Estimating purple-soil moisture content using Vis-NIR spectroscopy

Soil moisture is essential for plant growth in terrestrial ecosystems. This study investigated the visible-near infrared (Vis-NIR) spectra of three subgroups of purple soils (calcareous, neutral, and acidic) from western Chongqing, China, containing different water contents. The relationship between soil moisture and spectral reflectivity (R) was analyzed using four spectral transformations, and estimation models were established for estimating the soil moisture content (SMC) of purple soil based on stepwise multiple linear regression (SMLR) and partial least squares regression (PLSR). We found that soil spectra were similar for different moisture contents, with reflectivity decreasing with increasing moisture content and following the order neutral > calcareous > acidic purple soil (at constant moisture content). Three of the four spectral transformations can highlight spectral sensitivity to SMC and significantly improve the correlation between the reflectance spectra and SMC. SMLR and PLSR methods provide similar prediction accuracy. The PLSR-based model using a first-order reflectivity differential (R’) is more effective for estimating the SMC, and gave coefficient of determination root mean square errors of validation (RMSEV), and ratio of performance to inter-quartile distance (RPIQ) values of 0.946, 1.347, and 6.328, respectively, for the calcareous purple soil, and 0.944, 1.818, and 6.569, respectively, for the acidic purple soil. For neutral purple soil, the best prediction was obtained using the SMLR method with R’ transformation, yielding R2v, RMSEV and RPIQ values of 0.973, 0.888 and 8.791, respectively. In general, PLSR is more suitable than SMLR for estimating the SMC of purple soil.