Assessing the Use of Optical Satellite Images to Detect Volcanic Impacts on Glacier Surface Morphology

Globally, about 250 Holocene volcanoes are either glacier-clad or have glaciers in close proximity. Interactions between volcanoes and glaciers are therefore common, and some of the most deadly (e.g., Nevado del Ruiz, 1985) and most costly (e.g., Eyjafjallajökull, 2010) eruptions of recent years were associated with glaciovolcanism. An improved understanding of volcano-glacier interactions is therefore of both global scientific and societal importance. This study investigates the potential of using optical satellite images to detect volcanic impacts on glaciers, with a view to utilise detected changes in glacier surface morphology to improve glacier-clad volcano monitoring and eruption forecasting. Roughly 1400 optical satellite images are investigated from key, well-documented eruptions around the globe during the satellite remote sensing era (i.e., 1972 to present). The most common observable volcanic impact on glacier morphology (for both thick and thin ice-masses) is the formation of ice cauldrons and openings, often associated with concentric crevassing. Other observable volcanic impacts include ice bulging and fracturing due to subglacial dome growth; localized crevassing adjacent to supraglacial lava flows; widespread glacier crevassing, presumably, due to meltwater-triggered glacier acceleration and advance. The main limitation of using optical satellite images to investigate changes in glacier morphology is the availability of cloud- and eruption-plume-free scenes of sufficient spatial- and temporal resolution. Therefore, for optimal monitoring and eruption prediction at glacier-clad volcanoes, optical satellite images are best used in combination with other sources, including SAR satellite data, aerial images, ground-based observations and satellite-derived products (e.g., DEMs).


Introduction
Many (245 of 1413) Holocene volcanoes on Earth are currently ice-clad or have glaciers in close proximity [1,2], and interactions between volcanic activity and ice are therefore common (e.g., [3][4][5][6][7]). Such volcano-ice interactions are important since they often result in hazards, including floods, debris flows (lahars) and/or the distribution of fine-grained ash [3,4,8]. Some of the deadliest (e.g., Nevado del Ruiz, 1985) and most costly (e.g., Eyjafjallajökull, 2010) eruptions in recent history were associated with ice-clad volcanoes [9,10]. The former eruption killed~23,000 people [3], and the latter heavily impacted air travel in large parts of Europe and caused considerable costs of about 1.3 billion euros (£1.1 billion, US$1.7 billion), due to the widespread dispersal of fine-grained ash [8,9]. Therefore, the investigation of volcano-ice interactions is of considerable social and economic importance. They are also significant for glaciology, since predicting the future behaviour of glaciers requires knowledge of their response to different forcing mechanisms. A notable example is the potential for

