Introduction

Lava spines are magma extrusions at volcanoes that allow rare insights into the conduit and conduit-bounding shear processes. Spines can form tall structures and gravitationally collapse, potentially generating pyroclastic density currents that affect surrounding regions. Understanding the directionality of spine growth and collapse is therefore highly relevant for hazard assessment and early warning.

Lava spines can evolve by extrusion of low-porosity and rather solid-behaving (competent) material. The extrusion is accomplished by an underlying magma-filled conduit that is geometrically constrained by enveloping shear faults1,2. Extrusion is commonly slow and associated with pre-eruptive outgassing and melt crystallization3. Lava spines evolve to spectacular features, such as the 305 m tall spine at Mount Pelée, West Indies4, the ~40 m tall whaleback spines at Mount St. Helens (MSH), Washington, USA, in 2004–20085, the spines at Soufrière Hills volcano, Montserrat, UK, exceeding ~40 m in 19966, the ~40 m spines observed at Mount Unzen, Kyūshū, Japan in 1990–19957, or the ~110 m high 1968 Nautilus spine at Bezymianny, Kamchatka (Russian Far East)8,9. Spines that extrude to reach heights >100 m usually quickly collapse under their own weight10, as observed for the 300 m high and 100 m wide spine at Mount Pelée in 1902. This collapse was potentially associated with pyroclastic density currents and followed by other episodes of spine growth4,11.

Spine geometry may strongly depend on magma extrusion rates and the resulting magma rheology. Vertical spines form at low extrusion rates (below ∼1 m³/s) because these ascent rates provide more time for outgassing and crystallization. This allows the magma to solidify and strengthen, making it capable of sustaining tall vertical structures12. In turn, higher extrusion rates lead to more viscous rheology and blockier dome morphologies. At extrusion rates of ∼3 m³/s, massive stubby lava lobes develop6,13. At even higher extrusion rates of ∼8 m³/s, pancake-shaped lava lobes form before developing into explosive activity at rates >9 m³/s14. Growth rates of a spine can vary considerably during an eruption episode, from cm/h to dm/h at MSH in 2004–200815,16. A better understanding of the relationship between extrusion rates and conduit processes may advance the development of physics-based models. For instance, models simulating the extrusion of a solid plug modulated by rate-dependent friction at the conduit walls, may allow assessments or even predictions of the evolution and duration of eruptions17. Any better constraints on these parameters will also improve model simulations.

Heterogeneity in the rheology might also be expected in the area of the conduit and its shear zones, e.g., in the host rocks through which the spine protrudes. This will further contribute to the shape and directionality of a growing spine. Such heterogeneities may be expressed through microstructural and textural differences, for instance, rock permeability can span a range of seven orders of magnitude12. The permeability inside the spine is isotropic while at the spine margin shear zone it is highly anisotropic, showing a dominantly high vertical permeability1. Such heterogeneity may also play a fundamental role in volatile escape that regulates the explosivity of a volcanic system, and speculatively for shaping the extruding spines and directing their rotation and destabilization.

The role of the underlying conduit geometry on lava dome and spine morphology remains debated, for example, inclined conduits may produce inclined extrusions18. The development of fractures on spine extrusions are also poorly understood, and thought to be related to fluid accumulation at depth or to near-surface gravitational effects19. Spine extrusion is accomplished by alleged conduit-wall rock shear zones, which slip in a brittle mode, where higher porosity may facilitate shearing at low pressures and greater depths3. Previous studies highlight that shearing at shallow depths mainly occurs in a brittle way20, such that the growth of spines is associated with characteristic drumbeat seismicity and conduit-margin fault gouges1,2,5,6,7,18,21,22. Observations of these features are challenging and remain rare, as geophysical instruments must be placed very close to a spine, and remote sensing techniques require very high resolution.

At MSH, rare detailed observations of discrete and stacked seismic events indicated slip (~5 mm) along small faults at the conduit margins2,5,18. Close analysis of time-lapse camera data at MSH suggests that spine growth was repeatedly interrupted by short–term meter‐scale downward displacements, indicative of a gravity-driven response19. Spines at MSH that could be studied in detail expressed a 1–3 m thick mantle of faulted rock with striated fault gouge, cataclasites, and angular breccia15,22,23. This is indicative of pronounced strain localization at the conduit walls, leading to the conclusion that spines (and whaleback structures) develop if material strength is high due to a magma that has lost its porosity12,24. Details on the timing of the strain localization and resultant decoupling of the spine from the dome have not yet been measured geodetically, however. The fault gouge is variably densified, which at slow magma ascent rates (such as during spine extrusion) is capable of producing stronger and less permeable cataclasites25. Extrusion, therefore, at MSH and elsewhere, involves brittle and ductile conduit materials depending on rock porosity and crystallinity, where ductile behavior becomes more relevant at slow extrusion rates and brittle behavior dominates at fast extrusion rates12.

