Mariana Burkhardt

Impacts of natural disasters on supply chain performance

M. Burkhardt

### Impacts of natural disasters on supply chain performance

Mariana Burkhardt

**Impacts of natural disasters on supply chain performance**

#### Produktion und Energie

Karlsruher Institut für Technologie (KIT) Institut für Industriebetriebslehre und Industrielle Produktion Deutsch-Französisches Institut für Umweltforschung

Band 36

Eine Übersicht aller bisher in dieser Schriftenreihe erschienenen Bände finden Sie am Ende des Buches.

## **Impacts of natural disasters on supply chain performance**

by Mariana Burkhardt

Karlsruher Institut für Technologie Institut für Industriebetriebslehre und industrielle Produktion u. Deutsch-Französisches Institut für Umweltforschung

Impacts of natural disasters on supply chain performance

Zur Erlangung des akademischen Grades eines Doktors der Wirtschaftswissenschaften (Dr.rer.pol.) von der KIT-Fakultät für Wirtschaftswissenschaften des Karlsruher Instituts für Technologie (KIT) genehmigte Dissertation

von Dipl.-Kffr. Mariana Burkhardt

Tag der mündlichen Prüfung: 6. Februar 2020 Hauptreferent: Prof. Dr. Frank Schultmann Korreferent: Prof. Dr. Kay Mitusch

**Impressum**

Karlsruher Institut für Technologie (KIT) KIT Scientific Publishing Straße am Forum 2 D-76131 Karlsruhe

KIT Scientific Publishing is a registered trademark of Karlsruhe Institute of Technology. Reprint using the book cover is not allowed.

www.ksp.kit.edu

*This document – excluding the cover, pictures and graphs – is licensed under a Creative Commons Attribution-Share Alike 4.0 International License (CC BY-SA 4.0): https://creativecommons.org/licenses/by-sa/4.0/deed.en*

*The cover page is licensed under a Creative Commons Attribution-No Derivatives 4.0 International License (CC BY-ND 4.0): https://creativecommons.org/licenses/by-nd/4.0/deed.en*

Print on Demand 2021 – Gedruckt auf FSC-zertifiziertem Papier

ISSN 2194-2404 ISBN 978-3-7315-1020-8 DOI 10.5445/KSP/1000105982

#### **Impacts of natural disasters on supply chain performance**

Zur Erlangung des akademischen Grades eines

#### **Doktors der Wirtschaftswissenschaften (Dr.rer.pol.)**

von der Fakultät für Wirtschaftswissenschaften, des Karlsruher Instituts für Technologie (KIT)

genehmigte

#### **DISSERTATION**

von

#### **Dipl.-Kffr. Mariana Burkhardt**

geb. in Dresden

Tag der mündlichen Prüfung: 06.02.2020

Hauptreferent: Prof. Dr. Frank Schultmann Korreferent: Prof. Dr. Kay Mitusch

To Granny

## **Acknowledgements**

This dissertation was written during my time as a research associate at the Institute for Industrial Production (IIP) at the Karlsruhe Institute of Technology (KIT) under supervision of Prof. Dr. Frank Schultmann. The motivation for this topic was based on my work for the EU project MOWE-IT, funded under the 7th Framework Programme. Later my research was embedded in CEDIM - The Center for Disaster Management and Risk Reduction Technology, especially with the cooperation with the Institute of Economics (ECON), Chair of Network Economics of Prof. Dr. Kay Mitusch.

First of all I want to thank Prof. Dr. Frank Schultmann for his supervision and the opportunity to conduct my research, as well as the many insights I got in many interesting research fields. I also thank my former colleagues, especially Heike, Christina and Frank for our common project work.

A big thank you must go to Jan for his sporting support, motivation and friendship as well my many friends who accompanied me during the last years, especially: Ana, Alissa, Maren, Lea, Holger, Karin, Werner and Gerlinde. I also thank Kathrin - great to have somebody with the same dreams! Thank you to Franziska and Tomás too, for their support in completing this work! I also want to thank my family for their support, particularly my beloved grandma for inspiring me and for always being on and by my side. I love you!

Last but not (at all) least I want to thank my husband Michael for his contributing remarks, long discussions and understanding - without him this PhD would not have been possible!

Karlsruhe, July 2019 *Mariana Burkhardt*

## **Contents**




## **List of Figures**



## **List of Tables**



## **Acronyms**



## **1 Introduction**

### **1.1 Motivation and objective**

One of the best known natural disasters with global supply chain effects is the 2011 earthquake followed by a tsunami in Fukushima, Japan. Apart from the tremendous impact and suffering this event caused, especially where the flood wave hit, it also resulted in serious domestic and global supply chain disruptions ([214], p.76). This is not surprising, as Japan is a major worldwide manufacturing hub and the fourth largest export country in the world ([277]). Strongly noticeable had been shortages of critical components, operational shut downs, operations at reduced capacities and price increases for lacking materials ([201]).

Another event that stayed in memory is the volcanic eruption of Eyjafjallajökull 2010 in Iceland (here and following [180], p.93). Due to a massive ash cloud airlines were forced to ground their airplanes and also airports had to close for safety reasons. The event brought therewith a major shift of transport modes for continental freight, from air to rail, road or water transport. Even affected were intercontinental flights, as not all cargo could be rerouted to more southerly airports for reasons of capacity. So especially perishable goods, like flowers and fresh fruits did not find their way to Europe. Even on the other side of the globe the effects were noticeable, as for example in Japan, Nissan had to suspend production lines due to a lack of necessary parts.

As the aforementioned examples show, natural disasters affect businesses and supply chains in a severe manners, while business interruptions are the most frequently observed effect ([306], p.1519). Moreover, these performance impacts hampering normal operations can spread out all over the world and can even raise their magnitude. This is more than valid as through increased interconnectedness and more globalized supply chains local effects proceed from the place of origin to other connected points within the network ([383], p.327; [58], p.6; [115], p.70; [86], p.1116; [79], p.1; [372], p.305). So even businesses that are not directly affected can suffer indirect consequences, e.g. 'due to the failure of their suppliers or difficulties transporting supplies where needed' ([61], p.172).

Apart from those effects observed today, future developments seem even worse, as it is expected that the frequency and magnitude of natural disasters will increase in the next decades ([217], p.1159; [392], p.380; [78], p.90; [114], p.V; [194]), and that due to climate change even weather patterns are suspected to change ([115], p.68). Therefore a supply chain must be designed to meet those future developments, coping with changing environments in an appropriate way ([165], p.645). Thus a sustainable supply chain design takes the positive and negative potentials of a region into consideration ([86], p.1116), but only when the characteristics defining the potential for harm are known, risks can be reduced. And 'as our understanding of those [...] forces improves, so does our chance of developing more robust strategies for preparedness/ planning, response and recovery' ([392], p.381).

To assess the vulnerability of supply chains to natural disaster it is therefore necessary to investigate the attributes of the surrounding environment (the regions or places a supply chain is embedded in), which serve as explaining momentum for differences in vulnerability to natural disasters between distinct locations. The ambient system thereby consists of social, economic, physical and environmental aspects.

Nevertheless, impacts of natural disasters on businesses (performance) respectively supply chains - and so also supply chain vulnerability against natural threats - are relatively understudied, as the focus lies more on people's safety rather than on business continuity ([61], p.169ff; [379], p.103; [381], p.1f; [382], p.54). Also 'empirical observations on how businesses respond after a major catastrophe are rare' ([175], p.1007). But natural disasters 'can and do affect the performance of the supply chain' ([79], p.1), as the aforementioned examples demonstrate. This situation is aggravated by the fact that businesses do relatively little to prepare for catastrophes and if they do, measures are often site-specific, ignoring problems arising from outside the corporation and the aforementioned carry-over effects ([382], p.54). Thus despite the fact that people's safety is 'ranked' higher than business issues this neglect and the increased frequency and intensity of natural disasters can lead to more vulnerable supply chains regarding external threats ([375], p.121). This results in severe supply chain disruptions.

However, current supply chain risk assessment methods mostly lack a consideration of external threats, such as natural disasters, in general. If risks from outside a supply chain network are mentioned, the majority refers to risks associated with suppliers (for example [391]; [167]; [184]; [162] and [103]). That suggests that primarily the location of a supplier defines the vulnerability to natural forces. What also leads to the conclusion that production facilities in risk prone areas bear a high risk for supply chains. On the other hand, like supply chain risk assessments lack the incorporation of natural disasters, country and natural disaster risk approaches lack any consideration of supply chain impacts (for example [38]; [248] and [109]).

The main objective of this work is therefore to develop an approach that assesses the performance impacts of natural disasters on supply chain performance, as none currently exist. To consolidate those findings, an intensive literature review will be conducted. Moreover it is necessary to research the characteristics that build the potential for harm within distinct locations and to explain different levels of susceptibility. A definition of 'performance impacts' is required as well as the identification of an indicator that measures the effects. Based on the results of the developed approach implications for supply chain designs and procurement decisions are given. The development of the approach *SCperformND - Supply Chain performance impact assessment of Natural Disasters* and the necessary steps are explained in the following chapters of this work.

### **1.2 Structure**

In **Chapter 1** the motivation for the topic is firstly given. Raising awareness for the expected increase in frequency and intensity of natural disasters makes it inevitable to take those events also in supply chain risk management into consideration.

**Chapter 2** provides an overview on relevant definitions, clarifying risk, supply chain risk, vulnerability and exposure in the context of supply chain and natural disaster risk assessment. Additionally, factors that lead to an increased vulnerability of supply chains are explained. The chapter is concluded with the classification of supply chain risks, once without a concrete focus on natural events (a so called general classification) and otherwise in accordance to external risks. Additionally the risk drivers within each category are explained.

It is necessary to differentiate between natural hazards and disasters. Hazards are just the underlying natural processes that built the potential for harm but it needs a subject to experience the threat to become a disaster. As the focus is on supply chains the subject of the question here is the supply chain, which experience the impact of a natural catastrophe. The different types of natural disasters and their sub-types, as well as an explanation on each can be found in **Chapter 3**.

The impacts that natural threats can have on a supply chain and their performance are shown in **Chapter 4**. Beside the two major components where a supply chain is susceptible to natural disasters are explained and a literature review on currently used supply chain risks and country risk assessment approaches is conducted.

Based on the distinction between location and transport when it comes to supply chain vulnerability **Chapter 5** presents the concept of vulnerability of places in the context of this work. A literature review on vulnerability characteristics is conducted to identify explaining indicators for different levels of susceptibilities in different regions.

**Chapter 6** explains the necessary steps for the impact assessment method *SCperformND*, starting with a general description, followed by a mathematical foundation.

In **Chapter 7** the method *SCperformND* is demonstrated within a case study. Starting with the identification of vulnerability indicators with the greatest influence on performance, followed by a detailed analysis on country basis.

Finally **Chapter 8** concludes with a summary of this work, as well as a critical appraisal and outlook for future research.

## **2 Supply chain vulnerability**

### **2.1 Risk, vulnerability and related definitions**

There are numerous definitions of 'risk' existing, as well as with special focus on supply chains or disasters. To get a deeper understanding of the necessary background, definitions for the relevant terms are given, which shall apply for the rest of this work.

One possible reason for the high amount of differing definitions can be seen in the change of related contexts. That is for example the individual perception of risks changes over time, risks which have been perceived as relevant in history might have nowadays no or different meanings. Besides this, new risks came into place, for example from the use of state of the art IT-technology ([311], p.15, this paragraph is extracted from the authors Diploma-thesis p. 4).

Generally, **risk** can be defined as 'the probability of events that result in loss' ([134], p.120), 'danger, damage, [...], injury or any other undesired consequences' ([129], p.52). This definition concludes with the differentiation of risks regarding their probability of occurrence and their possible impacts. As can be seen in figure 2.1 are natural disasters located in the upper left quadrant and are therewith low probability, but high impact risks (to a supply chain). Given the apparently low probability of external risks, those risks are likely to be underestimated or are not in the focus of (supply chain) risk management. 'Most companies develop plans [just] to protect against recurrent, low-impact risks [...], but ignore high-impact, low-likelihood risks'([55], p.54). Nevertheless can external risks 'have [...] severe impact[s] in terms of magnitude in the area of their occurrence' ([373], p.305) as well as cascading effects along the supply chain ([58], p.11f; [160], p.13) - even so the majority of those risks are predominantly exceptional ([313], p.244). That is why an effective (supply chain) risk management must investigate not solely internal risks, but also external risks. This is more than important as 'many of the threats to business continuity lie outside the focal firm' ([58], p.6).

Figure 2.1: Risk map external events (sources: [313], p.244ff; [187], p.109; [206], p.172; [238], p.44)

In addition, the term 'risk' must be seen in the specific field of investigation, in this work supply chains and natural disasters. Following the previous definitions are **supply chain risks**'disruptions and disturbances of flows within the goods [...] networks [...], which have negative impacts on the objectives of single corporations respectively the whole supply chain regarding final customer benefits, costs, quality, time or value increases.' ([168], p.42)<sup>2</sup> . Those negative outcomes are summarized under **performance impacts**. The sources of those disruptive triggers can lie in organizational, supply chain or environmental characteristics ([203], p.535). Therewith is 'any disruption at any stage in a supply chain that can be linked to environmental causes [...] ascribable to external risks'([63], p.2). Since natural disasters occur outside the supply chain network, those risks are part of the category of external (supply

<sup>2</sup>Of course disruptions in the information flow can also occur, but as this is not the focus of investigation, it is just focused on the flow of goods.

chain) risks, as already mentioned. So are disruptions which can be associated with organizational characteristics assigned to internal risks of the corporation itself, while supply chain characteristics describe internal risks caused by the characteristics of the supply chain network (so called network risks). Further details on supply chain risk classifications can be found in chapter 2.3.

**Supply chain vulnerability** can consequently be defined as 'exposure to serious disturbances [or disruptions], arising from risks within the supply chain as well as risks external to the supply chain' ([58], p.3).

As this work focuses on impacts of natural disasters on supply chain performances, it is inevitable to provide also a definition of risk in the context of disaster management. From this understanding the interaction of a hazard (man-made or natural) and a vulnerable condition results in risk ([332], p.6). A vulnerable condition or **vulnerability** are the 'pre-event, [...] characteristics or qualities of [...] systems that create the potential for harm' ([70], p.599). 'System: [is thereby] the system of analysis, such as a coupled humanenvironment system, a population group, an economic sector, a geographical region, or a natural system' ([111], p.157). In the given case, the system is the supply chain network itself, as well as the related points (e.g. countries) the supply chain network is linked to. Here characteristics of the location and the supply chain, that define the potential impact, must be considered. The factors, explaining differences in supply chain vulnerability between distinct locations, are investigated in greater detail in chapter 5.2.

Generally speaking vulnerability depends deeply on topic, field of investigation and research question ([36], p.152f; [140], p.199). That is why the chosen explanation is just applicable for this specific context, and is subject to change for other research questions.

A counterpart to vulnerability, is **resilience**: 'the ability of a supply chain [or other systems] to overcome vulnerability' ([134], p.125) and 'return to its original state or move to a new, more desirable state after being disturbed' ([58], p.2). The capability of a system, person or community to reach those states is described as **recovery** ([89], p.4). Details on recovery can be found in chapter 4.3. Additionally it is also necessary to take the different types of disasters into consideration, as the vulnerability to different types can vary ([36], p.152f; [217], p.1149). A description on different types of natural disasters is therefore given in chapter 3.2.

Based on the given definitions are the reasons for increased supply chain vulnerability presented in the following chapter.

### **2.2 Reasons for increased supply chain vulnerability**

As stated in the previous chapter, supply chain risks can occur out of organizational, supply chain related or environmental causes. This three categories are the main factors explaining the increase in supply chain vulnerability, as can be seen in figure 2.2. All results presented here are findings of an intensive literature review, its results and sources can be seen in appendix A.1.


Figure 2.2: Reasons for increased supply chain vulnerability (sources see appendix A.1)

To start with the smallest entity, the corporation itself, reasons for ascending supply chain vulnerability lie in organizational aspects. Often cost reduction initiatives or attempts at being competitive in an increasingly globalized market are the reasons for research. As those investments in competitiveness can be valuable in the short-run, severe risk increases can hit in the long-term, mostly only recognized when the first major disaster strikes. Centralization is also an important aspect, as companies focus on a few distribution and production locations only and therewith often reduce the supplier base and their inventories ([221], p.35). This is very useful when costs are taken into consideration solely, but can be a very risky approach when unforeseen events disrupt normal processes and business continuity. Other initiatives are lean and just-in-time strategies, which must be weighed against probable higher risks.

The next category, the **network factors**, occur outside of interactions of several corporations within the supply chain network and are thereby defined through the charactersitics of the supply chain network. Reasons for increased supply chain vulnerability are, beside others, emerging globalization strategies, resulting in outsourcing to low cost countries or in general more international procurement processes ([221], p.35). That means a disaster somewhere in the world can have devastating impacts all over the connected points, not just for supply chains but also for people and societies ([333] p.5, [148], p.192). This is also valid for domestic supply chains, but the impacts increase with the expansion of a network ([148], p.198).

**External factors** affect the network from outside and are beyond the influence of the members of the network. Often stated factors are disasters or the expected increase in intensity and frequency of weather events ([182], p.6, 237; [207], p.7; [373], p.301; [192]; [58], p.1ff; [57], p.189). The analyzed natural disaster are thus part of this last class.

### **2.3 Classification of supply chain risks**

#### **2.3.1 - in general**

As well as for the definition of risk, numerous approaches to classify risks exist. Among others Pfohl 2002 ([220]), Hotwagner 2008 ([145]) and Lasch et.al. 2002 ([174]) distinguish between the following risk types in table 2.1.


