Statistical Emulation of Winter Ambient Fine Particulate Matter Concentrations From Emission Changes in China

Air pollution exposure remains a leading public health problem in China. The use of chemical transport models to quantify the impacts various emission changes on air quality is limited by their large computational demands. Machine learning can emulate provide computationally efficient predictions outputs based statistical associations with inputs. We developed novel emulators relating five key anthropogenic sectors (residential, industry, land transport, agriculture, and power generation) winter ambient fine particulate matter (PM2.5) concentrations across were optimized Gaussian process regressors Matern kernels. predicted 99.9% variance PM2.5 for given input configuration changes. are primarily sensitive residential (51%–94% first-order sensitivity index), industrial (7%–31%), agricultural emissions (0%–24%). Sensitivities generation all under 5%, except South West China where contributed 13%. largest reduction 68%–81%, down 15.3–25.9 μg m−3, remaining above World Health Organization annual guideline 10 m−3. greatest reductions driven reducing emissions, emphasizing importance these sectors. show that National Quality Target 35 m−3 unlikely be achieved during without strong from