空间分辨率与精度协同改进的卫星AOD产品降尺度模型

针对现有卫星气溶胶光学厚度(AOD)产品空间分辨率和精度往往难以满足大气污染精细治理实际需求,提出了一种耦合偏差校正的统计降尺度改进模型(SDBC).该模型基于“空间尺度不变性假设”引入相关驱动因子的额外空间信息实现AOD降尺度,并在此基础上通过偏差校正进一步提升降尺度产品的精度.以1km分辨率MAIAC AOD产品为例,在北京、大湾区、台湾岛3个典型地区开展模型验证.结果表明:(1)DEM、NDVI、人口数量和土地覆盖是影响AOD变化的细节因子,在SDBC空间降尺度过程中引入可将AOD产品的空间分辨率有效提升至500m,且降尺度产品验证R²最高可达0.88;(2)顾及卫星观测几何、质量标识、大气水蒸气柱、气溶胶模式等因子的偏差校正则可进一步提升降尺度AOD产品的精度,3个地区的验证R²均在0.85以上,最高可达0.93;(3)信息熵评估结果显示SDBC模型生成的500m AOD产品提高了原始MAIAC AOD产品的空间信息量.在保留了公里级产品AOD的空间分布格局的基础上,SDBC产品也增强了细节和纹理特征、改善了边界现象和马赛克效应.研究结果证实SDBC模型能有效协同改进现有卫星AOD产品的空间分辨率和精度,提升我国大气污染遥感精准监测的业务能力.

A downscaling model for satellite AOD product improvement in spatial resolution and accuracy

The spatial resolution and accuracy of existing aerosol optical depth (AOD) products cannot satisfy the demand for fine-scale air pollution control. To make up for such deficiencies, this study proposed a novel modeling approach named Statistical Downscaling model combined with Bias Correction (SDBC). Based on the hypothesis of "spatial scale invariance", this model introduced additional spatial information of driving factors to downscale AOD products and further improved the accuracy of downscaled results using bias correction. Take 1-km resolution MAIAC AOD products as an example, we examined the proposed model in three typical areas: Beijing, Greater Bay Area, and Taiwan Island. Results showed that: (1) Digital elevation model, normalized difference vegetation index, population, and land cover were the fine driving factors affecting AOD variations. Taking these factors into consideration, the spatial downscaling model can effectively improve the spatial resolution (1km) of the original product to 500m, and the highest validated R² was up to 0.88. (2) In addition, the accuracy of downscaled AOD products can be further improved by bias correction coupling with satellite observation geometry, quality flag, atmospheric water vapor column, aerosol model, and other factors. The validated R² of the three areas were all larger than 0.85, and the highest was 0.93. (3) The information entropy evaluation results showed the 500m AOD product generated by the SDBC model increased the spatial information of the original MAIAC AOD product. Based on retaining the AOD spatial distribution pattern of the MAIAC AOD product, the details and texture features were enhanced, and the boundary phenomenon and mosaic effect were also eliminated. These results confirm that the SDBC model can effectively improve the spatial resolution and accuracy of existing AOD products simultaneously, which can lift the operational capability of remote sensing precision monitoring of atmospheric pollution in China.