The extent to which climate change will exacerbate already growing inequality, both across and within locations, is an open question. The power that people have, or lack, to adapt to a warming climate is central to this issue. This dissertation includes three studies on potential climate adaptations. In the first, my coauthors and I study whether heat-induced use of energy-intensive adaptation technologies, such as air conditioning, can lead to financial distress. We test for this possibility with data from California on electricity use and disconnections, a consequence of utility bill non-payment. We find that hot weather increases electricity expenses in the current billing period and the relative risk of disconnection 51 to 75 days later. In the second study, my coauthors and I examine the impact of temperature-related crop losses on household diets in rural India, a setting with a high prevalence of small family farms. High temperatures during the growing season reduce crop yields, but it is unclear how these losses affect household diets. While we find no significant impact of heat on average calorie or iron consumption in the subsequent year, the number of extremely malnourished households increases. We also find suggestive evidence that households adapt to heat-induced losses of home-grown calories by purchasing more food, which helps to explain the lack of aggregate impacts. In the third study, I analyze worker responses to climate change. Past work in this area has largely focused on the on-the-job effects of extreme weather on environmentally exposed workers and potential short-term adaptations. Little is known about long-term adaptations, such as changing occupations. Well-documented challenges to occupational mobility, especially between occupations with different task requirements, suggest that this adaptation strategy may be prohibitively costly for many workers. Using individual-level panel data from France, I find that historically, inter-exposure mobility rates are low. The task composition of high exposure jobs provides a partial, but incomplete, explanation for this labor market segmentation.

With a growing population and climate-related disasters projected to become more severe, the need to anticipate and better prepare for food crises has never been higher. Status quo early warning systems perform well, by and large, but complex weather phenomena are a major source of uncertainty. In this dissertation, I explore whether statistical diagnostics associated with tipping point theory can be used to improve detection of major crises.

First, the accuracy of early warning systems is evaluated – with a focus on the Greater Horn of Africa. I discuss geographical disparities in early warning skill, and explore the various sources of uncertainty that impact food security projection accuracy. I quantify the contribution of climate forecast skill and conflict events in overall early warning skill to use as a baseline upon which to improve.

Second, the potential applications of tipping point theory are discussed in relation to the use of remotely-sensed environmental variables for food security early warning. I evaluate the different statistical diagnostics that can be applied to identify food security tipping points and the data requirements in terms of spatial and temporal resolutions, data availability, latency and geographic coverage in order to be maximally useful for early warning analysis.

Finally, I test whether statistical diagnostics of tipping points can be used for detection of food security transitions. Through this work, I demonstrate the utility of combining soil moisture measurements from the SMAP mission with food price data for identifying transitions, with a lead time of approximately 3 months. The model presented here was used to detect all major food security transitions (both towards and away from a crisis) in the observational record of soil moisture. Importantly, the diagnostics are also correlated with the magnitude of change in food security conditions which is critical information for food security planning.

Overall, the dissertation advances the science and capabilities of food security early warning, which will lead to an improvement in saving lives and resources.

The dissertation presents three papers examining exposure to extreme heat and wildfire in the Western United States. In the first chapter, I develop a framework for analyzing transit passenger exposure to extreme heat in Maricopa County and then implement an optimization algorithm for minimizing wait times through the reallocation of buses across the transit network. In simulating the reconfiguration of buses, I find the potential for small adjustments to produce large reductions in wait time for vulnerable populations. This work also formulates a way to measure passenger vulnerability with an activity-based model that accounts for the distinct demographics of transit riders.

In the second chapter, I study the prevalence of ground level wildfire smoke, specifically particulate matter 2.5�m in diameter, in California during the 2020 wildfire season - the most severe wildfire season ever recorded by the state. For the first time, I study how frequently extreme smoke levels at surface level coincide with extreme heat in space and time. These interactions can influence adaptive behaviors and studies show evidence of increased hospitalizations when these hazards co-occur. I find that a majority of Californians experienced at least one day of concurrent heat and smoke in 2020 and that these events were concentrated in more rural areas of the State. This case study motivates the integration of multi-hazard frameworks in both public and private sector risk planning.

In the final chapter, I examine wildfire risk factors for residential property in California. I leverage a dataset collected by CAL FIRE enumerators who record the features of a home and categorize the level of damage after every named incident. I enhance this dataset using remotely sensed detections of wildfire to impute the date when a home burned from which I then estimate time-varying weather risk factors like humidity, temperature and wind as well as fire intensity. I then use these features to train a predictive model to be used by homeowners or insurance carriers to better estimate the vulnerability of their property.