Deriving ozone and PM pollution vertical profiles using lightweight, cost-effective sensors and deep learning npj Climate and Atmospheric Science

We developed and deployed a drone-based air pollution measurement system composed of cost-effective and lightweight sensors. The system generates high-resolution vertical profiles of various pollutants. During campaigns conducted in 2023, we observed a diurnal cycle of ozone and analyzed extreme particulate matter events, including biomass burning and a rapid dust storm. Our analysis reveals consistent ozone depletion near the surface at night, an advection-related “knee” in the ozone vertical profile at ~100 meters, and significant differences in aerosol size distributions between background, biomass burning, and dust events. An ensemble of autoencoder-based deep learning models with prediction heads identified ground data and a novel combined factor as the most predictive variables for the ozone vertical profiles. These findings demonstrate the value of mobile vertical profiling systems for understanding pollutant distributions and tropospheric dynamics, including the distinction between local and regional ozone influences, with potential applications for air quality monitoring.