Content uploaded by Hosna Mohamadi
Author content
All content in this area was uploaded by Hosna Mohamadi on Jul 22, 2017
Content may be subject to copyright.
1
Determination of several soil properties based on ultra-violet, visible, and near-infrared reflectance
spectroscopy
Hosna Mohamdi Monavar
Department of Biosystem Engineering, Bu Ali Sina University, Hamadan, Iran.
*E-mail: mohamadihosna@gmail.com, hosnamohamadi@basu.ac.ir
Abstract- Soil is a fundamental natural resource which
people rely on for the production of food, fiber, and
energy. Given the importance of soils, there is a need for
regular monitoring to detect changes in its status so as to
implement appropriate management in the event of
degradation. Application of ultra-violet (UV), visible
(VIS) and near-infrared (NIR) spectroscopy for
prediction of soil properties offer a cost and time
effective approach for evaluation of soil structural
quality. The main objective of this study was to evaluate
the ability of reflectance spectroscopy in the UV, VIS and
NIR ranges to predict several soil properties
simultaneously. Soil samples (n=210 in 15cm depth from
surface) were used for simultaneous estimation of pH,
electrical conductivity (EC), air-dry gravimetric water
content, organic carbon (OC), total nitrogen (TN), free
iron, clay, sand, and silt contents, cation exchange
capacity (CEC), and exchangeable calcium (Ca),
magnesium (Mg), potassium (K), and sodium (Na). After
removal of outliers identified by principal components
analysis (PCA), 75% of the sample set was randomly
selected for calibration (n=157) and the remaining used
for validation. Modified partial least squares (PLS)
regression with cross-validation was used to develop
prediction models. The reliability of the models was
assessed using the coefficient of determination in
validation (R2V) and the ratio of standard deviation of the
reference data in the validation set to the standard error
of prediction (RPDV). Excellent models were achieved
for TN, OC and pH (RPD≥ 3.5, R2c ≥ 0.91). Good models
were obtained for EC and CEC (RPD from 2 to 3, R2c ≥
0.75), and moderate capability for prediction of particle
size and exchangeable cations (RPD from 1.5 to 1.99, R2c
≥ 0.68). Soil spectra produced acceptable models for
predicting relevant soil structural indicators, and the
mean soil spectra were different between soil structural
classes. Therefore VNIRS has the potential as a non-
destructive and cost-efficient tool for rapid
determination of soil quality indicators.
Key-words: NIR spectroscopy, PLS regression, soil
properties, UV-VIS spectroscopy.
I. INTRODUCTION
Chemical procedures used in the characterization of
soils may in fact further complicate the interpretation.
The extraction procedures can change the equilibrium
between solid and solution phases of soil and the
analyte as a result of the extractant interactions in the
solution and at the solution-particle interface. For this
reason, there is an increasing tendency towards the
development of techniques that preserve the basic
integrity of the soil system [1].There is a widespread
interest to assess soil parameters with quick and cheap
scanning methods instead of using more expensive
standard soil analysis procedures. Among the easy-to-
use alternative techniques, ultra-violet to visible (UV-
VIS) and visible to near-infrared (VIS-NIRS)
spectroscopy have experienced a boom in the last ten
years [2-10]. The ultraviolet (UV, 250–400 nm)
region, a zone of short-wavelength radiation that lies
between the X-ray (30–250 nm) and the visible
regions, has been used for identifying minerals such as
iron and titanium oxides and hydroxides [11].
Although the visible spectrum (VIS, 400–700 nm)
forms a small portion of the electromagnetic spectrum,
it has obvious significance for soil classification. Soil
visible reflectance, or colour, is a differentiating
characteristic for many soil classes in all modern
classification systems and it is an essential part of the
definitions for both surface and subsurface diagnostic
horizons [12]. Obukhov and Orlov reported that soils
with an elevated content of iron could be easily
distinguished by the inflection characteristic for pure
Fe2O3 [13]. They found the intensity of the reflection
in the region from 500 nm to 640 nm to be inversely
proportional to the iron content. In situ and laboratory
soil reflectance measurements in the visible–near-
infrared (Vis–NIR, 350–2400 nm) spectral range are a
promising analytical tool to determine soil properties.