These complexities during the growth and destabilization of spines must be well understood in order to better assess associated hazards. There is therefore an urgent need to find ways to safely monitor and better understand spine growth. In this work, we report on the extrusion of a spine at Shiveluch volcano, Kamchatka, Russian Far East, a volcano of dominantly andesitic composition26. The volcano is located to the north of the Klyuchevskaya volcano group and is affected by a regional SW-NE tectonic trend27. Shiveluch volcano is thought to have formed more than 80 ka before present (BP)28, and had at least 8 sector collapses, starting from 10 ka BP29. Shiveluch was active, then quiet for 15 years until a sector collapse occurred in 1964, followed again by a 16-year period of quiescence and degassing before activity resumed30,31. The 1964 sector collapse produced debris-avalanche deposits with a volume of 1.2 km³, that together with deposits of the associated Plinian eruption (0.8 km³) cover the volcano’s southern flank31 (Fig. 1). This eruption was one of the largest historical explosive eruptions of Kamchatka and left a deep amphitheater open to the SW. The amphitheater opening direction aligns with a debated SW-NE tectonic trend, traceable by an alignment of monogenic cones32, and by fractures and vents aligned high up at the summit33. The structural feature identified at the summit was interpreted as a fracture zone striking SW-NE and controlled repeated moderate-size collapses of a growing dome33. The subsequent eruptive activity changed to a spectacular episode of spine growth as described below.

Fig. 1: Spine growth at the Shiveluch volcano.
figure 1

a Satellite and helicopter image illustrating the scale of the Shiveluch volcano, its horseshoe-shaped amphitheater, the 1964 debris avalanche, and the 2005, 2010 and 2019 pyroclastic density current deposits on the southern flank (years of deposition shown in figure). b, c Photos from the south (top) and from the southeast (bottom) showing the whaleback spine on September 28, 2020 (photos by Yury Demyanchuk, see a for camera (Cam) location). Note the sharp cliff and fractures on the northern side, as well as striations on the whaleback. The map in a was created using ArcMap vs. 10.8.1.

Full size image

Results

Field observations and webcam imagery

Field photographs taken from a distance of 9 km with a DSLR camera and telephoto lens on the same day (September 28, 2020) reveal a rounded extrusion shape reminiscent of a whaleback when viewed from the SW, with striations aligned with the spine’s long axis (Fig. 1b). Side views (see also Supplementary Fig. 1) show that the northern portion of the spine developed tension fractures, open near a collapsing cliff surrounded by broad rocky talus (Fig. 1b). The first reports of changes were based on simple webcams (Fig. 2), which is a key observation tool for volcanoes in remote and inaccessible areas. Analysis of webcam images located ~43 km to the SSW of the volcano’s summit (Supplementary Fig. 2) revealed the growth of a lava dome and then a spine on April 6, 2020, which rapidly grew in height to over 100 m within 2 months by 9 June (Fig. 2). Maximum height tracking quantified an initially sharp vertical growth, which gradually slowed down after 18 June, accumulating a total height of 170 m by September 28, 2020.

Fig. 2: Spine growth tracking derived from webcam data at 43 km distance.
figure 2

Tracking results suggest a spine height growing quickly between March and June 2020, reaching 150 m in June (i.e., the first three months), and then only slowly increasing to 170 m over the following three months.

Full size image

Radar and photogrammetry observations

We closely monitored the Shiveluch spine extrusion by satellite radar and photogrammetry. Radar waves backscattered to the TerraSAR-X satellite allow monitoring of the dome and spine evolution at approx. 1 m pixel (px) resolution (Fig. 3). As the radar provides its own illumination, the sensor operates independent of sunlight and even penetrates ash and steam clouds, therefore providing a temporal resolution not approached by any other methods (Supplementary Fig. 3). The radar reflections confirm the emergence of a spine on April 24, 2020 (Fig. 3c), which then gradually gained in dimension and extended toward the north. Analysis of the pixel offset helps to identify the deformation field associated with spine formation. We identify a deformation area affecting a large part of the volcanic cone two months prior to spine extrusion, with 6–8 px (about 6–8 m) displacement within an 11-day repeat pass (Fig. 3a, b). Associated with the first emergence of the spine on April 24, 2020, the displacement area sharply decreased (see also Supplementary Fig. 4). Widespread deformation of the dome is virtually absent during spine extrusion, indicative of a mechanical decoupling of the spine and the dome.

