Deep Learning Application on Volcanic Deformation Detection and Blind Source Separation in InSAR Data
Abstract
Interferometric Synthetic Aperture Radar (InSAR) provides (sub-)centimetric measurements of ground surface displacements, which are key to characterize and monitor magmatic processes in volcanic regions. The abundant measurements of ground surface displacements in multitemporal InSAR data routinely acquired by SAR satellites, such as Sentinel-1 or COSMO-SkyMed,facilitate near real-time volcano monitoring on a global basis. However, the presence of atmospheric signals in time-series of interferograms complicates the interpretation of those measurements. Indeed, atmospheric effects may mislead identification of new deformation signal and misinterpret volcanic unrests. Given the vast quantities of SAR data available, an automatic data processing tool is required to separating volcanic and atmospheric signals and to reveal the true ground surface displacements. Deep learning (DL) techniques have progressively shown strong ability and enormous superiorities on various tasks in diverse fields such as: denoising, image and natural language processing. Benefiting from the evolution of DL, we design a fully convolutional neural network with an encoder-decoder architecture to eliminate the atmospheric noises by taking the multitemporal interferograms as inputs. After training with synthetic interferograms, the practical InSAR data are fed into the model to examine its capacity and generalization ability. The time series analysis of generated outputs indicates our proposed end-to-end network has a great power of suppressing atmospheric noises and revealing the true surface deformation, which are also validated against ground-based GPS data.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2019
- Bibcode:
- 2019AGUFM.G13C0556S
- Keywords:
-
- 1240 Satellite geodesy: results;
- GEODESY AND GRAVITY;
- 1241 Satellite geodesy: technical issues;
- GEODESY AND GRAVITY;
- 1294 Instruments and techniques;
- GEODESY AND GRAVITY;
- 1295 Integrations of techniques;
- GEODESY AND GRAVITY