3.1 The case of the 2018 Palu earthquake

Our compilation of the Twitter exchanges following the Palu earthquake and tsunami reveals how we can rapidly gain a first-order understanding of event characteristics within a few hours to 1 day and a more complete one in less than a week. After the initial tweets issued by responding agencies (e.g. the USGS in the USA and the BMKG in Indonesia), most of the exchanges we quote involved academic researchers from different countries and institutions (see Tables 1, S1) with specialities encompassing seismology, earthquake geology, tectonic geodesy, remote sensing, natural hazards, and science communication. We will not investigate the sociology of the people involved in the Twitter discussions in more detail, because it is outside the scope of the present study, but future work should address this critical subject.

Figure 2Screenshots of tweets chosen to illustrate how researchers shared and explained observations regarding the Palu earthquake and tsunami of 28 September 2018. A simplified timeline (from Fig. 1) is shown on the left for reference. (a) Within 2 h geoscientists described the seismotectonic context of the earthquake. (b) Geoscientists shared and translated the official validation of viral videos about the tsunami in Palu. (c) Researchers hypothesized supershear rupture. (d) Geoscientists shared satellite image correlation results showing a sharp rupture with a 5 m left-lateral offset across Palu town, and other researchers started to discuss these results. Refer to Fig. 3 for tweets about the tsunami warning.

The timeline built from the Twitter feeds (Figs. 1, 2, Table A1) shows that the geoscience community already knew the following about 1 day after the earthquake:

• i.

  the earthquake happened on the Palu-Koro Fault system, with a sharply localized strike-slip rupture directly beneath Palu City, and the epicentre was located in the Minahasa Peninsula on the north-east shore of Palu Bay (from earthquake location and moment tensor solutions provided by monitoring agencies, published papers on the seismotectonic context, and regional fault mapping; this information was shared via Twitter in the 2 h following the event – see Fig. 2a);

• ii.

  the rupture entered Palu Bay, but the geometry of its prolongation offshore toward the Minahasa Peninsula was uncertain (from early post-earthquake satellite imagery and preliminary image correlation using pre- and post-event data);

• iii.

  the aftershock zone extended ∼150 km in the north–south direction, and the mainshock hypocentre was located near the northern tip of this zone (from operational earthquake locations provided by monitoring agencies);

• iv.

  a tsunami with run-ups of several metres hit the shores of Palu Bay but was not recorded out of the bay (from reports and videos shared via social media by local people and the tide gauge records that were available in the hours following the event – see Fig. 2b);

• v.

  there was dramatic surface spreading and liquefaction in and south-east of Palu City (from photos and videos shared by locals).

The exchanges and discussions continued via Twitter, and by 5 d after the earthquake the geoscientific community had assembled a fairly accurate description of the event and its effects. The acquired common and credible knowledge was as follows:

• i.

  the earthquake ruptured two strands of the Palu-Koro Fault system for a total length of ∼150 km (from the aftershock distribution provided by monitoring agencies, radar and optical image analysis results, and early earthquake source models);

• ii.

  the strand south of Palu Bay had a sharp and extremely localized surface rupture with sinistral offsets of ∼5 m (from satellite imagery and state-of-the-art pre-and post-event image correlation, which was later confirmed by field observations posted on Twitter by Indonesian researchers ∼15 d after the event – see Fig. 2d);

• iii.

  the rupture started on an inland fault east of Palu Bay and then crossed Palu City from north to south (from satellite and InSAR imagery as well as early earthquake source models);

• iv.

  the earthquake rupture propagated unilaterally southward, likely at a supershear speed (faster than S waves), which was a fairly unique observation for earthquakes (from early earthquake source duration and rupture length estimates, with the latter based first on the distribution of early aftershocks and then on satellite images – see Fig. 2c);

• v.

  massive liquefaction and lateral spreading occurred in several sectors of Palu City (from aerial video footage shared by local government agencies, satellite imagery, and photos and videos shared by locals on social media);

• vi.

  tsunami waves hit the Palu Bay coast only a few minutes after the earthquake (from tide gauge records and videos shared on social media).

