EVALUATING REFLECTIVE SPECTROSCOPY FOR PREDICTING SOIL PROPERTIES IN GAJAPATI DISTRICT, ODISHA

Jilin Daxue Xuebao (Gongxueban)/Journal of Jilin University (Engineering and Technology Edition)
ISSN: 1671-5497
E-Publication: Online Open Access
Vol: 42 Issue: 10-2023
DOI: 10.5281/zenodo.8429553
Oct 2023 | 33
EVALUATING REFLECTIVE SPECTROSCOPY FOR PREDICTING SOIL
PROPERTIES IN GAJAPATI DISTRICT, ODISHA
RAHUL ADHIKARY*
Assistant Professor, Department of Soil Science, Centurion University of Technology and Management,
Odisha, 761211, India. *Corresponding Author Email: rahul.adhikary@cutm.ac.in,
ORCID ID: 0000-0003-0020-4161
SATYA NARAYAN SATAPATHY
Assistant Professor, Department of Entomology, Faculty of Agricultural Sciences, Siksha ‘O’ Anusandhan
Deemed to be University, Bhubaneswar, Odisha, India.
BISWAJIT LENKA
Assistant Professor, Department of Genetics and Plant Breeding, Faculty of Agricultural Sciences, Siksha
‘O’ Anusandhan Deemed to be University, Bhubaneswar, Odisha, India.
ASIT PRASAD DASH
Associate Professor, Department of Genetics and Plant Breeding, Faculty of Agricultural Sciences, Siksha
RAJESH KUMAR KAR
Assistant Professor, Department of Genetics and Plant Breeding, Faculty of Agricultural Sciences, Siksha
SUBHAM ACHARYA
Assistant Professor, Department of Agronomy, Faculty of Agricultural Sciences, Siksha ‘O’ Anusandhan
Deemed to be University, Bhubaneswar, Odisha, India.
Abstract
Visible near-infrared spectroscopy, renowned for its non-destructive nature, rapidity, cost-efficiency, and
minimal sample preparation requirements, holds promise as a substitute for in vitro techniques. This
ongoing study aims to evaluate the viability of reflective spectroscopy for predicting soil properties in ion
farming plains across Gajapati district Odisha. A meticulous collection of 110 soil samples from these
regions formed the basis, with their core attributes established using conventional in vitro methods.
Employing a land spectroscopic device, the soil samples underwent spectral analysis within the wavelength
band of 240 to 400 nm. Following spectrum recording, diverse pre-processing approaches were assessed,
paving the way for the application of PCA (Principal Component Analysis) and PLSR (Partial Least Squares
Regression) models to decipher pivotal soil properties. The superior model choice was subsequently
employed to formulate regressive functions, facilitating the prediction of targeted parameters through linear
regression. Findings spotlight the precision of both PCA and PLSR models in elucidating soil properties,
with the latter displaying heightened accuracy. Evaluated using the RPD (Ratio of Performance to
Deviation) metric, the most accurate estimations were achieved for minerals (RPD=9.34), pH (RPD=4.45),
and nitrogen (RPD>2), all classified within category A. In contrast, accuracy proved lower for variables like
clay, silt, gravel, phosphorus, potassium, calcium, magnesium, and gypsum, where RPD values ranged
between 0.01 and 0.28. These values collectively affirm the satisfactory precision of spectral regressive
functions in forecasting the targeted foundational properties. In summary, outcomes of this study
underscore the commendable precision of both PCA and PLSR models in determining crucial soil
parameters. Moreover, soil spectral data emerges as an effective indirect means to estimate the physical
and chemical attributes of soil. Compared to conventional laboratory methods, this technique emerges as

