Developing comprehensive and evidenced regional interaction frameworks requires diverse evidence to improve the systematic identification of relevant hazards and interactions. From our experience, in Table 2, we outline examples of different types of evidence to help construct and populate regional interaction frameworks and group these into five categories: (1) publications and other reports, (2) social and other media, (3) field evidence, (4) stakeholder engagement, and (5) miscellaneous. Some overlap exists between these categories, and not all the examples given are relevant in any given location.

In this paper, we use many of the examples of evidence in Table 2 to help construct and populate a regional interaction framework for Guatemala. We group our evidence into five broad types (A–E below) and use these five types throughout this paper. A.

International literature (publications and reports). This is a comprehensive synthesis of literature describing natural hazards in Guatemala and their interactions. This combines examples of both (1) publications and other reports and (2) social and other media, from Table 2, including peer-reviewed material, technical reports, databases, and media reports (93 peer-reviewed and 76 grey literature sources; Sect. 2.2).

Other evidence types (e.g. historical records or community knowledge) are included in the peer-reviewed and grey-literature publications we examined or may be particularly pertinent in other geographical locations. The use of multiple evidence types (vs. a reliance on one evidence type) facilitates a more comprehensive characterisation of hazards and hazard interactions. For each evidence type considered, we do not use all possible examples, methods, and sources; rather we use examples of key case studies from regions of interest. Collecting and interpreting this evidence requires engagement with a range of organisations engaged in research and practitioner work relating to natural hazards, disaster risk reduction, and disaster response. In Sect. 2.2 to 2.6 we characterise our data (evidence types) and the methods used to collect and unite this to address our research questions. We outline limitations associated with this evidence and the methods used to collect it in Sect. 2.7. In Sect. 2.8, we summarise how we integrate evidence types to develop our regional interaction frameworks in Sect. 3.

Internationally accessible publications and reports includes both peer-reviewed and grey literature, such as journal articles, edited volumes, master's and PhD theses, textbooks, technical reports, databases, and NGO disaster situation reports. This compilation of literature includes reports on hazard events in specific geographic regions, providing evidence of hazard interactions. For example, Rose et al. (2004) presents an edited volume of papers on natural hazards in El Salvador, and ReliefWeb (2018) presents a disaster situation report on the impact of Tropical Storm Nate in Central America. We identified multiple publication and report types with information about Guatemala. We prioritised literature giving a broad overview of natural hazards, synthesising multiple texts, or characterising hazard interactions. It is beyond the scope of this study to examine publications on every aspect of hazards in Guatemala or to review all publications on any one aspect of a hazard.

We primarily accessed literature using large Web databases (Google Scholar and Web of Science) for peer-reviewed articles and general online searches for other grey literature (e.g. media reports). We used Boolean search methods, including both “Guatemala” and keywords associated with a preliminary list of 21 natural hazards (from Gill and Malamud, 2014). For example, “earthquake”, “aftershock”, “seismic”, “tremor”, and “liquefaction” were searched for alongside “Guatemala” and “Central America” to identify relevant material. We evaluated results to determine their relevance and identify other keywords. We also identified specialist books, such as an edited volume on the geology of Central America (Bundschuh and Alvarado, 2007).

We examined literature in a systematic manner, collating references, maps, and figures for 17 (of the 21) natural hazards: earthquake, tsunami, volcanic eruption, landslide, flood, drought, regional subsidence, ground collapse, soil (local) subsidence, ground heave, storm, tornado, hailstorm, lightning, extreme temperature (heat), extreme temperature (cold), and wildfire. Snow avalanche and snowstorm have limited spatial relevance to Guatemala, and geomagnetic storm and impact event have little country-specific (vs. generically relevant) information. For each hazard considered, we cross-referenced diverse literature to characterise it at a level of detail appropriate to this study, including information on spatial and temporal distribution, triggering relationships, and impacts. We identified and used 169 sources as evidence, with 93 (55 %) of these being peer-reviewed and 76 (45 %) of these being grey literature.

Another evidence type to inform the development of regional interaction frameworks is locally accessible reports, such as government or NGO bulletins, newspapers, and emergency call-out records. Civil-protection information bulletins and newspapers can both give a focused overview of natural-hazard occurrences (e.g. Guzzetti et al., 1994; Trimble, 2008; Raška et al., 2014; Taylor et al., 2015), providing information on hazard interactions or noting triggering relationships.

