Responding to eruptive transitions during the 2020–2021 eruption of La Soufrière volcano, St. Vincent

Network strengthening

Overall emergency management and scientific support for La Soufrière volcano were coordinated by the staff of The University of the West Indies Seismic Research Centre (UWI SRC) in Trinidad, with assistance from the Montserrat Volcano Observatory (MVO) as well as regional and international collaborating partners. Initial observations and local support were provided by the SMU.

Limited resources and the COVID pandemic resulted in a much-reduced monitoring capacity at the onset of unrest (November 2020), with one working seismic station (SVB) 9 km from the volcano, and one continuous GPS station (SVGB) (Fig. 1). SRC reactivated the local observatory in late December 2020 and upgraded the monitoring network (Fig. 1). Eight broadband seismic stations were operating by the end of February 2021 and the ground deformation network was augmented by four continuous GPS sites (SVGR, SVGS, SVGF, SVGG) in addition to re-occupation of two campaign benchmark sites (JCWL and TBRK) (Fig. 1). A 9-prism EDM target was installed on the southern crater rim and weekly measurements attempted from six locations. Interferometric Synthetic Aperture Radar (InSAR) processing of available ALOS-2 and Sentinel-1 images augmented ground deformation monitoring. Sentinel-2 and PlanetLabs satellite imaged both the crater and colour changes of vegetation on the volcano’s flanks. Cameras installed at the Belmont Observatory (3 January 2021) and crater rim (24 January 2021), multispectral and radar satellite imagery, oblique aerial and terrestrial photographs and UAV aerial photography and photogrammetry allowed visual observations to document dome growth. From 14 January 2021, gas emissions were measured using a Multi-component Gas Analysing System (MultiGAS) and Ultra-Violet (UV) spectroscopy.

Seismic monitoring

Seismicity increased slightly on station SVB in November 2020, but remained modest until 23 December 2020, averaging two events per day, with a maximum magnitude of 3.3 Mt and no reported felt events (Fig. 2).

Although dome extrusion started on 27 December 2020, no seismicity was recorded until 6 January 2021, with an average of two events per day up to 17 January, when there was a sharp increase to 60 events per day. Subsequently, low frequency (0.5–5 Hz)^10,11 events were observed and interpreted as related to the dome emplacement; the events were recorded only by the closest stations indicating a shallow source. Volcano-tectonic (VT) swarms occurred during 23–24 March 2021 (226 events) with >95% located at depths shallower than 5 km; and 5–6 April 2021 (476 events) with an abrupt transition to deeper locations (Fig. 3), which was interpreted as a new volume of magma ascending from ~10 km depth.

Banded tremor¹², of increasing magnitude, began around noon (UTC time) on 8 April 2021, recorded by the closest stations at intervals of ~2.5 h (Fig. 4a). This change was interpreted as indicating an imminent explosive phase¹³, with a source attributed to the excitation of shallow gas and fluid pockets^14,15. The spectral content up to 10 Hz suggested that the banded tremor consisted of merging VT events (Fig. 4a). The 8^th cycle transitioned to continuous tremor with increasing amplitude and stable frequency content over time (Fig. 4), suggesting repetitive events at a constant rate¹³. The first explosion was recorded at 12:41 UTC on 9 April, followed by a period of sustained, but pulsing, explosive activity and tremor from 16:00 UTC on 9 April to 06:00 UTC on 10 April (Fig. 4).

The time series, RSAM and spectrograms of the explosive phase is shown in Fig. 4. The initially rapid rate of explosions and associated tremor made individual seismic events difficult to identify. Four stations stopped transmitting data during the first 36 h of the explosion sequence. Spectrograms for the explosive phase (Fig. 4a) show a larger amplitude but the same stable frequency as during the build-up. Each explosion lasted between 3–23 min, followed by 2–3 h of exponential decay in tremor amplitude (green arrows in Fig. 4a). Over the following two weeks, the pattern of seismic activity included episodes of short tremor bands accompanied by enhanced venting or explosive activity. Episodes of tremor were interspersed with long-period and hybrid earthquakes, with their rates of occurrence gradually decreasing prior to a period of high-level tremor on 22 April. The last explosion on 22 April was preceded by several hours of increasing-amplitude tremor (Fig. 4b), with an abrupt end of low-frequency tremor shortly after the explosion. Seismic activity steadily declined from 22 April to early May, from an average of 354 events/day to 24 events/day. Up to November 2021, the seismicity remained sparse, dominated by low-frequency events.