ISSN: 1671-5497
Vol: 42 Issue: 10-2023
DOI: 10.5281/zenodo.8429553
Oct 2023 | 34
a more cost-effective alternative, enhancing efficiency in terms of both time and cost while maintaining
heightened precision.
Keywords: Soil spectral pre-processing, soil properties, PCA and PLSR linear regressions
INTRODUCTION
In the current era, there is an undeniable global demand for spatial information about soil,
essential for environmental assessment, modeling, and effective soil management. The
complexity of soil, a dynamic and heterogeneous system governed by intricate processes,
underscores the need for advanced analytical techniques. While numerous analytical
methods have been proposed to establish relationships between physical and chemical
soil components, the mutual interactions among these components have often been
overlooked (Arnedo et al., 2017). Consequently, novel techniques are required to
efficiently collect soil spatial data across wide areas, capturing multiple properties in a
rapid and cost-effective manner (Stenberg et al., 2020). The urgency to streamline soil
analysis for time and cost efficiency has surged, driven by the demands of environmental
monitoring, soil quality assessment, and precision agriculture and afforestation (Viscarra
Rossel et al., 2006).
Reflective spectroscopy emerges as a viable solution, offering a potential alternative to
conventional analytical methods (Cohen et al., 2005). The utilization of visible-near
infrared reflective spectroscopy has gained traction over the past two decades. This
growth can be attributed to a multitude of reasons that align with spectral data's
capabilities. The advancement of multivariate statistics and data-mining techniques has
further expanded the application of this technology in pedological and agricultural
sciences. An inherent advantage of visible-near infrared reflective spectroscopy lies in its
non-destructive nature, ensuring minimal environmental impact (Guerrero et al., 2010).
As direct calibration of soil properties proves time-consuming, expensive, and subject to
temporal and spatial variability, the quest for indirect, cost-effective, and swift techniques
has become a pivotal research focus in pedology (Canbazoglu et al., 2013). In recent
years, telemetric methods, such as soil spectroscopic technology, have been harnessed
to enhance analytical efficiency, replacing cumbersome and costly laboratory procedures.
Spectroscopy involves the preparation of reflective curves within specific wavelength
ranges to quantify phenomena (Schneider et al., 2011).
The efficacy of near-infrared reflective spectroscopy rests on the sensitivity of soil organic
and mineral compounds to visible and infrared wave reflectance. This property has been
harnessed for agricultural and environmental soil studies. Over the past two decades, the
application of reflective spectroscopy in pedological sciences has grown, primarily
focusing on essential soil components, including organic matter, texture, mineralogy,
nutritional elements, structure, and microbial functions (Hassani et al., 2014).
Spectroscopy takes two forms: field and imaging. In both modes, electromagnetic energy
emitted from a light source interacts with a phenomenon, resulting in reflection,
absorption, and transmission. Spectroscopy enables quantitative analysis of these
processes. The visible (Vis) and near-infrared (NIR) spectral reflectance of soil spans
wavelengths ranging from 350 to 2500 nm (Iurian & Cosma, 2014). Furthermore, infrared

ISSN: 1671-5497
Vol: 42 Issue: 10-2023
DOI: 10.5281/zenodo.8429553
Oct 2023 | 35
spectroscopy, particularly in the near-infrared range, holds potential for identifying various
soil components. It facilitates the simultaneous estimation of more than 20 soil properties
through a single calibration, both in the field and in vitro. The spectral reflectance of soil
is influenced by attributes like humidity, texture, structure, organic matter content, and the
presence of various minerals (Soriano-Disla et al., 2014).
Critical soil properties, including organic carbon, mineral types, particle size distribution,
and carbonate minerals, significantly impact spectral reflectance. For instance, the
presence of organic carbon results in distinctive absorbent characteristics at specific
wavelengths due to complex chemical composition (Turhan et al., 2012). The shift in
spectral curve caused by increased organic carbon content in soil reduces reflectance,
especially in the visible and near-infrared ranges. Carbonate minerals, such as calcite
and dolomite, influence reflectance by affecting soil pH and CO2 levels (Hunt et al., 1997).
Soil particle size distribution also plays a role in spectral behaviour, with larger particles
absorbing lighter and exhibiting reduced reflectance. The influence of soil constituents,
such as iron oxides, clay, and organic carbon, on spectral curves and absorbent
characteristics underscores the significance of spectral behaviour in assessing soil
properties.
Given the overlapping nature of soil absorbent waves and their intricate adjustments,
statistical techniques and robust modeling approaches are essential for extracting
valuable spectral information. Approaches like Artificial Neural Network (ANN), Partial
Least Square Regression (PLSR), Principal Components Regression (PCR), and
Multiplicative Linear (and nonlinear) Regressions (MLR) have been employed to establish
connections between soil properties and spectral characteristics (Viscarra Rossel &
Behrens, 2010). Spectral curves have been effectively utilized to estimate various soil
physical and chemical properties, albeit with varying success (Lagacheri et al., 2008).
While most studies focus on the estimation of soil chemical properties through spectral
reflectance, a comprehensive understanding suggests that these properties are intricately
linked to the soil's solid structure and corresponding levels (e.g., clay, silt, organic matter,
electrical conductance, and pH), allowing for precise estimation using soil spectral data
(Warren, 2013a).
Despite limited research on the application of spectral data for estimating soil properties
in Odisha, India, this study seeks to evaluate the potential of visible-near infrared
spectroscopy in predicting select physical and chemical properties of soils in Gajapati
district, Odisha.
MATERIALS AND METHODS
Collection soil samples from study area
The study area encompassed the agricultural plains situated within Gajapati district in
south Odisha (18052’41.57” N, E 840 08’26.59” E ) from January 2021 to March 2023.
This region was situated at an average elevation of 250-1030 meters above sea level.
The annual temperature ranged from a maximum of 45°C to a minimum of 10°C, while

