Based on past events, we identify four key solutions to the challenges presented, alongside valuable lessons learnt: (i) assessing the threat; (ii) assessing and communicating uncertainty; (iii) establishing an early warning system; and (iv) developing and integrating decision-making tools.
3.1 Assessing the Threat
Attempts to assess the threat posed by volcanoes, relative to each other, began in the 1980s with three schemes created by: Bailey et al. (1983), Lowenstein and Talai (1984) and Yokoyama et al. (1984). The purpose of these was to identify the volcanoes most likely to generate destructive eruptions specific to the USA (Bailey et al. 1983) and Papua New Guinea (Lowenstein and Talai 1984) and, globally (UNESCO report, Yokoyama et al. (1984)). This would allow for preferential deployment of monitoring equipment for maximum threat mitigation. These three schemes were used as the basis for the U.S. National Volcano Early Warning System (NVEWS) (Ewert et al. 2005, 2007). NVEWS ranked 169 volcanoes in the USA in a combined assessment of 15 hazard and 9 exposure factors to generate a threat score. This scheme notably included a score for the potential exposure of aviation to an eruption of a specific volcano. Threat scores allow volcanoes to be ranked against each other and thus enable recommendations for varying levels of monitoring (Ewert et al. 2005; Moran et al. 2008). Monitoring efforts are therefore focused on the volcanoes most likely to generate significant risk.
It is important to note that all authors of these types of ranking systems recognize that such comparisons are dependent on the existing quality and quantity of information. If little is known of a volcano, then it is difficult to accurately calculate its threat, except through global comparisons to analogue volcanoes and rapid investigations and monitoring installations during a crisis (i.e., playing “catch up”). For example, the global assessment prepared for UNESCO by Yokoyama et al. (1984) failed to recognize the potential for Pinatubo to produce a large, explosive eruption. This was not an oversight, but rather a reflection of what was known at the time. Less than a decade later, this volcano produced one of the largest recorded eruptions of the century, highlighting the need for vigilance and thorough assessment of any volcanoes near population centres, even if they have appeared dormant for hundreds of years. Although, there was a remarkably successful response that saved an estimated 20,000 lives (Newhall and Punongbayan 1996), it is now widely recognized that playing “catch up” is not the best solution, as it puts the response team in danger and it may not always result in a positive outcome. Consequently, the importance of developing a Volcano Early Warning System (VEWS) well in advance of a crisis, for all high-risk volcanoes, is now widely accepted as best practice. This was one of the major motivations for NVEWS that was first implemented in 2005, and is now used in a number of nations.
3.2 Assessing and Communicating Uncertainty
A principal challenge is that the degree of certainty in forecasting varies widely with time before (Newhall 2000). It is possible to forecast eruptions with relative certainty at an intermittently active volcano over time scales of centuries, highly uncertain at intermediate time spans of months to a few years, and with greatly improved certainty at short time spans of days to hours. Yet, in spite of these restrictions, forecasts of eruptions have become relatively common. Volcano observatories worldwide issue alert levels, many of which include qualitative statements about the probability of an eruption (e.g., it is “likely”) within certain periods of time (e.g., “within days to weeks”). Repeated lava- dome eruptions were predicted successfully at Mount St. Helens (Swanson et al. 1983) and forecast at Montserrat (Voight et al. 1988) using changes in deformation and seismic rates. Similarly, based mainly on an escalation in seismicity and observations of physical changes at Pinatubo, the USGS-Philippine Institute of Volcanology and Seismology (PHIVOLCS) team estimated a 40% probability on 17 May 1991 for an eruption, 3 weeks before the eruption started. As levels and rates of unrest increased through early June, alert levels were used to warn that an eruption was likely to begin within 2 weeks and then within 24 hours (Punongbayan et al. 1996; Newhall and Pallister 2015). Based on seismic pattern-recognition during precursory activity and associated conceptual models of magma dynamics, successful forecasts have been made at many other volcanoes during the past several decades (McNutt 1996; Chouet 1996; White and McCausland 2016).
A problem with such forecasts, however, is that they typically use descriptive terms such as “likely” to convey the hazard (Doyle et al. 2014). This is a major shortcoming, because without a working understanding and effective communications of probability and uncertainty, emergency managers and the public may not be convinced of the potential hazard and urgency to take timely mitigation measures. In order to make forecasts more quantitative, probabilistic and statistical methods are now increasingly used. Probabilistic eruption forecasting typically utilizes Bayesian statistics, in which the probabilities of subsequent events depend on the outcomes of prior events; i.e., they are path dependent and increase in magnitude as the path is realized and the volcano progresses toward an event. These methods typically assign initial (a priori) probabilities on the basis of historical statistics and then update them into a posteriori probabilities that are based on interpretation of monitoring data and on the physical and chemical processes that are thought to be controlling the system.
