Journal of Volcanology and Geothermal Research xxx (2016) xxx–xxx

VOLGEO-05937; No of Pages 14

Contents lists available at ScienceDirect

Journal of Volcanology and Geothermal Research

j ourna l homepage: www.e lsev ie r .com/ locate / jvo lgeores

Gravity monitoring of Tatun Volcanic Group activities and inference forunderground fluid circulations

Maxime Mouyen a,⁎,1, Benjamin Fong Chao a, Cheinway Hwang b, Wen-Chi Hsieh c

a Institute of Earth Sciences, Academia Sinica, Academia Road, 11529 Taipei, Taiwan, ROCb Department of Civil Engineering, National Chiao Tung University, Hsinchu 300, Taiwan, ROCc Industrial Technology Research Institute, Hsinchu 310, Taiwan, ROC

⁎ Corresponding author.E-mail address: maxime.mouyen@univ-rennes1.fr (M

1 Now at Géosciences Rennes Université de Rennes 1, CCS74205, 35042 Rennes Cedex, France.

http://dx.doi.org/10.1016/j.jvolgeores.2016.10.0010377-0273/© 2016 Elsevier B.V. All rights reserved.

Please cite this article as: Mouyen, M., etcirculations, J. Volcanol. Geotherm. Res. (201

a b s t r a c t

a r t i c l e i n f o

Article history:Received 17 February 2016Received in revised form 31 August 2016Accepted 4 October 2016Available online xxxx

The Tatun Volcano Group (TVG), located on the northern coast of Taiwan adjacent to the city of Taipei, experi-ences active hydrothermalism but has no historical record of volcanic eruption. Yet recent studies suggest thatTVG is dormant-active rather than extinct. To monitor mass transfers and to gain further understanding of thisvolcanic area, gravity variations have been recorded continuously since 2012 using a superconducting gravime-ter, and once every few months since 2005 using absolute gravimeters. We analyze the continuous gravity timeseries and propose a model that best explains the gravity variations due to local groundwater redistribution. Bycorrecting these variations, we identify gravity changes as large as 35 μGal that occurred concomitantly to fluidpressure-induced earthquakes and changes in the gas composition at Dayoukeng, one of TVG's fumaroles, over2005–2007.Weexamine severalfluidmovements that canmatch the gravity observations, yet too few additionalconstraints exist to favor any of them. In particular, no significant ground displacements are observedwhen thesegravity variations occurred. On the other hand, the model of gravity changes due to local groundwater redistri-bution can be routinely computed and removed from the ongoing time gravitymeasurements in order to quicklyidentify any unusual mass transfer occurring beneath TVG.

© 2016 Elsevier B.V. All rights reserved.

Keywords:GravimetryGroundwaterHydrothermalismVolcanismTatun

1. Introduction

The Tatun Volcano Group (TVG) covers an area of 400 km2 in thenorthern tip of the island of Taiwan adjacent to themetropolis of Taipei(Fig. 1). It consists of more than twenty Quaternary volcanoes and hasbeen experiencing a sustained seismic and hydrothermal activity (Linet al., 2005; Lee et al., 2008). TVG is generally associated with theRyukyu volcanic arc to the east, which results from the subduction ofthe Philippine Sea plate beneath the Eurasian plate (Teng, 1996; Kimet al., 2005). An alternative hypothesis is that TVG resulted from apost-collisional lithospheric extension associated with the collapse ofthe northern Taiwan mountain belt (Wang et al., 1999). As for thetiming of TVG eruptions, rock dating revealed two main periods ofvolcanic activity, at 2.8–2.5 Ma and 0.8–0.2 Ma (Wang and Chen,1990; Song et al., 2000). However, Belousov et al. (2010) suggestedthat magmatic eruptions occurred as recent as 13,000–23,000 yearsago and that a late phreatomagmatic eruption happened only6000 years ago. Despite the fact that there is no record of historicaleruptions, Belousov et al. (2010) maintained that the last eruptions

. Mouyen).NRS INSU; Campus de Beaulieu,

al., Gravity monitoring of Ta6), http://dx.doi.org/10.1016

are recent enough to consider TVG as active. A magma chamber maystill exist beneath TVG (Konstantinou et al., 2007) and volcanoseismicsignals, such as tornillos, and earthquakes induced by variations offluid pressure have been recorded by the TVG seismic network(Konstantinou et al., 2007; Lin et al., 2005; Rontogianni et al., 2012).Nevertheless, Wen et al. (2012) suggested that the magma chamber isa remnant of a past magma intrusion, now undergoing a cooling pro-cess. This agrees with the study of Lee et al. (2008), claiming that newmagma supply in the chamber is unlikely. Wen et al. (2012) concludedthat a volcanic eruption may preferentially occur as a consequence ofthe rupture of one of the two active faults that cross TVG rather thanthe activity of the magma chamber itself.

These recent findings prompted the realization of a research projectcalled “Understanding the life of TVG” aiming at a better knowledge ofTVG. Aside from the importance of this project in terms of basic science,a better assessment of the eruption hazard of TVG is also a societal needsince this volcanic edifice overhangs the densely populated basin ofTaipei as well as two nuclear power plants. Under this project, asuperconducting gravimeter (SG, serial number T049, manufacturedby GWR Instruments, Inc.; SG49 for short) has been installed on theYangmingshan (Fig. 1, site name: YMSG), one of TVG's mounts, inApril 2012. An SG is a relative gravimeter, which provides continuousgravity time series with a long-term stability (Goodkind, 1999). Theanalysis of gravity time series is a way to monitor and quantify mass

tun Volcanic Group activities and inference for underground fluid/j.jvolgeores.2016.10.001

Fig. 1. Tatun Volcano Group (TVG) area. YMSG is the superconducting gravimeter site. TB-MW-23 is the groundwater well. There are three rainfall gauges in the area: C0AC4, C0A86 andC0A87 located at 6.2, 5.1 and 4.3 km fromYMSG, respectively. The two nuclear power plants are at Jinshan and Kuosheng. The lower left panel locates TVGat the Taiwan island scale. YMSGis collocated with one continuous GPS station (YMSM, not written on the map), five other continuous GPS stations (YM01 to YM05) are located around YMSG.

2 M. Mouyen et al. / Journal of Volcanology and Geothermal Research xxx (2016) xxx–xxx

redistributions. It has proven useful in various studies, includingearthquake-induced deformation (Imanishi et al., 2004), global isostaticadjustment, tectonics (Francis et al., 2004), hydrology (Creutzfeldt et al.,2008), active fault assessment (Lien et al., 2014), and volcanoesmonitoring (Jousset et al., 2000; Carbone et al., 2003). YMSG was setto determine whether or not magma flows exist beneath TVG. In addi-tion to superconducting gravity measurements, time-lapsed gravitymeasurements by absolute gravimeters (AGs) have also been collectedat the same location since 2005, providing a calibration factor andgravity references for YMSG data.

Numerous factorsmay cause gravity changes at a gravity site (Torge,1989). One critical factor is the seasonal groundwater redistribution,which may be difficult to quantify precisely yet often responsible forthe second largest gravity variations next to the tides. It is thus impor-tant to properly model the off-target sources before analyzing gravitychanges due to magma movements (Kazama and Okubo, 2009). Tothis aim, we take advantage of rainfall data and water table data bothacquired in the TVG area. We first detail how we process thesuperconducting and absolute gravity data and then describe how hy-drological data are used to analyze gravity time series. After correctingthis time series from seasonal hydrological effects, we discuss thegravity residuals in terms of volcanic and hydrothermal processes atTVG.

2. Data

2.1. Gravity data

The YMSG gravity time series that we use spans 3 years, from April2012 to April 2015 (Fig. 2a). We use TSOFT package (Van Camp andVauterin, 2005) to remove spikes from the gravity time series. TSOFTalso computes and removes the pole tide using pole coordinates provid-ed by IERS (the International Earth Rotation and Reference SystemsService at ftp://hpiers.obspm.fr/iers/eop/eopc04/). Tidal effects (both

Please cite this article as: Mouyen, M., et al., Gravity monitoring of Tacirculations, J. Volcanol. Geotherm. Res. (2016), http://dx.doi.org/10.1016

solid-Earth tides and ocean tide loading) were removed using TSOFTand tidal parameters obtained with the ETERNA package (Wenzel,1996). The effect of air pressure changes on gravity was removedbased on Boy et al. (2002) model and collocated pressure data. The AGvalues were measured with FG5 gravimeters (Niebauer et al., 1995) atthe same location as YMSG and corrected for polar motion and atmo-spheric pressure similarly as for the SG data. Tidal effects are removedusing the tidal parameters obtained from the SG tidal analysis, throughthe g software (Microg-LaCoste, 2008). Two different FG5 (serial num-bers #224 and #331) were used: #224 from 2005 to 2010 (except on7May2009) and#331 since 2011 and on 7May2009. One AGmeasure-ment session lasted 1 to 2 days, consisting of at least 30 sets. One setcontains 100 drops in 30 min (one drop every 5 s starting at half of anintegral hour).

