Precision Agriculture (2024) 25:2758–2775

https://doi.org/10.1007/s11119-024-10140-1

Hyperspectral sensing and mapping of soil carbon content

for amending within‑field heterogeneity of soil fertility
and enhancing soil carbon sequestration

Yoshio Inoue1 · Kunihiko Yoshino1 · Fumiki Hosoi1 · Akira Iwasaki1 ·

Takashi Hirayama2 · Takashi Saito2

Accepted: 3 April 2024 / Published online: 2 May 2024

© The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature 2024

Abstract
Soil fertility is one of the most critical bases for high productivity and sustainability in
crop production. Within-field heterogeneity is often problematic in both crop management
practices and crop productivity. Besides, appropriate soil management practices leads to
the effective carbon sequestration. Since the soil carbon content (SCC) is the most simple
and effective indicator of soil fertility, accurate and high-resolution mapping of SCC is
an essential basis for addressing these issues. Here, we developed a tractor-based hyper-
spectral sensing system for speedy and accurate mapping of SCC. A new hybrid spectral
algorithm linking normalized difference spectral index (h-NDSI) and machine learning
proved superior. Appropriate algorithms were implemented to generate diagnostic map and
prescription map from SCC map for the variable-rate application of pellet manure. The
field performance of the sensing/mapping system was tested in the farmers’ fields in the
Fukushima region of Japan where the within-field heterogeneity of soil fertility was disas-
trous due to the decontamination after the nuclear power-plant disaster. The structure and
functioning of the system proved promising. Moreover, the spatial simulation by linking
the SCC data and a dynamic simulation model clearly showed the significant impact of
variable-rate application of pellet manure on the chronosequential change of SCC, within-
field heterogeneity, and carbon stock. The systematic linkage of the sensing/mapping sys-
tem with the variable-rate spreader and dynamic simulation model would be effective for
improving soil fertility and soil carbon stock. Applicability of the system will be extended
through an extensive validation of the predictive models.

Keywords Hybrid algorithm · Machine learning · NDSI · Remote sensing · Soil carbon
content · Spatial simulation

Abbreviations
SCC Soil carbon content (%)

* Yoshio Inoue
[email protected]
1
The University of Tokyo, Tokyo, Japan
2
Fukushima Agricultural Technology Center, Minamisoma, Japan

Vol:.(1234567890)
Precision Agriculture (2024) 25:2758–2775 2759

SNC Soil nitrogen content (%)

NDVI Normalized difference vegetation index
NDSI Normalized difference spectral index
h-NDSI Hyperspectral NDSI
VNIR: Visible near infrared spectral region
SWIR Shortwave spectral region
FOV Field of view
GNSS Global navigation satellite system
RTK Real-time kinematic
KMZ ISO format of XML (Extensible Markup Language)
CV Coefficient of variation (%)
RMSE Root mean square error
PLSR Partial least-squares regression
iPLSR Interval partial least-squares regression
ML Machine learning
ANN Artificial neural network
SVM Support vector machine
XGB Extreme gradient boosting method

Introduction

Soil fertility is one of the most critical bases for high productivity and sustainability of
agriculture. Soil fertility is defined in chemical, physical, and biological aspects, but in
general, the chemical role (a potential source of nutrition) is most important. For example,
in rice, a large part (60–70%) of the nutrition absorbed by the plant is supplied by the soil
(Koyama et al., 1973).
The role of soil fertility is increasing under the global needs for less chemical-fertilizer
use and organic farming, especially in the recent frameworks such as European Green
Deal (EC, 2020), Agriculture Innovation Agenda (USDA, 2020), and Green Food System
Strategy (MAFF Japan, 2021). In general, the biological functioning (microbial activity)
and physical structure (soil aggregate) are also associated positively with the level of soil
organic matter (Six et al., 2004). Therefore, application of organic materials such as crop
residue and manure is one of the most traditional and effective practices. However, it is
well recognized that improving soil properties does need long and continuous effort (Poul-
ton et al., 2018).
The within-field heterogeneity of soil fertility is often significant due to various causes
such as consolidation of multiple fields, inconsistent soil management, and so forth (Hei-
jting et al., 2011; Schuster et al., 2023). One of the most disastrous heterogeneous condi-
tions was caused by radioactive decontamination; i.e., replacing the polluted surface soils
with non-polluted soils after the nuclear power plant accident in Fukushima (Tanigaki
et al., 2022). Hence, not only improving the fertility level but also amending the spatial
heterogeneity of soil fertility is required in many cases (e.g., Rhymes et al., 2023). The key-
practice for improving the heterogeneity is the site-specific application of organic matter.
However, the spatial information on soil fertility is prerequisite for the spatial management
such as variable-rate application of organic matter.
On the other hand, the soil layer, a thin skin of the terrestrial surface of a few
cm ~ m depths, is a significant carbon pool. The carbon in the soil layer is three times
2760 Precision Agriculture (2024) 25:2758–2775

