Rock glaciers, unique creeping periglacial landforms, have garnered research interest for their valuable insights into hydrological resources and representing potential geological hazards in the context of climate warming. However, owing to the scarcity of high-spatiotemporal-resolution kinematic observations, the mechanisms driving rock glacier movements remain poorly understood. This study introduces a strategy for estimating three-dimensional (3D) velocities of rock glaciers by combining ascending and descending interferometric synthetic aperture radar (InSAR) observations with a surface parallel flow model. We mapped the 3D surface creeping velocities of 1084 rock glaciers in Daxueshan, southwest Tibetan Plateau, from January 2019 to December 2020. Subsequently, we employed the generalized boosted model (GBM), a machine learning approach, to link rock glacier creep to environmental factors. The rock glaciers in Daxueshan exhibited spatially heterogeneous creep, with maximum and average annual velocities of 307.86 mm∙a and 40.37 mm∙a, respectively. About 92% of rock glaciers crept with a two-peak seasonal rhythm characterized by “spring acceleration, summer deceleration, autumn acceleration, and winter deceleration”. GBM modeling suggested that precipitation and snowmelt were the predominant environmental factors controlling the kinematics of rock glaciers, with relative importance of 32.6% and 31.7%, respectively. Land surface temperature (LST) was the third significant driver at 17.1% of the relative importance. Slope angle and topographic wetness index (TWI) had the least impact. The modeling also quantitatively revealed the distinct nonlinear responses of rock glacier creep to environmental changes. Rock glaciers initiate acceleration at precipitation and snowmelt levels of about 2 mm, LST close to −2 °C, slope angle and TWI higher than 12° and 4.6, and stabilize when these factors reach certain thresholds. Furthermore, the seasonal variations in rock glacier creeping velocity can be primarily attributed to hydrological changes. Our study highlights the value of combining InSAR and machine learning to elucidate the environmental controls on mass-wasting phenomena in mountainous regions.

中文翻译：

---

利用SAR干涉测量和广义增强模型表征青藏高原东南部大雪山活动岩石冰川的运动学


岩石冰川是独特的冰缘爬行地貌，因其对水文资源的宝贵见解以及气候变暖背景下潜在的地质灾害而引起了研究兴趣。然而，由于缺乏高时空分辨率的运动学观测，驱动岩石冰川运动的机制仍然知之甚少。本研究介绍了一种通过将上升和下降干涉合成孔径雷达 (InSAR) 观测与表面平行流模型相结合来估计岩石冰川三维 (3D) 速度的策略。我们绘制了2019年1月至2020年12月青藏高原西南部大雪山1084个岩石冰川的3D表面蠕变速度。随后，我们采用机器学习方法广义增强模型（GBM）将岩石冰川蠕变与环境因素联系起来。大雪山岩石冰川表现出空间异质蠕变，年最大流速为307.86 mm∙a，年平均流速为40.37 mm∙a。约92%的岩石冰川以“春季加速、夏季减速、​​秋季加速、冬季减速”为特征的双峰季节节律蠕动。 GBM模型表明，降水和融雪是控制岩石冰川运动的主要环境因素，相对重要性分别为32.6%和31.7%。地表温度 (LST) 是第三个重要驱动因素，占相对重要性的 17.1%。坡角和地形湿度指数（TWI）的影响最小。该模型还定量地揭示了岩石冰川蠕变对环境变化的独特非线性响应。 当降水量和融雪量达到约2毫米、地表温度接近-2℃、坡度角和TWI高于12°和4.6时，岩石冰川开始加速，并在这些因素达到一定阈值时稳定下来。此外，岩石冰川蠕动速度的季节性变化主要归因于水文变化。我们的研究强调了将 InSAR 和机器学习相结合来阐明山区大规模浪费现象的环境控制的价值。