The corrected YMSG and AG data are shown in Fig. 2b and c, respec-tively. The YMSG time series contains several gaps and steps, whichmaycome from electronical or power disruptions (Hinderer et al., 2007; VanCampand Francis, 2007). Each continuousYMSGdata segment betweentwo gaps can be offset independently along the gravity y-axis. (In Fig. 2bwe have applied nominal step corrections only to make the displaymore convenient. See later.)

2.2. Hydro-meteorological data

Taiwan's climate is influenced by the East Asianmonsoons and theirheavy rains (Chen and Chen, 2003). Therefore, local groundwater redis-tribution is likely to account for a large part of the residual gravity signal,as it has been shown for several others locations (e.g. Jacob et al., 2008;Longuevergne et al., 2009). We evaluate this contribution using twohydro-meteorological datasets recorded in the vicinity of YMSG (Fig. 1):

1. Hourly rain gaugemeasurements at station C0A86 (Fig. 2d) retrievedfrom theData Bank for Atmospheric Research at the Taiwan Typhoonand Floods Research Institute (https://dbar.ttfri.narl.org.tw/).

tun Volcanic Group activities and inference for underground fluid/j.jvolgeores.2016.10.001

Fig. 2. (a) Raw YMSG time series. Tidal effects account formost of the signal. Several steps exist in early 2012 andMarch 2015. (b) YMSG time serieswithout tidal and atmospheric effects.Gaps in the data are shownwith a grey band. The steps thatmay exist between each gapwill be corrected based on the hydro-gravitymodelling. (c) Collocated absolute gravity time-seriesfrom2005 to 2015. Datameasuredwith FG5#224and#331 are in black and red, respectively. (d)Hourly rainfall gaugemeasurements at C0A86. (e)Water table variations atwell TB-MW-23 (mean removed). The sampling interval is 15 min. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

3M. Mouyen et al. / Journal of Volcanology and Geothermal Research xxx (2016) xxx–xxx

2. Water table measurements at station TB-MW-23, at a 15-minutesampling interval (since April 2012, Fig. 2e). The water table variesin response to both thewater input (rainfall) and output (subsurfacerun-off, evapotranspiration) in the watershed.

These data are used below to compute a “hydro-gravitymodel”, thatis, a model of gravity variations due to local water flow only.

2.3. Ground displacements data

Ground displacements are another valuable information whenstudying volcanic areas; previous studies have shown that ground dis-placements can be generated either by magma (Rymer et al., 1995;Beauducel and Cornet, 1999; Jousset et al., 2003) or by hydrothermalwater movements (Battaglia et al., 2006; Fournier and Chardot, 2012).Such displacements can be measured by precise levelling surveys (e.g.Amaro and Chiarabba, 1995), interferometric synthetic aperture radar(InSAR, e.g. Solaro et al. (2010)), or GPS surveys (Puglisi et al., 2001).Six continuous GPS stations were operational at TVG in the entire timeperiod of this study (2005–2015); displacement data, computed in theIGS08 frame (Rebischung et al., 2012) are available through the GPSLABwebsite (gps.earth.sinica.edu.tw/). Ground displacements records forthe three components north, east and up are in Fig. 3.

Ground displacements at TVG are controlled to first order bytectonics (Yu et al., 1997; Shyu et al., 2005) which is responsible forthe long-term linear trends on the north and east components. Inaddition, seasonal displacements are attributed to atmospheric and hy-drologic loading (Dong et al., 2002). The long-term trend of the groundvertical displacements recorded by the YMSM permanent GPS station

Please cite this article as: Mouyen, M., et al., Gravity monitoring of Tacirculations, J. Volcanol. Geotherm. Res. (2016), http://dx.doi.org/10.1016

(Fig. 3, lower right panel), collocated with YMSG is 0.5 ± 0.1 mm yr−1.This is equivalent to −0.1 μGal yr−1 using the theoretical Bouguercorrected gradient of −0.2 μGal mm−1(De Linage et al., 2007; VanCamp et al., 2011). Since the YMSG time series is rigorously continuousfor at most 8 months (from October 2012 to July 2013), this long-termvertical displacement would make at most a deviation of 0.07 μGal. Wecan reasonably neglect this trend when analyzing the YMSG time series.

3. Hydro-gravity modelling

3.1. Porosity of the ground

Since aquifers are primarily recharged by rainfalls, we first investi-gate the correlation between rainfall measurements at C0A86 andwater table variations at TB-MW-23. We use Eq. (1) to estimate thewater table variations hest at time i (Crossley et al., 1998):

hest i ¼ r j 1−e− i− jð Þ=τ1� �

e− i− jð Þ=τ2 ð1Þ

That is, the thickness hest of water at a given time i depends on theamount of rain rj accumulated over past times j (given by COA86data), considering that this rainwater infiltrates the ground with acharge time constant τ1 and is released with a discharge time constantτ2, regardless of the processes responsible for this release. We firstcompute the correlation coefficient between this model of water layerthickness (hest) and the measured changes of water table (hmes), testingτ1 values ranging from 60 to 500 h and τ2 values ranging from 150 to2100 h. A maximum correlation coefficient of 0.92 between hest and

tun Volcanic Group activities and inference for underground fluid/j.jvolgeores.2016.10.001

Fig. 3. Ground displacements measured at six continuous GPS stations at TVG (see Fig. 1 for their locations, YMSM is the GPS station colocated with YMSG). Each column is for onecomponent: north, east and up from left to right. Each line is for one station.

4 M. Mouyen et al. / Journal of Volcanology and Geothermal Research xxx (2016) xxx–xxx

hmes is computed for τ1 = 190 h (8 days) and τ2 = 800 h (33 days). Thecorrelation coefficient is very close to 1 and shows that the rainfall dataat C0A86 and Eq. (1) are adequate to explain water table variations.However, one must multiply the results from Eq. (1), by a factor of 10in order to match the amplitude of the water table variations (Fig. 4).We conclude that the porosity of the aquifer is 10%; that is, the netmass variation of the actual water is 10% of the apparent well levelvariation. This porosity value is consistent with the dominantly andesiticconstitution of TVG (Belousov et al., 2010); fresh andesite having aporosity of approximately 8%, which can be increased by weatheringand fracturation (Jamtveit et al., 2011).

Note that two nearby rainfall gauges exist on TVG, C0A87 andC0AC4, that are located on different hill slopes (Fig. 1). We testedthem in the same way as for C0A86 but obtained lower correlationsbetween the estimated water table variation and the observed. Thuswe conclude that the rainfall at C0A86 better represents the localrainfall regime at the location of the well TB-MW-23.

Fig. 4.Water table level variation at TB-MW-23 (blue line) can bemodelled using the rain gaug(For interpretation of the references to colour in this figure legend, the reader is referred to th

Please cite this article as: Mouyen, M., et al., Gravity monitoring of Tacirculations, J. Volcanol. Geotherm. Res. (2016), http://dx.doi.org/10.1016

3.2. Hydro-gravity model constrained using YMSG data

The YMSG gravity time series (Fig. 2b) shows a signal of dailyvariability (high frequency) superimposed by a signal whichchanges over a period of several months (low frequency), al-though not strictly periodic. In this part we build a hydro-gravitymodel by summing the gravity effect of the water table variationsand the gravity effect of the rain soon after it reaches the ground.The rain is considered to be responsible for the high frequency sig-nal and the water table variation is responsible for the low fre-quency signal.

3.2.1. Gravity effect of the water table variations: low frequency modelWe use the water table variations measured at the well TB-MW-23.

