四川省PM2.5时空演化特征及定量模拟
2021-04-03周智敏孙丹喻琴昆
周智敏 孙丹 喻琴昆



摘 要:文章利用气象要素和PM2.5质量浓度数据,对四川省PM2.5的时空变化和定量模拟进行了研究。結果表明:西、北地区PM2.5污染最轻,污染呈改善趋势,四季变化不明显;东、南地区污染较重,四季变化明显,其中成都、巴中空气质量改善较大,宜宾、内江和自贡空气质量改善不明显。同时,混合效应模型会显著提高气象要素与PM2.5之间的相关性,绵阳模拟效果最好,交叉验证后,R2为0.82,RMSE为7.32 μg/m3,PRE为13.39%;宜宾最差,三者分别为0.74、23.61 μg/m3和34.54%;八地市平均为0.78、16.31 μg/m3和26.68%。
关键词:四川省;时空演化;定量模拟;混合效应模型;十折交叉验证
中图分类号:X513;TP391.9 文献标识码:A 文章编号:2096-4706(2021)19-0111-06
The Spatiotemporal Evolution Characteristics and Quantitative Simulation of PM2.5 in Sichuan Province
ZHOU Zhimin1, SUN Dan1, YU Qinkun2
(1.ABA Teachers University, Aba 623002, China; 2.Zigong Meteorological Office, Zigong 610072, China)
Abstract: This paper studies the spatiotemporal evolution and quantitative simulation of PM2.5 in Sichuan Province by using meteorological elements and PM2.5 mass concentration data. The results show: PM2.5 pollution in Sichuans western and northern regions is the lightest and displays a trend of continual improvement, and the change in four seasons is not obvious; the eastern and southern regions are heavily polluted, with obvious changes in four seasons. Among them, the air quality in Chengdu and Bazhong has improved greatly, the air quality in Yibin, Neijiang and Zigong has not improved significantly. At the same time, the mixed effect model will significantly improve the correlation between meteorological elements and PM2.5, Mianyang has the best simulation effect. After the cross validation, R2 is 0.82, RMSE is 7.32 μg/m3 and PRE is 13.39%; Yibin is the worst, the three index are respectively 0.74, 23.61 μg/m3 and 34.54%; the average value of eight cities are 0.78, 16.31 μg/m3 and 26.68%.
Keywords: Sichuan Province; spatiotemporal evolution; quantitative simulation; mixed effect model; 10-fold cross-validation
0 引 言
近数十年,我国社会经济及城市化的飞速发展,导致了环境空气质量日益恶化,其中以津京冀、长三角、珠三角以及四川盆地最为突出[1,2]。诸多研究表明,细颗粒物(PM2.5)是大气污染问题的决定性影响因素之一,且城市大气污染俨然已发展成重大民生问题之一[3]。对此,2017年,国务院常务会议部署了对大气重污染研究项目,并由环保部(现生态环境部)结合科技部、中科院、气象、高校等部委和多单位协同攻关,设立了大气重污染成因与治理攻关项目,旨在解析津京冀地区“2+26”城市大气重污染成因以及治理管控技术[4],随着攻关项目的开展,现已扩大到覆盖四川盆地等全部四大重污染区域。
PM2.5浓度的时空变化是研究大气污染的基础,但针对区域污染,通常只能采用“以点代面”的方法进行研究[5]。对此,发展了诸多反演技术,如卫星遥感反演[6,7]、气象要素反演[8,9]、能见度反演[10]、自回归预测[11,12]、机器学习[13-15]等。……
