A STEP-FURTHER TO THE AUTOMATIC IDENTIFICATION OF CO-SEISMIC DISPLACEMENTS ON THE EPOSAR DINSAR MAPS GLOBAL ARCHIVE

We present in this study an enhancement to our previous work, in which an automated method, based on Convolutional Neural Networks (CNNs), has been developed for identifying co-seismic ground deformation patterns within the Differen- tial SAR Interferometry (DInSAR) maps generated through the EPOSAR service of the European Plate Observing Sys- tem (EPOS) Research Infrastructure. The implemented im- provements have been achieved through two main lines of ac- tions. Firstly, the generation of the synthetic dataset, used to train the developed CNN, has been enhanced; this concerns the improvement of the simulation of possible atmospheric disturbance within the DInSAR interferograms, and the intro- duction of a simulator of phase unwrapping errors. Secondly, the original CNN-based deformation pattern detector layout, performing a binary classification, has been modified to ac- count for a multiclass deformation pattern classification. This allows us extending the capability of the system which, in addition to detect co-seismic deformation patterns, may also provide information on the earthquake source characteristics. The presented solution will be included in the EPOSAR DIn- SAR processing chain.