Final Report Summary - MATRIX (New Multi-HAzard and MulTi-RIsK Assessment MethodS for Europe)
Executive Summary:
Disaster risk reduction (DDR) activities generally treat different natural hazards and their associated risks separately. However, this approach ignores the spatial and temporal interactions that may arise along the disaster risk chain. For instance, an extreme event may trigger others, e.g. earthquakes causing tsunamis, or several different types may occur concurrently, e.g. severe weather and earthquakes. Considering vulnerability, an initial event would leave a community more susceptible to future, possibly different, hazards, e.g. an earthquake weakening buildings which are damaged further by windstorms. The temporal dimension may include changes in exposure, e.g. increased urbanisation, altering the total risk to an area, while repeated events lessen a community’s resilience. Meanwhile, although losses are estimated by usually only considering direct economic losses or casualties, this ignores less tangible losses such as reduced business activity or the loss of cultural heritage. In short, the total risk estimated when incorporating interactions between multiple hazards and risks is likely to be greater than the sum of the individual parts.
The Multi-HAzard and MulTi-RIsK Assessment MethodS for Europe or MATRIX project (01.10.2010 to 31.12.2013) set out to tackle some of the issues associated with multi-hazard and risk assessment. The focus was on the hazards that most affect Europe, namely earthquakes, landslides, volcanos, tsunamis, wild fires, storms and fluvial and coastal flooding. Interactions at the different levels were considered, such as cascading events and time dependency in vulnerability. Three test cases were considered, Naples, Italy, the French West Indies, and Cologne, Germany. Considerable interaction with end-users was undertaken, including identifying biases at the individual and institutional level which may hinder employing a multi-type framework.
The MATRIX project is recognised as being a step towards seeing multi-hazard and risk frameworks considered the standard in Europe, as evident by its representation at expert meetings and its inclusion in a recent EC document regarding the post-Hyogo framework. While it is acknowledged that multi-hazard and risk assessment are at an early stage, great interest has been shown by practitioners who note both their desire, and the potential barriers, to exploiting such a paradigm.
Some of the foremost outcomes of MATRIX may be summarised as follows with others outlined in detail in the rest of this report:
A framework was developed based on a multi-stage process (single qualitative, multi-qualitative, multi-semi-quantitative, multi-quantitative) that allows a practitioner to proceed with a hazard and risk assessment only as far as is necessary within a multi-risk environment.
A generic modelling tool that allows a practitioner to consider (or ignore) interactions between different hazards within a probabilistic framework (based on the Monte Carlo method) was developed. This may be used either for a given situation, or via a more general tool that employs a virtual city to allow practitioners to experiment with the relative importance of various interactions.
An assessment of the problems to the implementation of a multi-risk framework being incorporated into risk governance found the main barriers to include a lack of resources (time, personal, expertise, especially at the local level), the frequent disconnect between products developed by researchers and practitioners, and poor communication and interaction between the different “departments” who are effectively responsible for risk assessment for specific hazards.
Project Context and Objectives:
Many types of natural extreme events, including earthquakes, landslides, volcanic eruptions, tsunamis, river and coastal flooding, winter storms, and wildfires, threaten different regions of Europe. Disaster Risk Reduction (DRR) practitioners, decision and policy makers, and the experts who inform their judgment, usually treat such hazards and the associated risks separately. This means they follow a single-hazard risk framework, neglecting the frequent temporal and spatial interdependencies that often arise between them (see, for example, Carpignano et al., 2009; Kappes et al., 2012, Marzocchi et al., 2012).
Such dependencies arise at all levels of risk estimation (1). Considering the hazard level, situations may (and often do) arise where one hazard triggers another via so-called cascade or domino events, for example, tsunamis triggered by earthquakes, with the 2004 Indian Ocean and 2011 Töhoku earthquakes and tsunamis being the most prominent recent examples. Similarly, one type of hazard may increase the likelihood of another, e.g. droughts and wildfires, as in the case of the 2009 south eastern Australia bushfires, while there is the possibility of conjoint events, that is events occurring at the same time or, at least before the necessary repairs have been made as a result of the first, e.g. a windstorm soon following an earthquake.
At the vulnerability level, an initial event may increase the susceptibility of buildings to further damage from another, possibly different, incident, for instance, considering earthquakes and windstorms. Considering this example, the order of the events may also play a role. That is, one would expect a windstorm following an earthquake to be a more serious situation than the reverse, especially if the affected population are residing in temporary shelters such as tents, where they will be more exposed to the dangers posed by the meteorological events. Considering exposure, damage induced by one event may then lead to other parts being exposed to other hazards, such as an earthquake damaging flood protection dykes then leads to areas previously protected by these defences to be more exposed to flooding. There is also the situation that with ever increasing levels of urbanisation, society is more dependent upon sophisticated, and vulnerable, infrastructure and lifelines, while rural depopulation is leading to the breakdown of traditional landform management practices in some areas, further increasing the risk of certain disasters, such as landslides and wildfires (Shakesby, 2011).
When calculating risk, there is the current situation where usually only direct economic and human losses are found, generally ignoring indirect (e.g. reduced industrial production, increased transport costs) or intangible (e.g. loss of reputation, diminished potential for social development, loss of cultural or environmental heritage) losses. For example, as shown by examples considered within this project, even a single hazard (in this case drought in Iran, Keshavarz et al., 2013, Yazdanpanah et al., 2013) may lead to far reaching social consequences such as increased gender imbalance, distrust between sections of the community and increased prevalence of the “poverty trap” (see below).
All of these issues together (and individually) may lead to the underestimation of the total losses that can be expected from natural hazardous events in a given area, and the extent to which such events have an effect. Therefore, what is necessary is the development and implementation of a multi-hazard and risk framework that considers all of these points, and others, leading to a more comprehensive estimation of risk, which, in turn, will allow more effective mitigation strategies. A critical part of this also involves enhanced communication between the various actors involved in disaster mitigation, especially the need to bridge the gap between practitioners and researchers.
It was therefore within the context of such concerns that the European Commission, within its 7th Framework Programme, supported the New Multi-HAzard and MulTi-RIsK Assessment MethodS for Europe or MATRIX project (1st October 2010 to 31st December 2013) to develop methodologies and frameworks for multi-type hazard and risk assessment to confront the inadequacies currently existing in single-type assessments.
The goals of MATRIX may be summarised as follows:
Assess the state-of-the-art in single- and multi-type hazard and risk assessment methodologies and tools so that a solid foundation may be built upon.
Develop a multi-hazard and risk assessment framework that considers the various interactions and dependencies that may arise within such a multi-type environment.
Examine under what conditions multi-type risk assessments provide better (or not) results compared with single-type assessments.
Provide tools to support the decision-making processes required by civil protection and disaster management authorities on the basis of probabilistic information.
Assess existing governance frameworks and to investigate individual and institutional barriers to the implementation of multi-hazard and risk approaches.
Disseminate multi-hazard and risk concepts to potential end-users and relevant members of the broader community, e.g. civil protection and research establishments.
The hazards that were the primary concerns of MATRIX are those most likely to afflict Europe, as mentioned at the beginning of this section. However, it is also hoped that the results of MATRIX will be applicable to other environments and contexts, not only in terms of the natural environment, but also the specific social and political contexts.
The multi-disciplinary nature of MATRIX called for expertise not only in terms of the wide range of hazards considered (e.g. geology and geophysics, meteorology, forest ecology), but also how the resulting risks should be considered in terms of the metrics used, and the identification of any social or institutional issues that may hinder the following of a multi-type hazard and risk paradigm (e.g. behavioural science, structural engineering, economics) and the manner in which the developed methods and tools are produced and distributed to the relevant parties (information technology, disaster management and decision-making). This therefore called for a very heterogeneous consortium, made up of 12 partners, including ten research institutions (nine European and one Canadian), an end-user (the German National Contact Point for DRR) and a partner from private industry. 1 lists the consortium’s members, as well as their respective responsibilities in terms of the different components of work packages of the project, which will be outlined below.
To describe better the project’s objectives, it is perhaps useful to outline its work plan. This is shown in 2, which presents the project’s work package (WP) structure and their interactions. There were 8 work packages within MATRIX, including WP1 which is responsible for the management of the project. The more technical work packages follow what could be considered as a natural order for the work needed to be undertaken.
WP2 “Single-type risk assessment and comparability” serves as something of a starting point, given that its aim is the development of the means to allow the comparison of different single-type risks. Although MATRIX is strongly concerned with hazard and risk interactions, this does not mean that single-type assessments are ignored. In fact, they still play a critical role, as the first stage of any multi-hazard and risk assessment scheme or framework is to understand what are the hazards of concern, what are natures, what will their impact upon the human and natural environments be, and what information is required by end-users and stakeholders for their mitigation. Therefore, the principle goals of this work package include the harmonization of the methodologies used for the different types of natural hazard assessment, which means considering the different return periods and spatial scales, in order to compare risk estimates. It will also involve classifying and identifying the different sources of uncertainty that arise in risk estimates, in particular, the classification of these sources in terms of their aleatory and/or epistemic character (in collaboration with WP5, see below).
The issues associated with cascade (domino) effects within a multi-hazard environment were the subject of WP3 “Cascade effects in a multi-hazard approach”. It may be said that this work package dealt with the most obvious consequence of a multi-hazard environment. Due to the complexity of the interactions that arise at all levels of the risk chain, a multi-hazard and risk approach is potentially much more complex than the simple sum of the individual single-risk results. The objectives of WP3 therefore included the identification of the relevant temporal and spatial scales for multi-hazard assessment, the development of strategies to homogenize different types of hazard assessments (in collaboration with WP2), the identification of the plausible cascade events with respect to the MATRIX test cases, and the development of a probabilistic Bayesian and non-Bayesian framework to include cascade events within a multi-hazard assessment perspective.
Temporal-dependency in vulnerability was the topic of WP4 “Time-dependent vulnerability”, which considered this issue at the physical, systemic and socio-economic levels. The development of time-dependent vulnerability assessment procedures is an essential aspect of an integrated multi-risk approach, since it is safe to say that no environment is static. By this we mean that each hazardous event and/or possible mitigation action is likely to impact upon the vulnerability of the community, society, region, etc., to other future events. This work package therefore aimed at developing a framework that considers the history of a given element (which includes conjoint multiple disasters as well as successive ones) as an internal parameter of its vulnerability. The examples considered in this WP focused on how a given structure reacts to successive seismic events, and how an area’s systemic efficiency suffers as a result of not only the direct damage to infrastructure, such as power and water supplies and hospitals, but also in transport links.
WP5 “Framework for multi-type risk assessment” is the work package that set out to combine the efforts undertaken in work packages 2 to 4 in order to develop a consistent framework for multi-risk evaluation across different temporal and spatial scales. This included, as mentioned, the identification of the sources and types of uncertainties (following a similar scheme as done in WP2), demonstrate the application of a multi-risk assessment framework for different categories of tangible and intangible losses, develop fault trees and event trees for the assessment of environmental risks and the development of a framework within which end-users and stakeholders may place their multi-type hazard and risk assessments in order to know how much effort will need to be expended to obtain risk estimates appropriate for the resources at hand.
The work package that was concerned with risk governance and the implementation of multi-risk approaches was WP6 “Decision support for mitigation and adaption in a multi-hazard environment”. Its aim was to identify if cognitive and behavioural biases affected perceptions of probabilities for multiple risks and influenced choices for risk mitigation strategies. It involved an analysis of historical multi-risk disasters where not considering a multi-type hazard and risk environment lead to more severe consequences and increased vulnerability, as well as developing a Bayesian framework that identified most common behavioural biases, such as risk aversion, limited worry and loss aversion. The work package also endeavoured to identify existing social and institutional barriers to the implementation of a multi-risk approach (also dependent on features of governance systems such as centralized and decentralized decision-making) and proposed options for overcoming them. All of these efforts had the aim of identifying the benefits of multi-hazard and risk assessments compared to single-type assessment, and to provide recommendations as to how policies conducive to such an environment may be formulated.
The methodologies developed within the MATRIX project needed to be tested within real world settings to ensure their lasting legacy. This includes the development of a generic IT tool that allows practitioners to “experiment” within a controlled environment how the consideration, or neglect, of hazard and risk interactions affects the final results. This was dealt with by WP7 “’Virtual City’ and test cases”. WP7 saw the development of the MATRIX-Common IT sYstem (MATRIX-CITY), a multi-type risk assessment IT platform that performs and visualizes multi-hazard and multi-risk analyses for test cases. It also involved the so-called “Virtual City”, as its name suggests a virtual environment where different hazards and their risks may be analysed in terms of their effect on a predefined “virtual” community. The development of the system required the implementation of common data models, data exchange procedures and hazard and risk calculators, including time-dependent vulnerability. The test cases considered during the MATRIX project were Naples, Italy, Cologne, Germany, and the French West Indies. Each test case is exposed to a range of hazards and risks, with differing degrees of interaction.
The final work package, WP8 “Dissemination/end users” dealt with making the project’s results and aims known to as wide an audience as possible, and to interact with the end-user communities to determine their specific needs within the context of a multi-hazard and risk environment. The dissemination of MATRIX concepts and results was focused towards the scientific community at large, the disaster management community in European states, the commercial sector with an emphasis on insurance companies and companies in the information technology sector, and the press and media. The key objectives of this work package was therefore the development and implementation of a promotional and dissemination process, information products and a web site for the output from the project, and the involvement of stakeholders and user communities. The latter point proved to be most constructive and saw a great deal of interaction with WP6.
Project Results:
While the main scientific and technological results of the MATRIX project could be outlined in terms of their respective work packages, it needs to be remembered that in a project as complex as MATRIX, many work package often rely on the results of others. For example, WP5, dealing with the multi-risk assessment framework, calls upon results dealing with single-type assessment (WP2) and cascading events (WP3). Likewise, the test case studies (by definition) required the results from all of the work packages. Therefore, in the following, while we will generally proceed along the same path as suggested by the WP structure, the interactions and dependencies between them and the resulting outcomes will be emphasised. Naturally not all of results from a project as diverse as MATRIX can be presented in a report such as this, hence the focus (although not exclusively) will be on those presented in the MATRIX results and reference reports1.
Single-type hazard and risk assessment
Although the methodologies and issues surrounding a multi-hazard and risk framework were the foci of the MATRIX project, this by no means neglects the importance of single-type risk assessment. In fact, the framework defined by WP5 (see below) allows for cases where a single-type process can be employed where appropriate (i.e. cases where a multi-type process may either be not required or hindered by a lack of information and resources. Therefore, considerable effort went into reviewing the state of the art in single-type assessments, understanding the associated uncertainties (discussed below) and the harmonisation of risk results for different hazards.
From the review of the literature, some of the more prominent observations about the current state-of-the-art in single-type assessment methodologies (Parolai et al., 2014) are as follows:
For all hazards and risks examined, even when considering only the single-type framework, considerable further research is required. This mainly centres on the lack of understanding of the physical processes involved with each hazard. However, there appears to be an emphasis in the literature on the scarcity of data, limiting the ability to infer adequate time series. While such problems are present in all research, there are fairly significant disparities between the various components that make up risk assessment, as discussed below.
There appears to be a significant gap between the relatively high level of methodological development for hazard assessment and a lower level for analysing the vulnerability of the elements at risk, i.e. there seems to be a lack of vulnerability functions, except for the case of seismic hazards.
The focus in the literature seems to be on state-of-the-art research aspects of risk analysis, and not on risk management.
For the harmonisation of the different risks, two approaches are proposed within the literature:
Risk Curves, which requires the consideration of temporal constraints, spatial scale and resolution, and loss metrics.
Development of a common framework where harmonisation is carried out at each level, from hazard to risk.
During the time of the project, the first approach was the focus. The efforts made during MATRIX dealing with this issue considered two issues:
Combining individual but independent loss curves so as to gain an idea of the total loss an area may expect to experience.
How to compare two independent risks, that is how to determine if, for a given return period, they are the same and hence should be treated with the same level of concern or can some hierarchy be established.
For the first point, a simple formula was employed:
Ptot = 1 - ∏ (1 – Pi) (1)
where Ptot is the total annual probability of exceedance of a given risk (expressed as Euros), and Pi is the probability of exceedance of a given risk i (i.e. here represented by earthquakes, landslides and floods). An example of this is presented in 3a, where we show the individual risk curves from the work of Grünthal et al. (2006) for the case of Cologne (one of the MATRIX test cases). Considering as an illustrative example the situations where losses of 100 million Euros may arise, the combination of curves sees a significant increase in the probability of this level of loss, from 15 to 35% in 50 years for the individual hazards, to around 75% in 50 years when combined. Such results can be shown using an alternate format, the risk matrix (3b), which illustrates how different levels of risk arise when considering their likelihood and impact (e.g. Cox, 2008). Risk matrices are often used by civil protection authorities (e.g. BBK, 2011) and in fact, one of the statements to MATRIX members during the presentation of the MATRIX-CITY and Virtual-City support tools (see below) was to use such a format instead of risk curves. We see from Figure 3b how the combining of these risks leads to the final value being to the right (more likely), allowing decision makers a ready means of better understanding the total risk/likelihood of a level of loss.
