Harnessing machine learning and geospatial technologies for precise soil erodibility mapping and prediction

Abstract

Soil erosion threatens fertility and sustainability, with soil erodibility influencing erosion rates based on physical and chemical properties. This study aimed to estimate soil erodibility for various land uses using the K-factor from the Wischmeier equation, assess indicators such as the structural stability index, clay ratio, and dispersion ratio, and develop a predictive model for erosion risk using artificial neural networks (ANN) and geospatial technologies. High-resolution spatial maps of erosion risk were created to inform land management and conservation efforts. An ANN model in MATLAB R2024a predicted soil erodibility as well as indicators such as the dispersion ratio, crust formation, and clay ratio. Statistical analyses, including principal component analysis (PCA) and correlation assessment, were performed with OriginPro 2021b to explore relationships between soil properties. Spatial maps of observed and predicted erodibility were created using ArcGIS 10.7.1. Results showed that erodibility values ranged from 0.023 to 0.152 t·ha·hr·MJ⁻¹·mm⁻¹ for the observed data and 0.026 to 0.148 t·ha·hr·MJ⁻¹·mm⁻¹ for the predicted values. For different land uses, it included 0.09513t·ha·hr·MJ⁻¹·mm ¹ for cultivated land, 0.060796 t·ha· hr·MJ ¹ · mm ¹ for forest land, and 0.092685 t·ha·hr·MJ⁻¹·mm⁻¹ for pasture land. The ANN model demonstrated high accuracy, with R-values of 0.999 for soil erodibility, 0.996 for the structural stability index (SSI), 0.995 for the clay ratio (CR), and 0.904 for the dispersion ratio (DR). This study effectively combines machine learning and geospatial technologies to predict and map soil erodibility, providing insights for erosion control and sustainable land management.

Full text

Vol.:(0123456789)
Environmental Earth Sciences (2025) 84:291
https://doi.org/10.1007/s12665-025-12270-9
ORIGINAL ARTICLE
Harnessing machine learning andgeospatial technologies forprecise
soil erodibility mapping andprediction
WuduAbiye1,2 · EndalamawDessieAlebachew1,3· OrhanDengiz1
Received: 5 February 2025 / Accepted: 18 April 2025 / Published online: 19 May 2025
© The Author(s) 2025
Abstract
Soil erosion threatens fertility and sustainability, with soil erodibility influencing erosion rates based on physical and
chemical properties. This study aimed to estimate soil erodibility for various land uses using the K-factor from the Wis-
chmeier equation, assess indicators such as the structural stability index, clay ratio, and dispersion ratio, and develop a
predictive model for erosion risk using artificial neural networks (ANN) and geospatial technologies. High-resolution
spatial maps of erosion risk were created to inform land management and conservation efforts. An ANN model in MAT-
LAB R2024a predicted soil erodibility as well as indicators such as the dispersion ratio, crust formation, and clay ratio.
Statistical analyses, including principal component analysis (PCA) and correlation assessment, were performed with
OriginPro 2021b to explore relationships between soil properties. Spatial maps of observed and predicted erodibility
were created using ArcGIS 10.7.1. Results showed that erodibility values ranged from 0.023 to 0.152 t·ha·hr·MJ-1·mm-1
for the observed data and 0.026 to 0.148 t·ha·hr·MJ-1·mm-1 for the predicted values. For different land uses, it included
0.09513t·ha·hr·MJ-1·mm 1 for cultivated land, 0.060796 t·ha· hr·MJ 1 · mm 1 for forest land, and 0.092685 t·ha·hr·MJ-
1·mm-1 for pasture land. The ANN model demonstrated high accuracy, with R-values of 0.999 for soil erodibility, 0.996
for the structural stability index (SSI), 0.995 for the clay ratio (CR), and 0.904 for the dispersion ratio (DR). This study
effectively combines machine learning and geospatial technologies to predict and map soil erodibility, providing insights
for erosion control and sustainable land management.
Highlights
• Soil erodibility was accurately estimated for cultivated, forest, and pasture lands.
• Cultivated land has the highest value of soil erodibility (0.09513 t·ha·hr·MJ-1·mm-1).
• ANN models achieved high predictive accuracy (R = 0.99 for soil erodibility).
• PCA and correlation analysis revealed key relationships between soil properties affecting erodibility.
• The study provides valuable insights for improving soil conservation and sustainable land use planning.
Keywords ANN· Geospatial technologies· Land use types· RUSLE and Soil erosion risk
Introduction
Soil erosion is a critical environmental challenge that threat-
ens agricultural productivity, water quality, and the long-
term sustainability of ecosystems worldwide (Alebachew
etal. 2025). It results from the detachment and movement
of soil particles by water, wind, or tillage, causing the deg-
radation of fertile land and the sedimentation of water bod-
ies (Mirzaee and Ghorbani-Dashtaki 2018). The severity of
soil erosion is globally evident in its far-reaching impacts
on food security and environmental sustainability (Pimen-
tel 2006). The accelerated rate of soil erosion, driven by
* Wudu Abiye
wuduabiye@gmail.com
1 Faculty ofAgriculture, Department ofSoil Science andPlant
Nutrition, Ondokuz Mayıs University, Samsun, Turkey
2 Soil andWater Research, Amhara Agricultural Research
Institute (ARARI), Bahirdar, Ethiopia
3 Department ofSoil Resources andWatershed Management,
Hawassa University (HU), Wondo Genet College ofForestry
andNatural Resources, Hawassa, Ethiopia
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Environmental Earth Sciences (2025) 84:291291 Page 2 of 28
unsustainable land management practices and climate
change, has intensified land degradation, reduced soil fer-
tility, diminished arable land, and increased vulnerability to
natural disasters (Abiye etal.2023). One of the key factors
influencing soil erosion is soil erodibility, which reflects the
inherent susceptibility of soil to erosion and varies signifi-
cantly with land use types and soil properties. The precise
prediction of soil loss at the regional scale is paramount for
enhancing the understanding of erosion processes and for
the formulation of effective conservation practices (Mirzaee
and Ghorbani-Dashtaki 2021). In the context of the Black
Sea region, land use significantly influences soil erodibility.
Agricultural lands—particularly those subjected to intensive
cultivation—often exhibit higher erodibility due to frequent
soil disturbance and the removal of protective vegetation
cover (Luo etal. 2022).
Conversely, forested areas typically show lower erodibil-
ity due to the protective canopy and the stabilizing effect
of root systems (Zhang etal.2024). Recent findings sug-
gest that K-factor values (soil erodibility factor from the
Wischmeier equation) can vary markedly among land use
types, ranging from as low as 0.1 for undisturbed forests to
as high as 0.6 for intensively cultivated croplands (Taleshian
Jeloudar etal. 2018).
In addition to land use, soil health indicators such as the
soil structural stability index, dispersion ratio, and clay ratio
are closely linked to soil erodibility. These indices influ-
ence soil texture, aggregation, permeability, and resistance
to erosion. Organic matter, in particular, plays a pivotal role
in maintaining soil structure by promoting aggregation and
stability, thereby enhancing resistance to erosion (Liang
etal. 2022; Archibong etal. 2020; Özdemir & Gülser 2017;
Prout etal. 2021).
Soil erodibility is governed by several soil properties,
including structure, texture, organic matter content, and
permeability (Wischmeier & Mannering 1969). Soils with
poor structure, high silt content, low organic matter, and
poor permeability are more prone to erosion. The K-factor,
as a key component of the Universal Soil Loss Equation
(USLE), quantifies the erodibility of soil based on empirical
relationships. Understanding the spatial distribution of the
K-factor is essential for identifying erosion-prone areas and
prioritizing soil conservation efforts.
In recent years, both classical statistical and geostatistical
methods have been applied to assess the spatial variability of
the K-factor across diverse landscapes (Barchia etal.2023;
Zhang etal. 2011). Furthermore, machine learning tech-
niques, mainly Artificial Neural Networks (ANN), have
been increasingly utilized to model and predict soil erosion
due to their ability to handle complex, nonlinear interactions
among variables (Alzubaidi etal. 2021). The integration of
artificial neural networks (ANN) with geospatial technolo-
gies has enhanced the accuracy and resolution of spatial
predictions, enabling the generation of detailed erosion risk
maps (Woldemariam etal. 2018; Senanayake etal. 2020).
Despite growing efforts to understand soil erosion, a
knowledge gap remains regarding how soil erodibility var-
ies across different land-use types and how advanced mod-
eling techniques can enhance the accuracy of erosion risk
prediction. A better understanding of these variations and
the spatial dynamics of the K-factor is crucial for develop-
ing targeted and effective soil conservation strategies. The
objectives of this study were:(1)To estimate soil erodibil-
ity across different land use types in the study area using
the K-factor derived from the Wischmeier equation, (2).To
evaluate additional soil erodibility indicators, such as the
structural stability index, dispersion ratio, and clay ratio;
(3)To develop a predictive model for soil erosion risk using
Artificial Neural Networks (ANN); and (4) To generate high-
resolution spatial maps of erosion-prone areas by integrat-
ing ANN with geospatial technologies. This study presents
a novel approach that combines artificial neural networks
(ANN) with geospatial technologies to create a predictive
model for soil erosion risk. It also evaluates soil erodibility
indicators across different land use types, providing insights
for effective soil conservation strategies.
Material andmethods
Study area, sample anddataset
The study area is in a micro-watershed along the southern
coast of the Black Sea, in Samsun, Turkiye. Geographi-
cally, it extends between 41°20′15"and 41°21′31"North lat-
itude and 36°11′00"East longitude (Fig.1), with elevations
ranging from 127.6 to 565.2 m above sea level. The mean
annual precipitation is approximately 735 mm, resulting in
relatively humid conditions, while the mean annual tem-
perature is around 14 °C. Seasonal temperature extremes
range from approximately 6°C to 22 °C (Alebachew and
Dengiz 2024).
According to the World Reference Base (WRB) (2022) soil
classification system, the predominant soil types in the area
are cambisols, luvisols, and Leptosols, which contribute to
the region’s diverse soil characteristics https:// soilg rids. org/
For this study, 54 soil samples were collected from
three distinct land-use types within the watershed: cul-
tivated land, forest land, and pasture land. The sampling
sites were identified and distributed using Google Earth
during fieldwork. Several factors, including land use
type, landscape variation, and slope position, were con-
sidered to capture the variability within the study area.