Fig. 3: Radar amplitude observations.
figure 3

Color represents pixel offsets calculated at cross correlation windows of 32 × 32 pixels minimum, for data pairs of single, double or triple repeat pass (with 11 day, 22 d, and 33 d time window). Displacement U is shown in pixels [px]. a Up to 7 px displacement affects the summit region. b Displacement increases to 8 px in 11 days (d). c With first spine extrusion, the large displacements disappear. d Minor displacement of the dome, while the spine extrudes. e Spine extrusion continues, migration northwards. f The spine widens and/or bends further to the NE. Note the occurrence of doming first, then spine extrusion (without doming). Images shown in radar coordinates (azimuth in x-axis, range in y-axis), 1 px is 1 m² approximately, descending TSX track 11 spot 042 HH. North to the left, east is up, west is down.

Full size image

Helicopter aerial and Pleiades satellite stereo-photogrammetric data were acquired before and during spine growth. They provided decimeter-scale resolution, and allowed us to generate georeferenced orthophotos and digital terrain models (DTMs) that showed morphological details during three episodes of growth. Before spine growth, on October 22, 2019, the surface of the lava dome hosted small craters enclosing a 500 m wide depression (Fig. 4a). These craters together with a steep collapse scar to the eastern side developed ~1 year prior to the spine growth33. During spine growth, the first clear sky tri-stereo satellite image was acquired on October 01, 2020, revealing an SW-NE elongated extrusion body, 2684 m ASL, ~150 m wide and ~300 m long, with a 110 to 150 m wide talus at 2570 m ASL (Fig. 4b). The polished flat surface of the spine is well visible in satellite data, which is up to ~230 m long and 150 m wide in the Pleiades imagery. This agrees with estimates we also derived from Planet Labs satellite images, showing an increase of the polished spine surface long-axis, with 154 m on August 6, 2020, 171 m on August 16, 174 m on September 4, and 248 m on October 13, 2020 (Supplementary Fig. 5). Three-dimensional point cloud reconstruction (Supplementary Fig. 6) shows that the elongated spine is 114 m above the talus and 220 m above the previous 2019 crater, and at its highest point even 300 m above the 29 Aug 2019 collapse scar (Fig. 5). Another satellite dataset acquired 12 days later visually reveals only minor changes, but is particularly useful for analyzing late-stage spine extrusion rates.

Fig. 4: Digital terrain models (DTM) produced from stereo imagery.
figure 4

a A DTM on October 22, 2019 (before the spine growth) reveals the presence of a number of summit craters roughly aligned along the SW-NE structural trend defined in ref. 33. Note the deep collapse scar that formed in 2019 on the eastern part of the edifice. b DTM on October 01, 2020, shows the presence of a spine, partly overflowing the 2019 collapse scar. c DTM on October 13, 2020, shows continuous spine growth, yet cloud artifacts in the northeast. Section locations shown in c. Map created using ArcMap vs. 10.8.1.

Full size image
Fig. 5: Temporal evolution and dimension.
figure 5

Topographic profiles through the DTM data in a. SW-NE direction (profile A-A’) and b in NW-SE direction (profile B-B’) reveal the spine height is 220 m above the pre-existing topography, where a large amount of the elevation change is due to dome formation and talus materials. Note that the vertical cliff of the spine co-locates with the buried cliff of the 2019 collapse scar. Data highlights major topographic complexity prior to dome and spine growth, but also contains local data artifacts, e.g., at the NE part of profile A-A’.

Full size image

The Pleiades tri-stereo satellite images used to generate topography data (see Fig. 4b, c) were also used to study small scale displacement and fracture formation during spine extrusion (Fig. 6). Both images highlight the presence of along-axis striations and mainly across-axis open fractures (seen by dark and shadowed cracks). The striations are aligned with the spine’s long axis. Closer inspection of the striations reveals that they strongly resemble slickensides, with smooth rock surfaces and parallel grooves and scratches. The striation alignment is evident both in orthophotos and in topography, with ridges up to 2 m high, regularly spaced at 4–5 m at the spine apex, and larger 8–14 m spacing at the spine flanks (Supplementary Fig. 7). Similar to striations found elsewhere (Supplementary Figs. 8, 9), their documented presence at the Shiveluch spine characterizes pronounced shear localization and a kinematic decoupling from the dome. In this 12-day period, we further find an offset of fractures that evolved orthogonal to the striations. Existing fractures enlarged, and new fractures developed in the October 13, 2020, image (Fig. 6a, b). Pixel offset calculations suggest a mean displacement for the spine of 20.4 m in the 12 days (see inset in Fig. 6b), or a mean of ~1.7 m daily. Assuming this represents the extrusion rate and the satellite nadir-view represents the conduit outline (diameters circular 150 × 150 m², elliptical 150 × 300 m²), the extrusion volume was 0.3–0.7 m³/s, which is comparable to the rates observed at spines elsewhere6,13,14.

