Evaluation of Spatial Distribution of Soil Loss Using the RUSLE Model for the Akanyaru Sub-Catchment of Rwanda

Abstract

Soil erosion is a major environmental problem driven by land use change and unsustainable agricultural practices, particularly in Rwanda, where half of agricultural land is moderately to severely degraded. This study assessed soil loss in the Akanyaru Sub-catchment using the Revised Universal Soil Loss Equation (RUSLE) integrated with GIS and validated with field measurements from 876 rainfall events recorded on 12 runoff plots representing four land-use types. The model incorporated rainfall erosivity (R), soil erodibility (K), slope length and steepness (LS), cover management (C) for 2024 and 2025, and a constant conservation practice (P) factor. R values ranged from 1100 to 2400 MJ mm ha⁻¹ h⁻¹ yr⁻¹, K from 0.20 to 0.35 t h MJ⁻¹ mm⁻¹, LS from 0 to 256, and C from 0.01 to 0.99. Estimated annual soil loss ranged from 0 to 650 t ha⁻¹ yr⁻¹. From 2024 to 2025, mean and median soil erosion rates decreased by 5.21 and 18.7 t ha⁻¹ yr⁻¹, respectively, and total annual soil loss declined by 812,654 t (11.45%). Model validation showed strong agreement with field observations (R² = 0.97, NSE = 0.95, nRMSE = 26.32%, PBIAS = −12.47%), demonstrating the reliability of the RUSLE–GIS approach for soil erosion assessment and conservation planning.

1. Introduction

Soil erosion is one of the principal causes of environmental degradation on a global scale, with global erosion rates estimated to be 10 to 20 times higher than new soil formation rates [1]. Approximately 84% of global land degradation results from erosion, with mean rates ranging between 12 and 15 t/ha/year [2]. Soil erosion not only removes fertile topsoil but also reduces soil organic matter, nutrient availability, and water-holding capacity, resulting in declining agricultural productivity, reservoir siltation, deterioration of water quality, and degradation of aquatic ecosystems. Consequently, accurate assessment of soil erosion is essential for identifying erosion hotspots, prioritizing conservation measures, and supporting sustainable land management and watershed planning [3].

In sub-Saharan Africa, where over 50% of the population relies on agriculture for livelihood [4], soil erosion remains a critical challenge. In Rwanda, it is considered the most serious environmental problem in many catchment areas, influenced by factors such as topography, soil type, climate, and land use. National erosion control mapping revealed that out of 2.4 million hectares, about 45% are at high erosion risk, of which 7% are extremely high, 18% very high, and 28% high [5]. The conversion of forests into cropland has worsened the situation, with forest cover declining from 44% in 1990 to 28% in 2015 [6], causing an estimated annual loss of 1.4 million tons of soil and reducing the country’s food production capacity by the equivalent of feeding 40,000 people each year [7].

The main mechanisms behind soil erosion include intense rainfall, steep slopes, soil fragility, and poor land-use practices such as deforestation, over-cultivation, and overgrazing. Reduced vegetation cover weakens soil structure, making it more prone to being washed away by surface runoff. Between 1990 and 2005, Africa experienced net forest losses ranging from 1.1 to 2.7 million hectares per year [8], further increasing erosion rates. These processes lead to significant loss of topsoil and nutrients, declining soil fertility, and sediment deposition in rivers and marshlands [9], all of which threaten sustainable agricultural productivity.

To assess and quantify soil loss, several models have been developed, including APEX, SWAT, USLE, RUSLE, RUSLE2, MUSLE, EPIC, WEPP, and AGNPS [10]. Among these, the Revised Universal Soil Loss Equation (RUSLE) developed by the USDA [11] has become one of the most widely applied tools because of its improved performance and adaptability. Recent studies have demonstrated the effectiveness of integrating RUSLE with GIS for mapping erosion hotspots and supporting watershed management under diverse climatic and land-use conditions [12,13]. Comparative studies show that RUSLE provides 25–45% better prediction accuracy than USLE, especially in areas with slopes exceeding 15% [14].

Despite its widespread application, many RUSLE-based studies rely solely on model simulations without validating predicted soil loss using field observations. This limitation introduces uncertainty in model predictions, particularly in tropical mountainous environments where rainfall intensity, topographic complexity, soil properties, and land management practices strongly influence erosion processes [15]. In Rwanda, previous studies have successfully mapped soil erosion risk using RUSLE; however, most have focused on producing static erosion maps or estimating soil loss for a single period, with limited validation using measured soil loss data [6,16,17,18]. Recent reviews have identified the lack of field validation as one of the major limitations of RUSLE applications in sub-Saharan Africa, reducing confidence in model predictions and their application for land management decisions [19,20]. Moreover, long-term runoff plot studies have demonstrated that field measurements are essential for evaluating and improving the reliability of RUSLE predictions under different environmental conditions [21].

Therefore, this study addresses these limitations by integrating multi-temporal GIS-based RUSLE modeling with runoff plot measurements to assess the spatial and temporal response of soil erosion to land use and land cover (LULC) changes in the Akanyaru Sub-catchment of Rwanda between 2024 and 2025. Unlike previous RUSLE-based studies in Rwanda, which have largely focused on single-period assessments with limited field-based evaluation, the present study examines how recent LULC changes influenced the spatial distribution and magnitude of soil erosion while simultaneously evaluating model performance against observed soil-loss measurements. By comparing erosion patterns between 2024 and 2025, the study provides insight into the short-term response of soil erosion to changes in vegetation cover and land management, thereby improving understanding of recent landscape dynamics and their implications for erosion control.

Furthermore, RUSLE predictions were evaluated using measurements from twelve runoff plots representing four major land-use types, providing field-based evidence for assessing the reliability of modeled soil-loss estimates. The observed changes in mean soil erosion and total soil loss between the two years further demonstrate the potential for changes in vegetation cover and land management to produce measurable reductions in erosion over a relatively short period. Overall, by combining multi-temporal erosion assessment with field-based model evaluation, this study provides robust, field-validated evidence to support sustainable watershed management, strengthen land-use planning, improve the prioritization of soil and water conservation interventions, and inform erosion mitigation strategies in Rwanda.

2. Materials and Methods

The RUSLE parameters were derived using Geographic Information System (GIS) techniques after delineating the catchment with a 30 m resolution Digital Elevation Model (DEM) and defining the watershed outlet. Based on this delineation, other input factors were estimated specifically for the targeted Akanyaru Sub-catchment. Soil loss in the Akanyaru Sub-catchment was evaluated using the Revised Universal Soil Loss Equation (RUSLE) model at an annual temporal scale, which integrates rainfall, soil properties, slope length and steepness, land use, vegetation cover, and support practices. Geographic Information System (GIS) techniques were used to process these spatial datasets and calculate the five RUSLE factors: rainfall erosivity (R), soil erodibility (K), slope length and steepness (LS), cover-management (C), and support practice (P). By combining these factors, spatial maps of soil loss were generated, allowing identification of erosion-prone areas and providing a scientific basis for targeted soil and water conservation strategies in the sub-catchment. For model validation, twelve runoff plots were used for soil loss analysis with four land-use–land-cover types (maize, grass, forest with grass and forest without grass). Each land use has three replications.

2.1. Study Area

This study has been conducted in the Akanyaru Sub-catchment located in southwestern Rwanda, as shown in Figure 1. Erosion is very prominent in the upstream of the catchment (most of Nyaruguru district). The dominant soil types in the Akanyaru Sub-catchment are Nitisols, Acrisols, Alisols, Lixisols, Ferralsols, Cambisols, and Histosols (Figure 2). Nitisols, Acrisols, Alisols, and Lixisols cover most of the catchment, while Ferralsols occur mainly in the southwestern and eastern parts. Cambisols occur in localized areas, and histosols are predominantly found within the Akanyaru marshland. Pockets of cambisol and mineral soils with flat topography, partly plains, partly river valleys, complete this. The Akanyaru River valley contains histosols. Infiltration rates of these soils are generally high except for the mineral soils on flat topography; rare spots of clay are encountered in the center of the catchment and the urban domain of Huye. The catchment area is essentially covered by seven districts (Nyaruguru, Gisagara, Nyanza, Ruhango, Bugesera, Huye and Kamonyi). Some small parts of the catchment are located in the districts of Nyamagabe and Muhanga. The catchment experiences a tropical highland climate with a bimodal rainfall pattern, comprising the long rainy season from March to May and the short rainy season from September to December. The average annual rainfall is approximately 1225 mm and the maximum temperature is 32 °C, although rainfall and temperature vary spatially with altitude and local topographic conditions. A further subdivision of the elongated level 1 catchment of 3402 km² has been suggested as follows for the creation of three level 2 sub-catchments: Akanyaru hills (NAKN_1), Akanyaru valley left bank (NAKN_2) and Akanyaru Cyohoha (NAKN_3) [5]. The lower section of the Akanyaru, which is a flat and wide marshland with several wide tributaries, has substantial potential for agricultural development and peat production in the Akanyaru marshland. The Akanyaru marshland forms the boundary between Rwanda and Burundi (an area of 1561 km² in Rwanda) and the river receives a significant contribution from Burundi.

