Available online at:

http://journal.unila.ac.id/index.php/tropicalsoil
J Trop Soils, Vol. 16, No. 2, 2011: 93-98 93
DOI: 10.5400/jts.2011.16.2.93

Determination and Prediction of Some Soil Properties using

Partial Least Square (PLS) Calibration and Mid-Infra Red (MIR)
Spectroscopy Analysis

Irnanda Aiko Fifi Djuuna1, Lynette Abbott2 and Craig Russell2

1
Department of Soil Sciences, Faculty of Agriculture and Agriculture Technology, The State University of
Papua, Gunung Salju St. Amban Manokwari 98314, West Papua, Indonesia. Tel. +62-0986- 211830; fax:
+62-0986-213513 E-mail address: [email protected]
2
School of Earth and Geographycal Science, The University of Western Australia, Crawley 6009
Received 18 June 2010 / accepted 25 April 2011

ABSTRACT

Soil chemical, physical and biological analyses are a crucial but often expensive and time-consuming step in the
characterization of soils. Rapid and accurate predictions and relatively simple methods are ideally needed for soil
analysis. The objective of this study was to predict some soil properties (e.g. pH, EC, total C, total N,C/N, NH4-N,
NO3-N, P, K, clay, silt, and sand and soil microbial biomass carbon) across the Wickepin farm during summer
season using a Mid-Infra Red - Partial Least Square (MIR–PLS) method. The 291 soil samples were analyzed both
with soil extraction procedure and MIR Spectrometer. Calibrations were developed between MIR spectral data
and the results of soil extraction procedures. Results using the PLS-MIR showed that MIR-predicted values were
almost as highly correlated to the measured value obtained by the soil extraction method of total carbon, total
nitrogen and soil pH. Values for EC, NH4-N, NO3-N, C/N, P, K, clay, silt, sand, and soil microbial biomass carbon
were not successfully predicted by the MIR – PLS technique. There was a tendency for these factors to correlate
with the MIR predicted value, but the correlation values were very low. This study has confirmed that the MIR-PLS
method can be used to predict some soil properties based on calibrations of MIR values.

Keywords: MIR-Partial Least Square, MIR-Spectroscopy, soil properties

INTRODUCTION frequencies of this technique are generally in the

MIR 2,500-25,000 nm wavelengths or 4,000-250
Soils are r arely homogeneous and the cm-1 range (Janik et al. 1998). This method has
variability occurs both laterally and with depth. It been used to determine some macro and micro
can result from changes in the chemical balance of elements of soil including soil carbon (McCarty et
the soil associated with agricultural practices al. 2002), organic matter composition (Cheshire et
including nutrient uptake, crop rotations, fertilizer al. 1993; Spaccini et al. 2001) and many other soil
use, lime application (Viscarra Rossel and properties (Janik et al. 1998) as well as soil minerals
McBratney 1998) and leaching. To understand (Janik et al. 1995; Janik and Skjemstad 1995).
more about the variation in soil properties, spatially However, none of the above studies were conducted
dense soil analyses are often required. Soil under agricultural systems which had different land
analytical procedures also need to be able to cope use histories or at different occasions.
with a large number of samples. Mid infrared-partial least square (MIR-PLS)
An alternative method for assessing soil has not been widely used for soil property prediction
properties and their variability across a landscape in agricultural systems. However, some studies
is mid infrared (MIR) spectr oscopy. MIR have been conducted for (i) prediction of organic
spectroscopy has considerable advantages and carbon, nitrogen and other properties of peat soil
offers a possible alternative to conventional (Holmgen and Norden 1988), (ii) the contamination
methods through increased speed and sensitivity of hydrocarbon in wet soil (Hazel et al. 1997), (iii)
(Janik et al. 1995). The fundamental molecular a range of soil properties including mineralogical
analysis, organic carbon, nitrogen, carbonate, air-
dry moisture and cation exchange capacity (Janik
J Trop Soils, Vol. 16, No. 2, 2011: 93-98 and Skjemstad 1995; Janik et al. 1995; Murphy and
ISSN 0852-257X Milton 2004) and (iv) prediction of total carbon of
94 IA F Djuuna et al.: Determination and Prediction Soil Properties using PLS and MIR

