Prediction of soil macro- and micro-elements in sieved and ground air-dried soils using laboratory-based hyperspectral imaging technique

Hyperspectral image analysis in laboratory-based settings has the potential to estimate soil elements. This study aimed to explore the effects of soil particle size on element estimation using visible-near infrared (400–1000 nm) hyperspectral imaging. Images were captured from 116 sieved and ground soil samples. Data acquired from hyperspectral images (HSI) were used to develop partial least square regression (PLSR) models to predict soil available aluminum (Al), boron (B), calcium (Ca), copper (Cu), iron (Fe), potassium (K), magnesium (Mg), manganese (Mn), sodium (Na), phosphorus (P) and zinc (Zn). The soil available Al, Fe, K, Mn, Na and P were not predicted with high precision. However, the developed PLSR models predicted B (R2 CV = 0.62 and RMSECV = 0.15), Ca (R2 CV = 0.81 and RMSECV = 260.97), Cu (R2 CV = 0.74 and RMSECV = 0.27), Mg (R2 CV = 0.80 and RMSECV = 43.71) and Zn (R2 CV = 0.76 and RMSECV = 0.97) in sieved soils. The PLSR models using reflectance of ground soil were also developed for B (R2 CV = 0.53 and RMSECV = 0.16), Ca (R2 CV = 0.81 and RMSECV = 260.79), Cu (R2 CV = 0.73 and RMSECV = 0.29), Mg (R2 CV = 0.79 and RMSECV = 45.45) and Zn (R2 CV = 0.76 and RMSECV = 0.97). RMSE of different PLSR models, developed from sieved and ground soils for the corresponding elements did not significantly differ based on the Levene's test. Therefore, this study indicated that it was not necessary to grind soil samples to predict elements using HSI. © 2018 Elsevier B.V.