Vis-NIR spectroscopy coupled with PLSR and multivariate regression models to predict soil salinity under different types of land use

Abstract

Soil salinization is an important environmental challenging factor that threatens human life and food security. Therefore, accurate monitoring of soil salinization needs some low-cost measurement methods. One of the rapid and low costly methods is monitoring by using Vis-NIR spectroscopy methodology. For this aim, 84 points were selected in to take soil samples in different types of land uses such as bare lands, range lands, and farm lands. Soil salinity (ECe) was measured at all sampled soils. Moreover, spectral reflectance at various bands was recorded by using a portable spectrometer. The spectral reflectance at all wavelength ranges (i.e. from 400 to 2400 nm) and a combination of spectral reflectance from different wavelengths were used as input variables for developing PLSR (partial least squares regression) and MLR (multiple regression) models. The results indicated that the derived MLR model by using combinations of spectral reflectance from different wavelengths such as 1448, 1943 and 2400 nm as input variables performed better results to predict soil salinity (RMSE = 7.890 dS m−1, R2 = 0.728, and RPD = 2.024). In addition, the results of developed PLSR and MLR models showed that the MLR model performed better soil salinity predictions at farm lands and range lands, while the accurate prediction of soil salinity for range lands was found by using developed PLSR models. It was concluded that the employing Vis-NIR spectroscopy data under different types of land uses could predict soil salinity in a different level of accuracy.