Introduction

Soil erosion is an essential driver of soil degradation (Wei et al. 2007) and is a major threat to natural environments and to societies that depend on local resources or experience significant geological and meteorological hazards (Wei et al. 2007). Soil erosion threatens sustainability when it is greater than soil formation rates and when it damages soil productivity (Rodrigo-Comino and Cerdà 2018). Approximately, ten million hectares of cropland are lost to soil erosion annually, reducing crop production that leads to worsening economies and increasing malnourishment (Cerdà et al. 2017a). Soil erosion degrades land-use quality and is common in agricultural watersheds (Amiri et al. 2019). Globally, soil erosion rates are 10–40 times greater than soil formation rates and this threatens food security and environmental quality (Pimentel 2006).

More than 60% of eroded soil ends up in lakes or rivers, increasing the likelihood of flooding and contaminating fresh water with sediment and chemicals (such as fertilizers and pesticides) that adhere to particles (Pimentel 2006). Many agricultural regions are susceptible to soil erosion due to intense tillage, the absence of vegetative cover, and certain soil characteristics. The use of biocides that disrupt biological activities like plant development, microbial action, and decomposition can block soil-forming processes (Cerdà et al. 2017b). Irrational land use and harmful farming practices decrease efficient, sustainable, and profitable uses of soils (Cerdà et al. 2017b) and degrade soil quality (Mhazo et al. 2016). Centuries-long cultural exploitation of soils has decreased soil nutrients and organic matter in many regions of the world (Cerdà et al. 2017b). Tillage has increased the bulk density of soils over the long term due to structural rearrangement of clays in the soil; however, furrowing reduces short-term soil densification and breaks soil surfaces to increase infiltration rates (Rodrigo-Comino and Cerdà 2018). Combining tillage with biocide applications increase soil erosion rates due to the lack of natural processes to counteract the disruption and can lead to severe soil degradation (Cerdà et al. 2017a). Herbicides affect the structure of soils and cause compaction due to decreasing inputs of organic matter. Accordingly, there are incremental increases of sediment yield and runoff (García-Díaz et al. 2017). Researchers have shown that herbicide use increases susceptibility to soil erosion (Keesstra et al. 2016). More robust vegetation and thicker surface cover decrease soil erosion rates (Thornes1990; Morgan 1996). Many soil erosion models incorporate measures of vegetation cover as an important soil erosion susceptibility factor (e.g., Thornes 1985; Morgan et al. 1998), since plants protect the soil surface by intercepting raindrops, by improving infiltration, by adding organic substances to soil, and by providing surface roughness to slow winds (e.g., Morgan 1996).

In general, land use–land cover (LULC) is the most important predictor of soil erosion susceptibility as LULC will differentially impede or augment sediment yields and surface runoff rates (Chen et al. 2001). Irregular patterns of land use and vegetation diversity are strong determinants of hydrological responses of watersheds (Siriwardena et al. 2006). Management of LULC not only enhances soil properties that decrease soil degradation (Wei et al. 2007) but also influences vegetation quality and endurance (Kosmas et al. 2000). Rapid soil erosion removes topsoil and, therefore, soil quality is diminished, flooding is promoted, agricultural activities can be disrupted, and the security of communities and regions can be threatened (Lal 2001). Severe soil erosion can damage dams and diminishes reservoirs and it may eventually negate their original purposes.

Management must discern the spatial distribution and magnitude of soil erosion to develop effective remedies. Soil erosion susceptibility models (SESMs) can enable soil erosion assessment which examines the processes and their interactions that influence erosion. Over the last few decades, several models have been developed to forecast soil erosion rates (Lal 2001). Quantitative analyses are vital for soil erosion risk assessment and soil erosion management, as they can more effectively summarize soil erosion rates and compare them to the spatial processes and characteristics that may influence those rates (Park et al. 2011).

