Integrating Empirical and Participatory Approaches for Soil Erosion Assessment: A Comparative Study of USLE and AHP in Upland Central Vietnam

Abstract

Soil erosion threatens agricultural sustainability in tropical upland areas. This study evaluated soil erosion in Nam Dong district, Central Vietnam, using the Universal Soil Loss Equation (USLE) model and the Analytic Hierarchy Process (AHP) based on local stakeholder input. The USLE employed spatial data on rainfall, soil, topography, and land cover, while the AHP incorporated the perspectives of nine diverse community members. Both models identified the mountainous central region as most at risk; the USLE classified 62% of land as extreme erosion, whereas the AHP classified 82% as severe. These differences reflect the empirical approach of USLE versus the perception-driven results of the AHP. The study found that applying both methods independently and comparing their outcomes can yield different soil erosion scenarios. Furthermore, additional research is recommended to explore the use of the AHP as a tool for calibrating the relative importance of input factors in the USLE model. This approach could enhance the accuracy of soil erosion risk assessments and support more effectively targeted conservation strategies in complex upland landscapes.

1. Introduction

Soil erosion is considered as the primary cause of land degradation, severely affecting soil quality and reducing crop productivity, thereby causing significant economic losses, particularly in mountainous regions [1]. Globally, two main types of soil erosion have been identified: water erosion and wind erosion. Among these, water erosion predominates in tropical countries, where high rainfall and complex terrain exacerbate the erosion process [2]. In practice, numerous studies on soil erosion have been conducted at various scales to date. Soil erosion studies are largely based on technical models such as the Universal Soil Loss Equation (USLE), Soil and Water Assessment Tool (SWAT), and Water Erosion Prediction Project (WEPP) [3,4,5,6]. These models offer many advantages, such as providing quantitative data on soil erosion, integrating multiple environmental factors, and adapting to various spatial and temporal scales [7]. However, several studies have indicated that these models are constructed and calibrated using experimental data from specific regions. When these models are applied to areas with different ecological characteristics, they often yield significant errors [7,8]. For instance, in the USLE model, the support practice factor (P) is commonly assigned a default value of 1.0, effectively disregarding its influence [9]. This simplification arises due to the absence of available reference data that accurately reflect the land use practices of the studied area. Such omissions can substantially compromise the precision of soil erosion assessments.

Meanwhile, soil erosion is a complex process that varies across ecological regions. Because each region is unique, agricultural land use practices also differ, shaped by farmers’ knowledge and experience. Numerous studies have highlighted the importance of community participation in assessing soil erosion, particularly in relation to agricultural land use. Farmers possess intimate knowledge of local land conditions through their direct interaction with the environment. Integrating their insights with scientific methods results in more accurate assessments and more effective soil conservation strategies [10,11]. A widely used method in studying local community participation in soil erosion assessment is the focus group discussion combined with scoring [12]. To concretize this method, some techniques in multi-criteria decision-making have been applied recently, including the Analytical Hierarchy Process (AHP) [13,14]. In the Loess Plateau of China, AHP was used to develop an erosion potential map by weighting factors including slope, water flow accumulation, elevation, land use, and complex terrain features; the resulting map showed good agreement with historical erosion records and geomorphological data [15]. The AHP is considered effective for soil erosion mapping, as it can harmonize participants’ opinions while accounting for both natural and socio-economic factors. Additionally, when integrated with Geographic Information Systems (GISs), AHP facilitates the spatial representation of priority levels, enhancing the clarity and applicability of the results [14,15].

Thua Thien Hue province, located in the Central region of Vietnam, has over 50% of its total area classified as mountainous. The mountainous areas of Thua Thien Hue province are particularly vulnerable to soil erosion due to their topography and rainy climate. So far, several studies have investigated soil erosion in these regions using models such as the USLE and SWAT. However, these studies have overlooked the incorporation of indigenous knowledge in estimating soil erosion and in proposing sustainable mitigation strategies. Therefore, this study aims to assess the soil erosion situation in the agricultural production areas of Nam Dong district using the USLE model, as well as to integrate the AHP with local community knowledge.

2. Materials and Methods

2.1. Research Area

The Nam Dong district covers an area of 64,777 hectares and is geographically located between 107°28′ E to 107°54′ E longitude and 15°47′ N to 16°17′ N latitude, as shown in Figure 1. The district’s elevation ranges from 29 m to 1200 m above sea level, with the predominant elevation between 400 and 500 m. The average annual rainfall is approximately 3000 mm, with the rainy season occurring primarily from September to December. Temperatures vary seasonally, with the lowest average temperature of 13 °C in January and the highest reaching 41 °C in June. The district center is surrounded by high mountains, forming a valley where the local population resides and engages in agricultural activities. As of 2023, Nam Dong district had a population of approximately 27,000, of which over 60% belonged to ethnic minority groups. The total agricultural land area within the study is 21,996.9 hectares. Among this, production forest land (including both natural and planted production forests) constitutes the largest proportion at approximately 83.30%, followed by perennial crop land (rubber and fruit trees) at 15.15%. Land used for annual crops (such as rice and vegetables) account for 1.55%.

