基于卫星数据和深度学习方法预测京津冀地区地表 NO2浓度

摘要/Abstract

摘要： 二氧化氮 (NO2) 对人类健康和气候变化有着诸多负面影响，随着中国城镇化和工业化进程加速， NO2污染成为人们日益关注的问题。相关研究表明传统的单个站点监测结果只能代表数平方公里内的污染物水平，无法提供大尺度的污染物分布信息。相比于站点监测，卫星遥感可以提供大尺度且时空连续的数据，为研究大气污染提供了新的角度。基于哨兵5P卫星的NO2柱浓度数据和气象、人口密度等其他辅助数据，构建了对地表NO2进行预测的深度神经网络 (DNN) 模型。并使用两种交叉验证方法对该模型进行验证。在基于样本的验证中，模型的决定系数 R2、均方根误差 (RMSE) 和平均预测误差 (MAE) 分别为0.80、7.72 μg/m3和 5.31 μg/m3；在基于站点的验证中，模型的R2、RMSE 和MAE分别为 0.74、8.95 μg/m3和6.01 μg/m3，两种验证结果都表明DNN模型具有较好的整体预测能力和空间泛化性。此外，与经典的地学统计和机器学习算法对比结果表明DNN预测性能优于其它方法。最后用训练好的模型获得了京津冀地区 0.1° 的NO2分布图。

关键词: 二氧化氮, 机器学习, 哨兵5P, 遥感

Abstract: Nitrogen dioxide (NO2) has many adverse impacts on human health and climate change. With the acceleration of urbanization and industrialization in China, NO2 pollution has become a growing concern. However, releveant research shows that the traditional monitoring results of a single site can only represent the concentration of pollutants within a few square kilometers, and cannot provide large-scale pollutant distribution information. Compared with site monitoring, satellite remote sensing can provide large-scale and spatiotemporal continuous data. Based on NO2 column densities of Sentinel-5 Precursor and other auxiliary data such as weather and population density, a deep learning model (DNN) to predict groundlevel NO2 concentration is built in this work, and then the model is verified by two cross-validation strategies. In the sample-based cross validation, the determination coefficient R2, root mean square error (RMSE) and mean absolute error (MAE) of the model are 0.80、7.72 μg/m3 and 5.31 μg/m3, respectively, while in the site-based cross validation, they are 0.74、8.95 μg/m3 and 6.01 μg/m3, respectively. Both of the two cross-validation results indicate that the DNN model has excellent overall predictive performance and spatial generalization ability. In addition, the comparisons with the other classic geostatistics and machine learning algorithms also show that the predictive performance of the deep learning algorithm is better than that of the other methods. Finally, the trained model is applied to generate NO2 distribution with 0.1° spatial resolution across Beijing-Tianjin-Hebei region.

Key words: nitrogen dioxide, machine learning, Sentinel 5P, remote sensing

中图分类号:

X511

范轩硕, 吴海滨, 陈新兵, 宋伟 . 基于卫星数据和深度学习方法预测京津冀地区地表 NO2浓度[J]. 大气与环境光学学报, 2023, 18(2): 181-190.

FAN Xuanshuo , WU Haibin , CHEN Xinbing , SONG Wei . Deep learning architecture based on satellite remote sensing data for estimating ground-level NO2 across Beijing-Tianjin-Hebei Region[J]. Journal of Atmospheric and Environmental Optics, 2023, 18(2): 181-190.

