A Novel Framework for Gravity Anomaly Decomposition: Iterative Polynomial Fitting and K-Means Clustering with Application to the Zagros Collision Zone

This study presents an approach that integrates iterative polynomial fitting, dynamic weighting based on regional effects, K-means clustering, and redistribution of long-wavelength anomalies to isolate subsurface features. A high-degree polynomial is initially fitted to gravity data to capture the overall trend, followed by iterative reduction of the polynomial degree to highlight residual components. Tikhonov regularization prevents overfitting, while dynamic weighting adjusts residuals according to regional effects. K-means clustering classifies residuals into features such as sediment-basement interfaces, crustal roots, and the lithosphere-asthenosphere boundary, with clusters selected based on frequency characteristics and statistical robustness. The final anomaly for each feature is reconstructed by averaging cluster residuals, and remaining long-wavelength anomalies are redistributed to represent deeper layers accurately. The method, termed Iterative Polynomial Decomposition and K-means Clustering, is validated on three synthetic models simulating continental, oceanic, and rift environments. Results show improved decomposed layers, minimizing edge effects, dislocation, and over-smoothing. Applied to the Zagros collision zone, the crustal gravity response aligns with prior studies, e.g., a negative anomaly in the Zagros foreland basin (thick sediments) and a positive anomaly beneath the South Caspian Basin (dense lithosphere). Comparisons with seismic Moho models reveal deep Moho signatures in eastern Alborz and Kopeh Dagh, but discrepancies in northwestern Iran and Talesh suggest shallower gravity-derived Moho response. Across the Main Zagros Fault, gravity and seismic responses diverge, reflecting lateral density variations, partial melting, and method-specific sensitivities. In stable regions, gravity and seismic models converge, whereas in active collision zones, complex crustal gradients lead to differences in interpretations.