Impacts and prospective hazard analysis of rainfall-triggered lahars on St Vincent 2021–22 Open Access

‘Lahar’ is a term of Indonesian origin and describes flows that involve a mixture of loosely consolidated volcanic debris and water occurring on and around volcanoes (Vallance 2005). All of the volcanoes of the Eastern Caribbean are susceptible to lahars following explosive eruptions as a consequence of intense rainfall during rainy seasons in the tropical climatic zone where they are located. Lahars are generated by the interaction of intense, frequent rainfall with the resultant loosely consolidated deposits either from tephra or pyroclastic density current deposits. La Soufrière volcano (St Vincent) generated a series of explosions between the 9 and 22 April 2021, following approximately three months of effusive activity, confined to a summit crater without a substantive lake. There were at least 24 discrete explosions and one prolonged period of fluctuating but continuous explosive activity (Joseph et al. 2022a; Sparks et al. 2023). This generated island-wide tephra and, from approximately 18:30 UTC on the 10 April, these explosive events also generated pyroclastic density currents (PDCs) that flowed down multiple drainages leading from the summit crater (Cole et al. 2023). Tephra was unevenly dispersed across the island (in both thickness and median diameter) with distal coastal locations to the south and east only receiving a few millimetres of fine-grained ash during a subset of the early explosions. Some villages in proximal coastal locations in the NE received tephra for most of the eruptive sequence, resulting in deposits greater than 100 mm in thickness (Cole et al. 2023). PDCs reached to 4.2 km to the west and to 2.8 km to the south with tephra thicknesses at 3–4 km distance >100 mm, increasing rapidly towards the crater (Cole et al. 2023). These deposits and flows were also associated with substantive vegetation damage later evidenced in the make-up of the lahar deposits.

Both syn-eruptive and post-eruptive lahars have been observed in the historical and near-historical past across the Eastern Caribbean. For example, recent analysis of historical lahar occurrence on Mount Pelée and La Soufrière (Guadeloupe) recorded some 165 syn-, post- and non-eruptive lahars between 1530 and 2009, although substantial peaks in lahar occurrence are associated with Pelée's eruptive activity in 1902 and 1939 (Leone et al. 2019). Longitudinal analysis of catchment response to the perturbations associated with the 1995–2010 eruption of Soufrière Hills Volcano, Montserrat (Barclay et al. 2007; Christie et al. pers. comm.) demonstrate the long wavelength disturbance of these systems, responding variously to intense precipitation and seasonal variation in ground saturation, vegetation loss and the input of a heterogeneous variety of sediment. The difference between lahars and flash-flooding on these tropical islands in terms of outcome via damage impacts is indistinct so here we consider a ‘lahar’ to be any water-saturated sediment-charged flow associated with the transport and incorporation of loose volcanic material (whether recently erupted, ‘syn’ and ‘post’ or via excavation of unconsolidated volcanic deposits ‘non-eruptive’). The lahars considered here can be debris flows, hyperconcentrated streamflow or streamflow and often transition between these flows states during any one event.

On St Vincent both the 1902 and 1979 explosive events at La Soufrière were initiated in the presence of a substantive crater lake, and this has been implicitly attributed to the generation of syn-eruptive lahars. In 1902 contemporary accounts of the opening phases of the paroxysmal 7 May eruption involve various accounts of boiling flows, and some field evidence indicates that lahars are preserved locally near the base of the 1902 sequence (Cole et al. 2019). Following these eruptions initial rainfall percolated into the thick PDC deposits but only generated localized explosions rather than debris-charged water flow (Hovey 1909). By March 1903 the ‘immediate gorge of the Wallibou had been mostly cleared of its filling of ash’ (Hovey 1909, p. 421) by lahars, and dilute sediment-water flows were occurring, initiated by dry landslides from the destabilized slopes of the valley. A further visit in 1908, remarked on the still-disturbed geomorphology of these valleys and included the direct observations of debris-charged flows occurring after rainfall which ‘brought the river down on the surface of its bed to the sea in pulsations overloaded with debris and easily rolling along boulders in 30 or more centimeters in diameter’ (Hovey 1909, p. 421). These flows were observed in both the Rabaka (sic) and the Wallibou. At this point there was still substantive modification to the shoreline with ongoing processes of coastal erosion that Hovey (1909) inferred would eventually return to that of the pre-eruptive morphology. In 1979 ‘mudflows’ were reported during the initial vulcanian explosive phase (13–26 April) alongside the generation of base surges and minor PDCs. Initial explosions breached a pre-existing dome in the crater lake with the consequent ingress of water into the vent influencing the style of later explosions, until the crater lake disappeared (Fiske and Sigurdsson 1982). However, there is less record of laharic activity following these eruptions, although the five short-lived explosions only generated limited source material; PDC deposits were confined to valleys within 2 km of the crater and the cumulative deposit was only 30–40 mm at distances of 3–4 km from the vent (Sigurdsson 1982).

Lahars were observed during the short explosive phase of the 2020–21 eruption of La Soufrière volcano, and have continued to occur until the time of writing. The first lahar was reported on 11 April 2021 and, subsequently, lahars were detected by the seismic network and observed in all valleys around La Soufrière. (Joseph et al. 2022b) Their impacts were particularly felt by resident communities on the windward (east) side of the island, where the only road to the north around the coast crosses several rivers including the Rabacca and Overland rivers, and rivers in the settlements of Sandy Bay, Owia and Fancy (Fig. 1). The expectation from previous eruptions of La Soufrière and from climate projections showing more intense rainfall for the Caribbean (both during and outside of large weather systems associated with the May to November hurricane season; Taylor et al. 2012) is that lahars will continue to be a significant hazard on St Vincent into the future.

