Introduction

Abstract

Measurements of ground deformation have been used by geodesists to investigate natural and man-made phenomena since the 19th century. Since the early 1990s, Interferometric Synthetic Aperture Radar (InSAR) has changed the way that geodesists view the Earth, facilitating deformation monitoring in remote and inaccessible locations, and providing insight into spatially and temporally complex processes. Unsurprisingly, InSAR observations have pushed forth a new era in volcano deformation studies from both a monitoring and research perspective. InSAR observations of volcanic ground deformation enable the detection of subsurface magma, and therefore play a key role in eruptive hazard assessments. However, these observations have also shed new light on all aspects of volcanic behaviour, contributing to our understanding of the eruption cycle, hydrothermal systems and the formation of continental crust. One of the most significant achievements of volcano InSAR has been the identification and characterisation of magma inputs into the crust that would have otherwise gone unnoticed (e.g. Lu and Dzurisin 2014). Such observations of magma movement and other slow, long-term processes are crucial in determining how magmatic plumbing develops and evolves beneath a volcano (Dzurisin 2003; Cashman and Biggs 2014), but the mechanism responsible is often ambiguous.

Keywords

These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.

Measurements of ground deformation have been used by geodesists to investigate natural and man-made phenomena since the 19th century. Since the early 1990s, Interferometric Synthetic Aperture Radar (InSAR) has changed the way that geodesists view the Earth, facilitating deformation monitoring in remote and inaccessible locations, and providing insight into spatially and temporally complex processes. Unsurprisingly, InSAR observations have pushed forth a new era in volcano deformation studies from both a monitoring and research perspective.

InSAR observations of volcanic ground deformation enable the detection of subsurface magma, and therefore play a key role in eruptive hazard assessments. However, these observations have also shed new light on all aspects of volcanic behaviour, contributing to our understanding of the eruption cycle, hydrothermal systems and the formation of continental crust. One of the most significant achievements of volcano InSAR has been the identification and characterisation of magma inputs into the crust that would have otherwise gone unnoticed (e.g. Lu and Dzurisin 2014). Such observations of magma movement and other slow, long-term processes are crucial in determining how magmatic plumbing develops and evolves beneath a volcano (Dzurisin 2003; Cashman and Biggs 2014), but the mechanism responsible is often ambiguous.

1.1 Motivation

The application of InSAR in volcanic environments is often challenging, and in order for these observations to be a useful tool in volcanology we require (1) better ways of detecting low magnitude deformation signals and (2) physically reasonable ways of interpreting commonly observed deformation signals. A significant number of volcanoes have exhibited periods of subsidence over decadal time-scales (Fig. 1.1), but the mechanisms driving such deformation remain poorly understood. Whilst long-term volcanic subsidence is not immediately linked to eruption, it can be used to provide insight into magma storage conditions and longer-term processes related to tectonic setting, hydrothermal systems, and the building of crustal plutons. The Cascades Volcanic Arc contains three volcanoes that have exhibited subsidence for>10 years, and therefore presents an ideal place to add to observations of, and constrain the driving mechanisms behind, long-term volcanic subsidence.

In this thesis I therefore focus on the application of InSAR to the Cascades Volcanic Arc and the processes that cause long-term volcanic subsidence. In this introduction, I first provide a brief description of InSAR (Sect. 1.1) and its role in volcanology (Sect. 1.2), before introducing the Cascades Volcanic Arc (Sect. 1.3). In Chaps. 2 and 3, I address the challenges involved in making InSAR measurements in this setting. In Chaps. 4 and 5, I then use two examples from the southern Cascades to add to the known examples of volcanic subsidence, and develop a new modelling approach to characterise the long-term geodetic response to cooling magma intrusions.

