This paper combines research and commentary to reinforce the importance of
integrating hazard interactions and interaction networks (cascades) into
multi-hazard methodologies. We present a synthesis of the differences between
In this article we present both research and commentary to support the integration of hazard interactions and their networks (cascades) into multi-hazard methodologies. Building on the work of others (Delmonaco et al., 2007; Kappes et al., 2010, 2012; Marzocchi et al., 2012; Gill and Malamud, 2014) we advocate for a multi-hazard approach that goes beyond the simple overlay of multiple single hazards to an approach that also encompasses interactions between these hazards. We present here an enhanced framework for considering such interactions and integrating these into multi-hazard methodologies, supporting efforts to improve management of those aspects of the Earth system that are relevant to disaster risk reduction. Examples from primary research and published literature, together with commentary about multi-hazard approaches, are included throughout.
Following this introduction, Sect. 2 examines the differences between single-hazard, multi-layer single-hazard, and full multi-hazard risk approaches. In Sect. 3 we define and describe three distinct hazard and process groups (natural hazards, anthropogenic processes, technological hazards/disasters) that can be considered in multi-hazard methodologies. This is followed by Sect. 4, which discusses and visualises three principal interaction relationships between these hazards and processes (triggering relationships, increased-probability relationships, catalysis/impedance relationships), with a detailed description of their differences and examples of each. Then in Sect. 5 we discuss how individual interactions can join together to form networks of hazard interactions (cascades), using four case studies (one from Nepal and three from Guatemala) and two theoretical examples to consider different features of interaction networks and how these can be visualised using hazard interaction matrices and hazard/process flow diagrams. We also comment on the benefits of assessing networks of hazard interactions to support disaster risk reduction. Conclusions are outlined in Sect. 6.
Single-hazard approaches to assessing hazard potential, in which hazards are
treated as isolated and independent phenomena, are commonplace. Their
prevalence, however, can distort management priorities, increase
vulnerability to other spatially relevant hazards or underestimate risk
(Tobin and Montz, 1997; ARMONIA, 2007; Kappes et al., 2010; Budimir et
al., 2014; Mignan et al., 2014). If a community is susceptible to more than
one hazard, management decisions will benefit by reflecting the differential
hazard potential and risk from each of these, and not just focus on them as
individual entities. Focusing on a small portion of the whole Earth system,
rather than the dynamics of its entirety, may result in decisions being made
that increase people's vulnerability to other, ignored hazards. The
development of enhanced
Multi-hazard approaches are widely encouraged in key government and intergovernmental initiatives and agencies but are rarely defined. For example, the Hyogo Framework for Action (2005–2015) called for “an integrated multi-hazard approach to disaster risk reduction” (UNISDR, 2005, p. 4). The Sendai Framework for Disaster Risk Reduction (2015–2030) states that “disaster risk reduction needs to be multi-hazard” (UNISDR, 2015, p. 10). Despite the emphasis on multi-hazard approaches within these international agreements, both the Hyogo and Sendai frameworks do not define what a multi-hazard approach involves. At the time of writing, the term multi-hazard also does not appear in the most recent descriptions of terminology published by UNISDR (2009). Further examples of multi-hazard approaches being advocated for, but not clearly defined, can be found in United Nations (2002) and Government Office for Science (2012).
The term
The identification of all possible and spatially relevant hazards is an
important feature of a full multi-hazard assessment, but we believe it should
not be the sole defining characteristic of such an approach. Multi-hazard
assessments may also recognise the non-independence of natural hazards
(Kappes et al., 2010), noting that significant interactions exist between
individual natural hazards. In a previous study (Gill and Malamud, 2014) we
took 21 different natural hazards and identified 90 possible interactions
between the 441 (
We now highlight five possible types of hazard interactions that may occur if
an inhabited location is susceptible to multiple hazards, using
four natural hazards (tropical storms, floods, landslides and volcanic
eruptions) as exemplars:
The above five interaction types, based on just four natural hazard
exemplars, are taken from a much broader range of possible hazard
interactions and their networks. Even with these limited examples, they
demonstrate the limitations of assuming independence of single hazards within
a multi-layer single-hazard approach.
