利用PCA-kNN方法改进广州市空气质量模式PM_(2.5)预报
|
摘要
为了提高广州市PM_(2.5)客观预报能力,采用主成分分析结合机器学习算法k近邻(PCA-kNN)方法,基于空气质量模式(CMAQ)预报产品、中尺度天气模式(GRAPES-MESO)预报产品和2017年上半年广州PM_(2.5)观测实况,试验确定PCA-kNN方法的最佳参数方案,建立广州市空气质量模式PM_(2.5)预报客观订正方法。结果表明:与CMAQ模式的PM_(2.5)预报相比,在第1～3天预报时效上,PCA-kNN订正结果与实况的相关系数分别提高20%、15%、29%,均方根误差分别降低17%、16%、20%,平均偏差更接近0,PM_(2.5)浓度等级TS评分接近或优于CMAQ预报,PCA-kNN订正结果优于CMAQ预报。机器学习算法PCA-kNN方法可有效改进广州市空气质量模式PM_(2.5)预报,本研究对其他地区、其他污染物客观预报研究具有借鉴意义。
In order to improve the ability of objective forecasting of PM_(2.5) in Guangzhou, a bias correction method combining principal component analysis and k nearest neighbor regression in machine learning(PCA-kNN) was proposed based on the outputs of air quality model(CMAQ), mesoscale numerical weather model(GRAPES-MESO) and PM_(2.5) observations in the first half of 2017. The parameter scheme for PCA-kNN was determined by parametric tests and then the relevant data of Guangzhou were used to evaluate the performance. Results showed that compared with CMAQ PM_(2.5) products, the PCA-kNN method performed better in 1～3 d forecast of PM_(2.5) daily mean:(1) the correlation coefficients between PCA-kNN outputs and PM_(2.5) observations increased by 20%, 15% and 29% respectively;(2) root mean square errors decreased by 17%, 16% and 20% respectively;(3) mean biases were closer to zero; and(4) TS scores of different PM_(2.5) concentration levels were comparable or better. By applying machine learning algorithm, the PCA-kNN method can effectively improve air quality model PM_(2.5) forecasting. This study has implications for objective forecasting and research of other areas and other pollutants.
In order to improve the ability of objective forecasting of PM_(2.5) in Guangzhou, a bias correction method combining principal component analysis and k nearest neighbor regression in machine learning(PCA-kNN) was proposed based on the outputs of air quality model(CMAQ), mesoscale numerical weather model(GRAPES-MESO) and PM_(2.5) observations in the first half of 2017. The parameter scheme for PCA-kNN was determined by parametric tests and then the relevant data of Guangzhou were used to evaluate the performance. Results showed that compared with CMAQ PM_(2.5) products, the PCA-kNN method performed better in 1～3 d forecast of PM_(2.5) daily mean:(1) the correlation coefficients between PCA-kNN outputs and PM_(2.5) observations increased by 20%, 15% and 29% respectively;(2) root mean square errors decreased by 17%, 16% and 20% respectively;(3) mean biases were closer to zero; and(4) TS scores of different PM_(2.5) concentration levels were comparable or better. By applying machine learning algorithm, the PCA-kNN method can effectively improve air quality model PM_(2.5) forecasting. This study has implications for objective forecasting and research of other areas and other pollutants.
引文
[1]吴兑,毕雪岩,邓雪娇,等.珠江三角洲大气灰霾导致能见度下降问题研究[J].气象学报, 2006, 64(4):510-517.
[2]吴兑,邓雪娇,毕雪岩,等.细粒子污染形成灰霾天气导致广州市能见度下降[J].热带气象学报, 2007, 23(1):1-6.
[3]吴兑.近十年中国灰霾天气研究综述[J].环境科学学报, 2012, 32(2):257-269.
[4] LV B, COBOURN W G, BAI Y. Development of nonlinear empirical models to forecast daily PM2.5and ozone levels in three largeChinese cities[J]. Atmos Environ, 2016, 147(12):209-223.
[5] PEREZ P, SALINI G. PM2.5forecasting in a large city:comparison of three methods[J]. Atmos Environ, 2008, 42(35):8 219-8 224.
