G 3. AN ELECTRONIC JOURNAL OF THE EARTH SCIENCES Published by AGU and the Geochemical Society

Transcription

1 Geosystems G 3 AN ELECTRONIC JOURNAL OF THE EARTH SCIENCES Published by AGU and the Geochemical Society Article Volume 11, Number 1 19 January 2010 Q01003, doi: ISSN: Click Here for Full Article Duration, magnitude, and frequency of subaerial volcano deformation events: New results from Latin America using InSAR and a global synthesis T. J. Fournier, M. E. Pritchard, and S. N. Riddick Department of Earth and Atmospheric Sciences, Cornell University, Snee Hall, Ithaca, New York 14853, USA (mp337@cornell.edu) [1] We combine new observations of volcano deformation in Latin America with more than 100 previous deformation studies in other areas of the world to constrain the frequency, magnitude, and duration of subaerial volcano deformation events. We discuss implications for eruptive hazards from a given deformation event and the optimum repeat interval for proposed InSAR satellite missions. We use L band (23.6 cm wavelength) satellite-based interferometric synthetic aperture radar (InSAR) to make the first systematic search for deformation in all volcanic arcs of Latin America (including Mexico, Central America, the Caribbean, and the northern and southern Andes), spanning We combine L and C band (5.6 cm wavelength) InSAR observations over the southern Andes volcanoes to extend the time series from 2002 to 2008 and assess the capabilities of the different radars: L band gives superior results in highly vegetated areas. Our observations reveal 11 areas of volcano deformation, some of them in areas that were thought to be dormant. There is a lack of observed deformation at several erupting volcanoes, probably due to temporal aliasing. The total number of deforming volcanoes in the central and southern Andes now totals 15 (from observations between 1992 and 2008), comparable to the Alaska/Aleutian arc. Globally, volcanoes deform across a variety of time scales (from seconds to centuries) often without eruption and with no apparent critical observation time scale, although observations made every minute are sometimes necessary to see precursors to eruption. Components: 18,269 words, 19 figures, 3 tables. Keywords: InSAR; Andes; volcano; Caribbean; Central America; deformation. Index Terms: 8485 Volcanology: Remote sensing of volcanoes. Received 15 April 2009; Revised 13 October 2009; Accepted 13 November 2009; Published 19 January Fournier, T. J., M. E. Pritchard, and S. N. Riddick (2010), Duration, magnitude, and frequency of subaerial volcano deformation events: New results from Latin America using InSAR and a global synthesis, Geochem. Geophys. Geosyst., 11, Q01003, doi:. 1. Introduction [2] Deformation of the Earth s surface at volcanoes provides clues to the myriad processes occurring below and above the surface [e.g., Dvorak and Dzurisin, 1997], and might provide warning of an imminent eruption [e.g., Swanson et al., 1983; Klein, 1984]. Unfortunately, experience has shown that volcanoes have different behaviors before eruptions. Some volcanoes give little obvious warning that they are about to erupt [e.g., Hall et al., 2004], or may give many indications of Copyright 2010 by the American Geophysical Union 1 of 29

2 Figure 1. Global map of volcanoes (black triangles) (Smithsonian Institution, Global volcanism report, available at with areas of observed volcanic deformation shown as red triangles (Table 1). Several areas have not been completely surveyed for deformation (see section 1). impending eruption, but do not actually erupt (e.g., restless calderas like Long Valley, CA). Because different volcanoes have these different personalities, deformation must be monitored at all volcanoes, and a history of precursory activity and eruption should be established for each volcano. [3] Of the more than 1500 potentially active volcanoes around the world, the past performance of only a few dozen is well documented [e.g., Simkin and Siebert, 1994; Dvorak and Dzurisin, 1997]. Another problem is that the list of 1500 potentially active volcanoes is incomplete: occasionally, volcanoes that are not believed to be active can erupt (e.g., Mt. Pinatubo, Philippines, 1991) or at least show some sign of seismic or deformation activity [e.g., Pritchard and Simons, 2002]. For example, as of 1997, surface deformation had been observed at only 44 different volcanoes using ground-based methods (e.g., traditional surveying, tiltmeters, or the Global Positioning System, GPS) [Dvorak and Dzurisin, 1997]. In the last decade or so, observations of deformation at volcanoes have more than doubled to 110 (Figure 1; for a complete listing of observations of volcano deformation see Table 1) and are due largely to the use of satellite-based interferometric synthetic aperture radar (InSAR). Many of the deforming volcanoes discovered with InSAR were not thought to be active [e.g., Lu et al., 2000; Amelung et al., 2000a; Lu et al., 2002; Wicks et al., 2002; Pritchard and Simons, 2002]. The hazard from these presumed magma intrusions is unclear: will this magma accumulation result in an eruption, or is this a benign intrusion? [4] In an effort to address the question of potential hazard from volcano deformation events, we have compiled a database of deforming volcanoes from the literature (Table 1 and Figure 1). In order to fill in some regional and temporal gaps of volcano deformation, we also add new observations in Latin America from 2006 to 2008 using the Japanese L band radar instrument on board the ALOS satellite. Because 2 years of data are not sufficient to characterize volcanic activity, we include a longer time series of observations ( ) for a subsection of Latin America the Southern Volcanic Zone of Chile and Argentina. Our new observations reveal volcanic deformation in 11 2of29

3 Table 1 (Sample). Deforming Volcanoes Across the Globe Along With Some Information About the Type of Deformation Observed a [The full Table 1 is available in the HTML version of this article] Volcano Latitude Longitude Volcano Number Observation Frequency Duration Magnitude (cm/yr) Magnitude (mrad/yr) Type Method Aliased Reference Italy Campi Flegrei year >10 years 3 IE GPS, InSAR yes Gottsmann et al. [2006] Vesuvius <1 year >10 years 10 IE InSAR, leveling, tilt, trilateration yes Lanari et al. [2002] Stromboli continuous 2 days 10 IE GPS no Mattia et al. [2004] Vulcano year 18 months ±1 IE leveling yes Ferri et al. [1988] Etna year years 1 E InSAR, GPS yes Bonforte et al. [2008] continuous 6 days 730 PE GPS, tilt no Bonaccorso et al. [2002] Nisyros continuous minutes 10 IE GPS, InSAR no Gottsmann et al. [2007] Africa Gada Ale year 3 years 12 IE InSAR yes Amelung et al. [2000b] Dabbahu year 7 days 800 IE InSAR yes Wright et al. [2006] Asal-Ghoubbet years days 200 E leveling, trilateration yes Ruegg et al. [1979] Menengai year <3 years 1 IE InSAR yes Biggs et al. [2009] Longonot year <2 years 3.3 IE InSAR yes Biggs et al. [2009] Suswa year <3 years 1.5 IE InSAR yes Biggs et al. [2009] Lengai, Ol Doinyo months 3 months 20 E InSAR yes Baer et al. [2008] Paka months <9 months 25 IE InSAR yes Biggs et al. [2009] Nyamuragira year 2 months 28 E InSAR yes Cayol et al. [2007] Nyiragongo year months 15 E trilateration, yes Poland and Lu [2004] tilt, InSAR Indian Ocean and Arabia Piton de la Fournaise months 2 months 30 E InSAR yes Froger et al. [2004] Harrat Lunayyir year 2 days IE InSAR yes Gomez et al. [2009] New Zealand White Island months months-years ±20 IE leveling yes Peltier et al. [2009] Taupo year years 20 G InSAR, leveling yes Hole et al. [2007] Ruapehu year 4 years 2 trilateration yes Dvorak and Dzurisin [1997, and references therein] Indonesia and South Pacific Manam day 6 years 3.3 E tilt no Mori et al. [1987] Sulu Range year days? 100 IE InSAR yes Wicks et al. [2007] a The columns are as follows: volcano name, latitude, longitude, the volcano number assigned by the Smithsonian Institution, the relevant (or shortest) observation frequency, duration of deformation event, magnitude of deformation event (in cm/yr or mrad/yr), type of deformation event, observation method, whether or not the observation is aliased, and references. The different types of deformation are broken into seven broad categories: E, eruptive; IE, intereruptive; PE, preeruptive; G, geothermal; FD, flow deposit; GW, ground water; F, flank. We do not include the several hundred volcanoes that have robust observations of no deformation. The list of references is not complete. For each volcano, we cited a paper that supports the duration and magnitude of volcano deformation in the table. If multiple techniques have been used, we cite a paper that summarizes the observations or in a few cases, we have cited two papers that use different methods. Many volcanoes have complex temporal deformation patterns. To represent this complexity, we have more than one entry of deformation magnitude and duration for some volcanoes, but our list does not capture the full complexity of variation in types of volcano deformation. 3of29