In Guatemala, we use Spanish-language civil-protection information bulletins from the CONRED. Bulletins are issued when there is a threat to lives, livelihoods, and infrastructure and include information on hazards, their spatial and temporal extent, and their impacts, including triggering other hazards. Natural hazards occurring in remote regions or having a very low impact (e.g. very small landslides) are unlikely to be included in bulletins, and therefore bulletins do not provide a complete record of events. CONRED may issue multiple bulletins per day, depending on the evolution of, for example, a weather system or a disaster event. Bulletins are distributed to a mailing list of personnel, with some on their website (CONRED, 2018b) and ReliefWeb (2016). At the time of writing, CONRED bulletins were not systematically archived online. We therefore classify these bulletins as locally accessible.

CONRED made 291 information bulletins available to the authors (electronic format), published over a 127 d period between the 11 June and 15 October 2010, of which 24 were corrupt files and 267 were accessible. Based on their numbering, we believe that CONRED published 413 bulletins during this 127 d period. Additional information that characterises these bulletins is included in the Supplement (Table S1). We searched the 267 accessible bulletins for keywords, placing these into context by looking at the surrounding sentences. Taylor et al. (2015) used this approach to enrich the UK national landslide database by examining newspaper archives.

We selected and used the following six keyword verbs connecting two hazard types and suggesting an interaction between them (with an abbreviated Spanish verb base in parentheses): to trigger (desenca), to provoke (provoc), to generate (genera), to cause (caus), to produce (produ), and to catalyse (catal). We performed a keyword Boolean search in Spanish using the abbreviated form of the verb base to ensure the return of multiple derivatives of the verb. To check if there were other verbs of interest, we then searched for the following hazard keywords in Spanish form (both singular and plural): seismic, earthquake, volcano, eruption, landslide, flood, collapse, sinkhole, hurricane, storm, tsunami, drought, tornado, wind, and rain. We also searched for references to three active volcanoes (Pacaya, Santiaguito, and Fuego) in Guatemala. From these hazard keywords and three volcanoes, we looked for any further interaction verbs that might be included near these words and identified no additional keyword verbs. The number of keyword search results for each of the six keyword abbreviated verb bases connecting two hazard types are as follows: to trigger (desenca; 0 results), to provoke (provoc; 26 results), to generate (genera; 58 results), to cause (caus; 22 results), to produce (produ; 37 results), and to catalyse (catál; 0 results). In some cases, the same bulletin had more than one result.

In total, there were 143 results from 95 CONRED bulletins prior to us processing them based on their relevance to the theme of hazard interactions. By examining the context, we determined that 39 of the 143 results (from 36 different bulletins on 28 unique days) described unique events where interactions occurred between natural hazards. These results are presented in Table S2. The results in this section, although based on an incomplete dataset, demonstrate examples of the range of types of interactions that could occur. Further research could use a larger sample of bulletins to better characterise interactions in Guatemala or an event database such as EM-DAT (CRED, 2018). This would be necessary if the frequency of different types of events were a consideration, with a 4-month period being too short to analyse this.

Field observations can also help with understanding the relevance and dynamics of hazards and hazard interactions. For example, Havenith et al. (2003) describe field evidence of earthquake-triggered landslides in the northern Tien Shan of Kyrgyzstan. Approaches include reconnaissance visits to improve contextual understanding of the region; detailed geological, geomorphological or hazard mapping; and the application of technologies such as rain gauges, drones, and thermal-imaging infrared cameras.

In Guatemala, from January to March in 2014 (9 weeks total), the lead author visited regions in the southern Guatemalan Highlands (identified in Fig. 1) affected by multiple natural hazards and anthropogenic activity. This helped with familiarising the non-Guatemalan authors with the features of key locations and hazards in Guatemala, but primary field data (e.g. community interviews) were not gathered. Observing the spatial and temporal scales at which hazards and anthropogenic processes act enhanced understanding of Guatemala's multi-hazard environment. It also enriched interviews with expert participants (described in Sect. 2.5), with the interviewer making better use of examples, local places names, and descriptors used by participants to characterise and evidence natural-hazard interactions in Guatemala.