Background
The events presented in this study cover glaciovolcanic interactions from different geologic settings around the world. This implies variable ice thicknesses and eruption styles and intensities, which influence the way that ice masses are affected by volcanic activity. The ice masses studied in this paper range from comparably small, thin glaciers (commonly, but not exclusively, tens of metres thick) on top of stratovolcanoes, to hundreds of meters thick ice caps in Iceland. This section gives background information on the events presented in the Results section.
A common way that glaciers are affected by volcanoes is via subglacial heating and the formation of ice cauldrons e.g., [4,15,20]. The term ice cauldron covers the range from shallow, non-crevassed depressions in the ice surface to steep-sided, highly crevassed cylindrical chasms, which sometimes extend to bedrock (i.e., to the glacier bed). Cauldrons are generally caused by heat applied at the glacier base at the location of the vent. This rapid basal ice melting results in subsidence of the overlying ice and ice flow towards the vent. Some cauldrons are associated with meltwater accumulation that is rapidly released in jökulhlaups [20,21]. In contrast, openings are associated to meltwater flow through zones of weakness within the ice (e.g., pre-existing crevasses) due to high subglacial water pressures. The openings are formed locally if the meltwater finally reaches the glacier surface and continues flowing supraglacially [5,7]. The 2014-2015 rifting episode between Bárðarbunga and Askja in Iceland resulted in the formation of ice cauldrons (on Dyngjujökull glacier) along the path of a 45-km long subglacial dyke, and included the collapse of the Bárðarbunga caldera [15,[22][23][24]. These cauldrons lie within a sector of ice 430-670 m thick [25]. In 1999-2005, geothermal activity on Katla (Iceland), beneath Mýrdalsjökull glacier, resulted in the formation of very large ice cauldrons,~1-1.5 km in diameter above 400-700 m thick ice [21,26]. During the more violent eruption of Eyjafjallajökull (Iceland) in 2010, ice cauldrons formed in the summit crater and openings appeared in the ice surface Remote Sens. 2021, 13,3453 3 of 17 along a subglacial flood path down Gígjökull glacier [5,7]. Ice thicknesses are reported to be~200 m for the caldera region, and typically less than 100 m for Gígjökull glacier [7,27]. Cauldron formation is not exclusive to subglacial volcanic activity in Iceland but has been observed on volcanoes elsewhere globally. For instance, on Mount Redoubt (Alaska), ice cauldrons were observed approximately two months before its eruption in March/April 2009, alongside a variety of other precursors such as crevassing, ice-slurry/debris flows and the emission of ejecta [6]. One of the cauldrons formed in the summit crater of the volcano, where ice is~150 m thick [28], and was associated with near-vertical walls and clear concentric crevassing [6,29,30]. Cauldrons associated with the emission of gases and pyroclasts were also reported during the eruption of Mount Hudson (Chile) in October to December 2011 [31]. The March-April 2021 eruption of Mount Veniaminof in Alaska also produced a visible cauldron~1 km in diameter, east of the prominent and active intra-caldera cone within ice that is~100 m thick [32,33]. During this period, the emission of ash-plumes, (potential) lava effusion under the intra-caldera glacier and subsidence of glacier ice were reported [33].
Crevassing of glacier ice due to volcanic activity, but without cauldron formation has been reported from Mount Belinda (Montagu Island) in the South Sandwich Islands. Mount Belinda is a small volcanic cone in a 6-km wide ice-filled caldera, with a tidewater glacier draining its northern margin. Mount Belinda experienced a low-intensity volcanicallyactive period from 2001 to 2007. This included active lava flows and subglacial melt, which is presumed to have promoted glacier acceleration (due to basal lubrication), causing the outlet glacier's surface to become extensively crevassed [34,35]. An example of an ice mass being affected by subglacial dome growth comes from Mount St Helens (United States) during an active period from 2004 to 2006. At this time, a 1 km 2 and 200 m thick glacier occupied the volcano's "amphitheater"-like crater [36]. During the active period, a lava dome extruded in several spines and squeezed the glacier against the crater wall. The dome extrusion finally split the glacier apart and resulted in heavy crevassing and glacier bulging [37,38].

Materials and Methods
In this study,~1400 freely-available low-to medium-resolution, and a limited number of commercial high-resolution, optical satellite images were investigated. We focused on optical satellite data (rather than other datasets, including Synthetic Aperture Radar (SAR)) since only optical satellite data cover almost half a century and the past decade has seen a rapid increase in the variety of free and commercial datasets potentially suitable for glacier-volcano monitoring. Following Elliot et al. [39], we use the term "low-resolution" for image resolutions of 10-100 m, "medium-resolution" for image resolutions of 1-10 m and "high resolution" for images of less than 1 m resolution. The literature reveals~66 instances of documented volcanic impacts on glaciers at~42 volcanoes globally during the optical satellite era (i.e., 1972 to present) [13]. Following detailed analysis of each of these events, we focused on a selection of the best-documented examples that represent major potential volcanic impacts on glacier surface morphology and are therefore considered representative (see Figure 1 and Table 1). Supplementary Table S1 reports the characteristics of all investigated events as reported in the literature and provides an overview of all satellite image sources (commercial and non-commercial) used. Satellite images were obtained through geocento (https://geocento.com/, accessed on 18 November 2020), Google Earth Engine Code Editor" (GEE) [40], USGS EarthExplorer (earthexplorer.com, accessed on 24 February 2021) and Planet (planet.com, accessed on 26 May 2021), and images were viewed in the visible and near-infrared range of the electromagnetic spectrum (including panchromatic bands). Images were enhanced for better visibility using contrast enhancement/histogram stretch (where appropriate) in QGIS 3.    [2]) and volcanoes investigated in this study (red triangles, numbered). For some volcanoes, multiple events were investigated. Volcano numbers are detailed in Table 1. Basemap: "World Continents" (accessed on 10 March 2021).
Volcanoes and corresponding eruptive activity investigated in this study (volcano numbers relate to Figure 1). s illustrated in figures within the present study, are shown in bold.