4 different areas (10% of all known), including several areas that were thought to be dormant, demonstrating the current incompleteness of global volcano monitoring. Furthermore, many volcanic areas have still not yet been surveyed for deformation (e.g., Marianas, the Scotia arc, Kurile Islands, many areas of southwest Pacific, and ocean islands). In spite of the limitations of global observations, patterns of the frequency, magnitude, and duration of volcanic deformation events emerge. [5] In addition to assessing the potential hazard from volcano deformation events, another goal of our global compilation is to determine how frequently a satellite must image a given volcano to be able to detect precursory deformation before an eruption or to obtain a quality measurement. For example, within the next decade, NASA is planning to launch an L band radar satellite called DESDynI (Deformation, EcoSystems and Dynamics of Ice) with several observation targets, including volcanoes [Anthes et al., 2007], and several possible repeat intervals have been proposed. Most InSAR studies to date have used C band radars (5.6 cm wavelength) over 35 days that are not able to form usable interferograms over vegetated volcanoes [e.g., Zebker et al., 2000]. We assess whether the longer radar wavelength of ALOS (23.6 cm) allows coherent interferograms to be formed in Latin America in different seasons and over different time scales and investigate how different satellite repeat intervals will affect volcano monitoring capabilities. 2. Methods and Data Quality [6] We use the JPL/Caltech ROI_PAC software for data processing [Rosen et al., 2004] and digital elevation models (DEM) from the Shuttle Radar Topography Mission with 90 m pixel spacing to remove topography from the InSAR phase [Farr et al., 2007] (Tools for ALOS available at org/alos_palsar). [7] One advantage of the C band data over the L band data is that the shorter radar wavelength is more sensitive to small deformations. We expect that deformation rates above 3 5 cm/yr within coherent areas should be detectable with 1 year ALOS interferograms. For example at Lonquimay volcano in central Chile, a comparison of C and L band data of a subsiding lava flow suggests that the nine month ALOS interferogram is at the limit for detecting the subsidence which was clearly seen at C band (see section 3.1.3). [8] In areas where the deformation is of likely subsurface origin, we model the deformation data using simple elastic forward models and a Levenberg-Marquardt algorithm as implemented in the Matlab optimization toolbox for inversions. All inversions contain a deformation source (volcanic source), a static offset, and a bilinear ramp to account for interferometric artifacts. We use the following volcanic source models; an opening dislocation [Okada, 1992], isotropic point source [Mogi, 1958], and a prolate spheroid [Yang et al., 1988]. The best model is chosen based on misfit and realistic volcanic source geometries. We know that variations in source geometry [e.g., Pritchard and Simons, 2004a] and 3-D crustal structure [e.g., Masterlark, 2007] can effect source depths and volumes, but the simple modeling done here allows us to gain an order of magnitude estimate of these important quantities. More sophisticated modeling is not justified considering that there are limited InSAR acquisitions (we cannot reconstruct enough of the 3-D deformation field necessary to differentiate between models [Dieterich and Decker, 1975]), and there is no information on the subsurface material properties at these volcanoes (necessary to compute deformation in a layered elastic medium or in a finite element model). 3. New Results From Latin America [9] Latin America includes several distinct volcanic arcs, Mexico, Central America, the Caribbean, and the northern, central, southern, and austral Andes, encompassing about 300 volcanoes in the Smithsonian catalog of Holocene volcanism (Smithsonian Institution, Global volcanism report, available at Other volcano catalogs indicate 2500 volcanoes in the southern and central Andes alone, although the majority are dormant [de Silva and Francis, 1991; Gonzalez-Ferran, 1995]. While there are several well monitored volcanoes in the region, the activity of most is unknown. Our data will be of value in setting a baseline for assessing whether future bursts of activity are normal or may indicate escalating unrest that merits further attention. [10] We use the data from ALOS to survey the volcanoes of Central America, the Caribbean and the northern and southern Andes with data that spans (see Tables S1 S4). 1 While our survey is spatially comprehensive, it is quite 1 Auxiliary materials are available in the HTML. doi: / 2009GC of29

5 possible that we have missed deformation that is small in magnitude or spatial scale. We use long time periods to maximize sensitivity to small deformation rates (e.g., 1 year for ALOS), but this also means that the scattering properties of the ground might have changed between observations (causing decorrelation), and the observations at volcanoes with frequent changes in deformation rate and direction might be temporally aliased. Because data acquisitions are infrequent, data quantity and quality are not uniform at all volcanoes. [11] For the southern Andes, we combine C and L band observations in an effort to understand how effective the different sensors are. By using multiple satellites, we compare different time spans (6 years for C band, 1 year for L band) as an indication of what our 1 year ALOS derived results in other parts of Latin America might be missing. We do not present new observations of the central Andes because to date, C band observations provide good coverage ( ) of this arid region with little vegetation or population [e.g., Pritchard and Simons, 2002, 2004a; Froger et al., 2007; Ruch et al., 2008; Sparks et al., 2008]. We do not present new observations for the austral Andes, because of the limited number of ALOS scenes collected in the austral summer Southern Volcanic Zone of South America [12] The Southern Volcanic Zone (SVZ) of South America, (33 S 46 S) includes about 1300 volcanoes, with approximately 60 considered potentially active (having eruptions within the Holocene or last 10,000 years) [Gonzalez-Ferran, 1995; Smithsonian Institution, Global volcanism report, available at (Figure 2c). We compare results from the European Space Agency s ERS and Envisat C band radar satellites between the years , with the ALOS L band radar satellite from 2007 to 2008 (spanning January March 2007 to January March 2008). Since observations are limited to the austral summer, comprehensive coverage of all volcanoes is not possible. While a few volcanoes are monitored by the Southern Andes Volcano Observatory (OVDAS, administered by the Chilean Servicio Nacional de Geología y Mineria), the activity of the majority of these volcanoes is not well known. [13] A previous satellite-based InSAR study made observations of most volcanoes south of 40.5 S during the austral summer using C band radar, and found active deformation at two of them [Pritchard and Simons, 2004b]. Our combined ALOS and Envisat data analysis allows us to make measurements at about 40 Holocene volcanoes (Table S1) and extend observations north of 40.5 S for the first time. Most volcanoes showed no deformation within our detection threshold, including no deformation associated with small eruptions at Nevados de Chillán (2003), Villaricca ( ) and Llaima (2003, 2007) (Smithsonian Institution, Global volcanism report, available at volcano.si.edu), or the seismic swarm at Hornopirén (A. Pavez, personal communication, 2008). The lack of coeruptive deformation may indicate real constraints on eruptive processes or simply reflect poor temporal or spatial sampling, due to decorrelation. We find previously undocumented volcanic/hydrothermal deformation at Lonquimay, Llaima, Laguna del Maule, and Chaitén volcanoes, extend deformation measurements at Copahue, and illustrate temporal complexity to the previously described deformation at Cerro Hudson and Cordón Caulle. We next discuss the deformation in these seven volcanic areas from north to south, but do not discuss earthquake swarm deformation in the area near Puerto Aysén which might have a magmatic component [e.g., Fukushima, 2007]. Model results are given in Table Laguna del Maule [14] Laguna del Maule is a caldera on the Chile- Argentina border encompassing pyroclastic cones, stratovolcanoes, and lava domes of Pleistocene through Holocene age [e.g., Gonzalez-Ferran, 1995; Smithsonian Institution, Global volcanism report, available at but with no known historic activity. The observed deformation field (Figures 3b and 3c) has a maximum inflation rate of about 18.5 cm/yr in the radar line of sight from ALOS spanning January 2007 to January Using C band Envisat between March 2003 and February 2004, the interferogram is coherent but shows no deformation (Figure 3a), indicating the deformation rate is variable in time. We assume that the deformation is the result of a volume change caused by the inflation of a magma chamber or injection of hydrothermal fluids. A shallowly dipping sill is the best fit model located at 5 km depth with a dip of 20 and an approximate opening of 60 cm/yr (Table 2). The model residuals along with the model predictions are shown in Figures 3d 3f. The root mean squared error (RMSE) difference between the data and the model is 1 2 cm for most interferograms we 5of29