In this work we will synthesize a catalogue of lahar events during 2021 and review their impacts. Their detection on the seismic network cannot identify which valley they are flowing down, so we also use journalistic and social media records as well as discussions in local community WhatsApp chat groups. Recording small and relatively frequent lahar events is a challenge as the steep and sometimes denuded topography in the volcanic region of the island means that communities in St Vincent are used to the impacts of seasonal flash flooding which is not dissimilar to lahars. So, recording of events is typically unsystematic and variable in detail. There were also some challenges due to the short duration of the eruptive phase and in accessing the field areas during this time. Consequently, the focus of this paper is not on providing a detailed field stratigraphy of individual lahar events, rather we use parts of these field data as the basis for estimating the volume of future events for input to hazard models that predict their impacts.

The aim of this paper is to use our analysis of the drivers and outcomes from lahars to date, combined with models of lahar inundation within the context of past behaviour in this or analogue settings, to suggest a pathway for anticipating and potentially mitigating the impacts of future lahars in St Vincent. This builds on the first order-assessment presented by Miller et al. (2022). To do this we will construct a database of lahar events and correlate this with triggering rainfall, use observations of lahar events and their impacts to estimate the lahar volumes and volume fluxes to provide initial conditions for lahar hazard modelling. We present some preliminary results from this work with a discussion of how this can be used to contribute to early warnings and preparedness planning for future lahar impacts on St Vincent.

A database of lahar events from the start of the eruption until the end of 2021 was compiled to characterize their location and scale, using observations made by scientists working in the field, detections of lahars on the seismic network, and reports in newspapers and social media. The database indicated that lahars were observed or instrumentally detected on 25 days during this period. Lahar observations were made on 14 days by scientists from Belmont Observatory (Fig. 1) working in the field, and reported in Scientific Advisories and local newspapers (Scientific Advisories are official reports summarizing observations and their interpretation by the monitoring scientists, and those for the 2020–21 eruption are collected by Joseph et al. (2022b)). Since May 2021, lahars have also been detected on 16 days on the seismic network, and reported in Scientific Advisories. The first mention of possible lahar activity is in Scientific Advisory 202011_SVG_44 on 11 April, noting ‘steaming in the upper parts of the Rabacca valley’ as under investigation. Subsequent reporting of lahars in the Scientific Advisories mostly notes that lahars were detected on the seismic network, sometimes indicating the location of the strongest signals, but not identifying specific valleys down which lahars travelled. Lahars were also reported in the online newspapers iWitness News, Searchlight and the Barbados Advocate. These reports mostly quoted National Emergency Management Organisation (NEMO) officials or the University of the West Indies Seismic Research Centre (SRC) scientists describing lahar events, sometimes from the daily 8 am National Broadcasting Corporation radio programme ‘Eyeing the Volcano’. Photographs of lahars were occasionally included, typically uncredited and undated but sometimes including the location.

Community observations of lahars, shared on social media, were also an important source of information for constructing the timeline of lahar events. Between September 2021 and April 2022 a community engagement project, ‘Changing Landscapes’, was set up as part of the United Kingdom Research and Innovation Global Challenges Research Fund ‘Risk at the Margins’ research project. The project involved communities from the NE of St Vincent, the areas most affected by the eruption and lahars. One of the activities of the project included workshop discussions on the utility of documenting lahar events particularly at river crossings and the value of community knowledge for improving scientific understanding of St Vincent lahars. After the workshop, the ‘Changing Landscapes’ project participants formed a WhatsApp group where they shared social media links to photos and video recordings of significant lahar events from their community network. Between 12 April and 30 August 2021, video recordings of 18 lahars were shared on social media platforms by residents of the north windward communities across nine different river-crossing locations. From January to September 2022, five additional lahar observations were retrieved from the ‘Changing Landscapes’ WhatsApp Group including videos and written reports of lahars at the Overland River crossing (Fig. 1). These observations were correlated with rain gauge observations and instrumental data to populate the database and confirm locations of lahar occurrences. They were also useful for constraining timings and characteristics of lahar flow to support modelling of future impacts.

In our analysis, we compare timings of lahar occurrence with rainfall records, using daily rainfall totals from 25 rain gauges located across St Vincent and managed by the Central Water and Sewerage Authority (CWSA) of St Vincent and the Grenadines (Fig. 1). We explored the correlation between lahar occurrence and daily rainfall total, using both the daily rainfall total from the rain gauge nearest to the location of the lahar event (if known) and the maximum daily rainfall total recorded on any of the rain gauges. Daily total rainfall is the highest resolution available, so our analysis is consistent with some recent approaches for characterizing the initiation of lahars by rainfall, which use constant rainfall rates for the duration of rainfall events over daily or multi-day timescales (Mead et al. 2016; Baumann et al. 2018). We recognize the shortcomings of being limited to daily rainfall data since lahar initiation depends on both rainfall intensity and duration, as has been shown in previous studies (Rodolfo et al. 1996; Lavigne and Thouret 2003; Van Westen and Daag 2005). Jones et al. (2015) show that peak rainfall in one hour is a better predictor of lahar occurrence than total daily rainfall for temperate moderately vegetated land cover at Tungurahua volcano Ecuador. In line with our analysis, they also show that antecedent rainfall (1, 3 and 5 days before recorded lahar events) is not an effective predictor of lahar occurrence when used in isolation, but improves forecasts of lahar events when combined with peak one-hourly rainfall. The aim of our analysis is not to identify a formal threshold for issuing early warnings of lahar events, which would need higher resolution rainfall measurements (cf. Jones et al. 2015), rather here we use the comparison between lahar occurrence and daily rainfall totals to infer the initial conditions for lahar initiation and modelling, aimed at supporting longer-term preparedness and mitigation planning.

A timeline of lahar occurrence and daily rainfall is presented as Figure 2, which shows schematically the occurrence and observed extent of lahar events, and peak national and local daily rainfall, from which we can draw out broad trends in the correlation between rainfall and lahar occurrence in 2021. During 2021, lahars occurred in all valleys around the volcano. On the 14 days in which lahars were directly observed, on eight days lahars were reported around all or most valleys on the volcano, on two days lahars were reported only on the west side, on one day lahars were reported only on the east side, and for three days there is insufficient information to identify lahar locations. No information on lahar location or direction is reported in the Scientific Advisories. Over the period April to November 2021 there were two significant rainfall periods, on 27 April–5 May and 20–25 August, during which it is likely that lahar occurrence depended on antecedent rainfall on preceding days as well as on the days when events were observed or instrumentally detected. Lahar observations and rainfall measurements are presented as Table 1.