[6] EDER B, KANG D, MATHUR R, et al. An operational evaluation of the Eta-CMAQ air quality forecast model[J]. Atmos Environ, 2006,40(26):4 894-4 905.
[7] ZHANG Y, BOCQUET M, MALLET V, et al. Real-time air quality forecasting, part I:History, techniques, and current status[J]. AtmosEnviron, 2012, 60(12):632-655.
[8] ZHANG Y, BOCQUET M, MALLET V, et al. Real-time air quality forecasting, part II:State of the science, current research needs, andfuture prospects[J]. Atmos Environ, 2012, 60(12):656-676.
[9]薛纪善.和预报员谈数值预报[J].气象, 2007, 33(8):3-11.
[10]陈德辉,薛纪善.数值天气预报业务模式现状与展望[J].气象学报, 2004, 62(5):623-633.
[11]邓雪娇,邓涛,麦博儒,等.华南区域大气成分业务数值预报GRACEs模式系统[J].热带气象学报, 2016, 32(6):900-907.
[12] BORREGO C, MONTEIRO A, PAY M T, et al. How bias-correction can improve air quality forecasts over Portugal[J]. Atmos Environ,2011, 45(37):6 629-6 641.
[13] DJALALOVA I, DELLE MONACHE L, WILCZAK J. PM2.5analog forecast and Kalman filter post-processing for the CommunityMultiscale Air Quality(CMAQ)model[J]. Atmos Environ, 2015, 119(10):431-442.
[14] GLAHN H R, LOWRY D A. The use of model output statistics(MOS)in objective weather forecasting[J]. J Appl Meteorol, 1972, 11(8):1 203-1 211.
[15] KONOVALOV I B, BEEKMANN M, MELEUX F, et al. Combining deterministic and statistical approaches for PM10forecasting inEurope[J]. Atmos Environ, 2009, 43(40):6 425-6 434.
[16]王开燕,邓涛,邓雪娇,等.珠三角空气质量数值预报系统的检验与订正研究[J].环境科学与管理, 2017, 42(1):47-49.
[17]谢敏,钟流举,陈焕盛,等. CMAQ模式及其修正预报在珠三角区域的应用检验[J].环境科学与技术, 2012, 35(2):96-101.
[18] KANG D, MATHUR R, RAO S T. Real-time bias-adjusted O3and PM2.5air quality index forecasts and their performance evaluationsover the continental United States[J]. Atmos Environ, 2010, 44(18):2 203-2 212.
[19]黄思,唐晓,徐文帅,等.利用多模式集合和多元线性回归改进北京PM10预报[J].环境科学学报, 2015, 35(1):56-64.
[20] LYU B, ZHANG Y, HU Y. Improving PM2.5air quality model forecasts in China using a bias-correction framework[J]. Atmosphere, 2017, 8(8):147.
[21] MCGOVERN A, ELMORE K L, GAGNE D J, et al. Using artificial intelligence to improve real-time decision making for high-impactweather[J]. Bull Am Meteorol Soc, 2017, 98:2 073-2 090.
[22]陈德辉,薛纪善,杨学胜,等. GRAPES新一代全球/区域多尺度统一数值预报模式总体设计研究[J].科学通报, 2008, 53(20):2 396-2 407.
[23] WU X, KUMAR V, QUINLAN J R, et al. Top 10 algorithms in data mining[J]. Knowl Inf Syst, 2008, 14(1):1-37.
[24]周志华.机器学习:Machine learning[M].清华大学出版社, 2016:225-241.
[25] WOLD S, ESBENSEN K, GELADI P. Principal component analysis[J]. Chemometr Intell Lab Syst, 1987, 2(1-3):37-52.
[26] JOLLIFFE I T. Principal component analysis[M].Berlin:Springer-Verlag, 2002:20-24.
[27] BESSE P. PCA stability and choice of dimensionality[J]. Stat Probab Lett, 1992, 13(5):405-410.
[28] GANGOPADHYAY S, HARDING B L, RAJAGOPALAN B, et al. A nonparametric approach for paleohydrologic reconstruction ofannual streamflow ensembles[J]. Water Resour Res, 2009, 45(6):W06417.
