WIDE-AREA SENTINEL-1 DEFORMATION CLASSIFICATION FOR ADVANCED DATA EXPLOITATION

The WISE (Wide-Area Sentinel-1 Deformation Classification  for  Advanced  Data  Exploitation)  project focuses on developing advanced tools to exploit the large amount of satellite-borne Synthetic Aperture Radar (SAR) data. The main data source will be the Sentinel-1 SAR data: this Copernicus mission provides, with its fast revisit time, high-resolution and multi-polarization, an unprecedented amount of data, which need to be fully exploited.The  thematic  area  covered  by  this  project  is  the  study  of  the  ground  displacements  induced  by  different  phenomena, affecting the natural and the built environment. The deformation will be derived using Persistent Scatterer Interferometry (PSI). The project shall produce maps of labelled deformation areas, geo-located and categorized by their type. This is key to perform a systematic and comprehensive exploitation of deformation datasets over wide areas.  The project will exploit the PSI data containing the displacement time series of radar scatterers distributed all over Europe. A  relevant  data  source  that  will  be  processed  in  this  project  is  the re  cently-launched  European  Ground  Motion  Service  (EGMS) as part of the Copernicus Land Monitoring Services, based on Sentinel-1 PSI data. The European Environment Agency, deputing entity for the Service management, is intended as a fundamental stakeholder. The data analysis proposed in WISE  unfolds  following  a  rigorous  processing  chain.  First,  active  deformation  areas  (ADAs),  such  as  landslides, subsidence,  infrastructure  stability,  will  be  detected  employing  a  statistical  hypothesis  test.  This  operation  allows  preliminary  data  selection  and  removal  of  noisy  pixels,  yielding  clusters  of  pixels  whose  deformation  time  series  are correlated  in  both  space  and  time.  Second,  the  project  shall  develop  automatic  classification  methods  for  the  ADAs detected.  A  first  method  will  identify  the  features  necessary  to  uniquely  define  a  deformation  class,  then  traditional  classifiers will yield the first set of labelled deformation areas. Machine Learning techniques will be adopted to derive more advanced  classifiers.  An  accurate  deformation  classification  technique  cannot  leave  aside  the  temporal  information  associated to the persistent scatterers, hence classifiers using one or more memory layer will be employed to underpin this information.  The  novelty  of  such  methods  will  be  manifold,  as  they  have  been  little  utilised  in  the  Earth  Observation  domain, and certainly not over an area as wide as Europe. The developed algorithms will be tested on SAR data collected over smaller areas of interest by different sensors, such as COSMO-SkyMed and TerraSAR-X, enabling the evaluation of frequency-diverse data. The labelled ADAs will be associated with metrics quantifying the effective likelihood of the classification performed, supporting potential users of the producedmethods  and  datasets.  The  produced  maps  will  be  validated  using  inventories  and  in-situ  data  of  known  deformation  phenomena, by selecting representative test sites.The main output will be a database containing labelled deformation areas: a useful product for several end users of ground motion data, e.g. Civil Protection actors and risk management entities. Such a product will support the exploitation of theproducts  of  EGMS,  Sentinel-1  and  other  SAR  mission.  The  developed  software  will  be  freely  available  with  different  versions depending on the users’ experience in processing EO data.