A deformation decoupling and parameter inversion method considering mechanical causes was proposed, to address the difficulties and limited accuracy associated with direct parameter inversion in stress-deformation analysis and safety evaluation of earth-rockfill dams. Settlement monitoring data at the monitoring points on the dam crest were analyzed using a statistical model, whereby the settlement deformation was decoupled into water-level-dependent and time-dependent components, thus enhancing the accuracy of deformation component characterization. An optimization framework combining BP neural networks and genetic algorithms was then established for different components to separately invert the parameters of the wetting and creep models, thereby obtaining key parameters that reflect the actual mechanical behavior of the dam and improving the accuracy of parameter inversion through differentiated analysis. Based on the inverted parameters, finite element analysis was performed to analyze deformation characteristics and crack development of the dam. Results demonstrated that settlement values calculated with the inverted parameters showed close agreement with field monitoring data across the entire crest, with deviations controlled within a reasonable range, verifying the reliability and accuracy of the proposed method. Moreover, the crack patterns derived from the inverted parameters exhibited high consistency with observed conditions. The findings provide a new technical approach for mechanical parameter inversion of earth-rockfill dams.