You are on page 1of 11

Catena 196 (2021) 104902

Contents lists available at ScienceDirect

Catena
journal homepage: www.elsevier.com/locate/catena

Soil erosion modeling using erosion pins and artificial neural networks T
a,⁎ b c
Vahid Gholami , Hossein Sahour , Mohammad Ali Hadian Amri
a
Department of Range and Watershed Management and Dept. of Water Eng. and Environment, Faculty of Natural Resources, University of Guilan, Sowmeh Sara, Guilan,
Iran
b
Department of Geological and Environmental Sciences, Western Michigan University, Kalamazoo, MI 49008, USA
c
Department of Soil Conservation and Watershed Management, Mazandaran Agricultural and Natural Resources Research and Education Center, Agricultural Research,
Education and Extension Organization (AREEO), Sari, Iran

ARTICLE INFO ABSTRACT

Keywords: Assessment of soil erosion is crucial for any long-term soil conservation plan. Traditional in-situ measurements
Splash erosion provide a precise amount of erosion rate; however, the procedure is costly and time-consuming when applied
Surface erosion over an extensive area. This study aimed to investigate the use of erosion pins and artificial neural networks
GFF network (ANNs) to assess the spatial distribution of annual soil erosion rates in the mountainous areas of the north of
Hillslope
Iran. First, annual surface erosion and splash erosion were measured using two types of erosion pins. Next, the
Soil erosion map
variables affecting soil erosion (vegetation canopy, the shape of slope, slope gradient, slope length, and soil
properties) were identified and estimated through field studies and analysis of a digital elevation model (DEM)
and the data set were divided into three subsets of training, cross-validation, and testing. Seven artificial neural
network algorithms were used and evaluated to estimate the annual soil erosion rates for the areas without
recorded erosion data. Finally, the modeled values were mapped in GIS, and the longitudinal profiles of soil
erosion were extracted. Findings showed that (1) Consideration should be given to the generalized feed forward
(GFF) network, given the high accuracy rate (NMSE:0.1; R-sqr:0.9) compared to other tested ANN algorithms.
(2) Vegetation canopy was found to be the most significant variable in annual soil erosion rate (R: −0.75 to
−0.85) compared to other input variables. And (3) Annual measurements of erosion pins revealed that the
splash erosion is higher (contributing 62 percent to total erosion) compared to surface runoff erosion (con-
tributing 38 percent to total erosion).

1. Introduction and result in adverse effects on agricultural and food production as well
as water resources through contamination and sedimentation in dam
Soil erosion is one of the most important types of soil degradation, reservoirs (Aldrich et al., 2005; Pierson et al., 2007; Akay et al., 2008;
causing environmental concerns in many parts of the world Gholami and khaleghi, 2013).
(Luetzenburg et al., 2020; Li et al., 2020). Soil erosion is the movement Various factors affect soil erosion including the amount and in-
and transport of soil by various agents, such as water, wind, and mass tensity of rainfall, land-use, vegetation cover, soil properties, and length
movement (Bullock, 2005). In the humid and semi-humid climatic re- and gradient of slopes (Yair and Lavee, 1974; Masson, 1971; Pastor and
gions such as the north of Iran, the dominant form of erosion is water Castro, 1995; Martınez-Casasnovas, 1998; Uson, 1998; Di Stefano et al.,
erosion (Esmaeeli Gholzom and Gholami, 2012; Khaleghi and Varvani, 2000; Pickup and Marks, 2000).
2018). Water erosion may occur as splash erosion, surface erosion, or Soil erosion has been investigated in numerous studies. Descroix
stream erosion on the hillslopes and piping or gully erosions at foot and Poulenard (1995) demonstrated that soil erosion increases with the
slopes. Human activities can also result in erosion in two ways, directly slope to some certain gradient (27%). Bohm and Gerold (1995) de-
through soil moving operations, such as quarrying, and indirectly monstrated that vegetation cover is the most significant compared to
whereby activities such as cultivation destabilize slope materials and other controlling factors such as soil properties and slope gradient.
accelerate erosion through the wind, water, or gravity. Kirkby et al. (2005) concluded that the soil erosion rate on hillslopes
The above-mentioned factors exacerbate soil degradation by redu- generally is affected by the combination of slope and vegetation.
cing soil fertility and loss of vegetation cover in forests and rangelands There are several methods for measuring soil erosion on hillslopes,


Corresponding author.
E-mail address: Gholami.vahid@guilan.ac.ir (V. Gholami).

https://doi.org/10.1016/j.catena.2020.104902
Received 15 April 2020; Received in revised form 31 August 2020; Accepted 1 September 2020
0341-8162/ © 2020 Elsevier B.V. All rights reserved.
V. Gholami, et al. Catena 196 (2021) 104902

