基于地理加权回归克里金的广西雾霾预报模型研究

Frequent smog-haze has seriously affected the people's production and life. Therefore, the effective prediction of smog and haze has become a problem which needs to be studied and solved urgently. In recent years, based on the variations of GNSS ZTD/PWV, some scholars have carried out the researches of smog-haze monitoring and prediction using Neural Network, Linear Regression and other methods. However, these forecasting methods do not take into account the influence of meteorological factors such as temperature, pressure, relative humidity and wind speed. Moreover, these methods are difficult to effectively deal with the spatial non-stationary and autocorrelation characteristics between the main index of smog-haze and its influence factors. In order to solve these problems, Guangxi Tm model of non-meteorological parameter is established by using Guangxi sounding data, GGOS Atmosphere Tm grid data and ERA-Interim data. Combined with the ZTD derived from CORS station, high-precision PWV is obtained. According to the PM2.5/PM10, PWV and meteorological data, the correlations between PM2.5/PM10 and PWV, meteorological factors during smog-haze process are studied by using hourly analysis method. On the basis of correlations analysis, based on PWV variations and the influence of meteorological factors, Guangxi model of smog-haze forecasting is constructed by using Geographically Weighted Regression Kriging, which can provide technical support for regional smog-haze monitoring and early warning, promote the further development of GNSS meteorology.

频发的雾霾严重影响了人们的生产和生活，因而雾霾的有效预测已成为亟需研究和解决的问题。近年来，一些学者基于GNSS ZTD/PWV变化采用神经网络、线性回归等方法开展雾霾监测预报研究。然而这些方法没有考虑气温、气压、相对湿度以及风速等气象因子的影响，而且这些方法也难以有效处理雾霾主要指标与其影响因子之间的空间非平稳性、自相关性特征。针对这些问题，利用广西探空数据和GGOS Atmosphere Tm格网数据、ERA-Interim资料，建立广西地区非气象参数Tm模型；结合CORS站提取的ZTD，获得高精度PWV。依据PM2.5/PM10、PWV及气象数据，采用逐时分析方法，研究雾霾进程中PM2.5/PM10与PWV和气象因子的相关性。在此基础上，利用地理加权回归克里金方法，构建基于PWV变化并顾及气象因子影响的广西雾霾预报模型，为区域雾霾监测预警提供技术支持，促进GNSS气象学的进一步发展。

近年来，频发的雾霾严重影响了人们的生产和生活，因而雾霾的有效预测已成为亟需研究和解决的问题。针对常规的PM2.5浓度预测模型难以有效处理PM2.5分布的空间非平稳性和空间自相关性问题，基于研究区内的多种数据（气象、大气污染物、大气可降水量（Precipitable Water Vapor，PWV）、天顶对流层延迟、人口密度、交通、高程、归一化是植被指数、气溶胶光学厚度等），构建了多种组合模型进行区域PM2.5浓度的空间插值和时间序列预测。主要研究内容如下：（1）利用区域探空数据和全球大地观测系统大气加权平均温度（Tm）格网数据、ERA5再分析数据、MERRA-2再分析数据，建立了多种高精度的区域性非气象参数Tm模型，以确定利用全球导航卫星系统（Global Navigation Satellite System，GNSS）反演PWV的高精度区域转换系数，进而获取实时高精度的GNSS PWV；（2）采用Spearman和Pearson方法分析与PM2.5相关变量的相关性，依据相关性分析和共线性诊断的结果确定PM2.5浓度的影响变量；（3）以地理加权回归、神经网络等方法为基础，构建了PM2.5浓度的空间插值、时间序列预测、特殊时段预测（节假日、新冠疫情期）等多种组合模型，进行PM2.5浓度空间分布和时间序列预测，并对其预测精度进行了验证和评估，证明了模型的有效性与稳定性。研究成果可为区域GNSS PWV反演、雾霾监测与预警提供技术支持，促进GNSS气象学的进一步发展。