A spatial-temporal vulnerability assessment to support the building of community resilience against power outage impacts

TFS-18790; No of Pages 20 Technological Forecasting & Social Change xxx (2016) xxx–xxx

Contents lists available at ScienceDirect

Technological Forecasting & Social Change

A spatial-temporal vulnerability assessment to support the building of community resilience against power outage impacts Thomas Münzberg ⁎, Marcus Wiens, Frank Schultmann Karlsruhe Institute of Technology, Germany

a r t i c l e

i n f o

Article history: Received 22 March 2016 Received in revised form 30 September 2016 Accepted 30 November 2016 Available online xxxx Keywords: Decision making Spatial-temporal vulnerability assessment Monte-Carlo simulation Delphi survey Community resilience building Power outages

a b s t r a c t Power outages are among the most serious Critical Infrastructure (CI) disruptions and require effective disaster management with collaboration of affected CI providers and disaster management authorities. To support building community resilience, we introduce a vulnerability assessment which allows an enhanced spatial-temporal understanding of initial power outage impacts. Using the assessment enables planers to better identify which and when CIs become vulnerable and how important they are in comparison to other CIs before the overall crisis situation escalates and unmanageable cascading effects occur. The assessment addresses the initial phase of a power outage and corresponding early measures of local risk and crisis management organizations according to the German disaster management system. The assessment is an indicator-based approach which is extended to consider time-depending effects through time-referenced demand and the depletion of Coping Capacity Resources (CCR). The estimation of the relevance of CIs regarding the provision of vital services and products is addressed by a modified Delphi method. In addition, an expert survey was conducted to shed light on the evaluation of coping resources. In this paper, we describe the components of the assessment and propose different aggregation approaches which each enhances the understanding of spatial-temporal impacts of a power outage, and, hence, increases the forecasting capability for disaster management authorities. For demonstration purposes, the assessment is implemented for the case of the city of Mannheim, Germany. © 2016 Published by Elsevier Inc.

1. Introduction Within the Energiewende, the German electricity system is experiencing a historic turnaround towards a more sustainable, more efficient, and smarter energy system, including renewable and low-carbon energy generation. However, the change also implies new and unknown risks. As a consequence, the risks accompanying power outages became an even more important issue in disaster planning. For disaster management, the electric power grid is already the most “critical” infrastructure, since all Critical Infrastructures (CIs) (e.g. hospitals, pharmacies, General Practitioners (GP), etc.) depend on a reliable electricity supply (see, e.g., der Vleuten and Lagendijk, 2010; Kröger, 2008; Luiijf et al., 2009; Pescaroli and Alexander, 2016). According to the Sendai Framework, the United Nation (UN) emphasizes the strategic target to reduce disruptions of basic services and to strengthen the resilience of communities (UNISDR, 2015). Making communities more resilient against the impacts of power outages means to increase “the ability […] to resist, absorb, accommodate to and recover from the effects […] in a timely and efficient manner, including through ⁎ Corresponding author at: Karlsruhe Institute of Technology, Postfach 3640, 76021 Karlsruhe, Germany. E-mail address: [email protected] (T. Münzberg).

the preservation and restoration of its essential basic structures and functions” (UNISDR, 2009). In the present paper, the theory of vulnerability is a theoretical concept that is used as a functional measure instrument to enhance resilience. Vulnerability expresses the degree a region or CI can be affected by a power outage. It is a measure of the CI service losses due to a power outage taking into account criticality and resilience of (potentially) exposed CIs. In order to measure and enhance community resilience against power outage impacts, several substantial challenges have to be faced. The following list is a non-concluding enumeration of the challenges that will be discussed in the following sections in detail: ­ Disaster management authorities have to plan for the accompanying risks of missing CI services taking into account spatial-temporal characteristics of power outage exposed CIs and citizens. Therefore, planning tools are required, which in particular allow analyzing the impacts on local level. ­ Each CI provider is responsible for its own disaster preparation. However, an effective disaster management planning can only be ensured by a joint collaboration between CI providers and the responsible disaster management authorities. Effective planning is based on a harmonized and well-arranged coping approach. The problems here are the heterogeneous interests of CI providers and

http://dx.doi.org/10.1016/j.techfore.2016.11.027 0040-1625/© 2016 Published by Elsevier Inc.

Please cite this article as: Münzberg, T., et al., A spatial-temporal vulnerability assessment to support the building of community resilience against power outage impacts, Technol. Forecast. Soc. Change (2016), http://dx.doi.org/10.1016/j.techfore.2016.11.027

2

T. Münzberg et al. / Technological Forecasting & Social Change xxx (2016) xxx–xxx

disaster management authorities which results in a missing comparability of impacts. This makes it difficult to generate a common understanding of the potential local power outage consequences for a comprehensive decision base that is acceptable by all stakeholders. ­ The vast majority of power outages last for a maximum of several hours. Already in such short disruption periods the consequences can be severe in particular with regard to the provision of CI services. The underlying danger is within interdependencies between the CIs which allow propagation of failures into other CI systems (domino ad cascading effects). Therefore, it is essential to understand the initial impacts on CIs, to identify which and when CIs become vulnerable, which amount of CI service lost has to be avoided and how important the individual CIs are in comparison to other CIs before the overall crisis situation escalates. The vulnerability paths have to be identified and, as far as possible and appropriate, be managed before uncontrollable cascading effects arise. ­ The decisive point of leverage for building community resilience is the management of the initial and direct effects on CIs, although resilience comprises of preparing, adsorbing, and recovering from adverse consequences of a power outage. In addition, the initial and direct effects imply immediate sufferings that require instantaneous attention. The management of these impacts provide the most prompt effective and feasible basis to build resilience. To facilitate the enhancement of community resilience, we propose a spatial-temporal vulnerability assessment. The vulnerability assessment should allow an improved disaster risk governance by providing a better understanding of the initial impacts of a power outage taking into account the individual characteristics of CIs. The assessment aims at estimating the reduction or loss of CI services in a certain city or county under the impacts of a power outage. To this end, the assessment allows for a systematic and structured evaluation of the exposure of people and CIs, their coping capacities, and their criticality in discrete time steps. On this basis, time-depending effects can be assessed, vulnerability paths can be identified, and determined forecasting is possible. The assessment results should foster investments for mitigating and minimizing disaster risks triggered by power outages and strengthen the collaboration between CI providers and disaster management authorities. The assessment is based on a multi-indicator approach. A Delphi survey and decision maker evaluations are conducted to define weights and parameter values. The spatial-temporal vulnerabilities of districts and CIs are determined by aggregation taking into account the characteristics of the affected CIs. Specific attributes of a present or potential power outage scenario can be considered. This paper is structured as follows: The abovementioned challenges for enhancing community resilience in the context of power outages are discussed in the second section. In the third section we briefly present the current state of the research in simulating power outage impacts. In the fourth section, we introduce the development of the spatial-temporal vulnerability assessment. In the fifth section, the introduced assessment is applied to an exemplary case. The use case and its results are described and discussed. Furthermore, a sensitivity analysis of varying weighting values is conducted and the outcome of different coping strategies are calculated to demonstrate the benefits of the assessment. The paper closes with critical remarks and a conclusion in the last section.

2. Challenges in enhancing community resilience The core challenges for enhancing community resilience are already enumerated in Section 1. However, it is necessary to provide a comprehensive overview about the underlying problems, decision makers, and decision situations. To do so, the disaster management planning and the

collaboration between CI providers and disaster management authorities are discussed in detail from a practical perspective in this section. 2.1. Disaster management of power outage impacts As the German disaster management system is organized in a federal manner, each of the more than 400 local disaster management authorities has to manage the impacts of a power outage in the region for which they are in charge. Following the German incident command system (SKK, 1999), a crisis management group for major incidents is established in each city or county to manage any kind of disaster. Each crisis management group consists of the administrative crisis management team (“Verwaltungsstab”) and an operational crisis management team (“Führungsstab”). The group members come from the local fire brigade, the local administrative departments, and the disaster management authority. There are also special advisers (“Fachberater”) who represent aid organizations and CI providers, for instance. In the remainder of this paper, we always understand the members of the crisis management group as decision makers and end users of the assessment results. According to the planning for power outages, the decision makers' focus lies on the upcoming situation of a disaster (Ryan, 2013). In particular CIs services should be kept continuous to the greatest possible extent. The longer a disruption lasts and the more important and vulnerable the exposed CIs are, the more probable are life-threatening risks. The general life-threatening impacts are well-known and many publications cover the causal relation (see, e.g., Hiete et al., 2010; Petermann et al., 2010; Klinger et al., 2014; Berariu et al., 2015; Castillo, 2014; Barata et al., 2005; Nakayama et al., 2014; Wong et al., 2007; Zubin et al., 2007). However, academic research which establishes the link between electricity supply and resilience is still seen as insufficient (Kinn and Abbott, 2014). The information about potential escalations is often vague, generic, and does not take into account the unique specifications of a certain city or county. The requested decisive information for disaster planning is about which and when a CI becomes vulnerable and how important an individual CI is in comparison to other CIs during a power outage. The current foresight capacity of the disaster management authorities is often not sufficient to predict the concrete spatial-temporal impacts in the way required for an effective disaster risk management and contingency planning. The spatial-temporal impacts in turn depend on the individual characteristics of a CI. The CIs have different Coping Capacity Resources (CCR) whereby the ability to absorb the adverse effects and to maintain a vital supply for the population during a power outage varies. In addition, the CIs differ in its importance for the population. The importance depends on the CI's services and products, the size of the CI, and the daytime of the power outage. As a consequence, the CIs do not suffer in the same way and at the same time. To ensure a comparability, the specific character of each CI has to be put into relation with the characteristics of other CIs. The knowledge about how a power outage escalates and the comparability of exposed CIs can enhance the resilience in different ways. In general, it is not possible to determine the point in time at which the situation suddenly becomes life-threatening. Being aware of this tipping point is very important in disaster planning, because until this point the affected CI provider and population may rely on their own self-helping capacity without any additional assistance from emergency and disaster management units. Furthermore, it is essential to have spatialtemporal information about the impacts to ensure an appropriate and proper intervention (Ryan, 2013; Quarantelli, 1997). By this knowledge, weak and vulnerable CIs can be identified and the corresponding CI provider can be motivated to invest more in CCRs. During a power outage, the disaster management authorities can concentrate their activities to the CIs which have higher vulnerability values or are more important for the population as other CIs. During a present power outage, knowing the tipping point enables warnings and further preparations which may lead to a better resistance. Sometimes the disaster management

Please cite this article as: Münzberg, T., et al., A spatial-temporal vulnerability assessment to support the building of community resilience against power outage impacts, Technol. Forecast. Soc. Change (2016), http://dx.doi.org/10.1016/j.techfore.2016.11.027

T. Münzberg et al. / Technological Forecasting & Social Change xxx (2016) xxx–xxx

authorities and energy utilities can coordinate a small amount of mobile emergency power units. By using spatial-temporal information, these units can be distributed to the CIs or districts in an adequate and reasonable way. Spatial-temporal information is also requested in those cases, in which CIs or regions need to be prioritized. This applies exemplary during load reduction situations (load shedding) when districts have to be disconnected from the supply to stabilize the electricity grid (Münzberg et al., 2013). After a power outage, the districts are restored step by step. In such situations, a ranking of districts could help speed up the recovery of the most important CIs, minimizing the outage duration of vulnerable CIs and making them available to the population more quickly (prioritized recovery). This helps to ensure a recovery in an effective manner. 2.2. Collaboration between CI providers and disaster management authorities

3

information about missing common pictures in collaborative foresight see e.g. Weber et al., 2015, about the importance of a common understanding see e.g. Laakso and Palomäki, 2013). However, a common understanding is necessary to ensure a harmonized and well-arranged disaster management planning. This can only be achieved by an early involvement of the actors (ibid.) and by a joint collaboration in particular between CI providers and the responsible disaster management authorities (for collaborations in disaster management see e.g. Turoff et al., 2004). This is also important regarding the data about the CI's characteristics. This data can only be collected in a trustful collaboration between the CI provider and the authorities. To ensure a successful implementation of the assessment, it is important to keep the effort for the data collection in an appropriate and reasonable relation to the gain resulting from the assessment results. A too extensive data collection would inhibit practical implementation of the assessment. 3. Simulation and assessments of power outages impacts

The disaster management authorities aim at a continuous supply of CI services during a power outage. In Germany, CIs are legally defined as organizations or institutions which provide vital services and products to the population (see, e.g., BMI, 2011). There is a commonly used but not legally defined list of CI sectors and branches (ibid.). Although the CI sector definition is an arrangement between the Federation (“Bund”) and the federal states (“Länder”), it is characterized by the high-level management of the German federal authority for civil protection that operationally is not in charge of disaster management on local level. The definition is in accordance with the EU Council Directive 2008/114/EC on a common approach for the identification and designation of European Critical Infrastructures (ECI). From this perspective, CIs are mainly understood as large-scale/wide-area CIs and supra-regional networks of national relevance (e.g. electricity transmission grids, shipping traffic, medical supply, etc.). However, in the majority cities and counties only some components of large-scale CIs are located. This might be one reason why the local disaster management authorities are rather focused on local CIs like hospitals, dialysis clinics, and pharmacies. In Germany, local CIs have not yet been defined precisely, but represent basic services in the understanding of the UNISDR (UNISDR, 2015). They can be found in every city and county in varying number and size. They are often irregularly distributed in a city and county. A comprehensive and well-defined list of concrete facilities of local CIs is still missing in Germany. Many local disaster management authorities compile so called local CI Cadaster/Land Register (“KRITIS-Kataster”) with data about the local CIs (e.g. contact person, location, size, hazardous properties, emergency back-up power capability, fuel tank capacity, emergency power infeed capability). We also learned from a number of discussions with representatives of disaster management authorities that there might be additional important facilities and assets which do not fully fit in with the CI definition (i.e. with the definition of both large-scale and local CIs), but are very important to disaster risk management. These are special facilities that often have an inherent increased hazard potential. These are, for instance, crematories (infection risks), large venues (used for evacuations), or plants working with hazardous materials (CBRN risks). Generally and from the practical and legal perspectives, only the local disaster management authorities can decide whether components of large-scale CIs, local CIs or other facilities should be considered in the vulnerability assessment. In this paper, we focus on local CIs and provide preliminary suggestions for a list of local CIs in the further development of the vulnerability assessment. Regarding the disaster preparation, each CI provider and citizen has its specific interests and preferences. Each is responsible for its own disaster preparation. This sometimes results in variously interpretations of the potential consequences due to the missing comparability of impacts. It is difficult to generate a holistic, common, and comprehensive understanding of the potential power outage consequences (for more

