Catalog-independent detection and removal of coseismic steps in InSAR time series using shared changepoints and spatial regularization

InSAR time series are widely used to characterize long-term surface deformation, yet coseismic steps can distort displacement histories and bias velocity estimates if they are not explicitly identified and separated. We present a catalog-independent framework to detect and remove multiple coseismic steps from large InSAR time series datasets by exploiting a characteristic earthquake signature: many pixels exhibit near-synchronous displacement steps, while step amplitudes vary spatially. After reducing slowly varying components (e.g., linear trend and seasonal terms), we identify a sparse set of shared changepoint times across displacement histories using a multi-signal shared-changepoint model, enabling recovery of multi-event sequences within a single observation period. For each detected changepoint, we estimate pixel-wise step amplitudes using robust windowed statistics and/or step regression, and then regularize each event’s step-amplitude field on a spatial neighborhood graph using total-variation regularization to enforce spatial consistency, suppress outliers, and preserve sharp gradients expected near faults. Subtracting the regularized steps from the original time series yields de-evented displacement histories and updated long-term deformation rates. The approach is scalable, supports repeated and closely spaced events via joint estimation of multiple steps, and does not require prior event timing information. Applied to multi-year regional InSAR products, the method produces cleaner time series, reduced residual variance, and more stable velocity estimates, improving characterization of gradual deformation in tectonic and volcanic settings.