Gravity Anomaly Decomposition Using Iterative Polynomial Fitting and K-Means Clustering: Application to the Zagros Collision Zone
Description
This study presents an approach for gravity anomaly decomposition that integrates iterative polynomial fitting, dynamic weighting based on regional effects, K-means clustering, and redistribution of broad regional trend. The gravity field is initially fitted using a high-degree polynomial, and the resulting residual is retained as a decomposed component and removed from the gravity field before fitting the next lower-degree polynomial; this procedure is repeated iteratively to generate a set of residual components with different wavelength characteristics. Tikhonov regularization is applied to control overfitting, while dynamic weighting adjusts the contribution of residual components according to their regional effects. K-means clustering is then used to group residuals according to their frequency characteristics and statistical properties. The selected clusters are averaged to reconstruct mathematically defined short-, intermediate-, and long-wavelength gravity components, while the remaining broad regional trend contribution is redistributed among the decomposed components according to their anomaly ranges. The method, termed Iterative Polynomial Decomposition and K-means Clustering (IPDKC), is evaluated using three synthetic models representing continental, oceanic, and rift environments. The synthetic tests assess the recovery of the prescribed wavelength components while limiting edge effects, spatial displacement, and excessive smoothing. The resulting components are interpreted in terms of their potential geological sources rather than as direct representations of specific geological interfaces. Application to the Zagros collision zone separates the Bouguer anomaly into short-, intermediate-, and long-wavelength components. The intermediate-wavelength component exhibits a generally negative response across the Zagros Mountains and a positive response beneath the South Caspian Basin. These responses reflect the combined effects of sedimentary thickness, crustal density and thickness variations, and deeper lithospheric structure. The repository includes: The main IPDKC script and documentation. A synthetic model generator for continental, oceanic, and rift settings. Comparison methods (bandpass, adaptive filtering, upward continuation, wavelet, and Wiener filtering) used in the manuscript. Processed gravity data and synthetic grids. All code is written in MATLAB. Detailed instructions and parameter guides are provided in the individual README files.