Table 2.1: Classification of supply chain risks ([220], p.10ff; [145], p.24 and [174], p.113f)

As stated in chapter 2.1 the following definition of supply chain risk shall apply: which are 'disruptions and disturbances of flows within the goods [...] networks [...], which have negative impacts on the objectives of single corporations respectively the whole supply chain regarding final customer benefits, costs, quality, time or value increases.' ([168], p.42). According to the classification of supply chain risks these risks are pure risks, as they are 'just' referring to a negative **deviation from the expected result** ([397], p.60). Speculative risks are not considered explicitly here, as they are not the focus of this work. Nevertheless it is necessary to be aware that there are also speculative risks and therewith not just a negative impact of risks, but also a 'chance' to reach a better state after the occurrence of a risk or hazard (or more general: to gain a positive deviation from the expected result) ([220], p.11). Concerning the **level of risk**, risks can be graded from very small risks up to risks that threaten the livelihood of people or their existence. Within this category of life and existence threatening events can also natural hazards be summarized, as natural hazards are 'natural processes, which can lead to loss of human lives, injuries or other health related impairments, damages to property, loss of livelihood, and several services as well as to disturbances to the social and economic conditions of a society'([166], p.188). It can therefore be assumed that natural hazards are more likely huge risks than small or medium sized risks (or one of the other gradations), since they are 'normally accompanied by grave consequences' ([313], p.244). Regarding the **point of origin**, there are supply chain internal risks and supply chain external risks to be observed. Whereas supply chain internal risks are the result from processes of interaction between corporations within the supply chain network, external supply chain risks are the result of the interaction of the supply chain with its environment ([294], p.275; [157], p.201). These external or environmental risks are caused by events outside the influence of the supply chain, or events that can just barely be influenced ([399], p.28; [313],p.244). Internal supply chain risks can be subdivided in risks within the organizational boundaries of a firm (risks within the functional areas), and network-related or cooperation risks ([313], p.244; [294], p.275). As mentioned before networkrelated risks are the result of interactions between corporations within the supply chain. As already stated one of the reasons for the lack of attention on environmental risks, they are regarded as less likely than internal supply chain risks ([313], p.244). Nevertheless, external risks can 'have [. . . ] severe impact[s] in terms of magnitude in the area of their occurrence' ([373], p.305) as well as cascading effects along the supply chain ([58], p.11f; [160], p.13). That means external supply chain risks and therewith natural hazards have a lower probability of occurrence than internal risks, but can have a much higher impact ([206], p.173; [238], p.44). Dependent on the size of a corporation and other related factors (e.g. industrial sector) a corporation has several **functional areas**, which can cause risk. It is worth mentioning that not necessarily a corporation has all above mentioned areas, but it can have more as well (dependent on the aforementioned influencing factors). **Insurability** is dependent on whether it is possible to identify and quantify (or at least estimate) the probability of occurrence of an event and its impact, and for the insurance company in addition the ability to set premiums ([171], p.15). To ensure the existence of a corporation it is useful to insure at least existencethreatening risks ([233], p.13f), like natural hazards. The problem is that meanwhile some insurers feel that natural hazards are uninsurable due to huge losses they have faced in recent years ([171], p.40). The **measurability** of risks is closely linked to their insurability as a prerequisite of measurability and therefore for the quantification of risks is the availability of data. Hence risks with a higher probability of occurrence are easier to quantify than risks with low occurrences ([233], p.9). The availability of (historical) data is essential to analyze the frequency and magnitude of natural hazards as well as their possible occurrence ([118], XV). Through their **scope**, risks can also be differentiated in single risks and the so called overall risk. Whereas a single risk is the result of a single decision or a chosen alternative, the aggregation of all single risks is the overall risk ([310], p.14f; [220], p.12). But it has to be considered that through interdependencies the sum of all single risks is not necessarily equal to the overall risk ([177], p.338 [145], p.24) The **decision level** differs between operative, tactical and strategic risks. Strategic risks influence the achievement of long-term goals (e.g. securing the existence of a corporation) and have an effect on the whole corporation or supply chain ([220], p.12; [172], p.532; [174], p.114). Operative and tactical risks affect short and mid-term goals and also 'just' a part of a corporation or supply chain ([113], p.35; [397], p.61; [220], p.13; [120], p.38). Concerning the **factors of production** labor, material, capital and resource risks can exist ([120], p.37). Depending on whether and to what extent these factors are used.

To summarize, natural disasters are pure risks with often life threatening impacts that affect the supply chain from externally (external risks). While insurability is hardly achieved, insurers had to face severe losses in recent years. Due to historical data records natural disasters are more easily measurable than other risks.

Beside the above mentioned types of risks, other classifications exist as well. Moreover is the systematization and relevance of risks dependent on factors like company size, industrial sector and others ([388], p.21). The given overview is an excerpt of the most relevant approaches in the context of supply chain risk management and introduces the classification of supply chain risks regarding environmental risks in the next chapter.

#### **2.3.2 - in accordance to external risks and hazards**

Regarding the topic of this work, to assess the impacts of natural disasters on supply chains it is necessary to provide a detailed overview of existing classification approaches that take those risks into consideration. In table 2.2 such an overview is provided. The classifications used are separated by slashes and those related to external risk are marked bold. Even though it is just a sample it shows that the majority of authors refer to internal versus external supply chain risks, but often under different names and descriptions.

The hierarchical degradation of supply chain risks and their explanation can be found in figure 2.3 to 2.5 and in the text below.


Table 2.2: Classification of supply chain risks in accordance to external risks


To start with internal supply chain risks, those risks can be divided in organizational and network risks ([187], p.100ff; [156], p.114; [58], p.4ff; [157], p.201f; [141], p.116f). Organizational risks are internal to the corporation and internal to the supply chain as well, so the source of risks lies within the boundaries of the corporation and can directly be influenced by the corporation ([313], p.244; [187], p.108). Under this category all risks are subsumed out of the functional areas, e.g. sales and procurement risks (see also chapter 2.3.1 for this classification). Processes and their control are then the connecting elements to other points within the supply chain network and that link their partners ([157], p.201f; [204], p.437). The source of risks for network risks lies outside the corporation but within the supply chain ([187], p.108). The upper part of figure 2.3 illustrates this classification, while the lower box explains organizational risks (which are internal to the supply chain and to the corporation).

Figure 2.3: Internal supply chain risks (based on [168])

On the other hand external risks occur, which are caused by sources outside the supply chain, but within the environment of that system, that have direct or indirect impact on the network ([141], p.116f; [298], p.60ff). That means environmental supply chain risks are 'risks beyond the influence of the members of the entire supply chain' ([187], p.108). However the effect of those external supply chain risks is not limited to the supply chain only; they can even affect the market places themselves ([58], p.4ff). That is why it is necessary to distinguish external supply chain risks further into supply chain risks related to specific procurement markets and supply chain risks not related to specific procurement markets ([168], p.68ff). Examples of risks that are related to the procurement markets are: social, political, cultural, legal, economic, technological and ecological risks ([168], p.68ff; [220], p.13ff; [58], p.4ff, [156], p.122f; [294], p.275f; [90], p.5), which can also be seen in figure 2.4. Supply chain external risks related to specific procurement markets are therewith 'the result of characteristics and circumstances of the global procurement market' whereby supply chain risks not related to specific markets 'cannot be assigned to a specific country' ([168], p.68ff) or market, even if they occur in specific countries or regions. In this category natural disasters can be classified, which is explained in detail later.

Figure 2.4: External supply chain risks (based on [168])

The different characteristics of the procurement market that build the potential for harm to a supply chain are the third layer in figure 2.4. Whereby the legal risk of a procurement market is determined by the legal circumstances in the sourcing country, through its legal and economic system ([168], p.71). To this category belong e.g. environmental legislation, tax law and the ability of a country to protect intellectual property (or more general: a lack of legal protection) ([220], p.15; [55], p.57; [386], p.11). The legal risk is linked to the economic system of a country and vise versa. The economic risk describes all 'macroeconomic influences within a spatially delimited area' ([130], p.17). Which are, among others, for a specific procurement market: exchange rates, economy, inflation, foreign trade, interest rates and infrastructure ([386], p.11; [241], p.305). This category entails also financial risks, which 'expose [...] a firm to potential loss through changes in financial markets' ([129], p.53). 'Unpredictable changes in political structures or ideologies' ([168], p.70) are seen as the political risk of the supply chain environment. Caused by a high degree of uncertainty it is very difficult to act in a proactive manner when it comes to political risks ([220], p.15). Single risks in this category are: expropriation, trading restrictions, white collar crime, terrorism (man-made hazards) and political (in)stability ([386], p.11; [397], p.71; [168], p.70f; [374], p.66; [241], p.305). Social and / or cultural risks reflect the 'values, norms, attitudes and beliefs of the social units within a country' ([168], p.72). They are not just occurring because of differences in the aforementioned categories, but could also be the result of 'inadequate knowledge about people, culture, and language' ([148], p.200). Social aspects or moreover demographics, the distribution of income and wealth and the quality of education ([386], p.11; [130], p.69; [398], p.12). Attributes of the technological environment are the information and communication technology, knowledge transfer, process and product innovations, as well as the technology for material flows ([386], p.11; [130],p.68; [168], p.72). The technological risk is therefore the state of the technological development within the respective country ([168], p.72). The last category of environmental supply chain risks with a relation to specific procurement markets are ecological risks, which are for example determined by environmental protection, ecological damage and recycling ([386], p.11) as well as 'the geographical location of suppliers or the availability of raw material' ([6], p.66). As mentioned before also external risks exist that are not related to specific procurement markets and cannot be assigned to a particular country or region. This category entails natural and man-made hazards (e.g. terrorism, acts of war). As the focus of this work is the investigation and assessment of natural threats, man-made hazards are not considered any further. Work on man-made hazards can for example be found in [60].

Figure 2.5: External supply chain risks - not related to specific procurement markets (based on [168])

Natural hazards can be defined as: 'natural processes, which can lead to loss of human lives, injuries or other health related impairments, damages of property, loss of livelihood, and several services, as well as to disturbances to social and economic conditions of a society' ([166], p.188) - and so to disruptions or disturbances of a supply chain. As this definition of natural hazards can result in all these negative outcomes, it needs a system or asset to be affected to become a disaster. Details on that differentiation can be found in chapter 3.1. Subcategories of natural hazards are meteorological hazards, hydrological hazards, geological hazards, climatological hazards, extra-terrestrial hazards and biological hazards ([166], p.188f; [86], p.1035).

It is again worth to mentioning that supply chain risks can be classified in various manners, which sometimes also overlap. The developed approach in this work fits the topic best, as it presents a distinction between external supply chain risks that are related to specific procurement markets and those risks which cannot be associated with a specific market or region.

To understand the differences between different disaster types the next chapter gives an explanation on the distinction between hazards and disasters, followed by an overview on the disaster types and their related subtypes.

## **3 Natural hazards and disasters**

#### **3.1 Difference between hazards and disasters**

As already stated natural hazards are 'natural processes, which can lead to loss of human lives, injuries or other health related impairments, damages of property, loss of livelihood, and several services, as well as to disturbances to social and economic conditions of a society' ([166], p.188). But for a hazard to become a disaster 'there has to be a subject to experience the hazard or the threat. For example, people, infrastructure and economic activities' ([333], p.30). In the following case the supply chain, respectively the location of a supplier or own production facilities are the affected subjects. So while there is a risk of being affected by natural processes, they do not necessarily result in a disaster. The more assets (subjects) at risk the higher the potential (perceived) impact of a disaster. Moreover the disaster type has an effect on the extent of damage, e.g. explained through a different speed of onset. While earthquakes, tornadoes and hurricanes have sudden onsets, drought realizes slowly ([380], p.476). Other differentiating factors are the duration of impact and the length of forewarning. An outline on differentiating factors can be found in table 3.1.


Table 3.1: Classification of disasters by duration and length of forewarning ([4], p.10)


The next chapter will explain the different types of natural disaster as well as their classification.

### **3.2 Types of natural disasters**

In this chapter an overview on different types of natural disaster is given, which can be classified according to figure 3.1. The types of classification are geophysical, hydrological, meteorological, climatological, extra-terrestrial and biological events.

Whereas natural hazard describes just the possible occurrence of a natural process, as explained in the previous sub chapter, 'it becomes a natural disaster if people or values are influenced negatively' ([86], p.1120). A disaster is therewith 'a situation or event which overwhelms local capacity, necessitating

Figure 3.1: Types of natural disasters

a request to a national or international level for assistance' ([236], p.2). A negative consequence is assumed on supply chain performance, it is hereafter referred to as disaster.

To investigate the possible occurrence of natural threats for a certain region (and the effects of natural disasters on supply chains) it is necessary to analyze historical disaster data for each disaster type. Historical data in this work is derived from *The Dartmouth Flood Observatory* for floods. There are large floods listed that caused 'significant damage to structures or agriculture, long (decades) reported intervals since the last similar event, and/or fatalities' for all regions worldwide ([74]). Hurricanes as well as earthquakes within the United States of America are sourced from *The U.S. Department of Homeland Security* ([101]), where the data is extracted from the disaster declarations. And earthquake data for New Zealand stems from GeoNet (Geological hazard information for New Zealand [117]). The frequency analysis to it can be found in chapter 6.1. Cyclones (as found for example in Australia) are here subsumed 'under' flood, as the major impact was felt through the massive flooding after the cyclone. Other disaster types are not considered, as no studies could be found in the literature, where data on businesses or supply chains after a disaster where analyzed. Details on that are shown in chapter 6.

A database for all disaster types on a global scale can for example be found at *EM-DAT: The Emergency Events Database* ([76]). Beside this often country specific data bases exist, if a special focus is needed. Examples here are the *Austrian Research Center for Forests* ([17]) for Austria or the *Instituto de Estudios Ambientales (IDEA) (Universidad Nacional de Colombia)* for Colombia ([99]).

For the analysis it should also be remembered that external supply chain risks, not related to a specific procurement market (see chapter 2.3.2), cannot be associated with a specific country. Nevertheless natural disasters and their possible negative effects are mostly reported on country level ([236], p.3). That is why the vulnerability of places approach is used (introduced in chapter 5), in order to get a more detailed picture of the regions of interest, meaning, that while starting with the country perspective, information is attempted to be broken down to a smaller focus area (the place or region).

In the following sub chapters single types of natural disasters are shortly described. A profound definition of all types and sub-types can be found in appendix A.7.

#### **3.2.1 Hydrological disasters**

Hydrological disaster are caused by the occurrence, movement, and distribution of surface and subsurface water ([166], p.188) and are divided in floods, landslides and wave actions with their related sub-types, which can be seen in table 3.2.

The first entry '**floods** [describe] [...] an overflow or inundation that comes from a river or other body of water and often threatens lives and properties' ([143], p.64). While **coastal floods** are 'higher-than normal water levels along the coast caused by tidal changes or thunderstorms that result in flooding, which can last from days to weeks', are **riverine floods** 'a type of flooding resulting from the overflow of water from a stream or river channel onto normally dry land in the floodplain adjent to the channel' (here and following [152], p.13ff). As the name implies, **ice jam floods** are 'the accumulation of floating ice restricting or blocking a river's flow and drainage. Ice jams tend to develop near bends and obstructions (e.g.,bridges)'. And **flash floods** are 'heavy or excessive rainfall in a short period of time that produce immediate runoff. Creating flooding conditions within minutes or a few hours during or after the rainfall'([152], p.13f).

More than two thirds of all reported disasters and one third of all damages can be associated with flooding, making it one of the most frequent natural disasters (beside storms) ([194]). This aspect is also illustrated in figure 3.2. Even within literature, sources dealing with supply chains or businesses in the aftermath of natural disasters in chapter 6, where more sources identified dealing with flooding than any other disaster type. This fact is aggravated as 'hydrometeorological hazard frequencies and magnitudes might also change in the near future due to climate change and/or environmental degradation' ([217], p.1159). That is why flooding is a key aspect of investigation in the following chapters.


Table 3.2: Hydrological disasters ([77])

Figure 3.2: Loss events worldwide [196]

**Landslides**, 'include [...] many downslope movements of soil, rock, or other Earth materials. Landslides can be activated by earthquakes, rapid snowmelt, intense rainstorms, groundwater rise, slope toe cutting by rivers, or volcanic eruptions, in conjunction with gravity and occur when driving forces, such as gravity, exceed the frictional strength of the slope materials' ([228], p.435). And **wave actions** can be defined as 'wind-generated surface waves that can occur on the surface of any open body of water such as oceans, rivers and lakes, etc.' ([152]). For those two types no literature sources could be identified referring to businesses or supply chains and they are therewith not part of the impact assessment. But for reasons of completeness these types are described as well, what is valid for all following, not included disaster types.

#### **3.2.2 Meteorological disasters**

Meteorological disaster are caused by short-lived, extreme weather and atmospheric conditions ([166], p.188). Types of meteorological disaster are storms and extreme temperature with their related sub-types, which can be found in table 3.3.


Table 3.3: Meteorological disasters ([77])


High wind speeds can have severe impacts on (critical) infrastructure, properties, humans and therewith also on supplier or production facilities within a supply chain network. Damage to constructions starts when the wind speed exceeds 72,4 km/h ([228], p.440). The terminology for tropical cyclones varies regionally, so this is just one possible way to describe this disaster types. A **tropical cyclone** is 'an organized, cyclonically rotating system of convection driven by fluxes of heat derived from the ocean'. While **Tropical storms** are 'tropical cyclones that have maximum sustained winds between 17 and 32 m/s (34-63 kts); intense tropical cyclones – those with winds of at least 33 m/s (64 kts) - are called hurricanes in the Atlantic and eastern North Pacific basins and typhoons in the western North Pacific' ([169], p.481). The impact assessment in chapter 6 also incorporates several hurricanes that affected the United States of America, while typhoons could not be identified in the literature research. An **extra tropical storm** is 'a type of low-pressure cyclonic system in the middle and high latitudes (also called mid-latitude cyclone) that primarily gets its energy from the horizontal temperature contrasts (fronts) in the atmosphere' ([152]). And **extreme temperature** is 'a general term for temperature variations above (extreme heat) or below (extreme cold) normal conditions' ([152]).

#### **3.2.3 Climatological disasters**

Climatological disasters are caused by long-lived atmospheric processes (climate variability) ([166], p.188). Types of disaster here are drought, glacial lake outburst and wildfire as can be seen in table 3.4.