2.2. RUSLE Modeling

The Revised Universal Soil Loss Equation (RUSLE) was applied to estimate the average annual soil loss within the Akanyaru Sub-catchment by integrating climatic, soil, topographic, vegetation cover, and land management factors within a GIS environment using ArcGIS 10.8. The model is expressed as

$$A=R\times K\times LS\times C\times P$$

(1)

where $A$ is the average annual soil loss (t ha⁻¹ yr⁻¹), $R$ is the rainfall erosivity factor (MJ mm ha⁻¹ h⁻¹ yr⁻¹), $K$ is the soil erodibility factor (t h MJ⁻¹ mm⁻¹), $LS$ represents the slope length and steepness factor (dimensionless), $C$ is the cover-management factor (dimensionless), and $P$ is the support-practice factor (dimensionless).

The RUSLE factors were derived from multiple spatial and field datasets, including rainfall observations from the Rwanda Meteorological Agency, soil data obtained from the National Bureau of Soil Survey and Land Use Planning (NBSS & LUP), a 30 m ASTER Digital Elevation Model (DEM), Landsat 8 OLI imagery for vegetation cover analysis, and field observations of soil and water conservation practices. The resulting spatial layers were processed and integrated within ArcGIS 10.8 to estimate the spatial distribution of annual soil loss across the study area. Detailed procedures used to derive each RUSLE factor are presented in Section 2.3, Section 2.4, Section 2.5, Section 2.6 and Section 2.7. Model predictions were subsequently validated using observed soil loss data collected from experimental runoff plots established under representative land-use and land-cover types. Below is the conceptual framework of soil loss estimation by the RUSLE Model, as shown in Figure 3.

2.3. Rainfall Erosivity Factor (R)

The R factor shows how rainfall can lead to soil erosion. The rainfall erosivity factor (R) reflects the effect of rainfall intensity on soil erosion and requires detailed, continuous precipitation data for its calculation [16]. In this study, rainfall data from eleven meteorological stations (Byimana, Cyeru, Gihinga, Cyili, Rubona, Gakoma, Save Paroisse, Kansi Paroisse, Kansi Ecole, Kigembe and Save) operated by the Rwanda Meteorology Agency (Meteo Rwanda) were accessed. However, only datasets from nine stations were used due to the incompleteness of the remaining two stations (Save and Kansi Ecole), as shown in

Table 1. Geographic coordinates of the weather stations in the Akanyaru Sub-catchment.

No.	Meteorological Station	Latitude (°)	Longitude (°)
1	Rubona Colline	−2.460384	29.764801
2	Gihinga	−2.100000	29.910000
3	Cyili	−2.500000	29.900000
4	Gakoma	−2.490000	29.920000
5	Save Paroisse	−2.549400	29.777705
6	Cyeru	−2.709903	29.788110
7	Kigembe	−2.743203	29.745612
8	Kansi Paroisse	−2.683398	29.800105
9	Byimana	−2.160469	29.772360

and Figure 1. The rainfall records covered the period of 2000–2025. The rainfall dataset consisted of 10 min time steps observations, and all records were subjected to quality control before analysis by checking for missing values, inconsistencies, and abnormal rainfall events. The completeness of the rainfall records for the selected stations was 100% after discarding the two stations with missing data.

Rainfall events were separated using a 6 h window, and erosive storms were identified according to the criteria proposed by [21]. Rainfall erosivity was calculated using the EI₃₀ method, in which the erosivity of each storm is obtained by multiplying the storm kinetic energy (E) by the maximum 30 min rainfall intensity (I₃₀). The I₃₀ value was determined as the maximum rainfall intensity occurring within any continuous 30 min interval during an erosive storm. The rainfall erosivity factor was calculated following the procedures described by [10,18] using Equations (2)–(5). The annual erosivity values were averaged over the 2000–2025 period to produce a long-term mean annual R-factor, which was used as a single rainfall erosivity layer in the RUSLE model rather than calculating separate R-factor layers for individual years. Since rainfall measurements were available only at the locations of the nine meteorological stations, the spatial distribution of rainfall erosivity across the catchment was generated using the Inverse Distance Weighted (IDW) interpolation method in ArcGIS 10.8. The interpolation employed a power parameter of 2, a variable search radius using the nearest 12, and an output raster cell size of 30 m, corresponding to the spatial resolution of the Digital Elevation Model (DEM) and the other RUSLE input layers to ensure spatial consistency. The original method of calculating erosivity is described by [10,17] and adopted by [18,21] using Equations (2)–(5).

$$IF I\le 76 \mathrm{m}\mathrm{m}/\mathrm{h} \to e_m=0.119+0.0873 {log}_{10}\left(i_m\right)$$

(2)

$$IF I>76 \mathrm{m}\mathrm{m}/\mathrm{h} \to e_m=0.283$$

(3)

$$E=\sum \left(e_m\times P\right)$$

(4)

where e_m = unit kinetic energy (MJ/ha/mm) and i_m = rainfall intensity (mm/h).

$$R={\displaystyle \frac{1}{N}}{\sum }_{i=1}^N{\left({EI}_{30}\right)}_i$$

(5)

where $R$ is the rainfall erosivity factor, $E$ = total storm energy (MJ/ha), $I_{30}$ = maximum 30 min intensity (mm/h), ${\left({EI}_{30}\right)}_i$ for storm i (MJ mm/ha/h), and i = number of storms in an N year period.

2.4. Erodibility Factor (K)

The soil erodibility (K) factor indicates the average long-term reaction of the soil and soil profile to the erosive force brought on by runoff and rainfall, and the vulnerability of soil to erosion is indicated by the K-factor [22,23]. For the estimation of the K factor, the soil texture map was obtained from the National Bureau of Soil Survey and Land Use Planning (NBSS & LUP) and subsequently scanned, rectified, and geometrically corrected. Soil texture polygons were extracted from the map and clipped to the boundaries of the Akanyaru Sub-catchment. Undisturbed soil samples were collected from 15 representative sites across the catchment (Figure 4) at depths of 0–20 cm, 20–40 cm, and 40–60 cm to determine the relevant soil properties. Soil samples collected from the 40–60 cm depth were used to characterize soil profile properties required for estimating soil erodibility (K-factor), whereas erosion primarily affects the topsoil layer. The measured soil properties are presented in

Table 2. Soil properties of Akanyaru Sub-catchment.

Points	Depth (cm)	Org. C (%)	Organic Matter (%)	Sand (%)	Silt (%)	Clay (%)	Bulk Density (g cm−3)	Permeability mm/h
Point 1	0–20	4.97	8.57	54	12	34	1.36	27
	20–40	2.70	4.66	50	12	38	1.46	25
	40–60	2.04	3.52	44	20	36	1.73	23
Point 2	0–20	4.49	7.74	52	14	34	1.30	23
	20–40	2.73	4.71	50	15	35	1.63	19
	40–60	1.76	3.04	47	14	39	1.71	20
Point 3	0–20	3.94	6.80	46	14	40	1.58	22
	20–40	2.45	4.22	49	11	40	1.61	15
	40–60	1.50	2.58	49	12	39	1.61	12.26
Point 4	0–20	3.54	6.10	50	12	38	1.37	13.45
	20–40	1.71	2.95	46	11	43	1.32	9.83
	40–60	1.00	1.72	43	12	45	1.50	10.95
Point 5	0–20	5.44	9.37	50	12	38	1.22	11.53
	20–40	2.61	4.50	48	13	39	1.34	11.32
	40–60	1.36	2.34	46	9	45	1.23	13.68
Point 6	0–20	4.41	7.60	48	12	40	1.50	14.20
	20–40	2.80	4.82	50	11	39	1.49	3.58
	40–60	2.26	3.90	53	12	35	1.41	3.36
Point 7	0–20	3.77	6.50	52	12	36	1.46	2.74
	20–40	1.76	3.04	50	11	39	1.53	9.46
	40–60	3.56	6.13	58	12	30	1.64	1.31
Point 8	0–20	2.45	4.22	62	10	28	1.38	1.28
	20–40	1.91	3.29	58	11	31	1.69	29
	40–60	2.03	3.50	56	12	32	1.77	26
Point 9	0–20	5.71	9.85	58	13	29	1.52	23
	20–40	2.70	4.66	51	11	38	1.53	37
	40–60	2.19	3.77	58	12	30	1.57	42
Point 10	0–20	3.25	5.61	56	11	33	1.42	36
	20–40	2.04	3.51	54	12	34	1.51	64
	40–60	1.81	3.13	55	14	31	1.56	60
Point 11	0–20	4.19	7.23	53	14	33	1.5	63
	20–40	2.68	4.63	54	12	34	1.53	47
	40–60	2.07	3.57	52	13	35	1.53	43
Point 12	0–20	3.16	5.44	53	12	35	1.54	40
	20–40	2.42	4.17	52	12	36	1.56	37
	40–60	1.88	3.24	58	10	32	1.56	39
Point 13	0–20	3.25	5.61	52	10	38	1.42	37
	20–40	2.04	3.51	48	12	40	1.51	56
	40–60	1.81	3.13	42	16	42	1.56	52
Point 14	0–20	4.19	7.23	52	14	34	1.5	51
	20–40	2.68	4.63	50	10	40	1.53	44
	40–60	2.07	3.57	48	10	42	1.53	41
Point 15	0–20	3.16	5.44	60	6	34	1.54	49
	20–40	2.42	4.17	52	12	36	1.56	60
	40–60	1.88	3.24	58	10	32	1.56	57

.