Fig. 1.1

Global map of subsiding volcanoes. Small triangles mark the Holocene volcanoes from the global volcanism database of the Smithsonian Institution. White triangles mark volcanoes that have subsided as observed using InSAR. Black triangles mark volcanoes where subsidence is attributed to cooling and crystallisation of magmatic material, which is the focus of Chap. 4

1.2 InSAR

InSAR is an active satellite-based remote sensing tool capable of measuring mm-scale ground displacements at a spatial resolution of <5–10 s meters over large (100 km-scale) regions. InSAR has captured ground deformation resulting from the earthquake cycle (Wright 2002); aquifer depletion (Galloway and Burbey 2011); landslides (Ye et al. 2004); mining collapse (Plattner et al. 2010); underground storage of CO$_2$ (Vasco et al. 2010); cryospheric processes (Goldstein et al. 1993); and the eruptive cycle (Pinel et al. 2014). Unlike ground-based geodetic techniques such as leveling or GPS campaigns, InSAR imagery is capable of capturing the full extent and spatial complexity of the deformation field. The technique is cost effective, replacing long field campaigns with repeat satellite imagery available in increasingly small time intervals (Table 1.1). The remote nature of InSAR also enables measurements to be made in locations that are inaccessible, remote, or hazardous (e.g. Wauthier et al. 2013). The success of InSAR in measuring many processes, notably earthquakes and sudden eruptions, has also lain in the ability to design experiments in retrospect due to the semi-regular acquisition schedule of satellites.

Here we focus on the use of Synthetic Aperture Radar (SAR) imagery to measure displacements of Earth’s surface due to subsurface volcanic processes, but advances have also been made in using SAR derived products to measure topographic change (Wadge et al. 2010), the emplacement of lava flows (Dietterich et al. 2012), and shallow-seated flank instabilities (Ebmeier et al. 2014). This thesis utilises a large archive of InSAR data acquired for the Cascade volcanoes between 1992–2011 by European Space Agency (ESA) satellites ERS-1/2 and ENVISAT, and the Japanese Aerospace Exploration Authority (JAXA) satellite ALOS. These systems are no longer operational, but lay the basis for future studies using the 5 SAR satellite systems currently in orbit (Table 1.1).

Table 1.1 SAR satellite systems used in this study and that are currently operational

Several reviews describe SAR and SAR interferometry in detail (e.g. Massonnet and Feigl 1998; Rosen et al. 2000; Hanssen 2001; Bürgmann et al. 2000; Simons and Rosen 2007), and standard processing packages are now available on both an open access (e.g. ROI_PAC: Rosen et al. 2004) and commercial (e.g. GAMMA: Werner et al. 2000) basis. Here I provide a brief overview of SAR, including the steps used to produce high resolution radar images. I describe how SAR images are then combined to produce interferograms, and introduce the limitations of using this method in volcanic settings.

Synthetic Aperture Radar

SAR satellites travel in a quasi-polar orbit that is either ascending (south - north) or descending (north - south). Imagery is acquired in a side-looking geometry to eliminate ambiguities between points on the left and right of the satellite (Fig. 1.2). The satellite looks down to the Earth’s surface at an oblique angle ($\sim$10–50$^\circ$), recording information in the satellite line of sight (LOS). Like conventional radar, SAR operates by illuminating the target (in this case Earth’s surface) with microwave pulses (frequency 0.3–300 GHz). The microwave pulses are reflected from scatterers on Earth’s surface and received back at the antenna as a set of complex numbers recording both an amplitude and a phase. The amplitude is dependent upon the reflectivity of the Earth’s surface, including terrain slope and surface roughness, whereas the phase records many different effects, and essentially varies randomly between 0–360$^\circ$ (Massonnet and Feigl 1998).

A focussed SAR image is produced by applying range (across track) and azimuth (along track) compression to focus the raw radar echoes (Elachi 1988; Curlander and McDonough 1991). Range resolution of the radar image, $\Delta R_g$, is limited by the radar pulse duration, $\tau$, according to:

$$\begin{aligned} \Delta R_g = \frac{c\tau }{2 sin \theta }, \end{aligned}$$

(1.1)

where c is the speed of light and $\theta$ is the incidence angle. To achieve a resolution of <5–10s m requires very short pulse durations (50 ns), but the power supply of such a short pulse is not large enough to give the returned pulse a sufficient signal-to-noise ratio. Instead, a chirped (frequency modulated) pulse is used, and the required range resolution is achieved using a technique called range compression, described by Hanssen (2001). The range compressed signal with bandwidth $B_r$ has a shorter effective pulse length, $\tau _e$, equal to 1/$B_r$. The resolution then becomes:

$$\begin{aligned} \Delta R_r = \frac{c}{2 B_R sin \theta }. \end{aligned}$$

(1.2)