They also have the great advantage to be rapid and
without hazardous chemicals; moreover several
properties can be determined from a single scan [8,14].
In particular, soil reflectance spectra are strongly
dependent on organic matter, as well as to other
properties such as soil moisture and texture [15-18].
However, analysis of soil using NIR has been used for
the determination of pH, electrical conductivity (EC),
soil moisture, organic carbon (OC), cation exchange
capacity (CEC), total carbon, total nitrogen, and
exchangeable cations [10,19]. Accurate VNIRS
2
prediction of total nitrogen (TN) and soil organic
carbon (SOC) has been extensively reported [20-27].
For this study, UV-VIS, VIS–NIR spectroscopic and
chemometric analysis were employed for
classification and assessment of soil structural quality.
To achieve these objectives, several mathematical or
statistical procedures, consisting of a series of
operations, were used, including a principal
component analysis (PCA) based on the soil physical,
biological, and chemical properties and a partial least
squares-regression (PLS-R) to relate spectral
reflectance for measuring soil properties. The
objectives were to relate soil structural quality to
indicators typically used to describe soil quality in
general, and to evaluate the efficiency of soil UV-
VIS–NIR spectra for prediction of those factors
closely related to soil structural quality.
II.MATERIALS and METHODS
A. Site characterization and soil samples
The study area (34°16′N/1°57′E–48°31′N/1°01′E) was
located in Avarzaman town in Hamadan, Iran, and
agro-pedological zones. The investigated arable fields
were conventionally used with regular ploughing,
typical crops being wheat and potato. Soil chemical,
physical and biological properties, which were
considered as potential indicators of soil quality based
on published literature [28-31], were measured in
order to explore the relationship between soil
structural and overall soil quality. Field sampling was
conducted in a 46*7 m2 plot laid out with random
orientation where soil and land cover were uniform
based on visual examination on arrival in the potato
field. Soil material was taken from the top horizon
(approximately 0–15 cm depth) after removing plant
debris; all five samples per plot were pooled to one
composite sample, and unusually dry or wet areas,
headlands, gateways and highly compacted areas were
avoided. Soil samples (n = 210) were taken from the
top soil, sealed and stored in cool, dark conditions
prior to laboratory analysis. The soil collected after
fertilizer application (in winter) and crop
establishment in spring and autumn of 2014,
respectively. The samples were passed through a 2-
mm mesh sieve in preparation for laboratory analyses.
B. Laboratory analysis
Soil samples were dried at 40°C in an oven, ground,
and then sieved to obtain the <2-mm fraction. Standard
laboratory methods were used to measure pH that was
determined by a standard pH meter in a 1:1 soil-water
ratio [32] air-dry gravimetric water content
(gravimetric water content method), and EC (1:5 soil
water extract). Organic carbon was measured by a
modified Walkley and Black method [33], free iron by
the citrate-bicarbonate-dithionate method [34],
particle size distribution by the hydrometer method
[35], CEC calculated using the sum of exchangeable
cations method [36]; extractable calcium (Ca),
magnesium (Mg) and potassium (K) determined using
an inductively coupled plasma–atomic emission
spectrograph [37]; exchangeable Na by the 0.01 M
silver-thiourea method [38]; Total nitrogen (TN)
measured according to the combustion method [39]
using a nitrogen analyzer (LECO Tru-Spec CHN
analyser).
C. Spectral analysis
The spectral reflectance of soil samples was measured
using a UV–VIS (200-1100nm) and VIS–NIR (700-
2400) spectrometer. An AvaSpec-ULS2048-UV-VIS
system (Avantes Co., Netherland) with1-nm intervals
was used and 2-nm intervals in the NIR range with FT-
NIR Spectrum 100N from Perkin Elmer was applied
for acquisition of soil reflectance spectra. VIS-NIR
diffuse reflectance spectra of all samples were
collected in the laboratory by a FT-NIR Rapid Content
Analyzer spectrometer (USA). The spectrometer was
calibrated by an internal white reference before
measurement. The samples were scanned statically by
placing them in Petri dish on the sample window. The
resulting spectra were recorded as log (reflectance−1)
in 2 nm intervals and averaged over 24 scans for each
sample. Spectra were used in the range from 700 to
2400 nm and resampled to 5 nm spectral resolution
which resulted in 141 spectral predictor variables in
the VIS-NIR domain. For the multivariate approaches
original spectra were used without mean-centering,
baseline correction or detrending. Pre-treatment
included scatter correction using the standard normal
variate and detrending (SNVD) to reduce the scatter
and particle size effect, and also to remove the linear
or curvilinear trend of each spectrum [40].