the atmospheric carbon and two times the global biomass carbon (Batjes, 1996). Since
farmland soils have a great potential for carbon sequestration, effective soil management
strategies and practices are strongly required (Demenois et al., 2020). The spatial distri-
bution of soil carbon content (SCC) is fundamental for understanding the carbon stock
dynamics and effective carbon sequestration. Since the humus content is the major indi-
cator of soil fertility and is known to be proportional to SCC (e.g., 1.724 × SCC; Soil
Survey Staff, 2011), the SCC is the most simple and effective indicator of soil fertility.
Hence, accurate geospatial assessment of SCC is critical for improving the soil fertil-
ity, amending the spatial heterogeneity, and designing the appropriate soil management
scenarios for effective carbon sequestration. However, to date, data acquisition methods
largely rely on conventional soil sampling and chemical analysis. Spatial assessment
of soil fertility is still challenging due to laborious soil sampling and costly chemical
analysis (AHDB, 2018).
Remote sensing is vital for efficient acquisition of crop and soil information since
it allows non-contact and/or wide-area data sensing (e.g., Inoue, 2020; Moran et al.,
1997). Many studies have suggested that spectral data are related to soil properties such
as soil texture and soil organic carbon (Bellon-Maurel & McBratney, 2011; Inoue et al.,
2020; Ladoni et al., 2010; Stenberg et al., 2010; Stevens et al., 2008). Remote sensing
can play a critical role in rapid and spatial assessment of SCC (Angelopoulou et al.,
2019; Ben-Dor et al., 1997; Gholizadeh et al., 2018; Inoue, 2020; Moran et al., 1997).
Satellite- and drone- remote sensing has proved useful for the wide-area assessment of
soil fertility under bare surface conditions. However, the predictive ability and appli-
cability are highly dependent on the dataset (soil type and location), sensor, or surface
conditions. The accuracy and applicability are often inadequate due to their inherent
constraints (Angelopoulou et al., 2019; Bellon-Maurel & McBratney, 2011; Castaldi
et al., 2023; Ladoni et al., 2010). Two major constraints are the limited spectral specifi-
cation of remote sensors and the non-bare soil surface conditions. Limited availability
of ground-based spatial SCC data may be another constraint for calibration and valida-
tion of satellite remote sensing.
Therefore, on-the-go hyperspectral sensing coupled with cultivation practice would
be an alternative approach to obtain accurate and spatially distributed SCC data. In many
regions, agricultural fields are cultivated using rotary cultivators at least once a year before
seeding or planting. Several case studies clearly showed the great potential of on-the-go
sensing of the soil properties (Adamchuk et al., 2004; Knadel et al., 2015; Tabatabai et al.,
2019; Vogel et al., 2022). Many types of electric and spectral sensors, system structures,
and data processing methods have been attempted for estimating different soil properties
(e.g., particle size, clay, organic carbon, pH, etc.). These attempts revealed the advantage
and caveats of each approach suggesting the necessity for further innovation in sensor
design, data processing, analytical algorithms, and their systematic integrations.
Thus, in this study, we developed a tractor-based hyperspectral sensing system focus-
ing on the speedy and accurate mapping of soil carbon content. A new hybrid spectral
algorithm was proposed based on the hyperspectral measurement of soils. Also, algorithms
to generate diagnostic map and prescription map from SCC maps were also developed for
variable-rate application of pellet manure. The field performance of the sensing/mapping
system was tested in the farmers’ farmland in the Fukushima region of Japan where the
within-field heterogeneity of soil fertility was disastrous due to the decontamination after
the nuclear power-plant disaster. In addition, the possible impact of different manure appli-
cation scenarios on the chronosequential change of SCC, within-field heterogeneity, and
carbon stock was investigated using a dynamic simulation model.
Precision Agriculture (2024) 25:2758–2775 2761

Materials and methods

Instrumentation for on‑the‑go sensing of SCC

The prototype hyperspectral sensing system was equipped with the two sets of coupled
spectroradiometers in visible to near infrared (VNIR) and shortwave (SWIR) spec-
tral regions (Hamamatsu Photonics, Hamamatsu, Japan). The wavelength region was
350–2500 nm, and the spectral resolution was 1 nm and 6 ~ 8 nm in VNIR and SWIR
regions, respectively. Each couple of identical radiometers with 5 m long optical fib-
ers enabled the measurement of upward and downward spectral irradiance. The spectral
reflectance of the soil surface was derived as the ratio of downward and upward sensor
readings using the calibration factors. The field of view (FOV) was 25°, so the diameter
on the soil surface was approximately 0.8 m at a height of 2 m (Inoue & Yoshino, 2023).
The sensing system was mounted on a tractor with a rotary cultivator using a flexible
outrigger frame (Fig. 1). During on-the-go measurements, the hyperspectra of soil sur-
faces were recorded continuously at 5 Hz. An antenna of GNSS (global navigation satel-
lite system; Mitsubishi Electric, Tokyo, Japan) was attached next to the fiber probe. The
global position data were recorded at 10 Hz. The calculation of positions was supported
by the network RTK (Real-Time Kinematic) system via Wi-Fi. The overall accuracy of
the global position data was around a few centimeters. The depth of rotary cultivation
was approximately 15 cm which was the normal practice in this region.

Algorithms for spectral assessment of SCC

SCC data

Soil samples were taken from 0 to 5 cm depth in a range of agricultural fields. The total
soil carbon content (SCC) and soil nitrogen content (SNC) were determined using a C/N
analyzer (MT-700MK-2, Yanako, Kyoto, Japan). The average of moisture content was
15.0% with standard deviation of 2.7%.

1) Hyperspectral radiometers (VIS~SWIR)

4) Optical fiber probes 2) PCs for recording/controling/processing/networking
5) GNSS antenna 3) Independent power (batteries)

FOV

Fig. 1  Prototype hyperspectral sensing system mounted on a tractor with a rotary cultivator. Spectral range:
350 ~ 2500 nm (resolution: 1 ~ 7 nm), FOV: 25 degrees, GNSS: network-RTK (10 Hz)
2762 Precision Agriculture (2024) 25:2758–2775

Hyperspectral measurements

Spectral measurements were made in a laboratory environment using a spectroradiometer

(FieldSpec-FR, ASD, Boulder, USA). The spectral range was 350 ~ 2500 nm with resolutions
of 3 nm (350 ~ 1000 nm) and 10 nm (1000 ~ 2500 nm). The absolute reflectance was deter-
mined using a white reference panel (Spectralon, Labsphere, North Sutton, USA). A total of
700 wavebands (401 ~ 2498 nm; 3-nm interval) were used for the analyses (Fig. 2a).

Analytical methods and modeling approaches

The relationship between spectral reflectance and SCC values was analyzed using the spectral
index approach, multivariable regression, and machine learning methods (Fig. 2b). The gener-
alized formulation of normalized difference index (NDSI) has been applied to hyperspectra to
assess the vegetation and land-surface parameters (Inoue et al., 2008, 2012, 2016, 2020). The
most conventional spectral index NDVI (normalized difference vegetation index) is a specific
one of NDSI using red and near-infrared wavebands (Tucker, 1979). Here, the NDSI approach
is designated as h-NDSI (hyperspectral NDSI) to more specifically represent the hyperspectral
feature of the method.
h−NDSI (Ri, Rj) = (Rj - Ri) ∕ (Ri + Rj) (1)

a) b)
0.6 9
8
0.5
7
Reflectance

0.4 6
CV (%)

5
0.3
CV 4
0.2 3
2
0.1
n = 120 [ Hyperspectral data ] 1
0.0 0
400 600 800 1000 1200 1400 1600 1800 2000 2200 2400
h-NDSI map
Wavelength (nm)

r2val = 0.883 c) Hyperspectral Normalized

Predeicted SCC (%)

RMSEval = 0.139
n = 120 Difference Index (h-NDSI)
- Waveband selection
- Error reduction

Machine Learning (SVM)

- Non-linear fitting
- Accuracy enhancement
Measured SCC (%)

Fig. 2  Hybrid spectral algorithm and predictive ability of the optimal model for SCC prediction
Precision Agriculture (2024) 25:2758–2775 2763

where Ri and Rj are reflectance at i and j nm over the entire hyperspectral region. Here,
the optimal indices were explored using h-NDSI maps of predictive ability such as the
coefficient of determination r2 (Fig. 2b).
Multivariable linear regression methods such as partial least-squares regression (PLSR)
have been widely used in hyperspectral modeling to increase accuracy by making the most
of a large number of hyperspectral wavebands (Stenberg et al., 2010). The interval PLSR
(iPLSR: PLSR with wavelength selection) was also applied. Machine learning methods
such as artificial neural network (ANN), support vector machine (SVM) and extreme gra-
dient boosting method (XGB) were used since they are useful for the dataset of unknown
structures. These non-linear fitting methods can provide high accuracy for a given dataset,
although the applicability of such models is often unstable (Ali et al., 2015).
The coefficient of determination (r2) and root mean square error (RMSE) were used as
statistical indicators to compare the predictive ability of the spectral algorithms and mod-
els. The stability of models was inferred using the ratio of RMSE for validation and cali-
bration ­(RMSEval/RMSEcal).