Based on laboratory analyses and field observations, thematic layers representing sand, silt, clay, and organic matter contents were developed. Soil structure and permeability classes were assigned according to particle size distribution and soil permeability characteristics, as summarized in Table 2,

Table 3. Structure code for different types of soil.

Code	Structure	Particle Size (mm)
1	Very fine granular	<1
2	Fine granular	1–2
3	Medium or coarse granular	2–10
4	Blocky, platy or massive	>10

and

Table 4. Permeability code for different types of soil.

Code	Description	Permeability Rate (mm/h)
1	Rapid	>130
2	Moderate to rapid	60–130
3	Moderate	20–60
4	Slow to moderate	5–20
5	Slow	1–5
6	Very slow	<1

, respectively. These thematic layers were integrated within the GIS Model Builder environment to derive the K-factor layer using the regression Equation (6). The resulting K-factor raster was converted to vector format and intersected with the 100 m × 100 m grid network to obtain grid-based soil erodibility values for the entire Akanyaru Sub-catchment. With consideration for permeability, organic matter content, soil texture, and soil structure, Equation (6) computes the Soil Erodibility Factor (K). The Equation was developed by [13] and is being used by [5,24,25] in the following equation:

$$K ={\displaystyle \frac{\left(2.1 \times {10}^{-4}\times M^{1.14}\times \left(12 - OM\right) + 3.25\times \left(P - 2\right)+ 2.5\times \left(S - 3\right)\right)}{100}}$$

(6)

where $M$ (% silt + % very fine sand) (100 − % clay); $OM$ = percentage of organic matter; $P$: permeability class; and $S$: structure class. Soil permeability and structural codes were assigned according to the soil’s particle size and permeability rate as shown in Table 3 and Table 4, following the methods proposed by [26].

2.5. Slope Length and Slope Steepness (LS) Factor

The LS factor represents the topographic component of the RUSLE model, integrating the effects of slope length (L) and slope steepness (S) on soil erosion. Areas with longer and steeper slopes accumulate greater runoff energy, thereby increasing the potential for soil detachment and sediment transport. The LS factor was derived from a 30 m × 30 m Digital Elevation Model (DEM) using the Spatial Analyst extension in ArcGIS 10.8. Before analysis, the DEM was hydrologically conditioned using the Fill tool to remove surface depressions and ensure continuous surface drainage. Flow direction was then generated from the corrected DEM using the deterministic eight-direction (D8) flow-routing algorithm, which routes runoff from each grid cell to the steepest downslope neighboring cell. The resulting flow direction raster was used as input for the Flow Accumulation tool to calculate the number of upslope contributing cells for each grid cell, representing the slope length component (L). Slope gradients were calculated from the DEM using the Slope tool and expressed in degrees. Both the flow accumulation and slope raster were then integrated in the Raster Calculator using the empirical equation developed by [27] to compute the final LS factor map.

$$LS={\left({\displaystyle \frac{\left(Flow Accumulation\times Cell Size\right)}{22.13}}\right)}^{0.4}\times {\left({\displaystyle \frac{sin\left(\theta \right)}{0.0896}}\right)}^{1.3}$$

(7)

The cell size parameter was obtained from the DEM properties to maintain spatial consistency during calculation. The resulting $LS$ raster represents spatial variability in slope length and steepness across the catchment, where higher $LS$ values indicate steeper and longer slopes with greater erosion potential. The exact ArcGIS Raster Calculator expression used was

Power ((FlowAcc × 30/22.13), 0.4) × Power ((Sin (Slope × 0.0174533)/0.0896), 1.3)

(8)

For the flow accumulation raster, slope is the slope raster in degrees, and 0.01745 is the conversion factor from degrees to radians. No flow accumulation threshold was applied to exclude stream channels during LS computation; therefore, high LS values may occur along drainage pathways because of their large upslope contributing areas. However, consistent with the assumptions of the RUSLE model, these values indicate only a greater topographic potential for sheet and rill erosion and should not be interpreted as estimates of gully or channel erosion, which involve different erosion processes and require separate modeling approaches.

2.6. Cover Management Factor (C)

The cover management factor (C) represents the effect of vegetation cover and land management practices on soil erosion by reducing the impact of raindrops and surface runoff [28,29]. The C-factor was estimated using the Normalized Difference Vegetation Index (NDVI) derived from remote sensing data because of its efficiency, reliability, and wide availability. NDVI values range from −1 to +1, with higher values indicating denser vegetation cover and, consequently, lower erosion susceptibility.

To derive the C-factor, Landsat 8 Operational Land Imager (OLI), (Ball Aerospace, Boulder, CO, USA) surface reflectance imagery from 2024 and 2025 was processed to generate NDVI maps for the study area. A temporal analysis was conducted to capture vegetation variability during the study period. NDVI values, ranging from −1 to +1, were calculated, with higher values indicating denser vegetation cover and lower susceptibility to soil erosion. To capture seasonal and interannual vegetation dynamics within the Akanyaru Sub-catchment, a temporal analysis of NDVI was performed. The resulting NDVI values were then converted into C-factor values using the rescaled NDVI equation (Equation (9)) proposed by Van der Knijff et al. [22], which was developed for tropical environments. This approach has been widely applied and validated for estimating cover-management conditions and producing realistic C-factor values in tropical regions [9,30,31]. Ground validation was conducted through field surveys at 15 geo-referenced sampling locations distributed across the Akanyaru Sub-catchment (Figure 3). During field visits, observations of vegetation density, land cover conditions, crop cover, and exposed soil surfaces were recorded and georeferenced using a Global Positioning System (Garmin GPSMAP 64, Olathe, KS, USA). Representative photographs were also collected to document the prevailing surface conditions. The field observations were compared with the corresponding NDVI values extracted from the satellite imagery. Areas characterized by dense vegetation cover, including forests and well-covered croplands, exhibited high NDVI values and low C-factor values, whereas sparsely vegetated and bare-soil areas showed lower NDVI values and higher C-factor values. The agreement between field observations and satellite-derived vegetation patterns confirmed the suitability of the NDVI-based approach for estimating the C-factor within the study area. The validated NDVI map was subsequently transformed into a spatially distributed C-factor layer and incorporated into the RUSLE model.

$$NDVI={\displaystyle \frac{\left(NIR-RED\right)}{\left(NIR+RED\right)}}$$

(9)

where $NIR$ is the surface spectral reflectance in the near-infrared band and $RED$ is the surface spectral reflectance in the red band.

$NDVI$ (Normalized Difference Vegetation Index of the area).

$$C=0.1\times \left({\displaystyle \frac{\left(-NDVI+1\right)}{2}}\right)$$

(10)

2.7. Support Practices Factor (P)

The support practice factor (P) represents the effectiveness of soil and water conservation measures in reducing soil erosion and is defined as the ratio of soil loss under a specific conservation practice to the corresponding soil loss from up-and-down slope cultivation [32]. The P-factor ranges from 0 to 1, where lower values indicate greater effectiveness of conservation measures in reducing erosion, while a value of 1 represents the absence of support practices [10].

Information on soil and water conservation practices was collected through field surveys conducted across the entire Akanyaru Sub-catchment. During the survey, conservation practices were identified through direct field observations and georeferenced using a handheld Global Positioning System (GPS). These observations were complemented by visual interpretation of recent satellite imagery and consultations with local agricultural extension officers to verify the spatial distribution of conservation practices within the watershed.

The field survey indicated that contour farming was the predominant support practice implemented within the study area. Accordingly, P-factor values corresponding to the observed conservation practices were assigned according to values recommended in the Revised Universal Soil Loss Equation (RUSLE) guidelines. Areas without conservation practices were assigned a P-factor value of 1.0. The assigned P-factor values for contour farming across different slope classes are presented in

Table 5. Support practice factor (P) for contour farming.

Slope Class (%)	Contour Farming
0–7	0.55
8–11.3	0.60
11.3–17.6	0.80
17.6–26.8	0.90
>26.8	1.00

. The mapped conservation practices were digitized and integrated within a GIS environment to generate the spatial P-factor layer used as input for soil erosion modeling. The Spatial datasets used in the RUSLE modelis shown in

Table 6. Spatial datasets used in the RUSLE model.