Ice Cauldron Formation
In principle, ice cauldrons are readily detectable with low-resolution satellite imagery, particularly within thick ice. With satellite sensors operating in the visible and near-infrared range (including multispectral and panchromatic bands), cauldrons appear as dark (shaded) areas against the bright ice surface. A period of geothermal activity beneath Mýrdalsjökull between 1999 and 2005 led to the formation of large ice cauldrons, which were identifiable on ASTER visible and near-infrared (VNIR) images (15 m resolution, Figure 2). During the 2014-2015 Bárðarbunga eruption, three cauldrons  [2]) and volcanoes investigated in this study (red triangles, numbered). For some volcanoes, multiple events were investigated. Volcano numbers are detailed in Table 1. Basemap: "World Continents" (accessed on 10 March 2021).

Ice Cauldron Formation
In principle, ice cauldrons are readily detectable with low-resolution satellite imagery, particularly within thick ice. With satellite sensors operating in the visible and nearinfrared range (including multispectral and panchromatic bands), cauldrons appear as dark (shaded) areas against the bright ice surface. A period of geothermal activity beneath Mýrdalsjökull between 1999 and 2005 led to the formation of large ice cauldrons, which were identifiable on ASTER visible and near-infrared (VNIR) images (15 m resolution, Figure 2). During the 2014-2015 Bárðarbunga eruption, three cauldrons developed on the surface of Dyngjujökull glacier in the northern part of the Vatnajökull ice cap. These cauldrons are clearly visible as shadows against the snow-covered ice in a true-colour composite EO-1 ALI image (30 m resolution; Figure 3b). However, in a snow-free EO-1 ALI image (Figure 3c), the cauldrons are not visible, since transmitted radiance from the cauldrons region is not distinguishable from the surrounding dirty ice (even though the cauldrons are present at the time of image capture, as shown by Reynolds et al. [15] using aerial-and TerraSAR-X imagery). An enlarged view of the northernmost cauldron shows evidence of faint concentric crevassing (Figure 3d).

Opening Formation
Volcanically triggered openings on the glacier surface are in principle observable with low-resolution images (>15 m resolution), however, they are best observed with at least medium-resolution images (~5 m resolution or better) due to the surface texture making it difficult to distinguish between openings and other features on the glacier surface. A notable example comes from Eyjafjallajökull (Figure 4a), where the 2010 eruption, besides melting through the entire ice column in the main crater, produced openings on Gígjökull glacier (Figure 4b,c). A view from April 2010 shows the intact surface of the glacier before the explosive phase of the eruption (inset Figure 4a). Several openings (without concentric crevassing) developed in the upper part of Gígjökull glacier during the eruption due to subglacial meltwater flow [7], and are visible in low-resolution ASTER images (Figure 4b), but better seen with WorldView-1 imagery (2 m resolution) acquired on the same day (red arrows, Figure 4c).

Glacier Crevassing
In this section, glaciovolcanic factors/events that lead to crevassing of glacial ice are considered. Ice cauldron formation is often associated with concentric crevassing (Section 4.3.1). Other volcanic factors that can lead to glacier crevassing include subglacial lava dome growth (Section 4.3.2), localized crevassing adjacent to supraglacial lava flows (Section 4.3.3) and widespread (glacier-wide) crevassing due to glacier acceleration/advance triggered by increased meltwater at the ice-bed interface (Section 4.3.4).