Focussing in the azimuth direction utilises the synthetic aperture of the radar, achieved by having a moving antenna. Azimuth resolution, $\Delta A_g$, is limited by the antenna length, $L_a$, and the slant range of the satellite orbit, $R_m$:

$$\begin{aligned} \Delta A_g = R_m \frac{\lambda }{L_a}, \end{aligned}$$

(1.3)

where $\lambda$ is the satellite wavelength (for values see Table 1.1). To achieve a resolution comparable to that in the across track direction would require an antenna on the order of 5 km for a satellite orbiting at $\sim$800 km (Bürgmann et al. 2000). Instead, a synthetic aperture is used by combining a series of consecutive radar pulses that illuminate the same point on the ground. The echoes from a given point are isolated using the two-way travel time of the radar pulse, which defines a circle of equal distances on the ground directly beneath the satellite (Fig. 1.2). Doppler frequency shifts, which occur as the antenna moves along track, are then used to define a hyperbola of equal Doppler on the ground (Fig. 1.2). The location of the target is identified by finding the overlap between the two. By using an antenna of synthetic aperture, the resolution in azimuth is improved by 3 orders of magnitude from 5–10 km to <5–10 m.

Fig. 1.2

SAR satellite geometry for a right-looking, ascending (south–north) orbit adapted from Massonnet and Feigl (1998). Position of the returned radar echoes are determined using the two way travel time, as shown by lines of equi-distance. Doppler frequency shifts are then used to sort samples in the azimuth direction

As described, targets on the ground are identified by the range between the target and the satellite. In regions of significant topography, geometrical distortions occur. Layover, where returns from scatterers at high elevations occur before returns from scatterers in low-lying areas, displaces the top of mountains, and shadow occurs as the radar beam is blocked by topography. These effects are particularly problematic at steep-sided stratovolcanoes, but may be reduced by using a larger look angle (Pinel et al. 2011).

Interferogram Formation and Sources of Error

Interferograms are produced using two focussed SAR images, a master and slave, that cover the same area and are aligned to within a fraction of a pixel. The master image is multiplied by the complex conjugate of the slave image, and the interferogram amplitude is therefore the amplitude of the master multiplied by the amplitude of the slave, whereas the interferogram phase is the phase difference between the two.

As previously described, the phase of a pixel within a SAR image is based upon the contribution from many scatterers distributed within the corresponding ground resolution element (Fig. 1.3) and is random. If the scattering characteristics of the ground remain the same in the time period between the two acquisitions, differencing the two SAR images would remove the random contribution. However, in reality changes in the phase response of scatterers causes the scatterer contributions to sum differently (Zebker and Villasenor 1992). This is quantified by interferometric correlation, $\gamma$, which measures the coherence of a group of pixels in, for example, a 3$\,\times \,$3 window. Stable targets act as coherent backscatterers with $\gamma =$ 1, whereas pixels with a phase response independent between two acquisitions are incoherent, $\gamma =$ 0 (Seymour and Cumming 1994). The most dominant sources of incoherence are environmental changes over time (temporal decorrelation), caused by vegetation, unstable volcanic deposits, or snow cover. Larger scatterers, such as outcrops or buildings, tend to be more stable than smaller scatterers, such as leaves. As more energy is returned from scatterers that are comparable to the radar wavelength, L-Band satellites (Table 1.1) exhibit the least temporal decorrelation (Massonnet et al. 1996; Rosen et al. 1996). Geometrical decorrelation occurs due to differences in the satellite viewing geometry between orbits. Total decorrelation occurs when the difference in path length across a pixel is greater than the radar wavelength, $\lambda$ due to a large topographic slope ($\zeta$), such as at a steep-sided stratovolcano, or when there is a large spatial separation (baseline) between satellite orbits. The critical baseline, $B_{crit}$, is the perpendicular baseline above which complete decorrelation occurs:

$$\begin{aligned} B_{crit} = \frac{\lambda R_m tan(\theta - \zeta )}{2\Delta R_m}. \end{aligned}$$

(1.4)

Thus interferogram decorrelation increases with the temporal- or spatial-separation between acquisitions. Small baseline (SB) approaches therefore use interferograms with short spatial and temporal baselines to minimise changes in the phase contributions from scatterers and overcome decorrelation (Berardino et al. 2002; Schmidt and Bürgmann 2003). This is the approach used to produce interferograms in this thesis. Decorrelation is further reduced by applying a power spectrum filter (Goldstein and Werner 1988). The result may then be multi-looked by averaging in range and azimuth to increase the signal to noise ratio and produce square pixels.