soil, clay content and soil N supply (Murphy and The objective of this study was to predict some
Milton 2004). soil properties across the farm during summer season
In relation to soil biological characteristics, the using MIR –PLS method.
MIR spectral contribution of in-situ soil
microorganisms is expected to be minor because MATERIALS AND METHODS
only about 5% of the total organic carbon in soil is
microbial carbon (Janik et al. 1998). The reason Location and Soil Sampling
for this is that the signatures of microorganisms are
usually difficult to detect in situ in most soils due Two hundred ninety one (291) soil samples
to masking of soil mineral and other organic peaks. were collected from a farm (Fairlawn) in the
However, a study by Pankhurst et al. (1997) showed Wickepin Shire in south-western Australia, about
that there was a correlation between GC-FAME 300 km south east of Perth, 32° 47' 4'’ S, 117° 30' 9'’
spectra and MIR spectra of the same soils using E during summer season in 2004. The farm covers
PLS regression analysis. The correlation values an area of 647 ha and was divided into 14 paddocks.
ranged between 0.5 to 0.75, showing that the MIR The region had a Mediterranean climate with hot,
spectra could explain 75% of the variability in the dry summers and cold winters, and elevations ranged
FAME data. This study suggested that MIR spectra from about 250 to 350m. The area was dominated
can be obtained much more rapidly compared with by flats with some rolling hills. The major soil
the conventional methods for measuring bacteria textures in this area were medium loam, sandy loam,
and fungi. Therefore, there is an opportunity to clayey loam and clay. The major land use was
use this predictive capacity of MIR-PLS models as principally sheep farming and wheat production,
a simple bio-monitor of soil health and for soil with annual pasture dominated by grass species.
biological mapping. Soil Analysis
Another study (Grube et al. 1999) showed that
the principal component concentration changes in The soil samples were oven dried, ground and
microbial cells can be studied by quantitative sieved to a size fraction smaller than 2 mm. The
estimation of the biomass MIR spectra. In addition, number of samples collected and the soil analyses
MIR absorption spectra had also been applied to performed are summarized in Table 1.
determine bacterial biomass (Zymomonas mobilis)
which was grown in sucrose or glucose medium MIR Analysis
(Grube et al. 2002). Two hundred ninety one (291) air dried soil
samples were ground in an agate mill at 750 rpm to

Table 1. Laboratory methods of soil analyses.

Soil property n Technique References

pH H2O 291 1:5 soil: H 2O extract White (1969)
Sand, silt, clay (%) 291 Pipette method Rayment and
Higginson (1992)
Nitrate–nitrogen (NO3-N) 291 Skalar method from Skalar Rayment and Higginson (1992)
and ammonium–nitrogen SANp lu s Segmented Flow
(NH4-N) (mg kg-1) Autoanalyzer
Available phosphorus 291 Bicarbonate extractable P Olsen et al (1954), Rayment and
-1
Colwell (PCol) (mg kg ) (Spectrophotometer) Higginson (1992)
Exchangeable potassium 291 Bicarbonate extractable K Rayment and Higginson (1992)
(K) (mg kg-1) (AAS)
Electrical conductivity 291 1:5 soil :H 2O extract Rayment and Higginson (1992)
-1
(EC) (dS m )
Microbial ciomass C (mg 291 Fumigation-Extraction (FE) Vance et al. (1987) and Sparling
kg-1) method (1990).
Total C and total N (%) 291 CHN-1000 Elemental Rayment and Higginson (1992)
Analyzer version 1.1 (1991)
LECO
 J Trop Soils, Vol. 16, No. 2, 2011: 93-98 95