Chemometric analysis
Calibration precision was not improved when the NIR
region only was used, compared to the whole
spectrum. Hence, the whole spectrum was included in
the calibration process. The studied samples (n = 210)
were divided randomly into calibration and validation
set. The calibration set was used to develop a
prediction equation, and the validation set was used to
validate the predictive equation. After selecting the
calibration and validation sets, soil samples were
analyzed for various chemical and physical properties
in the laboratory. The ability of the UV–VIS–NIR
technique to predict a soil property was evaluated
3
using statistical parameters commonly used for the
spectroscopic technique. One example is the
coefficient of determination of measured and
predicted values of soil samples (R2), which measures
the proportion of the total variation accounted for by
the model. The remaining variation is attributed to
random error. The standard error of calibration (SEC)
is the standard deviation of the difference between the
measured and the estimated values for samples in the
calibration set, whereas the standard error of
prediction (SEP) is the standard deviation of the
difference between the measured and the estimated
values for samples in the validation set.
Another statistical parameters used to evaluate
calibrations was the RPD. The RPD is the ratio of the
standard deviation (SD) of the measured values of soil
properties in the prediction set to the SEP.
The Unscambler software (version X10; CAMO
software, Oslo, Norway) was employed to analyze
spectral data and perform PLSR. PLS regression is
similar to PCR (principal component regression), as
both employ statistical rotations to overcome the
problems of high dimensionality and multicollinearity.
In PLSR, the X and Y variables are rotated relative to
the response variables to maximize predictive power
[43]. The second sampling run then starts with another
subset of samples and uses a smaller value of r for a
more rigorous variable selection. In the end, m sets of
selected spectral variables are then compared in a
leave-one-out (LOO) cross-validation (for the
complete set of samples) to identify the optimal subset
of variables with the lowest RMSE.
III. RESULTS and DISCUSSION
A. Soil properties
The reflectance technique requires a calibration set of
samples, which will represent the samples used for
validation. Soil properties estimated by UV–VIS–NIR
method and their values analyzed by laboratory
method are presented in the Table 1. It can be seen that
the range for samples in the validation set was within
the sample range for the calibration set for most of the
soil properties. For pH and Mg soil properties, the
maximum values for the validation set were larger
than the maximum values for the calibration set. Most
of the soil parameters were significantly correlated
with each other (Table 2). As expected, TN and OC
were positively correlated (r= 0.98). The EC was
better correlated with TN (r= 0.73), OC (r = 0.77) and
pH ratio (r = 0.46) than the other labile parameters. In
contrast, CEC was not significantly correlated with
TN or OC, and had a weak correlation with soil pH
ratio (r =0.25) and a moderate correlation with EC (r=
0.32). PLSR models based on UV-vis and VIS-NIR
spectra without spectral variable selection provided
the possibility to distinguish between high and low
values for TN, OC, free iron, CEC, pH and
exchangeable cations values; K and Na were not
estimated successfully in both regions. Higher
accuracies of the PLSR models were obtained with
VIS-NIR spectra, which is in agreement with most
reported studies [5]. Accuracies increased especially
for TN, but also for free iron, Ca and Mg (all
“approximate quantitative”), whereas results for Mg
did not change markedly (Table 3). However,
accuracies obtained in the full spectrum-PLS
approaches were rather moderate when compared to
other results published in numerous studies that
especially exist for OC and TN (see overviews with
standard values in [44,45]). The relatively poor results
may be due to the heterogeneity of the sample set, as
optimal calibration requires limited (but sufficient) set
heterogeneity [44]. R2 <0.50 was considered
unacceptable in this study, suggesting that prediction
for EC (only in UV-VIS), silt (only in VNIR),
exchangeable K, and exchangeable Na was poor.