The Bayesian methods may be used in both long-term and short-term forecasting. The most common applications of statistics and uncertainty analyses in short-term eruption forecasting is the Bayesian Event Tree (Newhall and Hoblitt 2002), which considers probabilities and uncertainties of occurrence at each node in a tree-like time series leading to a potential eruption. Monitoring information is often combined with pre-determined patterns or thresholds and with conceptual models pertaining to the dynamics of magmatic systems to forecast outcomes of volcanic unrest. Current practitioners of Bayesian Event Tree (BET) analysis use either the Cooke-Aspinall method (Cooke 1991; Aspinall 2006) or the INGV (National Institute of Geophysics and Volcanology) method (Marzocchi et al. 2004, 2008), although there are other implementations (e.g., Sobradelo et al. 2014; Jolly et al. 2014; Newhall and Pallister 2015). In addition, Bayesian Belief Networks (BBN), another graphic method that does not require the same type of linear time progression as in BET systems, may be used effectively in some situations (e.g., Lindsay et al. 2010; Hincks et al. 2014; Aspinall and Woo 2014). All of these methods integrate some form of elicitation of opinions from a team of experts to assign probabilities and uncertainties based on monitoring data, past eruptive behaviour and conceptual models. They vary with respect to whether monitoring thresholds are defined in advance for the volcano in question and in how uncertainties are established. In comparison, the USGS/Volcano Disaster Assistance Programme (VDAP) team (Newhall and Pallister 2015) uses group discussion and consensus to assign nodal probabilities. In the INGV method, probability distributions are established for each node in the event tree (Marzocchi et al. 2008). In this procedure, the parameters, weights, and thresholds are established through expert opinions, updated using data of past eruptions, and uncertainty is expressed as a probability density function for each node in the tree (Marzocchi and Bebbington 2012).
A daunting challenge for scientists who use any of these methods is to effectively communicate the results to emergency managers and the public; groups who are rarely well versed in statistics. A well-designed VEWS should utilize everyday terminology that is well-known to the population at risk, and be explicitly linked to any assigned numerical probabilities. For example, the USGS/VDAP team generally translates probabilities in terms of odds and rounds to the nearest 10%; e.g., “1 out of 3” or “9 out of 10” and terms such as “unlikely” are defined as <10%, “moderately likely” as 10–70% and “highly likely” as >70%.
3.3 Establishing an Early Warning System
Early-warning systems (EWS) are employed globally for a range of rapid onset hazards. The United Nations International Strategy for Disaster Reduction (UNISDR) recognises EWS as a core component of disaster risk reduction (DRR) measures both in the Hyogo Framework (2005) and the Sendai Framework for Disaster Risk Reduction (2015), stipulating the need to ‘substantially increase the availability of and access to multi-hazard early warning systems and disaster risk information and assessments to the people by 2030’ (UN ISDR 2015 p. 12). EWS can be defined as ‘the set of capacities needed to generate and disseminate timely and meaningful warning information to enable individuals, communities and organizations threatened by a hazard to prepare and to act appropriately and in sufficient time to reduce the possibility of harm or loss’ (UNISDR 2009, p. 12). This approach is comprised of four key sections: risk knowledge, monitoring and warning service, dissemination and communication, and response capacity (UNISDR PPEW 2006). This definition moves away from a traditional approach to EWS, as merely technical warnings through a siren or other simple warning method.
According to Leonard et al. (2008), VEWS are composed of five key components (Fig. 2): the early warning system itself, planning, co-operation, education and participation, and exercises. It is widely accepted that VEWS are part of a broader framework of DRR measures including: scientific knowledge and limitation, education, technology capabilities, and policy. EWS are arguably the process by which many DRR measures are implemented, often within a broader mitigation strategy.
The process of developing a VEWS requires cooperation and communication not only across different cultures, but also different languages and political regimes. Garcia and Fearnley (2012) highlight that, whilst an EWS may have four key components as outlined by the UNISDR, it is often the links between these categories that are the focus of systemic failure. With multi-national volcanic events or hazards, these links are likely to be highly stressed. Whilst there are excellent studies on EWS (e.g., Mileti and Sorenson 1990; Kuppers and Zschau 2002; Basher 2006; Golnaraghi 2012), few look beyond the individual case study to focus on more international scale implications of a hazard event (Fig. 2).
