Ozone (O3) is a reactive oxidant that causes chronic effects on human health, vegetation, ecosystems and materials. This study aims to create O3 isopleths in urban and suburban environments, based on machine learning with air quality data collected from 2001 to 2017 at urban (EA) and suburban (CC) monitoring stations from Madrid (Spain). Artificial neural network (ANN) models have powerful fitting performance, describing correctly several complex and nonlinear relationships such as O3 and his precursors (VOC and NOx). Also, ANN learns from the experience provided by data, contrary to mechanistic models based on the fundamental laws of natural sciences. The determined isopleths showed a different behaviour of the VOC-NOx-O3 system compared to the one achieved with a mechanistic model (EKMA curve): e.g. for constant NOx concentrations, O3 concentrations decreased with VOC concentrations in the ANN model. Considering the difficulty to model all the phenomena (and acquired all the required data) that influences O3 concentrations, the statistical models may be a solution to describe this system correctly. The applied methodology is a valuable tool for defining mitigation strategies (control of precursors' emissions) to reduce O3 concentrations. However, as these models are obtained by air quality data, they are not geographical transferable.
Keywords: Artificial neural networks; Isopleths; Ozone; Threshold regression; VOC-NO(x)-O(3) system.
Copyright © 2021 Elsevier B.V. All rights reserved.