Numerous techniques have been used to evaluate soil erosion at various scales—from local to regional and global (Poesen et al. 2003). Mainly two types of models (i.e., physical and empirical based model) have been used to assess the soil erosion rates at specific spatial scales. Examples of these are: the universal soil-loss equation (USLE) (Wischmeier and Smith 1978), the revised universal soil-loss equation (RUSLE) (Renard et al. 1991; Gayen et al. 2019), and the modified universal soil-loss equation (MUSLE) (Arekhi et al. 2012). Quantitatively precise predictions of soil erosion are difficult because soil erosion is a complex process that is influenced by land surface and soil characteristics, and environmental conditions (Oh and Jung 2005; Park et al. 2011). To contend with the challenge of spatially specific quantified predictions, different formulae have been devised to enhance statistical modeling (Oh and Jung 2005; Rahmati et al. 2017). Aerial photographs and very-high-resolution satellite images can also enhance assessment methods (Kropacek et al. 2016).

The quantitative relationships between soil erosion conditioning variables and their spatial distribution have been evaluated in GIS using fuzzy logic (Hembram and Saha 2018; Banerjee et al. 2018), artificial neural networks (Rahmati et al. 2017; Ghosh and Samantaray 2019), bivariate statistical models (Rahmati et al. 2016), sensitivity analysis (Mendicino 1999), multivariate adaptive regression spline (Conoscenti et al. 2018), analytical risk-evaluation methods (Wu and Wang 2007), the analytical hierarchy process(AHP) (Svoray et al. 2012; Arabameri et al. 2018a), logistic regression (LR) (Arabameri et al. 2018b),random forest model (Arabameri et al. 2019), weight of evidence (WofE) (Gayen and Saha 2017), frequency ratio models (Zabihi et al. 2018), classification and regression tree models (Gayen and Pourghasemi 2019), ensemble models, (Camilo et al. 2017), and index of entropy models (Ren et al. 2018).

This study evaluates the effectiveness of SESMs generated by fuzzy logic and AHP knowledge-based models for an agricultural catchment of Kunur River Basin. The novelty of this work is that it uses LR algorithms to determine the variables that are most consequential for predicting soil erosion. Soil erosion susceptibility modeling and mapping has never been conducted in the Kunur River Basin; therefore, the results provide useful insights for planners and policymakers who desire to conserve and manage soil resources.

Materials and methods

Study area

The Kunur Drainage Basin is a 6th-order sub-watershed of the Ajay River (Fig. 1a, b) which is situated in the Rarh Bengal region. This region is known as the Rarh Bengal riverine landscape and it lies between 23°25′50″N and 23°39′21″N and 87°16′26″E and 87°54′12.6″E. The elongated river basin has an area of about 75 km2 and is situated between the Khari River basin to the north and the Ajay River Basins into the south. The primary river channel of the Kunur is 112-km long. The middle portion of the watershed contains several reserves and protected forests. Geologically, the upper portion of this watershed is dominated by the Panchet formation, a formal Gondwan stratigraphic unit with sandstones, coals, shales, and clays. The soils are various types of loam and alluvium. The middle and upper parts of the basin are dominated by older and newer alluvium, respectively.

Fig. 1
figure 1

Location of the study area with soil erosion inventory map

Full size image

To begin, soil erosion locations were identified using global positioning system (GPS) and the areas of soil erosion were circumscribed as polygons using Google Earth. A total 80 soil erosion patches were identified in the study area. The characteristics of these patches were used to formulate the models and then to validate their outputs (Fig. 1c). This methodology was formulated as a flow chart (Fig. 2).

Fig. 2
figure 2

Flow chart of the proposed approach of soil erosion susceptibility assessment

Full size image

Soil erosion influence variables

To assess soil erosion susceptibility, 15 effective factors were selected for analysis as they were derived from a review of the published scholarship (Conoscenti et al. 2008; Rahmati et al. 2016; Zabihi et al. 2018), knowledge about the study area, and assessment of the data available in this region. These 15 variables are slope degree, slope aspect, elevation, soil type, overland flow, rainfall erosivity, normalized difference vegetation index (NDVI), sediment transport index (STI), topographic wetness index (TWI), annual rainfall, geomorphology, LULC, distance from lineament(faults, for instance), distance from river, and stream power index (SPI) (Table 1) and the data were derived from several sources and methods described below.