The well being located about 1300 m from YMSG (Fig. 1), we assumethat its data can be used as a proxy of water table variations underneathYMSG. Since there is no constrain on the geometry of the aquifer, we

e data at C0A86, using Eq. (1) with τ1= 190 h, τ2= 800 h and a porosity of 10% (red line).e web version of this article.)

tun Volcanic Group activities and inference for underground fluid/j.jvolgeores.2016.10.001

5M. Mouyen et al. / Journal of Volcanology and Geothermal Research xxx (2016) xxx–xxx

start by using the simple infinite Bouguer plate approximation to con-vert the water table variation into a gravity variation gBouguer:

gBouguer ¼ 2π G ρ Δhwt ð2Þ

where G = 6.67 × 10−11 m3 kg−1 s−2 is the universal constant ofgravitation, ρ = 1000 kg·m−3 the volumetric mass of water, and Δhwt

is the actual water table variation measured at TB-MW-23 divided by10 to account for the rock porosity (estimated from water tablevariations, rainfall data and Eq. (1), see Section 3.1). However, thismodel significantly overestimates the gravity changes observed atYMSG (Fig. 5a, red and green lines), with a root mean square (rms)error of 6.8 μGal. To avoid this overestimation, we refine the aquifermodel by making its lateral extension finite instead of infinite.Therefore, we model the water table variations as a cylinder of radiusa, thickness Δhwt and depth c (c is the distance from YMSG elevationand the bottom of the cylinder). The gravity effect gcyl is computed fora-values ranging from 10 to 200 m and c-values ranging from 10 to500 m, using (Hofmann-Wellenhof and Moritz, 2006):

gcyl ¼ 2πGρ hwt þffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffia2 þ c−hwtð Þ2

q−

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffia2 þ c2

p� �ð3Þ

The bestfit is obtainedwith a=120mand c=130m. The rms erroris reduced to 2.5 μGal with no presence of the overestimation of gravityvariation (Fig. 5a, red and purple curves). But this model does notcapture the high frequency component of the gravity time series,which will be computed using the rainfall data.

Fig. 5. (a) Comparison of theYMSGgravity data (red)with the Bouguer platemodel (green) andat TB-MW-23. (b) Comparison of the YMSGdata (red)with the hydro-gravitymodel (blue), conFig. 2b) were removed by adjusting the continuous YMSG segments to the hydro-gravity mod(green dots). (c) YMSG residuals after removing the hydro-gravity model, both in time and hgreen lines. (For interpretation of the references to colour in this figure legend, the reader is re

Please cite this article as: Mouyen, M., et al., Gravity monitoring of Tacirculations, J. Volcanol. Geotherm. Res. (2016), http://dx.doi.org/10.1016

3.2.2. Gravity effect of the rainwater at the surface: high frequency modelHerewe use the rainfall data and Eq. (1) to compute the thickness of

an equivalent layer of water huz, corresponding to the water located inthe unsaturated zone close to the surface (in Eq. (1), hest is replaced byhuz). It is then converted into gravity effect gi, using the infinite Bouguerplate approximation. Keeping the time index i alreadymentioned in Eq.(1) we have:

gi ¼ 2πGρhuzi ð4Þ

The values of the time constant τ1 and τ2 are unknown, thuswe haveto test them, with τ1 ranging from 0.1 to 5 h and τ2 ranging from 1 to220 h, until the model best fit the data. This method can create a highfrequency signal compatible with that observed in the YMSG data. Thelast step is to combine this high-frequencymodel with the low frequen-cy model (Section 3.2.1).

3.2.3. Combination of the low and high frequency models into one singlehydro-gravity model

This combination consists in estimating all the parameters of thetwo previous models, namely a and c for the low frequency model(Eq. (3)), and τ1 and τ2 for the high frequency model (Eqs. (1) and(4)). In practice, the entire ranges of τ1 and τ2 values are tested foreach set of a and c. Then we sum the high and low frequency modelsand compare this sum to the YMSG time series. The rms is computedfor each test of {a, c, τ1, τ2}. Eventually, we find the lowest rms(1.4 μGal), upon the selected parameters τ1 = 0.2 h (12 min), τ2 =50 h, a = 70 m and c = 90 m (Fig. 5b). The residuals have a Gaussiandistribution between −4.4 and 5.7 μGal and centered on 0, suggestingthat this hydro-gravity correction is adequate (Fig. 5c). To summarize,at any given time, the majority of the residual gravity signal (without

the cylindermodel (purple), both constrained only by thewater table variationsmeasuredstrained by thewater table variations and the rainfall data. The steps in the YMSGdata (seeel. The absolute gravity (AG) data, considered as reference values, are also superimposedistogram views. Times when typhoons made landfall on Taiwan are shown with verticalferred to the web version of this article.)

tun Volcanic Group activities and inference for underground fluid/j.jvolgeores.2016.10.001

6 M. Mouyen et al. / Journal of Volcanology and Geothermal Research xxx (2016) xxx–xxx

tidal, atmospheric and pole tide effects) at YMSG comes from watermass changes in two different parts of the ground: the unsaturatedzone located near the ground surface and the deeper aquifer.

As seen in Fig. 2b, due to several gaps and steps, the original YMSGtime series is not rigorously continuous and segments were kept sepa-rated from each other. In Fig. 5a we shift them along the gravity axisuntil they overlap the hydro-gravity model. This method allows us tocorrect steps in a non-arbitrary way that is reasonable since we assumethat gravity changes are primarily due to local water redistribution. Onthe other hand, it may also remove from the YMSG data any long-termgravity trend that would not be due to the computed groundwaterredistribution. The first long-term gravity trend to consider is theinstrumental drift of YMSG. It has no geophysical meaning and can beestimated by comparing the YMSG time series with AG data measuredat the same location, as a FG5 gravimeter does not suffer from instru-mental drift (Hinderer et al., 2007). The drift of a SG can be exponentialor linearwith time, although Van Camp and Francis (2007) showed thatan exponential model should be preferred for records longer than10 years. In our case, the only option is to compare YMSG and AG dataon a time period where the YMSG measurements are continuous,without gaps or steps. This corresponds to a 6-months period, fromOctober 2012 to April 2013, during which three AG measuringcampaigns were carried out (Fig. 5b). This small amount of AG data,combined with the AG's standard deviation, leads to an instrumentallinear drift of 9.2 ± 14.8 μGal·yr−1 that we decided not to remove

Fig. 6. (a) Original hydro-gravitymodel (that is, the same as Fig. 5b, computed using the entire Ycomputed using only YMSG data during summer 2012. We find τ2 = 130 h (instead of 50 h). (model computedwith summerdata only (red). (d, e, f) The same but for summer 2013.Wefindreferred to the web version of this article.)

Please cite this article as: Mouyen, M., et al., Gravity monitoring of Tacirculations, J. Volcanol. Geotherm. Res. (2016), http://dx.doi.org/10.1016

from the YMSG data because of its very large uncertainty. The conse-quence is that we cannot rely on the YMSG time series for identifyinglong-term gravity linear trends that may yet exist at TVG, as a result oftectonic or slow volcanic processes. This could be fixed by increasingthe frequency of AG measurements.

In Fig. 5b and c, the largest residuals are mostly found during thesummer season, which experiences extraordinarily heavy rains dueoften to typhoons. Time-independent parameters of the hydro-gravitymodel for the three years of YMSG data may not be proper to predictthe groundwater redistribution during the short periods of heavyrains. We test this hypothesis by assessing the hydro-gravity modelparameters in the summers (June to September) of 2012 and 2013,not in summer 2014 since YMSG data are missing in August 2014.There is no reason to change the location of the aquifer, hence wekeep a = 70 m and c = 90 m. Rather, we re-estimate the effect of thewater in the unsaturated zone by varying τ1 and τ2. We find thatsummer's gravity variations are better modelled (lower rms) withτ2 = 130 h in summer 2012 and τ2 = 150 h in summer 2013, whileτ1 stay unchanged (Fig. 6 and Table 1).

This implies that the processes responsible for the release of waterfrom the unsaturated zone take about 3 times longer in summer thanduring the rest of the year, likely as a consequence of the heavierinput of water. Thus the large gravity residuals during the 2012 and2013 summer seasons seem to be due to an imperfection of ourhydro-gravity model rather than due to a volcanic event, such as a

MSG time series), focused on the summer 2012 time period. (b) Newhydro-gravitymodelc) Comparison of the histogram of the residuals for the original model (blue) and the newτ2=150h. (For interpretation of the references to colour in this figure legend, the reader is

tun Volcanic Group activities and inference for underground fluid/j.jvolgeores.2016.10.001

Table 1Synthesis of the values of τ1 and τ2 for the original hydro-gravitymodel and for the hydro-gravitymodel computed only using the summer period YMSG data, in 2012 and 2013, with rmsin each case.

