Ground-level ozone forecast based on machine learning

In this paper we apply methods of machine learning to the problem of ground-level ozone forecasting, using measured data and data calculated by the numerical weather prediction model ALADIN (Aire Limitee Adaptation Dynamique developement InterNational). Our goal is to build a simple ozone-forecasting tool to predict the daily maximum of ground-level ozone concentration per day using meteorological and air quality data. Tropospheric ozone episodes in Slovenia are mainly due to the local traffic sources and the long-range transport of ozone and its precursors, generally originating from Western Europe. Ozone forecast model was developed for the purpose of issuing public alerts to avoid exposure to high ground-level ozone levels and so that concerned citizens, industrial organizations and local authorities could take action to reduce harmful emissions of ozone precursors. The results enable an empirical approach to the short term ozone level forecasts, estimating the ozone trends and increasing the scientific understanding of the underlying mechanisms.