Multi-hazard methodologies, therefore, should ideally evaluate all identified individual hazards relevant to a defined spatial area and characterise all possible interactions between these identified hazards. Figure 1, from Gill and Malamud (2014), shows four distinct factors required to transition from a multi-layer single-hazard assessment to a detailed, full multi-hazard risk assessment (which includes hazard interactions, vulnerability and exposure). In addition to identifying all hazards and their interactions, this working framework also proposes an assessment of concurrent hazards (such as a tropical storm and volcanic eruption coinciding spatially and temporally), and the recognition that vulnerability is dynamic (which we discuss more in Sect. 5.3).
Many current hazard assessments that are labelled as
The hazard interactions literature outlined in Sect. 2.1 includes studies for different spatial extents, including global (e.g. Gill and Malamud, 2014), continental (e.g. Tarvainen et al., 2006) and local or sub-national (e.g. De Pippo et al., 2008). The scale of interest for a particular multi-hazard approach determines how interactions are characterised. Approaches may be based on an examination of an individual event (e.g. a given earthquake triggering landslides in a given region) or draw on a large population of individual events to infer the probabilistic behaviour of a relationship (e.g. considering many earthquake-triggered landslide events over different regions, and from this the dependence of the number of landslides triggered based on the earthquake magnitude). The latter approach is used to consider in general how one hazard will influence another. Both approaches are beneficial in different contexts. For example, a probabilistic viewpoint is likely to support the characterisation of possible interactions in a general, globally relevant way, as we often consider them in our paper. When adapting global, multi-hazard approaches for use in regional and local contexts, a different population of individual events is required to infer the probabilistic behaviour of a relationship specific to that context. In many regions, although the database of events is likely to better reflect site-specific conditions, it may be small, consisting of just a few (sometimes zero) individual events, depending on the period of time considered.
Multi-hazard risk framework (from Gill and Malamud, 2014). Shown is the progression from a multi-layer single-hazard approach to a full multi-hazard risk approach that includes (i) hazard identification and comparison, (ii) hazard interactions, (iii) spatial/temporal coincidence of natural hazards, and (iv) dynamic vulnerability to multiple stresses (when progressing from the assessment of hazard to the assessment of risk).
Another possible contrast between globally relevant multi-hazard approaches
and location-specific, multi-hazard approaches is the forecasting time window
(Marzocchi et al., 2012) or temporal resolution (Kappes et al., 2012). In
globally relevant approaches that draw upon many individual events,
generalisations across forecasting time windows (both short- and long-term
time windows) are used to construct the multi-hazard framework, with the
inclusion of interactions relevant at all temporal resolutions. When
developing location-specific multi-hazard assessments, clear temporal limits
should be established (Selva, 2013), depending on the purpose of the
multi-hazard approach. When constructing location-specific assessments of
hazard potential, Marzocchi et al. (2012) propose that the modelling of
hazard interactions is more necessary in the short term (e.g. hours to days)
than the long term (e.g. many decades to centuries). They argue that, in the
short term, the occurrence of a primary hazard (e.g. storms) can
significantly modify the probability of secondary hazards (e.g. floods,
landslides), compared to the long-term, where primary hazards (e.g.
earthquakes, landslides) are already considered in the long-term assessment
of the secondary hazard (e.g. tsunamis). In other words, they discuss that in
a long-term perspective (e.g. the tsunami hazard over the next 50 years),
databases already contain information of the fact that most tsunamis are
triggered by earthquakes, and there is no need to make additional
calculations to calculate the long-term tsunami hazard. It is therefore less
necessary in the long term (compared to short term) to model possible
interactions as databases of past single-hazard occurrences already reflect
the triggered nature of these hazards. In the long term, however, it is
important to consider the temporal proximity of successively occurring
hazards (e.g. earthquake
As further multi-hazard approaches are developed, and integrated into research and practice, we believe that it is important to recognise that (i) natural hazards do not operate in isolation, (ii) the characteristics of a framework at global spatial scales may differ from more context/location-specific scales, and (iii) enhanced multi-hazard approaches would also likely benefit from considering how human activity can influence the triggering of hazards and initiation of networks of hazard interactions. We now proceed to define and describe three principal groups of hazards and processes that enhanced multi-hazard frameworks may consider including.
