Predicting Ground Level Ozone in Marrakesh by Machine-Learning Techniques

Abstract

This study was undertaken to produce local, short-term, artificial intelligence-based models that estimate the ozone level with special attention to the relationship between diurnal and nocturnal ozone variations of some primary pollutants and meteorological parameters in the city of Marrakesh, Morocco. Hourly data has been collected from the three air-quality monitoring stations in the city. This paper seeks to analyze the main factors that are associated with ozone formation, including the generation of different daytime and nighttime scenarios. The present work extends existing publications about the region by developing ozone prediction models from meteorological variables and primary pollutants. Several experiments were conducted to verify properties of the produced models, thus making it possible not only to describe but also to predict ozone pollution in this geographical area. The findings facilitate 48 hour forecasts that have root mean square errors as low as 20 g/m3. Our results highlight the importance of using such models for civil applications.

Keywords: ozone pollution, ozone diurnal concentration, ozone nocturnal concentration, machine learning, nonlinear models, Marrakesh