High-resolution Aerosol Optical Depth Retrieval And Its Aplication Over Urban Areas

With the social and economic continue developing rapidly,and the urbanization deepening,atmospheric particulate matter pollution is receiving increasing attention in China.Aerosol is a gaseous dispersion system composed of solid and liquid particles suspended in a gaseous medium,impacting both global radiation balance and mental health.Particulate matter with aerodynamic diameter less than 10μm（PM10）is of crucial importance for air quality monitoring and public environmental policy making,and therefore has aroused widespread concern.Satellite remote sensing has been an important technology for monitoring air pollution for its wide coverage,short revisit period,and low cost.However,due to available aerosol products are at low resolution（≥1 km）,further studies of aerosol and air pollution below city scales are untainable.Based on the high-resolution Sentinel-2 data,the developing of aerosol optical depth（AOD）retrieval algorithms and the application of the retrieved Sentinel-2 AOD are carried out in this dissertation.The main results and innovations of the dissertation are discribled below.1.A high-resolution AOD retrieval algorithm based on Sentinel-2 data over urban areas is developed.In the vegetated areas,the surface reflectance relationships between visible and shortwave infrared channels in the dark target algorithm are improved.In the bright areas,a simple method for preconstructing the high-resolution suface reflectance database is proposed.Combining the obtained surface reflectance and the look-up table generated by the 6SV2.1 model,AOD at 60 m are retrieved.Furthermore,difficulties in preconstructing surface reflectance database with low-temporal resolution surface reflectance products and in estimating surface reflectance over complex urban areas are solved.Validated against the AERONET AOD,the achieved Sentinel-2 AOD have a correlation coefficient of 0.882 and an expected error（EE）of 73.85%.Compared with kinds of MODIS aerosol products,the Sentinel-2 AOD has higher or comparable accuracy.The validation and comparison results suggest that the proposed algorithm is able to retrieve reliable AOD over urban area.2.An AOD retrieval algorithm based on the surface reflectance ratio and time-series signal is proposed.The proposed algorithm is able to retrieve high-resolution（60 m）AOD over urban areas without the directly estimation of surface reflectance by improving the background AOD value and the aerosol model in the study area,as well as preconstructing seasonal red-blue surface reflectance ratio database and minimum apparent reflectance database of Sentinel-2 blue channel.The non-smooth distribution in high-resolution AOD images is also solved by the proposed algorithm.The validation results show that the Sentinel-2 AOD has a correlation coefficient of 0.927 with the AERONET AOD and an EE value of 77.31%.Compared with a variety of MODIS aerosol products,the Sentinel-2 AOD not only has richer spatial information,but also superior in data quality.3.The influence of land cover types on aerosol below city scales is conducted.With the achieved Sentinel-2 AOD and 100 m surface classification product CGLS-LC100,the impact of land cover types on aerosol below city scales is analyzed to promote the understanding of the spatialtemporal distribution patterns of aerosol in Beijing.The analysis results find that AOD distribution varies both annually and seasonaly.Compared with suburban areas,urban areas tend to produce higher AOD.Usually,land cover type with dense vegetation and low building density prefers to have low AOD,and vice versa.Among the five major land cover types in the study area,urban and built-up land has the highest AOD,then followed by cropland and open shrubland,and finally grassland and forest.Urban and build-up land most positively contributes to AOD,whereas forest is fully opposite.Both positive and negative contributions peak in spring.4.A model based on model-agnostic meta learning（MAML）for estimation the high-resolution PM₁₀ over urban areas is proposed.Considering the PM₁₀-AOD relationship varies timely,a day-based drawing method of model task is used.After training the MAML model with numbers of historical tasks consisting of low-resolution data,high-resolution PM₁₀ are then estimated from unseen task with few Sentinel-2 AOD.The proposed model solves problem of indirect monitoring of ground-level PM₁₀ below city scales with insufficient high-resolution AOD data.Validated against the ground-based PM₁₀,the PM₁₀ estimated by the MAML model have mean absolute error and root mean square error of 16.95μg/m³ and 23.39μg/m³.Compared with the PM₁₀ estimated by the baseline,PM₁₀ estimated by the MAML model have higher data quality.Additionally,the MAML model takes less fine-tuning steps than the baseline.The validation and comparison results report that MAML is capable of estimating reliable PM₁₀ with few samples.