The first lahar was reported on 11 April (two days after the start of the eruption). There was relatively low rainfall on this day (national maximum of 9.4 mm in the central south of the island recorded at Montreal) and approximately 5 mm on both the west (Dallaway) and east (Jennings) sides of the island (all rain gauge locations are shown on Fig. 1). There was also little antecedent rainfall (maximum daily rainfall of 6.3 mm anywhere on St Vincent in the preceding two days), so this event is likely mobilization of fresh and unconsolidated pyroclastic deposits by relatively low rainfall.

For lahar events from late April 2021 onwards, we can identify consistent trends in the correlation of their occurrence and daily rainfall. On days on which lahars occur in one, or a few, valleys on the same side of the volcano, a minimum of about 20 mm of daily rainfall was measured in rain gauges local to the lahar location. On days when lahars are observed in all or most valleys around the volcano (associated with the significant rainfall periods in April and August), there is both daily rainfall in excess of 80 mm anywhere on St Vincent, and rainfall near the volcano (either west or east sides) of at least 55 mm. The exception to this is for three later days (5 and 6 May and 21 August) within the two significant rainfall periods mentioned earlier, when lahar occurrence was correlated with lower daily rainfall (minimum 5.8 mm daily anywhere on St Vincent), although the previous one or two days had high daily rainfall and indicated the importance of antecedent rainfall on lahar initiation. On days on which lahars were detected on the seismic network (with no information about the valley in which they occurred), there was a minimum daily rainfall of about 20 mm or more measured close to valleys known for lahar activity since the start of the eruption.

From this limited analysis we conclude that lahar initiation is associated with at least 20 mm daily rainfall measured close to the lahar route, and this threshold has not changed during the first seven months after the eruption. Larger rainfall events can trigger lahars in multiple valleys around the volcano, with the same threshold rainfall observed for April and August 2021. Typically, lahar initiation requires increasing rainfall totals with time after the eruption, as pyroclastic deposits harden and form less permeable crusts due to chemical weathering (Yamakoshi 2005; Baumann et al. 2018); this does not appear to be the case over the first seven months since the eruption in April 2021, and suggests that the lahar hazard may be significant on St Vincent for some time into the future.

Syn-eruptive lahars occurred in at least the Wallibou and Dry Wallibou valleys on 11th April. These valleys are to the west of the volcano (Fig. 1) and are so named because the Dry Wallibou only has an active channel during the rainy season (typically June to November), whereas the Wallibou has an active channel throughout the year. The deposits from these events were examined on 26th April, 4 days after the explosions had ended, and were locally still warm to touch. These deposits were relatively fine-grained, at least on most of their surfaces (Fig. 3a, b). The deposits had apparently extended across the whole valley rather than being channelized. Small wooded areas at the margins of the Wallibou valley were inundated by the lahars to depths of between approximately 1 and 2 m (Fig. 3a). Flow fronts were either similarly fine-grained with small lobes 10 s of cm high (Fig. 3a), or composed of collections of tree trunks and branches (Fig. 3b). Locally, particularly at flow fronts near the coast of the Wallibou areas were concentrated in scoria blocks c. 20 cm in size. The surface of the syn-eruptive lahar deposit both in the and Dry Wallibou valleys contained extensive ‘dessication cracks’ that appear to have been deformed by continued creep of the deposit following emplacement (Fig. 3c). All evidence indicates that these syn-eruptive lahars were wholly depositional rather than erosive.

Sections through the syn-eruptive lahar deposits showed a generally massive, non-stratified nature, although two or three separate units could be identified (Fig. 3d). These massive deposits were up to c. 10 m thick in total in the lower parts of the Wallbou valley. Thicknesses in the Dry Wallibou were thinner, c. 1 m thick. Tephra fallout from the explosive activity was observed below these lahar deposits in the mouths of both the Dry Wallibou and Wallibou valleys (Fig. 3c) confirming the timing of these deposits occurring at least after the initial activity on 9 and 10 April.

Significant lahar events were observed in the Wallibou valley on 29 April and 3 May 2021. The 29 April lahars generated significant erosion through the syn-eruptive lahar deposits in the lower 500 m of the Wallibou valley: originally flat fans at the shoreline of both the Wallibou and Dry Wallibou valleys formed by the syn-eruptive lahars (on 11 April) were deeply dissected. Those on 3 May were also notably erosive, widening the canyons and eroding the older (pre-1902) pre-existing pyroclastic deposits in the walls of the Wallibou valley.

Deposits formed on the 29 April, 3 May and later events differed significantly from those of the syn-eruptive lahars. These deposits filled channels and valleys cut into the extensive, valley-wide terraces formed by the syn-eruptive lahars on 11 April (Fig. 3e). The deposits typically showed centimeter- to millimeter-scale discontinuous lensoid bedding, characterized by coarser lenses up to c. 20 cm thick and finer-grained, millimetre-scale, cross-laminated lenses. Very fine ash was generally lacking from these deposits (Fig. 3e, f).

A lahar event on 3 May 2021 was observed in detail from a vantage point above the lower Wallibou valley at location [13.313, −61.231], close to Richmond Vale Academy (location shown as ‘Richmond’ on Fig. 1). The flow lasted approximately one hour from 11.30 to 12.30 local time, and video recordings of the flow were made intermittently during this time. The flow consisted of a baseflow during which the channel base was sometimes exposed, allowing flow depth to be estimated, and intermittent pulses of higher flow rate (Fig. 4). Every few minutes, small pulses of increased flow rate were observed. These lasted about 30 s during which the flow depth increased by a factor of about 2. Less frequently, larger pulses that lasted a few minutes and increased the flow depth by a factor of 2 to 4 were observed (Fig. 4b). The flow was predominantly fine-grained sediment with intermittently larger boulders of tens of centimetres to 1 m scale visible in the flow, primarily during higher flow-rate pulses. Bank undercutting and erosion of previous lahar deposits at the channel edges was observed.

