Meteorological forecasting of respiratory diseases incidence with geospatial and machine learning methods

Weather conditions play a significant role in shaping human health, particularly in the incidence of respiratory diseases. This study proposes a workflow to predict hospital admissions related to respiratory conditions by integrating heterogeneous environmental and health data sources. A 15-year time series of meteorological data was combined with geographic and demographic information to derive biometeorological indices associated with hypoxia and hyperoxia. To account for spatial variability, three clustering techniques—k-means, Local Indicators of Spatial Association (LISA), and Spatial ’K’luster Analysis by Tree Edge Removal (SKATER)—were applied to define intra-provincial regions for targeted data extraction from raster models. These approaches were compared with conventional aggregation at the provincial level. Interpretable machine learning models, specifically tree-based algorithms, were used to evaluate predictive performance. The proposed spatial clustering strategies achieved comparable or improved performance relative to the baseline approach. Experiments conducted in three Spanish provinces (Cantabria, Madrid, and Valencia) yielded mean absolute errors ranging from approximately 5.4 to 8.2 cases per million inhabitants, depending on the province. Results highlight the relevance of spatially refined environmental data in improving predictive accuracy. The integration of geospatial data, meteorological models, and machine learning provides an effective framework for forecasting respiratory disease incidence. This approach supports the development of early warning systems and risk management strategies based on environmental monitoring. In this article we present how to gather, transform and model datasets from several data sources to predict hospital admissions of respiratory diseases. We also draw the importance of geographical distribution of population in order to determine the appropriate method to extract meteorological data.