2.2. Materials

The input data for both the USLE and AHP models play a critical role in determining the accuracy and reliability of soil erosion estimation. In this study, we utilized spatial datasets as summarized in

Table 1. Data sources used in USLE and AHP modeling.

Data	Sources	Mapping Method
Rainfall	Data from 9 meteorological monitoring stations [16]	Inverse Distance Weighting (resolution at 30 m)
Elevation	Digital Elevation Model (DEM), resolution at 30 m [17]	Original data, resolution at 30 m
Slope		Calculate from DEM via ArcGIS 10.3, resolution at 30 m
Aspect		Calculate from DEM via ArcGIS 10.3, resolution at 30 m
Soil texture	Soil map of Thua Thien Hue province [18]	Convert from Mapinfo format (Tab) to ESRI format (Shp)
Soil depth
Land use type	Land use map of Nam Dong district [19]	Convert from MicroStation (dgn) to ESRI format (Shp)
Land surface cover	Landsat 8 imagery [20]	Used to calculate NDVI
Windspeed	Global Wind Speed Atlas [21]	Resampled from original data (resolution 250 m) into raster dataset (resolution 30 m)

.

2.3. Methods

2.3.1. Universal Soil Loss Equation

The Universal Soil Loss Equation (USLE), developed by Wischmeier and Smith in the 1960s, estimates long-term soil erosion from rainfall and runoff [22]. Widely used in tropical regions like Indonesia, Brazil, and Vietnam [9], the USLE supports land use planning and erosion control due to its simplicity and low data demands [23]. However, it mainly applies to sheet and rill erosion and may overlook gully erosion and sediment dynamics. Despite these limits, its ease of use and GIS compatibility make it a valuable tool in soil conservation [24]. The USLE combines five factors and can be calculated as follows:

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

(1)

where

A is the average annual soil loss (tons/ha/year);

R is the rainfall erosivity (MJ∗mm/(ha∗hour∗year);

LS is the topographic factor (dimensionless);

K is the soil erodibility factor (t∗hour/(MJ∗mm);

C is the cropping management factor (dimensionless);

P is the practice support factor (dimensionless).

• R factor:

There is a strong correlation between the R factor and rainfall, which has been modeled by numerous researchers worldwide using various empirical formulas. This study adopts the rainfall-based R factor calculation formula proposed by Nguyen Trong Ha [25], which has been widely applied across different ecological regions in Vietnam for soil erosion studies [26]. The formula was developed based on observational data collected from 253 meteorological stations over a continuous period of 54 years, from 1941 to 1994. The formula is as follows:

$$R=0.548257 \ast M \left(\mathrm{mm}\right)-59.9$$

(2)

where R is rainfall erosivity, and M is annual rainfall.

• LS factor:

To calculate the LS factor, Digital Elevation Model (DEM) data are used to determine flow length and terrain slope. In this study, we applied the LS calculation formula proposed by Moore and Wilson [27].

$$LS={\left({\displaystyle \frac{A_s}{22.13}}\right)}^m\ast {\left({\displaystyle \frac{Sin (Slope degree \ast 0.01744)}{0.09}}\right)}^n$$

(3)

In this context, LS represents the topographic factor, A_s is the flow length, and slope degree refers to the slope steepness measured in degrees. The value 22.13 corresponds to the slope length in the empirical USLE model, equivalent to 72 feet. The value 0.01744 is the conversion factor from degrees to radians for slope measurement. The value 0.09 represents the experimental slope steepness, equivalent to 9%. m and n are empirical exponents, with m = 0.5 and n = 1.3, as commonly adopted in studies conducted under similar topographic and climatic conditions.

• K factor:

The K factor represents the soil’s resistance to erosion and is closely related to soil texture and organic matter content. Therefore, obtaining an accurate K value requires soil sampling and laboratory analysis of particle size distribution and organic matter content—procedures that are both time-consuming and costly. In this study, K values were adopted from previous research conducted in mountainous regions of Vietnam [28] for four major soil groups: Ferralsols, Acrisols, Luvisols, and Fluvisols. These values were determined based on the soil map of Thua Thien Hue province, developed by the Vietnam Institute of Agricultural Planning and Design in 2005 [18].

