Geographically Weighted Regression (GWR) is a common method to estimate mass concentrations of fine particulate matter (PM2.5). However, some shortage like spatial resolution of the raster input model still exists widely in the model. Therefore, based on GWR model, we adopted spatial downscaling (SD) method to solve this problem. GWR and SD were constructed by using Aerosol Optical Depth remote sensing data, GEOF meteorological grid data of the Goddard Earth Observing System, and PM2.5 data from the ground environmental monitoring station. In this study, GWR and SD were used to estimate monthly PM2.5 mass concentrations of the Beijing–Tianjin–Hebei (BTH) region in 2017. The results showed that: the average annual PM2.5 in 2017 estimated by GWR and SD had the characteristics of high in the south and low in the north with the boundary of 40°N in the spatial distribution. We found that the natural proximity method was the optimal choice for the treatment of residual values through verification of the estimated results. At the 95% confidence level, the determination coefficient R2 is 0.903, the mean prediction error is 7.307 μg/m3, the root mean square error is 11.62 μg/m3, and the relative prediction error is 18.35%. These results suggest that the GWR and SD method could objectively estimate PM2.5 mass concentrations in BTH region in 2017 and process raster data into the same spatial resolution.

中文翻译：

---

基于地理加权回归和空间降尺度法的京津冀地区PM2.5质量浓度估算

地理加权回归 (GWR) 是估算细颗粒物 (PM2.5) 质量浓度的常用方法。然而，该模型中仍广泛存在栅格输入模型的空间分辨率等不足。因此，基于 GWR 模型，我们采用空间降尺度 (SD) 方法来解决这个问题。GWR和SD是利用气溶胶光学深度遥感数据、戈达德地球观测系统GEOF气象网格数据和地面环境监测站PM2.5数据构建的。本研究使用 GWR 和 SD 估算了 2017 年京津冀（BTH）地区的月 PM2.5 质量浓度。结果表明：年均 PM2.5 质量浓度为 GWR和SD估计的2017年5号在空间分布上具有南高北低的特征，以40°N为界。通过对估计结果的验证，我们发现自然邻近法是处理残差值的最佳选择。在95%置信水平下，决定系数R2为0.903，平均预测误差为7.307 μg/m3，均方根误差为11.62 μg/m3，相对预测误差为18.35%。这些结果表明，GWR 和 SD 方法可以客观地估计 2017 年 BTH 地区 PM2.5 的质量浓度，并将栅格数据处理成相同的空间分辨率。在95%置信水平下，决定系数R2为0.903，平均预测误差为7.307 μg/m3，均方根误差为11.62 μg/m3，相对预测误差为18.35%。这些结果表明，GWR 和 SD 方法可以客观地估计 2017 年 BTH 地区 PM2.5 的质量浓度，并将栅格数据处理成相同的空间分辨率。在95%置信水平下，决定系数R2为0.903，平均预测误差为7.307 μg/m3，均方根误差为11.62 μg/m3，相对预测误差为18.35%。这些结果表明，GWR 和 SD 方法可以客观地估计 2017 年 BTH 地区 PM2.5 的质量浓度，并将栅格数据处理成相同的空间分辨率。