including field plots, erosion pins, or using empirical models. Field plots identify the critical hillslopes for soil conservation plan operations.
are used to estimate soil erosion or runoff from specific rainfall events
or within a given time interval. The use of plots provides an accurate 2. Methods and materials
estimate with the choice of time intervals from a storm event to a
monthly or annual time scale. However, field plots need continuous 2.1. Study area
maintenance and reconstitution that makes the process costly and time-
consuming. The study area is a portion of the Alborz Mountains in the north of
Another method is to use erosion pins, which have been used in Iran, extending between 36°05′ N to 37° N and 53°05′ E to 53°10′ E and
previous studies and proven to be a simple and efficient tool (Schumm, has a semi-humid climate (Esmaeeli Gholzom and Gholami, 2012). The
1956; Emmett, 1965; Ranwell, 1964; Clayton and Tinker, 1971; Kirkby predominant pattern of precipitation in the area is rainfall and rarely
and Kirkby, 1974; Haigh, 1977; Shi et al., 2013; Boardman et al., 2015(. snow (Tehrani et al., 2019). The majority of precipitation events (about
The erosion pins are cheaper and easier to use compared to field plots, 70 percent) occur during the cold seasons (winter and autumn). Spring
and they are suitable for long time scales such as annual erosion rate; and winter precipitation contribute to 20 and 10 presents of annual
however, it is challenging to use erosion pins for estimating soil erosion precipitation, respectively. The mean annual precipitation is 600 mm,
in short time scales (e.g., a specific rainfall event or specific day) and the mean annual temperature is 10° C. The average slope of the
(Ireland, 1939; Emmett, 1965; Haigh, 1977; Loughran, 1989; Hancock study area is 17°. Major land-use types include forest lands, poor ran-
et al., 2008; Boardman and Favis-Mortlock, 2016; Kearney et al., 2017(. gelands, and dry farming. The mean depth of soil is about 60 cm, and it
The selection of technique or method of measurement depends on the mainly consists of silty clay loam and clay loam texture. In order to
type of erosion, the purpose, and the target accuracy of the measure- minimize the impacts of human activities, far from the residential and
ment. agricultural area, a hillslope with a length of almost 1400 m, which
Numerous studies have been conducted to measure various forms of consists of forest in uphill and rangeland in downhill, was selected to
soil erosion using erosion pins (Boardman and Favis-Mortlock, 2016). install the erosion pins (Fig. 1). Soil property is one of the main factors
According to Haigh (1977), the use of erosion pins in soil erosion of the soil erosion process. In the chosen hillslopes, three classes of soil,
measurements originated in the USA by Colbert (1956) and Schumm namely Entisols, Inceptisols, and Mollisols, were identified based on soil
(1956). They used erosion pins to estimate soil erosion rates in bad- characteristics and using the USDA’s keys to taxonomy (USDA, 2010)
lands. Erosion pins were also used to estimate various forms of erosions, and world reference base for soil resources WRB, 2014. Entisols were
such as riverbank erosion (Lawler, 1978, 1991, 1992; 1993), down-land observed in the steep slopes of the upper hill, and Entisols and Mollisols
paths (Streeter, 1975; Summer,1986), badland erosion (Clarke and were observed in the areas of gentle slopes in downhills. Soil organic
Rendell, 2006; Nadal-Romero et al., 2011; Boardman et al., 2015; matter in the study area was ranged between 1.5 and 3.6 % and 0.77 to
Hancock and Lowry, 2015), wind erosion in sand dunes (Jungerius 2.7% in the topsoil layer and the entire soil profile, respectively. Group
et al., 1981; Jungerius and Van der Meulen, 1989; Wiggs et al., 1995; C is the predominant soil hydrologic group in the study area. However,
Livingstone, 2003) and gully erosion (Harvey,1974). It was reported in group D can also be observed in the top hills. The group C (soil with
several case studies that erosion pins are an efficient method to estimate relatively heavy texture and a low infiltration) and group D (soil with
soil erosion on hillslopes (Tervuren, 1990; Harden et al., 2009; Keay- heavy texture and a very low infiltration) have a high runoff generation
Bright and Boardman, 2009). For example, Nadal-Romero et al. (2011) potential. The study area has a Udic soil moisture regime. The observed
used and tested various methods to estimate sediments in Mediterra- gravel was less than five percent in the topsoil layer.
nean badland areas. According to their results, erosion pins were the
second-best method after gauging stations. Kumar Ghimire et al. (2013) 2.2. Annual soil erosion measurement using erosion pins
used erosion pins to evaluate various types of erosion (e.g., gully ero-
sion, sheet erosion, bank erosion, and landslides) in the Himalaya re- In the present study, we used erosion pins with a length of twenty
gion. They demonstrated that the erosion rate in the Himalaya is sig- centimeters and a diameter of eight millimeters to measure the annual
nificantly high compared to other regions. rate of soil erosion. A hillslope with a length of 1400 m was selected in
Soil erosion measurement is an expensive and time-consuming the study area to install the erosion pins. The selected hillslope was
process, and it can not be performed over an extensive study area. To heterogeneous in terms of the intensity of erosion, slope, vegetation
address this issue, field-based measurement models can be used. These cover, and soil properties. Additionally, we selected eleven locations
models are based on establishing a relationship between estimated soil with a high erosion rate in surrounding hillslopes and installed erosion
erosion (as output or target) and affecting factors in soil erosion (e.g., pins for performance assessments.
slope, vegetation cover, soil properties) in the field plot or erosion pins In order to measure soil erosion, pins with a similar length of 20 cm
area and applying those relationships to the rest of the study area with were used. The depth of erosion pins in the soil (13 cm) was accurately
unknown erosion rates. Therefore, Modeling methods such as artificial measured, and the pins were colored. We had two types of erosion pins,
neural networks (ANN)s in conjunction with field data can reduce the pins with metal caps and pins without the caps (Fig. 1). The pins with
costs and workload of the process and enables us to estimate the soil cap were paired with the ones without caps and installed to a depth of
erosion throughout the study area. ANNs have been widely used in 13 cm in the soil with minimal disturbance of the soil conditions. The
hydrological and environmental modeling processes (Anctil and Rat, erosion pins were tested before installation on the hillslope. We tested
2005; Isik et al., 2013; Gholami et al., 2018; Sahour et al., 2020; pins with different sizes and angles of installation through simulation of
Alshehri et al., 2020). Rosa et al. (1999) evaluated the interaction be- rainfalls using a watering can. The optimum height (7 cm) was the
tween land-cover and land-use characteristics and soil erosion using minimum height from the earth's surface that could protect the soil
decision trees and ANNs. Harris and Boardman (1998) used expert from raindrops (According to the dimensions of the caps). If we had
systems in conjunction with ANNs as an alternative paradigm to mea- increased the heights of the pins, the metal caps should have been
surable process-based erosion modeling for the South Downs in Eng- larger. A significant difference in erosion rate was observed between
land. Moreover, geographic information system (GIS) is a useful tool for two types of pins (pins with caps and pins without caps) which shows
mapping and analyzing the soil erosion intensity when coupled with the effectiveness of the caps in protecting the soil beneath them. Also,
ANNs (Martınez-Casasnovas, 1998; Dixon, 2004; Zhao et al., 2009; the direction of rainfall is an important criterion in the installation of
Rosas and Gutierrez, 2020). the pins. The slope of the study area was gentle; therefore, the pins were
The objective of this study is to model, map, and analyze the soil installed almost in a vertical position.
erosion using erosion pins and an optimum ANN coupled to a GIS and In total, 109 pairs (218 pins) with an average distance of 10 m

2
V. Gholami, et al. Catena 196 (2021) 104902

Fig. 1. (A) Location of the study area. (B, C) Location of the erosion pins on the hillslopes. (C) A view of the erosion pins in the study area.

between pins were installed in the study area in April 2018. The in- after one year (April 2019) by measuring the soil surface level from the
tervals were ranging from 2 to 12 m, considering the changes in land erosion pins. We evaluated surface and splash erosions, but stream
cover, soil erosion intensity, and topography. The majority of pins, erosion was not considered in our study. Pins with no caps show the
however, were installed with 10 m intervals. The distance between the total soil erosion (sum of splash erosion and surface runoff erosion), and
two pins in each pair was 30 cm. Further, the distance between pins is pins with caps indicate the surface (runoff) erosion. The difference
determined based on the time intervals of recording and the purpose of between total erosion (from pins with no caps) and surface runoff
the study (Boardman & Favis-Mortlock, 2016). We recorded the soil loss erosion (from pins with caps) determines the rate of the splash erosion