• C factor:

The C factor ranges from 0 to 1 and represents the extent of ground surface cover provided by vegetation. It has a strong negative correlation with the Normalized Difference Vegetation Index (NDVI). Therefore, several studies have proposed using the NDVI to estimate the C factor [29,30], with the formula developed by Durigon [31] being widely applied in recent years. This formula is as follows:

$$C ={\displaystyle \frac{(-NDVI+1)}{2}}, in which, NDVI= {\displaystyle \frac{(NIR-RED)}{(NIR+RED)}}$$

(4)

• P factor:

This factor is considered the most difficult to quantify and the least certain in the USLE model, as it depends on local farming practices, which vary across regions and ethnic groups. Consequently, researchers often omit this factor by assigning it a default value of 1 [32]. However, in studies related to land use and soil erosion, this factor cannot be ignored. Therefore, several methods and inference principles for estimating the P factor have been proposed by scientists worldwide based on empirical data. Among these, the approach proposed by Shin in 1999 has been widely adopted by both Vietnamese and international studies [9,33], in which the P factor is determined based on the type of land use combined with the slope gradient as in

Table 2. The P values based on the land use type and slope.

Land Use Types	Slope (Degrees)
	0–5	5–8	8–10	10–15	>15
Natural forest, Unused Land	1.00	1.00	1.00	1.00	1.00
Plantation forest, perennial trees	0.55	0.60	0.80	0.90	1.00
Annual crops	0.27	0.30	0.40	0.45	0.50

.

• Soil erosion rate classification using the USLE:

Soil erosion classification exhibits considerable diversity, with substantial variation in erosion severity across different ecological zones. In this study, we adopted a soil erosion classification scheme that is widely applied in tropical mountainous regions [1,34], as presented in

Table 3. Soil erosion classification in upland regions in tropical countries.

Soil Erosion Levels	Soil Erosion Rate (ton/ha/year)	Surface Soil Loss (mm/year)
Slight soil erosion	0–10	0–0.8
Moderate soil erosion	10–75	0.8–6.25
Severe soil erosion	75–150	6.25–12.5
Extreme soil erosion	>150	>12.5

.

2.3.2. Analytical Hierarchy Process

The Analytic Hierarchy Process (AHP), introduced by Thomas L. Saaty in 1987 [35] is a method used in Multi-Criteria Decision Analysis (MCDA). It derives a ranking scale by conducting pairwise comparisons of various attributes, relying on the judgments of participants [36]. The steps involved in applying the AHP method are outlined below.

• Selection of participants

The selection of participants is a critical factor in conducting an AHP assessment. A sample size that is too small may overlook important criteria and community perspectives, while an excessively large group can introduce data noise and hinder the achievement of consensus. Several studies suggest that an optimal number of participants ranges from 5 to 15, as this allows for the collection of comprehensive and meaningful information [36,37]. In this study, we selected 9 participants with diverse backgrounds and roles relevant to soil erosion in the study area. A summary of participant characteristics is presented in

Table 4. Summary of participants in the AHP.

Participant Code	Gender	Age	Profession	Background
001	Male	45	Local government staff	Natural resources management
002	Female	36	Researcher	Soil and Land Management
003	Female	37	Women Union staff	Agricultural production
004	Male	60	Farmer	Agricultural production
005	Male	48	Researcher	Soil and Land management
006	Male	45	Local government staff	Agriculture Department
007	Male	46	Farmer	Agricultural production
008	Male	60	Community leader	Agricultural production
009	Female	42	Agricultural Scientific	Soil and crop sciences

.

All participants were provided with an introduction to AHP theory and engaged in practice exercises using Excel-based pairwise comparison forms with an increasing number of criteria. This approach helped them to acquire a solid understanding of the fundamental concepts and procedures of the AHP method.

• Selection of the criteria

An overview of soil erosion, including visual images, was presented to all participants to establish a common understanding. Following this, each participant was asked to independently identify factors they believed influenced soil erosion based on their personal knowledge and experience. The individual responses were then compiled, and factors mentioned by more than 50% of the participants were retained. This process resulted in the identification of nine key factors affecting soil erosion: average rainfall, wind speed, soil texture, soil depth, land use type, land surface cover, elevation, slope, and terrain aspects.

• Construction of pairwise comparison matrices according to the relative importance of each criterion

All identified factors were organized into a pairwise comparison matrix, in which each factor was systematically compared with the others using the numerical scale proposed by Saaty [37], as presented in

Table 5. Verbal and numeric scale for the pairwise comparison of criterion according to the AHP.

