Study on PM2.5 concentration prediction by integrating GNSS, ERA5 PWV, and atmospheric pollutants
-
摘要: 冬春季节的空气质量预测有助于公众合理安排出行和政府相关部门的交通治理. 细颗粒物(PM2.5)的浓度主要影响因素有大气污染物、水汽等. 为提高PM2.5浓度预测的精度，以京津冀地区为例，利用快速傅里叶变换(fast Fourier transform，FFT)与长短期记忆(long short term memory，LSTM)神经网络方法相结合，考虑GNSS、ERA5水汽、大气污染物等观测要素，构建PM2.5的浓度预测模型，预测研究未来24 h的PM2.5的浓度. 利用GNSS水汽校正区域ERA5水汽，并进行精度评定. 利用FFT取大气污染物、第五代大气再分析产品(ECMWF atmospheric reanalysis 5，ERA5)水汽等观测要素的公共变化周期，获得最佳公共周期为78 h；选取最佳公共周期长度的各类要素作为模型输入，24 h序列的PM2.5浓度作为模型输出. 通过均方根误差(root mean square error，RMSE)评价指标进行模型精度评价. 研究结果表明：基于GNSS的ERA5水汽校正模型在秋冬季节ERA5水汽精度优于2 mm. FFT-LSTM模型预测精度在平原地区、山地地区和高原地区为10.22 μg/m3、8.56 μg/m3 和 9.02 μg/m3，预测时效达到24 h. 可有效预测未来24 h的PM2.5浓度. 该模型可为相关部门大气污染治理提供参考.
-
关键词:
- 细颗粒物(PM2.5) /
- 大气污染物 /
- GNSS水汽 /
- ERA5水汽 /
- 快速傅里叶变换(FFT) /
- 长短期记忆(LSTM)
Abstract: The prediction of air quality during the winter and spring seasons can be used for the public to make reasonable arrangements for travel and traffic management by relevant government departments. The main influencing factors of PM2.5 concentration include atmospheric pollutants, precipitable water vapor (PWV), etc. To improve the accuracy of PM2.5 concentration prediction, taking Beijing-Tianjin-Hebei region as an example, it was combined fast Fourier transform (FFT) and LSTM neural network methods, considered observation elements such as GNSS, ERA5 PWV, and atmospheric pollutants, and constructed the PM2.5 concentration prediction model to predict the concentration of PM2.5 in the next 24 hours. It was used GNSS PWV to correct the ERA5 PWV in the region and evaluated the accuracy. The public change period of air pollutants, ERA5 PWV and other observation elements are extracted by FFT, and the optimal public period is 78 hours; Select various factors with the best common cycle length as the model input, and the PM2.5 concentration of the 24 hour sequence as the model output. Evaluate model accuracy through RMSE evaluation indicators. The research results are indicated that the accuracy of ERA5 PWV correction model based on GNSS is better than 2 mm in autumn and winter seasons. The prediction accuracy of the FFT-LSTM model is 10.22 μg/m3 in plain, 8.56 μg/m3 in mountainous, and 9.02 μg/m3 in plateau regions, while the predicted time limit of 24 hours. It can effectively predict the PM2.5 concentration in the next 24 hours. This model can provide reference for relevant departments in air pollution control. -
表 1 ERA5产品与GNSS PWV 之间的RMSE 的统计
mm 季节 区域1 区域2 区域3 区域4 RMSE 相关性 RMSE 相关性 RMSE 相关性 RMSE 相关性 春 1.35 0.895 2.07 0.963 3.12 0.871 2.43 0.979 夏 3.34 0.806 5.61 0.942 5.27 0.962 6.27 0.959 秋 1.85 0.979 3.87 0.956 3.18 0.976 5.02 0.958 冬 0.72 0.935 1.22 0.894 1.53 0.912 2.07 0.908 
下载: 导出CSV
表 2 校正后ERA5水汽RMSE
mm 区域类型 春 夏 秋 冬 区域1 2.1 5.8 1.0 1.1 区域2 2.0 4.9 1.1 1.4 区域3 1.8 3.5 0.8 0.8 区域4 1.5 2.6 0.3 1.0
下载: 导出CSV
表 3 PM2.5浓度与大气污染物、ERA5水汽的相关性统计
测站 PM2.5与PM10 PM2.5与SO2 PM2.5与NO2 PM2.5与CO PM2.5与O3 PM2.5与PWV 围场 0.772 0.764 0.587 0.604 −0.282 0.195 唐山 0.861 0.314 0.518 0.665 −0.221 −0.221 沧州 0.854 0.430 0.749 0.651 −0.338 −0.339 邯郸 0.885 0.389 0.573 0.581 −0.404 −0.404
