Natural hazards, disaster management and simulation: a bibliometric analysis of keyword searches

3.1 What hazards are the focus?

This analysis shows that natural hazards (73%) are more commonly researched than man-made hazards (14%) (Fig. 4a), and for these man-made hazards terrorism (83%) is the concern of the majority. In general, generic natural hazards (37%) are covered more frequently than any one specific hazard type (Fig. 4a). Generic natural hazards, in this context, are those papers where a specific hazard type is not allocated and instead reference is only made to natural disasters in general. For the specific hazard types identified, the most common research hazards are hydrological (20%) and geophysical (20%), followed by meteorological (15%).

Open image in new window

The papers in this study were limited to years between 2000 and 2017 (as previously discussed). From these papers, the peak number of papers published was during 2003 (Fig. 4b). Research on terrorism only appears in publications from 2001 onwards, the same year which saw the 9/11 USA terror attacks (Fig. 4b). Other than the link to terrorism, there seems to be no distinct pattern with the year of publication and the hazards researched in papers (Fig. 4b). For example, after the Hurricane Katrina event in 2005, there does not appear to be a significantly greater number of papers on meteorological hazards published.

From the 84 papers analysed, 48% of papers included a case study as part of their study. In agreement with the hazards covered in papers, the case studies included also primarily focus on natural hazard disasters (95%), rather than man-made hazards (Fig. 5a). Of the natural hazards covered, earthquakes (28%) and hurricanes/cyclones/typhoons (28%) are the most covered, followed by flooding (25%) (Fig. 5a). Other types of hazards, including landslides (4%), hailstorms (4%) or drought (2%), are not well covered in comparison (Fig. 5a). Generally, these are the hazards which are less prevalent worldwide and therefore the amount and availability of data for case studies are likely to be reduced. The pattern of case study hazards has not changed significantly over time, demonstrating a continual commitment to case studies on earthquakes, flooding and hurricanes/typhoons/cyclones (Fig. 5b).

Open image in new window

3.2 What research is undertaken?

The results of this study have shown that the peak number of publications was seen in 2003 (13%), when the number of journal publications peaked, before gradually decreasing over the subsequent years until 2014 (2%) (Fig. 6a). From the papers analysed, nearly 50% use a review methodology and include little original or novel work (Fig. 6b). However, nearly 40% included some form of modelling technique in the paper, with the remaining 10% of papers consisting of case studies or introductory articles (Fig. 6b).

Open image in new window

The exact study length of paper can be quite difficult to define, particularly in the case of review papers. Previous studies have tended to consider only case study papers and based the length upon the research period alone (Chappin and Ligtvoet 2014; Chappin and van der Lei 2014), with time periods defined as: short < 1 year, medium 1–10 years and long > 10 years. To adapt this technique for review papers, we consider the length of the paper to be based upon the references (e.g. we define the time period as the publication date of the paper itself to that of the earliest reference used). We also need to adapt the time periods used, as if the same criteria are applied we would have no short studies, 42% medium and 58% long. As such, we propose the following criteria: short < 10 years, medium 10–24 years and long > 25 years. Using this new timescale, we find that the analysed journal papers can be classified into: short (17%), medium (63%) and long (20%). Generally, papers with longer reference times tended to be review papers (Fig. 6c).

The papers were split into those that were qualitative (review or case study papers) and those that were quantitative (included modelling). Out of the 84 papers, over 50% were qualitative, which was interesting as one of the search terms was simulation, so it was anticipated there would be a larger number of quantitative papers (Fig. 6d). Out of the quantitative papers, 80% included a case study, whereas 80% of the qualitative papers do not contain a case study (Fig. 6e). Of the different methodologies used, over 80% of the review papers were qualitative and over 75% of the modelling papers were quantitative, which is as expected, for example if a paper uses review techniques it is more likely to be qualitative and vice versa (Fig. 6f).

