Merging Mobile Measurements, Traffic Emissions Models, and Land Use Regression Towards Real-Time Estimation of Traffic-Related Air Pollution

Abstract

Background: Current Land Use Regression (LUR) modeling approaches lack the spatiotemporal precision needed to match people’s time-activity patterns for exposure assessment. We developed spatiotemporal LUR models of Particle Number (PN) and Black Carbon (BC) concentrations using two pieces of information: (1) mobile monitoring and (2) on-road emissions estimates from the INTEGRATION microscopic traffic simulation model which tracks vehicle movements, fuel consumption, and emissions every second.Methods: We employed mobile monitoring to collect ~120 hours of PN and BC measurements in Blacksburg, VA. Data collection was stratified by time of day (~10 hours of monitoring for each hour during 7am-7pm). We calibrated the INTEGRATION traffic emissions model to estimate hourly PN and BC emissions (100m spatial scale). We combined these datasets, along with traditional LUR covariates (e.g., land use variables), for model building. We developed and compared three types of models: (1) daytime average models that pool all mobile measurements, (2) hourly models (i.e., single model for each hour of day) and (3) spatiotemporal models (single model incorporating hour of day as a predictor).Results: Model fit for the daytime average models (adj-R2 for PN [BC]: 0.72 [0.66]) were comparable to models developed using fixed-site measurements. The hourly models (i.e., 12 separate models) had modest model fit (mean PN [BC] adj-R2: 0.52 [0.36]). The spatiotemporal models had the poorest model fit among model-types (adj-R2 for PN [BC]: 0.46 [0.27]). Results for the temporally resolved models were comparable to other mobile monitoring campaigns that attempted to model long-term averages suggesting it is possible to develop models with both spatial and temporal precision for real-time dissemination.Conclusions: Our approach to model traffic-related air pollution allows for matching the temporal and spatial resolution of people’s time-activity patterns towards improved exposure assessment.