Using cross validations on a dataset generally gives an
over-optimistic indication of the actual performance of
the models. When new and totally independent
samples are predicted with a ‘young’ calibration, it is
very rare to get an SEP at the same level as the SECV
[46]. At the same time, researchers have used different
calibration strategies for their proposed methods; for
example, Chang et al. were able to get R2 values of 0.8
for soil moisture, sand, silt, CEC, and exchangeable
Ca [19]. The predictions for several soil properties in
this study can be compared with the prediction
obtained by Ben-Dor and Banin who investigated 91
soil samples for simultaneous prediction of clay,
specific surface area, CEC, air-dry hygroscopic
moisture, carbonate content, and organic matter using
separate validation sets of samples to test the
calibration accuracy [10]. The prediction of clay in our
study follows a similar pattern to that obtained by
those authors. The validation models for each
indicator before performing pre-processing methods
were summarized in Table. 3. Following initial
analysis, Air dry gravimetric water content, free iron
and TN showed the most accurate prediction models,
and exchangeable cations had the lowest prediction
model. The pre-processing techniques improved the
model accuracy for soil properties (Table 4), but
different techniques emerged for each indicator. Mean
center standard scale was effective for improving the
4
accuracy of pH, OC, TN and free iron; 1st derivatives
for EC and CEC; mean normalization and smoothing
for exchangeable cations (Mg, K, Ca and Na);
maximum normalization for particle size. An excellent
validation was achieved for pH, OC and TN (RPD >
3.5, R2c = 0.91). A good model was obtained for free
iron, CEC and EC (RPD from 2 to 3, R2c ≥ 0.75), and
moderate capability for particle size and exchangeable
cations (RPD from 1.5 to 1.99, R2c ≥ 0.68). Chang et
al. reported that the NIR reflectance spectroscopy
technique had the ability to predict various properties
of soil and they used 3 categories based on RPD in the
ranges >2.0, 1.4–2.0, and <1.4 to indicate decreasing
reliability of prediction using this technique [19]. The
RPD values obtained in the present study were within
1.56–4.51 for pH (3.67), OC (4.28), TN (4.51), CEC
(2.85), exchangeable cations (1.97) and particle size
(1.56) which suggests a less reliable prediction for
unknown soil samples.
B. Assessing of spectral methods for predicting
soil properties
Effective, low-cost, quantitative evaluation of soil
structural quality linked to overall soil quality would
allow better decisions for managing agricultural
systems. In a cost-benefit analysis for determination of
SOC that compared conventional verses VIS-NIR
spectroscopic techniques [47], spectroscopic methods
were more cost effective because of operator time
(representing 40% of total sample cost for dry
combustion, compared to 2% to 5% for spectroscopic
methods). Simultaneous and accurate prediction of
soil indicators using VIS-NIR spectroscopy is now
well established [48], and this study has shown that a
direct application of soil spectroscopic analysis can be
used for soil structure management and for monitoring
of soil structural quality. One of the advantages of
using spectroscopic techniques is to model the
integrative impact of soil properties for multiple
aspects of evaluation [49] that can be used for multi-
dimension prediction of soil quality. Since soil
structural quality was associated with some indictors
that could be predicted accurately by spectral data and
the difference of the mean absorbance spectra of UV-
VIS and VIS– NIR spectroscopy can offer reliable and
quantitative, low cost evaluation of soil structural
quality.
IV. CONCLUSION
A small fraction of the complete number of spectral
variables provided by VIS-NIR and UV_VIS spectral
measurements has shown to be sufficient for
successful cross-validation performances. Compared
to chemical methods it is costly and time consuming,
but compared to a complete array for soil sampling and
laboratory analysis it has the potential to be very
efficient. The approach might also be appealing
because it takes away issues of operator variation and
bias that might occur with large-scale deployment of
visual methods in the field. The results of this study
show that simultaneous prediction of pH, air-dry
gravimetric water content, OC, clay, CEC,
exchangeable cations is possible using the UV–VIS–
NIR technique. V.Acknowledgment
Author wishes to thank the Faculty of Agriculture and
Chemistry, University of Bu-ali Sina, for the
contribution in this research.
REFERENCES
[1] L. J. Janik, R. H. Merry, J.O. Skjemstad, “Can mid
infrared diffuse reflectance analysis replace soil
extractions?” Australian J. of Experimental
Agriculture, vol.38, pp.681–696, 1998.
