Soil organic carbon prediction by hyperspectral remote sensing and field vis-NIR spectroscopy: An Australian case study
Introduction
Research in environmental monitoring, modelling and precision agriculture need good quality and inexpensive soil data. Hence we need the development of more time- and cost-efficient methodologies for soil analysis. Visible and near infrared reflectance (vis–NIR) spectroscopy is a physical non-destructive, rapid, reproducible method that provides inexpensive prediction of soil physical, chemical and biological properties according to their reflectance in the wavelength range from 400 to 2500 nm (Ben-Dor and Banin, 1995, Reeves et al., 2000, Reeves et al., 2002, Dunn et al., 2002, Shepherd and Walsh, 2002, Islam et al., 2003). Reflectance signals are produced by vibrations in bonds between C, N, H, O, P, and S atoms. Weak overtones and combinations of fundamental vibrations due to the stretching and bending of NH, OH and CH groups dominate the NIR (700–2500 nm) and electronic transitions the visible (400–700 nm) portions of the electromagnetic (EM) spectrum (Ben-Dor et al., 1999). Soil organic carbon (SOC) plays a major role with respect to many chemical and physical processes in the soil environment and significantly affects the shape and nature of a soil reflectance spectrum. The wide spectral range found by different workers to assess SOC content suggests that SOC is an important soil constituent across the entire spectrum (Ben-Dor et al., 1999).
Spectroscopy has demonstrated its capability to accurately determine SOC contents in the laboratory (e.g. Reeves et al., 1999, Chang and Laird, 2002) and directly in the field with a portable spectrometer (e.g. Barnes et al., 2003). Imaging spectrometry can also be used to estimate soil properties. But the conditions of the soil surface can affect the spectral signal. Some of the properties that are subject to variation both in time and in space are: the degree of soil crusting as a result of rain-drop impact, soil texture, soil moisture, roughness and vegetation or crop residue cover. These perturbing factors induce changes in soil reflectance that approach or exceed the spectral response of organic matter (Barnes et al., 2003). In addition the soil properties estimation can also be subject to degradations due to radiometric and atmospheric effects, spectral and spatial resolutions (Lagacherie et al., 2008). Therefore because of these disturbing factors, few studies have demonstrated the capability to accurately determine SOC contents from airborne-hyperspectral sensors (e.g. Ben-Dor et al., 2002, Selige et al., 2006, Stevens et al., 2006) and none from satellite hyperspectral sensors. As remotely-sensed hyperspectral satellite data offer a synoptic view and a repetitive coverage which are two important advantages compared to ground observations and hyperspectral airborne data, the study of the potential of hyperspectral satellite data for SOC prediction becomes a major issue for digital soil mapping development.
Quantitative spectral analysis of soil using vis–NIR reflectance spectroscopy requires sophisticated statistical techniques to discern the response of soil attributes from spectral characteristics. Various methods have been used to relate soil spectra to soil attributes. Partial least-squares regression (PLSR) is one of the most common techniques for spectral calibration and prediction (e.g. McCarty et al., 2002, Chang and Laird, 2002). Viscarra Rossel et al. (2006) provide a review of the literature comparing quantitative predictions of various soil attributes using multivariate statistical techniques and spectral response in the visible, NIR and Mid infrared (MIR, 2500–25000 nm) regions of the electromagnetic spectrum. Among others, Viscarra Rossel et al. (2006) resume the literature comparing quantitative predictions of SOC using PLSR and spectral response in the visible, NIR and MIR regions of the electromagnetic spectrum.
The aims of this paper are to (i) evaluate the potential for measuring SOC using the Hyperion hyperspectral satellite remote sensor (400–2500 nm) and (ii) compare these to predictions of SOC made using field-collected vis–NIR spectra. In both instances partial least-squares regression (PLSR) was used to relate spectral measurements to SOC contents. This study was performed in the environs of Narrabri in north western New South Wales (NSW), Australia.
Section snippets
Soil samples
A total of 146 surface soil samples (0–10 cm) was collected in the Narrabri region in north western NSW, Australia (− 32°12'27”S, 149°36'31”E). This region is dominated by Vertisols. Eighty eight samples were collected in north western of Narrabri (− 30°11'45”S, 149°37'18”E) in October 2006 and fifty eight soil samples near the town of Narrabri (− 30°18'27”S, 149°45'4”E) in December 2006. Among the 88 soil samples collected in October 2006, 72 were collected on dry bare soils over cotton crops.