We estimate approximately the mean discharge over the event using field estimates of vegetation dimensions to constrain the channel width and flow depth from video imagery of the lahars. We estimate a channel width of between 15 and 30 m (Fig. 4a), flow depths and velocities of between approximately 0.3 and 0.5 m and 2 and 3 m s⁻¹ for the lower discharge continuous baseflow, and flow depths and velocities of approximately 1 m and 4 m s⁻¹ for intermittent pulses of higher discharge. These flow conditions correspond to discharge estimates from 10 to 45 m³ s⁻¹ for the baseflow, corresponding to a lahar volume estimate of between 4 × 10⁴ and 2 × 10⁵ m³ due to the steady component of the flow. Field observations suggested the flow conditions for the high discharge pulses were maintained for about 10% of the duration of the lahar event, giving an estimate of total lahar volume of between 5 × 10⁴ and 2 × 10⁵ m³. The pulsatory nature of the flows makes estimation of the lahar volume from flow observations challenging, so these volumes should be treated as order of magnitude estimates.

The explosive phase of the eruption drastically altered the hydrological system of the watersheds, as vegetation was damaged or destroyed by ballistic projectiles, PDCs and heavy tephra falls that resulted in destruction of the forest (particularly on the upper parts of the volcano) and removal of the protective ground cover. This caused increased runoff on the steep terrain, triggered by high-volume and long-duration rainfall events that occurred after the eruption. Rapid erosion and transport of loose volcanic materials downslope by lahars was both spatially and temporally controlled as most of the physical damage was concentrated in areas located on the flanks of the volcano and was most significant soon after the eruption ended when tephra thickness was highest.

The radial nature of the drainage and damage to vegetation on the upper flanks of the volcano resulted in lahars occurring on all flanks of the volcano. This meant that the primary coastal road was most vulnerable to lahar inundation along the multiple river ford and bridges that connect communities. As such, the physical impacts of lahars was most visible at river crossings where there was damage caused to houses, roads, bridges, fords and electrical infrastructure (∼US$12M for total damage; Government of St Vincent and the Grenadines 2021a). Less visible was damage caused to water and electricity supply and distribution infrastructure which rely on surface water intakes from rivers. Lahars caused damage to spring intakes, reservoirs, and pipes especially infrastructure that was exposed above ground or located near to flow paths. They destroyed the catchment systems for water at Sandy Bay, Owia and Fancy (all located on the volcano; see Fig. 1), including river intake, pipelines and treatment tanks. Damage to the water supply infrastructure at Fancy, Owia and Sandy Bay including major damage to catchment intake structures and pipelines amounted to an estimated $241K East Caribbean dollars (XCD$) (Government of St Vincent and the Grenadines 2021b). An additional XCD$63K in damages was reported to water catchment systems at Perseverance and Jennings (see Fig. 1) (Government of St Vincent and the Grenadines 2021b). Disruption of supply to residents in various catchment systems resulted in losses ranging from XCD$115K to XCD$210K (Government of St Vincent and the Grenadines 2021b). In addition to damage to electricity poles, lahars also affected power generation due to the clogging of intake structures leading to the Richmond, Cumberland and South Rivers Hydro-electric power stations (see Fig. 1) by debris brought down in the rivers. This impact was most severe at the Richmond Power Station which was out of commission for 16 weeks following the eruption due to physical damage from lahars.

In examining the systemic impacts of lahars consideration must also be given to the loss of functionality as well as cost and duration of repairs, which are, in turn, related to their direct physical impacts. Lahars damaged electrical poles and water intakes and so disrupted both water supply and electricity supply to communities (Government of St Vincent and the Grenadines 2021c). Loss of function occurred as catchment systems were shut down or compromised due to increased turbidity. The damage from such impacts was most significant north of the Rabacca River (see Fig. 1) where lack of road access due to lahar damage hampered repairs to poles and lines that were damaged and down. Lahar impacts also required clean-up support to allow access for repairs and cleaning. Of all the critical infrastructure sectors, transportation was the most severely impacted with an estimated cost of damage to bridges and river fords in the red and orange volcanic hazard zones (Fig. 5) amounting to approximately XCD$32.5 M (Government of St Vincent and the Grenadines 2021b).

Lahar damage to the Windward Highway (Fig. 5) severely limited the speed at which repairs could have been undertaken to water catchments and electricity supply. As such, the impact of lahars extended beyond the direct physical damage to infrastructure and lasted longer than the specific period over which the events occurred. Disruption of the transportation network exacerbated disruption of the electrical and water systems by increasing the time for recovery and restoration of these systems as the repair crews and equipment could not get to critical sites (e.g. watch catchments, storage tanks, fallen electricity poles) to conduct repairs.

In this section, we test the capability of two different modelling approaches to assess future lahar hazards and impacts on St Vincent. The first applies the widely-used lahar inundation model Laharz (Schilling 2014) with volume estimates derived from observations of the 3 May lahar in the Wallibou valley. The second uses a dynamic model for lahars with rainfall volume estimates for the Overland River catchment (see Fig. 1). Lahars in this location are typically dilute, so we do not attempt to model initiation by shallow landsliding as in previous studies (e.g. Mead et al. 2016; Baumann et al. 2018), but instead take a simplified approach using a runoff coefficient (the ratio of rainfall runoff to total rainfall), which is widely used for flash-flooding on steep catchments (e.g. Tarolli et al. 2012). We use runoff coefficients from 0.2 (as suggested for national-scale flood hazard assessment in St Vincent; CHARIM 2014) to 0.65 (as inferred for steep tropical catchments in Costa Rica; Garcia-Martino et al. 1996), noting that this value is poorly-constrained and will increase with pre-event rainfall or wetness (Longobardi et al. 2003; Ogden 2016).