Concentric Crevassing Due to Ice Cauldron Formation
Concentric crevasses related to cauldron formation on ice-covered stratovolcanoes are observed with medium-to high-resolution images, but are too small to be identified with low-resolution images alone. In principle, crevasses can be identified in satellite images as dark fissures on the bright coloured glacial ice and snow. Figure 5a shows Mount Redoubt shortly before its eruption in March 2009, and Figure 5b-d reveal the capabilities of medium-and high-resolution images for ice-cauldron/crevassing detection at this volcano. In this example, the low-resolution ASTER image ( Figure 5c) and high-resolution IKONOS-2 image (Figure 5d, 0.8 m resolution) were acquired on the same day. The ice cauldron and concentric crevasses are more clearly visible in the IKONOS-2 image than in the ASTER image due to its high spatial resolution. With high-resolution images, it is not only possible to identify concentric crevasses, but also to accurately monitor their evolution (including changes in their number, length and width). This becomes clear if Figure 5b is compared to Figure 5d. The satellite image in Figure 5b was acquired several days before those of Figure  5c,d and shows a smaller, less evolved cauldron and fewer/smaller crevasses (blue arrows). In contrast to the cauldrons on Mount Redoubt presented above, in some cases, cauldrons lack a clearly-identifiable central melt-pit and are therefore only indicated by concentric crevassing. This was the case for Mount Hudson during its 2011 eruption, as revealed by Rapid-Eye imagery (5 m resolution, Figure 6).  Figure 7b) and parts of East Crater glacier became crevassed due to the onset of lava dome growth ( Figure 7b). As the lava dome continued to grow, it extruded through the ice, resulting in the deformation and heavy crevassing of East Crater glacier (Figure 7c,d). The impact of dome growth (ice displacement, crevassing and melt-hole formation) on very small glaciers (~1 km 2 , 200 m thick) such as the one described are best viewed with at least mediumresolution images such as QuickBird 2 (Figure 7c, 2.4 m). High-resolution images (such as QuickBird-2 panchromatic, 0.6 m) might be required particularly when it comes to small scale crevassing (Figure 7b). The impacts mentioned are less clear in medium-resolution EO-1 ALI panchromatic images (10 m resolution, Figure 7d), and almost impossible to observe in low-resolution imagery (>10 m resolution).  (Figure 8b). This medium-resolution imagery also reveals localised crevasses surrounding the supraglacial lava flow (reported by [35]; see Figure 8b). These crevasses presumably indicate that the lava melted deep channels/gullies into the glacier surface [9,34,35]. These melt-gullies are difficult to observe directly, but their impact on the surrounding ice is observed in medium-resolution imagery.

Widespread Crevassing Due to Glacier Acceleration/Advance
During the 2001-2007 eruption at Mount Belinda, the outlet glacier which drains the volcano's northern flank experienced widespread (i.e., glacier-wide) crevassing. This crevassing is thought to reflect glacier acceleration and advance (i.e., "surging") in response to subglacial meltwater release during the eruption [9]. The Landsat-7 panchromatic image (15 m resolution) in Figure 8a shows the largely intact (i.e., minimally crevassed) glacier surface~9 months before the onset of volcanic activity. By contrast, the EO-1 ALI panchromatic images in Figure 8c,d, acquired during the eruptive period, shows widespread crevassing of the glacier surface. Thus, unlike localized crevassing related to ice cauldrons or volcanic dome growth (e.g., Sections 4.3.1 and 4.3.2), widespread glacier crevassing at Mount Belinda is identifiable even without access to high-resolution imagery.

Discussion
From our analysis of the preceding examples of volcano-glacier interactions using optical imagery, we have identified the most readily observed volcanic impacts on glaciers, identified several key challenges to making these observations, and developed some "best practice" guidelines.

Ice Cauldron/Opening Formation
Ice cauldrons are the most common volcanic impact on glaciers observable from optical satellite images. For small subglacial eruptions (e.g., Bárðarbunga 2014-2015) or above geothermally active regions (e.g., Mýrdalsjökull) with large ice-thickness, these cauldrons tend not to extend to bedrock (i.e., to the glacier bed), but form depressions that are much wider than deep, typically up to 3-km in diameter see [21]. In optical images, these glacier surface depressions are often observable due to shadowing, which results in areas (inside parts of the ice cauldrons) that are distinct from the usually brighter glacier surface. Due to these often-conspicuous contrasts in surface shading, large surface cauldrons are, in principle, readily observable even in low-resolution satellite images ( Figure 3). In some cases, more energetic eruptions can produce ice cauldrons that extend to bedrock. For example, at Gjálp in 1996 ice cauldrons melted through up to 500 m of ice [4]. Though large, these cauldrons can be difficult to observe during, or shortly after eruptions, since the explosive events that lead to their formation also produce ash-laden eruption plumes, preventing direct observation of the ice surface using optical remote sensing sources (e.g., no suitable optical images show the cauldron in the main caldera at Eyjafjallajökull in 2010). Ice cauldron formation is also common at volcanoes with thinner and less extensive ice cover, and cauldrons often extend to bedrock. Using optical satellite images, it is possible to detect these cauldrons on Mount Redoubt during and following its active period in 2009 ( Figure 5) and on Mount Hudson during and following its active period in 2011 (see Figure 6). The detection of cauldrons on only partially icecovered volcanoes creates some difficulties due to an often restricted cauldron size (see Figure 5), combined with a highly variable surface texture, topography and colour at the summit (also see Section 5.2). However, cauldrons are often visible in medium-and highresolution imagery and appear as either dark spots (<a few pixels) or regions (multiple pixels), depending on image resolution (see Figures 3 and 4). Particularly large lahars might be generated by eruptions through crater lakes depending on the amount of lake water involved [12]. Similarly, large amounts of water stored in ice cauldrons formed before a large eruptive event might increase the likelihood of large lahars, emphasising the importance of identifying ice cauldrons.

