Societal responses to a changing climate: insights from machine learning methodologies

Climate change poses a unique and vital challenge to humanity. The pressures of climate change, partially exemplified through extreme weather events, have become larger in recent years, and are expected to continue to increase. It has therefore become imperative to understand the societal responses to these pressures and sequelae of climate change, as well as the mechanisms underlying these responses, from the global, to regional and local levels. As both climate change and the nature of societal response to it are complex, the integration of both established and innovative empirical methods is required to broaden and deepen our understanding of these phenomena. This dissertation aims to advance our understanding of societal responses to climate change by answering the following research question: How might we better understand how humans have heretofore responded to the diverse changes incurred by climate change through the use of integrated statistical, and machine learning methods? These response dynamics are examined through concrete examples related to the sequelae of climate change: the quantification of global economic e!ects to economic growth incurred via wildfires, the dynamics of recovery from hurricanes in the southern US, and the city-level contagion of climate mitigative behaviors through the spread of solar panel implementation. The methodological contributions of this thesis are woven throughout the three articles contained herein. Each article integrates the application of statistical and machine learning methods to satellite data towards providing a rigorous, granular analysis of the phenomena investigated. In several instances, these methods allow for a uniquely granular temporal or geographic perspective on the topic of study, or are able to provide deeper insight into the mechanisms underlying the plausibly causal relationships investigated. Each article of this thesis addresses societal response to climate change on a di!erent spatial scale. Beginning at the global level, Chapter 2 examines the temporal evolution of the response to fires through economic growth rates. Uniquely combining iterative statistical and machine learning methods, this study finds that the direction of the e!ects of fire on economic growth have changed over the time period of study, with data from more recent years indicating that fires are associated with declines in the economic growth rate. Moving from the global to the regional, Chapter 3 addresses the dynamics of recovery after hurricanes in the southern US, using nighttime light radiance levels as an index of economic activity and population density. Again matching statistical and machine learning methodologies, this study uncovers those factors most important in determining regional recovery from hurricanes (or lack thereof), finding not only that aid is instrumental in determining short-run recovery, but that it is unable to motivate more sustained recovery in the long term. Chapter 4 examines societal response to climate change on a micro scale of analysis, the local (household) level. Data on the locations of solar panels derived from satellite imagery are matched with machine learning methodologies to provide insight into those factors most important in determining the likelihood of a household having implemented solar panels. We find that the likelihood of having a solar panel is best predicted by proximity to an already installed solar panel, outweighing all demographic and socioeconomic factors. Finally, Chapter 5 discusses the relevance of the articles presented in this dissertation, particularly their implications for the future management of climate impacts. It concludes by discussing potential limitations of the findings of this dissertation, before suggesting potential areas for future work.