[2] M. Vohland, M. Ludwig, S. Thiele-Bruhn, B.
Ludwig, “Determination of soil properties with visible
to near- and mid-infrared spectroscopy: Effects of
spectral variable selection”, Geoderma, 223–225
pp.88–96, 2014.
[3] T. Paz-Kagan, M. Shachak, E. Zaady, A. Karnieli,
“A spectral soil quality index (SSQI) for
characterizing soil function in areas of changed land
use”, Geoderma 230–231, pp.171–184, 2014.
[4] M. Conforti, G. Buttafuoco, P. A. Leone, P. C.
Aucelli, G. Robustelli, F. Scarciglia, “Studying the
relationship between water-induced soil erosion and
soil organic matter using Vis–NIR spectroscopy and
geomorphological analysis: A case study in southern
Italy”, Catena, vol.110, 44–58, 2013.
[5] V. Bellon-Maurel, A. McBratney, “Near-infrared
(NIR) and mid-infrared (MIR) spectroscopic
techniques for assessing the amount of carbon stock in
soils — critical review and research perspectives”,
Soil Biol. Biochem. Vol.43, pp.1398–1410. 2011.
[6] B. Stenberg, R. A. Viscarra Rossel, A. M.
Mouazen, and J. Wetterlind, “Visible and Near
Infrared Spectroscopy in Soil Science”, Advances in
Agronomy, vol.107, Elsevier Inc. 2010
[7] R. A. Viscarra-Rossel, “Robust modelling of soil
diffuse reflectance spectra by bagging-partial least
squares regression”, Near Infrared Spectroscopy,
vol.15, pp. 37–47, 2007.
[8] D. J. Brown, K. D. Shepherd, M. G. Walsh, M.
Dewayne Mays, T. G. Reinsch, “Global soil
haracterization with VNIR diffuse reflectance
5
spectroscopy”, Geoderma, vol. 132, pp.273–290.
2006.
[9] K. D. Shepherd, M. G. Walsh, “Development of
reflectance spectral libraries for characterization of
soil properties”, Soil Science Society, vol. 66, pp.
988–998, 2002.
[10] E. Ben-Dor, A. Banin, “Near-infrared analysis as
a rapid method to simultaneously evaluate several soil
properties”, Soil Science Society of America Journal
vol.59, pp. 364–372, 1995.
[11] R. G. J. Strens, B. J. Wood, “Diffuse reflectance
spectra and optical properties of some iron and
titanium oxides and hydroxides”, Mineralogical
Magazine, vol.43, pp. 347–354, 1979.
[12] M. F. Baumgardner, L. F. Silva, L. L. Biehl, E. R.
Stoner, “Reflectance properties of soils”, Advances in
Agronomy, vol.38, pp.1–44, 1985.
[13] A. I. Obukhov, D. S. Orlov, “Spectral reflectivity
of the major soil groups and possibility of using
diffuse reflection in soil investigations”, Soviet Soil
Science, vol.2, pp.174–184, 1964.
[14] R. A. Viscarra Rossel, D. J. J. Walvoort, A. B.
McBratney, L. J. Janik, J. O. Skjemstad, “Visible, near
infrared, mid infrared or combined diffuse reflectance
spectroscopy for simultaneous assessment of various
soil properties”, Geoderma vol.131, pp.59–75, 2006.
[15] W. Schwanghart, T. Jarmer, “Linking spatial
patterns of soil organic carbon to topography — a case
study from south-eastern Spain”, Geomorphology
vol.126, pp.252–263, 2011.
[16] M. Ladoni, H. A. Bahrami, S. K. Alavipanah, A.
A. Norouzi, “Estimating soil organic carbon from soil
reflectance: a review”, Precision Agriculture, vol.11,
pp. 82–99, 2010.
[17] H. Aïchi, Y. Fouad, C. Walter, R. A. Viscarra
Rossel, Z. L. Chabaane, M. Sanaa, “Regional
predictions of soil organic carbon content from
spectral reflectance measurements”, Biosystem
Engineering, vol.104, pp. 442–446, 2009.