With respect to the second point, that is comparing the estimated losses for a given return period of two (or more) different hazards, this was examined by employing the Wilcoxon’s test, a distribution free ranking test that asks the specific question “Are the medians of the two distributions the same?” (Barlow, 1989). A new series of loss estimates for the case of Cologne were produced for the hazards of floods, earthquakes and windstorms (Fleming et al., 2014). We then compared these ranges of values for each pair of hazards (earthquake – flood, earthquake – windstorm, flood – windstorm) and apply a null hypothesis (to 0.05) that the question’s answer is in the affirmative. The results of these tests are shown in 4. Considering first the earthquake distribution, its bimodal character (a product largely of the choice of ground motion predictive equations) immediately adds an additional element of uncertainty as to whether the risks it is compared to are equivalent. We note for the 200 year return period (4a) that earthquakes and floods are not equivalent, but can be considered comparable for 500 years. For the windstorms and floods (4c-d), for both the 200 (4c) and 500 (4d) years return periods, it is obvious (even without applying this test) that windstorms and floods are not equivalent, with floods being of greater concern in both cases. Finally, for earthquakes and windstorms (4e-f), for the 200 year return period (4e), these appear to be of equivalent importance, while for 500 years (4f), this does not appear to be the case (with earthquakes being of greater importance).
The relevance of such an exercise is to do with the decision making process, whereby if the risk associated with two types of hazard is “equivalent”, then the required mitigation schemes may need to consider both, or at least help decision makers when deciding on how to allocate resources. It also shows that one needs to accommodate uncertainties, since simply using, for example, average curves, may yield misleading conclusions about the relative importance of a given combination of hazard types (e.g. the earthquake results). However, it is also important to note that the actual results would vary as the range of employed input models and parameters are updated and refined (as would be apparent in the earthquake case). Similarly, the uncertainty associated with the risk arising from a given hazard may be very wide, leading to the potential range of risk levels to cover from being of little concern to being of great importance.
Cascading events
One of the most obvious aspects of a multi-hazard and risk perspective is the need to accommodate cascade events. This involves the identification of the various interactions that may lead to such occurrences and to the increased vulnerability and amplification of damage in an area. In particular, two possible kinds of interactions, namely: (1) interactions at the hazard level, where a given initial ‘triggering’ event leads to a modification of the probability of the occurrence of a secondary event, and (2) interaction at the vulnerability (or damage) level, in which the main interest is to assess the effects that the occurrence of the first event occurring in time may have on the response of the exposed elements against additional events, which may or may not be of the same kind. Implicitly, a combination of both kinds of interactions is also possible.
The first step in assessing cascading effects is the identification of possible scenarios, or rather, scenarios that are plausible. To obtain a complete set of scenarios for a given area, different strategies can be adopted, ranging from event-tree to fault-tree strategies. In many applications, an adaptive method combining both kinds of approaches is applied in order to ensure the exhaustive exploration of scenarios. From the multi-risk assessment point of view, the cascading effects scenarios of primary interest are those that produce an amplified total risk when compared to the effects produced by the individual events (a case of 1 + 1 > 2). With an appropriate set of cascading scenarios, their quantification can be achieved by adopting different strategies, for example, analysing databases of past events, performing physical modelling for the propagation of the intensity measures of interest, and/or by performing expert elicitations in order to obtain information for extremely complex problems, or in these cases with poor data or needing rapid analysis.
To define some possible cascade scenarios, the ‘primary’ interactions between hazards need to be identified. These can be understood as the pairs of hazards where it is theoretically possible to define an event that has the capacity to directly trigger another (interaction at the hazard level), or in which the additive effects of the loads may lead to a risk amplification. In the matrix-like 2, the different hazards considered in the MATRIX project are classified as triggering (running in the x-axis) against the ‘triggered’ (running in the y-axis) events. In this case, all the possible ‘direct’ triggering effects are considered. It would also be obvious that it is physically impossible for some hazards to trigger another, e.g. wildfires and volcanoes (although the other way around is certainly a concern, especially for Naples). Note also from 2 that interactions at the vulnerability level may also be observed. Such a table can be expanded to cases where additional ‘levels’ of interactions can be inferred, simply by adapting an appropriate numbering scheme (e.g. a flood may be labelled 2, if a river is blocked by a landslide (labelled 1) caused by an earthquake (which would be labelled 0).
5 illustrates how one can identify the possible cascading scenarios that may arise for the MATRIX test cases. It therefore allows us to understand better the existing relationships between the different kinds of events and their relative level in the chain. In this way, the occurrence of different phenomena may be considered from the possible triggering factors. Such a diagram would be useful for DRR practitioners to at least visualise to themselves potential scenarios of cascading events. In fact, based on this analysis, examples of cascading scenarios were selected for more detailed analysis as part of the test case studies, namely:
Naples: The cases that received most attention were volcanic earthquakes and seismic swarms triggered by volcanic activity, and the simultaneous occurrence of ash-fall and earthquake hazards.
Guadeloupe: The scenario here was concerned with earthquake-triggered landslides after a cyclonic event or a heavy rainfall period.
Cologne: This scenario involved earthquake-triggered embankment/flood defence dyke failures and the subsequent inundation of the City of Cologne.
A further point that needs to be made concerning cascading events is that while they are usually considered at the hazard level, as alluded to in 2, they also occur at the vulnerability level, and not only the physical. Taking again the example mentioned above of the Iranian drought, one can easily consider the consequences of inappropriate mitigation actions, e.g. the limiting of support loans to the wealthier farmers, who could then improve their infrastructure, hence now needing fewer workers and hence deriving poorer farmers of a source of income, exasperating their problems. This leads to health (mental and physical) problems, mistrust developing within the community, the frequent need to withdraw children from education (usually the girls) leading to enhanced gender inequality and the so-called poverty trap (Hochrainer, 2013).
Uncertainties within a multi-hazard and risk framework
Uncertainties in risk estimates, while widely acknowledged as being critical, are rarely considered. In fact, taking the otherwise comprehensive review of Kappes et al. (2012) as an example, there is no mention at all of uncertainties. Therefore, WP2 and WP5 set out to identify and classify the different types and sources of uncertainty relevant to hazard and risk assessment, with efforts to identify those that were particular to multi-type environments.
The focus in WP2 (Rohmer, 2012) was first to devise a qualitative analysis for each step of the state of the art in single-type risk assessment. This saw uncertainties classified between aleatory (i.e. the general randomness of a system) and epistemic (inadequate knowledge about the processes involved, lack of data, and poorly defined input models and parameters). The latter was further divided into four additional types:
epistemic – data: This refers to the difficulties associated with measuring the properties of the system of interest or concern, e.g. issues with the instrumentation, methods of processing, length of time for data acquisition, and so forth.
epistemic – models: Models are by definition simplifications of a phenomena, leading to uncertainties arising from (a) uncertainty in the form/structure of the model, that is, what variables, processes, etc. are considered relevant and prominent, and (b) the actual choice of the “best” model, that is, the one that reproduces most closely the processes of interest. However, a problem may arise when more than one model gives equally good results.
epistemic – parameters: This source arises from the difficulties in estimating the required input parameters for models or an analysis due to the limited size, poor representativeness (caused by time, space and financial limitations), and imprecision of the available data.
epistemic – science: This class reflects our state of knowledge (or more importantly, ignorance) of the processes under investigation.
The qualitative analysis of uncertainties should be viewed as a tentative mapping of the state of our knowledge for risk assessment and of the major challenges for quantitative uncertainty treatment. More specifically, it highlights, at all stages of risk assessments, not only the prevailing presence of uncertainties of type “epistemic – parameter” (related to the difficulties in estimating the values of model parameters), but also of type “epistemic – model” (related to the uncertainty in the structure/form of the model and to the unambiguous choice of the “best” model to be used). It also appears that most work dealing with uncertainties was concerned with the hazard level rather than the risk level. The same taxonomy of uncertainties was also employed in WP5 (Vidar Vangelsten, 2012), where the most suitable methodologies for the quantitative treatment of uncertainty in multi-risk assessments were reviewed.
Time-dependency
For a comprehensive multi-hazard and risk framework, the temporal dimension in such interactions would need to be taken into account. By this we mean interactions at the hazard and vulnerability levels, which might occur with different delays. Such temporal dependencies might involve the following situations:
The repetition of events over time (these may be the same, e.g. aftershocks following an earthquake, or different events).
The concomitance of simultaneous-yet-independent events. By simultaneous or conjoint, one may also consider successive events occurring before the impacted upon area has adequately recovered from the initial event.
The succession of dependent phenomena (cascading events).
The fact that the future vulnerability of the risk bearer will depend upon the realization of risk in the past.
For the first of these, the effects of the repeat of a specific type of event were considered by examining a seismic example. The effects of fatigue due to the repetition of seismic shocks within a physical vulnerability assessment have been analysed through two mechanical methodologies. The first approach, proposed by BRGM (Réveillère, 2012), developed damage-state dependent fragility functions (Figure 6), while the second approach, performed by AMRA (Iervolino et al., 2014), analysed the multiple shock capacity reduction for non-evolutionary structural system (Figure 7).
The next two situations listed above were considered within the test case of Guadeloupe and will be mentioned below. In brief, the issues involved two hazards, an earthquake and an earthquake-triggered landslide. While these specific examples indicate a cascading scenario, the case study included the occurrence of the earthquake following a period of intense rain, hence a situation where one event (rain) increased the likelihood of another (landslide) in case of a third (earthquake). One obviously sees the cascading element here, i.e. earthquake/landslide, however, it was the rainfall aspect that led to this cascading scenario to be more likely.
Time-dependent vulnerability was also analysed within a study involving wildfires in Portugal using advanced statistical methods and databases to determine which factors were most important within a multi-risk framework. It was found that in disaster situations, financing and coping strategies are interrelated over time, with higher risks potentially emerging and possibly overriding the available resources. As in the Portugal case, communities as well as regional authorities play an important role in diverting help and giving assistance. The macro level analysis additionally showed that repeated events in the past will have consequences in the future, especially for less-wealthy countries. It was therefore proposed to use a systems perspective to capture these relationships and to use a process-based approach to tackle the issue of time- as well as scale-dependency.
Similar to the comment above about cascading events within the societal aspects of risks, cascading issues have also be considered within a systemic vulnerability framework. This involved the application of the I2Sim platform (see below), a tool for simulating interdependencies between infrastructures and lifelines in response to loading such as that arising from natural hazards (Marti et al., 2008). The idea was to examine how the multi-hazard and risk framework will help to more realistically estimate the impact upon issues such as the capacity for the treatment of injured.
Multi-risk frameworks and decision support tools
Multi-risk framework
As part of the MATRIX project, frameworks and tools for use within a multi-hazard and risk environment were to be developed. This may be divided into two aspects. The first is the framework which sets a context within which to conduct such assessments. While developing a fully rigorous mathematically approach that addresses all challenges and uncertainties at all steps of the analysis would be extremely complicated and resource and expertise intensive, there are many situations where the decision-maker can identify the optimum alternative among the possible options without undertaking a detailed, rigorous multi-risk analysis. Therefore, a framework was developed (Nadim and Liu, 2013, Liu and Nadim, 2013) based on a multi-level approach where there is no need to undertake a more sophisticated assessment than is necessary given the problem at hand, or what would be reasonable with the available information and resources.
The developed framework consists of the following stages (see Figure 8):
Level 0: It is assumed that the end-user has identified the relevant threats and has carried out an assessment of the risk(s) at the highest level of sophistication required for the problem at hand.
Level 1: This involves a qualitative assessment of the relevant single hazards.
Level 2: Here requires a qualitative multi-risk analysis.
Level 3: The final and most comprehensive stage, a quantitative multi-risk analysis.
A critical point as illustrated in Figure 8 in the specific inclusion of Monitoring and Reviewing and Communicating and Consulting. As will be discussed later, interactions with end-users and stakeholders are crucial to understand what is actually required and expected with regards to the input to DRR activities from researcher.
Considering first the Level 1 analysis, illustrated in 9, the process involves a flow chart type list of questions that helps the end-user to decide whether or not a multi-type assessment approach is required. These questions explicitly account for cascading hazards and dynamic vulnerability within the context of conjoint or successive hazards. If the Level 1 results strongly suggest that a multi-type assessment is required, then the end-user moves on to Level 2 to make an initial assessment of the effects of dynamic hazard and time-dependent vulnerability. If cascading events are potentially a concern, then user goes directly to the Level 3 analysis.
In the Level 2 analysis, the interactions among hazards and dynamic vulnerability are assessed approximately using semi-quantitative methods (Figure 10a). To consider hazard interactions and time-dependent vulnerability, use is made of a matrix approach based on system theory, as shown by Figure 10b-f (Modified after de Simeoni et al., 1999, Kappes et al., 2010). First, a matrix is developed by defining the hazards of concern along the diagonal (Figure 10b). Considering the example in Figure 10c, the interactions are defined by a clockwise scheme of interactions, the nature of which are described in Figure 10d. These descriptions are assigned numerical codes varying between 0 (No interaction) and 3 (Strong interaction) with intervals of 1, as a function of their degree of the interaction intensity (Figure 10e,f). From this, the degree of the impact of each hazard on the others and the effect from other hazards can be resolved. We employ a hazard interaction index HI, which is the sum of the codes for all the off-diagonal terms. The resulting index value is then compared with some threshold (e.g. HI,max= 50%) and then the decision made as to whether to go to a Level 3 analysis. For example, ff the hazard interaction index is less than this threshold, a Level 3 analysis may not be recommended because the additional accuracy gained by the detailed analyses is most likely within the uncertainty bounds of the simplified multi-risk estimates. Otherwise, if the hazard interaction index is greater than the threshold value, a detailed Level 3 analysis is recommended.
In the Level 3 analysis, the interactions among hazards and dynamic vulnerability are assessed quantitatively with as high accuracy as the available data allow. One of the methodologies developed within the MATRIX project involved the use of Bayesian networks (Nadim and Liu, 2013). This scheme allows both the estimation of the probability of a triggering/cascade effect and the modelling of the time-dependent vulnerability of a system exposed to multi-hazard. An example of a Bayesian network multi-risk model is shown in Figure 11. Such networks would need to take into account events that are:
Independent, but threatening the same elements at risk, with or without chronological coincidence (the column marked in orange in Figure 11), or
Dependent on one another or caused by the same triggering event or hazard; this is mainly the case for cascading or domino events (i.e. the column marked in green in Figure 11).
The MATRIX-CITY and Virtual-City
The above discussed a framework within which a decision maker or analyst can decide upon what level of assessment they need to do towards. In the following, we will present an IT tool, the MATRIX Common IT sYstem (MATRIX-CITY, see Mignan, 2012, 2013 and Mignan et al., 2014). MATRIX-CITY is a first step towards a more general use of multi-risk tools in decision-making, and encompasses 3 major advances in the implementation of a multi-risk framework:
1. The development of a generic probabilistic framework based on the sequential Monte Carlo method to implement coinciding events and triggered chains of events, as well as time-dependent vulnerability and exposure (Mignan et al., 2014).
2. The proposition of guidelines for the implementation of multi-risk, using the concept of the “Virtual City” to test basic multi-risk concepts in a controlled, yet realistic, environment (Mignan et al., 2014).
3. A better understanding of the IT requirements for the widespread use of multi-risk tools, based on the lessons learned from the development of the MATRIX-CITY platform prototype and interactions with stakeholders.
A sequential Monte Carlo method was proposed to generate a large number of risk scenarios (i.e. the generation of hazardous events and the computation of associated losses, while considering the impact upon vulnerability and exposure, e.g. 12). The analysis of these simulated risk scenarios then allowed us to assess losses in a probabilistic way and to recognize more or less probable risk paths, including extremes or low-probability high-consequences chains of events. We found that “black swans”, referring to unpredictable outliers, can only be captured by adding more knowledge about potential interaction processes to the computation process. However, this can only be achieved over time by following a “brick-by-brick” approach, given the considerable effort required.
To quantify hazard interactions, we introduced the concept of the hazard correlation matrix (13a). We considered three categories of interactions: event repeat (e.g. Ai Ai; C C), intra-hazard interaction (e.g. Ai Aj) and inter-hazard interaction (e.g. Ai Bj). The effect could be positive (i.e. probability increase) or negative (i.e. probability decrease), and temporary or long lasting. Time-dependent vulnerability and exposure are not described here, but are taken into account within the framework at a later stage of the calculations. To evaluate how multi-risk participates in the emergence of extremes, we introduced the concept of the risk migration matrix (a more detailed version of the risk matrix discussed above) and showed that risk migration and risk amplification are the two main causes for the occurrence of extremes (13b).
The multi-risk framework was developed and tested based on generic data and processes generated following the heuristic method. This strategy, which involves the use of intuitive judgment and simple rules, allows for the solving of problems that are otherwise difficult to consider. Our approach follows the existing recommendations on extreme event assessment, which involves the use of inductive generalizations and "scientific imagination" to include known examples of extremes, as well as potential "surprise" events within the same framework. However, abstract concepts, such as the definition of generic perils, remain difficult to comprehend and we therefore proposed some guidelines to help risk modellers and decision-makers apply this approach to realistic cases.
