Bayesian Gravity Inversion for Salt-Dome Localization Using an Adaptive Differential-Evolution Markov Chain Algorithm: Synthetic Tests and Application to Real Data from the Ajichay Area

Salt domes, due to their significant density contrast with host rocks, are suitable targets for gravimetric studies. However, inverting gravity data to recover their geometry remains challenging due to inherent non-uniqueness and nonlinearity. This study revises a Bayesian inversion algorithm based on the adaptive differential evolution Markov chain method for 2D modeling of a salt dome as a trapezoidal polygon, using the Talwani formula as the forward model. Two fundamental issues were identified in the existing code: failure to transition the chain state upon candidate rejection, causing chain trapping at zero, and incorrect array dimension ordering, mixing time and parameter axes when extracting posterior samples. After fixing these issues, the algorithm was enhanced with a robust L1-norm likelihood function, along with adaptive tuning of noise scale and jump step size. The revised method was validated on synthetic data from a salt dome with known parameters across four noise levels (0–15%). The method was then applied to real gravimetric data from the Aji-Chay region, comprising 110 stations across ten survey lines. After regional trend separation and independent inversion of each line, two unreliable edge lines were discarded using a model-independent criterion. Results from the remaining five lines revealed a structure with negative density contrast and a roughly north–south trend within a specific geographic range. The estimated top depth, root depth, and density contrast show reasonable agreement with two previous published studies on the same dome.