All analyses in this study were conducted using standard
methodologies appropriate for each soil property. Soil
texture was determined using the hydrometer method
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Environmental Earth Sciences (2025) 84:291 Page 3 of 28 291
(Bouyoucos 1951). Aggregate stability was assessed
through the wet sieving method (Kemper and Rosenau
1986). Organic matter content was measured using the
Walkley–Black method (1934). In the micro-watershed,
soil samples were collected from designated sampling
points to analyze various physical and chemical proper-
ties, including sand, silt, clay fractions, aggregate stabil-
ity, dispersion ratio, and organic carbon content. These
properties were assessed using the Revised Universal
Soil Loss Equation (RUSLE) model to determine soil
erodibility. An artificial neural network (ANN) model
was developed to predict soil erodibility, incorporating
seven soil properties as input variables. The network
architecture comprised a hidden layer with 10 neurons
and an output layer representing the soil erodibility fac-
tor (Fig.2). The ANN utilized the Levenberg–Marquardt
(LM) algorithm, with the feed-forward backpropagation
technique applied to optimize the predictive performance.
Determination ofsoil erodibility factor (K‑factor)
andits indicators
The RUSLE The model was used to determine the sen-
sitivity of soil loss. The soil erodibility factor (K) in the
RUSLE quantifies the susceptibility of soil particles to
detachment and transport by rainfall and runoff. This fac-
tor is influenced by several soil properties, including tex-
ture, organic matter content, structure, and permeability
(Renard 1997) (Eq.1). Soils with high silt content, low
organic matter, and poor structure tend to be more erod-
ible. The K factor is crucial for predicting soil loss and
implementing effective soil conservation practices, as it
helps identify soils that are more vulnerable to erosion and
require more intensive management (Wischmeier, 1969).
(1)
K=
(2.1X1O
−4
)(12 −OM)M1.14 +3.25(S−2)+2.5(P−3
)
100
Fig. 1 Location map of the study area
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Environmental Earth Sciences (2025) 84:291291 Page 4 of 28
OM = Organic matter content (%) of the top 20 cm of soil
S = Soil structure code, with values based on soil
aggregation (1 = very fine granular, 2 = fine granular,
3 = medium or coarse granular, 4 = blocky, platy, or mas-
sive) and P Permeability class of the soil (ranges from 1 =
rapid to 6 = very slow).
The permeability and structural code of the soil were
assigned based on various literature sources. For soil
with a sand textural class, the permeability is 1, and the
structure code is 1; for silt, the permeability is 5, and the
structure code is 2; and for clay, the permeability is 6, and
the structure code is 4(Table1).
Clay ratio (CR)
The clay ratio is an estimation of the amount of binding
agent clay that tightly binds the soil particles, making it
difficult for external forces to detach the particles in the
presence of a high clay content. CR is inversely related
to soil erodibility (ErdoğanYükse and Yavuz 2024).
Earlier studies stated the correlations among soil prop-
erties, which reveals that the modified clay ratio (MCR)
is also another index for erodibility (Agbai etal.2022)
(2)
M
=
(%silt + %ver yf inesand)(100 − clay%)
100
(3)
CR = %
sand
+%silt
%clay
Fig. 2 Flow chart of the study
Table 1 Assignment of soil
structural and permeability code
based on textural class
Texture Class Soil description Permeability Structure Reference
Sand Fast and very fast 1 1
Loamy sand Moderate fast 2 1
Sandy loam Moderate fast 2 1 Panagos
etal.2014
Loam Moderate fast 2 2
Silt loam Moderate 3 2 Efthimiou2020
Morgan, 2001
Silt Slow 5 2 Baillie 2001
Sandy clay loam Moderate low 4 2
Clay loam Moderate low 4 3
Silty clay loam Slow 5 3
Sandy clay Slow 5 2
Silty clay Very slow 6 4
Clay Very slow 6 4
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Environmental Earth Sciences (2025) 84:291 Page 5 of 28 291
Critical level oforganic matter
A critical level of organic matter (CLOM) is also an
indicator of erosion susceptibility (Parwadaa and Van
Tol2017). It is related to soil aggregate formation capabil-
ity, which offers resistance to soil erosion. If the value of
CLOM is less than 5%, it is indicated that the soil loses its
structure and becomes highly susceptible to erosion. Soil
is said to be moderately susceptible to erosion if the value
lies between 5 and 7%. A CLOM value of more than 9%
indicates that the soil is stable and offers more resistance
to erosion (Kusre etal. 2018).
Aggregate stability (AS)
High aggregate stability enhances soil structure and reduces
susceptibility to erosion. In terms of soil erodibility, soils
with higher aggregate stability are less prone to erosion
because the stable aggregates can better withstand the forces
that cause soil particles to detach and be transported (Ncii-
zah and Wakindiki2015). Thus, low aggregate stability typi-
cally leads to higher soil erodibility, making the soil more
vulnerable to erosion. The Yoder type was evaluated using
the wet sieve method (Khan etal. 2007).
Dispersion ratio (DR)
DR is calculated as the percentage of silt and clay particles
that remain in suspension without the use of a dispersing
agent, compared to the total silt and clay content measured
during mechanical analysis (where a dispersing agent is
used). This ratio reflects the degree of particle aggregation
in the soil. A higher dispersion ratio indicates weaker soil
aggregates, meaning that the soil particles are more prone
to dispersion and, thus, more susceptible to erosion. Con-
versely, a lower DR suggests more substantial soil aggre-
gates and reduced erodibility (Panda 2022).
(4)
MCR = %sand +%silt
%clay +%OM
(5)
CLOM = OM
clay + silt
(6)
As(%) = Mass of aggeregate −Mass of sand
Mass of soil −Mass of sand
×
100
(7)
DR(%) = (a
b)
∗
100
where a is the percentage of silt plus clay in suspension, b
is the percentage of silt plus clay dispersed with a chemical
agent (Bryan 1968).
Crust formation (CF)
Refers to the development of a hard, compacted surface layer on
the soil, typically caused by the impact of raindrops or irrigation
water. This process seals the soil surface, reducing its permeabil-
ity and increasing the likelihood of runoff. Crust formation can
significantly impact soil erodibility by limiting water infiltration
and promoting the accumulation of surface water, which in turn
leads to higher erosion rates. Soils prone to crust formation are
generally more susceptible to erosion, as the crust weakens the
soil structure and enhances the detachment and transport of soil
particles by water (Sumner and Miller 1992).
where CF; Crust formation, OM; Organic matter.
Structure stability index (SSI)
The SSI (%) is a measure of soil's resistance to disintegration
and erosion, indicating the stability of soil aggregates (Usta
2022). It can be calculated based on hydrometer measure-
ments using the following equation;
where:
• Ʃn refers to the sum of silt (%) and clay (%) fractions
obtained through mechanical analysis (using a dispersing
agent to break down aggregates).
• Ʃb refers to the sum of silt (%) and clay (%) fractions that
naturally disperse from soil aggregates into suspension
(without the use of a dispersing agent).
A higher SSI value indicates greater aggregate stability,
meaning the soil is more resistant to erosion. Conversely, a
lower SSI suggests weaker aggregates and higher suscepti-
bility to soil erosion.
Artificial neural network (ANN)
Machine learning (ML) is a transformative tool in research,
enabling the development of machine learning models that can
analyze vast amounts of data with minimal human intervention.
One of the most powerful techniques in machine learning is the
(8)
CF
=
(OM(%)×100)
%Clay + %Silt
(9)
SSI
=

n−b

n
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Environmental Earth Sciences (2025) 84:291291 Page 6 of 28
Artificial Neural Network (ANN), which is modelled after the
human brain's structure and function (Goodfellow etal.2016).
Artificial Neural Networks (ANNs) consist of interconnected lay-
ers of nodes, or"neurons,"that process input data, learn patterns,
and make predictions. The importance of ANNs in research pre-
diction lies in their ability to manage complex, nonlinear relation-
ships in data that traditional statistical methods often struggle to
model (Li and Pan 2020). ANNs can be trained on large datasets
to identify patterns, trends, and associations that are not imme-
diately apparent. This capability is especially valuable in fields
such as climate science, soil hydrology, and environmental man-
agement, where systems are inherently complex and influenced
by numerous interacting factors. By applying artificial neural
networks (ANNs), researchers can improve the accuracy and reli-
ability of their predictions, leading to more informed decisions
and effective strategies for addressing challenges such as climate
change, resource management, and environmental sustainability
(Li and Pan2020). To predict soil erodibility, an artificial neural
network (ANN) model was developed, incorporating seven soil
properties as input variables (Fig.3). The network architecture
consisted of a hidden layer with 10 neurons and an output layer
representing the soil erodibility factor. The ANN utilized the
Levenberg–Marquardt (LM) algorithm, with the feed-forward
backpropagation technique applied to optimize the predictive
performance.
Interpolation techniques andstatistical methods
In this study, we used descriptive statistics to summarize
and simplify the dataset, providing insights into patterns,
central tendencies, and variability. Next, we used Pearson
correlation to evaluate the strength and direction of linear
relationships between pairs of continuous variables (Rah-
nenführe etal.2023). Finally, we employed a biplot based
on principal component analysis (PCA) to visually represent
relationships between variables and observations, facilitat-
ing an intuitive understanding of complex multivariate rela-
tionships, patterns, and correlations. Various interpolation
methods were used, including Inverse Distance Weighting
(IDW), Radial Basis Functions (RBF) such as Thin Plate
Spline, Completely Regularized Spline, Spline with Ten-
sion, and Kriging (Ordinary, Simple, Universal), which
were assessed to generate distribution maps using ArcGIS
10.7.1 software. The most suitable interpolation model was
determined based on the Root Mean Square Error (RMSE)
evaluation. To assess model uncertainty, the variance of
predictions generated for each model for soil properties
was calculated (Malone etal., 2017; Sharififar 2022). The
square root of the variance (standard deviation) was then
computed (Sharififar 2022). The models'important variables
were calculated as FAO (2022) described. The predictions
were performed using Matlab® R2024a (MathWorks, 2023).
Results anddiscussion
Soil anderodibility properties
The soil properties examined in this study reveal key find-
ings that both align with and differ from previous research.