pass through a 0.2 mm mesh. A small amount of mean standard error of prediction (RMSEP), and
the samples was poured into 10 mm diameter the standard error of prediction (SEP). The values
stainless steel cups and the top of the powders were in the prediction set were recorded and this
leveled without applying pressure. Samples were procedure was repeated two times with different
scanned using MIR from 7,000 to 500 cm-1 at 2 samples in the validation set until all samples were
cm-1 resolution on a Spectrum One Fourier- predicted.
Transform mid Infrared (FT-IR) Spectrometer Vers. The best calibration and prediction model were
B Model L120000B. Each sample was scanned in then judged against the one with the lowest standard
duplicate and the spectra was averaged. The error of prediction and the highest correlation
infrared spectra were recorded by the diffuse coefficients. Two other statistical analysis were
reflectance (DRIFT) sampling accessory. Peak used to evaluate the calibration: (i) RPD, the ratio
areas from spectra were used for statistical analysis. of the standard deviation (SD) of the measured or
references value in the prediction set to root mean
Data Analysis square of error prediction (RMSEP), and (ii) RER,
The data files containing the spectral data the ratio of the range of measured/references value
points were imported as GRAMS format for the in the prediction set to the RMSEP (Malley et al.
PLS calculation. The MIR calibrations were 1999).
performed by Partial Least Squares (PLS)
RESULTS AND DISCUSSION
regression using the computer program for
multivariate modeling ‘Unscrambler’ Version 7.5 DRIFT Spectra for Whole Soil
(Camo A/S, Trondheim, Norway). The PLS was
used to determine the best correlation between The DRIFT spectra for a typical soil from the
chemical reference data and spectral data to develop Wickepin farm are shown in Fig. 1. The high peak
a calibration. The PLS regression model was built around 900-500 cm-1 reflected other clay mineral
using a calibration set with 2/3 of the spectra of characteristics (Janik and Skjemstad 1995; Nguyen et
213 and 291 for the overall data randomly selected al. 1991). This showed that mid infrared spectra
observations. Cross validation using partial least contain definable peaks which could be used in spectral
regression was performed for each soil variable. interpretation and differentiate between two or more
The predictive model was performed using samples.
validation of 1/3 of the unused data set of
independent 78 and 70 of the whole observations. Soil Characteristics
The calibration and predicted model were evaluated The summary of soil characteristics across the
from their correlation coefficients (r), r2, the root farm is presented in Table 2.

2. 01

1. 9
1. 8

1. 7

1. 6
Absor bance

1. 5

1. 4

1. 3
A
1. 2

1. 1

1. 0

0. 9

0. 8

0. 7

0. 55
4000. 0 3000 2000 1500 1000 500.0
cm -1
Wa ve n u m b er

Figure 1. Mid-infrared spectra of all soil samples from the farm at Wickepin in southwestern Australia.
96 IA F Djuuna et al.: Determination and Prediction Soil Properties using PLS and MIR

Table 2. Summary statistics of 291 soil samples for 13 soil characteristics at a farm scale at the
Wickepin farm (soils sampled in summer).

Standard
Soil Variables Mean Min Max
Deviation
-1
NH4 -N (mg kg ) 2.11 1.03 0.24 7.19
-1
NO3 -N (mg kg ) 6.54 4.08 0.00 25.87
pH H2O 5.80 0.33 5.21 7.51
EC (µS cm-1) 148.22 319.99 20.40 5021.00
Total C (%) 1.48 0.53 0.41 5.27
Total N (%) 0.11 0.04 0.00 0.42
C/N 13.79 7.62 8.50 136.33
-1
P (mg kg ) 72.39 69.78 11.04 569.01
-1
K (mg kg ) 45.12 17.25 8.20 118.00
Microbial Biomass-C (mg kg-1) 443.47 272.58 8.88 1672.51
Clay (%) 5.707 2.811 0.00 18.37
Silt (%) 7.272 9.004 0.00 55.01
Sand (%) 87.038 9.884 39.57 99.50

Calibration of Results also followed by acceptable RPD and RER values

(Table 3); RPD>3 and RER>10 are considered
The results obtained from the cross validation acceptable. This indicated that the infrared spectra
procedure are presented in Table 3 and confirmed could explain 76% of EC and soil pH, 86% of total
the feasibility of predicting some soil variables. nitrogen and 90% of total carbon of the variability
Total C, total N, EC and soil pH were successful in the laboratory extraction analysis data. Other
predicted compared to other soil properties for soil properties were not successfully predicted using
which the correlation values were low. The best this technique because the correlation values were
correlation for total C, total N and soil pH were lower than 50%.

