Abstract: The morphological characteristics and compositional patterns of cross-sections in mountainous river channels constitute a fundamental basis for remote sensing-based runoff estimation, regional riverine ecological management, and flood control strategies. In this study, water surface vector files of mountainous rivers were extracted from remote sensing data, upon which a comprehensive index system for cross-sectional morphological features was constructed. Using the K-means clustering method, we identified classifications of individual cross-sections, river reaches, and their combined morphological assemblages.The results indicate that the upper reaches of the Jinsha River exhibit five distinct types of cross-sectional morphology. Among them, the symmetric gorge-controlled type (Category 0) dominates, accounting for 76.98%, and shows a strong correlation with the prevailing hydrodynamic conditions. Furthermore, the complexity of river reach morphology increases with reach length; clustering categories rise from 5 to 19 as reach lengths extend from 300 m to 1500 m. Short reaches tend to be governed by singular hydrodynamic processes, while longer reaches are significantly influenced by topographic–hydrologic interactions.During the dry and flood seasons, the mean ratio of low-flow to high-flow water surface width is 0.52. In the flood season, a rightward flow deviation is observed (mean right-bias index rx = 1.18), consistent with the Coriolis effect. These findings provide scientific support for runoff estimation in data-scarce regions using remote sensing, hydrodynamic simulation in mountainous rivers, ecological restoration and flood risk management, as well as rapid identification of monitoring sites under emergency monitoring scenarios.