The short life of the volcanic island New Lateʻiki (Tonga) analyzed by multi-sensor remote sensing data

Submarine volcanic eruptions can create new islands whose longevity depends on the composition of erupted material. Long-lived islands (e.g., Surtsey, Nishinoshima, Hunga Tonga–Hunga Ha'apai) enable studies of ecological colonization and may affect national maritime claims, while others are eroded within months. Late'iki (Metis Shoal) on the Tonga Volcanic Arc has repeatedly produced ephemeral islands (1851–1894, 1967–1968, 1979, 1995, 2019). The 1967 and 1979 islands of pyroclastic debris were short-lived; in 1995 a hardened lava dome formed an island that persisted for over 25 years. In mid-October 2019 a new surtseyan eruption occurred at Late'iki. This study aims to provide a comprehensive, multi-sensor satellite analysis (thermal, optical, SAR) of the 2019 eruption, the collapse of the 1995 island, and the birth, evolution, and rapid erosion of the new island (New Late'iki). The work highlights the importance of spaceborne monitoring for remote volcanoes lacking in situ networks.

The paper reviews satellite-based volcano monitoring methods: thermal remote sensing since the 1980s with Landsat TM and NOAA AVHRR supports detection of lava flows, lava lakes, fumaroles, and domes; geostationary platforms (GOES, Meteosat/SEVIRI, MTSAT) and systems (HOTSAT, HOTVOLC, RSTVOLC, MIROVA, MODVOLC) enable near-real-time hotspot detection. VIIRS offers enhanced thermal anomaly capability, and combined MODIS–VIIRS data can estimate lava discharge rates. High-resolution optical imagery supports mapping of lava and pyroclastic deposits. SAR provides day/night, weather-independent observations using amplitude and interferometry to monitor morphology and deformation, with cross-polarized data advantageous at sea. Prior multi-sensor studies integrated optical, thermal, and SAR to analyze eruptions (e.g., Puyehue-Cordón Caulle 2011, Nabro 2011, Kadovar 2018/19). These inform the present multi-sensor approach to Late'iki.

Data: The study used 24 Sentinel-2 MSI high-resolution optical datasets (5 Sep 2019–17 Feb 2020), one Landsat-8 OLI/TIR image, all available MODIS (Aqua/Terra) and VIIRS (Suomi NPP, NOAA-20) acquisitions from 1 Sep 2019 to 29 Feb 2020, 15 Sentinel-1 C-band dual-pol (VV/VH) SAR images, and two TerraSAR-X X-band Spotlight scenes. Seismic catalogs (Global CMT, GEOFON, USGS) were queried due to lack of local stations. Multispectral analysis: Daytime visual inspections used RGB composites (MODIS 1/4/3 sharpened; VIIRS 11/M4/M3 sharpened; Sentinel-2 12/4/2; Landsat-8 7/4/2) to detect plumes and SWIR thermal signatures. Thermal anomaly detection on Sentinel-2 and Landsat-8 employed the Normalized Hotspot Indices (NHI) algorithm, which computes NHISWIR and NHISWNIR from TOA radiances at ~0.8, 1.6, and 2.2 µm (Sentinel-2: 8A/11/12; Landsat-8: 5/6/7). Pixels with NHISWNIR>0 indicate strong thermal emission; NHISWIR>0 indicates less intense hotspots. Total SWIR radiance and hotspot areas were estimated by combining detections from both sensors. Primary processing used the Google Earth Engine NHI Tool (v1.4) and complementary offline analysis via ESA SNAP to mitigate misregistration artifacts. Island outlines and areas (old Late'iki and New Late'iki) were mapped over time from Sentinel-2 and Landsat-8. Thermal low-resolution monitoring: Daily hotspots from MODIS (MIR 21/22; TIR 31) and VIIRS (MIR 14; TIR 15) were used to compute volcanic radiant power (VRP) via the MIR approach. For VIIRS, VRP from M13 (750 m) was apportioned across coincident 375 m hotspot pixels. The maximum daily VRP per sensor overflights was retained. SAR processing: Sentinel-1 SLC and TerraSAR-X SSC data underwent radiometric calibration to sigma nought, refined Lee speckle filtering (7×7), and geocoding to UTM/WGS84 Zone 1S. Due to the ephemeral island, no DEM was used. Dual-pol Sentinel-1 data were subjected to polarimetric processing: refined Lee polarimetric filtering, H/α decomposition (dual-pol), followed by unsupervised Wishart classification (initialized by nine H–α segments; three iterations) to delineate scattering classes. Results were mapped to UTM. Seismic analysis: Events within 100 km radius and <100 km depth around Late'iki were compiled to assess any temporal correspondence, acknowledging the nearest station is ~770 km away.