6 Geochemistry Geosystems 3 G fournier et al.: duration and magnitude of volcano deformation Figure 2. Interferometric coherence from ALOS for the volcanic arcs of Latin America draped over shaded topography. (a) Central America and Me xico, (b) the northern Andes, (c) the southern Andes, and (d) the Caribbean. The time period for most interferograms is 1 year, and the baseline is about 1 km, although in a few cases (especially the Caribbean), only shorter time periods are available. For a complete list of interferograms used, see Tables S1 S4. Coherence is calculated in a 5 5 pixel moving box. Large triangles with black outlines are volcanoes from the Smithsonian Institution (Global volcanism report, available at and yellow triangles show volcanoes mentioned in the text. (e) A reference map of the study area. have studied, probably within the range of the inherent InSAR noise [e.g., Pritchard and Simons, 2004a] Copahue [15] Copahue is an active stratovolcano along the Chile-Argentina border with an active hydrothermal system, a VEI 2 eruption in 2000 [Gonzalez-Ferran, 1995; Smithsonian Institution, Global volcanism report, available at and volcanic tremor [Iba nez et al., 2008]. Euillades et al. [2008] combined 16 scenes of an ascending track of Envisat data (spanning December 2002 and April 2006) with a time series approach to infer subsidence at a rate of 2 cm/yr. We confirm their result with the ascending data, and in addition, we use a descending track of Envisat data to document a similar rate of deformation up to January 2008 (Figure 4). A 1 year ALOS interferogram (Figure 4d) does not show any discernable signal, most likely because the deformation rate is too slow for this short time period. Further, we model the data shown in Figures 4b and 4c with an isotropic source (Figure 4e). The models for the 2 interferograms indicate a relatively constant contraction rate of about km3/yr (Table 2). The shallow depth of the source, 4 km, is consistent with the expected depth of hydrothermal activity Lonquimay [16] Lonquimay is a dominantly andesitic stratovolcano in central Chile with 5 historical eruptions [Gonzalez-Ferran, 1995], the most recent being from 1988 through 1990 when about 0.2 km3 of 6 of 29

7 Table 2. Model Results From Deforming Volcanoes in the Andes a Chaiten Cordon Caulle Maule Copahue Tungurahua Dates Apr 2007 to Jul 2008 Apr 2007 to Jul 2008 Model form dike normal fault Apr 2007 to Jul 2008 Apr 2007 to Jul 2008 Jan 2007 to Feb 2008 fault Mogi first Mogi Jan 2007 to Feb 2008 second Mogi Jan 2007 to Jan 2008 Feb 2007 to Dec 2007 joint Mar 2003 to Feb 2005 Feb 2004 to Jan 2008 joint Dec 2007 to May 2008 sill sill sill Mogi Mogi Mogi dike/crack MSE b 3.85 c Longitude d (deg) Latitude d (deg) X position d (km) Y position d (km) Depth d (km) , 4.1 DV (10 6 m 3 ) Length (km) c Width (km) c Dip (deg) c Strike (deg) c Open/slip (m) Static shift (cm) x ramp (cm/km) y ramp (cm/km) c , b 3.6 c , , b 0.20 c , , b 0.02 c , , a The time span of each interferogram is shown in the date row, and joint refers to an inversion that uses both available interferograms. All models include a static shift and bilinear ramp for each interferogram. b Values are for both fault and Mogi models. c Values are for both first and second Mogi models. d For fault and sill models the longitude, latitude, and depth values indicate the center of the updip edge of the plane. 7of29

8 Figure 3. At Laguna del Maule in southern Chile, (a) an Envisat image spanning March 2003 to February 2004 shows no deformation, but (b and c) recent ALOS images show inflation. Both of the ALOS interferograms are used to model the deformation source as an inflating sill like structure (Table 2). (d) The residual from the image spanning January 2007 to January 2008 and (e) the model prediction for that image. The surface projection of the model is drawn as a black rectangle with the updip edge of the sill shown in bold (Figure 3e). (f) The model prediction for February 2007 to December The black line shows the Chile-Argentina border. White triangles show Holocene volcanoes (Smithsonian Institution, Global volcanism report, available at while black triangles are from the catalog of [Gonzalez-Ferran, 1995]. The line of sight (LOS) between the satellite and the ground is shown by the arrow. The interferograms have been unwrapped and then rewrapped at different intervals to highlight different magnitudes of deformation. blocky andesitic lavas erupted from the NE flank vents [Naranjo et al., 1992; Smithsonian Institution, Global volcanism report, available at An Envisat interferogram spanning December 2002 to January 2008 shows deflation with a maximum rate of about 2 cm/yr (Figure 5). The deformation is likely due to the subsiding lava flows from the eruption, which have subsided by as much as 20 m in some places (H. Moreno, personal communication, 2008) Llaima [17] Llaima is a compound stratovolcano in Chile and is one of its largest and most active volcanoes [Gonzalez-Ferran, 1995; Smithsonian Institution, Global volcanism report, available at volcano.si.edu]. InSAR shows temporally complex deformation of uplift and subsidence up to 11 cm on the eastern flank (Figure 6). The reason for the deformation is unknown, but appears to begin in December 2007 and be related to the January/ February 2008 eruption. The 2008 eruption produced lahars and/or pyroclastic flows (Smithsonian 8of29

9 Figure 4. Subsidence at Copahue, located in southern Chile, is too slow to be seen in single year interferograms: (a) December 2002 to February 2004 and (d) January 2007 to February Multiyear interferograms show approximately 2 cm/yr of maximum subsidence: (b) February 2003 to February 2005 and (c) February 2004 to January Interferograms shown in Figures 4b and 4c are used to model the deformation source as an isotropic point force (Table 2). (e) The model prediction for February 2004 to January The source location is drawn as a black dot. (f) The residual of the model and data in Figures 4e and 4c, respectively. The black line shows the Chile- Argentina border. Symbols are the same as in Figure 3. Institution, Global volcanism report, available at that traveled down the eastern flank in the general location of the deformation. These deposits seem to be an unlikely cause of the deformation, however, because (1) the flows did not travel very far down the flank and (2) the area of deformation shows high correlation, while decorrelation would be expected on freshly deposited surfaces. We suspect that the deformation is related to an observed sector collapse and a kind of creep movement (H. Moreno, personal communication, 2008). 9of29

