Ongoing Dispersal of the 7 August 2019 Pumice Raft From the Tonga Arc in the Southwestern Pacific Ocean

2.1 Satellite Surveillance

Freely available moderate resolution natural reflectance images captured by sun-synchronous satellites such as MODIS Terra and Aqua (250-m pixel), Sentinel-3 (300-m pixel), and Suomi-VIIRS (375-m pixel) are very efficient for tracking large pumice rafts in real time. Their sequential passing over of the same area within a few hours daily increases the chances of identifying rafts in clear to partly cloudy weather. High-resolution satellite images from Sentinel-2 (10-m pixel) and LANDSAT-8 (30-m pixel) are excellent for raft tracking and identification of small features, although imaging frequency is ca. 3 days. The radar-equipped Sentinel-1 satellite (10-m pixel) in Interferometric Wide Swath mode (single VV polarization) is excellent at identifying nondilute rafts in any weather, although imaging frequency is low. Coastal ocean images obtained from the commercial fleet of Planet satellites are down to 0.8-m pixel. Our satellite image analysis and dispersal maps use combined data from all these satellites.

2.2 Oceanographic Simulations

We conducted short-term (1 month) forecast and long-term (2 year) hindcast simulations based on data from CMEMS (see acknowledgments and supporting information Text S1). These data include ocean currents and Stokes drift on 1/12° horizontal grid, as well as wind data on 1/4° horizontal grid, which each provide different dispersal patterns (Onink et al., 2019). Particles were released at (18.307°S 174.395°W) on the 7 August. Particle trajectories were calculated using Parcels (Delandmeter & van Sebille, 2019) (OceanParcels.org), a rapid, customizable framework for Lagrangian particle tracking simulations using output from Ocean Circulation models. The full code of these simulations is available at https://github.com/OceanParcels/PumiceEvent.

A key factor in pumice dispersal is windage, included as a fraction of the wind vector from the ocean model simulation. If a typical value of 1% windage is added to the surface current vector (e.g., Putman et al., 2018), then for usual wind speeds 5–10 m/s (trade winds) and current speeds 0.1–0.2 m/s (surface flow in the tropics), windage is about half the current speed. While previous estimates of pumice drift incorporated this wind percentage (around 50%) of current speed (Bryan et al., 2004), we more accurately include a windage effect that is independent of the current.

3.1 Close Encounter With the New Pumice Raft

On the afternoon and evening of the 9 August (all times are local, UTC + 13), SV Finely Finished and MY Olive crossed and sampled the main raft only 2 days after the eruption, providing invaluable information on the products of this submarine eruption (Figures 1, S1–S6; Movies S1 and S2). The crews' logbooks, photos, and videos described the main raft to be a compact, 15–30 cm thick layer of pumice clasts. The diameters of most clasts are a few centimeters, but range 0.3–80 cm. Patchy raft ribbons (elongated rafts) alongside or behind the main raft were composed of small (<1 km), one-clast thick rafts (Figure 2). These measurements combined with the raft area imply a volume of at least 30 million m³ 2 days after the eruption. On the 11 August, SV Barbarossa crossed the 2-km-wide main raft (Figure 3a). The reported raft was one-clast thick and predominantly comprised subrounded pumice clasts ca. 6 cm in size, with outliers at 0.2 and 40 cm, matching reports from all vessels that encountered the raft, including SY Arka and SY Roam on the 14 and 15 August, respectively (Figure 3a). All sampled pumice clasts are pale grey to pale brown, include coarse phenocrysts, and are highly vesicular (Figure 1). Chemically, the pumice is andesitic (Table S1), less evolved than the 2001 dacite (Bryan et al., 2004). The pumice clasts collected over the first week are well rounded to subangular, and display polyhedral joints, weak cauliflower texture, and a gradational rim-to-core increase in vesicle size typical of water quenching (McPhie et al., 1993) (Figure 1). The average pumice size (2–6 cm) matches the size of clasts that are able to reach the ocean surface from a submarine vent at ~200 mbsl prior to ingesting enough water to sink (Fauria & Manga, 2018) (Text S2; Figure S7).