Creation and visualization of SCC, diagnostic, and prescription maps

A program was developed for processing the sensing and position data and for converting
the reflectance data to SCC values using the predictive models. The SCC map was cre-
ated as mesh data of arbitrary size (2 ~ 3 m) from a large number of point data based on
distance-weighted interpolation. The format of the mesh files was KMZ (ISO-XML) which
can be used easily by GoogleEarth and many GIS applications. The program also enables
the conversion of a SCC map to a diagnostic map (fertility deficiency class) and a prescrip-
tion map (the amount of pellet manure to be applied) based on a given weighting logic
of manure application. Prescription maps are transferred to the controlling system of the
spreader for variable-rate application of pellet manure via wireless network or cloud server.
Generally, pellet manure is organic granular fertilizers made of chemical materials and
livestock manure. The typical percentage of water (­H2O), carbon (C), nitrogen (N), phos-
phorous ­(P2O5), and potassium ­(K2O) is 10%, 30%, 3.5%, 3.5%, and 1.5%, respectively,
although various combinations of chemical compositions are available. The diameter and
length of individual pellets are usually 6–7 mm and 20–30 mm, respectively, which allows
easy and precise spreading using spreaders (Hara, 2001).

Test of the tractor‑based sensing system in farmer’s fields

The performance of the prototype sensing/mapping system was tested in the coastal region
of Fukushima Prefecture where the farmland was heavily contaminated by the Fukushima
Daiichi Nuclear Power Plant disaster in 2011. The farmland in the area was polluted by
radionuclides such as radioactive cesium. Subsequently, the polluted fields were decon-
taminated by replacing the surface soil layer (5–10 cm) with new soils. More than 40,000
fields have been decontaminated to date. Their soil fertility is low, and within-field spa-
tial heterogeneity is disastrous (Inoue et al., 2020; Tanigaki et al., 2022). In some of such
fields, soil samples (n = 60) were taken and analyzed to validate the mesh-SCC values pre-
dicted by the sensing system. The original soil was categorized as Skeletal Haplic Gray
Lowland Soil, while the dressed soil was sandy soil. SCC values by the sensing/mapping
system were compared with those by chemical analysis. The reproducibility of the SCC
2764 Precision Agriculture (2024) 25:2758–2775

data by the sensing/mapping system was tested under different running and environmental
conditions using two types of tractors and rotary cultivators.

Simulation of SCC dynamics under different manure application scenarios

In general, amending or enhancing soil fertility takes a long time because of the slow bio-
chemical and microbial transformation within the soil layer (Poulton et al., 2018). Hence,
we used a simulation model to assess and predict the impact of different manure appli-
cation practices on the within-field heterogeneity fertility and carbon stock. The simula-
tion model is a process-based model developed by Rothamsted Research, UK (Coleman &
Jenkinson, 2014; Shirato & Yokozawa, 2005). Figure 3 illustrates the structure and func-
tion of the model. The model considers five carbon pools within the soil layer (decompos-
able plant material, resistant plant material, microbial biomass, humified organic matter,
inert organic matter) as well as the organic inputs (e.g., crop residue, weeds, manure). The
model simulates the carbon stock in a given soil layer and ­CO2 emission at a monthly inter-
val with the input data of monthly climate, soil properties, and monthly input of organic
matter. The carbon content of the pellet manure for variable-rate spreading was set to be
30% based on the chemical analysis of the manure. The monthly input of organic matter
from rice cropping was estimated from the typical grain yield and management practices.

Results

Comparison of ability and stability of spectral algorithms and models

Under open field conditions, the effective wavelength regions are limited to 401 ~ 1328
nm, 1481 ~ 1760, and 1970 ~ 2231 nm due to the sensitivity of sensors and the influence of
atmospheric vapor. Hence, predictive models were explored using the reflectance values
at an interval of 3 nm within the effective wavelength regions (a total of 492 wavebands).

Fig. 3  Simulation model of soil carbon dynamics

Precision Agriculture (2024) 25:2758–2775 2765

The predictive ability and stability of algorithms and models are summarized in Table 1.
The results suggest, first, the best predictive ability was achieved by the machine learning
methods (ANN, SVM, XGB) using h-NDSIs with r2val of 0.86 ~ 0.88. According to the
­ MSEval/RMSEcal, XGB had the worst stability although it showed similar predic-
ratio of R
tive ability (r2val = 0.87) as SVM. SVM was more stable than ANN or XGB. ANN or XGB
may be good at fitting, but the applicability seemed worse. Second, multivariable methods
(PLSR, iPLSR) had a comparable level of predictive ability with machine learning meth-
ods. iPLSR using a smaller number of wavebands proved superior to PLSR using all wave-
bands. This confirmed the importance of waveband selection. The linearity of predictive
models of ANN and SVM was superior to the linear regression models by PLSR or iPLSR.
The curvi-linear relationship between measured and predicted values suggested the limi-
tation of such linear regression methods. Third, the explored spectral indices (h-NDSIs)
using only two wavebands had a moderate ability (r2val = 0.84), and their stability was supe-
rior among all algorithms. In most algorithms, predictive models showed better ability and
stability when h-NDSIs were used as input instead of raw reflectance values. Although
Table 1 lists the best cases for each algorithm, these tendencies were common to all differ-
ent sets of wavebands (Ri, Rj) and h-NDSIs.
As a result, the most accurate and stable model was obtained by the hybrid algorithm
of h-NDSI and SVM (Fig. 2c). The coefficient of determination (r2 = 0.883; n = 120) sug-
gested the good potential of the hybrid algorithm and predictive models.