[29] HOPSON T M, WEBSTER P J. A 1-10-day ensemble forecasting scheme for the major river basins of Bangladesh:Forecasting severefloods of 2003-07[J]. J Hydrometeorol, 2010, 11(3):618-641.
[30] WU W, LIU Y, GE M, et al. Statistical downscaling of climate forecast system seasonal predictions for the Southeastern Mediterranean[J].Atmos Res, 2012, 118(11):346-356.
[31]中华人民共和国环境保护部.环境空气质量指数(AQI)技术规定(试行)[S]. HJ 633-2012.
[32]麦健华,于玲玲,邓涛,等.基于GRAPES-CMAQ的中山市空气质量预报系统预报效果评估[J].热带气象学报, 2018,34(1):78-86.
[33] PEDREGOSA F, VAROQUAUX G, GRAMFORT A, et al. Scikit-learn:Machine learning in Python[J]. J Mach Learn Res, 2011, 12(10):2 825-2 830.
[34]侯梦玲,王宏,赵天良,等.京津冀一次重度雾霾天气能见度及边界层关键气象要素的模拟研究[J].大气科学, 2017, 41(6):1 177-1 190.
[35]张艳霞,蒙伟光,戴光丰,等.城市冠层模式在GRAPES模式中的应用[J].热带气象学报, 2016, 32(3):311-321.
[36]许建明,徐祥德,刘煜,等. CMAQ-MOS区域空气质量统计修正模型预报途径研究[J].中国科学(D辑), 2005, 35(1):131-144.
[2]吴兑,邓雪娇,毕雪岩,等.细粒子污染形成灰霾天气导致广州市能见度下降[J].热带气象学报, 2007, 23(1):1-6.
[3]吴兑.近十年中国灰霾天气研究综述[J].环境科学学报, 2012, 32(2):257-269.
[4] LV B, COBOURN W G, BAI Y. Development of nonlinear empirical models to forecast daily PM2.5and ozone levels in three largeChinese cities[J]. Atmos Environ, 2016, 147(12):209-223.
[5] PEREZ P, SALINI G. PM2.5forecasting in a large city:comparison of three methods[J]. Atmos Environ, 2008, 42(35):8 219-8 224.
[6] EDER B, KANG D, MATHUR R, et al. An operational evaluation of the Eta-CMAQ air quality forecast model[J]. Atmos Environ, 2006,40(26):4 894-4 905.
[7] ZHANG Y, BOCQUET M, MALLET V, et al. Real-time air quality forecasting, part I:History, techniques, and current status[J]. AtmosEnviron, 2012, 60(12):632-655.
[8] ZHANG Y, BOCQUET M, MALLET V, et al. Real-time air quality forecasting, part II:State of the science, current research needs, andfuture prospects[J]. Atmos Environ, 2012, 60(12):656-676.
[9]薛纪善.和预报员谈数值预报[J].气象, 2007, 33(8):3-11.
[10]陈德辉,薛纪善.数值天气预报业务模式现状与展望[J].气象学报, 2004, 62(5):623-633.
[11]邓雪娇,邓涛,麦博儒,等.华南区域大气成分业务数值预报GRACEs模式系统[J].热带气象学报, 2016, 32(6):900-907.
[12] BORREGO C, MONTEIRO A, PAY M T, et al. How bias-correction can improve air quality forecasts over Portugal[J]. Atmos Environ,2011, 45(37):6 629-6 641.
[13] DJALALOVA I, DELLE MONACHE L, WILCZAK J. PM2.5analog forecast and Kalman filter post-processing for the CommunityMultiscale Air Quality(CMAQ)model[J]. Atmos Environ, 2015, 119(10):431-442.
[14] GLAHN H R, LOWRY D A. The use of model output statistics(MOS)in objective weather forecasting[J]. J Appl Meteorol, 1972, 11(8):1 203-1 211.
[15] KONOVALOV I B, BEEKMANN M, MELEUX F, et al. Combining deterministic and statistical approaches for PM10forecasting inEurope[J]. Atmos Environ, 2009, 43(40):6 425-6 434.
