基于Copula函数的成都夏季O3污染潜势模型

利用成都市2016~2019年6~8月O₃浓度的逐时监测数据以及该时段同时次的地面气象观测资料,构建了O₃污染潜势3维(紫外辐射、相对湿度和气温)Copula联合概率分布模型,并开展了模型的适用性研究.首先,通过对SciPy库概率分布函数的优选,确定了不同O₃浓度等级条件下紫外辐射、气温和相对湿度的最优边缘概率分布函数(均通过了显著性水平a=0.05的K-S检验);其次,计算了3种Copula联合概率分布函数的均方根误差(RMSE值)、赤池信息准则(AIC值)、贝叶斯信息量(BIC值),并借助Anderson-Darling检验,发现非对称3维frank Copula联合概率分布函数(M3Copula)可以最佳地表征不同O₃浓度等级条件下紫外辐射、相对湿度和气温的联合概率分布特征;最后,将不同O₃浓度等级条件下M3Copula联合概率密度作为对应O₃浓度等级的隶属度,O₃污染潜势的分类结果对实际O₃浓度等级具有较好的指示意义,模拟的平均准确率为63%,其中优等级、良等级、轻度污染等级以及中度及以上污染等级的模拟准确率分别为82%、64%、48%和75%.

Summer O3 pollution potential model based on copula function in Chengdu

The evolution of Ozone (O₃) concentration near the ground is closely related to the coupling effect of multiple meteorological factors, but the complexity and uncertainty keep still unclear . In order to explore the problem mentioned above, the hourly monitoring data of O₃ concentration as well as the surface meteorological observation data during the same time period in Chengdu from 2016 to 2019 during summer were collected, a three-dimensional copula joint probability distribution model of O₃ pollution potential (including UV radiation, relative humidity, and temperature) was constructed, and the applicability of the model was further explored. Firstly, the optimal marginal probability distributions of UV radiation, relative humidity and ambient temperature under different O₃ concentration levels were determined at significant level of a=0.05 in K-S test based on the optimization of probability distributions belonging to SciPy package. Secondly, the root-mean-square-error (RMSE), Akaike Information Criterion (AIC), and Bayesian Information criterion (BIC) of three kinds of joint probability distribution functions were calculated, respectively. With the help of Anderson-Darling test (A-D test), it was found that asymmetric three-dimensional frank Copula joint probability distribution function (M3Copula) has better fitting effects on the joint probability distribution characteristics of UV radiation, relative humidity, and ambient temperature, respectively, under different O₃ concentration levels. Finally, Taking the joint probability density of M3Copula under different O₃ concentration levels as the membership of O₃ concentration levels, the classification results of O₃ pollution potential has a fairly good indication of the actual O₃ concentration levels, and the M3Copula can simulate the O₃ concentration levels with 63% accuracy, of which the simulation accuracy of excellent level, good level, light pollution level and moderate or higher pollution level were 82%, 64%, 48%, and 75%, respectively. Our findings demonstrated that the classification results of O₃ pollution potential have a fairly good instruction significance to actual O₃ concentration levels.