参考文献 27

[1]	Song G X, Jiang L L, Chen G H, et al. A time-series study on the relationship between gaseous airpollutants and daily
	mortality in Shanghai [J]. Journal of Environment and Health, 2006, 23(5): 390-393.
	宋桂香, 江莉莉, 陈国海, 等. 上海市大气气态污染物与居民每日死亡关系的时间序列研究 [J]. 环境与健康杂志, 2006,
23	(5): 390-393.
[2]	Zhang T Y, Shen N C, Zhao X, et al. Spatiotemporal variation caracteristics of ozone and its population exposure risk assessment in Chengdu-Chongqing urban agglomeration during 2015 to 2019 [J]. Acta Scientiae Circumstantiae, 2021, 41
(10)	: 4188-4199.
	张天岳, 沈楠驰, 赵雪, 等. 2015—2019 年成渝城市群臭氧浓度时空变化特征及人口暴露风险评价 [J]. 环境科学学报,
20	21, 41(10): 4188-4199.
[3]	Jiang D S. Air quality assessment and exposure risk of different groups in Fujian Province in 2016―2018 [J]. Low Carbon
	World, 2019, 9(3): 12-13.
	蒋冬升. 福建省2016—2018年空气质量评价及不同人群暴露风险研究 [J]. 低碳世界, 2019, 9(3): 12-13.
[4]	Wang Q Q, Ye Y J, Zhang J Y, et al. Multi-site analysis of acute effects of air pollutants combination exposure on mortality
	in Jiangsu Province, China [J]. Chinese Journalof Preventive Medicine, 2019, 53(1): 86-92.
	汪庆庆, 叶云杰, 张嘉尧, 等. 江苏省大气污染物联合暴露对人群死亡风险急性效应的多中心研究 [J]. 中华预防医学杂
	志, 2019, 53(1): 86-92.
[5]	Shi X Q, Zhao C F, Jiang J H, et al. Spatial representativeness of PM2.5 concentrations obtained using observations from
	network stations [J]. Journal of Geophysical Research: Atmospheres, 2018, 123(6): 3145-3158.
[6]	Briggs D J, Collins S, Elliott P, et al. Mapping urban air pollution using GIS: A regression-based approach [J]. International
	Journal of Geographical Information Science, 1997, 11(7): 699-718.
[7]	Zou Y X, Wu Z F, Cao Z. Assessing PM2.5 exposure risk by coupling land use regression model andpopulation weighted
	model [J]. Journal of Geo-Information Science, 2019, 21(7): 1018-1028.
	邹雨轩, 吴志峰, 曹峥. 耦合土地利用回归与人口加权模型的PM2.5暴露风险评估 [J]. 地球信息科学学报, 2019, 21(7):
10	18-1028.
[8]	Zhao X, Hou L L, Wang X L, et al. Simulation of spatial distribution of PM2.5 and PM10 concentrations in Beijing in 2019
	based on LUR model [J]. Acta Scientiae Circumstantiae, 2020, 40(11): 4060-4069.
	赵雪, 侯丽丽, 王鑫龙, 等. 基于LUR模型的2019 年北京地区PM2.5与PM10浓度空间分异模拟 [J]. 环境科学学报, 2020,
40	(11): 4060-4069.
[9]	Feng C L, Li R K. Spatiotemporal variation analysis of air pollution from 2013 to 2019 in Beijing based on land use regression
	model [J]. Acta Scientiae Circumstantiae, 2021, 41(4): 1231-1238.
	冯春莉, 李润奎. 基于土地利用回归模型的北京市2013—2019 年大气污染时空变化分析 [J]. 环境科学学报, 2021, 41
(4)	: 1231-1238.
[10]	Wei S M, Pan J H, Tuo W L. Estimation and spatial-temporal distribution characteristic of PM2.5 concentration by remote
	sensing in China in 2015 [J]. Remote Sensing Technology and Application, 2020, 35(4): 845-854.
	魏石梅, 潘竟虎, 妥文亮. 2015 年中国PM2.5浓度遥感估算与时空分布特征 [J]. 遥感技术与应用, 2020, 35(4): 845-854.
[11]	Deng Y, Liu J P, Liu Y, et al. Spatial distribution estimation of PM2.5 concentration Beijing by applying Bayesian geographic
	weighted regression model [J]. Science of Surveying and Mapping, 2018, 43(10): 39-45.
	邓悦, 刘纪平, 刘洋, 等. 北京PM2.5浓度空间分布的贝叶斯地理加权回归模拟 [J]. 测绘科学, 2018, 43(10): 39-45.
[12]	Zhan Y, Luo Y Z, Deng X F, et al. Satellite-based estimates of daily NO2 exposure in China using hybrid random forest and
