Crowdsourcing human observations expands and enhances volcano monitoring records

The study investigates whether crowdsourced human observations can expand and enhance volcano acoustic monitoring records, particularly for audible-frequency signals (>20 Hz) that are often missed by instruments due to filtering and anthropogenic noise. Following the 15 January 2022 Hunga Tonga–Hunga Ha'apai eruption, audible booming was reported worldwide, presenting a rare opportunity to analyze far-field audible acoustic waves. The authors aim to test the reliability of crowdsourced observations by comparing them with geophysical monitoring data, assess spatial-temporal patterns of audibility across New Zealand, and explore added insights about human and environmental impacts. The work situates crowdsourcing as an interdisciplinary bridge between social and physical sciences, addressing perishable data collection, public engagement, and improved transparency in volcano monitoring.

Prior geophysical research documented the Hunga eruption’s atmospheric and tsunami waves globally, particularly Lamb waves and infrasonic signals detected by instruments and satellites. However, audible components were largely absent from many monitoring datasets due to noise mitigation and filtering for frequencies below human hearing. Crowdsourcing has been recognized in geohazards for augmenting scientific data and community engagement, typically via imagery, recordings, and quantitative observations. This study extends that literature by collecting qualitative recall-based observations (timing, loudness, count, descriptors) and integrating them with sensor data. Calls for interdisciplinary approaches to study Earth systems motivate combining social sensing with geophysical monitoring to fill gaps, capture perishable information, and potentially validate emerging physical models of long-range acoustic propagation.

Survey: An online, 39-question multi-hazard post-event survey was co-designed by social scientists, volcanologists, and tsunami scientists. It targeted eruption sounds and tsunami observations across New Zealand, focusing on timing, location, number of sounds, loudness, and qualitative descriptors. Questions included quantitative elements (e.g., count of booms, binned times, loudness choices mapped to indicative dB values) and open-ended responses. Ethical review classified the project as low risk. The survey was hosted on SurveyMonkey and collected responses from 21 January to 13 February 2022 (6–29 days post-eruption) via convenience/public sampling through social and traditional media. Timing data were binned (5–15-minute intervals); spatial analyses used buffers (e.g., 100 km around key stations; 300 km radial bins from the eruption). Qualitative data underwent inductive thematic and content analysis. Loudness used a categorical scale linked to approximate SPL values. Monitoring data and processing: Open-access datasets included: (1) 29 MetService weather stations (minute-sampled barometric pressure) to identify Lamb wave arrival via deviations from a 60-minute moving average; (2) GeoNet network: 168 seismic stations (short-period and broadband) and 19 acoustic stations; seismic and acoustic data sampled at 100 sps. Audible ground-coupled acoustic signals were visually picked in seismic data as sharp-onset, rapidly decaying transients. GeoNet acoustic sensors primarily recorded <20 Hz; audible components were not captured due to noise reduction at >20 Hz. (3) Raspberry Shake & Boom citizen stations at Hamilton (RF356), Wellington (R9066), and Christchurch (R20014), sampled at 100 sps without typical high-frequency noise suppression; data were high-pass filtered (10–20 Hz) to focus on audible frequencies. For RF356, sound pressure level (SPL) was derived from pressure using SPL = 10 log(P/P0), P0 = 20 µPa. Interdisciplinary integration: Crowdsourced and geophysical datasets were compared for trends in number of sounds, loudness, arrival times, and spatial distributions across North and South Islands, with particular attention to alignment with Lamb wave timing and north-to-south propagation.