Ensuing Twitter exchanges during the next weeks focussed on the surface rupture description in the field by Indonesian scientists, the bathymetry of Palu Bay, the possible fault geometry across it, and hypotheses about the tsunami source. These hypotheses explored whether the tsunami was due to the seismic rupture itself, due to underwater landslides and coastal collapse, or due to a combination of the two.

In this process of common knowledge building, geoscientists used a diverse range of data types that were openly shared and discussed on Twitter: published papers and maps about the seismotectonic context, teleseismic data, local seismic waveforms, high-resolution optical satellite images, synthetic aperture radar (SAR) satellite data analysis, tide gauge records, and field observations from both science groups and local residents. Data sharing and social interaction via Twitter appeared to be an effective way of getting prompt and diverse feedback from fellow researchers on early scientific ideas. The satellite image correlation results, available on Twitter 1–2 d after the earthquake, were then rapidly shared as a more formal report via the open repository zenodo.org (Valkaniotis et al., 2018). Some ideas and initial hypotheses about a supershear rupture and the offshore fault geometry in Palu Bay, both discussed on Twitter, provided impetus for accelerated development of in-depth scientific papers (Bao et al., 2019; Ulrich et al., 2019). Indonesian geoscientists, absent from the earlier scholarly exchanges on Twitter (only official agencies were providing advice), progressively joined the discussion, providing, for example, tide gauge records and field observations of fault surface rupture and offsets. This created the opportunity to develop new international collaborations. Further highlighting the use of social media, an analysis of the tsunami source by Carvajal et al. (2019) used videos posted on social media platforms such as Twitter and YouTube.

The spread of information via Twitter was not restricted to a small group of geoscience scholars. Journalists used and quoted these Twitter discussions in their articles (e.g. Andrews, 2018a; Wei-Haas, 2018a), using the thread to identify academic experts to interview for their articles. However, some journalists were not interested in the full range of geophysical observations and focussed instead on a “failed tsunami alert” (Fountain, 2018; Wright, 2018). Based on an Associated Press (AP) dispatch, on 1 October 2018, quoting some scientists (Wright, 2018), there were inaccurate reports from international media outlets about a “failed” tsunami warning. According to these reports, a network of tide gauges and buoys would have been able to issue an early tsunami warning after the earthquake, thereby saving lives. The media were quick to blame the Indonesian authorities, saying that such a warning would have been impossible because the Indonesian buoy network was not well maintained. However, geoscientists realized that there was not enough time to issue any warning given the very short distance between the earthquake source and the areas exposed to the tsunami in the very narrow Palu Bay (Fig. 3). As stated by Carjaval et al. (2019), “the most remarkable features of the tsunamis that devastated Palu were the very short, nearly instantaneous arrival times”. The first tsunami waves indeed hit the coast between 1 and 2 min after the earthquake. After the evidence-based explanation given by scholars on Twitter (Fig. 3), the process of fact-checking by some journalists took only a few hours after the publication of the AP dispatch (e.g. Morin, 2018).

Figure 3Screenshots of selected tweets about the tsunami warning in the case of Palu. (a, b) Geoscientists quote media articles regarding a possible “failed” tsunami warning and explain that such a warning was extremely difficult in the case of the Palu earthquake (see text for more details). (c) An example of geoscientists engaging in discussion with local people. (d) A geoscientist reports that Indonesian agencies issued an alert in due time and cancelled it only after the tsunami hit Palu.

As described above, the case of the Palu earthquake and tsunami provides an excellent example of how scholarly discussions on Twitter can provide initial and rapid scientific results, whilst also reinforcing local official authorities on the ground and helping to guide journalistic outputs.

3.2 The Mayotte 11 November 2018 rumble event

On 11 November 2018, more than 6 months after the start of an earthquake swarm between Madagascar and the Comoros archipelago in the Indian Ocean, a peculiar seismic signal radiated from the region of Mayotte. The signal was recorded worldwide by seismic networks, but it was not detected by their automatic event identification algorithms because of its odd spectral characteristics. It was an unusually long, low-frequency, highly monochromatic signal, like a low-pitched hum that travelled as seismic waves across the Earth.

