| 摘要: | 氣膠光學厚度 (Aerosol Optical Depth, AOD) 為表示空氣汙染程度的指標之一，透過衛星觀測能有效偵測大範圍的氣膠光學厚度。部分氣膠具有短生命週期以及區域性汙染的特性，然而以現有技術來說並無單一衛星能同時提供高空間且高時間解析度的資訊。透過時空影像融合法將具有不同優勢的影像進行融合，能獲得高空間及高時間解析度的資訊，如自適應時空反射率融合模式 (Spatial Temporal Adaptive Reflectance Fusion Model , STARFM)。然而STARFM僅適用於地表的相關訊息，無法提供大氣之訊息。為了獲取高時空解析度之氣膠資訊，本研究針對地球同步衛星向日葵八號 (Himawari-8) 以及繞極軌道衛星大地衛星8號 (Landsat-8) 之影像進行融合，並改進STARFM，進而得到高時空解析度的大氣層頂反射率影像，應用於氣膠光學厚度之反演。另外，本篇研究使用同時輻射率定法 (Simultaneous Radiation Solution, SRS) 進行氣膠光學厚度之反演，但該方法並不適用於高地表反射率的地區。為了突破此限制，本研究亦對同時輻射率定法進行修正，再結合影像融合得到的影像，反演具高時空解析度之氣膠光學厚度資訊，以掌握具有短生命週期且地區性汙染的氣膠特性。
研究結果應用於三個在台灣的大氣汙染事件中。反演結果在空間分佈上與NASA/MODIS 暗物法 (Dark Target) 3公里解析度氣膠光學厚度產品相當一致。透過AERONET (AErosol RObotic NETwork) 的資料進行驗證後，顯示高時空解析度SRS反演之氣膠光學厚度於三個個案中分別有63%、75%以及80%的反演結果落於期望誤差之內。與個別的AERONET測站比較下來相對誤差也大多小於國際上標準的20%。然而在部分特定區域中，反演結果相對較差。初步分析後發現主要原因來自地表反射率以及雙向反射特性的掌握不夠準確。若能建立各種地表種類下的雙向反射分佈函數 (Bidirectional Reflectance Distribution Function, BRDF)，將可能提升高時空解析度SRS反演氣膠光學厚度的精準度。
;Aerosol Optical Depth (AOD) is an important indicator of air quality. Through satellite observation, we can obtain comprehensive information on AOD in broad spatial distribution. However, since the characteristics of aerosols are both short-lived and regional, and that getting high spatial and temporal AOD information by single satellite observation is not feasible as well. Adopting the spatial-temporal image fusion technique, like Spatial-Temporal Adaptive Reflectance Fusion Model (STARFM), becomes one of the desirable approaches to deal with the dilemma.
Nevertheless, this model was designed for image fusion on surface reflectance data. Hence, revising the STARM model to retain the information from the atmosphere is necessary if we want to apply it on air quality monitoring. Moreover, we use an algorithm called Simultaneous Radiation Solution (SRS) to retrieve AOD. To correct the limitation of SRS, which is only applicable for low surface reflectance area, we modify the SRS in this research and further apply it on a higher surface reflectance area. In short, we get high spatial and temporal AOD information through STARFM and SRS.
After applying the new high spatial and temporal algorithm on three air pollution cases in Taiwan, the spatial distribution of the results correspond with the MODIS Dark Target AOD product in 3km resolution. We further use the AERONET data to validate our retrieval, and it shows that there are 63%, 75% and 80% of retrieved AOD for each case located in expected error respectively. However, a significant error appears in some specific areas. By preliminary analysis, we assume that the error comes from the wrong estimation of surface reflectance and poor handling on bidirectional reflectance distribution function (BRDF) in this research. If the BRDF for different land cover types is constructed, the retrieval of AOD should be improved. |
