Remote sensing of albedo over inhomogeneous surfaces is an important topic that involves core problems such as assuring incidence angle, reflection angle and spectrum continuity. Hyper-spectral remote sensing is a good way to investigate the spectral continuity effect on albedo retrieval. In this research, synchronized observations of a wide range of soil and vegetative land cover were performed using the hyper-spectral instrument ASD FieldSpec Pro FRTM. Differences between spectral reflectance and broad albedo were analyzed. Spectral curves show that there are large differences at visible and near-infrared wavelengths; reflectance is low from 400 to 450 nm and from 650 to 700 nm due to strong chlorophyll absorption; reflectance is high, from 750 to 1300 nm due to cell structure reflectance; and reflectance is low from 1360 to1470 nm, 1830 to 2080 nm, and 2350 to 2500 nm, due to water vapor absorption. In addition to the sun zenith angle, chlorophyll, the cell structure of vegetation and water content are shown to be the main factors affecting the spectral reflectance of underlying surfaces, which are determined by the different growth stages and conditions of the vegetation. Therefore, chlorophyll, the cell structure of vegetation and water content are considered to be the important distinguishing indices for describing the spectral reflectance of sparse vegetation and inhomogeneous surfaces. They can be described by the normal difference vegetation index (NDVI), the normal difference vegetation water index (NDWI), and the soil water content capacity (SWCC), respectively.
There is a good negative logarithmic relationship between albedo and SWCC when SWCC is larger than 0.2, and a negative logarithmic relationship and linearity between albedo and both NDVI and NDWI. Based on the relationships, an albedo retrieval model is estimated that relies on the reference spectral reflectance of the vegetation and soil. Spectral wavelengths between 0.3 and 2.35 μm are divided into 12 sections; the broad albedo is estimated by combining the 12 reflectance bands with a weighting relationship. Testing shows that the model can readily reflect the spectral reflectance and albedo, with an estimated value very closed to the observed value. The error in the albedo is less than 0.02, and relative error is from 8% to 13%, which shows that the model is fit for the retrieval of albedo. However, the model still needs to improve. The estimated spectral reflectance at near infrared wavelengths (750-1000 nm) is a little large, and the maximum absolute error is 0.015, which means that the model cannot correspond to the cell structure of the vegetation. The Aster spectrum library provides many kinds of land cover and spectral curves, which can be used for the model application.
As with other land process parameters, the theory of albedo retrieval for inhomogeneous surfaces is a basic problem involving parameterization, scale conversion, and scale coupling. Regions of low vegetation coverage are assumed to be mixed cells, and spectral reflectance can be expressed according to a model that may include reference vegetation, soil spectral reflectance curves, NDVI, NDWI and SWCC. The model determines the effect factor and mechanism, and at the same time, can be dependent on the retrieval of satellite albedo information. In addition, with the development of satellite detection technologies, hyper-spectral remote sensing has good prospects. This work can support the application of hyper-spectral remote sensing techniques to scale conversion and scale coupling of land process parameters over inhomogeneous surfaces.