Dataset	Source	Spatial Resolution	Temporal Resolution/Coverage	Purpose in the Study
Rainfall	Rwanda Meteorological Agency (Meteo Rwanda), 9 meteorological stations	Point observations (interpolated to 30 m using IDW)	10 min rainfall records (2000–2025)	Rainfall erosivity (R-factor) calculation
Digital Elevation Model (DEM)	ASTER GDEM	30 m	Static	Watershed delineation, slope, flow accumulation, and LS-factor calculation
Soil Data	National Bureau of Soil Survey and Land Use Planning (NBSS & LUP) with laboratory analysis of soil samples	Soil polygons	Static	Soil texture, organic matter, permeability, and K-factor estimation
Satellite Imagery	Landsat 8 Operational Land Imager (OLI) Surface Reflectance	30 m	2024 and 2025	NDVI generation, vegetation dynamics assessment, and C-factor estimation
Conservation Practices	Field survey, GPS observations, satellite image interpretation, and consultation with agricultural extension officers	Field observations	2024–2025	P-factor mapping and assignment

.

2.8. Runoff Plots

Twelve experimental runoff plots, each measuring 2200 cm × 400 cm, were established across four land-use types: cropland fields, grasslands, forests without grass cover, and forests with grass cover, as shown in Figure 5 and Figure 6. Each land-use type was represented by three replicates, and all plots were equipped with runoff collection tanks with dimension of 210 cm width, 210 cm length and 1 m depth as shown in Figure 6. Soil loss data were collected over a two-year period, from 2024 to 2025, during which 876 rainfall events were recorded and analyzed.

After each rainfall event, the runoff volume from each plot was measured, thoroughly mixed, and a 500 mL composite sample was collected. From this, a 100 mL subsample was oven-dried at 105 °C for 24 h to determine the dry sediment weight. Soil loss was calculated according to Equations (11)–(14) [32,33].

Cropland and grassland areas have lower to moderate slopes, relatively better soil fertility indicators, and higher infiltration capacity in grassland than in cropland. Cropland shows higher bulk density and moderate hydraulic conductivity, while grassland has lower bulk density and improved infiltration.

Forested areas occur on steeper slopes. Forest without grass has the most degraded soil conditions, including very low hydraulic conductivity, strongly acidic pH, and lower nutrient retention. Forest with grass shows improved soil structure and infiltration compared to forest without grass due to vegetation cover, although soils remain acidic, as shown in

Table 7. Physical and chemical properties of soils under different land-cover types in twelve experimental plots within the Akanyaru Sub-catchment.

Land Use/Land Cover	Slope (%)	Bulk Density (g cm−3)	Organic Matter (%)	Sand (%)	Silt (%)	Clay (%)	pH (H2O)	Hydraulic Conductivity (cm h−1)	CEC (meq 100 g−1)
Cropland	11.80	1.52	3.24	64.66	14.00	21.34	5.61	1.3261	11.05
Cropland	11.29	1.55	3.00	64.66	11.34	24.00	5.28	1.7902	9.77
Cropland	12.77	1.60	2.63	60.66	13.34	26.00	5.44	2.0415	10.46
Grassland	12.12	1.39	2.09	52.66	10.67	36.67	5.35	3.6914	9.99
Grassland	14.36	1.26	3.14	56.70	12.00	31.30	5.60	3.3139	10.40
Grassland	13.78	1.47	3.16	63.33	13.34	23.33	5.79	3.5853	9.81
Forest without Grass	23.18	1.54	3.04	57.33	11.34	31.33	4.42	0.6085	8.59
Forest without Grass	22.86	1.61	2.13	60.67	9.33	30.00	4.26	0.4993	7.98
Forest without Grass	22.19	1.54	3.54	59.67	10.33	30.00	4.32	0.5116	6.04
Forest with Grass	23.05	1.50	2.36	51.33	12.67	36.00	4.61	2.0731	7.74
Forest with Grass	23.93	1.52	2.98	54.00	11.33	34.67	4.81	2.3463	7.54
Forest with Grass	22.80	1.55	2.48	60.67	9.33	30.00	4.58	2.2467	3.96

.

$$S_c={\displaystyle \frac{W_s}{V_s}}$$

(11)

$S_c$ = sediment concentration kg/m³; where $W_s$ = dry sediment weight (g); $V_s$ = volume of sampled runoff water (L);

$$S_T=S_c\times Q$$

(12)

where $S_T$ = total sediment yield from the plot (kg); $S_c$ = sediment concentration kg/m³; $Q$ = total runoff (m³);

$$E_{plot}={\displaystyle \frac{ST}{A}}$$

(13)

where $E_{plot}$ = soil erosion per unit area (kg/m²); $ST$ = total sediment yield from the plot (kg); $A$ = plot area;

$$E_{annual}=\sum _{i=1}^n{Event}_i$$

(14)

where $E_{annual}$ = annual soil loss (t/ha⁻¹ y⁻¹); ${Event}_i$ = soil loss from individual rainfall event i (tons/ha); n = number of rainfall events in a year.

2.8.1. Field Measurements from Runoff Plots, RUSLE Model Predictions, and Model Validation

To validate the performance of the Revised Universal Soil Loss Equation (RUSLE) model, observed and predicted soil loss data were compared across different land-use and land-cover (LULC) types. Each plot was geographically located using its coordinates and overlaid on spatial maps of RUSLE parameters in a GIS environment. Parameter values were extracted from raster layers by matching plot coordinates with pixel values, allowing localized estimation for each LULC type, and predicted soil loss was directly compared with observed data. Model performance evaluation using Percent Bias (PBIAS), Normalized Root Mean Square Error (nRMSE), Coefficient of Determination (R²) and Nash–Sutcliffe Efficiency (NSE) were computed using Equations (15)–(18) [34]. Model performance was assessed using the coefficient of determination (R²), Root Mean Square Error (RMSE), Percent Bias (PBIAS) and Nash–Sutcliffe Efficiency (NSE). Performance ranges for hydrological and erosion models were shown in

Table 8. Performance ranges for hydrological and erosion models.

Performance Statistic	Very Good/Excellent	Good/Satisfactory	Unsatisfactory
Coefficient of Determination (R2)	>0.75	0.60–0.75	<0.60
Nash–Sutcliffe Efficiency (NSE)	>0.75	0.50–0.75	<0.50
Percent Bias (PBIAS)–Sediment	±15%	±15% to ±30%	>±30%
Normalized Root Mean Square Error (NRMSE)	<10%	10–20%	>20%

as proposed by [34,35].

$$PBIAS=100\times {\displaystyle \frac{\left({\sum }_{i=1}^n\left(O_i-P_i\right)\right)}{{\sum }_{i=1}^n{O}_i}}$$

(15)

where $PBIAS$ = Percent Bias, $O_i$ = observed value, $P_i$ = predicted values, n = number of observations.

$$nRMSE={\displaystyle \frac{RMSE}{\overline{O}}}\times 100\%$$

(16)

where $nRMSE$ = Normalized Root Mean Square Error, $RMSE$ = Root Mean Square Error, $\overline{O}$ = mean of observed values.

$$RMSE=\sqrt{ {\displaystyle \frac{1}{n}}\sum _{i=1}^n{\left(O_i-P_i\right)}^2}$$

(17)

where $RMSE$ = Root Mean Square Error, $O_i=\mathrm{o}\mathrm{b}\mathrm{s}\mathrm{e}\mathrm{r}\mathrm{v}\mathrm{e}\mathrm{d} \mathrm{v}\mathrm{a}\mathrm{l}\mathrm{u}\mathrm{e}\mathrm{s}, P_i=\mathrm{p}\mathrm{r}\mathrm{e}\mathrm{d}\mathrm{i}\mathrm{c}\mathrm{t}\mathrm{e}\mathrm{d} \mathrm{v}\mathrm{a}\mathrm{l}\mathrm{u}\mathrm{e}\mathrm{s}$, n = number of observations.

$$NSE=1-{\displaystyle \frac{\left({\sum }_{i=1}^n{\left(O_i-P_i\right)}^2\right)}{{\sum }_{i=1}^n{\left(O_i-\overline{\mathrm{O}}\right)}^2}}$$

(18)

where $NSE$ = Nash–Sutcliffe Efficiency, $O_i$ = observed values, $P_i$ = predicted values, $\overline{\mathrm{O}}$ = mean of observed values, n = number of observations.

$$R^2 = ({\displaystyle \frac{\sum \left(O_{\mathrm{i}}-\overline{\mathrm{O}}\right) \times \sum (P_i-\overline{P})}{\sqrt{\left[\sum (O_{\mathrm{i}}-\overline{\mathrm{O}}{)}^2\right]} \times \sqrt{\left[\sum (P_{\mathrm{i}}-\overline{P}{)}^2\right]}}}{)}^2$$

(19)

where $O_{\mathrm{i}}$ = observed values, $P_{\mathrm{i}}$ = predicted values, $\overline{P}$ = mean of predicted values, $\overline{\mathrm{O}}$ = mean of observed values, n = number of observations.

2.8.2. Statistical Analysis

Statistical analyses were conducted in R version 4.3.0. Soil loss data were tested for normality and homogeneity of variance prior to analysis. A mixed-effects model was performed to assess the effects of land use, rainfall intensity, water level, slope, bulk density, organic matter, and clay content on soil loss. Post hoc pairwise comparisons were conducted using an adjusted post hoc comparisons test to identify significant differences between land-use types. Boxplots were generated using the ggplot2 version 4.0.3 to visualize the distribution of soil loss across plots and land use categories.