[18] A. Stevens, B. Van Wesemael, H. Bartholomeus,
D. Rosillon, B. Tychon, E. Ben-Dor, “Laboratory,
field and airborne spectroscopy for monitoring organic
carbon content in agricultural soils”, Geoderma,
vol.144, pp. 395–404, 2008.
[19] C. W. Chang, D. A. Laird, M. J. Mausbach, C.R.
Hurburgh, “Near-infrared reflectance spectroscopy –
principal components regression analyses of soil
properties”, Soil Science Society of America Journal
vol.65, no.2, pp.480–490, 2001.
[20] T. Shi, L. Cui, J. Wang, T. Fei, Y. Chen, G. Wu,
“Comparison of multivariate methods for estimating
soil total nitrogen with visible/near-infrared
spectroscopy”, Plant Soil, vol. 366, pp. 363–375,
2013.
[21] M. St. Luce, N. Ziadi, J. Nyiraneza, G. F.
Tremblay, B. J. Zebarth, J. K. Whalen, M. Laterrière,
“Near infrared reflectance spectroscopy prediction of
soil nitrogen supply in humid temperate regions of
Canada”, Soil Science Society of America, vol.76,
pp.1454–1461, 2011.
[22] H. T. Xie, X. M. Yang, C. F. Drury, J. Y. Yang,
X. D. Zhang, “Predicting soil organic carbon and total
nitrogen using mid- and near-infrared spectra for
Brookston clay loam soil in Southwestern Ontario,
Canada”, Canadian Journal of Soil Science, vol. 91,
pp. 53–63, 2011.
[23] C. Nduwamungu, N. Ziadi, L. É. Parent, G. F.
Tremblay, L. Thuriès, “Opportunities for, and
limitations of, near infrared reflectance spectroscopy
applications in soil analysis: a review”, Canadian
Journal of Soil Science, vol. 89, pp. 531–541, 2009.
[24] Zornoza, R., Guerrero, C., Mataix-Solera, J.,
Scow, K.M., Arcenegui, V., Mataix-Beneyto, J., 2008.
Near infrared spectroscopy for determination of
various physical, chemical and biochemical properties
in Mediterranean soils. Soil Biol. Biochem. 40, 1923–
1930.
[25] Brunet, D., Barthès, B.G., Chotte, J.-L., Feller, C.,
2007. Determination of carbon and nitrogen contents
in Alfisols, Oxisols and Ultisols from Africa and
Brazil using NIRS analysis: effects of sample grinding
and set heterogeneity. Geoderma 139, 106–117.
[26]Chang, C.-W., Laird, D.A., 2002. Near-infrared
reflectance spectroscopic analysis of soil C and N. Soil
Sci. 167, 110–116.
[27] Fystro, G., 2002. The prediction of C and N
content and their potential mineralisation in
heterogeneous soil samples using Vis-NIR
spectroscopy and comparative methods. Plant Soil
246, 139–149.
[28] Lima, A.C.R., Brussaard, L., Totola, M.R.,
Hoogmoed, W.B., de Goede, R.G.M., 2013. A
functional evaluation of three indicator sets for
assessing soil quality. Applied Soil Ecology 64, 194–
200.
[29]Qi, Y., Darilek, J.L., Huang, B., Zhao, Y., Sun,
W., Gu, Z., 2009. Evaluating soil quality indices in an
agricultural region of Jiangsu Province, China.
Geoderma 149 (3– 4), 325–334.
[30] Masto, R., Chhonkar, P., Singh, D., Patra, A.,
2008. Alternative soil quality indices for evaluating
the effect of intensive cropping, fertilisation and
manuring for 31 years in the semi-arid soils of India.
6
Environmental Monitoring and Assessment 136 (1–3),
419–435.
[31] Ditzler, C.A., Tugel, A.J., 2002. Soil quality field
tools. Agronomy Journal 94 (1), 33–38.
[32]Thomas, G.W., 1996. Soil pH and soil acidity. In:
Sparks, D.L. (Ed.), Methods of Soil Analysis. Part 3.
Chemical Methods. Soil Science Society of America
Book Series No. 5, pp. 475–490.