Here we discuss the characterisation of hazard potential for an applied
multi-hazard approach that includes an assessment of at least three distinct
groups: (i) natural hazards, (ii) anthropogenic processes and
(iii) technological hazards/disasters. All of these can be considered to be
processes and/or phenomena with the potential to have negative impacts on
society. In the context of this article, these terms are defined as follows:
A more detailed list of examples for each of these three groups (natural
hazards, anthropogenic processes, technological hazards/disasters), based on
the definitions set out above, are given in Table 1. We now discuss in more
detail (Sect. 3.1, 3.2, 3.3) each of these three groups, particularly
potential overlap between the groups
Examples of hazard/process types, grouped into three categories: natural hazards (classification of 21 hazards from Gill and Malamud, 2014), anthropogenic processes and technological hazards/disasters.
The meaning of the phrase
Anthropogenic processes are less well defined and characterised as a group,
compared to the group labelled
Although often referred to in the context of disaster studies (e.g.
Fleischhauer, 2006; Tarvainen et al., 2006; Bickerstaff and Simmons, 2009),
technological hazards/disasters are also less well defined and characterised
than the group
The UNISDR (2009) definition of technological hazards also states that hazards originate from technological or industrial conditions, including human activities that may cause environmental damage, health impacts, economic disruption and other negative consequences. This could include human activities such as subsurface mining, groundwater abstraction and vegetation removal. Therefore, the UNISDR (2009) definition of technological hazards also appears to include examples that we have categorised as anthropogenic processes.
Other authors make a clearer distinction between anthropogenic processes and
technological hazards. For example, Kasperson and Pijawka (1985) outline
three categories of technological hazards:
The last two (technology failures, technological disasters) are distinguished
based on the scale of impact, with technological failures able to evolve into
technological disasters if losses are sufficiently large. Although included
within the broad category of technological hazards in Kasperson and Pijawka
(1985), there is significant overlap between their definition of “routine
hazard events of technology” and our definition of anthropogenic processes,
outlined in Sect. 3.2. For example, in Table 1 we note surface mining to be
an anthropogenic process. This classification is based on our definition of
anthropogenic processes being intentional human activities that are
non-malicious but may have a negative impact on society through the
triggering or catalysing of hazardous processes (Sect. 3). Surface mining can
also be considered to be a “routine hazard event of technology” as defined
by Kasperson and Pijawka (1985), in that the mining is a technological
process where there is exposure to underlying chronic hazardous activity,
which can be managed by established procedures.
Whereas technological failures and disasters are generally
In Sect. 3.1, 3.2 and 3.3 it was noted that both anthropogenic processes and
technological hazards/disasters are non-malicious; the negative consequences
are not the desired outcome. Events that are malicious or deliberately
destructive (e.g. terrorism, arson, aspects of warfare and criminal activity)
are not included within either
In the context of the rest of this article we focus on interaction relationships between the three groups just discussed – natural hazards, anthropogenic processes and technological hazards/disasters – and the development of possible networks of hazard interactions (cascades).
Multiple interactions exist between the hazard and process examples outlined
in the three groups (natural hazards, anthropogenic processes, technological
hazards/disasters) discussed above. Kappes et al. (2012) note a wide variety
of terms used to describe such interactions (e.g. interrelationships,
interconnections, coupled events) and specific sets of interacting hazards
(e.g. coinciding hazards, triggering effects). Here we continue to use the
term
We use the term As an illustrative example, we will take an earthquake as the primary hazard
and a triggered landslide population event as the secondary hazard and will
discuss perspectives from before the primary hazard occurs (
While we distinguish
When generalising across these three time periods (1, 2A, 2B), recognising that an earthquake (primary
hazard) can both trigger and increase the likelihood of landslides (secondary
hazard) occurring in (
In summary, while causal triggering relationships can only be “known” retrospectively, there is still a good justification for distinguishing between triggering and increased-probability relationships when using forward-looking approaches. For any given period of time after a primary hazard, those interested in hazard interactions (e.g. scientists, hazard managers) may want to know what the likelihood is of landslides occurring (being triggered), as well as whether there is a change in the likelihood of landslides beyond this period of time (increased probability). Although attribution or identifying a causal relationship between a specific primary hazard (e.g. a given earthquake) and a specific secondary hazard (e.g. a given tsunami) is clear in some cases, other times attribution is not so clear (e.g. the increase in probability of landslides as a result of a wildfire). This challenge of attribution is currently in the forefront of the climate change community, where attempts are made to determine the existence of causal relationships between anthropogenic climate change and specific extreme events (Stott et al., 2013; Shepherd, 2016).
We now discuss each of these three interaction relationships in more detail, giving examples and introducing two visualisations. These interaction relationships are also used in Sect. 5 when discussing networks of interactions (cascades).
