Interpretable machine learning-based analysis of driving factors for inverted channel morphology in the Qaidam Basin

Weng, Xuhua , Chen, Ninghua , Gao, Bowen , Jiang, Xinhao

2026-05-01 GEOMORPHOLOGY 2026   496(卷), null(期), (null页)

查看原文

  • JCR分区:

    影响因子:

  • Inverted fluvial channels serve as important paleoenvironmental indicators and are extensively distributed in arid regions of Earth and on other terrestrial planets such as Mars. The Qaidam Basin, which is considered an important analogue environment for Mars, contains abundant inverted channels. However, a comprehensive understanding of their distribution characteristics and the influence of environmental factors on their development is still lacking. This study focuses on revealing the driving factors of the inverted channels and their implications for paleoenvironments. The morphological characteristics of inverted channels in the Qaidam Basin show spatial heterogeneity. To unravel the governing mechanisms behind this spatial pattern, this study employed an integrated machine learning and SHAP interpretation approach with statistical techniques, including Partial Least Squares Structural Equation Modeling (PLS-SEM) and a Geographical Detector model. Our analysis reveals a multi-stage evolution governed by shifting environmental controls. The initial formation of inverted channels is predicated on substrate contrast that the differential erodibility between competent channelfill deposits and the surrounding weaker substrate. After the inverted channels are formed, the preserved length of channel segments is determined by a balance between initial fluvial scale and subsequent aeolian destruction, with topographic complexity providing critical shielding against erosional forces. High channel sinuosity is a well-preserved relic of meandering systems from wetter paleoclimates, while low sinuosity is a feature imposed by intense aeolian erosion that straightens the channel. This framework of controls directly explains the spatial patterns. This study enhances our understanding of inverted channel formation and evolution under multi-factor control, providing insights for paleo-hydrological reconstruction in the Qaidam Basin.