Increasingly extreme precipitation on the Tibetan Plateau (TP) threatens infrastructure, ecosystems, and livelihoods. However, traditional linear models are limited in capturing nonlinear drivers and projecting long-term changes. To address this, we applied K -means clustering and developed an explainable machine learning framework. Using six extreme precipitation indices from 1961–2022, we identified significant upward trends in R95pTOT, RX1day, and RX5day, increasing by 0.19 mm, 0.12 mm, and 0.12 mm per decade, respectively. Attribution analysis revealed that upper-level specific humidity (q400) generally serves as the most influential factor, contributing 45.72% to RX5day (north TP), 20.51% to R10 (south and east TP), and 32.6% to RX5day (southeast TP), although upper-level temperature (t400) occasionally dominates. Compared with CMIP6 simulations, the machine learning models project slightly lower magnitudes of future extreme precipitation but exhibit strong spatial and temporal consistency, effectively capturing large-scale patterns and reducing CMIP6 biases. Under SSP1-2.6, extreme precipitation shows a modest increase, whereas under SSP5-8.5, a pronounced intensification is projected. Specifically, R95pTOT and CWD are expected to rise by 0.28 mm and 0.16 days per decade, respectively. This study establishes a computationally efficient, reproducible, and interpretable framework, providing reliable scientific support for disaster risk management and climate adaptation in high-altitude, data-sparse regions. 青藏高原极端降水事件日益增多, 对基础设施,生态系统和民生构成威胁.然而, 传统的线性模型难以捕捉非线性驱动因素并预测长期变化.为解决这一问题, 我们应用K均值聚类算法, 并构建了一个可解释的机器学习框架.利用1961年至2022年的六个极端降水指数, 我们发现R95pTOT,RX1day和RX5day均呈现显著上升趋势, 分别以每十年0.19毫米,0.12毫米和0.12毫米的速度增长.归因分析表明, 高空比湿 (q400) 通常是影响最大的因素, 对RX5day (青藏高原北部) 的贡献率为45.72%, 对R10 (青藏高原南部和东部) 的贡献率为20.51%, 对RX5day (青藏高原东南部) 的贡献率为32.6%, 高空温度 (t400) 有时也起主导作用.与CMIP6模拟相比, 机器学习模型预测的未来极端降水强度略低, 但空间和时间一致性强, 有效捕捉了大尺度模式并减少了CMIP6的偏差.在SSP1-2.6情景下, 极端降水略有增加;而在SSP5-8.5情景下, 极端降水强度预计显著增强.具体而言, R95pTOT和CWD预计分别以每十年0.28毫米和0.16天的速度增加.本研究建立了一个计算高效,可复现且易于解释的框架, 为高海拔,数据稀少地区的灾害风险管理和气候适应提供了可靠的科学依据.
Li et al. (Sun,) studied this question.