The Drop of Relative Velocity Variation and Coherence Values Prior to Sinabung 2013 Eruptions

Abstract The cross-correlations of ambient seismic noise at Sinabung Volcano were analyzed from February 2013 to February 2014. Many eruptions occurred during these periods, started on 15 September 2013. Looking at the variations in the coda of the correlations, two types of measurements can be distinguished associated to two types of changes: relative velocity variation and waveform decoherence. The drop of relative velocity variations and waveform decoherence were observed for each station pair of Sinabung one to two months before the first eruption. These changes in accordance to the deformation of the Sinabung edifice were estimated from geodetic measurements, since an analysis of baseline change between GPS stations indicated an inflation of the volcanic edifice prior to September 2013 eruption. The monitoring of relative velocity variations and decoherence provides insights into the ongoing processes in the volcanic edifice to assist in determining the level of volcanic activity.


Introduction
Sinabung is a stratovolcano located in the northern part of the Sumatra Island, Indonesia. Sinabung used to be dormant until its eruption in 2010. After three quiet years, renewed activity occurred on September 2013. The first eruption started on 15 September 2013, and the volcano continued to be active with several eruptions that produced ash plumes as high as 0.5−10 km above sea level, ejected materials, and water favour emision. On December 16, 2013, The Indonesian Centre of Volcanology and Geological Hazard Mitigation (CVGHM) confirmed that a new lava dome occupied the crater. On December 26, 2013, the dome had the height of 56 m and diameter of 105 m. Its volume was estimated to be about 1 million m 3 , and the growth rate was 3.5 m 3 /sec. The lava dome started its partial collapse on December 30, 2013, generating pyroclastic flows down to the southeast slope of the volcano. The lava dome grew and flowed down to the southeast, repeating its partial collapse. Sinabung continued effusively erupting viscous lava, further contributing to both the growth of lava dome and lava flow that descended the southeast flank. The erupted lava was determined to be andesite, typical of many stratovolcanoes in subduction settings.
Seismicity has long been one of the most commonly monitored aspects of active volcanoes. Seismic monitoring can give real-time data, and correlations have been established between magma movement, eruptive phenomena, and seismicity. Sinabung awakened from centuries of dormancy on August 2010. It is difficult to predict the behaviour of a volcano like Sinabung that has been dormant for a long period. Decisions regarding alert levels and evacuation orders rely on numerous parameters currently monitored by the CVGHM. The limited number of seismic station is the main problem for monitoring volcano seismicity in Indonesia. Monitoring variation in the coda of ambient noise correlations seems to be a promising tool that can be set up on active volcanoes due to the simplicity of its implementation and potential capacity to forecast eruptions even with a limited number of stations (Duputel et al., 2009). Looking at the variations in the coda of the correlations, two types of measurements associated with two types of changes can be distinguished, such as relative velocity variations associated with changes in macroscopic elastic properties of the medium; and waveform decoherence caused by changes in the geological structures, and hence the scattering properties of the medium (Obermann et al., 2013). This paper discusses the method of monitoring using seismic noise cross correlations at Sinabung as a tool for identification of volcanic activity level. The result was compared to deformation of the Sinabung edifice estimated from GPS measurement.

Data Processing
This study uses the vertical component signal of four seismic stations locating around the edifice of Sinabung Volcano (Figure 1), those are Sukanalu (SKN), Sukameriah (SKM), Mardingding (MRD), and Lau Kawar (KWR). The data are continuously recorded by the CVGHM network equipped with the short period L4C seismometers (1 s natural period) at a sampling rate of 100 Hz. The data had been processed since February 2013 until the end of Februay 2014.