Some ground elements may contain a persistently dominant scatter with a phase response that varies little over time, despite changes in surrounding scatterers. This is referred to as a persistent scatterer (PS) (Hooper et al. 2004, 2007), (Fig. 1.3). PSInSAR describes a different way of selecting pixels by assessing phase variance in either time (e.g. Ferretti et al. 2001) or space (e.g. Hooper et al. 2004). The former requires an a priori model of the deformation in time. The latter utilises iterative spatial bandpass filtering to estimate the spatially correlated component of phase. The spatially correlated component is subtracted, and residual DEM errors for the whole dataset are modelled and then also removed. This isolates the contribution due to noise, which is not correlated in either space or time (Hooper et al. 2004). PS pixels are those where the noise term is small enough that it does not completely obscure the signal (Hooper et al. 2007). This approach of selecting PS pixels provides better coverage of PS in rural environments (Sousa et al. 2011), and is more suitable for measuring deformation that is non-linear in time (Hooper et al. 2012).

Fig. 1.3

Schematic comparing the phase response of distributed and persistent scatterers. Adapted from Hooper et al. (2007)

PS pixels are not necessarily coherent in SB interferograms, and distributed scatterer pixels coherent in SB interferograms are not necessarily identified as PS. Methods have therefore been developed to combine both approaches to optimise interferogram coherence (Hooper 2008; Ferretti et al. 2011). PSInSAR has been successfully applied in a number of volcanic settings (e.g. Pinel et al. 2011; Ofeigsson et al. 2011) including Three Sisters in the Oregon Cascades (Riddick and Schmidt 2011, and this method is implemented in Chap. 2.

The interferogram phase, $\phi _{LOS}$, is dependent upon path length, which can be used to measure ground deformation. In addition to the component caused by deformation of Earth’s surface ($\Delta \phi _{def}$), $\phi _{LOS}$ contains several nuisance terms:

$$\begin{aligned} \phi _{LOS} = \Delta \phi _{def} + \Delta \phi _{geom} + \Delta \phi _{topo} + \Delta \phi _{atm} + \phi _{error}. \end{aligned}$$

(1.5)

The term $\Delta \phi _{geom}$ arises due to differences in the satellite viewing geometry. Precise orbits are used to estimate the perpendicular baseline between satellite orbits, and the effects of baseline separation are initially removed during processing by approximating the surface of the earth to be a smooth ellipsoid. However, errors in the predicted satellite orbit occur due to e.g. solar radiation pressure (Ziebart et al. 2005), resulting in a residual, long wavelength phase ramp (Zebker et al. 1994). Empirical corrections used to remove this error contribution are discussed in Chap. 2. A digital elevation model (DEM) is also used during processing to remove the phase arising from topography. However, any discrepancies between the DEM and the true topography generates phase errors that also contribute to $\Delta \phi _{geom}$.

The largest sources of error are phase changes due to variations in the refractivity of the atmospheric between acquisitions ($\Delta \phi _{atm}$). Methods of identifying and quantifying atmospheric errors are introduced in Chap. 2, with a more thorough discussion of cause, effects, and mitigation in Chap. 3. $\phi _{error}$ is an additional noise term describing phase noise due to variations in scattering and instrument thermal noise.

