Remote sensing of aerosols over boreal forest and lake water from AVHRR data

A complete set of advanced very high resolution radiometer (AVHRR) data and ground-based measurements of aerosol and water-vapor content are used to test an algorithm for the retrieval of aerosol properties over dense vegetation in the red and over lake water in both the red and the near-infrared AVHRR channels. With the assumptions of a weak and reasonably constant surface reflectance and an appropriate aerosol model in the radiative transfer code, the remaining variance in the satellite signal is interpreted in terms of aerosol optical thickness. From theoretical computations, it appears that the algorithm is particularly sensitive to the surface albedo and that an uncertainty of 0.01 in reflectance leads to an error of +/-0.1 in the retrieved optical thickness. This theoretical estimate is confirmed by data acquired over a boreal forest region in Canada and over one of the Great Lakes (Ontario). In particular, channel 1 observations over vegetation in the forward scattering direction are well suited for retrievals because vegetation pixels appear darker owing to shadowing effects. Conversely, the forward scattering geometry over lakes introduces large errors in both channels owing to specular reflections (glint effects). Even for observations well removed from the forward scattering principal plane, lake surface reflections due to sky radiance glint have to be taken into account. Because the accuracy of the retrieval algorithm is affected by water-vapor absorption in channel 2 and by variations in lake-water turbidity in channel 1, the optimal retrieval configuration is to employ vegetation observations in channel 1. Bidirectional effects have to be considered, however, for observations in the backscatter directions. (C) Elsevier Science Inc., 1997.