INTRODUCTION

Soil erosion, characterized by the detachment and transport of soil particles, is driven by complex processes involving erosive agents like wind and water. Water erosion, particularly intricate, leads to substantial soil loss and sedimentation (Raza et al., 2021;Tachi et al., 2020). Human-induced environmental impacts have exacerbated erosion issues, causing severe consequences, including nutrient depletion, declining soil fertility, and negative impacts on biodiversity (Bhandari et al., 2021; Sidi Almouctar et al., 2021). On the other hand, the impacts of erosion out-site includes sedimentation of lakes and dams which can reduce their water storage capacity and efficiency of hydraulic structures (Andualem et al., 2023; Sirikaew et al., 2020). For these reasons, it is essential to implement soil erosion management measures to prevent negative consequences both on-site and off-site.

Algeria faces significant challenges related to soil erosion, a major cause of land degradation and desertification which threatens agronomic productivity and the environment, with huge amount of fertile soil is being lost annually (Gliz et al., 2015; Koussa & Bouziane, 2019). Overgrazing, poorly managed agricultural practices and the expansion of settlements are identified as factors contributing to erosion in Algeria (Halefom et al., 2018; Martínez-Murillo et al., 2020). The precarious situation of an imbalance between rural and urban areas is the result of a lack of concordance of agricultural policies and self-management of agricultural lands. Therefore, addressing the problem of soil erosion in the selected study area was crucial to implement effective measures and to control and manage soil erosion.

Various model-based methods, including empirical models like the Revised Universal Soil Loss Equation (RUSLE), were developed for soil erosion spatial assessment (Hategekimana et al., 2020; Morán-Tejeda et al., 2015; Räsänen et al., 2023). RUSLE, recognized for its consistency, flexibility, and data accessibility, is widely used for estimating annual soil loss at the watershed scale (Alewell et al., 2019; Bouzeria et al., 2023; Kumar et al., 2022; Saoud & Meddi, 2023). While effective in estimating soil loss, RUSLE does not quantify sediment migration specifically from a particular watershed by river (Serbaji et al., 2023). The RUSLE model focuses exclusively on evaluating soil erosion and does not consider sediment yield, which is a recognized limitation of the model (Alexiou et al., 2023; Kumar et al., 2022; Phinzi & Ngetar, 2019). To address this limitation, the sediment delivery ratio (SDR) is integrated into the RUSLE model, forming the RUSLE-SDR combination. The integration of the RUSLE and SDR models improves the accuracy of sediment yield estimation by identifying high-yield areas and providing reliable annual estimates, as demonstrated in various watershed studies in the world (Ebrahimzadeh et al., 2018; Efthimiou et al., 2017; Kanito et al., 2023)

Providing data on the origins of sediment within a watershed can serve as an indicator of the extent of erosion taking place in that specific watershed (Jain & Kothyari, 2000). Furthermore, understanding sediment yield and its influencing factors provides valuable insights for landform evolution such as dynamic soil quality assessments, sediment budgets, fluvial dynamics studies, and net erosion intensity estimation within river basins (Rajbanshi & Bhattacharya, 2020). For this reason, the quantification of the sediment expelled through the outflow of the watershed is employed with the RUSLE-based Sediment Delivery Ratio (SDR) approach. The combined RUSLE-SDR model has found its widespread application in recent years for the comprehensive assessment of both average annual soil loss (A) and sediment yield (SY) at the watershed scale (Colman et al., 2018; Ebrahimzadeh et al., 2018; Rajbanshi & Bhattacharya, 2020; Sampath & Radhakrishnan, 2023; Saoud & Meddi, 2023; Singh et al., 2019; Tsegaye & Bharti, 2021a).

In the small agricultural zone of the Bou Rouina watershed, this study presented a unique opportunity for proactive research and conservation efforts. Despite the current absence of significant erosion challenges, the scarcity of measured data on soil erosion and sediment yield in this region emphasized the importance of comprehensive studies. By applying the RUSLE-SDR model to estimate sediment yield, the research aimed to establish baseline data and insights for future conservation initiatives. Emphasizing the proactive role of small agricultural zones, the study contributed not only to the specific context of the Bou Rouina watershed but also to the broader discourse on sustainable land management in subhumid environments.

The aims of the present study were followed by: a) estimating spatial soil erosion patterns using RUSLE, b) determining the spatial Sediment Delivery Ratio (SDR), c) mapping sediment yield distribution along watershed channels, and d) validating the generated sediment yield map using Receiver Operating Characteristic (ROC) analysis. The study stands out as novel, given the lack of prior investigations using any modeling framework in this specific watershed context.