3. Results

3.1. Rainfall Erosivity (R)

The rainfall erosivity (R-factor) in the Akanyaru Sub-catchment shows clear spatial variation, ranging from approximately 1100 to 2400 MJ mm ha⁻¹ h⁻¹ yr⁻¹. Figure 7 indicates that higher R-factor values are mainly concentrated in the northern and north-western parts of the sub-catchment. Moderate values dominate the central zone, while the southern part of the sub-catchment is characterized by lower R-factor values.

3.2. Soil Erodibility (K)

The soil erodibility (K) factor in the Akanyaru Sub-catchment ranged from 0.20 to 0.35 (t h MJ⁻¹ mm⁻¹), indicating moderate to high susceptibility to erosion. Finer-textured soils such as silty clay and clay loam showed higher K-values (≈0.30–0.35), while clay-rich and sandy clay loam soils had lower values (≈0.20–0.28). Loam and sandy loam soils exhibited intermediate erodibility (≈0.25–0.30).

The spatial distribution of the K factor varied across the sub-catchment, with values ranging from 0.20 to 0.35 as shown in

Table 9. Basic soil properties in the Akanyaru Sub-catchment.

Soil Texture	Org. C (%)	Organic Matter (%)	Sand (%)	Silt (%)	Clay (%)	Bulk Density (g cm−3)	Permeability (mm/h)	K Factor	Area (km2)	Percentage (%)
Sandy clay	1.2	2.1	55	15	30	1.45	5	0.20	603.80	38.68
Sandy clay loam	1.5	2.6	60	15	25	1.40	8	0.2	269.55	17.27
Clay	2.0	3.4	20	20	60	1.30	2	0.22	204.86	13.12
Silty loam	1.8	3.1	20	65	15	1.35	10	0.30	137.80	8.83
Clay loam	1.7	2.9	30	35	35	1.35	6	0.25	117.11	7.50
Sandy loam	1.3	2.2	65	25	10	1.50	15	0.24	87.40	5.60
Loamy sand	0.8	1.4	80	10	10	1.60	25	0.20	74.44	4.77
Silty clay	1.9	3.3	10	45	45	1.30	3	0.28	62.64	4.01
Loam	1.6	2.8	40	39	21	1.35	12	0.22	3.41	0.22
Total									1561.00	100.00

and Figure 8. Higher K-values were observed in the western and central zones, while lower values were dominant in the eastern and northern areas.

3.3. Slope Length and Slope Steepness (LS)

The slope length and steepness (LS) factor in the Akanyaru Sub-catchment ranged from 0 to 256, indicating varying topographic influences on soil erosion across the area. The study area is mainly dominated by gentle to moderate slopes (0–14%), with some steep areas (22–64%), as shown in

Table 10. Area occupied by different slope classes in the Akanyaru Sub-catchment.

Slope Class (°)	Slope Category	Area (km2)	Percentage of Total Area (%)
0–7	Gentle	452	29.0
8–14	Moderate	516	33.0
15–21	Moderately steep	344	22.0
22–30	Steep	187	12.0
31–64	Very steep	62	4.0
Total		1561	100

and Figure 9. Steeper slopes are concentrated in the central and southern parts of the catchment.

The LS factor is generally low to moderate across most of the area, with some localized zones showing higher values. Higher LS values are associated with steeper slopes, indicating areas with greater potential for soil erosion.

3.4. Cover Management (C)

The cover and management factor (C) in the Akanyaru Sub-catchment, as shown in Figure 10, exhibited significant spatial heterogeneity and localized temporal improvements between 2024 and 2025. In 2024, the C-factor values ranged from 0.01 to 0.99, whereas the range expanded to 0–0.99 in 2025. While the structural persistence of the maximum value at 0.99 across both years isolates unmitigated erosion hotspots, the drop in the minimum boundary baseline to 0 signifies a measurable interannual increase in protective canopy density and vegetative ground cover. Spatially, low C values of 0 were geographically clustered within the southern and central zones of the sub-catchment, where dense permanent forests and grasslands effectively mitigated raindrop kinetic energy and surface runoff velocities. Conversely, high C values approaching 0.99 were concentrated across the northern and eastern portions, identifying highly vulnerable, exposed cultivated lands and degraded bare soils subjected to severe detachment risk. The localized cover improvements observed in 2025 reflect positive canopy dynamics likely driven by progressive agroforestry or seasonal biomass accumulation, though the persistent maximum values emphasize that critical zones still require targeted, long-term soil conservation measures.

3.5. Conservation Practice (P)

The support practice factor (P) in the Akanyaru Sub-catchment as shown in Figure 11 remained constant between 2024 and 2025, with values ranging from 0.55 to 1.0. The identical P-factor distribution in both years reflects the permanent nature of structural conservation measures. Areas with low P-factor values (0.55) were mainly located in the southern and south-central parts of the sub-catchment, indicating the presence of effective conservation practices that reduce runoff velocity and soil loss. In contrast, high P-factor values close to 1.0 were concentrated in the eastern and northeastern areas, where conservation measures are limited or absent, making these zones more susceptible to erosion.

The stability of the P-factor suggests that differences in soil loss between 2024 and 2025 were not influenced by changes in support practices. Instead, variations were likely caused by changes in vegetation cover and rainfall conditions, while conservation structures provided a consistent level of erosion control throughout the study period.

3.6. Predicted Soil Loss Under Different Land-Use–Land-Cover Types from RUSLE in Akanyaru Sub-Catchment

The predicted soil loss in the Akanyaru Sub-catchment, as shown in Figure 12 and

Table 11. Area occupied by each soil erosion class.

Class (t ha−1 yr−1)	Area (km2)	Catchment (%)
0–100	962	61.63
101–300	414	26.52
301–650	185	11.85
Total	1561	100.00

, ranged from 0 to 650 t ha⁻¹ yr⁻¹ in 2024 and 2025, showing strong spatial variation in erosion intensity across the landscape. The classification into three erosion severity classes indicates that a large portion of the catchment experiences moderate soil loss, while specific areas show very high erosion rates.

The lowest soil loss class (0–100 t ha⁻¹ yr⁻¹) is mainly found along the eastern and western outer fringes, as well as the extreme northern and southern tips of the catchment, typically associated with forested and grass-covered areas where vegetation protects the soil and reduces runoff; it occupies an area equal to 61.63%. Moderate soil loss (101–300 t ha⁻¹ yr⁻¹) is widespread across the middle and transitional slopes, mostly associated with gently to moderately sloping cultivated land.

The highest soil loss class (301–650 t ha⁻¹ yr⁻¹) is clearly localized as a prominent central longitudinal band running through the middle of the catchment, with additional high-intensity clusters situated in the northern and southern sections; it occupies an area equal to 11.85%. These hotspots occur in steep croplands, degraded hillslopes, and along drainage lines, where conditions favor increased runoff and soil detachment.

3.7. Changes in Soil Erosion Across the Catchment Between 2024 and 2025

In

Table 12. Summary statistics of changes in soil erosion across the catchment between 2024 and 2025.

Indicator	Absolute Change	Relative Change (%)
Mean soil erosion rate (t ha−1 yr−1)	−5.21	−11.45
Median soil erosion rate (t ha−1 yr−1)	−18.7	−10.80
Total annual soil loss (t/yr−1)	−812,654	−11.45

, the results show a consistent decline in soil erosion, with the mean and median erosion rates decreasing by 5.21 and 18.7 t ha⁻¹ yr⁻¹, respectively. Likewise, the total annual soil loss decreased by 812,654 t (11.45%), indicating an overall improvement in catchment conditions and a substantial reduction in soil degradation during the study period. The majority of the catchment experienced a reduction in soil erosion between 2024 and 2025. The catchment falls within the two negative change classes (−150 to −50 and −50 to −10 t ha⁻¹ yr⁻¹), indicating widespread decreases in soil erosion. In contrast, only 6.4% of the area exhibited little or no change (−10 to 10 t ha⁻¹ yr⁻¹). As shown in

Table 13. Area by soil erosion change class.

Change Class (t ha−1 yr−1)	Area (km2)	Catchment (%)
−150–−50	700.30	44.86
−50–−10	761.03	48.75
−10–10	99.67	6.39
Total	1561	100.00

and Figure 13 confirming that the observed reduction in soil loss was spatially extensive across the catchment.

3.8. Predicted Soil Loss and Observed Soil Loss for 12 Plots

The results showed that cropland experienced the highest soil loss at 36.1 t ha⁻¹ yr⁻¹, indicating high vulnerability due to limited vegetation cover, while grassland recorded the lowest soil loss at 1.3 t ha⁻¹ yr⁻¹, reflecting the protective effect of dense herbaceous vegetation. Forest with grass showed 4.5 t ha⁻¹ yr⁻¹, whereas forest without grass recorded 6.1 t ha⁻¹ yr⁻¹, highlighting the importance of understory vegetation in reducing erosion. The predicted RUSLE outputs were consistent with observed values, with cropland at approximately 41 t ha⁻¹ yr⁻¹, grassland at 1.23 t ha⁻¹ yr⁻¹, and forest conditions ranging between about 5–7 t ha⁻¹ yr⁻¹ depending on ground cover as shown in

Table 14. Predicted soil loss and observed soil loss for 12 plots and for each land use/land cover.