Triggering relationships are one form of causal relationship, where the occurrence of a primary event can result in secondary events occurring. For example, a tropical storm or hurricane (a primary natural hazard) may trigger many landslides (a secondary natural hazard) due to the rapid increase in ground saturation, such as in the case of Hurricane Mitch in 1998, where heavy rain associated with the hurricane resulted in thousands of landslides being triggered in Guatemala (Bucknam et al., 2001). As noted in Sect. 4.1, feedback mechanisms can also exist where a triggered secondary hazard exacerbates the primary hazard and results in further occurrences of the primary and/or secondary hazard being triggered.
Triggering interactions can occur between a diverse range of hazards and
processes. Gill and Malamud (2014) considered just natural hazards and
identified 78 possible triggering pairings between 21 natural hazards (the
same natural hazards as those given in Table 1). The inclusion of both
We also highlight that each triggering relationship identified will have
different likelihoods associated with it. In any given location, the
likelihood will be dependent on site-specific conditions (e.g. geology,
hydrology, neotectonics, the extent of human activity). From a probabilistic
viewpoint, generalising across multiple individual events for each triggering
relationship, we can also infer that some triggering relationships are more
likely to occur than others. For example, Gill and Malamud (2014) use a
nine-point scale to classify the
Of importance in the context of characterising triggering relationships are
the spatial and temporal scales of interest. When considering interactions in
a specific local/regional setting, different interaction behaviours will
occur at different spatial and temporal scales. For example, an anthropogenic
process, such as agricultural practice change, could occur at multiple
scales. An individual farmer ploughing a new field (approximate spatial scale
of 0.1–1
Another important factor for consideration when characterising triggering
relationships is the relative timing of different stages. For example, some
anthropogenic processes may involve multiple stages, including an initial
decision-making or survey stage before ground disturbance. In this example,
it is possible that a given anthropogenic process may trigger other processes
to occur before, simultaneously with,
or after any ground disturbance has occurred. Where an associated process is
stated to occur
Interaction relationships (triggering) framework using a
hazard/process flow diagram. A framework for hazard/process interactions is
given here, which highlights triggering relationships between three groups:
(A) natural hazards, (B) anthropogenic processes and (C) technological
hazards/disasters. Arrows are used to illustrate interaction relationships,
with the arrow fill colour indicating the source or initiation of the trigger
(medium grey: natural hazards; dark grey: anthropogenic processes; light
grey: technological hazards/disasters). We use a prime (A
When considering combinations between the three groups of hazards/processes
(natural hazards, anthropogenic processes, technological hazards/disasters),
we identify nine possible triggering relationships between these groups and
visualise these in Fig. 2, a hazard/process flow diagram. Triggering
relationships are illustrated using block arrows, with the internal arrow
fill colour indicating the group of hazards or processes to which the
Another type of causal relationship can be observed when a primary natural hazard, anthropogenic process or technological hazard increases the probability of another such event occurring. These situations involve a primary hazard or process altering one or more environmental parameters so as to change the temporal proximity or specific characteristics of an individual or population of secondary hazards or processes (Kappes et al., 2010; Gill and Malamud, 2014). Examples relating to specific natural hazards include an earthquake increasing the susceptibility of a slope to landslides, regional subsidence increasing the probability of flooding, or wildfires increasing the probability of ground heave. In Gill and Malamud (2014), we took the 21 different natural hazards identified in Table 1 and identified 75 possible relationships where a primary natural hazard could increase the probability of a secondary natural hazard. The inclusion of anthropogenic processes and technological hazards/disasters will also result in many more increased-probability relationships.
Interaction relationships (triggering, catalysing and impeding)
framework using a hazard/process flow diagram. Interactions in the form of
triggering relationships (Fig. 2) and catalysing/impedance interactions are
possible between (A) natural hazards, (B) anthropogenic processes and
(C) technological hazards/disasters. We use a prime (A
We have discussed above that one hazard/process may trigger another hazard/process. It is possible that further hazards and processes may cause these triggering relationship pairings to be catalysed or impeded. For example, tropical storms can often trigger floods. This triggering relationship can be catalysed by other specific anthropogenic processes (e.g. vegetation removal, urbanisation), natural hazards (e.g. wildfires) or technological failures (e.g. blocked drainage). Conversely, a volcanic eruption can trigger wildfires, but this triggering relationship may be impeded by other specific anthropogenic processes (e.g. deforestation) or natural hazards (e.g. tropical storms).
