Towards Resilient Vital Infrastructure Systems : Challenges , 1 Opportunities , and Future Research Agenda

16 Infrastructure systems are inextricably tied to society by providing a variety of vital services. These 17 systems play a fundamental role in reducing the vulnerability of communities and increasing their 18 resilience to natural and human-induced hazards. While diverse definitions of the resilience engineering 19 concept exist for the infrastructures, analysing resilience of these systems within cross sectoral and 20 interdisciplinary perspectives remains limited and fragmented in research and practice. This review 21 synthesizes and complements existing knowledge in designing resilient vital infrastructures with the aim 22 to assist researchers and policy makers by identifying: (1) key conceptual tensions and challenges that 23 arise when designing resilient infrastructure systems; (2) engineering and non-engineering based 24 measures to enhance resilience of the vital infrastructures, including the best recent practices available; 25 and (3) opportunities for future research in this field. Results from a systematic literature review 26 combined with expert interviews are integrated into a conceptual framework in which infrastructures are 27 defined as a conglomeration of interdependent social, ecological, and technical systems. Our results 28 indicate that conceptual and practical challenges in designing resilient infrastructures continue to exist, 29 hence these systems are still being built without taking resilience explicitly into account. A review of 30 available measures and recent applications shows that these measures have not been widely applied in 31 designing different systems. To advance our understanding of the resilience engineering concept for 32 infrastructure systems, main pressing topics to address evolve around the: (i) integration of the combined 33 social, ecological and technical resilience of infrastructure systems, focusing on cascading effects of 34 failures and dependencies across these complex systems; and (ii) development of new technology to 35 identify the factors that create different recovery characteristics for these socio-ecological-technical 36 systems. 37


Introduction
infrastructures that has emerged from an architectural context (Oxman, 2008;Mosalam et al., 2018;165 Hickford et al., 2018). This approach is broadly applied at the design stage (Hickford et al., 2018), and 166 is based on capability of infrastructures to function and perform well in response to an expected 167 pressure or disturbance. The performance-oriented approach, which is also referred to as "control 168 approach" (Hoekstra et al., 2018) or "robust control" (Anderies et al., 2007;Rodriguez et al., 2011), 169 focuses on a system's performance to provide benefits for economic functions. More details on this 170 approach and its application within infrastructure systems is beyond the scope of this study, since this 171 review is grounded on the capacity-oriented (resilience) approach as a different rationale in designing 172 infrastructure systems. 173 174 Capacity-based approach focuses on a system's capacity to adjust its functioning prior to, during, or 175 following changes and disturbances. This approach that has become the dominant discourse in the 176 study of complex systems (Underwood and Waterson, 2013) refers to the resilience approach that 177 examines the capability of systems to recognize and sustainably adapt to unexpected changes (Leveson 178 et al., 2006;Madni and Jackson, 2009;Siegel and Schraagen, 2014;Woods, 2015). Therefore, in the 179 resilience approach the focus is on maximizing capacity of the system to be able to cope with, and 180 adapt to changes and disturbances (Berkes et al., 2003;Folke, 2006). 181 182

183
The emerging concept of resilience engineering within infrastructures (originated from the capacity-184 oriented approach) is one of the main concerns in managing these systems (LRF, 2014; in which 185 complex mechanisms are involved for planning, financing, designing and operating systems (Hickford 186 et al., 2018). There is a wide range of definitions available in the recent literature for the concept of 187 resilience engineering (e.g., Woods, 2015;Sharma et al., 2017;Hollnagel, 2017;Hickford et al., 2018;188 https://doi.org/10.5194/nhess-2020-12 Preprint. Discussion started: 10 March 2020 c Author(s) 2020. CC BY 4.0 License. Gardoni and Murphy, 2018;Bene and Doyen, 2018). These definitions are varied, depending on which 189 aspect of the infrastructure system is under consideration. According to Hickford et al. (2018), while 190 some definitions focused on the ability of the organisations to anticipate the threat and rapidly recover 191 (e.g., Hale and Heijer, 2006), some other studies define the resilience engineering as the ability of the 192 socio-ecological system to absorb changes, and still keep the same function (e.g., Meerow et al., 193 2016). Among the available definitions, and in line with previous studies (i.e., Woods, 2015;194 Hollnagel, 2011;Connelly et al., 2017;Hickford et al., 2018), we distinguish between five 195 principles that are commonly shared within most of the definitions. These principles relate resilience 196 engineering to the ability of the system to: (1) anticipate; (2) absorb; (3) adapt/transform; (4) recover; 197 and (5) learn from prior unforeseen events. These five principles are translated for the infrastructure 198 systems as the system's ability to (i) monitor and anticipate the disruptive events; (ii) function at 199 thresholds of service delivery; (iii) cope with unexpected changes either by its adaptive or 200 transformative capacity; (iv) either return to its normal (steady) condition or re-organize after a 201 depends on how human actors play a role in managing and adapting physical components of the 209 system such as the structure of dikes or embankments. Thus, resilience of the flood protection system 210 relies on the degree to which the system is able to be self-organizing (social resilience), and is capable 211 of increasing its capacity for adapting to changes. Notably, within the social resilience perspective, 212 sustainable governance of the infrastructure systems either through adaptive or transformative 213 approaches plays a pivotal role in enhancing the system's resilience. More details of these two 214 approaches are provided in sections 4 and 5. 215