Fig. 6: Displacement and fracture analysis.
figure 6

The 2020 Shiveluch whaleback spine, shown for two satellite image datasets (October 1 and 13, 2020). a Striations have developed in a NE-SE direction, sub-parallel to the N54°E azimuthal elongation direction of the spine. Fractures show opening perpendicular to the elongation axis. b The following satellite data (12 days later) reveals similar striations, but fractures have been displaced to the NE by a cumulative mean of 20.4 m, as determined by pixel tracking. c, d Striation density analysis reveals clustering at the lower SE flank of the whaleback, with orientation NE-SW shown by the 48 polyline features weighted by length to 2940 elements, the mean directions at 56° in rose plot, and an angular deviation of only 5.1°. e, f Fracture density concentrates near the spine collapse region at higher elevations (above 2660 m). Statistics and rose plots show the 72 features weighted by length to 1649 elements, the mean directions at 132° with an angular deviation of 59.6°. Pleiades TriStereo data was analyzed in ArcMap vs. 10.8.1., rose diagrams generated using Polar Plots and Circular Statistics60.

Full size image

Our deformation analysis suggests a mechanical decoupling of the spine from the dome, with high displacements of the spine and very low displacements of the dome (Fig. 3). The presence of striations may suggest a high strain rate, and a polished spine surface may indicate fault gauge formation. Our lineament analysis also reveals striations located on the lower (southwestern) proportion of the whaleback. These are constant in both datasets with a striation azimuth of N040°−060° (Fig. 6c, d) for weighted polylines of 5651 and 5816 m cumulative fracture lengths for the October 01 and October 13 images, respectively. The analysis of the fractures dissecting across the spine reveals azimuths N120°−155° estimated for 7188 and 5816 m cumulative fracture lengths for the October 01 and October 13 images, respectively. Most fractures are located near the steep cliff in the NE on October 01, 2020, and are bimodally distributed near the lower plateau and also near the steep cliff on October 13, 2020 (Fig. 6e, f). The upper fracture region may be associated with rock falls and unloading of the frontal steep slope of the spine. The lower region of fractures is more evident in later images, and may be associated with bending or tilting of the spine as a whole. Therefore, the two satellite images reveal the occurrence of both fractures and striations, as well as the presence of a cliff and collapse of the current spine above an underlying collapse scar.

Modeling directed conduit extrusion

To study the dynamics and asymmetry of spine growth and NE-directed collapses and to understand the mechanical decoupling of the spine and the dome, we designed both numerical and sandbox models (Fig. 7a, b). The numerical models were designed in 2D and well constrained, while the sandbox models were in 3D and were less well constrained but allow natural variations in the used materials. Both model setups were complimentary so that an initial doming episode is followed by a spine-extrusion episode. A necessity for the transition from normal dome growth to spine extrusion is the extrusion of a highly competent and cohesive material24. We tested a large set of geometric and dynamic models, with end members ranging from homogeneous conduit material with horizontal topography, up to heterogeneous conduit material and topography-affected models (see Supplementary Figs. 10, 11). Generally, the models begin with extrusion of a blocky material and endogenous growth of the dome, with marginal slopes depending on the angle of repose and a mound-shaped dome top. Then the spine protrudes on top of a mound-shaped dome. Talus slopes continue to approach the same internal friction angle so that additional slumps on the margins are observed, while the central spine remains symmetrical and vertical. Once shear fractures delimiting the spine evolve, the dome deforms at a reduced rate. The difference in surface textures between the spine and the dome could result from material heterogeneity, but could also relate to changes in magma-conduit wall dynamics when more competent rock is extruded.

Fig. 7: Computational particle flow and analog modeling.
figure 7

a 2D PFC model and b analog sandbox model, both showing high and low cohesion material extruded from a conduit, with a stable buttress to the left side of the conduit exit. c, d Velocity vectors showing primarily upward movement above the conduit, with lateral flank motion. Spine emplacement can be seen at the top of the dome, where the cohesion contrast is forcing direct lateral motion. e, f Strain accumulation showing fault formation broadly in line with the conduit exit. Data in f is displayed as vorticity, which represents shear strain.