The lead author conducted multiple field visits alongside INSIVUMEH, with support from the University of Bristol, and one field visit with CONRED. This helped with developing constructive relationships, establishing the mutual trust and respect required for subsequent data-rich interviews (Kitchin and Tate, 2000). Examples of principal field locations and relevant interactions are (i) Lake Atitlán (e.g. tropical storms triggering landslides, landslides triggering flooding, and landslides triggering lake tsunamis), (ii) Fuego volcano (e.g. lahars triggering floods), and (iii) Santiaguito volcano (lahars triggering flooding).

Interviews provide additional evidence to construct and populate regional interaction frameworks. Participants often come from diverse backgrounds, with differing understanding of natural hazards and geographic regions. Participants with relevant evidence can include hazard and disaster professionals and local people who might be impacted by hazards (e.g. farmers, local government, and communities). Selecting participants based on their experience and relevance to a research question (purposeful sampling) can result in data-rich interviews (MacDougall and Fudge, 2001; Longhurst, 2003; Suri, 2011; Palinkas et al., 2015). Semi-structured interviews provide one means by which to have this dialogue, with questions focused on hazards and hazard interactions. This style gives enhanced freedom in exploring areas of interest and pursuing emerging lines of enquiry (Qu and Dumay, 2011).

Prior to stakeholder engagement in Guatemala in 2014, we obtained ethics approval (reference REP(GSSHM)/12/13-18) from King's College London for research with human participants. At the start of each interview (conducted by the authors Joel C. Gill or Joel C. Gill and Bruce D. Malamud) we explained the purpose of our work and sought informed, prior consent to use data generated. All participants gave permission for us to use their data and identify their institution unless this would identify the individual. We interviewed 21 hazard and civil-protection professionals in Guatemala during 19 interviews. Table S3 characterises the interview participants. Participants came from academia, the private sector, INSIVUMEH, and CONRED. We selected interview participants from diverse professional backgrounds in terms of hazard speciality (e.g. earthquakes, landslides, and floods) and engagement in the disaster cycle (e.g. early warning, mitigation, and recovery). We identified contacts before travelling to Guatemala through their online profiles and professional engagement in other projects and through introductions once in Guatemala.

Interviews ranged from 30 to 120 min, following a semi-structured approach (Longhurst, 2003; Qu and Dumay, 2011). Interviews included opportunities for participants to talk about (i) their background and training, (ii) their consideration and use of information on hazard interactions, (iii) examples of existing networks of multi-hazard interactions, and (iv) hazard interaction visualisations. All interviews aimed to cover these key themes; however there were differences in the order in which they were introduced and the specific questions asked. Interviews were conducted in Spanish (with a translator), in Spanish (without a translator), and in English, depending on the context. To reduce possible power relations, we ensured that participants were at ease with the interviewer and participant setting (Kitchin and Tate, 2000; DiCicco-Bloom and Crabtree, 2006; Qu and Dumay, 2011).

Table S4 presents key statements relating to natural hazards, hazard interactions, and anthropogenic processes, extracted from these 19 semi-structured interviews. Multiple participants highlighted specific interaction examples. These include ones already noted in accessible peer-reviewed or grey-literature sources (e.g. lahars from Santiaguito triggering flooding; Harris et al., 2006) and interactions not described in other evidence types (e.g. Pacific coastal flooding due to simultaneous high tides and river sedimentation).

Workshops are another form of stakeholder engagement and are designed to generate data through activities and focused discussion. We organised a 3 h workshop in Guatemala involving 16 civil-protection professionals at CONRED. Participants included senior and junior staff working in diverse departments. Table S3 characterises the 16 workshop participants (of whom 4 also took part in interviews), with all giving permission for us to use their data in an anonymised form. During our workshop, participants independently completed two tasks.

Task 1: network linkage diagram for 21 natural hazards (16 participants). Participants used this to record triggering relationships that they believed to be relevant to Guatemala. We did not expect any participant to map out all relevant interactions.

Using all 16 completed network linkage diagrams (Task 1 above), we can represent the combined knowledge of the workshop participants and use this as evidence when constructing regional interaction frameworks for Guatemala. In Fig. 2, we overlay evidence from 16 completed network linkage diagrams on a blank interaction framework, showing the number of participants (out of 16) proposing each triggering relationship. This interaction framework has 21 primary natural hazards on the vertical axis and the same 21 secondary natural hazards on the horizontal axis, using an approach laid out in Gill and Malamud (2014).