A first-order assessment of lahar hazard footprints was undertaken using Laharz (Iverson et al. 1998; Schilling 2014), a widely-used tool for understanding potential lahar inundation zones, to obtain an overall picture for future lahar impacts within the high hazard zone in the northern part of St Vincent (Fig. 6). The Laharz toolkit is a simplified tool for estimating inundation zones that is based on a statistical analysis of field-constrained lahar extents/volumes (Iverson et al. 1998) and flume experiments. It does not account for dynamic lahar processes, but rather assumes all of the parameters are constant during a given lahar event. Despite being a simplified model, Laharz has been widely implemented due to its ability to rapidly generate first-order inundation models with multiple user-defined lahar volumes simultaneously (e.g. Darnell et al. 2012; Castruccio and Clavero 2015). Here we employ the Laharz plugin for the ArcGIS software (Schilling 2014).

The required inputs for Laharz are a digital elevation model (DEM) and input lahar volume(s) to produce a drainage network and initiation points; the model then uses a pair of statistically derived empirical equations that relate the volume with the cross-sectional and planimetric areas to estimate potential inundation areas. The ability of the DEM to adequately represent topography is a function of its resolution and age (i.e. how much the land surface may have changed since the DEM was developed). Initial lahar volume can be estimated from measured events at a particular volcano and/or using a range of estimated volumes based upon historical lahar events in similar settings. Individual initiation points can be identified and used in the modelling when detailed knowledge of the drainage system is available, with this providing the additional advantage of reducing the computational requirements of the simulation. A proximal hazard zone is calculated in Laharz based on the height/runout length (H/L) slope value, whereby the intersection of this boundary with river channels provides initiation points. This (H/L) value is often linked to breaks in slope, which should define the area that is most likely to be affected by primary volcanic activity (Schilling 2014). Griswold and Iverson (2008) described H/L as the boundary where lahar behaviour changes from predominantly erosive to mainly depositional.

Here the 15 m DEM generated in 2005 by SRC (Robertson 2005) was used for scenario estimates of future lahar hazard to ensure a consistent resolution for the entire modelled area. The DEM was padded with 0 m pixel values beyond the shoreline of St Vincent to prevent the premature termination of the Laharz simulation, which can occur if there is an inadequate runout distance for larger lahar volumes that propagate along the DEM surface. Laharz creates surface hydrology rasters to define the downstream movement of the hydrologic flows over the DEM surface using a stream threshold value. A proximal hazard zone was delimited using H/L, as defined by the energy cone, which essentially represents the slope of the volcanic flanks. Several estimates were undertaken using a suite of input lahar volumes (5.0 × 10³, 5.0 × 10⁴, 1.0 × 10⁵, 5.0 × 10⁵ m³) to create a range of low to high estimates. Full details are provided in Miller et al. (2022), and we note here that this progression of increasing lahar volumes is not geometric, as would be ideal for an empirical model calibrated on an exponential relationship, as is the case for Laharz (see Iverson et al. 1998).

The modelled inundation using Laharz illustrates that the lahar hazard is confined to the north of St Vincent, the area of high volcanic hazard close to the volcano (Fig. 6). A comparison of the estimates of the calculated lahar flow for the Wallibou event on 3 May 2021 (4 × 10⁴ to 2 × 10⁵ m³) are consistent with the Laharz-derived runouts for 5.0 × 10⁴ and 1.0 × 10⁵ m³ moderate volumes. We see that a volume of 5.0 × 10⁴ m³ is unable to reach the lower reaches of the Wallibou valley (Fig. 6b), whereas a volume of 1.0 × 10⁵ m³ can propagate to the coastline. In the absence of further volume estimates for lahars across the north of St Vincent, we can only consider these estimates as a guide for potential future scenarios in the other channels across the north of St Vincent until lahar volume estimates are obtained for the other drainages. An examination of the Laharz estimates for the southernmost drainages on the eastern coastline indicate a similar pattern, whereby the events in the lower end of the simulated volume range (<1.0 × 10⁵ m³) do not propagate to the lower portions of either the Waribishy or Rabacca channels, the Rabacca channel requiring much higher volumes to propagate to the coastline (Fig. 6a). However, the channels further north require much smaller initiation volumes to propagate to the lower reaches of their respective channels, which is where the population centres are typically located.

We note that the lahar inundation extent from a specific initiation point requires the mobilization of volcanic material via a minimum amount of localized rainfall; therefore, a given rainfall event is not expected to generate lahars from each of the identified initiation points, such that only a subset of the channel will be impacted at a given time. Again, it is important to note that the Laharz results should only be used as a guide due to the simplicity of the model, and the uncertainties in the underlying topographic model (DEM) and defined initiation points. If an updated DEM becomes available with more recent topography changes, then this analysis should be re-run. Manually selected initiation points for individual catchments would ideally be utilized as well. That said, the Laharz modelling approach is a useful tool for obtaining a first-order approximation of the likely hazard across this broad volcanic zone, which can be used as a guide for island-wide hazard mapping purposes and the creation of lahar awareness zones. However, more complex modelling tools should be considered if there is a requirement for detailed land-use planning or time-dependent outputs regarding laha impacts.

Recently available dynamic models for lahars and debris flows offer the capability to predict lahar flow depths and speeds and flow front arrival times, in addition to inundated area. These models, including LaharFlow (Woodhouse et al. 2016; Tierz et al. 2017), D-CLAW (Iverson and George 2014), R-Avaflow (Mergili et al. 2017), Titan (Córdoba et al. 2015), Flo-2D (O'Brien et al. 1993) are formulated from conservation of mass and momentum of the flow, with a range of closures for frictional and material exchanges with the base, or rheologies. These models permit calculation of flow depths, densities and speeds, from which static pressure (the product of flow density, depth and gravitational acceleration) and dynamic pressure (the product of flow density and the square of the flow velocity) can be derived, to allow their impacts to be estimated (Jenkins et al. 2015; Gentile et al. 2022). In this study, we use the dynamic model LaharFlow (Woodhouse et al. 2016) which was developed and calibrated for lahars of a range of sizes, and explicitly parameterizes processes of substrate erosion and deposition that are characteristic of lahars, and which are explicitly linked to evolution of topography and flow dynamics. Full details of the model have been presented elsewhere (Woodhouse et al. 2016), although we present the formulation in summary form here for completeness. The model stability has been analysed (Langham et al. 2021), and it has been implemented as a webtool for use by operational hazard assessment agencies, available for download at http://www.laharflow.bristol.ac.uk.