3
V. Gholami, et al. Catena 196 (2021) 104902

because the caps protected the soil below from splash erosion. More- The PCA is an unsupervised linear method that finds a set of un-
over, the soil texture was mapped by determining soil units and drilling correlated features, from the inputs (Beghdad, 2008). The TLRN net-
soil profiles. Soil units were determined through field studies by drilling work is an MLP based algorithm with short term memory structures.
ten soil profiles and coupling geomorphologic and geologic maps. Soil The TLRNs are advanced in nonlinear time series forecasting, system
texture (clay, silt, and sand), as well as soil specific weight, were de- identification, and temporal pattern classification (Sattari et al., 2011).
termined in a soil laboratory, and the mean soil specific weight was In the modeling process, the data were first normalized and then
estimated to be 1.4 ton.m3. After determining the rate of splash erosion, divided into three subsets of training (65% of the data), cross-validation
surface runoff erosion, and total erosion, the annual soil erosion rate (in (10% of the data), and testing (25% of the data). A trial-and-error ap-
kg.m2) was estimated by applying the specific weight in a 1 m2 of the proach was used to define the optimal structure of the network (the
surface area. optimum inputs, learning algorithm, and the number of neurons) and
the number of the training epochs in the training stage (Isik et al.,
2.3. Annual soil erosion modeling using ANN 2013). The optimal structure of the selected networks was determined
using the trial–error approach and evaluating the error. In the tria-
We applied and evaluated several ANNs to find the optimal model l–error approach, each time, one of the network components is changed
for estimation of soil erosion in the areas with missing data (area while others are fixed. The optimal network parameters are the ones
without erosion pins). The output (or target variable) in our ANN with the minimum error values. The input variables were analyzed
models was the annual soil erosion rate. The independent variables (or using the correlation coefficient (SPSS software). The optimal inputs are
the predictors) were the slope gradient, slope length, curvature (shape vegetation canopy, slope shape (curvature), slope gradient, slope
of the slope), soil properties (clay percentage, silt percentage, and sand length, and soil properties (percentage of clay and sand). Rainfall is
percentage), and vegetation cover. another important factor that can trigger soil erosion. However, the
As we explained earlier, the soil erosion (output) was determined amount and the intensity of rainfall is the same throughout the study
through field studies by establishing the erosion pins. For the location area, and therefore, it could not be incorporated in our modeling pro-
of each pin, the slope gradient, the shape of slopes (curvature), and cess. The models were evaluated by comparing the observed and the
length of slopes were estimated from a digital elevation model (DEM) simulated values using the mean squared error (MSE) and the coeffi-
with a 10-meter spatial resolution in the ArcGIS. cient of determination (R-squared). Finally, the tested models were
The percentage of clay, silt, and sand were derived from the soil used to estimate the annual soil erosion in the rest of the study areas
texture map in the GIS environment. We had ten soil profiles on the with no erosion pins.
selected hillslope and twenty profiles on the rest of the study area.
Vegetation cover was ranging from woodlands in the upper hill and a 2.4. Mapping annual soil erosion rate in GIS
combination of woodlands and rangelands in the middle to rangeland
(grasslands) in the downhill. The spatial distribution of 109 pair pins After the modeling process (training, cross-validation, and test), the
regarding the types of land use included 10 pins in forest lands, 40 pins ANN and GIS were coupled to simulate and map annual soil erosion
in wooden rangelands, 12 pins in rich rangelands, and 47 in poor rates. The GFF network was selected as the optimum network for soil
rangelands. In order to provide the vegetation canopy percentage, erosion modeling. One of the objectives of our study was to couple the
homogeneous vegetation units were identified by field studies in the ANN to GIS in order to map the annual soil erosion rates and to extract
study area. Then, 10 random field-sampling plots with dimensions of the longitudinal profile of soil erosion on the hillslopes. For this pur-
1 × 1 m (for grass species) and 5 × 5 m (for tree species) were used to pose, the numerical values were extracted from inputs in raster formats,
determine the vegetation canopy percentage for each unit. To measure and the soil erosion rates were simulated using the input variables and
the percentage of canopy, the ratio of the vegetated area to the non- the validated GFF network for the entire study area. Next, the simulated
vegetated area in each plot was calculated. Finally, the canopy per- erosion rates were transferred back to GIS to convert the numerical
centage for each unit was estimated based on the mean canopy per- values to the raster format to map the annual soil erosion rate. In order
centage of ten used plots. to assess the accuracy of mapping, the soil erosion rates of the eleven
Different ANNs including multilayer perceptron (MLP), generalized validation pins were compared with the corresponding simulated va-
feed forward (GFF), jordan and elman network (JE), radial basis func- lues on the final map. Finally, the longitudinal profiles of the studied
tion (RFB), modular feed forward (MFF), principal component analysis hillslope and annual soil erosion were extracted from DEM and the final
(PCA) and time lagged recurrent (TLRN) were used for estimating an- map (raster layer of soil erosion) in GIS, respectively. The workflow of
nual soil erosion. MLP network is a layered feed forward network ty- the ANN-based soil erosion modeling was given in Fig. 2.
pically trained with the backpropagation algorithm. These networks
have found their way into numerous applications requiring static pat- 3. Results
tern classification. The major drawbacks of those networks are that they
are trained slowly and require lots of training data (Gholami et al., 3.1. Measurement of the annual soil erosion using erosion pins
2018). The GFF network is a generalization of the MLP such that con-
nections can jump over one or more layers. In theory, an MLP can re- As we mentioned earlier, we used erosion pins to measure annual
solve any problem that a GFF network does. In practice, however, GFF soil erosion on the hillslope. We recorded the measurements one year
often solves the problems much more efficiently (Beghdad, 2008). The after we established the pins in the study area. The values for surface
JE network extends the MLP with context units, which are processing erosion, splash erosion, and total erosion, as well as their corresponding
elements (PEs) that remember earlier activity. Context units provide the input values for some arbitrary points, were presented in Table 1. Ac-
network with the ability to derive temporal information from the data cording to the measurements, annual soil erosion in the study area was
(Mohamed and Atta, 2010). The RBF network is a nonlinear hybrid ranged from 0 to 33.6 kg.m−2 (average: 9.4 kg.m−2). The annual sur-
network typically comprising a single hidden layer of the processing face runoff erosion was ranged between 0 and 14 kg.m−2, and the
elements (PEs). The learning process in the RBF is much faster com- splash erosion was ranged from 0 to 19.6 kg.m−2. According to the
pared to MLP. The inputs in the MFF network are processed using results, the erosion caused by the raindrops was more pronounced (62
several parallel MLPs and then merge the results. This tends to build percent of the total mean soil erosion) compared to erosion caused by
some structure within the topology, which develops specialization of runoff (38 percent of the total mean soil erosion). We also used corre-
function in each sub-module (Beghdad, 2008). The PCA network lation analysis to investigate the role of the independent variables in
combines unsupervised and supervised learnings in the same topology. soil erosion rates, and the results were given in Table 2. In general, the

4
V. Gholami, et al. Catena 196 (2021) 104902

Fig. 2. The workflows of soil erosion modeling using erosion pins and ANNs.

