| Summary: | Effective monitoring of soil moisture (θ) by non-destructive means is important for crop irrigation management. Soil bulk density (ρ) is a major factor that affects potential application of θ estimation models using remotely-sensed data. However, few researchers have focused on and quantified the effect of ρ on spectral reflectance of soil moisture with different soil textures. Therefore, we quantified influences of soil bulk density and texture on θ, and evaluated the performance from combining spectral feature parameters with the artificial neural network (ANN) algorithm to estimate θ. The conclusions are as follows: (1) for sandy soil, the spectral feature parameters most strongly correlated with θ were S<sub>g</sub> (sum of reflectance in green edge) and A_Depth<sub>780–970</sub> (absorption depth at 780–970 nm). (2) The θ had a significant correlation to the R<sub>900–970</sub> (maximum reflectance at 900–970 nm) and S<sub>900–970</sub> (sum of reflectance at 900–970 nm) for loamy soil. (3) The best spectral feature parameters to estimate θ were R<sub>900–970</sub> and S<sub>900–970</sub> for clay loam soil, respectively. (4) The R<sub>900–970</sub> and S<sub>900–970</sub> showed higher accuracy in estimating θ for sandy loam soil. The R<sub>900–970</sub> and S<sub>900–970</sub> achieved the best estimation accuracy for all four soil textures. Combining spectral feature parameters with ANN produced higher accuracy in estimating θ (R<sup>2</sup> = 0.95 and RMSE = 0.03 m<sup>3</sup> m<sup>−3</sup>) for the four soil textures.
|
|---|