For this purpose, we developed the concept of the Virtual City (Mignan, 2013, 14). Within this concept or tool, the perils A, B, C, D and E are no longer simply abstract concepts, but are replaced, for instance, by earthquakes, volcanic eruptions, tsunamis, fluvial floods and storms. Hazard, exposure and vulnerability data, as well as details about possible interacting processes, are based on real examples obtained from the scientific literature.
Although these tools were based on state-of-the-art software engineering and a Python-based code, it was rapidly observed that multi-risk software would need to have all the functionalities of existing risk tools, on top of the multi-risk framework. Such a task would require significant resources and a commitment of modellers used to other types of risk modelling tools (including various procedures and formats). Therefore, at this stage, we recommend the exporting of the method developed for this IT tool to existing risk tools, which would facilitate its implementation and potentially encourage the widespread use of the proposed approach. Hence, the present work should be seen as a proof-of-concept, as we did not intend to fully resolve the complex problem of low probability-high consequence events. We only considered a selected number of possible interactions, where naturally adding more perils and interactions would yield more complex risk patterns. We thus recommend a brick-by-brick approach to the modelling of multi-risk, to progressively reduce epistemic uncertainties. A more realistic modelling of low-probability high-consequences events would also require the consideration of additional aspects, such as uncertainties, domino effects in socio-economic networks and long-term processes such as climate change, aging infrastructure and exposure changes. While the concepts developed in the present study outline the theoretical benefits of multi-risk assessment, identifying their real-world practicality will require the application of the proposed framework to real test sites.
Decision support tool
Another tool for decision support and presented to end-users also involved using the risk matrix as described above, that is it followed the definition of risk as a combination of the consequences of an event or hazard and the associated likelihood of its occurrence, again adapting the BBK (2010) framework (see the review in Wenzel, 2012). The risk matrix methodology was implemented into decision-support software based on the principles of Multi-Criteria Decision Analysis (MCDA), and tested with a group of stakeholders to communicate and transfer the information contained for the different risk scenarios in the risk matrix to the various stakeholders involved. The decision-support tool allows the stakeholders to display the total risk index ranking of different risk scenarios and to attribute different weights to the loss components (e.g. an extremely rare offshore earthquake which can trigger a tsunami, or a release of toxic material with severe impacts on the local environment, etc.) affecting a region in terms of expected losses that are quantitatively derived in different sectors (human, environment, economy, infrastructure, intangibles) for each scenario. An example of such a series of scenarios and their consideration within the MATRIX decision support tool (employing a risk matrix) is presented in 15. Following this approach, the sectorial losses are combined as a weighted sum into a single aggregated loss score for each scenario, as shown in Figure 16. Together, these two steps (i.e. severity and loss scores) are combined to produce a total risk index for each scenario.
For example, in Figure 16a, it can be seen that the offshore earthquake triggering a tsunami is deemed to have a much greater risk score than the toxic spill. As the total risk index for each scenario is determined as the aggregate weighted sum of each of the loss measures in each of the different sectors, the risk index ranking will also depend on the weights given to each sector. Through a participatory approach, the stakeholders assign the relative importance (weights) to the losses for the different sectors for each of the scenarios likely to occur in the region. Next, the decision support software is used in a group setting to discuss the weighting outcomes and interactively examine the variability of the ranking results. For example, a sensitivity graph can be used to see the effect on the rankings as the weights are changed. In Figure 16b it can be seen that as more weight is given to the “People” criteria (i.e. casualties, short- and long-term mass care), the risk score for the toxic spill decreases considerably. This is due to the fact that the toxic spill scenario produces none to very few fatalities and has a minor impact on mass care. As a result, when all the weight is given to only one measure, in this case human losses, the risk score for this scenario is minimal. On the other hand, the risk score of all other scenarios goes up, but importantly the relative rankings between them stays the same. Using various visualization tools in the decision support software, such as sensitivity graphs, stacked bars, scatter plots, and one by one comparisons between scenarios, the stakeholders are able to evaluate the total risk from different scenarios by considering many variables at once, which enables them to separate facts from value judgments, and better communicate their choice to others.
Tangible and intangible loss treatment
Usually within risk assessment, most attention is paid towards direct losses, referring to the physical components of society. However the real situation is much more complicated. Losses can be tangible or intangible, as well as being characterized as being immediate/direct (during the event), and being short- and long-term (indirect) losses. Some types of risks are relatively straightforward to estimate, while others can only be estimated approximately. In addition, data limitations must often be taken into account. A further complication, and an important analytical step, is the presence of risk aversion and extremely large risks. It is important to combine and compare risks from low probability, high consequence events with risks from more frequent, but low to medium consequence events.
These issues were examined within MATRIX by exploiting a risk analysis chain perspective (Pflug and Römisch, 2007), that is a framework involving the measurement, modelling, and management of risk. It was found, perhaps unsurprisingly, that different risk measures and modelling approaches are required for each of the components of the framework just outlined. Furthermore, for the case of extreme risk, traditional risk management strategies, successfully applied for frequent risks, are likely to fail. However, while there are fundamental differences between frequent and extreme risks in various dimensions, from a risk management point of view, they can be assessed together, given the right measures and decision support approaches (Hochrainer, 2011).
Regarding the tangible and intangible loss assessment component, it is argued that from a methodological perspective, it is beneficial to distinguish between dependent risks, i.e. risks which, if realized, are changing the likelihood of other risks, and independent ones. Additionally, based on a reflexive modernity point of view, e.g. seeing the decision making as a participatory rather than hierarchical process, different risk bearers exposed to the same risks, but over different scales, have to be considered separately. This is important as one cannot assume that the weights (or importance) proposed for the different components of risk stay constant over different scales (e.g. household to national), especially when the intangible contributions are considered. A multi-criteria analysis approach was therefore suggested for risk bearers on different levels to sort out some of the most relevant dimensions which need to be considered in such a setting (Hochrainer, 2011).
The MATRIX test sites
In order to verify the concepts and tools developed within MATRIX, it is necessary to apply them to real world situations where conjoint and cascading events and interactions between different hazards and risks need to be considered. It is for this reason, and matching the expertise of the consortium, that the MATRIX test cases were chosen. All three are under threat from multiple hazards. Naples (Garcia-Aristizabal et al., 2013a) and Guadeloupe (Monfort and Lecacheux, 2014) are the most threatened (and complex) examples, with both endangered by volcanic eruptions, earthquakes, as well as hurricanes (Guadeloupe), landslides (Naples and Guadeloupe) and forest fires (Naples). Each case is also susceptible to cascading events, in particular rain- and earthquake-induced landslides and volcano-earthquake interactions. Cologne (Fleming et al., 2014) on the other hand is not as exposed to such a range of hazards, nonetheless it must still contend with threats from earthquakes, floods and windstorms (Grünthal et al., 2006), with the possibility of earthquake-induced damage to its dyke system increasing the flood risk to the city.
During the project, the work on each test case tended towards a particular aspect of the multi-hazard and risk problem. For example, Cologne became the focus of efforts towards the comparison and harmonisation of single-type hazards and risks while considering uncertainties (3 and 4). However, another problem confronted involved the possible consequences of seismic loading on flood defences, leading to a potential increase in risk. For Naples, the focus tended towards cascade effects, of which two scenarios were examined in detail (see below). Finally for the French West Indies, a systemic assessment of the consequences of a multi-hazard and risk scenario on Guadeloupe’s infrastructure was made.
In addition, a preliminary effort was made to employ the multi-risk framework described above. Considering first Level 1, we present the answers to the raised questions for each test case in Table 3, we note immediately that for each example, we must proceed from the initial “More than one hazard?” question to dealing with the various interactions, with the need for at least a Level 2 analysis. However, even if this were not the case, i.e. only one hazard of concern, there is also the possibility of events of the same kind reoccurring during a given time period, which may be taken as the time required to carry out the necessary repairs/recovery from the original event (e.g. a series of storms separated by short periods of time). We also note that for all three cases, we would probably need to proceed to a quantitative Level 3 analysis, based on the fact that cascade events may arise. However, the fact that cascade events in Naples and Guadeloupe are more likely than in Cologne cannot, at this stage of an assessment (or comparison), be resolved. In addition, the cascade example for Cologne, i.e. an earthquake damaging flood defences, hence increasing flood risk, would also fit within the context of conjoint events. Therefore, it would appear that even the most “quiet” territories may be exposed to several hazards, with interactions potentially always present (for example, Na-Tech - Natural Technological - interactions are in many industrialised districts a major concern, although they are not dealt with in detail in MATRIX). Hence, one may expect the situation where only a Level 1 assessment is required would be fairly rare.
Considering the Level 2 assessment, the aim is to describe the various relationships between the assorted hazards, following the matrix method outlined in 10 and presented in 17. To reiterate, to read these figures, consider first that along the diagonal, the hazards of concern are listed. Then, moving in a clockwise manner, the level of interaction (scored between 0 and 3 with intervals of 1, where 3 indicates a strong interaction and 0 indicates none) and the nature of such interactions between each hazard pair are identified.
For Naples and Guadeloupe (17a and 17b), for the purpose of this work, we simply refer to three hazards, although obviously a larger matrix would be needed for a thorough study. We note the strong (3) interactions between some hazards, e.g. earthquakes and volcanoes for Naples, landslides and earthquakes for Guadeloupe, as well as hazards where no interaction would arise (e.g. hurricanes and earthquakes). Considering Cologne (17c), we identify few interactions between hazards, i.e. windstorms potentially bringing heavy rain, although for Cologne, more localised heavy precipitation causes little widespread flooding, and an earthquake damaging flood defences. However, it is also recognized that if we considered this at the risk level, then a windstorm may damage a building, increasing its susceptibility to a later earthquake, while considering the reverse (an initial earthquake followed by a windstorm) would most likely be more serious. Based on the numbers presented in each square, a so-called hazard interaction index may be inferred (found by adding all results row by row, representing causes, then column by column, representing effects), and as mentioned above comparing to some criteria (e.g. a predefined percentage of the maximum possible index for a given site) may decide whether or not to proceed to the more resource intensive Level 3 analysis. For example, Naples has a score of 16 and Guadeloupe 12, while Cologne has a value of 4, indicating the much great importance of such interactions for the first two cases.
In the following, we will now consider the specific aspects of the test cases introduced above. However, we will not do so in detail, with the reader referred to the associated deliverables.
Naples
As mentioned above, two scenarios were the focus of the cascading studies for this test case (the results of which are presented in deliverables D3.3 Garcia-Aristizabal and Marzocchi, 2013, and D7.3 Garcia-Aristizabal et al., 2013a):
Scenario 1: Volcanic earthquakes and seismic swarms triggered by volcanic activity (interaction at hazard level).
Scenario 2: Simultaneous occurrence of ash-fall and earthquake hazards (interaction at the vulnerability level).
For the first scenario, during the evolution of a volcanic system from its quiescent to eruptive state, a large number of small- to moderate-sized earthquakes occur. In fact, volcanic unrests are normally accompanied by (volcanic) seismic activity located directly below the volcano and/or also by triggered seismic swarms (sometimes categorized as ‘distant volcano-tectonic events’. In this scenario, the problem is to assess the possible effect that local seismicity (normally characterized by many shallow micro-earthquakes) triggered during volcanic unrests may have in the area where the volcano is located, and compare it with the general seismic hazard analysis. In the Neapolitan area, this case may be of particular interest given the highly urbanized area in the surroundings of Campi Flegrei (see Figure 18). The interactions involved in this case are at both the hazard and vulnerability levels, and at time scales from seconds to possibly years, although the focus was on the hazard level. It was observed how the contribution of volcanic earthquakes may be significant in an area located in the Campi Flegrei caldera (around Pozzuoli municipality), whereas the effect quickly decreases with distance, being negligible for distances greater than a few kilometres (e.g. for most of the area of Naples).
The second scenario, earthquakes and eruption-related hazards, can result in the increased vulnerability of exposed elements. The objective is therefore to estimate the “Loss distribution”, and/or “Expected Loss” while considering the possibility of the simultaneous occurrence of two hazards. In our case, these are: (1) volcanic ash-fall hazard and (2) seismic hazard, in the Arenella area (see 18). The interactions here are more towards the vulnerability level, acting over seconds to days.
It was found (see 19) that for the combined effect of seismic and volcanic ash (deposited over the roofs), the loss distribution and expected loss considering different values of ash load (3 and 5 kPa of ash load over the roofs) for 3 kPa are very close to those obtained for the pure seismic case (i.e. with no ash), whereas a greater difference (i.e. increased risk) is observed for the case of 5 kPa. Taking in consideration the median of the estimated curves, for an ash load variation of 3 kPa (from 0 to 3 kPa), there was an increase of the expected losses of about 19%. However, in the second case, an increase of 2 kPa (from 3 to 5kPa) of ash load implied an increase of about five times in the expected losses, revealing a strong non-linear (and more complex) arrangement.
French West Indies
The first results discussed here concern the establishment of a framework to compare risks within the context of the Pointe-à-Pitre municipality, considering storm surges and seismic risk. 20 shows the results of loss estimates as determined during MATRIX (see deliverable D7.4 Monfort and Lecacheux (2014) for details) for seismic risk for Pointe à Pitre and 5 return periods (100 years, 475 years, 975 years, 1975 years and 5000 years), and for storm surges, although for only a 100 years return period was considered, but for different maximum wave heights. 20a shows the results for economic losses, where it is apparent that earthquakes are significantly more serious than storm surges. However, if one looks at “no shelter”, we see for the 100 year return period the levels are very similar. It should also be commented that casualties may not be the best measure to compare earthquakes and storm surges, since there is an element of warning in surges (allowing people to make their own mitigating actions) while no or few such opportunities arise for earthquakes.
Another study performed over the Guadeloupian territory was the assessment of systemic vulnerability of infrastructures exposed to cascading events. This work aimed at analysing the effects of cascading events on the behaviour of interdependent systems and on the capacities of the health care system to treat the victims, with Guadeloupe being a useful example given its context as a closed system. The hazard cascading scenario considered involved a M6.3 earthquake striking Basse-Terre, and triggering landslides in the mountainous areas where antecedent precipitations had made the area prone to such events, leading to the closure of road links. Damage due to earthquakes was estimated for the 5 considered systems (buildings, healthcare system, electrical network, water supply network and transportation).
The inter- and intra-dependencies of the systems were found using the Infrastructure Interdependencies Simulator (I2Sim, Liu, 2007). I2Sim provides a platform to represent all critical infrastructures and their operating interdependencies. For example, 21a shows a simple system containing a power station, a water station, a hospital and a residential area under normal conditions. Even though it is a simple system, it is already has feedback loops and interdependencies. The consequences of disrupting, for example, the power supply (21b) then leads to decisions to be made as to how to distribute power, that is, balancing the requirements of the hospital, water station, etc.
Analyses were performed for different strategies of resources allocations, and one of the final results was the impact of the induced landslides on the health care treatment capacity. Based on the simulations, the time required to transport all injured people to a functioning hospital has increased by a factor of 2.5 if one takes into account the consequences arising from lifeline disruptions. 22 shows an example of how the various lifelines and their relative effectiveness may be determined from such a tool as I2Sim.
Cologne
Details of the risk estimates arising from the three hazards that mainly affect Cologne are presented in deliverable D7.5 (Fleming et al., 2014). As we have already presented some results for Cologne, namely the combination and comparison of risk types, here we will simply present the new results for the three hazard types. This is presented in 23 for earthquakes (23a), floods (23b) and windstorms (23c). For seismic risk, a logic tree approach was employed, leading to a total of 180 estimates of risk arising from a range of input parameters (see Tyagunov et al., 2014). For floods, use was made of the hybrid probabilistic-deterministic coupled 1D (dike breach) 2D hydrodynamic model IHAM (Vorogushyn et al., 2010) run in a Monte Carlo simulation. For windstorms, results from the Vienna Enhanced Resolution Analysis or VERA tool were used (Steinacker et al., 2006). All risk calculations for the three hazards used the same metric and total costs (i.e. direct damage to residential buildings). For seismic risk, we plot the mean estimate with the 95th and 5th percentiles, along with the seismic risk curve of Grünthal et al. (2006). For the flood and windstorm risk curves, we plot the median plus the 5th and 95th percentiles and again the curves of Grünthal et al. (2006). The major point to note is the substantial uncertainty arising for each hazard type. This makes it difficult to identify for given return periods which hazards should be of more concern to decision makers. It also again emphasises the results presented earlier employing the Wilcoxon’s test (4)
Interactions with the DRR community
End-user and stakeholder interaction was a very important and fruitful aspect of the MATRIX project. Details of some outcomes will be given in the following section dealing with dissemination activities, but here we will review a few of the main findings and responses. To undertake these activities, MATRIX instigated an external group referred to as the Cooperation partners. These were stakeholders (National Platforms for Disaster Risk Reduction, UN-ISDR, EU, CoE, and the private sector) who were prepared to participate in the project’s meetings, workshops, and to complete questionnaires about various aspects of multi-risk assessment, in particular its status in the different European member states. Considering the National Platforms, of the 22 European platforms, 13 interacted with MATRIX2.
