Hyperspectral remote sensing-based prediction of the soil pH and salinity in the soil to water suspension and saturation paste extract of salt-affected soils of the west coast region

Abstract

The study aimed to investigate hyperspectral remote sensing to predict the soil pH and salinity in the soil to water suspension and saturation paste extract of salt-affected soils of the west coast region. About 216 soil samples were collected from the salt-affected areas of the west coast region of India (State of Maharashtra, Goa, Karnataka and Kerala). The soil pH and electrical conductivity (EC) in fixed ratios of 1:1, 1:2, 1:2.5 and 1:5 soil to water extracts and soil saturation paste extract were determined and the spectral data measurement in the wavelength range of 350–2500 nm was carried out. The data was divided into a calibration dataset (70% of the total) and a validation dataset (30% of the total dataset). The spectral data (raw spectral reflectance averaged at 10 nm) was modeled using multivariate analysis techniques - partial least square regression, principal component regression and support vector regression. The average values of soil pH of 1:1, 1:2, 1:2.5, 1:5 soil: water extract and soil saturation paste (pHe) extract were 4.32, 4.96, 5.01, 5.15 and 5.33, respectively, whereas, the corresponding EC values were 14.00, 7.65, 6.37, 3.52 and 19.96 dS m−1, respectively. The pHe and EC of soil saturation paste extract (ECe) were in the range of 3.067.39 and 0.61–59.18 dS m−1. The coefficient of determination (R2) of pHe with pH of 1:1, 1:2, 1:2.5 and 1:5 soil extract ratio was 0.73, 0.73, 0.72 and 0.73 (p<0.05) respectively. The ECe had a R2 of 0.85, 0.84, 0.85 and 0.84 (p<0.05) with EC of 1:1, 1:2, 1:2.5 and 1:5 soil: water extract ratio, respectively. All these relationships were linear and significant. Thus, the regression relation of ECe with EC1:2, ECe = 2.5272 x (EC1:2) + 2.77 (R2=0.84, p<0.05) could be more useful for studies related to the salinity of salt-affected soil of the west coast region of India where soil salinity is typically associated with the acidity. A good agreement between the actual and predicted pH and EC for different ratios and the extract were exhibited by the coefficient of correlation ranging from 0.48–0.79 and 0.70–0.87, respectively. Among different multivariate techniques tested, partial least square regression outperformed principal component regression and support vector regression. Among all the parameters, the best prediction was achieved for ECe accuracy as r=0.87, R2=0.76, RMSE=5.65 and rank=4 (lowest) with partial least square regression. Thus, the soil saturation paste extract salinity of the salt-affected soils of the west coast region can be monitored using visible near-infrared remote sensing