STUDY AREA

The Bou Rouina watershed is a sub-basin of the great basin of Seybouse (6,745 km²), located in the North-East of Algeria. The study watershed has a surface area of 32.22 km² and is drained by the Bou Rouina Wadi. The surrounding terrain includes hills such as Fedj el Mzara (626 m), Ras Kef Bardou (770 m), Kef es Settah (832 m), Koudiat Seba Mzaier (972 m), Koudiat Thalsta (408 m), Koudiat el Arnel (323 m). The elevation of the area ranges between 157 m and 972 m (Figure 1).

The geological composition of the study area covers different rock types with mainly marls and marly limestone of Campanian to Maestrichian age. These rocks, highly eroded in the north-northwest, cover an area of 14.45 km² (44.85% of basin area). The southwestern part, covering 11.11% of the surface (3.58 km²), is composed of Numidian sandstone of Oligocene age. The Senonian microbreccias are less common, located in the north and covering only 0.68 km² (2.11% of area). The Quaternary formations, covering 41.95% (13.51 km²) consist of alluvium and colluviums deposits. The alluvial formations are identified as terraces of Soltanian and Tensiftian age. The colluvium deposits (e.g. scree deposits and glacis) are located at the base of the marly limestone and sandstone formations.

The study area exhibits three soil types: Chromic Cambisols, Calcic Cambisols, and Podzolic soils. Calcic Cambisols, featuring a permeable limestone-rich horizon, support various vegetation types, while Chromic Cambisols lack limestone and may have surface clay. Podzolic soils, with acidic horizons, support shrubs and grasslands. The climate is sub-humid, characterized by distinct wet (October-April) and dry (June-September) seasons, with a mean annual rainfall of 604.16 mm. The study basin has predominantly steep slopes (3-15%), and agricultural land use dominates (12.56 km²) with cereals as the main crop. Grasslands and shrubs cover significant areas, but the limited protective plant cover makes the soil vulnerable to water erosion (Figure 2).

METHODS AND DATA SOURCE PROCESSING

Sources of the data

The input datasets for RUSLE were obtained from various sources such as remote sensing imagery, digital elevation models (DEM), soil maps, and weather data. For computing the rainfall and runoff erosivity factor (R), spatial distribution of the annual mean rainfall were computed using the daily rainfall data of five rainfall stations (Ain Berda, el Kerma, Kef el Mourad, Nechmaya, Heliopolis) from 1978 to 2022. The data were collected from the National Agency of Hydraulic Resources and the National Office of Meteorology.

The geological map of Guelma (scale: 1:50,000) was used to extract information about lithology. The DEM with 30 m resolution was obtained from the website of the United States Geographical Survey (USGS) in order to calculate the topographic factor (LS).

Soil data were relatively scarce and obtained from a soil map of Constantine at scale of 1:200,000 and extracted from FAO Digital Soil Map of the World (DSMW). The soil map is considered essential to compute the soil erodibility factor (K). Land cover and land use maps were realized using satellite imageries of Landsat-8 and Google Earth Professional images at high resolution. Geospatial data from various sources were gathered and processed using ArcGIS 10.4 software to create the thematic layers required for the study.

Revised Universal Soil Loss Equation

The RUSLE model considers five main factors that affect soil erosion. They were selected based on the geo-environmental characteristics of the study basin and a comprehensive literature review (Bouamrane et al., 2021). The RUSLE model is advantageous because it is flexible in modeling, it utilizes accessible input data and can be easily managed and visualized within a GIS application (Singh & Kansal, 2023). The RUSLE model can be expressed as follows Equation 1:

where: A: the computed soil loss per unit area per year (t∙ha⁻¹∙yr⁻¹); LS: the topographic or slope length and steepness factor (dimensionless); K: the soil erodibility factor (t.ha.h.ha⁻¹.Mj⁻¹.mm⁻¹); R: the rainfall erosivity factor (MJ.mm.ha^-1.h^-1.yr^-1); C: the vegetal cover and management factor (dimensionless); P: the support practice factor (dimensionless).

The conditioning factors were derived individually in raster format, and then combined through map superposition using the mathematical equation of the RUSLE model as described above. The methodology used in this study is illustrated in (Figure 3).

Rainfall erosivity factor (R)

The rainfall erosivity factor is an important driving force of local climate on generating splash, sheet and rill erosion. This factor is based on the rainfall amount and intensity, whether for single storms or a series of storms, and reflects the effect of rainfall intensity on soil erosion (Koirala et al., 2019).

The R-factor is an index developed by Wischmeier and Smith (Wischmeier & Smith, 1978) using the total kinetic energy of rain and the maximum intensity during a 30 min interval. Due to the lack of the rain intensity data within the study area, the equation of Arnoldus (1980) was utilized. The parameter involved only annual and monthly precipitation to determine the R-factor with the following Equation 2:

where R is the annual rainfall erosivity factor (MJ.mm.ha^-1.h^-1.yr^-1), P is the mean annual precipitation in mm, Pi is the mean monthly precipitation in mm.

