Using remote sensing to measure suspended sediment concentration (SSC) in mountainous rivers can compensate for the scarcity of in situ sediment observations, providing valuable direct supplementation to observational records. However, for inland rivers, remote sensing SSC assessments face challenges such as data quality, long-term water body changes, environmental noise, flood events, and the transferability of local calibrations. Here, we introduce and apply remote sensing big data techniques using 12,445 cloud-free Landsat 5, 7, and 8 satellite images to calibrate SSC in the source region of the Yangtze River (SRYR). Utilizing Google Earth Engine, we implemented a series of image preprocessing techniques and water fraction methods to extract precise inland river water masks. Then we used unsupervised K-Means clustering and machine learning algorithms to model the relationship between water optical properties and SSC. By integrating these methodologies, we achieved an average relative calibration error of 0.26 for each optical cluster, and an average relative station deviation of 0.24 based on in situ measurements, minimizing SSC calibration to acceptable levels. Additionally, our results reveal that geomorphic patterns significantly influence sediment yield and transport by regulating sediment sources and sinks, fluvial morphology, and water-sediment connectivity. Over the past two decades, approximately 35.73 % of the sediment relative to the basin outlet discharge in the SRYR has been temporarily stored or confined within sediment sinks. These methods and findings hold significant implications for assessing and projecting fluvial sediment dynamics and the associated ecological and environmental issues in ungauged cold headwater regions.

中文翻译：

---

使用 Google Earth Engine 绘制和重建青藏高原长江源区悬浮沉积物动力学（1986-2021 年）


利用遥感技术测量山区河流中的悬浮沉积物浓度 （SSC） 可以弥补原位沉积物观测的稀缺性，为观测记录提供有价值的直接补充。然而，对于内陆河流，遥感 SSC 评估面临数据质量、长期水体变化、环境噪声、洪水事件以及本地校准的可转移性等挑战。在这里，我们介绍并应用了 12,445 张无云 Landsat 5、7 和 8 卫星影像的遥感大数据技术来校准长江源区 （SRYR） 的 SSC。利用 Google Earth Engine，我们实施了一系列图像预处理技术和含水分数方法，以提取精确的内陆河流水掩膜。然后，我们使用无监督 K-Means 聚类和机器学习算法来模拟水的光学特性与 SSC 之间的关系。通过集成这些方法，我们实现了每个光集群的平均相对校准误差为 0.26，基于原位测量的平均相对校准误差为 0.24，从而将 SSC 校准最小化到可接受的水平。此外，我们的结果表明，地貌模式通过调节沉积物源和汇、河流形态和水-沉积物连通性，显着影响沉积物的产量和运输。在过去的二十年里，相对于 SRYR 中流域出口排放的沉积物中，大约 35.73% 的沉积物被临时储存或限制在沉积物汇中。这些方法和发现对于评估和预测河流沉积物动态以及未测量的冷源头地区的相关生态和环境问题具有重要意义。