The average organic matter (OM) content of 2. 2.3% falls
below the critical threshold of 3–5%, essential for sustaining
soil fertility and structure. This finding corroborates earlier
studies emphasising organic matter's role in improving soil
health, especially in degraded and semi- arid regions where
organic input is vital for productivity (Lal 2021). The low
organic matter (OM) content here indicates a need for organic
amendments to enhance soil fertility, consistent with conclu-
sions drawn in other studies. The average clay content of
38. A 9% clay content demonstrates good moisture retention
and nutrient-holding capacity, aligning with the conclusions
of Carter and Gregorich (2007), who noted that soils with
higher clay contents are more stable in terms of moisture
retention and erosion resistance. However, the substantial
variation in the modified clay ratio (MCR) and clay ratio
(CR) with coefficients of variation exceeding 100% sug-
gests considerable differences in structural stability across
the study area. This contrasts with Amsili etal. (2021), who
found more consistent structural stability in less heterogene-
ous landscapes. This level of variability may reflect local-
ized soil- forming processes or land- use practices affecting
soil texture and stability. The average dispersion ratio (DR)
of 22. 8% indicates moderate dispersion, which may pose
risks of soil instability and erosion, particularly in areas with
high clay content. This aligns with findings by Young (1972),
who noted similar levels of DR in soils vulnerable to surface
sealing and erosion under varied rainfall. Furthermore, the
mean aggregate stability (Agg.St) of 58. 2% surpasses values
reported in comparable soils by Sharma etal. (2017), sug-
gesting better erosion resistance and improved infiltration
capacities in the soils examined. Crust formation (CF) with
an average value of 3. 3.9% points to potential surface seal-
ing risks that could adversely affect seedling emergence and
water infiltration. This finding aligns with earlier research
by Liu etal. (2024), which indicates that crust formation
correlates with the ratios of sand and silt, potentially reduc-
ing water infiltration in agricultural soils. Considering the
variability in soil properties like CR and DR, site- specific
management practices, such as organic matter incorporation
and erosion control methods, may be essential to enhance soil
structure and functionality (Table2).
Pearson correlation ofsoil properties
Analyzing soil properties and their correlation with the
soil erodibility factor reveals several significant findings.
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Environmental Earth Sciences (2025) 84:291 Page 7 of 28 291
Positive correlations were observed between soil erodibil-
ity and various factors, including clay content (r = 0.8), silt
content (r = 0.56), soil permeability code (r = 0.89), soil
structure code (r = 0.91), and crust formation (r = 0.31)
(Fig.4). These results are consistent with the established
scientific understanding that finer particles, such as clay
and silt, contribute to lower permeability and poorer soil
structure, increasing soil erodibility. Clay's high correlation
suggests its significant role in influencing soil erodibility,
though its effects may vary depending on soil structure
and aggregation. The strong positive correlation with soil
permeability suggests that under certain conditions, such
as soil saturation or intense rainfall, even permeable soils
can experience increased erosion due to enhanced runoff.
The correlation with soil structure aligns with the notion
that less cohesive and poorly aggregated soils are more
susceptible to erosion.
Conversely, negative correlations were found between
soil erodibility and sand percentage (r = −0.83), CR
(r = −0.66), MCR (r = −0.67), critical level of organic mat-
ter (r = −0.52), soil structural index (r = −0.59), and dis-
persion ratio (r = −0.52) (Fig.4). These findings reflect the
well-documented effect of these properties in reducing soil
erodibility. Higher sand content, as indicated by the strong
negative correlation, is associated with lower erodibility,
consistent with the notion that sandier soils are less prone
to erosion due to their larger particle size and lower cohe-
sion. The negative correlations with clay ratio, modified clay
ratio, and organic matter support the idea that improved soil
structure and increased organic matter contribute to reduced
erodibility. Likewise, the negative relationship between the
soil structural index and the dispersion ratio confirms that
enhanced soil aggregation and stability are associated with
lower susceptibility to erosion levels.
Principal component analysis (PCA)
The Principal Component Analysis (PCA) biplot illustrates
the distribution and influence of several variables on the first
two principal components, PC1 and PC2, which together
explain a substantial portion of the variance in the data-
set. Specifically, PC1 accounts for 45.66% of the total vari-
ance, while PC2 explains an additional 25.78%, resulting
in a combined explanation of 71.44% of the total variation
(Fig.5). This two-dimensional representation effectively
captures the underlying structure of the data by summarizing
the most influential variables. PC1 is primarily influenced
by variables such as OM and sand, as indicated by their
longer vectors. These variables contribute significantly to
the variance along PC1, suggesting that soil organic mat-
ter and sand content are key differentiating factors among
the samples. These variables are positively correlated, as
evidenced by their similar vector directions, indicating that
higher levels of OM are associated with increased sand con-
tent. In contrast, Coarse Fraction (CF), Permeability, and
Silt show a negative association with PC1, meaning that
as organic matter and sand content increase, the values for
CF, Permeability, and Silt decrease. This inverse relation-
ship highlights the opposing effects of these variables on
the principal components. Similarly, PC2 is influenced by a
different set of variables, with CF, Aggregate Stability (Agg.
St), and Permeability being the most influential. The vectors
for CF and Agg.St indicate a strong correlation with PC2,
contributing to the vertical distribution of the samples in the
plot. The association between CF and Agg.St suggests that
soil structural stability and particle size play substantial roles
in explaining the variability along this axis.
The relationships between the variables are evident in
the direction of their vectors. For instance, variables such
as OM and Sand are positively correlated, while CF, Per-
meability, and Silt exhibit negative correlations with them.
These interrelationships demonstrate the opposing effects
of organic matter and sand content compared to CF, Per-
meability, and Silt. Furthermore, CF and Agg. St distinctly
influence the data along PC2, differentiating the soil sam-
ples based on their structural and textural properties. The
PCA biplot effectively summarizes the most influential
variables contributing to the dataset's variation. OM and
Sand are the dominant contributors to PC1, highlighting
their importance in explaining the soil's organic content
and texture. Meanwhile, CF and Agg. St are prominent in
PC2, indicating their significant roles in describing soil
structural stability and coarser fractions. Together, these
variables capture the essential characteristics of the soil
samples, providing valuable insights into their physical and
compositional variability.
Fig. 3 Single hidden layer
network model
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Environmental Earth Sciences (2025) 84:291291 Page 8 of 28
Principal Component Analysis (PCA) reveals that
organic matter (OM) and clay content are the primary
variables explaining variability in the soil dataset. OM
accounts for 55.28% of the total variance, indicating
its primary influence on soil properties, such as struc-
ture, moisture retention, and erosion risk. Clay adds
20.49%, raising the total explained variance to 75.76%.
This underscores OM and clay as critical factors in soil
behavior analysis. Silt and sand contribute 9.43% and
8.30%, respectively, accounting for a cumulative vari-
ance of 17.73%. Aggregate stability and permeability add
minimal variance (5.18% and 1.33%) (Table3), while soil
structure contributes none. These findings emphasise the
prioritization of OM and clay in modeling efforts, while
less influential variables, especially soil structure, can be
excluded without a significant impact.
Table 2 Descriptive statistical
value of soil properties
OM organic matter, Agg.St Aggregate stability, CR clay ratio, MCR CLOM: critical level of organic matter,
SSI Soil structure index, DR dispersion ratio, Kurt Kurtosis, Skew Skewness
Soil Properties Mean S.D Min Max Range CV Skew Kurt
OM% 2.3 1.0 0.2 4.4 4.2 42.7 0.0 −0.6
Clay% 38.9 13.9 6.2 70.1 63.9 35.7 0.2 0.0
Silt% 19.6 6.6 2.4 36.2 33.7 33.8 −0.2 0.3
Sand% 41.6 17.9 7.4 85.6 78.2 43.0 0.1 −0.2
Agg.St% 58.2 14.3 15.5 84.7 69.2 24.6 −1.0 1.1
Permeability % 4.4 1.0 2.0 6.0 4.0 21.9 0.3 0.2
Structure 2.9 0.9 1.0 4.0 3.0 32.9 −0.1 −1.2
CR% 2.2 2.5 0.4 15.2 14.8 112.1 4.1 18.7
MCR% 2.0 2.1 0.4 13.0 12.5 104.3 4.0 17.9
CLOM% 0.0 0.0 0.0 0.1 0.1 49.8 0.2 −0.9
SSI% 6.4 3.4 0.4 15.2 14.8 53.5 0.6 −0.1
DR% 22.8 8.5 10.0 48.9 38.8 37.4 1.2 1.4
CF% 3.9 1.7 0.5 6.8 6.3 42.8 −0.2 −0.9
K 0.1 0.0 0.0 0.2 0.2 55.5 0.2 −0.3
Fig. 4 Correlation between soil
properties
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Environmental Earth Sciences (2025) 84:291 Page 9 of 28 291
The land use and land cover distribution in the study area
have significant implications for soil erodibility, particularly
in cultivated land, which occupies nearly 40% of the total
area (Table4). Agricultural practices, such as ploughing and
tilling, disturb soil structure, increase surface runoff, and
expose the soil to erosion. This is supported by several stud-
ies that have found cultivated lands more susceptible to ero-
sion due to the breakdown of soil aggregates and a reduction
in organic matter content (Lal 2001; Pimentel etal. 1995).
A higher dispersion ratio, which measures the extent to
which soil particles disperse in water, is often linked to cul-
tivated lands, further indicating weakened soil stability and
increased erodibility (Bryan 1968). In contrast, forest land,
which covers 50.02% of the study area, naturally protects
against soil erosion.
The forest canopy mitigates the impact of rainfall, while
the root systems enhance soil cohesion, improving aggregate
stability and reducing the dispersion ratio (Morgan 2005).
Forested soils tend to have higher organic matter content,
which fosters better soil structure and reduces erosion risks.
Studies have consistently shown that forests significantly
reduce soil erosion compared to agricultural or degraded
lands (Wang etal.2018; Shi etal. 2002).
Pasture land, comprising 8.98% of the total area, presents
a moderate risk of soil erosion (Fig.6). Overgrazing can lead
to soil compaction and reduced vegetation cover, both of
which can increase surface runoff and soil erosion (Teague
etal. 2011). However, Well-managed pastures tend to have
lower erosion rates due to vegetative cover stabilizing soil
and reducing surface runoff (Franzluebbers 2002). Water
bodies, covering 1.03% of the area, serve as accumulation
zones for sediments eroded from adjacent land. While not
directly influencing soil erosion, the presence of water bod-
ies may reflect the sediment transport and deposition pro-
cesses within the watershed, as noted by various sediment
yield studies (Walling 1983).
Soil erodibility factor inrelation toLULC
This research evaluates the actual and predicted values
of important soil properties and erosion indicators across
three land use types and land cover (LULC): cultivated
land, forest, and pasture. For the soil erodibility factor
(K), the actual values recorded for cultivated land (0.10 t
h MJ − 1 mm − 1), forest (0.06 t h MJ − 1 mm − 1), and
pasture (0.09 t h MJ − 1 mm − 1) align closely with the
model's predictions (Table5) (Chen etal. 2023). This find-
ing is consistent with earlier findings that underline bet-
ter structural integrity in forest soils, which often exhibit
greater resilience thanks to the presence of organic matter
and root structures that facilitate soil aggregation (Cerdà,
2000). In contrast, cultivated lands show comparatively
lower stability (56.87% actual versus 56.76% predicted),
reinforcing prior research findings that soil disturbance
from tillage and other agricultural practices diminishes
aggregate stability.