Monthly precipitation datasets were collected from the five rainfall stations. Ultimately, the data were imported to ArcGIS software and the Inverse Distance Weighted (IDW) method was used to interpolate the rainfall erosivity. It is a method of interpolation that estimates cell values by averaging the values of sample data points in the neighborhood of each processing cell. The closer a point is to the center of the cell being estimated, the more influence (or weight) it has in the averaging process. The (IDW) method was chosen for this research due to its superior fit with the present data and relatively lower error rate.

Soil erodibility factor (K)

Soil erodibility is a key parameter to measure soil susceptibility to water erosion. The soil erodibility factor (K) depends on the soil characteristics and the ability of soil or surface material to persist against the erosion (Fayas et al., 2019). It is a function of the soil permeability, soil organic matter content, and most importantly, the soil texture and structure (Bou-imajjane et al., 2020).

According to Wischmeier & Smith (1978), soil erodibility factor is calculated by using soil organic matter, particle size parameter, soil structure and permeability as shown below Equation 3:

where: K: soil erodibility factor (t.ha.h.ha⁻¹.Mj−1.mm⁻¹); M: textural factor with M = (msilt + mvfs) x (100 – mc); mc (%): clay fraction content (< 0.002 mm); msilt (%): silt fraction content (0.002–0.05 mm); mvfs (%): very fine sand fraction content (0.05–0.1 mm); OM (%): organic matter content; s: soil structure class (s = 1: very fine granular, s = 2: fine granular, s = 3, medium or coarse granular, s = 4: blocky, platy or massive); p: Permeability class (p = 1: very rapid,... p = 6: very slow). The value of 0.1317 is a conversion factor from US units to international units.

A total of 48 samples were chosen out of 3 different types of soil that were previously identified in the study area. The samples were physico-chemically analyzed in the laboratory to obtain the soil composition based on the percentage of sand, clay, silt, and organic matter. Finally, the soil erodibility datasets were converted into a rasterized surface through IDW interpolation technique.

Topographic Factor (LS)

The impact of soil erosion susceptibility is heavily influenced by slope length (L) and slope steepness (S), which are the two most critical topographic attributes considered in the LS-factor of the RUSLE model (Phinzi & Ngetar, 2019). The L represents the effect of slope length on erosion and the S represents the effect of slope steepness on erosion. Both parameters were determined from the STRM DEM with 30 m resolution obtained from USGS sources. The method was firstly conducted by generating flow-accumulation, creating the slope map; and then computing LS-factor by a raster layer in GIS environment. The later procedure was done by calculating the LS -factor for each raster cell in a DEM, essentially by treating each pixel as its own segment of uniform slope. The empirical equation developed by Wischmeier & Smith (1978) was provided by the following formula:

where: L is the slope length in meters, S is the angle of slope in percent, m is a constant dependent on the value of the slope gradient: 0.5 if the slope angle is greater than 5%, 0.4 on slopes of 3% to 5%, 0.3 on slopes of 1 to 3%, and 0.2 on slopes less than 1%.

The equation of LS-factor, used as a single index, expressed the ratio of soil loss. In order to implement LS-factor in ArcGIS, the below formula of Bizuwerk et al. (2008) was used as:

The method of using flow accumulation, upslope contributing area, and slope in a GIS environment has an advantage in accounting for convergence and divergence of flow; thus, capturing more complex topography (Benavidez et al., 2018).

Land cover factor (C)

Vegetation cover is regarded as one of the most important protection measures for controlling soil erosion caused by water (Vatandaşlar & Yavuz, 2017). Factor C is a variable that represents the average loss of soil over time, which is influenced by the variables such as slope, length and erosivity. These variables are weighted as a function of the amount of erosion caused by rainfall during the same time period (López-García et al., 2020).

The use of an annual variation in vegetal cover was considered in the present study, as there is no cultivation between June and December in the study area, with no regular rainfall. Making changes to the C-factor can allow easy comparison of the relative impact of management options on soil erosion. The C-factor ranges from near zero for a well-protected land cover to 1 for barren areas. When the land use and land cover of the study area predominantly consists of high distribution of forest and crop plantation, the impact of the C- factor on soil erosion becomes not significant (Ganasri & Ramesh, 2016). Normalized Difference Vegetation Index (NDVI) approach was carried out for the assessment of C-factor. Concerning NDVI, it computed the red band (RB) and near-infrared band (NIRB) by using the Landsat-8 as follows:

RB and NIRB correspond to Band 5 and Band 4 respectively. NDVI values range between -1,0 and +1,0.To estimate the C-factor, the used Landsat-8 image is acquired on 29 December 2021 with a spatial resolution of 30 m. Equation 6 and the following Equation 7 as well as the ArcGIS software were used to estimate the NDVI vegetation index and C-factor. The relationship between the C-factor and NDVI is as follows (Van der Knijff et al., 1999):

where α and β are parameters that determine the shape of the curve of NDVI as a function of C-factor, with values of 2 and 1 given for parameters α and β. The C-factor ranges between 1 and 0, while closeness to 0 indicates the well protected land.