Numeric Scale	Response Alternatives of Participants
9; 1/9	Criterion i is extremely more/less important than criterion j
7; 1/7	Criterion i is strongly more/less important than criterion j
5; 1/5	Criterion i is more/less important than criterion j
3; 1/3	Criterion i is slightly more/less important than criterion j
1	Criterion i is equally important as criterion j
8;6;4;2;1/8;1/6;1/4;1/2	These are intermediate values

.

Each participant independently completed a pairwise comparison matrix. These individual matrices were then aggregated to form a collective comparison matrix representing the entire group. The values in the final group matrix were calculated as the geometric mean of the corresponding elements from the ten individual matrices. The outcome is a pairwise comparison matrix (Matrix A):

$$A=\left(\begin{array}{ccccc}{A}_{11}& A_{12}& A_{1i}& A_{1j}& A_{1n}\\ A_{21}& A_{22}& A_{2i}& A_{2j}& A_{2n}\\ A_{i1}& A_{i2}& A_{ii}& A_{ij}& A_{in}\\ A_{j1}& A_{j2}& A_{ji}& A_{jj}& A_{jn}\\ A_{n1}& A_{n2}& A_{ni}& A_{nj}& A_{nn}\end{array}\right)$$

(5)

$$A_{ij}={({\prod }_{k=1}^p{a}_{ijk})}^{{\displaystyle \frac{1}{p}}}$$

(6)

$A_{ij}$ is an important level of criterion $i$ compared to criterion $j$;

$a_{ijk}$ is an important level of criterion $i$ compared to criterion $j$ by $k^{th}$ participant;

$p$ is the number of participants in the discussion.

Matrix A is a primitive matrix, meaning that the importance of the factors is expressed in absolute values. Therefore, it must be normalized into Matrix B. The normalized matrix B was derived from the original matrix A using the standard normalization technique in the Analytic Hierarchy Process, in which each element is divided by the sum of its respective column [38].

$$B=\left(\begin{array}{ccccc}{\overline{A}}_{11}& {\overline{A}}_{12}& {\overline{A}}_{1i}& {\overline{A}}_{1j}& {\overline{A}}_{1n}\\ {\overline{A}}_{21}& {\overline{A}}_{22}& {\overline{A}}_{2i}& {\overline{A}}_{2j}& {\overline{A}}_{2n}\\ {\overline{A}}_{i1}& {\overline{A}}_{i2}& {\overline{A}}_{ii}& {\overline{A}}_{ij}& {\overline{A}}_{in}\\ {\overline{A}}_{j1}& {\overline{A}}_{j2}& {\overline{A}}_{ji}& {\overline{A}}_{jj}& {\overline{A}}_{jn}\\ {\overline{A}}_{n1}& {\overline{A}}_{n2}& {\overline{A}}_{ni}& {\overline{A}}_{nj}& {\overline{A}}_{nn}\end{array}\right)$$

(7)

$${\overline{A}}_{ij}={\displaystyle \frac{A_{ij}}{\sum _{i=1}^n{A}_{ij}}}$$

(8)

${\overline{A}}_{ij}$ is the normalized value of $A_{ij}$;

$\sum _{i=1}^n{A}_{ij}$ is the sum of $A_{ij}$ by column $j$ from matrix A;

$n$ is the number of compared criteria.

From matrix B, the criteria weights can be derived as follows:

$$w_i={\displaystyle \frac{\sum _{j=1}^n{\overline{A}}_{ij}}{n}}$$

(9)

$$W=\left(\begin{array}{c}{w}_1\\ w_2\\ w_i\\ w_j\\ w_n\end{array}\right)$$

(10)

$w_i$ is the weight of criteria $i$;

$\sum _{j=1}^n{\overline{A}}_{ij}$ is the sum of $A_{ij}$ by row $j$ from matrix B.

• Validation of the prioritized level

Matrix B is considered valid for comparing the criteria since the consistency ratio (CR) is below 0.1, which aligns with the threshold recommended by Saaty for the AHP method.

$$CR={\displaystyle \frac{CI}{RI}}$$

(11)

$CR$ is the consistency ratio;

$RI$ is the random index, introduced by Saaty (2008) [37], and is presented in

Table 6. Random Index based on number of criteria.

n	1	2	3	4	5	6	7	8	9	10
RI	0	0	0.58	0.90	1.12	1.24	1.32	1.41	1.45	1.49

below.