Ground deformation monitoring

Forewarning of the effusive eruption was not recognised on the existing continuously operating GPS network (Fig. 2). However, a <10 cm line-of-sight shortening signal was observed in the crater area using ALOS-2 and Sentinel-1 radar, sometime between 19 and 31 December 2020. The associated deformation source was modelled as a ~63,000 m³ dike intrusion, shallower than ~500 m deep (Fig. 5). Subsequently, no deformation was detected from the SAR platforms. No deformation was detected on the EDM time series during the effusive phase.

Onset of the explosive phase was accompanied by a rapid deflation recorded on the continuous GPS network on 9 April 2021 (Fig. 2). Between 9 and 22 April 2021, the SVGB station (Fig. 2) measured an overall cumulative horizontal displacement of ~43 mm northward and ~37 mm eastward and a subsidence of ~81 mm. Using a Mogi point source¹⁶, the associated surface deformation was modelled by migration of ~50 × 10⁶ m³ of magma from a source at ~6 km depth. After explosive activity ended, slow deflation was observed over several months.

Gas and geochemical monitoring

Gas measurements in January 2021, using UV spectrometer and MultiGAS instruments_, detected no SO₂. The concentration ratio (ppm) of carbon to total sulphur (C/S_t) was measured by MultiGAS at the summit (Fig. 2). A C/S_t (=CO₂/H₂S) ratio of 85 and 30.6 (Fig. 2) was obtained on 14 and 15 January, respectively. Minor SO₂ (<1 ppm) was detected in February 2021, with a C/S_t (=SO₂ + H₂S) concentration ratio of 10 measured on 1 February and 11 on 18 February, before increasing to ~20 on 23 March 2021. Plume compositions during the effusive phase were dominated by a hydrothermal signature (Fig. 6). On the afternoon of 8 April, a coastal traverse yielded the first detection of SO₂ in the gas plume with a mass flux of 80 tonnes/day. The TROPOMI instrument on board Sentinel-5P also detected SO₂ during an overpass on 8 April at 17:25 UTC, confirming the change in plume composition. During the explosive phase (9–22 April) only satellite (Sentinel-5P) SO₂ measurements were possible, with values ranging from 2.76 × 10⁵ tonnes/day on 10 April to 331 tonnes/day on 22 April. Over the two months following the explosive phase, coastal traverse measurements of SO₂ flux decreased from ~800 to ~200 tonnes/day, and then maintained this average through to November 2021.

On 16 January 2021, samples were collected from the front of the lava dome. In April 2021, scoria and clasts from pyroclastic density currents emplaced during the explosive eruptions were sampled. XRF analysis of major elements shows both have basaltic andesite bulk compositions (Fig. 7).

Preliminary petrographic analyses of the dome rocks indicated a phenocryst assemblage similar to past eruptions, consisting of plagioclase, clinopyroxene and Fe-Ti oxides with sparse olivine and abundant gabbroic clots¹⁷ (Fig. 8a, b). Where present, olivine is invariably heavily altered, with symplectites intergrowth of forsterite and Fe-oxides common (Fig. 8b). Groundmass textures show evidence of late-stage disequilibrium, including groundmass crystals with localised alteration of orthopyroxene microlites to Fe and Mg oxides (Fig. 8c). Dome rocks are vesicular with textural evidence for fracturing and annealing of fluid pathways (Fig. 8a, d).

Dome growth and other visual observations

Initially, the new dome grew uniformly in all directions, reaching 70 m in height, subsequently elongating in the NW-SE direction (Fig. 9). Gas vented through a small depression in the dome’s summit. The shape evolved to an elliptical lava coulee with two distinct lobes, confined within the moat between the 1979 dome and the inner wall of the Summit Crater (Fig. 9). Rock fall activity from the margins was very limited, while no deformation of the crater floor was observed in flow fronts. Distinct marginal levees developed with radial and linear flow patterns appearing on the lava surface. Thermal images on 16 January 2021 yielded surface temperatures of up to 600 °C.

Extrusion rates calculated for periods from 1 to 34 days varied between 0.95 and 2.65 m³/s ± 0.59 m³/s with a long-term average of ~1.85 ± 0.14 m³/s (Fig. 2). The cumulative volume reached ~13 × 10⁶ m³ by 19 March 2021, when the dome measured 912 m long, 243 m wide and 105 m high. Extrapolating the linear trend through to 9 April suggests a final volume of ~18 × 10⁶ m³. On 6 April, observation via the installed camera indicated a rapid increase in dome height with incandescence becoming visible over the crater rim from the Belmont Observatory (Fig. 1) on the evening of 8 April. Cyclic gas emissions at the central vent occurred correlated with the banded tremor. During the explosive activity, the new lava dome and significant parts of the 1979 lava dome were destroyed, as confirmed by satellite imagery on 10 April from ICEYE (02:02 UTC) and Capella (14:03 UTC).