10 Figure 5. Interferograms showing subsidence of the lava flows from Navidad crater near Lonquimay volcano from 2002 to (a and b) Subsidence in several areas of the lava flow are seen in Envisat data. We have interpolated the SRTM 90 m DEM to 30 m here to reveal details of the deformation field. (c) An ALOS interferogram that spans a shorter time interval only shows deformation in a limited part of the lava flow, which appears to be subsiding the fastest. Symbols are the same as in Figure Cordón Caulle [18] We observe varying deformation at Cordón Caulle, a 17 km long by 2.5 km wide zone of volcanic fissures, domes and craters between Cordillera Nevada caldera and Puyehue volcano [Gerlach et al., 1988] in southern Chile. The most recent eruption occurred in 1960 following 2 days after the M9.5 Chilean earthquake [Gonzalez- Ferran, 1995; Lara et al., 2004; Walter and Amelung, 2007] (called Cordon Gaulle by Walter and Amelung [2007]). Previous work by Pritchard and Simons [2004b] showed a possible deflation rate of 3 cm/yr between February 1996 and February 1999 (Figure 7a), but at the time only a single interferogram could be made of this area. Our new analysis of Envisat data (Figures 7b and 7c) shows that from 2003 to 2005, there was a mean inflation rate of 1 cm/yr, and from 2004 to 2006 the rate increased to 3 cm/yr. An ALOS interferogram made from January 2007 to February 2008 (Figure 7d) exhibited a marked increase in inflation rate, to 19.8 cm/yr. [19] The complex deformation pattern at Cordón Caulle was fit using two isotropic point sources (Figure 7f), one at 7 km depth and the other at 4 km depth (Table 2). The cumulative volume change of the two sources is 0.06 km 3. The shallower source is 0.02 km 3 and the deeper source is 0.04 km 3. The deformation may be a result of hydrothermal activity (Cordón Caulle is the largest active hydrothermal area of the SVZ [e.g., Gonzalez-Ferran, 1995; Smithsonian Institution, Global volcanism report, available at and/or an inflating magma chamber. The area has been under yellow alert due to abnormal seismic swarms from May 2008 to September 2008 and off and on alert status prior to that ( Chaitén [20] Chaitén is located in Southern Chile and is composed of a small caldera, 3 km in diameter, with a growing lava dome [e.g., Gonzalez-Ferran, 1995; Smithsonian Institution, Global volcanism report, available at Thirty-eight hours of intense seismic activity started on 30 April 2008, and culminated in the first eruption in 9400 years [Lara et al., 2008; Smithsonian Institution, Global volcanism report, available at The volume of the ash deposits is between 1 and 5 km 3 and the dome volume, which continues to grow, was about 0.5 km 3 by December 2008 [Lara et al., 2008]. In the previous study by Pritchard and Simons [2004b], the InSAR phase at Chaitén was only coherent on the young lava within the caldera, but no deformation was observed (upper limit of about 3 cm/yr). Also, no deformation was observed in several ALOS interferograms between April 2007 and 16 April 2008 (Figure 8b), suggesting that any precursory deformation above about 3 cm/yr occurred less than 2 weeks before the eruption, perhaps coincident with the seismicity 38 h before the eruption. This suggests rapid magma accent at this long dormant volcano. [21] Unfortunately, the earliest posteruptive ALOS scene is from July 2008 and includes a combination of preeruptive, coeruptive and posteruptive deformation. Between April 2007 and July 2008, Chaitén showed 22 cm of subsidence (Figure 8a). We have made only one successful ERS interferogram because of few acquisitions, slow delivery 10 of 29

11 Figure 6 11 of 29

12 Figure 7. Cordón Caulle is located in southern Chile and shows temporal variation in its deformation. (a) Pritchard and Simons [2004b] reported subsidence. (b and c) Envisat interferograms between 2003 and 2006 are mostly incoherent but suggest that deformation might have occurred. (d) An ALOS interferogram between January 2007 and February 2008 shows uplift. Note that the wrap rate is different in each interferogram. The recent deformation (Figure 7d) is modeled using two isotropic sources (Table 2). (e) The residuals and (f) the model prediction (the source locations are shown as black dots (Figure 7f)). Symbols are the same as in Figure 3. from the ground station, and problems with the Doppler ambiguity. The ERS interferogram does not reveal any deep deformation during 35 days of the dome building phase (track 468 spanning 6 September to 2 August 2008), although it is incoherent in the summit area (Figure 8c). No successful Envisat interferograms are possible because of a lack of acquisitions. Further complicating interpretation of the deformation at Chaitén is that the eastern part of the ALOS interferogram is incoherent, probably a result of ash deposition Figure 6. During several periods of volcanic activity between 2002 and 2007, InSAR shows no obvious deformation at Llaima, located in southern Chile: (a) February 2003 to January 2005 and (c) February 2007 to November (d, e, g, and h) Multiple interferograms show subsidence along the east flank related to activity in late 2007 and early 2008 (circled region). (b and f) Less obvious, the possible deformation on the east flank is interpreted to be related to a slow landslide. Symbols are the same as in Figure of 29

13 Figure 8. Chaitén volcano, located in southern Chile, erupted violently in early May (a) Coeruptive subsidence is captured in an ALOS interferogram that spans the onset of the eruption and several months following. (b) A preeruptive interferogram shows no precursory deformation up to 2 weeks prior to the first explosive eruption. (c) A posteruptive ERS interferogram is not coherent enough to show any significant deformation. (d) The coeruptive deformation is modeled with a collapsing dike (Table 2), drawn as a rectangle with the updip edge in bold, along with the model prediction. (e) The residual. Symbols are the same as in Figure 3. during the time span (Figure 8a). This leaves an ambiguity of the cause of deformation. [22] We are able to match the deformation pattern with either a collapsing dike, dip-slip faulting, or a combination of faulting and magma chamber deflation. The model prediction from the best fit model (a dike) is shown in Figure 8d. The residual between the data and models is rather large (4 5 cm; the residuals for the best fit model are shown in Figure 8e), perhaps reflecting additional unmodeled deformation sources. A swarm of moderate sized (Mw ) earthquakes that occurred in April May 2008 (NEIC catalog) might have contributed to the observed deformation field and reflects the complicated tectonic region surrounding Chaitén [Lange et al., 2008]. The volume changes required by the geodetic data, km 3 cannot account for the large volume of erupted material, km 3 [Lara et al., 2008]. Even accounting for the compressibility of the magma it is difficult to reconcile these differences. Given the short accent time and long time span of the observation it is conceivable that many deformation processes are super imposed in the single observation. It is also possible that some of the magma rose from depth and passed through the magmatic system without leaving a geodetic signal. 13 of 29

14 Figure 9. Located in southern Chile, Cerro Hudson has been inflating since at least 1993 [Pritchard and Simons, 2004b]. (a) An Envisat interferogram spanning April 2004 to April 2007 is largely decorrelated but suggests that inflation continues. (b) An ALOS interferogram spanning March 2007 to December 2007 is decorrelated due to the large baseline (1.7 km) and steep topography in this area. Symbols are the same as in Figure Cerro Hudson [23] Cerro Hudson is an ice-filled caldera in southern Chile that produced one of the largest eruptions of the 20th century in 1991 (Smithsonian Institution, Global volcanism report, available at Pritchard and Simons [2004b] found inflation of 5 cm/yr (perhaps decreasing in time between 1993 and 1999) which they modeled with a spherical point source at 5 km depth. Deformation at Cerro Hudson is difficult to observe because of decorrelation in both the C and L band data (Figure 9), but Envisat data spanning shows about 2 cm/yr of inflation (Figure 9a). The rate and shape of deformation is consistent with the declining rate of deformation and the depth of the magma chamber from Pritchard and Simons [2004b] Northern Volcanic Zone of South America [24] Previous studies of the Northern Volcanic Zone (NVZ) of South America were not possible with C band SAR satellites [e.g., Zebker et al., 2000; Stevens and Wadge, 2004], although Bonvalot et al. [2005] found deformation at Reventador, Ecuador. Ground surveys have also revealed deformation at two volcanoes in Colombia [Dvorak and Dzurisin, 1997], and Guagua Pichincha, Ecuador [Garcia-Aristizabal et al., 2007] (Table 1). Using data that covers the entire volcanic arc (Figure 2b), we find two volcanoes that have deformation during this time period; Tungurahua is deflating and Reventador shows subsiding lava flows. Mothes et al. [2008] used ALSO InSAR data to observe deformation at these volcanoes plus Antisana, but our data for Antisana were not definitive regarding the existence of deformation. The SAR scenes processed for this study are described in Table S Tungurahua [25] Inflation at Tungurahua is observed in several interferograms spanning from December 2006 to August Loss of coherence at the volcano summit obscures most of the deformation except in one image from December 2007 to May 2008 (Figure 10b). During this period eruptive activity occurred continually, characterized by strombolian eruptions that reached maximum altitudes of 5 14 km (Smithsonian Institution, Global volcanism report, available at The inflation is likely associated with the injection of magma into the volcanic edifice. There are several interferograms that partially constrain the evolution of this deformation episode. An image spanning from December 2006 to December 2007 shows no significant deformation (Figure 10a). Two partially overlapping interferograms, December 2007 to May 2008 and March 2008 to August 2008, show a decreasing trend in the magnitude of deformation (Figures 10b and 10c). The earlier 14 of 29