Operational performance of the prototype sensing/mapping system

Figure 4 illustrates the overall structure and functioning of the sensing/mapping system.
The sensing/mapping system allows high-speed processing of hyperspectral data with
GNSS data, generating SCC, diagnostic, and prescription maps as well as the visualization
of within-field distribution of SCC by high-resolution map and histogram with coefficient
of variation (CV %).
The hyperspectral and positional data were acquired smoothly by on-the-go meas-
urement using the system throughout the field tests in Fukushima. The operating speed

Table 1  Comparison of accuracy and stability of various algorithms for prediction of SCC
Algorithm Input data Number of Validation RMSEval/
wavebands RMSE-
r2 val RMSEval (%) cal(%)

Single waveband Ri 1 0.719 0.215 102

h-NDSI Ri, Rj 2 0.835 0.165 102
PLSR Ri 492 0.773 0.194 117
NDSIs 28 0.851 0.157 112
iPLSR Ri 18 0.826 0.170 111
ANN Ri 492 0.877 0.143 189
NDSIs 28 0.862 0.151 136
SVM Ri 492 0.862 0.151 127
NDSIs 28 0.883 0.139 112
XGB Ri 498 0.808 0.178 352
NDSIs 28 0.870 0.147 187
2766 Precision Agriculture (2024) 25:2758–2775

[ Sensing map: SCC ]

a) Speedy processing of
hyperspectral data & GNSS data

[ Diagnostic map: deficiency class ]

b) Generating and visualizing

three KMZ maps on GoogleEarth

[ Prescription map: amount of manure ]

c) KMZ
[ Variable-rate Spreader ]
Fig. 4  Overall structure and functioning of the hyperspectral sensing & mapping system. Three KMZ maps
are created and transferred to variable-rate spreader of pellet manure

was 3 ~ 4 km h-1, which is standard for rotary cultivation. Sensing data were processed
to generate the georeferenced reflectance data (Fig. 4a). The reflectance data were con-
verted to SCC values using the optimized predictive model. The mesh-maps of SCC,
fertility deficiency class, and the amount of manure to be spread were created for each
field shortly after rotary cultivation (Fig. 4b). The mesh size in this experiment was
2.4 m. These maps were visualized on GoogleEarth and transferred to a variable-rate
spreader in KMZ format (Fig. 4c).
Precision Agriculture (2024) 25:2758–2775 2767

Fig. 5  Capability of the sensing/mapping system tested in a decontaminated field in Fukushima. (a) predict-
ability, (b) reproducibility. (a) Comparison between the mesh-scale SCC data by the sensing system and
point SCC data by chemical analysis. (b) Comparison between the mash-scale SCC data obtained by differ-
ent tractor, speed, and light conditions

The predicted SCC values in mesh-basis were comparable with the SCC values by
chemical analysis of sampled soils within each corresponding mesh (Fig. 5a). The high
coefficient of determination (r2 = 0.856) suggests that the accuracy will be adequate for
operational applications. The mesh-SCC data acquired by repeated runs in a specific field
under different running and environmental conditions using two types of tractors and rotary
cultivators were also highly correlated with each other (r2 = 0.902; Fig. 5b).

Chronosequential change of SCC under different manure application scenarios

The simulated chronosequential change of SCC (%), within-field heterogeneity (CV %),
and carbon stock increment (tC ­ha−1) under different manure application scenarios are
shown in Fig. 6. Here, the six lines in the upper graphs represent the six typical meshes
within the actual field. Their initial SCC values cover the full range of SCC distribution
of the field. In the case of uniform application (Fig. 6a), the SSC values of six meshes
remained nearly parallel for 20 years. The CV value for the field decreased slowly due to
the increasing field average, but was still more than 15% even after 20 years (lower graph
of Fig. 6a). The field-scale carbon stock increase gradually and reached 8.4 tC h­ a−1 at 20
years.
Contrarily, in the case of variable-rate application (Fig. 6b), the SCC values in the six
meshes continuously converged to an average value. The CV decreased much faster and
reached 10% in 7 years and 5% in 12 years. Since the same total amount of manure was
applied each year in these two scenarios (Fig. 6a & 6b), the field scale increment of carbon
stock was also the same (8.4 tC ­ha−1 after 20 years). In the case that an additional amount
of manure could be applied (Fig. 6c), the CV value decreased to 10% in 3 years and 5% in
6 years.
2768 Precision Agriculture (2024) 25:2758–2775

a) Uniform spreading b) Variable-rate spreading c) Variable-rate spreading

䛊 Once/year 䛋 䛊 Once/year 䛋 䛊 Twice/year 䛋
SCC (%)

䛆Individual Mesh䛇 䛆Individual Mesh䛇 䛆Individual Mesh䛇

5 10 15 2200 5 10 15 20 5 10

䛆Field scale䛇 䛆Field scale䛇 䛆Field scale䛇

SCC (%)

(ΔCarbon stock tC/ha) (ΔCarbon stock tC/ha) (ΔCarbon stock tC/ha)

CV% (8.4) (9.3)
(8.4)

CV (%)
(7.0) (7.0)
(5.3) (5.4)
(3.1)
(3.1) (5.3)
SCC% CV% SCC%
SCC% CV%

5 10 15 2200 5 10 15 20 5 10
Year Year Year

Fig. 6  Simulated chronosequential change of SCC, within-field heterogeneity, and carbon stock under
different manure application scenarios. The six lines represent six typical meshes in a field. (a) Uniform
spreading, (b) Variable-rate spreading with the same total amount as in (a), (c) Variable-rate spreading with
double amount of manure

Prescription logics and prescription maps

The prescription map is the essential data for the variable-rate spreading. It gives the data
of the amount of pellet manure to be applied to individual meshes. The data are calcu-
lated from the SCC and diagnostic maps using a specific conversion logic (i.e., prescription
logic). Hence, the prescription logic also has a critical role for the efficient improvement
of within-field heterogeneity. Desirable prescription logics can be expressed by appropri-
ate equations, and utilized by the mapping programs. Figure 7 depicts three typical logics

SCC (%)

PS_mass by Eq1_linear
Frequency

Diagnostic deficiency class

PS_mass by Eq2_logistic
Frequency
Amount of manure
(mg m-2)

PS_mass by Eq3_exponential
Frequency

Diagnostic deficiency class (0~100)

Fig. 7  Three examples of prescription logics and resultant prescription maps. The diagnostic deficiency
class of each mesh is translated to the actual amount of pellet manure to be applied by linear, logistic, and
exponential equations. An appropriate equation can be selected from various equations incorporated in the
mapping system
Precision Agriculture (2024) 25:2758–2775 2769

(i.e., linear, logistic, and exponential) and resultant prescription maps. The prescription
logics can be designed flexibly considering the affordable amount of manure and desirable
amending scenarios considering the farmers’ soil management strategies.

