remote sensing

Article
Synergistic Use of Multi-Temporal Radar and Optical Remote
Sensing for Soil Organic Carbon Prediction
Sara Dahhani 1, *, Mohamed Raji 1 and Yassine Bouslihim 2

1 Faculty of Sciences Ben M’sik, Hassan II University of Casablanca, Sidi Othmane,

Casablanca P.O. Box 7955, Morocco
2 National Institute of Agricultural Research (INRA), CRRA Tadla, Rabat P.O. Box 415, Morocco;
[email protected] or [email protected]
* Correspondence: [email protected]

Abstract: Exploring soil organic carbon (SOC) mapping is crucial for addressing critical challenges in
environmental sustainability and food security. This study evaluates the suitability of the synergistic
use of multi-temporal and high-resolution radar and optical remote sensing data for SOC prediction in
the Kaffrine region of Senegal, covering over 1.1 million hectares. For this purpose, various scenarios
were developed: Scenario 1 (Sentinel-1 data), Scenario 2 (Sentinel-2 data), Scenario 3 (Sentinel-1 and
Sentinel-2 combination), Scenario 4 (topographic features), and Scenario 5 (Sentinel-1 and -2 with
topographic features). The findings from comparing three different algorithms (Random Forest (RF),
XGBoost, and Support Vector Regression (SVR)) with 671 soil samples for training and 281 samples
for model evaluation highlight that RF outperformed the other models across different scenarios.
Moreover, using Sentinel-2 data alone yielded better results than using only Sentinel-1 data. However,
combining Sentinel-1 and Sentinel-2 data (Scenario 3) further improved the performance by 6% to
11%. Including topographic features (Scenario 5) achieved the highest accuracy, reaching an R2 of 0.7,
an RMSE of 0.012%, and an RPIQ of 5.754 for the RF model. Applying the RF and XGBoost models
under Scenario 5 for SOC mapping showed that both models tended to predict low SOC values across
the study area, which is consistent with the predominantly low SOC content observed in most of the
Citation: Dahhani, S.; Raji, M.; training data. This limitation constrains the ability of ML models to capture the full range of SOC
Bouslihim, Y. Synergistic Use of variability, particularly for less frequent, slightly higher SOC values.
Multi-Temporal Radar and Optical
Remote Sensing for Soil Organic Keywords: soil organic carbon; Sentinel-1; Sentinel-2; multi-temporal data; radar imagery; optical imagery
Carbon Prediction. Remote Sens. 2024,
16, 1871. https://doi.org/10.3390/
rs16111871

Academic Editors: Xiaoling Wu, 1. Introduction

Chong Luo, Liujun Zhu and Soil organic carbon (SOC) constitutes an essential element within the global carbon
Xiaoji Shen cycle, playing an important role in mitigating climate change, improving soil health, and
Received: 25 March 2024
enhancing agricultural productivity. Quantifying and monitoring SOC content is essential
Revised: 11 May 2024 for evaluating soil quality, orienting sustainable land management practices, and achieving
Accepted: 13 May 2024 international climate change mitigation commitments [1]. Consequently, SOC mapping
Published: 24 May 2024 has garnered global interest as a means of addressing environmental and food security
challenges. Interest in SOC mapping has been particularly pronounced in Africa [2–4],
which faces a unique combination of challenges and opportunities in soil management. The
diverse climates and ecosystems of Africa present a varied soil landscape, where accurate
Copyright: © 2024 by the authors. SOC mapping can make a significant contribution to improving agricultural resilience,
Licensee MDPI, Basel, Switzerland. food security, and climate change adaptation efforts [5]. Furthermore, digital mapping of
This article is an open access article SOC in a sub-Saharan country like Senegal can make a significant contribution to achieving
distributed under the terms and several Sustainable Development Goals (SDGs).
conditions of the Creative Commons
In this context, the integration of machine learning (ML) algorithms with Earth obser-
Attribution (CC BY) license (https://
vation (EO) data has been recognized as a powerful approach for improving the accuracy
creativecommons.org/licenses/by/
and efficiency of SOC prediction and mapping [6,7]. According to Nenkam Mentho et al. [8],
4.0/).

Remote Sens. 2024, 16, 1871. https://doi.org/10.3390/rs16111871 https://www.mdpi.com/journal/remotesensing

Remote Sens. 2024, 16, 1871 2 of 18

among 110 studies conducted in Africa, 34 and 6 specifically focused on SOC and soil
organic matter (SOM), respectively, both with and without the consideration of other soil
attributes. For instance, Hengl et al. [9] demonstrated the utility of the Africa Soil Informa-
tion Service (AfSIS) in conjunction with Moderate Resolution Imaging Spectroradiometer
(MODIS) data for the mapping of various soil properties, including SOC and pH, at a
resolution of 250 m. Utilizing the same data source, Vågen et al. [5] employed a Random
Forest model for SOC mapping across the African continent. Furthermore, Hengl et al. [10]
generated 30 m resolution pan-African maps detailing various soil nutrients, such as SOC,
pH, total nitrogen (N), phosphorus (P), and potassium (K), among others, through the com-
bination of diverse EO datasets and ensemble ML algorithms. Bouasria et al. [11] explored
the feasibility of utilizing pan-sharpened Landsat-8 imagery (15 m resolution) for SOM
mapping via multiple linear regression and artificial neural networks. Similarly, Bouslihim
et al. [12] employed a Random Forest approach for SOM mapping using Landsat-8 imagery
at a 30 m resolution.
Recent advances in remote sensing technologies have expanded the opportunities for
digital soil mapping (DSM). Sentinel-1 (C-band synthetic aperture radar) and Sentinel-2
(multi-spectral optical data) satellites can provide unprecedented opportunities for de-
tailed and frequent monitoring of the Earth’s surface, including soil properties. While
Sentinel-2 provides high-resolution optical images useful for capturing surface features
and vegetation indices, Sentinel-1 radar data offer advantages by penetrating cloud cover
and providing information on soil moisture, which is closely linked to SOC content [13,14].
Within the African context, out of 110 studies, 11 have utilized Sentinel-2 data for DSM
purposes, yet only 2 have yielded SOC maps at a 10 m resolution [8]. In the first study,
Mponela et al. [15] used Sentinel-2 data to determine soil fertility (including SOC, NPK,
etc.) for a 0.45 ha area in Malawi. Additionally, Flynn et al. [16] predicted soil particle
size distribution and SOC content at a 10 m resolution over a 366 ha area in South Africa.
Despite the potential, the application of Sentinel data in Africa for SOC mapping remains
underexploited. Predominantly, global studies have employed Sentinel data from a single
date [17–21]. However, a limited number of investigations have harnessed multi-temporal
data from Sentinel-1 or Sentinel-2 for enhanced analysis [22–24].
This study investigates several hypotheses related to DSM for SOC prediction. Firstly,
we hypothesized that the combined use of multi-temporal Sentinel-1 and Sentinel-2 data
would outperform the individual use of either data source in predicting SOC content.
Secondly, we posited that incorporating topographic features as auxiliary environmental
variables would further enhance the accuracy of SOC prediction models. Finally, we
anticipated that different machine learning algorithms (RF, SVR, and XGBoost) would
exhibit varying performance levels depending on the specific combination of input variables
and the chosen scenario. To test these hypotheses, we evaluated the efficacy of these data
sources and algorithms across various scenarios, aiming to identify the optimal approach
for generating high-resolution SOC maps. This research contributes valuable insights
into the synergistic potential of Sentinel data and the role of environmental variables
and machine learning in advancing digital soil mapping techniques for SOC prediction.
In addition, this paper supports SDG 13 (Climate Action) by providing crucial data for
understanding and monitoring carbon sequestration capacities, thus informing climate
change mitigation strategies, and SDG 15 (Life on Earth) through its potential to improve
soil health, promote sustainable land use practices, and combat desertification, which
is particularly important in arid and semi-arid regions. In addition, by enabling better-
informed agricultural practices, this research indirectly contributes to SDG 2 (Zero Hunger)
and SDG 1 (No Poverty) by improving food security and livelihoods through improved
soil fertility and crop yields. Thus, digital mapping of soil organic carbon serves as a
multi-disciplinary tool that cuts across various environmental and socio-economic aspects
of sustainable development in the context of African countries.
Remote Sens. 2024, 16, x FOR PEER REVIEW 3 of 18

as a multi-disciplinary tool that cuts across various environmental and socio-economic

Remote Sens. 2024, 16, 1871 aspects of sustainable development in the context of African countries. 3 of 18

2. Materials and Methods

2. Materials
2.1. Methodologyand Methods
2.1. Methodology
The flowchart presented in Figure 1 outlines the process for predicting SOC using
The flowchart
Sentinel-1 (radar) andpresented in (multi-spectral)
Sentinel-2 Figure 1 outlines the topographic
data, process for predicting SOC
features, and using
ML al-
Sentinel-1 (radar) and Sentinel-2 (multi-spectral) data,
gorithms. The methodology is divided into three main stages.topographic features, and ML algo-
rithms. The methodology is divided into three main stages.

Figure 1. Methodological flowchart adopted to predict SOC under different scenarios.