In addition to the 9 triggering interaction relationships previously identified (Fig. 2), a further 12 possible catalysing and impedance relationships can be considered, which we visualise in Fig. 3, also a hazard/process flow diagram. In Fig. 3, we contrast triggering relationships (9 thick block arrows with solid outlines) and catalysing/impedance relationships (12 thin block arrows with dashed outlines). The internal arrow fill colour again indicates the group of hazards or processes to which the catalyst/impeder belongs (medium grey: natural hazards; dark grey: anthropogenic processes; light grey: technological hazards/disasters).
Figure 3 highlights the range of possible interaction relationships between
the three broad groups of hazards/processes, using a hazard/process flow
diagram. Within each type of interaction relationship there exist specific
interactions that are rare and others that are very common, with a wide
spectrum between these two end members. Location-specific conditions
influence the likelihood of any given interaction relationship. The
likelihood of each catalysing relationship will depend on (i) the likelihood
of the primary hazard/process occurring, (ii) the likelihood of the primary
hazard/process triggering a secondary hazard, and (iii) the likelihood of a
given hazard/process catalysing this interaction pairing. Consider, for
example, the unloading of slopes through road construction catalysing
earthquake or storm-triggered landslides (thin, dark-grey arrow from B to
A
Examples of some specific catalysing and impeding interaction relationships
are presented below. Here we state which hazard or process group (e.g.
anthropogenic process) is acting as the catalysing or impeding agent, whether
it is a catalysis or impedance relationship, and which triggering
relationship identified in Sect. 4.2 is being catalysed or impeded (e.g.
A
In Sect. 4, we discussed three different interaction relationships
(triggering, increased probability, catalysing/impedance) between specific
natural hazards, anthropogenic processes and technological hazards/disasters.
However, in addition to having a paired relationship (e.g. one primary
natural hazard triggering a secondary natural hazard) these interactions can
be joined together to form a network of hazard and/or process interactions.
For simplification of language, we will call these
In Sect. 5.1 we introduce four case study examples of networks of hazard interactions: one example from Nepal and three from Guatemala. In Sect. 5.2 we illustrate the wide variation in spatial and temporal extent, frequency and impacts of such networks of hazard interactions, using three of these case studies. In Sect. 5.3 we then use our hazard interaction matrix and hazard/process flow diagrams to visualise networks of hazard interactions, using two of these case studies and three theoretical examples. Finally, in Sect. 5.4, we discuss why we believe evaluating networks of hazard interactions is important.
Networks of hazard interactions are relevant in many locations around the
world. Guatemala is an example of a location where multiple different
networks of hazard interactions can be identified. We have identified
examples of the wide range of hazards and processes in Guatemala using 21
semi-structured interviews with Guatemalan hazard professionals and personal
field observations, during 2 months of fieldwork in 2014.
Specific natural hazards: earthquakes, volcanic eruptions,
landslides, floods, droughts, tropical storms, extreme temperatures,
subsidence, ground collapse and wildfires. Relevant anthropogenic processes: deforestation, inadequate drainage,
agricultural practices and building/road construction practices. Technological hazards/disasters of relevance: structural collapses, urban
fires, chemical pollution and transport accidents.
Specific hazards or processes influencing Guatemala may last for decades
(e.g. eruptive activity of Santiaguito; Bluth and Rose, 2004) or days
(e.g. Tropical Storm Agatha; Stewart, 2011), impacting large areas (e.g.
landslides across thousands of square kilometres; Harp et al., 1981) or small areas (e.g.
20 m ground collapses; Stewart, 2011). A wide range of possible interactions
exist in Guatemala between specific natural hazards, anthropogenic processes
and technological hazards/disasters. Here we present four case study
examples of networks of hazard interactions, with three examples from
Guatemala and one additional example from Nepal, ordered according to their
use in subsequent sections.
Many other examples of networks of hazard interactions (cascades) can be
observed in the published scientific literature, technical reports, press
releases and other forums. It is beyond the scope of this article to compile
a comprehensive list of these cascades; however, many can be found in the
references noted at the end of this article. We proceed to use the four case
study examples outlined above, together with three further theoretical
examples, to illustrate two important concepts relating to networks of hazard
interactions.