**Drought** is 'an extended period of unusually low precipitation that produces a shortage of water for people, animals and plants' ([152]). '**Glacial lake outburst** floods (GLOFs), also known as jokulhlaups, occur when there is a sudden release of water from beneath or behind a glacier' ([125], p. 398). A **wildfire** is 'any fire occurring in vegetation areas regardless of ignition sources, damages or benefits' ([332], p.7). For all three types no evidence in the literature research was found that underpins an impact of those types on supply chain performance. So these three types are not part of the analyzes, even though possible impacts are obvious, especially for crops (food supply chains) when a drought or wildfire takes place. The focus of this work lies within industrial supply chains.

#### **3.2.4 Geophysical disasters**

Geophysical disasters (sometimes also declared as geological disasters) are events that originate from solid earth ([166], p.187) and 'may cause the loss of life or injury, property damage, social and economic disruption or environmental degradation' ([332], p.4). Subtypes are earthquakes, mass movements and volcanic activities (see table 3.5).


Table 3.5: Geophysical disasters ([77])

An **earthquake** is here defined as 'a tectonic or volcanic phenomenon that represents the movement of rock and generates shaking or trembling of the Earth' ([48], p.208). While **ground shaking** is a 'general term referring to the qualitative or quantitative aspects of movement of the Earth's surface from earthquakes or explosions. Ground motion is produced by waves that are generated by sudden slip on a fault or sudden pressure at the explosive source and travel through the Earth and along its surface' ([232], p.59). 'Ground shaking is the primary cause of earthquake damage to man-made structures. When the ground shakes strongly, buildings can be damaged or destroyed and their occupants may be injured or killed' ([343]). For the quantification of earthquakes the Richter Scale is applied, which allows to scale earthquakes by its size ([26], p.13). Richter (1935) describes the gradation as following: 'In general, shocks of magnitudes 0,1,2 are not reported as felt; shocks of magnitudes 3 and 4 are felt, but cause no damage; magnitude 5 may cause considerable minor damage; magnitude 6 is usually destructive over a limited area; and magnitude 7 and 8 transgress the lower limit of major earthquakes' ([230], p.14). Following this classification were historical earthquake data in chapter 6 considered from magnitude 6 on, as it needs at least a destruction to impact supply chains in a noticeable manner.

**Mass movements** are 'a variety of processes that result in the downward and outward movement of slope-forming materials composed of natural rocks, soil, artificial fill, or combinations of these materials' ([240], p.657). A **volcanic activity** is any 'volcanic event near an opening/vent in the Earth's surface including volcanic eruptions of lava, ash, hot vapour, gas, and pyroclastic material ([152]). One major volcanic event that strongly affected supply chains all over the world was the eruption of Eyjafjallajökull in 2010 (see [242]), as stated in the introduction.

#### **3.2.5 Extra-terrestrial disasters**

Extra-terrestrial disasters are caused by asteroids, meteoroids, comets, and by changes in interplanetary conditions ([152], [166], p.188). Types are impact and space weather as can be seen in table 3.6.


Table 3.6: Extra-terrestrial disasters ([77])

**Impact** is defined as 'a type of extraterrestrial hazard caused by the collision of the Earth with a meteoroid, asteroid or comet' ([152]). '**Space weather** is the chain of processes from eruptions on the sun, their passage through interplanetary space, and the interaction with the Earth's magnetic field that leads to disturbances in the Earth's magnetosphere, ionosphere, and on the ground that represent a hazard to man-made technology and human life' ([30], p.937). As the focus of this work are supply chains without substantial literature that shows influences from extra-terrestrial disasters on supply chains, these types were excluded from further evaluations.

#### **3.2.6 Biological disasters**

Biological disasters are 'processes of organic origin or those conveyed by biological vectors, including exposure to pathogenic micro-organisms, toxins and bioactive substances, which may cause the loss of life or injury, property damage, social and economic disruption or environmental degradation' ([332], p.2). Main types are epidemics, insect infestation and animal accidents (see table 3.7). As the focus lies on industrial supply chains those disaster types are not investigated further, but stated for the purpose of completeness. Moreover it would be necessary to consider the biological processes behind the outbreak.

**Epidemic** is 'either an unusual increase in the number of cases of an infectious disease, which already exists in the region or population concerned; or the appearance of an infection previously absent from a region' ([75]). Under **insect infestation** we understand 'the pervasive influx, swarming and/or hatching of insects affecting humans, animals, crops, and perishable goods. Examples are locusts and African Bees' ([152], p.15).


Table 3.7: Biological disasters ([77])

And **animal accidents** are encountering of humans 'with dangerous or exotic animals in both urban and rural environments' ([152], p.12).

The impacts of the defined natural disasters on supply chains are described below.

## **4 Impacts of natural disasters on supply chains**

As mentioned earlier external risks to a supply chain are often underestimated, as their occurrence is more or less rare while internal risks have often high occurrences. One reason for that might be, that 'firms may find it difficult to justify certain costly strategies for mitigating supply chain disruptions that rarely occur' ([375], p.122). Nevertheless the impacts of external risks can be much more devastating than other categories of risk.

The impact of natural disasters is regarded here from an economic perspective, where three categories in figure 4.1 can be distinguished.

Figure 4.1: Economic impacts of disasters ([336], p.12)

Concerning supply chains effects are defined as indirect costs that lead to perturbations within the flow of goods. Services in this work are not considered explicitly, but can of course be affected as well. As stated, can those 'intangible and indirect effects may have an equal, if not greater, impact to an organisation's ability to operate than the direct damage to an organisation [...]' ([158], p.8). While under direct damage the immediate effects of a disaster are subsumed ([79], p.2). Those physical damages can include (but are not limited to) damage to or destruction of buildings, inventories and materials (here and following [379], p.104). Also lifeline disruptions, like the disruption of water supply, electricity and telecommunications are possible, on which business operations are strongly dependent. 'For example, [can] firms [...] be forced to close down for extended periods of time if they lack critical power supplies or access to natural gas. Furthermore, businesses might lose suppliers or buyers as a result of power outages in impacted areas.' ([379], p.104). And business closure is moreover the most frequent consequence natural disasters have on business performance, which can be reread in chapter 6.1.

Most natural disasters can be devastating, when they hit a critical path within the network. Those are defined through single sourcing, long lead times and high levels of risk ([58], p.8), resulting e.g. in non availability of material and resources, cost increases or sales damages ([58], p.53). That is why those paths should be precisely monitored and first of all considered in the impact assessment of natural disasters on supply chains. For not yet established supplier relationships a general assessment of all potential suppliers (locations) is recommended. As it is even more difficult to assess indirect than direct damages ([136], p.159), e.g. due to their intangibility, a method to quantify or estimate those impacts must be identified or developed and at least applied. Therefore a literature review on current supply chain risk and country risk approaches with a focus on natural disasters was conducted, whose results are presented in chapter 4.4. Beside this, it is important to consider that 'regional economies behave differently when they face a disaster' ([136], p.151f), what is also valid for supply chains or supply chain networks. That is also why differences in regions (see chapter 5) have to be taken into account when the impacts of natural disaster on supply chains are evaluated.

Further are indirect impacts differentiated in disturbances and disruptions, which are explained in chapter 4.1.

#### **4.1 Disturbances vs. disruptions**

Disasters affect a supply chain in different ways, namely disruptions and disturbances, as can be seen in figure 4.2.

The major differences between disturbances and disruptions are their horizon of time as well as their extend of impact. While disturbances do not last for long periods (for example several hours) and have a limited extent (for

Figure 4.2: Disruption vs. disturbance (based on [221], p.35)

example just on one supplier), disruptions can have a long lasting influence on certain factors (affecting a supply chain for months up to years) within a wider extent (where several partners in the network could be affected). Disturbances can be possibly problems in quality, delivery defaults or demand fluctuation, which normally can be prevented through risk management measures. On the other hand result disruptions in severe effects, also on the financial situation of a corporation ([221], p.34; [238], p. 43; [55], p.55; [58], p.1). What can be 'significant supply-chain delays triggering stock-outs, inability to meet customer demand and increase in costs' ([22], p.4068), as well as the 'inability to produce goods' ([148], p.199). Due to carry-over effects can disruptions in the chain of one corporation also easily result in disturbances in the chain of another dependent corporation, even so the corporation is not affected itself ([316], p.225). Both, disturbances and disruptions, can therefore negatively influence a corporations' performance to certain extends ([22], p.4068; [55], p.54). Details on impacts of natural disaster on performances can be found in chapter 4.3. Here are natural disasters (external risks) the triggering event (in figure 4.2 defined as risk) leading to the aforementioned perturbations. As stated in chapter 2.3 risks can also occur inside the supply chain ([98], p.184), but this is not the focus so internal risks are excluded from further investigations.

As a supply chain is a dynamic conjunction it is inevitable that there are always disturbances and disruptions (from different sources), which cannot completely be eliminated. What is important to note is, that corporations know their supply chain (network) well, thus disturbances and disruptions can be detected early and negative effects can quickly be determined and measures are implemented ([22], p.4067). This so called agility (sometimes also referred as flexibility) is 'the ability to respond rapidly to unpredictable changes in demand or supply. Many organisations are at risk because their response times to demand changes or supply disruption are too long.' ([58], p.10).

To gain an overall insight of risk situations of a supply chain, it is necessary to not just investigate the triggering event, but also the susceptibility (determined through different characteristics) of a supply chain respectively location ([375], p.122). The susceptibility is referred to as supply chain vulnerability, which was already defined in chapter 2. An overview of supply chain characteristics that influence supply chain vulnerability is given in chapter 4.3.1, following chapter 4.2 where the aspects of transport and location susceptibility are explained.

As described in chapter 2.2 it is expected that the frequency and intensity of disrupting events, like natural disasters, is increasing. Hence corporations have to deal more often with supply chain disruptions ([22], p.4068). The more a supply chain is able to absorb those external shocks and re-organize back to a functioning system, the more resilient the supply chain is considered to be (see the lower left part of figure 4.2) ([70], p.599). So the direct damage happens at a point in time, the disruption lasts from the triggering event until the system has recovered ([234], p.5).

For the management of disruptive triggers the following steps should be taken into account (taken from [22], p.4069):


As a first step, the discovery is very obvious when a natural disaster strikes with a sudden onset. Non-detection is nearly impossible. Natural threats with a slow onset are not that obvious, but are normally detected as well but of course dependent on the presence of any negative effects. The second step takes the implementation of measures to overcome the disruption into account, while the last target supply chain design improvements, reducing future possible negative impacts for a corporation respectively a supply chain. As the last step implies supply chain design is a vital part to effectively manage supply chain risks and to overcome supply chain vulnerability. That is why, aspects concerning the design of a supply chain must be considered from the first day on. Starting from supplier selection under consideration of all risk aspects (most of all including external risks), to effectively manage the existing supplier base and strengthen long-term relationships (just to state a view).

Following on below are the two major parts that can be affected - transport and location. The differences in their susceptibility are also explained.

#### **4.2 Transport vs. location**

Looking at a supply chain, it generally shows two major parts: locations and the transport of goods, which connect the different points within the network. Even though there are differences in the services to supply chains, the aspect of interconnectedness between the nods is still valid. Nevertheless here only industrial supply chains with flow of goods are considered.

It can be assumed that if a disaster strikes, before and after, it is mostly the location that is affected (see figure 4.3). The transport is consequently influenced through for example impossible supplies to and from a directly affected corporation or through interconnections also from and to dependent locations especially where the types of disaster are the aforementioned hydrological, meteorological, climatological and geophysical ones. But of course natural disasters can also impair the transport directly, but normally transport routes are checked in advance and firms change routes according to weather conditions. In contrast decisions on locations for production facilities or suppliers are generally more long-term oriented and cannot be changed permanently. Thus decisions on supply chain design should be made very carefully (as already stated in the previous chapter). Given all that, the location in this work is the primary field of investigation. This leads to the concept of vulnerability of places, which is explained in chapter 5.

Figure 4.3: Supply chain vulnerability - transport vs. location

Details about the impacts of natural disaster on supply chain performance are now shown in chapter 4.3.

### **4.3 Performance impact and recovery**

As explained earlier the effects that disasters can have on a supply chain are distinguished in disruptions and disturbances, while natural disasters more often result in disruptions. And 'any serious disruption will affect the performance of a company in predictable ways', which can be 'measured by sales, production level, profits, customer service or another relevant metric' ([238], p.42). 'Supply Chain Performance refers [therewith] to the extended supply chain's activities in meeting end-customer requirements, including product availability, on-time delivery, and all the necessary inventory and capacity in the supply chain to deliver that performance in a responsive manner.' ([131], p.61). While 'the full impact of some disruptions is felt immediately. [...] Other disruptions can take time to affect a company, depending on factors such as the magnitude of the disruption, the available redundancy, and the inherent resilience of the organization and its supply chain' ([238], p.42). Those factors are defined in the following chapter 4.3.1.

#### **4.3.1 Supply chain characteristics that influence disaster impact and recovery**

'Recovery is a measure of recovery from the disaster', e.g. measured as 'the number of days that passed before resuming production' ([316], p.219) and is influenced by different supply chain and business characteristics, as well as vulnerability factors within a region (for the factors of place vulnerability see chapter 5.2.2). From the supply chain perspective the structure of the network and the parties involved are major aspects of influence on supply chain vulnerability ([375], p. 123). For definitions on supply chain vulnerability see chapter 2.

Influencing aspects are **supplier characteristics**, as well as the **supplier base** ([375], p.123), with their individualising factors. Those relationships reveal the risk of **dependence**, e.g. 'when firms depend on processed materials, parts, or components from suppliers affected by a disaster, these downstream firms may have to shut down their operations even when they themselves are unaffected by the disaster' ([316], p.218). Todo et al. (2015) 'hypothesize [therefore] that a firm's recovery from a disaster becomes more difficult with an increasing **number of connections** with suppliers and clients within the impacted area'. This is also valid for connections to indirectly associated corporations, as the likelihood to be connected with an affected corporation increases with the number of nods ([316], p.221f). That also goes hand in hand with the **complexity** of a supply chain, which can, as well as the **sector** ([78], p.90), lead to higher impacts and longer recovery. But a positive effect of large networks can be financial, physical and psychological **support** from partners, for example in helping to find alternative suppliers more easily ([316], p.209f), when the original source is not going to reopen again. From a precautionary sense, the implementation of **alternative producers** or at least the evaluation whether there are alternatives can help to limit the extent of disasters ([136], p.152; [55], p.55). Also a **diversified supply chain** with partners not solely within a limited geographic area helps to enhance supply chain resilience and reduce recovery times ([316], p.211ff). The goal is, not solely to have partners within the affected area, because the '**physical proximity** of transaction partners has a negative effect on short-term recovery from region-specific shocks through the disruption of local supply chain networks' ([316], p.212). Suppliers with strong relationships to **partners outside the affected area** '[are] [...] less likely to face shortages of supplies or demands and more likely to receive support' ([316], p.221). In general the **distance** to the source of origin has a positive correlation to the value of assets damaged (here and following [78], p.87). Those characteristics can also be summarized under the term **topology**). Of course, a potentially affected area cannot completely be estimated, that is why decentralization aspects should be considered within supply chain design from the start in order to mitigate negative outcomes of disasters. Moreover the **duration** of the experienced effects plays a significant role, which is also influenced by the above mentioned characteristics, like the availability of second sources ([136], p.152). If available, **insurance** can reduce the time to recover, too. There is often 'in developing countries micro and small-business owners rarely [...] insurance to cover natural disasters' ([78], p.65). That is why each region should be analysed separately, considering individual region characteristics, as they can vary considerable between them (see chapter 5). To sum up, 'a proper structure of the chain can enable resiliency and a quicker or even partly a proactive response.' ([319], p.249), what makes supply chain design an integral part of supply chain risk management.

Since a more general approach on country / region level is pursued in this work, and no specific corporation cooperated on supply chain characteristics, the field of investigation would be too large, if all factors affecting the impacts of natural disaster would be included. Therefore the choice of a representing factor is presented in the next chapter.

#### **4.3.2 Recovery times and deviation in delivery times**

To assess the impact of natural disasters on supply chain performance, a value that portrays such influences best has to be found. Wildemann (2006a [387]) shows that for procurement risks especially the supplier risk is of relevance (see figure 4.4). This goes along with previous findings, where in supply chain risk assessment approaches the majority focused on supplier risk or supplier selection (what is presented in chapter 4.4.1). Based on that, the assumption comes into place that especially supplier characteristics shall be investigated further.

In more detail Wildemann (2006a) revealed that meeting the agreed quality and the adherence of delivery dates are the most governing factors when logistics risks' impact on performance is evaluated, as can be seen in figure 4.5. While the expected quality is specified very differently in each corporation

Figure 4.4: Performance impact of procurement risks ([387], p.74)

and context, the delivery time deviation is easily quantifiable (even without deeper knowledge of underlying production specifications). Moreover delivery dates are in most corporations tracked on an ongoing basis ([181], p.27), resulting in an appropriate data basis for further investigations. Beside this are also comparisons between regions and countries regarding delivery time deviations possible, which suits the approach to be developed in this work. Additionally the delivery date deviation is a key performance indicator (KPI - a value measuring the degree of fulfillment for a strategic important corporational activity) and therewith a suitable, and well established, estimator for performance impacts of natural disaster. Even so it must be noted that the survey conducted by Wildemann (2006a) is too small to be representative, the aforementioned remarks are still valid. So delivery time deviations are tracked on an ongoing basis in nearly every corporation, it is a key performance indicator and it enables performance comparisons on a global scale. That is why this indicator is also used within the impact assessment in chapter 6.

Summarized 'the delivery date deviation, [...] the period between the planned delivery date, i.e. the last delivery date accepted by the customer and confirmed by the supplier [...], and the actual date of delivery' ([370], p.23). In reference to Buscher et al. (2010) Λ is the realized delivery time, which is the planned delivery time λ plus the delivery time deviation *X*.

