.C., 2011. Human contribution to more-intense precipitation extremes. Nature 470, 378\u0026ndash;381. https://doi.org/10.1038/nature09763\u003c/li\u003e\n \u003cli\u003eOtto, F.E.L., 2023. Attribution of Extreme Events to Climate Change. Annu. Rev. Environ. Resour. 48, 813\u0026ndash;828. https://doi.org/10.1146/annurev-environ-112621-083538\u003c/li\u003e\n \u003cli\u003ePark, I.-H., Min, S.-K., 2017. Role of convective precipitation in the relationship between subdaily extreme precipitation and temperature. J. Climate 30, 9527\u0026ndash;9537. https://doi.org/10.1175/JCLI-D-17-0075.1\u003c/li\u003e\n \u003cli\u003ePeel, M.C., McMahon, T.A., Finlayson, B.L., Watson, F.G.R., 2002. Implications of the relationship between catchment vegetation type and the variability of annual runoff. Hydrological Processes 16, 2995\u0026ndash;3002. https://doi.org/10.1002/hyp.1084\u003c/li\u003e\n \u003cli\u003ePfahl, S., O\u0026rsquo;Gorman, P.A., Fischer, E.M., 2017. Understanding the regional pattern of projected future changes in extreme precipitation. Nature Clim Change 7, 423\u0026ndash;427. https://doi.org/10.1038/nclimate3287\u003c/li\u003e\n \u003cli\u003eRan, Q., Wang, J., Chen, X., Liu, L., Li, J., Ye, S., 2022. The relative importance of antecedent soil moisture and precipitation in flood generation in the middle and lower Yangtze River basin. Hydrol. Earth Syst. Sci. 26, 4919\u0026ndash;4931. https://doi.org/10.5194/hess-26-4919-2022\u003c/li\u003e\n \u003cli\u003eScussolini, P., Luu, L.N., Philip, S., Berghuijs, W.R., Eilander, D., Aerts, J.C.J.H., Kew, S.F., Van Oldenborgh, G.J., Toonen, W.H.J., Volkholz, J., Coumou, D., 2024. Challenges in the attribution of river flood events. WIREs Climate Change 15, e874. https://doi.org/10.1002/wcc.874\u003c/li\u003e\n \u003cli\u003eSivapalan, M., 2003. Process complexity at hillslope scale, process simplicity at the watershed scale: is there a connection? Hydrological Processes 17, 1037\u0026ndash;1041. https://doi.org/10.1002/hyp.5109\u003c/li\u003e\n \u003cli\u003eTorre Zaffaroni, P., Baldi, G., Texeira, M., Di Bella, C.M., Jobb\u0026aacute;gy, E.G., 2023. The Timing of Global Floods and Its Association With Climate and Topography. Water Resources Research 59, e2022WR032968. https://doi.org/10.1029/2022WR032968\u003c/li\u003e\n \u003cli\u003eVeatch, W., Brooks, P.D., Gustafson, J.R., Molotch, N.P., 2009. Quantifying the effects of forest canopy cover on net snow accumulation at a continental, mid‐latitude site. Ecohydrology 2, 115\u0026ndash;128. https://doi.org/10.1002/eco.45\u003c/li\u003e\n \u003cli\u003eWang, X., Chen, R., Li, H., Li, K., Liu, J., Liu, G., 2022. Detection and attribution of trends in flood frequency under climate change in the Qilian Mountains, Northwest China. Journal of Hydrology: Regional Studies 42, 101153. https://doi.org/10.1016/j.ejrh.2022.101153\u003c/li\u003e\n \u003cli\u003eWang, Y., Zheng, D., Zhang, G., Carrivick, J.L., Bolch, T., Ren, W., Guo, L., Su, J., Yuan, S., Li, X., 2025. Patterns and change rates of glacial lake water levels across High Mountain Asia. National Science Review 12, nwaf041. https://doi.org/10.1093/nsr/nwaf041\u003c/li\u003e\n \u003cli\u003eWu, B., Zhang, Z., Guo, X., Tan, C., Huang, C., Tao, J., 2022. Spatial and Temporal Analysis of Quantitative Risk of Flood due to Climate Change in a China\u0026rsquo;s Plateau Province. Front. Earth Sci. 10, 931505. https://doi.org/10.3389/feart.2022.931505\u003c/li\u003e\n \u003cli\u003eYang, H., Zhang, Xiaoqi, Yuan, Z., Hong, X., Yao, L., Zhang, Xiuping, 2025. Investigating the effects of spatial heterogeneity of multi-source profile soil moisture on spatial\u0026ndash;temporal processes of high-resolution floods. Journal of Hydrology 652, 132672. https://doi.org/10.1016/j.jhydrol.2025.132672\u003c/li\u003e\n \u003cli\u003eYang, Keke, Guo, D., Hua, W., Pepin, N., Yang, Kun, Li, D., 2022. Tibetan Plateau temperature extreme changes and their elevation dependency from ground‐based observations. JGR Atmospheres 127, e2021JD035734. https://doi.org/10.1029/2021JD035734\u003c/li\u003e\n \u003cli\u003eYao, Y., Wang, Y., Chen, Q., Liao, Y., 2025. Temporal and spatial variation characteristics of extreme precipitation in central and eastern Tibetan Plateau. Journal of \u0026nbsp; Arid Meteorology 41, 714\u0026ndash;722. https://doi.org/10.11755/j.issn.1006-7639（2023）-05-0714\u003c/li\u003e\n \u003cli\u003eYin, H., Sun, Y., Donat, M.G., 2019. Changes in temperature extremes on the Tibetan Plateau and their attribution. Environ. Res. Lett. 14, 124015. https://doi.org/10.1088/1748-9326/ab503c\u003c/li\u003e\n \u003cli\u003eYin, H., Sun, Y., Wan, H., Zhang, X., Lu, C., 2017. Detection of anthropogenic influence on the intensity of extreme temperatures in China. Int. J. Climatol. 