Examining Machine-Learning Methods to Develop a Correction Equation for Drone-Mounted PM2.5 Concentrations

Accurate airborne measurement of fine particulate matter (PM$_{2.5}$) is limited by aerodynamic disturbances that affect sensors mounted on unpiloted aerial systems (UASs). These disturbances introduce systematic measurement drift relative to stationary optical particle counters. This study evaluates whether telemetry variables and physics-informed aerodynamic features can be used within machine-learning frameworks to model and correct this drift. Three UAS flights were conducted under consistent conditions, and a suite of linear and nonlinear regression models was trained to predict the residual error of PM2.5 concentrations between airborne and stationary measurements. Modeling the residual produced stable and transferable correction equations, with Ridge Regression achieving R^2 values up to approximately 0.53 under group cross validation. Physics-derived variables provided modest but meaningful improvements, particularly for nonlinear models. These results support a residual-based correction approach as a viable method for aligning UAS-based PM2.5 measurements with ground-level references.