Use of diffuse reflectance spectroscopy and Nix pro color sensor in combination for rapid prediction of soil organic carbon
Introduction
Optimization of soil management via high-resolution extensive soil analysis is obligatory for meeting the global food demand (Krishna, 2014, Barthes et al., 2019). However, traditional approaches of soil physicochemical analysis which are cost-prohibitive, labor-intensive and time-consuming, cannot achieve this target. Various sensors report important parameters about soil properties and the most obvious technologies which offer several benefits over conventional methods of soil physicochemical analyses are the non-invasive proximal soil sensors (Stenberg et al., 2010, Aldabaa et al., 2015).
Emerging proximal sensor technologies such as visible near infrared (VisNIR) diffuse reflectance spectroscopy (DRS) and portable X-ray fluorescence spectrometry (PXRF) can efficiently quantify soil fertility, pH/salinity, total C/total N, and additional soil attributes (texture, structure) (Viscarra Rossel et al., 2006, Sharma et al., 2014, Sharma et al., 2015). With the elemental data collected via PXRF, multiple soil parameters of interest can be predicted using linear regression models. Hyperspectral VisNIR DRS has become well known for quickly and simultaneously quantifying multiple soil parameters by analyzing the soil surface reflected radiation (Viscarra Rossel et al., 2006, Wang et al., 2015b). This technology is gaining acceptance in soil science, owing to its cost-effectiveness over other spectroscopic and wet chemistry methods. Generally, VisNIR DRS uses a white light source to illuminate the soil surface and collects reflected energy. The energy is carried into a spectroradiometer via fiber optic cable and parsed at 1–10 nm intervals from 350 to 2,500 nm. Following appropriate spectral pre-treatment (e.g., first derivative, standard normal variate etc.), regression procedures (e.g., random forest, support vector regression, penalized spline regression; partial least squares regression) are used to predict unknowns from a training dataset (Chakraborty et al., 2010), in much the same way that unknown pH values are determined after first creating a calibration equation with several known standards.
The use of DRS for predicting soil organic carbon (OC) dated back to 1986 when Dalal and Henry (1986) used near-infrared spectra (110–2500 nm) to predict soil OC with an R2 value of 0.93. Since then, scientists have extensively used VisNIR DRS to rapidly predict soil OC with good accuracy (Shepherd and Walsh, 2002, Brown et al., 2006, Viscarra Rossel et al., 2003, Viscarra Rossel et al., 2006, Nocita et al., 2014). Chang and Laird (2002) used DRS to predict total carbon via partial least squares regression (PLSR) using the first derivative spectra of the optical density [log(1/R)]. Fidêncio et al. (2002) combined DRS spectra (1000–2500 nm) and radial basis function networks (RBFN) to predict soil organic matter. McCarty et al. (2002) exhibited that soil OC can be predicted using the 1100–2498 nm region of DRS spectra via PLSR. Islam et al. (2003) combined the 700–2500 nm region of DRS spectra and principal component regression (PCR) to predict soil OC with an R2 value of 0.68. Morgan et al. (2009) used VisNIR DRS to predict OC of 72 soil cores collected from 21 soil series via PLSR.
Since soils with higher organic matter are darker and less reflective than those soils with lower organic matter, scientists have used soil color as a proxy to predict soil OC (Lindbo et al., 1998, Konen et al., 2003). Sudduth and Hummel (1991) exhibited that VisNIR DRS spectra and commercial colorimeter data can be used to predict soil organic matter. Wills et al (2007) used Munsell sol color chart, chroma meter, and horizon matrix color to predict soil OC. With technological advancements, researchers have also calibrated soil organic matter and soil OC via image-extracted soil color as a proxy (Chen et al., 2000, Visacarra Rossel et al., 2008, Aitkenhead et al., 2018). Gregory et al. (2006) predicted soil organic matter content using a high-resolution digital camera system. Morais et al. (2019) used digital soil images with multivariate image analysis to predict soil OC. Recently, Fu et al. (2020) have shown that soil organic matter under variable soil moisture can be predicted from cellular phone images.
Recently, soil scientists have reported the usefulness of another inexpensive (~$350 USD) sensor, the Nix Pro color sensor (Hamilton, Ontario, Canada). This affordable, portable, and miniaturized LED-based sensor has been used to measure soil color (Stiglitz et al., 2016a, Stiglitz et al., 2016b) and soil OC (Stiglitz et al., 2017, Mikhailova et al., 2017, Raeesi et al., 2019) with reasonable accuracy, however, using a limited number of samples. Recently, Kagiliery et al. (2019) exhibited that Nix Pro in combination with PXRF can rapidly predict total S in lignite coal samples. Results from this sensor contain color results from multiple color spaces like RGB, XYZ, CIEL*a*b etc. Yet, the performance of this sensor has only been sparsely studied and more research is needed to assess its efficacy on a wider range of soil types.