• Eruption timing and activity: First VIIRS thermal signature on 12 Oct 2019; steam plume observed 14–23 Oct 2019 by MODIS RGB. Thermal activity lasted through 23 Oct 2019 (visual), with MODIS/VIIRS hotspots detected 13–22 Oct. High-resolution observations (Sentinel-2, Landsat-8, Sentinel-1) indicate activity 15–22 Oct. • New island formation: Sentinel-2 on 15 Oct shows New Late'iki northwest of the former island, initially ~350 m (NE–SW) by up to ~60 m (NW–SE). Strong SWIR signatures and water vapor indicated ongoing surtseyan activity in shallow water. • Thermal anomalies (NHI): 16 Oct Landsat-8 OLI detected a single hotspot pixel with radiances 15.96 W·m−2·sr−1·µm−1 (1.6 µm) and 17.28 W·m−2·sr−1·µm−1 (2.2 µm); hotspot area ~900 m². 20 Oct Sentinel-2 detected four anomalies totaling 53.815 W·m−2·sr−1·µm−1 (1.6 µm) and 61.167 W·m−2·sr−1·µm−1 (2.2 µm); hotspot area ~1,600 m². 15 Oct Sentinel-2 NHI detection was impeded by plume contamination. • SAR backscatter: Sentinel-1 VH showed increased backscatter over a ~700 m (N–S) by 500 m (E–W) area around old Late'iki on 17 and 22 Oct, aligning with thermal/SWIR anomalies. Wishart classification separated strong-backscatter regions (one class on 17 Oct; two classes on 22 Oct), consistent with evolving eruptive processes. VV was more affected by sea surface roughness than VH. • Island area evolution: New Late'iki reached a maximum area of ~21,000 m² on 30 Oct 2019. It eroded rapidly thereafter: 30 Oct–19 Nov erosion rate ~960 m²/day; after 19 Nov ~124 m²/day; overall average ~464 m²/day. The island was submerged by 14 Dec 2019; by Feb 2020 only subtle seabed features/discoloration remained visible. • Comparative erosion: The 1995 lava-dome island decreased from ~75,000 m² (late Jun 1995) to ~400 m² just before Oct 2019, implying an average erosion rate of ~8.4 m²/day—approximately 55 times lower than New Late'iki’s average rate, indicating more resistant composition in 1995. • VRP: During the first three days, VRP increased to a MODIS maximum of 26.4 MW; average ~15 MW over the next six days; final higher VRP up to 49 MW on 21–22 Oct. MODVOLC recorded alerts on 18, 20, and 22 Oct, consistent with these observations. • Plume and gases: Monitoring revealed steam plumes without ash (confirmed by VAAC Wellington advisories). Sentinel-5P SO2 data showed no strong SO2 enhancement; however, persistent seawater discoloration during and after the eruption suggests continued degassing (e.g., SO2, CO2, H2S) similar to observations at Home Reef (2006). • Sensor performance: Geostationary AHI/GOES imagery at the study location lacked sufficient spatial resolution; ASTER scenes were clouded. TerraSAR-X detected residual backscatter at the island location on 2 Dec but little by 4 Dec.

The multi-sensor analysis demonstrates that the mid-October 2019 Late'iki eruption produced a short-lived island composed likely of easily erodible pyroclastic material, explaining its rapid destruction compared to the long-lived 1995 lava-dome island. Integrating Sentinel-2/Landsat-8 SWIR thermal indicators with Sentinel-1 SAR backscatter provided a coherent picture of the eruptive area and its evolution despite plume and cloud masking in optical data. VH-polarized SAR was particularly effective for small island detection in oceanic conditions, whereas VV suffered from sea surface roughness. The absence of ash signals and the presence of steam plumes are consistent with surtseyan activity in shallow water. Although Sentinel-5P did not show distinct SO2 anomalies, persistent seawater discoloration implies ongoing degassing for months after the eruption. The findings underscore the value of combined thermal, optical, and SAR satellite observations to monitor remote submarine volcanoes lacking local seismic networks, where geostationary data and distant seismic stations are insufficient for reliable detection.

A comprehensive multi-sensor satellite assessment (MODIS, VIIRS, Sentinel-2, Landsat-8, Sentinel-1, TerraSAR-X) of the October 2019 Late'iki eruption documented the collapse of the 1995 island and formation of New Late'iki. The new island expanded to ~21,000 m² by 30 October and was completely eroded within two months, exhibiting an average erosion rate ~55 times higher than the 1995 lava-dome island, consistent with a composition of easily erodible pyroclastic debris. The study highlights the crucial role of coordinated, high-temporal and high-spatial resolution satellite sensors—particularly the synergy of SAR and optical/thermal data—in monitoring eruptions at remote volcanoes where terrestrial networks are absent. Future work could leverage night-time high-resolution thermal observations, improved gas retrievals, and dedicated small-satellite tasking to better constrain eruptive energetics, composition, and post-eruptive degassing.

• Lack of local monitoring: The nearest seismic station (~770 km away) prevented clear attribution of seismic signals to the eruption. • Observation conditions: No night-time high-resolution optical acquisitions during the eruption; thermal radiance estimates are affected by daytime solar contamination and plume obscuration. Many scenes suffered cloud cover; ASTER images were unusable. • Sensor constraints: Geostationary AHI/GOES spatial resolution at Tonga was too coarse to detect activity. No DEM was available for the nascent island. Sentinel-5P SO2 retrievals did not show clear enhancements, limiting gas quantification. • Oceanic SAR challenges: Interpretation is complicated by sea surface roughness and breaking waves, especially in VV polarization.