Figure 4.5: Performance impact of logistics risks ([387], p.74)

$$
\Lambda = \lambda + X
$$

As natural disasters often result in large delivery time deviations, it is also possible that *X* = ∞, describing the total failure of a supplier ([42], p.131f). Also a non-permanent closure can result in deviations, assuming no other measures, like safety stock are in place. Moreover closures are one of the most common effects natural disaster have on business operations ([295], p.22), which can also hamper the ability to remain viable ([314], p.4). That leads to the assumption that a delivery time deviation lasts at least as long as a corporation is closed (while no ex-ante actions are taken). Given all that, business closure times are also incorporated within the model development in chapter 6. Other reasons that can affect a corporations' ability to operate in a normal manner are ([158], p.3; [381], p.8):


In the end those reasons can result in delivery time deviations or any other performance impact. To define the research gap more precisely (the incorporation of natural disaster in supply chain risk assessments) the literature review on current supply chain and country risk approaches will be presented next.

### **4.4 Risk assessment approaches in the context of disasters**

To evaluate the need to develop a method to assess the impacts of natural disasters on supply chain performance, a literature review on supply chain risk assessment and country risk assessment approaches was completed. This differentiation is chosen, as the supply chain is embedded in the surrounding system of different places, e.g. countries (see chapter 5).

#### **4.4.1 Supply chain risk assessment**

Most identified approaches refer to the selection or evaluation of suppliers. This goes along with the aforementioned fact, that especially supplier risks influence the level of performance impacts (chapter 4.3.2).

The first author in table 4.1 Gaudenzi et al. 2006 ([116]) used the **Analytic Hierarchy Process (AHP)** and included four different objectives, which are 'on time delivery', 'order complete', 'order correctness' and 'damage / defect free' which are all presented through the risk areas 'transport', 'manufacturing', 'order cycle', 'ware housing' and 'procurement'. They revealed that 'on time delivery' is the most important sub-objective when it comes to the assessment of supply chain risk, while procurement has a high impact on it. Their findings support the selection of delivery time deviations as estimator for performance impacts of natural disaster on supply chains, as mentioned in chapter 4.3.2 as well. Other authors that used AHP are Bayazit et al. 2005 ([20]), Wu et al. 2006 ([391]) and Zaim et al. 2003 ([395]). For the selection of suppliers Bayazit et al. (2005) incorporated 'logistical performance', 'commercial structure' and 'production', while the first is differentiated in 'delivery performance' and 'cost analysis'. Within 'logistical performance' they found 'delivery performance' to be more important than 'cost analysis'. The authors found 'production', with the most important sub-criteria 'product specifications', the main factor in vendor selection. Wu et al. (2006) is beside this the only reviewed source that distinguishes explicitly between 'internal and external risk' values, and 'controllable', 'partially controllable' and 'uncontrollable' risks. In essence, the authors considered natural disasters explicitly within the assessment. As mentioned earlier natural disasters are part of the group of uncontrollable external risks (see chapter 2.3.2). A case study revealed that 'quality', 'cost', 'continuity of supply', 'on-time delivery' and 'engineering / production' are the key risk factors. But even though external uncontrollable risks were considered the importance of internal risks is rated much higher than external risks. This can lead to the problems already stated in chapter 2.1. Particularly because of their possibly more devastating impact, should external risks not be underestimated nor excluded from supply chain risk assessments.


Table 4.1: Supply chain risk assessment approaches


While the aforementioned authors used AHP, Jharkharia et al. 2007 ([154]) applied the **Analytic Network Process (ANP)**, what can be seen as an extension of the AHP, waiving the strict hierarchical order. Within the ANP Jharkhari et al. (2007) compared 'compatibility', 'quality', 'cost' and 'reputation', finding that compatibility between the corporation and service provider is the most important fact. For this reason the selection of suppliers should be a well-made decision within supply chain design, especially with the risk of natural disasters. However, the authors did not consider those risks in their approach.

A **real option approach** is done by Cucchiella et al. 2006 ([64]), and the considered risks divided in internal and external uncertainties, while external criteria refer most to characteristics of the procurement market (see 2.3.2), like the 'political environment'. Beside this they refer to the 'availability of capacity' the most important source of uncertainty, and therewith - like many other authors do - an internal originating risk aspect.

Moder 2008 ([187]) focuses on an overview of all relevant supply chain risks that should be considered in an effective supply chain risk management, i.e. an appropriate early warning system for supplies. As already mentioned also Moder (2008) discovered the risk of catastrophes irrelevant, having a lowish probability. That this could end up in disasters for the forthcoming of a corporation was also already stated (see graphic 2.1 in chapter 2.1). Disaster are at least considered by this source. Köglmayr and Bihler 2009 ([167]) apply an **Indicator based approach**. As they focus on supplier risk, all risks are considered external to the corporation. The risk of natural events is summarized as ecological risk and contains climatological and natural threats, but without any detail on the type of disaster. Even though not explicitly stated, natural threats are underestimated because the evaluation is again reasoned on the probability and potential impact of the triggering event. And as shown earlier, natural disasters occur relatively rarely in comparison to other events, leading to an underestimated risk of natural events.

Another approach focusing on supplier selection and assessment while using **Factor Analysis** is presented by Kannan and Tan 2009 ([162]). Their survey results reveal that aspects like 'geographical compatibility / proximity' as well as 'cultural match between the companies' (regarded as external risks) are ranked relatively low in their importance within supplier selection. While the 'ability to meet delivery due dates' is the most important factor. That fact again supports the choice of deviation in delivery times as estimator for performance impacts of natural disaster on supply chains. Even for the assessment of suppliers is 'on-time delivery' one of three prime aspects. Janker (2008) integrated mostly criteria of a suppliers' capability to meet certain customer demands, like quantity, quality logistics (contains also delivery time aspects) and so on. Although Janker 2008 ([153]) presents an extensive method with a variety of different assessment criteria for supplier evaluation, but external risks are neglected. Factor Analysis is also applied by Choi and Hartley 1996 ([54]), finding that 'consistency' is the most significant term in supplier selection. 'Consistency' consists of 'conformance to specifications, consistency in meeting delivery deadlines, quality philosophy, and prompt response to requests' ([54], p.337f). Again delivery details are ranked top for supplier selection. The only factor that could be classified as an external aspect is the 'geographical location', which goes along with the criteria that Kannan and Tan (2009) mentioned.

Further to that, Min 1994 ([184]) pursues a supplier selection using an **Multiattribute utility approach**. External facets are 'cultural and communication barriers' and 'trade restriction'. These aspects were already brought up in chapter 2.2, where the factors that lead to increased supply chain vulnerability are explained. However as this work focuses on natural disasters factors like cultural differences are not considered. Other aspects Min (1994) is referring to are the capabilities of a potential supplier in distinct categories. Again 'on-time delivery' is an aim within the category 'service performance', and the second most relevant in international supplier selection. Min (1994) summed it up when he wrote: 'in choosing the most appropriate supplier, the buyer should assess the length of the supply chain as well as the strength of the supplier's commitment for on-time delivery services [...]' ([184], p.27).

Zsidisin and Ellram 2003 ([400]) approved an **Agency theory** investigation. Again the authors are only indicating internal risk sources. Considered risk drivers are 'unanticipated changes in the volume requirements and mix of items', 'production or technological changes', 'price increases', 'product unavailability' and 'product quality problems'. 'Timely, accurate deliveries' are also part of the evaluation.

Within a **Scoring model** Blackhurst et al. 2008 ([21]) investigated internal as well as external supplier risk facts and they additionally incorporated natural disaster risks, like earthquakes, fire and flooding. Fire in this study is the most threatening risk when it comes to disruptions. It is here the first approach identified, that integrates external risks, especially natural events, on a nearly equal weight like internal harms (here 40% resp. 60%). But due to mentioned space limitations the authors just computed an example for quality and disruptions/disasters. That is why no statement on the importance of single risk categories can be found. Nevertheless are within 'logistics' 'on-time deliveries' considered. Thom 2008 ([312]) argues that the most important risk factors in production networks are globalization, transport, customer structure, supplier structure, coordination, natural disaster and terrorism. While natural disaster and terrorism for the surveyed managers play a minor role, as they are seen as less predictable and with a limited influential area. This position though can be very risky, because cascading effects from the area of impact throughout the connected points within a supply chain are totally neglected.

Raj Sinha et al. 2004 ([226]) identified 'standards', 'suppliers', 'technology' and 'practices' as main risk areas, through a **Generic aerospace supply chain operation reference model (SCOR)**. The only possible risks that occur from outside are the stated 'market uncertainties'. These are, following the classification in chapter 2.3.2, 'Supply-chain-external risks, related to specific procurement markets'. In the envisaged FMEA (Failure Mode and Effect Analysis) is the 'failure to deliver on time', reasoned for example through machine breakdown, after 'no clear market perception' from the category 'poor quality of incoming material' the second largest risk source. It is more over to emphasize that Raj Sinha et al. 2004 also differentiated between controllable and uncontrollable risks.

Fiege 2006 ([103]) gives an overview on regulations regarding the **KonTraG** (Gesetz zur Kontrolle und Transparenz im Unternehmensbereich - law that sets standards e.g. on risk management for publicly listed corporations). Even though natural disasters are referred to as relatively rare their devastating impacts are clearly stated and an insurance recommended, because a corporation will unlikely burden the losses on its own. Natural disaster are also defined as threats endangering businesses.

Within the **Supply network risk tool** Harland et al. 2003 ([129]) considered different types of risks from 'risks internal to the corporation' (like operation risks) to 'supply chain external risks, that are related to procurement markets' (like fiscal risks). Risks that are not related to specific procurement markets cannot be found in this literature source. But the different consequences in form of losses are classified, which are 'financial', 'performance', 'physical', 'social', 'psychological' and 'time'.

The review reveals that external risks are very seldom included in supply chain risk assessments. This may stem from the fact that external risks are regarded as less likely than internal risks, even though the impact might be worse which is often ignored. Only 4 out of 17 methods mentioned (stated as m in table 4.1) at least external risks. While only Wu et al. 2006 and Blackhurst et al. 2008 included natural disasters in their investigation (stated as x), even though they do not give information on performance impacts. It must be said additionally, that the provided literature review is just an excerpt on the variety of risk assessment approaches. But nevertheless it can be seen that most often suppliers are in focus when it comes to risk investigations, what can be justified on the fact, that a firm's performance is often correlated to supplier selection criteria ([162], p.15). That is also why the vulnerability of supply chains to different threats is strongly dependent on supplier characteristics (like the region they are operating in). Furthermore none of the presented approaches considered performance impacts. And what can be seen further, is the high relevance of delivery aspects, as those are stated by almost every author. Which, as already written earlier, emphasizes the choice of delivery time deviations as estimator for performance impacts of natural disaster (chapter 4.3.2). The related mathematics and calculations are presented in chapter 6.

Risk evaluation techniques for country assessment approaches are presented in the following chapter and are analyzed in regards to external risks and natural disaster considerations too.

#### **4.4.2 Country risk assessment**

In this sub chapter different approaches are presented that are well-known for several types of worldwide analysis, and there will also be an excerpt on national approaches given (an overview can be seen in table 4.2). The focus lies again on external risks and natural disasters.

The **Worldwide Governance Indicators** [305] provided by The World Bank give a glimpse on the governance quality of alternative countries. The categories that are investigated are: 'voice and accountability', 'political stability and absence of violence', 'government effectiveness', 'regulatory quality', 'rule of law' and 'control of corruption'. All relevant information can be found on the following web page: http://info.worldbank.org/governance/wgi/#doc, explaining the methodology and providing data sheets. As the name of the indicator already reveals natural disasters are not included in these calculations. The external risks that are considered, are, following the definition from chapter 2.3.2), associated to procurement markets resp. the location / country a corporation might be sourcing from.

Another aspect is provided by the **Global Competitiveness Report** [106] from the World Economic Forum. The actual report can be downloaded from: https://www.weforum.org/reports/the-global-competitiveness-report-2017-2018. Also previous reports can be found. Information on the methodology as well as a profound definition of all twelve pillars, that are part of the overall calculation are provided under: http://reports.weforum.org/globalcompetitiveness-index-2017-2018/appendix-a-methodology-and-computation -of-the-global-competitiveness-index-2017-2018/.

Considered in this index are 'institutions', 'infrastructure', 'macroeconomic environment', 'health and primary education', 'higher education and training', 'goods market efficiency', 'labor market efficiency', 'financial market development', 'technological readiness', 'market size', 'business sophistication' and 'innovation' (those are the twelve pillars). But even though the methodology is explained, it is sometimes not clear how exactly the values are calculated. Additionally, but of course as this is not the aim of this index, natural disasters are excluded. Some of the relevant values for the competitiveness index explain also a certain part of supply chain vulnerability and can therewith be found in the section 5.2.2 dealing with supply chain vulnerability factors.

The **Index of Economic Freedom** [109] persuades also a worldwide view, focusing on economic opportunities and prosperities within an economy. The web page can be reached at: http://www.heritage.org/index/. There are four key aspects incorporated, which are 'rule of law', 'government size', 'regulatory efficiency' and 'market openness'. Each category consists of different single factors that are used for calculation. 'Open markets' subsume for example: 'trade freedom', 'investment freedom', 'investment restrictions' and 'financial freedom'. A disadvantage might be, that for the calculation other indices or reports, like the World Competitiveness Report, are integrated. Moreover the concrete computation is not shown and some of the included values seem very hard to quantify, e.g. 'public trust in politicians'. Therefore it is not replicable where the final values derive from, but they give at least an impression about economic freedom which allows one to compare different countries / economies. Natural disasters are just stated (and just in the text of a country) when they had recently a major impact on the economy of a country, but are not part of the calculation.


Table 4.2: Country risk assessment approaches


The impact of terrorism is investigated through the Institute for Economics and Peace, publishing the **Global Terrorism Index** [151]. Information that is considered is for example, the socio-economic conditions under which an act of terrorism occurs, trends and the geopolitical drivers. The calculation is based on four weighted indicators for the last five years before the year of consideration. Those are: the 'total number of terrorist incidents in a given year', 'total number of fatalities caused by terrorists in a given year', 'total number of injuries caused by terrorists in a given year' and 'a measure of the total property damage from terrorist incidents in a given year'. The methodology is clearly and comprehensible stated in the appendix, which is a plus compared to other reports. The report is to be found on: http://economicsandpeace.org/reports/. Other reports that are provided through the Insitute for Economics and Peace are e.g. The Positive Peace Report, The Global Peace Index or The Risk Report (which assesses the risk of conflicts and violence). Through the completely different aim, are natural disaster not included.

A **World Risk Report** [38] is published by The United Nations University, Institute for Environment and Human Security and Bü-ndnis Entwicklung hilft. This report shows how infrastructure of a country influences the impact of natural disaster. Beside this the attached World Risk Index calculates a disaster risk by multiplying the exposure of a certain region to natural disaster with the vulnerability of a society. This report is therewith one of the few that incorporates not only external risks that are associated to procurement markets, but also those that are not, like natural disasters. Five types of natural threats are included in the report: earthquakes, floods, cyclones, droughts and sealevel rise. The report and other details can be found here: https://ehs.unu.edu/. For definitions on vulnerability and exposure please see chapter 2. Factors that are used in this work to define the vulnerability to natural disaster can be found in chapter 5.2.2. And an overview on vulnerability factors that are applicable for alternative disaster types can also be found in the appendices A.3 to A.6.

Very common country comparisons are based on the credibility or likelihood of a country to default. A provider is for example Standard & Poors. Ratings from Standard & Poors extend from AAA to D, while AAA means that there is an 'extremely strong capacity to meet financial commitments', means D that there are 'payment default on a financial commitment or breach of an imputed promise; also used when a bankruptcy petition has been filed or similar action taken' ([248]). For a credit rating are five key aspects investigated: 'credit quality of the securitized assests', 'legal and regulatory risks', 'payment structure and cash flow mechanics', 'operational and administrative risk' and 'counterparty risk'. A detailed explanation of all criteria can be found in [247], whereas the exact calculations are not revealed. Germany, for example is rated AAA, The United States of America AA+ (state January 2017). Again natural disasters are not included, as financial aspects come first here. Nevertheless **Credit ratings** [248] enable a country comparison based on financial issues.

**Statistics on natural disasters** on a worldwide basis are done by MunichRe, called NatCatSERVICE [193]. Disasters are classified in 'geophysical events', 'meteorological events', 'hydrological events' and 'climatological events'. With this analysis tool it is possible to generate maps and graphs that suit the relevant topic best, e.g. the number of flood events for a specified region. Details on disaster types and sub-types can be found in chapter 3.

More granular are indices and **reports on national level**, like the probability of natural hazards for Columbia from IDEA (Instituto de Estudios Ambientales, Universidad Nacional de Colombia) [99], for Mexico from ERN (Especialistas en Evaluación de Riesgos Naturales) [97] or for Austria from the Austrian Research Centre for Forests, Department of Natural Hazards [17]. These three institutions are just an example to show that reports from and for single countries also exist. It is obvious that those investigations are often related to natural disasters, while worldwide approaches are often not.

One of the key findings on worldwide and national reports and indices is therefore: natural disasters are more likely investigated on a minor scale, while worldwide reports focus on other topics than natural disaster. Moreover none of the mentioned approaches has a relation to supply chains or the impact of natural disaster on supply chain performance. It is revealed that for country risk methods the same problem exists like for supply chain risk assessments. While supply chain risk management techniques tend to neglect natural disaster impacts, country approaches have no relation to supply chains. This insight supports the need for development of an assessment approach that evaluates the impact of natural disaster on supply chain performance. The details on the methodology used in this work can be found in chapter 6. It must additionally be emphasized that this review can just be an excerpt, but they show the current state of the art and the methodical gap.

