For several decades now, the study of environmental impacts on human well-being has been informed by what are called "People and Pixels'' methods: the combining of remotely sensed data about environmental conditions with geolocated data from household surveys about health and nutrition. However, much of this work has been conducted at the scale of individual countries and often relies on only one or two survey waves, which creates substantial issues around spatial autocorrelation and endogeneity. Furthermore, much of this work uses simple linear regression as its analysis technique, which is limited in its ability to describe spatial variation as well as non-linearities in the relationship between the environment and human well-being. Thus, this dissertation uses several insights from the emerging field of data science to advance these methods. First, this analysis draws on large, multinational datasets from dozens of surveys, making it possible to better estimate the non-linear effects of climate extremes on human well-being as well as examine spatial heterogeneities in vulnerability. Secondly, this analysis uses techniques at the boundary between traditional econometric regression models and more complex machine learning models, such as using Generalized Additive Models (GAMs) as well as LASSO estimation. This permits the creation of spatially-varying terms as well as nonlinear effects. Applying these techniques, the dissertation has yielded several insights that could be beneficial to policymakers in governments, non-profits, and multinational organizations. The initial chapters analyze the effects of rainfall anomalies on food security and malnutrition, finding that the effect of an anomaly varies considerably depending on the local socioeconomic and environmental contexts, with low-income, poorly-governed, and arid countries, such as Somalia and Yemen, being the most vulnerable. The latter chapters look at the role of ecosystem services in improving human livelihoods, as well as how land cover is associated with dependence on local provisioning ecosystem services.