Comparison of Cation Exchange Capacity Estimated from Vis–NIR Spectral Reflectance Data and a Pedotransfer Function

Abstract

Core Ideas Conventional methods for measuring CEC are time consuming and costly. A Vis–NIRS model and pedotransfer function (PTF) were developed for CEC determination. The Vis–NIRS model estimated CEC accurately for different soil types. The Vis–NIRS CEC model performed better than the PTF based on clay and organic C contents. Knowledge of the cation exchange capacity (CEC) for soils or other porous media is very important for civil engineering and agricultural applications. However, the standard laboratory methods to measure CEC are costly and laborious. The aim of this research was to develop a visible–near‐infrared spectroscopy (Vis–NIRS, 400–2500 nm) calibration model to predict CEC based on multivariate analysis and to compare the predictive ability of Vis–NIRS with that of a pedotransfer function (PTF). For this purpose, reference CEC was measured by the ammonium acetate method for 235 soil samples, collected from 21 countries. Diffuse spectral reflectance data were also collected by using a NIRSTM DS2500 spectrometer. The model was constructed on a calibration subset (80%) and evaluated with a validation subset (20%) using partial least squares regression. The Vis–NIRS calibration model was sufficiently robust based on the cross‐validation results [R2 = 0.79, RMSE of cross‐validation values of 7.9 cmolc kg−1 and bias = −0.14]. The independent validation of the Vis–NIRS model showed good prediction accuracy, regardless of sample origin (RMSE of prediction value of 5.0 cmolc kg−1 and ratio of performance to interquartile distance value of 4.5). Moreover, the Vis–NIRS prediction performance was superior to that of the PTF, which was influenced by the sample origin (RMSE values of 11.5 cmolc kg−1). The better prediction of CEC by the Vis–NIRS calibration model suggests that it is due to the co‐variation of CEC with clay (type and content) and organic C content.

Combining Vis–NIR spectroscopy and advanced statistical analysis for estimation of soil chemical properties relevant for forest road construction

“…The good results for CEC in this study, conform with Santra et al (2009), Bilgili et al (2010), andRehman et al (2019). The PLSR model for CaCO 3 did not yield meaningful results (RPIQ = 0.38) (Figure 3).…”

Estimating coefficient of linear extensibility using Vis–NIR reflectance spectral data: Comparison of model validation approaches

“…Further, the SRMSE (Equation ) was used to make a comparison with previous studies, since the RMSE is strongly influenced by the range of the measured data (Rehman et al., 2019): where “Range” represents the largest to the smallest values of the COLE for the respective dataset or from published studies. Smaller SRMSE values denote more accurate estimations of COLE, and vice versa.…”

“…The magnitude of COLE is determined by the cation exchange capacity, specific surface area, soil OC (Smith, Hadas, Dan, & Koyumdjisky, 1985), water content, clay mineralogy (Vaught et al., 2006), and soil texture (Gomboš & Tall, 2012). All these soil properties have been successfully determined by Vis–NIRS because they possess spectrally active components and clear responses in the visible and near‐infrared (Vis–NIR) regions (Goetz, Chabrillat, & Lu, 2001; Knadel et al., 2018; Rehman et al., 2019; Viscarra Rossel et al., 2006). Thus, the Vis–NIRS technique has the potential to indirectly estimate COLE from spectral characterization of fine‐grained soils, as reported by the very few studies currently available (Hallmark, Morgan, & Hutchison, 2011; Kariuki et al., 2004; Șenol & Akgul, 2012; Waruru, Shepherd, Ndegwa, Kamoni, & Sila, 2014).…”

Estimating Atterberg Limits of Fine‐Grained Soils by Visible–Near‐Infrared Spectroscopy

“…The RMSE value is strongly influenced by the range of measured values of the considered soil property. Thus, to compare the cross‐validation model and performance of the model with other studies, the SRMSE (Arthur, 2017; Rehman et al, 2019) was used: where Range represents the largest minus the smallest SL, PL, or LL for the respective dataset or published study. Smaller SRMSE values denote better estimations and vice versa.…”