Influence of Temperature and air Pollution on Provincial Mortality in Spain: A Seasonal AI Approach

In this study, we analyze weekly mortality series at the provincial level for all seasons of the year. Using an artificial intelligence model, we examine the importance of various meteorological and atmospheric pollution variables on the temporal variability of mortality. The variables analyzed include maximum and minimum temperatures (Aemet ROCIO 5km), concentrations of ozone, PM10, PM25, and NO2 (CAMS 10km) for both the concurrent week and the previous week. To obtain the provincial series, we developed a methodology that involves calculating the weighted average of all grid points within the province, weighted by the population of the covered area.

In summer, the results show that the artificial intelligence model can reasonably explain the variability in mortality in many Spanish provinces, especially in large cities such as Madrid and Barcelona, where the temporal correlations between predicted and observed values exceed 0.6. The minimum temperature is the most important variable in most provinces, followed by the maximum temperature for both the current and previous weeks. The results indicate that atmospheric pollution plays a significant role. However, there is considerable interprovincial variability. Analyzing extreme events reveals that, in most cases, temperature extremes coincide with atmospheric pollution episodes.

In winter, the most significant variable is the minimum temperature of the previous week in almost all provinces, with correlation indices above 0.5. Regarding extremes, we again find coincidences of several factors, especially at the beginning of major mortality episodes, with the most notable being high levels of nitrogen oxides and particulate matter.

We conclude that the proposed methodology is capable of reasonably explaining the temporal variability of mortality and that a significant portion of extreme mortality events have a composite nature (compound events), where the coincidence and succession of several factors seems to play a determining role.