Pave, Rain, Leptospira Gain - The Impact of Climate and Land‑Use Change on Leptospirosis in Brazil 2007–2023

As proven by COVID-19 and the Spanish Flu of 1918, the architecture of international security is not merely affected by a state’s behavior and war, but also pathogens, environmental change, and the invisible forces that destabilize a society from within. Rapid urban expansion and a changing climate are reshaping the global map of zoonotic disease risk—turning once-remote pathogens, as Ebola did when it leapt from forest village to crowded cities during the 2014 West African outbreak, into sudden regional and international crises—and the implications extend from local health systems to international security. Focusing on leptospirosis—a zoonotic, flood-linked disease concentrated in urban, low-income areas— this thesis traces how Brazil’s land-use and climate dynamics from 2007 to 2023 have altered where, when, and whom the disease strikes. Leveraging monthly municipality data on environmental and land-use variables and reported cases, I first develop a high-performing XGBoost machine learning model with high predictive accuracy that generalizes well across space and time (mean AUC = 0.882). Then, using model interpretation techniques, I show that growth in urban area is the single most powerful driver of disease risk, followed by other land-use variables; precipitation and temperature exert smaller, nonlinear effects. Building on these insights, I quantify the evolving health burden: Disability-Adjusted Life Years have risen fastest in municipalities undergoing recent urbanization, signaling that the economic and social costs of leptospirosis are shifting toward emerging urban peripheries. A counterfactual simulation shows that Brazil’s post-2007 urban growth alone is responsible for a substantial share (16.7\%) of the predicted case load for 2023. Combined, these findings demonstrate that unchecked land-use change can amplify climate-sensitive zoonoses, deepen inequality, strain urban infrastructure, and create latent security risks that transcend national borders. By combining machine-learning prediction with policy-relevant burden estimates, the thesis provides both a methodological prototype and an evidence base for proactive urban planning and health-system preparedness in an era of accelerating environmental change.