Receive a weekly summary and discussion of the top papers of the week by leading researchers in the field.

In Chemosphere

Modelling photochemical pollutants, such as ground level ozone (O3), nitric oxide (NO) and nitrogen dioxide (NO2), in urban terrain was proven to be cardinal, chronophagous and complex. We built linear regression and random forest regression models using 4-years (2015-2018; hourly-averaged) observations for forecasting O3, NO and NO2 levels for two scenarios (1-month prediction (for January 2019) and 1-year prediction (for 2019)) - with and without the impact of meteorology. These flexible models have been developed for, both, localised (site-specific models) and combined (indicative of city-level) cases. Both models were aided with machine learning, to reduce their time-intensity compared to models built over high-performance computing. O3 prediction performance of linear regression model at the city level, under both cases of meteorological consideration, was found to be significantly poor. However, the site-specific model with meteorology performed satisfactorily (r = 0.87; RK Puram site). Further, during testing, linear regression models (site-specific and combined) for NO and NO2 with meteorology, show a slight improvement in their prediction accuracies when compared to the corresponding equivalent linear models without meteorology. Random forest regression with meteorology performed satisfactorily for indicative city-level NO (r = 0.90), NO2 (r = 0.89) and O3 (r = 0.85). In both regression techniques, increased uncertainty in modelling O3 is attributed to it being a secondary pollutant, non-linear dependency on NOx, VOCs, CO, radicals, and micro-climatic meteorological parameters. Analysis of importance among various precursors and meteorology have also been computed. The study holistically concludes that site-specific models with meteorology perform satisfactorily for both linear regression and random forest regression.

Shukla Komal, Dadheech Nikhil, Kumar Prashant, Khare Mukesh


Linear regression, NO and NO(2), O(3), Pollutant forecasting, Polynomial transformation, Random forest regression