$CI$ is the Consistency Index (CI) and is obtained by using the following equation:

$$CI={\displaystyle \frac{{\lambda }_{max}-n}{n-1}}$$

(12)

$${\lambda }_{max}={\displaystyle \frac{\sum \frac{\sum _{j=1}^n{w}_i\mathrm{\ast }{A}_{ij}}{w_i}}{n}}$$

(13)

• Scoring for impacts of criteria on soil erosion and soil erosion classification

Based on their individual knowledge and experience, each participant independently evaluated the attributes of each criterion. The scoring scale used during the focus group discussion comprised four levels—slight, moderate, severe, and extreme—to correspond with the soil erosion classification derived from the USLE model. The final score for each attribute was determined as the geometric mean of all participants’ scores. The soil erosion rate for a specific area was then calculated using the following formula:

$$S=\sum _{i=1}^n{W}_i \mathrm{\ast } X_{ia}$$

(14)

where $S_a$ is the score of area $a$; $W_i$ is the weight of criterion $i$; $X_{ia}$ is the score of attributes of criteria $i$ for land area $a$; and $n$ is the number of criteria. The value of $X_{ia}$ ranges from 1 to 9, with higher values indicating a greater influence of the corresponding factor on soil erosion. The scoring of each soil erosion classification is shown in

Table 7. Scale for scoring according to AHP method.

$\mathbf{Score} \left(\mathbf{S}\right)$	Soil Erosion Classes
1–3	Slight soil erosion
3–5	Moderate soil erosion
5–7	Severe soil erosion
7–9	Extreme soil erosion

.

3. Results

3.1. The Input Factors of USLE

Based on the methods described above, combined with the data in Table 1 and processed with ArcGIS 10.3 software, five types of maps of the input data of the USLE model were created and are shown in Figure 2.

• R factor:

The rainfall erosivity (R) values gradually increase from the southwest to the northeast of the study area; however, the variation is minimal, with a difference in only about 1.04%. This indicates that rainfall contributes insignificantly to the spatial variability of soil erosion across the region.

• LS factor:

The LS factor, representing the combined effect of slope length and steepness, ranges from 0 to 7.48. High LS values are primarily concentrated in mountainous regions located in the western and southern parts of the study area.

• K factor:

The study area comprises four soil types, each assigned a specific soil erodibility (K) value. The dominant soil group is Acrisols, covering 15,152.58 ha with a K value of 0.23. The remaining soil types include Luvisols (4634.82 ha, K = 0.28), Ferralsols (1197.45 ha, K = 0.21), and Fluvisols (1012.05 ha, K = 0.15).

• C factor:

The cover management (C) factor ranges from 0.22 to 0.52, with an average of 0.31 and a standard deviation of 0.031. Areas with low C values are typically covered by natural forests or perennial vegetation and account for a considerable portion of the study area. In contrast, higher C values are observed in riparian zones, degraded forest areas, and recently planted lands with limited vegetation cover.

• P factor:

The support practice (P) factor predominantly has a value of 1, covering an area of 16,020.18 ha (approximately 72.83% of the total study area), which corresponds to natural forest areas with minimal human intervention. Lower P values (0.6–0.8) are observed in deforested regions with steep slopes and elevated terrain. The lowest P values are found in agricultural zones, such as paddy fields and areas under annual crop cultivation, where soil conservation measures are more commonly applied.

3.2. Community Perspectives on the Importance of Soil Erosion Factors

• The importance of the influenced factor to soil erosion

Table 8. The pairwise comparison matrix and weight of each factor on soil erosion assessment using the AHP method.

	Rainfall	Wind Speed	Soil Texture	Soil Depth	Land Use Types	Land Surface Cover	Elevation	Slope	Aspect	Weight
Rainfall	1.00	2.50	1.66	5.00	2.05	2.13	1.31	0.34	0.84	14.3
Wind speed	0.40	1.00	0.90	1.47	1.41	0.90	1.00	0.25	2.00	8.5
Soil texture	0.60	1.12	1.00	2.17	2.17	1.41	2.00	0.33	0.65	10.2
Soil depth	0.20	0.68	0.46	1.00	1.36	0.20	0.81	0.27	0.61	5.2
Land use types	0.49	0.71	0.46	0.74	1.00	1.32	0.80	0.25	0.67	6.2
Land surface cover	0.47	1.12	0.71	5.00	0.76	1.00	2.00	0.20	2.00	10.6
Elevation	0.76	1.00	0.50	1.23	1.25	0.50	1.00	0.32	0.87	7.2
Slope	2.98	4.01	3.01	3.68	4.01	5.00	3.15	1.00	2.91	28.7
Aspect	1.18	0.50	1.53	1.64	1.49	0.50	1.15	0.34	1.00	9.1
Consistency Index: 0.06 < 0.1 (The pairwise comparison result is accepted)