Conservation practice factor (P)

P-factor expresses the effects of the conservation practices that reduce soil erosion potential from the water runoff. The P-factor values range from 0.2 for reverse-slope bench terraces to 1.0 where there are no erosion control practices. In the study area, where farmers do not use antierosion development and conservation tillage practices, the P factor was assumed to be equal to 1.

Determination of sediment yield and Sediment Delivery Ratio

Sediment yield (SY) denotes the volume of sediment transported by a basin over a specific period and the quantity that reaches a reservoir or basin outlet due to soil erosion or surface wear caused by water, wind, ice, and gravity (Ayele et al., 2021). However, obtaining direct measurements of sediment yield in watersheds without comprehensive sediment data can be challenging (Tsegaye & Bharti, 2021b). The application of an accurate estimation of the Sediment Delivery Ratio (SDR), which represents the capacity to transport a fraction of gross erosion from a specific area within a given time frame, can be an effective method for estimating sediment yield (Kanito et al., 2023).

Sediment yield calculation

The calculation of average sediment yield (SY) involves using the following formula:

where SDR is the sediment delivery ratio (dimensionless), and A is the average soil loss of the watershed (t∙ha⁻¹∙yr⁻¹) (Da Ouyang, 1997).

Sediment delivery ratio

The sediment delivery ratio (SDR) represents the proportion of sediment yield at a specific stream cross section to the total erosion occurring in the watershed upstream from that point (Melese et al., 2019). Limited to a range between 0 and 1, the SDR is consistently less than the erosion rate itself (Sampath & Radhakrishnan, 2023). In order to identify the most suitable SDR model, three models were selected from various options: (a) Williams & Berndt (1972), (b) Cai & Fan (2004), and Zhao & Shi (2002) (Table 1).

Following Boufeldja et al.'s methodology (2020), the Erosion Potential Method (EPM) and the sedimentation coefficient (Ru) were employed to facilitate the best model selection, using primary comparative methods such as standard error (SE) and coefficient of variation (CV). The variability in factors used by each method contributes to explain the differences in SDR results. Additionally, for comparative purposes, we computed the sediment coefficient ratio (Ru) of the EPM model using the equation provided below:

where L is the length of the straight line joining the two ends of the watershed; P is the perimeter of the catchment area in km; D is the difference between the mean and minimum elevations of the watershed, which is given as follows:

Where: D₀ is the elevation at the outlet in km and D_ar is the average elevation of the watershed in km.

Subsequently, a comprehensive evaluation of model performance was conducted using diverse statistical tests including adaptive comparisons, standard error (S.E.) Equation 11, standard deviation (SD) Equation 12, and coefficient of variation (CV) Equation 13. The statistical parameters were chosen in relation to the type, nature, and relevant data for analysis and selection of the optimal model for the study area (Boufeldja et al., 2020). The selection of the best model is determined by assessing the models according to SE, SD, and CV, with preference given to the model exhibiting the lowest values for these metrics.

Concerning the standard error, it was calculated as follows:

with: SE: Standard Error (%); (SDR)_B: base sediment delivery ratio; (SDR)_E: estimated sediment delivery ratio by model.

In order to determine the appropriate SDR model or models, attempts were made to assess the accuracy of models by calculating CV of each model, based on measured SDR.

where: CV is the coefficient of Variation, SD is the standard deviation; X_o is the observed SDR (SDR_o); X_est is the estimated SDR (SDR_est).

Validation

The ROC (Receiver Operating Curve)/AUC (Area Under Curve) ratio serves as a quantitative measure to assess the performance of the stochastic deterministic identification and prediction system. In this study, using Google Earth Pro and field observations in erosion-prone zones, 17 erosive points (True Positive points, TP) were randomly selected from the study area. These points were then incorporated into ArcMap software to create an ROC / AUC curve using the ArcSDM (Statistical Data Modular) extension tool.

The ROC/AUC plot, which ranges from 0 to 1, provides insights into the accuracy of the developed spatial map RUSLE model. A value closer to 0 indicates low accuracy, while a value closer to 1 signifies high accuracy. According to Mandrekar (2010), different levels of accuracy are categorized, such as excellent (0.9 to 1.0), very good (0.8 to 0.9), good (0.7 to 0.8), satisfactory (0.6 to 0.7), and unsatisfactory (0.5 to 0.6).The True Positive Rate (TPR) and False Positive Rate (FPR) were determined by Equations 14 and 15, respectively.

