Abstract: The complex terrain characteristics of mountainous areas can increase the estimation error of surface reflectance, reduce the accuracy of land-atmospheric decoupling, and then affect the accuracy of aerosol retrieval. Currently, the widely used retrieval methods include the dark target algorithm (DT) based on the band relationship to estimate the surface reflectance, the deep blue algorithm (DB) based on the regional surface reflectance library, and the generalized retrieval of atmosphere and surface properties (GRASP) algorithm based on the bidirectional reflectance distribution function (BRDF) and bidirectional polarization distribution function (BPDF) model to estimate the surface reflectance. To explore the landatmospheric decoupling methods suitable for aerosol remote sensing in mountainous areas of China, the accuracy and applicability of GRASP aerosol optical depth (AOD_G) of polarization and directionality of the earth's reflectances (POLDER-3), DT aerosol optical depth (AOD_DT) and DB aerosol optical depth (AOD_DB) of moderate resolution imaging spectroradiometer (MODIS) in China from 2005 to 2013 were compared and analyzed using AErosol RObotic NETwork (AERONET) aerosol optical depth (AOD_A). The results show that the overall correlation between AOD_G and AOD_A at non-mountainous sites is the highest (with correlation coefficient R = 0.921), and the overall accuracy of AOD_DT and AOD_DB is not much different. However, the proportion of AOD_G higher than the expected error in mountainous areas is 79.87%, and the overestimation proportion of AOD_DT and AOD_DB increases by nearly 30% and 20%, respectively. Seasonal validation at two mountain sites, Xinglong in Hebei Province and Seimi-Arid Climate and Environment Observatory of Lanzhou University (SACOL), shows that the accuracy of all three satellite products tend to decrease in the autumn and winter seasons when the vegetation cover is low, indicating that the mountain topography can affect the accuracy of land-atmosphere decoupling after removing the influence of surface vegetation on reflectance. Further analysis shows that the mountainous terrain has a great influence on the land-atmosphere decoupling method based on BRDF and BPDF model. The multi-angle observation of surface band relationship is more conducive to accurate estimation of surface reflection over undulating surface such as mountainous areas, while there is no significant difference of estimating surface reflection between BRDF/BPDF models and band relationships in urban areas. The results provide a new direction for further optimizing the multi-angle observation (such as Gaofen-5 DPC) aerosol retrieval algorithm in mountainous areas.

Key words: mountainous areas, multi-angle polarization, aerosol optical depth, land-atmospheric decoupling