15 Figure 10. In central Ecuador, Tungurahua volcano is uplifting on its western flank. (a) An ALOS interferogram spanning December 2006 to December 2007 shows very little, if any, deformation, but (b) in the following 5 months (December 2007 to May 2008) 12 cm of uplift has occurred. (c) In an overlapping interferogram (March 2008 to August 2008) the magnitude and area of uplift has decreased by approximately half. A shallow dike (Table 2, see section for details) is used to model the deformation in Figure 10b. (d) The model prediction and projection of the model (black line). (e) The residual of the model (model data). Symbols are the same as in Figure 3. images show 12 cm line of sight inflation and the later image shows roughly half that amount. The areal extent of the deformation field is also smaller in the latter image. [26] The deformation pattern is elongated in the east-west direction and is modeled with a shallow dike that extends from the summit to the west 7 km (Figure 10d and Table 2). The opening dislocation is nearly vertical and reaches to within 100 m of the surface and extends to approximately 500 m depth. Without additional information about the seismicity or eruptive chronology it is difficult to speculate how realistic a dike model is for describing the current activity Reventador [27] A sector collapse at Reventador formed a horseshoe shaped crater which is now filled with an andesitic cone and pyroclastic and lava flow deposits (Smithsonian Institution, Global volcanism report, available at si.edu). An interferogram that spans January 2008 to January 2009 shows as much as 20 cm/yr of subsidence of these deposits, likely associated with cooling and compaction (Figure 11) Central America and Mexico [28] There have been several previous InSAR studies of selected volcanoes in Central America and Mexico. Null deformation was observed at a few volcanoes with C band data [Zebker et al., 2000], deformation due to surface loads was seen at Colima, Mexico, with persistent scatterer InSAR [Pinel et al., 2008], and deformation has been suggested at San Miguel Volcano, El Salvador by Schiek et al. [2008], but our data regarding deformation at this volcano is inconclusive. Ground surveys revealed deformation at three different volcanoes (Table 1) [Dvorak and Dzurisin, 1997]. 15 of 29

16 Figure 11. Reventador is located in northern Ecuador. An ALOS interferogram is coherent on lava flows inside the horseshoe shaped crater and reveals subsidence of some of those flows (circled). Symbols are the same as in Figure 3. [29] Our data between 2007 and 2008 reveal subsidence of a lava flow at the base of Parícutin cinder cone, but show no obvious subsurface volcanic processes occurring at any of the 113 volcanic centers. The lack of volcano deformation in Central America is somewhat surprising considering the level of eruptive activity (Table 3), but may be related to the short time window of the observations and the relatively poor coherence in the heavily vegetated region. The ALOS scenes processed for this study are described in Table S3. [30] Parícutin is the famous volcano born from a corn field [Pioli et al., 2008] in central Mexico in The eruption lasted nearly 10 years and produced voluminous lava deposits around the central vent. A total eruptive volume of roughly 1.4 km 3 [Stasiuk et al., 1993] is deposited over a circular area of 2.5 km radius, giving an average flow thickness of 70 m. The nonuniform thickness of the flows means that the deposits are likely to be much thicker in some areas. The lava flow deposits are believed to be subsiding due to thermal contraction and compaction, and the subsiding Figure 12. An ALOS interferogram spanning August 2007 to May 2008 shows subsidence of the lava flow at Parícutin in central Mexico. The lava was erupted between 1943 and 1952 and continues to show subsidence in the thickest part of the flow. Symbols are the same as in Figure 3. area most likely indicates the thickest part of the flow (Figure 12). [31] Lu et al. [2005] explored subsiding lava flows inside Okmok Caldera and concluded that thermoelastic contraction of the flows was the largest contributor to the deformation. The subsiding flows at Okmok are comparable in age to the Parícutin deposits, the Okmok flows were emplaced in 1945 and The subsidence rate at Okmok, 1.5 cm/yr, is considerably less than the cm/yr observed at Parícutin. The increased subsidence rate at Parícutin is likely the result of differences in the deposit thicknesses. At Table 3. Interarc Comparison of Volcano Deformation Volcanic Arc Number of Volcanoes a Number With Historic Eruptions InSAR Time Span Number of Volcanoes Deforming Central America 113 b Caribbean , c , 1 Northern Andes 35 b Central Andes Southern Andes 63 b , c , 7 Austral Andes Alaska/Aleutians d c 15 a Smithsonian Institution (Global volcanism report, available at b Only includes mainland volcanoes. c Only a portion of the arc is covered in this time interval. d Based on Lu et al. [2007, and references therein]. 16 of 29

17 similar to that found during an earlier time interval with ERS data [Wadge et al., 2006]. [33] Caution should be taken at volcanic centers with active deposition of material or dome growth. Changes to the topography from these processes can manifest as apparent ground deformation if care is not taken to use up-to-date DEMs. For instance the interferogram in Figure 13 shows 7 cm of line of sight uplift of the recent deposits from SHV, but also a large amount of apparent motion at the summit of the volcano (circled region), likely associated with changes in the topography of the active dome. 4. Global Compilation Figure 13. Soufrière Hills Volcano on Montserrat island in the northern part of the Caribbean Islands has been erupting for more than a decade. A 46 day ALOS interferogram shows deformation of recent pyroclastic deposits on the east flank, similar to the deformation observed in a 35 day ERS interferogram in 1999 [Wadge et al., 2006]. Apparent deformation at the summit (circled) may be the result of DEM errors associated with topography changes caused by dome growth. Symbols are the same as in Figure 3. Okmok the flow deposits are 25 m thick, while at Parícutin, although the thickness is not well constrained, it likely exceeds 70 m Caribbean [32] The ALOS data availability for this region is much sparser than the other areas in Latin America. We process scenes from 2006 to 2008, but some interferograms only span a few months because of the limited data (Table S4). Temporal decorrelation in the Caribbean may be the largest obstacle to using InSAR in this region. Images with a temporal baseline of about 1 year and longer are almost completely decorrelated. We only find volcano deformation associated with Soufrière Hills Volcano (SHV) on Monserrat Island. Several images show potential deformation associated with volcanic deposits on the flanks of SHV (Figure 13), [34] We have attempted to compile all known deforming volcanoes and some of the properties of their volcano deforming events (duration, magnitude, and relation to eruptive processes) in Table 1 and Figure 14. We have not attempted to catalog every volcano deformation event (Kilauea only would consume the entire table), but try to sample the range of different types of deformation events. We have mixed together many processes that can cause surface deformation, because it is not simple to separate these into different categories. The types of phenomena that can cause ground displacements include: hydrothermal circulation, subsidence of volcanic deposits, loading by deposits, loading and unloading by snow and ice, eruptive conduit processes, degassing, thermal and phase changes in magma, dome growth/collapse, dike emplacement and others. We categorize the observed deformation into seven general categories; eruptive, intereruptive, preeruptive, deformation of a volcanic deposit, geothermal, groundwater and flank deformation. We also indicate whether the observation is aliased or not. Any observation that does not show temporal evolution of the event is considered to be aliased. Figure 14 shows that these processes occur at a spectrum of different time scales and many occur at multiple time scales (from seconds to centuries), and can occur either with or without eruption. It should be noted that tilt events are included in Figure 14 for completeness and that the tilt magnitude (on the right axis) is not meant to correspond to an equivalent displacement rate (given on the right axis), instead tilt is scaled to fit within Figure 14. [35] On a world wide basis, it is not clear if the duration or magnitude of deformation events provide an indication of the potential for future 17 of 29