Discussion

Potential and caveats of spectral sensing of SCC

The overall pattern of spectral reflectance was typical of lowland gray soils (Fig. 2a).
The strong absorption around 1400 and 1940 nm is due to water, hydroxyls, and clay
(Ben-Dor et al., 1997). The CV was relatively high in the visible (VIS) to near-infrared
(NIR) wavelength regions (500–900 nm) compared to the shortwave infrared (SWIR)
region (1300–2500 nm). In the NIR and SWIR regions, some weak absorption occurred
due to stretching and bending of the N–H, O–H, and C–H bonds; such spectral changes
are related to SCC values via the differences in lignin, cellulose, and other carbohy-
drates (Stenberg et al., 2010).
Normalization of spectral signatures using multiple wavebands is effective to reduce
the impacts of various error sources (e.g., atmospheric vapor, disturbed surface con-
ditions), so a number of vegetation indices have been devised (Daughtry et al., 2000;
Garbulsky et al., 2011; Gitelson et al., 2005; Huete, 1988; Mcfeeters, 1996; Vogelmann
et al., 1993). The NDSI is the generalized version of such normalization for multiple or
hyperspectral data and proved powerful to explore the effective combinations of wave-
bands (Inoue et al., 2008, 2016, 2019). The analytical results from the h-NDSI map
(Fig. 2b) indicated that the best NDSI using only two wavebands had better predic-
tive ability (r2 = 0.835) than the PLSR model using 492 bands (r2 = 0.773). iPLSR was
also better than PLSR and comparable to the best NDSI. The h-NDSI map suggested
the spectral absorption of soil components such as lignin, cellulose, and carbohydrate,
which are directly related to SCC, may not strongly affect the waveband selection in this
type of soil. Conversely, the changes in VNIR wavebands (i.e., soil color) were more
significant, presumably because of their close relationship with those carbon constitu-
ents. Therefore, systematic use of multiple NDSIs with high predictive ability.
The PLSR regression has been widely used for the prediction of soil properties to
make the most of abundant hyperspectral data (Stenberg et al., 2010; Tabatabai et al.,
2019; Udelhoven et al., 2003). However, the use of many continuous wavebands often
has negative effects due to the issue of multi-collinearity (Inoue et al., 2016). It was also
confirmed in this study. Also, reflectance data have been directly used for PLSR regres-
sion in most studies. However, using the reflectance values would not be preferable for
predictive purposes since it is not easy to obtain precise absolute reflectance under open
field conditions. We have to be careful about those two points in application of PLSR
regression algorithms.
The machine learning (ML) methods could have superior accuracy because it is non-
linear iterative fitting with no theoretical assumptions. However, their applicability and
interpretability are largely limited because of the same reason (Ali et al., 2015; Hol-
loway & Mengersen, 2018). The comparisons in Table 1 confirmed that ML algorithms
were helpful to improve the predictive accuracy. However, we have to be careful about
the applicability of the model. The ratio of RMSEval/RMSEcal clearly suggests unsta-
ble applicability of the ML models. Among the ML algorithms, SVM models proved
2770 Precision Agriculture (2024) 25:2758–2775

superior in both predictive accuracy and applicability. ANN and XGB could have some-
what higher or comparable accuracy with SVM, but the robustness of the model, i.e.,
applicability, was much lower than SVM.
Thus, the hybrid spectral algorithm, i.e., linkage between h-NDSI and SVM proved
superior to all other algorithms (Table 1, Fig. 2). The superiority of this hybrid algo-
rithm is well supported by the technological consideration above. Since the hybrid algo-
rithms is relatively simple and robust, it would be applied to the satellite hyperspectral
sensors.
Nevertheless, it is well recognized that the applicability of predictive models is highly
dependent on the individual calibration datasets, i.e., soil types and sites (Stevens et al.,
2008; Vaudour et al., 2019). Therefore, careful calibration of sensors and models is essen-
tial in any application. For example, Inoue and Zhi (2012) showed that the predictive accu-
racy was improved significantly by dividing the diverse types of soils into Andisols and
non-Andisol groups. Therefore, it is a reasonable strategy to prepare a few sets of predictive
models with intensive calibration for different soil-type groups of similar physico-chemical
characteristics. The present on-the-go system would apply to various regions by referenc-
ing the geospatial information on soil types provided by the soil information system.

Capability and caveats of the prototype sensing/mapping system

The experimental results in the decontaminated fields in the Fukushima region (Figs. 4
and 5) suggested the practical capability of the hyperspectral sensing/mapping system. The
system and workflow proved promising for amending the within-field heterogeneity in the
region. Nevertheless, further improvement in robustness and applicability is necessary.
Many error sources have to be considered, especially in on-the-go sensing of the soil car-
bon content (Angelopoulou et al., 2019; Bellon-Maurel & McBratney, 2011; Inoue & Zhi,
2012; Rodionov et al., 2014; Stevens et al., 2008).
Various surface factors such as crop residues, weeds, surface roughness, and soil mois-
ture affect the accuracy and consistency of prediction (Bogrekci & Lee, 2006; Lobell &
Asner, 2002; Nocita et al., 2013; Rodionov et al., 2014; Udelhoven et al., 2003). Since
the present sensing system is coupled with the rotary cultivator, the issues of crop resi-
dues, weeds, and roughness may not be critical compared to the other remote sensors. In
general, the change of reflectance spectra due to water content is significant. According to
the hyperspectral data obtained under a wide range of soil moisture conditions (i.e., oven
dry ~ field capacity) in diverse soil types, reflectance curves changed proportionally with
moisture contents in a parallel manner (Inoue & Zhi, 2012). The change of soil moisture
could be precisely assessed from the reflectance spectra. In addition, the actual range of
surface soil moisture just after rotary cultivation is relatively narrow, so its impact would
be minor.
The high reproducibility of SCC values under different solar and surface conditions
(Fig. 5b) suggests the stability of the hyperspectral estimation of SCC. However, the accu-
racy and robustness of the predictive models should be further improved based on addi-
tional training datasets towards application of the sensing system to diverse soils and
conditions.
Precision Agriculture (2024) 25:2758–2775 2771

Impact of variable‑rate application of manure on SCC, spatial heterogeneity,

and carbon stock

The simulation results (Fig. 6) clearly showed the chronosequential change of SCC, within-
field heterogeneity, and carbon stock under different manure application scenarios. The
slow response of soil fertility to soil management practices is well recognized (Poulton
et al., 2018), but the long-term impact achieved by the variable-rate application scenarios
proved significant. Since the within-field heterogeneity caused by the decontamination is
disastrous, the satisfactory level of uniformity would never be achieved by the traditional
uniform application of manure. Contrarily, the systematic variable-rate application using
the prescription maps will substantially shorten the periods to achieve the satisfactory level
of heterogeneity (e.g., 5%).
The systematic linkage of the spatial SCC data obtained from the sensing/mapping sys-
tem with the process-based simulation model proved helpful in predicting the long-term
change of the heterogeneity as well as the carbon stock in the soil. While many studies
attempted mapping the within-field heterogeneity (Heijting et al., 2011; Rhymes et al.,
2023; Schuster et al., 2023), such maps might have never been used for spatial simulation
to assess the long-term impact of soil management practices.
Spatial simulations under various management conditions allow optimization of the
efficient application scenarios. In addition, the present mapping system can utilize various
prescription logics (Fig. 7). The prescription logic is critical to determining the amount of
pellet manure to be applied to individual meshes by translating the deficiency class data
and SCC data. Both optimization of the soil management scenario and prescription logic
will be effective for the efficient amendment and enhancement of soil fertility.