Of a total possible 441 (21×21) interactions, there are 86 different interactions proposed in Fig. 2 as being relevant in Guatemala (by 1–16 participants), equivalent to 20 % of the 441 possible interactions. Consequently, 355 interactions (80 % of the 441 possible interactions) were determined by all 16 participants as not being relevant in Guatemala. Some of the proposed interactions may not be relevant (false positives), and others not proposed by participants may be relevant (false negatives) in Guatemala. We present detailed statistics resulting from Task 1 and 2 of this workshop, and an analysis of the resultant hazard interaction matrices, in the Supplement (p. 11–16). The results of these two workshop tasks highlight different opinions on which hazard interactions are relevant in Guatemala. There is strong consensus on the occurrence of some interactions but weak consensus on others. The workshop results demonstrate the need for communication across hazard disciplines and the value of comprehensive, systematic, and evidenced frameworks to enhance understanding of relevant interactions.

Evidence types A–E, characterised in Sect. 2, are each associated with limitations and uncertainties. i.

Information accuracy. Based on working with blended sources of grey-literature evidence, we found that it can sometimes be difficult to verify information sources, including media articles and textbooks, civil-protection bulletins, and personal perspectives offered through interviews and workshops. Where possible, we evaluated authenticity by cross-referencing grey and older literature with peer-reviewed and recent literature. Including grey-literature, however, broadens the scope of reviews and provides comprehensive access to available published evidence (Mahood et al., 2014).

The construction of comprehensive and systematic regional interaction frameworks requires three components for a region of interest, each bringing together diverse strands of evidence and unifying them within a formal structure, supported by expert knowledge (Neri et al., 2008): i.

information on relevant single hazards and appropriate ways to classify these, using the evidence in Sect. 2.2 to 2.6 and the classification of 21 natural hazards in Gill and Malamud (2014);

In Table 3, we explore, in the context of Guatemala, six themes laid out by Gill (2016) to guide the generation of regional interaction frameworks: spatial scale, temporal scale, likelihood–magnitude relationships, selection and classification of natural hazards, identifying relevant hazard interactions, and visualisation style and user communities. We integrate perspectives from hazard and civil-protection professionals in Guatemala (from semi-structured interviews and the workshop; see Sect. 2.5 and 2.6). Professional organisations have an understanding of local culture, language, and knowledge and have the mandate to adapt interaction frameworks into suitable forms for other stakeholders (e.g. policymakers and communities).

Gill and Malamud (2014) propose a broad classification of 21 natural hazards into six hazard groups (geophysical, hydrological, shallow Earth, atmospheric, biophysical, and space). This, or an alternative, comprehensive classification can be adapted to develop a regionally specific classification, using available evidence. We use this approach to propose a detailed, location-specific classification of natural-hazard types in Guatemala, building on evidence in Sect. 2. We begin by identifying which of the 21 natural hazards listed in Gill and Malamud (2014) are relevant in Guatemala and sub-divide selected hazards where evidence supports an expanded classification. We present our evidenced classification scheme in Table 4, including six natural-hazard groups, 19 hazard types, and 37 hazard sub-types. We also include an indication of the evidence types supporting this classification, using identifying letters A–E introduced in Sect. 2.1 and specific referenced publications and reports where appropriate. The 37 detailed natural-hazard sub-types in Table 4 help with improving the detail by which we can characterise interactions in regional interaction frameworks (e.g. see Sect. 3.3).

Our classification is one way of grouping relevant natural hazards, with alternative classifications possible. Other natural-hazard types may exist in Guatemala that have been missed from our classification, including those occurring less frequently or having a smaller impact than those we consider. We reduce the likelihood of missing key hazards by reviewing multiple evidence types to ensure a comprehensive and evidenced classification. We include 26 to 32 more hazard sub-types than existing regional interaction frameworks (e.g. Tarvainen et al., 2006; Kappes et al., 2010; Liu et al., 2016). In addition to the 37 natural-hazard sub-types in Table 4, we could also consider how a changing climate influences natural hazards (see McGuire and Maslin, 2012, for a full discussion) or include other hazard groups or processes, such as biological hazards (e.g. epidemics), technological hazards (e.g. structural collapse), or anthropogenic processes (e.g. vegetation removal). The latter are discussed in Sect. 3.5.

