Using a multivariate regression model and hyperspectral reflectance data to predict soil parameters of Agra, India

Abstract

The diffuse reflectance spectroscopy was found to be a promising technique for soil parameters’ estimation. Moreover, the integration of multivariate regression models and Vis-NIR hyperspectral reflectance data proved high efficiency for soil characterization. Thus, this study aimed to estimate pH, ECe and CaCO3 soil parameters using partial least square regression (PLSR) and soil hyperspectral signature. Surface soil samples were collected from Agra, Uttar Pradesh, India. Soil samples were prepared and analyzed for examined parameters. In hyperspectral remote sensing laboratory conditions, soil hyperspectral signatures were collected using an analytical spectroradiometer devise in the spectral range from 350 to 2500 nm. The PLSR model was applied to soil spectra and soil parameters’ data to develop the calibration and validation models. The obtained results showed that pH and CaCO3 parameters were having high predictability whereas R2 values of prediction were 0.69 and 0.83 with RPD values were 1.70 and 2.06, respectively. The PLSR prediction model did not perform well for predicting ECe parameter whereas R2 and RPD values were 0.31 and 1.20, respectively. These techniques can be applied in both laboratory and field conditions by using spectroradiometers. It is rapid, time and cost-effective, and friendly to the environment. Furthermore, it can estimate many soil parameters at the same time with minimum or without samples preparation.