Abstract: Land subsidence in loess-filled areas exhibits complex evolutionary mechanisms. Conventional linear models based on time-series interferometric synthetic aperture radar (InSAR) frequently face challenges in monitoring the nonlinear creep of loess. In this context, this study proposes an InSAR-based method that considers the nonlinear creep behavior. In this method, an adaptive thresholding technique based on tangent angle variations is used to detect and segmentally estimate the fill distribution and thickness. In the small baseline subset InSAR (SBAS-InSAR) framework, the constitutive equations for calculating the nonlinear creep of loess based on the fractional calculus theory are integrated with a differential interferometric phase model. The Chernobyl disaster optimizer (CDO) algorithm is employed to estimate creep parameters, followed by deformation calculation. Experiments carried out in the Lanzhou New Area suggest that, compared to the SBAS-InSAR method based on conventional linear models, the method developed in this study increased the residual-phase modeling accuracy by 41.2% and the degree of agreement with benchmark data by 20.1%. Therefore, this study can provide data support for analyzing long-term deformations and performing early warning of hazards for loess high-fill engineering.