18 Figure 14. Duration of deformation versus rate calculated for a variety of volcanoes around the world using the information from Table 1. Even if deformation only lasts a few minutes, the rate is calculated in units per year. Tilt observations use the scale on the right side of the graph, while other types of deformation observations (GPS, InSAR, EDM, etc.) use the scale on the left side. Observations that include coeruptive deformation have open symbols. Very few observations are taken frequently enough to be considered nontemporally aliased. eruption. The only obvious trend to be drawn from Figure 14 is that the magnitude of deformation appears to decrease with the duration of the deformation event. This is not a surprising trend to find in this type of figure and it should not be interpreted as having any significant implication for volcano deformation. At the short time scale end of Figure 14, we lack observations of small magnitude deformation because current technology does not allow for measurements of such minute changes. At the long time scale end of Figure 14 there is an upper limit for how long any deformation rate can be sustained. 5. Discussion 5.1. Comparison of C and L Band [36] Because of the longer wavelength, we expect L band data to have higher coherence and fringe visibility than C band, and on a regional scale this seems to be true in the southern Andes (compare Figure 2c to Figure 1 of Pritchard and Simons [2004b]). However, it is hard to directly compare the C and L band coherence because the baselines and time periods are not equal. At a few volcanoes in the southern Andes, we have processed both C and L band data so a qualitative comparison can be made. In some areas, coherence at C and L bands is similar, for example; Laguna del Maule (Figure 3), Copahue (Figure 4), Lonquimay (Figure 5), and Llaima (Figure 6). Regions of coherence are larger in L band than C band in interferograms at Cordón Caulle and Chaitén (Figures 7 and 8). At Cerro Hudson, the C band interferogram is more coherent, probably because the large baseline of the ALOS interferogram causes topographic decorrelation in the areas of high relief (Figure 9). This decorrelation can be reduced by using a DEM during image coregistration [Yun et al., 2008]. [37] At some volcanoes, deformation is only detected with C band data (Copahue, Hudson, maybe Lonquimay), because of the short time span 18 of 29

19 or large baseline of the ALOS data. Although a direct comparison at Chaitén is not possible because C band data spanning the eruption does not exist, we get a sense of the data quality from previous C band interferograms that show decorrelation over 1 year. Deformation is detected with both C and L band at Llaima and Cordon Caulle, but L band has superior coherence, revealing more of the deformation pattern. [38] L band is more sensitive to ionospheric effects than C band, and we observe such effects in the Northern, Central and Southern Andes (see Appendix A). The L band data has higher spatial resolution than the available C band data, and seems to reveal errors in the DEM (see Appendix A). [39] In the northern Andes, coherence seems to be highest at high elevations where there is no snow and there is less vegetation (Figure 2b). A 46 day repeat provides coherent interferograms in the vegetated lowlands when the spatial baseline is small and/or when spatial averaging is applied. For longer time spans the lower-elevation regions become decorrelated. In the southern Andes, it is necessary to avoid austral winter, and even then coherence is the lowest in the foothills western side of Cordillera (Figure 2c) which receives the most precipitation and has the most vegetation. The coherence in the vegetated southern Andes depends on baseline length, even in low-lying areas; for example, compare the region west of Cordón Caulle in Figure 2c (path 119, perpendicular baseline 1.8 km) which is more coherent than the regions at similar longitude to the north (path 118, perpendicular baseline 2.6 km), even though both interferograms span 1 year. In the Caribbean (Figure 2d), the coherence is the lowest of any region we studied, and short time period interferograms with small baselines are essential. Vegetation and terrain cause the most problems, with the inland areas of most of the islands becoming decorrelated the fastest. In the northern Andes, Central America, and the Caribbean the L band coherence is clearly superior to C band results from these areas [e.g., Zebker et al., 2000; Wadge et al., 2006]. [40] With good baseline control (<250 m), we conclude that at L band sufficient coherence is maintained for solid earth applications in vegetated areas at 46 days, most of the time. We do not have enough data yet to quantify most of the time, but we know it is not 100%. A 7 day repeat interferogram from UAVSAR (L band) in San Francisco is incoherent in vegetated areas presumably because one scene was acquired right after a rain event (S. Hensley, personal communication, 2009). R. Ahmed et al. (A survey of temporal decorrelation from spaceborne L-band repeat-pass InSAR, submitted to Remote Sensing of Environment, 2009) show that wind/rain causes loss of coherence at L band in 1 day repeats with SIR-C data (from the Space Shuttle) over the eastern U.S. An unanswered question is how quickly coherence is recovered after a rain/wind storm. [41] We know that 46 day interferograms do not work in most snow or ice areas (e.g., the southern and austral Andes during austral winter). A significant fraction of volcanoes around the world have snow cover for at least part of the year. A short repeat time will maximize the chance that there will not be a precipitation or melting event that might diminish coherence. We do not know how this coherence diminishes with time (e.g., it depends on the quantity and properties of the snow), but a 7 day repeat interferogram at Mt. St. Helens from UAVSAR was incoherent when there had been a snow event between acquisitions (S. Hensley, personal communication, 2009) Interarc Comparison and Global Synthesis [42] Throughout Latin America, we find that most deforming volcanoes are not erupting and we do not observe deformation at most of the erupting volcanoes. How representative are these results? A comparison between Latin America and the Alaska/Aleutian arc (almost completely surveyed by InSAR, [e.g., Lu et al., 2007]) is illustrative of interarc variations (Table 3). The Aleutian arc has about the same number of historic eruptions (about 40) and actively deforming volcanoes (about 15) as the central and southern Andes combined, although there are many more volcanoes in the central and southern Andes (91 in the Aleutians and 132 in the central and southern Andes of Holocene age (Smithsonian Institution, Global volcanism report, available at and about 2500 total [e.g., Gonzalez-Ferran, 1995]). It is perhaps not fair to group the central and southern Andes together as these arcs have many compositional and structural differences (e.g., the crustal thickness approaches 70 km in the CVZ but is about 40 km in the SVZ; and volcanism is more basaltic in the SVZ (Smithsonian Institution, Global volcanism report, available at si.edu)). For example, the magnitude of deformation (of magmatic and/or hydrothermal origin) is 19 of 29