It is possible to establish some of the complexities that VEWS have to deal with by applying the concept of classification of mitigation strategies to VEWS (Day and Fearnley 2015). This depends on how the VEWS has been designed. Responsive mitigation strategies prescribe actions after a hazard-source event has occurred, such as evacuations to avoid lahars, which require capacities to detect and quantify the hazard and to transmit warnings fast enough to enable at risk populations to decide and act effectively. Permanent mitigation strategies prescribe actions such as construction of SABO dams or land use restrictions: they are frequently both costly and ‘‘brittle’’ in that the actions work up to a design limit of hazard intensity or magnitude and then fail. Permanent warning systems exist on volcanoes, whereby a warning is triggered, for example, by an automated lahar warning system. Anticipatory mitigation strategies, used in the mitigation of volcanic hazards more than for any other type of hazard, prescribe use of the interpretation of precursors to hazard source events as a basis for precautionary actions. However, challenges arise from uncertainties in hazard behaviour and in the interpretation of precursory signals. For example, evacuating vulnerable populations who live in areas susceptible to pyroclastic density currents prior to the onset of an eruption, pose hard questions about whether an early warning is based on forecasts, or on current activity and observations only, as well as our dependency on technology and statistical methods to make potentially life and death decisions.
Many countries operate so that their early warnings are based only and exclusively on scientific data and probabilistic forecasts. Other countries explicitly consider the social risks involved, alongside the scientific data and forecasts. There is potential for skewing of alert level assignment, intentionally or unintentionally, when there is prior-knowledge of the risks involved, and when scientists rely upon non-probabilistic decision making (Fearnley 2013). Papale (2017) presents an argument that warnings may be flawed by implicit vested interests, and he recommends that observatories should rely on pre-established thresholds and communication of scientifically based probabilistic forecasts for hazard communication. Dependant on the context, differing approaches may be taken in either adopting a top-down (government led initiative) or bottom-up approach (driven by community based approaches).
What remains a challenge is to define whether a VEWS has been successful or not; this also depends on how success is measured. Paton et al. (1998) state that effectiveness of an integrated response can be constrained by communication and coordination across stakeholders, training experience, and organisational capabilities. It is imperative that all warning communication has one consistent message, with no contradiction to cause confusion. This is essential to establish trust between the public and other users that the information is correct (Mileti and Sorenson 1990). Further challenges can arise from the accumulation of multiple disasters, e.g., the impact of Typhoon Yunya in the Philippines during the 1991 Pinatubo eruption significantly exacerbated lahars, ashfall distribution and loading (Newhall and Punongbayan 1996). It is also challenging to determine the cost benefit of a VEWS prior to the impact of the event and as a result, many disregard the value of the system, particularly for events with a long-return frequency.
Science is a necessary evidence base for making decisions and has become a key component in EWS or Incident Command Systems (ICS). In some cases, EWS have become ‘hazard-focused, linear, top-down, expert-driven systems, with little or no engagement of end-users or their representatives’ (Basher 2006, p. 2712). However, there are many examples where major efforts are being made to engage with end users via community outreach and educational activities such as PHIVOLCS (Philippine Institute of Volcanology and Seismology), the USGS, and CVGHM (Center for Volcanology and Geological Hazard Mitigation). Typically, government institutions that manage potential disasters use simple prescriptive policy. Within this they recognise that decision-making is more complex and that local practitioners and vulnerable populations are increasingly managing disasters relevant to them using community-based warning and emergency response systems (UN ISDR PPEW 2006). Such community-based warning and response systems are based upon local capabilities and technologies where communities can have ownership, generating a bottom-up approach. Although initially considered a radical approach when introduced by Hewitt (1983), community-based early warning and response systems have gained momentum and have been proven effective and empowering during crises (Andreastuti et al. 2017). Subsequently it is suggested by the UN ISDR PPEW (2006) that these community-based approaches develop people-centric early warning and emergency response systems.