ISSN: 1671-5497
Vol: 42 Issue: 10-2023
DOI: 10.5281/zenodo.8429553
Oct 2023 | 36
the mean annual precipitation ranged between 1000-1200mm. The predominant soil
textures within this zone were sandy loam with acidic pH.
A total of 110 soil samples from figure 1 were collected randomly from the study zone,
spanning depths of 0-30 cm. These soil samples were air-dried and then sieved through
a 2mm screen. Following sample preparation, various physical and chemical attributes of
the soil were measured, including percentages of Available Nitrogen, Phosphorus,
Potassium organic matter, nitrogen, gypsum, sodium, calcium, magnesium. These
measurements were conducted using standard water and soil analysis techniques.
Fig 1: Soil samples collection area
Soil's spectral response and the preprocessing of spectral data related to soil
characteristics
The spectral reflectance of the soil samples under study underwent calibration through
the utilization of a spectroscopic device. This process involved 110 soil samples which
were air dried earlier and sieved through a 2mm sieve, after which their spectroscopic
curves calibrated across the visible-near infrared ranges (250-400nm) employing
established spectrometric techniques (Viscarra Rossel, 2008).
Calibration of the spectral curves was executed utilizing a handheld device with an
attached lens (8.4°), creating a visual field of 2cm in diameter. Illumination was provided
by a halogen lamp (150Watt) positioned at a 45-degree angle with respect to the vertical
axis. To enhance precision during reflectance level calibration, noise reduction, and

ISSN: 1671-5497
Vol: 42 Issue: 10-2023
DOI: 10.5281/zenodo.8429553
Oct 2023 | 37
device calibration during measurements, a single reading was obtained for all three soil
samples from a blank standard page (reflectance: 100%) positioned in a configuration
similar to that of the soil samples. These curves were captured using RS3 software
connected to the device and stored on a portable computer. Each soil sample was
represented by the mean value of its spectral curves, calculated using View-Spec
software. Subsequently, these spectral curves were saved as spectral libraries in text file
format for subsequent analyses. Each recorded spectral curve boasted a spectral
resolution of 1nm, contributing to a total of 2050 spectral reflectance points (2050
wavelengths) within the range of 400-2450nm (Dematte et al., 2017).
The primary objective of pre-processing the spectral data is to eliminate external
influences and optimize the utilization of spectral data in the approximation process.
Given the presence of significant spectral distortions (noise) within certain wavelength
ranges, specifically (350-399 and 2451-2500nm), these wavelengths were excluded from
modelling efforts (Terra et al., 2015). Similarly, to mitigate noise in soil spectra—such as
nonlinear behaviour of bands, spectral normalization, filtering, and spectral detection—
various pre-processing techniques were applied to the spectral data using software tools
such as Parles 3.1 and Unscrambler 10.3.
Soil calibrated properties and its modelling:
The calibrated soil properties were employed as the foundation for modelling purposes.
The initial model was constructed using the Principal Components Analysis (PCA)
technique, while the second model was developed through the application of the Partial
Least Squares Regression (PLSR) method. All procedural steps for both methods were
carried out using the MATLAB and SPSS software platforms, following the guidelines
provided by Farifteh et al. (2007) and Zheng (2008).
To assess the accuracy of the models, two key criteria, namely R^2 (coefficient of
determination) and RMSE (root mean square error), were employed. These metrics
served as indicators of the precision and effectiveness of the developed models.
Principal Components Analysis (PCA)
The utilization of the Principal Components Analysis (PCA) model becomes particularly
relevant when dealing with a high quantity of variables within a modeling framework. Due
to the substantial number of variables present, this approach becomes essential for
parameter reduction (Kodaira & Shibusawa, 2013). The Unscrambler X program was
used to perform PCA models, scores plots, and loadings plots of each soil order and of
all soil orders together. The data was created in a tables in excel and then uploaded into
the program to begin the statistical modeling. Data was then transposed into row sets and
column sets to create the scores and loadings plots. All zeroes, negatives (if any), or no
detections (ND), are to be left as blank cells as the Unscrambler X program does not
recognize negatives. Graphs and plots were designed with colour coordination,
leverages, and groupings.
Principle Component Analysis (PCA) is a multivariate statistical technique used to
analyze relation-ships between a large number of variables and smaller number of