As noted by journalist Maya Wei Haas in her National Geographic article “only one person noticed the odd signal on the U.S. Geological Survey's real-time seismogram displays. An earthquake enthusiast […] saw the curious zigzags and posted images of them to Twitter” (Wei-Haas, 2018b). This image (Fig. 4a) was then retweeted by a citizen earthquake researcher, Jamie Gurney, who initiated an active discussion between academic researchers (Fig. 4), with some interactions from the media and the public. Analysis of openly accessible seismic waveform data from around the world by seismologists then confirmed the signal originated in the Mayotte region (e.g. Hicks, 2018a).

Figure 4Screenshots illustrating early Twitter exchanges about the very long-period seismic signal near Mayotte on 11 November 2018. The selected screenshots show that Twitter discussion was initiated by citizen scientists (a–c) and then progressively involved academic researchers (d–f). Those researchers then started an active discussion about the seismic signal and its possible origin (e–j).

The Twitter discussion involved a group of seismologists but also specialists in earthquake geology, volcanology, tectonics, geodesy, geomechanics, hazards, and science communication. Their exchanges eventually co-built a rapid appraisal of the 11 November signal and of its broader geophysical and geological context. The nature of the researchers' interactions are exemplified by the three successive Twitter moments (Lacassin, 2018a, b, c) that regroup our compilation of tweets (see Fig. 4 for a choice of tweets illustrating the early discussion between researchers). A simple content analysis of the selected tweet threads, illustrated by the two successive word clouds in Fig. 5, shows how the exchanges started with questions about the odd seismic signal itself (using words such as signal, event(s), wave(s), seismic, and frequency) and its geographic origin (using words such as Mayotte and location) (Fig. 5a) and then moved to a discussion more focussed on the event's geophysical source (using words such as source, signal, CMT, CLVD, and deformation) and data processing (using words such as data, model, InSAR, and inversion) (Fig. 5b). While many things remain to be understood about the geophysical processes at work offshore of Mayotte, the preliminary waveform modelling shared via Twitter (Hicks, 2018b) and the related discussion resulted in the consensus hypothesis that the 11 November seismic signal was due to a deflation event in a large and deep magmatic chamber combined with resonance and amplification of the seismic waves. This early hypothesis discussed on Twitter was subsequently supported by later in-depth analyses (Lemoine et al., 2019; Cesca et al., 2020; Feuillet et al., 2020).

Figure 5Word clouds illustrating the evolution of topics discussed on Twitter after the Mayotte 11 November 2018 very long-period (VLP) seismic event. The top word cloud (a) illustrates the first 60 tweets of the selected Twitter moment with most frequent words regarding the VLP signal (signal, event(s), wave(s), seismic, and frequency) and its geographic origin (Mayotte and location). The bottom word cloud (b), which corresponds to the subsequent 60 tweets, shows a discussion more focussed on the geophysical source of the VLP event (source, signal, CMT, CLVD, and deformation) and data processing (data, model, InSAR, and inversion).

The Twitter interactions on Mayotte drew the attention of the global geoscience community. Before the 11 November event, the long-standing earthquake swarm near Mayotte had been largely ignored; the swarm had only been studied by only a few researchers, mainly French, because Mayotte is a French territory. As noted by Lemoine et al. (2019) , the 11 November event “awakened the interest of the seismological community and the media”. We understand that the rapid “explosion” of the informal Twitter discussions that we report played a pivotal role in this awakening and helped hasten needed research in the region (Hicks, 2019). A few days after the 11 November event, at a meeting between the French geoscience community and stakeholders (funding agencies and ministry representatives), the Twitter exchanges were used to demonstrate the urgency with respect to funding research and surveys on the Mayotte earthquake swarm (Nathalie Feuillet, personal communication to RL, 2018).

The full interactive process on Twitter was the subject of two long articles in National Geographic (Wei-Haas, 2018b) and Gizmodo (Andrews, 2018b), with journalists gathering information and contacting researchers via Twitter before interviewing them via email or phone (Fig. 6). These articles were then used as primary sources by other media and stimulated stand-alone reports by more traditional news organizations (e.g. Sample, 2018).

Figure 6Screenshots of tweets by journalists Maya Wei-Haas and Robin George Andrews. After promoting their media article on the Mayotte 11 November 2018 event (a, d), the journalists acknowledged academic researchers who were first identified and contacted via Twitter and then interviewed via email or phone (b, c, e).