The interferogram phase is initially wrapped (modulo 2$\pi$), where a phase change of 2$\pi$ is equal to ground motion with magnitude of $\lambda /2$. For the SB interferograms produced in this thesis, phase unwrapping is implemented by masking pixels with $\gamma$ < 0.1 and integrating the phase gradient using a branch-cut algorithm (Goldstein and Werner 1988). The integrated field should be conservative i.e. the integral around a closed loop should be zero. Non-zero path integrals are identified, and regions with adjacent positive and negative residues are cut and cannot be integrated across, with phase integration continuing along another path (Goldstein and Werner 1988). Where coherence is discontinuous, the phase difference between neighbouring coherent patches can be estimated and bridged. To convert to displacement, the phase is multiplied by $-\lambda /{4\pi }$, where $\lambda$ is the satellite wavelength, and the negative sign follows the convention that movement towards the satellite (uplift) is positive.

1.3 InSAR Observations of Ground Deformation at Volcanoes

InSAR has been used to make observations of ground deformation at over 500 volcanoes worldwide (Biggs et al. 2014). These observations have been used in making timely hazard assessments, and to better understand the behaviour of volcanoes throughout the eruptive cycle. A recent review by Pinel et al. (2014) summarises the volcanological insights and advances resulting from the last 2 decades of volcano InSAR studies. This includes the detection of magma storage regions and characterisation of storage geometries, which may involve multiple components (e.g. Kilauea: Baker and Amelung 2013). Pathways of magma transport have also been mapped (e.g. Yellowstone: Wicks et al. 1998), with measurements showing the role of faults in segregating magmatic fluids (e.g. Chaiten: Wicks et al. 2011). Prior to eruption, magmatic intrusions may link to open conduits, where viscous flow induces near field displacements (e.g. Mount St Helens: Anderson et al. 2010). Surface deformation may also originate from shallow hydrothermal systems, where cyclical displacements reveal the formation and breaching of a self sealing zone (e.g. Campi Flegrei: Chiodini et al. 2015).

In all cases, the interpretation of volcano deformation signals requires the use of modelling. Simple analytical source models embedded in an isotropic, elastic half space are an established, quick to use, and widely applicable way of providing first order constraints upon the source of deformation (Mogi 1958; Okada 1985; Fialko et al. 2001a). Whilst these models do not account for crustal heterogeneites or rheological variations, they provide a remarkably good fit to deformation signals at volcanoes in a range of tectonic settings (Dzurisin 2007; Segall 2010). The results of analytical source modelling have also been used as input to other models, which is the approach taken in Chap. 4. Examples include linking intrusion geometry with coulomb stress models to investigate the stresses induced by magma bodies (e.g. Jónsson 2009). This provides a mode of combining deformation measurements with other datasets such as seismicity or petrology.

When more information about crustal structure is available, finite element modelling (FEM) may be used to allow for irregular source geometries, crustal layering, and non-elastic crustal rheology. This approach has been successful in detailed studies of individual cases such as Uturuncu, Bolivia (Hickey et al. 2013) and the Socorro magma body, New Mexico (Pearse and Fialko 2010). However, these models require a large number of parameters such that testing the parameter space, obtaining useful error estimates, and accounting for non-unique solutions, remains challenging (Pyle et al. 2013). Other important factors to consider when modelling magmatic sources include the compressibility of the magma (Mastin et al. 2009), and the role of pre-existing structures that act to focus crustal stresses.

Of the catalogue of volcano deformation measurements, an increasing number are derived from systematic surveys on the scale of volcanic arcs e.g. Central America: (Ebmeier et al. 2013b); Central Andes: (Pritchard and Simons 2004a); and Sunda: (Chaussard and Amelung 2012). Measurements made on these scales are not only advantageous for regional hazard monitoring, but also facilitate investigations into volcano-tectonic interactions (e.g. Biggs et al. 2009b), global patterns of magma storage depth (Biggs et al. 2014), and the fundamental controls upon magma storage (Chaussard and Amelung 2014). Regional studies have also identified unrest at volcanos previously thought to be quiescent (e.g. Biggs et al. 2009a) and off edifice deformation that may not have been identified by ground based surveys centered on the edifice (e.g. Three Sisters, Oregon: Riddick and Schmidt 2011). Recent increases in the number of deformation observations means we are beginning to address questions in volcanology on either a case-by-case or generalised basis, identifying possible commonalities in behaviour (Chaussard and Amelung 2014) whilst also highlighting the unique and changing behaviour of individual systems (Cashman and Biggs 2014).

