|摘要: ||大氣的氣膠不僅模糊了地表的資訊,亦嚴重地影響大氣的環境及氣候 的變遷,在地球資源的遙測與大氣環境的變遷是相當重要的參數之一,因 此有許多利用衛星資料求取氣膠參數的方法陸續地被提出。本文之研究重 點即在於離散係數法(Dispersion coefficient method, DCM)之測試改進與應 用,利用臺北市與北京市十三年的 Landsat7 資料,探討離散係數反演法之 特性,並針對其主要誤差來源進行分析與改進,以增進此方法在氣膠參數 反演之準確性與適用性,包括太陽入射角的影響、地表反射之正規化移動 室窗之大小以及影像間幾何定位等重要影響因子。最後則以 AERONET 地 面測站資料檢驗本研究在離散係數法改進之成效。|
研究結果顯示,在波段之平均反射率作為離散係數法之正規化參數之 選擇上,利用短波紅外比可見光及近紅外波段好。其誤差差異在經過太陽 入射角之調整後平均可達 53.5%。在反演視窗大小之選擇上,以大於 7x7 像元視窗大小之反演結果趨向穩定且準確度最佳,如藍光頻道在 9x9 像元 視窗之誤差為 28%。綠光頻道在 7x7 像元視窗的誤差為 29%。另外,紅光 及近紅外光頻道反演之結果由於誤差較大而不適合應用於此方法反演氣膠 光學厚度。;Atmospheric Aerosol Optical Depth (AOD) isn’t burning the information from surface but also impact the environment of atmosphere and climate change. AOD is one parameter as an important character in the field of remote sensing of earth resources and the change of atmospheric environment. This study focuses on Dispersion coefficient method (DCM), which availability, improvement and application. The effect from solar zenith angle and the normalization with visible, near infrared (IR) and short wave infrared (SWIR) band were investigate for DCM retrieval. These results were validated by AERONET ground station and the error range for this method will examined. In addition, SWIR indicated highest sensitivity in solar zenith angle when compare to visible, IR and SWIR bands for DCM normalization. The results show that the AOD spatial distribution from DCM can contribute the region’s air quality monitoring.
This research conducts 13 years data from sunphotometer and LANDSAT-7 ETM+ for AOD retrievals in Taipei and Beijing. The main error sources for DCM’s were analyzed and improved. For the normalize band, the results indicated that using SWIR band was better than other spectral bands. The mean error different was about 53.5%. By using retrieval window, we conclude that the window size after 7x7 tend to be stable. The best results for blue and green band were 9x9 pixel window size and 7x7 pixel window size whereas the error is 28% and 29% respectively. Red and near-infrared bands had the high mean error, therefore not suitable for this method.