ISSN: 1671-5497
Vol: 42 Issue: 10-2023
DOI: 10.5281/zenodo.8429553
Oct 2023 | 38
objects. In this study, the objects are the soil samples and the variables are the soil
characterization components (i.e. pH, EC, Fe, OC%). The interrelationship among the
two is explained through variables called principle components (PC) (Esbensen, 2010).
The soil variable data is obtained from the characterizations was transformed, auto
scaled, and evaluated using the PCA to geochemically distinguish soils. Relationships
among samples was demonstrated by data points in the score plots, and important
variables loaded on the samples were demonstrated by complementary PC subspace
distributions in the loading plot. Highly clustered samples in the score plot allows for the
down-selection of statistically distinguished samples to avoid redundancies in the future
experiments. Covariance measures the strength of joint variability between two or more
variables, indicating how much they change in relation to each other. To find the
covariance we can use the formula:
𝑐𝑜𝑣 (𝑥1, 𝑥2) = ∑
(𝑥1𝑖 − 𝑥1) + (𝑥2𝑖 − 𝑥2)
𝑛 − 1
𝑛
𝑖=1
The value of covariance can be positive, negative, or zeros.
Positive: As the x1 increases x2 also increases.
Negative: As the x1 increases x2 also decreases.
Zeros: No direct relation
Partial Least Square Root (PLSR) Model
In this study, we employed Partial Least Squares Regression (PLSR) to estimate soil
characteristics (anticipated variables) based on spectral indications (predictive variables).
This choice was made due to the interconnection between diffuse reflectance spectra and
soil attributes, as previously detailed. Since soil spectra exhibit an overlap of feeble
overtones and amalgamations of fundamental vibrational bands, it necessitated the
application of multivariate calibration techniques to accurately ascertain soil
characteristics. PLSR serves as a broader form of linear multiple regression, effectively
reducing an extensive array of collinear variables, such as reflectance values, into a
handful of uncorrelated latent variables or factors Wold et al. (2001) for comprehensive
PLSR descriptions.
To pinpoint the most appropriate procedure, we fitted numerous models. The three
datasets under consideration encompassed unprocessed data, as well as data processed
through Support Vector Machines (SVM) and Decision Trees (DT). Furthermore, in
pursuit of simpler and more efficient models, we established separate models for each
dataset, employing the ensuing subsets as independent variables in PLSR: Visible (VIS
- 350–700 nm), Near-Infrared (NIR - 701–1000 nm), Short-Wave Infrared (SWIR - 1001–
2500 nm), and the entire range (350–2500 nm).
We evaluated the resulting models based on criteria essential for constructing a robust
PLSR model: a minimal number of factors, minimal errors in leave-one-out cross-
validation (CV), and a high R2. Given the limited quantity of soil samples, we implemented

ISSN: 1671-5497
Vol: 42 Issue: 10-2023
DOI: 10.5281/zenodo.8429553
Oct 2023 | 39
the leave-one-out CV procedure to validate the regression models. For each model, we
calculated R2 and the root mean square error (RMSE) values during CV to assess
prediction accuracy. We also computed standard error (SE) values during CV. The
performance-to-deviation ratio (RPD), represented by the standard deviation (SD) to SE
ratio, was employed to gauge the usability of the calibrated models Williams et al. (2017).
The findings underscore that a significant proportion of the total variance of bands can be
accounted for by considering 20 factors. Moreover, the calibration phase resulted in an
R2-value of 0.89. The outcomes derived from the PLSR regression affirm the expectation
that this model has demonstrated commendable performance within the realm of soil
organic minerals estimation. This superiority can be attributed to the involvement of PCA-
dependent variables vis-à-vis PLSR in the band categories. These discoveries align with
the research conducted by various scholars who have delved into the estimation of soil
properties, either through manual spectrometers or super-spectral aerial images, yielding
promising predictions for soil property variables (e.g., Chen et al., 2008; He et al., 2007;
Zheng et al., 2008; Wang et al., 2010).Many statistical techniques hinge on the
assumption of a normal distribution for data. Hence, the normality status of each soil
property was evaluated using the Kolmogorov-Smirnov test, conducted at a significance
level of 5%, following the calculation of descriptive statistics. For variables exhibiting non-
normal distributions, suitable transformations were applied to achieve normalization.
Subsequently, the degree of correlation between spectral reflectance values at various
wavelengths and the physical and chemical properties of the soil was examined using
Pearson's correlation coefficient.
RESULTS AND DISCUSSION
Spectral data processing
Figure 2 presents the average raw spectral reflectance curves for the soils. This graphical
representation highlights a distinct peak in reflectance within the wavelength range of
500-700nm. Additionally, four notable absorption features are evident at wavelengths
990nm, 1414nm, 1915nm, and 2212nm. These absorbent characteristics are associated
with factors such as free water presence, hygroscopic properties (1414nm), (O-H factor)
or hydroxyl groups in clay mineral lattice (1915nm), and their interaction with iron,
aluminum, and magnesium (2212nm) within the clay mineral structure (Clark et al., 1990).
Several prevailing soil attributes influence the spectral reflectance levels, including soil
color, moisture content, organic carbon concentration, particle size distribution, and the
presence of iron and aluminum oxides. Generally, calcium carbonate tends to elevate
reflectance, while organic carbon and soil moisture have a dampening effect (Stenberg
et al., 2000).