One important activity carried out, which lead to a publication (Komenantova et al., 2014) was an investigation into how stakeholders reacted to the tools developed within MATRIX, that is the MATRIX-CITY/Virtual-City and the decision support tool of KIT (discussed above), as well as what their actual needs were. Existing risk assessment methods integrate large volumes of data and sophisticated analyses, as well as different approaches to risk quantification. However, the key question is why do losses from natural disasters continue to grow if our scientific knowledge on multi-risk is increasing? (White et al., 2001). The MATRIX work here dealt with the questions of communication and the transfer of scientific knowledge on multi-risk and its underlying drivers to stakeholders within the decision-making process. A two-way communication process allowed feedback to be collected from stakeholders (i.e. civil protection offices) across Europe on the usability of the multi-risk decision–support tools that have the potential to benefit decision-makers and to provide them with information on mitigation measures. This would also allow the integration of their feedback into improving the tools themselves.
The theoretical background of this work involved the concept of risk governance, which takes into account cultural and political factors when implementing risk mitigation measures and emphasizes the role of participation and communication. The risk governance concept is concerned with such issues as how information is perceived, collected and communicated, and, based on these factors, how management decisions are made (IRGC, 2005). Participatory modelling is an important part of risk governance and allows us to take into consideration not only facts, but also values by collecting feedback from stakeholders (Forester, 1999). The process of interacting with stakeholders leads to an enhanced understanding of the views, criteria, preferences and trade-offs employed in decision-making (Antunes et al., 2006). Also, since the development of scientific tools is also a social process, it is essential to involve relevant stakeholders who will be using the tools in the design process through the collection and integration of their feedback (Tesh, 1990).
There is also a gap in the scientific literature about feedback with respect to the usability of decision-support tools. While the use of feedback for the development of decision-support tools for environmental issues has been reported (Constanza and Ruth, 1998), as well as there being multi-risk decision–support tools that have the option of collecting feedback (T6, 2007), there is no evidence or analysis of the feedback from stakeholders from practice on the usability of multi-risk decision-support tools. During our work, we not only collected such feedback from civil protection officers, but we also used this information to improve the developed decision-making tools, directly integrating stakeholders’ perceptions into the model by attributing different weights to loss parameters according to preferences from stakeholders. The information was gained during two workshops, namely a MATRIX stakeholders’ meeting in Bonn (July, 2012, discussed further below) and a workshop on urban multi-hazard risk assessment in Lisbon3 (October, 2012), and from a questionnaire distributed prior to the first workshop.
The general results show that for the usage of multi-risk decision-support tools, two areas are most problematic. These are (1) the absence of clear definitions and (2) the lack of information on the added value of multi-risk assessment. Multi-risk is not systematically addressed among the EU countries for all hazards, but is only singularly integrated into risk assessment approaches. Some examples include the superposition of existing single-hazard risk prevention plans for all hazards, for example, combining flood and landslide hazards and flood risks with wind effects, the application of which has been within the context of risk assessment for critical infrastructure (in particular the combination of meteorological and technological risks). Generally, multi-risk analysis is barely or not at all integrated into decision-making processes, and only around half of stakeholders were aware of methodologies and tools to assess multi-risk.
The reaction of stakeholders to the multi-risk assessment and decision-making tools presented at the workshops was optimistic. Several stakeholders invited the tools’ developers to give presentations and conduct training at their home institutions. However, the usability of the tools in practice is complicated by such factors as the required large volume of input parameters (involving cumbersome data gathering) and that their possible application is limited to only a narrow number of experts, since a high-level of expertise is required to assess the dynamic multi-hazard and multi-risk processes, taking into account the complexity of the models and the required parameters.
The consultation process with stakeholders also showed significant variation in perceptions between stakeholders in academia and in practice. While both academicians and practitioners agreed that the decision-support tools are useful for understanding losses and their contributions in a risk scenario, differences arise between how practitioners viewed the usefulness of the tools when it came to prioritizing risk and developing risk management strategies. Similarly, practitioners found the tools less useful than academics when it came to preparing for disasters and allocating resources. For example, Figure 24 shows the almost opposite view of academics and stakeholders with respect to the results of the survey that asked for impressions as to how the decision support tool helps with the understanding of losses and their contribution in a risk scenario (Figure 24a) and how this tool helps with preparing for multi-risk disasters and optimizing the allocation of resources (Figure 24b). As can be seen, while both academics and practitioners thought the tool was useful for understanding (although less so for practitioners), practitioners did not see it as useful for actually preparing for multi-risk disasters and allocating resources, a critical finding for those (usually academics) who are developing the tools.
Recommendations with respect to two areas involving the application of decision-support tools involved convincing stakeholders involved in the decision-making process of the usefulness of the multi-hazard approach, as well as dealing with the problem of disseminating any results to the general public (hence confronting public acceptance issues). Some stakeholders expressed the opinion that politicians could also use such models as training to see what the consequences of a multi-hazard situation could be. Another general recommendation was that the decision-support tools could be used for education and training purposes.
Risk governance and multi-hazard and risk frameworks
It was identified during MATRIX that in risk assessment research and policy, there is much debate on multi-type hazard and risk assessment and the definition and use of realistic scenarios. This debate has been evoked, not least, by specific disasters in recent years, namely the Super Typhoon Haiyan (the Philippines, November 2013), causing floods and landslides, and the Töhoku earthquake (Japan, March 2011), with the resulting devastating tsunami and nuclear accident. Therefore, in MATRIX efforts were made to synthesise the identified benefits and barriers to multi-hazard mitigation and adaption, the target audience being practitioners within the public/private sector working in communities exposed to multiple risks as well as to those active at the science-policy interface, thus including researchers, policy and decision makers in risk and emergency management (Scolobig et al. 2013; Komendantova et al., 2013; Scolobig et al., 2014).
It was found that within current single-risk-centred governance systems, both centralized and decentralized, (which have evolved in parallel with the single-risk-centred assessment processes), practitioners hardly ever have the opportunity to discuss multi-risk issues, including triggered events, cascade effects and the rapid increase in vulnerability resulting from successive hazards. However, as revealed by the results from various workshops (see the following section) as well as several interviews with stakeholders, that risk and emergency managers clearly see the benefits of including a multi-risk approach in their everyday activities, especially in the urban planning sector, but also in emergency management and risk mitigation.
One example where practitioners believe a multi-risk viewpoint would be of value concerns decisions surrounding building restrictions for urban planning, as well as the correct cognitive and behavioural biases. A multi-risk approach is considered particularly useful also for gaining a holistic view of all of the possible risks that may affect a territory. For example, such an approach can show that focusing only on the impacts of one hazard could result in raising the vulnerability of the area to another type of hazard. An example of this is in the older buildings of Kobe, Japan, which were built with relatively heavy roofs. While this helped mitigation against frequent typhoons, it enhanced their vulnerability to rarer earthquakes (Otani, 1999).
Other benefits that are considered to be particularly crucial by practitioners include:
The cost reductions and improvements in the efficiency of proposed risk mitigation actions.
The development of new partnerships between agencies working on different types of risks.
An awareness of the potential for expected losses being exceeded (i.e. the total risk is possibly greater than the sum of the individual parts), as well as the lives and property saved and better protected by the use of a multi- vs. single-risk approach.
However, further research is still needed in order to better understand the extent of some of these benefits, as well as the need to consider aspects of the mitigation problem, such as the different time scales involved between the events themselves, response, initial recovery and ongoing mitigation. Our results also reveal that practitioners and researchers have in mind different agendas for future research on multi-risk assessment. Therefore, a transparent process to reach a compromise on the required priorities is needed.
Barriers to an effective implementation of multi-risk assessment can be found in both the science and practice domains. For example, considering scientific contributions to risk assessment research, the process has evolved differently in the fields dealing with geological versus meteorological hazards, with the different scientific development paths representing a major barrier to understanding and communicating between different “risk communities”. Accompanying this is the lack of open access to databases and research results, which is particularly worrying for risk managers. Overarching these problems are the matters of the lack of interagency cooperation and communication, which are particularly difficult for risks that are managed by authorities acting at different levels (e.g. in Italy, national bodies are responsible for volcanic risk, while river basin authorities deal with flood risk). The lack of capacities at the local level and unsatisfactory public-private partnerships are also major barriers that need to be confronted.
As a result of our interactions and discussions with stakeholders, some priority actions have been identified:
Encourage knowledge exchange and dialogue between the risk communities dealing with geological and meteorological hazards;
Identify new options for mitigation, e.g. multi-risk insurance schemes, new forms of public-private responsibility sharing for households exposed to multi-risks;
Develop territorial platforms for data and knowledge exchange between researchers and practitioners;
Create an inter-agency environment, where the different departments at the national and/or regional governmental level can exchange information, develop complementary protocols, and serve to provide consistent information and responses to the relevant stakeholders;
Create commissions for discussion at the local/municipal level ("local multi-risk commissions") in order to gain a common understanding of what multi-risk assessment actually is, what kind of cooperative actions can be undertaken to implement it, what are the priorities for future research etc.. Members of these commissions should be decision and policy makers, researchers and local natural hazard advisors, the latter acting as the liaising bodies between local communities and practitioners.
Potential Impact:
The first point to make with respect to the “impact” of the MATRIX project is that multi-hazard and risk is a relatively new field of endeavour, and that MATRIX itself can only realistically be considered as contributing to the expansion of research in this direction. Having said this, one of the more prominent areas where MATRIX has had an impact is in raising the awareness of the issue of multi-hazard and risk assessment. In the following, we will therefore begin with an overview of where this awareness of multi-hazard and risk assessment has grown as a result of MATRIX (recognising of course that MATRIX is by no means the only driver of this point), to be followed by summary of the dissemination activities undertaken, and some proposed ways in which the results may be exploited. For the latter point, mention will be made of current projects that have the potential to benefit from the MATRIX results.
The impact of the MATRIX project, in fact, is linked to the dissemination activities, in terms of representatives of the project being invited to various expert meetings, in particular those associated with the European Commission (see below for some comments as to the response to such presentations), the aim being, as mentioned, to encourage DRR practitioners to consider the need for a multi-hazard and risk framework. In fact, in a recent report from the European Commission on the post-2015 Hyogo Framework for Action1, it was commented that a number of Member States in their National Risk Assessments (NRA) mention “’consequential’ and ‘cascade-effects’” of risk (incidentally, MATRIX is specifically mentioned in this document). While it is difficult to assess to what extent MATRIX’s efforts have had on the recognition of this issue, acknowledging that multi-hazard and risk as understood within MATRIX was discussed in an early European Commission document outlining risk assessment and mapping guidelines2, the MATRIX project was specifically mentioned as an effort to develop the means of dealing with this issue. Another impact in terms of raising the discussion about multi-risk in DRR activities involves the production for the next Global Assessment Report (GAR 15) of a document outlining the issues involved in disaster risk governance when considering a multi-hazard and risk framework (Scolobig et al., 2014), as also outlined in Komendantova et al. (2014).
The dissemination activities conducted during MATRIX may be divided into three main types:
“Standard” dissemination practices in the form of academic publications, conferences and workshops, and representing MATRIX at other projects’ meetings, as well as the project’s website, fact sheets and its listing on other relevant websites (e.g. GRIPWEB3). Representatives of other projects and of Civil Protection were also invited to the project’s annual meetings.
“Expert” and similar meetings within the context of the various branches of the European Commission (commented upon above in terms of the impact of the project).
“End-user and stakeholder” meetings, which endeavoured to not only inform such groups about the results and goals of MATRIX, but also to encourage the adoption of a multi-hazard and risk framework in general. This part of the dissemination activities was possibly the most important, as it involved making contact with the European National Platforms for Disaster risk Reduction, as well as interacting with such parties as the European Commission, the UN-ISDR and the private sector, such as insurance companies.
The publications produced during the project are listed in Template A1 of Section A. In order to make these works available in an open source manner, efforts were expended in making them available via the OpenAire4 (Open Access Infrastructure for Research in Europe). Note these manuscripts are made available in an unformatted form in the “accepted version” to comply with the copyright regulations of the journals involved.
MATRIX was represented at a number of project meetings. These included the closing meeting of the KULTURisk project, which aimed to develop a culture of risk prevention by evaluating the benefits of different risk prevention initiatives5, and the Task C EUSBSR (EU Strategy for Baltic Sea Region) Flagship project 14.36 whose objective was to develop scenarios and identify gaps in the response to all of the main hazards that affect the Baltic Sea Region. In both projects, while multiple types of hazards are accommodated, a genuine multi-hazard and risk framework as considered within MATRIX was still lacking.
The presentation of MATRIX at various expert meetings in Brussels not only served to advertise the project’s goals, but to bring to the DRR community’s attention the potential value of a multi-hazard and risk approach. There was (and probably remains) some scepticism as to the value of such an approach. For example, following one of the first of these meetings where MATRIX was presented (July 2011, sponsored by the European Commission Humanitarian Aid and Civil protection, ECHO), one attendee commented “I would be happy if I could manage a simple risk assessment. Multi-risk is far away from the reality on the ground.” It was therefore the intention of subsequent MATRIX presentations to argue that this was not the case, and that a multi-hazard and risk assessment regime is plausible at this time. In fact, it was regularly proposed during similar presentations that on one level, there is no choice, but to adopt a multi-type approach given the increasing complexity of human society, and the recognition of the interactions that arise at all levels of disaster mitigation. Similarly, at a workshop organised by the European Commission Directorate General for Research and Innovation in Brussels dealing with the contribution of the socio-economic studies to hazard and risk research (22nd to 23rd October, 2012), at which the MATRIX project was presented, it was emphasised that research into the actual processes involved (that is, fundamental natural science research) was still required, especially within the context of a multi-hazard environment.
The most important (and productive) of the dissemination activities carried during the MATRIX project involved two end-user and stakeholder meetings, organised by DKKV (together with KIT and GFZ), one of the MATRIX partners and the German contact point for Disaster Risk Reduction. These meetings (held in July 2012 and June 2013) aimed to present the aims and results of the MATRIX project to various members of the DRR community, including representatives of the European National Platforms for Disaster Risk Reduction, UN-ISDR, the European Commission and other bodies such as insurance companies. The primary aims of these meetings may be summarised as follows:
Outlining of the aims and work program of the MATRIX project.
Presenting the concepts behind multi-hazard and risk assessment, as understood within the project. In particular, the case needed to be made for the importance and potential benefits of considering a multi-hazard and risk framework, where the spatial and temporal interactions and dynamics at the hazard, exposure, and vulnerability levels were accommodated, while also accounting for socio-economic impacts.
The presentation of various decision support tools developed within MATRIX were presented. This in particular involved showing the MATRIX-CITY and the Virtual City tools.
The presentation of the cross-cutting results from the test cases.
Presentations by representatives of the National Platforms and civil protection groups and insurance associations, covering themes ranging from the DRR status in their respective countries to test cases which may also involve a multi-hazard and risk environment.
An issue future projects will need to confront is that getting the private sector involved may prove to be very challenging, as it was during MATRIX. While the German Insurance Association was represented at the second workshop, the minutes of that meeting made available to other groups (Icelandic Natural Hazard Insurance Company and Austrian Railway), and a number of contacts to potentially interested enterprises, significantly greater effort would required to involve the private sector in future projects, especially within the Horizon 2020 (H2020) scheme. Another point raised by the attendees of the 2nd meeting was that, rather than simply contacting end users during a project or even before, it was advised that they should be involved with the actual preparation of future project proposals, rather than simply being “added on”.
With regards to the exploitation of the results of MATRIX, one point is the consideration of MATRIX in future research proposals. It is now expected that in the current round of hazard and risk proposals being (within which many members of the MATRIX consortium are involved), a multi-hazard and risk perspective is assumed and incorporated into any planning.
Finally, one of the MATRIX deliverables was a vision statement, outlining what the next steps should be within the context of multi-hazard and risk research (Wenzel, 2014). In summary, the main suggestion is for the redirection of future multi-risk research from risk assessment to risk mitigation, and for it to be considered within the context of resilience, which in fact is the case for a number of H2020 (and other funding bodies’) calls currently being prepared for. Some other issues that need to be considered in terms of research and implementation include:
The need to move from human and economic loss estimation to the efficiency of response and the speed and sustainability of recovery.
Continue work on the various forms of interactions (system, natural processes, etc.), and treating multi-hazard and risk over different spatial and temporal scales. This includes ongoing work on methods for predicting the consequences of, for example, cascading effects.
The importance of ensuring that the software tools developed within projects such as MATRIX are “not lost” (i.e. that resources are found to move them beyond the prototype stage).
Expansion of different means of data collection, and exploitation of different types and sources of data, including ground-based and remote sensing platforms and monitoring systems, crowd sourcing and “big data”.
The development of methods for multi-hazard and risk assessment should be integrated with those that allow for the incorporation of utility functions, cost-benefit analysis, and life-cycle analysis.
