3种区域空气质量空间插值方法对比研究

Comparative study of the three spatial interpolation methods for the regional air quality evaluation

空间插值方法可实现空气质量空间可视化展示,也是可视化研究空气质量区域分布特征的重要方法。由于区域空气质量受污染物种类及季节的影响明显,不同空间插值方法的可靠性和适用性存在差异。利用珠三角区域62个大气环境自动监测站的空气污染物浓度数据,分别采用反距离加权、规则样条函数2种确定性插值方法和地统计学的克里金法进行空间插值计算,结合交叉验证法,综合对比分析不同空气污染物、不同季节的空间插值效果。结果表明,克里金法能得到整体最优的插值精度,插值结果可靠性最高,但在表达空气质量空间分布特征上过渡平滑性欠佳。不同空气污染物空间插值效果不同,季节性特征也对插值结果影响显著。3种空间插值方法的ρ（PM2.5）、ρ（O3）插值效果均明显优于一次污染物,其插值结果的MRE（平均相对误差）在0.186~0.313,决定系数R2基本在0.8以上,但一次污染物的MRE均在0.352以上。3种方法 PM2.5的夏季插值结果均明显优于其他季节：PM2.5夏季ME（平均误差）均值在-0.1~0.3μg/m3,且95%置信区间较小,误差分布集中;而非夏季高污染时期的空间插值效果相对较差。

The paper is engaged in a comparative study of the three spatial interpolation methods for the regional air quality evaluation,whose reliability and applicability has been soundly verified.As a matter of fact,the daily concentration of the air pollutants（PM2.5,SO2,O3,NO2,CO） in the 62 automatic monitoring stations based on the Pearl River Delta,Guangdong,has been made available from 2012 to 2013,which helps us to compare the three different interpolation methods in testing and measuring various air pollutants on the said delta area.Moreover,it is necessary to point out that the three spatial interpolation methods can actually be made of the two deterministic methods（Inverse Distance Weighted and Completely Regularized Spline） and the other one known as geo-statistical method（Kriging）.What is more,we have to use the cross-validation method for evaluating the accuracy and reliability of the three methods on the various air pollutants in different seasons.With the aforementioned methods we can compare the interpolation methods and gain the four indices,such as the mean error（ME）,the mean relative error（MRE）,the root mean squared error（RMSE）,and the correlation coefficient（R2）.The results of our comparative study of the three spatial interpolation methods for the regional air quality evaluation demonstrate that Kriging method can always help to draw highly accurate evaluation with great reliability,though the application may not be done smoothly in the spatial distribution of air quality in the transitional or boundary areas.In addition,the effects of the spatial interpolation methods can vary as a result of the existing of different sorts of air pollutants and the effects of different seasons on the interpolation results.For example,the functions or roles of the three spatial interpolation methods on PM2.5and O3 can be made better than on the primary pollutants.For example,for PM2.5and O3,the mean relative error of interpolation results has been worked out within the range of 0.186 to0.313,with the correlation coefficient being above 0.8,whereas the mean relative error for the primary pollutants has been worked out just a bit over 0.352.As to the seasonal variation of PM2.5,or example,the interpolation results in summer turn out to be significantly better than in other seasons,for which the average value of mean error in summer has to be put in a range from-0.1μg/m3 to 0.3 μg/m3,much less than in other seasons.What is more,the average value of the mean error for PM2.5in summer tends to be at a confidence level at a 95% rate,a bit less than in other seasons,and so is the error concentrated distribution.And,finally,the featuring performance of the spatial interpolation turns to be rather poor in the periods of high concentration of air pollutants.Thus,it can be seen that the above results can be taken as a valuable reference for choosing the spatial interpolation methods in the study of the air quality distribution.