Ambient Seismic Noise Cross-correlation
The noise cross-correlation function was computed by using MsNoise software (Lecocq et al., 2014). After removing the mean of the traces, the data were merged into one-day trace, bandpassed between 0.01-8 Hz to prevent the low frequency content caused by temperature variation, then decimated to 20 Hz, and split into 30-min-long data. Those 30-min-long data were clipped to 3 time RMS and whitened between 0.1 to 1 Hz. The cross-correlation function (CC) was computed between signals of all possible pairs of stations, and daily CC were obtained by stacking 30-min CC. Figure 2a shows a frequency-time analysis of the CC for station pair in Sinabung. The spectrogram is dominated by the first arrival with low frequencies corresponding to direct Rayleigh wave between two stations Duputel et al., 2009). To observe temporal velocity variations, the daily CCs were compared with a fixed reference. As a reference CC (CC ref ), the average CC was computed from February 1 to April 30, 2013. This period was chosen because it was a 'quiet' period without eruption or strong tremor. Figure 2b shows the correlation coefficients between CC ref and a CC averaged over an increasing number of days for different frequency bands. The correlation coefficient defines the similarity between the current cross-correlation and the reference for a given stacking length. The similarity of the two time series was assessed using the cross coherence between energy densities in the frequency domain (Lecocq et al., 2014): in which the overline here represents the smoothing of the energy spectra for F cur and F ref and the spectrum of X, while X (v) is cross spectrum.
The decoherence is the difference in the coherence from one date to the others. A good coherency is reached for the 0.5 to 1 Hz frequency band ( Figure 2b). The lack of coherency in other frequency bands may be due to the instrument rensponse of L4 and L4C seismometer below 0.5 Hz (Duputel et al., 2009). The degree of similarity increased rapidly for a short stacking duration, and then it tended to stabilize after ten days of stacking length. Thus, computing the 'current' CC (CC cur ) was chosen using ten-day averaged CC and 0.5 -1 Hz frequency band of interest for calculating the time shift (δt) between CC cur and CC ref using Moving Window Cross Spectral (Clarke et al., 2011). This frequency band also minimized the effect of eruption tremors which had dominant frequencies, usually ranging between 2 Hz and 4 Hz.
The time shift for each window between two signals is the slope of a weighted linear regression (WLS) of the samples within the frequency band of interest (Lecocq et al., 2014). The time shift (δt) is computed for 5 sec. sliding windows overlapping by 50%. Figure 3a shows the δt calculation result for each window on a few days during March 2013. The δt random value was seen at a lag time of less than -60 sec. and more than 60 sec. Lecocq et al. (2014) advised to avoid working around zero lag time, because they might be sensitive to the changing position of the noise source (Froment et al., 2010). For δt/t calculations, both causal and acausal parts were used, lag times between + 5 and + 60 s were selected (Figure 3b). Remaining selection parameters are 0.7 minimum coherence, 0.1 s maximum error, and 0.5 s maximum δt. If one assumes a relative velocity variation δv/v homogeneous in space and a relative time shift δt/t between the reference and the current CF, it had been demonstrated by Ratdomopurbo and Poupinet (1995) that δv/v = -δt/t. The accuracy of the linear trend measurements is significantly improved by averaging local time shifts for different receiver pairs, assuming that the seismic velocities are perturbed uniformly within the sampled medium (Brenguier et al., 2008). The average of relative velocity variations and coherence value over all station pairs ( Figure  5) had showed the velocity drop coinciding with decoherence since the end of July 2013, about fifty days before September 15, 2013 eruption.

