Improving UAV-based air quality measurements through machine learning calibration of downwash effects

Air pollution is a major global health risk requiring high-resolution monitoring for effective management. Unmanned Aerial Vehicle (UAV) monitoring systems with Low-Cost Sensors (LCS) have attracted significant attention as an efficient approach for three-dimensional, high-resolution air quality monitoring. However, the downwash generated by UAV propellers induces airflow disturbances, leading to measurement errors. This study aimed to quantitatively analyze the downwash effect during hovering and to develop a Machine Learning (ML)-based algorithm for effective calibration. To achieve this, nine algorithms were evaluated, and meteorological and seasonal variables were integrated to capture complex nonlinear relationships. The results show that, in pollutant-specific analysis, fine particulate matter (PM 2.5 ) maintained strong agreement even within the downwash-affected region, indicating that additional calibration is unnecessary. In contrast, gaseous nitrogen dioxide (NO 2 ) exhibited clearer disturbance due to turbulence and dilution effects. Among the evaluated algorithms, a tree-based ensemble algorithm showed the lowest numerical error, reducing the NO 2 error to a level comparable to inter-sensor variability. These findings suggest that tree-based calibration methods can improve the reliability of high-resolution air quality data obtained from UAV monitoring systems.