Although research activities on simulation and assessing power outage impacts have significantly increased in the last decades, there is still a need for tools for disaster management (see e.g. Pescaroli and Alexander, 2016). In this section, a brief overview on the current state of research should be provided. This includes discussion views into CI interdependencies, cascading effects, the theory of value of lost load, and spatial-temporal vulnerability assessments. 3.1. CI interdependencies and cascading effects In the last decade, many models and simulations were developed to obtain a clearer understanding of the role of interdependencies in network systems and among large-scale CIs. This is motivated by the concerns about cascading effects which are based on the toppling domino theory. The toppling domino theory suggest that an initiating event like a CI disruption sets a sequence of disruption on other CIs (Luiijf et al., 2009; Van Eeten et al., 2011; Kadri et al., 2014; Pescaroli and Alexander, 2016). In this way, a failure in one CI can propagate through multiple escalation levels so primary, secondary, and tertiary cascading effects can occur (Kadri et al., 2014). To describe the cascading interconnections, some authors additionally distinguishes between cascade initiating and cascade resulting events (e.g. Luiijf et al., 2009). Many CI interdependency models and simulations can be found in literature that address the dependence understanding of Rinaldi et al. (2001) who distinguishes between (i) physical dependencies, (ii) cyber dependencies, (iii) geographical dependencies, and (iv) logical dependencies. The models and simulations were developed in several projects, such as DIESIS (Rome et al., 2009; Usov et al., 2010), I2Sim (Marti et al., 2008), IRRIIS (Klein et al., 2008), CIPS/DSS (Bush et al., 2005), or CARVER2. The predominated used modelling methods are agent-based modelling (e.g. Casalicchio et al., 2008; Casalicchio et al., 2010; Bagheri et al., 2007), system dynamics (e.g. Min et al., 2007; Cavallini et al., 2014), Bayesian (e.g. Di Giorgio and Liberati, 2012; Jha and Keel, 2012), and input-output modelling (e.g. Haimes et al., 2005; Oliva et al., 2011; Setola et al., 2013). Reviews of the projects, more insights, and the classification of CI interdependency methods, models and simulations are provided by e.g. Ouyang (2014), Pederson et al. (2006), Yusta et al. (2011), Theoharidou et al. (2011) and Eusgeld et al. (2008), Giannopoulos et al. (2012). In the context of disaster management, such models and simulations are still faced with a deep-seated and substantive reservations by the decision makers. The causes of these reservations can be find in the diverging purposes of the models, the necessary extensive data collection, the time consuming modelling extent, the difficult interpretability of complex model output which is often not possible without the modeler, the often too abstract and unspecific results which frequently do not provide enough reasoning and insights, and the fact that CI disruptions rarely cascade deeply in practice.

Please cite this article as: Münzberg, T., et al., A spatial-temporal vulnerability assessment to support the building of community resilience against power outage impacts, Technol. Forecast. Soc. Change (2016), http://dx.doi.org/10.1016/j.techfore.2016.11.027

4

T. Münzberg et al. / Technological Forecasting & Social Change xxx (2016) xxx–xxx

The purpose of the majority of CI interdependency models and simulations is to enhance network resilience and to mitigate or reduce the probability of interruptions in a particular CI, one or multiple CI sectors or a network of CIs. To use these models and simulations, a large amount of heterogeneous data and assumptions is required (for deeper discussion, see e.g. Kozik et al., 2015). The collection of the necessary data is often time-consuming and then sometimes still incomplete. In some cases, historical and empirical data are used in combination with expert surveys to model CI interdependencies (Stergiopoulos et al., 2016). In practice, however the CI providers and scientists have no or only limited access to detailed data on interdependencies. The actual challenge is to provide decision support approaches based on a limited amount of data that are easy to collect without vast experience by the disaster management authorities. Even though CI interdependency models and simulations are usually very complex, their results have a high level of abstraction that focuses on a general CI protection rather than a concrete design of an effective disaster management. With all remarkable efforts of the available models and simulations, their findings are often too general and vague for the purposes of disaster management. They frequently address a too high management level and, hence, do not sufficiently consider individual structures of a certain region. Often, high-level/wide-area CIs are addressed instead of local CIs. This makes it extreme difficult to understand the role of particular facilities or to derive and identify appropriate coping measures. There is still no single “silver bullet” approach (Kröger and Nan, 2014; Kröger, 2008) and still a need for appropriate analysis tools for the purposes of disaster management (Pescaroli and Alexander, 2016). Some results from empirical studies about cascading effects are remarkable because their findings are in contrast to the toppling domino theory. Luiijf et al. (2009) and Van Eeten et al. (2011) found that cascading events occur more frequently as expected. Interestingly, the overwhelming majority of cascading effects are caused by disruptions in the energy and telecom sectors. Only very exceedingly isolated cascading events were caused due to other CIs. This is not surprising as many theoretical analysis (e.g. Laugé et al., 2015; Stergiopoulos et al., 2016; Buldyrev et al., 2010; Setola et al., 2009) and other empirical analysis (e.g. Kunz et al., 2013; Luiijf et al., 2009; Van Eeten et al., 2011; Blake et al., 2013) have shown that the most of the CI services rely on electricity and on the information and communication infrastructures. Through this, both events have high potentials for becoming a cascade initiating event. The empirical studies have also shown that domino events do rarely cascade deeply (Luiijf et al. (2009) and Van Eeten et al., 2011). In 2009, for instance, Luiijf et al. analyzed more than 1700 CI disruptions and discovered that only 24% of the disruptions caused resulting cascading effects in other CI services. Four per cent of the total CI disruptions escalated in secondary events and only 0.22% in tertiary cascade events. Although it is not clear whether coping measures were effective in other cascading events or other cascading events were simply not be reported, the findings indicate that CI disruption do not frequently cascade deeply. This result is in sharp contrast with the predominated domino theory and also point out that cascading effects should not be overstated or exaggerated. Whether a power outage become a cascade initiating event or not is mainly depending on the outage duration and the characteristics of the exposed CIs. For an effective disaster management, the individual structure and characteristics of specific local CIs have to be taken into account. Weak points in a certain region have to be identified, counter measures compared, and stakeholder collaboration fostered. Related to the complex challenge of modelling the functionalities of CIs, Pescaroli and Alexander (2016) recommend to rather analyze potential escalation points and the sensitive nodes that may generate secondary resulting events. They suggest to “shift attention from risk scenarios based on hazard to vulnerability scenarios” based on paths for cascading events. Such a shift would stress the role of identification of tipping points of disruptions and the amplification of crises. Given the findings

of the empirical studies of cascading effects, this means in particular to analyze the directional paths for secondary cascading events that may are initiated by a power outage. The proposed assessment is addressing this with an emphasis on electricity dependencies and on the potential cascading resulting events of a power outage. 3.2. The value of lost load in disaster management The most popular approach for assessing the consequences of power outages is the Value of Lost Load (VoLL) (for a general overview, see Billinton, 2001; Billinton et al., 1993; Ajodhia et al., 2002). The VoLL is an estimation of the customer's willingness to pay for avoiding a power outage, which implies the monetary value of electricity lost for a concrete hour and customer. Some studies particularly focus on the conditions in Germany (see, e.g., Praktiknjo, 2016; Piaszeck et al., 2013; Growitsch et al., 2013). Although VoLL analysis provides information about power outage consequences, there are multiple reasons why the use of VoLL analysis results can be recommended for disaster management with certain limitations only. So far, studies to estimate the VoLL have addressed industrial and economic sectors, households, and the purposes of the energy and electricity sector exclusively. It was not possible to find a study of VoLL of CIs and CI sectors. This is probably due to the results of a VoLL analysis that reaches its limits when considering non-monetary, immaterial, and social implications of the loss of (CI) services. Another reason why VoLL analyses are still infeasible in disaster management is that the VoLL analyses are focused on the value of industrial sectors or households in certain cities (“kreisfreie Stadt”) and counties (“Landkreis”). The results are usually regionalized in a macroscopic way. For an effective disaster management, a higher resolution is requested. But the necessary data on the municipal level are often lacking. In addition, the VoLL can estimate the consequences of short-term losses of some hours to a very limited extent. Consequently, some of the methods (see, e.g., Leahy and Tol, 2011; Growitsch et al., 2013; de Nooij et al., 2007, de Nooij et al., 2009) use electricity (use) or load profiles of consumers to estimate an hourly VoLL. In this way, outage costs can be computed as a function of time. The resulting hourly VoLL is used, for instance, in reducing the monetary consequences of load reduction processes (see, e.g., Khujadze, 2014; Praktiknjo, 2016; Conejo et al., 2010). However, only the extension by load profiles is not enough for taking into account further dynamic changes triggered by power outages. This makes the VoLL still inappropriate in particular for estimations which address the impacts of a disruption lasting several hours. 3.3. Spatial-temporal vulnerability assessments Vulnerability assessments are widely used in disaster management to estimate harms (Birkmann and Wisner, 2006) and to identify weaknesses in the preparation stage (Bogardi, 2004). Vulnerability is generally recognized as an unobservable phenomenon (Moss et al., 2001; Patt et al., 2009). Therefore, different theoretical concepts were developed which use varying definitions for the term “vulnerability” (for more insights on the definitions of vulnerability see also Thywissen, 2006; Manyena, 2006; Hufschmidt, 2011). Generally, vulnerability is understood as an inherent character of a system to suffer from adverse effects under the impact of hazards (e.g. see Thywissen, 2006; Cardona, 1999; Birkmann and Wisner, 2006). Depending on the theory concept and the specific context, discipline, hazard, and assessment purpose, the vulnerability of a system is often derived from dimensions of e.g. exposure, threats, element at risks, susceptibility, criticality, and/or CCRs (see, e.g. Birkmann, 2006; Villagrán de León, 2006). These are frequently measured indirectly by indicators and indicator frameworks (see e.g. Kasperson et al., 1995; Brooks, 2003; Adger, 2006). The measurability is often difficult because of lacking or uncertain data. Data can be obtained empirically, by simulation or modelling or from expert estimations. To deal with lacking or uncertain data, Fuzzy

Please cite this article as: Münzberg, T., et al., A spatial-temporal vulnerability assessment to support the building of community resilience against power outage impacts, Technol. Forecast. Soc. Change (2016), http://dx.doi.org/10.1016/j.techfore.2016.11.027

T. Münzberg et al. / Technological Forecasting & Social Change xxx (2016) xxx–xxx

Logic (e.g. Claudio et al., 2011; Akgun et al., 2010), Monte-Carlo simulation (e.g. Perdikaris et al., 2011), probabilistic approaches (e.g. Blauhut et al., 2015) or Delphi surveys (e.g. Bañuls et al., 2013; Lee et al., 2013) are frequently used. The indicator values are aggregated to vulnerability indices by simple sums, weighted sums of normalized indicator values or multiplication of the indicator values. From a methodological point of view, this is approach is similar to multi-attributive value theory. The selection of the indicators and their aggregation depends on the assessment purpose and have to be chosen with care. Recently, there is a growing number of vulnerability assessments which allow an assessing of spatial-temporal impacts of hazards. This makes it possible to assess long-term changes such as from the impacts of climate change (see, e.g., Aubrecht et al., 2013; Gaillard, 2010) or of sea level rises (e.g. Sahin and Mohamed, 2010; Sobiech, 2013) but also of short-term or time specific vulnerability changes such as the time-depending earthquake risks (e.g. Debnath, 2013) or flood risks (e.g. Rodríguez-Gaviria and Botero-Fernández, 2013). 4. Developing a spatial-temporal vulnerability assessment In this section we introduce and discuss the development of the spatial-temporal vulnerability assessment. We firstly give a brief overview about the general theory and the components of the assessment. The components will then be discussed thereafter in detail. 4.1. The basic theory of the spatial-temporal vulnerability assessment As partly mentioned before, we understand the spatial-temporal vulnerability as a measure of the potential sufferings of a population in a certain city or county under the effects of a reduction or loss of CI services caused by a power outage. The assessment aims at estimating the vulnerability of districts of a certain city or county by focusing on the exposure of CIs which are located in the districts. In a city or county under consideration, there are l CIs, where each CI has an individual number i ∈ Infrastructures = {1,2, … , l}. For the further assessment, it is important to categorize and assign CIs into different CI types. As discussed before, there is no comprehensive and welldefined list of concrete facilities of local CIs in Germany. However, CI types can be found as attributes in CI Cadasters. Their determination is derived from the German laws for disaster management, civil protection, and emergency legislation. Fig. 3 displays a list of types for local CIs that can be find in nearly every city or county. This list is non-exhaustive but provides an adequate basis for a discussion for decision makers about what CI types should be considered in the assessment. Some of the CI types may be excluded. As already discussed, there are sometimes cases of facilities that are very specific for a city or county and from great practical interest although an assignment into the given CI categories is not possible. Such facilities have, for instance, an inherent increased hazard potential. The assessment is flexible to create new CI types and, hence, allow an integration of such special facilities. Finally, the decision on which CI types should be taken into account is up to the decision makers. The selected CIs that should be considered in the assessment are assigned to the respective CI types CI_type u where u ∈ Types = {1, 2, … , h} and h number of CI types taken into account. All hospitals, for instance, are assigned to the CI type “hospitals”, all pharmacies to the CI type “pharmacies”. It is not always sufficient to consider all CI types in a single assessment. To enhance sector specific insights or to focus on a particular aspects, it can be useful to conduct multiple assessment runs. The CIs are usually irregularly distributed in a city or county. This causes different impacts depending on which geographical regions are affected by a power outage and which local CIs are located in the affected regions. The considered city or county consists of multiple districtr,