No	Land Use	Plots	R	K	LS	C	P	Predicted (t ha−1 yr−1)	Observed (t ha−1 yr−1)	Average Predicted per Land Use (t ha−1 yr−1)	Average Observed per Land Use (t ha−1 yr−1)
1	Soybean & Maize	SP1	2000	0.33	1.58	0.07	0.55	40	35.6	41	36.1
2	Soybean & Maize	SP2	2000	0.34	2.6	0.04	0.58	41	31.4
3	Soybean & Maize	SP3	2000	0.33	1.26	0.09	0.56	42	42.6
4	Grass	GP1	1800	0.27	0.18	0.02	0.7	1.2	0.8	1.4	1.3
5	Grass	GP2	1800	0.21	0.14	0.03	0.9	1.4	1.7
6	Grass	GP3	1800	0.28	0.18	0.02	0.9	1.6	1.4
7	Forest without Grass	FP1	2200	0.22	0.45	0.05	0.6	6.5	6.8	7	6.1
8	Forest without Grass	FP2	2200	0.26	0.44	0.04	0.65	7	5.8
9	Forest without Grass	FP3	2200	0.22	0.5	0.05	0.62	7.5	5.7
10	Forest with Grass	FGP1	2100	0.20	0.7	0.02	0.75	4.5	4.4	5	4.5
11	Forest with Grass	FGP2	2100	0.23	0.7	0.02	0.74	5	5.5
12	Forest with Grass	FGP3	2100	0.21	0.81	0.02	0.77	5.5	3.4

. Model performance evaluation using R², nRMSE, PBIA and NSE showed values of 0.97, 26.3%, −12.86% and 0.95, respectively, indicating good model performance and acceptable accuracy.

3.9. Statistical Analysis of Measured Soil Loss from Experimental Erosion Plots

Annual soil loss varied significantly among land-use types Figure 14, with maize plots showing the highest soil loss, followed by forest without grass and forest with grass, while grassland recorded the lowest erosion rates. This indicates clear variation in erosion intensity across land-use types, with cultivated maize areas being the most vulnerable and well-vegetated areas being the most protected.

A mixed-effects model was performed using the soil loss measurements obtained from the experimental erosion plots. The results in

Table 15. Results for factors influencing soil loss by using mixed-effects model.

Factor	Sum Sq	df	F-Value	p-Value	Estimates	Confidence Intervals
Land use	149,074	3	6.169	<0.001	0.26	[−42.8071, 190.5417]
Rainfall intensity	25,174.81	1	3.1253	<0.001	−46.24	[−93.8882, 1.4016]
Precipitation amount	28,892	1	3.5869	<0.001	1.29	[−93.8882, 1.4016]
Slope	538.49	1	0.0669	0.079	6.98	[−8.4368, 11.0141]
Bulk density	63.2023	1	0.0078	0.929	−1.16	[−146.8728, 160.8414]
Organic matter	86.80	1	0.0108	0.917	−1.167	[−23.1067, 20.7723]
Clay content	7.59	1	0.0009	0.976	−0.043	[−2.7845, 2.6983]

showed that land use had a highly significant effect on measured soil loss (p < 0.001). Rainfall characteristics, including rainfall intensity (p < 0.001) and precipitation amount (p < 0.001), also significantly influenced soil loss from the plots. Slope had a marginally significant effect (p = 0.079), whereas soil properties, including bulk density, organic matter, and clay content, did not significantly affect the measured soil loss (p > 0.05).

Adjusted post hoc comparisons (

Table 16. Tukey’s HSD post hoc pairwise comparisons of soil loss among different land-use types.

Comparison	Difference (Estimate)	SE	t-Value	p-Value
Forest with Grass—Forest without Grass	−0.52	8.05	−0.06	1.00
Forest with Grass—Grassland	5.51	8.10	0.68	0.90
Forest with Grass—Maize	−43.52	8.21	−5.30	<0.001
Forest without Grass—Grassland	6	8.09	0.74	0.88
Forest without Grass—Maize	−43.00	8.27	−5.19	<0.001
Grass—Maize	−49	8.50	−5.76	<0.001

) indicated that maize plots had significantly higher soil loss than all other land uses (p < 0.001). No significant differences were observed among forest with grass, forest without grass, and grassland (p > 0.05). Boxplots (Figure 14) further confirmed that maize plots had the highest median and greatest variability in soil loss, while vegetated plots showed lower and more stable values.

4. Discussions

4.1. Rainfall Erosivity (R)

The observed variation reflects spatial differences in rainfall intensity and distribution across the catchment, which is consistent with patterns in tropical highland environments. Based on the rainfall erosivity of Akanyaru Sub-catchment (R = 1100–2400 MJ mm ha⁻¹ h⁻¹ yr⁻¹), the Akanyaru Sub-catchment falls within the medium to high erosivity class, typical of humid climates [36].

The higher R-factor values suggest a considerable risk of soil erosion, particularly in cultivated and steep slope areas, where intense rainfall can accelerate rill and gully formation. These findings are consistent with previous studies. Similar ranges have been reported in tropical highland regions, including studies by [13,37] in Ethiopia, and [38] in Nigeria.

In Rwanda, the results align with findings by [39], who reported rainfall erosivity values ranging from 1000 to 3000 MJ mm ha⁻¹ h⁻¹ yr⁻¹, influenced by altitude and rainfall patterns. Additionally, ref. [40] observed that Rwanda’s mid-altitude zones experience bimodal rainfall erosivity linked to short and long rainy seasons, with values between 1000 and 2500 MJ mm ha⁻¹ h⁻¹ yr⁻¹.

Overall, the classification of the Akanyaru Sub-catchment within the medium to high erosivity range highlights the importance of incorporating rainfall Erosivity into RUSLE-based soil erosion modeling and implementing climate-resilient watershed management strategies to support sustainable land use under current and future climate variability.

4.2. Soil Erodibility (K)

The observed variability in soil erodibility reflects differences in soil texture, structure, organic matter content, and permeability. Soils with higher silt content and lower organic matter are more prone to detachment and transport by rainfall and runoff, resulting in higher K-values [10,13]. Land use also plays an important role. Upland cultivated areas tend to have higher K-values due to frequent tillage, reduced organic matter, and exposure to raindrop impact, whereas forested and lowland areas show lower erodibility because of better soil structure and higher organic content [41,42].

The spatial heterogeneity observed in the Akanyaru Sub-catchment highlights the need for site-specific soil conservation strategies. The range of K-values found in this study is consistent with findings from other regions. Globally, K-values range from about 0.05 in clay soils to over 0.40 in silt loams [13]. In Europe, values between 0.15 and 0.40 have been reported [43], and in Africa, studies in Ethiopia show similar ranges (0.20–0.38). In Rwanda, previous studies also align with these findings, including [5], which reported values between 0 and 0.24, and [44], which found values between 0.25 and 0.40 in the Sebeya watershed.

Overall, the Akanyaru Sub-catchment falls within the typical erodibility range of humid tropical regions, where strong weathering and organic matter depletion influence soil erosion processes.

4.3. Slope Length and Slope Steepness (LS)

The variation in LS values shows that topography plays a significant role in soil erosion. In many parts of the catchment, the combined effect of slope length and steepness is limited, resulting in lower erosion risk. However, in areas with steep and long slopes, runoff velocity increases, leading to higher soil detachment and transport. These findings are consistent with other studies in Rwanda and the region. Ref. [5] reported LS values ranging from 0 to 322 at the national level, especially in western Rwanda. Similarly, ref. [44] found LS values between 0 and 470 in the Sebeya watershed, where steep slopes contributed to high erosion rates. In Ethiopia, ref. [45] reported even higher LS values (up to 3726), confirming that topography is a major driver of soil erosion in East Africa.

The LS range observed in the Akanyaru Sub-catchment falls within that of humid tropical highlands, where steep slopes accelerate surface runoff and increase erosion risk. This suggests that steep and cultivated upland areas are particularly vulnerable. Therefore, appropriate soil and water conservation measures such as terracing, contour farming, and vegetative barriers are necessary to reduce slope length, slow runoff, and minimize soil loss, especially in high-risk zones.

4.4. Cover Management (C)

The C-factor values in the Akanyaru Sub-catchment ranged from 0.01 to 0.99 in 2024 and 0 to 0.99 in 2025. The lower minimum value observed in 2025 indicates improved vegetation cover and greater soil protection compared to 2024. However, the maximum value remained unchanged at 0.99, suggesting that areas with sparse vegetation cover or bare soil continued to exist and remained highly susceptible to soil erosion. Overall, the decrease in the minimum C-factor value reflects a slight improvement in land cover conditions and a potential reduction in soil loss within well-protected areas.

