DeepQuake — An application of CNN for seismo-acoustic event classification in The Netherlands. (February 2022)

• Record Type: Journal Article
• Authors: Trani, Luca; Pagani, Giuliano Andrea; Zanetti, João Paulo Pereira; Chapeland, Camille; Evers, Läslo
• Is Part Of: Computers & geosciences. Volume 159(2022)
• Publication Date: 2022-02
• Subjects: Deep learning -- Explainable AI -- Low-magnitude event classification -- Machine learning -- Seismo-acoustic source detection; Environmental policy -- Periodicals; 550.5
• DOI: 10.1016/j.cageo.2021.104980
• Languages: English
• ISSNs: 0098-3004

Abstract

Abstract: Recent developments in instrumentation and methodology are major driving forces in the advances of solid Earth sciences. The deployment of large and dense sensor networks enables data centres to acquire data of increased volume and quality. The analysis of such data provides scientists with a better understanding about natural phenomena in the subsurface. At the same time new challenges arise to exploit the growing information potential. Innovative methods based on Artificial Intelligence offer concrete opportunities to tackle those challenges. In this paper we investigate Convolutional Neural Networks (CNN) for seismo-acoustic event classification in The Netherlands. A supervised deep learning approach is applied to continuous seismic waveforms in order to automatically detect seismo-acoustic events and discriminate their sources. Our region of interest is characterised by high seismic noise conditions and intense anthropogenic activity which deeply influence the recorded signals. Site conditions and our aim for a tool that could support operational earthquake analysis motivated us to devise a novel specialised CNN architecture. Building on existing approaches, we designed, trained and evaluated two CNN models: arch-time and arch-spect . Our results suggest that as CNN inputs, spectrograms are more suitable than continuous waveforms. Therefore, we propose a model (arch-spect ) that achieves a performance among the highest in modern source detection systems. Furthermore, we focus on explainability aspects and propose a method to gain insights in the behaviour of a 'black-box' deep learning classification system. By generating visualisations of activations we aim to empower domain experts with a useable cross-checking tool and pave the way for a broader operational uptake of Artificial Intelligence based methods. Highlights: Assessing feasibility of DL-based detection of low-magnitude induced seismo-acoustic events. Evaluating performance in comparison with an operational seismic detection system. Introducing a visualisation mechanism for the inspection of CNN (intermediate) results.