Figure 1. Methodological flowchart adopted to predict SOC under different scenarios.
(1) Data preparation: Multi-temporal data from Sentinel-1 and Sentinel-2 were pro-
(1) Data preparation: Multi-temporal data from Sentinel-1 and Sentinel-2 were pro-
cessed for various radiometric and geometric image corrections, Sentinel-2 bands were
cessed
used to for various
extract radiometric
various remoteand geometric
sensing indices,image corrections, features
and topographic Sentinel-2 bands
were were
prepared,
used to extract various remote sensing indices, and topographic
and all these data were combined with SOC content ground samples (952).features were prepared,
and all(2)these
Datadata were combined
pre-processing: Thewith SOC content
prepared data wereground samples
considered (952).
under four scenarios to
evaluate the suitability of Sentinel products for SOC prediction: Scenariofour
(2) Data pre-processing: The prepared data were considered under scenarios
1 (only to
Sentinel-1
evaluate the suitability of Sentinel products for SOC prediction: Scenario 1 (only Sentinel-
data), Scenario 2 (only Sentinel-2 data), Scenario 3 (Sentinel-1 and -2 combination), Scenario
4 (topographic features), and Scenario 5 (Scenario 3 and Scenario 4). Also, feature selection
was used to identify the most relevant variables for SOC prediction.
(3) Modeling and evaluation: Three ML algorithms were applied, Random Forest
(RF), XGBoost, and Support Vector Regression (SVR), using a 70/30% split for training and
 1 data), Scenario 2 (only Sentinel-2 data), Scenario 3 (Sentinel-1 and -2 combination), Sce-
nario 4 (topographic features), and Scenario 5 (Scenario 3 and Scenario 4). Also, feature
selection was used to identify the most relevant variables for SOC prediction.
Remote Sens. 2024, 16, 1871 (3) Modeling and evaluation: Three ML algorithms were applied, Random Forest 4 of 18
(RF), XGBoost, and Support Vector Regression (SVR), using a 70/30% split for training and
testing. All models were evaluated using the coefficient of determination (R2), the Root
Mean Square
testing. Error (RMSE),
All models and the Ratio
were evaluated usingofthe
Performance
coefficient to
of Inter-Quartile
determinationRange (RPIQ).
(R2 ), the Root
(4) Finally, the best models were used for SOC spatial prediction.
Mean Square Error (RMSE), and the Ratio of Performance to Inter-Quartile Range (RPIQ).
(4) Finally, the best models were used for SOC spatial prediction.
2.2. Study Area Description
2.2. Study Area Description
The Kaffrine region covers an area of 11,181 km2 (≈1.1 million ha), representing ap-
The Kaffrine
proximately 5.6% ofregion covers
Senegal. an area in
It is situated of central km2 (≈1.1
11,181 Senegal, millionbyha),
bounded therepresenting
coordinates
approximately
14°43’46.6′′N 5.6% of Senegal.
15°51’40.2′′W and It13°45’31.4′′N
is situated in14°34’00.7′′W
central Senegal, bounded
(Figure by the
2). The coordinates
region serves
14◦a43
as
′ 46.6′′ N 15◦ 51′ 40.2′′ W and 13◦ 45′ 31.4′′ N 14◦ 34′ 00.7′′ W (Figure 2). The region serves as
transitional zone between the Sahelian and Sudanian climatic domains. The topogra-
a transitional
phy zone between
is predominantly the Sahelian
flat, with and Sudanian
a gentle slope descendingclimatic
fromdomains. The topography
north to south. The area
is characterized by three primary soil types: tropical ferruginous, hydromorphic, The
is predominantly flat, with a gentle slope descending from north to south. area
and hol-
is characterized by three primary soil types: tropical ferruginous, hydromorphic,
omorphic soils. Climatically, Kaffrine experiences high temperatures throughout the year, and
holomorphic soils. Climatically, Kaffrine experiences high temperatures
with notable fluctuations, and has a distinct seasonal pattern comprising a short rainy throughout the
year, with notable fluctuations, and has a distinct seasonal pattern comprising
season from July to October and a prolonged dry season lasting from eight to nine months. a short rainy
season
The from annual
average July to October and a prolonged
rainfall recorded dry season
for the period fromlasting
2016 tofrom
2021eight to nine months.
was approximately
The average
702.6 mm. annual rainfall recorded for the period from 2016 to 2021 was approximately
702.6 mm.

Figure
Figure2. Limitsof
2.Limits ofKaffrine
Kaffrineregion
regionand
and geographical
geographical localization
localization of
of soil
soil samples.
samples.

2.3. Soil and Remote Sensing Data Preparation

Soil data: As a first step, a soil sampling design was structured to ensure that the
sampling accurately represented the soil properties across the study area. For that, a
stratified random sampling design was applied, and the entire study area was partitioned
into different blocks (10 × 10 km), which enabled a systematic organization of the sampling
effort and ensured extensive coverage of the study area. Out of these blocks, 45 were
Remote Sens. 2024, 16, 1871 5 of 18

selected through a random selection process to ensure that our sampling represented the
various landforms and soil types within the study area, thereby minimizing any potential
bias that could arise from selectively choosing specific blocks. Subsequently, in each of these
45 randomly selected blocks, soil sampling was conducted at 23 distinct sites, and some
sites were eliminated due to access constraints. Between 2018 and 2019, soil samples were
collected at each site from the top 20 cm. After collection, the soil samples were transported
to the laboratory for preparation and analysis. The preparation involved drying the
soil, removing all plant debris, and sieving through a 2 mm mesh to achieve a uniform
soil fraction for analysis, and the SOC content was measured using the Walkley–Black
method [25].
Remote sensing data: The multi-temporal dataset included images from Sentinel-1,
obtained from https://search.asf.alaska.edu/ (accessed on 22 Decembre 2023), and Sen-
tinel 2, obtained from the Copernicus Data Space Ecosystem (https://browser.dataspace.
copernicus.eu/, accessed on 15 January 2024). For Sentinel-1, the dataset included a se-
ries of synthetic aperture radar (SAR) images extending from May 2018 to March 2019,
including 4 scenes to cover the study area. These images featured dual polarization modes
(VH and VV) and were all captured in an ascending orbit. The pre-processing steps for
Sentinel-1 imagery were performed using SNAP (8.0.0) software, encompassing calibration
to convert digital number (DN) values into backscatter coefficients, multi-looking to reduce
speckle noise, and filtering to further improve image quality, and, since SAR images have
side view imaging characteristics, SAR image geometric misrepresentation may appear in
relief displacement. The Radar Geometric Terrain Correction tool was chosen to apply the
Range Doppler method for image registration [26,27]. In total, we obtained 22 images (11
for VH polarization and 11 for VV polarization).
Furthermore, Sentinel-2 L1C multi-spectral images were acquired from May 2018 to
March 2019 (July and September were excluded due to unfavorable weather conditions). A
total of 9 acquisition dates were obtained and atmospherically corrected using the sen2cor
processor in the SNAP (8.0.0) software [28]. Sentinel-2 bands at each date were used
to calculate various remote sensing indices, such as Brightness Index (BI), Coloration
Index (CI), Modified Normalized Difference Water Index (MNDWI), MERIS Terrestrial
Chlorophyll Index (MTCI), Normalized Difference Vegetation Index (NDVI), Normalized
Difference Water Index (NDWI), Redness Index (RI), and Soil-Adjusted Vegetation Index
(SAVI) values, and the formulas used for the index calculations are detailed in Table 1. The
index labels were coded as follows: Index_Month_Year; for instance, NDVI_5_18 refers to
the NDVI for May 2018.

Table 1. List of remote sensing indices calculated from Sentinel-2 bands.

Index Full Name Formula Reference

BI Brightness Index sqrt ((Red2 /Green2 )/2) [29]
CI Coloration Index (Red − Blue)/Red [30]
MNDWI Modified Normalized Difference Water Index (Green − SWIR)/(Green + SWIR) [31]
MTCI MERIS Terrestrial Chlorophyll Index (Red Edge 2 − Red Edge 1)/(Red Edge 1 − Red) [32]
NDVI Normalized Difference Vegetation Index ((NIR − Red)/(NIR + Red)) [33]
NDWI Normalized Difference Water Index (Green − NIR)/(Green + NIR) [34]
RI Redness Index (Red − Green)/(Red + Green) [33]
SAVI Soil-Adjusted Vegetation Index ((NIR − Red)/(NIR + Red + L)) × (1 + L) [35]

The digital elevation model was obtained from ASTGTM (version 3 with a 30 m reso-
lution) and used to extract different topographic features, such as elevation, slope, aspect,
Topographic Wetness Index (TWI), profile curvature, plan curvature, and Multi-Resolution
Index of Valley Bottom Flatness (MRVBF), using the SAGA program (version 9.1.2). The
Remote Sens. 2024, 16, 1871 6 of 18

elevation band was resampled to a 10 m resolution using the bilinear interpolation method
in QGIS Desktop (version 3.34.0) before the calculation of other topographic features.