List of Websites:
Project website: http://matrix.gpi.kit.edu/
Project coordinator: Professor Dr. Jochen Zschau (zschau@gfz-potsdam.de)
Centre for Disaster Management and Risk Reduction Technology
Disaster risk reduction (DDR) activities generally treat different natural hazards and their associated risks separately. However, this approach ignores the spatial and temporal interactions that may arise along the disaster risk chain. For instance, an extreme event may trigger others, e.g. earthquakes causing tsunamis, or several different types may occur concurrently, e.g. severe weather and earthquakes. Considering vulnerability, an initial event would leave a community more susceptible to future, possibly different, hazards, e.g. an earthquake weakening buildings which are damaged further by windstorms. The temporal dimension may include changes in exposure, e.g. increased urbanisation, altering the total risk to an area, while repeated events lessen a community’s resilience. Meanwhile, although losses are estimated by usually only considering direct economic losses or casualties, this ignores less tangible losses such as reduced business activity or the loss of cultural heritage. In short, the total risk estimated when incorporating interactions between multiple hazards and risks is likely to be greater than the sum of the individual parts.
The Multi-HAzard and MulTi-RIsK Assessment MethodS for Europe or MATRIX project (01.10.2010 to 31.12.2013) set out to tackle some of the issues associated with multi-hazard and risk assessment. The focus was on the hazards that most affect Europe, namely earthquakes, landslides, volcanos, tsunamis, wild fires, storms and fluvial and coastal flooding. Interactions at the different levels were considered, such as cascading events and time dependency in vulnerability. Three test cases were considered, Naples, Italy, the French West Indies, and Cologne, Germany. Considerable interaction with end-users was undertaken, including identifying biases at the individual and institutional level which may hinder employing a multi-type framework.
The MATRIX project is recognised as being a step towards seeing multi-hazard and risk frameworks considered the standard in Europe, as evident by its representation at expert meetings and its inclusion in a recent EC document regarding the post-Hyogo framework. While it is acknowledged that multi-hazard and risk assessment are at an early stage, great interest has been shown by practitioners who note both their desire, and the potential barriers, to exploiting such a paradigm.
Some of the foremost outcomes of MATRIX may be summarised as follows with others outlined in detail in the rest of this report:
A framework was developed based on a multi-stage process (single qualitative, multi-qualitative, multi-semi-quantitative, multi-quantitative) that allows a practitioner to proceed with a hazard and risk assessment only as far as is necessary within a multi-risk environment.
A generic modelling tool that allows a practitioner to consider (or ignore) interactions between different hazards within a probabilistic framework (based on the Monte Carlo method) was developed. This may be used either for a given situation, or via a more general tool that employs a virtual city to allow practitioners to experiment with the relative importance of various interactions.
An assessment of the problems to the implementation of a multi-risk framework being incorporated into risk governance found the main barriers to include a lack of resources (time, personal, expertise, especially at the local level), the frequent disconnect between products developed by researchers and practitioners, and poor communication and interaction between the different “departments” who are effectively responsible for risk assessment for specific hazards.
Project Context and Objectives:
Many types of natural extreme events, including earthquakes, landslides, volcanic eruptions, tsunamis, river and coastal flooding, winter storms, and wildfires, threaten different regions of Europe. Disaster Risk Reduction (DRR) practitioners, decision and policy makers, and the experts who inform their judgment, usually treat such hazards and the associated risks separately. This means they follow a single-hazard risk framework, neglecting the frequent temporal and spatial interdependencies that often arise between them (see, for example, Carpignano et al., 2009; Kappes et al., 2012, Marzocchi et al., 2012).
Such dependencies arise at all levels of risk estimation (1). Considering the hazard level, situations may (and often do) arise where one hazard triggers another via so-called cascade or domino events, for example, tsunamis triggered by earthquakes, with the 2004 Indian Ocean and 2011 Töhoku earthquakes and tsunamis being the most prominent recent examples. Similarly, one type of hazard may increase the likelihood of another, e.g. droughts and wildfires, as in the case of the 2009 south eastern Australia bushfires, while there is the possibility of conjoint events, that is events occurring at the same time or, at least before the necessary repairs have been made as a result of the first, e.g. a windstorm soon following an earthquake.
At the vulnerability level, an initial event may increase the susceptibility of buildings to further damage from another, possibly different, incident, for instance, considering earthquakes and windstorms. Considering this example, the order of the events may also play a role. That is, one would expect a windstorm following an earthquake to be a more serious situation than the reverse, especially if the affected population are residing in temporary shelters such as tents, where they will be more exposed to the dangers posed by the meteorological events. Considering exposure, damage induced by one event may then lead to other parts being exposed to other hazards, such as an earthquake damaging flood protection dykes then leads to areas previously protected by these defences to be more exposed to flooding. There is also the situation that with ever increasing levels of urbanisation, society is more dependent upon sophisticated, and vulnerable, infrastructure and lifelines, while rural depopulation is leading to the breakdown of traditional landform management practices in some areas, further increasing the risk of certain disasters, such as landslides and wildfires (Shakesby, 2011).
When calculating risk, there is the current situation where usually only direct economic and human losses are found, generally ignoring indirect (e.g. reduced industrial production, increased transport costs) or intangible (e.g. loss of reputation, diminished potential for social development, loss of cultural or environmental heritage) losses. For example, as shown by examples considered within this project, even a single hazard (in this case drought in Iran, Keshavarz et al., 2013, Yazdanpanah et al., 2013) may lead to far reaching social consequences such as increased gender imbalance, distrust between sections of the community and increased prevalence of the “poverty trap” (see below).
All of these issues together (and individually) may lead to the underestimation of the total losses that can be expected from natural hazardous events in a given area, and the extent to which such events have an effect. Therefore, what is necessary is the development and implementation of a multi-hazard and risk framework that considers all of these points, and others, leading to a more comprehensive estimation of risk, which, in turn, will allow more effective mitigation strategies. A critical part of this also involves enhanced communication between the various actors involved in disaster mitigation, especially the need to bridge the gap between practitioners and researchers.
It was therefore within the context of such concerns that the European Commission, within its 7th Framework Programme, supported the New Multi-HAzard and MulTi-RIsK Assessment MethodS for Europe or MATRIX project (1st October 2010 to 31st December 2013) to develop methodologies and frameworks for multi-type hazard and risk assessment to confront the inadequacies currently existing in single-type assessments.
The goals of MATRIX may be summarised as follows:
Assess the state-of-the-art in single- and multi-type hazard and risk assessment methodologies and tools so that a solid foundation may be built upon.
Develop a multi-hazard and risk assessment framework that considers the various interactions and dependencies that may arise within such a multi-type environment.
Examine under what conditions multi-type risk assessments provide better (or not) results compared with single-type assessments.
Provide tools to support the decision-making processes required by civil protection and disaster management authorities on the basis of probabilistic information.
Assess existing governance frameworks and to investigate individual and institutional barriers to the implementation of multi-hazard and risk approaches.
Disseminate multi-hazard and risk concepts to potential end-users and relevant members of the broader community, e.g. civil protection and research establishments.
The hazards that were the primary concerns of MATRIX are those most likely to afflict Europe, as mentioned at the beginning of this section. However, it is also hoped that the results of MATRIX will be applicable to other environments and contexts, not only in terms of the natural environment, but also the specific social and political contexts.
The multi-disciplinary nature of MATRIX called for expertise not only in terms of the wide range of hazards considered (e.g. geology and geophysics, meteorology, forest ecology), but also how the resulting risks should be considered in terms of the metrics used, and the identification of any social or institutional issues that may hinder the following of a multi-type hazard and risk paradigm (e.g. behavioural science, structural engineering, economics) and the manner in which the developed methods and tools are produced and distributed to the relevant parties (information technology, disaster management and decision-making). This therefore called for a very heterogeneous consortium, made up of 12 partners, including ten research institutions (nine European and one Canadian), an end-user (the German National Contact Point for DRR) and a partner from private industry. 1 lists the consortium’s members, as well as their respective responsibilities in terms of the different components of work packages of the project, which will be outlined below.
To describe better the project’s objectives, it is perhaps useful to outline its work plan. This is shown in 2, which presents the project’s work package (WP) structure and their interactions. There were 8 work packages within MATRIX, including WP1 which is responsible for the management of the project. The more technical work packages follow what could be considered as a natural order for the work needed to be undertaken.
WP2 “Single-type risk assessment and comparability” serves as something of a starting point, given that its aim is the development of the means to allow the comparison of different single-type risks. Although MATRIX is strongly concerned with hazard and risk interactions, this does not mean that single-type assessments are ignored. In fact, they still play a critical role, as the first stage of any multi-hazard and risk assessment scheme or framework is to understand what are the hazards of concern, what are natures, what will their impact upon the human and natural environments be, and what information is required by end-users and stakeholders for their mitigation. Therefore, the principle goals of this work package include the harmonization of the methodologies used for the different types of natural hazard assessment, which means considering the different return periods and spatial scales, in order to compare risk estimates. It will also involve classifying and identifying the different sources of uncertainty that arise in risk estimates, in particular, the classification of these sources in terms of their aleatory and/or epistemic character (in collaboration with WP5, see below).
The issues associated with cascade (domino) effects within a multi-hazard environment were the subject of WP3 “Cascade effects in a multi-hazard approach”. It may be said that this work package dealt with the most obvious consequence of a multi-hazard environment. Due to the complexity of the interactions that arise at all levels of the risk chain, a multi-hazard and risk approach is potentially much more complex than the simple sum of the individual single-risk results. The objectives of WP3 therefore included the identification of the relevant temporal and spatial scales for multi-hazard assessment, the development of strategies to homogenize different types of hazard assessments (in collaboration with WP2), the identification of the plausible cascade events with respect to the MATRIX test cases, and the development of a probabilistic Bayesian and non-Bayesian framework to include cascade events within a multi-hazard assessment perspective.
Temporal-dependency in vulnerability was the topic of WP4 “Time-dependent vulnerability”, which considered this issue at the physical, systemic and socio-economic levels. The development of time-dependent vulnerability assessment procedures is an essential aspect of an integrated multi-risk approach, since it is safe to say that no environment is static. By this we mean that each hazardous event and/or possible mitigation action is likely to impact upon the vulnerability of the community, society, region, etc., to other future events. This work package therefore aimed at developing a framework that considers the history of a given element (which includes conjoint multiple disasters as well as successive ones) as an internal parameter of its vulnerability. The examples considered in this WP focused on how a given structure reacts to successive seismic events, and how an area’s systemic efficiency suffers as a result of not only the direct damage to infrastructure, such as power and water supplies and hospitals, but also in transport links.
WP5 “Framework for multi-type risk assessment” is the work package that set out to combine the efforts undertaken in work packages 2 to 4 in order to develop a consistent framework for multi-risk evaluation across different temporal and spatial scales. This included, as mentioned, the identification of the sources and types of uncertainties (following a similar scheme as done in WP2), demonstrate the application of a multi-risk assessment framework for different categories of tangible and intangible losses, develop fault trees and event trees for the assessment of environmental risks and the development of a framework within which end-users and stakeholders may place their multi-type hazard and risk assessments in order to know how much effort will need to be expended to obtain risk estimates appropriate for the resources at hand.
The work package that was concerned with risk governance and the implementation of multi-risk approaches was WP6 “Decision support for mitigation and adaption in a multi-hazard environment”. Its aim was to identify if cognitive and behavioural biases affected perceptions of probabilities for multiple risks and influenced choices for risk mitigation strategies. It involved an analysis of historical multi-risk disasters where not considering a multi-type hazard and risk environment lead to more severe consequences and increased vulnerability, as well as developing a Bayesian framework that identified most common behavioural biases, such as risk aversion, limited worry and loss aversion. The work package also endeavoured to identify existing social and institutional barriers to the implementation of a multi-risk approach (also dependent on features of governance systems such as centralized and decentralized decision-making) and proposed options for overcoming them. All of these efforts had the aim of identifying the benefits of multi-hazard and risk assessments compared to single-type assessment, and to provide recommendations as to how policies conducive to such an environment may be formulated.
The methodologies developed within the MATRIX project needed to be tested within real world settings to ensure their lasting legacy. This includes the development of a generic IT tool that allows practitioners to “experiment” within a controlled environment how the consideration, or neglect, of hazard and risk interactions affects the final results. This was dealt with by WP7 “’Virtual City’ and test cases”. WP7 saw the development of the MATRIX-Common IT sYstem (MATRIX-CITY), a multi-type risk assessment IT platform that performs and visualizes multi-hazard and multi-risk analyses for test cases. It also involved the so-called “Virtual City”, as its name suggests a virtual environment where different hazards and their risks may be analysed in terms of their effect on a predefined “virtual” community. The development of the system required the implementation of common data models, data exchange procedures and hazard and risk calculators, including time-dependent vulnerability. The test cases considered during the MATRIX project were Naples, Italy, Cologne, Germany, and the French West Indies. Each test case is exposed to a range of hazards and risks, with differing degrees of interaction.
The final work package, WP8 “Dissemination/end users” dealt with making the project’s results and aims known to as wide an audience as possible, and to interact with the end-user communities to determine their specific needs within the context of a multi-hazard and risk environment. The dissemination of MATRIX concepts and results was focused towards the scientific community at large, the disaster management community in European states, the commercial sector with an emphasis on insurance companies and companies in the information technology sector, and the press and media. The key objectives of this work package was therefore the development and implementation of a promotional and dissemination process, information products and a web site for the output from the project, and the involvement of stakeholders and user communities. The latter point proved to be most constructive and saw a great deal of interaction with WP6.
Project Results:
While the main scientific and technological results of the MATRIX project could be outlined in terms of their respective work packages, it needs to be remembered that in a project as complex as MATRIX, many work package often rely on the results of others. For example, WP5, dealing with the multi-risk assessment framework, calls upon results dealing with single-type assessment (WP2) and cascading events (WP3). Likewise, the test case studies (by definition) required the results from all of the work packages. Therefore, in the following, while we will generally proceed along the same path as suggested by the WP structure, the interactions and dependencies between them and the resulting outcomes will be emphasised. Naturally not all of results from a project as diverse as MATRIX can be presented in a report such as this, hence the focus (although not exclusively) will be on those presented in the MATRIX results and reference reports1.
Single-type hazard and risk assessment
Although the methodologies and issues surrounding a multi-hazard and risk framework were the foci of the MATRIX project, this by no means neglects the importance of single-type risk assessment. In fact, the framework defined by WP5 (see below) allows for cases where a single-type process can be employed where appropriate (i.e. cases where a multi-type process may either be not required or hindered by a lack of information and resources. Therefore, considerable effort went into reviewing the state of the art in single-type assessments, understanding the associated uncertainties (discussed below) and the harmonisation of risk results for different hazards.
From the review of the literature, some of the more prominent observations about the current state-of-the-art in single-type assessment methodologies (Parolai et al., 2014) are as follows:
For all hazards and risks examined, even when considering only the single-type framework, considerable further research is required. This mainly centres on the lack of understanding of the physical processes involved with each hazard. However, there appears to be an emphasis in the literature on the scarcity of data, limiting the ability to infer adequate time series. While such problems are present in all research, there are fairly significant disparities between the various components that make up risk assessment, as discussed below.
There appears to be a significant gap between the relatively high level of methodological development for hazard assessment and a lower level for analysing the vulnerability of the elements at risk, i.e. there seems to be a lack of vulnerability functions, except for the case of seismic hazards.
The focus in the literature seems to be on state-of-the-art research aspects of risk analysis, and not on risk management.
For the harmonisation of the different risks, two approaches are proposed within the literature:
Risk Curves, which requires the consideration of temporal constraints, spatial scale and resolution, and loss metrics.
Development of a common framework where harmonisation is carried out at each level, from hazard to risk.
During the time of the project, the first approach was the focus. The efforts made during MATRIX dealing with this issue considered two issues:
Combining individual but independent loss curves so as to gain an idea of the total loss an area may expect to experience.
How to compare two independent risks, that is how to determine if, for a given return period, they are the same and hence should be treated with the same level of concern or can some hierarchy be established.
For the first point, a simple formula was employed:
Ptot = 1 - ∏ (1 – Pi) (1)
where Ptot is the total annual probability of exceedance of a given risk (expressed as Euros), and Pi is the probability of exceedance of a given risk i (i.e. here represented by earthquakes, landslides and floods). An example of this is presented in 3a, where we show the individual risk curves from the work of Grünthal et al. (2006) for the case of Cologne (one of the MATRIX test cases). Considering as an illustrative example the situations where losses of 100 million Euros may arise, the combination of curves sees a significant increase in the probability of this level of loss, from 15 to 35% in 50 years for the individual hazards, to around 75% in 50 years when combined. Such results can be shown using an alternate format, the risk matrix (3b), which illustrates how different levels of risk arise when considering their likelihood and impact (e.g. Cox, 2008). Risk matrices are often used by civil protection authorities (e.g. BBK, 2011) and in fact, one of the statements to MATRIX members during the presentation of the MATRIX-CITY and Virtual-City support tools (see below) was to use such a format instead of risk curves. We see from Figure 3b how the combining of these risks leads to the final value being to the right (more likely), allowing decision makers a ready means of better understanding the total risk/likelihood of a level of loss.
