Information depth of NIR/SWIR soil reflectance spectroscopy

Highlights

• A new model links the particle size distribution and soil NIR/SWIR reflectance.

• It partitions reflectance spectra into surface and volume contributions.

• It shows that light reflectance from dry soils is dominantly a surface phenomenon.

• The volume/total reflectance ratio is higher for coarse than for fine soils.

• The model opens new avenues for remote sensing of soil hydraulic properties.

Abstract

Proximal and remote sensing techniques in the optical domain are cost-effective alternatives to standard soil property characterization methods. However, the extent of light penetration into the soil sample, also termed soil information depth, is not well understood. In this study a new analytical model that links the particle size distribution and soil reflectance in the near infrared (NIR) and shortwave infrared (SWIR) bands of the electromagnetic spectrum is introduced. The model enables the partitioning of measured reflectance spectra into surface and volume (subsurface) contributions, thereby yielding insights about the soil information depth. The model simulations indicate that the surface reflectance contribution to the total reflectance is significantly higher than the volume reflectance contribution for a broad range of soils that vastly differ in texture, mineralogical composition and organic matter contents. The ratio of volume to total reflectance is higher for sandy soils than for clayey soils, especially at longer optical wavelengths, but the ratio rarely exceeds 15%. Therefore, the light reflection from dry soils is predominantly a surface phenomenon and the information depth in most soils rarely exceeds 1 mm. The results of this study reveal an intimate physical relationship between soil reflectance and the particle size distribution in the NIR/SWIR range, which opens a potential new avenue for retrieval of the particle size distribution from remotely sensed reflectance via a universal process-based approach.

Introduction

Recent advances in proximal sensing of soils have provided rapid and cost-effective methods for retrieving various soil properties in laboratory settings. For example, significant correlations between soil reflectance spectra in the optical domain (300–2500 nm) and soil mineralogy and chemical composition (Clark et al., 1990; Kruse et al., 2003; Mulder et al., 2013; Omran, 2017), soil organic matter content (Daniel et al., 2004; He et al., 2009; Ingleby and Crowe, 1999; Leue et al., 2017; Yuan et al., 2020), soil water content (Lobell and Asner, 2002; Sadeghi et al., 2015; Whiting et al., 2004; Zeng et al., 2016), soil particle size distribution (Bänninger et al., 2006; Hermansen et al., 2017; Sadeghi et al., 2018a), and soil hydraulic properties (Babaeian et al., 2015a; Babaeian et al., 2015b) have led to the development of promising data-driven or physics-based methods for soil property estimation.

Despite salient success of soil characterization with proximal sensing techniques, an inherent limitation of these methods is the shallow light penetration into the soil. In addition, the soil information depth, i.e., the maximum soil depth contributing to the measured signal, is not well understood. Consequently, it is challenging to determine the soil volumes that the estimated properties are representative for, which is key for many soil studies.

Light penetration into a particulate opaque medium such as soil is governed by two processes, light absorption and light scattering by particles. Models that explain these processes are mainly derived from approximations of the radiative transfer equation (Philips-Invernizzi et al., 2001). A main challenge of these models, when applied to natural soils, is their calibration for the soil optical properties. This requires information of not only the reflectance (i.e., reflected light intensity at the surface), but also the light intensity at a specified depth of the medium (Kortüm, 2012), which is challenging to measure in soils (Tian and Philpot, 2018; Baranoski et al., 2019). In addition, the inversion of some of the models, e.g., the beam-tracing model of Bänninger et al. (2006), are computationally - very demanding.

A number of researchers have previously attempted to indirectly determine the light penetration depth in soils via studying light-dependent soil chemical or biological activities at various soil depths (Balmer et al., 2000; Hebert and Miller, 1990) or germination of light-sensitive seeds planted at different depths (Benvenuti, 1995; Bliss and Smith, 1985; Woolley and Stoller, 1978). Ciani et al. (2005) were among the first to provide direct estimates of the information depth in the visible light range (300–700 nm). In this approach, the Kubelka-Munk two-flux radiative transfer model or its variants (Kubelka and Munk, 1931; de la Osa et al., 2020) can be used to relate the information depth to the soil optical properties determined with a dilution method, where the measured reflectance spectra of various soil mixtures with a white reference powder (e.g., barium sulphate) having known optical properties are analyzed. Although this method is computationally efficient, showing great potential for soil information depth estimation, it requires laborious soil measurements, and therefore is not feasible for large soil datasets.

In this paper we propose a novel physics-based analytical model that provides new insights about the information depth of NIR/SWIR soil spectroscopy. While the model is not capable of explicitly quantifying information depth, it estimates the contributions of surface reflectance (i.e., reflectance originating from the first layer of particles at the surface) and volume reflectance (i.e., reflectance originating from the underlying soil layers) to the total soil reflectance. The new model is based on the radiative transfer model introduced in Sadeghi et al. (2018a), and it is proposed for longer optical wavelengths (i.e., 700–2500 nm), which are of interest for several soil properties such as water content (Sadeghi et al., 2015) or particle size distribution (Sadeghi et al., 2018a). The proposed model requires easy-to-measure basic soil information including reflectance spectra and particle size distributions, and hence, it may be efficiently applied to large soil datasets.

In the next section, we briefly review the Sadeghi et al. (2018a) model for the reflectance-particle size functional relationship, R(D), which is valid for uniform soils (i.e., soils with a narrow particle size distribution). We then extend its applicability to nonuniform soils to allow partitioning of soil reflectance spectra into surface and subsurface contributions. In Section 3, we present experimental soil reflectance spectra and particle size distribution (PSD) data used for model evaluation. In Section 4, we introduce and discuss modeling results that unravel the contributions of surface and volume components to the total soil reflectance. A summary and conclusions are provided in Section 5.