Full size image

We find that spine extrusion can become asymmetric, or directed, if we consider stiffer material on one side, and weaker material on the other side of the conduit (Fig. 7a, b). Although this situation is hypothetical, we note that the extrusion models that consider both topographic complexity and a heterogeneous conduit material, are able to well reproduce directionality and growth of a whaleback-like spine. To simulate spine growth over a collapse scar (such as the 2019 scar at Shiveluch, Fig. 7a) we consider topographic effects and repeat the models. We now find an asymmetric talus, with the plateau on the buttressed side being higher than the unbuttressed side (Fig. 7). We also see a preferred lateral growth in the upper dome regions towards the unbuttressed side (Fig. 7c, d). Closer analysis of shear localization reveals that the spine is delimited by well-defined shear faults and initially extrudes vertically. In models, shear localization is most apparent on the buttressed side of the conduit, where striations and clear whaleback planes can be expected. With continued spine extrusion, the shear faults located closer to the buttress lead to mechanical decoupling of the spine, and the general extrusion becomes asymmetrical (Fig. 7e, f) with the spine increasingly inclined as lateral displacement dominates over vertical displacement.

Consequently, although the models are simplified, they highlight the importance of pre-existing topography and/or heterogeneous conduit material in controlling whaleback-like spine extrusion and the evolving internal deformation of a dome. We note that the relative scale of the modeled spine is smaller than that of Shiveluch. At a critical height, the spine partially collapses on one side (this can be both the buttressed or unbuttressed side), resulting in a detached and laterally inclined spine surface. It also shows a large portion of the spine to still be hidden in the shallow subsurface. With the emergence of the dominant shear faults, the internal dome deformation is rather small, which is in agreement with decoupling observations made at Shiveluch volcano.

The overall topographic expression of the models (Fig. 7) well matches the observed deformation (Fig. 3) and topographic profiles (Fig. 5), where the vertical cliff on the unbuttressed side co-locates with the buried collapse scar from 2019.

Discussion and conclusion

This study highlights that spine extrusions can provide a unique insight into hidden magma pathways and conduit geometry. The lifetime of a spine is short, because the structures are inherently unstable and often destroyed by gravitational collapse and/or volcanic explosions. Consequently, only a few spines worldwide have been investigated in greater detail, but the new Shiveluch activity provides a rare opportunity to study mechanisms controlling the extrusion and morphology of the spine. We studied the first episode of spine growth in 2020, which has collapsed by the end of 2020 and restarted again in early 2021. The Shiveluch lava dome and 2020 spine growth show a directionality towards a pre-existing depression that formed during the 2019 eruption and partial collapse, leaving a sharp and deep ravine. Previous work hypothesized that non-vertical spine extrusion occurs due to inclined conduits at depth beneath the MSH spine18 or variably inclined conduits beneath the Soufrière Hills spine6. Our study also suggests that inclined spines are affected by topography.

Such directionality of spine growth and associated collapse was also observed during spine extrusion at Mount Pelée34. A reassessment of the old records and topographic assessments at Mount Pelée4 indicate a strong asymmetry of the growing dome and collapse direction, with steeper slopes (40–50°) to the NE and shallow slopes (15°) to the south, thus a similar topographic control may have played a role. The southern direction was open and lavas gradually flowed or crumbled into a pre-existing deep valley of the Rivière Blanche4. Our analysis reveals spine collapse at a location where underlying topographic kinks in slope were present. Therefore, detailed knowledge of the previous topography is essential to understand the direction of collapses and pyroclastic hazards at Shiveluch, and to understand past events of directed spine growth elsewhere. Additionally, our modeling reveals that such a topographic configuration is directing broad internal shearing and lateral motion of materials towards the unbuttressed side.

Our kinematic modeling considering topographic and material effects convincingly explains the deformation, location of striations, and directed growth and collapse of the spine. Data on the morphology and extrusion rate of an active spine on a growing lava dome can be interpreted in terms of the parameters characterizing the shallow magma system which feeds the dome and spine. However, such interpretation strongly depends on the selection of the conceptual model of the magma-feeding system of the dome and spine. Currently, there are two principally different conceptual models of the magma systems feeding dome-forming eruptions: (1) a classic vertical or inclined cylindrical conduit, which is continuously filled with ascending magma35, and (2) a feeding system composed of several elongated magma batches that rise one-by-one36. If we hypothesize that the second model is valid and serves to explain the episodic dome growth at Shiveluch, an open question remains regarding the length and volume of the ascending elongated magma batches. Based on our measurements, we estimate that the length of the extruding and solid spine accumulates to as much as half a kilometer in length (excluding the dome material that erupted before). Furthermore, a considerable magma volume was added during the dome forming eruption, which may provide a first-order idea that the elongated magma batches are on the kilometer scale. Shortly after the observations reported here, a pronounced extrusion pause occurred, followed in 2021 by another extrusion event of similar duration and scale.

