Willy P Aspinall

A route to more tractable expert advice

There are mathematically advanced ways to weigh and pool scientific advice. They should be used more to quantify uncertainty and improve decision-making, says Willy Aspinall.

Evidence-based volcanology: application to eruption crises

Abstract The way in which strands of uncertain volcanological evidence can be used for decision-making, and the weight that should be given them, is a problem requiring formulation in terms of the logical principles of Evidence Science. The basic ideas are outlined using the explosion at Galeras volcano in Colombia in January 1993 as an example. Our retrospective analysis suggests that if a robust precautionary appraisal had been made of the circumstances in which distinctive tornillo signals were detected at Galeras, those events might have been construed as stronger precursory evidence for imminent explosive activity than were the indications for quiescence, given by the absence of other warning traits. However, whilst visits to the crater might have been recognised as involving elevated risk if this form of analysis had been applied to the situation in January 1993, a traditional scientific consideration of the available information was likely to have provided a neutral assessment of short-term risk levels. We use these inferences not to criticise interpretations or decisions made at the time, but to illustrate how a structured, evidence-based analysis procedure might have provided a different perspective to that derived from the conventional scientific standpoint. We advocate a formalism that may aid such decision-making in future: graphical Bayesian Belief Networks are introduced as a tool for performing the necessary numerical procedures. With this approach, Evidence Science concepts can be incorporated rationally, efficiently and reliably into decision support during volcanic crises.

The 1995–1998 eruption of the Soufriére Hills volcano, Montserrat, WI

Eruption of the Soufriere Hills volcano began on 18 July 1995 after three years of elevated seismic activity. Four months of increasingly vigorous phreatic activity culminated in mid-November 1995 with the initiation of dome growth. Growth rates increased unevenly through early March 1996, with fluctuations on time-scales from hours to months. Since March 1996, gravitational collapse of the unstable dome flank has affected an ever-increasing area with pyroclastic flows, surges and ashfalls. Major collapse of the eastern flank on 17 September 1996 resulted in a sub-Plinian explosive eruption later that day. By February 1997, the dome had outgrown the confines of the crater and begun to spill into the surrounding valleys. A large collapse on 25 June 1997 caused pyroclastic flows and surges on the northern flanks and resulted in the only deaths of the eruption. In August, September and October 1997, vulcanian explosions followed further collapses on the western and northern flanks. The largest event of the eruption occurred on 26 December 1997 with failure of the southwestern flank of the volcano producing a debris avalanche and large dome-collapse pyroclastic flows. Dome growth ceased in early March 1998, but residual volcanic activity has continued and consists of ash venting, mild explosions and dome-collapse pyroclastic flows.

Global link between deformation and volcanic eruption quantified by satellite imagery

A key challenge for volcanological science and hazard management is that few of the world’s volcanoes are effectively monitored. Satellite imagery covers volcanoes globally throughout their eruptive cycles, independent of ground-based monitoring, providing a multidecadal archive suitable for probabilistic analysis linking deformation with eruption. Here we show that, of the 198 volcanoes systematically observed for the past 18 years, 54 deformed, of which 25 also erupted. For assessing eruption potential, this high proportion of deforming volcanoes that also erupted (46%), together with the proportion of non-deforming volcanoes that did not erupt (94%), jointly represent indicators with ‘strong’ evidential worth. Using a larger catalogue of 540 volcanoes observed for 3 years, we demonstrate how this eruption–deformation relationship is influenced by tectonic, petrological and volcanic factors. Satellite technology is rapidly evolving and routine monitoring of the deformation status of all volcanoes from space is anticipated, meaning probabilistic approaches will increasingly inform hazard decisions and strategic development.

World Health Organization Estimates of the Relative Contributions of Food to the Burden of Disease Due to Selected Foodborne Hazards: A Structured Expert Elicitation