Possible contributions to smart farming and carbon sequestration

Carbon sequestration by soil, less use of chemical fertilizer, and organic farming are
strongly required worldwide for mitigation of climate change and higher sustainability of
crop production. Hence, the role of soil fertility will become even more critical. Although
improving soil fertility needs long-term and continuous efforts (Coleman & Jenkinson,
2014; Poulton et al., 2018), the present results clearly showed that the variable-rate appli-
cation of manure based on SCC sensing would contribute to the efficient improvement of
soil fertility, farmland quality, and carbon stock in the soil.
In the same context, various efforts have been spent on the wide area monitoring of
SCC using satellite imagery (Castaldi et al., 2023). However, accurate and high-resolution
assessment of SCC is still challenging due to many constraints. Since the limited surface
conditions (bare soil) and the limited availability of ground-based validation data are the
major constraints for spaceborne monitoring, the present system may be one of the useful
options for acquisition of spatial validation data (i.e., ground-truth data).
2772 Precision Agriculture (2024) 25:2758–2775

Conclusions

A prototype of a tractor-based hyperspectral sensing system was developed for on-the-go

spectral sensing of the soil carbon content concurrently with routine rotary cultivation. The
system enables acquisition of the hyperspectral reflectance data in 350 ~ 2500 nm wave-
length region concurrently with GNSS data.
The hybrid algorithm using the hyperspectral normalized difference index method
(h-NDSI) and machine learning method (SVM) proved to have superior predictive accu-
racy and stability in the hyperspectral assessment of soil carbon content. Since the soil car-
bon content is the most simple and reasonable indicator of soil fertility, the system allows
the spatial assessment of soil fertility and carbon stock in the surface soil layer at the same
time. Just after each cultivation run, the system creates and visualizes three KMZ maps,
i.e., carbon content-, diagnosis-, and prescription maps on GoogleEarth. The performance
of the sensing/mapping system proved promising through the application to farmers’ fields
in Fukushima where within-field heterogeneity of soil fertility is disastrous due to the
decontamination after the nuclear power plant accident.
The simulation study using a dynamic process model of soil carbon balance suggested
the enormous role of the site-specific management i.e., variable-rate application of pellet
manure based on high resolution soil carbon maps. The disastrous within-field heterogene-
ity of soil fertility would never be amended by uniform manure application. The system-
atic linkage of the sensing/mapping system and variable-rate manure spreader via KMZ
maps enables an efficient amendment of within-field heterogeneity of soil fertility and for
enhancing soil fertility. The system would apply to various regions with extensive training
of predictive models and support of soil information systems.
Acknowledgements This work was supported in part by the research fund (JP J009997) from Ministry
of Agriculture, Forestry and Fisheries, Japan (MAFF) and Fukushima Institute for Research, Education
and Innovation (F-REI), Japan. The collaborations with its research consortium and Dr. Minoru Tanigaki
(Kyoto University) are highly appreciated. We are grateful to Mr. Tadayuki Furudate and Mr. Yutaro Tan
(Tohoku branch YANMAR AGRI JAPAN CO., LTD.) for the collaborations in instrument settings and field
experiments. Our special thanks are also due to Mr. Noboru Watanabe for his kind cooperation in the field
experiments.

Funding This work was supported in part by the research fund (JP J009997) from Ministry of Agricul-
ture, Forestry and Fisheries, Japan (MAFF) and Fukushima Institute for Research, Education and Innovation
(F-REI), Japan.

Data Availability Data may be available from the authors upon reasonable request and with the permission
of the research participants.

Declarations
Conflict of interest The authors have no competing interests to declare that are relevant to the content of this
article.

References
Adamchuk, V. I., Hummel, J. W., Morgan, M. T., & Upadhyaya, S. K. (2004). On-the-go soil sensors for
precision agriculture. Computers and Electronics in Agriculture, 44, 71–91. https://​doi.​org/​10.​1016/j.​
compag.​2004.​03.​002
Precision Agriculture (2024) 25:2758–2775 2773