1.4 The Cascades Volcanic Arc

The Cascades Volcanic Arc dominates the topography of the western U.S., spanning over 1000 km from northern California to southern British Columbia. Throughout Washington, Oregon and California, the arc consists of 21 Holocene volcanos (as recognised by the global volcanism database of the Smithsonian Institution), but over 3400 Quaternary vents have been identified (Hildreth 2007). As is observed for other volcanic arcs (e.g. Central America: Stoiber and Carr 1973; Burkart and Self 1985), the Cascade volcanoes are separated into segments: (1) the Garibaldi Volcanic Belt, (2) Mount Rainier to Mount Hood, (3) the rest of the Oregon Cascades, (4) the Mount Shasta region, and (5) the Lassen region (Guffanti and Weaver 1988) (Fig. 1.4).

1.4.1 Tectonic Setting

The Cascades Volcanic Arc marks the western rim of the Pacific Ring of Fire at the boundary between the North American and Pacific plates. The relative motion between these plates is largely accommodated by spreading at the Juan de Fuca Ridge and subduction beneath Cascadia (Fig. 1.4), but 20–25 % of the motion is distributed within the overriding continental plate as block rotations and translations, the most significant of which is the clockwise rotation of Oregon (Wilson 1993; Wells et al. 1998), (Fig. 1.5).

In California, plate motion is dominated by the movement of the Pacific plate northwest relative to North America at a rate of $\sim$50 mm/yr (Atwater 1970). This motion is taken up on the San Andreas fault and the Walker Lane fault zone, a broad and discontinuous region of faulting that impinges upon the southernmost Cascade volcanoes (Hammond and Thatcher 2005; Wesnousky 2005). Between the Lassen and Shasta arc segments, dextral shear transitions to clockwise block rotation of Oregon (Simpson and Cox 1977), reflecting the transition from a strike-slip to convergent plate boundary (McCaffrey et al. 2007).

Convergence between the Juan de Fuca and North America plates increases from 30 mm/yr off the coast of Oregon, where subduction is margin-oblique, to 45 mm/yr off the coast of Washington, where subduction is arc-normal (Wilson 1993; McCaffrey et al. 2007). North of $\sim$48$^\circ$, the subduction zone exhibits trenchward-concave curvature, and the Juan de Fuca plate bends into an arc plunging eastwards, as imaged by seismicity and seismic tomography (Crosson and Owens 1987; Weaver and Baker 1988; Bostock and VanDecar 1995).

In the north of the arc, northward movement of the 400 km long Oregon block causes shortening (Wells et al. 1998) (Fig. 1.5). In the southern Cascades, the block rotation of Oregon creates extension along the trailing edge. This is further facilitated by westward migration of the east-west extensional Basin and Range province, located east of the Cascades axis (e.g. Wells et al. 1998; McCaffrey et al. 2007). Consequently, extension plays a significant role in Cascade volcanism (Guffanti and Weaver 1988), vent distributions and hydrothermal heat discharge (Ingebritsen and Mariner 2010), and is thought to be a controlling factor in segmentation of the arc (Hildreth 2007). For example in the southern Cascades, faulting in the gaps between arc segments indicates northwest-southeast dextral motion, whereas the volcanic centres are seen to be dominated by north-south orientated normal faults that facilitate extension (Blakely et al. 1997; Janik and McLaren 2010).

1.4.2 Magma Production and Storage

Magma production and extrusion throughout the history of the Cascades Volcanic Arc is thought to be a function of convergence rate and upper plate stress regime (Priest 1990): lower eruption volumes are linked to slower, more oblique subduction (Verplanck and Duncan 1987), and high erupted volumes are related to extension (Priest 1990). The relatively large across-axis width of the arc (up to 100 km) is though to imply the widespread availability and penetration of basaltic magma (Hildreth 2007). Where the arc is widest, e.g. the Lassen segment (Fig. 1.4), extreme subduction-component-enriched magmas occur in the fore-arc, and more intraplate magmas occur in the back-arc (Bacon et al. 1997).