The long thread about the Mayotte 11 November seismic event reveals the efficiency of knowledge building via scholarly online interactions, but it also outlines some pitfalls that are inherent in the informal aspect of exchanges via Twitter. While after the Palu earthquake and tsunami geoscientists were posting solid observations (i.e. “knowns”), for Mayotte they were trying to understand a peculiar event with large uncertainties, thereby opening many secondary discussions about “unknowns”. The resulting “bushy” nature of the thread makes it difficult to follow and comprehend in real time, and summarizing it a posteriori is challenging. Moreover, some of these secondary discussions were casual or humorous and were at risk of being seen as insensitive and taken out of context by the general public. We infer that scientific Twitter exchanges dealing with uncertainties and unknowns, as for Mayotte, are more prone to such pitfalls than those sharing knowns.

4.1 Argument 1: very rapid co-building of knowledge

The two case studies described above support previous work showing that Twitter allows the rapid building of knowledge (e.g. Choo et al., 2015; Hicks, 2019). In the case of the 2018 M_w 7.5 Palu earthquake, it took only 5 d to obtain a detailed description of the events and only a few days for the 11 November 2018 seismovolcanic event in Mayotte. It takes several months to years for scientific teams to gather relevant information, analyse it, and publish it in an academic journal after a long review–revision process. Thus, using Twitter makes information and basic explanations accessible to the scientific community and to the public more quickly. Communicating such ideas to the public may have a high impact in places where operational infrastructure and associated communication are limited. Moreover, Twitter provides direct and early scientific information for researchers, without any geographical and institutional barriers, acting as a “science news feed” that can be used to plan further in-depth research.

However, the knowledge built via Twitter is not exactly comparable to the knowledge built by a longer-term, classical academic approach. Even if a long practice of research allows scientists to estimate the quality of a data set or of a methodology almost immediately (if not intuitively), it does not substitute peer review as a process to check the validity of a result and “establish” knowledge. Therefore, a question arises regarding the credibility and legitimacy of the knowledge built rapidly and without peer review via Twitter: can it be believed and on what ground? The fact that the author of a tweet comes from a recognized expert institution increases his/her credibility. Nevertheless, this is not enough to ensure the scientific quality of his/her tweet – and the reverse is also true. As shown in the Mayotte example, nonpractising researchers and “hobby scientists” can develop a good scientific understanding and can legitimately discuss these topics (Fig. 4). Thus, the following question arises: how can we ensure that the most qualified comments receive the most attention?

Rapid dissemination of early scientific analysis products (for example, using up-to-date remote sensing data) to scientists working in the field is another aspect of using social media platforms. This use of social media is similar to the modern trends of using preprint servers for early sharing of scientific results. Twitter interaction is now also forming the basis of collaborations, leading to the development of ideas and the subsequent co-writing of papers within diverse, multidisciplinary teams, e.g. Hicks et al. (2019) and Ulrich et al. (2019) included co-authorships that were instigated from Twitter discussions. By widening stakeholder interactions, such open discussions may also help to enhance the scholarly value of open data sets.

A risk to sharing “breaking science” information on Twitter and social media is that this same information can enable publications by the global community before the local scientists who provided the initial information. There are vulnerabilities for those field teams who are committing resources as part of a response initiative and are required to, or feel a duty to, provide timely public information about an event. Elements of such a scenario unfolded following the 2016 Kaikōura earthquake in New Zealand, when tweets, blog posts, and media releases by the responding agencies were an important information source for an early publication by researchers without collaboration with the responding agency scientists. This publication (Shi et al., 2017) predated publications of field observations and analysis by teams on the ground by several months. This example raises questions about the ownership of scientific knowledge that is shared in the public domain and suggests that some scientists may choose to completely restrict, or be more selective about, publicly posting their scientific analysis into the public domain.

