摘要
通过卫星遥感反演的气溶胶光学厚度(Aerosol Optical Depth,AOD)数据存在显著的覆盖缺失问题,而现有的插补方法对AOD的地理时空异质性考虑不足,严重影响了插补关系的时空建模可靠性。该文在集成学习模型的非线性关系拟合基础上,针对遥感AOD的地理时空异质性特点,从地理时空位置特征构建与插补关系建模两方面入手,提出一种考虑地理时空异质性的AOD插补方法。该方法结合了地理时空位置编码策略和极端梯度提升树模型,有效解决了地理时空位置特征变化不均匀和非线性关系建模的问题,并在2019年中国区域日均1 km的AOD数据插补实验中取得了88.67%的拟合精度,与地面实测数据的相关性达0.842,表明该方法具有较高的插补精度及可靠性,可为空气污染的科学防治提供数据与方法支持。
The retrieval of aerosol optical depth(AOD)through satellite remote sensing provides an effective approach to comprehensively and promptly monitor the atmospheric environment,which is of great significance for regional air pollution control.However,AOD suffers significantly from non-random missing values and the existing gap-filling methods do not take into account its spatiotemporal heterogeneity,seriously affecting the reliability of spatiotemporal modeling of the gap-filling relationship.In this paper,a new AOD gap-filling method considering spatiotemporal heterogeneity is proposed based on the non-linear relationship fitting of machine learning models,which includes the location feature construction and gap-filling relationship modeling.The proposed method combines geographical spatiotemporal encoding strategy with extreme gradient boosting tree model,effectively solving the problem of uneven changes in geographic spatiotemporal location features as well as non-linear relationship modeling.It achieved 88.67% fitting accuracy and 84.20% correlation with ground-based measured data in the gap-filling experiment for China's regional data with the resolution of daily average 1 kilometer in 2019.The experimental results show that the proposed method has high gap-filling accuracy and reliability,providing support for the scientific prevention and control of air pollution.
作者
周涛
黄波
ZHOU Tao;HUANG Bo(Faculty of Geosciences and Environmental Engineering,Southwest Jiaotong University(SWJTU),Chengdu 611756;Department of Geography,The University of Hong Kong,Hong Kong 999077,China)
出处
《地理与地理信息科学》
CSCD
北大核心
2023年第4期29-36,共8页
Geography and Geo-Information Science
基金
国家自然科学基金项目(42271439)。
关键词
气溶胶光学厚度(AOD)
空气污染
时空插补
时空异质性
极端梯度提升树
aerosol optical depth(AOD)
air pollution
spatiotemporal gap-filling
spatiotemporal heterogeneity
extreme gradient boosting tree