Crevassing
Glacier crevassing is another volcanic impact on glaciers commonly observable from optical satellite images. These crevasses are often localized and concentric (e.g., encircling ice cauldrons), but can also be more widespread, and irregular or linear in form. In fact, all volcanic impacts on glaciers discussed in this paper are, to some degree, associated with crevassing. In principle, crevasses are visible as dark stripes in optical satellite images (Figure 8c), though their detection depends on a notable colour contrast between the crevasse and the surrounding glacial ice and snow. Various approaches have been used to detect widespread regions of crevassing on "non-volcanic" glaciers, using lowto high-resolution optical satellite images [41][42][43][44][45][46]. However, some volcanically-generated glacier crevasses (e.g., those forming around the margins of ice caldrons or those forming related to subglacial dome growth) are often small and comparatively isolated (rather than widespread), meaning that individual landform identification can be challenging, and requires medium-to high-resolution imagery. In addition, when investigating glacierized volcanoes, being able to monitor crevasse development (e.g., crevasses appearing where they were formerly absent) and evolution (e.g., growth) is important. This generally requires imagery with both high spatial and temporal resolution, and is easier for glaciers that are not already heavily crevassed (see Figure 8).

Difficulties with Observing Volcanic Impacts on Glaciers
In this study, we have identified some key challenges of using optical satellite images to study volcanic impacts on glaciers. Often, extensive cloud cover and eruption plumes limit the temporal resolution of useful imagery for a specific volcano. This is a problem common to optical remote sensing of many volcanoes globally, not just those that are glacier-clad. One potential solution (though not the focus of this study) is to use SAR satellite data operating in the microwave region of the electromagnetic spectrum, which can penetrate clouds due to a larger wavelength used. For example, crevassing of the glacier surface yields an increased backscatter of amplitude SAR data used to identify ice cauldrons and meltwater tunnels on Mýrdalsjökull [47]. Supraglacial snow cover also introduces difficulties with observing volcanic impacts on glaciers from optical imagery, since details of the ice surface are often obscured. This is particularly relevant for volcanic events that occur during winter (or the snow season). However, in some cases, snow cover can be an advantage, since it creates a homogeneous surface of low contrast, making subtle changes in topography (e.g., ice surface cauldrons) easier to identify (see Figure 3).
In general, low-resolution optical imagery is useful for identifying and monitoring some larger volcanic impacts on glacier surface morphology, but is unable to resolve smaller features including some crevasses, cauldrons, ice domes and supraglacial channels. Observing volcanic impacts on small (thin) glaciers can also be particularly challenging, since these glaciers typically have steep, and heavily crevassed surfaces (due to the influence of steep and undulating bed topography), with frequent bedrock protrusions sometimes difficult to distinguish from glacial ice and/or volcanic extruding features. Given these issues with scale, detailed investigation of volcanic impacts on glaciers often requires medium-resolution imagery of at least~5 m. At present, this has cost implications, since freely-available imagery often is of low resolution. A further problem of image availability is that historical eruptions are typically poorly covered by satellite imagery. This applies most notably to events that occurred prior to the widespread availability of imagery from Landsat 7 and ASTER satellites (both launched in 1999). This leads to a strong temporal bias towards more recent eruptions (see examples presented in this study).
In addition to these challenges with image availability, it is also difficult to assess the required (or optimal) frequency of image acquisition for successful monitoring of glacierclad volcanoes. The timescales that apply for the formation of the features discussed in this paper vary considerably from several minutes to months/years. For instance, the formation of new ice cauldrons in the summit area of Eyjafjallajökull in 2010 took place within minutes [5], whereas squeezing of glacier ice and related crevasse formation on the Crater glacier during the extrusion of a volcanic dome on Mount St. Helens continued for about two years [37,38]. The time-frame for widespread crevassing of a glacier due to increased generation of basal meltwater (as observed on Mount Belinda) likely lies somewhere between those two endmembers. Therefore, in some quickly evolving instances images would ideally be acquired very frequently (~several per day), whereas other instances could, in principle, be sufficiently well monitored based on the common repeat cycle of 16 days for Landsat or ASTER, though issues with cloud cover and eruption plumes remain.
Finally, when viewing optical satellite imagery, it is sometimes difficult to discern the precise nature of volcano-glacier interactions. For example, supraglacial flow paths of volcanically-triggered floods and lahars are typically easy to distinguish because the debris deposited contrasts with the bright glacier surface. However, using satellite imagery alone, it is difficult to establish whether these events incised channels into the ice, or just flowed across the glacier surface. This also applies to supraglacial lava flows, which are often notably darker than the surrounding ice, but optical images provide little information about whether the glacier surface has been melted/incised. In this latter example, crevasses radiating from a lava flow's terminus (e.g., Mount Belinda, Figure 8b) can provide some indication that, through melting, the lava has had an impact on the glacier surface morphology (also see [35]). Part of this challenge in observing volcanic impacts on glaciers using optical satellite imagery is the lack of quantitative information on changing ice surface elevations (a key indicator of ice melt, fracture, doming, etc.). This clearly points to digital elevation models (DEMs) as potentially useful sources (though not the focus of this study), particularly when multiple high-resolution DEMs representing conditions before and after periods of volcanic unrest are available. For example, Rossi et al. [24] used preand post-event DEMs to identify and measure supraglacial ice cauldrons formed during the 2014-2015 Bárðarbunga eruption.

