Analysis of real-time particulate matter (PM2.5) concentrations in Washington, DC, using generalized additive models (GAMs)

The objective of this paper is to report the results of a generalized additive model used to predict local particulate matter concentrations at a Washington, DC Department of Energy and Environment (DOEE) federal regulatory monitoring station. While the DOEE uses state-of-the-art federal equivalent method (FEM) equipment to demonstrate compliance with the clean air act for regulatory purposes, these measurements reflect regional, not neighborhood air quality. A GW student-led living lab project—Fresh Air DC—has been testing uRAD INDUSTRIAL low-cost air quality sensors that can be used to collect air quality data at the neighborhood level using LoRaWAN based smart city technology. Because low-cost sensors often lack the accuracy and sensitivity of FEM equipment, research indicates that low-cost sensor (LCS) monitoring networks require post- processing and data modelling in order to apply findings to educational and policy goals. Although LCS data processing has been conducted using linear and nonlinear models, nonlinear models tend to have a greater ability to capture the nuanced relationships between air pollutants and meteorological influences. In this paper, we post-process uRAD PM 2.5 sensor data using DOEE FEM equipment as a reference instrument in the development of three models to adjust uRAD data to the DOEE FEM data—ordinary least squares linear regression, generalized linear models (GLMs), and generalized additive models (GAMs). Our model includes meteorological variables such as temperature, humidity, and wind speed. Our statistical models for post-processing are evaluated on the basis of deviance and Akaike Information Criterion (AIC). We expect that the GLM and GAM will be useful for capturing nonlinear relationships between the PM2.5 measurements and meteorological variables.