Use of diffuse reflectance spectroscopy and Nix pro color sensor in combination for rapid prediction of soil organic carbon

https://doi.org/10.1016/j.compag.2020.105630Get rights and content

Highlights

  • Data from a simple color sensor and VisNIR DRS were combined to predict soil OC.

  • Variables measured by the color sensor showed negative and non-linear effects on OC.

  • PLSR produced satisfactory OC prediction using DRS spectra alone.

  • The combined model outperformed model from either sensor.

Abstract

Contemporary soil characterization is increasingly dependent upon proximal sensor data whereby high sample throughput and low-cost analysis are realized. Recent research studies have shown that combined sensor platforms generally offer greater predictive model stability and increased accuracy than the use of sensors in isolation. In this study, data from an inexpensive ($350 USD) Nix Pro color sensor, which measures the true color of an object by using red, blue, and green filters, was used with diffuse reflectance spectroscopy (DRS) (590–2490 nm) to predict soil organic carbon (OC) content in highly disturbed, landfill soils of India ex-situ. Generalized additive model (GAM) and partial least squares regression (PLSR) were applied to model DRS and Nix Pro data, respectively, both independently and by combining model predictions using a bilinear regression. Results showed that the combined model outperformed either sensor independently where the 30% external test set achieved a validation R2 of 0.95, residual prediction deviation (RPD) of 4.54, and the ratio of performance to interquartile range of 6.25 relative to laboratory-measured OC data. In contrary, the GAM-OC model using Nix Pro data alone and the PLSR-OC model using DRS data alone exhibited validation R2 values of 0.58 and 0.81, respectively. In sum, the addition of the inexpensive Nix Pro sensor substantively improved the prediction of soil OC relative to the use of DRS in isolation. Future studies should evaluate the effectiveness of such an approach on a wider variety of soil types (e.g., colors), its effectiveness in-situ under variable moisture conditions, and in possible combination with other proximal sensing platforms.

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)

  • H. Cheng et al.

    Estimating heavy metal concentrations in suburban soils with reflectance spectroscopy

    Geoderma

    (2019)
  • P.H. Fidêncio et al.

    Determination of organic matter in soils using radial basis function networks and near infrared spectroscopy

    Anal. Chim. Acta

    (2002)
  • Y. Fu et al.

    Predicting soil organic matter from cellular phone images under varying soil moisture

    Geoderma

    (2020)
  • Y. Hong et al.

    Cadmium concentration estimation in peri-urban agricultural soils: Using reflectance spectroscopy, soil auxiliary information, or a combination of both?

    Geoderma

    (2019)
  • S. Ibáñez-Asensio et al.

    Statistical relationships between soil colour and soil attributes in semiarid areas

    Biosyst. Eng.

    (2013)
  • J. Kagiliery et al.

    Rapid quantification of lignite sulfur content: Combining optical and X-ray approaches

    Int. J. Coal Geol.

    (2019)
  • B. Minasny et al.

    Removing the effect of soil moisture from NIR diffuse reflectance spectra for the prediction of soil organic carbon

    Geoderma

    (2011)
  • C.L.S. Morgan et al.

    Simulated in situ characterization of soil organic and inorganic carbon with visible near-infrared diffuse reflectance spectroscopy

    Geoderma

    (2009)
  • S. Mukhopadhyay et al.

    Assessment of heavy metal and soil organic carbon by portable X-ray fluorescence spectrometry and NixPro™ sensor in landfill soils of India

    Geoderma Regional

    (2020)
  • M. Nocita et al.

    Prediction of soil organic carbon content by diffuse reflectance spectroscopy using a local partial least square regression approach

    Soil Biol. Biochem.

    (2014)
  • S.M. O'Rourke et al.

    An assessment of model averaging to improve predictive power of portable vis-NIR and XRF for the determination of agronomic soil properties

    Geoderma

    (2016)
  • A. Sharma et al.

    Characterizing soils via portable X-ray fluorescence spectrometer: 3. Soil reaction (pH)

    Geoderma

    (2014)
  • A. Sharma et al.

    Characterizing soils via portable X-ray fluorescence spectrometer: 4. Cation exchange capacity (CEC)

    Geoderma

    (2015)
  • B. Stenberg et al.

    Visible and near infrared spectroscopy in soil science

    Adv. Agron.

    (2010)
  • R. Stiglitz et al.

    Evaluation of an inexpensive sensor to measure soil color

    Comput. Electron. Agric.

    (2016)
  • R. Stiglitz et al.

    Using an inexpensive color sensor for rapid assessment of soil organic carbon

    Geoderma

    (2017)
  • R. Sugita et al.

    Validity of color examination for forensic soil identification

    Forensic Sci. Int.

    (1996)
  • Y. Tang et al.

    Evaluating low-cost portable near infrared sensors for rapid analysis of soils from South Eastern Australia

    Geoderma Regional

    (2020)
  • R.A. Viscarra Rossel et al.

    Visible, near infrared, mid infrared or combined diffuse reflectance spectroscopy for simultaneous assessment of various soil properties

    Geoderma

    (2006)
  • D. Wang et al.

    Synthesized use of VisNIR DRS and PXRF for soil characterization: Total carbon and total nitrogen

    Geoderma

    (2015)
  • Y.B. Wang et al.

    Soil pH value, organic matter and macronutrients contents prediction using optical diffusereflectance spectroscopy

    Comput. Electron. Agric.

    (2015)
  • Y. Zhu et al.

    Characterizing surface soil water with field portable diffuse reflectance spectroscopy

    J. of Hydrol.

    (2010)
  • P. Bhattacharyya et al.

    Effect of heavy metals on microbial biomass and activities in century old landfill soil

    Environ. Monit. Assess.

    (2008)
  • S. Chakraborty et al.

    Rapid identification of oil-contaminated soils using visible near-infrared diffuse reflectance spectroscopy

    J. Environ. Qual.

    (2010)
  • Cited by (0)

    View full text