Original model Modified model (focused on Summer)

τ1 (hours) τ2 (hours) rms (μGal) τ1 (hours) τ2 (hours) rms (μGal)

Summer 2012 0.2 50 1.9 0.2 130 1.6Summer 2013 0.2 50 1.7 0.2 150 1.3

7M. Mouyen et al. / Journal of Volcanology and Geothermal Research xxx (2016) xxx–xxx

magma transfer at depth. Nevertheless, the overall agreement betweenthe YMSG data and our hydro-gravity model suggests that this model isadequate. That is, the gravity variations at YMSG are due to local season-al hydrology and have at least two components: the immediate effect ofthe rainwater flowing into the unsaturated zone, which triggers gravitychange as soon as the rain reaches the ground surface, and the effect dueto the water table variations, which is delayed by the time needed forthe rainwater to accumulate in an aquifer. The unsaturated zone actsas a low-pass filter for rain-induced gravity signal.

3.3. Hydro-gravity model before the installation of YMSG

To further test the hydro-gravity model, we compare it to the 23 AGmeasurements collected at the location of YMSG over 2005–2015. Nowell log data exist before 2012, but we have shown in Section 3.1 thatwater table variations can be predicted using rainfall data measured atC0A86 and Eq. (1), using τ1 = 190 h (8 days) and τ2 = 800 h(33 days). This can be used to estimate the missing water table mea-surements during 2005–2012, as C0A86 measurements started in1995. We thus have an estimation of the water table variations andthe rainfall data. Eventually we combine these two datasets using themethod and parameters {a, c, τ1, τ2} given in Section 3.2 and get thehydro-gravity model since 2005 (Fig. 7).

The hydro-gravity model is in agreement with 74% (17 out of 23) ofthe AG measurements. Outside the YMSG measurement time period(during which the YMSG data were used to constrain the hydro-gravitymodel), this model still matches 60% (9 out of 15) of the AG data. Themisfit in early 2008 seems due to the rain gauge failure that caused agap over 15 consecutive days marked with a thick grey vertical line inFig. 7. If this gap is artificially interpolated, for instance with the samerainfall as those following the gap, the hydro-gravity model matchesthe AG measurement within its standard deviation. The large gravityvariations measured in 2011 are well predicted by the hydro-gravitymodel and confirms the significant influence of water flows in theunsaturated zone and in the aquifer on the temporal gravity changesmeasured at TVG. But the hydro-gravitymodel cannot explain four con-secutive measurements between November 2005 and August 2007 and

Fig. 7. Hydro-gravity model computed since 2005 (blue line) using rain gauge C0A86 data anddots). The vertical grey lines correspond to 399 hourly rain gauge measurements failures betw15 days of January 2008. Small left-arrows point AG measurements that deviate the most frolegend, the reader is referred to the web version of this article.)

Please cite this article as: Mouyen, M., et al., Gravity monitoring of Tacirculations, J. Volcanol. Geotherm. Res. (2016), http://dx.doi.org/10.1016

one measurement in July 2010, which deviate from the model by up to21 ± 2 μGal. In particular, the 35 ± 3 μGal gravity change measuredfrom May 2006 to January 2007 is at least twice larger than any othergravity change over the entire period. This suggests that some largemass redistribution due to processes other than seasonal hydrologymay have occurred in this time period.

4. Large mass transfers between 2005 and 2007

4.1. Fluid circulation supported by seismological and geochemical studies

The 2005–2007 large gravity changes (Fig. 7) are concomitant toseismological and geochemical noticeable events (Fig. 8a and b). Leeet al. (2008, 2016) observed a significant increase of the SO2/H2S ratiofrom July 2005 to July 2007, only at the Dayoukeng fumarole. To explainthis increase, Lee et al. (2008) suggested that more magmatic gasescontributed to the fumaroles as water from the primary hydrothermalsystem, depicted as a deep water reservoir located close to the magmachamber, was supplied to the Dayoukeng fumaroles through fractureopenings. In the seismological records, Rontogianni et al. (2012)identified several spasmodic bursts, defined as “rapid-fire sequencesof brittle-failure earthquakes with overlapping coda” (Hill and Prejean,2005). They are induced by an increase of pore pressure due to fluidcirculation through cracks and most of them were recorded between2006 and 2007. It is interesting to note that these spasmodicbursts occurred in the region of a highly fractured former magmapassage (Fig. 8b) parallel to the Chinshan fault as evidenced by seismictomography (Wen et al., 2012). This passage lies beneath Dayoukengfumarole and the YMSG site, and extends between 0 and 4 km depthin this area.

4.2. No significant ground displacements

We mentioned in Section 2.3 that ground displacements at TVG arecontrolled to first order by tectonics and seasonal atmospheric andhydrologic loading. These effects are modelled using a combination oflinear and sinusoidal functions (Fig. 9, left panel), which are computed

compared to gravity variations measured using an absolute gravimeter (AG data, greeneen 2006 and 2008 (~2% of missing data). In particular, there is no data during the firstm the hydro-gravity model. (For interpretation of the references to colour in this figure

tun Volcanic Group activities and inference for underground fluid/j.jvolgeores.2016.10.001

Fig. 8. (a) Gravity residuals after removing the hydro-gravity model (Fig. 7) along with other evidences of volcanic or hydrothermal activity reported in geochemical and seismologicalstudies (Lee et al., 2008; Rontogianni et al., 2012; Pu et al., 2014). (b) Localization of the 2006–2007 spasmodic bursts (red dots). The magma passage runs all along the south east partof Chinshan fault (Wen et al., 2012), beneath the cluster of spasmodic bursts, YMSG and the fumaroles (blue triangles). Inverted black triangles are permanent GPS stations. One iscollocated with YMSG (name YMSM). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

8 M. Mouyen et al. / Journal of Volcanology and Geothermal Research xxx (2016) xxx–xxx

with the CATS program (Williams, 2008). Once these effects areremoved from the GPS solutions, the residual displacements observedin the 2005–2007 timeperiod, for the vertical or horizontal componentsare at the mm level (Fig. 9, right panel).

Please cite this article as: Mouyen, M., et al., Gravity monitoring of Tacirculations, J. Volcanol. Geotherm. Res. (2016), http://dx.doi.org/10.1016

Assuming an elastic rheology, we assess the magnitude of thedisplacements that might be associated with the gravity variationsobserved between 2005 and 2007. The average instrumental uncertain-ty δAG is 3.6 μGal from the standard deviation of the AG measurements.

tun Volcanic Group activities and inference for underground fluid/j.jvolgeores.2016.10.001

9M. Mouyen et al. / Journal of Volcanology and Geothermal Research xxx (2016) xxx–xxx

We add two uncertainty sources for the gravity changes in the 2005–2007 period, both related to the hydrological corrections describedin Section 3: The first one is the rms misfit δhyd = 1.4 μGal betweenthe hydro-gravity model and the YMSG data (see above). The secondis due to the misfit between the water table change estimated fromthe rainfall data (Fig. 4) and the actual one, or δwat = 3.1 μGal. Alto-

gether, they lead to a total uncertainty δTOT ¼ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiδ2AG þ δ2hyd þ δ2wat

q¼ 5

μGal , to be associated with the residual gravity changes during2005–2007.

This gravity change Δg is then decomposed into three time periodsegments:

• Period 1, from t1 (09-11-2005) to t2 (27-05-2006): Δg1 = −22 μGal• Period 2, from t2 to t3 (17-01-2007): Δg2 = 35 μGal• Period 3, from t3 to t4 (01-08-2007): Δg3 = −16 μGal

TheMogimodel gives relation between the vertical grounddisplace-ments (Δh) and the gravity changes as (Hagiwara, 1977; Jousset andOkada, 1999):

ΔgΔh

¼ β þ 43πGρ0 ð5Þ

where ρ0 is the mean density of the mass responsible for the gravitychange and β is the free-air effect due to the gravity measurement sitevertical displacements (β = 308.6 μGal·m−1). We compute Δh for theabove-listed three values of Δg and for a range of ρ0 from 1000 (forwater) to 2500 km·m−3 (for nominal magma).