presents the pairwise comparison matrix constructed using the AHP to determine the importance of nine factors affecting soil erosion. The results indicate that slope is the most important factor, with the highest weight (28.7%), underscoring its significant role in erosion intensity. This is followed by rainfall (14.3%), land surface cover (10.6%), and soil texture (10.2%), all of which are critical in influencing erosion through hydrological and soil physical mechanisms. Aspect (9.1%), wind speed (8.5%), and elevation (7.2%) are also considered moderately important. Land use type (6.2%) and soil depth (5.2%) are ranked lower but still contribute to the overall erosion risk evaluation. Overall, the AHP analysis highlights that slope, rainfall, land cover, and soil texture are the dominant variables impacting soil erosion in the study area.

• The score of influenced factors

The participatory assessment process yielded a comprehensive classification of soil erosion risk based on nine selected factors. Thresholds for each factor were defined through group discussion among local stakeholders and experts, ensuring that the classifications were context-specific and grounded in local knowledge. The scoring of each threshold, reflecting perceived erosion risk, was conducted independently by participants. The results were subsequently aggregated and analyzed to identify the most critical conditions contributing to soil degradation. The scoring of each factor was shown as Figure 3.

Annual rainfall showed a clear positive correlation with erosion risk. Areas receiving over 3500 mm of rainfall annually were assigned the highest score of 9.00, indicating very high erosion potential due to intense rainfall and associated runoff. This category also encompassed the largest area (16,553 ha). Wind speed was also influential, particularly in areas experiencing high-speed conditions. Scores increased from 2.59 for areas with wind speeds below 2 m/s to 8.27 for areas exceeding 6 m/s, although the latter represented a relatively small spatial extent (23.4 ha). Soil texture influenced erosion susceptibility based on particle cohesion and infiltration characteristics. Clay soils had the lowest erosion score (3.66), while silt-dominated areas, known for their loose structure and high detachment rate, received a score of 6.27. Loam-clay soils were moderately vulnerable (5.16) and occupied the largest area among texture classes (14,779 ha). Soil depth was inversely related to erosion risk, with shallower soils (<30 cm) assigned a high score of 7.76, indicating high vulnerability due to limited infiltration and reduced root support. In contrast, soils deeper than 100 cm were scored at just 1.15, emphasizing their protective role against erosion. Notably, shallow soils covered a substantial area (15,944 ha), highlighting a potential management priority. Land use type played a prominent role in erosion dynamics. Forested areas had the lowest score (2.49), confirming their function as natural erosion barriers. Although forests are often seen as natural erosion barriers, steep or degraded forested slopes can generate runoff that intensifies erosion in adjacent areas. Annual cropping systems, due to frequent soil disturbance and low canopy cover, were rated highest (7.57) but covered only a limited area (341.64 ha). Perennial crops were assigned an intermediate score of 5.42. Land surface cover revealed that sparse vegetation cover (NDVI < 0.2) was associated with the highest erosion risk (8.27), though this condition was limited in extent. Conversely, areas with dense vegetation cover (NDVI > 0.6) received the lowest score (1.23), reflecting strong protective functions. Elevation and slope were also crucial determinants. Higher elevations (>600 m) and steeper slopes (>15°) corresponded with high erosion risk scores of 8.49 and 8.79, respectively. Steep slope areas, in particular, covered a substantial portion of the landscape (14,940 ha), underscoring their role as erosion hotspots. Finally, aspect influenced microclimatic conditions and erosion exposure. North and northeast-facing slopes received high scores (5.16 and 6.64, respectively), potentially due to their orientation toward prevailing rainfall and wind directions. Flat areas and southwestern exposures were assigned lower scores (<2), suggesting lower vulnerability.

3.3. Soil Erosion Situation in Nam Dong District

The overall trends observed in the maps generated by the USLE and AHP (Figure 4a,b) have a similar erosion trend, despite variations in the classification values. In both methods, the central and eastern parts of the study area are characterized by higher erosion risk, while lower-risk areas are primarily located in the western and northwestern regions. This consistent spatial pattern across the two models indicates a shared recognition of the most erosion-prone zones within the landscape. Both the USLE and AHP approaches classified the study area into four categories of erosion severity: slight, moderate, severe, and extreme. The results of the classification are detailed in

Table 9. Statistics of soil erosion classes by the USLE and AHP.

Soil Erosion Classification	Area (Hectare)
	USLE	AHP
Slight soil erosion	1963.17	0
Moderate soil erosion	3459.69	3554.64
Severe soil erosion	2925.90	18,140.58
Extreme soil erosion	13,648.14	301.68

.