Although several studies have revealed reasonable to good prediction accuracies of soil physicochemical properties via an individual sensor, often diversity of soil matrix poses difficulty in development of robust prediction models. Combination of multiple sensors may provide a cost-effective, site-specific solution for soil health monitoring and management. In soil science, the combined use of VisNIR DRS and PXRF was first demonstrated for total C total N determination (Wang et al., 2015a), soil salinity (Aldabaa et al., 2015), and soil petroleum contamination (Chakraborty et al., 2015); which indicate the tremendous potential of PXRF/VisNIR DRS for quick soil characterization and associated environmental quality assessment. Coupled with georeferencing, the combined use of VisNIR DRS + PXRF enables the rapid spatial prediction of multiple soil attributes, on-site, and non-destructively.
Notably, there is a distinct difference in the working principles of Nix Pro and DRS where the former measures the true color of an object by using three filters Red (R), Blue (B), and Green (G) on the photodiodes used. Notably, RGB hues are considered as the common primaries for the RGB color space (McCamy, 1992). Since this sensor does not consider spectral band-limit, peak wavelength intensity, and full-width half maximum, RGB filters produce overlapping spectral curves. Conversely, DRS works on the principle of measuring the intensity of light at different wavelengths.
This present study aims to assess the performance of combining the Nix Pro and VisNIR DRS systems to predict soil OC. We hypothesize that the combined OC predictive model using both optical sensors will offer the best prediction accuracy, better than each sensor individually.
Section snippets
Soil sampling and laboratory analysis
In this research, soil samples from agricultural fields adjacent to a landfill were used, where the soils were highly disturbed due to intense anthropogenic activities. The area is characterized by the tropical wet-and-dry climate (Köppen climate classification Aw). The annual mean temperature is 24.8 °C; monthly mean temperatures range from 15 °C to 30 °C. A total of 200 georeferenced surface (0–15 cm) soil samples were collected randomly from 1 km2 agricultural land area near the Dhapa landfill
Descriptive statistics and trends
Table 1 exhibits the summary statistics of the laboratory measured soil OC along with Nix Pro reported L* and a* readings. Soil OC content showed a wide variation from 12.71 to 39.63 g kg−1 with a mean of 29.90 g kg−1 which can be attributed to the continuous deposition of composted materials. This trend was consistent with the findings of Chakraborty et al. (2017) who reported a high magnitude of soil organic matter for the same region. Each soil scan by Nix Pro simultaneously gathered
Conclusions
In this project, an algorithm combining predictions from DRS spectra and Nix Pro color space values yielded better soil OC prediction accuracy than using either sensor independently. While increases in predictive accuracy afforded by utilizing Nix Pro color data in tandem with DRS spectra were substantial (63% and 17% increase in R2 value as compared to the GAM-OC and PLSR-OC models, respectively), the cheapness (~$350) and dwell time (<2.5 sec) of the Nix Pro sensor make it a valuable
CRediT authorship contribution statement
Swagata Mukhopadhyay: Conceptualization, Methodology, Software, Data curation, Writing - original draft. Somsubhra Chakraborty: Supervision, Writing - review & editing.
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgments
The authors wish to thankfully acknowledge financial assistance from the Department of Science and Technology-Science & Engineering Research Board (DST-SERB), Govt. of India, under early career research scheme [No. ECR/2017/000510].
References (68)
- et al.
Digital RGB photography and visible-range spectroscopy for soil composition analysis
Geoderma
(2018) - et al.
Combination of proximal and remote sensing methods for rapid soil salinity quantification
Geoderma
(2015) - et al.
Performance comparison between a miniaturized and a conventional near infrared reflectance (NIR) spectrometer for characterizing soil carbon and nitrogen
Geoderma
(2019) - et al.
Critical review of chemometric indicators commonly used for assessing the quality of the prediction of soil attributes by NIR spectroscopy
TrAC, Trends Anal. Chem.
(2010) - et al.
The reflectance spectra of organic matter in the visible near-infrared and short wave infrared region (400–2500 nm) during a controlled decomposition process
Remote Sens. Environ.
(1997) - et al.
Global soil characterization with VNIR diffuse reflectance spectroscopy
Geoderma
(2006) - et al.
Non-saturated soil organic horizon characterization via advanced proximal sensors
Geoderma
(2017) - et al.
External parameter orthogonalization of Eastern European VisNIR-DRS soil spectra
Geoderma
(2019) - et al.
Rapid assessment of regional soil arsenic pollution risk via diffuse reflectance spectroscopy
Geoderma
(2017) - et al.
Development of a hybrid proximal sensing method for rapid identification of petroleum contaminated soils
Sci. Total Environ.
(2015)