With respect to the second point, that is comparing the estimated losses for a given return period of two (or more) different hazards, this was examined by employing the Wilcoxon’s test, a distribution free ranking test that asks the specific question “Are the medians of the two distributions the same?” (Barlow, 1989). A new series of loss estimates for the case of Cologne were produced for the hazards of floods, earthquakes and windstorms (Fleming et al., 2014). We then compared these ranges of values for each pair of hazards (earthquake – flood, earthquake – windstorm, flood – windstorm) and apply a null hypothesis (to 0.05) that the question’s answer is in the affirmative. The results of these tests are shown in 4. Considering first the earthquake distribution, its bimodal character (a product largely of the choice of ground motion predictive equations) immediately adds an additional element of uncertainty as to whether the risks it is compared to are equivalent. We note for the 200 year return period (4a) that earthquakes and floods are not equivalent, but can be considered comparable for 500 years. For the windstorms and floods (4c-d), for both the 200 (4c) and 500 (4d) years return periods, it is obvious (even without applying this test) that windstorms and floods are not equivalent, with floods being of greater concern in both cases. Finally, for earthquakes and windstorms (4e-f), for the 200 year return period (4e), these appear to be of equivalent importance, while for 500 years (4f), this does not appear to be the case (with earthquakes being of greater importance).
The relevance of such an exercise is to do with the decision making process, whereby if the risk associated with two types of hazard is “equivalent”, then the required mitigation schemes may need to consider both, or at least help decision makers when deciding on how to allocate resources. It also shows that one needs to accommodate uncertainties, since simply using, for example, average curves, may yield misleading conclusions about the relative importance of a given combination of hazard types (e.g. the earthquake results). However, it is also important to note that the actual results would vary as the range of employed input models and parameters are updated and refined (as would be apparent in the earthquake case). Similarly, the uncertainty associated with the risk arising from a given hazard may be very wide, leading to the potential range of risk levels to cover from being of little concern to being of great importance.
Cascading events
One of the most obvious aspects of a multi-hazard and risk perspective is the need to accommodate cascade events. This involves the identification of the various interactions that may lead to such occurrences and to the increased vulnerability and amplification of damage in an area. In particular, two possible kinds of interactions, namely: (1) interactions at the hazard level, where a given initial ‘triggering’ event leads to a modification of the probability of the occurrence of a secondary event, and (2) interaction at the vulnerability (or damage) level, in which the main interest is to assess the effects that the occurrence of the first event occurring in time may have on the response of the exposed elements against additional events, which may or may not be of the same kind. Implicitly, a combination of both kinds of interactions is also possible.
The first step in assessing cascading effects is the identification of possible scenarios, or rather, scenarios that are plausible. To obtain a complete set of scenarios for a given area, different strategies can be adopted, ranging from event-tree to fault-tree strategies. In many applications, an adaptive method combining both kinds of approaches is applied in order to ensure the exhaustive exploration of scenarios. From the multi-risk assessment point of view, the cascading effects scenarios of primary interest are those that produce an amplified total risk when compared to the effects produced by the individual events (a case of 1 + 1 > 2). With an appropriate set of cascading scenarios, their quantification can be achieved by adopting different strategies, for example, analysing databases of past events, performing physical modelling for the propagation of the intensity measures of interest, and/or by performing expert elicitations in order to obtain information for extremely complex problems, or in these cases with poor data or needing rapid analysis.
To define some possible cascade scenarios, the ‘primary’ interactions between hazards need to be identified. These can be understood as the pairs of hazards where it is theoretically possible to define an event that has the capacity to directly trigger another (interaction at the hazard level), or in which the additive effects of the loads may lead to a risk amplification. In the matrix-like 2, the different hazards considered in the MATRIX project are classified as triggering (running in the x-axis) against the ‘triggered’ (running in the y-axis) events. In this case, all the possible ‘direct’ triggering effects are considered. It would also be obvious that it is physically impossible for some hazards to trigger another, e.g. wildfires and volcanoes (although the other way around is certainly a concern, especially for Naples). Note also from 2 that interactions at the vulnerability level may also be observed. Such a table can be expanded to cases where additional ‘levels’ of interactions can be inferred, simply by adapting an appropriate numbering scheme (e.g. a flood may be labelled 2, if a river is blocked by a landslide (labelled 1) caused by an earthquake (which would be labelled 0).
5 illustrates how one can identify the possible cascading scenarios that may arise for the MATRIX test cases. It therefore allows us to understand better the existing relationships between the different kinds of events and their relative level in the chain. In this way, the occurrence of different phenomena may be considered from the possible triggering factors. Such a diagram would be useful for DRR practitioners to at least visualise to themselves potential scenarios of cascading events. In fact, based on this analysis, examples of cascading scenarios were selected for more detailed analysis as part of the test case studies, namely:
Naples: The cases that received most attention were volcanic earthquakes and seismic swarms triggered by volcanic activity, and the simultaneous occurrence of ash-fall and earthquake hazards.
Guadeloupe: The scenario here was concerned with earthquake-triggered landslides after a cyclonic event or a heavy rainfall period.
Cologne: This scenario involved earthquake-triggered embankment/flood defence dyke failures and the subsequent inundation of the City of Cologne.
A further point that needs to be made concerning cascading events is that while they are usually considered at the hazard level, as alluded to in 2, they also occur at the vulnerability level, and not only the physical. Taking again the example mentioned above of the Iranian drought, one can easily consider the consequences of inappropriate mitigation actions, e.g. the limiting of support loans to the wealthier farmers, who could then improve their infrastructure, hence now needing fewer workers and hence deriving poorer farmers of a source of income, exasperating their problems. This leads to health (mental and physical) problems, mistrust developing within the community, the frequent need to withdraw children from education (usually the girls) leading to enhanced gender inequality and the so-called poverty trap (Hochrainer, 2013).
Uncertainties within a multi-hazard and risk framework
Uncertainties in risk estimates, while widely acknowledged as being critical, are rarely considered. In fact, taking the otherwise comprehensive review of Kappes et al. (2012) as an example, there is no mention at all of uncertainties. Therefore, WP2 and WP5 set out to identify and classify the different types and sources of uncertainty relevant to hazard and risk assessment, with efforts to identify those that were particular to multi-type environments.
The focus in WP2 (Rohmer, 2012) was first to devise a qualitative analysis for each step of the state of the art in single-type risk assessment. This saw uncertainties classified between aleatory (i.e. the general randomness of a system) and epistemic (inadequate knowledge about the processes involved, lack of data, and poorly defined input models and parameters). The latter was further divided into four additional types:
epistemic – data: This refers to the difficulties associated with measuring the properties of the system of interest or concern, e.g. issues with the instrumentation, methods of processing, length of time for data acquisition, and so forth.
epistemic – models: Models are by definition simplifications of a phenomena, leading to uncertainties arising from (a) uncertainty in the form/structure of the model, that is, what variables, processes, etc. are considered relevant and prominent, and (b) the actual choice of the “best” model, that is, the one that reproduces most closely the processes of interest. However, a problem may arise when more than one model gives equally good results.
epistemic – parameters: This source arises from the difficulties in estimating the required input parameters for models or an analysis due to the limited size, poor representativeness (caused by time, space and financial limitations), and imprecision of the available data.
epistemic – science: This class reflects our state of knowledge (or more importantly, ignorance) of the processes under investigation.
The qualitative analysis of uncertainties should be viewed as a tentative mapping of the state of our knowledge for risk assessment and of the major challenges for quantitative uncertainty treatment. More specifically, it highlights, at all stages of risk assessments, not only the prevailing presence of uncertainties of type “epistemic – parameter” (related to the difficulties in estimating the values of model parameters), but also of type “epistemic – model” (related to the uncertainty in the structure/form of the model and to the unambiguous choice of the “best” model to be used). It also appears that most work dealing with uncertainties was concerned with the hazard level rather than the risk level. The same taxonomy of uncertainties was also employed in WP5 (Vidar Vangelsten, 2012), where the most suitable methodologies for the quantitative treatment of uncertainty in multi-risk assessments were reviewed.
Time-dependency
For a comprehensive multi-hazard and risk framework, the temporal dimension in such interactions would need to be taken into account. By this we mean interactions at the hazard and vulnerability levels, which might occur with different delays. Such temporal dependencies might involve the following situations:
The repetition of events over time (these may be the same, e.g. aftershocks following an earthquake, or different events).
The concomitance of simultaneous-yet-independent events. By simultaneous or conjoint, one may also consider successive events occurring before the impacted upon area has adequately recovered from the initial event.
The succession of dependent phenomena (cascading events).
The fact that the future vulnerability of the risk bearer will depend upon the realization of risk in the past.
For the first of these, the effects of the repeat of a specific type of event were considered by examining a seismic example. The effects of fatigue due to the repetition of seismic shocks within a physical vulnerability assessment have been analysed through two mechanical methodologies. The first approach, proposed by BRGM (Réveillère, 2012), developed damage-state dependent fragility functions (Figure 6), while the second approach, performed by AMRA (Iervolino et al., 2014), analysed the multiple shock capacity reduction for non-evolutionary structural system (Figure 7).
The next two situations listed above were considered within the test case of Guadeloupe and will be mentioned below. In brief, the issues involved two hazards, an earthquake and an earthquake-triggered landslide. While these specific examples indicate a cascading scenario, the case study included the occurrence of the earthquake following a period of intense rain, hence a situation where one event (rain) increased the likelihood of another (landslide) in case of a third (earthquake). One obviously sees the cascading element here, i.e. earthquake/landslide, however, it was the rainfall aspect that led to this cascading scenario to be more likely.
Time-dependent vulnerability was also analysed within a study involving wildfires in Portugal using advanced statistical methods and databases to determine which factors were most important within a multi-risk framework. It was found that in disaster situations, financing and coping strategies are interrelated over time, with higher risks potentially emerging and possibly overriding the available resources. As in the Portugal case, communities as well as regional authorities play an important role in diverting help and giving assistance. The macro level analysis additionally showed that repeated events in the past will have consequences in the future, especially for less-wealthy countries. It was therefore proposed to use a systems perspective to capture these relationships and to use a process-based approach to tackle the issue of time- as well as scale-dependency.
Similar to the comment above about cascading events within the societal aspects of risks, cascading issues have also be considered within a systemic vulnerability framework. This involved the application of the I2Sim platform (see below), a tool for simulating interdependencies between infrastructures and lifelines in response to loading such as that arising from natural hazards (Marti et al., 2008). The idea was to examine how the multi-hazard and risk framework will help to more realistically estimate the impact upon issues such as the capacity for the treatment of injured.
Multi-risk frameworks and decision support tools
Multi-risk framework
As part of the MATRIX project, frameworks and tools for use within a multi-hazard and risk environment were to be developed. This may be divided into two aspects. The first is the framework which sets a context within which to conduct such assessments. While developing a fully rigorous mathematically approach that addresses all challenges and uncertainties at all steps of the analysis would be extremely complicated and resource and expertise intensive, there are many situations where the decision-maker can identify the optimum alternative among the possible options without undertaking a detailed, rigorous multi-risk analysis. Therefore, a framework was developed (Nadim and Liu, 2013, Liu and Nadim, 2013) based on a multi-level approach where there is no need to undertake a more sophisticated assessment than is necessary given the problem at hand, or what would be reasonable with the available information and resources.
The developed framework consists of the following stages (see Figure 8):
Level 0: It is assumed that the end-user has identified the relevant threats and has carried out an assessment of the risk(s) at the highest level of sophistication required for the problem at hand.
Level 1: This involves a qualitative assessment of the relevant single hazards.
Level 2: Here requires a qualitative multi-risk analysis.
Level 3: The final and most comprehensive stage, a quantitative multi-risk analysis.
A critical point as illustrated in Figure 8 in the specific inclusion of Monitoring and Reviewing and Communicating and Consulting. As will be discussed later, interactions with end-users and stakeholders are crucial to understand what is actually required and expected with regards to the input to DRR activities from researcher.
Considering first the Level 1 analysis, illustrated in 9, the process involves a flow chart type list of questions that helps the end-user to decide whether or not a multi-type assessment approach is required. These questions explicitly account for cascading hazards and dynamic vulnerability within the context of conjoint or successive hazards. If the Level 1 results strongly suggest that a multi-type assessment is required, then the end-user moves on to Level 2 to make an initial assessment of the effects of dynamic hazard and time-dependent vulnerability. If cascading events are potentially a concern, then user goes directly to the Level 3 analysis.
In the Level 2 analysis, the interactions among hazards and dynamic vulnerability are assessed approximately using semi-quantitative methods (Figure 10a). To consider hazard interactions and time-dependent vulnerability, use is made of a matrix approach based on system theory, as shown by Figure 10b-f (Modified after de Simeoni et al., 1999, Kappes et al., 2010). First, a matrix is developed by defining the hazards of concern along the diagonal (Figure 10b). Considering the example in Figure 10c, the interactions are defined by a clockwise scheme of interactions, the nature of which are described in Figure 10d. These descriptions are assigned numerical codes varying between 0 (No interaction) and 3 (Strong interaction) with intervals of 1, as a function of their degree of the interaction intensity (Figure 10e,f). From this, the degree of the impact of each hazard on the others and the effect from other hazards can be resolved. We employ a hazard interaction index HI, which is the sum of the codes for all the off-diagonal terms. The resulting index value is then compared with some threshold (e.g. HI,max= 50%) and then the decision made as to whether to go to a Level 3 analysis. For example, ff the hazard interaction index is less than this threshold, a Level 3 analysis may not be recommended because the additional accuracy gained by the detailed analyses is most likely within the uncertainty bounds of the simplified multi-risk estimates. Otherwise, if the hazard interaction index is greater than the threshold value, a detailed Level 3 analysis is recommended.
In the Level 3 analysis, the interactions among hazards and dynamic vulnerability are assessed quantitatively with as high accuracy as the available data allow. One of the methodologies developed within the MATRIX project involved the use of Bayesian networks (Nadim and Liu, 2013). This scheme allows both the estimation of the probability of a triggering/cascade effect and the modelling of the time-dependent vulnerability of a system exposed to multi-hazard. An example of a Bayesian network multi-risk model is shown in Figure 11. Such networks would need to take into account events that are:
Independent, but threatening the same elements at risk, with or without chronological coincidence (the column marked in orange in Figure 11), or
Dependent on one another or caused by the same triggering event or hazard; this is mainly the case for cascading or domino events (i.e. the column marked in green in Figure 11).
The MATRIX-CITY and Virtual-City
The above discussed a framework within which a decision maker or analyst can decide upon what level of assessment they need to do towards. In the following, we will present an IT tool, the MATRIX Common IT sYstem (MATRIX-CITY, see Mignan, 2012, 2013 and Mignan et al., 2014). MATRIX-CITY is a first step towards a more general use of multi-risk tools in decision-making, and encompasses 3 major advances in the implementation of a multi-risk framework:
1. The development of a generic probabilistic framework based on the sequential Monte Carlo method to implement coinciding events and triggered chains of events, as well as time-dependent vulnerability and exposure (Mignan et al., 2014).
2. The proposition of guidelines for the implementation of multi-risk, using the concept of the “Virtual City” to test basic multi-risk concepts in a controlled, yet realistic, environment (Mignan et al., 2014).
3. A better understanding of the IT requirements for the widespread use of multi-risk tools, based on the lessons learned from the development of the MATRIX-CITY platform prototype and interactions with stakeholders.
A sequential Monte Carlo method was proposed to generate a large number of risk scenarios (i.e. the generation of hazardous events and the computation of associated losses, while considering the impact upon vulnerability and exposure, e.g. 12). The analysis of these simulated risk scenarios then allowed us to assess losses in a probabilistic way and to recognize more or less probable risk paths, including extremes or low-probability high-consequences chains of events. We found that “black swans”, referring to unpredictable outliers, can only be captured by adding more knowledge about potential interaction processes to the computation process. However, this can only be achieved over time by following a “brick-by-brick” approach, given the considerable effort required.
To quantify hazard interactions, we introduced the concept of the hazard correlation matrix (13a). We considered three categories of interactions: event repeat (e.g. Ai Ai; C C), intra-hazard interaction (e.g. Ai Aj) and inter-hazard interaction (e.g. Ai Bj). The effect could be positive (i.e. probability increase) or negative (i.e. probability decrease), and temporary or long lasting. Time-dependent vulnerability and exposure are not described here, but are taken into account within the framework at a later stage of the calculations. To evaluate how multi-risk participates in the emergence of extremes, we introduced the concept of the risk migration matrix (a more detailed version of the risk matrix discussed above) and showed that risk migration and risk amplification are the two main causes for the occurrence of extremes (13b).
The multi-risk framework was developed and tested based on generic data and processes generated following the heuristic method. This strategy, which involves the use of intuitive judgment and simple rules, allows for the solving of problems that are otherwise difficult to consider. Our approach follows the existing recommendations on extreme event assessment, which involves the use of inductive generalizations and "scientific imagination" to include known examples of extremes, as well as potential "surprise" events within the same framework. However, abstract concepts, such as the definition of generic perils, remain difficult to comprehend and we therefore proposed some guidelines to help risk modellers and decision-makers apply this approach to realistic cases.
