Automatic Detection of Turning Points for Partial Trend Analysis in InSAR Deformation Time Series: EGMS Ortho Products

Identifying turning points (TPs) in time series is essential for understanding changes in trend behaviour, especially in geodetic applications such as ground deformation monitoring. This study introduces a new automatic TP detection framework based on robust local regression models, specifically by contrasting Local Linear Regression (LLR) using Locally Weighted Scatterplot Smoothing (LOWESS) and Local Quadratic Regression (LQR) using Locally Estimated Scatterplot Smoothing (LOESS). The difference between these models is smoothed into a diagnostic signal, where extrema indicate potential TPs. A simulation-based calibration is developed to relate the prominence of each extremum to the magnitude and sign of gradient change, enabling both localisation and quantification of deformation transitions. The method is first validated using synthetic time series with known structural changes under varying noise and sample size conditions, demonstrating high accuracy in both TP positioning and gradient estimation. It is then applied to real-world InSAR deformation time series (DefTS) from the European Ground Motion Service (EGMS), confirming the method’s ability to detect significant changes in displacement trends across both vertical and horizontal components. While sensitive to periodic behaviour and sudden jumps, the approach offers a promising step toward statistically grounded, automatic and interpretable TP detection in large-scale InSAR DefTS.