3.3 Where is research focused?

An important aspect of disaster management is related to the location of threats. Certain locations are more vulnerable to specific hazards or more frequently experience event types. Therefore, there is a need to understand where current research is carried out and whether this correlates with the locations of the most frequent, deadly or costliest natural disasters across the globe. Locations feature in papers in numerous ways, including the author’s country, the research focus of papers and the location of case studies. By looking into these, it is possible to understand whether there is any correlation and if these locations are linked with the key disaster trends.

This study has found that corresponding authors are located across the globe, with the largest proportion in North America (59%), then Europe (18%), Oceania (12%) and Asia (11%) (Fig. 7a).The analysis also considered the focus of papers in terms of location. It was found that the focuses ranged from non-specific worldwide themes (44%) and developing countries (5%), to specifics on the USA (33%), others combined a country and a worldwide emphasis, for example USA/worldwide (10%), and some focused on specific locations (8%) (Fig. 7c). Also, a number of papers included case studies to help demonstrate their findings or for use within modelling techniques; the locations were North America (60%), Asia (18%), Europe (15%) and Oceania (7%) (Fig. 7e).

Open image in new window

The corresponding authors can be used to give an indication of the likely area covered by a paper. It appeared that this may not be the case as over 40% of the papers had a worldwide/generic focus (e.g. they were not specific to any one country or continent). However, over 40% of the papers are specifically linked to either the USA or a combined USA/worldwide approach (Fig. 7c). This demonstrates that for some locations, the author location does influence the focus of the paper (Fig. 7b). It can be deemed that in general if an author is based in the USA, their research will focus on the USA (Fig. 7b). It is to be expected that there would be large amounts of research carried out in the USA, as they are frequently affected by natural hazards and the financial implications in absolute values are large for the country.

The obvious exception to this are the authors who choose to carry out generalised research and to not focus on a specific location. There are also some examples where author location does not affect the research location choice. For example, a UK author researches on a case study in Greece (Fig. 7b), but in contrast this is still part of the same continent.

There are a wide range of case studies included within the papers, with locations across the globe in North America, Asia, Europe and Oceania (Fig. 7e). This location spread is to be anticipated due to the corresponding author locations. Asia is not well covered with case studies (18%), and there are no case studies in Africa. Commonly, it is seen that case studies are carried out in the same locations as authors or in close proximity (Fig. 7d). This is likely to be due to the availability, accessibility or reliability of information and/or data and the local knowledge of authors. Also if there is any requirement for field trips or data collection, this is generally easier to facilitate when in close proximity. Again, the largest number of case studies is focused on North America (60%), tying into the frequency of events and the financial costs (Fig. 7d).

It is also the case that hazards reflect the common hazards seen on a continent rather than extreme or one-off events, for example North America experiences flooding, earthquake and hurricanes, whilst Asia experiences flooding, earthquakes, typhoons and landslides (Fig. 7f). This is to be expected for case studies, as a quantity of reliable data is often required.

3.4 Where to find this research?

The vast majority of research is undertaken in academic institutions and in research facilities across the globe. With large volumes of research produced on different topic areas, it is necessary to understand whether this information is located in the same journals, what impact this research has and whether citations can help to increase the prevalence and impact of the research.

The most prevalent journal in this study was Natural Hazards (10%), Natural Hazards Review (8%) and Disaster Prevention and Management: An International Journal (7%) (Fig. 8). In total, 57 different journals were recorded in the study; out of a total of 84 papers, there were only 11 instances where there were repeated articles in the same journal publication. The highest impact factors do not correspond with the mostly commonly used journals in this study (Fig. 8). It is perhaps not surprising that all of the most prevalent journals in the study contain a keyword from the search criteria, e.g. “natural hazards” or “disaster management”.