- Participation and coverage: 1930 participants answered sound-related questions; 1751 reported hearing booms; 1375 observations were geolocated across both islands, concentrated in populated areas (likely sampling bias). No broadscale acoustic shadow zones were evident. - Timing: Four peaks in reported first-arrival times: 19:00, 19:30, 20:00, 20:30 NZDT. 69% (n=790) heard the first sound between 19:00–20:00 NZDT; 35% (n=401) reported 19:00 or 20:00. Mean arrival earlier in North Island (19:21) than South Island (19:29); within 50 km of Dunedin, mean 20:22. Sensor analysis (GeoNet, MetService, Raspberry Shake) showed two phases of audible signals: first arriving ~19:00 NZDT with Lamb wave (~310 m/s apparent velocity), second ~19:45 NZDT (~280 m/s), propagating north-to-south; arrivals at WHZ near Dunedin ~20:00 and ~21:00. - Count of booms: Average reported booms: North Island 5.6 (n=1086), South Island 4.9 (n=163), overall 5.5 (n=1350). Some reported >12 booms (n≥16). Counts generally decreased with distance from the eruption zone. - Loudness: Most frequent loudness was “moderate” (70 dB) at 29% (n=392), followed by “light” (60 dB) at 23% (n=313) and “moderately loud” (80 dB) at 20% (n=268). Near RF356, average reported loudness within 25, 50, 100 km was ~68, 74.4, and 76 dB respectively. Highest average loudness (83 dB) reported in Tairāwhiti Gisborne and Āhuriri Napier (~200 km from Hunga). RF356 identified many potential audible events; seven exceeded 60 dB, with maximum 67 dB. - Qualitative characteristics: Among predefined descriptors, “explosion” (37%, n=507) was most common, then “thumping” (29%, n=390), “thunder” (17%, n=231), “door slamming” (9%, n=124). Free-text frequently referenced fireworks (n=139), guns (n=73), cannons (n=64), and similar percussive sounds. - Physical and social impacts: Reports included building/window rattling and bodily sensations. Estimated maximum overpressures inferred from such effects were ~100–220 Pa. Animal behavior changes were noted by 5% (n=40). Information-seeking/social checking behaviors were common; 6% (n=44) discussed with others or checked social media; 4% (n=41) reported strong emotional responses. About 50% (n=971) were willing to be contacted for further discussion. - Overall: Crowdsourced data and sensors consistently indicated ~5–7 audible signals with peak amplitudes around 60–80 dB, arriving in two ~30-minute phases starting ~19:00 NZDT and ~19:45 NZDT, traveling north-to-south across New Zealand.

The study demonstrates strong concordance between human-reported audible observations and geophysical records during the Hunga eruption, supporting crowdsourcing as a reliable, low-cost complement to traditional monitoring, especially for perishable audible-frequency data often missed by instruments. Human observations added nuance about waveform character (rolling/rumbling clusters), spatial-temporal structure (two phases, north-to-south progression), and environmental and bodily impacts. The inferred overpressures (∼100–220 Pa) and widespread audibility thousands of kilometers from source highlight substantial far-field disturbance and suggest refinements to models of long-range acoustic propagation, including potential nonlinear energy transfer from low-frequency Lamb waves to audible bands. Social responses (information-seeking, comparisons to past eruptions) suggest that audible cues can inform preparedness messaging and risk communication. While sensor limitations (filtering, installation noise) constrained instrumental capture of audible signals, the combined approach verified timing windows, event counts, and relative amplitudes, filling gaps and expanding the monitoring record.

Crowdsourcing of human observations can robustly augment volcano monitoring by capturing audible-frequency signals and associated impacts that are frequently absent from instrumental datasets. For the Hunga eruption, crowdsourced observations aligned with sensor-derived timing and propagation, revealing two phases of ~30 minutes each, north-to-south travel, and typical peak loudness of 60–80 dB with ~5–7 audible booms across New Zealand. Beyond validation, human reports provided unique qualitative insights into sound character, built-environment effects, and social responses. Future work should: (1) develop standardized, ethical real-time and post-event reporting tools to systematically capture acoustic observations (including multiple arrival phases and quiet zones); (2) integrate crowdsourced datasets into benchmarking frameworks for validating physical models, including mechanisms for energy transfer from infrasonic/Lamb waves to audible frequencies; (3) expand distributed low-cost acoustic networks (e.g., citizen stations) and improve understanding of air–ground coupling to better quantify loudness from seismic records; and (4) investigate far-field impact thresholds on structures, humans, and animals to refine hazard assessments and communications.

- Survey design and timing: Data were collected 6–29 days post-event, relying on recall which may introduce timing imprecision (rounding, memory bias) and potential influence from media or social information. The survey was not optimized to capture reports of multiple distinct arrival windows or extended sequences (e.g., 12-hour booms), limiting temporal resolution. - Sampling bias: Convenience sampling led to geographic and self-selection biases; observations concentrated around populated areas, limiting representativeness and the ability to infer absence zones (quiet/shadow regions). - Instrumental constraints: Many professional acoustic sensors filtered out audible frequencies or were affected by anthropogenic noise, limiting direct instrumental records. Limited citizen acoustic stations in NZ reduced spatial coverage. Unknown site-specific air–ground coupling prevents accurate loudness estimation from seismic data. - Ethical and psychosocial considerations: Crowdsourcing during/after impactful events entails ethical risks and potential psychological burden; ensuring participant well-being, data sharing, and equitable access remains essential. - Data uncertainty: Loudness perception varies across individuals and frequencies; qualitative categorizations mapped to dB are approximate. Reported overpressure estimates are inferred from impact analogs and carry uncertainty.