Guidelines for Best Practice
For the purposes of monitoring glacier-clad volcanoes, being able to detect volcanic impacts on the surface morphology of glaciers is important. With that in mind, here we outline a set of general guidelines for anyone undertaking (or wishing to undertake) this task using optical satellite imagery.

Site Overview and Image Availability
For a particular volcano, glacier and/or event, satellite image availability is a key factor determining which volcanic impacts on glaciers are identifiable. For historical analysis (rather than monitoring), this is primarily determined by the period of interest. For example, for older events (but still within the satellite era), e.g., those that occurred between 1972 (the launch year of Landsat 1) and 1999, imagery is mostly of low spatial resolution, and choice of satellites is very limited (e.g., Landsat 1-5 and SPOT 1-4) [39]. For events that have occurred since 1999, the range and resolution (spatial and temporal) of available images improves dramatically, particularly in recent years. Some free, but lowresolution, sources (e.g., Landsat 7, ASTER) are available throughout this period. Mediumand high-resolution images are also available (e.g., since the launch of IKONOS in 1999), but are typically commercial, rather than accessible for free.
Given the costs associated with obtaining high-resolution imagery, for the purposes of observing volcanic impacts on glaciers, lower resolution, but freely available, imagery can serve as a foundation. These images can then be supplemented with "targeted" and judiciously selected (i.e., covering specific areas and/or time periods) commercial imagery.
Free and commercial imagery is available via various websites, and some image types (e.g., Landsat and ASTER) are often easiest to access through platforms such as Google Earth Engine [40]. In some cases, freeware virtual globes (e.g., Google Earth and NASA Worldwind) and web mapping services (e.g., Bing Maps and Google Maps) can provide access to high-resolution images (with plugins available to export images into GIS packages such as QGIS). Thus, virtual globes and web mapping services allow access to high-resolution imagery, without the costs associated with commercial data. However, due to the limited availability of images, and the lack of user choice in image selection, these platforms rarely have images (and certainly not multiple images) covering a particular time period of interest (c.f., [13]), and are of limited use when considering highly dynamic environments (such as those associated with active volcanism).

User Experience
The ability to identify volcanic impacts on glaciers using optical satellite imagery partly depends on user experience. For most users (particularly those with comparatively limited experience), examining "known" (or documented) examples, such as those highlighted in this paper, is likely to be useful. This might involve observing volcanic impacts in the field, from existing ground-based or aerial photographs, or by looking at medium-to high-resolution satellite imagery. This is likely to be particularly useful for distinguishing between bedrock protrusions and features related to volcanic activity. Once familiar with the types of features that are present, previously undocumented volcanic impacts on glaciers can be searched for, using lower resolution imagery if necessary.