Building upon the reflections in Sect. 3.1, and using the hazard classification in Sect. 3.2 and evidence in Sect. 2, we now construct and populate interaction frameworks for two different spatial extents in Guatemala:

National spatial extent (Sect. 3.3.1). We produce a 21×21 interaction framework (matrix form), with 16 relevant primary hazards and 15 relevant secondary hazards. We initially constrain interactions for a national spatial extent using the Gill and Malamud (2014) hazard classification (21 hazard types).

In addition to one hazard triggering or increasing the probability of another hazard, longer linear or non-linear networks of multi-hazard interactions (or cascades) can also occur (Han et al., 2007; Choine et al., 2015; Gill and Malamud, 2016; Pescaroli and Alexander, 2018). These network events have both spatial and temporal dimensions, include both high- and low-likelihood events, and have diverse impacts. Ciurean et al. (2018) outline a range of methods for qualitatively and quantitatively characterising such multi-hazard interaction network events. For example, event scenario trees can be used to assess the probabilities of specific hazard cascades (Neri et al., 2008, 2013).

The evidence we present in Sect. 2 includes many examples of multi-hazard interaction network events. For example, the internationally published literature characterising the 1976 Mw=7.5 Guatemala earthquake clearly articulates a set of triggered hazards. After the earthquake, there were multiple aftershocks and movement on other faults close to Guatemala City as well as rapid subsidence or ground collapse (Espinosa, 1976; Plafker et al., 1976). The earthquake triggered more than 10 000 landslides, rockfalls, and debris flows, blocking vital transport routes (Plafker et al., 1976; Harp et al., 1981) and blocking rivers to trigger upstream flooding (Plafker et al., 1976; Harp et al., 1981). Breaches of these landslide dams also resulted in further flooding (Harp et al., 1981).

The civil-protection bulletins characterised in Sect. 2.3 also include several examples of events consisting of multi-hazard interactions in Guatemala. These include events with primary, secondary, and tertiary hazards as well as events reporting primary hazards changing the likelihood of future hazards. Table 6 gives three diverse examples of network events derived from Table S2, demonstrating the complexity of hazard interaction networks in Guatemala. Table 6 also includes a simple visualisation of each example, showing the range of hazards and interaction relationships: i.

linear events where one primary hazard triggers one secondary hazard which triggers one tertiary hazard (Example 1),

In Sect. 3.2 to 3.4, we primarily consider interactions between natural hazards; however, anthropogenic processes can also trigger natural hazards and influence natural-hazard interactions (Glade, 2003; Knapen et al., 2006; Owen et al., 2008; Gill and Malamud, 2017). Information on relevant anthropogenic processes can support hazard and civil-protection professionals in evaluating how anthropogenic activity may trigger hazards and influence hazard interactions.

Using a classification of 18 anthropogenic processes (Gill and Malamud, 2017), and evidence from Sect. 2, we identify 17 relevant anthropogenic processes in Guatemala, listed in Table S6. Some of these processes are only relevant for small spatial extents (e.g. individual towns), with others being more widespread (e.g. in many populated regions). Table S6 includes the evidence types (A–E) from Sect. 2 used to justify their relevance to Guatemala. Some anthropogenic processes feature multiple times within one evidence type. For example, four interviewees noted road construction (infrastructure construction: unloading) and four noted deforestation (vegetation removal) in the context of triggering landslides. In contrast, only one interview participant mentioned groundwater abstraction as a potential trigger of subsidence.

The spatial and temporal relevance of these 17 anthropogenic processes will vary and could change over time. Anthropogenic processes can start and stop and both grow and shrink in their spatial extent. The anthropogenic processes in Table S6 should be regularly reviewed to assess their relevance, if other processes have started, and if there are any consequences of this variation on natural hazards and hazard interactions. For example, increased road construction may change the likelihood of landslides during heavy rain.

