Rapid assessment of soil water repellency indices using Vis-NIR spectroscopy and pedo-transfer functions

Abstract

Soil water repellency (SWR) is an important soil physical property that may restrict water infiltration and soil water retention. Common laboratory and field techniques for assessing SWR are laborious, time-consuming, and costly. Meanwhile, Visible–Near-Infrared (Vis-NIR) spectroscopy has been reported as a rapid, cost-effective, and alternative technique to estimate several soil properties. To investigate the efficacy of this technique for predicting SWR indices [soil water repellency index (RI) and soil–water contact angle (β)] at dry condition, 100 soil samples collected from farmlands, orchards, rangelands and forests of Zrêbar lake watershed in Kurdistan province, Iran, were measured by Vis-NIR spectroscopy within the 350–2500 nm range. The Savitzky–Golay first derivative method was applied for denoising the spectral data. The RI and β were measured by the intrinsic sorptivity method (using water and ethanol as absorbing liquids). The other basic soil properties were also measured by standard laboratory methods. Stepwise multiple linear regression (SMLR) and partial least squares regression (PLSR) were utilized to establish pedo-transfer functions (PTFs) and spectro-transfer functions (STFs) using basic soil properties and spectral absorbance data, respectively, to estimate soil organic matter (SOM) content and SWR indices. We obtained good predictions for SOM with R2 = 0.67 and RPIQ (the ratio of performance to interquartile range) = 1.92 using the PLSR-based STF. The results also revealed that although the SMLR-based STFs achieved slightly better estimates of the SWR indices (RI and β) than the SMLR-based PTFs (R2 values of 0.28 to 0.39 vs. 0.19 to 0.23, respectively); but, overall, none of these transfer functions for estimating these indices showed acceptable predictive capability. However, the PLSR-based STFs could provide a reasonable prediction for the studied SWR indices (R2 > 0.52 and RPIQ > 2.27). The majority of important adsorption bands in the Vis-NIR PLSR models for the SWR prediction was related to both the quantity and quality of SOM. Overall, the results demonstrated that the Vis-NIR PLSR could be applied to predict SOM and SWR indices rapidly, non-destructively, and with fair accuracy.