A Modeling Framework Using Markov Chain and Autologistic Regression to Adjust Temporal and Spatial Dependencies for PM2.5 Trajectory Risk Prediction

Data on various air-quality metrics are collected repeatedly by numerous monitoring stations worldwide to closely assess the severity of pollution. Particle pollution from fine particulate matter PM2.5 is one such measure used as an indicator of whether air quality is unhealthy. PM2.5 is a specific form of air contamination that negatively impacts the environment and human health when levels are above a certain threshold. As data are collected repeatedly over time at multiple locations, there may be a temporal dependence among repeated outcomes and spatial dependence between neighboring stations. Thus, it is important to assess the impact of risk factors on trajectory risk prediction. However, due to the temporal and spatial dependencies, trajectory risk prediction for such data becomes complicated, as both types of dependences must be accounted for during model building. In this paper, we propose a modeling framework that accounts for both types of dependences by incorporating Markov chains and Markov regression, using autologistics for model fitting and trajectory risk prediction. The proposed model fitting and trajectory risk prediction are illustrated using PM2.5 outdoor air pollution data from the United States from 2000 to 2020.