Comparison of Vis/SWNIR and NIR spectrometers combined with different multivariate techniques for estimating soil fertility parameters of calcareous topsoil in an arid climate

Abstract

The lab spectroscopy using Vis/SWNIR and NIR spectrometers was used to estimate the major fertility parameters in calcareous topsoil from arid region of central Iran. The fertility parameters included soil organic matter (SOM), total nitrogen (TN), available phosphorus (P -avl ), and available potassium (K -avl ). A total of 151 samples were collected from the depth of 0–20 cm, and the diffuse reflectance spectra of them were collected using charge-coupled device (CCD) and indium-gallium-arsenide (InGaAs) spectrometers. To extract the prediction models, partial least squares (PLS) as linear, and artificial neural networks (ANN) and support vector machines (SVM) as nonlinear multivariate techniques were used. The CCD-based spectra combined with the ANN model led to excellent predictability for SOM (residual predictive deviation (RPD) = 2.60), TN (RPD = 2.54), and P -avl (RPD = 3.01) and very good prediction for K -avl (RPD = 2.02). However, the InGaAs-based spectrometer outperformed the CCD one for predicting all the soil fertility parameters. The best prediction models were obtained from the nonlinear multivariate techniques where the excellent predictions were provided by the ANN model for SOM ( R 2 p = 0.941; RPD = 3.27), P -avl ( R 2 p = 0.902; RPD = 3.34), and K -avl ( R 2 p = 0.863; RPD = 2.69) and by SVM for TN ( R 2 p = 0.863; RPD = 2.91). This study revealed that despite the great potential of InGaAs spectrometers for predicting the soil fertility parameters, the low-cost CCD-based spectrometer could result in satisfactory performance when the appropriate nonlinear technique is used for data analysis. • The potential of two spectrometers was compared to estimate fertility parameters. • Linear and nonlinear techniques were used to extract the prediction models. • CCD spectrometer combined with the ANN model led to satisfactory performance. • Nonlinear multivariate techniques resulted in the best accuracy.