LaharFlow solves equations of mass and momentum conservation under the ‘shallow layer’ assumption that underpins many surface flow models; the flow is ‘shallow’ (its length is much greater that its depth). This permits us to neglect slope-normal accelerations and average the flow motion over its depth, so that the lahar may be simulated efficiently over two spatial directions. The flow configuration is shown in Figure 7; a layer of water containing solid particles flows down an arbitrary slope of elevation b. Its depth h, velocity u = (u, v), solids concentration C and the bed elevation b evolve as functions of space x = (x, y) and time t. We assume that the solids are well-mixed throughout the water column.

The factor of 1/cos θ on the right-hand side is a geometric correction, accounting for the fact that bed elevation is measured along the vertical, while transfer of bed material takes place along the slope-normal direction (Iverson and Ouyang 2015).

We applied the model to simulate lahar events in the Overland catchment on the east side of St Vincent (Fig. 8). The main road to the north windward crosses the Overland River through a ford, so this is one of the critical locations for lahar impacts on the island. Notable lahar events reported by the local population and on social media happened here on 25 August 2021, 25 December 2021, 27 April 2022 and 1 July 2022. The first of these events resulted from multiple days of antecedent rainfall, with 105 mm rainfall recorded on 23 August at the nearest CWSA raingauge at Jennings (Fig. 1; approximately 8 km from the Overland catchment). The other events resulted from intense rainfall on the day of the event, with 61 mm of rainfall on 25 December 2021 measured at Jennings, 41.2 mm rainfall on 27 April 2022, and 55 mm rainfall on 1 July 2022, both measured using a rain gauge that was installed at Mahorne Ridge ([13.3196, −61.2126]; approximately 0.5 km from the Overland catchment) on 13 April 2022. These lahar events are of comparable volume and flow characteristics to that observed at Wallibou on 3 May 2021 (Fig. 4), as shown in Figure 9 for the 1 July 2022 lahar at Overland. The lahar is predominantly fine-grained sediment, with a flow depth and velocity ranging from 0.5–1 m and 1–3 m s⁻¹ in the 20 m wide channel shown, and standing waves can clearly be seen on the right-hand side (downstream direction) of the image. We estimate typical volume flowrates for these events in the range 10–60 m³ s⁻¹.

The LaharFlow model uses an idealized initial condition of an instantaneous fixed volume release or a volume flux that varies over time (hydrograph), and can include multiple source locations and release times. However, unless there are detailed measurements of the lahar dynamics (including source locations and gauging of discharge, or quantitative observations from which to infer it), a standard approach as adopted here is to assume a simplified triangular (or ‘three-point’) hydrograph and explore sensitivity to that by model calibration (Woodhouse et al. 2016). The model requires a topographic map, and here we used a DEM generated as part of the Charim project (CHARIM 2014). This DEM had a resolution of 2 m for the channel in the lower part of the catchment (below the source location shown on Fig. 8) but was merged with coarser 30 m resolution mapping higher in the catchment. Simulations were run at a downscaled resolution of 5 m, which represents a compromise between longer run times at higher resolution, and sufficient grid cells across the typical channel width of 15–25 m.

To specify the source volume, we estimated the area of the Overland catchment and used a runoff ratio of 0.2 as suggested for national-scale flash flood assessment in St Vincent by the Charim project. Due to the DEM being merged from mapping at different resolutions, we could not use a GIS algorithm to automate the measurement of catchment area, so we estimated this using 3D projections of the catchment on Google Earth (Fig. 8) to identify the extent of the catchment based on downslope direction. This gave a catchment area of 2.82 km², which corresponds to a rainfall volume of 1.13 × 10⁴ m³ with our identified threshold of 20 mm daily rainfall and a runoff ratio of 0.2, increasing to 1.03 × 10⁵ m³ for the largest daily rainfall (of 61 mm) for the case study events and higher runoff ratio of 0.6 that has been inferred for more intense tropical storm rainfall in Dominica (Ogden 2016). This is comparable to the volume range estimated for the 3 May 2021 lahar in the Wallibou valley, but the lower end of the range is smaller for the Overland River. The Wallibou catchment has been estimated at 5.63 km² (Caribbean Conservation Association 1991) which is larger than estimated for the Overland catchment, consistent with potentially larger volumes in the Wallibou.

Simulations were made with the LaharFlow model using triangular input hydrographs of a range of volumes and peak source volume fluxes. We used two criteria to assess whether the simulations were a good match to observed lahar events: the lahar travel time from the source to the Overland crossing (Fig. 8), and the flow conditions (flow depths, velocities and estimates of volume flowrate derived from these) reported at that location. The distance from the source to the Overland crossing is approximately 3 km, so flow travel times in excess of 120 min correspond to mean flow velocities less than 0.4 m s⁻¹, which is inconsistent with both observations made at the Overland crossing (Fig. 9) and velocities typical of lahars in other locations in St Vincent (Fig. 4). Observations of lahar flow at the Overland crossing suggested that flow depths less than 0.5 m, flow velocities less than 1 m s⁻¹ and volume fluxes outside the range 10–60 m³ s⁻¹ were unreasonable. Using these criteria, we found that source volumes of 1.13 × 10⁴ m³ were too small and that a minimum source volume of 2.5 × 10⁴ m³ was needed, corresponding to a runoff coefficient of 0.44 and 20 mm of daily rainfall, or a runoff coefficient of 0.20 and 44 mm of daily rainfall. In addition, a peak source flux of 100 m³ s⁻¹ or more was needed to give reasonable travel times. Simulations at higher source volumes and fluxes up to 4.0 × 10⁴ m³ and 320 m³ s⁻¹ showed travel times in the range 80–110 mins. For the highest source volume, a rectangular hydrograph of constant source flux of 30 m³ s⁻¹ showed a travel time of 110 mins. In all cases, the source released water (a solids concentration of zero at the source), and the solids concentration changed during the flow as a consequence of erosion and deposition.