In the example case studies described in Sect. 5.1, we observe variation in
the spatial and temporal extent, frequency and impact of networks of hazard
interactions. Networks of hazard interactions (cascades) can vary over many
orders of magnitude both spatially and temporally. For example, a tropical
storm (lasting several days) may trigger landslides across a small localised
area or an entire region (e.g. Central America). One of these triggered
landslides may further block a river causing a small, localised flood or
weaken the structural integrity of a dam and cause a large regional flood. We
illustrate the wide variation in spatial and temporal extent, frequency of
networks of hazard interactions and impacts of such networks using Case Study
1 (
In the 2015
The regular eruptions of Santiaguito in Guatemala and subsequent
lahars/flooding (Case Study 2 in Sect. 5.1) also illustrate variation across
spatial and temporal scales. Volcanic activity may extend over a
sub-national, national or multi-national spatial level, and be either
short-lived or persist for many decades. The Santiaguito dome in Guatemala,
for example, has seen unsteady, extrusive activity since 1922 (Bluth and
Rose, 2004), mainly impacting the southwest of Guatemala. Volcanic activity
at Santiaguito, in combination with regular rainfall, results in lahars each
rainy season, which have an impact on the fluvial system at distances of up to
60
Finally, consider the example of Tropical Storm Agatha and the eruption of
Pacaya volcano (May 2010) in Guatemala (Case Study 3 in Sect. 5.1), which also
demonstrates variations in spatial and temporal scale. Tropical Storm Agatha
had an impact across multiple nations within Central America (a scale of
hundreds of thousands of square kilometres). In contrast, one of the secondary
hazards associated with this storm was a localised ground collapse event,
with a diameter of 20
Networks of hazard interactions (cascades) can also vary in terms of their
frequency and impact. For example, they can be observed in low-frequency,
high-impact events such as the 2015
As demonstrated through discussion of these case studies, networks of hazard interactions (cascades) are relevant at diverse spatial and temporal scales, can be both high- and low-frequency events, and have impacts ranging from fatalities to impacts on livelihoods.
Given the prevalence of networks of hazard interactions, we consider here how
these networks can be visualised to support multi-hazard assessments of
interacting natural hazards. In this section we present two ways of
visualising networks of hazard interactions, using Case Study 2
(
An example of a network of hazard interactions (a cascade system)
(from Gill and Malamud, 2014) using a hazard interaction matrix. A
Two examples of networks of hazard interactions (cascade systems)
using a hazard interaction matrix. Hazard interaction networks based on (top)
the 1976 Guatemala earthquake sequence and (bottom) lahar-triggered flooding
associated with Santiaguito, Guatemala. Both network examples are placed on a
In Gill and Malamud (2014), we developed one method of visualising networks
of hazard interactions through the use of a
Using the hazard interaction matrix visualisation framework illustrated in
Fig. 4, we can also represent two of the case study examples introduced in
Sect. 5.1. Figure 5 shows two examples of networks of hazard interactions
(cascades), both from the southern Guatemala Highlands. Figure 5 (top)
visualises some of the hazards and hazard interactions relevant to the 1976
Network of hazard interactions (example A) using a hazard/process flow diagram. Using the visualisation frameworks constructed in Figs. 2 and 3, an example of an interaction network (cascade) can be presented. Three hazard/process groups are included: (A) natural hazards, (B) anthropogenic processes and (C) technological hazards/disasters. Arrows are used to illustrate interaction relationships, with both triggering relationships (thick block arrows with solid outlines) and relevant catalysing/impedance relationships (thin block arrows with dashed outlines). For clarity of communication, those catalysing/impedance relationships not of relevance to the specific example are not included. See caption explanations in Figs. 2 and 3 for further details. Arrows within the example network of hazard interactions are labelled (1–4) and shaded dark blue to highlight the relevant pathway. In this example, a primary anthropogenic process catalyses (1) the triggering relationship between a primary and secondary natural hazard (2), with the secondary natural hazard then triggering (3) a primary technological hazard, which in turns triggers (4) a primary anthropogenic process to occur.