Fig. 5 Principal component
analysis (PCA) of soil proper-
ties
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Environmental Earth Sciences (2025) 84:291291 Page 10 of 28
The model effectively predicts the dispersion ratio (DR)
and crust formation (CF). Forest soils have slightly higher
DR (23.45 actual vs. 23.41 predicted) and CF (4.12 actual
vs. 4.16 predicted) compared to cultivated and pasture soils.
This aligns with previous studies indicating that forest soils
resist crusting and dispersion owing to their higher organic
matter content, though they remain vulnerable to heavy
rainfall (Chen and Duan 2015). Conversely, cultivated soils
exhibiting lower DR and CF values may benefit from strate-
gies like no-till farming to help combat soil crusting and
enhance water infiltration. The clay ratio (CR) and soil struc-
ture index (SSI) also emphasise superior structural health
in forest soils. They display the highest CR (2.97 actual vs.
3.11 predicted) and SSI (8.88 actual vs. 8.51 predicted),
indicating that forested areas preserve a more cohesive
soil structure due to less disturbance and a steady influx of
organic matter (Smith etal.2018). In contrast, cultivated
lands, with lower CR (1.25 actual vs. 1.31 predicted) and
SSI (4.21 actual vs. 3.99 predicted), demonstrate signs of
soil degradation. These observations underline the neces-
sity of sustainable land management practices in agricul-
tural regions to enhance soil resilience and decrease erosion
risks (Critchley etal. 2023). The model's reliable predictions
regarding soil erodibility and structural indicators across
different land use and land cover (LULC) types affirm its
capability to simulate the effects of land use on soil health.
These findings reinforce the broader scientific consensus that
forested areas provide better protection against soil erosion.
At the same time, cultivated regions are more vulnerable
and demand improved management strategies to maintain
soil integrity quality.
The impact of land use and land cover on soil proper-
ties reveals notable differences across various land use
types. The observed soil erodibility factor for cultivated
land is 0.10, which is slightly higher than the predicted
value of 0.09, suggesting that cultivation may marginally
increase soil erodibility. Aggregate stability in cultivated
areas is 56.87, which is closely aligned with the predicted
value of 56.76, indicating relatively stable soil aggre-
gates. The dispersion ratio and crust formation are lower
than other land uses, potentially reducing erosion rates.
However, the observed clay ratio and structural stability
index are slightly lower, which might impact long-term
soil stability. In contrast, forested land shows the lowest
observed soil erodibility factor of 0.06, consistent with
the predicted value, reflecting the protective effect of for-
est cover against soil erosion. Forest soils also exhibit
the highest aggregate stability at 60.75 and the highest
structural stability index at 8.88, indicating superior soil
structure and erosion resistance. The dispersion and clay
ratios are also high, suggesting that forest cover maintains
soil structure and nutrient retention. Pasture land presents
an intermediate observed soil erodibility factor of 0.09,
with a slightly lower predicted value. Aggregate stability
is comparable to cultivated land, but the structural stability
index is lower than in forested areas, potentially affecting
soil resilience to erosion. The dispersion and clay ratios
are moderate, reflecting the influence of pasture manage-
ment on soil properties.
Spatial distribution andANN process ofsoil
erodibility factors
Various interpolation techniques, aligned with appropriate
models, have been employed to map the spatial distribution
of soil erodibility. These techniques utilize observed and pre-
dicted values, evaluated based on the root mean square error,
to delineate the influencing factors. The results indicate that
the Radial Basis Function (RBF) with Completely Regular-
ized Spline (CRS) performed better than the Inverse Distance
Weighting (IDW) method, as reflected by the lower root mean
square error (RMSE) values. The IDW method, applied to the
observed data, yielded an RMSE of 0.043, indicating a reli-
able interpolation for the observed dataset. This low RMSE
suggests that IDW was effective in modelling the spatial dis-
tribution of the observed values, although it may not fully
capture more complex spatial patterns. On the other hand, the
RBF-CRS method, applied to the predicted data, produced a
slightly lower RMSE of 0.04017%, indicating better accu-
racy in capturing the spatial variability of the predicted data
(Table6) (Chen etal. 2020). The ability of RBF-CRS to gen-
erate a smoother and more continuous surface, coupled with
its lower error, suggests that it is more suited for datasets with
complex spatial structures. Both methods performed well, but
Table 3 Eigenvalue, percentage of variance and cumulative
Factors %Eigen value %Variance %Cumulative
OM 3.86941 55.28% 55.28%
Clay 1.43399 20.49% 75.76%
Silt 0.6602 9.43% 85.19%
Sand 0.58079 8.30% 93.49%
Agg.St 0.36236 5.18% 98.67%
Permeability 0.09325 1.33% 100.00%
Structure 0 0.00% 100.00%
Table 4 Land use types and area coverage
Land use type Area (ha) Ratio (%)
Cultivated 194.31 39.97
Forest 243.13 50.02
Water bodies 5.01 1.03
Pasture 43.67 8.98
Total 486.12 100.00
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Environmental Earth Sciences (2025) 84:291 Page 11 of 28 291
the RBF-CRS method was more accurate in predicting spatial
patterns, as evidenced by the lower RMSE.
Soil erodibility‑k factor
The K-factor indicated high spatial variability among
the different land use types in the study area, showing
values of 0.095 t·ha·hr·MJ−1·mm−1 for cultivated land,
0.060 t·ha·hr·MJ−1·mm−1 for the forested area, and 0.092
t·ha·hr·MJ−1·mm−1 for pasture land (Fig.7). This means that
cultivated lands, which have the highest K-factor, are very
prone to erosion. This is in agreement with other studies
such as Cerdan etal. 2010, stating that intensive agriculture
practices destroy the soil structure and therefore the soil
becomes more vulnerable to erosion. Moreover, Pimentel
and Kounang, 1998, had observed that conventional systems
of tillage primarily contributed to an increase in the rate
of soil erosion by breaking down the soil aggregates and
reducing organic matter content in it. The lowest K-factors
are observed in the forested sites, indicating that the soil
there is well stabilized and thus highly resistant to erosion.
Cohesive soils, a higher degree of soil cohesiveness, and less
surface runoff are the characteristics of forest lands with the
protective canopy cover, root systems, and organic matter all
combining to mitigate the erosive process. These findings
are in agreement with Morgan (2005) who had proven that
forests protect soil from erosive forces. Moreover, Nyssen
etal. (2004) described that the value of forest cover was pre-
defined by the importance of forested landscapes to diminish
soil erosion, especially in mountain areas.
The pasture, with the value of K being 0.092
t·ha·hr·MJ−1·mm−1, lies somewhat between the cultivated and
forested. While some protection is afforded by the pasture veg-
etation, compaction due to grazing often lowers infiltration
and increases surface runoff, rendering these systems more
vulnerable to erosion. Sharpley and Syers (1979) previously
stated similar points to suggest that, under grazing pressure,
the chances of soil erosion can be increased. Similarly, Trimble
Fig. 6 Land use map of the
study area
Table 5 Observed and predicted values of some soil properties in dif-
ferent land uses
Ac Actual; soil erodibility factor; Agg.St aggregate stability; Pred Pre-
dicted; DR Dispersion ratio; CR Clay ratio; SSI Structural stability
index; CF crust formation
Erodibility Parameter LULC
Cultivated Forest Pasture
W-K 0.10 0.06 0.09
Pred-K 0.09 0.06 0.08
Ac -Agg.St 56.87 60.75 56.83
Pred-Agg.St 56.76 60.42 55.55
Ac-DR 21.88 23.45 23.50
Pred -DR 21.43 23.41 23.58
Ac-CF 3.89 4.12 3.79
Pred -CF 3.89 4.16 3.80
Ac -CR 1.25 2.97 2.39
Pred-CR 1.31 3.11 2.40
Ac-SSI 4.21 8.88 6.17
Pred -SSI 3.99 8.51 5.84
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Environmental Earth Sciences (2025) 84:291291 Page 12 of 28
and Mendel (1995) highlighted that overgrazing can lead to
the breaking down of soil structure, causing an increase in the
erodibility of pasture ecosystems. Above all, these findings fur-
ther underlie the importance of land use for determining erod-
ibility. The high K-factor values in the cultivated areas argue
for better land management practices through approaches such
as conservation tillage and crop rotation to lessen erosion risk.
On the other hand, a lower K-factor for forested areas implies
that maintenance and expansion of forest cover should be pur-
sued to enhance soil stability and reduce erosion.
Artificial Neural Networks (ANN) provide a robust
framework for predicting soil erodibility factors due to their
ability to model complex, non-linear relationships between
soil properties and erosion risk. By leveraging a variety of
input data, including soil texture, organic matter content,
structure and permeability, ANNs can generate more accu-
rate predictions compared to traditional empirical methods.
The variation in soil erodibility values observed in this
study, ranging from 0.023 to 0.152 t h MJ −
1mm −
1 for
the USLE-derived values and 0.026 to 0.148 t h MJ − 1mm
− 1 as predicted by the artificial neural network (ANN),
aligns with previous findings in soil erosion modelling
research (Fig.7). The slight differences between observed
and predicted values suggest that the ANN model performed
effectively in estimating soil erodibility factors, a critical
parameter influencing soil loss predictions. These values fall
within a common range reported in similar studies on soil
erosion in agricultural landscapes. For example, in a study
by Fu etal. (2010), soil erodibility in hilly and cultivated
regions of China showed a similar range, with USLE-based
calculations yielding erodibility factors between 0.02 and
0.15 t h MJ −
1mm −
1, depending on the land cover and
soil type. This range is consistent with the values found in
the current study, further validating the reliability of using
ANN for predicting soil erodibility factors.
Additionally, Arabameri etal. (2021) reported compara-
ble ranges of K values in erosion-prone regions, highlighting
that using machine learning techniques, such as ANN, can
provide robust estimates when calibrated correctly, espe-
cially in complex landscapes. Furthermore, the ANN's pre-
diction performance is supported by findings from Nouri
etal. (2024), where ANN models were successfully used
to predict soil erodibility with minimal deviation from
observed values. Their results indicated that ANN can cap-
ture the nonlinear relationships between soil properties and
erosion factors, similar to the current study’s results, demon-
strating its applicability in modelling soil erosion processes.
The close match between observed and ANN-predicted soil
erodibility values in this study is consistent with existing
research, affirming that ANN models are effective tools
for estimating soil erosion risk in various landscapes. This
further underscores the importance of machine learning in
enhancing predictive accuracy for sustainable soil manage-
ment practices.