ISSN: 1671-5497
Vol: 42 Issue: 10-2023
DOI: 10.5281/zenodo.8429553
Oct 2023 | 40
Fig 2: The mean of spectral reflectance of studied soils, position of absorption
characteristics at wavelengths 990, 1450, 1946, and 2200 nm have been specified on this
figure including the dimensions, configuration, and distribution of particles and gaps within
it. These factors play a pivotal role in determining the distance light travels through a soil
sample, consequently exerting an impact on the observed spectral signatures.
Table1: Statistical description of soil properties
Soil chemical properties Max Min Mean Std deviation
AvlN kg/ha 280 110 195 18.9
Avl P (kg/ha) 22.5 18.6 9.19 15.9
EC(dS/m) 0.89 0.18 0.21 8.5
pH 6.11 7.3 7.11 0.3
O/C 0.76 0.30 0.25 0.2
Avl K (kg/ha) 273 184 240 237
Ca(mg/l) 1072 30 221 207
Mg(mg/l) 214 2.7 48 31.4
Gypsum(%) 33.6 0.02 4.6 8.7
As shown in Table 1, the soil properties in the studied samples exhibit a wide range of
variability. The levels of organic matter and total salinity were measured at 0.21 deci-
Siemens/m and 0.76%, respectively. Soil samples contained varying amounts of nitrogen
(195 kg/ha phosphorus (19.9 kg/ha), and potassium (240kg/ha). The concentrations of
calcium, magnesium, sodium, and gypsum were determined to be 221 ppm, 48 ppm, and
4.66%, respectively. Owing to the presence of carbonates, the pH of the studied soils
ranged from slightly acidic to neutral (6.1-7.3). The soils in Zone-2 exhibited a diverse
range of textural classes.

ISSN: 1671-5497
Vol: 42 Issue: 10-2023
DOI: 10.5281/zenodo.8429553
Oct 2023 | 41
Partial Least Squares Regression (PLSR) Model for Soil Chemical Properties
In this study, we employed Partial Least Squares Regression (PLSR) modeling to
investigate the relationship between various soil chemical properties and the predictor
variables, aiming to establish predictive models for the given soil attributes. The dataset
includes a range of soil properties, such as Available Nitrogen (Avl N), Available
Phosphorus (Avl P), Electrical Conductivity (EC), pH, Organic Carbon (O/C), Available
Potassium (Avl K), Calcium (Ca), Magnesium (Mg), and Gypsum content.
Model Performance
The PLSR model was assessed in terms of its predictive performance using several key
metrics, including R-squared (R²), Root Mean Square Error (RMSE), and Mean Absolute
Error (MAE). These metrics provide insight into how well the model captures the variation
in the observed data.
Variable Importance: The PLSR analysis allowed us to assess the importance of each
predictor variable in explaining the variation in soil chemical properties. For instance, the
model may have identified that pH and organic carbon are the most influential factors in
predicting certain soil properties, while others may be less important.
Among the parameters denoted as (r) in the Pearson correlation coefficient table, the
physical and chemical attributes of soil are associated with the values of spectral
reflectance within a range of wavelengths spanning from 400nm to 2450nm. In relation to
this observation, a substantial correlation is observed between calibrated physical and
chemical characteristics and the values of spectral reflectance within the visible to near-
infrared range. The findings demonstrate that the distribution of soil particle sizes shows
both positive and negative correlations across various wavelengths. The outcomes,
depicted in Table 2, present the root mean square error (RMSE) values and coefficient of
determination (R2) for soil attributes such as clay, silt, sand, and nutrient elements like
nitrogen, phosphorus, potassium, calcium, and magnesium, corresponding to
wavelengths 990nm, 1414nm, 1915nm, and 2212nm, respectively. The R2 values for
estimating the physical and chemical properties of the studied soil range from 0.76 to
0.95%, with the highest R2 value attributed to gypsum and the lowest to phosphorus
content. Importantly, the presence of such strong correlations between fundamental soil
attributes and rates of spectral reflectance has also been noted by other researchers
(Somers et al., 2010; Bilgli et al., 200). Based on the RPD statistic, the most accurate
approximations for the suggested regression functions are evident for organic matter
(RPD=9.34), pH (RPD=4.45), and nitrogen (RPD>2) across all three series-A categories.
The provided approximations, including clay, silt, gravel, phosphorus, potassium,
calcium, magnesium, and gypsum (RPD range of 0.01-0.28), indicate a reasonable
precision of spectral regression functions in predicting the fundamental properties under
study.
The validation results for the proposed functions are presented in Figure (3) as dispersion
diagrams (1:1), which demonstrate the estimation of various soil physical and chemical
properties against their calibrated values. The R2 values shown in both the figure and
table indicate the statistical performance of the tested data. Among the foundational

