水利水电技术（中英文）

【目的】开展气候变化背景下高寒水源区径流模拟与归因分析，可为高寒流域水文响应研究和水资源适应性管理提供理论依据。【方法】比选CMIP6方案并进行降尺度和Delta偏差矫正，选取CanESM5模式，采用四种SSP情景数据驱动VIC模型，模拟2021—2100年径流过程，并利用LSTM对长序列径流残差进行修正。采用Mann-Kendall检验、R/S分析、径流质心、不均匀系数、变化幅度、弹性系数和多元线性回归方法，评估径流年际、年内变化及气象因子敏感度和贡献度。【结果】未来流域气温和降水均增加，不同排放模式下气温增长率为0.05～0.37℃·(10a)^-1，降水增长率为1.70～3.60mm·(10a)^-1；LSTM残差修正后径流模拟精度较单一VIC模型高，桐子林站NSE由0.81提高至0.87，KGE由0.76提高至0.82；径流总体呈可持续上升趋势，且随排放强度增强由局部显著转为全流域显著。年内径流仍为单峰型，但来水重心前移，不同情景下C_t下降0.13～0.37，C_n和C_p分别下降0.17～0.68和0.90～5.85，丰枯差异缩小。径流对降水表现为全流域正敏感，对气温则呈现“上游正敏感、中游零敏感或弱负敏感、下游负敏感”的空间异质性。降水对未来径流变化的贡献率高达0.77以上。【结论】基于LSTM残差修正的VIC模型能有效克服高寒复杂流域径流模拟的局限性，提高径流预测精度。未来流域整体呈“暖湿化”发展，降水是驱动全流域径流增加的主导因子，气温起到调节作用。这种气候共同作用不仅增加了径流总量，更深刻改变了径流的年内分配节律。研究结果可为高寒水源区水安全管理与适应性调控提供理论支撑。

[Objective] Runoff simulation and attribution analysis in alpine water source areas under climate change scenarios are conducted， providing a theoretical basis for hydrological response studies and adaptive water resource management in alpine basins. [Methods] The CMIP6 schemes were compared and selected， followed by downscaling and Delta bias correction. The CanESM5 model was selected， and four shared socioeconomic pathway (SSP) scenario datasets were used to drive the variable infiltration capacity (VIC) model for simulating runoff processes from 2021 to 2100. The long-term runoff residuals were corrected using a long short-term memory (LSTM) network. Interannual and intra-annual runoff variations， along with the sensitivity and contribution of meteorological factors， were subsequently evaluated using the Mann–Kendall test， rescaled-range (R/S) analysis， runoff centroid， nonuniformity coefficient， variation amplitude， elasticity coefficient， and multiple linear regression. [Results] Air temperature and precipitation across the river basin in the future were projected to increase， with air temperature and precipitation increasing at rates of 0.05–0.37 °C per decade and 1.70–3.60 mm per decade， respectively， under different emission scenarios. After LSTM residual correction， the accuracy of runoff simulation was higher than that of the standalone VIC model， with the NSE and KGE values at Tongzilin station increasing from 0.81 to 0.87 and from 0.76 to 0.82， respectively. Runoff exhibited an overall persistent increasing trend， and the trend changed from locally significant to significant across the river basin as emission intensity increased. The intra-annual runoff remained unimodal， but the runoff centroid shifted earlier. Under different scenarios， Ct decreased by 0.13–0.37， while Cn and Cp decreased by 0.17–0.68 and 0.90–5.85， respectively， indicating a smaller difference between wet and dry seasons. Runoff was found to be positively sensitive to precipitation across the entire river basin， whereas its sensitivity to air temperature exhibited marked spatial heterogeneity， characterized by positive sensitivity in the upper reaches， near-zero or weakly negative sensitivity in the middle reaches， and negative sensitivity in the lower reaches. The contribution of precipitation to future runoff changes was estimated to exceed 77%. [Conclusion] The LSTM residual-corrected VIC model can effectively overcome the limitations of runoff simulation in complex alpine basins， thereby improving runoff prediction accuracy. The river basin is projected to exhibit an overall warming and wetting trend in the future. Runoff increases are primarily driven by precipitation， whereas air temperature plays a regulating role. The combined climatic effect increases total runoff volume and， more importantly， reshapes the intra-annual runoff distribution pattern. These findings offer theoretical support for water security management and adaptive regulation in alpine water source areas.