Table 6 The root means square error (RMSE) values for observed and predicted soil properties
K soil erodibility factor; Agg.St. aggregate stability; DR Dispersion ratio; CR Clay ratio; SSI Structural stability index; CF crust formation; Or
Ordinary; Si Simple; Uni Universal; Shp Spherical; Exp Exponential; Gau Gaussian
*The bold and underlined methods are the selected interpolation techniques based on their low RMSE values
Interpolation
Methods
RMSE values
K DR CF Agg.St SSI CR
Ob Pr Ob Pr Ob Pr Ob Pr Ob Pr Ob Pr
1 0.044 0.0402 8.906 8.053 1.769 1.750 13.770 15.328 3.139 3.347 0.0365 2.616
IDW 20.043 0.0406 9.001 8.389 1.787 1.765 13.471 15.156 3.113 3.303 0.0352 2.682
3 0.045 0.043 9.420 8.983 1.882 1.856 14.159 15.929 3.195 3.328 0.0347 2.807
RBF CRS 0.0436 0.0401 8.788 8.135 1.769 1.749 13.600 15.264 3.096 2.92 0.0412 2.651
SWT 0.0436 0.0401 8.780 8.082 1.763 1.745 13.596 15.258 3.098 2.922 0.0405 2.651
TPS 0.0538 0.0543 11.818 11.535 2.331 2.299 19.214 21.683 3.410 3.155 0.0413 3.553
Kriging Or Sph 0.9497 0.948 0.974 1.067 1.006 1.008 0.941 1.016 0.975 0.978 0.5194 2.624
Exp 0.9452 0.941 0.972 1.036 0.984 0.977 0.942 1.003 0.97 0.982 0.4188 2.635
Gau 0.9478 0.933 0.973 1.054 0.991 0.992 0.945 1.034 0.9724 0.974 0.0346 2.660
Si Sph 1.0227 1.039 1.091 1.1207 1.027 1.017 1.119 1.170 0.887 0.895 0.4975 2.452
Exp 1.0059 1.033 1.094 1.1208 0.973 1.000 1.061 1.092 0.888 0.897 0.265 2.453
Gau 1.0174 1.038 1.084 1.1202 0.999 1.014 1.134 1.173 0.889 0.891 1.839 2.456
Uni Sph 0.9497 0.948 0.974 1.0676 1.006 1.008 0.941 1.016 0.9753 0.975 0.5194 2.624
Exp 0.9452 0.941 0.972 1.0565 0.984 0.977 0.942 1.003 0.9791 0.979 0.4188 2.635
Gau 0.9478 0.936 0.973 1.0752 0.991 0.992 0.934 1.034 0.9724 0.972 0.3486 2.660
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Environmental Earth Sciences (2025) 84:291 Page 13 of 28 291
The ANN model developed in this study demonstrated
excellent predictive performance, with an R2 value of
0.99, indicating that it can explain 99% of the variance
in soil erodibility factors. This finding is consistent with
the results reported by Mirzaee etal. (2017) and Mir-
zaee etal 2020, who found that ANN-based SSPF mod-
els yielded the highest R2 and lowest RMSE values for
predicting WEPP erodibility parameters (Kib and Krb)
and also performed reasonably well for τcb, underscor-
ing the effectiveness of ANN in modeling complex soil
properties. This accurate prediction has essential impli-
cations for soil erodibility-K. As this study observes (up
to 9.28%), soils with a higher structure index are gener-
ally better aggregated and more erosion-resistant. Well-
structured soils improve water infiltration, enhance root
growth, and are less susceptible to soil detachment during
rainfall events. This relationship between soil structure
index and erosion resistance has been extensively docu-
mented, with Lal (2014) affirming that soils with more
substantial structures exhibit lower erosion rates. The
strong correlation between observed and predicted soil
structure indices demonstrates the model’s reliability in
assessing soil stability. The findings reinforce the impor-
tance of maintaining soil structure to mitigate erosion
risks, particularly in erosion-prone areas. Promoting soil
conservation practices that enhance aggregation, such as
incorporating organic matter and maintaining vegetative
cover, is essential for sustaining soil health. The model’s
ability to accurately predict soil structure provides valu-
able insights for soil conservation and land management
strategies to reduce erosion and preserve agricultural
productivity.
The R values for predicting soil erodibility-K and its fac-
tors are notably high across all datasets, reflecting strong
model performance. The training sample R of 1.00 indicates
a perfect fit with the training data, suggesting the model has
effectively captured the variance in soil erodibility factors
within this subset. The slightly lower but still impressive R2
values of 0.99 for validation and testing datasets demonstrate
the model's robust generalization capabilities, maintaining
Fig. 7 Soil erodibility-K maps (A): Observed and (B); Predicted
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Environmental Earth Sciences (2025) 84:291291 Page 14 of 28
high predictive accuracy on unseen data. The overall R of 0.99
consolidates these findings, affirming that the model reliably
predicts soil erodibility across various data sets (Fig.8). This
high degree of consistency highlights the model's effective-
ness in capturing the relationships between soil erodibility
and its influencing factors. However, the minor reduction from
training to validation/testing data is normal and underscores
the inherent variability in real-world applications.
The model achieved its best performance at epoch 11,
with the lowest RMSE, indicating an accurate prediction
of soil erodibility based on properties such as texture and
organic matter. The Levenberg–Marquardt algorithm effec-
tively captured non-linear relationships, and further training
showed no improvement, highlighting the model's optimal
performance at this point (Fig.9). This result suggests that
the model generalizes well and that prolonged training may
lead to overfitting, emphasizing the importance of stopping
at the optimal epoch for precise environmental predictions.
Figure10 illustrates a strong agreement between the
predicted and observed values derived from the K-USLE
model, as well as the predictions generated by the Artificial
Neural Network (ANN). The close alignment of these values
is supported by a high coefficient of determination (R2 =
0.98), indicating that the ANN model explains 98% of the
variability in the observed data. Furthermore, the regression
equation (y = 0.9992x + 2E-05) demonstrates an almost per-
fect 1:1 relationship between predicted and observed values,
suggesting minimal bias in the ANN model's predictions.
These findings highlight the ANN’s robust performance in
modeling soil erodibility and confirm its reliability as a com-
plementary approach to traditional models like K-USLE.
The aggregate stability
The aggregate stability findings from this study reveal a
range of 40.47% to 68.57% for predicted values and 42.6%
to 71.9% for observed values (Fig.11).
This disparity suggests that observed values tend to be
higher than predicted, which could be attributed to sev-
eral factors (Fig.12). Predictive models might not fully
account for the variability in soil conditions or manage-
ment practices that influence aggregate stability. In con-
trast, observed values reflect real-world conditions and
may benefit from local soil management practices, such
as conservation tillage or cover cropping, which enhance
aggregate stability (Sarker 2023). Additionally, the higher
organic matter content often found in well-managed soils
can improve aggregate stability, aligning with findings
by Rui etal. (2022). As discussed by Bronick and Lal
(2005), soil texture also plays a significant role, with
coarser-textured soils typically exhibiting lower aggregate
stability compared to finer-textured soils. Environmental
factors, such as rainfall intensity, further contribute to this
variability, as noted by Khan etal. (2017). Thus, the higher
observed values might reflect better management practices
and real-world conditions not fully captured by the predic-
tive models, highlighting the need to incorporate these
factors into future model improvements.
The model achieved its best validation performance at
epoch 5 from the total, with a mean squared error of 2.017
for predicting aggregate stability using various soil proper-
ties (Fig.13). This suggests the model effectively learned
the relationships between inputs and the crust formation by
this point, with further training offering no improvement and
risking overfitting. The result emphasizes stopping at the
optimal epoch to maintain accurate predictions.
The crust formation
The crust formation values observed in this study, rang-
ing from 0.97 to 6.40, and predicted values from 0.59 to
6.73, provide significant insights into soil erosion dynamics
(Fig.14).
Crust formation impacts soil erodibility by affecting
surface structure and infiltration rates. Higher crust for-
mation values generally indicate reduced soil porosity and
increased runoff, leading to greater erosion susceptibil-
ity. Observed values surpassing predicted ones suggest
that real-world conditions may exacerbate crust formation
beyond what models anticipate. This finding is consistent
with research by Yeoh (2023), which indicates that crusted
soils experience higher erosion rates due to decreased
infiltration. The influence of land use and cover types on
crust formation is also notable; intensive agricultural prac-
tices and soil disturbance often result in increased crust-
ing compared to natural vegetation or cover crops, which
enhance soil structure and organic matter content (Yuan
etal. 2023). Previous studies, such as those by Zaady
etal (2013), have similarly found that agricultural lands
exhibit more significant crusting compared to undisturbed
lands. This aligns with the higher observed values in your
study if the observed areas involved intensive land use.
Furthermore, research by Saby etal. (2017) supports the
observation that real-world crust formation often exceeds
predictions due to complex field conditions. These com-
parisons underline the need to refine predictive models to
better account for actual soil management practices and
environmental conditions (Fig.15).
The model achieved its best validation performance at
epoch 9, with a mean squared error of 0.0114 for predict-
ing the crust formation index using various soil properties
(Fig.16). This suggests the model effectively learned the
relationships between inputs and the crust formation by this
point, with further training offering no improvement and
risking overfitting. The result emphasizes stopping at the
optimal epoch to maintain accurate predictions.
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Environmental Earth Sciences (2025) 84:291 Page 15 of 28 291
Fig. 8 Results of regression
between output data and targets
for the levenberg–marquardt
Fig. 9 Levenberg–marquardt
combination performance graph
of soil erodibility -K factor
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Environmental Earth Sciences (2025) 84:291291 Page 16 of 28
Clay ratio
The research results indicate that the observed clay ratio
ranges from 0.79 to 4.24, while the predicted clay ratio spans
a broader range of 0.44 to 9.01 (Fig.17).
This discrepancy suggests that although the prediction
model captures the general trend, it overestimates the clay
ratio at the upper end and underestimates it at the lower
end. Such variations may stem from model limitations, input
data quality, or unaccounted variability in soil properties.
Refining the model by incorporating additional predictor
variables, improving calibration, or employing advanced
methods to better represent non-linear soil processes could
address these discrepancies (Fig.16).
Comparing these findings with published results, it is
common for prediction models to exhibit varying degrees
of accuracy in estimating soil properties like clay con-
tent. Studies using machine learning approaches, such
as Artificial Neural Networks (ANNs) and Random For-
ests, have reported deviations of 10–30% from observed
Fig. 10 Observed versus pre-
dicted value of soil erodablity
Fig. 11 Soil aggregate stabil-
ity (A); observed and (B);
Predicted
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Environmental Earth Sciences (2025) 84:291 Page 17 of 28 291
values, with extreme values posing significant challenges
(Stenberg and Rossel 2010; Padarian etal. 2019). Simi-
lar discrepancies have been observed in heterogeneous
landscapes where soil properties vary significantly, align-
ing with the wider predicted range in the current study
(McBratney etal. 2003). Improvement strategies sug-
gested in the literature include integrating high-resolution
spatial or temporal data to enhance prediction accuracy
(Minasny and McBratney 2016). Additionally, employing
more advanced machine learning techniques or combin-
ing multiple data sources, such as remote sensing and
ground-based observations, may reduce discrepancies and
improve model performance (Grunwald 2009).