Fig. 9. Vertical, East and North components at the six continuous GPS stations on TVG (Fig. 8b)seasonal (sinusoidal) trend (blue line). The de-trended data are on the right panel, with a 90-dalarge gravity changes are observed at YMSG and the vertical dotted black lines show the exact timreferences to colour in this figure legend, the reader is referred to the web version of this artic

Please cite this article as: Mouyen, M., et al., Gravity monitoring of Tacirculations, J. Volcanol. Geotherm. Res. (2016), http://dx.doi.org/10.1016

Fig. 10 shows that several centimeters of vertical displacements areexpected from this model. In particular, the maximum gravity change,between May 2006 and January 2007, should result in as much as9 cm of uplift at YMSM. Such a vertical displacement is not observed;the ground displacements at YMSM fluctuated within less than±1 cm (Fig. 9, bottom right panel). The model for Eq. (5) considers aspherical shape for the material moving beneath the surface. SinceTVG is located in a tectonically active and faulted area, alternativelythematerial responsible for the gravity changes at YMSG can be consid-ered to move through fractures as theorized by Sasai (1986, 1988)where specific fault configurations can return lower ground displace-ments for a given gravity change than the Mogi model. Jousset et al.(2003) demonstrated that the fracture filling model was more efficientthan the Mogi model in reconciling Δg and Δh data acquired at Usuvolcano (Japan) before and after its 2000′s eruption. Still, uplift of theorder of several cm and horizontal displacements up to 10 cm shouldbe present for the model due to the elastic deformation of the rock inwhich some material is forcing its way. Thus, to explain the absence ofsurface displacement at TVG, we hypothesize that the material canmove in an existing network of open fractures. Evidences of such amechanism were found at Etna volcano between June 1990 and June1991, when magma entered a pre-existing fracture system that can re-main open between eruptions (Carbone et al., 2003; Rymer et al., 1993).At TVG, this fracture system can be identified with the above-men-tioned highly fractured magma passage (Wen et al., 2012). It is parallelto the Chinshan fault, an active normal fault materializing the exten-sional tectonics of the north-eastern Taiwan (Chang et al., 2003b;Huang et al., 2012; Shyu et al., 2005). Although no fracture orientationmeasurements are available in this area, one can reasonably suppose

. The original data are on the left panel, along with a fit for both a long-term (linear) and ays smoothing (red line). The vertical black lines bound the 2005–2007 time period wheree of the four AG gravitymeasurements done in this time period. (For interpretation of the

le.)

tun Volcanic Group activities and inference for underground fluid/j.jvolgeores.2016.10.001

Fig. 9 (continued)

10 M. Mouyen et al. / Journal of Volcanology and Geothermal Research xxx (2016) xxx–xxx

Please cite this article as: Mouyen, M., et al., Gravity monitoring of Tatun Volcanic Group activities and inference for underground fluidcirculations, J. Volcanol. Geotherm. Res. (2016), http://dx.doi.org/10.1016/j.jvolgeores.2016.10.001

Fig. 10.Grounddisplacements computed using Eq. (4) for the three largest gravity changesmeasured at YMSG(blue lines), assuming aMogi sourcemodel. a, b and c refer to the gravity changeslisted as steps 1, 2 and 3, respectively. Black lines correspond the 5 μGal uncertainty on the gravitymeasurements, converted into displacements. (For interpretation of the references to colour inthis figure legend, the reader is referred to the web version of this article.)

11M. Mouyen et al. / Journal of Volcanology and Geothermal Research xxx (2016) xxx–xxx

that this extensional tectonics builds the stress needed for maintainingthe fractures open. Barton et al. (1995) showed that the permeability ofcritically stressed faults is much higher than that of faults that are notoptimally oriented for failure in the current stress field. As a result,such a fractured medium can be considered as a network of efficientfluids conduits, allowing significant material transfers in the ground. Itis not possible with our data to rigorously demonstrate that this fluidis magma or hydrothermal water. Otherwise, the coupling grounddisplacements and gravity changes would have been an efficient wayto infer the density contrast of the fluid with the host rocks (Jousset etal., 2003; Sasai, 1986).

In the rest of this study we will thus use the generic term “fluid”, in-stead of water or magma. Given that, we note that previous TVG studiesconducted in 2005–2007 have in most cases suggested hydrothermalcirculations rather than magma movements (Konstantinou et al.,2007; Lee et al., 2008; Rontogianni et al., 2012).

4.3. Fluid circulation model

Fluid ascent beneath TVGwas believed to be responsible for both thegas composition changes in Dayoukeng fumarole and the occurrence ofspasmodic bursts recorded during 2005–2007 (Lee et al., 2008;Rontogianni et al., 2012). We adopt this hypothesis as model A (fluidascension) to explain the residual gravity changes, according to whichmost of the spasmodic bursts had occurred in a radial distance (x)between 1000 and 2000 m from YMSG and a depth (z) between 0 and2 km. These geometric parameters are used to compute themass trans-port that is responsible for the observed gravity change Δg under the

Table 2Horizontal (x) and vertical (z) location of the fluid that circulates during the 2005–2007 largegravity changes, relative to the gravity site (where AG and SG data are acquired) and at four dif-ferent times (t1 to t4) in the gravity time series. If several locations canmatch a gravity change,we only give the average value of these locations,with its standard deviation. They are comput-ed(forwardcomputation)by testingwhat combinationmass/x/z canfit thegravity changesΔg1,Δg2 and Δg3 measured at YMSG. In the ascending model, we constrain the fluid to ascend to-ward the surface. In the cycling model, the fluid does a cycle, that is, its initial (t1) and final(t4) locations are the same. The values in the table are only those found for theminimummassof fluid able to explain the gravity changes, that is, no solution exists for lower masses.

Model A: Ascending fluid Model B: Cycling fluid

x (m) z (m) x (m) z (m)

Time t1 1100 ± 100 1200 ± 200 1100 ± 100 1200 ± 200Time t2 2000 750 ± 50 2000 750 ± 50Time t3 1000 650 ± 50 1000 650 ± 50Time t4 1100 ± 100 400 ± 100 Same as at t1Minimum massof fluid (Mt)

15 15

Please cite this article as: Mouyen, M., et al., Gravity monitoring of Tacirculations, J. Volcanol. Geotherm. Res. (2016), http://dx.doi.org/10.1016

point-source model that gives the amplitude of the vertical componentof the gravity as measured by gravimeters:

Δg ¼ GΔmz

x2 þ z2ð Þ3=2ð6Þ

Our aim is to find a mass change (Δm) and its radial x and vertical zdistances fromYMSG thatfit the gravity changesΔg1,Δg2 andΔg3, withintheir uncertainties. While setting these gravity constraints in Eq. (6), wetest a range of possible x and z values that are compatible with the loca-tion of the cluster seismic burst, that is x ∈ [1000; 2000 m] and z ∈ [0–2000m]. Given those conditions and the fact thatwe force the fluid to as-cend from t1 to t4,we obtain aminimummass of 15megatons (Mt) need-ed to match the gravity changes (Table 2, “Ascending fluid” column).Increasing themass of moving fluidmostly enlarges the vertical and hor-izontal domains where the fluid could be, yet the pattern of the transportis identical. We also test a model B, whose x domain is the same as formodel A, but the fluid eventually goes back to its initial location at timet1 (Table 2, “Cycling fluid” column). This cycling model is based on thefact that the gravity value, after experiencing several large changes, goesback to the 2005 value as if the fluid returned down to its initial location,under gravitational forces, after several displacements in the ground.Models A and B are summarized in Fig. 11. Note that since x is the radialdistance fromYMSG, the fluid can be anywhere on a circle of radius x andcentered on YMSG. Thus at t1 and t2 we represent the fluid in the area ofthe spasmodic bursts, west of YMSG and, at t3 (and t4 in case b) beneaththe Dayoukeng fumarole, located north-east of YMSG and at a distancethat match the computations. This is only a qualitative interpretation of

Fig. 11. Scheme of the fluid circulation models that fit the gravity changes observed atYMSG. Values x (radial distance from the gravimeter) and z (depth) are given in Table 2.Numbers 1 to 4 refer to the time of the absolute gravity measurements and the values inμGal are the gravity changes measured between each time. Areas experiencingspasmodic bursts are in red (see Fig. 8b for a map view). Here we suggest that the fluidascended toward Dayoukeng fumarole through a dense network of existing fractureslocated in a former magma passage. At time 3, the fluid either continues upward eithergoes downward, back to its initial location at time 1. (For interpretation of thereferences to colour in this figure legend, the reader is referred to the web version ofthis article.)

tun Volcanic Group activities and inference for underground fluid/j.jvolgeores.2016.10.001

12 M. Mouyen et al. / Journal of Volcanology and Geothermal Research xxx (2016) xxx–xxx

the geochemical changes observed in this fumarole, suggesting that the2005–2007 fluid circulation affected it.