#### **4.5 Résumé**

The impact natural disasters can have on supply networks can be manifold. Perturbations in the flow of goods are generally referred to as indirect economic impacts, which was explained in chapter 4. Furthermore supply chain influences are differentiated in disruptions, which are longer lasting and have a wider extent and disturbances. Due to the expected scope it is more likely that natural threats cause supply chain disruptions instead of disturbances. To examine the impact of natural disasters on supply chain performance it is necessary to define what it is exactly that makes a supply chain vulnerable to those events. Therefore the two major components a networks consists of must be analyzed, which are location and transport. As vulnerability while goods are under transportation is very hard to examine, the location can be analyzed by the concept of vulnerability of places (chapter 5). The vulnerability of a production facility or supplier is defined through characteristics of the system / location in question (e.g. a country or state). Those factors can be of social, economic, physical and environmental nature and are quantified using indicators. All factors and indicators used in this work can be found in chapter 5.2.2. Apart from the criteria raised from the surrounding of a location there are also characteristics of the chain that can increase or decrease their vulnerability to certain risks. Since the location shall be in focus here, and no concrete supply chain can be used for evaluations, attributes of the supply chain itself are not incorporated. To assess the aforementioned performance impacts the key performance indicator *deviation in delivery times* is chosen, as it is easily quantifiable, monitored in nearly every corporation and can even be interpreted without deeper knowledge of the underlying production process. Other KPIs, like quality, are too different between corporations. A quality standard that might be acceptable for construction companies might not be so for car manufacturers. The literature review on supply chain risk and country risk approaches revealed that just a few of the existing methods refer to natural disasters perspectively supply chains and none to performance impacts. Therefore a new approach needs to be developed, which is presented in chapter 6.

## **5 Vulnerability of places - i.e. location**

As stated in chapter 4.2 the location is the primary source of investigation here. The concept of vulnerability of places, as well as the application to the research question is explained in the following sub-chapters, followed by a detailed overview of the identified vulnerability factors and their related indicators.

### **5.1 Concept**

Generally speaking 'place' means 'location' ([320], p.8076), which in the context of this work is the location of a production facility or a supplier. In a wider context the location could be a country or narrowed down even further a county or city. Depending on the field of investigation and the system in question the extent of 'place' may vary. As stated earlier, several indicators are available on a country level only (see chapter 3.2). That is why it is useful to start research with a country perspective and try to get a more detailed overview in the next step. Additionally, some factors are only reasonable when they are considered on country level and are not applicable for smaller areas ([383], p.326). The two main components that define place vulnerability are 'those factors of the environment that lead to increased potential for hazardous events to occur, [...]; and those characteristics of the people and places that make them less able to cope with and rebound from disaster events' ([68], p.106). Those factors can be classified in social, economic, physical and environmental vulnerability criteria, as can be seen in figure 5.1 ([29], p.932; [332], p.41; [333], p.32; [166], p.193). The four categories enable thereby an explanation of differences or inequalities between places ([67], p.243). Even for nearly similar systems the vulnerability can be divergent. This makes it necessary to analyze every place separately and to not expect similarities ([67], p.242; [320], p.8078).

#### **Vulnerability**

'pre-event [...] characteristics or qualities of [...] systems that create the potential for harm' ([68], p.2006)

#### **Social vulnerability**

'susceptibility of social groups to the impacts of hazards, [...] their resilciency, or ability to adequately recover from them' ([68], p.2006)

#### **Economic vulnerability**

'country's proneness to exogenous shocks [...] from a number of inherent economic features' ([34], p.2009)

#### **Physical vulnerability**

'refer mainly to [...] susceptibilities of location and the built environment' ([332], p.41f)

#### **Environmental vulnerability**

'describe the state of the environment within a region' ([111], p.157f)

#### Figure 5.1: Vulnerability of places

The overall vulnerability contains all 'pre-event, [...] characteristics or qualities of [...] systems that create the potential for harm' ([68], p.2008). On one hand it can be the characteristics of the region a supply chain is embedded in, and / or on the other hand it can be the characteristics of the supply chain itself (see chapter 4.3.1). So the term 'system' can have different meanings. This work refers to the location as the focal point. **Social vulnerability** is understood as the 'susceptibility of social groups to the impacts of hazards, [...] their resiliency, or ability to adequately recover from them.' ([68], p.2006). As this definition implies 'social vulnerability [...] [is] linked to the level of well-being of individuals, communities and society' ([332], p.42). Often social susceptibility is the result of social inequalities and is described using 'individual characteristics of people [like] [...] age, race, health, income, type of dwelling unit [and] employment' ([67], p.243). A detailed review on social factors can be found in chapter 5.2.2. From an **economic** point of view the **vulnerability** is a 'country's proneness to exogenous shocks [...] from a number of inherent economic features' ([34], p.2009). Of course smaller entities than countries can also be considered, but this requires respective data. Factors are the 'economic status of individuals, communities and nations' ([332], p.42) here, which are defined in more detail in chapter 5.2.2. The **physical factors** 'refer mainly to considerations and susceptibilities of location and the built environment. It may [also] be described as 'exposure" ([332], p.41f). Single criteria are for example building and infrastructure characteristics, as shown in chapter 5.2.2. 'Exposure describes the elements at risk to a natural disaster' (here and following [383], p.329f), while physical proximity to the source enhances the risk. For example 'the susceptibility of a human settlement to be affected by a dangerous phenomenon due to its location in the area of influence of the phenomenon' ([46],p. 13). Thus, location is strongly related to topology and shall not be mistaken by the total vulnerability of a location. The '**environmental vulnerability** describes the state of the environment within a region' ([111], p.157f), which is the ecosystem that surrounds the system under consideration. Examples for relevant factors are the state of environmental degradation or the number and area of protected territories (see chapter 5.2.2). For the analysis of different locations, 'each factor is characterized by a set of proxy indicators that are generalized at the national and sub-national scales' ([47], p.113). The chosen indicators are, apart from the aforementioned vulnerability factors, presented in the chapters 5.2.2 to 5.2.2. A complete overview of identified factors and indicators (from the intensive literature review) can be found in the appendices A.3 to A.6.

To gain a profound understanding of the inherent vulnerability of a place, it is important to note that 'social, economic, [physical] and environmental sectors are all interlinked, [and] reliable indices should take all these factors into consideration' ([121], p.3). The interaction of vulnerability factors can be seen in figure 5.2.

Additionally the characteristics of the disaster, like probability (frequency), intensity, geographical extent and duration are part of the investigation, as the possible negative outcomes are also dependent on those ([332], p.36; [166], p.192; [383], p.329). Moreover, the type of the disaster can be of relevance ([59], p.16; [333], p.32; [35], p.3), caused by differences for example in onset and duration. For example droughts are the result of long-term developments (time with low precipitation), while floods have a very sudden onset. Details on that can be found in chapter 3.1. The necessary information on natural disaster can be derived from historical data ([29], p. 932; [238], p.43).

After defining the prerequisites for appropriate indicators, subsequently the factors of the four vulnerability categories with their related indicators will be presented.

Figure 5.2: Interaction of vulnerability factors ([332], p.41)

### **5.2 Vulnerability indicators**

To evaluate the vulnerability of a location, it is firstly necessary to identify the relevant factors and indicators, which are presented in the following chapters. As this work is not focusing on specific disasters, a more generic approach was chosen. This means that the factors shown in chapter 5.2.2 to 5.2.2 are those mentioned in the literature for all types of disasters. And 'there are certain factors that are likely to influence vulnerability to a wide variety of hazard' ([36], p.153). Additionally a more general method is useful, when 'we wish to undertake comparative assessments of vulnerability at the national level' ([36], p.153), which is intended as a first step, as already mentioned in chapter 5.1. Investigations on county or city level might be step two. Nevertheless, a detailed overview on disaster specificities in vulnerability indicators is given in appendix A.3 and the following.

#### **5.2.1 Prerequisites for indicators**

Firstly it must be clear which prerequisites indicators have to be fulfilled, to be applicable. As mentioned several times the selection of factors and indicators generally 'depends on the purpose of the study, the research discipline being explored and the final application' ([89], p.14). Indicators are, despite other methods, used here, as they 'seem to be useful media, because they synthesis complex state-of-affairs such as the vulnerability of regions, households or countries into a single number that can be easily used [...]' ([140], p.198). They enable decision makers to reduce 'complexity, measuring progress, mapping, and setting priorities' ([70], p.603) and are 'useful to communicate complex issues from science to policy or the general public' ([140], p.204). But it is important to note, that 'indices reflect only the current state of the environment and must be constantly reviewed to ensure accuracy.' ([121], p.2f). Therefore, the historical values are used in chapter 6 only as a starting point. Moreover, 'vulnerability indicator[s] [do] [...] not give us information on when in the future harm will occur' ([140], p.201), they express merely the suspectibility to possible future harm from natural disasters.

To choose suitable indicators the following aspects must be taken into account ([91], p.20 <sup>1</sup> ; [155], p.188; [89], p.15f; [383], p.330):


As not all items in the stated literature are suitable for the research focus, only fitting characteristics were chosen. Data availability and quality refers to the aspect that data for all indicators must be available and that the source must be reliable. Apart from this it is essential that the indicators already 'recognised by researchers [are] as important'([89], p.15f) for the evaluation of vulnerabilities. Quantitative values are more comprehensive and reduce the influence of subjectivity within analysis, leading also to the aspect of objectivity. Moreover quantitativeness encourages a wider indicator acceptance ([89], p.15f). As a comparison between different regions (e.g. countries) is planned, the indicators must be available for all regions and times in question, enabling an international or inter-regional comparison. Additionally the indicators must

<sup>1</sup>The original source, stated in UNEP 2006 is not available anymore. That is why the information is taken from this source.

be easy to understand, so that they are 'useful to the public, policy-makers, and programme administrators' ([91], p.20), or in general the decision makers which are probably the supply chain risk managers in the context of this work. Last but not least data must be affordable. It is not useful when the costs for data purchases exceed the benefits.

The identified 'vulnerability indices can [therefore] help [to] identify and prioritise vulnerable regions, [...] raise awareness, and can be part of a monitoring strategy' ([202], p.23). All vulnerability factors and their indicators presented in the following chapters are derived from an intensive literature review, taking the aforementioned prerequisites into consideration. Following the classifications found in the literature sources the identified vulnerability factors are also differentiated in social, economic, physical and environmental aspects.

#### **5.2.2 Indicators**

#### **Factors of social vulnerability**

One important aspect to evaluate the differences in vulnerabilities to natural disasters, are social factors. 'Social vulnerability is [often] the product of social inequalities. It is defined as the susceptibility of social groups to the impacts of hazards, as well as their resiliency, or ability to adequately recover from them' ([68], p.103). To understand dissimilarities it is therefore necessary to evaluate the system (place) in question. 'What needs to be analyzed is how the structure of a society determine[s] the way in which a hazard is likely to affect it' ([45], p.26). Social facets can thereby explain the contrasts between or within regions (in combination with the later mentioned economic, physical and environmental factors) concerning the extent of natural disaster impacts ([68], p.102, [45], p.14, [392], p.382, [333], p.18, [244], p.101). Furthermore, the social system itself can be the source of vulnerability ([45], p.14). Factors contributing to the vulnerability of people / societies are shown in table 5.1. It is important to note that single factors cannot explain the overall vulnerability of a person, it is more often the interaction or combination of several factors ([89], p.17). Figure 5.3 portrays the percentage of authors referring to a certain category of social factors <sup>2</sup> . What can be seen is, that individual's

<sup>2</sup>The values presented in the graphic show the percentages for all identified literature sources, not just the ones stated in table 5.1. Moreover are different sources counted once within each factor

Figure 5.3: Social vulnerability factors: percentage of authors referring to different characteristics

characteristics and population aspects are stated most. These and all other single entries are now discussed in detail.

**Age** and **gender** are the factors that are stated most in the literature, and it is assumed that they can in fact explain the majority of differences in social susceptibilities. Often gendered vulnerability is the result of 'historically and culturally specific patterns of relations in social institutions, culture and personal lives' ([93], p.159). It is expected that women are more vulnerable, due to their role as caregiver, caring for children, elderly, whole families, or disabled, as well as due to lower wages and sector-specific employment ([133], p.7, [107], p.36, [71], p.21, [189], p.238f, [332], p.42, [67], p.246).

In addition, this role results in some societies lack of education and also restricted mobility, enhancing their vulnerability even more. The indicator used is 'gender per percentage of population', while a higher percentage of females concludes with more vulnerability. In addition elderly and adolescents are in need for help when a disasters strikes, but are lacking the necessary physical resources to respond ([176], p.812, [71], p.21). This goes along with the aforementioned aspect that women rather than men care for them, making women more vulnerable. Another aspect is 'the elderly [...] tend to be more reluctant to evacuate their homes in a disaster' ([71], p.21). The higher the percentage of younger or older people, the higher a society's vulnerability.

category and not just once for the main category (here social), resulting in a sum greater than one in the pie chart. This aspects are valid for all percentage analysis.

Furthermore **population growth** and **density** can be used to describe the social vulnerabilities within a region or place. With faster population growth, the inherent susceptibility increases as well, as the surrounding structures are often too slow to adapt ([67], p.248). Apart from that, the need for general care and post-disaster help increases in areas with high population densities. And on top of that more people are affected when they are densely populated near a triggering event. This is accompanied by the fact that 'the top countries at risk in terms of killed per year are the most populated countries (China, India, Indonesia, Bangladesh), whereas small islands states [...] come first in terms of killed per million inhabitants per year' ([217], p.1156). Indicators are the number of housing units (giving an overview on the density of built structures), population density (explaining how much people live in a certain area) and birth rates.


Table 5.1: Factors of social vulnerability


Social vulnerability also depend on the level of **education**, measured by governmental expenditures, showing the social development of a society. A higher educational level is linked to greater expected earnings (here and following [67], p.248), which is enabling a society a better compensation of disaster impacts. While low educational levels may restrain the ability to understand warnings, and in the aftermath of an external shock may complicate applications for financial relief. The status of **employment** also influences social vulnerability, as people that are already in the need of help, suffer more often, e.g. due to a lack of financial resources for recovery. In general 'poor people have less money to spend on preventative measures, emergency supplies, and recovery efforts. Although the monetary value of the economic and material losses of the wealthy may be greater, the losses sustained by the poor are far more devastating in relative terms' ([71], p.20). But even 'the potential loss of employment following a disaster exacerbates the number of unemployed workers in a community, contributing to a slower recovery from the disaster' ([67], p.247). As '**health** care providers including physicians, nursing homes, and hospitals are important post-event sources of relief' ([67], p.248f), this aspect is pictured in access to medical services. 'The lack of proximate medical services will lengthen immediate relief and longer-term recovery from disasters.' ([67], p.2488f).

In summary, social vulnerability is mostly linked to the level of development ([217], p.1149), which is also dependent on economic factors, which are explained in the following chapter 5.2.2.

#### **Factors of economic vulnerability**

Economic vulnerability is described as 'a country's proneness to exogenous shocks [...] from a number of inherent economic features' ([34], p.232). In contrast to chapter 5.2.2, in which individual characteristics of people and society were discussed the focus will now shift to economic features (see table 5.2). Those factors can refer to economic attributes of the whole community, population groups or individuals ([111], p.157) and explain the varying impacts of disaster on economic activities ([316], p.211).

Figure 5.4: Economic vulnerability factors: percentage of authors referring to different characteristics

Figure 5.4 shows that especially the (socio)economic status and the economy itself are the prime contributing factors to economic vulnerability. As 'wealth enables communities to absorb and recover from losses more quickly' (here and following [67], p.246ff) the **economic capacity** is a good indicating factor of economic strength. But even though a high capacity suggests enough financial resources, there are normally more assets at risk to natural disasters, resulting in higher material losses. Meanwhile poorer societies often face higher losses in terms of affected or killed people and livelihoods. That is why 'in less developed regions of the world, low losses reflect a deficit of infrastructure and economic assets rather than a low impact' ([332], p.13). 'Obviously, similar exposures with contrasting levels of development lead to drastically different tolls of casualties' ([217], p.1149). Infrastructural aspects are hence incorporated as **transport structures** and **accessibility of resources**. Moreover, the availability of transport infrastructure (even in the case of disasters) is an integral asset to prevent deviations in delivery times (please see chapter 4.3.2). Apart from this wealth can be expressed in **health expenditures**, showing if economies can afford to care to a certain extent for their population.


Table 5.2: Factors of economic vulnerability


The **size of the domestic market** and the **availability of resources** is also of relevance, as a high dependence on other countries enhances the occurrence of side-effects. Even if an economy is not directly affected, the indirect effects to the flow of goods can be tremendous (see chapter 4). Meanwhile 'countries with a relatively small domestic market have very few options but to resort to exports, [...] those with limited natural resources tend to be highly dependent on imports' ([34], p.232). Based on that, the size of an economy, as well as **import and export concentration** are included in further investigations. Additionally the **type of economic activity** is of relevance as economies are more vulnerable when they are dependent on primary commodities. Which are often the first affected when a disaster strikes. Also 'a singular reliance on one economic sector [...] creates a form of economic vulnerability for countries' ([67], p.253). Indicators are the percentage of urban and rural population as well as the percentage of arable land.

The stability of economies also has a strong influence on procurement policies and strategic sourcing decisions ([72], p.113f), as it is always desirable to source in more stable countries, minimizing the own supply chain risk. As can be seen in comparison to chapter 5.2.2 some of the sub-categories double, like health. If this is the case, the indicator does refer to different aspects of vulnerability, which can be social, economic, physical and environmental vulnerability. In this example, the 'public expenditures on health' are related to economic capabilities (so economic vulnerability) and the 'number of physicians' explain the cover of medical care (so social vulnerability).

The third category, physical vulnerability is next explained.

#### **Factors of physical vulnerability**

Physical factors explain the susceptibilities of locations (topology) and the built environment ([332], p.41f, [111], p.157). Once again population aspects in the literature are stated most, followed by infrastructure, as can be seen in figure 5.5. The relevant vulnerability factors and their indicators are stated in table 5.3.