RESULTS AND DISCUSSION

The obtained results using GIS techniques and RUSLE model have displayed the mean annual soil loss of the Bou Rouina watershed. The computed soil loss map is helpful in the management and control of erosion in the study basin.

RUSLE input factors

Rainfall erosivity (R-factor)

R-factor map, created from the mean annual rainfalls and IDW method, has given a mean annual rainfall that ranges between 562 mm and 639 mm. The average R-factor among all the chosen rainfall stations is 148.02 MJ.mm.ha^-1.h^-1.yr^-1, with the highest value of 155.11 MJ.mm.ha^-1.h^-1.yr^-1, recorded at Ain Berda station. On the other hand, the lowest value of 143.05 MJ.mm.ha^-1.h^-1.yr^-1 is observed at Nechmaya station. The erosivity (R) distribution in the study basin exhibits a low variability, as indicated by a standard deviation of 3.92 and a coefficient of variation of 2.65%. This implies that the erosivity values across the basin are relatively consistent and do not display significant fluctuations (Table 2).

Based on the erosivity map, the northern region adjacent to the basin area outlet exhibits the highest values, covering an area of approximately 13% of the total basin. Conversely, the lowest erosivity is predominantly found in the west part, corresponding to R value of 22.40%. The analysis of rainfall dataset revealed that the spatial patterns of the highest erosivity values are coincident with the corresponding patterns of high rainfall events of sufficient duration, where intense storms based on the convective vertical velocity of rain are dominant (Figure 4).

Soil erodibility (K)

The comprehensive dataset, integrating both soil composition and land use specifics, offers a refined perspective on soil erodibility within the study area. Focusing on the three primary soil types: Calcic Cambisols, Chromic Cambisols, and Podzols, it becomes clear that each type has unique features that affect how easily they erode (Figure 5).

The Calcic Cambisols, representing 28.55% of the basin's surface area, exhibit erodibility levels ranging from 0.036 to 0.056 t.ha.h.ha⁻¹.Mj⁻¹.mm⁻¹ (Table 3). Predominantly employed for agriculture, these soils display higher percentages of clay (26.245%) and silt (36.165%). Despite the presence of organic matter (3.12%), the Calcic Cambisols exhibit higher erodibility, suggesting that agricultural activities may contribute to this increased susceptibility. Recognizing the high erodibility class of these soils (9.30 km²), targeted conservation efforts are imperative to mitigate potential land degradation in agriculturally dominated regions.

The Chromic Cambisols show slightly lower erodibility range, varying from 0.031 to 0.036 t.ha.h.ha⁻¹.Mj⁻¹.mm⁻¹. These soils, covering 46.25% of the basin area, are highly sandy content (with 47.70%) and primarily occupied by degraded shrubs. The organic matter content (2.95%) and the specific land use contribute to the moderate erodibility distribution (14.90 km²).

The Podzolic soils exhibit the lowest erodibility (0.016− 0.031 t∙ha⁻¹∙yr⁻¹) with an area of almost 25% and a vegetal cover of grassland and sparse shrubs. These soils benefit from lower clay and silt contents, having 25.46% and 30.63% respectively. The organic matter content (3.56%) and specific land cover result in less soil erodibility extension in the study basin (8.02 km²). While these soils present a reduced risk of erosion.

The correlation between soil characteristics, land use, and erodibility is evident in the spatial distribution. Podzolic soils, with their lower erodibility, align with areas of grassland and degraded shrubs in the steeper slopes of the south-southwestern region. Chromic Cambisols, occupying less steep slopes, display a moderate erodibility distribution and are primarily covered by sparse shrubs. In contrast, the agriculturally dominated northern part of the basin, which is characterized by calcic cambisol soils, faces a higher risk of water erosion.

These findings show the importance of considering both soil properties and land use patterns in soil erodibility assessment. Targeted conservation measures should address the specific vulnerabilities of each soil type, acknowledging the interplay between natural factors and human activities. This refined understanding lays the groundwork for effective strategies to mitigate the environmental risks associated with soil erosion by ensuring sustainable land management practices across the diverse landscapes of the study area.

Topographic factor (LS)

The length and slope of each segment were measured and the LS-factor for each segment was computed using RUSLE method and Arcgis. It is known that the LS-factor varies throughout the watershed depending on the existing C and P factors, but generally, the steeper the slopes the higher the LS-factor. The slope percentage ranges between 0 and 50% in the study area, while the LS-factor has reached 44.10 in some places, indicating a significant difference in the topographic status of the area (Table 4). A mean value of LS-factor is equal to 3.58.

