Effects of spatial sampling schemes and preferential sampling: a focus on air quality

Abstract

Air quality monitoring is essential for environmental management and public health. The strategic placement of monitoring stations and accurate assessment on the Air Quality Index remains unsolved due to geographical and socio-economic factors. This project aims to explore the various sampling schemes, preferential factors, and predictive methodologies in assessing AQI distribution and optimizing air monitoring networks. As we adopt stratified and random sampling, followed by kriging interpolation, we were able to analyze AQI variability across different counties. While stratified sampling allows us to capture the diverse environmental characteristics of each county, random sampling provided a wider view of AQI distribution. In addition, the use of machine learning approach, Random Forest Regression model was trained to include preferential factors as a covariate to predict AQI values. Stratified Sampling coupled with kriging revealed specific pollution hotspots, indicating potential areas for a station placement. Furthermore, random sampling provided a general AQI spread, identifying a wider region for possible air monitoring placement. On the other hand, Random Forest Regression model offers predictive insights, which is useful to make predictions in unseen locations. This study offers valuable insights in the effects of sampling strategies and offer valuable guidance in the placement of air monitoring stations.