2.4. Data Pre-Processing and Machine Learning Algorithms

The Recursive Feature Elimination (RFE) method was employed, utilizing a Random
Forest Regressor as the estimator [36]. RFE selects features by recursively considering
progressively smaller sets of features; it initiates with all predictors in the dataset and
sequentially removes the least significant feature at each iteration. A Random Forest Regressor
was configured with 100 trees and a fixed random state (42) to ensure reproducibility and
was instructed to select the top 10 features for Scenario 1 and Scenario 2 and the top 20
features for Scenario 3 and Scenario 5 based on the training dataset. For Scenario 4, all
seven topographic features were used.
Furthermore, three different ML algorithms were compared: (1) Random Forest (RF),
an ensemble learning method that constructs a multitude of decision trees at the training
time and outputs the average prediction (for the regression task) of the individual trees [37].
RF is highly recommended for remote sensing applications due to its ability to handle
large datasets and its robustness against overfitting, which makes it a powerful tool for
land cover classification [38], estimation of soil properties [39], and biomass prediction [40],
among other applications. (2) XGBoost (Extreme Gradient Boosting) is an efficient and
scalable implementation of gradient-boosted decision trees, designed for speed and per-
formance. Developed by Chen and Guestrin [41], XGBoost has gained popularity through
its performance in ML challenges and has been noted for its ability to handle sparse data
and for its scalability and regularized boosting technique that helps prevent overfitting.
(3) Support Vector Regression (SVR) applies the principles of support vector machines
(SVMs) to regression problems. The SVR model aims to fit the best line within a prede-
fined or epsilon margin of tolerance with the goal of minimizing error and fitting the
model within the defined threshold [42]. RF and SVR models were selected due to their
widespread application in DSM applications and their history of yielding diverse results.
Comparing these established methods allowed us to assess which is more suitable for this
specific case. Additionally, we included XGBoost, which can be considered a newer ML
algorithm, to explore its potential benefits in SOC prediction.
Each model was developed using 70% of the data (n = 671), and the remaining 30%
(n = 281) was used for model testing. Also, the hyperparameter tuning approach was
applied for every model based on its parameters, and all these parameters are listed
in Table 2 with descriptions. The Google Colab platform [43] was used for all steps
related to data pre-processing, predictive modeling, and SOC mapping. Due to the limited
performance of the free version of Google Colab, the layer stack prepared for SOC mapping
was divided into 10 parts, and each part was used with the desired model to predict SOC;
afterwards, the 10 parts were mosaicked to return a single SOC raster. The complete script
utilized in Google Colab for this research is accessible on GitHub (links are available in the
Data Availability Statement).
All developed models were evaluated using three metrics: the coefficient of deter-
mination (R2 ) (Equation (1)), the Root Mean Square Error (RMSE) (Equation (2)), and the
Ratio of Performance to Inter-Quartile Range (RPIQ) (Equation (3)). R2 provides a measure
of how well the observed outcomes are replicated by the model, based on the proportion of
the total variation in outcomes explained by the model [44]. An R2 of 1 indicates perfect
correlation, while an R2 of 0 indicates that the model does not explain any of the variability
in the response data around its mean. The RMSE is a standard way to measure the error
of a model in predicting quantitative data [45]. It represents the square root of the second
sample moment of the differences between predicted values and observed values or the
quadratic mean of these differences. RMSE is particularly useful when large errors are
particularly undesirable, as it squares the errors before averaging, thus giving a relatively
high weight to large errors. The RPIQ is calculated by dividing the interquartile distance
(IQR) by the RMSE [46]; higher RPIQ values indicate better model performance, as they
Remote Sens. 2024, 16, 1871 7 of 18

suggest that the model’s predictions are accurate relative to the natural variability of the
data, and lower RPIQ values suggest that the model’s predictions are less accurate, with
prediction errors that are large in comparison to the variability of the dataset.
2
∑(yi − ŷi )
R2 = 1 −  2 (1)
∑ yi − y
r
1 n
n ∑ i =1 i
RMSE = (y − ŷi )2 (2)

RPIQ = IQR RMSE


(3)
where yi is the actual value of the dependent variable for the ith observation, ŷi is the
predicted value of the dependent variable for the ith observation, and y is the mean value
of the dependent variable. The IQR represents the range between the first (25th percentile)
and third quartiles (75th percentile) of the observed data.

Table 2. List of hyperparameters used for RF, SVR, and XGBoost model tuning.

Model Hyperparameter Description

n_estimators The number of trees in the forest
max_features The number of features to consider when looking for the best split
RF max_depth The maximum depth of the tree
min_samples_split The minimum number of samples required to split an internal node
min_samples_leaf The minimum number of samples required to be at a leaf node
C Regularization parameter
SVR epsilon Specifies the epsilon tube
gamma Kernel coefficient for r’bf’, p’oly’, and s’igmoid’.
learning_rate (or eta in XGBoost documentation) Step size shrinkage used to prevent overfitting
max_depth Maximum depth of a tree
gamma Minimum loss reduction required to make a further partition on a leaf node of the tree

XGBoost Control the subsample ratio of columns for the tree building at different levels of tree
colsample_bytree
building
min_child_weight Minimum sum of instance weight (hessian) needed in a child
subsample Subsample ratio of the training instances
n_estimators Number of gradient boosted trees, equivalent to the number of boosting rounds

3. Results
3.1. Statistical Description
For the training dataset with 671 samples (Table 3), the SOC content ranges from a
minimum of 0.11% to a maximum of 0.72%, with a mean value of approximately 0.22%. The
standard deviation is 0.0725, indicating a moderate spread around the mean. The 25th, 50th
(median), and 75th percentiles are 0.175%, 0.21%, and 0.26%, respectively, showing a slight
skew towards lower SOC values (Figure 3). Comparatively, the test dataset (281 samples)
shows a slightly tighter range of SOC values, from 0.12% to 0.57%, with a mean value very
close to that of the training set, at about 0.22%. The standard deviation in the test set is
slightly lower at 0.0692, suggesting a slightly less varied set of SOC percentages than in the
training dataset. Percentile values are also similar to those of the training set, with the 25th,
50th, and 75th percentiles at 0.18%, 0.21%, and 0.26%, respectively. Overall, both datasets
show a relatively consistent range of SOC percentages, with a central tendency around
0.22%. The slight differences in spread and range between the training and test datasets
 For the training dataset with 671 samples (Table 3), the SOC content ranges from a
minimum of 0.11% to a maximum of 0.72%, with a mean value of approximately 0.22%.
The standard deviation is 0.0725, indicating a moderate spread around the mean. The
25th, 50th (median), and 75th percentiles are 0.175%, 0.21%, and 0.26%, respectively,
Remote Sens. 2024, 16, 1871 8 of 18
showing a slight skew towards lower SOC values (Figure 3). Comparatively, the test da-
taset (281 samples) shows a slightly tighter range of SOC values, from 0.12% to 0.57%,
with a mean value very close to that of the training set, at about 0.22%. The standard de-
suggestin
viation minor variations
the test in soil lower
set is slightly organic
at carbon
0.0692, content across
suggesting the twoless
a slightly datasets,
variedbut, overall,
set of SOC
they exhibit similar statistical properties, with a low SOC content.
percentages than in the training dataset. Percentile values are also similar to those of the
training set, with the 25th, 50th, and 75th percentiles at 0.18%, 0.21%, and 0.26%, respec-
Table 3. Summary statistics for train and test SOC (%) data.
tively. Overall, both datasets show a relatively consistent range of SOC percentages, with
a central
Data tendency around 0.22%.
Count Min The slight Max
differences in spread and
Mean range between
Standard Deviationthe
training
Train
and test datasets
671
suggest0.11
minor variations
0.72
in soil organic
0.224
carbon content
0.072
across
the two datasets, but, overall, they exhibit similar statistical properties, with a low SOC
Test 281 0.12 0.57 0.223 0.069
content.

Figure 3. Distribution of SOC content for train and test data.

3.2. Feature Selection and Correlation Analysis

Table 3. Summary statistics for train and test SOC (%) data.
The Recursive Feature Elimination (RFE) method was used to select the most influen-
tialData Count
features across Min
four scenarios. Maximportant
The Mean Standard
variables identified forDeviation
different scenarios
areTrain 671 4. For 0.11
listed in Table Sentinel-10.72
data in 0.224 0.072 features, 5 VH
Scenario 1, out of 10 selected
Test were selected
features 281 for0.12
different0.57 0.223
months (VH_5_18, 0.069 VH_9_18, and
VH_6_18, VH_8_18,
VH_3_19) and 5 VV features were selected for July, September, and December 2018 and
for February
3.2. and March
Feature Selection 2019. For Analysis
and Correlation Sentinel-2 data in Scenario 2, three MNDWI features
(MNDWI_6_18, MNDWI_12_18,
The Recursive Feature Elimination and MNDWI_3_19),
(RFE) methodtwo wasSAVI
usedfeatures
to select(SAVI_8_18 and
the most influ-
SAVI_12_18), two MTCI features (MTCI_11_18 and MTCI_3_19), and two CI
ential features across four scenarios. The important variables identified for different sce- features
(CI_5_18
narios areand CI_3_19)
listed were
in Table selected
4. For with one
Sentinel-1 dataBIinfor June 2018.
Scenario In of
1, out the10third scenario,
selected which
features, 5
combined Sentinel-1 and Sentinel-2 data, 20 features were selected to equally represent both
VH features were selected for different months (VH_5_18, VH_6_18, VH_8_18, VH_9_18,
datasets and to see if the same features would be selected. The results revealed that, from
and VH_3_19) and 5 VV features were selected for July, September, and December 2018
the 20 features, 15 were selected from Sentinel-2 data and only 5 from Sentinel-1, including
and for February and March 2019. For Sentinel-2 data in Scenario 2, three MNDWI fea-
3 VH variables and 2 VV variables. For Scenario 4, all seven topographic features were
tures (MNDWI_6_18, MNDWI_12_18, and MNDWI_3_19), two SAVI features (SAVI_8_18
used, and the high importance of elevation was noticed. For the last scenario, which com-
and SAVI_12_18), two MTCI features (MTCI_11_18 and MTCI_3_19), and two CI features
bined Sentinel-1, Sentinel-2, and topographic features, 1 topographic feature was selected
(CI_5_18 and CI_3_19) were selected with one BI for June 2018. In the third scenario, which
(elevation) and 4 and 15 features were selected from Sentinel-1 and Sentinel-2, respectively.
The correlation analysis revealed that the backscatter coefficient in VV polarization
for March 2019 (VV_3_19) showed the highest correlation with SOC, with a value of 0.202,
followed by VV_12_18 polarization with a negative correlation of −0.17 and VH_6_18 with
a positive correlation of 0.16. This indicates that the relationship between SOC content and
radar backscatter in both VV and VH polarizations is not uniformly positive or negative;
rather, it is variable. Both polarizations exhibit correlations with SOC, albeit with differing
magnitudes and directions (positive and negative). For Sentinel-2 multi-spectral data, the
MNDWI exhibited a generally positive correlation with SOC, with coefficients ranging
from 0.15 to 0.27. Conversely, a negative correlation was observed between SOC content
and SAVI, with coefficients from −0.25 to −0.28. Similarly, the CI demonstrated a negative
correlation with SOC, with values between −0.27 and −0.36. Lastly, elevation showed the
highest correlation value of 0.42.
Remote Sens. 2024, 16, 1871 9 of 18