We used a combination of field cameras, satellite radar backscattering amplitude images, and stereo photogrammetry to characterize the evolution of an extruding spine and quantify its growth, fracturing, and instability which are mostly reproduced by both analog and numerical models. The ascent rate of magma in the conduit (or, accordingly, the ascent rate of the uppermost magma batch) is about 1.7 m/day. Hence, the average discharge rate (0.3–0.7 m³/s) of magma during the period of observation was up to 6 × 104 m3/day. The upper part of the conduit is filled with highly viscous magma that probably still contains some melt phase that allows the spine to deform plastically under its own weight to obtain the whaleback or nautilus appearance.

The great spine at Mount Pelée that extruded through the dome from 3 to 4 November 1902, assuming it had not suffered collapses, would have accumulated to a cylinder that measures 850 m in height and 150 m in diameter34. The 2020 spine at Shiveluch grew 135 m within the first 3 months (Fig. 2), which translates to a daily vertical growth of 1.5 m. Later analysis in October suggested a mean displacement of 20.4 m within 12 days, translating to a daily displacement of 1.7 m. Assuming these numbers represent a constant growth rate since March 2020 and assuming no material is lost by collapse and bending, the spine could have reached 315–590 m in height by October 2020. This highlights that there is (i) a significant loss of material by collapses and (ii) the depth extent of the uprising magma plug is up to half a kilometer depth. A similar rapid initial vertical growth, followed by a decaying vertical growth rate (as shown in Fig. 2), has been observed during previous dome and spine emplacement episodes at other volcanoes (e.g., St. Vincent37; Volcán de Colima38; Soufrière Hills Volcano39), as well as being well replicated by previous particle-based numerical models (similar to those used here) that use a cohesive material38.

We have inferred that the upper part of the conduit of Shiveluch has a circular, or maybe oval, cross-section with a diameter of about 200 m. This may be a simplification, as satellite imagery suggests the whaleback is elongated SW-NE, which is also the same alignment of a structural trend described earlier33. For instance, located only 400 m to the NE of the spine is an explosion crater, aligned with the structural trend. The TerraSAR-X data reveals a gradual shift of the spine width and location towards the NE, which could be indicative of a rather elliptic conduit.

The Shiveluch spine has some striking geometric and structural similarities to some of the seven spines that were sequentially extruded during the 2004–2008 eruption at MSH15,40. All of these spines were bounded by a 1–3 m thick gouge zone15 and extrusion was accompanied by drumbeat seismicity2 and small sporadic volcano-tectonic earthquakes41. At Shiveluch, seismic networks were likely too far away or not sensitive enough to record seismicity. Our modeling shows extrusion of spines through a dome if the material is stiffer or mechanically more competent (cf. ref. 24). This is in line with mechanical studies at MSH suggesting that the strength of lavas increased during the 2004–2008 eruption, becoming more crystalline and less porous, so that shear zones could localize42, fault gauges develop and spines evolve43. This is further supported by our analog models, where we can directly observe spine formation as a result of frictional faulting originating from the conduit walls. Striations and slickensides on an exposed face of a spine demonstrate contact of the magma plug with the conduit wall44. Such structures are well described at MSH2,45 or at the Unzen spine46, commonly developing asymmetrically at the side exposed and depicting the whaleback. This structure is well reproduced by our models, which reveal a main shear zone on the side of the buttress and topographic high.

Observed fractures are consistent with extension as a result of bending of the spine, suggesting less brittle (or even partially viscous) behavior of the inner spine material. The fractures are located at the upper parts of the spine and are mostly oriented parallel to the collapse scar. We note that material heterogeneities of the Shiveluch spine are speculative, highlighting the need for more rigorous mechanical testing in the future. Different material strengths can result from compositional or textural variations, for example as a result from gas pressure or alteration-induced porosity changes47. While we recognize in our models that topography alone can explain an inclined extrusion direction of a dome and/or a spine, we find a similar or contributing effect results from heterogeneous material within the conduit. One may hypothesize that such material heterogeneity could be explained by differential degassing rates throughout the dome. Commonly, the conduit margins are highly permeable1, so the magma at the margins can degas more effectively compared to magma located towards the center of the conduit. Additionally, heterogeneity may arise from previous eruptive history. For instance, we find that the spine is located to the S and SSW of previous eruptions and craters as described earlier33. Therefore the N and NNE side of the conduit and extruding spine is at or close to the location of an earlier conduit, which may have shattered or thermally affected the surrounding rocks.