20 larger in the SVZ than in the CVZ. While the magnitude of deformation observed in the SVZ for 5 of the 7 deformation areas is greater than 5 cm/yr, in the CVZ no volcanic deformation has yet been observed to exceed 5 cm/yr. Deformation caused by surface processes, such as the subsidence of recent volcanic deposits at Lonquimay, Parícutin, and Reventador, is not unusual in volcanic arcs [e.g., Pritchard and Simons, 2004a; Lu et al., 2005; Masterlark et al., 2006; Whelley et al., 2008], but the activity at Llaima defies simple explanation. [43] One goal of this study is to assess the range and variability of deformation styles observed worldwide. It is clear that volcano deformation events occur frequently without eruption and are usually short lived. In the CVZ and SVZ alone, we have observed deformation to start or cease at 7 of the 15 areas of deformation (Chaitén, Cordón Caulle, Llaima, Laguna del Maule, Ticsani, Hualca Hualca, and Lazufre) during the few years of InSAR observation. On the other hand, some areas have been continuously deforming for decades, and even millennia, without eruption [e.g., Newhall and Dzurisin, 1988]. Centuries long patterns of deformation alternating between uplift and subsidence at Campi Flegrei [Cinque et al., 1985], Yellowstone [Pierce et al., 2002], and possibly Long Valley [Reid, 1992], have yielded net uplift of only m. Uplift at Socorro has been continuous between at least , but proposed evidence for continuous uplift lasting 10,000 years or more is dubious [Fialko et al., 2001; Finnegan and Pritchard, 2009] Optimum Observation Interval for Volcano Deformation [44] There are two key questions related to the optimum interval for making deformation observations, which translates into a satellite repeat overflight time interval: [45] 1. How frequently do observations need to be made to observe deformation precursors before eruptions? There is incomplete data to answer this question in a global sense (and different volcanic systems seem to have different behaviors), but there is anecdotal evidence that precursors occur before eruptions on time scales of hours to weeks. Klein [1984, p. 3059] notes that at Kilauea time scales of less than 20 days are important: Tilt level is an eruption precursor significant to better than 99.9% when averaged over any interval from 1 to 20 days. For basaltic systems like Kilauea, there might be an absence of deformation immediately before an eruption as a magma chamber asymptotically approaches some fixed pressure [Dvorak and Okamura, 1987]. Yet, while a decrease in deformation rate was observed before the 2008 eruption of Okmok volcano, Alaska, similar to that at Kilauea, continuous GPS measurements still found precursory deformation in the weeks/months before the eruption [Larsen et al., 2009]. The hours to weeks time scale for precursors also seems to hold at Piton de la Fournaise [Collombet et al., 2003], Sakurajima, Suwanosejima and Semeru [Iguchi et al., 2008], Etna [Bonaccorso et al., 2002], Soufrière Hills [Voight et al., 1999] and Mt. St. Helens [Swanson et al., 1983]. [46] Seismic precursors before eruptions have been studied at more volcanoes than deformation precursors. For example, a global compilation of 191 swarms associated with eruptions have a mean time scale of 8 days although this time scale includes precursory, coeruptive and posteruptive swarms [Benoit and McNutt, 1996]. There is anecdotal evidence, at Long Valley, California, for example, that [Sorey et al., 2003, p. 171] episodes of accelerated deformation generally precede increases in earthquake activity by several weeks to months. Therefore, seismic information seems to indicate that deformation observations on a time scale of days to weeks could be useful for eruption precursors. [47] Sometimes there is an uptick in deformation starting months to years before an eruption (e.g., Sierra Negra, Galapagos [Chadwick et al., 2006], and Okmok, AK [Fournier et al., 2009]), but eruptions that are triggered by a dike intrusion may only start deforming 24 h or less before an eruption (e.g., Hekla, Iceland, 2000 [Sigmundsson, 2006], and Sierra Negra, 2005 [Geist et al., 2008]; Chaitén, Chile, was dormant for 9400 years until a few days before the 2008 eruption (this work)). Thus, some basaltic and more silicic volcanoes require deformation observations every minute (or hour) to catch dike intrusions. [48] 2. How frequently do observations need to be made to understand the physical processes occurring at a particular volcano, independent of eruption? Another way of asking this question is what is the time scale for variations in volcano deformation? In the Aleutians, more than 60% of eruptions lasted less than 24 days during the past decade (Z. Lu, personal communication, 2009), which implies that observations on a daily to weekly basis are necessary in order to monitor deformation 20 of 29

21 during an eruptive crisis. Globally (Table 1), it is clear that some volcanoes change their rate and pattern of deformation on a daily to weekly time scale. It is worth noting that observations that span months-years might measure some deformation, but are sometimes temporally aliased and not capturing all physical processes involved. [49] The danger of aliasing the deformation signal is exemplified very well in the case of the 2006 eruption of Augustine Volcano, Alaska. The common mode for InSAR analysis in Alaska (and other snowy areas) is to use images acquired only in the summer months [e.g., Lu et al., 2007] due to snow cover and the long repeat intervals of the existing SAR satellites. The cumulative deformation at Augustine from the summers bracketing the eruption would show no deformation of the volcano, even though the GPS record shows a dynamic deformation history during the yearlong interval [Cervelli et al., 2010]. Since deformation associated with various volcanic processes occurs at different temporal and spatial scales the problem of aliasing should always be considered in regions where very little is known about the volcanoes. The lack of deformation observed during several volcanic eruptions (for example, Nevados de Chillán, Villaricca and Llaima (this work); Shishaldin [Moran et al., 2006]; Korovin, Pavlof, Cleveland, Chiginagak, Augustine, and Veniaminof, Alaska [Lu et al., 2007]; Kliuchevskoi, Sheveluch, and Bezymianny, Kamchatka [Pritchard and Simons, 2004c]; Lascar, Irruputuncu, Aracar, and Sabancaya, central Andes [Pritchard and Simons, 2004a]) is at least partly explained by temporal aliasing. 6. Conclusions [50] 1. Our global compilation (and previous work [e.g., Dzurisin, 2003]) indicates that the deformation episodes at dormant volcanoes can be short lived (lasting weeks to months), and commonly occur without eruption. Geochemical analysis also infers that there might be several small intrusions spanning the decades to centuries before an eruption [e.g., Zellmer et al., 2003], so that any eruption related to current volcano deformation might be many years from now. On the other hand, there are several examples where volcano deformation has been used to indicate that an eruption might occur within a few months [e.g., Larsen et al., 2009] or even predict an eruption a few minutes to days before eruption [Swanson et al., 1983; Klein, 1984; Voight et al., 1999; Bonaccorso et al., 2002; Collombet et al., 2003; Iguchi et al., 2008]. Deformation precursors before eruption might occur only days before an eruption at a volcano that has been dormant for thousands of years (e.g., Chaitén). [51] 2. In terms of data quality, both C band and L band observations have proved useful when combined together to maximize spatial and temporal coverage of the SVZ. Given previous studies [Rosen et al., 1996], it is not surprising that we conclude that L band data quality is superior to available C band data in highly vegetated areas (especially Central America, the northern Andes and the Caribbean). The repeat interval necessary to maintain L band coherence is regionally variable in Latin America depending on distance to the ocean, elevation, and type of vegetation, but for volcanic applications, a 46 day repeat seems to maintain coherence as long as there is good baseline control. [52] 3. From our comprehensive survey of all volcanoes in Latin America (minus the central Andes), we document 11 centers of volcanic deformation, including discovery of deformation in 4 locations. It seems that there are significant variations in deformation characteristics between volcanic arcs. For example, the few deforming volcanoes found in the northern Andes, the Caribbean and Central America seems contradictory to the high level of eruptive activity in these regions (Table 3). There also seems to be a low number of deforming volcanoes relative to eruptions in Kamchatka (P. R. Lundgren, personal communication, 2009). While 1 year of observations may not be sufficient to find all deforming volcanoes in these arcs, our study in the southern Andes over a similar time from ALOS found four areas of deformation. It is possible that no deformation is recorded because magma may not refill every chamber before or after eruption, or might not even be stored in a shallow chamber before erupting [Dzurisin, 2003]. It seems that the nature of the magma plumbing in the northern Andes, Kamchatka, and Central America is different from the central and southern Andes, perhaps because of the tectonic regime, magmatic flux, magma or crustal composition or some other factor. Nonetheless, based on continuous GPS observations at a few Alaskan volcanoes during eruptions (Okmok and Augustine), we suspect that the frequency of InSAR observations has not been sufficient to measure coeruptive deformation. 21 of 29

22 geodetic stations) are an important complement to InSAR, especially in cases where magma movements are undetectable by the satellite (because the magma moves without deforming the surface, the deformation time scale is short, or the deformation is too small). Veniaminof, Alaska and Suwanose-jima, Japan are examples where deformation was seen with GPS instruments [Iguchi et al., 2008; Fournier et al., 2009], but not with InSAR [Lu et al., 2007; Aoki et al., 2008]. [54] 4. Globally, volcanoes deform across a variety of time scales (from seconds to centuries) with no apparent critical observation time scale, other than to note that observations should be made as frequently as possible. For dike intrusions, the necessary time scale is seconds to minutes, which is Figure A1. ALOS interferogram from ascending path 109 (acquired around 2200 local time) spanning December 2006 to December While the fringes could be caused by errors in the orbital parameters, we have only seen similar distortions in this area. Through pair-wise logic of making multiple interferograms with each date, we have determined that the data on 6 December 2006 were corrupted, possible by a large phase change in the ionosphere. Unlike other examples in the central and southern Andes, this phase distortion does not have associated streaks of decorrelation. Symbols are the same as in Figure 3. [53] Even when we do observe coeruptive deformation, the currently available repeat intervals leave significant gaps in observation that leave many processes poorly understood. For example, InSAR observations of the 2008 Chaitén eruption are ambiguous as to the nature of deformation (diking, faulting, or some complex combination) and cannot account for the observed eruptive volume, probably because of temporal aliasing of the deformation signal. Future InSAR satellites with more frequent revisits (like NASA s DESDynI) or constellations of satellites are necessary to truly determine the existence or nonexistence of coeruptive deformation (as well as the nature of this deformation and the existence or nonexistence of precursory deformation). In any case, ground-based sensors (particularly gravimeters, and continuous Figure A2. ALOS interferogram from ascending path 105 (acquired around 2200 local time) spanning August 2007 to February Fringes in the northern part of the interferogram are from the M w 7.7 earthquake on 14 November 2007, but the fringes in the southern part of the image are associated with streaks of decorrelation and are presumably of ionospheric origin. 22 of 29

