OOS 55-4 - Impacts of land use change on the landscape of vector-borne diseases

Humans have pervasively modified land cover and use across the majority of the Earth’s surface for purposes including livelihoods, transportation, resource extraction, and recreation. These intensifying pressures have consequences for human health, including the transmission of infectious diseases. Here, we test the hypothesis that there are distinct land use niches for vector-borne diseases that depend on the ecology of vectors, reservoir hosts, and their contact with susceptible humans. To do so, we use a remotely sensed human footprint index (HFI) to capture multidimensional aspects of land use change: built environments, transportation and energy infrastructure, agriculture, and human density. With extreme gradient boosted regression tree (XGBoost) models, we ask whether the occurrence of six vector-borne diseases—malaria, cutaneous and visceral leishmaniasis, dengue, Zika, and chikungunya—in municipalities in Brazil is predicted by human footprint, along with climate and land cover covariates. We then ask whether the relationship between human footprint and disease occurrence is nonlinear and varies by disease, reflecting the underlying transmission ecology.

We found that human footprint was among the top predictors for all six focal diseases, suggesting that it captures the multidimensional impacts of land use change. The diseases exhibited distinct, sharp thresholds in occurrence across a gradient of human footprint. Malaria and cutaneous and visceral leishmaniasis dropped steeply above thresholds of 5 (marking the transition from intact to agriculture) and 10-11 (intensifying agriculture into intense pressure), respectively. By contrast, dengue, chikungunya, and Zika steeply increased as human footprint crossed thresholds of 8, 11, and 13, respectively, at the transition from intensifying agriculture to intense urban pressure. We then aimed to better understand the relationship between land use and leishmaniasis by expanding on the already large range of known reservoir hosts and vectors using machine learning models. We predicted 136 new mammalian hosts from 12 orders, including many bats and rodents, which tended to occur in high human density, crop cover, urban cover, and temperature and 13 new sandfly vectors, which tended to occur in close association with humans but high forest canopy height, low human modification, and a broad range of ecoregions. Together, these results describe the land use niches for vector-borne diseases to better anticipate how environmental policy change can affect human health.