ISSN: 1671-5497
Vol: 42 Issue: 10-2023
DOI: 10.5281/zenodo.8429553
Oct 2023 | 42
properties, attributes such as cation exchange capacity, sand, organic carbon, and silt
have an average R2 value of approximately 75. Notably, clay demonstrates a particularly
favourable approximation within the suggested regression functions for estimating these
properties (McGann et al., 2023)
Fig 3: Diagram of precision for results of validation to calibrated and estimated values
The figure 3 highlights the soil organic carbon (0.89), available nitrogen (0.92), available
phosphorus (0.92), potassium (0.78), calcium (0.91), magnesium (0.93), and gypsum
percentage (0.95) R2-statistic values that stand out for the group of fundamental features.
These figures represent how well the suggested regression functions approximate these
specific traits. A recent study by Dematte et al. (2014) found significant agreement
between estimates made using spectral data and calibrated values of important soil
properties. For characteristics such as clay (0.83, 4.03), silt (0.32, 5.06), sand (0.7, 5.67),

ISSN: 1671-5497
Vol: 42 Issue: 10-2023
DOI: 10.5281/zenodo.8429553
Oct 2023 | 43
cation exchange capacity (0.64, 1.94 centi-mol.kg), organic matter (0.73, 0.31%), and
calcium carbonate (0.64, 0.70%), the researchers reported interpretation coefficients (R2)
and RMSEA. It's described same by Bilgli et al. (2010) also voiced support for leveraging
spectral data in estimating specific soil physical and chemical attributes. Another study
conducted by Babaeian et al. (2015) ventured into predicting basic soil properties, along
with soil hydraulic attributes, utilizing spectral data. Collectively, these researchers
underscored the potential of soil spectral properties as an effective indirect means to
estimate the examined physical and chemical attributes with relative accuracy.
Principal Components Analysis (PCA) models for result analysis
The Principal Components Analysis (PCA) model is particularly useful when dealing with
a high number of variables in a modeling context. Given the abundance of variables, this
technique becomes indispensable for the purpose of reducing the number of parameters,
as demonstrated by Kodaira and Shibusawa in 2013. Upon conducting a PCA analysis
on the independent variables or bands, it was observed that out of the initial set of
variables, 20 principal components were retained. Remarkably, these five components
managed to encapsulate a significant percentage of the variance within the independent
variables. In other words, a substantial number of bands were effectively condensed into
20 components, and these 20 components accounted for a considerable portion of the
variance across the bands.
The results obtained through the application of the PCA technique align with those
obtained using previous methods. However, it is important to note that the predicted
values did not yield satisfactory regression coefficients across in south Odisha region.
CONCLUSION
The outcomes of this study underscore the general efficacy of both PCA and PLSR
models in accurately determining key soil parameters within the provinces of Gajapati,
Odisha region. Notably, the PLSR model demonstrates heightened precision compared
to the PCA model. This study evaluated the potential of using soil spectral data within the
visible-near infrared range (400-2400nm) to estimate specific soil physical and chemical
properties. The preprocessing curve was delineated for spectral data after its recording
and calculation of means. The findings underscored the superiority of the method for
spectral data preprocessing. Considering the strong correlation and significance between
calibrated soil properties and their spectral reflectance values, regression functions were
formulated to estimate these properties. Various statistical measures were employed to
assess the accuracy of these functions, including Root Mean Square Error (RMSE),
interpretation coefficient (R2), Evaluation of these statistics revealed RMSE values of 27
and 56.7 for the total range of basic properties studied. Conversely, lower approximations
were found (RPD=0.01-0.028) for phosphorus, potassium, calcium, magnesium, and
gypsum, indicating a satisfactory precision of spectral regression functions in predicting
the studied basic properties. In summation, the findings of this study highlight the potential
of using soil spectral data as an indirect method for estimating soil's physical and chemical
properties. However, given the empirical nature of spectral regression functions and the
spatial-temporal variability of soil properties in diverse zones (like: problematic soils at

