高分辨率卫星遥感影像灰霾颗粒物反演算法与估算模型研究

The retrieval of haze aerosol optical depth (AOD) in China is not included the mainstream satellite remote sensing product due to the misidentification of haze, and the inappropriate assumptions about the surface reflectance and the aerosol model. Many uncertainties such as aerosol vertical distribution, hygroscopic growth, satellite resolution and spatial instability have a large impact on the relationship between surface-level PM2.5 and the satellite-retrieved column AOD. This approach first studies the spectral characters of cloud, fog, haze, and land cover pixels, and develops a haze detection algorithm based on the thresholds of apparent reflectance and brightness temperature, and the spatial distribution from China’s high-resolution satellite data. Second, the alternative Normalized Difference Vegetation Index (NDVI) and the Bidirectional Reflectance Distribution Function (BRDF) are used to separate the surface contribution over the clear days’ dense dark, and haze days or urban bright targets, respectively. Since lacks of enough ground-based observation, the aerosol vertical distribution, particle component and relative humidity are extracted from the atmospheric chemistry model data to build the high spatial and temporal aerosol model, and vertical-humidity correction method. In order to better reflect the local variation of PM2.5 fitting parameters, the spatial structure is embedded in the linear regression equation to build the Geographically Weighted Regression (GWR) model. Finally, long-term haze distribution, AOD and PM2.5 products in China will be generated and validated by the Aerosol Robotic NETwork (AERONET) and ground-based PM2.5 observations.

目前国外主流遥感产品由于无法解决霾识别、地表贡献与气溶胶模式等问题，缺乏我国霾天的气溶胶光学厚度(AOD)。AOD-PM2.5之间的关系也受垂直分布、湿度、卫星分辨率和空间不稳定性等因素影响。本项目基于国产高分卫星系列数据，首先通过研究不同像元在敏感通道的光谱特性，构建基于反射率、亮温阈值和空间分布特征的霾识别算法。其次采用植被指数、耦合地表二向反射等方法分别进行晴天暗像元、雾霾和城市等亮目标区域的地表噪声去除。针对地基数据不足的问题，AOD反演过程中所需的气溶胶模型，以及垂直和吸湿订正模型所需的气溶胶廓线和相对湿度等数据通过大气化学模式获取。为了更好的反映PM2.5拟合系数的局地变化特征，构造PM2.5地理加权回归计算模型。最后，基于AERONET和PM2.5地基观测数据对反演和估算结果进行验证，在此基础上实现我国长时间序列灰霾分布、AOD和PM2.5的生产。

本课题首先基于国产高分5号卫星多角度偏振传感器、及其类似的国外载荷，开展了云雾霾等不同像元的光谱特征分析，并构建了耦合偏振信息的像元识别算法。其次，通过研究国产高分一号、风云三号等载荷的波段响应函数，构建了红蓝波段的地表反射率经验关系、以及蓝波段的地表反射率库，分别实现了在浓密植被和城市亮目标等情况下的气溶胶光学厚度（AOD）反演。再次，借助主动激光雷达提供的气溶胶廓线信息，对柱浓度的AOD进行分层校正，获取了近地面的消光系数。通过耦合气象参数，构造了二阶段的广义相加空间统计模型，实现了PM2.5的精确估算。在大量地面数据验证的基础上，本课题生产了成渝地区和中原经济区等典型区域的长时间序列的大气遥感产品，并深入分析了上述区域的年际、季度和空间分布特征。本课题相关技术挖掘了国产卫星数据在大气遥感领域的应用潜力，能够为我国大气污染防治提供技术和数据支撑。