Table 4. List of selected features (bands) across different scenarios.

Scenario Selected Features

VH_5_18, VH_6_18, VH_8_18, VH_9_18, VH_3_19, VV_7_18, VV_9_18,
Scenario 1 (Sentinel-1)
VV_12_18, VV_2_19, VV_3_19
BI_6_18, CI_5_18, CI_3_19, MNDWI_6_18, MNDWI_12_18, MNDWI_3_19,
Scenario 2 (Sentinel-2)
MTCI_11_18, MTCI_3_19, SAVI_8_18, SAVI_12_18
BI_5_18, BI_6_18, CI_5_18, CI_3_19, MNDWI_5_18, MNDWI_6_18,
MNDWI_12_18, MNDWI_3_19, MTCI_10_18, MTCI_11_18, MTCI_3_19,
Scenario 3 (Sentinel-1 + Sentinel-2)
NDWI_8_18, SAVI_8_18, SAVI_12_18, SAVI_3_19, VH_5_18, VH_6_18,
VH_9_18, VV_9_18, VV_3_19
Scenario 4 (Topography) Elevation, slope, aspect, TWI, profile curvature, plan curvature, MRVBF
Elevation, BI_5_18, BI_6_18, CI_5_18, CI_3_19, MNDWI_5_18, MNDWI_6_18,
MNDWI_12_18, MNDWI_3_19, MTCI_10_18, MTCI_11_18, MTCI_3_19,
Scenario 5 (Sentinel-1 + Sentinel-2 + Topography)
NDVI_8_18, SAVI_8_18, SAVI_12_18, SAVI_3_19, VH_5_18, VH_6_18,
VH_9_18, VV_3_19

3.3. Machine Learning Performance

Table 5 lists the fitted values of the different hyperparameters used and the R2 , RMSE,
and RPIQ results for the different models in four scenarios. The RF model consistently
outperformed the other models across all scenarios, indicating its superior predictive ability
for this dataset (Table 6 and Figure 4). In Scenario 1, the RF model achieved an R2 of
0.36, an RMSE of 0.042, and an RPIQ of 1.644, surpassing XGBoost (R2 : 0.34, RMSE: 0.046,
RPIQ: 1.501) and SVR (R2 : 0.21, RMSE: 0.054, RPIQ: 1.279). This trend continued, with the
RF model exhibiting the highest performance in Scenario 2 (R2 : 0.49, RMSE: 0.037, RPIQ:
1.866), Scenario 3 (R2 : 0.61, RMSE: 0.024, RPIQ: 2.877), Scenario 4 (R2 : 0.65, RMSE: 0.02,
RPIQ: 3.45), and Scenario 5 (R2 : 0.70, RMSE: 0.012, RPIQ: 5.754). Comparatively, the SVR
model consistently showed the lowest performance metrics across all scenarios, indicating
that it might be less suitable for SOC prediction under these study conditions. Scenario
Remote Sens. 2024, 16, x FOR PEER REVIEW 5
11 of 18
represented the best outcome for all models, suggesting that combining all data (Sentinel-1,
Sentinel-2, and topography) was most conducive to predictive modeling. The RF model’s
superior performance in Scenario 5, with an R2 of 0.70, an RMSE of 0.012, and a higher
XGBoost 0.64 0.017 4.061
RPIQ of 5.754, demonstrated its robustness and efficiency in handling the relationship
SVR
between different predictors and SOC content. 0.56 0.023 3.002

Figure 4.
Figure Scatter plots
4. Scatter plots of
of measured
measured vs.
vs. predicted
predictedSOC
SOC%
%for
forRF
RF(A),
(A),XGBoost
XGBoost(B),
(B),and
andSVR
SVR(C)
(C)models
mod-
under
els Scenario
under 5. 5.
Scenario

In Scenario 5, which exhibited the highest performance among all scenarios, the
importance of various predictors was analyzed for three models: RF, XGBoost, and SVR
(Figure 5). Elevation stands out as the most influential variable within all three models.
Following elevation, the CI for the date 3_19 is the next most prominent variable for the
Remote Sens. 2024, 16, 1871 10 of 18

RF and XGBoost models, suggesting its repeated importance. As we delve further into
the hierarchy, VV and VH radar bands from Sentinel-1, acquired at different time points,
consistently rank high in importance, particularly for the RF and XGBoost models. This
pattern also holds true for MNDWI and MTCI, where different dates yield a consistently
high ranking across these two models, reflecting their key roles as predictors. In contrast
to the RF and XGBoost models, which display a concurrence in the importance ranking
of these variables, the SVR model also assigns high importance to elevation, indicating
its cross-model relevance. However, its pattern of importance for other variables differs,
allocating varying degrees of importance to the radar bands and spectral indices.

Table 5. Fitted values of different hyperparameters.

Model Hyperparameter Scenario 1 Scenario 2 Scenario 3 Scenario 4 Scenario 5

Number of
selected 10 10 20 7 20
features
n_estimators 100 500 100 100 100
max_features Log2 Log2 Log2 Log2 Log2
RF max_depth 10 10 15 5 16
min_samples_split 2 2 5 2 5
min_samples_leaf 2 2 2 2 2
C 0.1 10 1 0.1 0.5
SVR epsilon 0.01 0.01 0.01 0.01 0.01
gamma 0.01 1 0.01 0.01 0.01
learning_rate 0.1 0.05 0.1 0.05 0.1
max_depth 7 7 4 5 5
gamma 0 0 0 0 0
XGBoost colsample_bytree 0.5 0.5 1 0.5 1
min_child_weight 5 10 10 5 5
subsample 1 0.5 0.5 0.5 0.5
n_estimators 50 50 50 50 50

Table 6. Validation accuracy for the three models across different scenarios.

Scenario Model R2 RMSE RPIQ

RF 0.36 0.042 1.644
Scenario 1 XGBoost 0.34 0.046 1.501
SVR 0.21 0.054 1.279
RF 0.49 0.037 1.866
Scenario 2 XGBoost 0.45 0.039 1.770
SVR 0.35 0.049 1.409
RF 0.61 0.024 2.877
Scenario 3 XGBoost 0.51 0.028 2.466
SVR 0.38 0.047 1.469
 XGBoost 0.64 0.017 4.061
SVR 0.56 0.023 3.002

Remote Sens. 2024, 16, 1871 11 of 18

Table 6. Cont.

Scenario Model R2 RMSE RPIQ

RF 0.65 0.02 3.45
Scenario 4 XGBoost 0.62 0.023 3
SVR 0.47 0.035 1.971
RF 0.7 0.012 5.754
Scenario 5 XGBoost 0.64 0.017 4.061
Figure 4. Scatter plots of measured vs. predicted SOC % for RF (A), XGBoost (B), and SVR (C) mod-
SVR 0.56 0.023 3.002
els under Scenario 5.

Figure 5.
Figure Featureimportance
5. Feature importance for
for RF,
RF, XGBoost,
XGBoost, and
and SVR
SVR models
models under
under Scenario
Scenario 5.
5.