The LS class of low values represents the low slopes (0−5%) and characterizes more than 86.76% of the total basin area. The high values represent only 13.24% (Table 4), that it is observed in the southern part of the study area where the relief is high Figure 6. In fact, the slopes in the study basin exhibit a different hillslope profile that is convex along the upper portion and concave along the lower portion. The shape of the hillslope profile affects soil loss rates due to the changes in the length and gradient characteristics along the hillslope surface.

Soils of the Bou Rouina watershed are classified as prone to rill or interrill erosion because they have low percentage of clay (<39%) (Renard et al., 1997). Based on the steepness of the Bou Rouina hillslopes, it appears that the basin is characterized by dominance of low relief. Nevertheless, it is observed that threshold length at which rilling may start to occur at fairly low slopes is frequent in the study basin.

Land cover factor (C)

The C-factor, representing the positive impact of vegetation cover on soil particle stability, plays a crucial role in minimizing soil loss. Plant cover effectively reduces soil erosion by absorbing kinetic energy from raindrops and decreasing runoff. Consequently, different land cover types can be associated with distinct estimated C values.

The study area was categorized into five land cover classes by using supervised image classification and NDVI method. The C-factor values range between 0.15 and 0.95, with a mean value of 0.54, which is relatively high Figure 7.

The spatial distribution of C-factor shows that the most sensitive area to soil erosion is located downstream of the study area (north half) with a low vegetation cover, where almost 33% of the watershed area has C values greater than 0.61. This area represents agricultural land such as cultures, often associated with sparse shrubs or barren land. The high C-factor surely results from variation of vegetation cover throughout the year, during the rainy seasons (climatic variation). Consequently, the sensitivity to soil erosion is different from one season to another.

The moderate C values (0.49−0.61) occupy 22.34% of the basin area. The corresponding zone is distinguished by mainly sparse shrubs and grassland distributed as smaller extensions throughout the basin. The low to very low values (south half), representing 32.73% of the basin area, are found in areas occupied essentially by degraded shrubs and grassland.

Soil loss assessment

The GIS-produced soil erosion map of the Bou Rouina watershed was generated by extracting the data layers for R, K, LS, and P factors and integrating them within the raster calculator of the Arcgis10.4 spatial analyst tools. The synthetical map of soil loss was also performed to quantify and estimate the erosion risk for the study watershed (Figure 8).

Erosion rates within the watershed vary across different areas due to the influence of various factors that govern erosion. These factors include slope, climatic aggressiveness, and the type and density of vegetation cover (Elaloui et al., 2017; Mohammadi et al., 2021). Steep slopes with fast runoff generally result in substantial erosion, with the extent depending on factors such as geology, soil characteristics, and the protective role played by vegetation cover (Jourgholami et al., 2020). This landform character highlights the crucial role of the LS-factor.

Low vegetation cover played a vital role in reducing soil protection by deteriorating infiltration and the physical and chemical properties of the soil, which leads to increasing runoff and soil erosion (Luo et al., 2020; Vásquez-Méndez et al., 2010). Unfortunately, this situation did not help maintaining soil cohesion and reinforcing its mechanical properties. Therefore, the significance of the vegetation cover factor surpassed that of other factors influencing erosion, regardless of climate aggressiveness, slope, or soil type.

The average annual soil loss rate is calculated to be 6.81 t∙ha⁻¹∙yr⁻¹, with a maximum loss of 93.06 t∙ha⁻¹∙yr⁻¹. Table 5 and Figure 8 present the soil erosion severity classes based on the average erosion rate and the total annual soil loss (E) in the study area. Concerning the severity classes. The soil loss severity classes were adopted using the Classification of the National Bureau of Soil Survey (NBSS) and Land Use Planning (LUP) (Loukrakpam & Oinam, 2021).

The results indicate that approximately 51% of the study area is classified as having a low potential erosion risk (E ≤ 5 t∙ha⁻¹∙yr⁻¹), while 26% of the area experiences a moderate erosion rate. The remaining portion of the basin, with 21.17% of the total area, is classified as having high and very high soil erosion by running water.

It is worth noting that in Algeria, there is no officially announced classification for soil loss. However, some researchers suggest that the tolerable threshold for soil loss should not exceed 5 t∙ha⁻¹∙yr⁻¹(Prasannakumar et al., 2012). On the other hand, according to other studies, if the soil loss surpasses a threshold of 7 t∙ha⁻¹∙yr⁻, the erosion risk level is considered high (Benchettouh et al., 2021; Haregeweyn et al., 2017; Karamage et al., 2017).

