长三角对流层甲醛柱浓度时空变化及驱动因素

通过OMI卫星数据分析了2005~2016年长江三角洲对流层甲醛柱浓度的时空变化规律.同时结合2008年和2010年各部门VOCs人为源排放量，利用BP神经网络和RBFN神经网络模型对对流层甲醛柱浓度进行了县域尺度上的回归模拟和各部门排放量贡献度分析.结果表明：长三角城市群对流层甲醛柱浓度在2005~2010年存在着增加趋势，2011~2016年甲醛浓度有下降的趋势.高值区域分布在皖北苏北、上海及其附近，低值区域分布在浙西南一带.人为源排放使得经济发达地区的甲醛柱浓度显著增高.工业源在长三角的分布较为广泛，电力源分布稀疏且VOC排放量远小于工业源排放量，居民源的VOC排放量介于工业源和电力源之间，有明显的南北差异.交通源主要集中在苏南、浙北和上海附近，少部分沿交通线条状分布.机器学习算法可以较好地利用人为源排放数据对甲醛柱浓度进行模拟.神经网络的拟合精度可以达到0.6~0.8，比线性回归的拟合精度超出0.3~0.4.模型变量重要性计算显示各部门中居民源对甲醛柱浓度的贡献程度最高.研究对流层甲醛柱浓度的长期时空变化及其影响因素有利于深入研究臭氧污染，同时也为大气治理和政策制定提供了科学依据.

Spatiotemporal variation of tropospheric formaldehyde concentration and its driving factors in Yangtze River

This study uses OMI satellite data to analyse the temporal and spatial changes of the tropospheric formaldehyde column concentration in the Yangtze River Delta from 2005 to 2016. The BP and RBFN neural network models are used to perform regression simulation on the tropospheric formaldehyde column concentration at the county scale and analysis of the proportion of emissions from various departments using non-methane volatile organic compounds (NMVOC) data in 2008 and 2010. The results show that the tropospheric formaldehyde column concentration in the Yangtze River Delta urban agglomeration has an increasing trend from 2005 to 2010 and a downward trend from 2011 to 2016. The concentrations are higher in northern Anhui, northern Jiangsu, Shanghai and nearby areas, while those in southwestern Zhejiang are lower. In addition, NMVOC have significantly increased the concentration of formaldehyde in economically developed areas. The industrial sector's emissions are widely distributed in the Yangtze River Delta, and the VOC emissions from the power sector are much smaller than those from the industrial sector, and the distribution is also very sparse. The amount of VOC emissions generated by residents' lives is between the above two, with a clear North-South differentiation. Those from the transportation sector are mainly concentrated in southern Jiangsu, northern Zhejiang and Shanghai, and are distributed in strips along the transportation lines. What's more, the fitting accuracy of the neural network can reach 0.6~0.8, which is 0.3~0.4 higher than that of the linear regression, which proves that machine learning algorithms can better simulate the concentration of the formaldehyde column with NMVOC. The VOC emissions generated by residents' lives contribute most to the tropospheric formaldehyde column concentration. Studying the long-term temporal and spatial changes of the tropospheric formaldehyde column concentration and its influencing factors is conducive to in-depth study of ozone pollution, and it also provides a scientific basis for atmospheric governance and policy making.