InSAR Displacement Time Series Mining: A Machine Learning Approach

Abstract

Interferometric Synthetic Aperture Radar (InSAR)-derived surface displacement time series enable a wide range of applications from urban structural monitoring to geohazard assessment. With systematic data acquisitions becoming the new norm for SAR missions, millions of time series are continuously generated. Machine Learning provides a framework for the efficient mining of such big data. Here, we focus on unsupervised mining of the data via clustering the similar temporal patterns and data-driven displacement signal reconstruction from the InSAR time series. We propose a deep Long Short Term Memory (LSTM) autoencoder model which can exploit temporal relations in contrast to the commonly used shallow learning methods, such as Uniform Manifold Approximation and Projection (UMAP). We also modify the loss function to allow the quantification of uncertainties in the time series data. The two approaches are applied to the Lazufre Volcanic Complex located at the central volcanic zone of the Andes and thereby compared.

Trend Classification of InSAR Displacement Time Series Using SAE–CNN

“…Recent works have explored machine-learning (ML) approaches to handle massive InSAR data [11][12][13][14] and identify abnormal InSAR time-series displacement signals [15]. Ansari et al [16] and Martin et al [17] proposed a deep long short-term memory (LSTM) autoencoder model and a deep temporal clustering (DTC) algorithm respectively to learn and cluster time-series displacements. A method for monitoring surface deformation was proposed by van de Kerkhof et al [18] using t-distributed stochastic neighbor embedding (t-SNE) dimension reduction and density-based clustering non-parametric algorithm (DBSCAN) clustering.…”

A Clustering Approach for the Analysis of InSAR Time Series: Application to the Bandung Basin (Indonesia)

“…However, until a few years ago InSAR interpretation was mainly limited to analysis of the average displacement rates [19], but advances in innovative big data analysis methods change this and enable exploitation of the full displacement time series [20][21][22] In the case of large InSAR datasets, traditional manual analysis is a complex and timeconsuming process and more automated techniques are necessary. Previous work has used a variety of methods to help interpret InSAR time series, ranging from semi-automatic [21,23,24] and automatic statistical approaches [25], to the use of supervised [26,27] and unsupervised [20,28,29] machine learning algorithms. These studies have aggregated MPs based on their average velocities [27] but inevitably fail to identify non-linear movements.…”

“…We implemented an unsupervised clustering technique to identify patterns and features in the data that may not be visible by traditional clustering methods: our approach involves the use of Uniform Manifold Approximation and Projection (UMAP) [32] for dimensionality reduction followed by the Hierarchical Density-based Spatial Clustering of Applications with Noise (HDBSCAN) [33] algorithm. Our study expands on a recent application of this approach [29] in two ways: (1) combining both ascending and descending satellite observation geometries to retrieve the vertical and east-west displacement time series to get a more accurate estimation of the vertical movement, and (2) introducing a change detection method to determine when and in which manner the deformation trend changes over time to aid in further interpretation of the data. It is often the case that a single linear model cannot adequately describe the evolution of displacement over time; instead, multiple linear/polynomial relationships applying to different time spans may be more appropriate [34][35][36][37].…”

Use of LSTM for Sinkhole-Related Anomaly Detection and Classification of InSAR Deformation Time Series

“…Modelling schemes such as Multiple Hypothesis Testing (MHT) are well established in using regression to model such patterns. However, a major problem is to classify such patterns, such that the models scale well over large deformation datasets [23]. Classification of these patterns can be done using deep-learning methods which have the advantage of transforming time series in a multi-dimensional feature space where the samples can be separated using a multi-dimensional plane efficiently.…”

“…RNN such as Long Short Term Memory (LSTM) [28] have been used for forecasting of deformation time series using TSIn-SAR derived data [29]. Unsupervised classification of InSAR deformation time series using an LSTM auto-encoder model has been attempted by [23]. Furthermore, simulated data has been used to train LSTM networks to identify anomalies in InSAR deformation time series over Italy using Sentinel-1 data [22].…”