Machine learning based quantification of VOC contribution in surface ozone prediction.

Area of Science:

• Environmental Science
• Atmospheric Chemistry
• Data Science

Background:

• Surface ozone poses significant risks to human and environmental health.
• Volatile organic compounds (VOCs) are key drivers of ozone formation, particularly in urban settings.
• Limited VOC measurements impede understanding the VOC-ozone relationship.

Purpose of the Study:

• To apply machine learning for accurate temporal forecasting of surface ozone in an Indian metropolitan city.
• To identify the most influential parameters for surface ozone prediction.
• To evaluate model performance under different input data scenarios.

Main Methods:

• Utilized machine learning algorithms for temporal forecasting of surface ozone.
• Analyzed continuous data on VOCs, meteorology, and other pollutants (2014-2016).
• Performed simulations with varied input combinations to determine optimal predictor sets.

Main Results:

• The optimal model combined isoprene, meteorology, nitrogen oxides (NOx), and carbon monoxide (CO) (Isop + MNC), achieving RMSE of 4.41 ppbv and MAPE of 6.77%.
• This Isop + MNC model demonstrated superior performance during the summer season (RMSE: 5.86 ppbv, MAPE: 7.05%), effectively capturing high ozone events.
• Using all available data did not necessarily yield the best prediction outcomes.

Conclusions:

• Machine learning models, particularly with specific VOC and pollutant inputs, can effectively predict surface ozone levels.
• Isoprene data is crucial for accurately forecasting ozone peaks, especially during high-ozone seasons like summer.
• Careful selection of input parameters and critical evaluation of model results are essential for reliable ozone prediction.