ISSN: 1671-5497
Vol: 42 Issue: 10-2023
DOI: 10.5281/zenodo.8429553
Oct 2023 | 44
separate zone shows different spectral behaviour), it is advisable to develop and assess
these functions for a wider array of zones and soil types.
Acknowledgement
I thank the following individuals for their expertise and assistance throughout all aspects of our study and
for their help in writing the manuscript.
References
1) Arnedo, M.A., Rubiano, J.G., Alonso, H., Tejera, A., Gonz alez, A., Gonz alez, J., Gil, J.M.,
Rodríguez, R., Martel, P., Bolivar, J.P., (2017). Mapping natural radioactivity of soils in the
eastern Canary Islands. J. Environ. Radio act, 166, 242– 258.
2) Babaeian, E., M. Homaee, C. Montzka, H. Vereecken, and A. A. Norouzi. (2015). Towards
retrieving soil hydraulic properties by hyperspectral remote sensing. Vadoze Zone J, 14(3),
doi: 10.2136/ vzj2014.07.0080.
3) Bilgili, A. V., Van Es, H. M., Akbas, F., Durak, A., & Hively, W. D. (2010). Visible-near infrared
reflectance spectroscopy for assessment of soil properties in a semi-arid area of Turkey.
Journal of Arid Environments, 74(2), 229-238.
4) Canbazoglu, C., Turhan, S., Bakkal, S., Ugur, F.A., Goren, E., (2013). Analysis of gamma
emitting radionuclides (terrestrial and anthropogenic) in soil samples from Kilis province in
south Anatolia, Turkey. Ann. Nucl. Energy, 62, 153–157.
5) Chappell, Mark A., Jennifer M. Seiter, and Brandon J. Lafferty. "A Pedo-Informatic Approach
for Universal Predictions of Complex Soil Environmental Processes." Soil Science Society of
America. N.P., 17 Nov. 2015. Web. 6 June 2017.
6) Chen, F., D.E. Kissel, L.T. West, W. Adkins, D. Rickman and J.C. Luvall. (2008). Mapping
soil organic carbon concentration for multiple fields with image similarity analysis. Soil
Science Society of America Journal, 72(1), 186-193.
7) Cohen MJ, Prenger JP, DeBusk WF. (2005). Visible-Near infrared reflectance spectroscopy
for rapid, nondestructive assessment of wetland soil quality. Journal of Environmental
Quality. 34, 1422–1434. [PubMed: 15998865].
8) Demattê, J. A. M., Bellinaso, H., Romero, D. J., & Fongaro, C. T. (2014). Morphological
Interpretation of Reflectance Spectrum (MIRS) using libraries looking Scientia Agricola
towards soil classification, 71(6), 509-520.
9) Dematte, J. A. M., Horak-Terra, I., Beirigo, R. M., Terra, F. D. S., Marques, K. P. P., Fongaro,
C. T., Vidal-Torrado, P. (2017). Genesis and properties of wetland soils by VIS-NIR-SWIR as
a technique for environmental monitoring. J Environ Manage, 197, 50-62. doi:
10.1016/j.jenvman.2017.03.014
10) Esbensen, K. H. (2010). "Multivariate Data Analysis - In Practice," CAMO Software
11) Farifteh, J., F.D. Van der Meer, C. Atzberger and J.M. Carranza. (2007). Quantitative analysis
of salt affected soil reflectance spectra: a comparison of two adaptive methods (PLSR and
ANN). Remote Sensing of Environment, 110, 59-78.
12) Guerrero, C., R. A. Viscarra Rossel and A. M. Mouazen.( 2010). Diffuse reflectance
spectroscopy in soil science and land resource assessment. Geoderma. 158, 1-2.