As explosive activity intensified close observation of discrete events became more difficult. By 12 April pyroclastic density currents (PDCs) had descended several valleys on the southern and western flanks of the volcano and reached the sea. Following that, enhanced venting or Vulcanian-style explosive activity episodically occurred until 22 April. Some explosions generated PDCs in valleys on the western flanks of the volcano.

Crisis response: warning and decision-making systems

The SRC supports local authorities for strengthening preparedness and communicating volcanic hazards through product development (e.g., integrated volcanic hazard maps). The map for St. Vincent (Fig. 1) is a colour-coded depiction of the expected impact of volcanic hazards across the island⁷. The volcanic alert level system (VALS, see Supplementary Information) translates the volcanic activity level into required actions during volcanic unrest. Operational constraints in country meant that any increased likelihood of an explosive eruption needed to be communicated 24–48 h before onset to enable successful evacuation.

Distinguishing rapid accelerations in activity after the onset of an eruption, in particular, transitions to an explosive eruption, are a globally recognised forecasting challenge¹⁸. It is also important to recognise when the probability of explosions decreases, to lower the alert level. Here we identify the practical challenges in recognising and communicating these changes. We highlight the importance of preparedness, diverse forms of communication, and structured approaches to the interpretation of scientific data, development of evidence-informed forecasts and assessment of risk during volcanic crises.

Short-term contribution to decision-making: assessment of risk

Current uncertainty in understanding volcanic processes contribute to a variety of opinions on causative mechanisms and prognoses, particularly during an evolving crisis. In addition, the aleatoric uncertainty associated with the complex behaviour of volcanic systems requires caution against a deterministic interpretation that over-emphasises one specific outcome¹⁹. SRC used the framework of a structured expert elicitation^20,21 around a range of scenarios, to generate both consensus (a collective ‘most likely’ prognosis) and to represent the diversity of opinions.

Weekly elicitations (January to early March 2021) favoured continuation of effusive activity (~80%) each time. The likelihood of an escalation to explosive activity in the following weeks remained at median probability of ~10%. However, following the first VT swarm (23–24 March), elicited estimates for a transition to explosive activity doubled to a median probability of ~20%. With the appearance of banded tremor (8 April), elicited probabilities of explosive activity tripled to a median value of ~60%. The authorities of SVG were alerted to this increase in volcanic activity. The alert level was raised to Red on 8 April at 18:00 UTC triggering the evacuation of ~16,000 persons from the Red and Orange Zones, prior to the start of explosive activity on 9 April at 12:41 UTC, with no reported serious injuries or loss of life.

The visual observations of declining surface activity, lowered seismic activity and declining gas output, coupled with the slow deflation signal observed since 22 April, were key drivers for the lowering of the alert level to Orange on 6 May.

Longer-term contribution to decision-making: risk awareness, preparedness and communication

Hazard assessments and analysis of past events have continuously been updated in response to new understanding^{22,23,24,25,26}. Further, improvements in communication of improved understanding of hazards have been assessed and implemented in hazard planning by NEMO and SRC. The Volcano Ready Community Project (VRCP) led by SRC in collaboration with NEMO, launched in April 2018 and completed in April 2021, targeted twelve northernmost communities of St. Vincent in the Red and Orange hazard zones of the most recent volcanic hazard map⁷. The VRCP, enabled community plans to be drafted and integrated into the national response mechanisms prior to the 2020–2021 eruption.

Communication pathways: transition from extrusive to explosive

Communication of messaging between SRC and NEMO was harmonised. A continuous flow of near real-time information was provided to the public and stakeholders about volcanic activity, hazards, and risk reduction. These communications maintained credibility in the monitoring capability of SRC^27,28. A similar communication strategy employed by the USGS, in response to the 2014–2015 Kilauea volcano lava-flow crisis, was shown to be a highly effective approach²⁹ and aligns with volcano observatory best practices for operations during crises⁶.