A high concentration of people (**population density**) is associated here with unsafe physical settlements ([243], p.63), making them more vulnerable to the impacts of disaster. But as the factor 'population density' was already discussed in chapter 5.2.2 it will only be mentioned here and will not be repeated. Strongly related to dense concentrations of people is also the level of **urbanization**.

Figure 5.5: Physical vulnerability factors: percentage of authors referring to different characteristics


Table 5.3: Factors of physical vulnerability

Reasons for the impairing effect of urbanization are, the subsequent growth of urbanized areas, which can be seen in higher risk areas, as for instance in coastal metropolises like Houston (here and following [186], p.27). Houston also recently came into focus when Hurricane Harvey made its way through Texas ([191]). Moreover, the high concentration of assets is reinforcing vulnerability, as more value can be affected within a limited geographical area. Additionally the replacement rate of old buildings can often not keep up with urban growth, resulting in a high level of dwellings that do not meet current standards.

To gain the true coverage with infrastructure, the available kilometers (already stated in chapter 5.2.2) are here set in ratio to the land area. The greater the coverage, the lower a region's vulnerability. The aspect of **dwelling location** is included, because floods are the most frequent natural disasters (see chapter 3.2.1) all over the world. And as more people continue living in areas below 5 meters (N.N.) there is a higher risk of flooding and thus higher expected disaster impacts for that region. The last of the four described vulnerability categories - environmental vulnerability - contains aspects of the environment, which are discussed in the next paragraph.

#### **Factors of environmental vulnerability**

Figure 5.6: Environmental vulnerability factors: percentage of authors referring to different characteristics

Environmental vulnerability 'describe[s] the state of the environment within a region' ([111], p.158) and 'is related to the risk of damage to the natural environment' ([195], p.182). All categories within this category can be found in table 5.4. Degradation in this category is the most important facet of vulnerability and 21 % of the authors in the literature review referred to it (figure 5.6). Damage or irreversible change to the ecosystem is incorporated as **environmental degradation**, lowering the resistance to natural disaster. Another factor of environmental stress is **ecology**, referring to the marine and terrestrial protected areas in relation to total land area. Where 'reserves may be one of the few ways managers could off-set some other environmental damage and build resilience against natural events that can damage the environmental support system' ([100]).

The number of **borders** captures the risk of neighboring effects. The more adjacent countries a country has, the greater the risk of trans-boundary impacts (here and following [100]). Also, the country dispersion is of importance, as highly dispersed countries are also more prone to effects from other countries and often lack natural barriers to environmental risks. Regarding to floods also the length of coastlines is included, leading to higher susceptibility the longer a coast line ([333], p.3). Similar to the indicator *population living in areas where elevation is below 5 meters (N.N.)* from the previous chapter, the land area under this sea level is at risk here. The higher the percentage of land area below this value, the higher the risk for environmental damage through flooding within the observed area. Additionally a larger **land** mass enables a better compensation of negative effects from disasters, because more non-affected area is available. Likewise **population density** is for social and physical susceptibilities, an important variable of environmental vulnerability, what makes population the most influencing factor for the overall vulnerability. Population density here is 'a proxy measure for pressure on the environment resulting from the number of humans being supported per unit of land. A higher number of people increases pressure on the environment for resources, for the attenuation of waste and physical disturbance of the environment' ([100]).


Table 5.4: Factors of environmental vulnerability

Finally it can be stated that there are numerous factors for the different vulnerability levels of distinct places and that the research focus, as well as the underlying system, must be considered. While those aspects that are most often considered in the literature and that are related to the topic - impact assessment of natural disaster on supply chain performance - are discussed here, other indicators are in place as well. A complete overview over the factors identified within the literature review can therefore be found in the appendices A.3 to A.6. The influence of the mentioned vulnerability indicators on supply chain performance will be tested in chapter 6.

## **6 Impact assessment**

Below are the steps of the **S**upply **C**hain **Perform**ance Impact Assessment of **N**atural **D**isaster (SCperformND) described in general, followed by a mathematical foundation of each step. The calculation follows the Loss Distribution Approach (LDA), quantifying delivery time deviations as estimator for the performance impact caused by natural disasters. Steps are:


### **6.1 Methodology**

The KPI 'deviation in delivery time' will be used as an estimator for the performance impact natural disaster can have on supply chains. Please see the references made in chapters 4.3 and 4.3.2 for further explanations. To compute those impacts a Loss Distribution Approach (LDA) was chosen, as this method allows a combination of severity (the extent of delivery time deviations) and probability (the frequency of deviations), resulting in a compound distribution which states the frequency of certain deviations for distinct locations.

Subsequently necessary steps for the methodology will be described in general, followed by a mathematical foundation in the next chapter.

#### *Step 1 - Identification of business closure times*

To repeat, 'the delivery date deviation, [...] is the period between the planned delivery date, i.e. the last delivery date accepted by the customer and confirmed by the supplier [...], and the actual date of delivery' ([370], p.23). Since the deviation is referring to a time horizon, the loss / severity (distribution) to be determined is also expressed in entities of time ([393], p.123). The severity distribution is one part of the Loss Distribution Approach mentioned above, stating the extent of delivery time deviations.

As there was no corporation specific delivery data available, due to already stated reasons, and in the conducted literature review no source on delivery time deviations caused by disasters could be found, the assumption is made that the deviation lasts at least as long as a corporation is closed after a disaster hit. A consequence is the business closure time representative for the deviation in delivery time. Business closure time is understood as the time a corporation needs to reopen after being affected by a disaster, while during the interruption 'no business activities, such as production or service, are possible' ([170], p.12). Apart from the assumption it is necessary that no preventive measures are in place, as no statement about it could be made. A reason for that could be that measures are not known and their influence on delivery time deviation reductions is not jet analyzed. Even within the extensive literature review done in this work, just a few authors could be identified that investigated affects of natural disasters on business closures, resulting in 32 studies that are used for the approach in this work. The small number of sources is not surprising as the affects of natural disasters on businesses are generally underestimated and mostly people's safety instead of business continuity is addressed when it comes to prevention (see chapter 1.1). But even though no individual delivery time data could be used in this work, it is again important to note that this information is tracked in nearly every corporation and can easily be applied within the introduced method.

#### *Step 2 - Identification vulnerability indicators*

To explain the differences in delivery time deviations after a catastrophe in distinct locations, it is necessary to identify the factors which can lead to varying vulnerability levels. All factors and their indicators used in this work have been presented in chapter 5.2.2 and are based on an intensive literature review. The four major categories are social, economic, physical and environmental factors with the relating subcategories shown in figure 6.1.

Figure 6.1: Vulnerability factor categories

#### *Step 3 - Relation between business closure times and vulnerability indicator*

To find indicators for different business closure times for varied places, a linear regression model is used. All vulnerability indicators identified in step 2 (out of the four vulnerability factor categories - social, economic, physical and environmental) are tested as explaining variable. The detailed computation for that step can be found in chapter 6.4.

#### *Step 4 - Frequency of natural disaster*

For the performance impact assessment the probability of the triggering event must be determined, which is defined as 'the number of times it occurs within a specified time interval' ([178], p.359). The distribution is referred to as frequency and is the second part of the LDA. The frequency distribution indicates the probability of business closure, respectively delivery time deviations. All relevant data for the frequency analysis is obtained from statistical time series ([79], p.3; [119], p.xv).

#### *Step 5 - Compound distribution*

The compound distribution is calculated combining the severity distribution (the business closure times from step 1) and the frequency distribution (the frequency of natural disaster from step 4). The distribution explains the probability of certain extents of closure times after a natural disaster. Whereby the business closure times are determined through the different vulnerability indicators from step 2, leading to varied distributions for distinct locations.

Furthermore the mathematical formulations are demonstrated for the above introduced steps.

#### **6.2 Step 1 - Identification business closure times**

The relevant data on business closure times was, according to the methodology description in chapter 6.1, gathered in a literature review, giving the results shown in table A.8 in the appendix. The general procedure, using a probability mass function, to reach comparability between the stated values in the literature review is described below. Exemplary figure 6.2 shows the determination of a probability mass function and a cumulative distribution function. Part (a) states the given data - the time a certain amount of corporations closed, for example did 50% not close while 10% closed for 3 days. The number of days a business closed after a natural disaster hit is defined by the discrete random variable *Z*. The probability mass function (illustrated as (b) in figure 6.2) is a function given by

$$f\_{\mathbf{Z}}(z) = P(\mathbf{Z} = z) \tag{6.1}$$

where *f<sup>Z</sup>* presents the probability that a business closed for *z* days. The cumulative distribution function of *Z* (part (c) in figure 6.2) is a function given by

$$F\_{\mathbf{Z}}(z) = P(\mathbf{Z} \le z) \tag{6.2}$$

where *F<sup>Z</sup>* defines the probability that a business is not closed for more than *z* days. For example is *FZ*(2) = 0.9, meaning that with 90% probability closure time is 2 days or less.

Figure 6.2: Visualisation step 1

Because the varied literature sources used partially very different levels of detail to describe business closure, ranging from a single percentage of closed corporations to several gradations within one day, the mean closure time (MCT) is introduced and used for further calculations. As the MCT could be calculated for all sources, it enables the aforementioned comparability between the differing studies. Other values, for example the standard deviation etc. were tested, but could not be computed for all sources.

Given *f<sup>Z</sup>* as the probability mass function of *Z*, the *MCT<sup>i</sup>* is defined as the expected value of *Z<sup>i</sup>* and is calculated for every data source *i* as follows:

$$\text{MCT}\_{i} = \text{IE}[\mathbf{Z}\_{i}] = \sum\_{z \in \mathbf{IR}} z f \mathbf{Z}\_{i}(z) \tag{6.3}$$

Figure 6.3: Mean closure time per event-ID

If more than one data set was available for the same event, the MCT is the arithmetic mean over all studies. With *n* as the number of sources utilized, the MCT is calculated as:

$$MCT = \frac{1}{n} \sum\_{i=1}^{n} MCT\_i \tag{6.4}$$

Figure 6.3 illustrates the mean closure times for all investigated events, as well as the respective disaster type and the affected country. For a repetition of disaster types, please see chapter 3.2. The mean closure time values are shown in appendix A.8.

As a next step factors which serve as explaining momentum for varied mean closure times between distinct locations are identified.

#### **6.3 Step 2 - Identification vulnerability indicators**

According to step 1, the indicators (the values that measure the factors in figure 6.1) for the nine countries indicated, for which business closure times were investigated in the 32 studies mentioned above. Considered countries are The United States of America, Germany, New Zealand, South Africa, Pakistan, Thailand, Australia, Canada and The United Kingdom. The underlying disasters and the year the event took place can be found in table 7.2 in chapter 7.1 and the literature sources utilized in appendix A.10.

As the indicators change over time and the data on business closure times are from a specific event year, it was necessary to search the data from the respective event year, too. That is all the more important, as finally a statement shall be given on future delivery time deviations, given the current vulnerability indicators. The intensive research resulted in more than 3500 single data entries, for the indicators in table 6.1 which had been already explained in chapter 5.2.2. For collection and calculations is Microsoft Excel®2010 used. Despite the effort to gather all indicator values from the respective event year, it was sometimes not possible. In those cases the next data year was chosen, if the deviation was one year maximum, as it is assumed that a possible change is acceptable within one year. For greater deviations the available data has been used to update or backdate to the event year, up to a deviation of eight years give or take. Tests with a wider time interval resulted in a total change of the underlying regression (step 3), so a range of eight years was chosen. Figure 6.4 illustrates the realized linear interpolation to calculate the missing data points. For a value in time α between a time *a*<sup>0</sup> and *a*<sup>1</sup> is *f*(α) with *f*(*a*0) and *f*(*a*1) approximated through ([384], p.16):

$$f(\mathfrak{a}) = f(a\_0) + \frac{f(a\_1) - f(a\_0)}{a\_1 - a\_0} \left(\mathfrak{a} - a\_0\right). \tag{6.5}$$

Unfortunately it was still not possible to gather the entire value set for the United Kingdom from 1978 and 1979, consequently the United Kingdom had to be excluded from further investigations.


Table 6.1: Factors of vulnerability and explaining indicators


### **6.4 Step 3 - Relation between business closure times and vulnerability factors**

As can be seen in figure 6.3, the mean closure times for the investigated countries and natural disaster types are varied. To explain those differences

Figure 6.4: Linear interpolation for missing data points

a linear regression is used, testing the relation between mean closure times and factors of place vulnerability (out of step 2). The dependent variable *Y* = *mean closure time* (*MCT*) is thereby explained through one or more explanatory variable(s) *X* - the vulnerability factors and their related indicators.

First a simple linear regression was computed, so a single vulnerability indicator *X* and an absolute term β<sup>0</sup> explains the mean closure time *Y*. The values *xi* , *y<sup>i</sup>* for *i* = 1,...,*n* are displayed through a linear model, where the *x<sup>i</sup>* are non stochastic and *y<sup>i</sup>* realizations of stochastic variables *Y<sup>i</sup>* (here and following ([146], p.155ff).

$$\begin{aligned} Y\_i &= \beta\_0 + \beta\_1 X\_i + U\_i & \quad &i = 1, \dots, n\\ \beta\_0, \beta\_1 &\in \mathbb{R} \end{aligned} \tag{6.6}$$

The *x*1,..., *x<sup>n</sup>* are known and fixed (non stochastic) values, given that all *x<sup>i</sup>* are non-identical. For the error variable *U<sup>i</sup>* applies:

$$\begin{aligned} \text{IE}[U\_i] &= 0 \\ \text{Var}[U\_i] &= \sigma^2 \text{ with } 0 < \sigma^2 < \infty & \quad i = 1, \dots, n \\ U\_1, \dots, U\_n &\text{ are not correlated} \end{aligned} \qquad \text{i} \qquad \text{i} = 1, \dots, n $$

The simple linear regression enabled a first glimpse on the relation between mean closure times and a single indicator, but it is expected that more than one indicator is needed to explain the connection. This is supported by the statements already made in previous chapters, where for a holistic vulnerability assessment, factors from different categories must be considered (social, economic, physical and environmental). Therefore also multiple linear regression models are tested.

Within the multiple linear regression with *K* non stochastic explanatory variables, the observed values (*xi*1,..., *xiK*, *yi*) for *i* = 1,...,*n* are described by a linear model with all *xi j* being non stochastic, while all *y<sup>i</sup>* are realizations of a random variable *Y<sup>i</sup>* with

β0,β1,...,β*<sup>K</sup>* ∈ I*R* (here and following [146], p.161). All *xi j* for *i* = 1,...,*n* and *j* = 1,...,*K* are known and fixed, so that the following equation applies:

$$Y\_{l} = \beta\_{0} + \beta\_{1}X\_{l1} + \dots + \beta\_{K}X\_{iK} + U\_{l} \qquad \qquad \qquad l = 1, \dots, n \tag{6.8}$$

For the error variable *U<sup>i</sup>* the above remarks from equation 6.7 are still valid.

For the selection of potential explanatory variables the following 4 groups were chosen and analyzed:


At last the aggregation of four values was chosen, because the regression with one indicator out of each category (what is recommended based on the remarks already made) requires a minimum of four values. That is why, less than four indicators are not tested nor more than four to avoid overfitting of the underlying regression model. The statistics degree of freedom is therewith 16. To automate the model fit and due to the complexity with more than 3500 single data points a Python script was written and applied ([303]). Figure 6.5 show exemplary the results calculated with the Python script.

The final selection of a regression model is based on the goodness of fit, focusing on the coefficient of determination *R* 2 and an appropriate condition


Figure 6.5: Output of multiple linear regression

number (which 'measures the sensitivity of the estimates to changes in *X*' (here and following ([18], p.86)). 'A large condition number indicates that small changes in *X* can cause large changes in the estimated coefficients'. 'Generally, if the condition number is less than 100, there is no serious problem with multicollinearity. Condition numbers between 100 and 1000 imply moderate to strong collinearity, and if [...] [it] exceeds 1000, severe multicollinearity is indicated' ([188], p.298).

### **6.5 Step 4 - Frequency of natural disaster**

Following the already made remarks on the LDA apart from the severity of events a frequency distribution is also necessary, which is calculated, using historical data for the countries and event types mentioned in table 6.2.


Table 6.2: Natural disaster types considered for the different regions

'The frequency *N<sup>T</sup>* with a probability function *p<sup>N</sup>* explains [thereby] the distribution of the number of events within a specific time interval *T* ([292], p.50)'.

As *N<sup>T</sup>* can only be an integer value, the modeling with a discrete distribution is recommended, whereby possible distributions can be: Poisson, Exponential, Gamma, Binomial, Negative binomial and Panjer ([292], p.50; [62], p.48ff). Whereby 'the chosen distribution needs to be the distribution that gives the closest approximation to the observed data' ([144], p.73). The suitability of a distribution can be evaluated taking the following aspects into consideration ([144], p.73ff):


For the analysis of natural disasters especially the tail of a distribution is of interest, thus it is expected that this aspect has a high influence on the choice of a distribution in those cases. If a frequency distribution for future time intervals is modeled in dependence to the already realized loss events, a counting process can be introduced, which allows to consider the progression of loss events and not only the very number within a given time interval ([62], p.48ff). Within the high range of possible processes especially the Poisson process is used to describe highly random behavior ([164], p.1; [43], p.5), like natural disasters. Additionally the Poisson process is seen as a standard ([2], p.7) and it is also one of the simplest time-continuous processes ([377], p.61) to apply. An easy application is necessary particularly for the acceptance and comprehensibility in an applying corporation. Therefore the Poisson process is used and also described here.

