Particulate Matter (PM10) Impact on Health and Environment: A Statistical Learning Approach for Analyzing Spatiotemporal Data

Source abstract

Inhalable airborne particulate matter (PM), including components (nitrate, sulfate, ammonium dust, pollen, mold, soot, iron, aluminum, and sea salt, among others) with an aerodynamic diameter of ≤10 micrometers (µm), referred to as PM10, is one of the indicators of air pollution. The particulate matter PM10 levels above normal have been shown to have consistent, long-term harmful effects on human health and the environment, which in turn can cause severe ecosystem disruption. Ironically, sustainability can fail when disruptions to ecosystems upset their natural balance, reducing their ability to provide essential resources over time. Severe ecosystem disruptions can trigger permanent and damaging shifts in regional habitability and agricultural productivity. Moreover, chronic exposure to PM10 drives higher mortality and morbidity rates than short-term exposure alone. These PM10-based air pollution data are often collected repeatedly at many monitoring stations worldwide. These ongoing data collections from many stations often exhibit spatial and/or temporal dependencies, which introduce additional complexity in conducting appropriate statistical analysis. These dependencies warrant multivariate models across many time points, which are largely unavailable. In addition, relating time-varying covariates to PM10 levels at different time points and estimating covariate risks are other complexities that typically arise during the analysis. In this research, we propose marginal–conditional approaches based on joint statistical modeling to predict the trajectory risk of spatiotemporal air pollution data. Our approach will be illustrated using repeatedly collected PM10 data from the Environmental Protection Agency (EPA) in the USA between 2000 and 2025 at 1110 monitoring stations. The proposed model overcomes several barriers to analyzing large, complex spatiotemporal data by introducing a unified modeling framework.