216
In addition to interaction between social and technical systems, there is also an interplay between 217 physical and ecological systems. From a technical-ecological perspective, infrastructure systems 218 encompass the surrounding built environment (Wolch et al., 2014), and therefore a physical systems' 219 resilience is also related to the natural systems' resilience. Such an interaction with nature highlights 220 the degree to which natural assets (e.g., wetlands ecosystems such as mangroves and urban green 221 areas) can increase the capacity of the whole system to cope with shocks and stresses (ecological 222 resilience). From a socio-ecological perspective, social and ecological systems are also interlinked 223 systems (Adger, 2000). Ecosystems as natural resources, also referred to as "natural infrastructures", 224 provide a variety of services and goods (e.g., flood protection, food provision) that directly or 225 https://doi.org/10.5194/nhess-2020-12 Preprint. Discussion started: 10 March 2020 c Author(s) 2020. CC BY 4.0 License.
indirectly contribute to human well-being (Mehvar et al., 2019a; and, therefore, contribute to the 226 resilience of societies. 227

228
In this article, we define vital infrastructures as a conglomeration of interdependent social, ecological, 229 and technical systems. Within this perspective, a conceptual framework is developed, indicating that 230 resilience of the infrastructures to disturbances depends on the resilience level of each sub-system and 231 the mutual interactions therein (see Figure 1). Notably, applying the resilience engineering concept for 232 designing VIS here does not mean to "engineer" the social and ecological sub-systems, therefore, the 233 socio-ecological aspects are not separately considered than the technical one. This implies that the 234 infrastructure systems are integrated socio-ecological-technical systems, the performance of each sub-235 system has effects on the other one. Thus, this perspective is different than the engineering one in 236 which infrastructures are first of all defined as technical systems.

244
Apart from the inter-relations between the socio-ecological-technical sub-systems, there is also a cross 245 sectoral inter-dependency between different types of VIS (see Figure 2). This cross sectoral relation 246 refers to the mutual effects that function/malfunction of a specific type of VIS may have on other 247 types. Such an inter-dependency is also called "cascading effects" of failure between infrastructures in 248 different sectors. For example, power outage can considerably affect function of transport systems, 249 and other infrastructures, e.g., in the tele-communication sector. This inter-relation is also seen in the 250 https://doi.org/10.5194/nhess-2020-12 Preprint. Discussion started: 10 March 2020 c Author(s) 2020. CC BY 4.0 License. flood protection structures as any failure in these systems may result in sever damages to roads or any 251 other types of infrastructure systems (more details on cascading effects of failure are provided in 252 The inter/cross-sectoral dependencies considered within VIS here are in line with emerging 255 approaches in analysis of VIS resilience such as "system-of-systems" perspective. Such an integrated 256 approach has been used in the recent years to explore the relation between different components of an 257 infrastructure system (e.g., user, physical asset, and network). Using these approaches can also help to  concept of resilience engineering, as well as the applications in which this concept is applied.  The conceptual and practical challenges indicated in Table 1 arise from different components of 289 infrastructure systems, including physical asset, environment, and actor/user, referring to the technical, 290 ecological, and social aspects, respectively (i.e., sub-systems in Figure 1). Figure 3 illustrates the 291 relation of these challenges within these components. This relation is shown through positioning these 292 challenges in the figure depending on whether the challenge arises mostly from a particular 293 component, or is it related to the two/three components. In particular, physical asset here refers to the 294 physical and technical characteristics of the system, environment refers to the natural settings and 295 surrounding of the systems in which a system functions and provides services, and actors/users refers 296 to the policy makers (e.g., government) and users of the infrastructure services (i.e., people). Figure 3  297 shows that most of the challenges are pertaining rather equally to the integration of the three 298 components, while some of them arise mostly from the actors/users of the systems (e.g., units of 299 analysis), or from coupled inter-connections between asset/environment and actor/user (e.g., 300 predicting long term pressures). 301  focused approach, resilience is perceived as the ability of a system to return to its normal (steady) 318 condition after a disturbance (Coaffee, 2013), representing the resilience concept positively (assuming 319 that the normal condition of the system is steady and desirable). However, a system can be resilient, Within such different interpretations, there is also a challenge arising from the resilience engineering 323 concept which is related to the idea of bouncing back (returning to the pre-disaster state). This is in 324 contradiction with the resilience goal of promoting justice among societies (Nagenborg, 2019). 325

