Influence of spatial resolution of PM_2.5 concentrations and population on health impact assessment from 2010 to 2020...

Ambient PM_2.5 pollution is a leading environmental health risk factor worldwide. The spatial resolution of PM_2.5 concentrations and population strongly impacts PM_2.5-related health impact estimates. However, long-term variations and regional differences in this impact have rarely been explored, particularly in China. Here, by aggregating satellite-derived PM_2.5 concentration and population datasets at 1-km resolution in China to coarser resolutions (10, 50, and 100 km), we evaluated decadal changes in the impact of resolution on health assessments at national and local scales. For the sensitivity of population-weighted mean (PWM) PM_2.5 concentrations to resolution, we found that the national PWM PM_2.5 concentration decreased with coarser resolutions; this pattern was widely observed and was more obvious in southern and central China and the Sichuan Basin. The results showed that the sensitivity of national PWM PM_2.5 concentrations to resolution continuously weakened from 2010 to 2020, likely due to a reduction in the spatial heterogeneity of PM_2.5 concentrations in regions with high sensitivity. This weakness caused a large underestimation of the long-term trend of national PWM PM_2.5 using a 100-km resolution, which was 7% lower than the trend at 1 km. Regarding the sensitivity of PM_2.5-attributable mortality to resolution, most of China exhibited a pattern in which attributable mortality decreased with coarser resolution. The sensitivity of the estimated PM_2.5-attributable mortality to resolution also weakened over time on a national scale and in most parts of China. Nevertheless, the weakness for mortality sensitivity was not as apparent as for PWM PM_2.5 sensitivity. This was likely because different drivers played distinct roles in the temporal variation of the mortality sensitivity: population aging enhanced the sensitivity, and variations in PM_2.5 concentrations and population distribution both weakened the sensitivity. However, the national attributable mortality trend at a 100-km resolution was still underestimated by 1.75% relative to the 1-km resolution.