Based on best practice and evidence, risk communication products were developed to target different learning styles, media platforms and preferences²⁸. These products included visual, print and audio products, and were combined with live scientific presentations during media interviews and to special interest groups. Social media schedules and posts were coordinated, while SRC scientists on island participated in daily activity updates on local television and radio stations, and provided cabinet briefings and updates to decision-makers. Scientists participated in virtual and drive-through community meetings for Red Zone residents with live online streaming and simulcast on local television and radio. An important facet of uncertainty during eruptions is dealing with misinformation and rumours. The strategy of maintaining a continuous presence on social media (Fig. 10) and the use of FAQs and short interviews allowed growing concerns or misapprehensions to be addressed. The frequency of these communications was influenced by changes in the ongoing activity. International scientists were also encouraged to amplify existing messages and use SRC materials in discussing the eruption with their local media.

Another important dimension was systematised internal communication. External contributions of data were facilitated by a team lead who was responsible for internal communication and coordinating data requests. This approach also facilitated international collaborations and engagement with academic scientists, which supported the SRC to develop conceptual models.

With the start of the explosive phase of the eruption, social media posts were still the primary tool used by SRC to communicate with the public. Scientific bulletins were shared directly on these platforms, with the addition of voice notes shared via mobile networks. Daily activity updates on local radio and television stations continued. Visual, print and audio products now also focussed on explanations of, and recommended responses to, the primary volcanic hazards (pyroclastic density currents, ash fall and lahars) being observed.

Crisis management

Volcano monitoring data enable scientists to provide short-term forecasts or advise of possible changes during an ongoing eruption³⁰. However, for successful crisis management, monitoring data and interpretations need to be: (a) framed within the context of wider scientific knowledge, (b) presented in the context of decision-making (‘useful, usable and used’)³¹ and (c) effectively communicated to diverse audiences³². The St. Vincent case provides an important demonstration of how these principles were integrated, complementing synoptic analyses of the state-of-the-art in volcano observatory crisis operations^6,33. Next, we discuss the key lessons from our analysis of response to the unfolding events, particularly the eruptive transition, and assess the role data and models played in decision-making. We also reflect on the constraints on best practice imposed by finite resources, and how this can be improved.

Conceptual models and their value in forecasting eruptive transitions

Historically, La Soufrière volcano can produce both explosive and effusive eruptions over time intervals of weeks to months. However, transitions in behaviour can occur over only a few hours⁶ and pose acute challenges to risk management; particularly when decisions to evacuate are exacerbated by resource or space constraints that affect the tolerability of evacuations. Analysis of previous eruptions in St. Vincent has demonstrated that compliance with long-duration evacuations will dissipate, a feature shared with crises at other volcanoes⁴.

A working conceptual model of volcanic behaviour was created and developed in real-time, which was used to inform the scientific response to emergency management and advise the authorities. Critically, during the effusive phase our evolving working model was used to anticipate explosive transition or other significant changes in activity.

In early January 2021, we interpreted onset of the eruption as the consequence of the injection of fresh gas-rich magma into a sub-volcanic reservoir, making its way to the surface³. However, this interpretation could not explain the comparatively low seismicity rates, lack of surface deformation and near-constant extrusion of lava (Fig. 2). The presence of a ductile well-connected magma ascent pathway was proposed to reconcile these early seismic observations³⁴. The absence of deformation and steady extrusion could be explained by either (i) a magma source that maintains a near-constant overpressure³⁵ or (ii) that a large magmatic source, relative to the material extruded, resulted in pressure decay in the reservoir being too small to be detected^36,37 or (iii) that hot magma mush surrounding the source region, in combination with the viscous flow in the crust, maintained the high pressure³⁷, or (iv) some combination of these processes. The diversity of explanations, informed by monitoring observations, was important for assessing the potential for explosive activity, and the timescale over which this might happen. At that time, there were no evacuations in place, but existing hazard assessment and outcomes from past simulations (e.g., Tradewinds Exercise 2019³⁸) demonstrated that risk to the northern population could rapidly become high, with a 24–48 h interval needed for full evacuation.

By early February, the absence of detectable SO₂, however, led us to infer the presence of degassed magma remaining within the conduit, following the 1979 eruption, confined by a strong ‘cap’, was slowly being pushed up by a new injection of gas-rich magma. By early March, the similar composition and petrographic characteristics of the extruded dome rocks, in comparison to past eruptive products, had reinforced this model. Similar behaviour has been inferred at volcanoes such as Kilauea in Hawai’i, where in 2018 near real-time geochemical analysis of lava indicated magma characteristics consistent with progressive flushing of residual magma in the conduit³⁹.