Under the USLE model, the largest area falls into the extreme soil erosion category, which covers 13,648.14 hectares. This is followed by moderate erosion at 3459.69 hectares, severe erosion at 2925.90 hectares, and slight erosion at 1963.17 hectares. These results suggest that extreme and moderate erosion are the most dominant classes in terms of area distribution in the USLE output. In contrast, the AHP analysis identified the severe soil erosion class as the most extensive, with an area of 18,140.58 hectares. The moderate erosion class followed, covering 3554.64 hectares, while extreme erosion occupied a much smaller area of 301.68 hectares. Notably, no areas were classified under the slight erosion category in the AHP result.

4. Discussion

4.1. Soil Erosion Assessment Using USLE and AHP

Both models identified similar spatial trends in erosion distribution, with higher erosion risks concentrated in the mountainous central parts of Nam Dong district. However, they differed markedly in their classification of severity levels. The USLE model indicated that 13,648 ha (about 62%) of land area experienced extreme erosion. In contrast, the AHP approach classified the majority of the area (18,140 ha, 82%) into the severe erosion category, with only 301 ha marked as extreme, and no areas classified as experiencing slight erosion.

This divergence can be attributed to the fundamental differences in model structure and input methodology. While the USLE provides a long-term average based on empirical equations and pixel-level spatial data sensitive to cumulative slope and land cover effects, the AHP reflects current stakeholder perceptions of landscape degradation at the parcel level, incorporating recent land use changes often missed by remote sensing. The USLE model is a quantitative, empirical model built upon decades of field experimentation. It integrates factors such as rainfall erosivity, soil erodibility, slope length and steepness, land cover, and conservation practices to estimate long-term average soil loss. Its standardized, equation-based structure makes it well-suited for generating comparable results across different regions, provided localized parameters are carefully calibrated. Prior research has validated the robustness of the USLE in tropical environments when appropriate spatial data and soil inputs are available [8,32].

By contrast, the AHP model is a qualitative decision-support tool that relies on expert or stakeholder judgment to assign weights and scores to influencing factors. In this study, nine participants with diverse backgrounds—including local government staff, researchers, community leaders, and farmers—provided input. The group was gender-diverse (three females, six males) and ranged in age from 36 to 60, encompassing both institutional expertise and indigenous knowledge. This diversity likely contributed to a more holistic perspective on erosion risk, as participants could incorporate not only physical features but also practical experiences related to land use and farming systems. The tendency of the AHP to classify more land as “severely eroded” may reflect the heightened sensitivity of stakeholders to observable landscape degradation, especially in areas recently affected by intense rainfall or land use change. This has been echoed in recent studies, which found that farmers tend to assess erosion severity based on visible rill formation, crop loss, or surface runoff, while empirical models like the USLE estimate long-term averages [11]. Similarly, Hayatzadeh et al. (2023) demonstrated that AHP models can be influenced by socio-cultural perceptions, especially when involving older or more experienced participants [13].

Thus, while the USLE provides standardized and reproducible results useful for regional planning and policy formulation, the AHP captures contextual knowledge and localized erosion dynamics, especially valuable in areas lacking robust spatial data. In upland Vietnam, where topographic complexity and socio-ecological diversity are high, the participatory insights from the AHP can complement the USLE’s empirical estimates.

4.2. Limitations and Potential Integration of AHP and USLE for Soil Erosion Assessment

The USLE model, while quantitative and well-established, presents several shortcomings. First, it is designed primarily for estimating sheet and rill erosion and fails to account for other forms such as gully or mass movement erosion, which are often prominent in steep, mountainous terrains [7,8]. Second, its reliance on empirical factors such as the support practice factor (P) and the cover management factor (C) can be problematic in regions with limited empirical data. In this study, the P factor was derived from general land use and slope-based values due to limited field data. This may not fully reflect local soil conservation practices such as terracing or perennial cropping, potentially leading to overestimated erosion. Future studies should prioritize field surveys to localize the P factor more accurately [32,39]. Although we used a highly detailed soil classification map, the K factor values were derived from regional-scale secondary data and assigned to the corresponding soil types. While this approach is considered acceptable in some recent studies, it may not accurately reflect in situ soil properties, which can vary significantly due to micro-topography and land use history [7,24]. To address these limitations, several improvements can be implemented. First, integrating complementary physically based models or gully-specific erosion models is recommended to capture erosion processes—such as gully and mass movement erosion—that are not adequately assessed by the USLE in steep, mountainous environments [7,8,40,41]. Second, it is crucial to develop localized datasets for key USLE parameters. Field surveys and laboratory analyses can provide site-specific data for the P and C factors, reducing the reliance on default or literature-based values and enhancing the model’s sensitivity to local conservation practices and land management [24,32,42,43]. Moreover, direct laboratory analysis and detailed field sampling of local soils should be used to derive accurate K factor values, which reflect real in situ soil properties and account for micro-topographic variability [8,24]. Lastly, periodic calibration and validation of USLE outputs with observed soil loss measurements or sediment yield data are essential for increasing the credibility and predictive reliability of soil erosion assessments in complex terrains [44,45].

