Application of Wavelet Transform to Predict Soil Volumetric Water Content from Surface Measurements.

Numerical models are widely used to simulate soil water content over time and space, but they require information on soil hydraulic properties and data defining conditions at the boundaries of the domain. The purpose of this research was to use a wavelet-based model (WaveReg) to predict soil water content at various depths from observations of water content near soil surface. Soil moisture and meteorological observations were obtained for the Cream Ridge (CR) and Upper Deerfield (UD) sites of the New Jersey Weather and Climate Network and for seven sites from the Soil Climate Analysis Network (SCAN). Soil water contents were obtained from sensors installed between 0.1 and 3 m (CR and UD) and 0.05 to 1.0 m (SCAN) depth from the surface. Continuous soil water content data covering at least 4 years were decomposed into a series of orthogonal components at 47 scales using wavelet transform. Transfer coefficients were obtained using linear regression between the decomposed signals at the surface and each depth for all 47 scales, and used to predict soil moisture at that depth from surface measurements. Principal component analysis was used to group the transfer coefficients of the 47 scales into four groups. The final reconstruction was completed using the average transfer coefficient of each group. Soil hydraulic properties and texture were investigated as predictors of transfer coefficients. WaveReg predicted soil water content measurements with Nash–Sutcliffe efficiency coefficients ranging between 0.06 and 0.94, except at CR site where a low-permeability horizon attenuated the oscillations of soil moisture.