The epicentres of the intense seismic swarms before the explosions show a concentric distribution of earthquakes around the volcano (Fig. 3a). This suggested magma ascending through the volcanic conduit. Most (>95%) of the estimated epicentres were located above 5 km until 18:00 UTC on 5 April, when seismicity migrated to deeper levels (Fig. 3a, b). This sudden transition in depth was considered evidence of increased deviatoric stresses around the conduit, possibly related to a new intrusion of gas-rich magma.

In our evolving model, obstruction by the cap material and overlying 1979 dome prevented fresh magma from reaching the surface and limited SO₂ flux to volumes low enough to be scrubbed by the volcanic hydrothermal system, until April 2021. We speculated that the accelerated extrusion rate observed after 6 April, was after the high-viscosity magma cap was displaced by new lower viscosity gas-rich magma. Banded tremor, consisting of merging VT events, attributed to the excitation of fluids at relatively shallow levels, was observed one day prior to the onset of the explosive phase^14,15. This observation suggested possible pressure oscillations within the ~6 km deep magma reservoir as the trigger of highly periodic tremor events. This strongly implied the imminent passage of gas-rich material from depth into the shallower edifice, consistent with the first detection of SO₂ flux on 8 April (Fig. 6). Then on 9 April, gas rich magma reached the surface and conditions for explosive fragmentation were realised, which correlated with the observation of syn-eruptive deflation (Fig. 2).

The working conceptual model provided a robust framework against which these rapidly emerging data could be interpreted and understood. This working model illustrates the need to interpret scientific data in real-time to inform rapid emergency decision-making and the difference between theoretical models and critical interpretations that trigger real-world, life-preserving decisions. The conceptual models, synthesised from quantitative data, were necessary for decision-making and formed a framework to create actionable evidence for responding to an acceleration in activity. Nonetheless, the conceptual model was also strongly informed by quantitative outputs from generic models for different aspects of volcanic behaviour and the input of boundary conditions obtained from new knowledge of the magma composition and observations of dynamic behaviour. Further scientific analysis with longer-term research programmes and quantitative modelling will test and improve these models. An important dimension of fully quantitative models is the recognition of generalizable insights relevant to other settings worldwide where rapid transitions in activity occur, that can be derived from empirical observations in real time.

The role of uncertainty and impact of monitoring in a resource-constrained setting

Interpretation of the monitoring data and development of a preliminary conceptual model were associated with large uncertainties when anticipating eruptive behaviour throughout the unrest episode. These uncertainties created both temporal (when) and spatial (how big) challenges. Specific uncertainties included: (i) interpreting the extent that seismic unrest patterns were similar to historical background seismic activity at La Soufrière volcano; particularly long episodes of unrest preceding >VEI4 explosions; (ii) during steady-state dome growth, distinguishing monitoring signals indicative of a potential acceleration of activity from normal behavioural fluctuations, in the absence of any significant measured deformation; and (iii) during the explosive phase anticipating the likely duration and peak intensity of explosions, given the range in size and documented intensity of the previous eruptions⁸. This contributed to uncertainties in interpreting signals that might represent the onset of an explosive phase and reduced timescales over which accelerations or decelerations in the intensity of activity could be confidently attributed to changing behaviour. Coping with uncertainties framed our conceptual model and attendant different scenarios. Our combined expert view of likelihoods captured via expert elicitation, fed into decision focussed advice (e.g., VALS for La Soufrière). This approach avoided interpretations dependent on single outcomes¹⁹, which inadvertently minimised aleatoric uncertainties or ignored ambiguities in datasets. The St. Vincent case demonstrates the benefits of the structured expert elicitation methodology to capture uncertainty.

An important factor in generating epistemic uncertainty was the relatively sparse monitoring network at the onset of the eruptive episode, which was a direct consequence of financial constraints on the monitoring operations. Limitations in the density of the network challenged our ability to definitively say whether the onset of the effusive eruption would have been instrumentally detected. However, as the monitoring network strengthened, observed signals were interpreted against improvements in data volume and accuracy. For example, additional GPS stations installed during the eruption greatly improved the sensitivity of the network. A sensitivity study⁴⁰, demonstrated that the network had no significant azimuthal gaps, but suffered from a lack of near-field stations to capture shallow deformation sources. In addition, interpretation of the low amplitude banded tremor detected on 8 April, reinforced by the observation of a detectable SO₂ gas flux later that day, was the most salient information to feed into changed views on explosion likelihood. Similarly, detailed seismic analysis and near real-time satellite and deformation measurements contributed to the anticipation of waning explosive activity during the acute phase (9–11 April).