Background The Foodborne Disease Burden Epidemiology Reference Group (FERG) was established in 2007 by the World Health Organization (WHO) to estimate the global burden of foodborne diseases (FBDs). This estimation is complicated because most of the hazards causing FBD are not transmitted solely by food; most have several potential exposure routes consisting of transmission from animals, by humans, and via environmental routes including water. This paper describes an expert elicitation study conducted by the FERG Source Attribution Task Force to estimate the relative contribution of food to the global burden of diseases commonly transmitted through the consumption of food. Methods and Findings We applied structured expert judgment using Cooke’s Classical Model to obtain estimates for 14 subregions for the relative contributions of different transmission pathways for eleven diarrheal diseases, seven other infectious diseases and one chemical (lead). Experts were identified through international networks followed by social network sampling. Final selection of experts was based on their experience including international working experience. Enrolled experts were scored on their ability to judge uncertainty accurately and informatively using a series of subject-matter specific ‘seed’ questions whose answers are unknown to the experts at the time they are interviewed. Trained facilitators elicited the 5th, and 50th and 95th percentile responses to seed questions through telephone interviews. Cooke’s Classical Model uses responses to the seed questions to weigh and aggregate expert responses. After this interview, the experts were asked to provide 5th, 50th, and 95th percentile estimates for the ‘target’ questions regarding disease transmission routes. A total of 72 experts were enrolled in the study. Ten panels were global, meaning that the experts should provide estimates for all 14 subregions, whereas the nine panels were subregional, with experts providing estimates for one or more subregions, depending on their experience in the region. The size of the 19 hazard-specific panels ranged from 6 to 15 persons with several experts serving on more than one panel. Pathogens with animal reservoirs (e.g. non-typhoidal Salmonella spp. and Toxoplasma gondii) were in general assessed by the experts to have a higher proportion of illnesses attributable to food than pathogens with mainly a human reservoir, where human-to-human transmission (e.g. Shigella spp. and Norovirus) or waterborne transmission (e.g. Salmonella Typhi and Vibrio cholerae) were judged to dominate. For many pathogens, the foodborne route was assessed relatively more important in developed subregions than in developing subregions. The main exposure routes for lead varied across subregions, with the foodborne route being assessed most important only in two subregions of the European region. Conclusions For the first time, we present worldwide estimates of the proportion of specific diseases attributable to food and other major transmission routes. These findings are essential for global burden of FBD estimates. While gaps exist, we believe the estimates presented here are the best current source of guidance to support decision makers when allocating resources for control and intervention, and for future research initiatives.

Quantifying uncertainties in the measurement of tephra fall thickness

The uncertainties associated with tephra thickness measurements are calculated and implications for volume estimates are presented. Statistical methods are used to analyse the large dataset of Walker and Croasdale J Geol Soc 127:17-55, 1971 of the Fogo A plinian deposit, São Miguel, Azores. Dirichlet tessellation demonstrates that Walker and Croasdale’s measurements are highly clustered spatially and the area represented by a single measurement ranges between 0.5 and 10 km2. K-means cluster analysis shows that lower thickness uncertainties are associated with closely spaced measurements. Re-examination and analysis of Fogo A fall deposits show thickness uncertainties are about 9% for measured thickness while uncertainty associated with natural variance ranges, between 10 and 40%, with an average error of 30%. Correlations between measurement uncertainties and natural variance are complex and depend on a unit’s thickness, position within a succession and distance from source. Normative error increases as tephra thickness decreases. The degree to which thickness measurement error impacts on volume uncertainty depends on the number of measurements within a given dataset and their associated uncertainty. The uncertainty in volume associated with thickness uncertainty calculated herein for Fogo A is 1.3%, equivalent to a volume of 0.02 km3. However uncertainties associated with smaller datasets can be much larger; for example typically exceeding 10% for less than 20 data points.

Quantifying volcanic hazard at Campi Flegrei caldera (Italy) with uncertainty assessment: 1. Vent opening maps

Campi Flegrei is an active volcanic area situated in the Campanian Plain (Italy) and dominated by a resurgent caldera. The great majority of past eruptions have been explosive, variable in magnitude, intensity, and in their vent locations. In this hazard assessment study we present a probabilistic analysis using a variety of volcanological data sets to map the background spatial probability of vent opening conditional on the occurrence of an event in the foreseeable future. The analysis focuses on the reconstruction of the location of past eruptive vents in the last 15 ka, including the distribution of faults and surface fractures as being representative of areas of crustal weakness. One of our key objectives was to incorporate some of the main sources of epistemic uncertainty about the volcanic system through a structured expert elicitation, thereby quantifying uncertainties for certain important model parameters and allowing outcomes from different expert weighting models to be evaluated. Results indicate that past vent locations are the most informative factors governing the probabilities of vent opening, followed by the locations of faults and then fractures. Our vent opening probability maps highlight the presence of a sizeable region in the central eastern part of the caldera where the likelihood of new vent opening per kilometer squared is about 6 times higher than the baseline value for the whole caldera. While these probability values have substantial uncertainties associated with them, our findings provide a rational basis for hazard mapping of the next eruption at Campi Flegrei caldera.

