**1. Introduction**

Many active volcanoes, about 200 worldwide [1], generate lava domes that are often characterised by hazardous explosive eruptions that involve flank instability [2]. As domes grow, the outer flanks oversteepen until they collapse and perilous pyroclastic flows are produced that purge down the slopes, affecting regions at kilometres distance to the dome [3]. Lava domes are thought to grow by interactions between magma injections into the dome (i.e., endogenous dome growth) and the addition of extrusion layers on the top of the carapace (i.e., exogenous dome growth) [4–7]. So far, these two styles of growth are considered as endmembers, with few examples showing higher complexity that could be instrumentally recorded in nature, such as at the dome building volcanoes Mount St. Helens, Unzen, or Soufrière Hills [5–7]. Geophysical sensors often observe short-term precursors, such as seismicity, enhanced rockfall intensity, alternating volcanic gas emissions, or localised deformation

when magma reaches shallow depths prior to imminent eruptions [8]. Interferometric Synthetic Aperture Radar (InSAR), for example, provides an estimation of precursory deformation on the mm to cm scale over short [9] and long [10] periods of time, yet the technique is affected by atmosphere and it is less effective when volcanoes are covered in snow or when ground motion exceeds the maximum detectable deformation gradient [11]. Moreover, determining deformation that is associated with dome building volcanoes, and therefore identifying the particularities of lava dome growth, is challenging due to the small dimensions and the hazardous access of most domes. Successful approaches barely include in-situ and, more often, remote sensing approaches, such as ground-fixed cameras [12,13] or satellite radar amplitude images [14–16]. A noteworthy case of the strength of camera monitoring for tracking deformation is that of Mount St. Helens during the 2004–2008 dome growth episode, which allowed for the spatial and temporal quantification of endogenous and exogenous growth [17]. The value of satellite radar observation, on the other hand, was underlined during the 2010 cataclysmic eruption at Merapi, where Synthetic Aperture Radar (SAR) amplitude scenes provided vital support in the early detection of dome growth and the associated hazard assessment [14]. The weaknesses of SAR, in turn, come from the poor revisit period (several days), and geometric distortions that limit interpretations due to the regions of shadow, foreshortening, and layover effects.

Here, we integrate the strengths of these techniques to better understand the current dome growth mechanisms acting during the January 2016–June 2017 eruption sequence at Bezymianny. We use camera monitoring to roughly identify topographic changes at Bezymianny's flank, and we employ a pixel offset tracking algorithm on high-resolution TerraSAR-X amplitude images to quantify ground motion in range and azimuth direction. We show the details of plug extrusion that were identified at least seven months before the first documented effusive eruption, and that exogenous growth at Bezymianny was likely preceded by intrusions into the northern part of the composite dome. The complexity of the observed cascade suggests that this finding may also provide a basis for dome growth observation at other dome building volcanoes, ultimately promoting the understanding of dome growth and related hazard assessment.

#### **2. Bezymianny**

#### *2.1. Volcanological Background*

Bezymianny is an andesitic, dome building volcano (~3000 m a.s.l.) that is located within the Klyuchevskoy Group of Volcanoes (KGV) in Kamchatka, Russia (Figure 1a). It is thought that Bezymianny, Klyuchevskoy, but also the further south located Tolbachik, derive their fluids from a common deep parental magma chamber at 30 km depth [18,19]. During ascent beneath Bezymianny, the volatile-rich magma arrests at different levels, which are likely associated with magma chambers, at approximately 15 km and 5 km, but also possibly at 1.5 km depth [18,20–22].