For this purpose, we developed the concept of the Virtual City (Mignan, 2013, 14). Within this concept or tool, the perils A, B, C, D and E are no longer simply abstract concepts, but are replaced, for instance, by earthquakes, volcanic eruptions, tsunamis, fluvial floods and storms. Hazard, exposure and vulnerability data, as well as details about possible interacting processes, are based on real examples obtained from the scientific literature.
Although these tools were based on state-of-the-art software engineering and a Python-based code, it was rapidly observed that multi-risk software would need to have all the functionalities of existing risk tools, on top of the multi-risk framework. Such a task would require significant resources and a commitment of modellers used to other types of risk modelling tools (including various procedures and formats). Therefore, at this stage, we recommend the exporting of the method developed for this IT tool to existing risk tools, which would facilitate its implementation and potentially encourage the widespread use of the proposed approach. Hence, the present work should be seen as a proof-of-concept, as we did not intend to fully resolve the complex problem of low probability-high consequence events. We only considered a selected number of possible interactions, where naturally adding more perils and interactions would yield more complex risk patterns. We thus recommend a brick-by-brick approach to the modelling of multi-risk, to progressively reduce epistemic uncertainties. A more realistic modelling of low-probability high-consequences events would also require the consideration of additional aspects, such as uncertainties, domino effects in socio-economic networks and long-term processes such as climate change, aging infrastructure and exposure changes. While the concepts developed in the present study outline the theoretical benefits of multi-risk assessment, identifying their real-world practicality will require the application of the proposed framework to real test sites.
Decision support tool
Another tool for decision support and presented to end-users also involved using the risk matrix as described above, that is it followed the definition of risk as a combination of the consequences of an event or hazard and the associated likelihood of its occurrence, again adapting the BBK (2010) framework (see the review in Wenzel, 2012). The risk matrix methodology was implemented into decision-support software based on the principles of Multi-Criteria Decision Analysis (MCDA), and tested with a group of stakeholders to communicate and transfer the information contained for the different risk scenarios in the risk matrix to the various stakeholders involved. The decision-support tool allows the stakeholders to display the total risk index ranking of different risk scenarios and to attribute different weights to the loss components (e.g. an extremely rare offshore earthquake which can trigger a tsunami, or a release of toxic material with severe impacts on the local environment, etc.) affecting a region in terms of expected losses that are quantitatively derived in different sectors (human, environment, economy, infrastructure, intangibles) for each scenario. An example of such a series of scenarios and their consideration within the MATRIX decision support tool (employing a risk matrix) is presented in 15. Following this approach, the sectorial losses are combined as a weighted sum into a single aggregated loss score for each scenario, as shown in Figure 16. Together, these two steps (i.e. severity and loss scores) are combined to produce a total risk index for each scenario.
For example, in Figure 16a, it can be seen that the offshore earthquake triggering a tsunami is deemed to have a much greater risk score than the toxic spill. As the total risk index for each scenario is determined as the aggregate weighted sum of each of the loss measures in each of the different sectors, the risk index ranking will also depend on the weights given to each sector. Through a participatory approach, the stakeholders assign the relative importance (weights) to the losses for the different sectors for each of the scenarios likely to occur in the region. Next, the decision support software is used in a group setting to discuss the weighting outcomes and interactively examine the variability of the ranking results. For example, a sensitivity graph can be used to see the effect on the rankings as the weights are changed. In Figure 16b it can be seen that as more weight is given to the “People” criteria (i.e. casualties, short- and long-term mass care), the risk score for the toxic spill decreases considerably. This is due to the fact that the toxic spill scenario produces none to very few fatalities and has a minor impact on mass care. As a result, when all the weight is given to only one measure, in this case human losses, the risk score for this scenario is minimal. On the other hand, the risk score of all other scenarios goes up, but importantly the relative rankings between them stays the same. Using various visualization tools in the decision support software, such as sensitivity graphs, stacked bars, scatter plots, and one by one comparisons between scenarios, the stakeholders are able to evaluate the total risk from different scenarios by considering many variables at once, which enables them to separate facts from value judgments, and better communicate their choice to others.
Tangible and intangible loss treatment
Usually within risk assessment, most attention is paid towards direct losses, referring to the physical components of society. However the real situation is much more complicated. Losses can be tangible or intangible, as well as being characterized as being immediate/direct (during the event), and being short- and long-term (indirect) losses. Some types of risks are relatively straightforward to estimate, while others can only be estimated approximately. In addition, data limitations must often be taken into account. A further complication, and an important analytical step, is the presence of risk aversion and extremely large risks. It is important to combine and compare risks from low probability, high consequence events with risks from more frequent, but low to medium consequence events.
These issues were examined within MATRIX by exploiting a risk analysis chain perspective (Pflug and Römisch, 2007), that is a framework involving the measurement, modelling, and management of risk. It was found, perhaps unsurprisingly, that different risk measures and modelling approaches are required for each of the components of the framework just outlined. Furthermore, for the case of extreme risk, traditional risk management strategies, successfully applied for frequent risks, are likely to fail. However, while there are fundamental differences between frequent and extreme risks in various dimensions, from a risk management point of view, they can be assessed together, given the right measures and decision support approaches (Hochrainer, 2011).
Regarding the tangible and intangible loss assessment component, it is argued that from a methodological perspective, it is beneficial to distinguish between dependent risks, i.e. risks which, if realized, are changing the likelihood of other risks, and independent ones. Additionally, based on a reflexive modernity point of view, e.g. seeing the decision making as a participatory rather than hierarchical process, different risk bearers exposed to the same risks, but over different scales, have to be considered separately. This is important as one cannot assume that the weights (or importance) proposed for the different components of risk stay constant over different scales (e.g. household to national), especially when the intangible contributions are considered. A multi-criteria analysis approach was therefore suggested for risk bearers on different levels to sort out some of the most relevant dimensions which need to be considered in such a setting (Hochrainer, 2011).
The MATRIX test sites
In order to verify the concepts and tools developed within MATRIX, it is necessary to apply them to real world situations where conjoint and cascading events and interactions between different hazards and risks need to be considered. It is for this reason, and matching the expertise of the consortium, that the MATRIX test cases were chosen. All three are under threat from multiple hazards. Naples (Garcia-Aristizabal et al., 2013a) and Guadeloupe (Monfort and Lecacheux, 2014) are the most threatened (and complex) examples, with both endangered by volcanic eruptions, earthquakes, as well as hurricanes (Guadeloupe), landslides (Naples and Guadeloupe) and forest fires (Naples). Each case is also susceptible to cascading events, in particular rain- and earthquake-induced landslides and volcano-earthquake interactions. Cologne (Fleming et al., 2014) on the other hand is not as exposed to such a range of hazards, nonetheless it must still contend with threats from earthquakes, floods and windstorms (Grünthal et al., 2006), with the possibility of earthquake-induced damage to its dyke system increasing the flood risk to the city.
During the project, the work on each test case tended towards a particular aspect of the multi-hazard and risk problem. For example, Cologne became the focus of efforts towards the comparison and harmonisation of single-type hazards and risks while considering uncertainties (3 and 4). However, another problem confronted involved the possible consequences of seismic loading on flood defences, leading to a potential increase in risk. For Naples, the focus tended towards cascade effects, of which two scenarios were examined in detail (see below). Finally for the French West Indies, a systemic assessment of the consequences of a multi-hazard and risk scenario on Guadeloupe’s infrastructure was made.
In addition, a preliminary effort was made to employ the multi-risk framework described above. Considering first Level 1, we present the answers to the raised questions for each test case in Table 3, we note immediately that for each example, we must proceed from the initial “More than one hazard?” question to dealing with the various interactions, with the need for at least a Level 2 analysis. However, even if this were not the case, i.e. only one hazard of concern, there is also the possibility of events of the same kind reoccurring during a given time period, which may be taken as the time required to carry out the necessary repairs/recovery from the original event (e.g. a series of storms separated by short periods of time). We also note that for all three cases, we would probably need to proceed to a quantitative Level 3 analysis, based on the fact that cascade events may arise. However, the fact that cascade events in Naples and Guadeloupe are more likely than in Cologne cannot, at this stage of an assessment (or comparison), be resolved. In addition, the cascade example for Cologne, i.e. an earthquake damaging flood defences, hence increasing flood risk, would also fit within the context of conjoint events. Therefore, it would appear that even the most “quiet” territories may be exposed to several hazards, with interactions potentially always present (for example, Na-Tech - Natural Technological - interactions are in many industrialised districts a major concern, although they are not dealt with in detail in MATRIX). Hence, one may expect the situation where only a Level 1 assessment is required would be fairly rare.
Considering the Level 2 assessment, the aim is to describe the various relationships between the assorted hazards, following the matrix method outlined in 10 and presented in 17. To reiterate, to read these figures, consider first that along the diagonal, the hazards of concern are listed. Then, moving in a clockwise manner, the level of interaction (scored between 0 and 3 with intervals of 1, where 3 indicates a strong interaction and 0 indicates none) and the nature of such interactions between each hazard pair are identified.
For Naples and Guadeloupe (17a and 17b), for the purpose of this work, we simply refer to three hazards, although obviously a larger matrix would be needed for a thorough study. We note the strong (3) interactions between some hazards, e.g. earthquakes and volcanoes for Naples, landslides and earthquakes for Guadeloupe, as well as hazards where no interaction would arise (e.g. hurricanes and earthquakes). Considering Cologne (17c), we identify few interactions between hazards, i.e. windstorms potentially bringing heavy rain, although for Cologne, more localised heavy precipitation causes little widespread flooding, and an earthquake damaging flood defences. However, it is also recognized that if we considered this at the risk level, then a windstorm may damage a building, increasing its susceptibility to a later earthquake, while considering the reverse (an initial earthquake followed by a windstorm) would most likely be more serious. Based on the numbers presented in each square, a so-called hazard interaction index may be inferred (found by adding all results row by row, representing causes, then column by column, representing effects), and as mentioned above comparing to some criteria (e.g. a predefined percentage of the maximum possible index for a given site) may decide whether or not to proceed to the more resource intensive Level 3 analysis. For example, Naples has a score of 16 and Guadeloupe 12, while Cologne has a value of 4, indicating the much great importance of such interactions for the first two cases.
In the following, we will now consider the specific aspects of the test cases introduced above. However, we will not do so in detail, with the reader referred to the associated deliverables.
Naples
As mentioned above, two scenarios were the focus of the cascading studies for this test case (the results of which are presented in deliverables D3.3 Garcia-Aristizabal and Marzocchi, 2013, and D7.3 Garcia-Aristizabal et al., 2013a):
Scenario 1: Volcanic earthquakes and seismic swarms triggered by volcanic activity (interaction at hazard level).
Scenario 2: Simultaneous occurrence of ash-fall and earthquake hazards (interaction at the vulnerability level).
For the first scenario, during the evolution of a volcanic system from its quiescent to eruptive state, a large number of small- to moderate-sized earthquakes occur. In fact, volcanic unrests are normally accompanied by (volcanic) seismic activity located directly below the volcano and/or also by triggered seismic swarms (sometimes categorized as ‘distant volcano-tectonic events’. In this scenario, the problem is to assess the possible effect that local seismicity (normally characterized by many shallow micro-earthquakes) triggered during volcanic unrests may have in the area where the volcano is located, and compare it with the general seismic hazard analysis. In the Neapolitan area, this case may be of particular interest given the highly urbanized area in the surroundings of Campi Flegrei (see Figure 18). The interactions involved in this case are at both the hazard and vulnerability levels, and at time scales from seconds to possibly years, although the focus was on the hazard level. It was observed how the contribution of volcanic earthquakes may be significant in an area located in the Campi Flegrei caldera (around Pozzuoli municipality), whereas the effect quickly decreases with distance, being negligible for distances greater than a few kilometres (e.g. for most of the area of Naples).
The second scenario, earthquakes and eruption-related hazards, can result in the increased vulnerability of exposed elements. The objective is therefore to estimate the “Loss distribution”, and/or “Expected Loss” while considering the possibility of the simultaneous occurrence of two hazards. In our case, these are: (1) volcanic ash-fall hazard and (2) seismic hazard, in the Arenella area (see 18). The interactions here are more towards the vulnerability level, acting over seconds to days.
It was found (see 19) that for the combined effect of seismic and volcanic ash (deposited over the roofs), the loss distribution and expected loss considering different values of ash load (3 and 5 kPa of ash load over the roofs) for 3 kPa are very close to those obtained for the pure seismic case (i.e. with no ash), whereas a greater difference (i.e. increased risk) is observed for the case of 5 kPa. Taking in consideration the median of the estimated curves, for an ash load variation of 3 kPa (from 0 to 3 kPa), there was an increase of the expected losses of about 19%. However, in the second case, an increase of 2 kPa (from 3 to 5kPa) of ash load implied an increase of about five times in the expected losses, revealing a strong non-linear (and more complex) arrangement.
French West Indies
The first results discussed here concern the establishment of a framework to compare risks within the context of the Pointe-à-Pitre municipality, considering storm surges and seismic risk. 20 shows the results of loss estimates as determined during MATRIX (see deliverable D7.4 Monfort and Lecacheux (2014) for details) for seismic risk for Pointe à Pitre and 5 return periods (100 years, 475 years, 975 years, 1975 years and 5000 years), and for storm surges, although for only a 100 years return period was considered, but for different maximum wave heights. 20a shows the results for economic losses, where it is apparent that earthquakes are significantly more serious than storm surges. However, if one looks at “no shelter”, we see for the 100 year return period the levels are very similar. It should also be commented that casualties may not be the best measure to compare earthquakes and storm surges, since there is an element of warning in surges (allowing people to make their own mitigating actions) while no or few such opportunities arise for earthquakes.
Another study performed over the Guadeloupian territory was the assessment of systemic vulnerability of infrastructures exposed to cascading events. This work aimed at analysing the effects of cascading events on the behaviour of interdependent systems and on the capacities of the health care system to treat the victims, with Guadeloupe being a useful example given its context as a closed system. The hazard cascading scenario considered involved a M6.3 earthquake striking Basse-Terre, and triggering landslides in the mountainous areas where antecedent precipitations had made the area prone to such events, leading to the closure of road links. Damage due to earthquakes was estimated for the 5 considered systems (buildings, healthcare system, electrical network, water supply network and transportation).
The inter- and intra-dependencies of the systems were found using the Infrastructure Interdependencies Simulator (I2Sim, Liu, 2007). I2Sim provides a platform to represent all critical infrastructures and their operating interdependencies. For example, 21a shows a simple system containing a power station, a water station, a hospital and a residential area under normal conditions. Even though it is a simple system, it is already has feedback loops and interdependencies. The consequences of disrupting, for example, the power supply (21b) then leads to decisions to be made as to how to distribute power, that is, balancing the requirements of the hospital, water station, etc.
Analyses were performed for different strategies of resources allocations, and one of the final results was the impact of the induced landslides on the health care treatment capacity. Based on the simulations, the time required to transport all injured people to a functioning hospital has increased by a factor of 2.5 if one takes into account the consequences arising from lifeline disruptions. 22 shows an example of how the various lifelines and their relative effectiveness may be determined from such a tool as I2Sim.
Cologne
Details of the risk estimates arising from the three hazards that mainly affect Cologne are presented in deliverable D7.5 (Fleming et al., 2014). As we have already presented some results for Cologne, namely the combination and comparison of risk types, here we will simply present the new results for the three hazard types. This is presented in 23 for earthquakes (23a), floods (23b) and windstorms (23c). For seismic risk, a logic tree approach was employed, leading to a total of 180 estimates of risk arising from a range of input parameters (see Tyagunov et al., 2014). For floods, use was made of the hybrid probabilistic-deterministic coupled 1D (dike breach) 2D hydrodynamic model IHAM (Vorogushyn et al., 2010) run in a Monte Carlo simulation. For windstorms, results from the Vienna Enhanced Resolution Analysis or VERA tool were used (Steinacker et al., 2006). All risk calculations for the three hazards used the same metric and total costs (i.e. direct damage to residential buildings). For seismic risk, we plot the mean estimate with the 95th and 5th percentiles, along with the seismic risk curve of Grünthal et al. (2006). For the flood and windstorm risk curves, we plot the median plus the 5th and 95th percentiles and again the curves of Grünthal et al. (2006). The major point to note is the substantial uncertainty arising for each hazard type. This makes it difficult to identify for given return periods which hazards should be of more concern to decision makers. It also again emphasises the results presented earlier employing the Wilcoxon’s test (4)
Interactions with the DRR community
End-user and stakeholder interaction was a very important and fruitful aspect of the MATRIX project. Details of some outcomes will be given in the following section dealing with dissemination activities, but here we will review a few of the main findings and responses. To undertake these activities, MATRIX instigated an external group referred to as the Cooperation partners. These were stakeholders (National Platforms for Disaster Risk Reduction, UN-ISDR, EU, CoE, and the private sector) who were prepared to participate in the project’s meetings, workshops, and to complete questionnaires about various aspects of multi-risk assessment, in particular its status in the different European member states. Considering the National Platforms, of the 22 European platforms, 13 interacted with MATRIX2.