AHDB. (2018). Measuring and managing soil organic matter. Retrieved https://​ahdb.​org.​uk/​knowl​edge-​
libra​ry/​measu​ring-​and-​manag​ing-​soil-​organ​ic-​matter. from 15 Feb 2024
Ali, I., Greifeneder, F., Stamenkovic, J., Neumann, M., & Notarnicola, C. (2015). Review of machine learn-
ing approaches for biomass and soil moisture retrievals from remote sensing data. Remote Sensing, 7,
16398–16421. https://​doi.​org/​10.​3390/​rs712​15841
Angelopoulou, T., Tziolas, N., Balafoutis, A., Zalidis, G., & Bochtis, D. (2019). Remote sensing techniques
for soil organic carbon estimation. Remote Sensing, 11, 676. https://​doi.​org/​10.​3390/​rs110​60676
Batjes, N. H. (1996). Total carbon and nitrogen in the soils of the world. European Journal of Soil Science,
47, 151–163.
Bellon-Maurel, V., & McBratney, A. B. (2011). Near-infrared (NIR) and midinfrared (MIR) spectroscopic
techniques for assessing the amount of carbon stock in soils - Critical review and research perspectives.
Soil Biology and Biochemistry, 43, 1398–1410.
Ben-Dor, E., Inbar, Y., & Chen, Y. (1997). The reflectance spectra of organic matter in the visible near-
infrared and shortwave infrared region (400–2500 nm) during controlled decomposition process.
Remote Sensing of Environment, 61, 1–15.
Bogrekci, I., & Lee, W. S. (2006). Effects of soil moisture content on absorbance spectra of sandy soils in
sensing phosphorus concentrations using UV–VIS–NIR spectroscopy. Transactions ASABE, 49, 1175–
1180. https://​doi.​org/​10.​13031/​2013.​21717
Castaldi, F., Koparan, M. H., Wetterlind, J., Zydelis, R., Vinci, I., & Savas, A. O. (2023). Assessing the
capability of Sentinel-2 time-series to estimate soil organic carbon and clay content at local scale in
croplands. ISPRS Journal of Photogrammetry and Remote Sensing, 199, 40–60. https://​doi.​org/​10.​
1016/j.​isprs​jprs.​2023.​03.​016
Coleman, K., & Jenkinson, D.S (2014) RothC - A model for the turnover of carbon in soil: Model
description and users guide. Rothamsted Research (Harpenden, Herts, AL5 2JQ, UK.), 44p. https://​
www.​rotha​msted.​ac.​uk/​sites/​defau​lt/​files/​RothC_​guide_​WIN.​pdf
Daughtry, C. S. T., Walthall, C. L., Kim, M. S., de Colstoun, E. B., & McMurtrey, J. E. I. I. I. (2000).
Estimating corn leaf chlorophyll concentration from leaf and canopy reflectance. Remote Sensing of
Environment, 74, 229–239. https://​doi.​org/​10.​1016/​S0034-​4257(00)​00113-9
Demenois, J., Torquebiau, E., Arnoult, M. H., Eglin, T., Masse, D., Assouma, M. H., et al. (2020). Bar-
riers and strategies to boost soil carbon sequestration in agriculture. Frontiers in Sustainable Food
Systems, 4, 37. https://​doi.​org/​10.​3389/​fsufs.​2020.​00037
European Commission (2020) The European Green Deal. Retrieved from https://​commi​ssion.​europa.​eu/​
strat​egy-​and-​policy/​prior​ities-​2019-​2024/​europ​ean-​green-​deal_​en15 Feb 2024.
Garbulsky, M. F., Peñuelas, J., Gamon, J., Inoue, Y., & Filella, I. (2011). The photochemical reflec-
tance index (PRI) and the remote sensing of leaf, canopy and ecosystem radiation use efficiencies.
Remote Sensing of Environment, 115, 281–297. https://​doi.​org/​10.​1016/j.​rse.​2010.​08.​023
Gholizadeh, A., Žižala, D., Saberioon, M., & Boruvka, L. (2018). Soil organic carbon and texture
retrieving and mapping using proximal, airborne and Sentinel-2 spectral imaging. Remote Sensing
of Environment, 218, 89–103.
Gitelson, A. A., Viña, A., Rundquist, D. C., Ciganda, V., & Arkebauer, T. J. (2005). Remote estimation
of canopy chlorophyll content in crops. Geophysical Research Letters, 32, L08403. https://​doi.​org/​
10.​1029/​2005G​L0226​88
Hara, M. (2001). Fertilizer pellets made from composted livestock manure. Food & Fertilizer Technol-
ogy Center, Extension Bulletin, 506, 1–12.
Heijting, S., de Bruin, S., & Bregt, A. K. (2011). The arable farmer as the assessor of within-field soil
variation. Precision Agriculture, 12, 488–507. https://​doi.​org/​10.​1007/​s11119-​010-​9197-y
Holloway, J., & Mengersen, K. (2018). Statistical machine learning methods and remote sensing for
sustainable development goals: a review. Remote Sensing, 10, 1365. https://​doi.​org/​10.​3390/​rs100​
91365
Huete, A. R. (1988). A soil vegetation adjusted index (SAVI). Remote Sensing of Environment, 25, 295–
309. https://​doi.​org/​10.​1016/​0034-​4257(88)​90106-X
Inoue, Y., Peñuelas, J., Miyata, A., & Mano, M. (2008). Normalized difference spectral indices for esti-
mating photosynthetic efficiency and capacity at a canopy scale derived from hyperspectral and
­CO2 flux measurements in rice. Remote Sensing of Environment, 112, 156–172. https://​doi.​org/​10.​
1016/j.​rse.​2007.​04.​011
Inoue, Y., Sakaiya, E., Zhu, Y., & Takahashi, W. (2012). Diagnostic mapping of canopy nitrogen con-
tent in rice based on hyperspectral measurements. Remote Sensing of Environment, 126, 210–221.
https://​doi.​org/​10.​1016/j.​rse.​2012.​08.​026
2774 Precision Agriculture (2024) 25:2758–2775

