Processing of massive seismic datasets at Campi Flegrei (Italy) through Convolutive Indepedent Component Analysis
A novel procedure is proposed in order to analyse continuous seismic signal on hourly scale to have a prompt discrimination among the different sources. The case study regards the Campi Flegrei caldera during the year 2006 when a swarm of volcano-tectonic earthquakes occurred. The necessity to analyse a massive data set has required applying a robust methodology and the introduction of suitable parameters to be monitored over the time. Specifically we apply the Convolutive Independent Component Analysis to the seismic recording at four broadband stations. As a result, we obtain a clear separation among meteo-marine, anthropogenic noise, hydrothermal tremor in absence of volcano-tectonic activity, whereas in non-stationary conditions a contribution connected to the corner frequency of the earthquakes emerges. We introduce a coarse-grained variable to be monitored continuously, i.e. the frequency associated with the maximum amplitude of the power spectral density of the deconvolutive independent components. This parameter is sensitive to the variation in the frequency bands of interest (e.g. that corresponding to the corner frequencies of volcano-tectonic events) and can be used as marker of the insurgence of seismic activity. We propose the following procedure to be applied routinely in the observatory practice: namely, estimate CICs on hourly series; then represent the distribution of the FMPSDA. Significative variations in the frequency bands of interest can be indicative of the insurgence of a renewed activity (e.g. VTs). Once individuated the “hot” periods, then one can go deeper with finer distinctions at a single event scale by using a simple STA/LTA (Short Time Average vs. Long Time Average) technique in order to detect events. This coarse-grained procedure on massive data through CICA would provide fast alert on the occurrence of even very-small VTs and FMPSDA may represent a suitable “observable” to monitor in the observatory practice. Finally, this approach can be employed for the prompt detection in massive data of other kinds of seismic signals such as LP, tremor, fluid-induced seismicity buried in noisy environments.
Coarse grained parameters for detection of volcano-tectonic activity inferred by CICA
The main aim of this work is to recognize the occurrence of seismic VT activity and its discrimination among other active and persistent natural and/or artificial sources, such as meteo-marine and anthropogenic noise. This task is of interest because it would support the routine practices of the observatories, considering that on the hourly scale events of very small energy are often completely hidden and not even the tedious work of operators can be resolved. To achieve that goal we adopt a robust automatic method, namely the Convolutive Independent Component Analysis (CICA), which in involves higher-order statistics in frequency domain. This technique is successful in seismological framework in the case of seismic signals, which can be considered as the convolution of time delayed source signals. In this work, we focus on Volcano Tectonic (VT) activity at Campi Flegrei Caldera (Italy) during the 2006 ground uplift. The activity is characterized approximately by 300 low-magnitude (Md < 2; for the definition of duration magnitude, see Petrosino et al., [2008]) VT earthquakes. Most of them were concentrated in distinct seismic sequences with hypocentres mainly clustered beneath the Solfatara-Accademia area, at depths ranging between 1 and 4 km b.s.l.. CICA is fruitfully applied to massive data on hourly scale obtaining a separation among different independent sources. Specifically, the identification of meteo-marine (< 1 Hz), anthropogenic noise (mainly affecting [8-14] Hz frequency range), and hydrothermal tremor (at about 0.8 Hz) is achieved in absence of VT activity, defining the background level. Variations of that underlying condition appear in approaching to the period of intense low-energy VT activity. Namely, a further component in 13-15 Hz compatible with the typical corner frequency of VTs is extracted. We propose a coarse-grained procedure directly applied to massive data separated through CICA, which would provide fast alert on the occurrence of even very-small VTs representing a suitable “observable” to monitor in the observatory practice.