The relatively late detection of effusive eruption onset and the importance of new data during the eruptive transition, clearly demonstrate that well-resourced multi-parametric networks are of high value. The reality in settings like St. Vincent is often different and network strengthening took place during the eruption, creating important safety concerns. SRC used a fieldwork life-safety risk assessment⁴¹ with an estimated hourly risk of fatality exceeding 10⁻⁴ during the initial fieldwork period (see Supplementary Information). This procedure gave strong justification for the use of a helicopter and provided an opportunity for monitoring scientists to express any concerns and contribute recommendations on the best field practices.

The value of collaborative preparedness, awareness and communication

Identifying local needs and obtaining evidence of the efficacy and impact of the SRC’s risk communication in the vulnerable communities⁴² was a persistent challenge largely due to the Education and Outreach (E&O) team’s remote operations. Harmonisation of messaging was essential⁴³. Close collaborations with NEMO strengthened communication efficacy and reinforced local capacity for effective communication. In turn, this provided SRC with insight into appropriate content for its communication products. These types of relations take time and resources. The groundwork was essential to the success of managing the crisis in a rapidly changing volcanic situation with a requirement for the implementation of advice into action.

Wide acceptance of the risk information was indicated by the authorities acting decisively on advice provided by SRC, resulting in increased alert levels and the issuance of evacuation orders 24 h ahead of the first explosion. Furthermore, the public understood the increased volcanic activity and complied with evacuation orders.

Integrated approach: value in anticipating eruptive transitions

Effective crisis science and consequent volcanic risk reduction is a partnership between scientists, response agencies and the affected communities^44,45, It begins with the robust gathering and interpretation of scientific data, before, during and after a crisis. Our analysis provides an excellent case study of the principles outlined in recent synoptic analyses^6,33.

However, important challenges arise in the acute crisis phase where decision-making timescale appropriate to the lifetime of the eruption (typically weeks to months) contract into minutes and hours with the growing prospect of a change in behaviour. As a transition threatens, uncertainty rises and demands dynamic interpretation of emergent datasets. Thus, our analysis here particularly reflects on the important drivers of risk in this moment.

An important dimension was the capacity to interpret data against a flexible conceptual model that expresses and formalises uncertainty. Further, the understanding from previous research^7,8,9,13,22 and agency-to-agency interactions that framed social context and societal constraints were important. Monitoring agencies need to be responsible for interpreting datasets and anticipating changes on societally relevant timescales. This responsibility also underpinned our communication strategies and timescales. The long-term relationship we described here increased the chances that advice given during an eruptive transition was more readily translated into actions by local emergency managers, and in turn, the populace at risk.

Research that accounts for the realities of managing crises could further improve effective decision-making. In volcanology, counterfactual analysis is a powerful way to understand what might have transpired⁴⁶. A counterfactual analysis to include the range of possible scenarios and outcomes using the ‘real time’ evolving knowledge gathered, would assess whether the decision-making strategy here was robust to all eruptive outcomes. For example, considering situations where explosive activity happened at an earlier stage or explosions that generated larger pyroclastic density currents. Similarly, a focus on emerging petrological techniques that allow rapid forensic examination of timescales of disruption, degassing and ingress prior to other eruptive episodes would have significantly helped with the interpretation of changing monitored signals at the acute crisis point.

Finally, it is important to acknowledge the implicit risk to monitoring scientists during the intra-eruptive network strengthening. Our analysis demonstrates the value of the strengthened network, as well as remotely observed data, to data interpretation despite the risk in this particular case. Research that improves understanding of the effectiveness of monitoring networks would help identify strategies that best minimise risk, while maximising data benefit.

Operating in a resource-constrained setting influenced scientific response and emergency management. The steady global growth of disaster risk, volcanic or otherwise, compels disaster response agencies to fortify disaster preparedness capabilities and to ensure that institutional capacities are in place to optimize effective planning, response, and mitigation. Our assessment of the 2020–2021 La Soufrière eruption demonstrates the critical controls, produced over longer timescales, of an effective response during an acute crisis at the moment of eruptive transition.

Confidence in our conceptual models of reactivation via a gas-rich magma at depth was improved through the strengthening of the seismic network, real-time deformation and dome monitoring, changes in gas composition and petrological sampling. Nonetheless, as monitored signals shifted and the likelihood of transition increased, longer-term preparedness measures allowed us to disseminate rapidly changing information effectively on short timescales and contextualise our advice on timescales appropriate for actions to prevent loss of life, while minimising impacts on livelihoods.