We have integrated diverse evidence types regarding hazards and hazard interactions in Guatemala and unified them in a formal structure, supported by expert knowledge. We have collated information on relevant single hazards and appropriate ways to classify these in Guatemala and information on relevant hazard interactions. Using a comprehensive and systematic approach, we have constructed evidenced national and sub-national interaction frameworks in matrix form, considering hazard interaction networks and relevant anthropogenic processes. We have demonstrated that our approach is scalable (with national and sub-national applications described) and therefore suggest that it is reproducible in diverse geographical contexts and at multi-national to local scales. Regional interaction frameworks provide a comprehensive overview of potential hazard interactions that allow agencies responsible for hazard monitoring and response to assess if current disaster risk reduction and response strategies, and communication and collaboration mechanisms, can be enhanced to recognise the complexity represented.

The interdisciplinary, multi-method approach we have laid out in Sects. 1 to 3 is scalable and can be applied in diverse geographical settings to generate a comprehensive, systematic, evidenced review of potential hazard interactions. A synthesis of available evidence in any given context (e.g. multi-national, national, and sub-national) is necessary for underpinning the construction of regional interaction frameworks. Our approach first develops an extensive location-specific hazard classification and then populates a customised matrix with information about relevant hazard interactions. This contrasts with many existing studies of multi-hazards which are often focused on the layering of single hazards but not looking at the potential interactions. When potential hazard interactions are considered, most studies are not systematic and are selective about which hazards they include. The studies often do not describe the evidence for including or excluding certain hazards or interactions between hazards. The regional interaction frameworks we present in Sect. 3 include 21 to 33 natural hazards, compared to 6 to 11 natural hazards in the examples summarised in Table 1.

Other countries in Central America (e.g. Nicaragua, El Salvador, and Costa Rica) have similarities to Guatemala in their multi-hazard landscape. Their national interaction frameworks would likely be similar, although not identical, to Guatemala. Interaction frameworks for other countries may look very different, shaped by the tectonic and meteorological setting. Regional interaction frameworks can also be developed for sub-national scales, including large geographical domains, municipalities, or localised sites important to the development of critical infrastructure.

We propose that comprehensive, systematic, and evidenced regional interaction frameworks can improve awareness of complex multi-hazard landscapes and assessment of potential networks of hazard interactions, thus informing disaster risk reduction and response strategies. Detailed and evidenced reviews of multi-hazard interactions are a fundamental first step in understanding the complexity of the multi-hazard landscape and therefore understanding risk (Sendai Framework, Priority for Action 1). In particular, regional interaction frameworks can be a powerful tool for scenario discussions between hazard managers and those responsible for single hazard preparedness and response. Through sitting down and discussing together the potential multi-hazard scenarios that may occur, decisions can be made about the preparedness steps required and how different actors would work together to respond. It may be possible to indicate which scenarios have a high likelihood vs. a low likelihood and which could have a large impact vs. a small impact. When the regional interaction frameworks were used by us in this way in Guatemala during a visit of the first author in 2018, some participants questioned the inclusion of particular hazards and/or hazard interactions in the interaction frameworks (e.g. landslides triggering tsunamis). Following discussion of the evidence used to populate the matrix for this scenario, participants reported changes in opinion about the relevance of these interactions and their need for inclusion within planning.

Further examples of how the information within regional interaction frameworks, and generated scenarios, can be used by agencies responsible for hazard monitoring, DRR, and disaster response are as follows.

Scenarios to ensure hazard preparedness and disaster response systems are effective. The occurrence of one hazard (e.g. a volcanic eruption) may result in the movement of people or assets to another region. Ensuring comprehensive awareness within decision-making agencies of how this hazard has changed the likelihood of other hazards (e.g. lahars and landslides) is necessary for ensuring that exposure and vulnerability of displaced people are not increased. Developing and discussing scenarios of triggered hazard scenarios, particularly with diverse single-hazard actors all taking part in the discussion, can help explore dynamic vulnerability between successive hazard events and the steps needed to prevent compounding impacts.

Hazard interactions cut across multiple disciplines and so require input from diverse specialisms (Kappes et al., 2012; Scolobig et al., 2017). Interaction frameworks could therefore help with facilitating enhanced cross-institutional dialogue about hazard interactions and their likelihoods and potential impacts. This could help with strengthening collective knowledge of hazard interactions and the ability of an individual to access this knowledge. By contrasting results from our workshop (Fig. 2) with our Guatemala national interaction framework (Fig. 3), we can examine and quantify congruence between the two matrices and create a new figure. Figure 7 is a 21×21 interaction matrix that combines Figs. 2 and 3 to indicate the number of workshop participants (from a total of 16) that identified an interaction as being relevant to Guatemala (numbers) and the interactions identified within our national interaction framework (grey shading; from Fig. 3).