Using Multi-Temporal Images
Since observing volcanic impacts on glaciers effectively involves change detection (e.g., the development of glacier crevasses where they were previously absent), observing satellite images from a range of time periods (pre-, syn-and post-volcanic activity) is important. This is often possible by generating a time-series of images (in native resolution), e.g., using Google Earth Engine Code Editor, which requires some java script knowledge [40]. When producing such a time series, image availability (see Section 5.3.1) might be a limiting factor; however, where possible, images which show the glacier surface in its "natural" (i.e., non-volcanically impacted) state are key. This is particularly true of images from the same season (but perhaps a different year) as the period of volcanic unrest. How long before, or after, a period of activity images are useful/relevant is unclear, though some glaciers are known to have experienced changes in surface morphology months prior to eruptions (e.g., Mount Redoubt, 2009), and many impacts were still observable years after [13].
As the spatial and temporal resolution of imagery continue to improve over coming years, monitoring volcanic impacts on glaciers will likely be easier (particularly obtaining cloud-free images) and possible with greater detail than for historical events. In many cases, images are likely to be available on cloud-based platforms such as Sentinel Hub Playground (https://apps.sentinel-hub.com/sentinel-playground/, accessed on 18 March 2021), which allow data visualisation and manipulation (e.g., producing different band combinations) without having to download.

Mount Veniaminof: A Recent Example
Section 5.3, presented guidelines for best practice for monitoring of glacier-clad volcanoes. In this section, we apply these guidelines to document changes in glacier surface morphology on Mount Veniaminof (Alaska) during its very recent active period from March-April 2021. Emission of ash-plumes, (potential) lava effusion under the intracaldera glacier and subsidence of glacier ice were reported [33]. Both low spatial resolution Landsat-8 panchromatic images (15 m resolution) and medium spatial resolution Sentinel-2 images (10 m resolution) are sufficient to detect a newly developed ice cauldron and concentric crevasses in the glacier surface ( Figure 9). Due to repeated image acquisition (i.e., the 5-day repeat cycle of the Sentinel-2 satellites), ice-cauldron evolution could be monitored with a much better temporal and spatial resolution than most of the other cases presented in this work. This example clearly demonstrates the improved potential of recently launched optical satellites for monitoring of glacier-clad volcanoes.

Conclusions
Recent years have seen a rapid increase in freely available and commercial optical satellite images, covering large parts of the Earth. In this study, this global data source has been searched for evidence of volcanic impacts on glacier surface morphology, with a focus on well-known (documented) examples from the literature. The main conclusions are:

•
The most common volcanic impact on glacier surface morphology observable in optical satellite imagery is the formation of ice cauldrons (i.e., depressions in the glacier surface). Depending on eruption energy/type, ice thickness and image resolution, these cauldrons often appear as dark points/areas or as regions of distinct shadowing which contrast with the usually bright glacier surface.

•
Other observable volcanic impacts on glacier surface morphology are widespread glacier crevassing, localized crevassing (e.g., adjacent to supraglacial lava flows or ice cauldrons) and ice bulging and fracturing due to subglacial dome growth. • All volcanic impacts on glaciers investigated in this work are, to some degree, associated with crevassing, which is commonly visible as dark stripes on optical satellite images.

•
The key challenge when using optical satellite images to study volcano-glacier interactions is the availability of cloud-and plume-free images with sufficient spatial and temporal resolution to observe (and monitor) the evolution of (sometimes) small-scale changes in glacier surface morphology. Useful imagery is increasingly available, but this does not help when observing historical eruptions, and the widespread use of commercial imagery is usually prohibitively expensive. Lower resolution imagery is often freely available, but identifying and interpreting changes in glacier morphology from these sources requires greater user experience.

•
Overall, optical satellite imagery is a useful source for studying changes in glacier morphology caused by volcanic activity, particularly in remote and inaccessible parts of the world. This suggests that global-scale analysis and monitoring are possible. However, due to the costs of high-resolution (spatial and temporal) imagery, and the limited quantitative information that can be extracted, optical satellite imagery is best used in combination with DEMs, radar data, aerial images (derived from planes and/or drones) and ground-based observations.