[16]王开燕,邓涛,邓雪娇,等.珠三角空气质量数值预报系统的检验与订正研究[J].环境科学与管理, 2017, 42(1):47-49.
[17]谢敏,钟流举,陈焕盛,等. CMAQ模式及其修正预报在珠三角区域的应用检验[J].环境科学与技术, 2012, 35(2):96-101.
[18] KANG D, MATHUR R, RAO S T. Real-time bias-adjusted O3and PM2.5air quality index forecasts and their performance evaluationsover the continental United States[J]. Atmos Environ, 2010, 44(18):2 203-2 212.
[19]黄思,唐晓,徐文帅,等.利用多模式集合和多元线性回归改进北京PM10预报[J].环境科学学报, 2015, 35(1):56-64.
[20] LYU B, ZHANG Y, HU Y. Improving PM2.5air quality model forecasts in China using a bias-correction framework[J]. Atmosphere, 2017, 8(8):147.
[21] MCGOVERN A, ELMORE K L, GAGNE D J, et al. Using artificial intelligence to improve real-time decision making for high-impactweather[J]. Bull Am Meteorol Soc, 2017, 98:2 073-2 090.
[22]陈德辉,薛纪善,杨学胜,等. GRAPES新一代全球/区域多尺度统一数值预报模式总体设计研究[J].科学通报, 2008, 53(20):2 396-2 407.
[23] WU X, KUMAR V, QUINLAN J R, et al. Top 10 algorithms in data mining[J]. Knowl Inf Syst, 2008, 14(1):1-37.
[24]周志华.机器学习:Machine learning[M].清华大学出版社, 2016:225-241.
[25] WOLD S, ESBENSEN K, GELADI P. Principal component analysis[J]. Chemometr Intell Lab Syst, 1987, 2(1-3):37-52.
[26] JOLLIFFE I T. Principal component analysis[M].Berlin:Springer-Verlag, 2002:20-24.
[27] BESSE P. PCA stability and choice of dimensionality[J]. Stat Probab Lett, 1992, 13(5):405-410.
[28] GANGOPADHYAY S, HARDING B L, RAJAGOPALAN B, et al. A nonparametric approach for paleohydrologic reconstruction ofannual streamflow ensembles[J]. Water Resour Res, 2009, 45(6):W06417.
[29] HOPSON T M, WEBSTER P J. A 1-10-day ensemble forecasting scheme for the major river basins of Bangladesh:Forecasting severefloods of 2003-07[J]. J Hydrometeorol, 2010, 11(3):618-641.
[30] WU W, LIU Y, GE M, et al. Statistical downscaling of climate forecast system seasonal predictions for the Southeastern Mediterranean[J].Atmos Res, 2012, 118(11):346-356.
[31]中华人民共和国环境保护部.环境空气质量指数(AQI)技术规定(试行)[S]. HJ 633-2012.
[32]麦健华,于玲玲,邓涛,等.基于GRAPES-CMAQ的中山市空气质量预报系统预报效果评估[J].热带气象学报, 2018,34(1):78-86.
[33] PEDREGOSA F, VAROQUAUX G, GRAMFORT A, et al. Scikit-learn:Machine learning in Python[J]. J Mach Learn Res, 2011, 12(10):2 825-2 830.
[34]侯梦玲,王宏,赵天良,等.京津冀一次重度雾霾天气能见度及边界层关键气象要素的模拟研究[J].大气科学, 2017, 41(6):1 177-1 190.
[35]张艳霞,蒙伟光,戴光丰,等.城市冠层模式在GRAPES模式中的应用[J].热带气象学报, 2016, 32(3):311-321.
[36]许建明,徐祥德,刘煜,等. CMAQ-MOS区域空气质量统计修正模型预报途径研究[J].中国科学(D辑), 2005, 35(1):131-144.
© 2004-2018 中国地质图书馆版权所有 京ICP备05064691号 京公网安备11010802017129号 地址：北京市海淀区学院路29号 邮编：100083 电话：办公室：(+86 10)66554848；文献借阅、咨询服务、科技查新：66554700 |