5

where r ∈ Districts = {1, 2, … , g} and g number of districts in a city or county. Fig. 1 shows the general assessment's aggregation framework. Each CIi , r , u is defined according to their individual number i, to their assignment to the district r where it is located, and to their respective CI type u. For each CIi , r , u a CI Vulnerability Index is calculated using a measurement for the CI's criticality and coping capacity. The criticality of a CI is measured by the Relevance Criticality Weight, the Size Criticality Indicator, and the Time Dependent Criticality Indicator. A Coping Capacity Depletion Indicator is used to take into account the CI's coping capacity. The Relevance Criticality Weight and Size Criticality Indicator are static values and do not change over time. The Time Dependent Criticality Indicator and the Coping Capacity Depletion Indicator are time dependent and change during an ongoing power outage. A detailed explanation and derivation of the CI vulnerability measuring will be provided in the upcoming sections. Each CI has an individual role in the entire CI system of a city or county. Depending on the CI's criticality, a missing service from a certain CI can have a severe impact in the entire city or county. To reflect this, the vulnerability measurement assesses the vulnerability of one CI in relation to the characteristics of all other CIs in the city or county. The relation addresses the relevance of the CI type (measured by the Relevance Criticality Weight) and the size of the certain CI in relation to other CIs from the same CI type (measured by the Size Criticality Indicator). Through this, the amount of a CI's vulnerability index also expresses the criticality of the CI for the whole city or county. Temporalspatial analyses are realized by the calculation of district vulnerability indexes. This index is an aggregation of vulnerability indexes of those CIs that are located in the district under consideration. This aggregation allows temporal analysis based on the calculation of vulnerability values for different point in times during a power outage. For analyzing spatial effects, the district's vulnerability index displays the consequences of a power outage for the entire city which are caused by the missing or reduced services of the CIs located in the respective district. Higher districts vulnerability values mean that the CIs suffering from the effects of a power outage in the certain districts have a high criticality to the entire CI system. The assessment can be used before and during a power outage. For both application contexts, a power outage scenario has to be defined. A power outage can affect different areas, vary in its duration, and happen at any type of day (weekend, working day, etc.) and daytime. The effects of a present power outage can be analyzed by using the scenario attributes of its starting time. In the disaster preparation and mitigation phase, a reasonable reference power outage scenario is requested. The definition of this reference scenario is discussed in the upcoming section. The general structure and approach of the assessment is displayed in Fig. 2. The figure shows the assessment processes, how the decision makers are generally involved, and the origin of necessary data. Adequate preparations are necessary before the vulnerability assessment can be conducted. Depending on whether the assessment should facilitate preparedness/mitigation or response/recovery activities, the attributes of a reference power outage scenario or attributed of a present power outage (event scenario) have to be determined by the decision makers. This also includes the selection of CI types that should be analyzed in the assessment. Thus, the boundary conditions for conducting the assessment are defined by the decision makers. A minimum of data about the characteristics of CIs is necessary. The decision makers have to provide attributes about the CI's locations, the sizes, and the coping capacities. Usually, these data is provided by the local CI Cadasters which substantially reduces the efforts for data collection. The assessment is based on two estimations of decision makers. First is the evaluation of coping capacity depletion and the second is the evaluation of CI type relevancies. The evaluations are integrated in the

Please cite this article as: Münzberg, T., et al., A spatial-temporal vulnerability assessment to support the building of community resilience against power outage impacts, Technol. Forecast. Soc. Change (2016), http://dx.doi.org/10.1016/j.techfore.2016.11.027

6

T. Münzberg et al. / Technological Forecasting & Social Change xxx (2016) xxx–xxx

Fig. 1. The general assessment's aggregation framework of indicators and weights to calculate CI and district vulnerability indexes.

assessment by the Relevance Criticality Weight and the Coping Capacity Depletion Indicator. The results of the assessment are aggregated outputs. Usually, the first assessment run considers a baseline scenario in which no additional coping measures are taken into account. In further assessment runs, coping strategies could be applied. Multiple strategies can be considered by the decision makers. The benefit of the strategies can be measured and compared by the updated aggregated output. We believe that the general structure of the assessment is comparatively simple. The low efforts for data collection, the strong concentration on four parameters, and the integration of the decision maker's estimations allow rapid implementation and a well understanding of the results. However, the conduct itself and the evaluation request detailed methodological knowledge and skills. In particular exploration and integration of the decision maker's estimations are demanding and can lead to an overextension of the decision makers. It can be suitable that the decision makers are consulted by an analyst and moderator who leads them through the whole assessment process. However, the provision of the assessment results alone will not reach a higher level of resilience. Based on the results, it is only possible to identify strengths, weaknesses, and lacks of preparations. However, such results motivates for investing in CCRs, facilitate the disaster planning processes, and can lead to appropriate activities to enhance the resilience of CIs.

4.2. Definition of a reference power outage scenario Although the simulation provides valuable insights, the accurate prediction of power outage consequences is timely limited to the initial period. The reason for this is that power outages are, like any other disaster, exceptions to the rules (Turoff et al., 2004). However, the proposed approach is pre-deterministic and rule-based which makes it impossible to predict unforeseen developments which will particular arise the longer the outage lasts. The longer a power outage lasts, the more unpredictable and exceptional is the situation and the less it is possible to make reliable forecasts. Thus, only initial impacts are sufficiently predictable. To effectively disaster planning, it is necessary to define a reference scenario of a power outage which is relevant, reasonable, and still allows for an assessment based on predeterminations. Past power outages demonstrated that a disruption of electricity supply can happen any time, expose areas of various sizes, and last from seconds to days (see, e.g. Anagnostatos et al., 2013; UTCE, 2007; der Vleuten and Lagendijk, 2010; Howell, 2012). According to the size of the area that is exposed to a power outage, it is appropriate to consider the worst case in which the entire city or county is affected in the reference scenario. The definition of the duration is a balance between a too short and a too long outage. From a practical point of view, the duration must be suitable to address the initial impacts of a power outage.

Fig. 2. Visualization of the vulnerability assessment structure.

Please cite this article as: Münzberg, T., et al., A spatial-temporal vulnerability assessment to support the building of community resilience against power outage impacts, Technol. Forecast. Soc. Change (2016), http://dx.doi.org/10.1016/j.techfore.2016.11.027

T. Münzberg et al. / Technological Forecasting & Social Change xxx (2016) xxx–xxx

However, a definition of a too long outage duration (e.g. several days and weeks) may result in an unpredictable situation development. Based on this considerations, we propose a reference scenario as an outage which affects the entire city or county with an outage duration of one and a half day (or, in other words, 36 h corresponding to 2160 min). However, decision makers are enabled to define other reference scenarios. Beside the reference scenario, the decision makers may define other scenarios which may are also from interest. For an enhanced understanding it may also be interesting to vary the starting time, the duration, and/or the districts which are affected by a power outage, for instance.

4.3. Criticality of critical infrastructures The criticality of a CI can be measured through several dimensions. We limited the dimensions to the smallest reasonable number of three dimensions. The three dimensions still allow profound insights at simultaneous reduced expenses for modelling and data collection. This comparatively simplifies the vulnerability assessment and ensures a better interpretation and comprehension of the results. The first criticality dimension addresses the assignment of a CI to a specific CI type. The CI types differ in their relevance for providing vital products and services to the population. As an example, the disruption of a hospital is more severe for the population than the disruption of a GP or a pharmacy. This dimension is considered by the Relevance Criticality Weight. As this dimension refers to different types of CIs, we also call this property “inter-criticality”. The second criticality dimension is related to the size of a CI. CIs that are larger than others of the same type supply products to a higher amount of customers and are therefore more critical. Hence, a disruption of a larger CI have more fatal impacts than a disruption of a smaller infrastructure. This dimension is considered by the Size Criticality Indicator and also understood as “intra-criticality” of CIs of the same type. The third criticality dimension refers to the daytime at which a CI is exposed to a power outage. Some CIs are operated only during the day or the population's demand for the CIs fluctuates over daytime. To consider this, we introduce the Time Dependent Criticality Indicator. The following sections will describe how all of these dimensions are operated.

7

4.3.1. Relevance Criticality Weight As already mentioned before, the CIs can be assigned to different CI types such as hospitals, GPs, pharmacies, etc. The CI types differ in their relevance for providing vital products and services to the population. In the multi-attributive value theory, this is expressed by weights (for more information see e.g. Belton and Stewart (2002)). There are multiple ways to determine values for the weights (for a review see e.g. Zardari et al., 2015; Riabacke et al., 2012). In our case, the determination should be based on the estimations of decision makers. Due to there are multiple decision makers involved, an appropriate group weighting method has to be selected carefully. Because of the high number of CI types, the little experience of the decision makers in using weighting methods, and the high number of decision makers involved we selected the direct weighting technique as one of the most practical weighting method in this Delphi evaluation. In this paper, we used a modified Delphi evaluation to rate the CI type relevancies (for general Delphi evaluation see, e.g., Dalkey and Helmer (1963), Linstone and Turoff (1975)). Traditionally, a Delphi evaluation aims at aggregating and discussing evaluations of the same decision makers in multiple rounds until a consensus is reached. This origin approach is modified to ensure a higher level of practical orientation, better acceptance of the assessment results by decision makers, and greater legal certainty when exercising an administrative discretion. The modified structure is displayed in Fig. 3 and consists of minimum two rounds. In each round a panel of decision makers evaluate weights for the CI types. Deviating from the traditional approach, the panels of decision makers in the first and the second round are not the same. In the first round, we asked a couple of decision makers from different cities and for estimating the relevancies of CI types. The decision makers are coming from different cities or counties and each is in charge for managing power outage impacts in their area of responsibility. We are asked them separately and anonymously to evaluate each CI type of the set of CI types with respect to its importance for supplying the population with vital services and products. As the decision makers' judgements are independent of a concrete city or county, the intermediate summary of the first round provides a general statement without taking into account local specifications. Then, the intermediate summary of the first round is used in a second round to support those decision makers who are in

Fig. 3. Structure of the modified Delphi evaluation for determining the CI type relevancies.

Please cite this article as: Münzberg, T., et al., A spatial-temporal vulnerability assessment to support the building of community resilience against power outage impacts, Technol. Forecast. Soc. Change (2016), http://dx.doi.org/10.1016/j.techfore.2016.11.027

8

T. Münzberg et al. / Technological Forecasting & Social Change xxx (2016) xxx–xxx

charge of disaster management of the city or county under consideration. Like in practice, these decision makers have to discuss the weights in face-to-face meetings and have to find a consensus for the value they determine for each CI type. In this second round, they are allowed to use the intermediate summary or to revise it taking into account their knowledge of the concrete CIs, their experience, preferences, and knowledge about local disaster risks. This revision allows for a change of the previous weights as well as for the inclusion or exclusion of CI types. The direct weighting technique is used to determine the weights. The direct weighting is based on importance weighting scores wscore u which represent the importance of each CI type. The score varies between 0 for no importance and 100 for high importance. The considered set of CI type and the role of CI importance were discussed with the decision makers in workshops. Using a standardized questionnaire, each decision maker of the first round was asked to rate each CI type without considering other aspects, such as the potential CI size, its CCRs or the time of exposure. We used two workshops to launch the first round of the Delphi evaluation. In 2014, a first workshop took place which was accessible to any representative who were responsible for disaster management in a German city or county. The workshop had the main purpose to verify the list of relevant CI types. Eleven decision makers participated. In a second workshop, all disaster management authorities of the 44 cities and counties in the federal state of Baden-Wuerttemberg were invited to send representatives who are in charge of preparation for and coping with the effects of power outages in their area or responsibility. 31 decision makers participated in the second workshop. From both workshops, 32 responses of the participants were analyzed. The intermediate summary is displayed in Fig. 4. Fig. 4 shows the intermediate results from the first round which expresses the expert estimations of the pre-defined set of CI types. It displays how the responses for each CI types are distributed over a smaller or larger. The box plots illustrates how the majority of decision makers evaluate the CI types. It represents a general statement about the importance of CI types from a larger number of decision makers. The result was neither discussed with the decision makers nor changed in the first round and, consequently, the result has wide empirical variance. The empirical variance indicates common or different understandings of a CI type's importance. In the second round, this uncertainty is an

important decision support for those decision makers who are in charge of a city or county and analyze the results in the second round of the Delphi method. The decision makers of the second round can then discuss and analyze the intermediate summary taking account the individual structure and characteristics of the CIs in their city or county. They can relate their decision on relevance rates to the certain representative results of the first round or revise the previous evaluation, if objective reasons exist that make a change necessary. Compared to the practice of today in which the decision makers do not use a systematic and structure approach to rate CIs, the Delphi approach allow a determination of weights which is less arbitrary or random and, hence, can also better withstand critical judicial reviews. In the second round of the Delphi estimation, the decision makers from a certain city or district are also free to exclude some of the proposed CI types or to include new ones, where necessary. Ideally, the second round results in a consensus of importance weighting scores wscore for each CI type u. For the further calculation, u it is necessary to fulfil the condition that the sum of every weight is 1. This is done by a linear normalization of the importance weighting scores and results using a normalized relevance criticality weight wRelev u with wscore ¼ Xh u wRelev u wscore u¼1 u

ð1Þ

Besides the second round and the further consensus debate with the decision makers in charge, it is also possible to integrate the uncertainty from the intermediate summary into the further calculations. This results in an advanced understanding of the possible variance of the assessment outcome. An exclusion of CI types is still possible. In this case, the determination of the importance weighting factor wRelev deu pends on the combination of the importance weighting scores wscore u which each can be assumed as a statistical distribution. This can be solved by Monte Carlo simulations using a sample statistical distribution (for more information of uncertainty handling using Monte Carlo simulation see, e.g. Bertsch, 2008). Each score is sampled by its statistical distribution based on the intermediate summary. In each simulation a new random combination of importance weighting scores is formed and

Fig. 4. Intermediate summary of the importance weighting scores of CI types in a boxplot diagram based 32 evaluations from representatives of different disaster management authorities in the first round of the Delphi evaluation.

