两种空气质量数值模式的应用评估与集合改进研究

孙弦; 孙磊; 聂会文; 梁秀姬; 苏烨康; 王静; 夏冬

doi:10.16032/j.issn.1004-4965.2023.033

两种空气质量数值模式的应用评估与集合改进研究

doi: 10.16032/j.issn.1004-4965.2023.033

珠海市公共气象服务中心，广东珠海 519000

基金项目:

广东省气象局科学技术研究项目 GRMC2021M19

珠海市气象局科学技术研究项目 ZH202108

详细信息

通讯作者:
孙磊，男，江苏省人，工程师，主要从事环境气象研究。E-mail: ggqx_sl@zhuhai.gov.cn

中图分类号: X16
计量
- 文章访问数: 87
- HTML全文浏览量: 31
- PDF下载量: 17
- 被引次数: 0
出版历程
- 收稿日期: 2022-06-02
- 修回日期: 2022-12-20
- 网络出版日期: 2023-09-11
- 刊出日期: 2023-06-20

APPLICATION EVALUATION AND ENSEMBLE IMPROVEMENT OF TWO AIR-QUALITY NUMERICAL MODELS

Zhuhai Public Meteorological Service Center, Zhuhai, Guangdong 519000, China

摘要

摘要: 利用2018—2019年国控站观测，评估CAMx和CMAQ模式对广东珠海主要污染物时空分布与演变特征的预报能力，并引入多元线性回归和随机森林方法对预报结果进行集成，探究不同集合方法的改进能力。结果表明：CMAQ在各污染物浓度季节-日变化方面明显优于CAMx，但两者存在明显系统偏差，并对多数污染物（除O₃之外）的昼夜和空间变化的模拟能力仍存在明显缺陷。例如，CMAQ合理地还原了CO、PM_2.5、PM₁₀、SO₂、O₃和NO₂的季节变化，相关系数介于0.72~0.84，但NMB分别达到-0.58、-0.18、-0.30、1.52，-0.16和-0.20，RMSE分别达到0.40 mg/m³、6.86、16.02、10.71、25.05和10.21 μg/m³。同时，基于不同污染物构建的两种集合方法均有效移除了系统偏差，加强了CMAQ的模拟优势，并且随机森林方法明显优于多元线性回归，但两者均对模式缺陷无明显改进。进一步分析发现，CMAQ与CAMx模型的重要性基本相当，表明集合方法的预报能力与集合成员的线性偏差无关，主要取决于不同成员的代表性。最后，本研究揭示以随机森林为代表的集合方法虽有效提高了污染物的预报能力，但改进数值模式自身能力和增加具有代表性的集合成员对预报水平的进一步提升十分关键。
- 空气质量模式 /
- 应用评估 /
- 集合方法 /
- 随机森林
Abstract: This study evaluated the performances of two numerical air-quality models (i. e., CMAQ and CAMx) for the air pollutants forecast over Zhuhai based on the observations from national stations during 2018-2019. Moreover, two ensemble methods including Multiple Linear Regression (MLR) and Random Forest (RF) were employed to determine their improvement capabilities. The results show that the CMAQ outperformed the CAMx in reproducing the seasonal-daily variations of each air pollutant, but both of them showed obvious systematic biases, and presented poor performances in representing the diurnal and spatial variations of the air pollutants (except O₃). For example, the CMAQ reasonably reproduced the seasonal variations of CO, PM_2.5, PM₁₀, SO₂, O₃ and NO₂, with correlation coefficients ranging from 0.72 to 0.84, while the NMB (RMSE) reached -0.58 (0.40 mg/m³), -0.18 (6.86 μg/m³), -0.30 (16.02 μg/m³), 1.52 (10.71 μg/m³), -0.16 (25.05 μg/m³) and -0.20 (10.21 μg/m³), respectively. Meanwhile, the MLR and RF models, constructed based on all pollutants, effectively removed the systematic biases, and enhanced the strengths of the CAMx model, with the RF performing better. However, both of them did not improve these model defects much. Further analysis indicates that the impotency of CMAQ and CAMx was comparably important, suggesting that the prediction ability of the ensemble method is mainly related to the representativeness of different members, irrespective of their linear deviations. Finally, this study reveals that despite the effective improvement of the RF in predicting the pollutants, the key for further refinement of predicting capabilities is to improve numerical models'own abilities and incorporate more representative participating members.
- air quality model /
- application evaluation /
- ensemble methods /
- Random Forest