Predictions for soils containing low amounts of SOC
The “Cropping soils” were used to determine if low SOC contents can be estimated by the spectral resolutions of the AgriSpec and Hyperion hyperspectral instruments. Baumgardner et al. (1970) noted that if the SOC drops below 2%, it has only a minimal effect on spectral response. As the maximum value of the SOC for this “Cropping soils” is 1%, the spectral response of the studied soil samples should be affected poorly by the SOC content. The cross validations for SOC content for the “Cropping
Results of SOC prediction with Hyperion spectra
From Section 3, both AgriSpec and Hyperion spectral resolutions provided excellent cross validation when the soil sample set is more comprehensive. Using the AgriSpec spectra resampled to the low spectral resolution similar to that of the Hyperion data (152 spectral bands) were as useful as the high-spectral resolution of the AgriSpec instrument.
The “Soils over Hyperion images” were used to determine if SOC contents can be estimated by Hyperion hyperspectral data. The cross validations for SOC
SOC mapping from Hyperion data
When the amount of SOC dropped below 1%, reflectance spectra are not able to predict SOC contents, whatever the spectral resolution. As we observed on the field that bare cotton crops had very low SOC contents, we cannot perform SOC prediction from Hyperion data on an area covered by cotton crops. So the SOC prediction from Hyperion data was performed over mainly pastures, as described in the Section 2.6, which contains a wide SOC contents range. The cross-validation step was performed with the
Discussion
This paper shows that whatever the SOC ranges of the soil samples (between 0.54 and 1%, between 1.08 and 5.1%, or between 0.54 and 5.1%) and whatever the number of soil samples (56, 72 or 146) used in the prediction models, the spectral resolution did not change the accuracy of the model. So in our agricultural and pedological context of the Narrabri area, the use of Hyperion hyperspectral data should be as useful as the use of field vis–NIR data for SOC prediction. However, we observed the SOC
Conclusions
We used for the first time satellite hyperspectral data for SOC prediction by multivariate regression modelling. This remote sensing satellite approach shows the theoretical potential of satellite hyperspectral data for SOC prediction. The results indicate that whatever the SOC ranges of the soil samples and whatever the number of soil samples used in our prediction models, the spectral resolution did not change the accuracy of the model. The cross validation models using 146 soil samples with
Acknowledgment
Cecile Gomez was in receipt of a short-term postdoctoral fellowship from the University of Sydney.
References (25)
- et al.
Estimation of soil clay and calcium carbonate using laboratory, field and airborne hyperspectral measurements
Remote Sensing of Environment
(2008) - et al.
On digital soil mapping
Geoderma
(2003) - et al.
High resolution topsoil mapping using hyperspectral image and field data in multivariate regression modelling procedures
Geoderma
(2006) - et al.
The EnMAP hyperspectral imager—an advanced optical payload for future applications in Earth observation programmes
Acta Astronautica
(2007) ParLeS: software for chemometric analysis of spectroscopic data
Chemometrics and Intelligent Laboratory Systems
(2008)- et al.
Visible, near-infrared, mid-infrared or combined diffuse reflectance spectroscopy for simultaneous assessment of various soil properties
Geoderma
(2006) - et al.
Remote- and ground-based sensor techniques to map soil properties
Photogrammetric Engineering and Remote Sensing
(2003) - et al.
Effects of organic matter on the multispectral properties of soils
- et al.
Near infrared analysis as a rapid method to simultaneously evaluate several soil properties
Soil Science Society of America Journal
(1995) - et al.
Soil reflectance: remote sensing for the Earth Science
Mapping of several soil properties using DAIS-7915 hyperspectral scanner data—a case study over clayey soils in Israel
International Journal of Remote Sensing
Near-infrared reflectance spectroscopic analysis of soil C and N
Soil Science
Cited by (461)
Review – Plant nutritional status analysis employing the visible and near-infrared spectroscopy spectral sensor
2024, Spectrochimica Acta - Part A: Molecular and Biomolecular Spectroscopy