Please cite this article as: Münzberg, T., et al., A spatial-temporal vulnerability assessment to support the building of community resilience against power outage impacts, Technol. Forecast. Soc. Change (2016), http://dx.doi.org/10.1016/j.techfore.2016.11.027

T. Münzberg et al. / Technological Forecasting & Social Change xxx (2016) xxx–xxx

normalized. By conducting multiple simulations, it is possible to generate a normalized relevance criticality weight wRelev which fully capture u the uncertainties in weights. 4.3.2. Size Criticality Indicator Depending on the local structure of the city or county under consideration, there are multiple CIs of the same CI type or only one or no CI of a CI type. If there are multiple CIs of the same type, they may also have different sizes. The number and the sizes of CIs from the same type in a certain district are a dimension of criticality. To consider this, we introduce a CI type-specific Size Criticality Indicator to include this dimensions into the aggregation of the Vulnerability Index Value. The indicator is measured by a CI-specific size attribute ai,r,u. If multiple CIs of a CI type have similar sizes, it is sufficient to use the total number of CIs of the same type in a city or county. This, for instance, applies to pharmacies, because pharmacies in Germany have nearly similar floor spaces and numbers of customers. Consequently, for pharmacies further differentiations are not necessary. In other cases the size of CIs of a CI type vary greatly. The larger the CI, the more important is it for the provision of vital services and products. This, for instance, applies to hospitals, for which a higher degree of differentiation is necessary, because the hospital sizes vary. For each of such kinds of CI type a specific attribute ai , r , u has to be defined. An adequate sufficient size indicator of hospitals is the number of hospital beds. This is similar for dialysis clinics or nursing beds. Table 1 displays a proposal for some selected CI-specific attributes for some CI types which may need a size differentiation, if multiple CIs of the same CI type are located in the considered city or county. The CI-specific Size Criticality Indicator Si , r , u is calculated by a linear normalization taking into account all attributes values of all CIs of the same CI type u with u ∈ V = {1, 2, … , h} ai;r;u Xg

Si;r;u ¼ Xl i¼1

a r¼1 i;r;u

∀u ∈ Types

ð2Þ

The attribute values are obtained from statistical, census or descriptive data provided by the local statistical offices, local disaster management authorities, or own surveys. In the short-term perspective of a few weeks, these values will not change and are highly precise. Hence, it is reasonable to ignore the uncertainty of these data. However, the data display a situation at a discrete point in time which can change during the years. To ensure the validity of the model in the future, the attribute values have to be updated periodically. 4.3.3. Time dependent criticality indicator Electricity is consumed by customers in various and individual intensities throughout a day. Regularly, the variation of customer-specific consumption during a day is expressed by intraday load course of electrical power in a resolution of 15 min. Intraday load characteristics represent typical loads over time for the processes that are executed during a day at a CI facility. The processes correspond to business hours (like

Table 1 CI type and CI-specific attributes ai,r,u. CI type u

Name of CI type

CI type-specific attributes ai, r,u

1 4 5 6 7 14 15 16 17 21

Hospitals Shelters for refugees and homeless Dialysis clinics Nursing homes Assisted living Ambulance stations Fire stations Police stations Correctional facilities Drinking water facilities

Number of hospital beds Number of beds Number of dialysis beds Number of care beds Number of beds Number of staff per shift Number of staff per shift Number of staff per shift Number of prisoners Liters per day

9

those of pharmacies), periodic shift operations (like those in hospitals) or the human circadian rhythm (like the day-night-cycle). There are higher load periods during a day in which more processes are activated and other operations are underway to provide services and products. Accordingly, there also is a smaller demand indicated during those periods of a day in which the load course is relatively low. This implies that a power outage in the period with low values of the load course would lead to minor adverse impacts compared to those that occur in time periods with a higher load level. In this way, the intraday load course represents daytime-specific changes of the demand of vital services and products, the susceptibility to damage, losses, and expected but potentially lacking supply of CI functionalities and, hence, a daytime-depending criticality. The load course represent the electricity consumption for all process of a facility. In a CI facility, these processes can be those for the core services and productions but also for supporting tasks. The processes have different criticalities for keeping the CI services, have different demands on electricity, are conducted at different times at a day (process scheduling), and depend on each other (process interactions). These could be included in a vulnerability assessment by further micro models of single CI facilities. We are convinced, that this is not practical because this approach request many data which collection is time consuming and in some cases probably not sufficient at the moment. It also would lead to higher efforts in CI modelling and generates new uncertainties which are difficult to handle. These are the reasons why it is reasonable and proper to see a CI as one common unit comprises of all its processes. We assume, that a CI can only operate correctly without an undesired reduction or loss of CI services, if all types of processes work and are sufficiently supplied by electricity. In estimating the vulnerability, the integration of intraday load courses allows to consider the level of demand at the time at which a power outage starts. Although the intraday load course represents normal day-to-day conditions, specific demand changes that occur in the aftermath of a power outage are not displayed by load courses. Due to this limitation, indicating the vulnerability through an intraday load course is useful for measuring the short run effects of power disruptions. It can be assumed that in the beginning of an unexpected power outage the affected customers are not able to adjust their demand. In this initial situation, we assume that the impairments corresponds to the load course. This may change after a couple of days of an ongoing power outage. The longer an outage takes, the more significantly does the demand deviate from normal conditions. For considering long-term power outages, the use of load courses for impact estimations is not suitable. There are multiple ways to collect load course data. Sometimes, facility operators track their electricity consumption and generate daily load courses. In these cases, it might be possible to analyze the data and generate representative load courses for working days, Saturdays, and Sundays that are specific of the considered CI. However, this way of data collection is very time–consuming because of the high number of CIs in a city or county. In addition, the success depends on cooperation with the CI provider. Safety concerns and data protection may reduce the willingness to cooperate. Another way to collect load course data is the use of data from forecasting techniques. There are multiple techniques in literature that allow for a precise forecast of load courses, which is very important in planning and operating electric utilities (Fischer, Härtl, and Wille-Haussmann, 2015; Piwko et al. 2005). The techniques use historic data or synthetically generated data. Both techniques use statistic evaluations. In doing so, the customer's behavior is taken into account, which is influenced, for instance, by time, day, season, weather, and random effects. Reviews of forecast techniques and the associated expenditure are provided by e.g. Hippert et al. (2001), Alfares and Nazeeruddin (2002), Taylor and McSharry (2008), Singh et al. (2012). Up to now, the load courses of relatively small systems like households, retail shops and offices, and minor segments of customer clusters have been represented as so-called standard load profiles (SLPs) or as

Please cite this article as: Münzberg, T., et al., A spatial-temporal vulnerability assessment to support the building of community resilience against power outage impacts, Technol. Forecast. Soc. Change (2016), http://dx.doi.org/10.1016/j.techfore.2016.11.027

10

T. Münzberg et al. / Technological Forecasting & Social Change xxx (2016) xxx–xxx

synthetically generated profiles. SLPs are based on evaluations of historical data and are references for a typically reasonable and accurate load course. They are representative of facilities of the same kind. In Germany, SLPs are provided by local distribution grid providers or federal associations like the BDEW (German Association of Energy and Water Industries). SLPs are available for all CI types on the local level. SLPs represent an intraday load course with a time-depending electrical load value YDaySaisonu(tstep) for a CI type u of a particular day type (working day, Saturday, Sunday) and season (summer, winter, spring, autumn). The SLP displays a load in a resolution of 15 min. Therefore, the load value refers to a discrete point in time tstep with step= 1 , … , 96. For the modelling of a daytime-depending CI criticality, a normalized value is necessary, which varies between 0 for no load and no criticality and 100 for maximum load and maximum criticality. We propose a simple normalization function LoadDaySaisonu(tstep) with respect to the maximum load of the intraday load phase maxYDaySaisonu({t1, … , t96}): 

Saison

LoadDay

u

 t step ¼







Y Saison Day u t step maxY Saison Day u ðft 1 ; …; t 96 gÞ

fuel consumption. However, the CCR is regularly designed for constant full-load operation. Often, load banks and protective switches are installed to prevent engines and equipment from being damaged during overload and underload operations. For this reason, it can be assumed that the CCR is consumed constantly, even in times during a power outage in which processes are reduced, rescheduled or skipped. The vulnerability can be expressed by a CCR depletion function Ck(tstep) for a CI k for different time steps tstep. The function implies an evaluation of vulnerability that is derived from the CC depletion and the loss of operational abilities. The function expresses a score between 0 for full CCR or not vulnerable and 1 for fully depleted CCR or highly vulnerable for different CCR consumptions. If CCR is implemented, the score 0 is reached in the beginning of a power outage, because enough CCR is available to absorb the adverse effects. The score 1 is reached at the point in time tCC i . After this point, the CCR is fully exhausted and reaches the score 1: 

C i t step ¼ ð3Þ

Meanwhile, the use of SLPs has been criticized by some researchers, such as Hayn et al. (2014), who point out that the ongoing transformation process of electricity systems and changed consumer behaviors are not adequately taken into account by SLPs. To overcome these weaknesses, bottom-up load models are used with raising popularity (ibid.). Such models aim at providing synthetically generated profiles and take into account multiple parameters like consumer's behaviors, attitudes, lifestyles, statistical data like census data, and the capacity demand of used electric appliances at different times of a day. The increasing trend of using bottom-up load models is also pushed by the availability of new technologies for metering electricity consumption. Those technologies emerged in the past years (Neenan and Hemphill, 2008). The use of bottom-up load models may provide additional opportunities to estimate intraday load course-triggered vulnerability in the future, but for the moment, the number of synthetically generated profiles for CI types is not sufficient. 4.4. Coping capacity resource depletion indicator Some CI providers use emergency power units or batteries to continue their business even in the situation of a power outage. However, the loading capacity of batteries or the volume of fuel tanks for generators is restricted. Continued supply is limited in time, although these CCRs enable the provider to absorb adverse effects of a power outage. Batteries and fuel tanks can be recharged and refilled. However, in the first instance the model should focus on the baseline scenario without any further response measures. The more batteries discharge and the more the filling level of the fuel tanks drops, the more CCR is consumed and the lower is the ability to cope with the adverse effects (communicating with others, refilling CCR, preparing for a full disruption, preparing evacuation, etc.). Because of the depletion of CCR, the vulnerability of the considered CI increases with the duration of a power outage. The vulnerability increase depends on how long the CCR can last, which is expressed by ΔTCCR . At the time i tCCR with tCC = t0 + ΔTCCR in which the CC is fully depleted, the CI fai cilities are at high risk to break down, to suffer damage, and to lose performance in providing vital services and products to the population. Until this point in time, recharging or refueling, the recovery of the primary electricity supply or alternative measures that keep a population supplied are required. The increase in the CI's vulnerability depends on the speed at which CCR is depleted. Whether the CCR ability comes from a battery or a power unit with a fuel tank, the CC can be operated under overload, normal load, and underload conditions. This may lead to different speeds of

8 <

0 if t step −t CCR ¼0 i   t if t step bt CCR Depl i step i : 1 else

ð4Þ

In the baseline scenario, a refilling of the CCR is not considered. The vulnerability increase up to the point in time tCCR expresses a subjective perception. The increase depends on how decision makers percept and evaluate the CCR consumption rate. It could be expressed by different types of depletion graphs Deplk(tstep), which may be linear, quadratic, and exponential or of another type. To determine a depletion graph Depli(tstep), another decision maker survey using a standardized questionnaire was conducted during the second workshop mentioned before. As it is not possible to ask the decision makers for depletion graphs for every possible ΔTCCR and every CI type, the survey is simplified to i consider the vulnerability level for those time steps in which 25, 50, and 75% of the CCR is consumed. The decision makers were asked to estimate vulnerability for these three time steps using a score between 0 for not vulnerable and 1 for highly vulnerable. The estimations of the decision maker vary and result in a specific distribution for each considered time step. This distribution is used to express the uncertainty of the subjective evaluations of the vulnerability increase depending on the CC consumption rate. This uncertainty can be integrated into vulnerability modelling by using a Monte-Carlo simulation. For this purpose, the distributions for each time steps have to be aggregated to functions that represent the vulnerability increase for the whole period and for different probabilities. To do so, quantiles are defined for each distribution of the three considered time steps. Including the fixed values for the further time steps t0 and tCCR with Depli(t0) = 0 and Depli(tCCR ) = 1, a regression i function for each quantile can be derived. Each regression function then allows for the calculation of a vulnerability value for each time step. To conduct the Monte-Carlo simulation, each regression is sampled by its probability distribution. Random variables are used that follow a discrete distribution of the regression. 4.5. Spatial-temporal vulnerability aggregation The weights and indicators are aggregated to indexes. The CI Vulnerability Index VulnCI i , r , u(tstep) expresses how a CI suffers at a specific time step during a power outage taking into account the CI's Size Criticality Indicator, Time Dependent Criticality Indicator, and Coping Capacity Resource Depletion Indicator. It is calculated by a multiplicative aggregation:     CI  Saison  Vulni;r;u t step ¼ C i t step  LoadDay u t step  Si;r;u ∀i ∈ Infrastructure ð5Þ The District Vulnerability Index Vulnidistrict , r , u (tstep) is aggregated by the weighted sum of the CI Vulnerability Index Values of those CIs which are

Please cite this article as: Münzberg, T., et al., A spatial-temporal vulnerability assessment to support the building of community resilience against power outage impacts, Technol. Forecast. Soc. Change (2016), http://dx.doi.org/10.1016/j.techfore.2016.11.027

T. Münzberg et al. / Technological Forecasting & Social Change xxx (2016) xxx–xxx

located in the considered district r and the corresponding CI-type-specific Relevance Criticality Weight: district 

Vulnr

h h   CI  t step ¼ ∑ ∑ Vulni;r;u t step wRelev u i¼1 u¼1

∀ r∈ Districts

ð6Þ

The CI Vulnerability Index Values and the District Vulnerability Index Values are computed for 15-minutemin time steps for the considered power outage scenario. The sequence of values for multiple time steps expresses a time series that display vulnerability profiles. The profiles show the development of the vulnerability for the duration of the considered power outage scenario. In the third step, the function for the vulnerability profiles can also be used to calculate a static Resilience Value R which represent a comparable time-independent metric for the resilience of a city or county taking into account all CI Vulnerability Index values. According to Bruneau et al. (2003), Zobel and Khansa (2014), and Pant et al. (2014), the resilience can be defined as the sum of all integrals of inverse District Vulnerability Index values that includes all temporal vulnerability values of CI facilities for the period under consideration that lasts from t0 as the beginning of a power outage to tp, the point in time where the period under consideration ends: Z

g

Rðα Þ ¼ ∑ r¼1

tp h t0

district 

1−Vulnr

i t step dt

ð7Þ

In this way, resilience values can be calculated for the baseline scenario and additional scenarios α in which the level of preparation for different CI facilities is changed. The result enhances insight into the role of CCR in disaster planning. We propose a rescaling of the resilience values to scale the range in [0, 1]: R0ðα Þ ¼ Z

