Abstract: Since the 19th century, environmental aspects have been the subject of several studies, and initially, only natural phenomena were characterized. From this perspective, climate change was not considered responsible for the spread of diseases. However, after the advent of Biometeorology in the 1930s, these concepts began to be rethought so that increasingly, the relationship between the physical and chemical variables of micro and macro environments, atmospheres, and the living organisms characteristic of each region, whether animals and plants. Consequently, it is now possible to state that there is currently a consensus that many existing diseases present specific interactions with the environment in which they are inserted, often due to the life cycle of vectors or the interaction of human beings with the environment, since the climate impacts human health through three primary means: extreme weather events, changes in the environment (modifying predominant health factors), and social processes affected by climate conditions. Brazil has a complex regional diversity regarding the spatial and temporal distributions of specific morbidities directly associated with the country’s considerable cultural, climatic, and ecological diversity, specifically affecting individuals and social groups when exposed to particular climatic changes. Therefore, national public health requires analyses based on different social, environmental, and economic perspectives. Although the Brazilian Unified Health System (SUS) has a national database (DataSUS), it lacks information that makes it possible to understand the dynamics of morbidities in different regions, especially those associated with environmental degradation, such as respiratory diseases. Furthermore, the poorest areas and, consequently, most exposed to climate vulnerability usually suffer from a lack of data and, consequently, possible underreporting. Within this scenario, the project entitled: “Application of Deep Learning in Landscape Epidemiology using DataSUS data for spatial prediction of endemic and chronic non- communicable diseases in Brazil: CAIPORA” will develop, test and validate a predictive model that, based on the time series of meteorological variables, concentration of pollutants and social indicators is capable of predicting, for all Brazilian capitals, cases of mortality and hospital admissions due to cardiovascular, cerebrovascular and respiratory diseases, as well as hospitalizations of patients with mental illnesses, registered in the various database systems of the SUS IT Department (DATASUS), taking ICD-10 as a reference. The project involves two stages: first, analyzing data and creating a predictive model with Artificial Neural Networks (ANN) for São Paulo, and second, replicating the methodology in other Brazilian capitals using transfer learning. The initial focus on São Paulo leverages its abundant data to ensure a robust ANN training, which will then serve as a reference for achieving reliable results in regions with less data.

Keywords: Air Pollution, Human Health, Machine Learning, Landscape Epidemiology, Climate change.