Type of challenge Challenge / limitation / debate
According to Nagenborg (2019), understanding resilience and the recovery process as a window of 326 https://doi.org/10.5194/nhess-2020-12 Preprint. Discussion started: 10 March 2020 c Author(s) 2020. CC BY 4.0 License. opportunity (bouncing forward) would promote justice. Of particular relevance here is that poor 327 communities are more vulnerable to shocks, and therefore likely to be less resilient. However, there 328 are cases such as slum areas in which communities have very strong social networks and ties that 329 increase resilience of these groups. Yet, calling communities or individuals "resilient" may be an 330 excuse of not changing anything in the environment. In such a context, resilience can become a 331 concept that promotes conservative, bouncing back-oriented policies (maintaining status quo is the 332 epitome of conservatism). stated that robustness is a characteristic of the control approach that aims to increase safety of the 347 system by resisting to changes and eliminating risks; therefore, it contradicts the resilience approach 348 which refers to responding (adapting) to unexpected changes. Markolf et al. (2018) state that 349 effectiveness of the robustness (also named as control) approach can be reduced due to the current 350 infrastructure-related challenges and pressures such as climate variability and unpredictability, as well 351 as interdependency between the systems. Another reason why robustness cannot be equated with 352 resilience is that robustness only works in situations where disturbances are well-modelled, whereas 353 resilience applies to a set of disturbances that is not well-modelled and that changes (Woods, 2015). provided by an infrastructure system, analysing the system resilience can be done, for example, for an 386 individual (person), team, organization (e.g., company), or society as a whole. Notably, the complexity 387 level increases from a lower (i.e., individual) to a higher (i.e., society) level, and the main challenge is 388 how to connect these levels within a resilient system, given that a system is constrained by a level 389 above and below. 390 391

f) Risk versus resilience 392
In general, risk and resilience concepts are viewed differently. One may consider resilience as a 393 distinct concept from the traditional risk management approach that is used to mitigate or even avoid 394 likely risks. Within this perspective, in resilience engineering, the aim is to become less risk-averse, 395 implying that a certain level of risk is accepted; however, the big question is: what is the acceptable 396 risk? On some accounts, resilience engineering is considered as a related concept to risk management, 397 reflecting the idea that if there is no risk, there is no need to be resilient. Resilience is a function of the 398 present hazard type(s) and their magnitude (which it has in common with risk). Within this 399 perspective, risk assessment including risk identification, prioritization, and mitigation processes is a 400 https://doi.org/10.5194/nhess-2020-12 Preprint. Discussion started: 10 March 2020 c Author(s) 2020. CC BY 4.0 License. basis for designing resilient infrastructure systems, representing risk as an exponent of resilience. 401 However, with respect to the risk and resilience related studies, there is a shift in some terminologies 402 used. For example, in the current literature, the term "resilience" sounds more positive than the 403 traditional term "fault tolerance". 404 405 From a risk assessment perspective, a key question is whether priority should be given to reducing 406 hazard impacts or hazard risks. This dilemma is particularly relevant for infrastructures that aim to 407 protect people against natural hazards. For example, investments in flood protection structures (e.g., 408 dikes, seawalls) in vulnerable coastal areas may help to reduce hazard impacts. However, protective 409 measures may also be counterproductive since they may allude people to move and live closer to the  Table 1 are explored and discussed below. 423 424