Table 1 Techniques of production of various thematic data layers
Full size table

Soil degradation is a threshold-dependent process under the influence of numerous effective variables (Valentin et al. 2005). The 15 erosion-promoting variables were mapped in the study area (Fig. 3a–o). Elevation (from an ASTER digital elevation model (DEM) at a 30-m resolution) is one of the most important topographical variables influencing the susceptibility of soil to erosion. The elevation in the basin varies from 28 to 131 m above mean sea level. The extent and rates of variation of elevation leads to different soil development and erosion processes (Hongchun et al. 2014) due to it changes of climates and vegetation (Aniya 1985). Slope aspect is another key factor (Nagarajan et al. 2000) as it influences rainfall, wind patterns, sunlight, and morphological structure of the watershed. Each of these variables also influences soil erosion rates (Komac 2006). Slope gradient is important (Lee and Min 2001) as steepness influences runoff, drainage intensity, and detachment of soil particles (Ercanoglu and Gokceoglu 2002; Komac 2006). Rainfall erosivity is a parameter in the universal soil-loss equation (USLE) and revised universal soil-loss equation (RUSLE) models. The rainfall erosivity factor (R) is a driving agent for sheet and rill erosion. Rainfall characteristics, like annual amount, seasonal distribution, and intensity, also influence soil erosion (Wischmeier and Smith 1978). The TWI is an important secondary topographical factor (Rahmati et al. 2016) that indicates water accumulation at a location. TWI is calculated by the applying Moore and Wilson (1991) equation.

Fig. 3
figure 3figure 3

Soil erosion conditioning factors a slope, b distance from lineament, c distance from river, d elevation, e aspect of slope, f soil texture, g stream power index, h topographical wetness index (TWI), i sediment transportation index, j length of overland flow, k rainfall, l land use/land cover (LULC), m NDVI, n geomorphology, o rainfall erosivity index

Full size image

This watershed includes six major LULC types: dense vegetation, sparse vegetation, agricultural land, water bodies, wetlands, and built-up areas. The LULC classification of the study area was produced with Landsat-8 OLI satellite data (from Earth Explorer at 30-m resolution) and field verification using a maximum likelihood classification algorithm (Hembram and Saha 2018) in ArcGIS. For validation of this LULC classification, the Cohen’s Kappa index test indicated 88% accuracy. NDVI was also developed from 24th March 2016 Landsat-8 OLI satellite data. Generally, NDVI is inversely related to soil erosion—for instance, wind erosion is reduced significantly when forests cover more than 20–35% of an area (Mirmousavi 2016). Soil types also influence erodibility, nutrient levels, and organic content of soils. There are eight soil types covering significant proportions of the study area: fine loamy, fine alluvium, fine–coarse loamy, fine–fine loamy, fine loamy sandy, fine loamy–coarse loamy, and gravely loamy. The coverage of these units was extracted from a 1:50,000-scale National Bureau of Soil Survey and Land Utilization Planning (NBSS&LUP) map (Fig. 3f). The distances of sites of erosion from the nearest linear features and rivers were computed in GIS from river and lineament vectors (Fig. 3b, c). Distance from a river is important due to the adverse influence of streams on slope strength as materials are removed from the bases of slopes (Saha et al. 2002). Surface lineaments represent weak zones that are highly permeable and low in resistance, capable of affecting slope stability and soil degradation. The length of overland flow of the Kunur River Basin was measured on a river-network map generated from a 1:50,000-scale topographical map. The SPI and STI are important predictors of soil erosion and both were mapped from an ASTER DEM in a GIS (SAGA-GIS). The SPI is a multi-topographic attribute (Moore and Wilson 1991) that is useful for computing the force of running water and the propensity to erode soil aided by the convex and concave shapes of the land surface (Gayen and Saha 2017). The geomorphic landforms in the basin are depositional and denudational plains. Geomorphic characteristics influence relief, slope, soil types and parent material, and the depositional nature of alluviums (Fig. 3n).

