ANALYTICAL METHODS FOR COMPUTING THE RESILIENCE, RECOVERY, AND TRANSFORMATION OF COMMUNITIES AND THEIR CONSTITUENT SYSTEMS IN THE AGE OF BIG DATA
Communities are increasingly vulnerable to climatic risks which are estimated to cost $1.8 trillion and lead to 2 million deaths annually by the end of the century. To minimize this vulnerability in the face of the increasing climatic risks, resilience is used as an organizing principle by all scale of governments, decision makers, and international organizations to address climatic risks. Resilience is conceptualized across many fields and is broadly meant to represent the ability of a system to maintain critical functionality, adapt, and ‘bounce back’ after a shock or disruption.
Moving from theoretical conceptualizations of resilience to operational decisions which aim to foster adaptive capacity in communities, requires consideration of the dynamics of engineered, social, ecological, economic, and political systems among others. This dissertation develops analytical techniques to leverage ‘big data’ to understand the multifaceted aspects of how communities and engineered systems are impacted by and recover from major disruptions in an effort to bridge the gap between resilience in theory and resilience in practice.
In the light of the disciplinary variations in conceptualization and operationalization of resilience, the introduction to this dissertation begins by unpacking the myriad of resilience definitions and how they relate to communities and engineered systems; describing analytical techniques which are used to model and quantify communities and engineered systems.
Chapters 2-5 summarize the articles included as a component of the dissertation. First (Chapter 2) I analyze the characteristics of large-scale disruptions in network-based infrastructure systems. There is a large body of work which utilizes graph-theoretic representations of engineered systems to model resilience to shocks. However, the way by which shocks or disruptions are simulated in the system are either based on random failures –indicative of component aging– or targeted failures –based on an intentional threat like terrorism– and do not reflect the explicit spatial structure of natural hazards. To address this gap, I propose two methods for generating failures in network based infrastructure models which have a connected, spatial structure similar to that of a large-scale natural disaster such as a hurricane. When evaluating the performance of the system after a disruption using network-based performance metrics, the networks with spatially-distributed outages show statistically different measures of performance compared with similarly sized randomly-distributed outages. Additionally, when simulating the recovery of the system; the spatial characteristics of the outages drastically alter the way in which the network recovers. Of note, systems disrupted with random outages showed antifragile properties, while spatially-distributed outages do not. This work is extended to interdependent infrastructure systems in Chapter 3.In Chapter 4, I contribute to the nascent literature on harnessing social media data for resilience analytics. Specifically, I develop algorithms for analyzing how community members perceive the dynamics of their community during a crisis event, using twitter data during 14 major crises events. Grounded in theories of community resilience and sociological risk appraisal, these algorithms —called the Social Resilience Fingerprint— capture the patterns of discourse in communities related to the attributes of communities which contribute to its resilience, such as infrastructure, economic, and ecological systems. Using this framework, I show how different types of major disruptions (hurricanes, earthquakes, political events etc) have signatures identifiable in social media data and discuss the trends driving these similarities.
Finally, in Chapter 5 I formulate machine learning methods for evaluating the potential of communities to transform after major disruptions. The current paradigm of community resilience modeling aims to rapidly return to normal-operation following a disruption. By promoting the status quo, however, this modeling technique may be counteracting itself by reinforcing persistent, maladaptive states which inhibit the ability of communities to grow and transform. With this gap in mind, I have developed an alternative method for measuring community resilience, termed a Contrastive Community Network (CNN), which identifies key drivers of community transformation and quantifies how communities reorganize after major disruptions into alternative, stable equilibria. Using this improved methodology, I identify resilience traps: risk factors which, while critical for rapid recovery to the status quo, do not allow for any possibility of transformation and long-term adaptation. These traps clearly demonstrate some of the pitfalls present in current methodologies for quantifying community resilience.
The methodologies and algorithms developed in this dissertation can improve the ability of stakeholders and decision makers to understand and analyze how communities adapt and respond to major crisis events, allowing for data-driven decisions to be made to bolster the resilience of communities in response to climate change.