The counting process *N*(*t*),*t* ≥ 0 is a homogeneous Poisson Process (HPP), if the following requirements apply (here and following ([2], p.8); ([62], p.60)):

	- For all *t* > 0 applies 0 < *P*(*N*(*t*) > 0) < 1 (6.11)
	- *N*(*t* +*u*)−*N*(*t*) ∼ **Pois**(λ*u*) for any *t* ≥ 0, *u* > 0 (6.12)
	- The random variables *N*(*ti*+1)−*N*(*ti*),*i* = 0,...,*n*−1 (6.13)

are for any 0 = *t*<sup>0</sup> < *t*<sup>1</sup> < ... < *t<sup>n</sup>* independent of each other

**Pois**(α) expresses the Poisson distribution with parameter α, which means with α = λ*u*, where λ presents the expected number of losses within the time interval [0,*u*], apply:

$$P(N(\mu) = n) =: p\_n(\mu) = \frac{(\lambda \mu)^n}{n!} \cdot e^{-\lambda \mu} \tag{6.14}$$

$$\text{for } n \in \mathbb{N} \text{ and } \mu > 0$$

$$\text{IE}[N(\mu)] = \text{Var}[N(\mu)] = \lambda \mu \tag{6.15}$$

Based on this mathematical formulation the written Python Script (see figure 6.6) computes the frequency for all given events as difference between two successive events (in days) and draws the related histogram, like the exemplary one in figure 6.7.

Figure 6.6: Python script to create a frequency plot

Figure 6.7: Empirical distribution of differences between successive events for USA

### **6.6 Step 5 - Compound distribution**

The compound distribution (as the next step of the LDA) is now calculated as the combination of the severity distribution from step 1 and the frequency distribution out of step 4, giving the total loss distribution for a specific time interval *T* ([292]). The process is graphically illustrated in figure 6.8 and mathematically described below.

Figure 6.8: Compound distribution (based on [3], p.286)

The convolution (*f* ∗ *g*)(*x*) of two ordinary functions *f*(*x*) and *g*(*x*) in I*R n* is a function defined through the following integral (taken from [147], p.300):

$$(f\*g)(\mathbf{x}) = \int\_{\mathbb{R}^n} f(\mathbf{y})g(\mathbf{x}-\mathbf{y})d\mathbf{y} \tag{6.16}$$

Whereby for both functions apply, that the respective random variables are stochastically independent. That means the severity (the extent of impact) has no influence on the probability (frequency) and vise versa. The cumulative loss *L*(*t*) within a time interval [0,*t*] can be expressed as an accumulated claim process ([2], p.4f), which can be seen in figure 6.9.

Figure 6.9: Accumulated claim process ([2], p.5)

'The random variables *X*1, *X*2, ... , *X<sup>n</sup>* correspond the extent of loss (per loss event), [while] *X<sup>i</sup>* > 0 quantifies the extent of the i-th loss event' ([2], p.5). With *N*(*t*) as counting process (see step 4), which refers to the number of losses within the time interval [0,*t*], applies (here and following [2], p.4ff):

$$L(t) = \sum\_{i=1}^{N(t)} \mathbf{X}\_i \tag{6.17}$$

$$\text{with } L(t) = 0 \text{ for } N(t) = 0 \tag{6.18}$$

'As *N*(*t*)is a homogeneous Poisson Process (HPP), *L*(*t*)is a compound Poisson process (CPP)', and it applies:

$$\mathbb{E}[L(t)] = \mathbb{E}[N(t)]\mathbb{E}[X] \tag{6.19}$$

$$\text{Var}[L(t)] = \text{IE}[N(t)]\text{Var}[X] + \text{IE}[X]^2\text{Var}[N(t)] \tag{6.20}$$

Figure 6.10: Loss distribution and Value at risk (taken from [110], p.3)

With *F n*∗ as n-th convolution of the distribution function F, it applies for the distribution function *Gt*(*x*):

$$G\_t(\mathbf{x}) = P(L(t) \le \mathbf{x}) = \sum\_{n=0}^{\infty} P(N(t) = n) P(L(t) \le \mathbf{x} | N(t) = n) \tag{6.21}$$

$$\xi = \sum\_{n=0}^{\infty} P\_n(t) F^{n\*}(x) \text{ with } x \ge 0, \ t \ge 0 \tag{6.22}$$

'Of particular interest [for the Loss Distribution Approach] are the (1 - α) quantiles *Q*1−α[*L*(*t*)] = *G* −1 *t* (1 − α) for the given confidence levels 0 < α < 1[...] α, as they correspond the Value at Risk of the compound distribution with a confidence level α (for the regarded period [0,*t*])' ([2], p.17). <sup>1</sup> The graphical illustration can be found in figure 6.10.

In the context of this work the VaR (Value at risk) states the delivery time at risk from natural disasters. Below is the approach presented within a case study.

<sup>1</sup>The term *L*(*t*) is equivalent to *S*(*t*), which is used in the literature source. Due to notations this term was changed in this work.

## **7 Application SCperformND approach**

### **7.1 Vulnerability indicators with highest relevance to explain delivery time deviations**

Following the steps (here **step 1**) introduced in the previous chapter, first the business closure times were extracted from the 32 studies identified within the literature review. The complete overview on the data set can, as said, be found in appendix A.8. Due to a lack of space and for reasons of comprehensibility only just an excerpt of the complete table (see table 7.1) is shown and explained ([382]; [379]).


Table 7.1: Business closure times

For easier computability for each given disaster a so called event-ID (*k*) is introduced, which is number 1 in table 7.1. That event-ID refers to the Loma-Prieta Earthquake, which hit California in 1989. The complete overview on events and their associated event-ID is given in table 7.2. Additionally the literature sources are classified with ID, which was necessary as sometimes more than one author investigated the same event. The results Webb et al. 2002 ([382]) (ID 1) revealed on business closure are, that of the 910 survey respondents, 24,5% (stated as 0,245 in table 7.1) did not close. Therefore the interval span is 0, as well as the mean closure time. The following row shows that 44,6% closed for one hour (left-bound) up to 3 days (right-bound). One hour is 1 <sup>24</sup> *day* <sup>=</sup> <sup>0</sup>,042, as all values are stated in days<sup>1</sup> . The values can of course also be converted to other units, like hours, if needed. It must only be ensured that all values have the same unit.


<sup>1</sup>Sources that used another time resolution had been converted to days.


To gain the mean closure time (the expected value of the distribution) for each row, the left-bound is multiplied with % businesses closed within interval.

*MCTrow* = % businesses closed within intervall*row* ·left-bound [days]*row* (7.1)

Within the example for the second row the MCT is: 0.019 = 0.446 · 0.042. If the closure time was stated as *inf*, referring to a business that closed forever and no information about the point in time when that happened was made, it is assumed that the final closure took place after one year. In those cases the leftbound is multiplied with 365. The interval center was tested for calculation too, but resulted in approximately the same values, that is why the left-bound was kept. The sum over the last column for each ID gives the mean closure time for each literature source and the investigated region. For the given ID 1, the MCT is around 16 days. As already mentioned, sometimes more than one author investigated the same event, in these cases the mean closure times are averaged over the event-ID, providing the mean time the businesses were closed during / after the Loma-Prieta earthquake of 1989 here.

Let k be the k-th event with Event-ID k and n the number of data sources investigating the closure times for the event with Event-ID k. The MCT for Event-ID k is stated as follows:

$$\text{MCT}\_{k} = \frac{1}{n} \sum\_{i=1}^{n} \text{MCT}\_{i} = \frac{1}{n} \sum\_{i=1}^{n} \sum\_{\text{row}=ID} \text{MCT}\_{\text{row}} \tag{7.2}$$

For event-ID 1 the MCT is:

$$MCT\_1 = \frac{1}{2} \left( \left( 0 + 0.019 + 0.78 + 0.592 + 14.6 \right) + \left( 0 + 6.008 \right) \right) = 11 \text{ days} \tag{7.3}$$

The MCT for all events is graphically illustrated in figure 6.3 in chapter 6.2.

Within **step 2** are the vulnerability indicators identified which shall be part of the analysis. These are (in this work) the ones already mentioned in table 6.1 in chapter 6.3. Depending on the context of analysis it is also possible to exclude or include other indicators. As stated earlier those in table 6.1 are the ones the majority of authors mentioned and those that fitted the topic of this work best. So here is the complete set of indicators from table 6.1 tested within **step 3** - the relation between business closure times and vulnerability indicators. Therefore the indicators from the respective event year and the MCTs were condensed within a table, of which an extract can be seen in table 7.3. To prevent negative results the values are logarithmized.

To identify the vulnerability indicators with the greatest influence on business closure several linear regression models were tested. The comparison between all tested regressions revealed that the regression selecting one indicator out of each category has the best 'combination' of a high coefficient of determination *R* 2 , with condition numbers that are better than in the other cases. Even though the condition numbers are in any tested case higher than recommended (the problem can be explained out of the relatively few data sets), this regression model shows the best results. With a coefficient of determination *R* <sup>2</sup> of 0.575 the 4 indicators in table 7.4, which are also pictured in the extract of the Python Script in figure 7.1, show the greatest impact on business closure times respectively delivery time deviations. To repeat, the equation in 7.4 explains


Table 7.3: Extract of data for regression 7.1 Vulnerability indicators with highest relevance to explain delivery time deviations


Table 7.4: Regression coefficients

once more the calculation of the MCT. The terms β stand for the respective values of the regression which are multiplied with the indicator values *x*.

$$\mathbf{y} := \ln{\mathbf{M}CT} \quad = \begin{array}{c} \mathcal{B}\_1 \ln{\mathbf{x}\_1} + \mathcal{B}\_2 \ln{\mathbf{x}\_2} + \mathcal{B}\_3 \ln{\mathbf{x}\_3} + \mathcal{B}\_4 \ln{\mathbf{x}\_4} \\ \end{array} \tag{7.4}$$

$$\begin{array}{rcl}\text{MCT} &=& e^{\text{y}} = e^{\beta\_1 \ln x\_1 + \beta\_2 \ln x\_2 + \beta\_3 \ln x\_3 + \beta\_4 \ln x\_4} \end{array} \tag{7.5}$$

$$MCT \quad = \quad e^{\beta\_1 \ln x\_1} \cdot e^{\beta\_2 \ln x\_2} \cdot e^{\beta\_3 \ln x\_3} \cdot e^{\beta\_4 \ln x\_4} \tag{7.6}$$

$$\begin{array}{rcl}\text{MCT} &=& \boldsymbol{\alpha}\_1^{\beta\_1} \cdot \boldsymbol{\alpha}\_2^{\beta\_2} \cdot \boldsymbol{\alpha}\_3^{\beta\_3} \cdot \boldsymbol{\alpha}\_4^{\beta\_4} \end{array} \tag{7.7}$$

A high birth rate (indicator *x*1) indicates high population growth, which 'focuses on the potential for damage relating to expanding human populations. It signals increasing rates of habitat damage, exploitation of natural resources and disposal of wastes [...]' (here and following [100]). 'The greater numbers of people increases pressure on the environment' and enhances the vulnerability of people and businesses. Other challenges related to large population growth are that housing often lack quality, that 'social services networks may not have had time to adjust to increased populations' ([67], p.248) and that it can 'result in a lack of infrastructure and therefore of disaster management capacity' ([237], p.94).

The second variable *x*<sup>2</sup> is referring to forest areas, focusing on 'the loss of natural vegetation cover in a country' (here and following [100]). 'By affecting people's livelihood, environmental degradation increases the vulnerability of some communities and can also contribute to increasing the vulnerability of others through migrations' ([229], p.124). 'The existing level of environmental degradation is of particular relevance for evaluating the vulnerability [e.g.] of

Figure 7.1: Regression results case study

floods, droughts and cyclones. The effects of environmental degradation might vary with climate conditions and affect areas differently [...]' ([237], p.96). But deforestation in general 'leads to soil erosion, loss of nutrients and marginality of agriculture' and 'can lead [...] [to] new pattern of flood, drought, fire or landslide hazards' ([333], p.5). 'Areas of natural vegetation are viewed as refuges [...]. Natural forests and vegetated areas are also likely to be important areas for groundwater intake, soil production, *CO*<sup>2</sup> - oxygen relationships and attenuating air and water pollution. A country's resilience to future hazards will be related to this rate.

The road network (indicator *x*3) is part of the 'institutional infrastructure [that] provide[s] the framework for disaster mitigation, preparedness and response activities' ([237], 88ff). 'Loss of sewers, bridges, water communications, and transportation infrastructure compounds potential disaster losses. The loss of infrastructure may place an insurmountable financial burden on smaller communities that lack the financial resources to rebuild' ([67], p.247). The larger the transport network the easier corporations can also reroute if a certain area is affected through natural disasters.

The fourth variable *x*<sup>4</sup> is the population living in areas where elevation is below 5 meters (as % of total population). The elevation influences the vulnerability to floods and cyclones strongly ([237], p.91ff), as 'areas of lowlands are those that will tend to be the first to flood' ([100]). The relevance of this indicator might stem from the fact that the majority of data sources investigated floods. The higher the percentage of population living in areas under 5 meters (per cent), the higher their vulnerability. The high susceptibility of those areas is also related to the fact that coastal regions (which are often areas of lower elevation) are the most productive living areas of a country ([100]), and because of that also the most densely populated ones ([102], p.1590; [213], p.1390). This results also in a high concentration 'of residential housing, transport and energy supply infrastructure' ([80], p.478) and also in a high amount of businesses located. Moreover coasts are often popular tourist areas and an important asset of economic activity ([213], p.1390). Thus due to their specific location, rapid urbanization and high concentration of assets those areas are highly vulnerable to natural disasters ([296], p.1) and with it the population and businesses located.

Coming up next are the results of SCperformND compared to the observed data from the literature studies.

#### **Model validation**

To validate the model a in sample validation was done, testing the fit between the observed closure times and the closure times calculated with SCperfomND on the basis of the four indicators identified above. The literature values as well as the calculated values of the MCTs can be found in table 7.5 and are graphically illustrated in figure 7.2.

In a perfect model all data points lie directly on the orange line through the point of origin, the further a data point is away from this line the larger the respective model error.


Table 7.5: Model validation


$$\frac{\partial MCT}{\partial x\_{1}} \quad = \beta\_{1} \boldsymbol{x}\_{1}^{\beta\_{1}-1} \cdot \boldsymbol{x}\_{2}^{\beta\_{2}} \cdot \boldsymbol{x}\_{3}^{\beta\_{3}} \cdot \boldsymbol{x}\_{4}^{\beta\_{4}} \quad = 0.006412 \cdot \boldsymbol{x}\_{1}^{2.243193} = 1.85 \quad (7.8)$$

$$\text{for} \qquad \boldsymbol{x}\_{1} = 12.5$$

$$\frac{\partial MCT}{\partial \boldsymbol{x}\_{2}} \quad = \boldsymbol{x}\_{1}^{\beta\_{1}} \cdot \beta\_{2} \boldsymbol{x}\_{2}^{\beta\_{2}-1} \cdot \boldsymbol{x}\_{3}^{\beta\_{3}} \cdot \boldsymbol{x}\_{4}^{\beta\_{4}} \quad = 0.382412 \cdot \boldsymbol{x}\_{2}^{-0.251711} = 0.1575 \quad (7.9)$$

$$\text{for} \qquad \boldsymbol{x}\_{2} = 33.9$$

$$\frac{\partial MCT}{\partial \boldsymbol{x}\_{3}} \quad = \boldsymbol{x}\_{1}^{\beta\_{1}} \cdot \boldsymbol{x}\_{2}^{\beta\_{2}} \cdot \beta\_{3} \boldsymbol{x}\_{3}^{\beta\_{3}-1} \cdot \boldsymbol{x}\_{4}^{\beta\_{4}} \quad = -32500 \cdot \boldsymbol{x}\_{3}^{-1.571718} = -6.123 \cdot 10^{-7} \quad (7.10)$$

$$\text{for} \qquad \boldsymbol{x}\_{3} = 6662841.41 \tag{7.10}$$

$$\frac{\partial MCT}{\partial \boldsymbol{x}\_{4}} \quad = \boldsymbol{x}\_{1}^{\beta\_{1}} \cdot \boldsymbol{x}\_{2}^{\beta\_{2}} \cdot \boldsymbol{x}\_{3}^{\beta\_{3}} \cdot \boldsymbol{\beta}\_{4} \boldsymbol{x}\_{4}^{\beta\_{4}-1} \quad$$

Figure 7.2: Model validation

For the observed closure times the overall MCT is 52.95 days in the mean per event, while SCperfomND calculates 39.47 days. The mean error, defined as difference between mean closure time from SCperformND and the observed closure time is therefore -13.48 days and the mean absolute error 22.02 days. The deviations between all calculated and observed values can also be seen in figure 7.3.

Additionally a sensitivity analysis is done, testing the change of the MCTs when the underlying vulnerability indicator is changed infinitesimal. Therefore each of the four indicators was tested for the influence a change will bring on MCT, while all other indicators remain steady. The partial derivations 7.8 to 7.11 for *x*1, *x*2, *x*<sup>3</sup> *and x*<sup>4</sup> form the analytical basis for the plots in figures 7.4 to 7.7. When the rate of birth (per 1000 population) changes one percent the MCT changes about 1.85 days (see figure 7.7 and equation 7.8). This indicator shows also the greatest influence on business closure in comparison the other three used within this work. From the figure and mathematics it can also be derived that a change of forest area per one percent results in a change of MCT by 0.16 days. While the influence of the road network change is −6.123·10−<sup>7</sup>

Figure 7.3: Comparison MCT from literature and SCperformND

on MCT and the percentage of people living in areas where elevation is below 5 meters has an influence of 0.37 days when changed for one per cent.

Figure 7.4: Sensitivity analysis Rate of birth (per 1000 population)

Figure 7.6: Sensitivity analysis roadways in km

Population living in areas where elevation is below 5 meters [% of total population]

Figure 7.7: Sensitivity analysis Population living in areas where elevation is below 5 meters (% of total population)

Below are now the identified indicators used within the application of SCperformND for a concrete sample region.