Spatial modeling of soil erosion susceptibility

Analytical hierarchy process (AHP)

The AHP method is a semi-quantitative, multi-objective, and multi-criteria method that was first used by Saaty (1980). This method is a multi-criteria decision-making approach in which decisions are made at specific scales by selecting preferences from a set of alternatives (Saaty 1980). This is a widely used model for susceptibility analysis, decision-making, and regional planning (Kayastha et al. 2013). The AHP model involves several steps: (1) define the unstructured problem and objectives of a study; (2) identify the variables that influence the problem and rearrange them into a hierarchical sequence; (3) rank values according to their subjective relevance to determine the relative importance of every factor, and (4) adjust factor ratings based on the priorities determined by the decision-makers (Saaty and Vargas 2001). In the construction of a pair-wise comparison matrix, every factor is rated against each of the others by assigning a value of relative dominance (ranging from 1 to 9) to the intersecting cell (Table 2). Finally, variables are ranked based on their relative weights from the pair-wise comparison matrix (Table 3).

Table 2 Scale of preference between two parameters in AHP (Saaty and Vargas 2001)
Full size table
Table 3 Rating of sub-classes of individual parameters
Full size table

The variables on the horizontal axis are less important than the variables on the vertical axis which are assigned values from 1 to 9. The importance of every factor to soil erosion susceptibility relates one to all the others and is made tangible by dual evolution reflected in the comparative importance of pairs. The values are determined by the relevance of variables in each location and these determinations are based on the choices made by decision-makers.

The consistency ratio (CR) is used to evaluate the probability that the matrix judgments were randomly chosen (Saaty 1977, 1980):

$${\text{CR}} = {\text{CI}}/{\text{RI,}}$$
(1)

where RI denotes the mean of the resulting consistency index depending on the matrix order provided by Saaty (1980) and CI denotes the consistency index expressed as:

$${\text{CI}} = \left( {\lambda_{ \text{max} } - n} \right)/\left( {n - 1} \right),$$
(2)

where λmax is the largest eigen value of the matrix, it can be easily calculated from the matrix; ‘n’ represents the matrix order. The CR is a ratio between the random index and matrix consistency index, and its value ranges from 0 to 1. A CR of 0.1 or less is interpreted to be a reasonable level of consistency and more than 0.1 represent indicates that revision is needed due to an inconsistent treatment for individual factor ratings (Malczewski 1999).

Fuzzy logic

Fuzzy logic (or fuzzy inference) is also known as a fuzzy expert system. It is a system constructed as a multi-valued, logic-based mathematical tool. Fuzzy logic systems simulate whole systems in a model to depict conjecture out of data inputs (Pradhan and Pirasteh 2010; Banerjee et al. 2018). The Mamdani method (Yanar 2003) is the most well-known fuzzy logic system. It uses min–max and if–then rules to improve classification and accountability (Akgun et al. 2012).This system is normally comprised of four stages: (1) fuzzification, (2) rule evaluation, (3) inference of fuzzy products, and (4) de-fuzzification (Tien Bui et al. 2011, 2012).

This fuzzy logic process was achieved using the fuzzy-inference tool in ArcGIS (v.10.3). Two tangible tools were employed: the membership fuzzy tool and the fuzzy-overlay tool. All data for all sites were converted into fuzzy sets and than classified by the expert into four categories based on expert judgement. These classes ranged from high susceptibility to low susceptibility. The large fuzzy membership operation was calculated using Eq. 3.

$$\mu (x) = \frac{1}{{1 + \left( {\frac{x}{{f_{j} }}} \right)^{{ - f_{i} }} }},$$
(3)

where fi = spread and fj = midpoint. The fuzzy if–then method was used to create zones of soil erosion vulnerability.

