| Summary: | The large archive of NOAA AVHRR data, covering more than 25 years, bears the potential to detect aerosol trends. However, remote sensing of aerosol optical depth over land is a difficult task, because the aerosol signal gets weaker over bright surface targets and the quality of the retrieval mainly depends on the quality of the surface reflectance estimate. In this study, we test the sensitivity of an existing multitemporal method for the AOD retrieval over land from NOAA AVHRR channel 1 (0,63 μm) radiance. In the original approach the surface reflectance is determined by evaluating the minimum reflectance within a certain moving time period using the concept of the convex hull. Consequently, no a priori knowledge of the bidirectional distribution function (BRDF) and no additional spectral information is necessary. However, the resulting surface reflectance estimate turns out to be especially sensitive to negative short time variations which generally lead to an overestimation of AOD. Using measurements from the Aerosol Robotic Network (AERONET), the initially retrieved AOD can be corrected. With the aim to improve the initial surface reflectance estimate before applying a post-processing technique, we test two other methods (minimum per interval, Rahman-Pinty-Verstraete BRDF model) using different aerosol models and varying time periods. Despite an improved accuracy with the minimum per interval method over bright surface targets, the convex hull scheme turns out to be the most suitable approach considering all different surface types. In addition, combining continental and urban aerosol types enhances the accuracy of the AOD retrieval. We use NOAA-18 AVHRR and AERONET data from nine different sites in Central Europe (40.5°N-50.0°N, 0°E-17°E) and evaluated the results for the year 2006. 1.
|
|---|