There are at least 3 end-member primitive magma sources throughout the Cascades Volcanic Arc (Bacon et al. 1997). Primitive high-Al olivine tholeiite (comparable to mid-oceanic ridge basalt) is erupted from northern California as far north as southern Washington, with vents marked by low shield summits, pit craters and small spatter cones (Bacon et al. 1997). These magmas are related to high-temperature, extensional environments (e.g. Donnelly-Nolan 1988), and are thought to result from nearly anhydrous melting of depleted mantle that had previously been enriched with a subduction component (Donnelly-Nolan et al. 1991; Baker et al. 1994). More silicic calc-alkaline (arc) basalts and basaltic andesites are abundant at the arc-axis (Bacon et al. 1997). These higher viscosity, lower temperature lavas, have higher H$_2$O contents due to the transport of H$_2$O to the mantle wedge during subduction (Brandon and Draper 1996). Variations in the concentration of SiO$_2$ and incompatible elements varies with latitude throughout the arc, and is thought to be associated with the enrichment of a third, subduction component (Bacon et al. 1997). More evolved erupted products including dacites (e.g. Mount St Helens: Pallister et al. 2008) and rhyolites (e.g. Medicine Lake Volcano: Grove et al. 1997), are thought to originate from mafic crustal inputs (Hildreth 1981) and have been produced by magma mixing, fractional crystallisation and crustal melting (the FARM model: Baker et al. 1991).

Fig. 1.4

Tectonic map of the Cascades Volcanic Arc with Quaternary vent locations (black dots) and major faults (black lines) from Hildreth (2007) and plate velocities from McCaffrey et al. (2007). The 13 main volcanic edifices are labelled and shown by white triangles. On the right are the arc segments after Guffanti and Weaver (1988)

Fig. 1.5

GPS velocities (1993–2011) in Washington and Oregon showing block-wise rotation of Oregon. GPS velocities are relative to North America and error ellipses mark 70 % confidence. Adapted from McCaffrey et al. (2013). The 13 main volcanic edifices are shown by white triangles as in Fig. 1.4

Petrological and geophysical studies throughout the Cascades provide evidence of magma storage at depths of $\sim$7–10 km e.g. Mount St Helens: (Rutherford et al. 1985; Pallister et al. 1992; Scandone and Malone 1985; Moran 1994); cinder cones near North Sister: (Ruscitto et al. 2010); Mount Rainier: (Venezky and Rutherford 1997); Three Sisters: (Riddick and Schmidt 2011). Magma storage is thought to occur on a relatively small scale. For example at Mount St Helens, melt-rich magma is thought to be stored in a widened conduit at 5–10 km depth with a volume of $\sim$4 km$^3$ (Gardner et al. 1995; Pallister et al. 2008). Small-scale magma storage is also supported by compositional variability between neighboring vents separated by only 1–12 km (Hildreth 2007).

1.4.3 Monitoring

Cascade volcanoes threaten major populations and development (Ewert et al. 2005). Consequently, monitoring of the arc is undertaken by the U.S. Geological Survey’s Volcano Hazards Program at the Cascades and California Volcano Observatories. The main monitoring infrastructure in the Cascades consists of seismic and continuous GPS networks, with variable coverage throughout the arc.

In the northern Cascades, routine seismic monitoring has been undertaken by the Pacific Northwest Seismic Network from the University of Washington since the 1970s. In the southern Cascades, the Northern California Seismic Network is operated by the U.S. Geological Survey and the University of California, Berkeley. Levels of seismicity are variable throughout the arc, as are detection thresholds of the seismic network (Moran 2004). In addition to volcano-tectonic events, seismicity is attributed to glaciers e.g. Mount Baker: Nichols et al. 2011), hydrothermal systems (e.g. Lassen Volcanic Center: Janik and McLaren 2010), and seasonal groundwater recharge (e.g. Mount Hood: Saar and Manga 2003). Deep, long-period earthquakes are also observed, and between 1980 and 2009, 31 events were detected at Mount Baker, 9 at Glacier Peak, 9 at Mount Rainier, 9 at Mount St Helens, 1 at Three Sisters, and 1 at Crater Lake (Nichols et al. 2011). Long-period events are also detected beneath Lassen Volcanic Center, with 11 events on average recorded each year between 2003–2011 (A. M. Pitt, unpublished data in Clynne et al. 2012). These events are thought to represent fluid/magma transport along pre-existing tectonic structures (Nichols et al. 2011).