Variable importance using logistic regression (LR)

Logistic regression is a widely used multivariate statistical method for calculating the probability of an event (Rahmati et al. 2016; Lombardo and Mai 2018). LR seeks the best-fitting model to represent the relationship between a dichotomous dependent variable and a set of independent explanatory variables (Rahmati et al. 2016). The relative contribution of a number (n) of independent variables (x1, x2, x3···xn) on a dependent (y) is used to predict a logit transformation of probability based on the presence or absence of an event. LR does not directly explore susceptibility, but instead it provides probability values for each point or pixel from which a susceptibility map can be produced. There are no rules to determine selection of the independent variables, but the variables should be measurable and have some degree of contextual relationship to the dependent variable over the entire study area (Gayen and Saha 2018).

The maximum likelihood regression coefficient of an LR model (Süzen and Doyuran 2004) can be expressed as:

$$P = 1/1 + e^{z} ,$$
(4)

where P denotes the chances of occurrence of a phenomena and factor “z” may be expressed as:

$$Z = b_{0} + b_{1} x_{1} + b_{2} x_{2} + b_{3} x_{3} + \cdots + b_{n} x_{n}$$
(5)

where b0 is the intercept and n is the total number of independent variable. bi (i  = 0, 1, 2, 3…n) is the slope coefficient.

Results

The AHP and fuzzy logic SESMs were mapped using the GIS. The probabilities and the resulting classes for each unit area in the study region were regarded as the outputs from each method. The selected soil erosion variables were qualitatively compared and judged by whether and how much they positively enhanced the accuracy of the SESMs.

Soil erosion susceptibility mapping using AHP

The comparison matrix was built on the expert knowledge of decision-makers. The software produced individual weights for each of the erosion-promoting variables according to their relative importance. The software generated the CR value for each variable. The CR was 0.082, an acceptable value for pair-wise comparisons of consistency in the SESMs. NDVI with AHP priority (0.113), distance from river with AHP priority (0.110), LULC with AHP priority (0.107), rainfall erosivity index AHP priority (0.106), and elevation with AHP priority (0.099) were the most important variables for this watershed, whereas TWI, distance from lineament, slope aspect, and SPI were of least importance (Table 3). The individual variables and their pair-wise comparison matrix, CR, and AHP priority were also calculated (Table 3).

The AHP SESM was developed with the help of following equation:

$$\begin{aligned} {\text{ESMahp}} = \left( {S \times {\text{Wahp}}} \right) + \left( {{\text{DFL}} \times {\text{Wahp}}} \right) + \left( {{\text{DFR}} \times {\text{Wahp}}} \right) + \left( {E \times {\text{Wahp}}} \right) + \left( {S_{o} \times {\text{Wahp}}} \right) \hfill \\ + \left( {L_{of} \times {\text{Wahp}}} \right) + \left( {A \times {\text{Wahp}}} \right) + \left( {{\text{SPI}} \times {\text{Wahp}}} \right) + \left( {{\text{LULC}} \times {\text{Wahp}}} \right) + \left( {{\text{STI}} \times {\text{Wahp}}} \right) \hfill \\ + \left( {{\text{TWU}} \times {\text{Wahp}}} \right) + \left( {R_{ei} \times {\text{Wahp}}} \right) + \left( {{\text{NDVI}} \times {\text{Wahp}}} \right) + \left( {R \times {\text{Wahp}}} \right) + \left( {{\text{Gmp}} \times {\text{Wahp}}} \right), \hfill \\ \end{aligned}$$
(6)

where Wahp is the weight of each soil erosion conditioning factor. Finally, based on AHP combination, a soil erosion susceptibility map was generated for the Kunur River basin. The soil erosion susceptibility values were subdivided into four classes by applying natural-break statistics (Shahabi et al. 2014). These are low, moderate, high, and very high susceptibility zones and they were mapped (Fig. 4). Measurements of the classes on the mapped AHP SESM showed that 38.1% of the total area has low susceptibility to soil erosion. The medium and high zones comprised 30.9% and 21.0% of the watershed. The zone of very high soil erosion vulnerability is 10.1% of the watershed (Fig. 5).