3.4. Soil Organic Carbon Mapping

3.4. Soil Organic Carbon Mapping
Figure 6 shows the spatial distribution of SOC using the RF and XGBoost algorithms
Figure 6 shows the spatial distribution of SOC using the RF and XGBoost algorithms
with Scenario 6, which has been defined as the optimal combination of model and scenario
with Scenario 6, which has been defined as the optimal combination of model and scenario
configurations. The RF algorithm predicted SOC concentrations ranging from 0.12% to
configurations. The RF algorithm predicted SOC concentrations ranging from 0.12% to
0.42%, corresponding to the dominant low SOC content in the majority of the 671 soil
0.42%, corresponding to the dominant low SOC content in the majority of the 671 soil
samples used for model training. In this dataset, a small number of samples (n = 35) had
samples used for model training. In this dataset, a small number of samples (n = 35) had
SOC levels above 0.35%, and only seven samples exceeded the 0.45% value. Consequently,
SOC
the RFlevels above 0.35%,
prediction modeland only seven samples
satisfactorily capturedexceeded the 0.45%
the general value.
pattern Consequently,
of SOC values. In
the RF prediction model satisfactorily captured the general pattern
contrast, the XGBoost algorithm predicted a more restricted range of SOC of SOC values. In from
values, con-
trast, the XGBoost algorithm predicted a more restricted range of SOC values,
0.15% to 0.32%, indicating a lower degree of heterogeneity in its predictions than RF. from 0.15%
to 0.32%,these
Despite indicating a lower
variations, bothdegree of heterogeneity
algorithms in its predictions
generally reflected the bias ofthan RF. Despite
the training data
these variations, both algorithms generally reflected the bias of the training
towards the lowest SOC values, reducing the ability of the models to reflect the full data towards
range
the lowest
of SOC SOC values,
variability, reducingless
particularly the frequent
ability ofsamples.
the models Thetorestricted
reflect thepredictive
full rangerange
of SOCof
the models suggests a reduced sensitivity to the complex relationships between SOC and
influencing covariates, leading to a potential underestimation of SOC levels in areas where
they naturally exceed the dominant range of the training dataset.
 variability, particularly less frequent samples. The restricted predictive range of the mod-
els suggests a reduced sensitivity to the complex relationships between SOC and influenc-
Remote Sens. 2024, 16, 1871 ing covariates, leading to a potential underestimation of SOC levels in areas where12theyof 18
naturally exceed the dominant range of the training dataset.

Figure 6.
Figure 6. Spatial
Spatial distribution of SOC
distribution of SOC content
content (%)
(%) for
for RF
RF and
and XGBoost
XGBoost models.
models.

4. Discussion
To
To thoroughly
thoroughly discuss the findings of this study, study, three main aspects were were considered:
considered:
(i) feature
feature importance
importancein inSOC
SOCprediction,
prediction,(ii)(ii)the
theperformance
performance ofof
the various
the variousscenarios
scenarios using
us-
Sentinel-1
ing Sentinel-1 andand
Sentinel-2 and and
Sentinel-2 topographic
topographic data,data,
and and
(iii) (iii)
the effectiveness
the effectiveness and and
comparative
compar-
analysis of theof
ative analysis three ML algorithms.
the three ML algorithms.
Firstly,
Firstly, the RFE method was
the RFE method was used
used toto select
select thethe most
mostimportant
importantvariables/features
variables/features for for
SOC prediction. For that, 10 variables were identified for Scenarios
SOC prediction. For that, 10 variables were identified for Scenarios 1 and 2, 20 variables 1 and 2, 20 variables
were
were identified
identified forfor Scenarios
Scenarios 33 and
and 5, and 77 variables
5, and variables werewere identified
identified variables
variables forfor Scenario
Scenario
4. The number of variables for Scenarios 3 and 5 was increased
4. The number of variables for Scenarios 3 and 5 was increased to assess whether the RFE to assess whether the
RFE
model model
would would
extractextract identical
identical variables
variables from from Sentinel-1,
Sentinel-1, Sentinel-2,
Sentinel-2, and topographic
and topographic data,
data, or if one dataset would predominate over the others.
or if one dataset would predominate over the others. The variables identified The variables identified
as being as
being significant were MNDWI, SAVI, and MTCI, each with more
significant were MNDWI, SAVI, and MTCI, each with more than three variables from dif- than three variables from
different months,
ferent months, indicating
indicating their
their relevance
relevance overover different
different timetime periods.
periods. TheThe importance
importance of
of these variables is explained by the fact that SAVI and MTCI
these variables is explained by the fact that SAVI and MTCI reflect vegetation [47,48], reflect vegetation [47,48],
which
which is is indirectly
indirectly correlated
correlated with
with soil
soil health
health and and fertility
fertility [5,49]
[5,49] and
and consequently
consequently serves
serves
as a proxy for soil organic matter content [50]. This association has been supported by
as a proxy for soil organic matter content [50]. This association has been supported by
numerous studies that have identified vegetation indices, such as SAVI, NDVI, and others,
numerous studies that have identified vegetation indices, such as SAVI, NDVI, and others,
to predict SOC or SOM [51–56]. The link between MNDWI and SOC is more indirect and
to predict SOC or SOM [51–56]. The link between MNDWI and SOC is more indirect and
complex. Similarly, SOC affects soil physical and chemical properties, including color,
complex. Similarly, SOC affects soil physical and chemical properties, including color,
texture, and moisture retention capacity. These properties can influence soil reflectance
texture, and moisture retention capacity. These properties can influence soil reflectance
characteristics in different spectral bands, including green and SWIR bands, and may
characteristics in different spectral bands, including green and SWIR bands, and may in-
indirectly highlight the importance of soil moisture parameters in SOC prediction [57–59],
directly highlight the importance of soil moisture parameters in SOC prediction [57–59],
as moisture-rich environments can facilitate the preservation and accumulation of organic
as moisture-rich environments can facilitate the preservation and accumulation of organic
carbon in soil [1,60,61]. Furthermore, our results align with those of Lu et al. [62], who
carbon in soil [1,60,61]. Furthermore, our results align with those of Lu et al. [62], who
highlighted the importance of MNDWI alongside other soil moisture indices such as the
highlighted the importance of MNDWI alongside other soil moisture indices such as the
Topographic Wetness Index (TWI) for SOC prediction. CI and BI showed a significant
TopographictoWetness
contribution Index (TWI)
SOC prediction due tofor SOC
their prediction.
ability to capture CI variations
and BI showed in soil acolor,
significant
which
contribution to SOC prediction due to their ability to capture variations
are often indicative of SOM content and other soil properties [63,64]. The correlation in soil color, which
are often SOC
between indicative
and CI of and
SOMBIcontent and other
was already soil properties
highlighted [63,64].
in previous The correlation
studies, such as Saha be-
tween SOC and CI and BI was already highlighted in previous studies,
et al. [65], which demonstrated that different spectral color indices, especially CI, are such as Saha et al.
important for SOC prediction and mapping.
The Sentinel-2-derived indices used in Scenario 2 contributed more significantly than
the Sentinel-1 dual-polarization indices (VV and VH). This can be attributed to the superior
ability of Sentinel-2 variables to predict SOC compared with Sentinel-1, which is reflected
in the performance differences between the models. In detail, Scenario 2 showed higher
Remote Sens. 2024, 16, 1871 13 of 18

performances for RF (R2 = 0.49, RMSE = 0.037%) and XGBoost (R2 = 0.45, RMSE = 0.039%)
compared to Scenario 1, for which the RF performance was R2 = 0.36 and RMSE = 0.042%
and the XGBoost performance was R2 = 0.34 and RMSE = 0.046%. In addition, the com-
bination of the two scenarios resulted in an even higher performance for RF (R2 = 0.61,
RMSE = 0.024%) and XGBoost (R2 = 0.51, RMSE = 0.028%), with a significant contribution
from Sentinel-2 variables. This advantage of Sentinel-2 has been confirmed by various
studies, such as Nguyen et al. [54], who found that SOC prediction performance using
Sentinel-2 was superior to that using Sentinel-1, with R2 values of 0.44 versus 0.25. Zhang
et al. [66] obtained similar results, with an R2 of 0.47 for Sentinel-2 versus 0.26 for Sentinel-1.
In addition, Fatholoumi et al. [67] and Wang and Zhou [68] pointed out that the use of
multi-temporal variables improved prediction performance due to the dynamic relation-
ship between SOC and vegetation across a longer period compared to using data from a
single date. Furthermore, the improvement in performance observed from the combination
of the two scenarios was further validated by Zhang et al. [66], who reported an improve-
ment in accuracy ranging between 2% and 5%. Similarly, Zhou et al. [69] highlighted that
combining Sentinel-1 and Sentinel-2 data led to an increase in SOC prediction accuracy by
5 to 6% and a reduction in error by 5% to 7%. Including topographical features increased
the performance of all models, with a significant contribution from elevation, the highest
performance being reached by the RF model with an R2 of 0.7, an RMSE of 0.012%, and an
RPIQ of 5.754. The importance and contribution of topographic features were highlighted
by Zhou et al. [70], who showed that elevation, slope, and TWI contributed more than 27%
to the model’s explanation. Additionally, Li et al. [71] showed that relief and TWI were
the most important variables controlling SOC. The same was demonstrated by Gibson
et al. [72], indicating that topographic features have an impact on SOC modeling at different
resolutions. Furthermore, the same reasoning for grouping environmental covariates was
demonstrated by Duarte et al. [73], based on Landsat-8 and various other covariates, such
as climate and topography, and yielded the best results for SOC stocks in forested land.
The comparison of ML algorithms revealed that RF and XGBoost outperformed the
SVR model, mainly due to their ensemble nature, which offers greater adaptability in
addressing complex, non-linear relationships within data. Across all scenarios, RF and
XGBoost consistently demonstrated higher R2 values compared to the SVR model, indi-
cating a greater proportion of variance explained by the dependent variable, as well as
lower RMSE values. These results are also reflected in other studies, such as that of Nguyen
et al. [54], who highlighted that XGBoost and RF surpassed the SVR model in predicting
SOC content using Sentinel-1 and Sentinel-2 data, achieving a higher performance with
an R2 value higher than 0.7. Similarly, Siewert [74] compared various algorithms for SOC
prediction and identified a superior performance of RF models over others. Moreover,
Zhang et al. [66] observed that RF could outperform XGBoost when using separate Sentinel
data, which is in line with our findings of an RF with R2 values of 0.61 and 0.7 for Scenarios
3 and 4, respectively, versus R2 values of 0.51 and 0.64 for XGBoost and 0.38 and 0.56 for
SVR. The performance results obtained in the present study are similar to those reported
by Pouladi et al. [75], who used only Sentinel-2, and Nguyen et al. [54], with R2 values
around 0.72 for RF; however, these values were higher than those obtained in other studies
that demonstrated low performance, such as Shafizadeh-Moghadam et al. [23] and Tajik
et al. [76], with performance being characterized by R2 values less than 0.5. The low perfor-
mance in these studies can generally be attributed to factors such as high heterogeneity
with an extensive study area size and the low density of sampling points [70]. In our case,
the reasons for the low performance for Scenario 1 and Scenario 3 may be attributed to
the low variability in SOC content (min = 0.11%, max = 0.72%), which could introduce
complexity into the modeling process [12]. The SOC distribution also revealed that the
XGBoost algorithm predicts a lower SOC value than the RF model. This could reflect more
conservative estimation or potential underfitting where the XGBoost model does not fully
capture the higher SOC values present in the training data, perhaps due to model com-
plexity or regularization parameters. Clearly, both models have limitations in representing
Remote Sens. 2024, 16, 1871 14 of 18

