Predictive Machine Learning Modelling of Tropsopheric Ozone Using Geospatial and Atmospheric Data

Efficient prediction of tropospheric ozone $\left({O}_{3}\right)$ is crucial to monitor its harmful effects on the ecosystem and thereby take necessary actions. Owing to the complex formation mechanisms and interrelated meteorological parameters, accurate prediction of $O_{3}$ using physics-based model is highly complex. Thus, machine learning techniques are currently being exploited to understand the underlying correlations and develop a good prediction model. This study thus focuses on predicting the long-term ground level ozone using a multitude of geographical features and finally analyzing the contributing factors in the prediction of ozone. To achieve this, various Exploratory Data Analysis (EDA) techniques are performed to visualize the data, followed by data modelling using linear regression, random forest (RF), and artificial neural network models. Finally, different Explainable AI (XAI) techniques are employed to identify the most contributing features in the ozone prediction. The results show that RF has a better performance and that latitude, altitude, NO<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</inf> emissions, population density, and nightlight are the significant features.