Rðα Þ

ð8Þ

tp

1 dt t0

This allows for a better comparison of the resilience from different scenarios α. In the following section, we apply the proposed approach to a use case. 5. Case study 5.1. The City of Mannheim and its CIs The results of an application of an exemplary vulnerability assessment are shown and discussed in this section to demonstrate the benefits and limits of the vulnerability assessment. The assessment is applied to the city of Mannheim, Germany, and aims at assessing the six central (Innenstadt/Jungbusch, NeckarstadtWest, Neckarstadt-Ost, Schwetzingerstadt/Oststadt, Lindenhof, and Neuostheim/Neuhermsheim) and eleven peripheral city districts. For demonstration purposes, we are interested in the impacts of the reference power outage scenario. The assessment is focused on the impacts to the CI facilities in the health sector and on households. Hence, 256 individual facilities and app. 176.000 households are Table 2 Types and numbers of considered CIs in the use case. CI type u

Types

Number of facilities per type

1 2 3 5 6 7 27

Hospitals GPs Pharmacies Dialysis clinics Nursing homes Assisted living Households

5 109 93 6 30 13 176.539

11

considered. (Table 2). Relating to the Delphi discourse and the weighting process discussed before, we skip the second Delphi round and use the intermediate summary of the first round in this demonstration. This approach allows for taking into account the weights' uncertainties. Although all values relating to the locations, the sizes, and the preparation levels of the considered CIs are available, we slightly changed the size values and the preparation levels of the considered CIs in this paper for secrecy and privacy reasons. 5.2. Assessment results for the reference power outage scenario Based on the simulation of all possible scenario parameter combinations, it was possible to identify the combination with the highest vulnerability integral. In the considered case, this is a power outage that lasts two working days starting at 8 o'clock a.m. in a winter season. For this worst case scenario, we calculated the aggregated vulnerability values for each city district (Fig. 5) and the overall vulnerability value for the whole city (Fig. 6). In addition, we calculated the vulnerability values of all hospitals for a more detailed analysis of CIs of the same type. In Fig. 6, the districts' vulnerability profiles express the assessment results for the worst case scenario. The vulnerability course is calculated in 15-minutemin time steps. The increasing of the districts' vulnerabilities imply the level of loss or at least reduction of the quality of supply of vital services and products of CIs for the population. In Fig. 7, the results are also shown in a Geographic Information System (GIS) for selected time steps. The vulnerability profiles display a slight vulnerability increase in most districts in the first hours of the power outage. In most districts, the values reach a plateau with relatively high values from 11 a.m. to 5 p.m. For this period, the districts can be distinguished into two groups: One group of districts with relatively high values (comprising Lindenhof, Innenstadt/Jungbusch, Neckarstadt-Ost, Schwetzingerstadt/ Oststadt) and one group with relatively moderate or low values (e.g. Sandhofen, Wallstadt, Seckenheim, Friedrichsfeld etc.). Although the vulnerability values of the first group are comparatively high, the values significantly increase in the further course of time. Hence, the period can be understood as a retarding effect. The retarding effect and the level of the plateau are determined by the CIs which have a relatively high level of preparation. In the aftermath of the retarding effect, the vulnerability of all districts decreases to a very low plateau from 00:00 a.m. to 5 a.m. of the next morning. This effect is mainly caused by the relatively low level of demands for CI products and services at night times. The effect is determined by the load courses which reflect a relatively low level of CI operations for this period. In the next morning from 5:00 a.m. to 9:00 a.m., the vulnerabilities of the districts of the first group dramatically increase (Fig. 8). This increase represents a significant change of the situations in the districts, but also in the whole city. This period can be understood as a tipping point. The tipping point is determined by the increasing demand for CI services and products, but also by the CCRs which is fully depleted in most CIs. Hence, nearly no resources are available at the CIs to adsorb the effect of the tipping point. In the further course of the second day, the vulnerability has to peaks which are separated by the slightly lower demand at midday. Later, at about 7 p.m., the course collapse again although the most CIs run out of CCRs. This effect is mainly caused by the decreasing demand at the night time. The individual vulnerability profile of each district can also be assessed. This is particularly important in cases of anomalies of the vulnerability course. As an example, the profile of the district NeckarstadtOst reaches the highest values compared to the other profiles. A more detailed view of the vulnerability profile, including its uncertainty and the CI types which influence the course, is worthwhile. Fig. 10 shows the profile which has a relatively small uncertainty. The profile is also characterized by a comparatively small vulnerability plateau on the first day, a slight decrease during night time, and a dramatic increase

Please cite this article as: Münzberg, T., et al., A spatial-temporal vulnerability assessment to support the building of community resilience against power outage impacts, Technol. Forecast. Soc. Change (2016), http://dx.doi.org/10.1016/j.techfore.2016.11.027

12

T. Münzberg et al. / Technological Forecasting & Social Change xxx (2016) xxx–xxx

Fig. 5. Cumulative overall Vulnerability Index Value representing the vulnerability of the whole city of Mannheim expressed by the mean value and its simple standard deviation for the considered time steps.

on the next morning with much higher values than on the day before. A special characteristic is a second peak on the second day after the tipping point effect. This second peak is the highest of all vulnerability values of the other districts and mainly caused by the vulnerability profile of the hospital that is situated in Neckarstadt-Ost. In Fig. 10, the vulnerability profile of the district Neckarstadt-Ost is shown with the influence of CI types of the CIs that are situated in Neckarstadt-Ost. Facilities for assisted living are not located in the district. It can be identified easily that the hospitals and dialysis clinics are main drivers of the tipping point effect in the morning of the second day. The vulnerability of the hospital in Neckarstadt-Ost increases nearly linearly. The vulnerability of the dialysis clinics starts to increase in the next morning. In combination with the other CIs and their vulnerability courses, the district's vulnerability escalates dramatically on the second day. The second peak on the second day is mainly influenced by the vulnerability increase of the hospital.

The spatial-temporal vulnerability values and their variabilities can be compared using the vulnerability integral calculation. Fig. 11 displays the comparison of the vulnerability integrals of all districts for the standard reference scenario, the selected starting time, and the considered type of days for which the power outage lasts. The result shows that the vulnerability integrals of the districts Innenstadt/Jungbusch, Neckarstadt-Ost, Schwetzingerstadt, and Lindenhof reach similar values. However, the degrees of variability are different. The highest variability is seen for the vulnerability of the Neckarstadt-Ost.

5.3. Sensitivity analyses Sensitivity analyses aim at identifying the influences of parameter variations on the final results. This frequently allows interesting findings on the robustness of the assessment results and the role of the

Fig. 6. Vulnerability profiles of each city district for a reference power outage lasting two working days in the winter season starting at 8 o'clock.

Please cite this article as: Münzberg, T., et al., A spatial-temporal vulnerability assessment to support the building of community resilience against power outage impacts, Technol. Forecast. Soc. Change (2016), http://dx.doi.org/10.1016/j.techfore.2016.11.027

T. Münzberg et al. / Technological Forecasting & Social Change xxx (2016) xxx–xxx

08:00 a.m.

18:30 p.m.

13

05:00 a.m.

05:00 p.m.

Fig. 7. GIS visualizations of the vulnerability for different time steps.

individual parameters. Both enhances the understanding of the potential impacts and the assessment itself. Referring to the proposed assessment, sensitivity analyses are possible. However, it is important to point out that there is no common assessment sensitivity. Each use case has its own sensitivity which depends on the individual input parameter values. They are specific to the characteristics of the considered CI's in the city or county under consideration. This makes the findings on sensitivity specific to the chosen use case. Multiple input parameter values can be changed and analyzed regarding their impact to the final results. It is possible to consider value changes of the Relevance Criticality Weight, the Size Criticality Indicator, the Coping Capacity Depletion Indicator, and the Time Dependent Criticality Indicator. The weights in multi-attributive assessments have generally great influences on the final results. In addition, the Relevance Criticality Weights are subjectively determined by expert estimations in the proposed assessment. This makes the weight sensitivity one of the most interesting aspect in the sensitivity analysis. We discuss this type of

06:00 a.m.

07:00 a.m.

analysis in detail to demonstrate how sensitive analysis provide valuable insights, although there also other types of analysis. To analyze the sensitivity of weight changes, we use the Overall Vulnerability Index OverallVuln which represent the sum of all integrals of District Vulnerability Indexes taking into account all time steps under consideration: g

OverallVuln ¼ ∑ r¼1

Z

tp t0

district 

Vulnr

 t step dt

ð9Þ

(tstep) The calculation of the District Vulnerability Index Vulndistrict r comprises the normalized relevance criticality weight which is in turn calculated by the absolute importance weighting score wscore . To identiu fy major influences on the variation of weight values, we exemplarily assumed an extreme change of the importance weighting score of a CI type under consideration u_sens by minus 95% of the previous score sens wuscore ¼ 0; 05wscore u sens sens

08:00 a.m.

ð10Þ

09:00 a.m.

Fig. 8. GIS visualizations of the vulnerability of each city district at different time steps during the turning point effect.

Please cite this article as: Münzberg, T., et al., A spatial-temporal vulnerability assessment to support the building of community resilience against power outage impacts, Technol. Forecast. Soc. Change (2016), http://dx.doi.org/10.1016/j.techfore.2016.11.027

14

T. Münzberg et al. / Technological Forecasting & Social Change xxx (2016) xxx–xxx

Fig. 9. Vulnerability profile of the city district Neckarstadt-Ost expressed by the mean value and its simple standard deviation for the considered time steps.

The Overall Vulnerability Index is then calculated using the changed importance weighting score for the CI type under consideration. The extreme adjustment enables the identification of key drivers for vulnerability and resilience in a city (see Fig. 12). Fig. 12 shows the changes of the Overall Vulnerability Index after importance weighting scores of one CI type have been modified by minus 95% of the initial weight value. The results show the median values and the simple standard deviations. For some CI types, such as hospitals and households, the resulting overall vulnerability increased. This means that a minor assumed relevance of these CI types would lead to higher vulnerability values. Hence, these CI types can be identified as resilience drivers. In contrast to this, the resulting overall vulnerability of pharmacies is smaller compared to the results of the baseline. Hence, pharmacies can be identified as vulnerability drivers in this use case. It should be noted that although a change of the other CI types has no high impact on the overall vulnerability integral for the city of

Mannheim, a change may lead to different results of the vulnerability profiles of districts. There are further multiple ways to change input parameters and to analyze the sensitivity on the final results. An evaluation about changes on the Relevance Criticality Weight and the Coping Capacity Depletion Indicator, for instance, is also possible through the use of the MonteCarlo simulation. Usually, the Monte-Carlo simulation allows the consideration of uncertainty. However, the results also inherently provide information about the sensitivity of changes in the final result (see Figs. 5, 9, and 11). Another aspect is the change of values of the Size Criticality Indicator. The Size Criticality Indicator is a relation of the size of a certain CI and the sizes of the CIs from the same type. Through this, there is either a change in the overall vulnerability nor in the resilience. The changes can only be identified by focusing on the results of individual CIs or the districts. However, a change of this indicator values is not possible

Fig. 10. Vulnerability profile of the district Neckarstadt-Ost in the form of a stacked bar chart displaying the values for different CI types.

Please cite this article as: Münzberg, T., et al., A spatial-temporal vulnerability assessment to support the building of community resilience against power outage impacts, Technol. Forecast. Soc. Change (2016), http://dx.doi.org/10.1016/j.techfore.2016.11.027

T. Münzberg et al. / Technological Forecasting & Social Change xxx (2016) xxx–xxx

15

Fig. 11. Normalized vulnerability integrals for the city district taking into account variability through simple standard deviations.

during an assessment, in practice. Insights about the sensitivity have therefore only limited implications. The Influence of value changes of the Coping Capacity Depletion Indicator is from great interest for disaster management. This addresses the closing of gaps in preparation and may helps to identify the CI's and district which have high influence on the overall resilience because of a lack or sufficiency of coping capacity. To consider this, it is also possible to conduct what-if analyses which are discussed in the next section. 5.4. Comparing strategies by “what-if” analyses The vulnerability assessment also allows for a comparison of threat scenarios and alternative preparation strategies. Since this is realized inter alia by changing input parameters for the CCRs, this can also be understood as a sensitivity analysis of the Coping Capacity Depletion

Indicator. In the following section, we calculate three sample threat scenarios and their potential impacts. In addition, we select five fictitious preparation strategies and calculate their outcomes for demonstration purposes. Each calculation run is defined as one scenario which starts with the first minute of the considered power outage. A power outage can affect the whole city, but also only some districts. It is also possible that the power outage starts at another time. In this way, outage scenarios can vary regarding the spatial expansion and time. While retaining the reference power outage, we exemplarily assess the results of sample threat scenarios described in Table 3. The overall vulnerability profiles for the scenarios a, b, and c are displayed in Fig. 13. For the disaster management authorities, the results allow to adjust the response activities referring to the divergent spatialtemporal effects of the individual scenario. The results show that a power outage which affects the central districts would lead to higher vulnerability values in comparison to a power outage which only affects

Fig. 12. Sensitivity analysis to display the influence of variations of the importance weighting score for a CI type under consideration. The figure shows the change of the Overall Vulnerability Index after the importance weighting score was changed by minus 95% compared to the baseline score.

Please cite this article as: Münzberg, T., et al., A spatial-temporal vulnerability assessment to support the building of community resilience against power outage impacts, Technol. Forecast. Soc. Change (2016), http://dx.doi.org/10.1016/j.techfore.2016.11.027

16

T. Münzberg et al. / Technological Forecasting & Social Change xxx (2016) xxx–xxx

Table 3 Description of sample threat scenarios. Scenario α

Description

Baseline

Worst case power outage scenario affects all districts. It starts at 8 a.m. and lasts two working days in winter. A power outage affects all central districts. It starts at 8 a.m. and lasts two working days in winter. A power outage affects all peripheral districts. It starts at 8 a.m. and lasts two working days in winter. A power outage affects all districts. It starts at 6 p.m. and lasts two working days in winter.