The correlation coefficients of 0.99 for the training, test,
and entire datasets, along with 0.97 for cross-validation
(Fig.18), indicate that the model performs exceptionally
well. It shows a strong positive relationship between observed
and predicted values, with minimal overfitting, as the per-
formance on cross-validation is slightly lower but still high.
The model achieved its best validation performance
at epoch 3, with a mean squared error of 0.0410 for
predicting the clay ratio using various soil properties
(Fig.19). This suggests the model effectively learned
the relationships between inputs and clay ratio by this
point, with further training offering no improvement and
risking overfitting. The result emphasizes the importance
of stopping at the optimal epoch to maintain accurate
predictions.
Dispersion rate
The study found that the observed soil dispersion rate ranged
from 14.8 to 33.6, while the predicted values fell between 14.8
and 33.6 (Fig.20). These results indicate that the model accu-
rately captured mid-range dispersion rates but underestimated
higher extremes and overestimated lower values. This variation
may be due to model limitations or the inherent variability in
soil properties, such as texture, organic matter, and mineral
composition, which significantly influence dispersion rates.
Comparing these findings with published results, pre-
diction models often face challenges in estimating extreme
values due to the complex nature of soil processes. Zhang
etal. (2018) noted similar difficulties in soils with high clay
content or organic matter, which strongly impacts soil stabil-
ity and aggregation. The predicted range in this study aligns
with mid-range dispersion rates but does not fully capture
Fig. 12 Results of regression
between inputs and targets soil
aggregate stability for the leven-
berg–marquardt
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Environmental Earth Sciences (2025) 84:291291 Page 18 of 28
Fig. 13 Levenberg–marquardt
combination performance
graphs of aggregate stability
Fig. 14 Crust formation map
(A); observed and (B); Pre-
dicted
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Environmental Earth Sciences (2025) 84:291 Page 19 of 28 291
Fig. 15 Results of regression
between inputs and targets crust
formation for the levenberg–
marquardt
Fig. 16 Levenberg–marquardt
combination performance graph
of crust formation
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Environmental Earth Sciences (2025) 84:291291 Page 20 of 28
the broader variability, especially at the upper end of the
observed values.
Soil dispersion is a key indicator of soil erodibility.
Higher dispersion rates are linked to increased suscepti-
bility to erosion, as water and wind transport dispersed
soils more easily. Studies have shown that factors like clay
mineralogy and organic carbon content significantly affect
soil dispersion (Levy etal. 2003). The higher observed dis-
persion rates (up to 33.6) suggest that soils with poor struc-
ture or low organic matter are more vulnerable to erosion.
This is consistent with research indicating that soils with
higher dispersion rates have reduced aggregate stability and
are more prone to erosion (Lal2014). The implications of
these findings for soil erodibility are significant. Soils with
higher observed dispersion rates are more susceptible to
erosion, leading to increased sediment loss and land degra-
dation. The observed dispersion rates can serve as an indi-
cator of the soil erodibility factor (K-factor), with higher
values suggesting a greater need for conservation measures.
Darboux etal. (2023) highlight those soils with high dis-
persion rates often experience structural breakdown, reduc-
ing agricultural productivity and increasing erosion risk.
The findings emphasize the importance of soil conservation
practices, particularly in areas with high dispersion rates.
Improving model accuracy by incorporating additional var-
iables such as clay mineralogy and organic matter content
could enhance the prediction of soil erosion risk, leading
to more effective management strategies (Fig.21).
In the soil dispersion rate simulation over 10 epochs, the
best validation performance was achieved at epoch 4, with a
mean squared error (MSE) of 21.5 (Fig.22). This indicates
that the model was most effective at predicting the disper-
sion rate at this point. The MSE of 21.5 reflects the average
error between predicted and actual values. Further training
beyond epoch 4 did not improve performance and may have
led to overfitting. This result highlights the importance of
Fig. 17 Clay ratio map (A): observed and (B): Predicted
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Environmental Earth Sciences (2025) 84:291 Page 21 of 28 291
Fig. 18 Results of regression
between inputs and targets clay
ratio for the levenberg–mar-
quardt
Fig. 19 Levenberg–marquardt
combination performance graph
of clay ratio
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Environmental Earth Sciences (2025) 84:291291 Page 22 of 28
stopping training early to maintain accuracy and avoid over-
fitting, with epoch 4 being the optimal training duration.
Soil structure index
The study found that the soil structure index for the observed
data ranged from 2.94% to 9.28%, while the predicted values
ranged between 2.77% and 8.90% (Fig.23).
The minimal deviation between observed and expected
results suggests that the model effectively captures the key fac-
tors influencing soil aggregation and stability, such as organic
matter content and soil texture. This strong alignment reflects
the model’s robustness in estimating soil structural index,
which is crucial for understanding soil health and its impli-
cations for erosion. When compared with previous studies,
these findings are consistent with research emphasizing the
role of soil structure in soil stability and erosion resistance.
Bronick and Lal (2005) highlighted that soil aggregation is
largely influenced by organic matter and mineral composition,
which contribute to improved soil stability. Similarly, Borrelli
etal. (2017) emphasized the significance of organic matter in
maintaining well-structured soils. The model’s accuracy in this
study aligns with these findings, demonstrating its ability to
predict soil structure based on key influencing factors.
The regression analysis using the Levenberg–Marquardt
approach yielded high R2 values across training (0.99), vali-
dation (0.99), and test sets (0.97), indicating the model's
strong predictive capability and minimal variance unex-
plained. The model effectively generalized across unseen
data, maintaining robust accuracy. Key soil parameters,
including aggregate stability (R2 = 0.96), crust formation
(R2 = 0.98), critical organic matter levels (R2 = 0.98), and
clay ratio (R2 = 0.98), demonstrated near-perfect correla-
tions, suggesting that the model accurately captured the
influence of soil texture, organic matter, and structural prop-
erties. The lower R2 value for the dispersion ratio (0.90)
Fig. 20 Dispersion rate map (A); observed and (B); Predicted
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Environmental Earth Sciences (2025) 84:291 Page 23 of 28 291
Fig. 21 Results of regression
between inputs and targets
dispersion rate for the leven-
berg–marquardt
Fig. 22 Levenberg–marquardt
combination performance
graphs of dispersion rate
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Environmental Earth Sciences (2025) 84:291291 Page 24 of 28
indicates more complex variability in the input–output
relationship, likely due to the multifactorial influences on
soil dispersion. However, the soil erodibility factor (R2 =
0.99) and soil structure index (R2 = 0.99) displayed high
predictive accuracy, underscoring the model’s capability
to predict soil stability and erosion susceptibility. The high
R2 values across most parameters confirm the model's effi-
cacy in integrating key soil properties, making it a reliable
tool for assessing soil health and guiding soil management
practices (Fig.24).
The model achieved its best validation performance at
epoch 5, with a mean squared error of 0.18106 for predict-
ing the soil structural index using various soil properties
(Fig.25). This suggests the model effectively learned the
relationships between inputs and the structural index by
this point, with further training offering no improvement
and risking overfitting. The result emphasizes the impor-
tance of stopping at the optimal epoch to maintain accurate
predictions.
The analysis revealed varying Mean Squared Error (MSE)
values, indicating the model's performance in predicting dif-
ferent soil parameters. Parameters like aggregate stability,
crust formation, and critical organic matter levels had very
low MSE values, suggesting high prediction accuracy. In
contrast, the dispersion ratio (MSE = 3.16) and soil erod-
ibility factor (MSE = 10.0) showed higher errors, reflecting
the model's challenges in capturing the complexity of these
properties. The number of training epochs was effective for
parameters with low MSE, while those with higher errors
may require additional adjustments. Overall, the model per-
formed well but showed room for improvement in predicting
certain soil characteristics.
Fig. 23 Soil structural index (SSI), (A); Observed and (B); Predicted
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Environmental Earth Sciences (2025) 84:291 Page 25 of 28 291
Fig. 24 Results of regression
between inputs and targets soil
structural stability index for the
levenberg–marquardt
Fig. 25 Levenberg–marquardt
combination performance graph
of structural index -k factor
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Environmental Earth Sciences (2025) 84:291291 Page 26 of 28
Conclusion
This study primarily aimed to assess soil erosion sensitivity
within a micro-catchment using the K-factor in the Revised
Universal Soil Loss Equation (RUSLE) model, as well as
other erodibility factors, and, secondly, to evaluate their pre-
dictability through artificial neural networks (ANNS). Soil
erodibility (K-factor) was calculated for three land use-land
cover types: cultivated land (0.095 t·ha·hr·MJ−1·mm−1), for-
est land (0.061 t·ha·hr·MJ−1·mm−1), and pasture land (0.093
t·ha·hr·MJ−1·mm−1). The key soil properties influencing the
erodibility factor included organic matter (OM), soil texture
(sand, silt, and clay), soil permeability, and soil structure
codes. Additionally, other indicators of soil erodibility, such
as the dispersion ratio, clay ratio, soil aggregate stability,
soil structure index, modified clay ratio, and critical lev-
els of organic matter, were derived from these parameters.
The application of ANN for predicting soil erodibility and
related soil quality indicators demonstrated high prediction
accuracy. The correlation coefficients between observed and
predicted variables were 0.98 for soil erodibility, 0.96 for
aggregate stability, 0.98 for crust formation, 0.98 for critical
organic matter levels, 0.81 for dispersion ratio, and 0.98 for
the soil structure index. These results suggest that soil erod-
ibility can be reliably predicted through direct soil analysis
and ANN without requiring extensive consideration of envi-
ronmental and pedogenic factors. However, a limitation of
the study was the focus on a single micro-catchment and the
relatively small sample size, which could introduce errors in
ANN predictions. Additionally, this study didn not consider
other environmental variables as predictors, such as climate
data, terrain data, and vegetation indices. Future research
should expand the dataset to include different regions, land
use types and environmental variables to validate the gen-
eralizability and accuracy of ANN models in predicting soil
erodibility. This approach could enhance the robustness of
ANN in predicting soil properties across diverse landscapes.
Acknowledgements This research was supported by Ondokuz
Mayıs University under project code PYO.ZRT.1914.23.002. The
authorswould like to thank all contributors for their valuable support.