3.4 Decision-Making Tools
The way that people perceive information that has been communicated to them is vitally important, as it will shape how they frame problems and make decisions. There is significant progress in the role of various tools to assist in applying new knowledge making use of communicative products such as: map making, messages in preparedness products, infograms, and the simple verbal conveyance of crisis communication. Equally there are numerous new challenges and benefits to effective communication, For example there may be too little monitoring data, which increases the uncertainties in forecasts. In a few select cases where there are many different types of monitoring methods available, it may be difficult for scientists to synthesise all the information into a forecast in a timely manner. This suggests that there are optimal levels of monitoring and/or procedures for timely data processing and interpretation if the aim is to forecast future activity. Equally, the expansion of social media has opened lines of communication both to and from volcano observatories in new transparent and engaging ways, as seen via Twitter feeds, new citizen science apps, and community based monitoring (e.g., Stone et al. 2014), and in the sharing of knowledge. However, it also has placed pressure on the credibility of information, raising the risk of false data and interpretations that require careful management, and new levels of trust and engagement that must be built between the volcano observatories and the publics.
Maps are increasingly being used as a tool in conveying uncertainty, risk, and warnings. Volcano hazard maps are widely used to graphically portray the nature and extent of hazards and vulnerabilities and, in a few cases, the societal risk. Such maps may also be used to designate prohibited, restricted entry, or warning zones. They vary widely in style and content from nation to nation, and from volcano to volcano. In the most basic form, a volcano hazard map consists of hazard zones based on the underlying geology and history of past eruptions to define the extent of past flows and tephra falls. More sophisticated hazard maps utilize detailed geologic mapping and modelling of potential flow paths, often using Digital Elevation Models (DEMs) and statistical or numerical models that simulate flows of varying volume and duration. Some new approaches use automatic GIS-based systems that incorporate numerical model results and display the results in a GIS format (Felpeto et al. 2007) or that display the results of spatial probability for potential vent locations and flow inundation (Bevilacqua et al. 2015; Neri et al. 2015). These automated methods provide the capability to quickly modify the hazard map during a rapidly developing crisis. In addition, a new generation of numerical models have enabled near-real-time probabilistic forecast maps of ash cloud and ash fall hazards (Schwaiger et al. 2012 and references therein). Regardless of their degree of sophistication, hazard maps are a fundamental means to convey the spatial distribution of danger zones to emergency managers and the public. Although not everyone can effectively read a topographic map, shaded relief and 3D oblique projections using DEMs provide more effective means to communicate map information (Newhall 2000; Haynes et al. 2007).
To date there has been little evaluation of the influence of institutional organisation and the flow of information between different actors in a crisis when deciding what to do with the ‘threat’. Fearnley (2013) investigated the role of decision-making in the USGS when assigning a volcano alert level, which established that informal communication is essential to enable key user groups to determine the extent of risk and likelihood of events. This was commonly achieved via face-to-face meetings, workshops and exercises, and telephone conversations, alongside web resources. Interactions are conducted in a multi-directional manner as various stakeholders may discuss relevant issues, moving away from typical one or two-way communication models. Evidence suggested that the ability to develop dialogue enabled key decision-makers to gauge the volcano’s behaviour and forecast in terms relevant to their own geographical, and temporal relations to the hazard. Today, observatories have developed a number of institutional communication tools, whether they are simply telephone calls or meetings that enable dialogue, or a one-way tool of information from the observatory via standardised messages targeted to specific users, such as the Volcano Activity Notice (VAN) or Volcano Observatory Notice for Aviation (VONA). Information can be communicated via daily, weekly, or monthly formal updates, status or information reports, or via Tweets, social networking, and the Smithsonian Weekly Updates. With so many options available it is up to the observatory and their stakeholders to establish what tools best serve their purpose.
In addition, during times of crisis, most observatories also participate in National Incident Command systems, or other similar civil protection procedures. For example, the USGS volcano observatories contribute scientific information to the National Incident Management System (NIMS), which was developed over many decades in response to inter-agency responses to wildfires, and is now used for all types of crises and disasters. The fundamental element of NIMS is the Incident Command System (ICS) system, which is used to structure and organize responses by federal, state and local agencies with responsibility for responding to natural as well as man-made crises and disasters. Figure 3 shows how the USGS contributes to the ICS system during volcanic crises. For example, in a disaster response, USGS scientists serve as technical advisors in the Planning Section to provide information about hazards (e.g., forecasts regarding eruptive activity, information about areas likely to be affected, extent and duration of impacts, etc.). They may also have a role in the Operations Section (e.g., in helping coordinate aviation operations). During an ICS response, a Joint Information Center (JIC) and a Joint Operations Center (JOC) are established. Through the JIC, press briefings and other media events are planned and conducted (Dreidger et al. 2004). The JIC and JOC are places where representatives of all involved agencies meet to coordinate information and crisis operations.