| 基于GTWR-XGBoost模型的四川省PM2.5小时浓度估算 |
| 摘要点击 1740 全文点击 455 投稿时间：2022-07-17 修订日期：2022-09-12 |
| 查看HTML全文
查看全文 查看/发表评论 下载PDF阅读器 |
| 中文关键词 PM2.5 Himawari-8 AOD GTWR-XGBoost模型 机器学习 时空分布 |
| 英文关键词 PM2.5 Himawari-8 AOD GTWR-XGBoost model machine learning spatial and temporal distributions |
|
| 中文摘要 |
| 卫星气溶胶光学厚度（AOD）和气象数据已被广泛用于估算空气动力学直径≤2.5 μm的地表颗粒物（PM2.5）浓度.研究高时间分辨率、高精度的PM2.5浓度估算方法，对及时准确的空气质量预报和大气污染的预防及缓解具有重要意义.使用Himawari-8 AOD小时产品和ERA5气象再分析资料作为估算变量，提出GTWR-XGBoost组合模型，对四川省PM2.5小时浓度进行估算.结果表明：①提出的组合模型运用于全数据集的性能优于KNN、RF、AdaBoost、GTWR、GTWR-KNN、GTWR-RF和GTWR-AdaBoost模型，拟合精度指标R2、MAE和RMSE分别为0.96、3.43 μg ·m-3和5.52 μg ·m-3，验证精度指标R2、MAE和RMSE分别为0.9、4.98 μg ·m-3和7.92 μg ·m-3.②该模型作用于PM2.5浓度小时估算上，具有较高的拟合优度（全数据集的R2为0.96，不同时刻的R2介于0.91~0.98之间），表明该模型对于小时估算具有较好的时间稳定性，能为区域空气质量评估提供精准的估算信息.③在时间上年均PM2.5小时浓度估算变化总体上呈现"倒U"型趋势，09：00开始升高至11：00达到峰值［ρ（PM2.5）为44.56 μg ·m-3］后逐渐降低；且季节性变化非常明显，呈现冬季>春季>秋季>夏季的趋势.④在空间分布上总体呈现东高西低和局部污染程度较高的特点，高值区主要分布在城市快速发展和人口密集的东部地区，低值区主要分布在经济发展落后和人口稀疏的西部地区. |
| 英文摘要 |
| Aerosol optical depths of satellites and meteorological factors have been widely used to estimate concentrations of surface particulate matter with an aerodynamic diameter ≤ 2.5 μm. Research on a high time resolution and high-precision PM2.5 concentration estimation method is of great significance for timely and accurate air quality prediction and air pollution prevention and mitigation. Himawari-8 AOD hour product and ERA5 meteorological reanalysis data were used as estimation variables, and a GTWR-XGBoost combined model was proposed to estimate hourly PM2.5 concentration in Sichuan Province. The results showed that:① the performance of the proposed combination model was better than that of the KNN, RF, AdaBoost, GTWR, GTWR-KNN, GTWR-RF, and GTWR-AdaBoost models in the full dataset; the fitting accuracy indexes R2, MAE, and RMSE were 0.96, 3.43 μg·m-3, and 5.52 μg·m-3, respectively; and the verification accuracy indexes R2, MAE, and RMSE were 0.9, 4.98 μg·m-3, and 7.92 μg·m-3, respectively. ② The model had a high goodness of fit (R2 of the whole dataset was 0.96, and R2 of different times ranged from 0.91 to 0.98) when applied to the estimation of PM2.5 concentration hour. It showed that the model had good time stability for hourly estimation and could provide accurate estimation information for regional air quality assessment. ③ In terms of time, the annual average PM2.5hourly concentration estimation showed an inverted U-shaped trend. It began to increase gradually at 09:00 am to a peak of 44.56 μg·m-3 at 11:00 and then gradually decreased. Moreover, the seasonal variation was very obvious, with winter>spring>autumn>summer. ④ In terms of spatial distribution, it showed the characteristics of high in the east and low in the west and a high degree of local pollution. |
|
|
|