Spines grow and collapse quickly, potentially leading to the formation of dangerous pyroclastic density currents. It is for this reason that we need to know more about their growth, as well as ways to safely monitor their growth (using remote sensing techniques, for example). Analysis of the Shiveluch spine may help to understand past volcanic activity and directionality of growth and collapse elsewhere. The collapse of the spine at Mount Pelée in mid-1903 produced a lethal pyroclastic density current34. With the new understanding of the relevance of previous topography and growth style, future spine extrusions may be more closely monitored with necessary actions being implemented to assess their hazard. It is important to precisely and safely measures the topography prior to spine formation, for which similar remote sensing techniques shown in this study may be used. For most of the 200 dome-building volcanoes worldwide, this information is still lacking and not regularly updated.

Methods

Time-lapse camera analysis

A camera was set up in Klyuchi town at 56.31888° 160.85687° at a distance of 43 km from the Shiveluch dome. The camera is a 3-megapixel Internet Stardot Netcam XL, taking images at 1024 × 768 px resolution every 1 min. Due to the large distance of the camera, the horizontal field of view is 7 km, translating to a 1 px dimension of 6.8 m. In other words, a spine growing to 100 m would be visible in only ~15 px. Most of the images are not usable due to poor visibility. Despite this and the low resolution, we find the information extracted for the growth rate is important. We first align the images using a digital image correlation approach48. The image stack was then preprocessed by brightness adjustment, normalizing the brightness of each image relative to the first image. This way, illumination changes are significantly reduced. Third, we apply a manual pixel tracking method, by picking and tracing the highest peak identified as the lava spine. Lastly, we converted all pixel units to meter scale, by picking locations of known distance in the field of view. Results show changes in elevation of the lava spine from March to September 2020. To account for possible measurement uncertainties, we repeated the tracking four times at different starting points, yielding a maximum vertical rise of 170.8, 167.5, 172.5, and 172.5 m, respectively, with a mean of 170.8 m and a standard deviation of 2.35 m. Uncertainties may be even higher, however, as these measurement errors do not take data errors resulting from camera shaking, optical deficiencies, and other artifacts into account.

Satellite radar observations

We used the synthetic aperture radar (SAR) data acquired every 11 days by the German radar satellite TerraSAR-X. The satellite is able to acquire data in high-resolution spotlight mode, which provides us with an acquisition type yielding a spatial resolution close to 1 m per pixel (px), which is about one order of magnitude better than most other radar data available. For Shiveluch we found a major advantage of the SAR technique was that it is an active sensor, thus the ground is illuminated by the sensor independent of day or night. The X-band wavelength passes through eruption clouds, and hence allows views of the cratered landscape, which is especially useful during a volcanic crisis49. We analyzed the acquisitions mainly in descending mode, with an 11-day repeat pass. Due to the complex topography and the side-looking radar line-of-sight, a geometric distortion may affect not only pixel locations but also the offsets of pixels. Accordingly, even symmetric deformations may appear asymmetrically. We note that ascending mode data also exists and was analyzed, but due to the large geometric distortion the growing spine is located in the foreshortening region (see Supplementary Fig. 3). An interferometric technique (InSAR) was not found to be valuable due to ash coverage, rapid reflectivity changes, and resulting decorrelation. Therefore, we exploited the amplitude information38; we stacked and coregistered the images using the GAMMA software. As the satellite sees the backscattered electromagnetic microwave radiation, each reflected pixel represents the proportion of power relative to the energy scattered and transmitted back. The X-band (wavelength of 31 mm and frequency of 9.6 GHz) amplitude depends on surface geometry, therefore the morphology and associated changes of the topography is expressed by amplitude images50. The measured amplitude accordingly depends on the sum of all scatterers within a ground resolution pixel. To avoid geocoding warping artifacts in the steep summit region, we analyzed the amplitude in radar coordinates. We used a particle image velocimetry (PIV) approach24,48 to quantify offsets of groups of pixels in this amplitude image dataset.