Scenario a Scenario b Scenario c

Table 4 Scenario description of sample strategies. Scenario α

Description

Baseline

Worst case power outage scenario affects all districts. It starts at 8 a.m. and lasts two working days in winter. All CIs are enabled to keep their business operated for at least 6 h during a power outage All CIs are enabled to keep their business operated for at least 12 h during a power outage All CIs which are located in the central districts are enabled to keep their business operated for at least 18 h during a power outage All CIs which are located in the central districts are enabled to keep their business operated for at least 12 h during a power outage. All CIs which are located in the peripheral districts are enabled to keep their business operated for at least 6 h during a power outage. All nursing homes and facilities for assisted living are enabled to keep their business operated for at least 24 h during a power outage.

Scenario d Scenario e Scenario f Scenario g

the peripheral districts. The fact can be taken into account in the disaster preparedness policy making process or in designing standard operation procedures which address the initial disaster response reaction to power outages. The results of the scenario c expresses the sensitivity referring to the starting time of a power outage. Scenario c starts at 6 p.m. and the first outage hours are in the night time. Corresponding to the low level of demand at this time, the vulnerability reaches comparatively low values. Unlike to the baseline scenario, there is only a single escalation which occur earlier. However, the vulnerability values in scenario c are slightly lower because there are still some CCRs available. The decision makers may consider different preparation strategies. We exemplarily selected five fictitious preparation strategies, each of which is to enhance the resilience in dealing with the effects of a power outage by enhanced CCRs (Table 4). The scenarios are applied for the reference power outage starting at 8 a.m. and lasting two working days in the winter season. The vulnerability profiles of the scenarios d, e, f, g, and h are shown in Fig. 14. The results of the scenario display a change in the vulnerability course of the first hours. The core finding of the vulnerability profiles in Fig. 13 is that the scenario courses are depended and very sensitive on the CCRs of the affected CIs. The CIs vulnerability profiles escalate to a lower extant or are delayed in comparison with the baseline scenario if the CI are enabled to keep their operations at least for 6 or 12 h. Through this, the courses crosses in some cases. Afterwards this duration, the vulnerability paths are the same as in the baseline scenario. In scenario d, for instance, all CIs are enabled to keep their business for six hours minimum. The vulnerability nearly increase linearly till the sixth outage hour and then takes the same course of the baseline scenario afterwards. In comparison to the baseline scenario, there is no first vulnerability peak after 4 h in scenario d. This also applies in

scenario e in which all CIs are enabled to keep their business for 12 h. The course is also lower and the escalation considerably weaken compared to the baseline scenario. Interestingly, a similar effect occurs in scenario g in which all CIs located in the central districts are enabled to keep their business operated for at least 12 h and all CIs which are located in the peripheral districts are enabled to keep their business for at least 6 h. If only the CIs located in the central districts focused and enabled to keep their business for at least 18 h like in scenario f, the vulnerability course is higher in the beginning of the outage. However, the further course is slightly lower as in the scenarios d and g. In scenario h, a CCR adjustment for specific CI types was conducted. The results show, that the course is on a lower extant but still on the same course as the baseline scenario. The scenario results are interesting demonstrations of the considerable spatial influences of the CIs that are located in the central or peripheral districts. Such scenario comparisons provide important insights for the disaster management authorities. The temporal impacts of changed preparation levels of CIs are visualized. The different benefits are transparent. Moreover, “what-if” analysis allow simulations and comparisons of different preparation scenarios for certain CI types like demonstrated in scenario h. This also allows testing and understanding the role of CI types on the overall vulnerability of a city or county, but also enables multi-level scenarios in which regions from different character (e.g. central/peripheral) can be summarized and assessed together.

Fig. 13. Vulnerability profiles for the scenarios a, b, and c.

Fig. 14. Vulnerability profiles for the scenarios d, e, f, g, and h.

Scenario h

Please cite this article as: Münzberg, T., et al., A spatial-temporal vulnerability assessment to support the building of community resilience against power outage impacts, Technol. Forecast. Soc. Change (2016), http://dx.doi.org/10.1016/j.techfore.2016.11.027

T. Münzberg et al. / Technological Forecasting & Social Change xxx (2016) xxx–xxx

5.5. Implication to resilience building process and decision support The proposed assessment can have beneficial implications for the resilience building process. The assessment results allow decision makers to identify the specific temporal development of CI's and district's vulnerabilities during a power outage in a city or county under consideration. The CIs located in a city or county are evaluated regarding their specific preparation level, social relevance, and size. The parameter uncertainties due to variation of parameter values are expressed in the assessment results. Temporal consequences of different power outage scenarios can be simulated and assessed. The results display the progress of vulnerability, its peaks and tipping points. The assessment outcomes can be visualized by diagrams and geographic information systems. The spatial insight about the outage effects is enhanced by the underlying evaluation of individual CI's characteristics and spatial what-if analysis. The CI's characteristics and the potential outage effects are placed in relation to the function of the entire CI system. The spatial aggregation of the CI's vulnerability values and the spatial what-if analysis facilitate comparisons of regional power outage consequences that have impacts to the whole city or county. The assessment's findings provide valuable insights and can assist decision making processes during the disaster mitigation and preparedness phase. The outcome of the assessment clearly point out the influence of certain CIs and CI types to the vulnerability progress. On the one hand side, this helps CI providers in better reflecting their role in the entire CI system of a city or county. It is possible to compare the preparation levels of CIs from the same type which may also motivate some CI providers to invest into more CCRs. On the other hand, the disaster management authorities can identify lacks of preparation and the most vulnerable CIs and districts. This may also increase their effectiveness because it is possible to concentrate all mitigation activities at the highest vulnerable CIs and regions instead of giving too much attention on CIs that are less relevant. In this way, the vulnerability assessment results provide transparency and foster the building of resilience. The CI specific knowledge about the temporal vulnerability and the district specific spatial-temporal findings can also be used in designing standardized and appropriated reactions to a power outage. The simulation and assessment result help to define suitable and specific initial response actions. In this way, standard operation procedures, contingency plans, and backup strategies can be refined by using the assessment results. A favorable gain of the assessment is the integration of data from CI providers and of decision maker's estimations. Both facilitate the collaboration between CI providers and disaster management authorities. To conduct an assessment, it is necessary that CI providers supply disaster management authorities with essential information about their infrastructures. Due to the construction of assessment, the exchange of data is structured and reduced to a comparatively minor expense for data collection. The decision maker's estimations on CI's importance and CCR depletion are included in the assessment by a Delphi like collaboration technique. In this way, different and sometimes conflicting views can be considered systematically. The integration of decision maker's estimation makes it possible to create a comprehensive and common picture about the impacts of a power outages, the potential lacks in disaster planning, and potential losses and reductions of CI services during a power outage. Another problem in the mitigation and preparedness phase is that it is unknown on what type of day and at which daytime the next severe power outage starts and how long it will lasts. However, a concrete scenario is necessary for disaster planning. This is solved by the definition of a reference power outage scenario that represents a reasonable case of a power outage. The definition of a reference scenario allows to calculate the worst case of a power outage. The day types and the starting time of the worst case are very important findings for the development of disaster plans. The recognition can also raise the awareness for power outage consequences within the activities of risk communication.

17

The assessment and its results can also be used during a particular power outage in the disaster response and recovery phases. If the assessment is already implemented the assessment provides forecast capabilities which enable a better reaction to an ongoing power outage. In this way it can be used as decision support. The scenario characteristics are known during an ongoing power outage. This makes it possible to estimate the specific initial spatial-temporal consequences for a present power outage. The assessment results can used for the development of action plans but also in all cases in which it is necessary to prioritize CIs or districts. The evaluation of CI's and district's susceptibility and criticality enables to identify those CIs and districts which should be better protected against a power outage or restored as soon as possible after an outage event. The resulting prioritization is helpful in cases in which a limited amount of mobile emergency power units are available and have to be distributed among the affected CIs and districts. In other cases of power outages it is still possible to supply some CIs or some districts of a town or county with a limited amount of electricity. In these cases, the CIs or districts can be ranked for a prioritized electricity supply. In other cases such as load shedding procedures, it is necessary to select districts which have to be decoupled from the electricity grid to ensure network stability. By using the results of the vulnerability assessment, it is possible to identify those regions that have no CIs or CIs with low criticality and which are less susceptible for the consequences of a power outage. A vulnerability based ranking of districts can provide a list with those districts which should be decoupled and be affected by a power outage first. Depending on the cause of a power outage, a prioritization is also sometimes necessary in the recovery phase. After a power outage it is often the case that districts are successively recoupled to a recovered electricity grid. There are also cases in which electrical substations have to be restored and for this purpose manually recoupled to the grid. Since these procedures are very time consuming and the affected CI should not be kept longer without electricity than necessary, the assessment results can provide a prioritization and a list of districts which should be restored first. In all cases in which prioritizations are a helpful decision support, the assessment results can be used to enhance resilience of CIs and through this of cities or counties. 6. Critical remarks and conclusion In this paper, we introduced an indicator-based spatial-temporal vulnerability assessment to enable crisis management groups, disaster management authorities, and CI providers to enhance their understanding of the initial impacts of a power outage. The assessment results provide insights into the resilience of certain CIs and districts, and, hence, allow for simulating the effectiveness of considered preparation and response. The implementation of the assessment was demonstrated for the CIs of the health sector in the city of Mannheim. We demonstrated that the findings are valuable in providing an enhanced spatial-temporal understanding and in facilitating decision processes. The assessment can also be applied to other cities and districts. The proposed assessment considers the legal situation in Germany (regional focus, focus on local CIs, selection of decision makers). If the assessment is applied to regions in other countries, there might be slight adjustments necessary according to the legal regulations. If desired by the decision makers, also other CIs can be stressed. The assessment is generally easy to apply because only a few data about the local CIs is requested which is often publicly available or can be collected by the disaster management authorities in charge with calculable expense. The vulnerability assessment takes into account the location of potentially affected CIs, their criticality for providing vital services and products to the population, and their CCRs which allow them to continue the CI operations during a power outage for a short time (Coping Capacity Resource Depletion Indicator). The criticality has three dimensions: The relevance weights of CI types (Relevance Criticality Weight), the CI size (Size Criticality Indicator), and the time-depending demand of CIs (Time Dependent Criticality Indicator). We limited the

Please cite this article as: Münzberg, T., et al., A spatial-temporal vulnerability assessment to support the building of community resilience against power outage impacts, Technol. Forecast. Soc. Change (2016), http://dx.doi.org/10.1016/j.techfore.2016.11.027

18

T. Münzberg et al. / Technological Forecasting & Social Change xxx (2016) xxx–xxx

dimensions of criticality to the smallest reasonable number. Additional criticality dimensions and adjustments of the used indicators could provide slightly more concrete or detailed results, but this would also request more data and would lead to higher efforts in conducting the vulnerability assessment. The weighting process for the definition of relevance weights is of crucial importance and requires a group decision. In practice, the relevance weights are often defined subjectively and random way. We proposed a modified Delphi survey, which provides a more systematic and better structured definition of the relevance weights. In the first round of the Delphi survey, two workshops were conducted, during which decision makers from different disaster management authorities were asked to weigh the relevance of CI types using a questionnaire and the direct weighting method. The resulting intermediate summary displays the weight value dispersions for the considered CI types and, hence, is a helpful means for those decision makers who apply the vulnerability assessment in a certain city or county in the second round. Using the intermediate summary in the weight definition process produces a higher legal certainty. This also minimizes ambiguity in the consensus finding process. However, it is still possible to adjust weights, if necessary in the second round. To consider the sizes of CIs we proposed a CI-type specific attribute. A CI's attribute value is linearly normalized referring to the sum of all CI's attributes from the same type. Also other normalization could be applied. In addition, there might be other or more attributes for one CI type which are also suited for measuring the CI's sizes but may also enhance the efforts in implementing the assessment. According to the Time Dependent Criticality Indicator, we assumed that a CI can only work correctly if all processes are supplied with the requested electricity. The current approach, the indicator values are derived from SLPs which have a temporal resolution of the results of 15 min. From a practical point of view, also lower resolution from 30 to 60 min would also be appropriate. More research is needed to identify the most suitable temporal granularity of results. The use of SLPs do not allow the consideration of shifting processes. This flexibility will be increased in the future by the implementation of smart technologies. Therefore, new vulnerabilities as well as new ways for dealing with power outage impacts arise. Another important component of the vulnerability assessment is the depletion of CCRs. CI vulnerability increases, while its CCR is consumed during a power outage. The increase modelling is based on subjective estimations of experts from different disaster management authorities. To enhance the insight into this estimation, another expert evaluation was conducted in a workshop. This was the first expert evaluation on this topic as far as we know. There might also be other resources which allow for a fast reaction, an enhanced problem solving capability, a business continuity or a faster recovery after a disruption. These capabilities often are of high interest in risk and crisis management, too. They may be integrated by additional attributes to characterize further dimensions of a CI's CCR. In this case, it is important to note that these additional attributes need to be of general character because it must be guaranteed that the attributes can be applied and measured for all CIs. The weighting process and the evaluation of vulnerability increase depending on the CCR depletion include statistical uncertainties. These parameter and weighting uncertainties are integrated into the final vulnerability aggregation by using Monte-Carlo approaches. However, the results of the use case have shown that the uncertainties for certain time steps of the vulnerability profiles are relatively low. This may be different in other cases. However, calculations without the consideration of parameter uncertainties may also be reasonable for decision support. In some cases it is possible to warn the population, the disaster management authorities, and the CI provider of an upcoming power outage. This would minimize the impacts to the CIs and the population because a better and early preparation is possible, food, drinking water, and fuel

stocks could be refilled, and processes that rely highly on electricity can be prematurely shutdown to avoid further damages, etc. However, there is no data available how warnings influence the resilience of CIs and the population and, hence, this is difficult to integrate. The proposed assessment is of course still no “silver bullet”. It can neither forecast the unpredictable nor solve all problems in disaster planning for a power outage. There are always aspects which cannot be tackled. In this case, this applies, for instance, for the coordination of fuel for the emergency power units during a power outage, aspects of the occurrence probability of power outages, CI's recovery capabilities or the CI interdependencies. In combination with other models and analysis approaches which address these issues, additional insights are possible. The main added value of the vulnerability assessment is its straightforward approach that can be easily applied to any city or county with a manageable effort. Through this, it is possible to increase the community disaster resilience according to the aims of the UN Sendai Framework.