Fig. 4
figure 4

Critical soil erosion areas a by AHP, b by fuzzy logic

Full size image
Fig. 5
figure 5

Percentage of soil erosion susceptibility areas that fall into the various classes of AHP and Fuzzy logic Susceptibility map

Full size image

Soil erosion susceptibility mapping using fuzzy logic

In the fuzzy environment, the membership function for each conditional factor was determined by its values across the entire watershed. The membership function is used to generate the fuzzy layers of control points or input variables. Each predisposing factor changes the input variables into a 0–255 range which is dimensionless and unidirectional (Bharath et al. 2014). It is a common practice to overlay data layers with different unit values. Several methods are used to combine fuzzy layers to maintain mathematical consistency (Lisetskii 2008). The map for each factor was layered to create a soil erosion map using the fuzzy operations OR, AND, an algebraic sum, an algebraic product, and a gamma operation in which the gamma value was set within the range of 0.1–0.9 to assess its effect on the SESM (Fig. 6).

Fig. 6
figure 6

Membership value plots of four soil erosion susceptibility zones of the fuzzy logic model

Full size image

The fuzzy-operation-generated SESM was derived and the final fuzzy membership values were calculated (Table 3). The output values were grouped into 4 classes of susceptibility: low, moderate, high, and very high (Fig. 4). The fuzzy logic SESM classified 36.2% of the watershed as having low susceptibility to erosion. Moderately and highly susceptible zones amounted to 37.2% and 14.8% of the watershed, respectively. The zone of very high soil erosion susceptibility was covered 11.9%.

Validation of soil erosion maps

Validation is an important stepin assessing a soil erosion prediction model (Ayalew and Yamagishi 2005). Sixty-five soil erosion locations were used to test the prediction models. An ROC curve was used to validate both models. This ROC curve graphs the true-positive rates (TPR) with respect to false-positive rates (FPR) at assorted thresholds. The area under the curve (AUC) is defined as the rate of goodness. So, ROC curve can be considered as the measure of goodness. This ROC curve is generated by plotting the points with respect to the of proportions of 1-specificity and sensitivity. This curve has been used to measure the accuracy of a predictive rule (Pradhan 2010; Pourghasemi et al. 2012, 2018). Rashid et al. (2016) classified the values of AUC into five levels of accuracy: excellent (0.90–1.00), good (0.80–0.90), satisfactory (0.70–0.80), poor (0.60–0.70), and failing (0.50–0.60). SPSS statistics software (v.17.0) was used to produce the ROC curve and generate the results (Fig. 7). The results showed that the AUC of the fuzzy logic model is 0.914 (91.4%) and is 0.897 for the AHP model. The fuzzy logic model’s results would provide excellent capacity to assess soil erosion susceptibility.

Fig. 7
figure 7

ROC curve for both susceptibility models (the AHP and fuzzy logic)

Full size image

Discussion

The soil erosion susceptibility results are discussed in two parts: an evaluation of the models’ performances and comparison of their results and the identification of the characteristics most likely to increase susceptibility to soil erosion throughout the study region.

The model’s performance and their comparison

Knowledge-based and machine-learning models reveal two aspects of susceptibility. The AHP SESM generated an AUC of 0.88, whereas the fuzzy logic SESM generated an AUC of 0.91. The fuzzy logic model was more accurate in this region. Previous research revealed similar results with regard to landslide susceptibility assessment (Pourghasemi et al. 2012). Fuzzy logic provides better results as it allows more flexible integration of weighted maps and is also readily incorporated into a GIS modeling environment (Pradhan 2010). The fuzzy logic model contrasts with data-driven approaches like LR, weights of evidence, and AHP, which use soil erosion locations or known objects to estimate coefficients (Pradhan 2010). The main drawbacks of the AHP model are: the several pair-wise comparisons become very large in this model and can sometimes generate good scores for some criteria and bad scores for the others, obscuring or losing important information (Arabameri et al. 2018a). Rahmati et al. (2016) produced a gully erosion susceptibility map using two data-driven models—WofE and FR. They concluded that the FR model (AUC = 78.11%) had better prediction capacity compared to the WofE model (AUC = 70.07%). Based on these results, we can conclude that both models provide adequate results (i.e., both models provide good prediction power) for susceptibility assessment in this watershed; however, the fuzzy logic model exceeded the accuracy of the AHP model.