5. Discussion

5.1. Limitation of the gravity modelling

The gravity modelling in Section 4.2 was done using purposely apoint source gravity effect in order tominimize the amount of unknownparameters to be sought: mass, radial distance and depth of this massrelative to YMSG. Since there is only one single site of gravity measure-ments and no ground displacements, multiple solutions are possiblewhile none can be favored. Moreover, the geometry of the undergroundfluid reservoir is probably more complex and can result in considerabledifferences between actual and estimated volumes (Fialko et al., 2001).As a test, we build penny-shaped reservoirs, whose radius is arbitrarilyset to be 10 times larger than their thickness while keeping the centersand volumes the same as for the point-source models given in Table 2.Their gravity effect is computed using analytic formulas from Singh(1977).We find that the computed gravity valueΔg1,Δg2 andΔg3 devi-ates by 9%, −12% and 58%, respectively, from Model A, and by −4%,−6% and 6% from Model B. This deviation can result in either changesof the location or the mass of the fluid flowing in the ground, especiallyfor Model A, where the fluid keeps ascending during the entire 2005–2007. Ground displacements, if exist, together with the gravity changes,provide useful constraints on the path ofmaterialmovement at depth interms of its volume (e.g. Bartel et al., 2003; Beauducel and Cornet, 1999)and density (Battaglia et al., 2006, 1999; Jousset et al., 2003, 2000). Theirvery low amplitudes here prevent us from considering the elastic rheol-ogy and hence an unequivocally revelation of the scenario of the fluidmass transport. One way to overcome these uncertainties is to measuregravity time series at several other sites around YMSG, a possibility assoon as YMSG gravity time series shows significant deviations fromthe hydro-gravity model proposed in Section 3. At present, while thelarge gravity changes show little doubt that mass transports did occurin TVG, the suggested models in Section 4.2 must be regarded astentative.

5.2. Timing of the absolute gravity measurements

Before 2012, the gravity time series is sporadic, for which we in-terpret the large gravity changes between 2005 and 2007 using thefirst AG measurement done in November 2005 as a reference. How-ever, Lee et al. (2008) have observed changes of the Dayoukeng fu-marole composition and temperature as early as August 2004 priorto any gravity measurement. One could thus question the validityof using the first AG measurement as a reference, even though ithas the same value (within the uncertainty) as the AG values mea-sured after the 2005–2007 event. It is unknown whether mass trans-port was already ongoing during November 2005 but the gravityvalue was the same as during the rest of the time period after August2007. On the other hand, we note that, because of the discontinuityof the gravity measurements, the actual extreme values of the2005–2007 gravity changes could be higher than the apparent mea-sured values.

Pu et al. (2014) identified a swarmof earthquakes related to volcanicor hydrothermal activity nine months before, in October 2009, and theSO2/H2S ratio increased again between February 2010 and February2011 (Lee et al., 2016). The single gravity measurement of July 2010which showed an apparent gravity drop is the only one over a periodof nearly 17months (from 9 February 2010 to 20 June 2011). Assessingamass transport fromone single gravitymeasurement is not reasonableand the question whether another fluid movement occurred within theTVG hydrothermal system in 2010 remains open. YMSG provides nearlycontinuous gravity measurements that will enable us to observe a com-plete cycle of gravity changes if a new hydrothermal cycle occurs in the

Please cite this article as: Mouyen, M., et al., Gravity monitoring of Tacirculations, J. Volcanol. Geotherm. Res. (2016), http://dx.doi.org/10.1016

coming years. To finish, the process that triggered this fluidmigration isnot known but it might be related to Taiwan's high seismicity under itsactive tectonics. Indeed, Pitt and Hutchinson (1982) proposed that anincrease of the seismic activity beneath the hydrothermal system ofthe Mud Volcano at Yellowstone National Park could have expandedthe pre-existing fracture system, thus enhancing fluid flows. Similarly,at the Juan de Fuca mid-oceanic ridge, Johnson et al. (2000) observedthe activity of hydrothermal vents a few days after an earthquakeswarm of tectonic origin. However, this hypothesis is difficult to testas Taiwan experiences tectonic earthquakes on a daily basis, preventingfrom finding relevant time correlations between these earthquakes andthe fluid circulations.

6. Concluding remarks

The gravity variations recorded at TVG since 2005 primarily reflectthe local redistribution of water. They can be considered as the sumof a daily signal, due to the immediate effect of rains infiltrating theunsaturated zone and a pluri-months signal controlled by the watertable variations. Besides, the water table variations can also be properlyreconstructed only using rainfall data, and it suggests that the aquifer'sporosity is 10%. Once this hydrological effect is removed from thegravity time series we find large gravity residuals between 2005 and2007, which show the occurrence of mass transfers not related toseasonal hydrology. These gravity changes are concomitant to changesof the geochemical composition of the Dayoukeng fumarole (Lee et al.,2008) and to an increase of the occurrence of fluid-pressure inducedspasmodic bursts (Rontogianni et al., 2012). These gravity changes areinterpreted as the consequence of fluid mass transport, probably dueto hydrothermal circulations, although the data presented in thisstudy cannot rigorously show that this fluid iswater ormagma.We pro-pose that fluids circulate through a network of existing fractures hostedin a former magma passage that runs beneath TVG, parallel to theChinshan normal fault (Wen et al., 2012). The existence of such frac-tures is hypothesized because no clear surface ground displacementsare observed by GPS during the gravity changes, suggesting that thematerial could move without increasing the ambient pressure. Thisabsence of displacements results in a lack of constraints for invertingthe mass and path of the material, which are thus inferred only usingthe value of the gravity change and the location of the spasmodic bursts.Several models are possible and we find that at least 15 Mt of materialmight have transported in the first 2000 m of the surface and at most1000 m away (horizontally) from YMSG, among other possible scenar-ios. Aside from the confirmation of TVG's 2005–2007 mass transport,the interest of this study is to propose a suitable hydro-gravity modelthat can help to rapidly notice any departure from the usual local gravityeffect and identify future mass transfers due to hydrothermalism ormagmatic processes at depth. Then, measuring time-variable gravitydata at other sites on TVG, as soon as such a departure is identified,should permit to better unravel the path and mass associated to theseprocesses.

Acknowledgements

The data acquisition and the main analysis by M. M. in this workwere performed under the full support of the Theme Project of theAcademia Sinica of Taiwan. The facility of superconducting gravimeterand absolute gravimeter was graciously granted by the Ministry ofInterior of Taiwan. The work is further supported by the Ministry ofScience and Technology of Taiwan via grants #103-2116-M-001-024and 104-2611-M-009-001. Figs. 1 and 8b were done with GMTsoftware (Wessel and Smith, 1991).We thankHsiao-Fen Lee and TefangLan for insights about Tatun volcano's geochemistry, and LaurentLonguevergne for helpful discussions on groundwater flows. We alsothank Daniele Carbone and Philippe Jousset, whose reviews improvedthis manuscript.

tun Volcanic Group activities and inference for underground fluid/j.jvolgeores.2016.10.001

13M. Mouyen et al. / Journal of Volcanology and Geothermal Research xxx (2016) xxx–xxx

Appendix A. Supplementary data

Supplementary data associated with this article can be found in theonline version, at http://dx.doi.org/10.1016/j.jvolgeores.2016.10.001.These data include the Google map of the most important areasdescribed in this article.

References

Amaro, A., Chiarabba, C., 1995. Recent uplift of the Alban Hills Volcano (Italy): evidencefor magmatic inflation? Geophys. Res. Lett. 22, 1985–1988.

Bartel, B.A., Hamburger, M.W., Meertens, C.M., Lowry, A.R., Corpuz, E., 2003. Dynamics ofactive magmatic and hydrothermal systems at Taal Volcano, Philippines, fromcontinuous GPS measurements. J. Geophys. Res. 108, 2475. http://dx.doi.org/10.1029/2002JB002194.

Barton, C.A., Zoback, M.D., Moos, D., 1995. Fluid flow along potentially active faultsin crystalline rock. Geology 23, 683–686. http://dx.doi.org/10.1130/0091-7613.

Battaglia, M., Roberts, C., Segall, P., 1999. Magma intrusion beneath Long Valley Calderaconfirmed by temporal changes in gravity. Science 285 (80-.), 2119–2122. http://dx.doi.org/10.1126/science.285.5436.2119.

Battaglia, M., Troise, C., Obrizzo, F., Pingue, F., De Natale, G., 2006. Evidence for fluidmigration as the source of deformation at Campi Flegrei caldera (Italy). Geophys.Res. Lett. 33, L01307. http://dx.doi.org/10.1029/2005GL024904.