Acknowledgement The work is embedded in the Helmholtz Association's (HGF) portfolio project “Security Research” and the critical infrastructure protection activities of the Center for Disaster Management and Risk Reduction Technologies (CEDIM) of the Karlsruhe Institute of Technology (KIT). The HGF's and CEDIM's financial support of the work is gratefully acknowledged. We would like to thank all participants of the mentioned workshops for their collaboration and dedicated support. Some of the figures above show maps made by using QGIS (more information about the Open Source Geospatial Foundation Project see http://qgis.osgeo. org) and data derived from ©OpenStreetMap for visualizing the assessment results. References Adger, N., 2006. Vulnerability. Glob. Environ. Chang. 16, 268–281. Ajodhia, V., van Gemert, M., Hakvoort, R., 2002. Electricity outage cost valuation: a survey. Discussion Paper, DTe, Den Haag, the Netherlands. Akgun, I., Kandakoglu, A., Ozok, A.F., 2010. Fuzzy integrated vulnerability assessment model for critical facilities in combating the terrorism. Expert Syst. Appl. 37 (5): 3561–3573. http://dx.doi.org/10.1016/j.eswa.2009.10.035 (May 2010). Alfares, H.K., Nazeeruddin, M., 2002. Electric load forecasting: literature survey and classification of methods. Int. J. Syst. Sci. 33 (1), 38–48 (2002). Anagnostatos, S.D., Halevidis, C.D., Polykrati, A.D., Bourkas, P.D., Karagiannopoulos, C.G., 2013. Examination of the 2006 blackout in Kefallonia Island, Greece. Int. J. Electr. Power Energy Syst. 49:122–127. http://dx.doi.org/10.1016/j.ijepes.2012.12.003 (ISSN 0142-0615, July 2013). Aubrecht, Steinnocher, K., Köstl, M., Züger, J., Loibl, W., 2013. Long-term spatio-temporal social vulnerability variation considering health-related climate change parameters particularly affecting elderly. Nat. Hazards 68, 1371–1384. Bagheri, E., Baghi, H., Ghorbani, A.A., 2007. An agent-based service-oriented simulation suite for critical infrastructure behaviour analysis. Int. J. Bus. Process. Integr. Manag. 2 (4). http://dx.doi.org/10.1504/IJBPIM.2007.017756. Bañuls, V.A., Turoff, M., Hiltz, S.R., 2013. Collaborative scenario modeling in emergency management through cross-impact. Technol. Forecast. Soc. Chang. 80, 1756–1774 (2013). Barata, I.A., Llovera, I., Ward, M.F., Riccardi, D., Mayerhoff, R.M., Litroff, A., Sam, A., 2005. Are individual households prepared for a disaster such as fire, flood, earthquake, blackout or a terrorist attack in their community? Ann. Emerg. Med. 46 (3, Supplement), 93–94 (September 2005). Belton, V., Stewart, T., 2002. Multiple Criteria Decision Analysis – An Integrated Approach. Kluwer Academic Press, Boston (2002). Berariu, R., Fikar, C., Gronalt, M., Hirsch, P., 2015. Understanding the impact of cascade effects of natural disasters on disaster relief operations. Int. J. Disaster Risk Reduct. 12, 350–356 (June 2015). Bertsch, V., 2008. Uncertainty Handling in Multi-Attribute Decision Support for Industrial Risk Management. Universitätsverlag Karlsruhe http://dx.doi.org/10.5445/KSP/ 1000007378. Billinton, R., 2001. Methods to Consider Customer Interruption Costs in Power System Analysis. Brochures thématiques/CIGRE Paris. Conseil International des Grands Réseaux Électriques (CIGRE). Billinton, R., Tollefson, G., Wacker, G., 1993. Assessment of electric service reliability worth. Electr. Power Energy Syst. 15 (2), 95–100. Birkmann, J., 2006. Measuring vulnerability to promote disaster-resilient societies: conceptual frameworks and definitions. In: Birkmann, J. (Ed.), Measuring Vulnerability to Natural Hazards. United Nations University Press, New York (2006).

Please cite this article as: Münzberg, T., et al., A spatial-temporal vulnerability assessment to support the building of community resilience against power outage impacts, Technol. Forecast. Soc. Change (2016), http://dx.doi.org/10.1016/j.techfore.2016.11.027

T. Münzberg et al. / Technological Forecasting & Social Change xxx (2016) xxx–xxx Birkmann, J., Wisner, B., 2006. Measuring the un-measurable. The Challenge of Vulnerability. UNU Institute for Environment and Human Security (UNU-EHS), Bonn (2006). Blake, E.S., Kimberlain, T.B., Berg, R.J., Cangialosi, J.P., Beven, J.L., 2013. Tropical Cyclone Report Hurricane Sandy (AL182012), 11–29 October 2012. National Hurricane Centre, Miami, Florida (http://www.nhc.noaa.gov. Accessed 3 Sep 2015). Blauhut, V., Stahl, K., Vogt, J., 2015. Assessing risk by impacts: a probabilistic approach for drought assessment in Europe. EGU General Assembly 2015, Vol. 17, EGU2015-688. Bogardi, J.J., 2004. Hazards, risks and vulnerabilities in a changing environment: the unexpected onslaught on human security? Glob. Environ. Chang. 14, 361–365. Brooks, N., 2003. Vulnerability, risk and adaptation: a conceptual framework. Tyndall Centre for Climate Change, Working Paper 38, p. 16. Bruneau, M., Chang, S., Eguchi, R., Lee, G., O’Rourke, T., Reinhorn, A., Shinozuka, M., Tierney, K., Wallace, W., von Winterfelt, D., 2003. A framework to quantitatively assess and enhance the seismic resilience of communities. EERI Spectra Journal 19 (no.4), 733–752. Buldyrev, S.V., Parshani, R., Paul, G., Stanley, H.E., Havlin, 2010. Catastrophic cascade of failures in interdependent networks. Nature 464 (7291), 1025–1028. Bundesministerium des Innern (BMI), 2011. Sektoren- und Brancheneinteilung Kritischer Infrastrukturen https://www.bmi.bund.de/SharedDocs/Downloads/DE/Themen/ Sicherheit/BevoelkerungKrisen/Sektoreneinteilung.pdf?__blob=publicationFile, (2011 Web. 20 July, 2015). Bush, B., Dauelsberg, L., LeClaire, R., Powell, D., Deland, S., Samsa, M., 2005. Critical infrastructure protection decision support system (CIPS/DSS) project overview. Proceedings of the Twenty-Third International Conference of the System Dynamics Society, pp. 17–21. Cardona, O.D., 1999. Environmental management and disaster prevention: two related topics: a holistic risk assessment and management approach. In: Ingleton, J. (Ed.), Natural Disaster Management. IDNDR-Tudor Rose, London, UK, pp. 151–153. Casalicchio, E., Galli, E., Tucci, S., 2008. Modeling and simulation of complex interdependent systems: a federated agent-based approach. Crit. Inf. Infrastruct. Secur. Volume 5508 of the Series Lecture Notes in Computer Science:pp. 72–83 http://dx.doi.org/10. 1007/978-3-642-03552-4_7. Casalicchio, E., Galli, E., Tucci, 2010. Agent-based modelling of interdependent critical infrastructures. Int. J. Syst. Syst. Eng. 2 (1) (2010). Castillo, A., 2014. Risk analysis and management in power outage and restoration: a literature survey. Electr. Power Syst. Res. 107, 9–15 (February 2014). Cavallini, S., d'Alessandro, C., Volpe, M., Armenia, S., Carlini, C., Brein, E., Assogna, P., 2014. In: Butts, J., Shenoi, S. (Eds.), A System Dynamics Framework for Modeling Critical Infrastructure Resilience, Critical Infrastructure Protection VIII. Springer:pp. 141–154 http://dx.doi.org/10.1007/978-3-662-45355-1_10. Claudio, M., Rocco, S., Ramirez-Marquez, J.E., 2011. Vulnerability metrics and analysis for communities in complex networks. Reliab. Eng. Syst. Saf. 96 (10):1360–1366. http:// dx.doi.org/10.1016/j.ress.2011.03.001 (October 2011). Conejo, A.J., Carrión, M., Morales, J.M., 2010. Decision making under uncertainty in electricity markets. Int. Ser. Oper. Res. Manag. Sci. 153. http://dx.doi.org/10.1007/ 978–1–4419-7421-1 (2010). Dalkey, N.C., Helmer, O., 1963. An experimental application of the Delphi method to the use of experts. Manag. Sci. 9 (3):458–467. http://dx.doi.org/10.1287/mnsc.9. 3.458. de Nooij, M., Koopmans, C., Bijvoet, C., 2007. The value of supply security: the costs of power interruptions: economic input for damage reduction and investment in networks. Energy Econ. 29 (2007), 277–295. de Nooij, M., Lieshout, R., Koopmans, C., 2009. Optimal blackouts: empirical results on reducing the social cost of electricity outages through efficient regional rationing. Energy Econ. 31 (2009), 342–347. Debnath, R., 2013. An assessment of spatio-temporal pattern of urban earthquake vulnerability using GIS: a study on Dhaka City. Ann. GIS 19 (2), 63–78. der Vleuten, E.v., Lagendijk, V., 2010. Transnational infrastructure vulnerability: the historical shaping of the 2006 European “Blackout”. Energ Policy 38 (4):2042–2052. http://dx.doi.org/10.1016/j.enpol.2009.11.047 (ISSN 0301-4215, April 2010). Di Giorgio, A., Liberati, F., 2012. A Bayesian network-based approach to the critical infrastructure interdependencies analysis. IEEE Syst. J. 6 (3):2012. http://dx.doi.org/10. 1109/JSYST.2012.2190695. Eusgeld, I., Henzi, D., Kröger, W., 2008. Comparative Evaluation of Modeling and Simulation Techniques for Interdependent Critical Infrastructures Scientific Report. ETH Zürich, Laboratorium für Sicherheitsanalytik. Gaillard, J., 2010. Vulnerability, capacity and resilience: perspectives for climate and development policy. J. Int. Dev. 218–232. Giannopoulos, G., Filippini, R., Schimmer, M., 2012. Risk assessment methodologies for critical infrastructure protection. Part I: A State of the Art, EUR 25286 EN - 2012. Growitsch, C., Malischek, R., Nick, S., Wetzel, H., 2013. The costs of power interruptions in Germany - an assessment in the light of the Energiewende. Institute of Energy Economics at the University of Cologne (EWI), Working Paper, No 13/07. Haimes, Y., Horowitz, B., Lambert, J., Santos, J., Lian, C., Crowther, K., 2005. Inoperability input-output model for interdependent infrastructure sectors. I. Theory and methodology. J. Infrastruct. Syst. http://dx.doi.org/10.1061/(ASCE)1076-0342(2005)11: 2(67), 67-79. Hayn, M., Bertsch, V., Fichtner, W., 2014. Electricity load profiles in Europe: the importance of household segmentation. Energy Res. Soc. Sci. 3, 30–45. Hiete, M., Merz, M., Trinks, C., Grambs, W., Thiede, T., 2010. Krisenhandbuch Stromausfall Baden-Württemberg – Krisenmanagement bei einer großflächigen Unterbrechung der Stromversorgung am Beispiel Baden-Württemberg, Innenministerium BadenWürttemberg (Hrsg.).