g) Design with minimum/maximum capacity 425
Infrastructures are often constructed to their minimum limit/capacity. For example, loading capacity of 426 bridges needs to cope another 100 years, but the systems are frequently designed and constructed to 427 cope to the current load traffic. On the one hand, there is a need to expand roads by using all traffic 428 management approaches to accommodate more cars on the roads; while using the maximum capacity 429 of roads may result in losing natural buffering capacity of the system at the time of a 430 disaster/disruption. As a result, a small disruption in such systems that function with top capacity can 431 propagate immediately throughout the entire system. Therefore, one of the challenges in increasing 432 resilience of VIS is often trade-off between resilience and efficiency of the system as especially 433 prominent in the transport systems. There might also be controversies within social and technical aspects. For example, in the "smart city" 488 initiative which is designed to increase the security of urban areas, it is proposed to place security 489 cameras. But this proposal has its own disadvantages, since such a monitoring system affects people's 490 privacy as they are continuously traced. Therefore, equipping new infrastructures with such tools may, Multi-functionality of the infrastructure systems may increase or decrease the resilience of the system. 508 On the one hand, multi-functionality may decrease resilience of a system, since this characteristic 509 decreases the adaptability of the system to changes because of difficulty of some functions to change 510 in a long run. For example, with respect to the flood protection structures, repairing, re-constructing, 511 https://doi.org/10.5194/nhess-2020-12 Preprint. Discussion started: 10 March 2020 c Author(s) 2020. CC BY 4.0 License. and raising dikes decreases the system's resilience. On the other hand, if an infrastructure system still 512 provides multi-functions after a failure/damage occurs, but different ones than initially aimed for, this 513 system still represents an example of resilient infrastructure, since it adapted to changes while 514 providing different functions. For instance, closure dikes in the Netherlands initially aimed at 515 poldering to create farming area, however the structure led to protection against floods, as well as a 516 fast road transport connecting North Holland and Friesland provinces. Therefore, there might be some 517 resilience hidden anyhow in constructing the infrastructures, since the system might be more resilient In this section, potential opportunities and measures to enhance resilience of VIS are identified. These 567 measures are divided in two categories: (1) Engineering; and 2) Non-engineering, given that proper 568 governance plays a key role in parallel to these measures to ensure that infrastructure services are 569 constantly available to users. Figure 4 shows these opportunities and their linkage to the five main 570 system's capabilities required for a resilient VIS as previously mentioned in section 3.3. 571 572

Engineering-based measures 573 a) Emerging techniques in pre/post disaster anticipation/identification 574
With respect to the pre-disaster anticipation, and preparedness to potential hazards, early warning 575 systems play a pivotal role in raising social awareness, quick evacuation and much lower social 576 disruptions after a disaster occurs. Also remote sensing-based methods that support every aspect of 577 risk assessment, routine surveillance, early warning and event monitoring, have been developed 578 (Kerle, 2015). In terms of post-disaster recovery, automatic and accurate damage identification can be 579 done by first obtaining actionable, accurate, and timely disaster data/information, which is a necessity 580 at the time of disaster. The term "timely" depends on the location and type of devastating event, and 581 can be interpreted in different time scales (e.g., in case of an earthquake in Japan, there are hourly 582 data/information updates). The required data can also be obtained by using space-borne remote 583 sensing, providing satellite images that serve as a basis for an inventory to show the extent of the 584 https://doi.org/10.5194/nhess-2020-12 Preprint. Discussion started: 10 March 2020 c Author(s) 2020. CC BY 4.0 License. affected area and critical hotspots. However, in particular satellite images have been shown to have 585 severe limitations in damage mapping (Kerle, 2010), mainly due to their comparatively limited spatial 586 detail (resolution is at best 30 cm for commercial imagery), but also their vertical perspective that 587 severely limits the damage evidence that can be detected. Damage data can also be provided by 588 drones, which yield more local observations that can be incorporated further in 3D modelling of the De-centralization and detaching physical components of a networked infrastructure is another way of 682 creating resilience for these systems. This measure is often applicable for power supply, thanks to the 683 widespread introduction of renewable energy sources such as wind, solar and biomass (Birkmann et 684 al., 2017). De-centralization is also a solution to promote resilience of the water infrastructures 685 referring to small and medium-sized systems (e.g., wastewater recycling, and rainwater harvesting 686 infrastructure), which rely on locally available water sources (Leigh and Lee, 2019). Notably, all three 687 measures of "redundancy creation", "diversification", and "de-centralization" can contribute to the 688 three system's abilities of absorbing, responding, and recovering. of damage identification models (i.e., fragility curves) for the urban bridges, tunnels, main roads, and 697 metro stations affected by earthquakes to provide a better insight on applicability of these models in 698 seismic vulnerability and resilience assessments. Such damage identification models are extended to 699 damage recovery scenarios to explore the resilience of VIS for a given post-disaster recovery scenario 700 (see Do and Jung, 2018). Enhancing the resilience of the VIS can also be achieved in other ways, e.g., 701 by improving the information flow across organizational levels (from individual to society) and 702 adapting new technology such as social media in order to coordinate data for use (Shittu et al., 2018).