Open image in new window

Google Scholar as part of its search results provides a record of the number of citations a paper has received. The majority of the papers in this study have been cited 0–100 times (33%), 101–200 (25%) or 201–300 (11%). This shows that as the number of citations increases, the number of papers achieving the level required decreases; however, there are examples of some journals being cited over 1000 times and one is cited over 3000 times (Fig. 9). It is also interesting to consider the number of citations given by Google Scholar and the position of the journal paper in the Google Scholar search results. This shows that Google Scholar does not identify journal papers on the number of citations alone (Fig. 10). In the search that uses simulation, there are a number of papers with higher numbers of citations in the initial results compared to the “agent based simulation” search where the number of citations is spread more evenly amongst the search position.

Open image in new window

Open image in new window

Of the most prevalent journals, the Google Scholar citations indicate that Natural Hazards has a range between 0 and 300 citations, Natural Hazards Review has a larger range of 0–700 citations and Disaster Prevention and Management has a range between 0 and 500 times (Fig. 11). This shows that journals with a higher number of published papers (in this study) do not correlate with a higher number of citations per paper, according to Google Scholar, nor does having a higher journal impact factor link to a larger number of citations. For example, Environmental Modelling and Software has one paper with 0–100 citations and another with 101–200 citations, whilst in Tourism Management one paper has 101–200 citations and one has 501–600 citations (Fig. 11). However, the exception to this is Global Environmental Change which has one paper cited 1001–2000 times and one cited over 3000 times (Fig. 11), which ties well with the high impact rating. This indicates that the topic and contents of the paper are the driving factors behind citations, rather than the high impact factor of the journal.

Open image in new window

3.5 Keyword search terms

With any journal publication, authors are required to select several keywords to reflect the overall themes of the paper. This can be a good indication of the overarching themes of the work and can be influential in whether a paper appears in a search or not. The keywords from all the papers were collected together as part of this study. It is interesting to note that not all papers provided keywords with 26% not including keywords. From the keywords, the search terms of “natural hazards”, “disaster management” and “simulation” were removed. “Agent based simulation” was not removed, as although this was one of the search criteria, it was decided to determine how prevalent the term was within the keywords.

After removing the search terms, the most prevalent words were “risk” (5%), “modelling” (3%), “flood” (3%), “GIS” (3%), “evacuation” (3%) and “emergency” (3%) (Fig. 12a, c). It is interesting to note that “agent” (1%) was not in the top keywords, even though it had been one of the search criteria for the papers.

Open image in new window

Although the keywords can be useful for determining the themes of a paper, keywords can often be limited to a selection list (set by the journal), meaning that important themes may be missing. Therefore, after analysing the 84 papers, the key themes and terms were extracted by the reader from the papers to determine whether this matched with the author’s keyword selection. The most common terms were “risk” (4%), “vulnerability” (2%), “evacuation” (2%), “emergency” (2%) and “systems” (2%) (Fig. 12b, c). Again, the term agent (0.8%) was not in the top keywords, even though it had been one of the search criteria for the papers. This shows that there is correlation between the author’s selection and those chosen after reading the papers, suggesting that the use of lists for keyword selection does not adversely affect the selection of keywords.

The search terms selected for the literature search were aimed at producing results relating to modelling and simulation techniques in a disaster management context for natural hazards. However, when the search term “simulation” was used, over 60% of the papers included no modelling, and the modelling papers did not relate to agent based modelling (Fig. 12d). Using the search term “agent based simulation”, 50% of the papers generated contained no modelling, 35% contained reference to a modelling technique and only 15% of the papers specifically used agent based modelling (Fig. 12d). There were a number of papers that appeared as results for both “simulation” and “agent based simulation”; of these papers, over 60% had no modelling included, with just under 20% having a modelling technique in and the same amount (< 20%) having a specific mention of agent based modelling (Fig. 12d). Despite the use of specific search terms to extract modelling- and simulation-based papers, the use of these search terms was not particularly successful at producing relevant results.

