The impact of different spatial sampling strategies on the spatial extrapolation prediction accuracy of PM2.5 concentrations in the Beijing-Tianjin-Hebei region, China

Abstract

PM_2.5 monitoring station networks play a critical role in retrieving ground-level PM_2.5 concentrations using satellite remote sensing technology. However, the optimal spatial distribution of PM_2.5 monitoring stations is frequently overlooked during satellite-based PM_2.5 retrieval. Here, we employ the space-time linear mixed effects (STLME) model to assess the impact of different spatial sampling strategies on the spatial extrapolation prediction accuracy of PM_2.5 concentrations in the Beijing-Tianjin-Hebei region in 2020. Results demonstrate that the all-station strategy (national + provincial stations) achieves superior performance, with station- and county/district-based CV-R² values measuring 0.87 and 0.78 respectively, compared to the national-station strategy's corresponding values of 0.80 and 0.73. These findings suggest that both the all-station and national-station strategies generally reflect the model's spatial extrapolation prediction accuracy at the county and city scales. Furthermore, six strategies based on the all-station framework were developed through spatial stratified sampling strategy, demonstrating that strategic expansion of monitoring stations enhances model performance. However, model performance exhibits limited improvement when the number of stations exceeds 300. That indicate the all-station strategy can meet the estimation accuracy requirements for spatial extrapolation prediction models in this area. These findings suggest that the all-station strategy offers a relatively robust framework for PM₂.₅ extrapolation modeling in the BTH region, making it a preferred choice for supporting future air quality management and monitoring network optimization.