下载: 导出CSV
表 4 各类观测要素的公共周期
观测站点 频率/cpd 周期/h 观测站点 频率/cpd 周期/h 0.309 78 0.309 78 围场 0.225 107 唐山 0.021 1142 0.156 154 0.045 533 0.309 78 0.309 78 沧州 0.157 153 邯郸 0.033 727 0.229 105 0.203 118
下载: 导出CSV
表 5 不同频率下PM2.5与其他观测要素的相关性
观测站点 频率/cpd PM2.5与PM10 PM2.5与SO2 PM2.5与NO2 PM2.5与CO PM2.5与O3 PM2.5与PWV 0.309 0.771 0.819 0.666 0.737 −0.517 0.067 围场 0.225 0.789 0.757 0.699 0.769 −0.517 −0.084 0.156 0.872 0.749 0.707 0.820 −0.361 0.298 0.309 0.904 0.319 0.687 0.612 −0.349 −0.349 唐山 0.021 0.847 0.402 0.590 0.434 −0.053 −0.053 0.045 0.834 0.540 0.685 0.510 −0.159 −0.159 0.309 0.908 0.294 0.876 0.222 −0.103 −0.103 沧州 0.157 0.695 0.362 0.752 0.280 −0.196 −0.196 0.229 0.889 0.375 0.861 0.314 −0.166 −0.166 0.309 0.887 0.764 0.691 0.686 −0.536 −0.536 邯郸 0.033 0.692 0.464 0.551 0.532 −0.293 −0.293 0.203 0.847 0.725 0.635 0.630 −0.493 −0.493
下载: 导出CSV
-
[1] 郦嘉诚, 高庆先, 李亮, 等. 对首要污染物所揭示的京津冀环境空气质量状况的认识启迪与对策建议[J]. 环境科学研究, 2018, 31(10): 1651-1661. [2] 顾芳婷, 胡敏, 王渝, 等. 北京2009-2010年冬、春季PM2.5污染特征[J]. 中国环境科学, 2016, 36(9): 2578-2584. [3] TANG B Y, XIN Y Y, GAO W K, et al. Characteristics of complex air pollution in typical cities of North China[J]. Atmospheric and oceanic science letters, 2018, 11(7): 1-8. DOI: 10.1080/16742834.2018.1394158 [4] JIANG N, LI Q, SU F C, et al. Chemical characteristics and source apportionment of PM2.5 between heavily polluted days and other days in Zhengzhou, China[J]. Journal of environmental sciences, 2018(66): 188-198. DOI: 10.1016/j.jes.2017.05.006.Epub2017May10 [5] 贾佳, 韩力慧, 程水源, 等. 京津冀区域PM2.5及二次无机组分污染特征研究[J]. 中国环境科学, 2018, 38(3): 801-811. [6] WANG G, CHENG S Y, LANG J L, et al. Characteristic of PM2.5 and assessing effects of emission reduction measures in the heavy polluted city of Shijiazhuang, before, during and after the ceremonial parde 2015[J]. Aeroso land air quality research, 2017, 17(2): 499-512. DOI: 10.4209/aaqr.2016.05.0181 [7] ZHANG Y, BOCQUET M, MALLET V, et al. Realtime air quality forecasting[J]. Atmospheric environment, 2012(60): 632-655. DOI: 0.1016/j.atmosenv.2012.06.031 [8] 王平, 张红, 秦作栋, 等. 基于wavelet-SVM的PM10浓度时序数据预测[J]. 环境科学, 2017, 38(8): 3153-3161. [9] 戴李杰, 张长江, 马雷鸣. 基于机器学习的PM2.5短期浓度动态预报模型[J]. 计算机应用, 2017, 37(11): 3057-3063. [10] 宋国君, 国潇丹, 杨啸, 等. 沈阳市PM2.5浓度ARIMA-SVM组合预测研究[J]. 中国环境科学, 2018, 38(11): 4031-4039. [11] 朱亚杰, 李琦, 侯俊雄, 等. 基于支持向量回归的PM2.5浓度实时预测[J]. 测绘科学, 2016, 41(1): 12-17. [12] FENG X, LI Q, ZHU Y J, et al. Artificial neural networks forecasting of PM2.5, pollution using air mass trajectory based geographic model and wavelet transformation[J]. Atmospheric environment, 2015(107): 118-128. DOI: 10.1016/j.atmosenv.2015.02.030 [13] 高愈霄, 汪巍, 黄永海, 等. 基于神经网络和数值模型的重点区域PM2.5预报比较分析[J]. 环境科学, 2022, 43(2): 663-674. [14] 程春英, 尹学博. 雾霾之PM2.5的来源、成分、形成及危害[J]. 大学化学, 2014, 29(5): 1-6. [15] 王勇, 刘严萍, 李江波, 等. 水汽和风速对雾霾在PM2.5/PM10变化的影响[J]. 灾害学, 2015, 30(10): 5-7. [16] 王勇, 刘严萍, 李江波, 等. GPS和无线电探空的水汽变化与PM2.5/PM10变化的相关性研究[J]. 武汉大学学报(信息科学版), 2016, 41(12): 1626-1630. [17] 王勇, 王泓易, 刘严萍, 等. 融合GNSS水汽、风速与大气污染物的河北省冬季PM2.5浓度预测研究[J]. 大地测量与地球动力学, 2020, 40(11): 1145-1152. -
点击查看大图
图(3) / 表(5)
计量
- 文章访问数: 89
- HTML全文浏览量: 27
- PDF下载量: 15
- 被引次数: 0