Inoue, Y., & Zhi, X. (2012) Estimating soil carbon content in diverse ecosystems based on hyperspectral
analysis. Proceeding of the International Geoscience and Remote Sensing Symposium (IGARSS
2012), Munich, WEP.P204, 1–4.
Inoue, Y., Guérif, M., Baret, F., Skidmore, A., Gitelson, A., & Schlerf, M. (2016). Simple and robust
methods for remote sensing of canopy chlorophyll content: A comparative analysis of hyperspectral
data for different types of vegetation. Plant, Cell & Environment, 39, 2609–2623. https://​doi.​org/​
10.​1111/​pce.​12815
Inoue, Y., Darvishzadeh, R., & Skidmore, A. (2019) Hyperspectral assessment of ecophysiological func-
tioning for diagnostics of crops and vegetation. In Hyperspectral Remote Sensing of Vegetation
2nd Ed. Vol. 3 (eds. Thenkabail, P.S., Lyon, J., Huete, A.), London, CRC Press-Taylor and Francis
group, 25–72.
Inoue, Y. (2020). Satellite- and drone-based remote sensing of crops and soils for smart farming - a
review. Soil Science and Plant Nutrition, 66, 798–810. https://​doi.​org/​10.​1080/​00380​768.​2020.​
17388​99
Inoue, Y., Saito, T., Iwasaki, A., Nemoto, T., & Ono, T. (2020). Hyperspectral assessment of soil fertil-
ity in farm fields in Fukushima decontaminated after the radioactive fallout. Soil Science and Plant
Nutrition, 66, 820–827. https://​doi.​org/​10.​1080/​00380​768.​2020.​17532​37
Inoue, Y., & Yoshino, K. (2023) Hyperspectral sensing and mapping of soil fertility for amending
within-field heterogeneity. Precision Agriculture ’23 (ed. J. V. Stafford), Wageningen Academic
Publishers, Amsterdam, 925–932.
Knadel, M., Thomsen, A., Schelde, K., & Greve, M. H. (2015). Soil organic carbon and particle sizes map-
ping using vis–NIR, EC and temperature mobile sensor platform. Computers and Electronics in Agri-
culture, 114, 134–144. https://​doi.​org/​10.​1016/j.​compag.​2015.​03.​013
Koyama, T., Chammek, C., & Niamsrichhand, N. (1973). Nitrogen application technology for tropical rice
as determined by field experiments using 15N tracer technique. Tropical Agriculture Research Center,
Japan, Technical Bulletin, 3, 13–19.
Ladoni, M., Bahrami, H. L., Alavipanah, S. K., & Norouzi, A. A. (2010). Estimating soil organic carbon
from soil reflectance: A review. Precision Agriculture, 11, 82–99.
Lobell, D. B., & Asner, G. P. (2002). Moisture effects on soil reflectance. Soil Science Society of America
Journal., 66, 722–727. https://​doi.​org/​10.​2136/​sssaj​2002.​0722
MAFF Japan (2021) Strategy for Sustainable Food Systems. Retrieved from https://​www.​maff.​go.​jp/e/​polic​
ies/​env/​env_​policy/​meadri.​html15 Feb 2024.
Mcfeeters, S. K. (1996). The use of normalized difference water index (NDWI) in the delineation of open
water features. International Journal of Remote Sensing, 17, 1425–1432.
Moran, M. S., Inoue, Y., & Barnes, E. M. (1997). Opportunities and limitations for image-based remote
sensing in precision crop management. Remote Sensing of Environment, 61, 319–346. https://​doi.​org/​
10.​1016/​S0034-​4257(97)​00045-X
Nocita, M., Stevens, A., Noon, C., & van Wesemael, B. (2013). Prediction of soil organic carbon for differ-
ent levels of soil moisture using Vis-NIR spectroscopy. Geoderma, 199, 37–42.
Poulton, P., Johnston, J., Macdonald, A., White, R., & Powlson, D. (2018). Major limitations to achieving “4
per 1000” increases in soil organic carbon stock in temperate regions: Evidence from long-term experi-
ments at Rothamsted research. United Kingdom. Global Change Biology, 24, 2563–2584. https://​doi.​
org/​10.​1111/​gcb.​14066
Rhymes, J., Chadwick, D. R., Williams, A. P., Harris, I. M., Lark, R. M., & Jones, D. L. (2023). Evaluating
the accuracy and usefulness of commercially-available proximal soil mapping services for grassland
nutrient management planning and soil health monitoring. Precision Agriculture, 24, 898–920. https://​
doi.​org/​10.​1007/​s11119-​022-​09979-z
Rodionov, A., Pätzold, S., Welp, G., Pallares, R. C., Damerow, L., & Amelung, W. (2014). Sensing of soil
organic carbon using visible and near-infrared spectroscopy at variable moisture and surface rough-
ness. Soil Science Society of America Journal, 78, 949–957.
Schuster, J., Mittermayer, M., Maidl, F., Nätscher, L., & Hülsbergen, K. (2023). Spatial variability of soil
properties, nitrogen balance and nitrate leaching using digital methods on heterogeneous arable fields
in southern Germany. Precision Agriculture, 24, 647–676. https://​doi.​org/​10.​1007/​s11119-​022-​09967-3
Shirato, Y., & Yokozawa, M. (2005). Applying the Rothamsted Carbon Model for long-term experiments
on Japanese paddy soils and modifying it by simple tuning of the decomposition rate. Soil Science and
Plant Nutrition, 51, 405–415.
Six, J., Bossuyt, H., Degryze, S., & Denef, K. (2004). A history of research on the link between (micro)
aggregates, soil biota, and soil organic matter dynamics. Soil Tillage Research, 79, 7–31. https://​doi.​
org/​10.​1016/j.​still.​2004.​03.​008
Precision Agriculture (2024) 25:2758–2775 2775

Soil Survey Staff (2011) Soil Survey Laboratory Information Manual. Soil Survey Investigations Report
No. 45, Version 2.0. R. Burt (ed.). U.S. Department of Agriculture, Natural Resources Conservation
Service. 412p.
Stenberg, B., Rossel, R. A. V., Mouazen, A. M., & Wetterlind, J. (2010). Visible and near infrared spectros-
copy in soil science. Advances in Agronomy, 107, 163–215. https://​doi.​org/​10.​1016/​S0065-​2113(10)​
07005-7
Stevens, A., van Wesemael, B., Bartholomeus, H., Rosillon, D., Tychon, B., & Ben-Dor, E. (2008). Labora-
tory, field and airborne spectroscopy for monitoring organic carbon content in agricultural soils. Geo-
derma, 144, 395–404.
Tabatabai, S., Knadel, M., Thomsen, A., & Greve, M. H. (2019). On-the-go sensor fusion for prediction of
clay and organic carbon using pre-processing survey, different validation methods, and variable selec-
tion. Soil Science Society of America Journal, 83, 300–310.
Tanigaki, M., Inoue, Y., Momota, S., Saito, T., Nemoto, T., Ono, T., et al. (2022). Development of a robot
for the measurement of radioactive contamination and fertility of the soil in farmland. Radiation Pro-
tection Dosimetry, 198, 964–970. https://​doi.​org/​10.​1093/​rpd/​ncac0​20
Tucker, C. J. (1979). Red and photographic infrared linear combinations for monitoring vegetation. Remote
Sensing of Environment, 8, 127–150. https://​doi.​org/​10.​1016/​0034-​4257(79)​90013-0
Udelhoven, T., Emmerling, C., & Jarmer, T. (2003). Quantitative analysis of soil chemical properties with
diffuse reflectance spectrometry and partial least-square regression: A feasibility study. Plant and Soil,
251, 319–329. https://​doi.​org/​10.​1023/A:​10230​08322​682
USDA (2020) Agriculture Innovation Agenda. Retrieved from https://​www.​usda.​gov/​aia15 Feb 2024.
Vaudour, E., Gomez, C., Loiseau, T., Baghdadi, N., Loubet, B., & Arrouays, D. (2019). The impact of
acquisition date on the prediction performance of topsoil organic carbon from Sentinel-2 for croplands.
Remote Sensing, 11, 2143.
Vogel, S., Bönecke, E., Kling, C., Kramer, E., Lück, K., Philipp, G., Rühlmann, J., Schröter, I., & Geb-
bers, R. (2022). Direct prediction of site-specific lime requirement of arable fields using the base neu-
tralizing capacity and a multi-sensor platform for on-the-go soil mapping. Precision Agriculture, 23,
127–149. https://​doi.​org/​10.​1007/​s11119-​021-​09830-x
Vogelmann, J. E., Rock, B. N., & Moss, D. M. (1993). Red edge spectral measurements from sugar maple
leaves. International Journal of Remote Sensing, 14, 1563–1575.

Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and
institutional affiliations.

Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under
a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted
manuscript version of this article is solely governed by the terms of such publishing agreement and applicable
law.