[33] McCleod S (1973) Studies on wet oxidation
procedures for the determination of organic carbon in
soil. In‘Notes on soil techniques’. pp. 73–79. (CSIRO
Division of Soils: Melbourne)
[34] Mehra OP, Jackson ML (1960) Iron oxide
removal from soils and clays by a dithionite-citrate
system
buffered with sodium bicarbonate. Clays and Clay
Minerals 7, 317–327.
[35] Gee GW, Bauder JW (1986) Particle size
analysis. In ‘Methods of soil analysis’. Part 1, 2nd edn
(Ed. A Klute) pp. 383–411. (American Society of
Agronomy: Madison, WI)
[36] Sumner, M.E., Miller, W.P., 1996. Cation
Exchange Capacity and Exchange Coefficients. In:
Methods of Soil Analysis. Part 3 Chemical Methods.
Soil Science Society of America Inc., Madison, pp.
1201–1229.
[37] Soltanpour, P.N., Johnson, G.W., Workman,
S.M., Jones, J.B., 1996. Inductively Coupled Plasma
Emission Spectrometry and Inductively Coupled
Plasma-Mass Spectrosmetry. In: Methods of Soil
Analysis. Part 3 Chemical Methods. Soil Science
Society of America Inc. pp. 91–139.
[38] Rayment GE, Higginson FR (1992) ‘Australian
laboratory handbook of soil and water chemical
methods.’(Inkata Press: Melbourne)
[39] Matejovic, I., 1997. Determination of carbon and
nitrogen in samples of various soils by the dry
combustion. Communications in Soil Science and
Plant Analysis 28 (17–18), 1499–1511.
[40] Barnes, R.J., Dhanoa, M.S., Lister, S.J., 1989.
Standard normal variate transformation and de-
trending of near-infrared diffuse reflectance spectra.
Appl. Spectrosc. 43, 772–777.
[41] Savitzky, A. and Golay, M. J. E. 1964. Smoothing
and differentiation of data by simplified least squares
procedures. Analytical Chemistry, 36(8): 1627-1639.
[42] Martens H, Naes T (1987) ‘Multivariate
calibration by data compression.’ (John Wiley and
Sons: Chichester, UK)
[43] Wold, S., Sjöström, M., Eriksson, L., 2001. PLS-
regression: a basic tool of chemometrics. Chemom.
Intell. Lab. 58, 109–130.
[44] Ce´ cillon, L., Barthe` s, B.G., Gomez, C., Ertlen,
D., Genot, V., Hedde, M., Stevens, A., Brun, J.J.,
2009. Assessment and monitoring of soil quality using
near-infrared reflectance spectroscopy (NIRS).
European Journal of Soil Science 60 (5), 770– 784.
[45] Kuang, B., Mahmood, H.S., Quraishi, M.Z.,
Hoogmoed, W.B., Mouazen, A.M., van Hentent, E.J.,
2012. Sensing soil properties in the laboratory, in situ,
and on-line: a review. Adv. Agron. 114, 155–223.
[46] Dardenne P, Sinnaeve G, Baeten V (2000)
Multivariate calibration and chemometrics for near
infrared spectroscopy: which method? Journal of Near
Infrared Spectroscopy 8, 229–237.
[47] O’Rourke, S.M., Holden, N.M., 2011. Optical
sensing and chemometric analysis of soil organic
carbon – a cost effective alternative to conventional
laboratory methods? Soil Use and Management 27 (2),
143–155.
[48] Kinoshita, R., Moebius-Clune, B.N., van Es,
H.M., Hively, W.D., Bilgilis, A.V., 2012. Strategies
for soil quality assessment using visible and near-
infrared reflectance spectroscopy in a Western Kenya
Chronosequence. Soil Science Society of America
Journal 76 (5), 1776–1788.
[49] Cohen, M.J., Prenger, J.P., DeBusk, W.F., 2005.
Visible-near infrared reflectance spectroscopy for
rapid, nondestructive assessment of wetland soil
quality. Journal of Environmental Quality 34 (4),
1422–1434.
Table1. Statistical summary of soil properties used for calibration and validation sets
Soil properties
Calibration set (157)
Validation set (53)
Mean
Min.
Max.
SD
Mean
Min.
Max.