Importantly, shallow seismicity is not necessarily a strong indicator of magmatic activity at Cascade volcanoes (e.g. Mount Baker: Crider et al. 2011; Three Sisters: Riddick and Schmidt 2011). Geodetic observations are therefore a key component of monitoring the arc. Of the 12 high - very high threat Cascade volcanoes (as ranked in the National Volcano Early Warning System: Ewert et al. 2005), 9 are monitored by at least 1 continuous GPS station (Table 1.2). Continuous GPS (cGPS) data for the Cascade volcanoes is primarily collected by UNAVCO and is available online in near-real time via the U.S. Geological Survey Earthquakes Hazard Program. In addition to cGPS, 10 of the 12 high - very high threat volcanoes have undergone additional geodetic surveys, with baseline measurements established as far back as the 1930s (e.g. Newberry: Yamashita and Doukas 1987), and campaign GPS measurements made as recently as 2014 (e.g. Mount St Helens) (Table 1.2).

The archive of InSAR data for the Cascades is large. ENVISAT data for Mount St Helens alone includes three ascending satellite tracks and one descending track, totalling over 140 separate data acquisitions between 2003–2010. Accordingly, the existence of baseline InSAR imagery is considered to be the most basic level of deformation monitoring of Cascade volcanoes (Ewert et al. 2005). However, as this thesis describes, the application of InSAR data in the Cascades has been limited by incoherence caused by dense vegetation and snow cover, and extensive atmospheric artefacts (e.g. Mount St. Helens: Poland and Lu 2008).

Table 1.2 Geodetic monitoring of high - very high threat Cascade volcanoes

1.5 Thesis Structure

This thesis contains two ongoing themes (1.) the application of InSAR data in the Cascades, an environment associated with incoherence and high levels of noise; and (2.) causes of long-term volcanic subsidence, with a focus on two volcanoes in the southern Cascades. Chapter 2 is published in Geophysical Journal International (Parker et al. 2014), and a modified version of Chap. 3 is published in Remote Sensing of Environment (Parker et al. 2015). Chaps. 4 and 5 are currently in preparation for submission to journals and are therefore subject to peer-review.

Chapter 2 describes the application of multi temporal analysis techniques to improve the application of InSAR at Medicine Lake Volcano. Past geodetic studies show that the volcano has subsided at a steady rate since the 1950s, and the results of InSAR analysis are used to extend the geodetic record to 2011 and show that deformation continues at historical rates. The high spatial resolution of deformation measurements from InSAR are then used to provide improved constraints upon the source geometry, and therefore the driving mechanism of subsidence.

Chapter 3 continues to focus on the application of InSAR data in the Cascades, and uses large-scale atmospheric models to investigate atmospheric uncertainties on an arc-wide scale. Rather than focusing on correcting atmospheric uncertainties retrospectively, this chapter describes a strategy to produce a priori estimates of atmospheric uncertainties, that can be used in near-real time. This is achieved by investigating the influence of volcano topography and geography upon atmospheric uncertainties, which are used to estimate detection thresholds for long-term monitoring of small magnitude signals, and short-term monitoring of pre-eruptive ground deformation in the Cascades.

Chapter 4 investigates the long-term geodetic response to cooling and crystallisation of magmatic intrusions. Geometrical and petrological constraints from Medicine Lake Volcano are input to a thermal model, which is used to assess the influence of intrusion geometry and composition upon the magnitude and time-scales of resulting surface deformation. Comparing the results to the geodetic history at Medicine Lake Volcano yields a suite of best-fitting models that are used to constrain the timing of intrusion. Consideration is given to the effects of convection, incremental intrusions, and volume loss due to degassing, and the impacts of these factors upon interpretations of the timing of intrusion.

Chapter 5 uses the methods described in Chaps. 2 and 3 to apply InSAR data at Lassen Volcanic Center. Multiple datasets are used to constrain the spatial and temporal characteristics of the deformation field, and by incorporating past geodetic datasets, the onset of deformation is inferred. The style of deformation and best-fitting source geometry are compared to those for Medicine Lake Volcano, and causes of subsidence are evaluated in light of seismic and hydrothermal activity at the volcano.