Developing Advanced PM2.5 Exposure Models in Lima, Peru

Abstract

Background: There is convincing evidence of adverse health effects induced by exposure to PM2.5 in the growing body of literature. Lima’s topography and aging vehicular fleet results in severe air pollution with limited amounts of monitors to effectively quantify measurements for epidemiologic studies.Objectives: We propose to develop a high-performance satellite-driving exposure model to estimate daily PM2.5 concentrations at a 1 km spatial resolution in Lima, Peru from 2010 to 2016 using a combination of ground measurements, aerosol optical depth (AOD), meteorological fields, parameters from atmospheric chemical transport models, and land use variables.Methods: Parameters from the Weather Research and Forecasting model coupled with Chemistry (WRF-CHEM) and the European Centre for Medium-Range Weather Forecasts (ECMWF) were evaluated against ground monitoring stations from Weather Underground. A random forest model was used to gap-fill non-random missing satellite AOD data due to cloud cover to enhance spatial coverage and quality. Both a linear mixed effects model and a random forest model was used to fit AOD, WRF-CHEM, ECMWF, meteorological fields, and land use parameters against ground measurements from 16 monitoring stations with available data between 2014 to 2016. Both models were then used to estimate daily PM2.5 concentrations from 2010 to 2016.Results: The overall cross-validation (CV) R2 value and (RMSE) for the linear mixed effects model and random forest model was 0.58 (7.08 μg/m3) and 0.73 (5.66 μg/m3), respectively. The random forest model’s robust ability to include more parameters outperformed the linear mixed effects model due to limited number of ground observations.Conclusions: Our models allow for construction of long-term historical daily PM2.5 levels to support fundamental and imperative epidemiological studies that will likely impact governmental policies on air quality in Lima, Peru.