correlation between independent variables and soil erosions are similar values using the GFF network was presented in Fig. 4.
for each type of erosion (soil erosion, surface erosion, splash erosion).
Spatial distribution of the percentage of vegetation canopy and the
shape of slopes (curvature index) in the study hillslopes are given in 3.3. Coupling ANN and GIS
Fig. 3. The results showed that the most important factors in soil ero-
sion were vegetation cover (correlation coefficient: −0.75 to 0.85) and The optimal inputs of soil erosion modeling are vegetation canopy,
curvature or shape of slope (correlation coefficient: 0.56 to 0.65). The slope shape (curvature), slope gradient, slope length, and soil properties
results also indicated that splash erosion contributed more to soil ero- (percentage of clay and sand). The percentage of silt was removed from
sion compared to surface runoff erosion. It is noteworthy to mention the sets of input variables given its insignificance and low correlation.
that rainfall is the predominant type of precipitation in the study area Using the optimum network (the GFF), optimum inputs and optimum
and rainstorm with high intensity and short duration is very common. structure (transfer function, learning technique, neuron number), we
Further investigations revealed that the sediments are trapped in the estimated the soil erosion for the entire study area, and the results were
area with concave slopes, and the convex slopes are the area of higher imported to GIS for mapping and analysis. Fig. 5 shows the annual soil
erosion (Di Stefano et al., 2000). erosion rate map for the study area. The verification pins were overlaid
on the final map and a QuickBird satellite image to evaluate the map
accuracy. We added the satellite image to show the estimation rate
3.2. Annual soil erosion rate modeling using ANN
associated with different land-use and land-cover (Fig. 6). Results
showed that the areas without erosion (zero value for soil erosion rate)
As we mentioned earlier, we applied and tested different ANN
on the map are associated with the area of densely vegetated area and
models through a trial–error approach, and results were evaluated to
low slope gradient. On the other hand, the high erosion rates were
select the optimum procedure. Evaluation of the adopted methods in
observed in the areas with minimum vegetation cover and a high slope
training and cross-validation was given in Table 3. The final decision on
gradient. The results are consistent with our correlation analysis that
selecting the optimum model should be based on the performance of the
was presented in Table 2. Therefore, adopted methodology in this study
model on the test subset (Samani et al., 2007; Gholami et al., 2018).
can be used for soil erosion mapping and analysis in the area with
The evaluation of the networks on the test subset was presented in
missing soil erosion data. And finally, using a DEM and the generated
Table 4. The performance assessment on the test subset showed that the
soil erosion map (both with a spatial resolution of 10 m × 10 m), we
GFF network performed better given to the higher R-squared, and lower
generated the longitudinal profiles of the slope and erosion rate for one
NMSE (R-squared: 0.9, and NMSE: 0.12) values compared to other ANN
of the studied hillslopes (Fig. 7).
networks. The comparison between the observed values and simulated

5
V. Gholami, et al. Catena 196 (2021) 104902

Table 1
Erosion rates and ANN inputs for a few arbitrarily selected pins.
Erosion height (mm) Surface erosion Splash erosion height (mm) Slope° Curvature Slope length (m) Clay Silt Canopy Sand
height (mm) Slope shape° (%) (%) (%) (%)

0.5 0.2 0.3 8 0.19 400 35.4 44.9 50 19.7


0.1 0 0.1 15.1 −0.47 1331 38.4 41.2 90 20.4
0.7 0.3 0.4 8 0.31 431 35.4 44.9 30 19.7
0.3 0.1 0.2 5.8 −0.05 1397 38.4 40.4 80 21.2
0.7 0.3 0.4 7.9 −0.11 467 35.4 44.9 32 19.7
0.6 0.2 0.4 8.2 0.28 481 35.4 44.9 40 19.7
0.5 0.3 0.2 8.5 0.309 495 35.4 44.9 42 19.7
1 0.3 0.7 11.6 0.55 645 35.4 44.9 20 19.7
0.3 0 0.3 10.75 −0.44 1346 38.4 40.4 90 21.2
0.9 0.4 0.5 18.7 0.72 540 35.4 44.9 25 19.7
1 0.5 0.5 21.5 0.85 552 37.9 39.4 20 22.7
1.2 0.5 0.7 27.3 0.67 653 37.9 39.4 15 22.7
1.1 0.5 0.6 24.6 0.19 600 37.9 39.4 15 22.7
1 0.4 0.6 20.4 0.03 622 37.9 39.4 25 22.7
0.7 0.3 0.4 17.7 −0.108 645 37.9 39.4 30 22.7
0.8 0.4 0.4 18.6 0.27 657 37.9 39.4 33 22.7
0.9 0.4 0.5 20.4 0.28 668 37.9 39.4 25 22.7
0.7 0.3 0.4 19.65 −0.08 688 37.9 39.4 30 22.7
0.6 0.2 0.4 14.88 −0.13 702 37.9 39.4 37 22.7
0.65 0.3 0.35 23.6 0.508 1247 33.41 43.06 35 23.53
1.1 0.5 0.6 25.5 0.76 1252 33.41 43.06 25 23.53
0.7 0.3 0.4 9 0.02 766 37.9 39.4 42 22.7
0.6 0.2 0.4 7.48 0.04 775 37.9 39.4 40 22.7
0.9 0.4 0.5 9.5 0.11 793 37.9 39.4 35 22.7
0.2 0 0.2 12.1 −0.46 1339 37.4 40.2 90 22.4
0.7 0.3 0.4 8 0.32 819 37.9 39.4 45 22.7
0.6 0.2 0.4 8.2 0.49 829 37.9 39.4 50 22.7
0.55 0.25 0.3 8.6 0.49 838 37.9 39.4 55 22.7
0.7 0.3 0.4 8 0.75 849 37.9 39.4 40 22.7
0.3 0.1 0.2 15.9 −0.13 1258 33.4 44.4 45 22.2
1.2 0.5 0.7 14.2 0.87 878 37.41 41.53 22 21.06
1.1 0.5 0.6 16.1 0.78 883 37.41 41.53 20 21.06
1 0.4 0.6 15.3 0.81 882 37.41 41.53 23 21.06
0.9 0.3 0.6 14.79 0.8 888 37.41 41.53 25 21.06
0.4 0.1 0.3 7 −0.36 1413 37.4 44.2 50 18.4
0.9 0.3 0.6 18.1 0.54 907 37.41 41.53 25 21.06
0.8 0.3 0.5 18.7 0.39 919 37.41 41.53 32 21.06
1.1 0.4 0.7 19.3 0.4 930 37.41 41.53 20 21.06
0.6 0.2 0.4 21.6 0.62 942 37.41 41.53 30 21.06
0.55 0.35 0.2 21.6 0.233 950 37.41 41.53 60 21.06
0 0 0 3.85 −0.33 1392 38.4 41.2 100 20.4
0.8 0.3 0.5 22.7 0.84 970 37.41 41.53 25 21.06
0.7 0.2 0.5 24 0.83 978 37.41 41.53 30 21.06
0 0 0 3.85 −0.33 1392 38.4 41.2 100 20.4
1 0.4 0.6 25.7 0.7 1018 37.41 41.53 20 21.06
0.9 0.3 0.6 24.8 0.79 1028 37.41 41.53 24 21.06