19

Hippert, H.S., Pedreira, C.E., Souza, R.C., 2001. Neural networks for short-term load forecasting: a review and evaluation. IEEE Trans. Power Syst. 16 (1), 44–55 (Feb 2001). Howell, J.P., 2012. When the lights went out: a history of blackouts in America. J. Hist. Geogr. 38 (1):95–96. http://dx.doi.org/10.1016/j.jhg.2011.11.015 (ISSN 0305-7488, January 2012). Hufschmidt, G., 2011. A comparative analysis of several vulnerability concepts. Nat. Hazards 58 (2), 621–643. Jha, M.K., Keel, R.A., 2012. In: Premchaiswadi, W. (Ed.), Using Dynamic Bayesian Networks for Investigating the Impacts of Extreme Events, Bayesian Network (ISBN 978-95351-0556-5, 2012). Kadri, F., Birregah, B., Châtelet, E., 2014. The impact of natural disasters on critical infrastructures: a domino effect-based study. J. Homel. Secur. Emerg. Manag. 11 (2), 217–242. Kasperson, J.X., Kasperson, R.E., Turner, B.L., 1995. Regions at Risk: Comparisons of Threatened Environments. United Nations University Press, New York. Khujadze, S., 2014. In: United States Agency for International Development (Ed.), A Study of the Value of Lost Load (VoLL) for Georgia. Kinn, M.C., Abbott, C., 2014. To what extent is electricity central to resilience and disaster management of the built environment? Procedia Econ. Financ. 18 (2014), 238–246. Klein, R., Rome, E., Beyel, C., Linnemann, R., Reinhardt, W., 2008. Information Modelling and Simulation in Large Interdependent Critical Infrastructures in IRRIIS. Klinger, C., Landeg, O., Murray, V., 2014. Power outages, extreme events and health: a systematic review of the literature from 2011–2012. PLOS Curr. Disasters (2014 Jan 2. Edition 1). Kozik, R., Choraś, M., Hołubowicz, W., Renk, R., 2015. Increasing protection and resilience of critical infrastructures – current challenges and approaches. J. Pol. Saf. Reliab. Assoc. Summer Saf. Reliab. Semin. 6 (3) (2015 79). Kröger, W., 2008. Critical infrastructures at risk: a need for a new conceptual approach and extended analytical tools. Reliab. Eng. Syst. Saf. 93 (12): 1781–1787. http://dx.doi.org/10.1016/j.ress.2008.03.005 (ISSN 0951-8320, December 2008). Kröger, W., Nan, C., 2014. In: D'Agostino, G., Scala, A. (Eds.), Addressing Interdependencies of Complex Technical Networks, Understanding Complex Systems, Networks of Networks: The Last Frontier of Complexity (ISBN: 978-3-319-03517-8). Kunz, M., Mühr, B., Kunz-Plapp, T., Daniell, J.E., Khazai, B., Wenzel, F., Vannieuwenhuyse, M., Comes, T., Elmer, F., Schröter, K., Fohringer1, J., Münzberg, T., Lucas, C., Zschau, J., 2013. Investigation of Superstorm Sandy 2012 in a multi-disciplinary approach. Nat. Hazards Earth Syst. Sci. 13, 2579–2598. Laakso, K., Palomäki, J., 2013. The importance of a common understanding in emergency management. Technol. Forecast. Soc. Chang. 80 (2013), 1703–1713. Laugé, A., Hernantes, J., Sarriegi, J.M., 2015. Critical infrastructure dependencies: a holistic, dynamic and quantitative approach. Int. J. Crit. Infrastruct. Prot. 8:16–23. http://dx. doi.org/10.1016/j.ijcip.2014.12.004 (January 2015). Leahy, E., Tol, R.S.J., 2011. An estimate of the value of lost load for Ireland. Energ Policy 39 (2011), 1514–1520. Lee, G., Jun, K.-S., Chung, E.-S., 2013. Integrated multi-criteria flood vulnerability approach using fuzzy TOPSIS and Delphi technique. Nat. Hazards Earth Syst. Sci. 13:1293–1312. http://dx.doi.org/10.5194/nhess-13-1293-2013 (2013). Linstone, H.A., Turoff, M., 1975. The Delphi Method: Techniques and Applications. Addison-Wesley, Reading, Mass (ISBN 978–0–201-04294-8). Luiijf, E., Nieuwenhuijs, A., Klaver, M., Eeten, M.V., Cruz, E., 2009. Empirical findings on critical infrastructure dependencies in Europe. In: Setola, R., Geretshuber, S. (Eds.), Critical Information Infrastructure Security, Lecture Notes in Computer Science. Springer Berlin/Heidelberg, Berlin, Germany 2009. Manyena, S.B., 2006. The concept of resilience revisited. Disasters 30 (4), 433–450. Marti, J., Hollman, J., Ventura, C., Jatskevich, J., 2008. Dynamic recovery of critical infrastructures: real-time temporal coordination. Int. J. Crit. Infrastruct. 4 (1/2), 17–31. Min, H.-S.J., Beyeler, W., Brown, T., Son, Y.J., Jones, A.T., 2007. Toward modeling and simulation of critical national infrastructure interdependencies. IIE Trans. 39 (1):57–71. http://dx.doi.org/10.1080/07408170600940005. Moss, R.H., Brenkert, A.L., Malone, E.L., 2001. Vulnerability to climate change: a quantitative approach. Technical Report PNNL-SA-33642. Pacific Northwest National Laboratories, Richland, WA. Münzberg, T., Mueller, T., Möhrle, S., Comes, T., Schultmann, F., 2013. An integrated multicriteria approach on vulnerability analysis in the context of load reduction. In: Comes, T., Fiedrich, F., Fortier, S., Geldermann, J., Müller, T. (Eds.), Proceedings of the 10th International Conference on Information Systems for Crisis Response and Management (ISCRAM 2013), Baden-Baden, Germany, May 2013, pp. 251–260 (ISBN 978-3923704-80-4). Nakayama, T., Tanaka, S., Uematsu, M., Kikuchi, A., Hino-Fukuyo, N., Morimoto, T., Sakamoto, O., Tsuchiya, S., Kure, S., 2014. Effect of a blackout in pediatric patients with home medical devices during the 2011 eastern Japan earthquake. Brain Dev. 36 (2), 143–147 (February). Neenan, B., Hemphill, R.C., 2008. Societal benefits of smart metering investments. 21 (8), 32–45 (October 2008). Oliva, G., Panzieri, S., Setola, R., 2011. Fuzzy dynamic input–output inoperability model. Int. J. Crit. Infrastruct. Prot. 4 (3–4):165–175. http://dx.doi.org/10.1016/j.ijcip.2011. 09.003 (December 2011). Ouyang, M., 2014. Review on modeling and simulation of interdependent critical infrastructure systems. Reliab. Eng. Syst. Saf. 121, 43–60 (2014). Pant, R., Barker, K., Zobel, C., 2014. Static and dynamic metrics of economic resilience for interdependent infrastructure and industry sectors. Reliability Engineering System Safety 05/2014; 125, 92–102.

Please cite this article as: Münzberg, T., et al., A spatial-temporal vulnerability assessment to support the building of community resilience against power outage impacts, Technol. Forecast. Soc. Change (2016), http://dx.doi.org/10.1016/j.techfore.2016.11.027

20

T. Münzberg et al. / Technological Forecasting & Social Change xxx (2016) xxx–xxx

Patt, A.G., Schröter, D., Klein, R.L.T., Vega-Leinert, A.C., 2009. Assessing Vulnerability to Global Environmental Change. Earthscan Publications, London, UK. Pederson, P., Dudenhoeffer, D., Permann, M., Hartley, S., 2006. Critical Infrastructure Interdependency Modeling: A Survey of US and International Research. Idaho National Laboratory. Perdikaris, J., Gharabaghi, B., McBean, E., 2011. A methodology for undertaking vulnerability assessments of flood susceptible communities. Int. J. Saf. Secur. Eng. 1 (2): 126–146. http://dx.doi.org/10.2495/SAFE-V1-N2–126-146 (2011). Pescaroli, G., Alexander, D., 2016. Critical infrastructure, panarchies and the vulnerability paths of cascading disasters. Nat. Hazards 82 (1):175–192. http://dx.doi.org/10.1007/ s11069-016-2186-3 (May 2016). Petermann, T., Bradke, H., Lüllmann, A., Poetzsch, M., Riehm, U., 2010. Gefährdung und Verletzbarkeit moderner Gesellschaften – am Beispiel eines großräumigen Ausfalls der Stromversorgung. Arbeitsbericht Nr. 141. Piaszeck, S., Wenzel, L., Wolf, A., 2013. Regional Diversity in the Costs of Electricity Outages: Results for German Counties. Hamburg Institute of International Economics (HWWI) (ISSN 1861-504X, 2013). Praktiknjo, A., 2016. The value of lost load for sectoral load shedding, measures: the German case with 51 sectors. Energies 2016 (9):116. http://dx.doi.org/10.3390/ en9020116. Quarantelli, E.L., 1997. Social aspects of disasters and their relevance to pre-disaster planning. Disasters 1 (1), 98–107 (1977). Rinaldi, S., Peerenboom, J., Kelly, T., 2001. Identifying, understanding and analyzing critical infrastructure interdependencies. IEEE Control. Syst. 21 (6), 11–25 (2001). Riabacke, M., Danielson, M., Ekenberg, L., 2012. State-of-the-Art Prescriptive Criteria Weight Elicitation, Advances in Decision Sciences. Volume 2012:24 http://dx.doi. org/10.1155/2012/276584 Article ID 276584. Rodríguez-Gaviria, E.M., Botero-Fernández, V., 2013. Flood vulnerability assessment: a multiscale, multitemporal and multidisciplinary approach. J. Earth Sci. Eng. 2, 102–108. Rome, E., Bologna, S., Gelenbe, E., Luiijf, E., Masucci, V., 2009. DIESIS: an interoperable European federated simulation network for critical infrastructures. Proceedings of the SISO European Simulation Interoperability Workshop, pp. 139–146. Ryan, M., 2013. Planning in the emergency operations center. Technol. Forecast. Soc. Chang. 80 (2013), 1725–1731. Sahin, O., Mohamed, S., 2010. Coastal vulnerability to sea level rise: a spatio-temporal decision making tool. Proceedings of the 2010 IEEE International Conference on Industrial Engineering and Engineering Management, 29 Oct–31 Oct 2010, Xiamen, China. Setola, R., De Porcellinis, S., Sforna, M., 2009. Critical infrastructure dependency assessment using the input–output inoperability model. Int. J. Crit. Infrastruct. Prot. 2 (4), 170–178 (December 2009). Setola, R., Oliva, G., Conte, F., 2013. Time-varying input-output inoperability model. J. Infrastruct. Syst. 19 (1):47–57 (2012). 10.1061/(ASCE)IS.1943-555X.0000099. Singh, A.K., Ibraheem, I., Khatoon, S., Muazzam, M., Chaturvedi, D.K., 2012. Load forecasting techniques and methodologies: a review. 2nd International Conference on Power, Control and Embedded Systems, Allahabad (978-1-4673-1049-9/12). Sobiech, C., 2013. Agent-based Simulation of Vulnerability Dynamics. A Case Study of the German North Sea Coast, 2013. Springer Theses. Ständige Konferenz für Katastrophenvorsorge und Katastrophenschutz (SKK), 1999. Führung und Leitung im Einsatz. Führungssystem, Köln. Stergiopoulos, G., Kotzanikolaou, P., Theocharidou, M., Lykou, G., Gritzalis, D., 2016. Timebased critical infrastructure dependency analysis for large-scale and cross-sectoral failures. Int. J. Crit. Infrastruct. Prot. 12:46–60. http://dx.doi.org/10.1016/j.ijcip.2015. 12.002 (ISSN 1874-5482, March). Taylor, J.W., McSharry, P.E., 2008. Short-term load forecasting methods: an evaluation based on European data. IEEE Trans. Power Syst. 22 (2213–2219), 2008. Theoharidou, M., Kotzanikolaou, P., Gritzalis, D., 2011. Risk assessment methodology for interdependent critical infrastructures. Int. J. Risk Assess. Manag. Thywissen, K., 2006. Components of risk: a comparative glossary. Publication Series of UNU-EHS, Source 2. Turoff, M., Chumer, M., Van de Walle, B., Yao, X., 2004. The design of a Dynamic Emergency Response Management Information System (DERMIS). J. Inf. Technol. Theory Appl. 5 (4), 1–35 (2004). Union for the Co-ordination of Transmission of Electricity (UTCE), 2007. Final Report System Disturbance on 4 November 2006. Union for the Co-ordination of Transmission of Electricity (UTCE).

United Nations International Strategy for Disaster Risk Reduction (UNISDR), 2009. Terminology on Disaster Risk Reduction (Geneva). United Nations International Strategy for Disaster Risk Reduction (UNISDR), 2015. Sendai Framework for Disaster Risk Reduction 2015–2030, Geneva. first ed. Usov, A., Beyel, C., Rome, E., Beyer, U., Castorini, E., Palazzari, P., Tofani, A., 2010. The DIESIS approach to semantically interoperable federated critical infrastructure simulation. Proceedings of the Second International Conference on Advances in System Simulation, pp. 121–128. Van Eeten, M., Nieuwenhuijs, A., Luiijf, E., Klaver, K., Cruz, E., 2011. The state and the threat of cascading failure across critical infrastructures: the implications of empirical evidence from media incident reports. Public Adm. 89 (2), 381–400. Villagrán de León, J.C., 2006. Vulnerability: a conceptual and methodological review. SOURCE Publication Series of UNU-EHS 4. United Nations University Institute for Environment and Human Security (UNU-EHS), Bonn, Germany. Weber, C., Sailer, K., Katzy, B., 2015. Real-time foresight — preparedness for dynamic networks. Technol. Forecast. Soc. Chang. (2015). 10.1016/j.techfore.2015.05.016. Wong, J.-J., Su, C.-T., Liu, C.-S., Chang, C.-L., 2007. Study on the 729 blackout in the Taiwan power system. Int. J. Electr. Power Energy Syst. 29 (8), 589–599 (October 2007). Yusta, J., Correa, G., Lacal-Arántegui, R., 2011. Methodologies and applications for critical infrastructure protection: state-of-the-art. Energ Policy 39 (10), 6100–6119. Zardari, N.H., Ahmed, K., Shirazi, S.M., Yusop, Z.B., 2015. Weighting methods and their effects on multi-criteria decision making model outcomes in water resources management. Springer Briefs in Water Science and Technology http://dx.doi.org/10.1007/ 978-3-319-12586-2_2. Zobel, C., Khansa, L., 2014. Characterizing multi-event disaster resilience. Computers and Operations Research - special issue on MCDM in Emergency Management 42, 83–94. Zubin, A., Craig Martin, J., Paul, A.M., 2007. Gregory, pharmacy practice in times of civil crisis: the experience of SARS and “the blackout” in Ontario, Canada. Res. Soc. Adm. Pharm. 3 (3), 320–335 (September 2007). Thomas Münzberg is a research associate at the Karlsruhe Institute of Technology (KIT). He studied Engineering with the specification on Disaster Management and Safety Engineering and received his Master of Science at the Technical University of Cologne. His research interests include risk, vulnerability, and resilience analyses and the development of decision support tools to assist disaster management authorities and critical infrastructure providers in managing disruptions of lifelines (esp. power outages and food supply). He is lecturer for risk analysis and disaster preparedness and a researcher at the Center for Disaster Management and Risk Reduction Technology (CEDIM) which is part of the IRDR International Centres of Excellence on Critical Infrastructures and Strategic Planning (IRDR ICoE-CISP). Thomas Münzberg is member of the German Red Cross and has further practical experiences in disaster management as Emergency Medical Technician and Rescue Squad Leader for Medical Services. Marcus Wiens is head of the research group on risk management at the Institute of Industrial Production (IIP) at Karlsruhe Institute of Technology (KIT) and research fellow of the Virginia Polytechnic Institute and State University (VA Tech) in Blacksburg. He studied Economics at the Universities of Bayreuth, Dijon and Saarbrücken and received his Ph.D. in Economics at the Bundeswehr University Munich. His fields of research are systemic risk management, behavioral economics, decision theory, game theory and experimental economics. He is member of the German Operation Research Society and researcher at the Center for Disaster Management and Risk Reduction Technology (CEDIM). Frank Schultmann is Professor at the Karlsruhe Institute of Technology (KIT), Germany, and the Director of the KIT's Institute for Industrial Production (IIP) and the FrenchGerman Institute for Environmental Research (DFIU). He studied Industrial Engineering and Management at the University of Karlsruhe (now KIT), completing his doctorate and Habilitation at the Faculty of Economics and Business Engineering. Prior to his present positions he was Professor at the University of Koblenz-Landau and at the University of Siegen. His research interests include sustainable production and logistics, decision support, supply chain management and optimization, systems modelling, project management, technology assessment and construction management.

Please cite this article as: Münzberg, T., et al., A spatial-temporal vulnerability assessment to support the building of community resilience against power outage impacts, Technol. Forecast. Soc. Change (2016), http://dx.doi.org/10.1016/j.techfore.2016.11.027