Comparison of two methods for estimation of soil water content from measured reflectance

Li, Y., Liu, S., Liao, Z. and and He, C. 2012. Comparison of two methods for estimation of soil water content from measured reflectance. Can. J. Soil Sci. 92: 845–857. Remote sensing (RS) technology has increasingly been used in soil water content estimation, but highly accurate estimates of soil water content are still difficult to obtain using this technique. This study aims to determine the wavelengths at which the reflectance is most sensitive to changes in soil water contents (θ). Four types of soils were selected and light reflectance (λ) was measured at different θ. Results showed that parabolic functions fit θ with measured λ very well but only for individual wavelengths. Multivariate linear functions of θ with measured λ at visually selected characteristic wavelengths led to improved predictions, but the coefficients of determination between the soil water content and measured reflectance (R² ranged from 0.788 to 0.925 for the four soils) were still not high for the studied soils. Stepwise multiple linear regressions between θ and measured λ showed higher coefficients of determination (R² increased to 0.99 when the number of the statistically selected wavelengths increased) than the multiple regression, but had lower coefficients of determination than the stepwise multiple linear regressions between θ and the normalized band depths (Dₙ). The multi-variable linear functions fitted the measured θ vs. Dₙ best with much higher R² values, even when a single wavelength was used. Re-sampling wavelengths of less than 20 nm preserved the main features of the original reflectance for the studied soils. Parameters were fitted for the quadratic functions using the re-sampled reflectance data at 20-nm wavelength intervals for further estimation of soil water content, which was considered potentially applicable in RS technology for estimating soil water content provided that soils are relatively homogeneous. In conclusion, stepwise multiple linear regression functions between θ vs. Dₙ are statistically precise and are recommended for estimating soil water content from reflectance. Parameters for quadratic functions relating soil water content to the observed reflectance could be potentially used in RS technology for estimating soil water content.