4. Discussion with soil erosion. Vegetation, silt, and clay percentage have a negative
correlation (inverse relationship) with soil erosion rate. According to
In this study, the annual erosion rates were presented as splash the past studies, vegetation, clay, and silt have a positive relationship
erosion, surface runoff erosion, and total soil erosion. According to our with soil erosion (Masson, 1971; Yair and Lavee, 1974; Bohm and
correlation analysis, the most significant variables in soil erosion are Gerold,1995; Foster, 2001) and slope, curvature, slope length, and sand
vegetation canopy percentage, curvature (shape of the slope), slope percent have a negative relationship with soil erosion (Descroix and
length, slope gradient, and soil texture (Clay and sand percent), re- Poulenard 1995; Kirkby et al., 2005). The shape of the slope (curvature)
spectively. Among those variables, curvature, slope gradient, slope is another crucial parameter affecting soil erosion. Di Stefano et al.
length, and sand value have a positive correlation (direct relationship) (2000) mentioned that shape of the slope could be a determinative

Table 2
Correlation coefficients between the input variables and different types of soil erosion.
Pearson correlation with surface erosion Pearson correlation with splash erosion Pearson correlation with soil erosion Parameter

0.2 0.36 0.30 Slope gradient


0. 43 0.32 0.40 Slope length
0.56 0.64 0.65 Curvature (slope shape)
−0.75 −0.83 −0.85 Vegetation canopy
−0.20 −0.22 -0.22 Clay %
−0.04 * −0.03 * −0.04* Silt %
0.25 0.23 0.25 Sand %
* Not Significant

6
V. Gholami, et al. Catena 196 (2021) 104902

Fig. 3. (A) The slope shapes (curvature index) in the study area. (B) Spatial distribution of percent canopy in the study area.

factor in runoff velocity and soil erosion process. Measurements showed of the pins in each pair had a metal cap (Fig. 1) to conserve the soil from
that soil erosion is relatively lower in the area with concave slopes the impact of raindrops. Therefore, the amount of recorded erosion for
compared to the surfaces with convex slopes (Gray, 2016). Also, the those pins was only related to surface erosion. Recording the paired
effect of curvature on the amount of soil erosion from surface runoff is erosion pins revealed that splash erosion is the dominant type of ero-
more significant than the erosion caused by splashing. In general, sion in the study area compared to the erosion caused by surface runoff.
spatial variation of soil erosion, the spatial distribution of rainfall and Therefore, rangeland vegetation can play an essential role in soil ero-
changes in soil properties (clay, silt, and sand values) have significant sion control, especially splash erosion (Chen et al., 2019).
effects on soil erosion (Yair and Lavee, 1974; Wischmeier and Smith, In our study, factors affecting soil erosion were used as inputs, and
1978; Bohm and Gerold, 1995; Keim et al., 2006). annual erosion rates were simulated as the neural network outputs.
As we mentioned earlier, we used paired erosion pins in which one Modeling results showed that high performance of the adopted

7
V. Gholami, et al. Catena 196 (2021) 104902

Table 3
Performance evaluation of the different ANNs at the training and cross-validation stages for modeling the annual soil erosion.
R-squared Training Standard Deviation Final MSE The mean MSE Modeling ANN
stage

0.97 0.0004 0.005 0.005 Training Generalized feed forward (GFF)


0.0003 0.001 0.001 Cross-validation
0.97 0.0004 0.005 0.005 Training Multilayer perceptron
0.0003 0.002 0.001 Cross-validation (MLP)
0.92 0.0001 0.007 0.007 Training Jordan and Elman networks (JEN)
0.0004 0.01 0.02 Cross-validation

0.9 0.003 0.004 0.004 Training Radial basis function


0.005 0.002 0.001 Cross-validation (RFB)
0.9 0.0009 0.007 0.007 Training Modular feed forward (MFF)
0.003 0.003 0.002 Cross-validation
0.93 0.001 0.007 0.007 Training Principal component analysis (PCA)
0.002 0.007 0.002 Cross-validation

0.93 0.009 0.008 0.008 Training Time lagged recurrent (TLRN)


0.001 0.05 0.08 Cross-validation

Table 4 other relevant datasets (DEM and satellite imagery) revealed that the
Performance evaluation of the different ANNs at the testing stage for modeling area of low erosion rates are associated with the areas with low slope
the annual soil erosion. gradient and densely vegetated areas, and the area with high slope
R-squared Max Abs Error Min Abs Error MAE NMSE MSE ANN gradient and low vegetation cover have relatively higher erosion rates
(Sun et al., 2016).
0.9 2.9 0.04 1.39 0.12 2.71 GFF Converting the simulated values to the raster format in the GIS and
0.77 6.4 0.21 1.8 0.21 5.1 MLP
evaluating the longitudinal profiles of land slope and longitudinal
0.81 4.6 0.1 1.71 0.19 4.2 EN
0.68 8.9 0.07 2.08 0.33 7.2 RFB profiles of soil erosion is an important applied result in the present
0.77 6.76 0.21 1.7 0.21 4.6 MFF studies. Monitoring changes in erosion rate along the slopes is an es-
0.81 4.4 0.18 2 0.24 5.2 PCA sential factor for planning soil conservation practices such as biological
0.8 5.6 0.33 1.65 0.19 4.19 TLRN treatments (Zhang et al., 2020). As shown in Fig. 5, soil erosion in-
creases as the length of slope and, consequently, the volume and ve-
locity of runoff increases. However, this amount of soil erosion de-
methodology in the estimation of soil erosion rates in a short time with
creases in the foot slope due to the change in the gradient of the slope.
minimal costs. This methodology can be used in other parts of the
(Las Heras et al., 2010). Therefore, it is crucial to identify the con-
world, especially developing countries with a high rate of soil erosion
trolling factors in the process of soil erosion to plan for soil conservation
and financial limitations (Samani et al., 2007; Licznar and Nearingb,
operations. Finally, monitoring of soil erosion changes along the slopes
2003; Zhao et al., 2009).
is a useful practice for identifying locations with higher soil erosion rate
Evaluation of the ANN performance in soil erosion modeling was
to prioritize the biological treatments or the other protective methods.
presented in Figs. 2 through 4. The evaluation was conducted in two
Compared to past studies, the adopted methodology and the obtained
steps of validation of the model and verification. In the test or valida-
results have several advantages, including estimation in a short time,
tion step, the simulated values were compared to the observed values in
high accuracy, adaptability of the methodology for similar settings
the test subset (25 percent of the data) using statistical coefficients
elsewhere, and providing practical results for mapping soil erosion rate
(MSE, NRMSE, R-squared). In the second step, the estimated soil ero-
for planning soil conservation projects.
sion rate throughout the study area was verified using the eleven ver-
ification pins that were randomly distributed on the surrounding hill-
5. Conclusion
slopes. The estimated rates from the soil erosion map are consistent
with the recorded values from the verification pins (R-squared = 0.9).
ANN-based models can be a powerful tool for estimating hydro-
In other words, the performance of the adopted methodology was
logical or soil erosion parameters. Nevertheless, what is essential in this
evaluated in two stages of testing and verification. In the majority of the
regard is choosing a suitable method or algorithm. In the first step of
ANN-based soil erosion modeling, the performances of the models were
estimating soil erosion, it is necessary to select the erosion measure-
only evaluated in the testing stage.
ment method according to the objectives of the study and the type of
Moreover, analysis of the spatial distribution of the soil erosion with
erosion that needs to be investigated. Both erosion pins and erosion

Fig. 4. Comparison between the predicted annual soil erosion (using the GFF network) and the observed values in the test subset (R-squared = 0.9).