Table 3. Independent validation set of 71 soil samples with the 213 samples
used to develop a calibration of 2002 soil samples from Wickepin farm
(MBC=microbial biomass carbon; RMSEP= root mean square of error
prediction; RPD= ratio of standard deviation to standard error of
prediction (SD/SEP); RER= ratio of range to SEP).

2
Soil variables R RMSEP RPD RER

pH 0.76 0.22 0.98 5.37

EC 0.76 3.05 1.09 13.60
Total C 0.90 0.23 2.00 15.00
Total N 0.86 0.02 1.96 16.66
C/N 0.01 8.81 0.12 0.73
N-NH4 0.43 0.93 0.60 2.89
N-NO3 0.42 3.65 0.55 2.11
P 0.48 64.04 0.50 2.01
K 0.03 17.33 0.14 0.61
Clay 0.50 6.33 0.54 2.83
Silt 0.26 39.12 0.18 0.95
Sand 0.01 15.99 0.18 0.95
Microbial Biomass-C 0.26 251.28 0.58 2.65
 J Trop Soils, Vol. 16, No. 2, 2011: 93-98 97

Among the soil variables measured in this applicable because only the calibration for soil pH
study, only total carbon, total nitrogen and soil pH data corresponded well to the other data, but this needs
were best predicted with a PLS-MIR calibration. to be further investigated.
Total soil carbon is the most common soil property
used in NIR and MIR studies and was reported to REFERENCES
have very high correlations with conventional soil
analytical data compared to other soil properties Brown DJ, RS Bricklemyer and PR Miller. 2005.
(e.g. McCarty et al. 2002). However, the study by Validation requirements for diffuse reflectance soil
Brown et al. (2005) to predict soil C using the same characterization models with a case study of VNIR
technique failed at some sites due to “pseudo- soil C prediction in Montana. Geoderma 129 (3-
4): 251-267.
independent” validation (random selection of non-
Cheshire MV, JD Russell and AR Fraser. 1993. The
independent test samples) that can overestimate decomposition of straw in soil: use of multiple
predictive accuracy relative to independent internal reflectance infrared spectroscopy. J Soil
validation. Sci 44: 693-702.
A major advantage of diffuse reflectance Grube M, J Zagreba, E Gromozova and M Fomina. 1999.
spectroscopy for soil analysis is that from a single Comparative investigation of the macromolecular
spectrum many properties may be (accurately) composition of mycelia forms Thielavia terrestris by
determined, thus offering the possibility for infrared spectroscopy. Vib Spectrosc 19: 301-306.
considerable cost savings and increased efficiency Grube M, M Bekers, D Upite, and E Kaminska. 2002.
over conventional laboratory analysis. Furthermore, IR-spectroscopic studies of Zymomonas mobilis and
levan precipitate. Vib Spectrosc 28: 277-285.
the technique is rapid, making it possible to analyze
Hazel G, F Bucholtz, GN Anngarwal and KJ Ewing. 1997.
a large number of samples in a practical and timely Multivariate analysis of Mid-IR spectra of
manner. These properties make spectroscopic hydrocarbon-contaminated wet soils. Appl
analyses combined with PLSR very attractive for Spectrosc 51: 984-989.
environmental monitoring, modeling and precision Holmgen A and B Norden. 1988. Characterization of
agriculture (Viscara Rossel et al. 2006). peat samples by diffuse reflectance FT-IR
Other studies have also shown that not all spectroscopy. Appl Spectrosc 42: 225-262.
conventional soil extractions can be predicted using Janik LJ and JO Skjemstad. 1995. Characterization and
the MIR-PLS technique (Janik et al. 1998). They analysis of soils using Mid infrared Partial Least-
showed that soil properties such as EC, K-available, Squares 2. Correlations with Some Laboratory Data.
Aust J Soil Res 33: 637-650.
P-available, P buffer, DTPA-extractable Zn and Cu,
Janik LJ, JO Skjemstad and MD Raven. 1995.
CaCl2 , exchangeable Na and K had very low Characterization and analysis of soils using mid-
correlations (r2 = <0.50) between conventional and infrared partial least squares. I. Correlations with
infrared assays. Reeves et al. (2001) found that XRF-determined major element composition. Aust
soil pH (r2 = 0.94) was significantly better than other J Soil Res 33: 621-636.
soil properties such as total carbon and nitrogen, Janik LJ, RH Merry and JO Skjemstad. 1998. Can mid
and some biological properties of soil. Other soil infrared diffuse reflectance analysis replace soil
properties measured in this study were less well extractions? Aust J Exp Agr 38: 681-696.
correlated using the MIR-PLS technique. Although Malley DF, L Yesmin, D Wray and S Edwards. 1999.
the prediction was less successful for these Application of near-infrared spectroscopy in analysis
of soil mineral nutrients. Comm Soil Sci Plant Anal
variables, MIR may be useful for screening
30: 999-1012.
purposes. McCarty GW, JB Reeves, VB Reeves, RF Follett and JM
Kimble. 2002. Mid-infrared and near-infrared diffuse
CONCLUSIONS reflectance spectroscopy for soil carbon measurement.
Soil Sci Soc Am J 66: 640-646.
The calibration values can be developed using Murphy DV and N Milton. 2004. Mapping biological
mid infrared spectra and conventional assays for soil nitrogen supply using mid infrared technology.
several soil parameters, particularly total carbon, In: Lines-Kelly (ed) Soil Biology in Agriculture.
total nitrogen and soil pH. This study also provides Proceeding of a workshop on current research into
evidence that total carbon, total nitrogen and soil soil biology in agriculture. Tamworth Sustainable
Farming Training Centre 11-12 August 2004. NSW
pH can be simultaneously measured by rapid and
Department of Primary Industries, Orange 2800, p.136
non-destructive mid-infrared spectroscopy. The Nguyen TT, LJ Janik and M Raupah. 1991. Diffuse
calibrations developed in this study have not been reflectance infrared fourier transform (DRIFT)
validated beyond this location. There is an spectroscopy in soil studies. Aust J Soil Res 29:
indication that the calibrations might not be widely 49-67.
98 IA F Djuuna et al.: Determination and Prediction Soil Properties using PLS and MIR