The AHP is subjective by nature, relying on stakeholder judgment for factor weighting and scoring, and while this allows the integration of local knowledge, it introduces bias and inconsistency, particularly if the participant group is small or lacks technical diversity. In this study, the sample consisted of nine participants with diversity in knowledge backgrounds and professions. This could have led to noise within the data and a less conclusive understanding of soil erosion drivers [36]. For example, in this study, most participants believed that forests act as natural barriers against soil erosion. While this is often true, especially under dense canopy conditions, in steep terrains, runoff generated from forested slopes can accumulate and concentrate downslope, potentially intensifying erosion in adjacent agricultural or disturbed areas—particularly where forest vegetation is sparse or degraded due to logging or land conversion [2,24]. Moreover, women’s participation in agricultural decision-making in Vietnam remains limited; a study using the Women’s Empowerment in Agriculture Index (WEAI) in ethnic minority communes of Central Vietnam found that women were disempowered in agricultural decision-making in up to 70% of households, compared to just 15% of men [46]. Although women accounted for 33% of participants in this study, gender imbalance may not have significantly influenced the results. Nevertheless, we recommend ensuring greater gender diversity in future AHP applications to capture more differentiated perspectives. Additionally, the AHP lacks the ability to simulate actual physical erosion processes. It provides a relative assessment of erosion risk but does not quantify soil loss in measurable units. This makes it challenging to calibrate or validate against observed erosion, limiting its predictive reliability [14]. There are some feasible solutions that have been suggested by recent studies to overcome these shortcomings such as expanding the number and diversity of participants and ensuring more balanced representation across key stakeholder groups, which could help reduce bias and increase the reliability of results, as a larger and more diverse group is likely to provide more stable and representative judgments [36,37]. Providing adequate training for participants on both the AHP methodology and the objects of study is also crucial for improving consistency and reducing errors associated with subjective judgments [47]. Additionally, incorporating fuzzy logic or sensitivity analysis within the AHP framework can help manage uncertainty and provide a more robust evaluation of expert input, particularly in complex, data-scarce environments [14].

Integrating the AHP and USLE presents a significant methodological challenge, as the AHP is inherently a weighted additive approach based on the linear combination of standardized criteria, while the USLE relies on a multiplicative model in which each factor is empirically derived and calibrated to yield an absolute estimate of soil loss. Therefore, directly incorporating AHP-derived weights into the original USLE is not scientifically valid and can compromise the accuracy of soil erosion assessment. Previous studies have attempted to apply AHP weights within the USLE framework; however, these efforts do not alter the original USLE multiplicative formula but rather use AHP-derived weights in a normalized or standardized version of the factors to produce relative erosion risk or susceptibility maps [48,49]. Consequently, it is advisable to apply the AHP and USLE as independent approaches. The results can then be compared, cross-validated, or used in combination to enhance the understanding of soil erosion dynamics in specific research areas [8,43]. Additionally, for site-specific studies, the AHP method can be employed to calibrate input variables of the USLE model—particularly those related to local farming practices, such as the P factor—rather than relying solely on secondary data. However, this approach requires thorough participant training and the provision of appropriate reference materials to ensure accurate understanding and consistent assessments. Expanding the research scope and involving multiple stakeholders is crucial for a comprehensive assessment of soil erosion, especially in evaluating downstream ecological impacts driven by land use-induced changes in local erosion and sediment transport through fluvial processes.

5. Conclusions

This study highlights the roles of the USLE and AHP models in assessing soil erosion in Nam Dong district, Central Vietnam. Both methods identified similar spatial trends, with the mountainous central region showing the highest erosion risk. However, the USLE classified 62% of land as extreme erosion, while AHP, based on community input from nine diverse participants, classified 82% as severe and only a small area as extreme. These differences reflect the empirical, quantitative nature of the USLE versus the perception-driven assessment of the AHP. Limitations include the USLE’s reliance on generalized factors and the AHP’s subjectivity. Integrating AHP-derived weights directly into the USLE formula is not scientifically justified. Instead, using both models independently and comparing their results provides a clearer, more reliable understanding of soil erosion risk, supporting targeted conservation planning in complex upland areas.