Predicción de valores del contaminante atmosférico benceno en Madrid con metodologías de Machine Learning, y análisis mediante georreferenciación de datos

Abstract

Air pollution in large cities represents a global problem for the environment and human beings, affecting the health of people, living beings, plants, degradation of materials, reducing the quality of life and with a social and economic cost. The legislation establishes the need to evaluate the concentration of pollutants, or determine it through models when measurement is not possible. The prediction through mathematical models and the relationship with other pollutants allowsto anticipate episodes of high pollution and thusthe authorities can make decisions, plan and establish prevention and control measures to try to improve air quality and protect health and well‐being of people. This thesis analyzes the data collected by the air quality monitoring networks of the stations of the Community and of the Madrid City Council from 2001 to 2020, studying the evolution of benzene over the years, its relationship with other pollutants and its prediction, at the local level. The data collected by 8 stations during different time periods are studied using regression‐based machine learning models and univariate and multivariate time series models, establishing the relationship between benzene and nitrogen oxides, nitrogen dioxide, toluene and particulate material smaller than ten microns and the prediction of benzene concentration from observations of the those latter. The relationship of benzene with the reference pollutants is also analyzed for 8 years in one of the stations and predictions of the benzene concentration are made from them. For each of the established models and for each station, the results are compared through performance measures. Using a geographic information system, the spatial distribution of the benzene concentration in the municipality of Madrid is studied from the observations recorded at its measurement stations, in different periods and time bands.