Estimation of Some Soil Properties Using Spectral Data Analysis (Vis-NIR) and Various Pre-Processing Methods

Soil spectroscopy has overcome many limitations of conventional soil analysis methods due to its rapid, accurate, cost-effective, and non-destructive nature. This study was aimed to investigate the capability of soil spectral data in estimating some key soil properties and comparing different spectral preprocessing methods in determining the performance of the partial least squares regression (PLSR) model. For this purpose, 100 soil surface samples were collected from the study area which was located between Neyriz and Estahban regions in the east of Fars Province. The samples were analyzed for organic carbon (OC), electrical conductivity (EC), calcium carbonate equivalent (CaCO3) and gypsum using standard laboratory methods. Then, the spectral reflectance of the soil samples was recorded in the range of 350-2500 nm and various spectral pre-processing methods were applied to the data. Afterwards, the soil properties were estimated using PLSR. The results indicated the desirable capability of PLSR method in estimating the amount of gypsum (RPD >2, R2 = 0.81, RMSE = 3.87) and its acceptable ability for OC, EC and CaCO3 (2< RPD < 1.4). Also, the best modeling systems for OC, gypsum and EC were obtained as the first derivative with Savitzky-Golay smoothing method (FD-SG), the standard normal variate method (SNV), and the second derivative with SG smoothing method (SD-SG), respectively. Besides, non-preprocessing data of soil CaCO3 provided better estimations than various pre-processing methods. Overall, the results revealed that the visible spectrum range provided the best performance for estimating of OC and EC, and the NIR range for CaCO3 and gypsum.