Bezymianny is a relative young volcano (5.5 ka) whose geologic history was characterised by major eruptive activity between 2400 and 1700 and 1350–1000 before present [23]. In 1955–1956, Bezymianny re-emerged with a phase that culminated in a cataclysmic sector collapse and directed lateral blast eruption, which left behind a horseshoe-shaped crater moat (Figure 1b) [24–26]. Eruption characteristics showed strong similarities to the catastrophic eruption at Mount St. Helens in 1980 [27–29]. After the 1956 eruption at Bezymianny, near-continuous, mostly endogenous dome growth started to fill the horseshoe shaped crater floor until 1965 [24,25,30,31]. Since 1977, on average, 1–2 explosive eruptions occurred, which showed a characteristic cyclic behaviour: initially, days to weeks long-lasting summit plug extrusions were followed by Vulcanian explosions and pyroclastic flows; eruptions then eventually ceased with lava flow emplacements and degassing until the volcano became quiet again [30,31]. Only few eruptions during the 1980 s and 1990 s were solely characterised by effusive activity, or lava flow emplacements prior to explosions [30]. By 2004, Bezymianny's dome was completely covered with lava flows, and multiple explosions on its top between 2005 and 2012 a new relatively stable summit crater [19,32,33]. After four years of quiescence, activity initiated in 2016 and was followed by

long effusive activity (5 December 2016–9 March 2017) and two strong explosive eruptions on 9 March and 16 June 2017 [31]. Today, the morphology of Bezymianny is characterized by the remnants of the 1956 sector collapse amphitheatre (the "somma") and the presence of an approximately 500–600 m high composite dome in the centre (Figure 1b).

**Figure 1.** (**a**) Shaded relief map (TanDEM digital elevation model from 2014) of Bezymianny and its closest neighbouring volcanoes in Kamchatka, Far East Russia (star in inset map). Location of the time-lapse camera and footprint of the TerraSAR-X (TSX) satellite are indicated by CAM and black box, respectively. Orthogonal arrows show flight and line-of-sight (LOS) directions of the descending (DSC) and ascending (ASC) TSX satellites, respectively. ASC is shown in dashed lines, as this paper focuses on the more regular DSC data. White box denotes area shown in (**b**). (**b**) Close-up shaded relief map of Bezymianny showing the 1956 collapse scar (somma), the subsequently evolved central composite dome, and its recent summit crater. Profile A-A' indicates approximate 1956 collapse plane (dashed line). Small letter profiles a–a' and b–b' show landmarks that are used for scale approximation of camera images.

#### *2.2. Monitoring Activities at Bezymianny*

Bezymianny is one of the most active volcanoes in Kamchatka that poses a risk to air traffic between North America and Asia. However, multiparametric and long-term monitoring is challenging due to the remoteness of the volcano. During the past two decades, seismic monitoring was realized by the Kamchatkan Branch of Geophysical Survey [34], allowing for eruption precursor identification days to weeks before eruptions [35]. Enhanced frequency of rockfalls from the central dome is easily identified and indicative of renewed activity at Bezymianny [35,36]. Besides characteristic tremors and high frequency seismic signatures, long-period (LP) seismicity may also identify heralding eruptions [19], but earlier studies suggest that only one out of four eruptions were preceded by LP events [36]. Eruptions at Bezymianny are sometimes concurrent with activity at Klyuchevskoy, which may strongly obscure the records of Bezymianny's seismic activity [35,36].

Besides the routine seismic monitoring, increasing importance has been ascribed to remote sensing data analysis. Remote sensing comes with two main motivations: first, general monitoring of the volcanic activities exploiting cost-free data and web portals; second, experimental and scientific in-depth analysis of selected volcanic crisis. For instance, the instruments of the Advanced Spaceborne Thermal Emission and Reflectance Radiometer (ASTER) have been episodically used to study heat radiation during eruptions of the last few decades [37–39]. Overall, these studies have identified enhanced ground temperature anomalies as a common precursor for Bezymianny's eruptions, although two eruptions were reported without a preceding change in the thermal level [39].

Yet, existing real-time monitoring methods, as well as event-based observations, could not assess detailed dome growth processes of Bezymianny. Here, we investigate webcam images and high-resolution satellite radar data that cover the December 2016–June 2017 eruption sequence at Bezymianny. The data catalogue enabled the observation of the volcano with unprecedented precision of precursory activity, as well as exogenous and endogenous dome growth.