Figure 7 combines information and knowledge from 16 participants to present something that is “owned” by no individual. It is collective knowledge, combining information and knowledge owned by multiple people (Antonelli, 2000). We do not expect an individual scientist or hazard professional to map out all relevant interactions. Assessing how an organisation rather than an individual understands interactions demonstrates their collective knowledge. For this knowledge to be truly collective, there must be effective communication between participants and a means by which this knowledge can be accessed, shared, and applied (Foray, 2000; Antonelli, 2000; Paton et al., 2008).

Multi-hazard research is complex and requires scientists and professionals operating in many different disciplines. Figure 7 demonstrates large variation in perspectives between participants on hazard interactions. There is a unanimous consensus (i.e. 16 participants) that an interaction exists in 2 (0.5 %) of 441 possible triggering interactions. To assess congruence between the participants' perspectives (numbers in Fig. 7) and national interaction framework (grey shading in Fig. 7), we use Matthews' correlation coefficient, or MCC (Matthews, 1975). MCC values are a function of true positives (TPs), true negatives (TNs), false positives (FPs), and false negatives (FNs) and can be expressed as follows (Matthews, 1975; Powers, 2011): 1MCC=(TP×TN)-(FP×FN)(TP+FP)(TP+FN)(TN+FP)(TN+FN). The MCC gives a value of congruence between -1.0 (zero overlap between the numbers and grey shading in Fig. 7) and +1.0 (perfect overlap between the numbers and grey shading in Fig. 7). MCC = 0.0 suggests that the amount of congruence is no better than a random average (Kaufmann et al., 2012). We use two different approaches: i.

All identified interactions. Where ≥1 people note an interaction to be relevant, we consider this to be part of the group's collective knowledge. From Fig. 7, a total of 86 interactions were identified by the 16 workshop participants. This is compared to 50 interactions in the national framework (Fig. 3).

Matthews' correlation coefficient is a simple indicator of agreement, which we use to examine differences between stakeholder perspectives and our national interaction framework (Fig. 3). When applying a small threshold (≥3 participants agreeing on a given interaction) to determine which interactions were analysed, the collective knowledge of 16 participants generated the closest agreement to the national interaction framework (MCC = 0.51). This MCC is based on 22 (44 %) of 50 interactions in Fig. 3, being identified by ≥3 participants and therefore 28 (56 %) of 50 interactions that ≤2 participants identified in the workshop. Of these 27 interactions identified by ≤2 participants, nobody identified 25 different interactions. These results suggest the following:

Enhanced communication within and across organisations involved in natural hazards and DRR in Guatemala could help when considering hazard interactions. When co-created by diverse stakeholders, interaction frameworks can help with facilitating communication across specialisms engaged in hazard monitoring and civil-protection. Interaction frameworks could also help elicit additional information for characterising interactions, such as which are most likely to occur and which could cause the greatest damage to interaction likelihoods and impacts. Ensuring that collective understanding of hazard interactions is operationalised to greatest effect will require strong institutions and cross-departmental and cross-disciplinary communication (Scolobig et al., 2017).

We lay out an approach in Sects. 1 to 3 that integrates diverse evidence sources from the natural and social sciences through a visual database to give a comprehensive, systematic, and evidenced review of the multi-hazard interactions for a regional spatial extent. We believe that this approach builds on and enhances existing forms of regional interaction framework, such as those described in Table 1. Additional research can further enhance regional interaction frameworks (Sect. 4.3.1), as can better understanding how to embed research outputs into relevant agencies through meaningful stakeholder dialogue (Sect. 4.3.2). Engagement with hazard and civil-protection professionals, academics, the private sector, and intergovernmental organisations in Guatemala informed our development of regional interaction frameworks. Understanding stakeholder requirements (e.g. terminology, spatial scales, and temporal scales) helps to ensure that frameworks are fit for purpose. Draft results were discussed with many of these stakeholders in Guatemala in 2018, prior to publishing. We shared our interaction frameworks through seminars, roundtable discussions, and interviews to document perspectives on (i) the structure and content of the interaction frameworks, (ii) use of the interaction frameworks, and (iii) future research and innovation opportunities. We highlight some of the common themes in the following two sub-sections.