Spectral information related to soil slaking: An example from Australia

Highlights

• Fe bearing minerals and clay minerals are related with soil slaking processes.

• It is possible to successfully predict soil slaking using spectroscopy.

• Predictability of soil slaking improves as more regions of spectra are added.

Abstract

Soil slaking is a highly relevant property in erosion risk management and soil physics in general. Despite its importance and close relation with known soil chemical and physical properties such as soil organic carbon content, pH, sodium content, and soil aggregate stability, it is still underutilized mainly due to the lack of practical methodologies for measuring and understanding of its controlling factors. To facilitate and encourage its use in decision-making, infrared spectroscopy was tested as a proxy for its assessment. Different modelling approaches using soil reflectance values from the visible, near (Vis-NIR) and mid-infrared (MIR) regions of the electromagnetic spectra were employed to predict slaking index coefficient (SIa). The dataset had 146 samples with measured SIa, covering an important part of the agro-ecological variability of NSW, Australia. It was observed that the model's accuracy (20% used as an independent validation) increased as different regions of the spectra were considered. The results showed low predictability (R² 0.45–0.54) using only the Vis-NIR region of the spectra, where only soil organic carbon (SOC) and Fe oxides spectral regions were identified as main predictors (~2000–2100 nm or 5000–4760 cm⁻¹ and ~700–900 nm or 14,280–11,100 cm⁻¹). As MIR regions were added, the modelling performance improved (R² 0.51–0.76), where SOC and clay minerals related regions were also identified as important predictors (4700–4900 cm⁻¹, 3500–3700 cm⁻¹ or ~2000–2100 nm and ~2700–2850 nm respectively). Since the uncertainty related to the models was relatively low (RMSE of ~1.0 slaking index units), it can be concluded that infrared spectroscopy can be used as a prospecting tool for predicting soil slaking.

Introduction

Traditional methods for evaluating soil aggregate stability require considerable time, work and are also expensive to outsource to commercial laboratories. Fajardo et al. (2016) presented a new methodology for assessing soil slaking that proved to be faster and less expensive. Furthermore, it provided new indices related to key processes of slaking of soil aggregates in the presence of free water.

It was shown that the new indices of soil slaking are correlated with known soil properties; considering these relationships, it is hypothesized that it should be possible to create predictive models with other sources of soil information. Many authors have demonstrated the benefits of Vis-NIR and MIR spectroscopy to predict soil properties as soil spectra have been proven to contain comprehensive information about the soil physico-chemical composition (Ben-Dor and Banin, 1995; Soriano-Disla et al., 2014; Stenberg et al., 2010).

Vis-NIR and MIR soil spectra measurement requires minimum sample preparation. When employed with multivariate analytical methods, it has been shown that several soil properties can be simultaneously assessed from a single spectrum. The main benefits from using hyperspectral information (e.g., Vis-NIR-MIR spectroscopy) are not only the fast acquisition of a particular soil property, or the reduced cost, but the scalability of the output i.e., use of hyperspectral imagery to predict a relevant soil property at unsampled locations.

Among the many soil properties that have proven predictability from spectroscopy i.e., soil organic carbon, pH, clay, cation exchange capacity, etc. (Armenta and de la Guardia, 2014), only a few that can predict a soil process (e.g., slaking of soil aggregates) hence to represent how the soil surface will withstand contact with free water. This becomes highly attractive in terms of its use in erosion risk programs and related planning activities. Minasny et al. (2008) explored the possibility of predicting (using spectroscopy) physical properties dependant on pore structure like bulk density and hydraulic conductivity. They found that these should not be predicted directly, since spectroscopic methods only provide surface composition information. Despite this, they foresaw a possible direct use in predicting other complex properties like Atterberg limits (Casagrande, 1932) or swelling behaviour.

On the other hand, Leelamanie et al. (2013) observed that since soil slaking is linked with rapid pressure build-up within aggregates, those properties or components (rather mineral and/or organic) related with water hydrophobicity may be the cause of hampering the pressure increase within aggregates by reducing their water-filled pore space.

Following this idea, as the SIa is a complex property not completely understood and linearly correlated with multiple chemical and physical properties (Fajardo et al., 2016), it would be expected that chemical and mineralogical properties of the soil matrix could control its behaviour, even more considering the nature of the method which replicates a soil surface phenomena.

Considering the novelty of the method and the lack of more diverse datasets, a mechanistic approach for soil slaking behaviour using this method seems still a distant task. Hence, this study follows a more pragmatic approach, which are first, to assess the predictability of SIa directly from Vis-NIR and MIR spectral information in Australian conditions (more specifically New South Wales). Second, this paper aims to report the specific spectral features related with the slaking process which can be then considered for predicting soil slaking at larger scales e.g., digital soil mapping using hyperspectral imagery.