Network of hazard interactions (example B) using a hazard/process flow diagram. Using the visualisations constructed in Figs. 2 and 3, an example of an interaction network (cascade) can be presented. In this example the network is more complex than in example A (Fig. 6), with three branches and five interaction relationships highlighted here. Three hazard/process groups are included: (A) natural hazards, (B) anthropogenic processes and (C) technological hazards/disasters. Arrows are used to illustrate interaction relationships, with both triggering relationships (thick block arrows with solid outlines) and relevant catalysing/impedance relationships (thin block arrows with dashed outlines). For clarity of communication, those catalysing/impedance relationships not of relevance to the specific example are not included. See caption explanations in Figs. 2 and 3 for further details. Arrows within the example network of hazard interactions are labelled (1–5) and shaded dark blue to highlight the relevant pathway. This example shows a primary natural hazard triggering (1) a primary technological hazard, which in turn triggers (2) a primary anthropogenic process. The same primary natural hazard may trigger (3) a secondary natural hazard. This secondary natural hazard could then trigger (4) a primary technological hazard and (5) tertiary natural hazards.
The hazard/process flow diagram visualisations previously introduced in Sect. 4 (Figs. 2, 3) can also be used to represent complex networks of hazard interactions involving a mixture of natural hazards, anthropogenic processes and technological hazards/disasters. We use the structure of Figs. 2 and 3, with appropriate replication within the same figure to allow for longer and more complex networks of hazard interactions, and give two theoretical examples (A and B, described further below) in Figs. 6 and 7 of a complex network of hazard interactions. The two hazard/process flow diagram examples in Figs. 6 and 7 show all possible triggering interactions (thick block arrows with solid outlines) and (for simplification) only relevant catalysing/impedance interactions (thin block arrows with dashed outlines). Possible networks of hazard interactions are visualised using light-blue boxes to highlight the relevant hazards/processes (i.e. nodes within a network), and dark-blue arrows to highlight the relevant interactions (i.e. links within a network).
The overlay of networks of hazard interactions from case studies in Sect. 5.1 on hazard interaction matrices (Figs. 4, 5), and the overlay of theoretical examples on hazard/process flow diagrams (Figs. 6, 7) can be complemented by other visualisation techniques. For example, when a quantitative evaluation of possible outcomes of interaction relationships is possible, probability trees can be used to assess networks of hazard interactions (e.g. Neri et al., 2008; Marzocchi et al., 2009; Neri et al., 2013). Probability trees are used to visually represent the possible outcomes of an event and add associated probabilities. All three methods are useful for communicating information about specific chains of events. The two visualisation techniques that we have presented here, together with existing probability trees, allow simple and more complex networks of hazard interactions to be evaluated and visualised.
We believe that the assessment and visualisation of possible interaction networks (cascades) within multi-hazard methodologies is of importance to both the theoretical and practical understanding of hazards and disaster risk reduction. Here we outline three principal reasons for identifying possible interaction networks.
An analysis of past occurrences of hazards and disasters shows that interaction networks are often part of the structure of disasters. The need to better match observed reality, by including interaction networks, is applicable to events of diverse spatial and temporal extent, frequency and impact, as has been discussed in Sect. 5.2. The frequency of occurrence of specific networks of hazard interactions demonstrates that more could be done to understand and characterise them. Following the 2015 Gorkha (Nepal) earthquake, the European Geosciences Union (EGU) issued a statement (EGU, 2015) calling for a multi-hazard, integrated approach to risk assessment and the management of natural hazards. This statement also notes the need for agreement within the geoscience community on how to model cascades of natural hazards. This call joins many previous calls (Delmonaco et al., 2007; Kappes et al., 2012; Marzocchi et al., 2012; Gill and Malamud, 2014; Liu et al., 2016) encouraging the assessment of interacting natural hazards and their integration into multi-hazard methodologies. Assessing interaction networks is therefore important as they are a fundamental part of hazard and disaster events.
Example of vulnerability changes within a network of hazard interactions (cascade). A representation of changing vulnerability during a hazard cascade, where the magnitude of vulnerability is proportional to the size of the box. Following a disaster event, pressures on society, infrastructure and coping capacity are likely to be increased, and thus the vulnerability of a community and its systems/assets to further shocks or hazards may increase.
As one progresses along a network of hazard interactions (cascade), aspects of social and/or physical vulnerability may change following the occurrence of a specific natural hazard, anthropogenic process or technological hazard/disaster. If there is a succession of hazard events (i.e. a network of hazard interactions), there may be progressive changes in vulnerability during this succession. While some aspects of vulnerability may remain at the same level before and after the occurrence of a specific event, it is also possible that other aspects of vulnerability may increase as pressure is placed on society and infrastructure, thus reducing coping capacity or decrease. Other aspects of vulnerability could also decrease, especially if there are significant time intervals between successive events in a cascade. This could, for example, help facilitate a growth in community awareness and preparation.