Beauducel, F., Cornet, F.H., 1999. Collection and three-dimensional modeling of GPS andtilt data at Merapi volcano, Java. J. Geophys. Res. 104, 725. http://dx.doi.org/10.1029/1998JB900031.

Belousov, A., Belousova,M., Chen, C.-H., Zellmer, G.F., 2010. Deposits, character and timingof recent eruptions and gravitational collapses in Tatun Volcanic Group, NorthernTaiwan: hazard-related issues. J. Volcanol. Geotherm. Res. 191, 205–221. http://dx.doi.org/10.1016/j.jvolgeores.2010.02.001.

Boy, J.-P., Gegout, P., Hinderer, J., 2002. Reduction of surface gravity data from globalatmospheric pressure loading. Geophys. J. Int. 149, 534–545. http://dx.doi.org/10.1046/j.1365-246X.2002.01667.x.

Carbone, D., Budetta, G., Greco, F., Rymer, H., 2003. Combined discrete and continuousgravity observations at Mount Etna. J. Volcanol. Geotherm. Res. 123, 123–135.http://dx.doi.org/10.1016/S0377-0273(03)00032-5.

Chang, C.-P., Chang, T.-Y., Angelier, J., Kao, H., Lee, J.-C., Yu, S.-B., 2003. Strain and stressfield in Taiwan oblique convergent system: constraints from GPS observation andtectonic data. Earth Planet. Sci. Lett. 214, 115–127. http://dx.doi.org/10.1016/S0012-821X(03)00360-1.

Chen, C.S., Chen, Y.L., 2003. The rainfall characteristics of Taiwan. Mon. Weather Rev. 131,1323–1341.

Creutzfeldt, B., Güntner, A., Klügel, T., Wziontek, H., 2008. Simulating the influenceof water storage changes on the superconducting gravimeter of the GeodeticObservatory Wettzell, Germany. Geophysics 73, WA95–WA104. http://dx.doi.org/10.1190/1.2992508.

Crossley, D., Xu, S., Van Dam, T., 1998. Comprehensive analysis of 2 years of SG data fromTable Mountain, Colorado. Proc. 13th Int. Symp. Earth Tides, Brussels, pp. 659–668.

De Linage, C., Hinderer, J., Rogister, Y., 2007. A search for the ratio between gravityvariation and vertical displacement due to a surface load. Geophys. J. Int. 171,986–994. http://dx.doi.org/10.1111/j.1365-246X.2007.03613.x.

Dong, D., Fang, P., Bock, Y., Cheng, M.K., Miyazaki, S., 2002. Anatomy of apparent seasonalvariations from GPS-derived site position time series. J. Geophys. Res. 107. http://dx.doi.org/10.1029/2001JB000573.

Fialko, Y., Khazan, Y., Simons, M., 2001. Deformation due to a pressurized horizontalcircular crack in an elastic half-space, with application to volcano geodesy. Geophys.J. Int. 146, 181–190.

Fournier, N., Chardot, L., 2012. Understanding volcano hydrothermal unrest from geodeticobservations: insights from numerical modeling and application to White Islandvolcano, New Zealand. J. Geophys. Res. 117, B11208. http://dx.doi.org/10.1029/2012JB009469.

Francis, O., Van Camp, M., Van Dam, T., Warnant, R., Hendrickx, M., 2004. Indication of theuplift of the Ardenne in long-term gravity variations in Membach (Belgium).Geophys. J. Int. 158, 346–352.

Goodkind, J.M., 1999. The superconducting gravimeter. Rev. Sci. Instrum. 70, 4131–4152.http://dx.doi.org/10.1063/1.1150092.

Hagiwara, Y., 1977. Gravity changes associated with seismic activities. J. Phys. Earth 25,S137–S146. http://dx.doi.org/10.4294/jpe1952.25.Supplement_S137.

Hill, D.P., Prejean, S., 2005.Magmatic unrest beneathMammothMountain, California. J. Volcanol.Geotherm. Res. 146, 257–283. http://dx.doi.org/10.1016/j.jvolgeores.2005.03.002.

Hinderer, J., Crossley, D.J., Warburton, R.J., 2007. Gravimetric methods – superconductinggravity meters. In: Herring, T., Schubert, G. (Eds.), Treatise on Geophysics, pp. 65–122.

Hofmann-Wellenhof, B., Moritz, H., 2006. Physical Geodesy. SpringerWienNewYork.Huang, H.H., Shyu, J.B.H., Wu, Y.M., Chang, C.H., Chen, Y.G., 2012. Seismotectonics of

northeastern Taiwan: kinematics of the transition from waning collision to subduc-tion and postcollisional extension. J. Geophys. Res. 117, B01313. http://dx.doi.org/10.1029/2011JB008852.

Imanishi, Y., Sato, T., Higashi, T., Sun, W., Okubo, S., 2004. A network of superconductinggravimeters detects submicrogal coseismic gravity changes. Science 306 (80-.),476–478. http://dx.doi.org/10.1126/science.1101875.

Jacob, T., Bayer, R., Chery, J., Jourde, H., Le Moigne, N., Boy, J.-P.J.-P., Hinderer, J., Luck, B.,Brunet, P., 2008. Absolute gravity monitoring of water storage variation in a karst

Please cite this article as: Mouyen, M., et al., Gravity monitoring of Tacirculations, J. Volcanol. Geotherm. Res. (2016), http://dx.doi.org/10.1016

aquifer on the Larzac plateau (Southern France). J. Hydrol. 359, 105–117. http://dx.doi.org/10.1016/j.jhydrol.2008.06.020.

Jamtveit, B., Kobchenko, M., Austrheim, H., Malthe-Soørenssen, A., Ryne, A., Svensen, H.,2011. Porosity evolution and crystallization-driven fragmentation during weatheringof andesite. J. Geophys. Res. 116. http://dx.doi.org/10.1029/2011JB008649.

Johnson, H.P., Hutnak, M., Dziak, R.P., Fox, C.G., Urcuyo, I., Cowen, J.P., Nabelek, J., Fisher, C.,2000. Earthquake-induced changes in a hydrothermal system on the Juan de Fucamid-ocean ridge. Nature 407, 174–177. http://dx.doi.org/10.1038/35025040.

Jousset, P., Okada, H., 1999. Post-eruptive volcanic dome evolution as revealed bydeformation and microgravity observations at Usu volcano (Hokkaido, Japan).J. Volcanol. Geotherm. Res. 89, 255–273. http://dx.doi.org/10.1016/S0377-0273(99)00003-7.

Jousset, P., Dwipa, S., Beauducel, F., Duquesnoy, T., Diament, M., 2000. Temporal gravity atMerapi during the 1993–1995 crisis: an insight into the dynamical behaviour of vol-canoes. J. Volcanol. Geotherm. Res. 100, 289–320. http://dx.doi.org/10.1016/S0377-0273(00)00141-4.

Jousset, P., Mori, H., Okada, H., 2003. Elastic models for the magma intrusion associatedwith the 2000 eruption of Usu Volcano, Hokkaido, Japan. J. Volcanol. Geotherm.Res. 125, 81–106. http://dx.doi.org/10.1016/S0377-0273(03)00090-8.

Kazama, T., Okubo, S., 2009. Hydrological modeling of groundwater disturbances to ob-served gravity: theory and application to Asama Volcano, Central Japan. J. Geophys.Res. 114, B08402. http://dx.doi.org/10.1029/2009JB006391.

Kim, K.-H.H., Chiu, J.-M.M., Pujol, J., Chen, K.-C.C., Huang, B.-S.S., Yeh, Y.-H.H., Shen, P.,2005. Three-dimensional VP and VS structural models associatedwith the active sub-duction and collision tectonics in the Taiwan region. Geophys. J. Int. 162, 204–220.http://dx.doi.org/10.1111/j.1365-246X.2005.02657.x.

Konstantinou, K.I., Lin, C.-H., Liang, W.-T., 2007. Seismicity characteristics of a potentiallyactive Quaternary volcano: the Tatun Volcano Group, northern Taiwan. J. Volcanol.Geotherm. Res. 160, 300–318. http://dx.doi.org/10.1016/j.jvolgeores.2006.09.009.

Lee, H.-F., Yang, T.F., Lan, T.F., Chen, C.-H., Song, S.-R., Tsao, S., 2008. Temporal variations ofgas compositions of fumaroles in the Tatun Volcano Group, northern Taiwan.J. Volcanol. Geotherm. Res. 178, 624–635. http://dx.doi.org/10.1016/j.jvolgeores.2008.06.005.