Stereophotos and digital terrain modeling

We acquired and analyzed high-resolution optical images taken from a helicopter on October 22, 2019 (before the spine growth) and by the Pléiades satellite PHR1B sensor on October 01 and 13, 2020 (during the spine growth). The acquisition and photogrammetric processing of the 2019 aerial images is described in detail in ref. 33. The tri-stereo Pleiades images were made during a single satellite pass, so that panchromatic 1 m resolution imagery could be used to develop orthophotos and DTMs. The data acquisition timing is mainly dependent on concurrent satellite passes and a (rare) clear sky condition. Processing of the data was realized in Erdas Imagine 2015 v15.19, where we used between 30 and 50 tie points for the relative orientation. A conversion from pixel scale to metric scale was employed using the Rational Polynomial Coefficients block adjustment approach. After the image orientation, we obtained photogrammetric models with a total root-mean-square error (RMSE) of 0.2 m. By using the Enhanced Automatic Terrain Extraction module (eATE) with normalized cross-correlation algorithm as implemented in the Erdas Imagine software, we were able to extract 2 m resolution point clouds referenced to the WGS84 coordinate system UTM57 zone. These point clouds were filtered with the CloudCompare v2.9.1 noise filter and then manually cleaned with the CloudCompare segmentation tool. The derived orthophotos and DEMs were used in the geoinformatics framework (ArcGIS 10.2.1) for analyzing shapes, azimuths, striations, and fractures. We manually traced all features and calculated the azimuthal direction for each lineament, necessary for rose diagram representations. We further computed the density of lineaments searching for the number of lines in a 70 m radius area, yielding up to 70 lineaments/area. The two orthophotos on October 01 and 12, 2020 were further used for manual tracking of relevant features and pixels, allowing us to derive the displacement in the field of view.

Modeling

We performed two complementary modeling strategies. Analog modeling was performed by extruding mixtures of sand and plaster powder onto a custom-built table. Experiments were recorded at a 30 s interval with a single camera facing horizontally onto a Plexiglas plate, enabling the observation of the dome and spine extrusion in a profile view. In separate tests, we found that boundary effects between the extruded sand and the plexiglas plate are apparent but negligible. The extrusion speed was set to 1.75 mm/min, although in the absence of viscous material the extrusion rate does not affect the extrusion and allows for a mainly geometric scaling24. To test the effect of heterogeneous material within the conduit, the experiments were modified to extrude cohesive and strong material from the conduit, corresponding to dense, degassed and brittle magma. Here, the sand mixture contained 30% plaster for the cohesive conduit24. In order to better reflect the local topography around the Shiveluch spine, one side was additionally buttressed by placing a barrier next to the conduit wall, thus preventing equal deposition on each side of the dome. To evaluate the experimental results, we employed Particle-Image-Velocimetry (PIV) on sequential images using the LaVision DaVis software package (version 10.0.5). Motions in the image field were extracted using multiple passes by applying a decreasing window size of 64 × 64 px at 50% overlap and 24 × 24 px at 75% overlap. This enabled the visualization of displacement vectors, particle strain, and vorticity to understand movement and structures during the dome and spine growth.

Numerical modeling was performed using commercially available software Particle Flow Code (PFC) from Itasca Consulting Group Ltd., and the method primarily follows that described earlier by51,52, without distinct material behavior for a fluid core and a solid talus (because observations suggest spine material is so competent that it exhibits solid-like behavior). A two-dimensional discrete element method (DEM) is used53 in this software. We consider heterogeneous conduit material with a cohesion on the left of 4 MPa and on the right of 1 MPa. Incorporation of cohesive strengths for the magma allows for simulation of non-Newtonian magmatic behavior54,55. The lower cohesion value is in line with estimates of bulk material cohesion for endogenous domes56, whilst the higher cohesion value is artificially increased in line with estimates that spines require much more cohesive material in order to maintain observed heights57. To construct the numerical model geometry, a topographic profile from 2019 is used, with an additional buttress next to the conduit wall. The boundary conditions in the model are frictionally-controlled at the interaction between the dome material and the topography, meaning that this material resists rolling along the topography. Conduit dynamics (e.g., frictional slip at conduit margins) are not simulated here. Results from the DEM modeling are visualized in two ways: (1) by velocity vectors, similar to the PIV approach used for the analog models; and (2) the computation of normalized finite shear strain. The finite shear strain is computed by inverse strain modeling58,59 and calculates the Cauchy–Green deformation tensor to compute maximum shear strain. This method highlights areas where groups of particles move as a coherent block relative to their neighbors, such that it picks out planes of failure or shear bands.