23 Figure A3. ALOS interferogram from ascending path 114 (acquired around 2200 local time) spanning December 2007 to March While some fringes in this interferogram may be caused by errors in the orbital parameters, the fringes in the southern part of the image are associated with streaks of decorrelation and are presumed to be of ionospheric origin. The black line shows the Chile-Argentina border. Symbols are the same as in Figure 3. characteristics seem to vary across the regions, with streaks of decorrelation and phase perturbation in the central and southern Andes (similar to those seen in polar regions [e.g., Gray et al., 2000]) and only long spatial wavelength phase perturbations observed in the northern Andes. No ionospheric artifacts have been observed in Central America or the Caribbean arcs. These potential ionospheric disturbances affect a small proportion of the acquired SAR images. For example, in the northern Andes only 2 of more than 30 processed images are negatively affected by the ionosphere. While ionospheric effects are thought to be more important at L band than C band, in the central Andes, we have seen phase distortions with a similar orientation to those in Figure A2 with C band radar data collected during a similar time of day [Pritchard, 2003]. [56] We have several examples where we think that errors in the 3 arcsec (90 m/pixel) SRTM DEM create a phase change in the ALOS interferograms, particularly in Ecuador (Figures A4 and A5). There are several potential error sources when using the 3 arcsec SRTM DEM with ALOS data. First, the ALOS fine beam pixel size at full resolution is about 10 m, so to get the interferogram resolution outside the realm of currently available or nearfuture satellite InSAR observations. Appendix A: Quality of L Band Interferograms [55] We see linear phase distortions oblique to the satellite track in a few interferograms in the northern, central and southern Andes that we attribute to ionospheric effects (Figures A1 A3). Because of the timing of acquisitions (between 2000 and 0200 local time, when ionospheric scintillation is most intense), the location within about 20 of the magnetic equator (where these scintillations are most intense outside of the polar regions), and the fact that the distortions are only associated with single SAR images, we suspect that the distortions are of ionospheric origin [e.g., Xu et al., 2004]. The Figure A4. ALOS interferogram from ascending path 109 spanning 46 days (8 June 2007 to 24 July 2007). The phase change on the west side of Imbabura volcano, Ecuador, is seen in several interferograms, but we suspect that the signature is due to a DEM error. The perpendicular baseline is about 300 m, which means that a DEM error of 200 m is necessary to cause 11.8 cm of phase change (corresponding to 2p radians); that is, the ambiguity height is about 200 m. 23 of 29

24 program. Thanks to F. Amelung and an anonymous reviewer for helpful comments that greatly improved the manuscript. The GMT program was used to create Figures 1 13 and A1 A5 [Wessel and Smith, 1998]. References Figure A5. ALOS interferogram from ascending path 110, spanning 23 December 2006 to 26 December 2007 (perpendicular baseline of about 100 m, with an ambiguity height of about 600 m). In several interferograms we see a phase difference at Cerro Altar and at several nearby mountains that are facing toward or away from the radar, probably due to a DEM error. and DEM resolution nearly comparable, we must look down (or spatially average) the interferogram and/or interpolate the DEM, and there might be errors in the interpolation. A second consideration is that the SRTM DEM was created using C band SAR which might scatter higher in the vegetation canopy than the L band ALOS SAR, yielding the wrong altitude. Given the phase differences observed (up to two fringes in Figure A4) and the ambiguity heights (200 to 600 m for the examples shown), this would require a DEM error of hundreds of meters, which seems too large to be due to vegetation. We suspect instead that these are just small-scale errors in the STRM DEM which are significantly outside the globally estimated absolute vertical accuracy of about 9 m, but which is known to be less accurate in regions of steep topography like the Andes [Farr et al., 2007]. Acknowledgments [57] We thank H. Moreno, A. Pavez, L. Lara, P. R. Lundgren, S. Hensley, P. Siquiera, Z. Lu, M. Poland, and C. Wicks for helpful discussions. N. Williams, E. Knight, R. B. Lohman, and B. Barnhart assisted with ordering and processing the data. ERS and Envisat data were acquired under an ESA Category 1 project. ALOS data were provided by JAXA and with support from ASF and NASA. This work was partly supported by NASA grant NNX08AT02G issued through the Science Mission Directorate s Earth Science Division. S.N.R. was partly supported by the NASA/New York Space Grant summer intern Amelung, F., and S. Day (2002), InSAR observations of the 1995 Fogo, Cape Verde, eruption: Implications for the effects of collapse events upon island volcanoes, Geophys. Res. Lett., 29(12), 1606, doi: /2001gl Amelung, F., S. Jónsson, H. Zebker, and P. Segall (2000a), Widespread uplift and trap-door faulting on Galápagos volcanoes observed with radar interferometry, Nature, 407, , doi: / Amelung, F., C. Oppenheimer, P. Segall, and H. Zebker (2000b), Ground deformation near Gada Ale volcano, Afar, observed by radar interferometry, Geophys. Res. Lett., 27, , doi: /2000gl Amelung, F., S. H. Yun, T. R. Walter, P. Segall, and S. W. Kim (2007), Stress control of deep rift intrusion at Mauna Loa volcano, Hawaii, Science, 316, , doi: / science Anthes, R. A., et al. (2007), Earth Science and Applications From Space: National Imperatives for the Next Decade and Beyond, Natl. Acad. Press, Washington, D. C. Aoki, Y., J. Oikawa, M. Furuya, and M. Iguchi (2008), Ground deformation of Suwanose-jimavolcanoasviewedfrom ALOS/PALSAR InSAR, Eos Trans. AGU, 89(53), Fall Meet. Suppl., Abstract V11B Baer, G., Y. Hamiel, G. Shamir, and R. Nof (2008), Evolution of a magma-driven earthquake swarm and triggering of the nearby Oldoinyo Lengai eruption, as resolved by InSAR, ground observations and elastic modeling, East African Rift, 2007, Earth Planet. Sci. Lett., 272, , doi: / j.epsl Banks, N. G., C. Carvajal, H. Mora, and E. Tryggvason (1990), Deformation monitoring at Nevado del Ruiz, Colombia October 1985-March 1988, J. Volcanol. Geotherm. Res., 41, , doi: / (90)90092-t. Bartel, B. A., M. W. Hamburger, C. M. Meertens, A. R. Lowry, and E. Corpuz (2003), Dynamics of active magmatic and hydrothermal systems at Taal Volcano, Philippines, from continuous GPS measurements, J. Geophys. Res., 108(B10), 2475, doi: /2002jb Beauducel, B., F. Cornet, E. Suhanto, T. Duquesnoy, and M. Kasser (2000), Constraints on magma flux from displacement data at Merapi volcano, Java, J. Geophys. Res., 105, , doi: /1999jb Benoit, J. P., and S. R. McNutt (1996), Global volcanic earthquake swarm database , U.S. Geol. Surv. Open File Rep., 96 69, 334 pp. Biggs, J., E. Anthony, and C. Ebinger (2009), Multiple inflation and deflation events at Kenyan volcanoes, East African Rift, Geology, 37, , doi: /g30133a.1. Bonaccorso, A., M. Aloisi, and M. Mattia (2002), Dike emplacement forerunning the Etna July 2001 eruption modeled through continuous tilt and GPS data, Geophys. Res. Lett., 29(13), 1624, doi: /2001gl Bonforte, A., A. Bonaccorso, F. Guglielmino, M. Palano, and G. Puglisi (2008), Feeding system and magma storage beneath Mt. Etna as revealed by recent inflation/deflation cycles, J. Geophys. Res., 113, B05406, doi: /2007jb of 29