One important activity carried out, which lead to a publication (Komenantova et al., 2014) was an investigation into how stakeholders reacted to the tools developed within MATRIX, that is the MATRIX-CITY/Virtual-City and the decision support tool of KIT (discussed above), as well as what their actual needs were. Existing risk assessment methods integrate large volumes of data and sophisticated analyses, as well as different approaches to risk quantification. However, the key question is why do losses from natural disasters continue to grow if our scientific knowledge on multi-risk is increasing? (White et al., 2001). The MATRIX work here dealt with the questions of communication and the transfer of scientific knowledge on multi-risk and its underlying drivers to stakeholders within the decision-making process. A two-way communication process allowed feedback to be collected from stakeholders (i.e. civil protection offices) across Europe on the usability of the multi-risk decision–support tools that have the potential to benefit decision-makers and to provide them with information on mitigation measures. This would also allow the integration of their feedback into improving the tools themselves.
The theoretical background of this work involved the concept of risk governance, which takes into account cultural and political factors when implementing risk mitigation measures and emphasizes the role of participation and communication. The risk governance concept is concerned with such issues as how information is perceived, collected and communicated, and, based on these factors, how management decisions are made (IRGC, 2005). Participatory modelling is an important part of risk governance and allows us to take into consideration not only facts, but also values by collecting feedback from stakeholders (Forester, 1999). The process of interacting with stakeholders leads to an enhanced understanding of the views, criteria, preferences and trade-offs employed in decision-making (Antunes et al., 2006). Also, since the development of scientific tools is also a social process, it is essential to involve relevant stakeholders who will be using the tools in the design process through the collection and integration of their feedback (Tesh, 1990).
There is also a gap in the scientific literature about feedback with respect to the usability of decision-support tools. While the use of feedback for the development of decision-support tools for environmental issues has been reported (Constanza and Ruth, 1998), as well as there being multi-risk decision–support tools that have the option of collecting feedback (T6, 2007), there is no evidence or analysis of the feedback from stakeholders from practice on the usability of multi-risk decision-support tools. During our work, we not only collected such feedback from civil protection officers, but we also used this information to improve the developed decision-making tools, directly integrating stakeholders’ perceptions into the model by attributing different weights to loss parameters according to preferences from stakeholders. The information was gained during two workshops, namely a MATRIX stakeholders’ meeting in Bonn (July, 2012, discussed further below) and a workshop on urban multi-hazard risk assessment in Lisbon3 (October, 2012), and from a questionnaire distributed prior to the first workshop.
The general results show that for the usage of multi-risk decision-support tools, two areas are most problematic. These are (1) the absence of clear definitions and (2) the lack of information on the added value of multi-risk assessment. Multi-risk is not systematically addressed among the EU countries for all hazards, but is only singularly integrated into risk assessment approaches. Some examples include the superposition of existing single-hazard risk prevention plans for all hazards, for example, combining flood and landslide hazards and flood risks with wind effects, the application of which has been within the context of risk assessment for critical infrastructure (in particular the combination of meteorological and technological risks). Generally, multi-risk analysis is barely or not at all integrated into decision-making processes, and only around half of stakeholders were aware of methodologies and tools to assess multi-risk.
The reaction of stakeholders to the multi-risk assessment and decision-making tools presented at the workshops was optimistic. Several stakeholders invited the tools’ developers to give presentations and conduct training at their home institutions. However, the usability of the tools in practice is complicated by such factors as the required large volume of input parameters (involving cumbersome data gathering) and that their possible application is limited to only a narrow number of experts, since a high-level of expertise is required to assess the dynamic multi-hazard and multi-risk processes, taking into account the complexity of the models and the required parameters.
The consultation process with stakeholders also showed significant variation in perceptions between stakeholders in academia and in practice. While both academicians and practitioners agreed that the decision-support tools are useful for understanding losses and their contributions in a risk scenario, differences arise between how practitioners viewed the usefulness of the tools when it came to prioritizing risk and developing risk management strategies. Similarly, practitioners found the tools less useful than academics when it came to preparing for disasters and allocating resources. For example, Figure 24 shows the almost opposite view of academics and stakeholders with respect to the results of the survey that asked for impressions as to how the decision support tool helps with the understanding of losses and their contribution in a risk scenario (Figure 24a) and how this tool helps with preparing for multi-risk disasters and optimizing the allocation of resources (Figure 24b). As can be seen, while both academics and practitioners thought the tool was useful for understanding (although less so for practitioners), practitioners did not see it as useful for actually preparing for multi-risk disasters and allocating resources, a critical finding for those (usually academics) who are developing the tools.
Recommendations with respect to two areas involving the application of decision-support tools involved convincing stakeholders involved in the decision-making process of the usefulness of the multi-hazard approach, as well as dealing with the problem of disseminating any results to the general public (hence confronting public acceptance issues). Some stakeholders expressed the opinion that politicians could also use such models as training to see what the consequences of a multi-hazard situation could be. Another general recommendation was that the decision-support tools could be used for education and training purposes.
Risk governance and multi-hazard and risk frameworks
It was identified during MATRIX that in risk assessment research and policy, there is much debate on multi-type hazard and risk assessment and the definition and use of realistic scenarios. This debate has been evoked, not least, by specific disasters in recent years, namely the Super Typhoon Haiyan (the Philippines, November 2013), causing floods and landslides, and the Töhoku earthquake (Japan, March 2011), with the resulting devastating tsunami and nuclear accident. Therefore, in MATRIX efforts were made to synthesise the identified benefits and barriers to multi-hazard mitigation and adaption, the target audience being practitioners within the public/private sector working in communities exposed to multiple risks as well as to those active at the science-policy interface, thus including researchers, policy and decision makers in risk and emergency management (Scolobig et al. 2013; Komendantova et al., 2013; Scolobig et al., 2014).
It was found that within current single-risk-centred governance systems, both centralized and decentralized, (which have evolved in parallel with the single-risk-centred assessment processes), practitioners hardly ever have the opportunity to discuss multi-risk issues, including triggered events, cascade effects and the rapid increase in vulnerability resulting from successive hazards. However, as revealed by the results from various workshops (see the following section) as well as several interviews with stakeholders, that risk and emergency managers clearly see the benefits of including a multi-risk approach in their everyday activities, especially in the urban planning sector, but also in emergency management and risk mitigation.
One example where practitioners believe a multi-risk viewpoint would be of value concerns decisions surrounding building restrictions for urban planning, as well as the correct cognitive and behavioural biases. A multi-risk approach is considered particularly useful also for gaining a holistic view of all of the possible risks that may affect a territory. For example, such an approach can show that focusing only on the impacts of one hazard could result in raising the vulnerability of the area to another type of hazard. An example of this is in the older buildings of Kobe, Japan, which were built with relatively heavy roofs. While this helped mitigation against frequent typhoons, it enhanced their vulnerability to rarer earthquakes (Otani, 1999).
Other benefits that are considered to be particularly crucial by practitioners include:
The cost reductions and improvements in the efficiency of proposed risk mitigation actions.
The development of new partnerships between agencies working on different types of risks.
An awareness of the potential for expected losses being exceeded (i.e. the total risk is possibly greater than the sum of the individual parts), as well as the lives and property saved and better protected by the use of a multi- vs. single-risk approach.
However, further research is still needed in order to better understand the extent of some of these benefits, as well as the need to consider aspects of the mitigation problem, such as the different time scales involved between the events themselves, response, initial recovery and ongoing mitigation. Our results also reveal that practitioners and researchers have in mind different agendas for future research on multi-risk assessment. Therefore, a transparent process to reach a compromise on the required priorities is needed.
Barriers to an effective implementation of multi-risk assessment can be found in both the science and practice domains. For example, considering scientific contributions to risk assessment research, the process has evolved differently in the fields dealing with geological versus meteorological hazards, with the different scientific development paths representing a major barrier to understanding and communicating between different “risk communities”. Accompanying this is the lack of open access to databases and research results, which is particularly worrying for risk managers. Overarching these problems are the matters of the lack of interagency cooperation and communication, which are particularly difficult for risks that are managed by authorities acting at different levels (e.g. in Italy, national bodies are responsible for volcanic risk, while river basin authorities deal with flood risk). The lack of capacities at the local level and unsatisfactory public-private partnerships are also major barriers that need to be confronted.
As a result of our interactions and discussions with stakeholders, some priority actions have been identified:
Encourage knowledge exchange and dialogue between the risk communities dealing with geological and meteorological hazards;
Identify new options for mitigation, e.g. multi-risk insurance schemes, new forms of public-private responsibility sharing for households exposed to multi-risks;
Develop territorial platforms for data and knowledge exchange between researchers and practitioners;
Create an inter-agency environment, where the different departments at the national and/or regional governmental level can exchange information, develop complementary protocols, and serve to provide consistent information and responses to the relevant stakeholders;
Create commissions for discussion at the local/municipal level ("local multi-risk commissions") in order to gain a common understanding of what multi-risk assessment actually is, what kind of cooperative actions can be undertaken to implement it, what are the priorities for future research etc.. Members of these commissions should be decision and policy makers, researchers and local natural hazard advisors, the latter acting as the liaising bodies between local communities and practitioners.
Potential Impact:
The first point to make with respect to the “impact” of the MATRIX project is that multi-hazard and risk is a relatively new field of endeavour, and that MATRIX itself can only realistically be considered as contributing to the expansion of research in this direction. Having said this, one of the more prominent areas where MATRIX has had an impact is in raising the awareness of the issue of multi-hazard and risk assessment. In the following, we will therefore begin with an overview of where this awareness of multi-hazard and risk assessment has grown as a result of MATRIX (recognising of course that MATRIX is by no means the only driver of this point), to be followed by summary of the dissemination activities undertaken, and some proposed ways in which the results may be exploited. For the latter point, mention will be made of current projects that have the potential to benefit from the MATRIX results.
The impact of the MATRIX project, in fact, is linked to the dissemination activities, in terms of representatives of the project being invited to various expert meetings, in particular those associated with the European Commission (see below for some comments as to the response to such presentations), the aim being, as mentioned, to encourage DRR practitioners to consider the need for a multi-hazard and risk framework. In fact, in a recent report from the European Commission on the post-2015 Hyogo Framework for Action1, it was commented that a number of Member States in their National Risk Assessments (NRA) mention “’consequential’ and ‘cascade-effects’” of risk (incidentally, MATRIX is specifically mentioned in this document). While it is difficult to assess to what extent MATRIX’s efforts have had on the recognition of this issue, acknowledging that multi-hazard and risk as understood within MATRIX was discussed in an early European Commission document outlining risk assessment and mapping guidelines2, the MATRIX project was specifically mentioned as an effort to develop the means of dealing with this issue. Another impact in terms of raising the discussion about multi-risk in DRR activities involves the production for the next Global Assessment Report (GAR 15) of a document outlining the issues involved in disaster risk governance when considering a multi-hazard and risk framework (Scolobig et al., 2014), as also outlined in Komendantova et al. (2014).
The dissemination activities conducted during MATRIX may be divided into three main types:
“Standard” dissemination practices in the form of academic publications, conferences and workshops, and representing MATRIX at other projects’ meetings, as well as the project’s website, fact sheets and its listing on other relevant websites (e.g. GRIPWEB3). Representatives of other projects and of Civil Protection were also invited to the project’s annual meetings.
“Expert” and similar meetings within the context of the various branches of the European Commission (commented upon above in terms of the impact of the project).
“End-user and stakeholder” meetings, which endeavoured to not only inform such groups about the results and goals of MATRIX, but also to encourage the adoption of a multi-hazard and risk framework in general. This part of the dissemination activities was possibly the most important, as it involved making contact with the European National Platforms for Disaster risk Reduction, as well as interacting with such parties as the European Commission, the UN-ISDR and the private sector, such as insurance companies.
The publications produced during the project are listed in Template A1 of Section A. In order to make these works available in an open source manner, efforts were expended in making them available via the OpenAire4 (Open Access Infrastructure for Research in Europe). Note these manuscripts are made available in an unformatted form in the “accepted version” to comply with the copyright regulations of the journals involved.
MATRIX was represented at a number of project meetings. These included the closing meeting of the KULTURisk project, which aimed to develop a culture of risk prevention by evaluating the benefits of different risk prevention initiatives5, and the Task C EUSBSR (EU Strategy for Baltic Sea Region) Flagship project 14.36 whose objective was to develop scenarios and identify gaps in the response to all of the main hazards that affect the Baltic Sea Region. In both projects, while multiple types of hazards are accommodated, a genuine multi-hazard and risk framework as considered within MATRIX was still lacking.
The presentation of MATRIX at various expert meetings in Brussels not only served to advertise the project’s goals, but to bring to the DRR community’s attention the potential value of a multi-hazard and risk approach. There was (and probably remains) some scepticism as to the value of such an approach. For example, following one of the first of these meetings where MATRIX was presented (July 2011, sponsored by the European Commission Humanitarian Aid and Civil protection, ECHO), one attendee commented “I would be happy if I could manage a simple risk assessment. Multi-risk is far away from the reality on the ground.” It was therefore the intention of subsequent MATRIX presentations to argue that this was not the case, and that a multi-hazard and risk assessment regime is plausible at this time. In fact, it was regularly proposed during similar presentations that on one level, there is no choice, but to adopt a multi-type approach given the increasing complexity of human society, and the recognition of the interactions that arise at all levels of disaster mitigation. Similarly, at a workshop organised by the European Commission Directorate General for Research and Innovation in Brussels dealing with the contribution of the socio-economic studies to hazard and risk research (22nd to 23rd October, 2012), at which the MATRIX project was presented, it was emphasised that research into the actual processes involved (that is, fundamental natural science research) was still required, especially within the context of a multi-hazard environment.
The most important (and productive) of the dissemination activities carried during the MATRIX project involved two end-user and stakeholder meetings, organised by DKKV (together with KIT and GFZ), one of the MATRIX partners and the German contact point for Disaster Risk Reduction. These meetings (held in July 2012 and June 2013) aimed to present the aims and results of the MATRIX project to various members of the DRR community, including representatives of the European National Platforms for Disaster Risk Reduction, UN-ISDR, the European Commission and other bodies such as insurance companies. The primary aims of these meetings may be summarised as follows:
Outlining of the aims and work program of the MATRIX project.
Presenting the concepts behind multi-hazard and risk assessment, as understood within the project. In particular, the case needed to be made for the importance and potential benefits of considering a multi-hazard and risk framework, where the spatial and temporal interactions and dynamics at the hazard, exposure, and vulnerability levels were accommodated, while also accounting for socio-economic impacts.
The presentation of various decision support tools developed within MATRIX were presented. This in particular involved showing the MATRIX-CITY and the Virtual City tools.
The presentation of the cross-cutting results from the test cases.
Presentations by representatives of the National Platforms and civil protection groups and insurance associations, covering themes ranging from the DRR status in their respective countries to test cases which may also involve a multi-hazard and risk environment.
An issue future projects will need to confront is that getting the private sector involved may prove to be very challenging, as it was during MATRIX. While the German Insurance Association was represented at the second workshop, the minutes of that meeting made available to other groups (Icelandic Natural Hazard Insurance Company and Austrian Railway), and a number of contacts to potentially interested enterprises, significantly greater effort would required to involve the private sector in future projects, especially within the Horizon 2020 (H2020) scheme. Another point raised by the attendees of the 2nd meeting was that, rather than simply contacting end users during a project or even before, it was advised that they should be involved with the actual preparation of future project proposals, rather than simply being “added on”.
With regards to the exploitation of the results of MATRIX, one point is the consideration of MATRIX in future research proposals. It is now expected that in the current round of hazard and risk proposals being (within which many members of the MATRIX consortium are involved), a multi-hazard and risk perspective is assumed and incorporated into any planning.
Finally, one of the MATRIX deliverables was a vision statement, outlining what the next steps should be within the context of multi-hazard and risk research (Wenzel, 2014). In summary, the main suggestion is for the redirection of future multi-risk research from risk assessment to risk mitigation, and for it to be considered within the context of resilience, which in fact is the case for a number of H2020 (and other funding bodies’) calls currently being prepared for. Some other issues that need to be considered in terms of research and implementation include:
The need to move from human and economic loss estimation to the efficiency of response and the speed and sustainability of recovery.
Continue work on the various forms of interactions (system, natural processes, etc.), and treating multi-hazard and risk over different spatial and temporal scales. This includes ongoing work on methods for predicting the consequences of, for example, cascading effects.
The importance of ensuring that the software tools developed within projects such as MATRIX are “not lost” (i.e. that resources are found to move them beyond the prototype stage).
Expansion of different means of data collection, and exploitation of different types and sources of data, including ground-based and remote sensing platforms and monitoring systems, crowd sourcing and “big data”.
The development of methods for multi-hazard and risk assessment should be integrated with those that allow for the incorporation of utility functions, cost-benefit analysis, and life-cycle analysis.
List of Websites:
Project website: http://matrix.gpi.kit.edu/
Project coordinator: Professor Dr. Jochen Zschau (zschau@gfz-potsdam.de)
Centre for Disaster Management and Risk Reduction Technology