Non-Engineering measures 714 a) Systems thinking -System of systems approach 715
In order to improve infrastructure resilience, a whole system view is required which includes the 716 physical assets, the users and stakeholders (Pearson et al., 2018). Therefore, there should be a holistic 717 approach focusing on the ways that the system's constituent parts interrelate and work over time 718 within larger systems. Infrastructure resilience might be neglected or sacrificed among the users due 719 to lack of having a systems view, which may highlight more immediately recognizable system 720 properties such as sustainability or productivity (Meadows, 2008). Analysis of the infrastructures 721 through a lens of systems thinking/approach provides a better insight towards understanding the 722 system's complexity and interconnectivity which is required to enhance its resilience 723 comprehensively and coherently (Field and Look, 2018). This approach can improve the 724 infrastructure system's ability in terms of better anticipating, absorbing, responding, and recovering 725 from changes at disruptive events. 726

727
The systems thinking perspective is similarly represented by "system-of-systems" approach which Risk assessment is a necessity for designing infrastructure systems within the context of resilience 759 engineering, however opinions are different in terms of the inter-connection between these two 760 concepts (as referred to in section 4.1-f). Risk assessment can be done by using different methods and 761 analysis including fault trees, four-eyes principle, and safe-fail mechanism. These methods provide 762 qualitative metrics highlighting the root causes of the system failure, and quantitative metrics dealing 763 with probability, cost, and impact of a disruption (Kumar and Stoelinga, 2017). For example, the fault 764 tree is a graphical method that models the propagation of failures through the system, investigating the 765 dependability of all components failures, to find out whether or not all failures lead to a system failure 766 (Ruijters and Stoelinga, 2015). Such risk-related methods can improve the ability of a system in 767 monitoring, anticipating, and absorbing disturbances. Risk assessment is more applicable for assessing 768 the high-tech infrastructure systems that are at risk of self-failure, cyber-attacks and human errors 769 (e.g., flood protection systems, power plants, tele-communication equipment). 770

e) "Human-centred design" approach 772
Human-centeredness is a core quality of systems design (van der Bijl-Brouwer and Dorst, 2017). 773 Human-centred design approach presents a framework which aims to empower all the actors, people, 774 stakeholders of an integrated system, by actively involving those who can interact with changes and 775 development processes. Applying this approach as a design and management framework to the 776 infrastructure systems, the technical and social aspects of the system can be integrated with a focus on 777 two goals: 1) To make sure that the human needs are addressed; and 2) To make sure that the 778 framework fulfils its purpose by continuously addressing the human needs in a changing environment. 779 Therefore, using this framework, the system has to adapt to changes and to recover addressing the 780 needs of people (contributing to the system's abilities "respond", and "recover"). Considering this 781 https://doi.org/10.5194/nhess-2020-12 Preprint. Discussion started: 10 March 2020 c Author(s) 2020. CC BY 4.0 License. objective, the resilience concept is already incorporated (as a goal) within this context, while also 782 being linked to the processes to ensure that all stakeholders are involved to achieve the goal. For 783 example, in the transport sector, van den Beukel and van der Voort (2017) conducted a study to assess 784 driver's interaction with partially automated driving systems. This was done by proposing an 785 assessment framework that allows designers to analyse driver-support within different simulated 786 traffic scenarios. 787 788