SD
pH
7.20
3.40
9.31
1.21
6.8
4.86
9.88
1.13
EC (mS/cm)
0.22
0.01
1.52
0.25
0.16
0.01
0.81
0.15
TN (%)
1.04
0.57
2.00
1.00
0.46
0.14
1.89
0.97
OC (%)
0.99
0.09
4.99
0.89
1.17
0.16
4.01
0.84
Air dry gravimetric
water content (g/g)
0.06
0.01
0.13
0.04
0.03
0.01
0.08
0.02
Free iron (%)
0.78
0.09
4.11
0.82
0.69
0.05
2.98
0.59
7
Clay (%)
38.93
2.01
69.85
19.01
30.63
6.67
67.04
16.58
Sand (%)
44.00
9.02
99.05
23.12
151.11
15.04
89.85
22.02
Silt (%)
17.61
0.01
40.01
8.89
18.74
3.46
43.32
9.58
CEC (mmolc/kg)
163.03
26.04
315.92
77.49
127.98
15.67
239.78
70.03
Ca (mmolc/kg)
35.62
0.01
94.73
23.45
25.64
0.13
80.94
22.47
Mg (mmolc/kg)
33.00
0.46
91.05
20.02
25.43
3.20
94.64
19.89
K(mmolc/kg)
8.88
1.32
40.09
7.06
8.02
0.72
25.62
5.51
Na(mmolc/kg)
17.09
0.23
107.87
22.42
14.87
0.22
72.83
18.02
Table 2. Correlation matrix (Pearson’s r) for studied soil properties (at p<0.001)
Indicators
Free iron
CEC
EC
TN
OC
pH
Free iron
1.00
CEC
0.22
1.00
EC
-0.23
0.32
1.00
TN
0.52
-0.05
0.73
1.00
OC
0.48
0.00
0.77
0.98
1.00
pH
0.16
0.25
0.46
0.57
0.66
1.00
Table 3. Statistics of estimates UV-VIS and VNIR spectra using PLS regression (LOO cross validation)
Soil properties
UV-VIS
VIS-NIR
R2
RMSE
RPDV
R2
RMSE
RPDV
pH
0.70
0.62
2.1
0.73
0.51
1.5
EC (mS/cm)
0.43
0.12
1.6
0.51
0.11
1.1
TN (%)
0.67
0.38
1.4
0.77
0.28
1.0
OC (%)
0.61
0.46
2.9
0.64
0.31
1.8
Air dry gravimetric
water content (g/g)
0.88
0.09
1.3
0.87
0.04
1.2
Free iron (%)
0.64
0.28
2.5
0.81
0.15
1.7
Clay (%)
0.78
0.39
1.8
0.74
0.37
1.3
Sand (%)
0.72
3.26
1.6
0.68
2.79
1.2
Silt (%)
0.59
5.18
2.9
0.43
5.61
2.2
CEC (mmolc/kg)
0.66
2.94
2.3
0.75
3.14
1.9
Ca (mmolc/kg)
0.69
1.49
2.8
0.79
0.98
2.0
Mg (mmolc/kg)
0.75
1.68
3.0
0.81
1.17
1.9
K(mmolc/kg)
0.12
2.05
2.7
0.34
1.89
1.6
Na(mmolc/kg)
0.44
7.32
3.6
0.49
3.05
2.1
Table 4. Optimum NIR spectral pre-processing for features related to soil structural quality
Features
Optimum pre-
processing
RPD
Model
R2
RMSE
pH
Mean center,
std scale
3.67
Calibration
0.93
0.44
Validation
0.85
0.47
OC
Mean center,
std scale
4.28
Calibration
0.91
0.17
Validation
0.83
0.26
TN
Mean center,
std scale
4.51
Calibration
0.94
0.07
Validation
0.89
0.12
EC
First
derivative
2.26
Calibration
0.75
0.32
Validation
0.64
0.32
CEC
First
derivative
2.85
Calibration
0.78
0.78
Validation
0.69
0.94
8
Free iron
Mean center,
std scale
3.19
Calibration
0.82
0.27
Validation
0.77
0.31
Particle size
Maximum
normalization
1.56
Calibration
0.68
2.05
Validation
0.57
2.18
Exchangeable
cations
Mean
normalization,
smoothing
1.97
Calibration
0.70
24.02
Validation
0.61
32.10