#### **3. Data and Methods**

This study concentrates on camera monitoring and satellite radar data acquired in 2016 and 2017. The eruption had a precursory phase, as identified by rockfalls and seismicity, then an effusive eruption between 5 December 2016 and 28 February 2017, followed by (i) the effusive and explosive eruption on 9 March, and then (ii) the explosive 16 June 2017 eruption. Details of these three stages (precursor–effusive–explosive) were identified in the data. We compare our camera and SAR results to the seismic records of Bezymianny [40].

### *3.1. Camera Monitoring and Mimatsu-Diagrams*

Previous studies have demonstrated the strength of time-lapse camera analysis for the determination of morphology changes at volcanoes, which substantially contributed to the spatial deformation monitoring. Mimatsu (1962) already highlighted the significance of optical volcano monitoring by constantly recording the volcano's changing shape on his office paper window during the dome growth episode at Showa Shinzan volcano, Japan. Johnson et al. [41] used video-derived imagery to track dome uplift at Santiaguito volcano, Guatemala, and successfully correlated the results with long-period seismic signals. A fixed camera network that was installed around Mount St. Helens, USA, permitted the precise estimation of growth and strain rates during the 2004–2008 spine extrusion [13,17]. Based on displacements between fixed time-lapse photographs, Walter et al. [42] showed that dome deformation at Merapi volcano, Indonesia, is strongly governed by the local topography. In this context, we used time-lapse imagery of Bezymianny during the 2016–2017 eruption series to record the changes at the summit following Mimatsu's approach.

The employed day and night capable network camera (Axis P1346) that is operated by the Kamchatkan Branch of Geophysical Survey [43] has a focal length of 4 mm and it produces one image (2048 × 1536 pixels) per second (Figure 2a). The time-lapse camera is located 7 km to the southeast (160.696E, 55.94N; Figure 1a) and it captures Bezymianny as well as the neighbouring Kamen and Klyuchevskoy volcanoes. The investigated period from May 2016 to August 2017 encompasses 579,180 time lapse images. The images were weeded out for night, no-operation, and cloudy records, but also records where the camera lens was covered with snow. The remaining dataset was then visually checked for clear view and high contrast images that were taken at approximately the same daytime, of which 12 representative images were manually selected and cropped to the area of Bezymianny (Figure 2a,e and Figure S1). From the image stack, we follow the silhouette of the volcano image-by-image, which is referred to as a Mimatsu diagram. Image offsets due to strong winds, recurrent snow cover on top of the camera, and/or temperature changes of the installed camera gear are corrected by manual alignment (translation in x and y, rotation around the image centre) of all the images with respect to the master scene (7 May 2016). We favoured manual over automatic alignment as Bezymianny's dome was recurrently covered with snow, limiting automatic algorithm performance.

To quantify topographic changes in the field-of-view (FOV), two scales were derived by measuring pixel distances between conspicuous landmarks on the master image (summit crater, 1956 collapse scar) that correspond to landmarks on the digital elevation model (Figures 1b and 2b). Thus, the scale varies between ~4.2–5.3 m/pixel, which is related to the reduction of the three-dimensional topography into the camera's two-dimensional FOV. As the pixels of the images are approximately squared, the scale is assumed to be valid for horizontal and vertical changes. Since most of the images unveiled insufficient contrast conditions and strongly varying colours (e.g., recurrent snow cover), the outlines of Bezymianny's composite dome were manually mapped and stacked in the resulting Mimatsu diagram.

**Figure 2.** (**a**) Monitoring camera image of Bezymianny and its neighbouring volcanoes on 7 May 2016. View is to the west. Black box denotes image details shown in: (**b**) Close-up view of Bezymianny. Pixel and metric distances between a–a' and b–b' are derived from a terrain model (cf. Figure 1b). (**c**) Descending non-geocoded spotlight-mode TSX amplitude image from 23 September 2016 (cf. Figure 1a). Flight direction (azimuth) and line-of-sight (LOS) or range direction of the satellite are indicated. White box shows area used for pixel offset tracking displayed in: (**d**) Close-up of Bezymianny. Small black boxes mark assumed stable areas (i.e., no deformation) referred to later in displacement analysis. (**e**) Cumulative number of seismic events in a 6 km radius to the volcano. Available TSX acquisitions with their perpendicular baselines (Baseline⊥) to adjacent acquisitions.