The study revealed that high levels of soil erosion risks are closely associated with steep slopes, degraded stream flow courses, and areas with low vegetation fraction, predominantly located in the upstream part of the watershed (Figure 8). The upstream part of the watershed is particularly vulnerable to erosion due to its moderate erodibility. This vulnerability can be attributed to the steep hilllslopes exceeding 23% and modified slope length (LS) value of 14 and higher. These steep slopes increase the potential for runoff and sheetwash leading to the erosion and transportation of soil particles downslope (Figure 9). Additionally, the insufficient vegetation cover (C ≥ 0.5) in these areas reduces infiltration, making thus soil to be more susceptible to the erosive force of rainfall. The visible consequences include gravel accumulation (Figure 9) with leads to a depletion of the upper soil layer, and exposure of rocky surfaces.

This scenario is in line with Morsli's et al. (2012) studies in the sub-humid Mediterranean mountains of north-west Algeria, which highlight the universal nature of soil erosion problems and the importance of implementing effective erosion control measures and sustainable land management practices.

Conversely, the northern part of the watershed, characterized by flat terrain and dominated by agricultural lands, exhibits the lowest values of erosive risk. On the findings in this area of low slopes (slope ≤ 15%), expectations for identifying more stable surfaces with less runoff generation that prevent sediment detachment and soil transport particles are plausible. Despite the absence of protective and permanent vegetation cover, the exposed areas are less susceptible to erosion and the associated risks display low erosion rates due probably to the lowland topography.

It can be seen from the analysis that distinguishable pattern of soil loss is correlated with almost all the considered erosion factors. Among the factors considered in the study, LS emerges as the most influential factor, contributing more significantly to the increase in erosive potential compared to vegetation cover. However, it is crucial to acknowledge the impact of land use on soil erosion reduction. Changes resulting from a combination of processes, as observed in certain southern areas of the study basin, can enhance flow energy and transport capacity, thereby promoting erosion along the stream and altering the drainage network.

Previous studies examining soil erosion worldwide revealed varying rates of soil loss attributed to different factors, depending on the land types and local conditions. For instance, in the Bekaa watershed in Lebanon, researchers have identified the slope's role as a significant contributor to erosion, which ranges from moderately favorable to highly favorable (Hassan et al., 2013). Additionally, the absence of protective vegetation cover or the presence of sparse vegetation cover has been identified as another cause of heightened soil sensitivity to erosion in the region. Similarly, in a study conducted in a Moroccan region, researchers such as El Jazouli et al (El Jazouli et al., 2017) have explained the influential factors of the Revised Universal Soil Loss Equation (RUSLE). The length/slope factor (LS) was found to have a substantial impact on soil erosion, along with the erodibility component of the soil (K) and the decline in vegetation cover. These factors directly contribute to the rate of soil erosion in the area under investigation.

The importance of the LS-factor influence can be shown in some other studies in the world (Das et al., 2022; García-Ruiz et al., 2015; Zhao et al., 2022). In karst areas, the interaction between land use type and slope significantly influenced soil erosion, with steep slopes exacerbating soil loss in agricultural lands (Wang et al., 2019). In the northeastern region of Algeria, researchers have observed severe soil erosion driven by a combination of natural and anthropogenic factors. Nehaï & Guettouche (2020) highlighted that the highest soil loss aligns with the increasing LS-factor towards the eastern and southern areas of the Jijel Province, where the terrain becomes more rugged Another study conducted in the Bouhamdane watershed by Bouguerra et al. (2017) reported a high-risk erosion rate. According to these authors, the primary factors driving soil erosion in the region are rainfall erosivity, soil erodibility, and topography. Additionally, Bouhadeb et al. (2018) have identified areas with predominantly moderate to very high erosion rates, characterized by erodible soils, steep slopes, and limited vegetation cover with an average erosion rate of 7.8 t ha⁻¹ y⁻¹ in the Bou Namoussa watershed. Furthermore, Meghraoui (2018) noted that 56.58% of the Sebaa Chioukh Mountains experiences annual soil loss between 150 and 200 t ha⁻¹, emphasizing the role of erodible soils and limited vegetation in intensifying erosion risks. For the most part, it is important to note that variations in erosion can be attributed to a result of combination of unfavorable natural factors and humans in certain situations such as soil conditions, land use, topography, and land management practices.

Sediment yield and sediment delivery

In this study, we conducted a comprehensive assessment of three sediment delivery ratio (SDR) models against the Erosion Potential Method (EPM) as a standard, with the aim of identifying the most suitable model for sediment yield estimation within the study area. The sediment input ratio (Ru) is recorded as 0.57 (Table 6).