3.2 A Submarine Eruption

This eruption is the first reported activity of this volcano since its pumice-raft forming eruption from a nearby vent in 2001 (Brandl et al., 2019; Bryan et al., 2004; Global Volcanism Program, 2019). Precursory activity included a 5.7 Mw earthquake at 10:01 pm on the 5 August, at 15 km south-southwest of the volcano and 10 km depth, followed by six events of around magnitude 4 over the next 2 days (Brandl et al., 2019; International Seismological Centre, 2019) (Figure S8). A Sentinel-1 satellite image on the 6 August, 7:07 pm showed no surficial evidence of eruptive activity, and no activity has been imaged since at least 2013 (Movies S3 and S4). High-resolution Sentinel-2 satellite imagery at 11:01 am on the 7 August shows two concentric rings of 225 and 1,150 m diameter emerging from a point source centered at (18.307°S 174.395°W) (Figure 2). This point source corresponds to a 200 mbsl bathymetric high on the northernmost rim of the 8 × 5 km, 700 mbsl unnamed submarine caldera volcano (Brandl et al., 2019; Global Volcanism Program, 2019; Taylor, 2012). This vent location is 1.5 km northwest from the supposed 2001 eruption site and situated on the same broad bathymetric high and on the caldera rim (Figure S9). The two rings strongly resemble breaching of billowing underwater eruption columns or steam cupolas generated during shallow submarine eruptions (Kokelaar & Durant, 1983; Somoza et al., 2017). No pumice raft is visible in the satellite image of the rings (pixel size 10 m), therefore pumice raft formation occurred subsequently. Raft velocity on the 9 August suggests pumice raft formation at midday on the 7 August 2019 (Text S3).

We observed ca. 10 m wide, white linear trails scattered above the concentric rings that we interpret to be condensed water vapor blown northwestward by surface wind (Figure 2a). This water vapor was likely derived from very coarse, warm pumice clasts and/or fluid that breached the water surface (Reynolds et al., 1980). A low-altitude, white atmospheric plume fanned out at 2–11 km northwest (downwind) from the inferred vent. This plume, presumably made of condensed water vapor, is attributed to the eruptive pulse that produced the largest billowing ring (Figure 2). No ash-rich or SO₂-rich (Sentinel-5 satellite) eruption plume is observed in any of the satellite images. The atmospheric wind profile (Figure S10) and plume direction constrain a plume height <1,000 m.

Available satellite images show minor activity at the vent at least on the 9 and 11 August, with emission of a trail of pale blue-green discolored water and a delayed, small (<1 km²) and dilute raft at 2–5 km from the vent (11 August). The eruption style cannot be ascertained without further textural and chemical analyses of the pumice clasts, a post-eruption bathymetry survey, and seafloor sampling of proximal volcanic products. The first satellite image shows a strongly irregular-shaped raft, suggesting successive raft-forming eruption pulses. The low viscosity of the erupted andesite hinders fragmentation in the conduit (Text S4) so that clasts may be generated by quenching in the ocean, consistent with some of the observed textures.

3.3 Pumice Raft Formation and Dispersal

The edges of the raft were digitized from georeferenced satellite images imported into QGIS software. Our time-series from the 9 August to the 3 September includes 61 satellite passes over the raft, allowing precise reporting of its geometry and areal distribution (Figure 3). The raft drifted 5–54.5 km/day (22.8 km/day average, matching global dispersal speeds; Jutzeler et al., 2014) westward overall but was entrained into ocean loops or eddies and changed direction (Figure 3; Text S5). The early raft was 195 km² (9 August) and sailing crew observations of raft thinning from 30–50 cm to one-clast thick (5 cm) from the 9 to the 11 August imply areal dispersal and pumice abrasion (Jutzeler et al., 2014). Over the next weeks of dispersal, the raft broke up into hundreds of small (<1 km²) rafts, which covered an area of <1,600 km² on the 30 August, not including rafts <10 m. Ribbons tend to get segmented into lenses of same wavelength and orientation as the swell (Figure 2c) suggesting that Stokes drift actively contributes to breaking up rafts. The raft passed over Oneata's lagoon and then stranded on Lakeba Island for several days until continuing its westward route toward Fiji. On the 22–26 September, the raft passed through the two main Fiji Islands and most of the raft got stranded 60 km to the west, on the Yasawa Islands (Figure 2f). A few episodes of wind changes allowed fractions of the raft to be released back into the ocean; the shores were eventually cleared on the 29 December (Text S4; Figures S11–S13).

