Comparing geospatial interpolation techniques for estimating ambient PM2.5 mass concentrations from sugarcane burning in South Florida.

Abstract

Few studies quantify air pollutant exposures from agricultural burning - a potentially significant source of fine particulate matter (PM_2.5) in rural regions - due to sparse air quality monitoring outside of urban areas. Using a dense network of low-cost sensors, we evaluated several interpolation-based approaches (e.g., inverse distance weighting and kriging) to estimate 24-hour average PM_2.5 exposures from sugarcane burning in the Everglades Agricultural Area (EAA) for future regional health studies. Measurements were taken at 46 unique locations from June 8th, 2024, to December 1st, 2025, capturing the EAA's burn and non-burn seasons. Model performance was evaluated using leave-one-out cross-validation and overall, varied by model and fit, meteorological season, and the presence of smoke, with R² values ranging from 0 to 0.99. Based on the placement of sensors and number of measurements each day, inverse distance weighting models performed best - maintaining the highest R² and the lowest root mean squared error and mean absolute error values for the most days. However, combining approaches, choosing the "best" model each day, provided the most robust exposure estimates for future studies. Independent of interpolation technique, estimated PM_2.5 mass concentrations were highest in the presence of sugarcane burning smoke, increasing 1.65 μg/m³ (standard error = 0.13 μg/m³), and were localized inland, near sugarcane burning regions and the EAA. Notably, PM_2.5 was lowest along the coast - highlighting air quality disparities in South Florida.