	spatiotemporal kriging model [J]. Environmental Science& Technology, 2018, 52(7): 4180-4189.
[13]	Luo A R. Research of Prediction Model on Atmospheric PM2.5 Concentration Using Support Vector Regression [D]. Beijing:
	Beijing University of Technology, 2018.
	罗奥荣. 基于支持向量回归机的大气PM2.5浓度预测模型研究 [D]. 北京: 北京工业大学, 2018.
[14]	Yang L. Optimal-Combined Model for Air Quality Index Forecasting—5 Cities in North China [D]. Lanzhou: Lanzhou
	University, 2019.
	杨玲. 最优-组合模型的空气质量指标预测: 以中国华北的5城市为例 [D]. 兰州: 兰州大学, 2019.
[15]	Yuan Q Q, Shen H F, Li T W, et al. Deep learning in environmental remote sensing: Achievements and challenges [J]. Remote Sensing of Environment, 2020, 241: 111716.
[16]	Zaytar M A, Amrani C E. Machine learning methods for air quality monitoring [P]. Networking, Information Systems &
	Security, 2020.
[17]	Rodríguez J D, Pérez A, Lozano J A. Sensitivity analysis of kappa-fold cross validation in prediction error estimation [J].
	IEEE Transactions on Pattern Analysis and Machine Intelligence, 2010, 32(3): 569-575.
[18]	Li T W, Shen H F, Zeng C, et al. A validation approach considering the uneven distribution of ground stations for satellitebased
	PM2.5 estimation [J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2020, 13:
13	12-1321.
[19]	Li X W, Zhang Y, Jiang S, et al. Preliminary application of atmospheric pollution monitoring in Jiangsu Province with
	TROPOMI sensor onboard sentinel-5P satellite [J]. Environmental Monitoring and Forewarning, 2019, 11(2): 10-16.
	李旭文, 张悦, 姜晟, 等. "哨兵-5P"卫星TROPOMI传感器在江苏省域大气污染监测中的初步应用 [J]. 环境监控与预
	警, 2019, 11(2): 10-16.
[20]	Baldi P. Gradient descent learning algorithm overview: A general dynamical systems perspective [J]. IEEE Transactions on
	Neural Networks, 1995, 6(1): 182-195.
[21]	Liu J W, Liu Y, Luo X L. Research and development on deep learning [J]. Application Research of Computers, 2014, 31(7):
19	21-1930.
	刘建伟, 刘媛, 罗雄麟. 深度学习研究进展 [J]. 计算机应用研究, 2014, 31(7): 1921-1930.
[22]	Qin W Z. The Basic Theoretics and Application Research on Geographically Weighted Regression [D]. Shanghai: Tongji
	University, 2007.
	覃文忠. 地理加权回归基本理论与应用研究 [D]. 上海: 同济大学, 2007.
[23]	Tang F M. Study of Support Vector Machines Algorithm Based on Statistical Learning Theory [D]. Wuhan: Huazhong
	University of Science and Technology, 2005.
	唐发明. 基于统计学习理论的支持向量机算法研究 [D]. 武汉: 华中科技大学, 2005.
[24]	Wei J Y, Ru F. Forecasting the traffic volume by the model of GRNN and studing it's realization [J]. Journal of Changsha
	Communications University, 2006, 22(2): 46-50.
	魏晋雁, 茹锋. 采用GRNN模型进行交通量预测及实现研究 [J]. 长沙交通学院学报, 2006, 22(2): 46-50.
[25]	Du X, Feng J Y, Lv S Q, et al. PM2.5 concentration prediction model based on random forest regression analysis [J].
	Telecommunications Science, 2017, 33(7): 66-75.
	杜续, 冯景瑜, 吕少卿, 等. 基于随机森林回归分析的PM2.5浓度预测模型 [J]. 电信科学, 2017, 33(7): 66-75.
[26]	Diego R J, Pérez A, Antonio L J. Sensitivity analysis of kappa-fold cross validation in prediction error estimation [J]. IEEE
	transactions on pattern analysis and machine intelligence, 2010, 32(3): 569-575.
[27]	Li T W, Shen H F, Yuan Q Q, et al. Geographically and temporally weighted neural networks for satellite-based mapping of
	ground-level PM2.5 [J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2020, 167: 178-188.