This changing vulnerability within a network of hazard interactions can be represented visually, as shown in Fig. 8, where a series of three hazard events occur in succession and an assumption is made that each hazard event will increase subsequent levels of vulnerability. Before and between these three hazard events, a representation of vulnerability is given, where we illustrate the vulnerability magnitude as proportional to the height of the rectangle. Figure 8 shows the dynamic nature of vulnerability as one progresses along a network of interacting hazards. In this representation, we have assumed that there are increases in vulnerability as the chain of events progresses, but we note that this will not always be the case. On the ground these changes to social and physical vulnerability may be observed in different ways. For example, buildings may have sustained significant damage so that they are more likely to collapse during an aftershock. Hospitals may be at maximum capacity following an earthquake and therefore not able to respond effectively if a subsequent typhoon results in further casualties. Injuries sustained by a community during an earthquake may mean that they have a reduced capacity to evacuate if a tsunami is subsequently triggered.
These examples demonstrate that existing assessments of vulnerability may rapidly become out of date following a hazard event. The identification of possible interacting hazard networks in a given region would allow improved planning of possible changes in vulnerability during successive events. In turn, this could help to improve preparedness efforts.
In addition to the risk reduction benefits that come from the last two points, understanding how chains of interacting hazards are initiated and propagated may help determine how to invest resources to minimise disruption should a specific network of interacting hazards occur. Scientific and management efforts can be focused on (i) preventing the initiation of interaction networks and (ii) reduce or eliminate specific interactions along the interacting hazard network. It may not always be possible to prevent an initial primary hazard from occurring, but sensible investments in structural and non-structural mitigation measures may reduce the likelihood of specific networks of hazard interactions propagating. While we cannot currently prevent a tropical storm from forming and hitting land, for example, measures may be taken to improve drainage and reduce flooding, reinforce certain slopes that are susceptible to failure, or improve urban management to reduce structural collapses, urban fires and water contamination.
In this research and commentary article, we have sought to
advance the understanding of enhanced multi-hazard frameworks, which we
believe to be of relevance to improved Earth systems management. We advocate
an approach that goes beyond multi-layer single-hazard approaches to also
encompass interaction relationships and networks of interactions (cascades).
This study has described this integrated approach, noting that to do
otherwise could distort management priorities, increase vulnerability to
other spatially relevant hazards or underestimate risk. The development of an
enhanced framework to assess and characterise interactions and networks of
interactions first required a description of three principal groups of
hazards/processes, including natural hazards, anthropogenic processes and
technological hazards/disasters. These three groups can interact in a range
of different ways, with three interaction relationships discussed in the
context of this article: triggering relationships, increased-probability
relationships, and catalysis/impedance of other hazard interactions. In
addition to those circumstances where one stimulus triggers one response, it
is highly likely that more than one of these interactions can be joined
together to form a network of interactions, chain or cascade event. We have
developed enhanced frameworks to visualise these interactions and networks of
interactions (cascades) in two different ways (hazard/process flow diagrams
in Figs. 2, 3, 6 and 7, and hazard interaction matrices in Figs. 4 and 5).
These frameworks, visualisations and associated commentary
reinforce the importance of enhanced multi-hazard approaches,
integrating hazard interactions and networks of interactions to better model
observed dynamics of the Earth system; offer a more holistic approach to assessing hazard potential,
helping to improve management of those aspects of the Earth system that are
relevant to disaster risk reduction; support the research community to consider future research
directions in the context of multi-hazard research in regional settings.
Better characterisation and integration of interactions and networks of
interactions into multi-hazard methodologies can contribute to an improved
theoretical and practical understanding of hazards and disaster risk
reduction.
This research was funded by a studentship grant from NERC/ESRC (grant: NE/J500306/1). The authors wish to thank INISVUMEH (Guatemala) and CONRED (Guatemala) for their assistance in the field when reviewing case studies. The authors also wish to thank Reik Donner (Potsdam Institute for Climate Impact Research, Germany) and one anonymous reviewer for their helpful and constructive reviews, and Christian Franzke for his support as editor for this paper. Edited by: C. Franzke Reviewed by: R. V. Donner and one anonymous referee