Among the models considered, the Williams & Berndt (1972) model emerged as a robust contender, demonstrating a SDR estimation of 0.45 with minimal deviation from the EPM base. The SDR of 45% indicates that 45% of the soil eroded from the landscape is being transported as sediment to the basin outlet. In other words, only 45% of the soil loss contributes to sediment yield. The remaining 55% may be retained within the landscape or deposited in other forms. (Table 7). Notably, this model exhibited low standard error (27.16%) and coefficient of variation (28.21%), indicating high accuracy and precision. This finding aligns reasonably well with the results reported by Saygın et al. (2014), who indicated SDR values varied between 0.106 and 0.506, which is determined by the length and slope of Saraykoy watershed in Turkey.

The corresponding sediment yield values, totaling 9890.27 t/an (or 3.07 t∙ha⁻¹∙yr⁻¹), indicates the efficiency of this model in capturing sediment dynamics.

The selection of the Williams and Berndt model as the preferred sediment delivery ratio (SDR) model has paved the way for the calculation of sediment yield and the generation of a map that correlates with the soil loss (Figure 10). This concordance of sediment distribution in both maps can be attributed to the close relationship between soil erosion and sediment yield processes.

Therefore, there is an interconnection of their spatial distribution. Areas identified as having higher soil loss are likely to contribute more sediment to water bodies, resulting in a proportional distribution on the sediment yield map. The spatial distribution of soil loss and sediment yield is shaped by topography, land use, lithology, and climatic factors. High slopes and degraded vegetation amplify sediment yield, as seen in Algeria’s Mina catchment and K’sob watershed, where rainfall erosivity and agricultural practices are critical (Benchettouh et al., 2021; Guesri et al., 2020). Similar patterns are observed globally; for instance, Brazil’s cultivated fields exhibit high erosion rates (Aneseyee et al., 2020; Lense et al., 2023) while Nepal's Triyuga watershed links steep slopes to sedimentation hotspots (Yigez et al., 2021). Lithology’s role in sediment distribution is evident in Morocco, underscoring its global relevance (Gourfi et al., 2018).

In Algeria, sediment delivery is among the highest globally due to intense rainfall and steep terrain (Achite & Ouillon, 2016; Probst & Suchet, 1992). SDR studies in the Maghreb confirm sedimentation patterns, with slope and catchment size playing significant roles (Remini & Hallouche, 2005). The application of the chosen model in estimating sediment delivery ensures that the derived map accurately represents the areas of the watershed that are more susceptible to sedimentation. This congruency between the soil loss and sediment yield maps enhances our understanding of the erosional processes within the watershed, facilitating targeted conservation efforts and sustainable land management practices in areas with higher sediment yield.

Validation

Given the interpretation provided by Mandrekar (2010) for levels of accuracy in ROC/AUC, with ranges categorized from excellent to unsatisfactory, the obtained AUC value of 0.736 falls within the “good” category (Figure 11). This suggests that the model's ability to discriminate between different sediment yield classes is reasonably effective but may still need improvement.

In the context of sediment yield mapping, a “good” level of accuracy (0.7 < AUC < 0.8) indicates that the model is capable of providing valuable insights into areas with varying levels of sediment yield. While not reaching the highest levels of accuracy, the model's discriminatory power is substantial enough to be considered useful for identifying areas of concern. However, it is important to note that further refinement and validation of the model might enhance its performance.

CONCLUSION

This study used primary and secondary attribute and spatial data. Field visits, rainfall, 30m Landsat imagery, 30m DEM and soil map were the datasets used for surveying and mapping based on geomatic techniques in the ArcGIS 10.4 software. The individual factor layers of R, K, LS, C were independently processed, analyzed, and standardized to UTM Zone 32N with the WGS 1984 datum. Following this standardization, the raster layers were integrated into the RUSLE framework and computed to estimate soil loss within the GIS environment. Moreover, this study pioneered the assessment of sediment yield in the Bou Rouina watershed using the integrated RUSLE-SDR approach. Finally, the performance of the model was validated employing the ROC curve performance.

The results revealed that the RUSLE model was highly sensitive to the LS factor. The mean annual soil loss was estimated to be 6.81 t ha⁻¹ yr⁻¹, identifying soil erosion as a significant threat to agricultural production in upstream areas. The mean sediment delivery ratio was estimated and found that about 45% of eroded material was transported to the outlet of the watershed. The result of this finding shows that the mean sediment yield of the watershed was equal to 3.07 t ha⁻¹ yr. These results emphasize the critical link between soil erosion severity and sediment yield, providing therefore the importance of prioritizing conservation strategies in high-risk areas.

The study provides crucial insights into soil erosion processes in the subhumid environments of northeastern Algeria, offering a valuable reference for addressing soil degradation challenges. Acknowledging limitations, such as data resolution and local variations, it highlights the need for refining methodologies to enhance prediction accuracy.

Beyond the Bou Rouina watershed, these findings contribute to global efforts in sustainable land management and erosion mitigation in similar environments.

REFERENCES

Edited by

Publication Dates

History