Governance 789
Governance is a key element of the infrastructure resilience which includes decision making 790 procedures, tools, and monitoring used by governmental organisations and the associated partners to 791 ensure that infrastructure services are available to people (OECD, 2015). For example, preparedness is 792 one of the important approaches to ensure that systems are able to cope with sudden shocks and future power, and tele-communication. In doing so, we include both studies that focus on initial phases of a 808 design process (e.g., assessment or analysis of resilience) as well as studies that design, analyse or 809 evaluate interventions to enhance or increase resilience. Table 2    infrastructure systems (i.e., performance and capacity-oriented) were discussed providing the basis to 848 conceptualize the resilience engineering for VIS. This conceptualization was done by defining VIS as 849 an integrated socio-ecological-technical system, highlighting the inter-sectoral, as well as cross-850 sectoral dependencies within these systems. The inter-sectoral dependency indicated that infrastructure 851 resilience is not only dependent on the technical resilience and engineering characteristics of the 852 system, but also relies considerably on the resilience level of the two other sub-systems (i.e., 853 ecological, and social) and their mutual interactions. The cross-sectoral dependency refers to the 854 mutual effects that function of a specific type of VIS may have effects on other types (as also referred 855 to as cascading effects). and analyzed in relation to the five main abilities required for a resilient system (i.e., anticipate and 870 monitor, absorb, respond, recover and learn from the past). This analysis showed that: (1) engineering-871 based measures (e.g., nature-based, redundancy creation, remote sensing techniques) contribute mostly 872 to the three system's capabilities; absorption, response, and recovery; (2) non-engineering methods 873 (e.g., systems thinking, knowledge sharing and team reflection, human-centered design) highlight 874 mostly the importance of the social aspects of the system, playing an important role in improving 875 system's ability especially in terms of anticipating and monitoring, responding and learning from the 876 previous experiences. Notably, governance and sustained investment can considerably facilitate better 877 implementation of both types of measures, and provide effective measures in promoting all the five 878 system's abilities mentioned above. 879 https://doi.org/10.5194/nhess-2020-12 Preprint. Discussion started: 10 March 2020 c Author(s) 2020. CC BY 4.0 License.
Finally, analysis of the selected 50 recent studies on improving infrastructure resilience resulted in the 880 following main observations: (1) transport systems (often with one mode of transport) and water 881 infrastructures are the most commonly studied systems; (2) knowledge sharing, risk assessment, 882 system-of-systems approach, and nature-based solutions constitute the approaches that are frequently 883 used in the recent applications; (3) natural hazards and climate change impacts represent the major 884 sources of shocks and pressures that have been studied. However, analysis of system resilience due to 885 the disruptions caused by human errors (e.g., accident in transport systems), cyber-attacks, terrorism, 886 and urbanization appears to be less-explored in current literature. It is expected that future standards for designing infrastructures (e.g., flood defences) will become less 904 conservative as soon as resilience thinking and post-disaster recovery of the infrastructures are 905 explicitly considered in the design regulations and decision making procedure. More inclusion of the 906 recovery process in designing and decision making procedure may result in replacing the long-term 907 standards (that may not be well applicable for a sudden shock) into short-term and urgent agreements 908 that can be accepted by both policy makers and stakeholders for better management of a very sudden 909 change/failure in the system. 910

911
There should also be more emphasis on the role of regular maintenance and understanding the 912 performance of the current infrastructure systems, especially the ones that are not supposed to work 913 well (due to their short lifetime), but are still functioning properly, even at the time of a short 914 disruption or big disasters. Therefore, one of the focuses of future studies in designing resilient 915 infrastructures should be on analysis of what worked well in the system rather than only looking at 916 https://doi.org/10.5194/nhess-2020-12 Preprint. Discussion started: 10 March 2020 c Author(s) 2020. CC BY 4.0 License.
what went wrong during a disturbance. Within this perspective, resilience engineering has to take a 917 larger view into account on human errors, but also on human capabilities and regular maintenance of 918 the infrastructure that would increase the efficiency/function of a system in many cases. A cognitive 919 approach that appears to have been less investigated in the current resilience literature, offers an 920 applicable measure for better understanding of this important issue. 921 922 It is also suggested to have a different way of thinking about the resilience of infrastructure systems. 923 Resilience should be considered as a relative quantity, rather than an absolute quantity. Infrastructure 924 systems are better to be designed in a way to become "more resilient", rather than being "resilient". 925 Therefore, instead of setting a threshold to call a system resilient, comparing a system with its 926 previous situation is suggested. In this context, the recovery speed represents a good measure to 927 indicate whether a system is "more resilient" than it used to be. However, the work described in this 928 review also demonstrates a challenge, in that resilience measured on the ground using conventional 929 assessment methods did not always correspond to effective recovery. for future studies on how to make the obtained data useful in identifying the factors that create 935 different recovery characteristics (i.e., quicker/slower, complete/partial). Work is now emerging to 936 couple image-based recovery assessment with macro-economic agent-based modelling that aims at 937 explaining better the observed recovery patterns. If successful this can be used to identify socio-938 economic, as well as legal and political measures to improve the process. Such efforts can provide 939 better insight into the little-known issue of differential impacts and recovery rates across communities, 940 as well as feedback processes and dynamic of the systems after a shock has occurred. This may also 941 serve as a government's tool to find out what are the most significant responsible parameters to inform 942 the success of recovery.