Discussion
In addition to the seismic velocity change time series, the results were compared with the displacements measured by three GPS stations (KWR, GRK, and SKN) of Sinabung, to obtain daily relative baseline changes between GPS stations ( Figure 6a). GPS data were processed using GAMIT/GLOBK software package, which took into consideration International GNSS Service (IGS) precise ephemeris, a stable support network of ten IGS stations around Sumatra Island. All baseline distances show the increased value, indicating an inflation of the volcano (Figure 6b   stored in transitory reservoirs (Tait et al., 1989), and a slow continuous filling (Blake, 1981). These mechanisms could lead to decreasing velocity produced by tensile stress during enhanced pressurization in the volcanic edifice creating fissures and changes in the elastic properties. The drop of relative velocity variation had been tried to be compared since the end of July 2013 and the inflation from GPS result since May 2013 with the number of volcanic earthquakes. It was found that within these periods, an increase of volcanotectonic earthquake on early July 2013 occurred. On Figure 7, the average relative velocity variation and waveform decoherence were compared with the daily number of volcano-tectonic (VT) earthquake, change of GPS baseline which across Sinabung edifice and Realtime Seismic Amplitude Measurement (RSAM). RSAM was recalculated from digital raw data of SKN seismic station as: where: A i is the signal amplitude, A is the mean amplitude in the calculation window, n is the number of samples of the window.
In order to get more detailed and to separate contributions from the different types of volcanic sources, the following procedure was applied: 1. Butterworth bandpass filter was applied among the following ranges: 1 -5 Hz; 5 -10 Hz, and 10−15 Hz. 2. Then RSAM of 5-min long window was calculated for each frequency range. 3. The resulting RSAM were then averaged using 2-hour-long window. Each frequency range roughly corresponds to the different types of volcanic events. 4. The RSAM in specific frequency bands provides additional information about the nature of seismicity that helps to highlight subtle frequency that can be related to changing dynamics of magma movement. High frequency earthquakes with peak frequencies between 5 and 15 Hz, for example are the result of brittle failure of rock within the volcanic edifice (McNutt, 1996). The predominance of high frequencies suggests that much of energy was released by fracturing rock. The low frequency earthquake is dominated by frequencies between 1 and 5 Hz, and is thought to result from the resonance of fluid-filled cracks (Chouet, 1996).
The increase of RSAM was interpreted as an increase in energy released resulted from magma pressurization within the volcano plumbing sys- tem. Figure 7a shows the RSAM of 5-min. and 2-hours averaged computed for SKN station. Starting in February 2013, the average value of RSAM is almost constant in spite of some burst energy related with seismic swarm at a certain frequency range. The increasing value of KWR-GRK GPS baseline distance observed since May 2013 (Figure 7b), indicated an inflation of the volcano, preceeded by the slight increase of RSAM on March and April 2013 where some earthquakes were recorded with high amplitudes. The average value of relative velocity variation and coherence (Figure 7c) had begun to drop since the end of July 2013 preceeded by volcano inflation and increasing of VT earthquakes on early July 2013 (Figure 7d). However, the amplitude of events were small, so there were no increasing of RSAM observed in these periods. The observed decrease in the seismic velocities could reflect the dilatation of the propagation medium due to the tensile stresses induced by over-pressurization of the magma reservoir within the volcano plumbing system (Patanè et al., 2006;Brenguier et al., 2008). The decoherence is mainly due to physical changes of the volcanic edifice that are directly associated with the volcanic activity. The permanent decrease in coherence as a structural modification of the medium is irreversible, whereas the recoverable loss of coherence could be associated with reversible displacements of scatterers or opening and closing of preexisting cracks (Obermann et al., 2013).
The measured relative velocity variations and the coherence values depend on the position of the stations relative to the change in the medium and are generally used to retrieve information about the strain time evolution and its spatial distribution (Obermann et al., 2013;Rivet et al., 2014). Different pattern of relative velocity variation are observed for each station pair prior to and during Sinabung eruptions (Figure 8a). This suggests that the velocity variations are spatially localized. The Kriging method was applied to represent the location of the corresponding velocity changes (Figure 8b). Once δv/v is estimated for each pair of stations, the obtained values are the δv/v value at the centre of pair stations. These results should not be considered as a completely realistic image of the changes spatial distribution. They help to visualize which station pairs were affected by δv/v variations.
The relative velocity change maps obtained at different times between July 2013 and December 2013 show that the negative value of δv/v started from the southwest and northeast flanks of Sinabung, and then were going to the  et al., 2000). The fault plane strikes 030 0 where it crosses the summit of the volcano (Prambada et al., 2011;Kriswati et al., 2017). Kriswati et al. (2017) suggested that the Bireun tectonic earthquake, 252 km from Sinabung, on 2 July 2013 (6.1 Mw) was sufficient to initiate magma ascent and the ensuing volcanic seismicity, which led to the initial of 2013 eruption two months later on 15 September 2013. The value of relative velocity variation for KWR-MRD, KWR-SKM, and MRD-SKM had begun to drop since the end of July 2013 (Figure 8a). The observed decrease velocity changes did not recover until the end of the observation period, interpereted to be due to long-lasting and continues intrusions of magma or to stress buildup within the reservoir that dilated the edifice on the southwest flank of Sinabung. The velocity drop for other station pairs (KWR-SKN; MRD-SKN; SKM-SKN) were seen at about one month before the first eruption on 15 September 2013. It is suggested that the stresses then localizing in the northeast part of edifice induced a decrease of velocity and a soil compaction, and hence increased the velocity in other areas. The similar events also occurred in Piton de la Fournaise Volcano (Duputel et al., 2009;Obberman et al., 2013). Positive values of δv/v (> 0.1%) were located at Sinabung crater on December 16 and 27, 2013. CVGHM reported that a lava dome had been observed at Sinabung crater since 16 December 2013.

Conclusion
In this study, continuous ambient seismic noise records had been analyzed prior to and during September 2013 -February 2014 Sinabung eruptions. The value of relative velocity variation and coherence began to drop about one to two months before the first eruption on September 15, 2013, which started from the southwest to northeast flank of Sinabung edifice and then moved to the central part/crater of Sinabung. Long transient velocity changes were observed on the station pairs located in the southwest flank of Sinabung, interpreted to be due to long-lasting and continues intrusions of magma or to stress buildup within the reservoir that dilated the edifice. Decoherence and recoverable loss of coherence value were seen, that could be associated with the opening and closing of preexisting cracks. It is concluded that the relative velocity variation and waveform decoherence can be used as an indicator of increasing levels of volcanic activity.

I J O G
The Drop of Relative Velocity Variation and Coherence Values Prior to Sinabung 2013 Eruptions (Y. Suparman et al.)