Outputs from the LaharFlow model are shown in Figure 10 for one of the source conditions that was a good match to observations at the Overland crossing. Figure 10a and b show the flow depth, which has maximum values of up to 5 m higher in the catchment. Figure 10b shows an enlarged view of the flow depths at the Overland crossing, with values up to 1 m (orange cells) but mostly in the range 0.3–0.5 m (red cells). Figure 10c shows the flow velocity (vector sum of along-channel and cross-channel depth-averaged velocities) with maximum values up to 3 m s⁻¹ and most cell values in the range 1.5–3 m s⁻¹. Figure 10d shows the topographic change at this time in the simulation, with red colours showing reduced cell heights (erosion), grey colours showing increased cell heights (deposition) and white showing height changes less than 10 cm. Erosion is observed primarily in the centre of channels, and on steeper topography, where there is faster flow, with erosion depths up to 2 m. There is relatively little erosion or deposition at the channel edges or on flatter topography close to the Overland crossing. Using the LaharFlow model, the increase of volume, or bulking, due to erosion, can be calculated from the computed erosion and a typical bed porosity. For simulations of lahars in the Overland catchment, bulking ranged up to a factor of approximately 1.3. This is on the lower end of the range typical for lahar bulking (Vallance 2000; Vallance and Iverson 2015), consistent with the relatively small volume and short travel distance for this simulated lahar.

The use of paired criteria of matching observed flow conditions and volume flowrates at the Overland crossing constrained the model source parameter space to provide a reasonable starting point for simulating future lahar events on St Vincent. Note here that we have not varied the LaharFlow model parameter values for basal drag, erosion and deposition from their default values calibrated for a range of historical lahars (Woodhouse et al. 2016), in order to focus on inferring source conditions that produce flows consistent with observations. These parameter values could also be varied, but this would not increase the constraints on source conditions unless there were additional specific observations of erosion and deposition. We do not have observations of the lahar flow conditions at the Overland crossing for individual events on 25 December 2021, 27 April 2022 and 1 July 2022, so we cannot link source constraints to the rainfall on those days, but we can use these to infer a runoff coefficient range of 0.19–0.28.

The April 2021 eruption of St Vincent created conditions under which rainfall-triggered lahars have occurred repeatedly during heavy rainfall events since the eruption. The destruction of vegetation by the volcanic ash from tephra fall and pyroclastic density currents has resulted in increased rainfall runoff since the eruption, and remaining tephra deposits now provide a significant source of lahar materials. Our direct observations of lahar events, video postings of events on social media, and limited access to deposits on the western side, all suggest that the lahars are predominantly hyperconcentrated or concentrated streamflow, and only weakly erosive, as compared to typical lahars from larger eruptions (Pierson et al. 1990; Newhall and Punongbayan 1996). Nonetheless, at the point of crossing the road the addition of sediment poses an additional extensive hazard to the populations in the NE of St Vincent.

With the exception of the first few weeks after the eruption (where the lahars were more highly sediment concentrated and erosive), lahar inundation patterns and typical volumes have not changed significantly over the 25 events that occurred during the first eight months after the eruption. A first-order assessment of lahar inundation, made using the Laharz model, suggests that these lahar volumes are sufficiently large to potentially cause impacts on populated coastal locations all around the volcano. Analysis of rainfall records shows that lahars can consistently be associated with as little as 20 mm daily rainfall proximal to the lahar source, although further work is needed to establish how robust this threshold rainfall is, as we have not analysed ‘false positives’, i.e. the number of times in the records that this rainfall is not correlated with a lahar event. Taken together, these results suggest that relatively small lahars can be triggered by relatively low daily rainfall, and their impacts on communities or infrastructure near river valleys are likely to constitute a significant hazard on St Vincent into the future.

The analysis and modelling conducted in this paper provides a starting point for mitigating impacts of future lahars through issuing of alerts, preparedness planning for areas likely to be inundated, and physical measures to protect infrastructure. In the immediate future, developing the existing CWSA rain gauge network to provide higher temporal resolution rainfall measurements would allow 20 mm rainfall to be used as an effective threshold for lahar initiation to provide alerts to communities living in coastal regions or using river crossings. Independently of our analysis, NEMO has reported at the UN Global Platform 2022 that 20 mm rainfall is sufficient to cause lahars (iWitness News 2022) so this threshold already has importance for national risk management. Installation of additional high temporal-resolution rain gauges close to critical infrastructure would improve capability to alert local communities. It has been demonstrated that in these small island settings significant orographic variation in rainfall can be anticipated for a range of weather systems (Barclay et al. 2006; Poulidis et al. 2016). By involving communities in the curation and collection of rain gauge data in this region some progress has been made with this for communities in the NE through the ‘Changing Landscapes’ project, but more coverage and long-term support for real-time rain gauge networks is needed. In the meantime, working with communities to understand their experience of rainfall thresholds is critically important. Rainfall in St Vincent can be highly localized and rainfall in the lahar source regions does not always extend to the populated areas on the coast, but informal conversations with communities in the NE in Sandy Bay, Owia and Fancy, and analysis of social media posts indicates that people do know how heavy the rain needs to be for ‘the river to come down’ (i.e. a lahar event). The communication of warnings using social media platforms such as WhatsApp is an important short-term step for mitigation of lahar impacts and risk to life.

