Datensatz

Zusammenfassung

The Global Navigation Satellite System (GNSS) is widely acknowledged for its ability to monitor ground deformation and provide guidance to assess associated hazards. However, noise in GNSS time-series can hide or even mask the actual ground deformation signals. Various denoising techniques have been developed to improve the signal-to-noise ratio and detect low amplitude signals. The Discrete Wavelet Transform (DWT) has proven to be one of the most effective techniques. However, the DWT requires extensive time-series data and it is therefore computationally expensive, making it unsuitable for real-time monitoring.In this research, we first assess the feasibility of using deep learning (DL) to perform the equivalent of wavelet analysis on GNSS data. Secondly, we explore the possibility of using DL to predict the future values of a wavelet-denoised GNSS time-series. The proposed method can be described as follows: i) wavelet analysis is applied to GNSS time-series coming from different sites in a permanent network; ii) a DL model is trained using the original time-series as input and the "Wavelet processed" series as target; iii) the trained model is used to perform real-time denoising on newly recorded GNSS data; iv) a separate model is trained to predict future values of the so-denoised GNSS data.We tested our approach on GNSS data collected in the Campi Flegrei area (Naples, Italy), an active volcanic caldera well-known for its ongoing deformation. The preliminary results are promising, as the models show good accuracy in both tasks: simulating past denoised signals and predicting future ones.