4.1 Overview

The UN Office for Disaster Risk Reduction (UNISDR) considers disasters as: “a serious disruption of the functioning of a community or a society involving widespread human, material, economic or environmental losses and impacts, which exceeds the ability of the affected community or society to cope using its own resources” (UNISDR 2009). The data in this section have been collated to demonstrate the frequency of natural disasters and how communities have been affected (including fatalities and economic impacts), to enable an exploration of the current disaster trends across the globe. The data have been obtained from the EM-DAT, the International Disaster Database produced by the Centre for Research on the Epidemiology of Disasters (CRED) (Guha-Sapir et al. 2016a, b) and from the “The Human Cost of Disasters: A Global Perspective” report (CRED 2015). For a disaster to be included within the EM-DAT database it must meet one of the following criteria: (1) 10 or more people died, (2) 100 or more people were affected, (3) there was a declaration of a state of emergency or (4) there was a call for international emergency assistance (EM-DAT 2016a, b).

The prevalence of natural disasters has remained relatively static over the past 20 years; however, the number of people affected, fatalities and economic costs continue to grow (CRED 2015). During the study period (1994–2013), the largest number of people, around 600–700 million, affected in 2002, and the second highest in 1998 around 300–400 million people (CRED 2015). Generally, the number of people affected is around 200–300 million people per year (CRED 2015), from which it can be calculated that approximately 2–3% of the world’s population is affected by a natural disaster each year.

The lowest frequency of natural disasters occurs in low-income countries (17%), whilst the other income groups have more of an even split: high income (26%), upper-middle income (30%) and lower-middle income (27%) (CRED 2015). However, the lowest number of fatalities from natural hazards occurs in the high-income group (13%), followed by upper-middle-income group (19%); conversely, the highest fatalities are in the lower-middle-income group (35%) and low-income group (33%) (CRED 2015). Therefore, it could be argued that low-income countries are disproportionately impacted by natural disasters, as with each disaster that occurs the community is severely impacted, causing the development of these countries to be broken by the cycle of recovery.

In terms of economic damage by country’s income group, the economic damage in absolute terms shows that high-income countries experience the largest losses (64%), followed by upper-middle-income countries (26%), whereas low-income countries experience much smaller (3%), followed by lower-middle-income countries (7%) (CRED 2015). Alternatively, if the economic damage is expressed as a percentage of GDP, then for the low-income countries the losses are the greatest (5.1%) compared to the lower-middle-income (0.2%), high-income (0.3%) and upper-middle-income country groups (0.6%) (CRED 2015). Therefore, it could be argued that again low-income countries are disproportionately impacted, as the disasters that occur result in funds being diverted to recovery efforts rather than allowing communities to continue developing; this can also compound debt problems for low-income countries.

4.2 What hazards are there?

Natural hazards are primarily categorised into either hydrological (e.g. flooding), geological (e.g. earthquakes), meteorological (e.g. cyclones) or climatological (e.g. drought). Other types of disasters do occur, and these are predominantly caused by human events, such as war and terrorism. There are also biological events, which are normally a result of nature but can also be caused or aggravated by human actions.

In terms of natural disasters, the disaster types occur in different proportions, affect different numbers of people, cause differing amounts of fatalities and incur different financial costs, as shown in Fig. 13a. The disaster types also affect infrastructure including housing, health facilities and schools differently (Fig. 13b). From these two figures, it is possible to draw several conclusions: hydrological events are the most frequent, geophysical hazards are the deadliest, hydrological events affect the greatest number of people and meteorological hazards are the costliest. In terms of damaged houses and health/school facilities, hydrological events are the worst, but for destroyed health and school facilities the worst is from meteorological events.

Open image in new window

4.3 Where are the disasters?

Natural disasters are experienced across the globe, with Asia experiencing the largest number of disasters followed by the Americas, and then Africa and Europe, and Oceania experiencing the smallest numbers (CRED 2015). In terms of specific countries, India, China, the USA and Philippines have each experienced the largest number of natural disasters, around 243–509 disasters across a 20-year period (CRED 2015).