HTML全文

图 1 珠海市地形高度空间分布

其中红星代表四个国控站（分别为唐家、吉大、前山和斗门）所在位置。

下载: 全尺寸图片幻灯片

图 2 两种数值模式（CAMx和CMAQ）及其集合方法（MLR和RF）2018—2019年珠海市六种空气污染物浓度月均值变化与观测（OBS）对比

下载: 全尺寸图片幻灯片

图 3 两种数值模式（CAMx和CMAQ）及其集合方法（MLR和RF）2018年珠海市六种空气污染物浓度日均值变化与观测（OBS）对比

下载: 全尺寸图片幻灯片

图 4 两种数值模式（CAMx和CMAQ）及其集合方法（MLR和RF）2018—2019年珠海市六种空气污染物（a~f）浓度日均值（x轴）与对应观测（y轴）对比散点图（不同颜色代表不同模式或方法）

其中k和R²分别为拟合线的斜率和决定系数（两者越接近于1，模拟效果越好，颜色与点相对应）。

下载: 全尺寸图片幻灯片

图 5 观测（OBS）、两种数值模式（CAMx和CMAQ）及其集合方法（MLR和RF）给出的珠海市2018—2019年六种空气污染物浓度（已减去自身均值）昼夜变化对比

下载: 全尺寸图片幻灯片

图 6 两种数值模式（CAMx和CMAQ）及其集合方法（MLR、RF和RF-sta）预报的吉大站（第1列）、斗门站（第2列）、前山站（第3列）、唐家站（第4列）2018—2019年四种主要空气污染物年日浓度均值（减去站点平均，柱状线，对应左侧纵坐标，单位为μg/m³）、标准差（除以站点平均，三角形，对应右侧纵坐标，单位为μg/m³）与实测对比

下载: 全尺寸图片幻灯片

图 7 随机森林集合方法中CAMx和CMAQ模型对于各空气污染物的重要性

下载: 全尺寸图片幻灯片

表 1 CMAQ与CAMx模式设置

模式选项	CMAQ	CAMx
气象条件来源	区域GRAPES模式	区域GRAPES模式
排放源清单	多种源清单融合，并结合卫星遥感和观测进行优化	多种源清单融合
网格嵌套模式	三重嵌套，单向	三重嵌套，单向
水平分辨率	27-9-3 km	27-9-3 km
垂直分层数	25	25
水平平流	PPM	PPM
垂直对流	PPM	隐式对流
水平扩散	ACM2	ACM2
垂直扩散	涡流扩散	涡流扩散
干沉降	M3DRY	WESELY89
气象化学机理	SAPRC07	SAPRC07
气象化学算法	EBI	CMC
网格烟雨模块	关闭	关闭

下载: 导出CSV

表 2 珠海市六种污染物季节变化统计参数

污染物	RMSE				R				NMB
污染物	CAMx	CMAQ	MLR	RF	CAMx	CMAQ	MLR	RF	CAMx	CMAQ	MLR	RF
CO	0.34	0.40	0.08	0.08	0.61	0.77	0.76	0.81	-0.49	-0.58	0.01	0.01
PM_2.5	11.02	6.86	6.42	5.17	0.82	0.90	0.91	0.93	-0.33	-0.18	0.01	0.02
PM₁₀	25.28	16.02	10.86	8.68	0.82	0.84	0.85	0.90	-0.53	-0.30	0.01	0.01
SO₂	14.16	10.71	1.75	1.57	0.25	0.78	0.71	0.78	1.99	1.52	0.01	0.02
O₃	28.56	25.05	25.33	22.44	0.41	0.72	0.73	0.76	-0.15	-0.16	-0.16	-0.12
NO₂	27.61	10.21	9.93	9.37	0.49	0.74	0.73	0.78	-0.87	-0.20	0.04	0.05

下载: 导出CSV

参考文献(47)