Over the first two weeks of drifting, low-level cloud trails formed above the large (>6 km) raft and often extended downstream (Figure 2b). We infer that these clouds formed by retention of solar radiation on the rafted pumice, favoring evaporation of sea water, similar to an island effect (Yang et al., 2008). This implies that pumice rafts, if big enough, can create their own microclimate.

3.4 Degradation of the Raft Through Mechanical Abrasion

Rafted pumice clasts undergo continuous mechanical abrasion by the constant movement of waves and swell causing clasts to move past each other. All sailing crews reported intense clinking noises emitted from the raft. Many of the pumice clasts were subrounded on day 2 of dispersal, and rounded on day 4, attesting to their rapid abrasion (Jutzeler et al., 2017) and thus decrease in raft volume (Figure 1). Satellite images show further evidence of significant clast abrasion as seen in the immense pale-blue discolored water patches that envelope and follow the largest rafts for several weeks post eruption (Figure 2). This pale-blue discolored water is not connected to the submarine eruption plume (Mantas et al., 2011), and likely consists of a fine suspension of glass shards generated by abrasion in the raft, in addition to isolated sinking pumice clasts that will eventually settle to the seafloor (Jutzeler et al., 2014; Jutzeler et al., 2017). Further, blockage of the raft on the eastern shores of the Yasawa Islands for >2 months reveal areal shrinking of the rafts by stranding and/or abrasion by surf. The volume loss of pumice by waterlogging is difficult to calculate without seafloor samples. Sampled pumices from week 1 of dispersal show subangular to rounded pumice clasts, for a minimum volume loss of 10 vol.% (Jutzeler et al., 2014), equivalent to 3 million m³. Two yacht crews mentioned a pronounced smell of sulfur crossing the rafts on days 2 and 6 of raft dispersal. Sulfur dioxide and hydrogen sulfide are volcanic gases that are trapped within isolated pore space of pumice. Its detection by the crews far from the vent and several days after the eruption attests of its release by fragmentation of the delicate pumice vesicle walls by rafting-induced abrasion, rather than originating directly from an eruption plume that would have been rapidly dispersed by surface wind.

3.5 Forecasting Raft Dispersal by Ocean Currents, Waves, and Winds

We initially simulated dispersal of the raft for 1 month, under a range of windage values and Stokes drift (Figures 4a, 4b, and S14). The dispersal simulation reproduces actual raft drift relatively well. The simulation shows a very good match in westward dispersal direction, and includes sinuosities that broadly but not completely match those observed in satellite images. Simulated dispersion matches satellite data for the first week, but then becomes 20–30% faster over the following weeks. Among the four simulations presented (Figure S14), the best match is with a 1% windage and no Stokes drift (Figure 4a).

The long-term simulations show predominant westward dispersal by the South Equatorial Current, assisted by south-easterly trade winds (Figures 4, S15–S17). Earliest arrival times on the eastern Australian coast fall in the range 300–540 days (Figure 4a), or March 2020 to January 2021, matching raft dispersal rates in this region (Bryan et al., 2004; Jutzeler et al., 2014). There is a trend for generally later arrival times as we move southward along the eastern seaboard of Australia. These averages “hide” considerable interannual variability in the hindcast (Figure S17), most evident in the wide meridional spread and the common occurrence of an additional eastward dispersal. Approaching the eastern coast of Australia, particles move either northward and further westward through the Indonesian Archipelago, or southward with the East Australian Current. These results are sensitive to the choice of windage parameter, with higher windages leading to a generally more westward drift. Adding of Stokes drift led to a slightly more northward pathway. In all cases, however, the modeled plume was transported faster than in the satellite observations (Figures 4a, 4b, and S14). Of consequence here, we note looping of the observed trajectory in the longitude range 175–177°W during days 5–15, broadly consistent with inertial motion at this latitude under a sudden change in winds. The discrepancy between simulated and observed drift may thus be a consequence of inertial currents that are unresolved in the CMEMS data, although additional unresolved motion may be attributed to small-scale eddies. The El Niño–Southern Oscillation was relatively neutral at time of raft dispersal (see https://www.esrl.noaa.gov/psd/enso/mei/). We therefore expect the dispersal pathways to be similar to e.g., 2003–2006. For the model hindcast, we consider the cumulative coverage over 14 years (2000–2013) to be representative of much of the variability in ocean currents.