ISSN: 1671-5497
Vol: 42 Issue: 10-2023
DOI: 10.5281/zenodo.8429553
Oct 2023 | 45
13) Hassani, A., Bahrami, H., Noroozi, A., & Oustan, S. (2014). Visible-near infrared reflectance
spectroscopy for assessment of soil properties in gypseous and calcareous soils. Journal of
Watershed Engineering and Management. 6(2), 125- 138.
14) He, Y., M. Huang, A. Garcia, A. Hernandez and H. Song. (2007). Prediction of soil
macronutrients content using near-infrared spectroscopy. Computers and Electronics in
Agriculture, 58, 144-153.
15) Iurian, Andra-Rada, Cosma, Constantin, (2014). A practical experimental approach for the
determination of gamma-emitting radionuclides in environmental samples. Nucl. is unknown
and the path forwards. Plant and Soil, 763, 132- 136.
16) Kodaira, M., Shibusawa, S. (2013). Using a mobile real-time soil visible-near infrared sensor
for high resolution soil property mapping. Geoderma. 199, 64-79.
17) Lagacherie, P., Baret, F., Feret, J.B., Netto, J.M and Robbez-Masson, J.M. (2008).
Estimation of soil clay and calcium carbonate using laboratory, field and airborne
hyperspectral measurements. Remote Sensing of Environment, 112, 825-835.
18) McGann, D., William,D., Barshop, Jesse D., Canterbury, Chuwei Lin., Wassim, Gabri-el.,
Jingjing, Huang., Bergen, David., Zabrouskov, Vlad., Rafael D. Melani, Wilhelm, Mathias.,
Graeme C. McAlister, and Devin K. Schweppe J. (2023). Real-Time Spectral Library
Matching for Sample Multiplexed Quantitative Proteomics. Proteome Res. 22, 9, 2836–2846.
19) Özdemir, T., 2012. Distribution of terrestrial and anthropogenic radionuclides in Turkish
surface soil samples. Geoderma, 187–188, 117–124.
20) Schneider, W. E. and R. Young. (1998). Spectroraidometery Methods: a guide to photometry
and visible spectroradiometery. Application note (A14), Optronic Labratories, INC. PP. 47.
21) Soriano-Disla, J.M., Janik, L.J., Viscarra Rossel, R.A., Macdonald, L.M., McLaughlin,
M.J.(2014). The Performance of Visible, Near-, and Mid-Infrared Reflectance Spectroscopy
for Prediction of Soil Physical, Chemical, and Biological Properties, Applied Spectroscopy
Reviews, 49 (2), 139-186.
22) Stenberg, B., Rossel, R. A. V., Mouazen, A. M., & Wetterlind, J. (2010). Chapter five-visible
and near infrared spectroscopy in soil science. Advances in Agronomy, 107, 163-215.
23) Summers, D., Lewis, M., Ostendorf, B., and Chittleborough, D. (2011). Visible near-infrared
reflectance spectroscopy as a predictive indicator of soil properties. Ecological Indicators,
11(1), 123-131.
24) Viscarra Rossel RA, Walvoort DJJ, McBratney AB, Janik LJ, Skjemstad JO. (2006). Visible,
near infrared, mid infrared or combined diffuse reflectance spectroscopy for simultaneous
assessment of various soil properties. Geoderma. 131, 59–75.
25) Viscarra Rossel, R., McGlynn, R., McBratney, A. (2006). Determining the composition of
mineralorganic mixes using UV–vis–NIR diffuse reflectance spectroscopy. Geoderma, 137,
70-82.
26) Williams, P.; Dardenne, P.; Flinn, P. (2017). Tutorial: Items to Be Included in a Report on a
near Infrared Spectroscopy Project. J. Near Infrared Spectrosc. JNIRS, 25, 85–90.
[CrossRef]
27) Willmott, C.J. (1981). On the validation of models. Physical Geography, 2, 184–190

ISSN: 1671-5497
Vol: 42 Issue: 10-2023
DOI: 10.5281/zenodo.8429553
Oct 2023 | 46
28) Wold, S.; Sjöström, M.; Eriksson, L. (2001). PLS-Regression: A Basic Tool of Chemometrics.
Chemom. Intell. Lab. Syst, 58, 109–130. [CrossRef]
29) Xuemei, Liu, & Jianshe, Liu. (2013). Measurement of soil properties using visible and short
wave-near infrared spectroscopy and multivariate calibration. Measurement, 46(10), 3808-
3814. doi: 10.1016/j.measurement.2013.07.007
30) Zheng, B. (2008). Using satellite hyperspectral imagery to map soil organic matter, total
nitrogen and total phosphorus. MSc. Thesis, Indiana University, 81 pages.
31) 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(7), 1923-1930. doi:
10.1016/j.soilbio.2008.04.003

EVALUATING REFLECTIVE SPECTROSCOPY FOR PREDICTING SOIL PROPERTIES IN GAJAPATI DISTRICT, ODISHA

More Related Content

Similar to EVALUATING REFLECTIVE SPECTROSCOPY FOR PREDICTING SOIL PROPERTIES IN GAJAPATI DISTRICT, ODISHA

More from indexPub

Recently uploaded

EVALUATING REFLECTIVE SPECTROSCOPY FOR PREDICTING SOIL PROPERTIES IN GAJAPATI DISTRICT, ODISHA