| 基于CatBoost-LSTM模型的成渝城市群近地面O3浓度估算 |
| 摘要点击 1891 全文点击 152 投稿时间：2024-05-19 修订日期：2024-08-07 |
| 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 近地面O3 Sentinel-5P TROPOMI数据 成渝城市群 CatBoost-LSTM模型 时空分布 |
| 英文关键词 near-surface ozone Sentinel-5P TROPOMI data Chengdu-Chongqing urban agglomeration CatBoost-LSTM model spatial and temporal distribution |
| DOI 10.13227/j.hjkx.20250602 |
|
| 中文摘要 |
| 受臭氧源及影响因子时空差异性的影响，作为大气中重要空气污染物的臭氧（O3）往往呈现出空间异质性和时域相关性. 为提升O3的空间分辨率和估算精度，以成渝城市群为研究区，以O3地面观测站点数据为响应变量，Sentinel-5P TROPOMI离线数据、ERA5再分析气象资料和地形等为解释变量，协同CatBoost和LSTM构建了一种高精度的近地面臭氧浓度估算模型（CatBoost-LSTM模型）. 结果表明：①整体模型中，CatBoost-LSTM模型相较于文中其它模型的估算精度最高，在验证集上的决定系数（R2）、均方根误差（RMSE）和平均绝对误差（MAE）分别为0.965、5.81 μg·m-3和4.42 μg·m-3. ②由于顾及了O3浓度及其影响因子季节上的差异性，基于CatBoost-LSTM的季节模型较CatBoost-LSTM整体模型在验证集上的精度均得到了不同程度的改善，其中冬季模型精度提升最为显著. ③研究区近地面O3月均浓度整体呈倒“V”趋势，其中在5月份出现小幅度下降趋势，8月O3浓度达到最高（89.08 μg·m-3），12月降到最低（29.30 μg·m-3）；近地面O3浓度存在明显的季节性差异，由高到低依次为夏季（84.59 μg·m-3）、春季（72.62 μg·m-3）、秋季（53.59 μg·m-3）和冬季（35.23 μg·m-3）. ④空间分布上，近地面O3浓度高值区主要分布在西部海拔较高、山脉密布、工业活动频繁、交通密集度高、人口密集和污染源较多的地区. 由于工业活动和交通密集度较低，加之相对较少的污染源排放和较为平坦的地形等原因，东部海拔较低地区的O3浓度整体较低. |
| 英文摘要 |
| Due to spatio-temporal differences in ozone sources and influencing factors， ozone （O3）， an important air pollutant in the atmosphere， often exhibits spatial heterogeneity and temporal correlations. To enhance the spatial resolution and estimation accuracy of ozone， a high-precision ozone estimation model （CatBoost-LSTM） was constructed by combining CatBoost and LSTM. The model utilized ground-based ozone observations as response variables， Sentinel-5P TROPOMI offline data， ERA5 reanalysis data， and elevation as explanatory variables. The experimental results of Chengdu-Chongqing urban agglomeration showed that： ① In the overall model， CatBoost-LSTM exhibited the highest estimation accuracy compared to that of other models in the study. The coefficient of determination （R2）， root mean square error （RMSE）， and mean absolute error （MAE） on the verification set were 0.965， 5.81 μg·m-3， and 4.42 μg·m-3， respectively. ② Considering the seasonal differences in ozone concentration and its influencing factors， the CatBoost-LSTM seasonal model demonstrated varying levels of improvement in accuracy on the verification set， with the winter model showing the most significant enhancement. ③ The average monthly near-surface ozone concentration in the study area exhibited an inverted “V” trend， with a slight decrease observed in May， followed by reaching its peak concentration in August （89.08 μg·m-3）， and then declining to its lowest level in December （29.3 μg·m-3）. ④ There were obvious seasonal differences in the near-surface ozone concentration， with higher levels observed in summer （84.59 μg·m-3）， followed by those in spring （72.62 μg·m-3）， autumn （53.59 μg·m-3）， and winter （35.23 μg·m-3）. ⑤ In terms of spatial distribution， the areas with higher near-surface ozone concentration were mainly distributed in the western regions characterized by higher elevations， dense mountains， frequent industrial activities， high traffic density， dense population， and numerous pollution sources. Due to the limited intensity of industrial activities and traffic， as well as the relatively minor emissions from pollution sources and the flat topography， the overall ozone concentration in the eastern region at lower altitudes was comparatively lower. |