Longer-term preparedness planning requires detailed characterization of the dynamics of lahars to forecast inundation and impacts. The preliminary lahar modelling presented in this paper provides a framework for a more complete assessment of lahar threat in St Vincent in the future. The inundation modelling conducted using Laharz shows reasonable agreement with the observed lahar event in the Wallibou river on 3 May 2021 and provides a rapid assessment of areas potentially inundated by future lahars, to enable prioritization of higher-resolution dynamic modelling. Tests using the LaharFlow dynamic model on the Overland River catchment show that a simplified hydrological approach assigning a constant runoff ratio for rainfall onto a catchment can be used to set the initial conditions for simulations that show good agreement with observations of lahars at the coast. Using the 20 mm rainfall threshold, runoff ratios in the range 0.2–0.4 could be used as the basis for dynamic modelling of lahars in other catchments. In the absence of detailed hydrological models, which require comprehensive stream and rain gauging to implement, this simplified approach provides a starting point for characterizing the dynamics and potential impacts of lahars. The dynamic modelling also provides a link between instrumental detection of lahars on the seismic network and warnings, as the models compute the travel time of the lahar through the catchment. Two of the Overland case study lahars (25 August 2021 and 1 July 2022) were detected on the seismic network at times before the flow reached the road crossing, and so there is potential with more detailed time recording of future events and modelling to quantify the warning time available from seismic detection of lahars.

Lahar impacts to the road network continue to be a significant source of disruption to communities in the NE of St Vincent. Although these communities experience flash flooding from heavy rainfall that occurs during the rainy season, impacts from lahars are more significant, and the frequency and magnitude of lahar events since the end of the eruptive period on 22 April 2021 has increased. Government response has been to construct bridges at the major river crossings, in essence shifting from the use of fords which was previously the dominant mechanism for roads crossing river valleys in the areas north of the Rabacca River. While the new bridges will alleviate the short-term issues, it is still important to assess potential lahar impacts to these bridge structures and other critical infrastructure include electricity poles, drinking water supply inlets and hydroelectric power equipment situated in river channels. Dynamic modelling of sediment transport can anticipate impacts of future lahar events including bridge scour, loading on structures due to static and dynamic pressures exerted by the lahar flow, and blocking of pipework inlets.

Mitigating future lahar impacts will always require collaboration between scientific and risk management agencies and affected communities. Identifying which valleys and downstream communities lahars will impact can be a challenge when they are detected on the University of the West Indies Seismic Research Centre seismic network, so additional eyewitness observations are essential. Detailed studies of lahar deposits in the immediate aftermath are compromised by the rapid removal and clean-up response necessary to reduce disruption and allow population movement. The results presented in this study depend critically on numerous eyewitness reports of lahars captured on social media (including shared WhatsApp videos). These reports were mainly from people living in communities on the eastern side of the volcano and also increased in frequency after 23 August 2021 when the all-clear was given for residents to return to the occupied areas on the volcano. Video sharing made use of connections made between residents and communities in the ‘Changing Landscapes’ and VolcanoReady projects. These observations were vital in enabling us to compile a comprehensive record of lahar activity and also to test dynamic modelling to a high fidelity. In the future they will remain useful to augment instrumental data from the seismic network and rain gauges, and also provide additional information for understanding impacts at the community and infrastructure level, particularly at bridges and river fords in the north windward communities of St Vincent. Continuing to build strong positive relationships with these communities will be critical for underpinning successful long-term lahar risk management in St Vincent.

Lahars have been a persistent hazard on St Vincent since the April 2021 eruption of La Soufrière, with 25 recorded events in 2021 and ongoing activity, particularly in the rainy season. Typically, the lahars are relatively small volume, dilute or hyperconcentrated, weakly erosive and can occur on all flanks of the volcano. Daily rainfall of 20 mm local to a catchment can be sufficient to trigger a lahar event, although further work is needed to quantify the reliability of this as a threshold. Lahars are likely to be a significant hazard on St Vincent into the future, because of widespread ash deposition and vegetation damage on the flanks of the volcano caused by the 2020–21 eruption of La Soufrière volcano.

The impacts of lahars have resulted in total losses estimated at US$12M. Most of the impacts have occurred to the primary coastal road and the communities in the NE of St Vincent. Damage has been done to their drinking water supplies, road crossings and transportation and electrical distribution equipment. Hydro power plant has been affected by increased turbidity on both sides of the volcano. Damage to the road transport network has also hampered repairs to other infrastructure impacted.

Observations of lahars on St Vincent have been used to inform hazard modelling for both rapid identification of inundated areas using Laharz, and computation of lahar properties and arrival times using the dynamic model LaharFlow. A simplified catchment hydrology approach using the 20 mm daily rainfall threshold seems to perform well in reconstructing a well-observed lahar event, and provides a basis for future more detailed hazard assessment.

Community observations of lahars communicated by social media have been vital in completing the record of lahar events, partly because instrumental detection of lahars by the seismic network cannot identify locations or flow properties. Engagement by scientists with affected communities has, in part, supported the creation of a WhatsApp group as a way of communicating occurrence and impacts of lahars across the NE of St Vincent, which helps those communities reduce risk to life and manage disruption to transportation.

We thank community members from the ‘Changing Landscapes’ community group for their ongoing engagement and support with investigation of lahar occurrence and impacts in their communities. We thank Jennie Gilbert and Jim Vallance for their detailed and helpful reviews that have greatly improved our paper.

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

JP: conceptualization (lead), data curation (lead), formal analysis (lead), funding acquisition (supporting), investigation (lead), methodology (lead), writing – original draft (lead); JB: conceptualization (supporting), funding acquisition (supporting), investigation (supporting), writing – original draft (supporting), writing – review & editing (supporting); PC: conceptualization (supporting), investigation (supporting), writing – original draft (supporting), writing – review & editing (supporting); MJ: conceptualization (supporting), investigation (supporting), writing – review & editing (supporting); VM: conceptualization (supporting), investigation (supporting), writing – original draft (supporting); RR: conceptualization (supporting), investigation (supporting), methodology (supporting), writing – original draft (supporting).

This research was partly funded by the ‘Risk at the Margins’ United Kingdom Research and Innovation Global Challenges Research Fund project EP/T024747/1. JCP acknowledges support from a University of Bristol Research Fellowship. Fieldwork for PDC and JB was funded by the Royal Society APX/R1/180094.

The datasets generated during and/or analysed during the current study are not publicly available due to lack of secure storage repositories for all individual authors but are available from the corresponding author on reasonable request.