基于AOD数据和GWR模型估算京津冀地区PM2.5浓度

利用MODIS气溶胶光学厚度（AOD）数据针对不同土地覆盖类型的适用性，提出了一种基于土地覆盖类型的AOD融合方法，生成了一种新的3km AOD数据集.在此基础上，通过地理加权回归（GWR）模型估算了京津冀地区2016年PM_2.5浓度，并用交叉验证的方法对模型性能进行评价.结果表明：利用融合后的AOD数据建立的模型可解释PM_2.594.85%的浓度变化，交叉验证R²为0.94，RMSE为9.27μg/m³，MPE为6.72μg/m³，明显优于多元线性回归（MLR）模型；基于GWR模型估算的京津冀地区2016年年均PM_2.5浓度为58.57μg/m³，其中冬季PM_2.5浓度最高，春秋季次之，夏季浓度最低，PM_2.5月均浓度变化范围32.78~140.83μg/m³，8月份浓度最低，12月份浓度最高；空间分布南北差异显著，衡水市PM_2.5污染最为严重，张家口市PM_2.5浓度较低.利用此方法成功弥补了PM_2.5空间缺失，为城市尺度的健康效应和环境流行病学研究提供数据支持.

Estimation of PM2.5 concentration in Beijing-Tianjin-Hebei region based on AOD data and GWR model

In this study, a MODIS AOD (Moderate Resolution Imaging Spectroradiometer Aerosol Optical Depth) combination method was proposed based on land cover type, and a new 3km AOD data set was generated. PM_2.5 concentration in the Beijing-Tianjin-Hebei region for the year of 2016 was estimated by the Geographical Weighted Regression (GWR) model, and was evaluated by the method of cross-validation. The results showed that the model established by AOD data after combination could explain 94.85% of the PM_2.5 concentration change, with the cross-validation R² of 0.94, the RMSE of 9.27μg/m³, and the MPE of 6.72μg/m³. These values were significantly better than that of the multiple linear regression (MLR) model; Based on the GWR model, average annual PM_2.5 concentration in Beijing-Tianjin-Hebei region was 58.57μg/m³, with the highest concentration of PM_2.5 in winter, followed by spring and autumn, and the concentration in summer was the lowest. The monthly average concentration of PM_2.5 varied from 32.78 to 140.83μg/m³, the lowest and the highest concentrations were estimated in August and in December, respectively. The spatial distribution was significant compared with north part from south, the PM_2.5 pollution in Hengshui City was the most serious, PM_2.5 concentration was the lowest in Zhangjiakou. This method successfully compensated for the lack of PM_2.5 space and provided data support for urban-scale health effects and environmental epidemiological studies.