8
V. Gholami, et al. Catena 196 (2021) 104902

Fig. 5. Map of annual soil erosion rate (kg/m2) for the study area.

plots are types of field-based measurements that have proven to be very artificial intelligence-based models, such as neural networks, precise
effective for measuring annual erosion. However, on the areas with and fast modeling is an advantageous feature. The study provides a
steep slopes, it is more beneficial to use erosion pins due to the cost of feasible conventional technique to assess the erosion rate and identify
operations and performance ease, drastic topographic and vegetation the critical areas for prioritizing the conservation plans such as biolo-
changes (Kirkby and Kirkby, 1974; Las Heras et al., 2010; Shi et al., gical treatments using the generated erosion map and longitudinal soil
2013; Boardman et al., 2015). profile. Additionally, identifying the main factors of erosion and the
Combining an ANN-based model, soil erosion pins, and GIS-based dominant form of erosion is crucial in managing and controlling soil
outputs was a reliable, inexpensive, and easy to use approach for ac- erosion. If the erosion is mainly caused by raindrops (splash erosion),
curate modeling of the annual soil erosion across the study area. The we should focus on rangeland vegetation across the hillslope, and if the
present methodology has some advantages in estimating soil erosion. In erosion is due to surface runoff flow, priority should be given to the foot
terms of measurements, the ease of use, the speed of the process, and slopes due to the maximum velocity and volume of runoff (Descroix and
the low cost of operation can be noted. GIS as a pre-processor and post- Poulenard, 1995; Kirkby et al., 2005; Gholami et al., 2018). Moreover,
processor is also a powerful tool for analyzing the results. Another the results showed that land use changes and vegetation degradation
advantage is the ability to integrate the neural network inputs and are the most important factors of soil erosion increase (Khaleghi, 2017;
outputs with GIS for spatial analysis or soil erosion mapping. In Naghdi et al., 2017). The maximum increase in soil erosion was in the

Fig. 6. The map of annual soil erosion rate (kg/m2) overlaid with satellite image and validation erosion pins (the observed erosion values) for evaluating the results.

9
V. Gholami, et al. Catena 196 (2021) 104902

Fig. 7. Longitudinal profiles of slope and erosion for one of the hillslopes in the study area. High soil erosion rates on the profile (length 500–900 m) are associated
with the areas with a steep slope and poor vegetation cover (lower half of the study slopes). Minimum soil erosions in the sites are associated with the area of dense
vegetation cover and minimum slope gradient, or concaved surfaces.

