School of Information Science and Technology, Yunnan Normal University, Yunnan, 650500, China; GIS Technology Research Center of Resource and Environment in Western China, Ministry of Education, Yunnan Normal University, Yunnan 650500, China.
GIS Technology Research Center of Resource and Environment in Western China, Ministry of Education, Yunnan Normal University, Yunnan 650500, China; Smith School of Business, Queen's University, Toronto, Canada.
Sci Total Environ. 2019 Dec 15;696:133983. doi: 10.1016/j.scitotenv.2019.133983. Epub 2019 Aug 19.
High concentration of fine particulate matter (PM) has been shown to be a major contributor to haze weather, which has been associated with an increased prevalence in lung cancer. An accurate estimation and predication of PM historical levels, and its spatial-temporal variability can assist in strategically improving regional air quality and reducing its harmful effects on population health. This paper targets Beijing, Tianjin, and Hebei province (BTH), three northeast province of china (TNPC), Yangtze river delta (YRD) and pearl river delta (PRD) as the study areas. Data used in this study include PM measurements from April 2013 to December 2016, MODIS AOD raster imageries and five meteorological factors from 2000 to 2016. By combining back propagation artificial neural network (BPANN) and ε-support vector regression (ε-SVR), a novel hybrid model was constructed to impute the historical PM missing values in the long time series from 2000 to 2012, and to predict the concentration of PM from April 2014 to December 2017. The hybrid model produced results superior to BPANN and ε-SVR with a higher accuracy, lower error rate, and a stable performance. This model can be applied to the other four regions with consistent results. Results of spatial-temporal analysis indicated that the PM concentration has increased along with a pollution range expansion in BTH from 2000 to 2010. In addition, the PM concentration decreased slowly in PRD. The concentration and pollution range of PM in TNPC and YRD showed a stable trend. In 2012, the four research areas all showed decreased trend, and the pollution range narrowed. From 2013 to 2016, the PM concentration increased shortly then decreased; in particular, the high pollution areas saw a decrease in PM concentration, which correlated with control measures adopted by the state during the same time period. The hot spots of PM were mainly distributed in the inland cities.
高浓度的细颗粒物（PM）已被证明是导致雾霾天气的主要因素之一，而雾霾天气的出现与肺癌发病率的上升有关。准确估计和预测 PM 的历史水平及其时空变化，有助于有策略地改善区域空气质量，减少其对人口健康的有害影响。本文以北京、天津和河北省（BTH）、中国东北三省（TNPC）、长江三角洲（YRD）和珠江三角洲（PRD）为研究区域。本研究使用的数据包括 2013 年 4 月至 2016 年 12 月的 PM 测量值、MODIS AOD 光栅成像仪以及 2000 年至 2016 年的五个气象因素。通过结合反向传播人工神经网络（BPANN）和ε-支持向量回归（ε-SVR），构建了一种新的混合模型，用于插补 2000 年至 2012 年长时间序列中的历史 PM 缺失值，并预测 2014 年 4 月至 2017 年 12 月的 PM 浓度。该混合模型的结果优于 BPANN 和 ε-SVR，具有更高的准确性、更低的错误率和更稳定的性能。该模型可应用于其他四个具有一致结果的区域。时空分析结果表明，2000 年至 2010 年，BTH 地区的 PM 浓度随着污染范围的扩大而增加。此外，PRD 地区的 PM 浓度下降缓慢。TNPC 和 YRD 地区的 PM 浓度和污染范围呈稳定趋势。2012 年，四个研究区域均呈下降趋势，污染范围缩小。从 2013 年到 2016 年，PM 浓度先增加后减少；特别是高污染地区的 PM 浓度有所下降，这与同期国家采取的控制措施有关。PM 的热点主要分布在内陆城市。