Author contributions WA: Conceptualization; Data Analysis; Investi-
gation; Methodology; Model, Map Preparation; Validation; Visualiza-
tion, Review, and Editing.EDA: Conceptualization, Methodology, Map
Preparation; Visualization, Review and Editing.OD: Conceptualiza-
tion, review, editing, writing, methodology and supervision.
Funding Open access funding provided by the Scientific and Tech-
nological Research Council of Türkiye (TÜBİTAK). No funding was
received for conducting this research.
Data availability No datasets were generated or analysed during the
current study.
Declarations
Ethics approval and consent to participate Not applicable.
Consent for publication Not applicable.
Competing interests The authors declare no competing interests.
Open Access This article is licensed under a Creative Commons Attri-
bution 4.0 International License, which permits use, sharing, adapta-
tion, distribution and reproduction in any medium or format, as long
as you give appropriate credit to the original author(s) and the source,
provide a link to the Creative Commons licence, and indicate if changes
were made. The images or other third party material in this article are
included in the article’s Creative Commons licence, unless indicated
otherwise in a credit line to the material. If material is not included in
the article’s Creative Commons licence and your intended use is not
permitted by statutory regulation or exceeds the permitted use, you will
need to obtain permission directly from the copyright holder. To view a
copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
References
Abiye W, Waltner I, Kindie H (2023) Spatiotemporal dynamics of
soil erosion response to land use land cover dynamics and cli-
mate variability in Maybar watershed, Awash basin, Ethiopia.
Geol Ecol Landsc:1–23
Agbai WP, Efenudu U, Kosuowei M, Ehigiator EF (2022) Determination
of soil erodibility status under different land-use types using some
erodibility indices. Ethiop J Environ Stud Manage 15(2):201–214
Alebachew ED, Dengiz O (2024) Multi-criteria decision analysis
(MCDA) and artificial neural network (ANN) for assessing soil
quality variation under different land use and land cover in Sam-
sun, Türkiye (Msc thesis, un published)
Alebachew ED, Abiye W, Dengiz O, Turan İD (2025) Soil erosion
estimation and risk assessment based on RUSLE in Google
Earth Engine (GEE) in Turkiye. Annals of GIS:1–19
Alzubaidi L, Zhang J, Humaidi AJ, Al-Dujaili A, Duan Y, Al-
Shamma O … Farhan L (2021) Review of deep learning: con-
cepts, CNN architectures, challenges, applications, future direc-
tions. J Big Data 8:1–74
Amsili JP, van Es HM, Schindelbeck RR (2021) Cropping system and
soil texture shape soil health outcomes and scoring functions. Soil
Security 4:100012
Arabameri A, Chandra Pal S, Costache R, Saha A, Rezaie F, Seyed
Danesh A, ... Hoang ND (2021) Prediction of gully erosion
susceptibility mapping using novel ensemble machine learn-
ing algorithms.Geomatics, Natural Hazards and Risk,12(1),
469–498
Archibong GA, Sunday EU, Akudike JC, Okeke OC, Amadi C (2020)
A review of the principles and methods of soil stabilisation. Int J
Adv Acad Res| Sci 6(3):2488–9849
Baillie IC (2001) Soil survey staff 1999, soil taxonomy: a basic sys-
tem of soil classification for making and interpreting soil surveys.
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Environmental Earth Sciences (2025) 84:291 Page 27 of 28 291
Agricultural Handbook 436. Natural Resources Conservation Ser-
vice, USDA, Washington, DC, pp 869
Barchia MF, Sulistyo B, Putri EL, Herman W (2023) Spatial prediction
of soil erodibility indices of the sensitive landscape of Bengkulu
Watershed, Indonesia. Int J Environ Sci Dev 14(3):190–194
Borrelli P, Robinson DA, Fleischer LR, Lugato E, Ballabio C, Alewell
C … Panagos P (2017) An assessment of the global impact of
21st-century land use change on soil erosion. Nat Commun 8:2013
Bouyoucos GJ (1951) A recalibration of the hydrometer method for
makingmechanical analysis of soils. Agron J 43:434–438
Bronick CJ, Lal R (2005) Soil structure and management: a review.
Geoderma 124(1–2):3–22
Bryan RB (1968) The development, use, and efficiency of indices of
soil erodibility. Geoderma 2(1):5–26
Carter MR, Gregorich EG (2007) Soil sampling and methods of analysis.
CRC Press
Cerdà A (2000) Aggregate stability against water forces under different
climates on agriculture land and scrubland in southern Bolivia. Soil
Tillage Res 57(3):159–166
Cerdan O, Govers G, Le Bissonnais Y, Van Oost K, Poesen J, Saby
N, Gobin A, Vacca A, Quinton J, Auerswald K, Kliks A, Kwaad
JPM, Raclot D, Ionita I, Rejman J, Rousseva S, Muxart T, Rox
MJ, Dostal T (2010) Rates and spatial variations of soil erosion
in Europe: a study based on erosion plot data. Geomorphology
122(1–2):167–177
Chen X, Duan Z (2015) Impacts of soil crusts on soil physicochemical
characteristics in different rainfall zones of the arid and semi-arid
desert regions of northern China. Environ Earth Sci 73:3335–3347
Chen Y, Song L, Liu Y, Yang L, Li D (2020) A review of the artifi-
cial neural network models for water quality prediction. Appl Sci
10(17):5776
Chen S, Zhang G, Zhu P, Wang C, Wan Y (2023) Impact of land use type on
soil erodibility in a small watershed of rolling hill northeast China. Soil
Tillage Res 227:105597
Critchley W, Harari N, Mollee E, Mekdaschi-Studer R, Eichenberger
J (2023) Sustainable land management and climate change adap-
tation for small-scale land users in Sub-Saharan Africa. Land
12(6):1206
Darboux F, Räsänen T, Johannsen L, Schmaltz E, Moussa R, Pellegrini
S, Deproost P (2023) Guideline on the current implementation
of erosion measures and other connectivity elements depending
on scale and modelling approach. Doctoral dissertation, EJP Soil
Efthimiou N (2020) The new assessment of soil erodibility in Greece.
Soil Tillage Res 204:104720
Erdoğan Yüksel E, Yavuz G (2024) Evaluating the soil properties of
different land use types in the Deviskel Watershed in the Hilly
Region of Northeast Türkiye. Sustainability 16(22):9732
FAO (2022) Soil Organic Carbon Mapping Cookbook. Available
online: https:// www. fao. org/documents/card/en/c/I8895EN/
Franzluebbers AJ (2002) Water infiltration and soil structure related to
organic matter and its stratification with depth. Soil Tillage Res
66(2):197–205
Fu X, Shao M, Wei X, Horton R (2010) Soil organic carbon and total
nitrogen as ffected by vegetation types in Northern Loess Plateau
of China. Geoderma 155(1–2):31–35
Goodfellow I, Bengio Y, Courville A, Bengio Y (2016) Deep learning,
vol 1, no 2. Cambridge: MIT Press
Grunwald S (2009) Multi-criteria characterization of recent digital soil
mapping and modeling approaches. Geoderma 152(3–4):195–207
Kemper WD, Rosenau RC (1986) Aggregate stability and size distribu-
tion. In: Methods of soil analysis: part 1 physical and mineralogi-
cal methods, vol 5, pp 425–442
Khan KY, Pozdnyakov AI, Son BK (2007) Structure and stability of
soil aggregates. Eurasian Soil Sci 40:409–414
Khan AU, Jiang J, Wang P, Zheng Y (2017) Influence of watershed
topographic and socio-economic attributes on the climate sensitiv-
ity of global river water quality. Environ Res Lett 12(10):104012
Kusre BC, Ghosh P, Nath K (2018) Prioritization of soil conservation
measures using erodibility indices as criteria in Sikkim (India). J
Earth Syst Sci 127:1–13
Lal R (2001) Soil degradation by erosion. Land Degrad Dev
12(6):519–539
Lal R (2014) Soil conservation and ecosystem services. Int Soil Water
Conserv Res 2(3):36–47
Lal R (2021) Fate of soil carbon transported by erosional processes.
Appl Sci 12(1):48
Levy GJ, Mamedov AI, Goldstein D (2003) Sodicity and water quality
effects on slaking of aggregates from semi-arid soils 1. Soil Sci
168(8):552–562
Li Y, Pan Y (2020) A novel ensemble deep learning model for stock
prediction based on stock prices and news.https:// doi. org/ 10. 48550/
arXiv. 2007. 12620
Liang C, Wu Y, Liu J, Chen D, Zhu Y (2022) A critical review on soil
structure: research methods, structured indexes, and constitutive
models. Arab J Geosci 15(18):1509
Liu J, Chen L, Wang B, Peng X (2024) Effects of physical crust on soil
detachment by overland flow in the loess plateau region of China.
Int Soil Water Conserv Res 12(1):107–120
Luo T, Liu W, Xia D, Xia L, Guo T, Ma Y, ... Hu Y (2022) Effects of
land use types on soil erodibility in a small karst watershed in
western Hubei.PeerJ10:e14423
Malone BP, Minasny B, McBratney AB, Malone BP, Minasny B,
McBratney AB (2017) Some methods for the quantification of
prediction uncertainties for digital soil mapping. In: Using R for
digital soil mapping, pp 169–219
McBratney AB, Santos MM, Minasny B (2003) On digital soil mapping.
Geoderma 117(1–2):3–52
Minasny B, McBratney AB (2016) Digital soil mapping: a brief history
and some lessons. Geoderma 264:301–311
Mirzaee S, Ghorbani-Dashtaki S (2018) Deriving and evaluating
hydraulics and detachment models of rill erosion for some cal-
careous soils. Catena 164:107–115
Mirzaee S, Ghorbani-Dashtaki S (2021) Calibrating the WEPP model
to predict soil loss for some calcareous soils. Arab J Geosci
14:1–10
Mirzaee S, Ghorbani-Dashtaki S, Mohammadi J, Asadzadeh F, Kerry
R (2017) Modeling WEPP erodibility parameters in calcareous
soils in northwest Iran. Ecol Ind 74:302–310
Mirzaee S, Ghorbani-Dashtaki S, Kerry R (2020) Comparison of a
spatial, spatial and hybrid methods for predicting inter-rill and
rill soil sensitivity to erosion at the field scale. Catena 188:104439
Morgan RPC (2001) A simple approach to soil loss prediction: a revised
Morgan–Morgan–Finney model. Catena 44(4):305–322
Morgan RPC (2005) Soil erosion and conservation, 3rd edn. Blackwell
Publishing Company
Nciizah AD, Wakindiki II (2015) Physical indicators of soil erosion, aggre-
gate stability and erodibility. Arch Agron Soil Sci 61(6):827–842
Nouri A, Esfandiari M, Eftekhari K, Torkashvand AM, Ahmadi A
(2024) Development support vector machine, artificial neural
network and artificial neural network–genetic algorithm hybrid
models for estimating erodible fraction of soil to wind erosion.
