Multi-Temporal Anomaly Detection for SAR Earth Observations

Principal Investigator (PI): Hook Hua, NASA's Jet Propulsion Laboratory

Project Summary

Synthetic aperture radar (SAR) uses remote sensing to produce fine-scaled measurements of the Earth's surface. SAR-based geodetic imaging has revolutionized Earth science research in many areas, including studies of the solid earth, ecosystems, and cryosphere. When used correctly, SAR data can also be used to monitor and detect warning signs of natural hazards such as volcano inflation or precursory signals to landslide events.

Yet the ability to effectively utilize SAR data for research, for long-term monitoring of spatial areas of interests (AOIs), and for rapid hazard response has been limited. Barriers include large data volumes, processing complexity, and long latencies. For example, barriers to rapid hazard response include the lack of automated data triggers from forecasts, the need for specialized processing parameters that currently rely on expert intervention, and the manual delivery of actionable science data products to decision support communities.

Our team's AIST-2011 and AIST-2014 efforts towards an Advanced Rapid Imaging and Analysis (ARIA) data system successfully demonstrated the capability to automate high-volume SAR image analysis in a cloud computing environment. However, a limiting factor has been the continued need for expert analysis for detection of features in the Level 2 data products as well as transients in the Level 3 time-series data products.

ACCESS 2017 Multi TemporalDecision support products, which are most useful if they are generated rapidly and with simplified information (e.g., "damaged or not damaged" or "flooded or not flooded"), require change detection-based approaches that utilize before and after event scenes. These change detection approaches are often processed with threshold values set on the underlying SAR measurement values of either amplitude or coherence. The steps in change detection, which require a "human-in-the-loop," have become a bottleneck for rapid and reliable exploitation of geodetic SAR data for both long-term monitoring and event rapid response.

Automating these time domain-based feature detection procedures is challenging because of the complexity of processing, the need to process large temporally co-registered data stacks, and the human expertise needed to assess the time domain signals. Machine learning approaches for Earth science data have typically been applied to single scene feature detection. For example, in our AIST-2014, we prototyped automated classification of phase unwrapping features in processed L2 interferograms from the Sentinel-1A/ B data streams. However, for change detection techniques, most methods have focused on paired "before/after" observations.

With the greater availability of low-latency and global multi-temporal remote sensing data, opportunities exist to exploit detection of time-dependent features of highly temporal Earth science observations. Multi-temporal spatial prediction techniques that leverage long-term historical observations can yield more accurate and more interpretable predictions than more commonly used pair-wise change detection techniques.

This project will 1) on-ramp onto the cloud computing platform, an automated and large-scale pre- processing of multi-temporal data stacks, and 2) develop automated large-scale machine learning analysis of multi-temporal transient detection and precursory signal analysis of multi-temporal stack datasets. We will demonstrate real science value by exploiting NASA's Sentinel-1A/B archive in preparation for the upcoming NASA-ISRO SAR (NISAR) mission.

All input data will be acquired by discovering and accessing NASA's Earth Science Data and Information System (ESDIS) Distributed Active Archive Center (DAAC) data holdings via queries to the EOSDIS Common Metadata Repository (CMR). We will demonstrate this by pre-processing co-registered stacks of single looks complex data in the Amazon Web Services cloud. The generic time-dependent anomaly detection approach will exercise example use cases such as automated landslide detection, automated volcanic uplift early detection, and/or automated detection of pre-event time-series patterns vs co-event flood detection.

Midterm Update

First year accomplishments:

  • Set up an AWS-based SAR analysis system using open source software
  • AOI-based usage of ESDIS DAAC mirror of Sentinel-1A/B in NGAP in AWS us-east-1
  • Co-registered L1 S1-SLC stack generation
  • Persistent Scatterer (PS)-based L3 time series generation from Sentinel-1A/B
  • Machine Learning (ML) model training
    • Landslides
    • Floods
  • Initial prediction code applied to L3 PS time series processing pipeline

Page Last Updated: Feb 18, 2020 at 1:38 PM EST