#### **7.2 Case study - United States of America**

In the previous chapter the indicators with the highest influence on business closure times had been identified. This section aims to present the SCperfomND approach for a sample region - the United States of America (USA). This explanatory location has been chosen, as the majority of sources (out of the 32 studies mentioned) investigated the USA, consequently the available data basis is broader than for the other regions (Germany, Canada, Australia, Thailand, Pakistan and South Africa). Following the introduced methodological sequence (see chapter 6) the first part is determined through the identification of business closure times. The indicators with the greatest impact on business closure times had been identified in chapter 7.1, are now for the calculation of a severity distribution (here the MCT) the four indicators: birth rate (number of birth per 1000 population), forest area (% of land area), roadways in kilometer and population living in areas where elevation is below 5 meters (% of total population) used. Different from the calculations already made, for the assessment of future business closure times, the latest values available are now applied. The indicator values can be found in table 7.6 and the respective mean closure time is calculated as follows:

$$\begin{aligned} \text{InMCT}\_{\text{USA}} &= 3.2432 \cdot 2.5257 + 0.7483 \cdot 3.5234 - 0.5717 \cdot 15.7121 + 0.1306 \cdot 0.9203 \\\\ &= 1.9654940798956 \end{aligned} \tag{7.12}$$

$$MCT\_{\text{USA}} = e^{1.9654940798996} = 7.14 \text{ days} \tag{7.13}$$

The second step (the identification of vulnerability indicators) and the third step (the relation between business closure times and vulnerability indicators) are already done as the indicators identified as most relevant in chapter 7.1 are utilized. Depending on the research question, the branch or corporate specific requirements can also have other indicators tested within regression, which can result in a different set of explaining indicators.


Table 7.6: Indicators USA

As the USA is the region of interest the frequency of natural disaster in the USA has to be calculated as frequency distribution (step 4). Therefore the sources mentioned in chapter 3.2 were utilized. The respective mathematics can be found in chapter 6.5. Within 33 years (1985-2017)<sup>2</sup> 486 natural disasters occured, of which were 431 floods, 17 hurricanes and 38 earthquakes. This is resulting in an average amount of 14.73 = 486 <sup>33</sup> events per year. Moreover, the distribution of the counting process *N*(*u*) has to be determined, following the below mentioned mathematics.

$$
\lambda \mu = \frac{K}{N} \tag{7.14}
$$

$$
\lambda = \frac{K}{Nu} = \frac{K}{U} \tag{7.15}
$$

$$U = N \cdot \mu \tag{7.16}$$


<sup>2</sup>This time interval has been chosen as the available information is better in quality and quantity from 1985 on.


With the average number of events per year with 14.73 and the interval *u* of 365 days is λ = 14.73 <sup>365</sup> <sup>=</sup> <sup>0</sup>.04035. The counting process follows therewith a Poisson distribution *Pois*(14.73). The respective probability mass function is illustrated in figure 7.8.

Figure 7.8: Probability mass function for *N*(*u*) = *Pois*(14.73)

The cumulative loss *L*(*t*) is, as already said in the previous chapter, then expressed as an accumulated claim process (step 5). To repeat:

$$L(t) = \sum\_{i=1}^{N(t)} X\_i \tag{7.17}$$

$$\text{with } L(t) = 0 \text{ for } N(t) = 0. \tag{7.18}$$

The distribution of *X<sup>i</sup>* (the extent of loss), which is also called the severity distribution was stated as the MCT - which is a single value. Therewith *X<sup>i</sup>*

is a degenerate distribution ([231], p.369) and *L*(*t*) is simplified to *L*(*t*) = *N*(*t*)·(*MCT*). For the expected value the following applies:

$$\mathbb{E}\left[L(t)\right] = \mathbb{E}\left[N(t) \cdot (MCT)\right] = (MCT) \cdot \mathbb{E}\left[N(t)\right] \tag{7.19}$$

$$\mathbb{E}\left[N(t)\right] = \mathbb{X}u\tag{7.20}$$

with *N*(*t*) ∼ **Pois**(λ*u*)

*L*(*t*) follows then *Pois*(14.73)· 7.14, which is illustrated in figure 7.9. The expected value of *L*(365) = 105.15 states that within one year it is expected that a corporation closes for 105.15 days due to a disaster. This value can be perceived as relatively high, but can be based on the fact that the USA were considered as one entity.

Figure 7.9: Probability mass function for *L*(*t*) = *Pois*(14.73)· 7.14

For the 95% quantile, with a confidence level α = 0.05 the VaR is

$$\mathcal{Q}^{-1}\_{36\\$}(0.9\mathsf{S}) = 122 \, d\mathrm{sys}.\tag{7.21}$$


Table 7.7: VaR comparison for sample regions

95% of business closure times are less than 122 days after a disaster. The 5% quantile states that in just 5% of all cases business closure times account for less than 89 days. Coming back to the question of supplier or production location decisions, this value must now be compared to the VaR of all other available alternatives. Therefore the methodology of SCperformND (see chapter 6.1 for repetition of the necessary steps) applied to the other six countries is used within this work. The respective values can be found in table 7.7.

Based on the MCT per year Germany has the lowest value as well as the lowest VaR. Without any further information Germany would therefore be chosen as production or supplier location. The least advantageous country in this example is New Zealand. However it is useful to include additional information, like operational aspects, which were not included in this work due to its context. But for a holistic vulnerability analysis it is necessary to incorporate all aspects that build the potential for harm to a corporation. Moreover it is useful to investigate further smaller areas (like states or communities) for more detailed statements, as indicator values can of course be varied within a country, leading to different MCTs and VaRs. As SCperformND was estimated on country basis, for a deeper analysis data of required granularity would be necessary, which was not available. Nevertheless the functioning of SCperformND was shown and it can easily be applied within a corporation as data on delivery time deviations with respect to business closures are normally tracked on a daily basis. Therewith SCperformND closes the research gap of supply chain risk methods lacking the consideration of natural disasters and the gap of natural disaster assessments lacking the consideration of influences of natural disaster on supply chain performance through a combined approach.

Based on the achievements the work closes with a summary and critical appraisal in the next chapter.

## **8 Summary and conclusion**

#### **8.1 Summary**

As can be seen in daily life, the risks of natural disasters are increasing, which is not just demonstrated by the latest global recognized disaster like Hurricane Harvey in 2017. To be competitive, corporations must be aware of those developments and should take the potential risk of natural disasters within supply chain design into consideration from the first day onwards. In the case of strategic decisions, for example supplier selection or location decisions should be evaluated if there is a risk for natural disasters in a specific region. It should always be kept in mind that one disaster strike can easily outweigh the benefits of for example low cost sourcing. So even external shocks are relatively rare in comparison to internal risks, their impact can be much worse. Nevertheless, recent developments have shown that corporations have not jet recognized the importance of that risk category. Most often internal risks are analyzed in detail, while external risks are ignored or even neglected. This is based on the common perception that there is no such urgent need to secure against rarely insecure events, when there are risks with high probabilities. Due to those facts, and the expected increase in frequency and intensity, natural disasters are a growing threat to supply chain resilience. This is analyzed in an intensive literature review, stating factors of increased supply chain vulnerability in chapter 2.

To understand the differences between risk categories, a corporation has to deal firstly with a general classification of supply chain risks as demonstrated in chapter 2.3.1. To emphasize the need for a deeper analysis of external risks, especially natural disasters, a distinction is introduced that allows to separate risks in: risks that can be associated with specific procurement markets and those that are not market related (see chapter 2.3.2). A specialty in terms of risks that cannot be associated to specific procurement markets (like natural disasters) is that the participants within a supply chain network do not have any influence on their occurrence. Only possible impacts can be reduced through appropriate measures.

To reveal the current status on research about the influence of natural threats on supply chains and their performance, a second literature review is conducted, showing the research gap in that area. Within the analysis of supply chain risk assessment methods in chapter 4.4.1, evidence on the lack of information is given, as natural disasters are rarely included within assessments. If risks from outside the supply chain network are incorporated it is remarkable that those risks are usually associated to the supplier. Following the stated classification this may be valid for risks that can be connected to the procurement market and therewith partly to a supplier, but not for natural disasters. Moreover, only one source is identified that included external risks in the computation, all others - if at all - only mentioned that there are also external risks. None of the identified approaches investigated the impact of natural disasters on supply chain performance. On the other hand, there are several global and regional natural disaster risk indices, reports and assessment approaches in place, focusing on the threat itself (see chapter 4.4.2). But like supply chain risk assessment approaches lack the consideration of natural disasters, they do lack any relation to supply chains. At the moment no approach exists that evaluates the influence of natural disasters on supply chain performance. Therefore a new approach is developed and explained in chapter 6 in this work.

To evaluate the performance impact of natural disasters on supply chains it is necessary to focus on the region they take place, therefore the concept of vulnerability of places is used and introduced in chapter 5. To assess the susceptibility of a region and to enable comparability between different places, vulnerability factors and their explaining indicators had to be identified. The main categories the factors can be assigned to are social, economic, physical and environmental aspects. While social characteristics refer mainly to the social system within a region, economic criteria focus on the economic. Aspects of the built environment, like infrastructure, are subsumed within physical factors and ecological matters are classified as environmental vulnerability factors. Within each of the four categories the most referred to and suitable for the topic were chosen and used within the impact analysis in chapter 6.

To measure the performance impact, the value 'deviation in delivery times' is defined as indicating variable. As information about that key performance indicator is tracked in nearly every corporation, it is quantifiable and allows easy applicability for different corporations. The delivery deviation is the time between a requested date of delivery and the realized one. Due to the scare data basis the assumption is made that a corporation could at least not deliver as long it is closed after a disaster hit. There were 32 studies identified, dealing with business closure times after a natural threat. Based on the information given in the literature the severity distributions were calculated for the different countries stated. The severity distribution expresses thereby the extent of closure times. As the realization of an impact is also dependent on the probability of the triggering event, the frequency distributions were calculated with the help of historical data sets. To finally gain a total loss function the Loss Distribution Approach is chosen, as it enables the convolution of the frequency and the severity distribution to one distribution, which expresses the distribution of loss within a given time interval. The resulting value is also called delivery time at risk.

After the presentation of the necessary steps for the **S**upply **C**hain **perform**ance impact assessment of **N**atural **D**isasters (SCperformND) approach and the mathematical foundation in chapter 6, a case study emphasized the application of the developed approach. The first case study identifies the vulnerability indicators with the greatest influence on delivery time deviations, which are: birth rate (number of birth per 1000 population), forest area (% of total land area), roadways in km and population living in areas where elevation is below 5 meters (% of total population). The chapter closes with an in-sample model validation - evaluating the deviations between observed and calculated values. The key vulnerability indicators serve in the second case study as input variables to calculate the potential extent of delivery time deviations for the sample region which in this case is the USA. With the SCperformND approach it is therefore possible to compare different places based on the identified indicators and to calculate the delivery time deviations. Based on that information recommendations for supply chain design decisions are possible, like the selection of potential suppliers or location decisions for own production facilities. The higher the potential delivery time deviations after a natural disaster, the higher the potential risk to source / produce in this area. It must be taken into consideration that for example the benefits of low cost sourcing can easily be outweighed when a disaster strikes. The SCperformND approach hence suggests to choose a region with lower delivery time deviations if there are alternatives. If not, the potential impacts on supply chain performance should at least be known and appropriate measures should be implemented. With the SCperformND approach it is therefore possible to identify vulnerability characteristics with the greatest influence on delivery time deviations as well as the extent of delivery time deviations for a specific region and time horizon, enabling the evaluation of performance impacts of natural disaster on supply chain performance. The approach closes the research gap on supply chain risk assessment methods lacking the consideration of natural disaster, and the gap on natural disaster risk assessments lacking supply chain performance aspects. SCperformND is therefore a powerful tool, extending operational supply chain risk methods to a holistic assessment.

The work closes with a summary and critical appraisal as well as directions for future research.

### **8.2 Critical appraisal and directions for future research**

As mentioned several times previously a practice partner would be an asset to gain real time data for the approach presented in this work. Although the principles of the **S**upply **C**hain **perform**ance impact assessment of **N**atural **D**isasters (SCperformND) approach were shown, additional data sets would enable the derivation of supply chain design recommendations for a specific corporation. Additionally the assumptions made, such as that a business could not deliver as long it is closed and that there are no measures in place to mitigate possible impacts, are obsolete, as the respective supply chain could be mapped in detail. Moreover, the simplification, that only direct connections between different locations are considered, has to be suspended for modeling a complex real world supply chain.

The small data base itself could also distort the presented results. To overcome this problem more research on business closure respectively delivery time deviations in the aftermath of a natural disaster is necessary or at least evaluations of real time data of a corporation. Hence the chosen indicator 'delivery time deviation' is tracked in nearly every corporation and is classified as KPI this could easily be done within a corporation. Therefore the available delivery data must be related to natural disaster events in the region of question. Above that it is due to the lack of data not possible to investigate the relevance of vulnerability indicators for different natural disaster types. This distinction might be useful as disaster types vary such as on speed of onset and duration. A broader data set could as well enable the test of combinations of more than four indicators as explaining variables for delivery time deviations, which is not possible here.

Furthermore, characteristics of a corporation can influence the extent of disaster impacts itself, which were not investigated, because the focus of this work is the vulnerability of regions and the evaluation of performance impacts based on characteristics of that area and not of a corporation. However this can bring further insights about which factors affect the impacts of natural threats on businesses.

Aside from that the transport of goods shall also be incorporated within an analysis, as the routes goods are transported by are at risk from natural disasters too. This aspect is excluded within the analysis of place specific vulnerability aspects in this work.

Another field of research is the investigation of smaller regions, like states. Based on the presented methodology it is possible to evaluate places of minor extent, when the relevant vulnerability indicators are first identified. Within future research several regions / places could be compared to others.

It might also be useful to extend SCperformND to an industry perspective, as some of the indicators identified might change in importance due to different requirements in distinct branches. Also the incorporation of business specific weights for the indicators could be investigated. Even the selection of another indicating variable for performance impacts can be evaluated, for some corporations a focus on the delivered quality might be an example instead of the presented delivery time deviations.

Future research should also focus on the extension of the approach presented with risk methods for operational risks. This is to gain an overall supply chain risk assessment. Only if all categories with potential for harms are considered, can risks be reduced effectively.

# **A Appendix**

### **A.1 Reasons for increased supply chain vulnerability**


TableA.1:Reasonsforincreasedsupplychainvulnerability



### **A.2 Supply chain risk classification - sources**


TableA.4:Supplychainriskclassification






### **A.3 Factors in social vulnerability**


Table A.11: Factors social vulnerability














## **A.4 Factors in economic vulnerability**


Table A.25: Factors economic vulnerability














## **A.5 Factors in physical vulnerability**


Table A.39: Factors physical vulnerability A.5 Factors in physical vulnerability









### **A.6 Factors in environmental vulnerability**


Table A.50: Factors environmental vulnerability A.6 Factors in environmental vulnerability













## **A.7 Natural disaster types - definitions**


Table A.64: Natural disaster types - definitions (extracted from[152], p.12ff and extended with [75], [92], [16], [15], [88])






### **A.8 Business closure times**


Table A.70: Business closure times (complete)






### **A.9 Vulnerability indicator USA 1989**


Table A.71: Values vulnerability factors USA 1989


### **A.10 Sources vulnerability indicator**


Table A.72: Sources vulnerability indicator

## **References**


im Rahmen des Projekts" Wirtschaftliche Chancen der internationalen Klimapolitik"(FKZ 90511504)". In: *Wuppertal Papers* 171 (2008).


inland - waterways - in - use - in - germany/ (visited on 12/10/2017).


80/research/ehcs96/tables/taba1- 5.htm (visited on 12/12/2017).


Full\_Report/Volume\_1, \_Chapter\_1\_State\_Level/ Louisiana/st22\_1\_009\_010.pdf (visited on 11/27/2017).


#### Produktion und Energie

Karlsruher Institut für Technologie (KIT) Institut für Industriebetriebslehre und Industrielle Produktion Deutsch-Französisches Institut für Umweltforschung

ISSN 2194-2404



Die Bände sind unter www.ksp.kit.edu als PDF frei verfügbar oder als Druckausgabe zu bestellen.


Die Bände sind unter www.ksp.kit.edu als PDF frei verfügbar oder als Druckausgabe zu bestellen.

Band 19 Sophia Radloff Modellgestützte Bewertung der Nutzung von Biokohle als Bodenzusatz in der Landwirtschaft. 2017 ISBN 978-3-7315-0559-4

#### Band 20 Rebekka Volk Proactive-reactive, robust scheduling and capacity planning of deconstruction projects under uncertainty. 2017 ISBN 978-3-7315-0592-1

Band 21 Erik Merkel Analyse und Bewertung des Elektrizitätssystems und des Wärmesystems der Wohngebäude in Deutschland. 2017 ISBN 978-3-7315-0636-2


Band 30 Kai Mainzer Analyse und Optimierung urbaner Energiesysteme – Entwicklung und Anwendung eines übertragbaren Modellierungswerkzeugs zur nachhaltigen Systemgestaltung. 2019 ISBN 978-3-7315-0882-3

Die Bände sind unter www.ksp.kit.edu als PDF frei verfügbar oder als Druckausgabe zu bestellen.


#### Band 33 Felix Hübner

 Planung und Modellierung des Rückbaus kerntechnischer Anlagen unter der Berücksichtigung von Unsicherheiten – Ein Beispiel zur Planung von Großprojekten. 2019 ISBN 978-3-7315-0911-0

Band 34 Kira Schumacher Public acceptance of renewable energies – an empirical investigation across countries and technologies. 2019 ISBN 978-3-7315-0948-6

Band 35 Daniel Fehrenbach Modellgestützte Untersuchung des wirtschaftlichen Potenzials sektorgekoppelter Wärmeversorgung in Wohngebäuden im Kontext der Transformation des Energiesystems in Deutschland. 2019 ISBN 978-3-7315-0952-3

Band 36 Mariana Burkhardt Impacts of natural disasters on supply chain performance. 2021 ISBN 978-3-7315-1020-8

Die Bände sind unter www.ksp.kit.edu als PDF frei verfügbar oder als Druckausgabe zu bestellen.

institut fÜr industriEBEtriEBslEhrE und industriEllE Produktion dEutsch-franZÖsischEs institut fÜr umWEltforschung

issn 2194-2404 isBn 978-3-7315-1020-8 m. Burkhardt

impacts of natural disasters on supply chain performance