For the period 2000–2016, there were many high impact individual natural disasters; the costliest of these was the Tohoku earthquake in Japan (March 2011), with estimated damage of $210 billion, followed by Hurricane Katrina in the USA (August 2005) at $125 billion and then the Sichuan earthquake in China (May 2008) with costs of $85 billion (EM-DAT 2016a, b).

4.4 Governmental policy

Governmental policy can be a key driver for change across the globe. Examples of good policy practice can be used to drive forward changes in other nations as they strive to meet the benchmarks set out. In the UK, more effective disaster management has been driven through the Civil Contingences Act (2004), which provides a single framework for civil protection in the UK and introduces the duty to create Local Resilience Forums (LRFs) (Cabinet Office 2013). The LRFs have statutory duties as local authorities to prevent serious damage to their local communities. Each geographical area is based on police force boundaries and is “required to prepare to deliver an appropriate emergency response and to maintain normal services during a crisis” (Newcastle City Council 2014). To help this, risks need to be identified in each area, so community risk registers and frameworks have been set out (Northumbria Local Resilience Forum 2014; London Resilience Partnership 2017).

Once risks have been identified and management plans are in place, the validity of these plans needs to be tested to ensure they are adequate in dealing with the anticipated risks. Therefore, there is a requirement to run emergency planning scenarios. These are included within the Civil Contingencies Act, which states that Category 1 responders must include provision for carrying out exercises and training staff on their emergency plans (Cabinet Office and National Security and Intelligence 2013). Currently, three types of exercises are proposed: (1) discussion based, (2) tabletop and (3) live (Cabinet Office and National Security and Intelligence 2013). Discussion-based exercises are relatively cheap to run and easy to prepare so are often utilised for training purposes. Tabletop drills are based on scenarios, which are useful for validation purposes and exploring weaknesses, with low costs other than staff time, but more planning and preparation is required. Live exercises are a real-life simulation of an event, which is expensive to run, demands very extensive planning and can be disruptive to the general public. Currently, there is no provision to utilise computational modelling for planning and preparation. However, this could provide a more robust method for testing scenarios, allowing planners to test multiple runs without the resource and cost requirements. Previously, models would not have been capable of this but with the emergence of new techniques and additional computer power, it is now possible to test computationally.

The UK’s National Security Strategy states that “the security of our nation is the first duty of government” and that “it is the foundation of our freedom and our prosperity” (Cabinet Office and National Security and Intelligence 2010). Therefore, to supplement the Community Risk Registers produced by LRFs, the UK Government carries out a National Risk Assessment (NRA) annually. This is a classified document; however, the Government also produces an annual publically available version of this document, namely the National Risk Register (NRR) (Cabinet Office 2008). The NRA and NRR were first published in 2008 as a response to the National Security Strategy (Cabinet Office and National Security and Intelligence 2010) with the aim of capturing a range of emergencies that might have a substantial impact on all, or a significant part of the UK. These documents outline the larger, national picture of risks compared with the localised risks considered by the LRFs (Cabinet Office 2008). The 2008 NRR showed that the highest impact event was anticipated to be pandemic influenza, but that the most likely events were attacks on transport or electronic attacks, although the impacts were deemed to be smaller (Cabinet Office 2008). The most recent NRR was produced in 2015; this adapted the previous register and better quantified the likelihoods of risks and impacts, by indicating the relative likelihood of events occurring in the next five years such as “between 1 in 20 and 1 in 2”. The register was split into two parts: risks of terrorist or malicious attacks and other risks (Cabinet Office 2015). In terms of terrorist attacks, a catastrophic terrorist attack was seen as medium–low plausibility but had the highest impact, whereas cyber-attacks compromising data confidentiality are highly plausible but have low impact (Cabinet Office 2015). For the other risks, pandemic influenza has the highest impact and its relative likelihood of occurring in the next 5 years is between 1 in 20 and 1 in 2 (Cabinet Office 2015).

