数值模式及机器学习对兰州市近地面臭氧模拟适用性  

The applicability of numerical model and machine learning for near-surface ozone simulation in Lanzhou City

作　　者：周恒左 廖鹏 杨宏[1] 陈恒蕤 落义明 潘峰[1] 仝纪龙[1] 刘永乐 ZHOU Heng-zuo;LIAO Peng;YANG Hong;CHEN Heng-rui;LUO Yi-ming;PAN Feng;TONG Ji-long;LIU Yong-le(College of Atmospheric Sciences,Lanzhou University,Lanzhou 730000,China)

机构地区：[1]兰州大学大气科学学院,甘肃兰州730000

出　　处：《中国环境科学》2024年第1期15-27,共13页China Environmental Science

基　　金：国家自然科学基金资助项目(42075174)。

摘　　要：分别应用数值模式及机器学习模型对兰州市2019年7月近地面臭氧浓度进行模拟,以对比不同方法下模拟效果的差异.其中,数值模式部分选用了3种不同的化学机理(CBMZ、RADM2、CB06r3),结果显示CBMZ化学机理模拟效果优于其他2种化学机理,RADM2高估了兰州市近地面臭氧浓度,CB06r3则有些低估.机器学习部分则选用了两种模型(XGBoost、PSO-BP),结果表明在缺少大气污染物排放清单的情况下,仅使用气象数据,无论是单个站点还是空间分布,2种机器学习模型均表现较好,且XGBoost模型在模拟近地面臭氧空间分布上表现更优.整体来看,2种机器学习模型相较于数值模式计算速度更快,但受其输入数据的影响较明显,对于更高空间分辨率的模拟研究及污染过程分析仍然需要依靠数值模式.因此,应该根据不同的需求及数据条件选择更合适的方法进行近地面臭氧模拟.Numerical-and machine learning models with three different chemical mechanisms(CBMZ,RADM2,and CB06r3)were applied to simulate the near-surface ozone concentration in Lanzhou city in July 2019.Results show that CBMZ performed better than both RADM2 and CB06r3 did of which RADM2led to an overestimate of the near-surface ozone concentration while CB06r3 to a slightly underestimate.Then,the results from two machine learning models(XGBoost and PSO-BP)showed that in the absence of atmospheric pollutant emission inventory and only meteorological data were used,both two machine learning models performed better,regardless of single site or spatial distribution.In addition,the XGBoost model performed better for simulating the spatial distribution of near-surface ozone concentrations.Overall,the two machine learning models computed faster than the numerical models,but were significantly influenced by the input data,implying that the numerical models are more suitable for simulating pollution processes.Generally,a model suitable for simulating ground-level ozone should be selected according to the simulation requirements and data conditions.

关 键 词：臭氧 数值模式 化学机理 机器学习 

分 类 号：X511[环境科学与工程—环境工程]