Identifying important soil erosion conditioning variables using LR

In this study, soil erosion was defined by a dichotomous variable (erosion absent or erosion present) that potentially depends upon eight of the independent variables—slope steepness, slope aspect, SPI, TWI, LULC, elevation, slope length, and geology. After eliminating collinear independent variables, the test data were input into a LR algorithm in SPSS to calculate the correlation of each factor with the occurrence of landslides. LR was conducted a second time but incorporated only the variables that were correlated with and made a significant contribution to predicting the historical and slides in the watershed. The results revealed that soil erosion rates are noticeably and positively determined by relief (Wald = 553.84, Exp (B) = 1.10, df = 1), NDVI (Wald = 314.90, Exp (B) = 1.0, df = 1), annual rainfall (Wald = 150.01, Exp (B) = 0.98, df = 1), rainfall erosivity (Wald = 145.66, Exp (B) = 1.54, df = 1), distance from river (Wald = 129.75, Exp (B) = 0.97, df = 1), length of overland flow (Wald = 45.47, Exp (B) = 1.54, df = 1), fine–coarse loamy soil (Wald = 20.43, Exp (B) = 2.39, df = 1), and gravelly loamy (Wald = 22.87, Exp (B) = 2.63, df = 1), each having a measure of statistical significance (Table 4). Therefore, we can say that soil erosion is not caused by a single factor, but rather it can be predicted using a combination of several variables. These variables (relief, NDVI, distance from river, annual rainfall, rainfall erosivity, length of overland flow, and fine loamy and gravelly loamy soils) have the strongest relationships with soil erosion susceptibility in the basin. The other variables (TWI, SPI, slope aspect, and STI) are less important variables for the process of predicting the creation of or the soil erosion susceptibility of lands in this watershed.

Table 4 Result of logistic regression (LR)
Full size table

Conclusion

Soil erosion is a significant problem in the agriculturally developed river basins of eastern India. For a comprehensive regional development plan, a rigorous evaluation of soil erosion susceptibility is essential. To get a clear understanding of the distribution of soil erosion and degradation and to identify the areas at highest risk for erosion caused by development and agriculture, the drivers of soil erosion in the Kunur River Basin were studied using two different methods of modeling to evaluate soil erosion susceptibility. Fuzzy logic and the analytical hierarchy process were employed to model the relationships between soil erosion and fifteen important variables. The relationships revealed that the probability of soil erosion is highest in the upper portion of the basin due to human land-use activities (such as rapid deforestation) and site and situational variables (such as proximity to a lineament, and moderately steep slopes). The models were operationalized in GIS and were validated against a set of patches of soil erosion identified using analysis of remotely sensed data and field study. The resulting SESMs produced high-quality maps. Validation tests of the SESMs confirmed the satisfactory performance of the predictions of soil erosion compared the ground surface soil erosion locations. To evaluate the accuracy and predictive supremacy of the SESMs, ROC curves were scrutinized to confirm that the models are good predictors of soil erosion susceptibility.

Agricultural areas are more prone to soil degradation due to, among other things, periods during which there is a lack of vegetative cover and periods of intensive tilling, both of which reduce soil-forming processes, poor land management strategies and other activities that damage soils. The results show that factor selection combined with either fuzzy logic or AHP is capable of accurately identifying the areas that are likely to experience soil erosion. We suggest that future work should focus on the mechanisms of soil erosion in this agricultural watershed. Therefore, SESM-generated maps of soil erosion susceptibility can be valuable for soil conservation and sustainable planning of soil erosion-prone areas like the Kunur River basin, India.