the less frequent, slightly higher SOC values, which were few in the training data. This
skew towards lower SOC values is a common problem in machine learning, where model
performance is strongly influenced by the distribution of the training dataset. In practical
applications, this could potentially mean that areas with naturally higher SOC levels could
be underestimated.

5. Contributions, Limitations, and Future Research Directions

This study contributes significantly to the field of DSM by demonstrating the potential
of combining Sentinel-1 and Sentinel-2 data for high-resolution (10 m) SOC prediction,
offering valuable insights for stakeholders in African agriculture and beyond. Our scenario-
based approach sheds light on the influence of different environmental variables on model
performance, highlighting the importance of considering topography alongside remotely
sensed data. Additionally, the comparative analysis of machine learning algorithms pro-
vides guidance for selecting the most suitable method based on specific data and objectives.
However, some limitations were encountered. Despite achieving reasonable accuracy, the
models exhibited a bias towards the dominant low SOC values within the training data,
resulting in a reduced ability to capture the full range of SOC variability. This limitation
suggests a potential underestimation of SOC, which could impact land management de-
cisions. Additionally, computational limitations restricted the generation of uncertainty
maps, hindering a more comprehensive assessment of model reliability.
Future research should prioritize addressing these limitations and exploring new
avenues for improvement. Techniques like data augmentation or incorporating prior
knowledge about SOC distribution could mitigate the bias towards low values and enhance
the models’ ability to represent the full spectrum of SOC variability. Exploring alternative
feature selection methods, such as those based on expert opinion [77], as well as alternative
machine learning approaches, such as ensemble methods or meta-learners that combine
multiple algorithms with diverse structures, may improve prediction accuracy and over-
come the problem of the limit of singular models in predicting SOC values outside the
limit of dominant values. Furthermore, investigating computationally efficient methods
for generating uncertainty maps remains crucial for enhancing the interpretability and
reliability of SOC predictions. By addressing these challenges and building upon this
study’s foundation, future research can further advance DSM and provide increasingly
accurate and reliable high-resolution SOC maps. These maps will be invaluable tools for
stakeholders in African agriculture and other regions, supporting sustainable land man-
agement practices, soil conservation efforts, and informed decision making for improved
agricultural productivity and environmental sustainability.

6. Conclusions
This study evaluated the suitability of time-series radar (Sentinel-1), optical (Sentinel-2),
and topography data for SOC prediction across a variety of scenarios and predictive mod-
eling frameworks. In conclusion, this research demonstrates the feasibility of integrating
high-resolution EO data with ML algorithms to predict SOC in case of low-value content.
The key findings are as follows:
• Combining multi-temporal Sentinel-1 and Sentinel-2 data enhances the precision of
SOC prediction, with an improvement of R2 values and reduced error compared to
using single-source data. This underscores the benefit of multi-sensor data fusion for
DSM applications.
• Including topographic data improves the accuracy of different models and signifies
that the integration of all data inputs culminates in optimal model efficacy.
• RF and XGBoost algorithms outperform SVR in SOC prediction across different sce-
narios, highlighting the effectiveness of ensemble learning techniques in handling
complex spatial datasets.
• Despite the overall success, the models predominantly predict low SOC values, re-
flecting the inherent limitations in capturing the full range of SOC variability, which
Remote Sens. 2024, 16, 1871 15 of 18

suggests the need for further refinement of modeling approaches to better address less
frequent, high-concentration samples.
Finally, the generated SOC maps are crucial for informing sustainable land manage-
ment practices and climate change mitigation strategies. Furthermore, in future studies,
it will be interesting to test radar and optical data for other soil fertility parameters, or to
evaluate time series for other satellite products such as hyperspectral data.

Author Contributions: Conceptualization, S.D., M.R. and Y.B.; Data curation, S.D. and Y.B.; Formal
analysis, S.D. and Y.B.; Methodology, S.D., M.R. and Y.B.; Supervision, M.R.; Validation, S.D. and
Y.B.; Writing—original draft, S.D., M.R. and Y.B. All authors have read and agreed to the published
version of the manuscript.
Funding: This research received no external funding.
Data Availability Statement: The data that support the findings of this study are available from the
corresponding author upon reasonable request. The scripts used for this paper can be accessed at
https://github.com/yassinebos/SOC_prediction-mapping (accessed on 28 April 2024).
Conflicts of Interest: The authors declare no conflicts of interest.

References
1. Lal, R. Soil Carbon Sequestration Impacts on Global Climate Change and Food Security. Science 2004, 304, 1623–1627. [CrossRef]
2. von Fromm, S.F.; Hoyt, A.M.; Lange, M.; Acquah, G.E.; Aynekulu, E.; Berhe, A.A.; Haefele, S.M.; McGrath, S.P.; Shepherd, K.D.;
Sila, A.M.; et al. Continental-scale controls on soil organic carbon across sub-Saharan Africa. Soil Discuss. 2020, 2020, 1–39.
[CrossRef]
3. Schulze, R.E.; Schütte, S. Mapping soil organic carbon at a terrain unit resolution across South Africa. Geoderma 2020, 373, 114447.
[CrossRef]
4. Odebiri, O.; Mutanga, O.; Odindi, J.; Naicker, R. Modelling soil organic carbon stock distribution across different land-uses in
South Africa: A remote sensing and deep learning approach. ISPRS J. Photogramm. Remote Sens. 2022, 188, 351–362. [CrossRef]
5. Vågen, T.G.; Winowiecki, L.A.; Tondoh, J.E.; Desta, L.T.; Gumbricht, T. Mapping of soil properties and land deg-radation risk in
Africa using MODIS reflectance. Geoderma 2016, 263, 216–225. [CrossRef]
6. Al Masmoudi, Y.; Bouslihim, Y.; Doumali, K.; Hssaini, L.; Namr, K.I. Use of machine learning in Moroccan soil fertility prediction
as an alternative to laborious analyses. Model. Earth Syst. Environ. 2022, 8, 3707–3717. [CrossRef]
7. Wadoux, A.M.-C.; Minasny, B.; McBratney, A.B. Machine learning for digital soil mapping: Applications, challenges and suggested
solutions. Earth-Sci. Rev. 2020, 210, 103359. [CrossRef]
8. Nenkam Mentho, A.; Wadoux, A.M.C.; Minasny, B.; Silatsa, F.B.; Yemefack, M.; Ugbaje, S.; Akpa, S.; van Zijl, G.M.; Bouslihim, Y.;
Chabala, L.; et al. Applications and Challenges of Digital Soil Mapping in Africa. Available online: https://ssrn.com/abstract=47
25182 (accessed on 15 March 2024). [CrossRef]
9. Hengl, T.; Heuvelink, G.B.; Kempen, B.; Leenaars, J.G.; Walsh, M.G.; Shepherd, K.D.; Sila, A.; MacMillan, R.A.; de Jesus, J.M.;
Tamene, L.; et al. Mapping soil properties of Africa at 250 m resolution: Random forests significantly improve current predictions.
PLoS ONE 2015, 10, e0125814. [CrossRef] [PubMed]
10. Hengl, T.; Miller, M.A.E.; Križan, J.; Shepherd, K.D.; Sila, A.; Kilibarda, M.; Antonijević, O.; Glušica, L.; Dobermann, A.; Haefele,
S.M.; et al. African soil properties and nutrients mapped at 30 m spatial resolution using two-scale ensemble machine learning.
Sci. Rep. 2021, 11, 6130. [CrossRef]
11. Bouasria, A.; Namr, K.I.; Rahimi, A.; Ettachfini, E.M.; Rerhou, B. Evaluation of Landsat 8 image pansharpening in estimating soil
organic matter using multiple linear regression and artificial neural networks. Geo-Spat. Inf. Sci. 2022, 25, 353–364. [CrossRef]
12. Bouslihim, Y.; John, K.; Miftah, A.; Azmi, R.; Aboutayeb, R.; Bouasria, A.; Razouk, R.; Hssaini, L. The effect of covariates on Soil
Organic Matter and pH variability: A digital soil mapping approach using random forest model. Ann. GIS 2024, 30, 215–232.
[CrossRef]
13. Sayedain, S.A.; Maghsoudi, Y.; Eini-Zinab, S. Assessing the use of cross-orbit Sentinel-1 images in land cover clas-sification. Int. J.
Remote Sens. 2020, 41, 7801–7819. [CrossRef]
14. Urbina-Salazar, D.; Vaudour, E.; Baghdadi, N.; Ceschia, E.; Richer-de-Forges, A.C.; Lehmann, S.; Arrouays, D. Using sentinel-2
images for soil organic carbon content mapping in croplands of southwestern france. The usefulness of sentinel-1/2 derived
moisture maps and mismatches between sentinel images and sampling dates. Remote Sens. 2021, 13, 5115. [CrossRef]
15. Mponela, P.; Snapp, S.; Villamor, G.B.; Tamene, L.; Le, Q.B.; Borgemeister, C. Digital soil mapping of nitrogen, phosphorus,
potassium, organic carbon and their crop response thresholds in smallholder managed escarpments of Malawi. Appl. Geogr. 2020,
124, 102299. [CrossRef]
16. Flynn, T.; Rozanov, A.; Ellis, F.; de Clercq, W.; Clarke, C. Farm-scale digital soil mapping of soil classes in South Africa. S. Afr. J.
Plant Soil 2022, 39, 175–186. [CrossRef]
Remote Sens. 2024, 16, 1871 16 of 18

