Supervised Machine Learning Algorithms for Ground Motion Time Series Classification from InSAR Data

The increasing availability of Synthetic Aperture Radar (SAR) images facilitates the generation rich Differential Interferometric SAR (DInSAR) data. Temporal analysis DInSAR products, and in particular deformation Time Series (TS), enables advanced investigations for ground identification. Machine Learning algorithms offer efficient tools classifying large volumes In this study, we train supervised models using 5000 reference samples three datasets to classify TS five trends: Stable, Linear, Quadratic, Bilinear, Phase Unwrapping Error. General statistics features are also computed from assess classification performance. proposed methods reported accuracy values greater than 0.90, whereas customized significantly increased Besides, importance was analysed order identify most effective classification. were tested on 15000 unlabelled data compared a model-based method validate their reliability. Random Forest Extreme Gradient Boosting could accurately positively assign correct labels random samples. This study indicates efficiency management TSs, along with shortcomings nonmoving targets (i.e., false alarm rate) decreasing shorter TS.