Predicting water content using linear spectral mixture model on soil spectra

Abstract Remote sensing has been widely applied for soil moisture estimation. However, such estimates become difficult to obtain and can be inaccurate when applied to complex earth surfaces with more than one soil type because of the interference of spectral signals from different soil components. This study aims to develop a moisture prediction method that is insensitive to soil types; this is based on in situ samples collected from an intertidal zone in Jiangsu Province in China and on laboratory measurements of soil spectra. The results demonstrate that for a reflectance-based method, moisture content is closely related to reflectance on the three wavebands centered at 2143, 1760, and 742 nm for four types of soil (sand, silty sand, sandy silt, and silt) considered separately; the relationship is not close if all soil types are mixed together ( R 2 = 0.77 ). To develop the desired model, a linear spectral mixture model (LSMM) was employed to extract parameter water abundance (Wa: information on soil water content) in advance, while eliminating redundant information from other soil components. Wa has a relatively higher correlation ( R 2 = 0.82 ) than reflectance with moisture content for a mixed soil type. Thus, employing the LSMM helps realize a practical water content estimation model for predicting moisture over complex earth surfaces, because it has the potential of eliminating spectral effects from soil components.