17. Gholizadeh, A.; Žižala, D.; Saberioon, M.; Borůvka, L. Soil organic carbon and texture retrieving and mapping using proximal,
airborne and Sentinel-2 spectral imaging. Remote Sens. Environ. 2018, 218, 89–103. [CrossRef]
18. Castaldi, F.; Chabrillat, S.; Don, A.; van Wesemael, B. Soil Organic Carbon Mapping Using LUCAS Topsoil Database and Sentinel-2
Data: An Approach to Reduce Soil Moisture and Crop Residue Effects. Remote Sens. 2019, 11, 2121. [CrossRef]
19. Castaldi, F.; Hueni, A.; Chabrillat, S.; Ward, K.; Buttafuoco, G.; Bomans, B.; Vreys, K.; Brell, M.; van Wesemael, B. Evaluating the
capability of the Sentinel 2 data for soil organic carbon prediction in croplands. ISPRS J. Photogramm. Remote Sens. 2019, 147,
267–282. [CrossRef]
20. Wang, S.; Zhou, M.; Zhuang, Q.; Guo, L. Prediction Potential of Remote Sensing-Related Variables in the Topsoil Organic Carbon
Density of Liaohekou Coastal Wetlands, Northeast China. Remote Sens. 2021, 13, 4106. [CrossRef]
21. Tripathi, A.; Tiwari, R.K. Utilisation of spaceborne C-band dual pol Sentinel-1 SAR data for simplified regres-sion-based soil
organic carbon estimation in Rupnagar, Punjab, India. Adv. Space Res. 2022, 69, 1786–1798. [CrossRef]
22. Izurieta, J.E.A.; Santillán, C.A.J.; Márquez, C.O.; García, V.J.; Rivera-Caicedo, J.P.; Van Wittenberghe, S.; Delegido, J.; Verrelst, J.
Improving the remote estimation of soil organic carbon in complex ecosystems with Sentinel-2 and GIS using Gaussian processes
regression. Plant Soil 2022, 479, 159–183. [CrossRef] [PubMed]
23. Shafizadeh-Moghadam, H.; Minaei, F.; Talebi-Khiyavi, H.; Xu, T.; Homaee, M. Synergetic use of multi-temporal Sentinel-1,
Sentinel-2, NDVI, and topographic factors for estimating soil organic carbon. Catena 2022, 212, 106077. [CrossRef]
24. Zhou, T.; Geng, Y.; Chen, J.; Pan, J.; Haase, D.; Lausch, A. High-resolution digital mapping of soil organic carbon and soil total
nitrogen using DEM derivatives, Sentinel-1 and Sentinel-2 data based on machine learning algorithms. Sci. Total Environ. 2020,
729, 138244. [CrossRef] [PubMed]
25. FAO. Standard Operating Procedure for Soil Organic Carbon Walkley-Black Method Titration and Colorimetric Method; Food & Agriculture
Organization: Rome, Italy, 2019.
26. Dahhani, S.; Raji, M.; Hakdaoui, M.; Lhissou, R. Land cover mapping using sentinel-1 time-series data and ma-chine-learning
classifiers in agricultural sub-saharan landscape. Remote Sens. 2022, 15, 65. [CrossRef]
27. Loew, A.; Mauser, W. Generation of geometrically and radiometrically terrain corrected SAR image products. Remote Sens.
Environ. 2007, 106, 337–349. [CrossRef]
28. Main-Knorn, M.; Pflug, B.; Louis, J.; Debaecker, V.; Müller-Wilm, U.; Gascon, F. Sen2Cor for sentinel-2. In Image and Signal
Processing for Remote Sensing XXIII; SPIE: Bellingham, WA, USA, 2017; Volume 10427, pp. 37–48.
29. Escadafal, R.; Girard, M.-C.; Courault, D. Munsell soil color and soil reflectance in the visible spectral bands of landsat MSS and
TM data. Remote Sens. Environ. 1989, 27, 37–46. [CrossRef]
30. Escadafal, R.; Belghith, A.; Ben Moussa, H. Indices spectraux pour la télédétection de la dégradation des milieux naturels en
Tunisie aride. In Proceedings of the 6th International Symposium on Physical Measurements and Signatures in Remote Sensing,
Val d’Isère, France, 17–21 January 1994; pp. 17–21.
31. Xu, H. Modification of normalised difference water index (NDWI) to enhance open water features in remotely sensed imagery.
Int. J. Remote Sens. 2006, 27, 3025–3033. [CrossRef]
32. Dash, J.; Curran, P.J. The MERIS terrestrial chlorophyll index. Int. J. Remote Sens. 2004, 25, 5403–5413. [CrossRef]
33. Bannari, A.; Morin, D.; Bonn, F.; Huete, A.R. A review of vegetation indices. Remote Sens. Rev. 1995, 13, 95–120. [CrossRef]
34. McFeeters, S.K. The use of the Normalized Difference Water Index (NDWI) in the delineation of open water features. Int. J.
Remote Sens. 1996, 17, 1425–1432. [CrossRef]
35. Huete, A.R. A soil-adjusted vegetation index (SAVI). Remote Sens. Environ. 1988, 25, 295–309. [CrossRef]
36. Darst, B.F.; Malecki, K.C.; Engelman, C.D. Using recursive feature elimination in random forest to account for correlated variables
in high dimensional data. BMC Genet. 2018, 19, 65. [CrossRef] [PubMed]
37. Breiman, L. Random forests. Mach. Learn. 2001, 45, 5–32. [CrossRef]
38. Bouslihim, Y.; Kharrou, M.H.; Miftah, A.; Attou, T.; Bouchaou, L.; Chehbouni, A. Comparing Pan-sharpened Landsat-9 and
Sentinel-2 for Land-Use Classification Using Machine Learning Classifiers. J. Geovisualization Spat. Anal. 2022, 6, 1–17. [CrossRef]
39. John, K.; Bouslihim, Y.; Bouasria, A.; Razouk, R.; Hssaini, L.; Isong, I.A.; M’Barek, S.A.; Ayito, E.O.; Ambrose-Igho, G. Assessing
the impact of sampling strategy in random forest-based predicting of soil nutrients: A study case from northern Morocco. Geocarto
Int. 2022, 37, 11209–11222. [CrossRef]
40. Bouasria, A.; Bouslihim, Y.; Gupta, S.; Taghizadeh-Mehrjardi, R.; Hengl, T. Predictive performance of machine learning model
with varying sampling designs, sample sizes, and spatial extents. Ecol. Inform. 2023, 78, 102294. [CrossRef]
41. Chen, T.; Guestrin, C. XGBoost: A scalable tree boosting system. In Proceedings of the 22nd ACM SIGKDD International
Conference on Knowledge Discovery and Data Mining, San Francisco, CA, USA, 13–17 August 2016; pp. 785–794.
42. Drucker, H.; Burges, C.J.; Kaufman, L.; Smola, A.; Vapnik, V. Support vector regression machines. In Proceedings of the Advances
in Neural Information Processing Systems 9, NIPS, Denver, CO, USA, 2–5 December 1996.
43. Bisong, E. Building Machine Learning and Deep Learning Models on Google Cloud Platform: A Comprehensive Guide for Beginners;
Apress: Berkeley, CA, USA, 2019; pp. 59–64.
44. Piñeiro, G.; Perelman, S.; Guerschman, J.P.; Paruelo, J.M. How to evaluate models: Observed vs. predicted or predicted vs.
observed? Ecol. Model. 2008, 216, 316–322. [CrossRef]
45. Smith, J.; Smith, P.; Addiscott, T. Quantitative methods to evaluate and compare soil organic matter (SOM) models. In Evaluation
of Soil Organic Matter Models: Using Existing Long-Term Datasets; Springer: Berlin/Heidelberg, Germany, 1996; pp. 181–199.
Remote Sens. 2024, 16, 1871 17 of 18

