Unveiling the nonlinear relationships and co-mitigation effects of green and blue space landscapes on PM2.5 exposure through explainable machine learning

Abstract

Green-blue spaces are nature-based solutions to mitigate particulate matter pollution. However, the individual and co-mitigation effects of green-blue space landscapes on PM2.5 exposure risk remain poorly understood. This study employed an explainable machine learning framework to investigate the nonlinear relationships, interaction effects, and heterogeneity of green-blue space landscape patterns on population-weighted PM_2.5 exposure (PWP) in the Yangtze River Delta, China. Our findings highlight that (1) Greenspace coverage (G_PLAND), mean greenspace patch size (G_AREA_MN), blue space patch contiguity (W_CONTIG_MN), and mean distance between blue space patches (W_ENN_MN) are the four most influential landscape indicators. (2) G_PLAND and G_AREA_MN negatively influence PWP with thresholds of 40 % and 50 ha, respectively. W_CONTIG_MN (> 0.26) and W_ENN_MN (< 400 m) positively impact PWP. (3) Effects of green-blue space landscapes on PWP vary with different exposure levels: high (blue space is more important), medium (green and blue space are equally important), and low (green-blue spaces are not important). (4) Interactions of green and blue spaces can reinforce PWP mitigation under certain conditions (G_PLAND > 40 %, G_AREA_MN < 12 ha, W_ENN_MN and W_CONTIG_MN with thresholds of 200 m and 0.31, respectively). The findings can facilitate comprehensive planning and optimization of regional green-blue spaces to mitigate PWP.