Integrating machine learning and geomorphic indices to assess tectonic geomorphology in the Karoun River Basin, Zagros, SW Iran

The geomorphological expression of tectonic activity in the Karoun River Basin, a tectonically active region within the Zagros Mountains of Iran, was investigated. The basin's complex geological setting, influenced by the collision of the Arabian and Eurasian plates, provides an optimal environment for analyzing tectonic-geomorphic interactions. A combination of geomorphic indices, including the Stream-Gradient Index (SL), Asymmetric Factor (Af), Hypsometric Integral (Hi), and Valley Floor Width to Valley Height Ratio (Vf), was employed to assess tectonic influences on landscape development. Elevated SL values were interpreted to suggest active uplift, while asymmetric drainage patterns (Af) and narrow valley profiles (Vf) were found to further support tectonic dominance. Hypsometric analysis (Hi) indicated youthful landforms undergoing continuous tectonic modification. To augment conventional geomorphic assessments, advanced machine learning techniques, specifically Random Forest (RF) and Convolutional Neural Networks (CNNs), were utilized to model the spatial distribution of geomorphic indices. Interpretability methods such as SHAP (SHapley Additive exPlanations) were applied to elucidate the relationships between tectonic processes and geomorphic features, enhancing model accuracy and mechanistic interpretation. The Index of Active Tectonics (Iat), derived through GIS-based analysis, was used to categorize the basin into three tectonic activity classes: Class 1 (very high activity, 24% of the area), Class 2 (high activity, 63%), and Class 3 (moderate activity, 10%). The findings highlight the significance of tectonic geomorphology in natural hazard evaluation, particularly landslide susceptibility in steep, tectonically uplifted terrains. Additionally, the examination of river terraces contributes to understanding historical landscape responses to tectonic and climatic forcing, advancing knowledge of long-term geomorphic evolution. The integration of traditional geomorphic indices with machine learning establishes a robust analytical framework for future research in tectonically active regions, with implications for geological hazard assessment and environmental planning.