Olsen SR, CV Cole, FS Watanabe and LA Dean. 1954. Sparling GP. 1990. Soil biomass evaluation. In:
Estimation of available phosphorus in soils by the PACIFICLAND Workshop the Establishment of
extraction with sodium bicarbonate; Circ. 939; U.S. Soil Management Experiments on Sloping Lands.
Dep. of Agric. IBSRAM Technical Notes No. 4 International
Pankhurst CE, LJ Janik, CA Kirkby and BG Hawke. Board for Soil Research and Management,
1997. Application of GC-FAME and mid-infrared Bangkok, pp.163-184.
analysis of soil as measures of soil health. RIRDC Vance ED, PC Brookes, DS Jenkinson. 1987. An
Project CSO-10A, Final Report. extraction method for measuring soil microbial
Rayment GE and FR Higginson. 1992. Australian biomass C. Soil Biol Biochem 19: 703-707.
Laboratory Handbook of Soil and Water Chemical Viscarra Rossel RA and AB McBratney. 1998. Soil
Methods. (Inkata Press: Melbourne). Australia, 330 chemical analytical accuracy and costs: implication
pp. from precision agriculture. Aust J Exp Agr 38: 765-
Reeves JB and GW McCarty. 2001. Quantitative 775.
analysis of agricultural soil using near infrared Viscarra Rossel RA, DJJ Walvoort, AB McBratney, LJ
reflectance spectroscopy and a fiber-optic probe. Janik and JO Skjemstad. 2006. Visible, near
J Near Infrared Spectrosc 9 (1): 25-34. infrared, mid infrared or combined diffuse
Spaccini R, A Piccolo, G Haberhauer, M Stemmer and reflectance spectroscopy for simultaneous
MH Gezarbek. 2001. Decomposition of maize assessment of various soil properties. Geoderma
straw in three European soils as revealed by DRIFT 131: 59-75.
spectra of soil particle fractions. Geoderma 99: 245- White, RE. 1969. Measurement of soil pH. J Aust Inst
260. Agric Sci 35: 3-15.