areas of deforestation or the areas that rangelands changed to farming Descroix, L., Poulenard, J., 1995. Les formes d’érosion dans la Sierra Madre Ocidentale
lands. For future studies, we suggest using a combination of erosion (Nord Ouest du Mexique). Bull. Lab. Rhod. Géomorphol. 33-34, 1-19, Lyon.
Di Stefano, C., Ferro, V., Porto, P., Tusa, G., 2000. Slope curvature influence on soil
pins and erosion plots and artificial intelligence-based models in com- erosion and deposition processes. Water Resour. Res. 36 (2), 607–617.
plex hillslopes to model and map other types of soil erosion (e.g., gully Dixon, B., 2004. Prediction of groundwater vulnerability using an integrated GIS-based
erosion, stream erosion, and riverbank erosion). neuro-fuzy techniques. J. Spa. hydrol. 14 (12), 1–38.
Emmett, W.W., 1965. The Virgil Network: methods of measurement and a sampling of
data collected. Int. Assoc. Sci. Hydrol. Publ. 66, 89–106.
Declaration of Competing Interest Esmaeeli Gholzom, H., Gholami, V., 2012. A comparison between natural forests and
reforested lands in terms of runoff generation potential and hydrologic response (case
study: Kasilian Watershed). Soil Water Res. 4, 166–173.
The authors declared that there is no conflict of interest. Gholami, G., Khaleghi, M.R., 2013. The impact of vegetation on the bank erosion (case
study: the Haraz river). Soil Water Res. 8 (4), 158–164.
Gholami, V., Booij, M.J., Tehrani, E.N., Hadian, M.A., 2018. Spatial soil erosion estima-
Acknowledgments
tion using an artificial neural network (ANN) and field plot data. Catena. 163,
210–218.
We would like to thank the Natural Resources Organization and Gray, D., 2016. Effect of slope shape on soil erosion. J. civil environ. eng. Vol(6) 3,
Watershed Management of Mazandaran for providing the topographic 1000231. https://doi.org/10.4172/2165-784X.1000231.
Foster, G.R., 2001. Keynote: soil erosion prediction technology for conservation planning.
data. In: Stott, D.E., Mohtar, R.H., Steinhartdt, G.C. (Eds.), Proceedings of the Sustaining
the Global Farm. Selected papers from the 10th International Soil Conservation
References Organization Meeting. Purdue University and the USDA-ARS National Soil Erosion
Research Laboratory, 24–29 May 1999.
Haigh, M.J., 1977. The use of erosion pins in the study of slope evolution. In, Shorter
Akay, A.E., Erdas, O., Reis, M., Yuksel, A., 2008. Estimating sediment yield from a forest Technical Methods (ll). Technical Bulletin No. 18, British Geomorphological Research
road network by using a sediment prediction model and GIS techniques. Build. Group. Geo Abstracts: Norwich, UK; 31-49.
Environ. 43, 687–695. Hancock, G.R., Loughran, R.J., Evans, K.G., Balog, R.M., 2008. Estimation of soil erosion
Aldrich, G.A., Tanaka, J.A., Adams, R.M., Buckhouse, J.C., 2005. Economics of western using field and modelling approaches in an undisturbed Arnhem land catchment,
juniper control in central Oregon. Rangeland Ecol. Manage. 58, 542–552. Northern Territory, Australia. Geogr. Res. 46 (3), 333–349. https://doi.org/10.1111/
Alshehri, F., Sultan, M., Karki, S., Alwagdani, E., Alsefry, S., Alharbi, H., Sahour, H., j.1745-5871.2008.00527.x.
Sturchio, N., 2020. Mapping the Distribution of Shallow Groundwater Occurrences Hancock, G.R., Lowry, J.B.C., 2015. Hillslope erosion measurement–a simple approach to
Using Remote Sensing-Based Statistical Modeling over Southwest Saudi Arabia. a complex process. Hydrol. Process. 29, 4809–4816.
Remote Sensing 12 (9), 1361. Harden, C.P., Foster, W., Morris, C., Chartrand, K.J., Henry, E., 2009. Rates and processes
Anctil, F., Rat, A., 2005. Evaluation of neural networks streamflow forecasting on 47 of streambank erosion in tributaries of the Little River, Tennessee. Phys. Geogr. doi:
watersheds. J. Hydrol. Eng .ASCE 10(1), 85-88.DOI: 10.1061/ (ASCE) 1084- 10.2747/0272-3646.30.1.1.
0699(2005)10:1(85). Harris, T.M., Boardman, J., 1998. Alternative approaches to soil erosion prediction and
Beghdad, R., 2008. Critical study of neural networks in detecting intrusions. Computers & conservation using expert systems and neural networks. Modeling soil erosion by
Security. 27 (5), 168–175. water. NATO ASI Series I, vol. 55. Springer-Verlag, Berlin, 461 – 477.
Boardman, J., Favis-Mortlock, D.T., Foster, IDL., 2015. A 13-year record of erosion on Harvey, A.M., 1974. Gully erosion and sediment yield in the Howgill Fells, Westmorland.
badland sites in the Karoo, South Africa. Earth Surf. Proc. Land. doi: 10.1002/esp. In: Gregory, K.J., Walling, D.E. (Eds.), Fluvial Processes in Instrumented Watersheds.
3775. Institute of British Geographers, London, pp. 45–58.
Boardman, J., Favis-Mortlock, D., 2016. The use of erosion pins in geomorphology. Ireland, H.A., Sharpe, C.F.S. Eargle, D.H., 1939: Principles of gully erosion in the pied-
Geomorphological Techniques, Chap. 3, Sec. 5.3. mont of South Carolina. Technical Bulletins 167374, United States Department of
Bohm, P., Gerold, G., 1995. Pedo-hydrological and sediment responses to simulated Agriculture, Economic Research. Doi: 10.22004/ag.econ.167374.
rainfall on soils of the Konya Uplands (Turkey). Catena. 25 (1–41), 63–75. Isik, S., Kalin, L., Schoonover, J., Srivastava, P., Lockaby, B.G., 2013. Modeling effects of
Bullock , P., 2005. Climate change impact. Encyclopedia of Soils in the Environment. changing land use/cover on daily stream flow: An artificial neural network and curve
Cranfield University–Silsoe, Silsoe, UK. 254-262. https://doi.org/10.1016/B0-12- number based hybrid approach. J. Hydrol. 485, 103–112.
348530-4/00089-8. Jungerius, P.D., Verheggen, A.J.T., Wiggers, A.J., 1981. The development of blowouts in
Chen, Z., Wang, L., Wei, A., Gao, J., Lu, Y., Zhou, J., 2019. Land-use change from arable ‘De Blink’, a coastal dune area near Noordwijkerhout. The Netherlands. Earth Surf.
lands to orchards reduced soil erosion and increased nutrient loss in a small catch- Proc. Land. 6, 375–396.
ment. Sci. Total Environ. 648, 1097–1104. https://doi.org/10.1016/j.scitotenv.2018. Jungerius, P.D., van der Meulen, F., 1989. The development of dune blowouts, as mea-
08.141. sured with erosion pins and sequential air photos. Catena. 16, 369–376.
Clarke, M.L., Rendell, H.M., 2006. Process-form realtionships in southern Italian bad- Kearney, P., Fonte, S.J., Garcia, E., Smukler, M., 2017. Improving the utility of erosion
lands: erosion rates and implications for landform evolution. Earth Surf. Proc. Land. pins: absolute value of pin height change as an indicator of relative erosion. Catena.
31, 15–29. 163, 427–432. https://doi.org/10.1016/j.catena.2017.12.008.
Clayton, L., Tinker, J.R., 1971. Rates of hillslope lowering in the Badlands of North Keay-Bright, J., Boardman, J., 2009. Evidence from field-based studies of rates of erosion
Dakota. North Dakota University Water Resources Research Institute, Report on degraded land in the central Karoo, South Africa. Geomorphology 103, 455–465.
W1–221-012-71. W73.09121.N.T.I.S. PB 220 355, 1–36. Keim, R.F., Skaugset, A.E., Weiler, M., 2006. Storage of water on vegetation under si-
Colbert, E.H., 1956. Rates of erosion in the Chinle Formation. Plateau. 28 (4), 73–76. mulated rainfall of varying intensity. Adv. Water Resour. 29 (7), 974–986.