Int J River Basin Manage 22(3):379–388
Nyssen J, Poesen J, Moeyersons J, Deckers J, Haile M, Lang A (2004)
Human impact on the environment in the Ethiopian and Eritrean
highlands—a state of the art. Earth-Sci Rev 64(3-4):273–320
Özdemir N, Gülser C (2017) Clay activity index as an indicator of soil
erodibility. Eurasian J Soil Sci 6(4):307–311
Padarian J, Minasny B, McBratney AB (2019) Using deep learning for
digital soil mapping. Soil 5(1):79–89
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Environmental Earth Sciences (2025) 84:291291 Page 28 of 28
Panagos P, Meusburger K, Ballabio C, Borrelli P, Alewell C (2014) Soil
erodibility in Europe: a high-resolution dataset based on LUCAS. Sci
Total Environ 479:189–200
Panda S (2022) Soil properties responsible for soil loss. In: Soil and
water conservation for sustainable food production. Springer
International Publishing, Cham, pp 13–34
Parwadaa C, Van Tolab J (2017) Litter quality effects on soil stability
and erodibility of selected South African soils
Pimentel D (2006) Soil erosion: a food and environmental threat. Envi-
ron Dev Sustain 8:119–137
Pimentel D, Harvey C, Resosudarmo P etal (1995) Environmental and
economic costs of soil erosion and conservation benefits. Science
267(5201):1117–1123
Prout JM, Shepherd KD, McGrath SP, Kirk GJ, Haefele SM (2021) What is
a good level of soil organic matter? an index based on organic carbon
to clay ratio. Eur J Soil Sci 72(6):2493–2503
Rahnenführer J, De Bin R, Benner A, Ambrogi F, Lusa L, Boulesteix
AL, … & topic group “High-dimensional data” (TG9) of the
STRATOS initiative (2023) Statistical analysis of high-dimen-
sional biomedical data: a gentle introduction to analytical goals,
common approaches and challenges. BMC Med 21(1):182
Renard KG (1997) Predicting soil erosion by water: a guide to con-
servation planning with the revised universal soil loss equation
(RUSLE). US Department of Agriculture, Agricultural Research
Service
Rui Y, Jackson RD, Cotrufo MF, Sanford GR, Spiesman BJ, Deiss L,
... Ruark MD (2022) Persistent soil carbon enhanced in Mollisols
by well-managed grasslands but not annual grain or dairy forage
cropping systems.Proceedings of the National Academy of Sci-
ences,119(7), e2118931119.
Saby N, Minasny B, Arrouays D, de Gruitjer J, Malone BP, Mcbratney
AB (2017) Designing soil monitoring schemes for large areas
based on high resolution digital soil mapping products: a case
study from France. In: Pedometrics 2017, p 298
Sarker MMR (2023) Trade-off analysis of crop residue management
for improving conservation agriculture practices under a changing
climate in Bangladesh. Doctoral dissertation, University of Leeds
Senanayake S, Pradhan B, Huete A, Brennan J (2020) A review on
assessing and mapping soil erosion hazard using geo-informat-
ics technology for farming system management. Remote Sens
12(24):4063
Sharififar A (2022) Accuracy and uncertainty of geostatistical models
versus machine learning for digital mapping of soil calcium and
potassium. Environ Monit Assess 194(10):760
Sharma IP, Chandra S, Kumar N, Chandra D (2017) PGPR: heart
of soil and their role in soil fertility. In: Agriculturally impor-
tant microbes for sustainable agriculture: Volume I: Plant-Soil-
Microbe Nexus, pp 51–67
Sharpley AN, Syers JK (1979) Phosphorus inputs into a stream drain-
ing an agricultural watershed: II. Amounts contributed and rela-
tive significance of runoff types. Water Air Soil Pollut 11:417–428
Shi ZH, Cai CF, Ding SW, Wang TW, Chow TL (2002) Soil conserva-
tion planning at the small watershed level using RUSLE with GIS:
a case study in the three gorges area of China. Catena 47(2):85–99
Smith P, Lutfalla S, Riley WJ, Torn MS, Schmidt MW, Soussana JF
(2018) The changing faces of soil organic matter research. Eur J
Soil Sci 69(1):23–30
Stenberg B, Rossel RV (2010) Diffuse reflectance spectroscopy for
high-resolution soil sensing. In: Proximal soil sensing, pp 29–47
Sumner ME, Miller WP (1992) Soil crusting about global soil degrada-
tion. Am J Altern Agric 7(1–2):56–62
Taleshian Jeloudar F, Ghajar Sepanlou M, Emadi M (2018) Impact of
land use change on soil erodibility. Global J Environ Sci Manage
4(1):59–70
Teague WR, Dowhower SL, Baker SA, Haile N, DeLaune PB, Cono-
ver DM (2011) Grazing management impacts on vegetation, soil
biota, and soil chemical, physical, and hydrological properties in
tall grass prairie. Agr Ecosyst Environ 141(3–4):310–322
Trimble SW, Mendel AC (1995) The cow as a geomorphic agent—a
critical review. Geomorphology 13(1–4):233–253
Usta A (2022) Prediction of soil water contents and erodibility indices
based on artificial neural networks: using topography and remote
sensing. Environ Monit Assess 194(11):794
Walling DE (1983) The sediment delivery problem. J Hydrol
65(1–3):209–237
Wang Y, Ran L, Fang N, Shi Z (2018) Aggregate stability and asso-
ciated organic carbon and nitrogen as affected by soil erosion
and vegetation rehabilitation on the Loess Plateau. Catena
167:257–265
Wischmeier WH, Mannering JV (1969) Relation of soil properties to
its erodibility. Soil Sci Soc Am J 33(1):131–137
Woldemariam GW, Iguala AD, Tekalign S, Reddy RU (2018) Spatial
modelling of soil erosion risk and its implication for conservation
planning: the case of the Gobele Watershed, East Hararghe zone.
Ethiopia Land 7(1):25
Yeoh ZX (2023) Assessment of soil erosion based on satellite remote
sensing data. Doctoral dissertation, UTAR
Young RA (1972). The role of rainfall impact and surface flow in soil
detachment and transport
Yuan J, Liu X, Li H, Wang R, Luo X, Xing L, ... Zhao H (2023) Assess-
ment of Spatial–Temporal Variations of Soil Erosion in Hulunbuir
Plateau from 2000 to 2050.Land,12(6), 1214
Zaady E, Arbel S, Barkai D, Sarig S (2013) Long-term impact of agri-
cultural practices on biological soil crusts and their hydrological
processes in a semiarid landscape. J Arid Environ 90:5–11
Zhang JC, DeAngelis DL, Zhuang JY, Zhang JC, DeAngelis DL,
Zhuang JY (2011) Spatial variability of soil erodibility (K factor)
at a catchment scale in Nanjing, China. In: Theory and practice
of soil loss control in Eastern China, pp 101–113
Zhang X, Hu M, Guo X, Yang H, Zhang Z, Zhang K (2018) Effects of
topographic factors on runoff and soil loss in Southwest China.
Catena 160:394–402
Zhang Y, Zhang P, Liu Z, Xing G, Chen Z, Chang Y, Wang Q (2024)
Dynamic analysis of soil erosion in the affected area of the lower
Yellow River based on RUSLE model. Heliyon 10(1):e23819
Publisher's Note Springer Nature remains neutral with regard to
jurisdictional claims in published maps and institutional affiliations.
Content courtesy of Springer Nature, terms of use apply. Rights reserved.

1.
2.
3.
4.
5.
6.
Terms and Conditions
Springer Nature journal content, brought to you courtesy of Springer Nature Customer Service Center GmbH (“Springer Nature”).
Springer Nature supports a reasonable amount of sharing of research papers by authors, subscribers and authorised users (“Users”), for small-
scale personal, non-commercial use provided that all copyright, trade and service marks and other proprietary notices are maintained. By
accessing, sharing, receiving or otherwise using the Springer Nature journal content you agree to these terms of use (“Terms”). For these
purposes, Springer Nature considers academic use (by researchers and students) to be non-commercial.
These Terms are supplementary and will apply in addition to any applicable website terms and conditions, a relevant site licence or a personal
subscription. These Terms will prevail over any conflict or ambiguity with regards to the relevant terms, a site licence or a personal subscription
(to the extent of the conflict or ambiguity only). For Creative Commons-licensed articles, the terms of the Creative Commons license used will
apply.
We collect and use personal data to provide access to the Springer Nature journal content. We may also use these personal data internally within
ResearchGate and Springer Nature and as agreed share it, in an anonymised way, for purposes of tracking, analysis and reporting. We will not
otherwise disclose your personal data outside the ResearchGate or the Springer Nature group of companies unless we have your permission as
detailed in the Privacy Policy.
While Users may use the Springer Nature journal content for small scale, personal non-commercial use, it is important to note that Users may
not:
use such content for the purpose of providing other users with access on a regular or large scale basis or as a means to circumvent access
control;
use such content where to do so would be considered a criminal or statutory offence in any jurisdiction, or gives rise to civil liability, or is
otherwise unlawful;
falsely or misleadingly imply or suggest endorsement, approval , sponsorship, or association unless explicitly agreed to by Springer Nature in
writing;
use bots or other automated methods to access the content or redirect messages
override any security feature or exclusionary protocol; or
share the content in order to create substitute for Springer Nature products or services or a systematic database of Springer Nature journal
content.
In line with the restriction against commercial use, Springer Nature does not permit the creation of a product or service that creates revenue,
royalties, rent or income from our content or its inclusion as part of a paid for service or for other commercial gain. Springer Nature journal
content cannot be used for inter-library loans and librarians may not upload Springer Nature journal content on a large scale into their, or any
other, institutional repository.
These terms of use are reviewed regularly and may be amended at any time. Springer Nature is not obligated to publish any information or
content on this website and may remove it or features or functionality at our sole discretion, at any time with or without notice. Springer Nature
may revoke this licence to you at any time and remove access to any copies of the Springer Nature journal content which have been saved.
To the fullest extent permitted by law, Springer Nature makes no warranties, representations or guarantees to Users, either express or implied
with respect to the Springer nature journal content and all parties disclaim and waive any implied warranties or warranties imposed by law,
including merchantability or fitness for any particular purpose.
Please note that these rights do not automatically extend to content, data or other material published by Springer Nature that may be licensed
from third parties.
If you would like to use or distribute our Springer Nature journal content to a wider audience or on a regular basis or in any other manner not
expressly permitted by these Terms, please contact Springer Nature at
onlineservice@springernature.com