4.2 Argument 2: science beyond the laboratory

Twitter allows us to step outside the laboratory in many ways. First, it opens the door to professional networking and new academic collaborations between scientists from different disciplines, institutions, or even countries. In the case of the Palu earthquake, most of the early exchanges involved non-Indonesian academic researchers; Indonesian geoscientists then joined the discussion and provided data that could only be acquired locally (e.g. field observations about the earthquake rupture or liquefaction-induced landslides). This led them to engage in a discussion with other members of the international scientific community and paved the way for new collaborations, such as the sharing of tsunami source models for operational hazard analyses. In the short term, however, it might be difficult for local scientists to get involved in social media if they are busy with the management of the crisis and/or collecting the first information from the field. Furthermore, scientists from local monitoring organizations or universities may have strict social media usage and communication policies.

Twitter also opens the door to exchanges with the global public. The scientific value of contributions from nonacademics varies between examples, but there are always some external inputs that help to clarify or reframe the scientific questions and the way to explain them to the public. Nonacademics can launch important discussions. In the case of Mayotte, it was a citizen scientist who drew attention to a strange seismic signal (Fig. 4a), and it was the subsequent “explosion” of informal Twitter discussions that woke the scientists and the authorities up (Lemoine et al., 2019; Hicks, 2019). Among Twitter users, journalists “listening in” are particularly important as they can pass on some of the scientific content of the discussions in an understandable way. The challenge for them is to have access to information that is as fresh as it is credible. From this point of view, Twitter is an important resource because it can serve as a pool of potential experts to give in-depth comment (Fig. 6). Conversely, perhaps this trend reduces the diversity in these pools, with public comment favouring scientists on Twitter rather than those who avoid Twitter and/or use other social media platforms. Moreover, how much checking does a journalist do to assess a tweeter's scientific credibility?

4.3 Argument 3: opening the scientific process to the public

The process of knowledge building on Twitter is open and public, which may help to improve the general public's and the media's understanding of how scientific research works. The examples described above show that the process of knowledge co-construction is not linear. Some discussion threads might look like well-structured “trees” (e.g. the Palu earthquake) but others resemble “wild bushes” with many secondary branches of discussions opening up over time (e.g. the Mayotte seismovolcanic crisis). Scientists are seen by the public to use a wide variety of data and following indirect, unchronological, and unstructured thought paths before reaching a conclusion. As a window to the scientific process, Twitter also helps to clarify that scientific work is organized into disciplines and subdisciplines and that the knowledge and know-how of these groups may be difficult to articulate, but they are all necessary to build a global view of a subject. Scientists themselves are familiar with these aspects of their work but non-scientists may not be, largely because scientific knowledge is often presented retrospectively as having been constructed in a cumulative and chronological manner. Epistemologists have long denounced this misconception (e.g. Kuhn, 1996). Thus, Twitter can contribute to making the “messy part of science” more tangible and visible. Early information on Twitter can also provide excellent teachable material for educators.

One limitation is that the thread has to be “visible” on Twitter, using a proper hashtag for instance. Also, if the public is not aware of the sphere and the discussion is not ”visible” to them, they will not see it even though it is public. Moreover, some schools of thought, especially those from a public safety standpoint, may argue that scientists should concentrate on disseminating the certainty of known, well-established facts and interpretation about a hazardous event, rather than on communicating uncertainties and cutting-edge research (e.g. Jones, 2020). It is our view that scientists must reach a careful balance between knowledge building and being sensitive to a damaging geohazard event very soon after it has happened.

4.4 Argument 4: helping people to understand hazards and risk mitigation

Improving people's understanding of natural phenomena can help to improve risk mitigation, at least indirectly. For example, in the case of the Palu earthquake, the international media insisted that a ”failed” tsunami warning was responsible for the associated fatalities, but scientists quickly realized and explained that there was not enough time to issue an efficient alert because of the proximity of the earthquake (see above). In fact, the Indonesian agency in charge (BMKG) issued an alert a few minutes after the event and then cancelled it ∼30 min later (Fig. 3d, Table A1); in the meantime, the tsunami hit the Palu Bay coasts (Krippner, 2018). Later the same day, BMKG issued a press release to explain their alert management process. This contradictory information is likely to open a debate that will improve the general public's understanding of what to expect (or not) from early-warning systems. More generally, by bringing facts and evidence-based arguments into the public debate, the scientific community can contribute to the quality of people's information and, in the long-term, help people to prepare. Twitter discussions are opportunities to prevent confusion and misunderstanding by reinforcing and disseminating information and advice given by local government agencies (Bartel and Bohon, 2019).