The USA, like the UK, is subjected to a number of natural and man-made hazards every year, for example in 2015 the USA experienced 28 recorded natural disasters with financial costs of approximately $21 billion (Guha-Sapir et al. 2016a, b), so there is a need to develop and improve disaster management strategies. The USA Department for Homeland Security has approached this by forming a national preparedness goal. This is set out as “a secure and resilient nation with the capabilities required across the whole community to prevent, protect against, mitigate, respond to, and recover from the threats and hazards that pose the greatest risk” (US Department of Homeland Security 2015), deeming a shared responsibility across the entire nation (FEMA 2015). Overall, FEMA’s mission can be described as ensuring “that as a nation we work together to build, sustain, and improve our capability to prepare for, protect against, respond to, recover from and mitigate all hazards” (FEMA 2017). The National Preparedness Goal is capability based, with 32 core capabilities (identified as part of a strategic national risk assessment), which are organised into five mission areas, namely prevention, protection, mitigation, response and recovery (US Department of Homeland Security 2015). This has been used to identify the types of threats that posed the greatest risk to the USA’s security, including natural, technological/accidental and adversarial/human-caused hazards. Currently, natural hazards, pandemic influenza, technological hazards, terrorism and cyber-attacks are classified as a significant risk to the USA in their strategic national risk assessment (US Department of Homeland Security 2015).

New Zealand is also susceptible to many natural hazards, including the Christchurch earthquake (February 2011), which caused 65 fatalities and $3 billion worth of damage (BBC 2011). Therefore, to reduce this vulnerability, risks categorised by type (e.g. geophysical, social and technological) have been identified and analysed by the Institution of Professional Engineers New Zealand (IPENZ), to enable measures to be put in place to either eliminate or reduce their impacts. This analysis covers the characteristics of hazards, in order to understand their relationship with national planning measures, which includes a range of likelihoods and consequences. The indicative risks show that cyber-attacks affecting data confidentiality are likely to occur at least once a year but are likely to have only minor consequences (Institution of Professional Engineers New Zealand (IPENZ) 2012), whereas a very large volcanic eruption is only likely to occur once in a millennium; however, the consequences would be considered catastrophic (Institution of Professional Engineers New Zealand (IPENZ) 2012). In particular, the IPENZ highlights that for natural hazards, each threat has a different profile, thereby suggesting that it is not appropriate to “lump together” all natural hazards and it would be more appropriate to target hazard-specific reduction measures to each individual hazard type. The national risk framework also incorporates the localised risks, with a risk exposure calculated for major settlements in New Zealand. This shows that any measures to mitigate natural hazards need to recognise the regional differences in terms of risk. New Zealand currently has a number of acts included in a regulatory framework, similar to the UK, including the Resource Management Act 1991, the Building Act 2004, the Civil Defence Emergency Management Act 2002, the Local Government Act 2002 and the Local Government Official Information and Meetings Act 1987 (Institution of Professional Engineers New Zealand (IPENZ) 2012). However, these acts are inconsistent with their definitions of natural hazards and do not include some important threats, and the range of restrictions to be imposed is limited (Institution of Professional Engineers New Zealand (IPENZ) 2012).

Despite the lack of recognition across the globe of the merits of computational modelling for disaster scenarios, the UK Government noted that “modelling and simulation techniques are important ways of enabling complex systems to be understood and manipulated in a virtual environment” (Council for Science and Technology (CST) 2009). As such, there is an understanding of the benefits of completing computational modelling from a government perspective. This commitment was shown through the recommendations made in “A National Infrastructure for the 21st Century” report (Council for Science and Technology (CST) 2009). Recommendation Three recommended “stimulating better understanding of the complexity and resilience of the national infrastructure, by commissioning research into scenario planning and modelling national infrastructure systems, from physical, economic and social perspectives” (Council for Science and Technology (CST) 2009), underlining the government’s commitment to facilitating the simulation and modelling of disaster management systems in natural hazard scenarios.