Determination of influential variables and mapping of PM2.5 regional distribution using Neural Networks and MERRA-2 re-analysis model

Routine monitoring of atmospheric particulate matter with aerodynamic diameters less than or equal to 2.5 μm (PM2.5) is relevant because of their potential negative impacts on human health and the environment. These air pollutants can be accurately measured using ground, mass-based devices. However, the limited availability and spatial coverage of ground-monitoring networks restrict their use for the analysis of pollution at local and regional scales. The use of ground-based measurements combined with satellite and reanalysis models data is an excellent tool to analyze air pollution events on a regional scale. In this study, a Neural Network Model (NN) that relates Aerosol Optical Depth (AOD) and meteorological variables from the MERRA-2-reanalysis model was developed for the estimation and mapping of PM2.5 concentration distribution in Northeastern Mexico. The NN was calibrated using ground-based PM2.5 concentrations from the Monterrey Metropolitan Area (MMA), the main urban area in the region, for a period covering 2010 to 2014. AOD, temperature, and relative humidity, along PM2.5 components (dust, sea salt, black carbon, organic carbon, and sulfate) from MERRA-2 reanalysis model were found to be factors that significantly influence the regional distribution of PM2.5. Monthly-averaged PM2.5 concentrations were mapped, with maximum levels located over the MMA. The largest PM2.5 concentrations occurred in winter and spring, while moderate concentrations were observed in autumn and low concentration in summer. In conclusion, MERRA-2 reanalysis data can be used to study atmospheric pollution distribution at regional and local scales. The magnitude of PM2.5 pollution events of PM2.5 varies in space and time, showing large seasonality and heterogeneous distribution across the region. The methodology developed and results obtained here can be used to improve air quality management in places with few or no ground-based measurements.