46. Castaldi, F.; Palombo, A.; Santini, F.; Pascucci, S.; Pignatti, S.; Casa, R. Evaluation of the potential of the current and forthcoming
multispectral and hyperspectral imagers to estimate soil texture and organic carbon. Remote Sens. Environ. 2016, 179, 54–65.
[CrossRef]
47. Pastor-Guzman, J.; Brown, L.; Morris, H.; Bourg, L.; Goryl, P.; Dransfeld, S.; Dash, J. The Sentinel-3 OLCI Terrestrial Chlorophyll
Index (OTCI): Algorithm Improvements, Spatiotemporal Consistency and Continuity with the MERIS Archive. Remote Sens. 2020,
12, 2652. [CrossRef]
48. Vani, V.; Mandla, V.R. Comparative study of NDVI and SAVI vegetation indices in Anantapur district semi-arid areas. Int. J. Civ.
Eng. Technol. 2017, 8, 559–566.
49. Brevik, E.C.; Calzolari, C.; Miller, B.A.; Pereira, P.; Kabala, C.; Baumgarten, A.; Jordán, A. Soil mapping, classification, and
pedologic modeling: History and future directions. Geoderma 2016, 264, 256–274. [CrossRef]
50. Ngatia, L.W.; Moriasi, D.; Grace, J.M., III; Fu, R.; Gardner, C.S.; Taylor, R.W. Land use change affects soil organic carbon: An
indicator of soil health. In Environmental Health; Books on Demand: Norderstedt, Germany, 2021.
51. Crapart, C.; Finstad, A.G.; Hessen, D.O.; Vogt, R.D.; Andersen, T. Spatial predictors and temporal forecast of total organic carbon
levels in boreal lakes. Sci. Total Environ. 2023, 870, 161676. [CrossRef] [PubMed]
52. Bian, Z.; Guo, X.; Wang, S.; Zhuang, Q.; Jin, X.; Wang, Q.; Jia, S. Applying statistical methods to map soil organic carbon of
agricultural lands in northeastern coastal areas of China. Arch. Agron. Soil Sci. 2019, 66, 532–544. [CrossRef]
53. Kaya, F.; Keshavarzi, A.; Francaviglia, R.; Kaplan, G.; Başayiğit, L.; Dedeoğlu, M. Assessing Machine Learning-Based Prediction
under Different Agricultural Practices for Digital Mapping of Soil Organic Carbon and Available Phosphorus. Agriculture 2022,
12, 1062. [CrossRef]
54. Nguyen, T.T.; Pham, T.D.; Nguyen, C.T.; Delfos, J.; Archibald, R.; Dang, K.B.; Hoang, N.B.; Guo, W.; Ngo, H.H. A novel intelligence
approach based active and ensemble learning for agricultural soil organic carbon prediction using multispectral and SAR data
fusion. Sci. Total Environ. 2022, 804, 150187. [CrossRef]
55. Wang, S.; Zhuang, Q.; Jin, X.; Yang, Z.; Liu, H. Predicting Soil Organic Carbon and Soil Nitrogen Stocks in Topsoil of Forest
Ecosystems in Northeastern China Using Remote Sensing Data. Remote Sens. 2020, 12, 1115. [CrossRef]
56. Wang, K.; Qi, Y.; Guo, W.; Zhang, J.; Chang, Q. Retrieval and Mapping of Soil Organic Carbon Using Sentinel-2A Spectral Images
from Bare Cropland in Autumn. Remote Sens. 2021, 13, 1072. [CrossRef]
57. Liu, T.; Zhang, H.; Shi, T. Modeling and Predictive Mapping of Soil Organic Carbon Density in a Small-Scale Area Using
Geographically Weighted Regression Kriging Approach. Sustainability 2020, 12, 9330. [CrossRef]
58. Sodango, T.H.; Sha, J.; Li, X.; Noszczyk, T.; Shang, J.; Aneseyee, A.B.; Bao, Z. Modeling the Spatial Dynamics of Soil Organic
Carbon Using Remotely-Sensed Predictors in Fuzhou City, China. Remote Sens. 2021, 13, 1682. [CrossRef]
59. Pei, T.; Qin, C.-Z.; Zhu, A.-X.; Yang, L.; Luo, M.; Li, B.; Zhou, C. Mapping soil organic matter using the topographic wetness index:
A comparative study based on different flow-direction algorithms and kriging methods. Ecol. Indic. 2010, 10, 610–619. [CrossRef]
60. Davidson, E.A.; Janssens, I.A. Temperature sensitivity of soil carbon decomposition and feedbacks to climate change. Nature 2006,
440, 165–173. [CrossRef] [PubMed]
61. Scharlemann, J.P.; Tanner, E.V.; Hiederer, R.; Kapos, V. Global soil carbon: Understanding and managing the largest terrestrial
carbon pool. Carbon Manag. 2014, 5, 81–91. [CrossRef]
62. Lu, W.; Lu, D.; Wang, G.; Wu, J.; Huang, J.; Li, G. Examining soil organic carbon distribution and dynamic change in a hickory
plantation region with Landsat and ancillary data. Catena 2018, 165, 576–589. [CrossRef]
63. He, T.; Wang, J.; Lin, Z.; Cheng, Y. Spectral features of soil organic matter. Geo-Spat. Inf. Sci. 2009, 12, 33–40. [CrossRef]
64. Hossain, M.Z. Farmer’s view on soil organic matter depletion and its management in Bangladesh. Nutr. Cycl. Agroecosyst. 2001,
61, 197–204. [CrossRef]
65. Saha, S.K.; Tiwari, S.K.; Kumar, S. Integrated use of hyperspectral remote sensing and geostatistics in spatial pre-diction of soil
organic carbon content. J. Indian Soc. Remote Sens. 2022, 50, 129–141. [CrossRef]
66. Zhang, H.; Wan, L.; Li, Y. Prediction of Soil Organic Carbon Content Using Sentinel-1/2 and Machine Learning Algorithms in
Swamp Wetlands in Northeast China. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 2023, 16, 5219–5230. [CrossRef]
67. Fathololoumi, S.; Vaezi, A.R.; Alavipanah, S.K.; Ghorbani, A.; Saurette, D.; Biswas, A. Improved digital soil mapping with
multitemporal remotely sensed satellite data fusion: A case study in Iran. Sci. Total Environ. 2020, 721, 137703. [CrossRef]
[PubMed]
68. Wang, L.; Zhou, Y. Combining Multitemporal Sentinel-2A Spectral Imaging and Random Forest to Improve the Accuracy of Soil
Organic Matter Estimates in the Plough Layer for Cultivated Land. Agriculture 2022, 13, 8. [CrossRef]
69. Zhou, T.; Geng, Y.; Chen, J.; Liu, M.; Haase, D.; Lausch, A. Mapping soil organic carbon content using multi-source remote
sensing variables in the Heihe River Basin in China. Ecol. Indic. 2020, 114, 106288. [CrossRef]
70. Zhou, T.; Geng, Y.; Ji, C.; Xu, X.; Wang, H.; Pan, J.; Bumberger, J.; Haase, D.; Lausch, A. Prediction of soil organic carbon and the C:
N ratio on a national scale using machine learning and satellite data: A comparison between Sentinel-2, Sentinel-3 and Landsat-8
images. Sci. Total Environ. 2021, 755, 142661. [CrossRef] [PubMed]
71. Li, X.; McCarty, G.W.; Karlen, D.L.; Cambardella, C.A. Topographic metric predictions of soil redistribution and organic carbon in
Iowa cropland fields. Catena 2018, 160, 222–232. [CrossRef]
72. Gibson, A.; Hancock, G.; Bretreger, D.; Cox, T.; Hughes, J.; Kunkel, V. Assessing digital elevation model resolution for soil organic
carbon prediction. Geoderma 2021, 398, 115106. [CrossRef]
Remote Sens. 2024, 16, 1871 18 of 18

73. Duarte, E.; Zagal, E.; Barrera, J.A.; Dube, F.; Casco, F.; Hernández, A.J. Digital mapping of soil organic carbon stocks in the forest
lands of Dominican Republic. Eur. J. Remote Sens. 2022, 55, 213–231. [CrossRef]
74. Siewert, M.B. High-resolution digital mapping of soil organic carbon in permafrost terrain using machine learning: A case study
in a sub-Arctic peatland environment. Biogeosciences 2018, 15, 1663–1682. [CrossRef]
75. Pouladi, N.; Møller, A.B.; Tabatabai, S.; Greve, M.H. Mapping soil organic matter contents at field level with Cubist, Random
Forest and kriging. Geoderma 2019, 342, 85–92. [CrossRef]
76. Tajik, S.; Ayoubi, S.; Zeraatpisheh, M. Digital mapping of soil organic carbon using ensemble learning model in Mollisols of
Hyrcanian forests, northern Iran. Geoderma Reg. 2020, 20, e00256. [CrossRef]
77. Pullanagari, R.R.; Cavalli, D. Advances and applications of multivariate statistics and soil-crop sensing to improve nutrient use
efficiency and monitor carbon cycling. Nutr. Cycl. Agroecosyst. 2023, 127, 97–99. [CrossRef]

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual
author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to
people or property resulting from any ideas, methods, instructions or products referred to in the content.