37, 1229\u0026ndash;1237. https://doi.org/10.1002/joc.4771\u003c/li\u003e\n \u003cli\u003eYu, T., Ran, Q., Pan, H., Li, J., Pan, J., Ye, S., 2023. The impacts of rainfall and soil moisture to flood hazards in a humid mountainous catchment: a modeling investigation. Front. Earth Sci. 11, 1285766. https://doi.org/10.3389/feart.2023.1285766\u003c/li\u003e\n \u003cli\u003eZhang, G., Carrivick, J.L., Emmer, A., Shugar, D.H., Veh, G., Wang, X., Labedz, C., Mergili, M., M\u0026ouml;lg, N., Huss, M., Allen, S., Sugiyama, S., L\u0026uuml;tzow, N., 2024. Characteristics and changes of glacial lakes and outburst floods. Nat Rev Earth Environ 5, 447\u0026ndash;462. https://doi.org/10.1038/s43017-024-00554-w\u003c/li\u003e\n \u003cli\u003eZhang, S., Zhou, L., Zhang, L., Yang, Y., Wei, Z., Zhou, S., Yang, D., Yang, X., Wu, X., Zhang, Y., Li, X., Dai, Y., 2022. Reconciling disagreement on global river flood changes in a warming climate. Nat. Clim. Chang. 12, 1160\u0026ndash;1167. https://doi.org/10.1038/s41558-022-01539-7\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"nature-portfolio","isNatureJournal":true,"hasQc":false,"allowDirectSubmit":false,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"","title":"Nature Portfolio","twitterHandle":"","acdcEnabled":false,"dfaEnabled":false,"editorialSystem":"ejp","reportingPortfolio":"","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"Flood, Climate extreme, Scale-dependent response, Hydrological connectivity, Tibetan Plateau ","lastPublishedDoi":"10.21203/rs.3.rs-7153596/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7153596/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eUnderstanding of the dynamic interplay between plateau floods and climate extremes has long been constrained by scale fragmentation in existing studies. Our multi-scale analysis unravels and interprets the scale-dependent responses of floods to changes in climate extremes across the Tibetan Plateau (TP). Extreme precipitation, temperature, and snowmelt drive the average flood day increase (0.7 d/10a) at the plateau scale, while the rise in annual maximum daily discharge (Q\u003csub\u003emax\u003c/sub\u003e) (2.1 m\u003csup\u003e3\u003c/sup\u003e/s/10a) is modulated by extreme precipitation and drought indices. Watershed-scale analysis uncovers a distinct east-west partitioning of flood drivers, whereas river order-scale analysis reveals patterned shifts in flood drivers from main streams to tributaries. Cross-watershed analysis shows that upstream temperature changes contribute 5.3% to downstream flood frequency and 4.8% to magnitude variability via hydrological connectivity. The scale-specific disparities, shaped by the synergistic effects of watershed hydrological processes, underlying surface heterogeneity, climate factor sensitivities, and climate-cryosphere interactions, establish a framework for alpine flood attribution and predictive models.\u003c/p\u003e","manuscriptTitle":"Unraveling scale-dependent flood responses to changing climate extremes over the Tibetan Plateau","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-08-27 18:21:41","doi":"10.21203/rs.3.rs-7153596/v1","editorialEvents":[],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"communications-earth-and-environment","isNatureJournal":true,"hasQc":false,"allowDirectSubmit":false,"externalIdentity":"commsenv","sideBox":"Learn more about [Communications Earth and Environment](https://www.nature.com/commsenv/)","snPcode":"","submissionUrl":"","title":"Communications Earth \u0026 Environment","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"ejp","reportingPortfolio":"Communications Series","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"22a86ad9-0d73-41c1-96b7-60b49292dde4","owner":[],"postedDate":"August 27th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[{"id":52673248,"name":"Earth and environmental sciences/Natural hazards"},{"id":52673249,"name":"Earth and environmental sciences/Climate sciences/Climate change/Climate-change impacts"}],"tags":[],"updatedAt":"2026-03-19T07:13:23+00:00","versionOfRecord":{"articleIdentity":"rs-7153596","link":"https://doi.org/10.1038/s43247-026-03413-2","journal":{"identity":"communications-earth-and-environment","isVorOnly":false,"title":"Communications Earth \u0026 Environment"},"publishedOn":"2026-03-18 04:00:00","publishedOnDateReadable":"March 18th, 2026"},"versionCreatedAt":"2025-08-27 18:21:41","video":"","vorDoi":"10.1038/s43247-026-03413-2","vorDoiUrl":"https://doi.org/10.1038/s43247-026-03413-2","workflowStages":[]},"version":"v1","identity":"rs-7153596","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7153596","identity":"rs-7153596","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}