[1]	WANG F, QIU X, CAO J, et al. Policy-driven changes in the health risk of PM_2.5 and O₃ exposure in China during 2013-2018[J]. Science of the Total Environment, 2021, 757: 143775.
[2]	曾贤刚, 阮芳芳, 姜艺婧. 中国臭氧污染的空间分布和健康效应[J]. 中国环境科学, 2019, 39(9): 4 025-4 032.
[3]	AN Z, HUANG R J, ZHANG R, et al. Severe haze in northern China: A synergy of anthropogenic emissions and atmospheric processes[J]. Proceedings of the National Academy of Sciences, 2019, 116(18): 8 657-8 666.
[4]	NAM K M, ZHANG X, ZHONG M, et al. Health effects of ozone and particulate matter pollution in China: a province-level CGE analysis[J]. The Annals of Regional Science, 2019, 63(2): 269-293.
[5]	LIU H, LIU S, XUE B, et al. Ground-level ozone pollution and its health impacts in China. Atmospheric environment[J], 2018, 173: 223-230.
[6]	WILSON S R, MADRONICH S, LONGSTRETH J D, et al. Interactive effects of changing stratospheric ozone and climate on tropospheric composition and air quality, and the consequences for human and ecosystem health[J]. Photochemical & Photobiological Sciences, 2019, 18 (3): 775-803.
[7]	DENG T, WANG T, WANG S, et al. Impact of typhoon periphery on high ozone and high aerosol pollution in the Pearl River Delta region[J]. Science of the total environment, 2019, 668: 617-630.
[8]	周胜, 黄报远, 陈慧英, 等. 珠三角城市群PM_2.5和O₃污染特征及VOCs组分敏感性分析[J]. 环境工程, 2020, 38(1): 42-47.
[9]	王耀庭, 梁加成, 郑祚芳, 等. 基于地-空观测数据的粤港澳大湾区大气污染调查及时空演变特征[J]. 环境科学, 2020, 41(5): 2 075-2 086.
[10]	ZHANG Y, BOCQUET M, MALLET V, et al. Real-time air quality forecasting, part I: History, techniques, and current status[J]. Atmospheric Environment, 2012, 60: 632-655.
[11]	徐爱兰, 张再峰, 孙强, 等. 基于深度学习的人工智能空气质量预报系统构建[J]. 中国环境监测, 2021, 37(2): 7.
[12]	WANG P, ZHANG H, QIN Z, et al. A novel hybrid-Garch model based on ARIMA and SVM for PM_2.5 concentrations forecasting. Atmospheric Pollution Research[J], 2017, 8(5): 850-860.
[13]	YAFOUZ A, AHMED A N, ZAINI N A, et al. Ozone concentration forecasting based on artificial intelligence techniques: A systematic review[J]. Water, Air, & Soil Pollution, 2021, 232(2): 1-29.
[14]	TAYLAN O. Modelling and analysis of ozone concentration by artificial intelligent techniques for estimating air quality[J]. Atmospheric environment, 2017, 150: 356-365.
[15]	MA R, BAN J, WANG Q, et al. Random forest model based fine scale spatiotemporal O₃ trends in the Beijing-Tianjin-Hebei region in China, 2010 to 2017[J]. Environmental Pollution, 2021, 276: 116635.
[16]	张云海, 孙财涛, 杨洪斌. 基于逐步回归分析方法的PM₁₀浓度预报模型[J]. 环境科学与技术, 2009, 32(5): 100-102.
[17]	HU J, CHEN J, YING Q, et al. One-year simulation of ozone and particulate matter in China using WRF/CMAQ modeling system[J]. Atmospheric Chemistry and Physics, 2016, 16(16): 10 333-10 350.
[18]	王燕丽, 薛文博, 雷宇, 等. 京津冀区域PM_2.5污染相互输送特征[J]. 环境科学, 2017, 38(12): 4 897-4 904.
[19]	HONG J, MAO F, MIN Q, et al. Improved PM_2.5 predictions of WRF-Chem via the integration of Himawari-8 satellite data and ground observations[J]. Environmental Pollution, 2020, 263: 114451.
[20]	WANG Z, ITAHASHI S, UNO I, et al. Modeling the long-range transport of particulate matters for January in East Asia using NAQPMS and CMAQ[J]. Aerosol and Air Quality Research, 2017, 17(12): 3 065-3 078.
[21]	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.
[22]	王自发, 吴其重, GBAGUIDI Alex, 等. 北京空气质量多模式集成预报系统的建立及初步应用[J]. 南京信息工程大学学报: 自然科学版, 2009, 1(1): 19-26.
[23]	周广强, 谢英, 吴剑斌, 等. 基于WRF-chem模式的华东区域PM_2.5预报及偏差原因[J]. 中国环境科学, 2016, 8(9): 2 251-2 259.