The wide range of C-factor values demonstrates the significant influence of vegetation cover and land management on soil erosion processes. Areas characterized by low C-factor values are associated with dense vegetation cover, forests, grasslands, or well-managed agricultural systems that effectively reduce raindrop impact, surface runoff, and soil detachment. In contrast, areas with high C-factor values represent cultivated lands, degraded landscapes, or exposed soils where limited ground cover increases vulnerability to erosion.

This finding is consistent with previous studies. Ref. [46] reported significant variation in C-factor values in Rwanda, with values approaching 1 in poorly covered areas. Similarly, ref. [47] emphasized that the C-factor (ranging from 0 to 1) is strongly linked to land use and land cover, and its variability can lead to large differences in soil erosion estimates. Ref. [5] also found C-values ranging from 0.003 to 0.63 in Rwanda, showing that better vegetation cover reduces erosion risk.

The results indicate that poorly managed croplands and bare soils remain the most erosion-prone areas within the Akanyaru Sub-catchment, whereas areas with dense vegetation provide greater protection against soil loss. Therefore, strengthening sustainable land management practices such as agroforestry, mulching, cover cropping, and conservation agriculture is essential for enhancing vegetation cover and reducing soil erosion across the catchment. The observed reduction in the minimum C-factor value from 2024 to 2025 further suggests that improved land cover conditions can contribute to better soil conservation and watershed sustainability.

4.5. Conservation Practices (P)

The variation in p-values reflects uneven implementation of soil conservation practices across the catchment. Areas with lower values demonstrate the effectiveness of practices like terracing and contour farming in reducing soil erosion, whereas higher values indicate zones where soils are more vulnerable due to lack of such measures.

These findings are consistent with previous studies in Rwanda and similar regions [17]; reported higher p-values in steep croplands of the Satinskyi Catchment located in Ngororero District, Western Province, Rwanda due to weak or absent conservation practices, while lower values were associated with areas where structural interventions were implemented. Similarly, ref. [48] found that poor maintenance or absence of conservation measures in Rulindo led to increased erosion risk.

Other studies also support this pattern. Ref. [49] reported p-values ranging from 0.55 to 1.0, where higher values correspond to greater erosion vulnerability. Ref. [50] similarly found that lower p-values are linked to effective practices such as contour farming, reduced tillage, and cover cropping, which help reduce soil loss.

Overall, these results highlight that conservation practices are not uniformly applied, and scaling up effective soil and water conservation measures is essential to reduce erosion and sustain agricultural productivity.

4.6. Predicted Soil Loss Under Different Land-Use–Land-Cover Types from RUSLE in Akanyaru Sub-Catchment

The predicted soil loss in the Akanyaru Sub-catchment ranged from 0 to 650 t ha⁻¹ yr⁻¹ in both 2024 and 2025. Based on the RUSLE model, soil loss was classified into three categories: low (0–100 t ha⁻¹ yr⁻¹), moderate (101–300 t ha⁻¹ yr⁻¹), and high (301–650 t ha⁻¹ yr⁻¹), with spatial patterns reflecting the combined effects of vegetation cover, topography, and land management practices. Areas experiencing low soil loss were mainly associated with forests and grasslands. Moderate soil loss was predominantly observed in cultivated areas on gentle to moderately steep slopes. The highest soil loss values were concentrated in steep croplands, degraded hillslopes, and areas near drainage channels, particularly within a prominent central longitudinal band extending through the catchment and clusters at the northern and southern extremes. The observed spatial variability in soil loss in 2024 and 2025 is influenced by key factors such as rainfall intensity, soil properties, slope characteristics, and land management practices [51,52,53]. Areas with steep slopes, poor vegetation cover, and inadequate land management are particularly vulnerable to high erosion rates.

These findings are consistent with previous studies in Rwanda and beyond. For example, ref. [54] reported that about 15% of Rwanda’s land experiences extreme erosion (>300 t ha⁻¹ yr⁻¹) contributing disproportionately to national soil loss. Similarly, ref. [44] observed extreme erosion exceeding 1000 t ha⁻¹ yr⁻¹ in the Sebeya watershed, while ref. [55] identified hotspots above 900 t ha⁻¹ yr⁻¹ in the upper Nyabarongo catchment.

Globally, studies such as [10,56] show that although average soil loss may be relatively low, localized hotspots on steep or bare land can exceed 500 t ha⁻¹ yr⁻¹, similar to the high values observed in Akanyaru. In Europe, refs. [53,57] also reported very high erosion rates in mountainous and intensively cultivated regions, driven by steep slopes, poor vegetation cover, and intense rainfall.

Overall, these comparisons indicate that the Akanyaru Sub-catchment follows patterns typical of mountainous and intensively cultivated landscapes, where erosion risk is highly uneven. The spatial distribution of soil loss identified distinct erosion hotspots, particularly in steep cultivated areas, while lower erosion rates were associated with forests and grasslands that provide effective soil protection. These spatial patterns offer valuable information for environmental planning by enabling decision-makers to identify priority areas for soil conservation and watershed management. The generated soil loss maps can support the implementation of targeted interventions, including terracing, contour farming, agroforestry, vegetation restoration, and improved land management, in areas most susceptible to erosion. Furthermore, the comparison of the 2024 and 2025 soil loss maps provides a basis for monitoring the effectiveness of conservation measures and land-use changes over time. Overall, these findings demonstrate that the spatial assessment of soil erosion can guide efficient allocation of conservation resources, reduce land degradation, and support sustainable agriculture and watershed management in the Akanyaru Sub-catchment.

4.7. Changes in Soil Erosion Across the Catchment Between 2024 and 2025

Soil erosion between 2024 and 2025 indicates a measurable improvement in catchment conditions and demonstrates the positive influence of changes in land management and land cover on soil conservation. The mean and median soil erosion rates decreased by 5.21 and 18.7 t ha⁻¹ yr⁻¹, respectively, while the total annual soil loss declined by 812,654 t (11.45%). These reductions suggest that the catchment became less susceptible to soil detachment and transport during the study period, reflecting improved protection of the soil surface against rainfall impact and surface runoff. Similar reductions in soil erosion have been associated with improvements in vegetation cover and the implementation of conservation practices in several watershed studies [10,57].

The spatial distribution of erosion changes further supports this finding. Most of the catchment was classified within the negative change categories (−150 to −50 and −50 to −10 t ha⁻¹ yr⁻¹), indicating that reductions in soil erosion were widespread rather than localized. In contrast, only 6.4% of the catchment experienced negligible change (−10 to 10 t ha⁻¹ yr⁻¹), confirming that the improvement occurred across a large proportion of the watershed. This spatially extensive decline suggests that the reduction in erosion resulted from catchment-scale changes rather than isolated site-specific interventions. Similar spatial patterns have been reported in East African watersheds, where land use changes and conservation measures produced substantial reductions in soil loss over extensive areas [57,58].

Increased vegetation cover, expansion of forested and grassland areas, improved crop management practices, and the implementation of soil and water conservation measures such as contour farming, terraces, and agroforestry can substantially reduce runoff velocity and enhance rainfall interception. These processes increase infiltration, strengthen soil structure through root development, and reduce the erosive energy of overland flow, thereby lowering sediment detachment and transport [10,17].

The results are consistent with studies conducted in Rwanda and other tropical highlands, where improved land management has significantly reduced soil erosion despite persistent steep slopes and high rainfall erosivity [57,59]. Therefore, the observed decline in annual soil loss likely reflects the cumulative effects of improved land cover conditions and the adoption of soil and water conservation practices throughout the catchment. Such interventions are essential to further reduce sediment production, improve agricultural sustainability, and minimize downstream impacts on water quality and reservoir sedimentation [1,51].

Overall, the reduction in both the magnitude and spatial extent of soil erosion demonstrates progress toward sustainable watershed management. Nevertheless, areas that continue to exhibit relatively high erosion rates should remain priorities for targeted conservation interventions, including afforestation, terracing, contour cultivation, and other integrated soil and water conservation measures.

4.8. RUSLE Model Predictions and Model Validation

The results clearly show that soil loss is strongly controlled by land-use and vegetation cover, with croplands being the most vulnerable due to reduced soil protection, while grassland and forested areas effectively reduce erosion risk. The good agreement between observed and predicted soil loss confirms that the RUSLE model reliably captures spatial patterns of erosion in the Akanyaru Sub-catchment. The performance indicators, including a relatively high R², low bias, and acceptable error levels, demonstrate that the model provides robust estimates of soil loss. These findings are consistent with previous studies such as [58,59], which reported higher erosion in cultivated lands and lower erosion in well-vegetated areas, and they also align with [17,60], who emphasized the protective role of vegetation cover in reducing erosion rates. Similar conclusions were reached by [61], who highlighted that land cover is a key determinant of erosion intensity. Overall, the combined field and GIS-based RUSLE approach demonstrates reliable performance for estimating soil erosion and identifying vulnerable areas, providing a strong basis for targeted soil and water conservation planning in the Akanyaru Sub-catchment.