Lee, H.F., Yang, T.F., Lin, C., Lan, T., Pu, H., Mouyen, M., 2016. The association betweenvolcanic fluid variations and seismic events in the Tatun Volcano Group, northernTaiwan. European Geosciences Union General Assembly 2016, Vienne, Austria.Geophys. Res. Abstr. 18, pp. EGU2016–10740.

Lien, T., Cheng, C.-C., Hwang, C., Crossley, D.J., 2014. Assessing active faultingby hydrogeological modeling and superconducting gravimetry: a case study forHsinchu Fault, Taiwan. J. Geophys. Res. 119, 7319–7335. http://dx.doi.org/10.1002/2014JB011285.

Lin, C.H., Konstantinou, K.I., Liang, W.-T., Pu, H.-C., Lin, Y.-M., You, S.-H., Huang, Y.-P., 2005.Preliminary analysis of volcanoseismic signals recorded at the Tatun Volcano Group,northern Taiwan. Geophys. Res. Lett. 32, L10313. http://dx.doi.org/10.1029/2005GL022861.

Longuevergne, L., Boy, J.-P., Florsch, N., Viville, D., Ferhat, G., Ulrich, P., Luck, B.,Hinderer, J., 2009. Local and global hydrological contributions to gravity vari-ations observed in Strasbourg. J. Geodyn. 48, 189–194. http://dx.doi.org/10.1016/j.jog.2009.09.008.

Microg-LaCoste, 2008. g8 Absolute Gravity Data Acquisition and Processing Software.Micro-g LaCoste, Inc., Lafayette, Colorado, E.U. (www.microglacoste.com).

Niebauer, T.M., Sasagawa, G.S., Faller, J.E., Hilt, R., Klopping, F., 1995. A new generation ofabsolute gravimeters. Metrologia 32, 159–180.

Pitt, A.M., Hutchinson, R.A., 1982. Hydrothermal changes related to earthquake activity atMud Volcano, Yellowstone National Park, Wyoming. J. Geophys. Res. 87, 2762. http://dx.doi.org/10.1029/JB087iB04p02762.

Pu, H.-C., Lin, C.-H., Huang, Y.-C., Chang, L.-C., Lee, H.-F., Leu, P.-L., Hsiao, N.-C., Chang, C.-H., Yeh, Y.-H., 2014. The volcanic earthquake swarm of October 20, 2009 in the Tatunarea of northern Taiwan. Terr. Atmos. Ocean. Sci. 25, 625–635. http://dx.doi.org/10.3319/TAO.2014.04.11.02(T)1.

Puglisi, G., Bonforte, A., Maugeri, S.R., 2001. Ground deformation patterns on Mount Etna,1992 to 1994, inferred from GPS data. Bull. Volcanol. 62, 371–384. http://dx.doi.org/10.1007/s004450000112.

Rebischung, P., Griffiths, J., Ray, J., Schmid, R., Collilieux, X., Garayt, B., 2012. IGS08: the IGSrealization of ITRF2008. GPS Solut. 16, 483–494. http://dx.doi.org/10.1007/s10291-011-0248-2.

Rontogianni, S., Konstantinou, K.I., Lin, C.-H., 2012. Multi-parametric investigation of thevolcano-hydrothermal system at Tatun Volcano Group, Northern Taiwan. Nat. Haz-ards Earth Syst. Sci. 12, 2259–2270. http://dx.doi.org/10.5194/nhess-12-2259-2012.

Rymer, H., Cassidy, J., Locke, C.A., Murray, J.B., 1995. Magma movements in Etna volcanoassociated with the major 1991–1993 lava eruption: evidence from gravity and de-formation. Bull. Volcanol. 57, 451–461. http://dx.doi.org/10.1007/BF00300989.

Rymer, H., Murray, J.B., Brown, G.C., Ferrucci, F., McGuire, W.J., 1993. Mechanisms ofmagma eruption and emplacement at Mt Etna between 1989 and 1992. Nature361, 439–441. http://dx.doi.org/10.1038/361439a0.

Sasai, Y., 1986. Multiple tension-crack model for dilatancy: surface displacements, gravityand magnetic change. Bull. Earthquake Res. Inst. 61, 429–473.

Sasai, Y., 1988. Correction to the paper “Multiple tension-crack model for dilatancy: surfacedisplacements, gravity and magnetic change”. Bull. Earthquake Res. Inst. 63, 323–326.

Shyu, J.B.H., Sieh, K., Chen, Y.G., Liu, C.S., 2005. Neotectonic architecture of Taiwan and itsimplications for future large earthquakes. J. Geophys. Res. 110, B08402. http://dx.doi.org/10.1029/2004JB003251.

Singh, K.S., 1977. Gravitational attraction of a circular disc. Geophysics 42, 111–113.Solaro, G., Acocella, V., Pepe, S., Ruch, J., Neri, M., Sansosti, E., 2010. Anatomy of an unstable

volcano from InSAR: multiple processes affecting flank instability at Mt. Etna, 1994–2008. J. Geophys. Res. 115, 1994–2008. http://dx.doi.org/10.1029/2009JB000820.

tun Volcanic Group activities and inference for underground fluid/j.jvolgeores.2016.10.001

14 M. Mouyen et al. / Journal of Volcanology and Geothermal Research xxx (2016) xxx–xxx

Song, S.-R., Yang, T.-F., Yeh, Y.-H., Tsao, S., Lo, H.-J., 2000. The Tatun Volcano Group is ac-tive or extinct? J. Geol. Soc. China 43, 521–534.

Teng, L.S., 1996. Extensional collapse of the northern Taiwan mountain belt. Geology 24,949.

Torge, W., 1989. Gravimetry (465 pp.) Walter de Gruyter, Berlin.Van Camp, M., Francis, O., 2007. Is the instrumental drift of superconducting gravimeters

a linear or exponential function of time? J. Geod. 81, 337–344. http://dx.doi.org/10.1007/s00190-006-0110-4.

Van Camp, M., Vauterin, P., 2005. Tsoft: graphical and interactive software for the analysisof time series and Earth tides. Comput. Geosci. 31, 631–640. http://dx.doi.org/10.1016/j.cageo.2004.11.015.

Van Camp, M., de Viron, O., Scherneck, H.-G.G., Hinzen, K.-G.G.,Williams, S.D.P., Lecocq, T.,Quinif, Y., Camelbeeck, T., 2011. Repeated absolute gravity measurements formonitoring slow intraplate vertical deformation in western Europe. J. Geophys. Res.116, B08402. http://dx.doi.org/10.1029/2010JB008174.

Wang, K.-L., Chung, S.-L., Chen, K.-H., Shinjo, R., Yang, T.F., Chen, C.-H., 1999. Post-collisional magmatism around northern Taiwan and its relation with opening of the

Please cite this article as: Mouyen, M., et al., Gravity monitoring of Tacirculations, J. Volcanol. Geotherm. Res. (2016), http://dx.doi.org/10.1016

Okinawa Trough. Tectonophysics 308, 363–376. http://dx.doi.org/10.1016/S0040-1951(99)00111-0.

Wang,W.-S., Chen, C.-H., 1990. The volcanology and fission track age dating of pyroclasticdeposits in Tatun Volcano Group, northern Taiwan. Acta Geol. Taiwan. 28, 1–40.

Wen, S., Chang, Y.-Z., Chen, C.-H., Chen, Y.-G., Teng, T.-L., 2012. The seismic velocity andattenuation structure beneath the Tatun volcanic area, Taiwan. J. Asian Earth Sci.54-55, 182–191. http://dx.doi.org/10.1016/j.jseaes.2012.04.011.

Wenzel, H.-G., 1996. The nanoGal software: earth tide data processing package: Eterna3.3. Bull. Inf. Marées Terr. 124, 9425–9439.

Wessel, P., Smith, W.H.F., 1991. Free software helps map and display data. EOS Trans. 72,445–446.

Williams, S.D.P., 2008. CATS: GPS coordinate time series analysis software. GPS Solutions12, 147–153. http://dx.doi.org/10.1007/s10291-007-0086-4.

Yu, S.-B., Chen, H.-Y., Kuo, L.-C., 1997. Velocity field of GPS stations in the Taiwan area.Tectonophysics 274, 41–59. http://dx.doi.org/10.1016/S0040-1951(96)00297-1.

tun Volcanic Group activities and inference for underground fluid/j.jvolgeores.2016.10.001