Quantifying volcanic hazard at Campi Flegrei caldera (Italy) with uncertainty assessment: 2. Pyroclastic density current invasion maps

Campi Flegrei (CF) is an example of an active caldera containing densely populated settlements at very high risk of pyroclastic density currents (PDCs). We present here an innovative method for assessing background spatial PDC hazard in a caldera setting with probabilistic invasion maps conditional on the occurrence of an explosive event. The method encompasses the probabilistic assessment of potential vent opening positions, derived in the companion paper, combined with inferences about the spatial density distribution of PDC invasion areas from a simplified flow model, informed by reconstruction of deposits from eruptions in the last 15 ka. The flow model describes the PDC kinematics and accounts for main effects of topography on flow propagation. Structured expert elicitation is used to incorporate certain sources of epistemic uncertainty, and a Monte Carlo approach is adopted to produce a set of probabilistic hazard maps for the whole CF area. Our findings show that, in case of eruption, almost the entire caldera is exposed to invasion with a mean probability of at least 5%, with peaks greater than 50% in some central areas. Some areas outside the caldera are also exposed to this danger, with mean probabilities of invasion of the order of 5–10%. Our analysis suggests that these probability estimates have location-specific uncertainties which can be substantial. The results prove to be robust with respect to alternative elicitation models and allow the influence on hazard mapping of different sources of uncertainty, and of theoretical and numerical assumptions, to be quantified.

Expert Elicitation for the Judgment of Prion Disease Risk Uncertainties

There is a high level of uncertainty surrounding the potential for iatrogenic prion transmission through transplantation, medical instrument reuse, blood transfusion, and blood product use due to a lack of evidence-based research on this important risk issue. A group of specialists was enlisted to evaluate some of the knowledge gaps in this area using the “Classical Model,” a structured elicitation procedure for weighting and pooling expert judgment. The elicitation exercise was undertaken in March 2009 with 11 transmissible spongiform encephalopathy (TSE) experts who were first calibrated using a series of seed questions for which the answers are known; they were then asked to answer a number of target questions that are important for risk assessment purposes, but for which there remains high uncertainty at this time. The target questions focused on variant Creutzfeldt–Jakob disease (vCJD) prevalence, incubation times for vCJD, genetic susceptibility to prion disease, blood infectivity, prion reduction of blood and blood products, surgical instrument risks, and interspecies transmission of TSEs. The experts were also asked to perform pairwise risk rankings for 12 different potential routes of infection. Dura mater transplantation was seen as having the highest risk, while dental tissue grafts were viewed as presenting the lowest risk of iatrogenic transmission. The structured elicitation procedure provides a rational, auditable, and repeatable basis for obtaining useful information on prion disease risk issues, for which data are sparse.

Santorini unrest 2011–2012: an immediate Bayesian belief network analysis of eruption scenario probabilities for urgent decision support under uncertainty

Unrest at the Greek volcanic island of Santorini in 2011–2012 was a cause for unease for some governments, concerned about risks to their nationals on this popular holiday island if an eruption took place. In support of urgent response planning undertaken by the UK government, we developed a rapid evaluation of different eruption scenario probabilities, using the Bayesian Belief Network (BBN) formulation for combining multiple strands of scientific and observational evidence. Here we present three alternative BBN models that were devised in early 2012 for assessing the situation: (1) a basic static net for evaluating probabilities at any one moment in time, utilising just four key unrest indicators; (2) a compound time-stepping net, extending the basic net to update probabilities through time as the indicators changed; and (3) a more comprehensive net, with multiple lines of other data and observations incorporated, reflecting diversity of modern multi-parameter monitoring techniques. A key conclusion is that, even with just three or four basic indicators, it is not feasible, or defensible, to attempt to judge mentally the implications of signs of unrest – a structured probabilistic procedure using Bayes’ Rule is a rational approach for enumerating evidential strengths reliably. In the Santorini case, the unrest, and official anxiety, diminished quite quickly and our approach was not progressed to the point where detailed consideration was given to BBN parameters, analysis of data uncertainty or the elicitation of expert judgements for quantifying uncertainties to be used in the BBN. Had this been done, the resulting scenario probabilities could have been adopted to determine likelihoods of volcanic hazards and risks caused by possible eruptive activity, as identified in a concurrent assessment of the scale and intensities of potential volcanic impacts (Jenkins et. al., Assessment of ash and gas hazard for future eruptions at Santorini Volcano, Greece. forthcoming). Ideally, such hazard and risk assessments should be elaborated in detail and critiqued well before crisis-level unrest develops – not initiated and implemented within a few hours just when a situation looks ominous. In particular, careful analysis of all information is required to determine and represent parameter uncertainties comprehensively and dependably.