10
V. Gholami, et al. Catena 196 (2021) 104902

United States Department of Agriculture, Natural Resources Conservation Service., 2010. fromairbone gamma radiometrics and digital elevation models in a weathered
Keys to Soil Taxonomy. landscape. Earth Surf. Proc. Land. 25, 535–557.
Kirkby, A.V.T., Kirkby, M.J., 1974. Surface wash at the semi-arid break in slope. Pierson, F.B., Batees, J.D., Svejcar, T.J., Hardegree, S.P., 2007. Runoff and erosion after
Zeitschrift fur Geomorphologie Suppl. 21, 151–176. cutting western juniper. Rangeland Ecol. Manage. 60, 285–292.
Khaleghi,M.R., 2017. The influence of deforestation and anthropogenic activities on Ranwell, D.S., 1964. Spartina salt marshes in southern England 11: Rate and seasonal
runoff generation. J. For. Sci. 63, 2017 (6): 245–253. pattern of sediment accretion. J. Ecol. 52, 79–94.
Khaleghi, M.R., Varvani, J., 2018. Simulation of relationship between river discharge and Rosa, D., de la Mayol, F., Lozano, S., 1999. An expert system/neural network model
sediment yield in the semi-arid river watersheds. Acta Geophys. 66 (1), 109–119. (impelERO) for evaluating agricultural soil erosion in Andalucia region, southern
Kirkby, M.J., Bracken, L.J., Shannon, J., 2005. The influence of rainfall distribution and Spain. Agri. Ecosyst. Environ. 73 (3), 211–226.
morphological factors on runoff delivery from dryland catchments in SE Spain. Rosas, M.A., Gutierrez, R.R., 2020. Assessing soil erosion risk at national scale in devel-
Catena. 62, 136–159. https://doi.org/10.1016/j. catena.2005.05.002. oping countries: The technical challenges, a proposed methodology, and a case his-
Kumar Ghimire, S., Higaki, D., Prasad Bhattarai, T., 2013. Estimation of soil erosion rates tory. Sci. Total Environ. 703, 135474. https://doi.org/10.1016/j.scitotenv.2019.
and eroded sediment in a degraded catchment of the Siwalik Hills. Nepal, Land 2, 135474.
370–391. https://doi.org/10.3390/land2030370. Sahour, H., Sultan, M., Vazifedan, M., Abdelmohsen, K., Karki, S., Yellich, J.A.,
Las Heras, M.M., Nicolau, J.M., Martín, L.M., Bradford, P., Wilcox, B.W., 2010. Plot-scale Gebremichael, E., Alshehri, F., Elbayoumi, T.M., 2020. Statistical applications to
effects on runoff and erosion along a slope degradation gradient. Water Resour. Res. downscale GRACE-Derived terrestrial water storage data and to fill temporal gaps.
46, 1–12. https://doi.org/10.1029/2009WR007875. Remote Sens. 12, 533. https://doi.org/10.3390/rs12030533.
Lawler, D.M., 1978. The use of erosion pins in river banks. Swansea Geographer. 16, Samani, N., Gohari-Moghadam, M., Safavi, A.A., 2007. A simple neural network model for
9–18. the determination of aquifer parameters. J. Hydrol. 340, 1–11. https://doi.org/10.
Lawler, D.M., 1991. A new technique for the automatic monitoring of erosion and de- 1007/s10584-005-5922-3.
position rates. Water Resour. Res. 27 (8), 2125–2128. Sattari, M.T., Yurekli, K., Pal, M., 2011. Performance evaluation of artificial neural net-
Lawler, D.M., 1992. Process dominance in bank erosion systems. In: Carling, P., Petts, work approaches in forecasting reservoir inflow. Appl. Math. Model. 36 (6),
G.E. (Eds.), Lowland Floodplain Rivers: Geographical Perspectives. Wiley, Chichester, 2649–2657.
pp. 117–143. Schumm, S.A., 1956. Evolution of drainage systems and slopes in badlands at perth
Lawler, D.M., 1993. The measurement of river bank erosion and lateral channel change: a amboy, New Jersey. Geol. Soc. Am. Bull. 67, 597–646.
review. Earth Surf. Proc. Land. 18, 777–821. Shi, Z., Wen, A., Zhang, X., Yan, D., 2013. Comparison of the soil losses from Be-7
Licznar, P., Nearingb, M.A., 2003. Artificial neural networks of soil erosion and runoff measurements and the monitoring data by erosion pins and runoff plots in the Three
prediction at the plot scale. Catena. 51 (2003), 89–114. Gorges Reservoir region. China. Appl. Radiat. Isot. 69, 1343–1348.
Li, J., Ma, X., Zhang, C., 2020. Predicting the spatiotemporal variation in soil wind ero- Streeter, D.T., 1975. Preliminary observations on rates of erosion on Chalk Downland
sion across Central Asia in response to climate change in the 21st century. Sci. Total paths. Institute of British Geographers, Oxford.
Environ. 709, 136060. https://doi.org/10.1016/j.scitotenv.2019.136060. Summer, R.M., 1986. Geomorphic impacts of horse traffic on Montane landforms. J. Soil
Livingstone, I., 2003. A twenty-one-year record of surface change on a Namib linera dune. Water Conserv. 41 (2), 126–128.
Earth Surf. Proc. Land. 28, 1025–1031. Sun, J., Yu, X., Li, H., Yu, C.H., Wang, H., Tu, Z., Liang, H., 2016. Simulated erosion using
Loughran, R.J., 1989. The measurement of soil erosion. Prog. Phys. Geog. 13, 216–233. soils from vegetated slopes in the Jiufeng Mountains. China. Catena. 136, 128–134.
Luetzenburg, G., Bittner, M.J., Calsamiglia, A., Renschler, C.S., Estrany, J., Poeppl, R., Tehrani, E.N., Sahour, H., Booij, M.J., 2019. Trend analysis of hydro-climatic variables in
2020. Climate and land use change effects on soil erosion in two small agricultural the north of Iran. Theo. Appl. Climatol. 136 (1–2), 85–97. https://doi.org/10.1007/
catchment systems Fugnitz-Austria. Can Revull-Spain. Sci. Total Environ. 704, s00704-018-2470-0.
135389. https://doi.org/10.1016/j.scitotenv.2019.135389. Tervuren, J.M., 1990. Soil loss by rainwash: A case study from Rwanda. Z.
Martınez-Casasnovas, J.A., 1998. Soil-landscape-erosion. Gully erosion in the Alt Geomorphologie N.F. 1990, 34, 385–408.
Penedes-Anoia (Catalonia, Spain). A spatial information technology approach: spatial Uson, A., 1998. Medidas de control de la erosion en suelos de vina de lascomarcasAnoia-
databases, GIS andremote sensing. Ph.D. Thesis. University of Lleida, Lleida, pp. AltPenedes (Barcelona). Efectividad y viabilidad. Ph.D. Thesis. University of Lleida,
333 pp.. Lleida.
Masson, J.M., 1971. L’érosion des sols par l’eau en climatméditerranéen. Wiggs, G.F., Thomas, D.S., Bullard, J.E., Livingstone, I., 1995. Dune mobility and vege-
Méthodesexpérimentales pour l’étude des quantitésérodées ‘al’échelle du champ. tation cover in the southwest Kalahari Desert. Earth Sur. Proc. Land. 20, 515–529.
Thèse de Docteur-Ingénieur, USTL, Montpellier, 215 PP. Wischmeier, W.H., Smith, D.D., 1978. Predicting rainfall erosion losses: a guide to con-
Mohamed, M.A., Atta, M., 2010. Automated classification of galaxies using transformed servation planning. Agric. Handbook No. 282. US Department of Agriculture,
domain features. IJCSNS International Journal of Computer Science 86 and Network. Washington, DC.
Security 10, 86–91. World reference base for soil resources (WRB)., 2014. Food and agriculture organization
Nadal-Romero, E., Martinez-Murillo, J.F., Vanmaercke, M., Poesen, J., 2011. Scale-de- of the united nations, international soil classification system for naming soils and
pendency of sediment yield from badland areas in Mediterranean environments. creating legends for soil maps. pp193.
Prog. Phys. Geog. 35, 297–332. Yair, A., Lavee, H., 1974. Areal contribution to runoff on scree slopes in an extreme arid
Naghdi, R., Dalir, P., Gholami, V., Pourghasemi, H.R., 2017. Modeling of sediment gen- environment. A simulated rainstorm experiment. Zeitschr, Fur Geom.
eration from forest roads employing SEDMODL and its calibration for Hyrcanian Zhang, Z., Luo, J., Chen, B., 2020. Spatially explicit quantification of total soil erosion by
forests in northern Iran. Environ. Earth Sci. 76, 414 (2017). https://doi.org/10.1007/ RTK GPS in wind and water eroded croplands. Sci. Total Environ. 702, 134716.
s12665-017-6758-7. Zhao, Z., Chow, T.L., Rees, H.W., Yang, Q., Xing, Z., Meng, F.R., 2009. Predict soil texture
Pastor, M., Castro, J., 1995. Soil management systems and erosion. Olivae. 59, 64–74. distributions using an artificial neural network model. Comput. Electron. Agri. 65
Pickup, G., Marks, A., 2000. Identifying large-scale erosion and deposition processes (2009), 36–48.

11

You might also like