4.9. Statistical Analysis Results

The mixed-effects model results (Table 15) demonstrate that land use and rainfall-related factors were the dominant controls of soil loss variability in the Akanyaru Sub-catchment, whereas soil physical properties such as bulk density, organic matter, and clay content showed limited influence during the study period. The significant effect of land use (F = 6.169, p < 0.001) highlights the important role of vegetation cover and management practices in regulating erosion processes. Cropland areas, particularly maize fields, exhibited higher soil loss because of frequent tillage, limited surface residue during some crop growth stages, and reduced protection against raindrop impact and runoff generation.

Rainfall intensity was also a significant predictor of soil loss (F = 3.125, p < 0.001), confirming the strong relationship between erosive rainfall events and sediment detachment. The negative coefficient estimate suggests that variations in rainfall intensity contributed to changes in erosion response among land-use systems. High-intensity rainfall events increase raindrop kinetic energy, surface sealing, runoff generation, and transport capacity, thereby accelerating soil detachment and sediment movement. This finding agrees with previous erosion studies indicating that rainfall intensity, particularly short-duration high-intensity storms, is often more influential than total rainfall amount in determining erosion rates.

Precipitation amount also showed a significant influence on soil loss (F = 3.587, p < 0.001), with an estimated effect of 1.29. Although rainfall quantity alone does not directly determine erosion magnitude, greater seasonal rainfall generally increases cumulative runoff production and the frequency of erosive events. The combined effects of rainfall amount and intensity explain why erosion rates vary considerably between seasons, especially in tropical highland environments such as Rwanda where intense convective rainfall events are common.

Slope showed a relatively weaker but marginal effect on soil loss (F = 0.067, p = 0.079). The positive estimate (6.98) indicates that increasing slope steepness tends to increase erosion, although the relationship was not statistically significant at the 5% level. This may be attributed to the fact that slope effects were partly controlled by land use distribution; steep areas in the catchment were often associated with forests or grass-covered areas that reduced runoff and soil detachment. Therefore, slope alone was insufficient to explain soil loss variation without considering vegetation cover and rainfall conditions.

Soil properties including bulk density (p = 0.929), organic matter (p = 0.917), and clay content (p = 0.976) did not significantly affect observed soil loss variability. Although these properties influence soil erodibility by controlling infiltration capacity, aggregate stability, and resistance to detachment, their effect may have been reduced by the strong influence of land management and rainfall characteristics. In addition, the relatively small variation in soil properties within the study area may have limited their statistical contribution. Similar findings have been reported in tropical agricultural catchments where erosion dynamics are frequently controlled more strongly by vegetation cover and rainfall erosivity than by individual soil characteristics.

The Tukey HSD pairwise comparisons (Table 16) further clarify the influence of land-use type on soil loss patterns. Significant differences were observed only between maize cropland and the other land-use categories. Soil loss in maize fields was significantly higher than in forest with grass (difference = −43.52 t ha⁻¹ yr⁻¹, p < 0.001), forest without grass (difference = −43.00 t ha⁻¹ yr⁻¹, p < 0.001), and grassland (difference = −49.00 t ha⁻¹ yr⁻¹, p < 0.001). These results confirm that maize cultivation represents the most erosion-prone land use system in the catchment. The higher erosion rates observed in maize fields are likely associated with intensive cultivation practices, exposed soil surfaces after land preparation, and reduced vegetation cover during early crop development stages.

In contrast, no significant differences were detected among forest with grass, forest without grass, and grassland (p > 0.05). The difference between forest with grass and forest without grass was very small (0.52 t ha⁻¹ yr⁻¹) and statistically insignificant (p = 1.00), suggesting that both forest systems provide comparable soil protection. Similarly, forest and grassland areas showed similar erosion responses, indicating that continuous vegetation cover effectively reduces runoff velocity, enhances infiltration, and minimizes soil detachment. These findings demonstrate the importance of maintaining permanent vegetation cover as a key soil conservation strategy.

Overall, the statistical analysis confirms that land-use management and rainfall characteristics are the primary drivers of soil loss in the Akanyaru Sub-catchment. The significantly higher erosion rates observed under maize cultivation emphasize the need for improved agricultural conservation practices, including mulching, minimum tillage, contour farming, and maintenance of crop residue. Meanwhile, the similarity between forest and grassland areas highlights the protective role of permanent vegetation cover in reducing erosion risk. These results support watershed management interventions focused on sustainable land-use planning and climate-resilient agricultural practices to reduce soil degradation in Rwanda’s highland catchments.

Uncertainty and Model Limitations

The soil loss estimates presented in this study are subject to uncertainties associated with the RUSLE model inputs. Uncertainty may arise from the estimation of the rainfall erosivity (R), soil erodibility (K), topographic (LS), cover management (C), and support practice (P) factors, as well as from the spatial resolution of the input datasets. During the calculation of LS, a Digital Elevation Model (DEM) of 30 m resolution was accessed as secondary data and may be prone to errors. The small errors in the elevation model, slope gradient, or flow accumulation may have propagated into substantial differences in predicted soil loss. Similarly, the C and P factors were assigned based on land-use classes and conservation practices and may not fully capture local management conditions or seasonal variations in vegetation cover. Although the R and K factors were derived from the best available rainfall and soil datasets, uncertainties in the spatial distribution and temporal variability of these data may also influence the predicted erosion rates. Furthermore, model validation was based on twelve runoff plots representing four land-use types, which provides confidence in the model performance but may not fully represent the spatial heterogeneity of the entire 1561 km² Akanyaru Sub-catchment. Consequently, the predicted soil loss values should be interpreted as estimates of potential sheet and rill erosion rather than exact measurements, and additional long-term monitoring using a larger number of validation sites would further improve model reliability. Despite these limitations, the model successfully identified the spatial distribution of erosion-prone areas and provides a robust basis for prioritizing soil and water conservation interventions in the Akanyaru Sub-catchment.

5. Conclusions

This study assessed the spatial and temporal patterns of soil erosion in the Akanyaru Sub-catchment using the Revised Universal Soil Loss Equation (RUSLE) integrated with GIS and validated using soil loss measurements from 876 rainfall events collected on 12 experimental runoff plots representing four land-use types. The results demonstrated substantial spatial variability in soil erosion, primarily controlled by rainfall erosivity, topography, land use, vegetation cover, and conservation practices. The rainfall erosivity (R) factor ranged from 1100 to 2400 MJ mm ha⁻¹ h⁻¹ yr⁻¹, the soil erodibility (K) factor from 0.20 to 0.35 t h MJ⁻¹ mm⁻¹, the LS factor from 0 to 256, the C factor from 0.01 to 0.99 in 2024 and 0–0.99 in 2025, and the P factor from 0.55 to 1.00, reflecting the spatial variability of erosion drivers across the catchment.

The estimated annual soil loss ranged from 0 to 650 t ha⁻¹ yr⁻¹ during both 2024 and 2025, with the highest erosion rates concentrated in steep, sparsely vegetated croplands and the lowest rates occurring in forests and grasslands. Field observations confirmed that cropland experienced the highest average soil loss (approximately 36 t ha⁻¹ yr⁻¹), followed by forest without grass (6.1 t ha⁻¹ yr⁻¹), forest with grass (4.5 t ha⁻¹ yr⁻¹), and grassland (1.3 t ha⁻¹ yr⁻¹). The RUSLE predictions closely followed the observed patterns, and model validation demonstrated good agreement between predicted and measured soil loss (R² = 0.97, NSE = 0.95, nRMSE = 26.32%, and PBIAS = −12.47%), confirming the reliability of the integrated RUSLE-GIS approach for soil erosion assessment in the study area.

Comparison of the 2024 and 2025 soil erosion maps revealed an overall improvement in catchment conditions. Mean and median soil loss decreased by 5.21 and 18.7 t ha⁻¹ yr⁻¹, respectively, while total annual soil loss declined by approximately 812,654 t (11.45%). Most of the catchment experienced reductions in soil erosion, particularly within the −150 to −50 and −50 to −10 t ha⁻¹ yr⁻¹ change classes, indicating that improvements in vegetation cover represented by the C factor were the primary driver of reduced erosion, whereas the P factor remained unchanged during the study period.

Statistical analysis further demonstrated that land use and rainfall characteristics significantly influenced soil loss (p < 0.001), while measured soil properties, including bulk density, organic matter, and clay content, did not show significant effects (p > 0.05). These findings emphasize that vegetation cover and land management practices are the dominant controls of soil erosion under the environmental conditions of the Akanyaru Sub-catchment.

Overall, the study highlights the high vulnerability of cultivated lands to soil erosion and the important protective role of forests and grasslands. The spatial soil erosion maps generated in this study provide valuable information for identifying erosion hotspots and prioritizing conservation interventions. The findings support the implementation of targeted soil and water conservation measures, including terracing, contour farming, agroforestry, vegetative buffer strips, and integrated watershed management, to reduce soil degradation and promote sustainable agricultural production. The validated RUSLE-GIS framework also provides a practical and reliable decision-support tool for environmental planning, watershed management, and future monitoring of soil erosion in the Akanyaru Sub-catchment and other highland catchments with similar environmental conditions.