[24]	EDER B, KANG D, MATHUR R, et al. An operational evaluation of the Eta-CMAQ air quality forecast model[J]. Atmospheric Environment, 2006, 40(26): 4 894-4 905.
[25]	邓涛, 邓雪娇, 吴兑, 等. 珠三角灰霾数值预报模式与业务运行评估[J]. 气象科技进展, 2012, 6(7): 38-44.
[26]	李婷苑, 吴乃庚, 邓雪娇, 等. 华南区域大气成分数值模式GRACES预报性能评估[J]. 热带气象学报, 2021, 37(2): 207-217.
[27]	SHEN J, WANG X, LI J, et al. Evaluation and intercomparison of ozone simulations by Models-3/CMAQ and CAMx over the Pearl River Delta[J]. Science China Chemistry, 2011, 54(11): 1 789-1 800.
[28]	MA S, ZHANG X, GAO C, et al. Multimodel simulations of a springtime dust storm over northeastern China: implications of an evaluation of four commonly used air quality models (CMAQ v5. 2.1, CAMx v6. 50, CHIMERE v2017r4, and WRF-Chem v3. 9.1)[J]. Geoscientific Model Development, 2019, 12(11): 4 603-4 625.
[29]	VAUTARD R, SCHAAP M, BERGSTRÖM R, et al. Skill and uncertainty of a regional air quality model ensemble[J]. Atmospheric Environment, 2009, 43(31): 4 822-4 832.
[30]	李勋, 秦墨梅, 李琳, 等. 基于空气质量模拟的江苏省大气污染物排放清单比较研究[J]. 南京信息工程大学学报: 自然科学版, 2020, 12 (6): 705-713.
[31]	LI M, LIU H, GENG G, et al. Anthropogenic emission inventories in China: a review[J]. National Science Review, 2017, 4(6): 834-866.
[32]	BASSETT R, YOUNG P J, BLAIR G S, et al. A large ensemble approach to quantifying internal model variability within the WRF numerical model[J]. Journal of Geophysical Research: Atmospheres, 2020, 125(7): e2019JD031286.
[33]	KITAYAMA K, MORINO Y, YAMAJI K, et al. Uncertainties in O₃ concentrations simulated by CMAQ over Japan using four chemical mechanisms[J]. Atmospheric Environment, 2019, 198: 448-462.
[34]	VAN LOON M, VAUTARD R, SCHAAP M, et al. Evaluation of long-term ozone simulations from seven regional air quality models and their ensemble[J]. Atmospheric Environment, 2007, 41(10): 2 083-2 097.
[35]	DJALALOVA I, WILCZAK J, MCKEEN S, et al. Ensemble and bias-correction techniques for air quality model forecasts of surface O₃ and PM_2.5 during the TEXAQS-Ⅱ experiment of 2006[J]. Atmospheric Environment, 2010, 44(4): 455-467.
[36]	MALLET V, STOLTZ G, MAURICETTE B. Ozone ensemble forecast with machine learning algorithms[J]. J Geophy Res: Atmos, 2009, 114(D5).
[37]	潘锦秀, 晏平仲, 孙峰, 等. 多元线性回归方法对北京地区PM_2.5预报的改进应用[J]. 中国环境监测, 2019, 35(2): 43-52.
[38]	杨关盈, 石春娥, 邓学良, 等. 多模式集成方法在安徽地区PM_2.5预报中的应用研究[J]. 环境科学学报, 2021, 41(3): 1-11.
[39]	李娟, 尉鹏, 戴学之, 等. 基于机器学习方法的西安市数值模拟优化研究[J]. 环境科学研究, 2021, 34(4): 872-881.
[40]	汤静, 谭浩波, 邓雪娇, 等. 利用PCA-KNN方法改进广州市空气质量模式PM_2.5预报[J]. 热带气象学报, 2019, 35(1): 125-134.
[41]	BREIMAN L. Random forests[B]. Machine learning, 2001, 45(1): 5-32.
[42]	WANG Y, DU Y, WANG J, et al. Calibration of a low-cost PM_2.5 monitor using a random forest model[J]. Environment International, 2019, 133: 105161.
[43]	WANG M, TANG G, LIU Y, et al. The difference in the boundary layer height between urban and suburban areas in Beijing and its implications for air pollution[J]. Atmospheric Environment, 2021, 260: 118552.
[44]	PANDAY A K, PRINN R G. Diurnal cycle of air pollution in the Kathmandu Valley, Nepal: Observations[J]. J Geophy Res: Atmos, 2009, 114(D9): D09305.
[45]	赖安琪, 陈晓阳, 刘一鸣, 等. 珠江三角洲高质量浓度PM_2.5和O₃复合污染特征[J]. 中山大学学报(自然科学版), 2018, 258(4): 36-42.
[46]	GRÖMPING U. Variable importance assessment in regression: linear regression versus random forest[J]. The American Statistician, 2009, 63(4): 308-319.
[47]	沈劲, 王雪松, 李金凤, 等. Models-3/CMAQ和CAMx对珠江三角洲臭氧污染模拟的比较分析[J]. 中国科学: 化学, 2011, 41(11): 207-217.