Influence of land use activities on predicted soil loss in a semi-arid river basin

Soil erosion is a complex phenomenon occurring on the Earth (Judson et al. 1987) assessed the world soil erosion rate and found it to be 26.5 billion tonnes per year. During the late 20th century, the world lost one-third of arable land productive soil by erosion (Pimentel et al. 1995) and the estimate suggests that the loss continues at mean rates between 12 and 15 t/ha/year (FAO and ITPS 2015). Soil erosion has negative impact on soil's ability to operate as a biodiverse organism that endorses life forms (NRCS 2019) and results in serious consequences for environmental and human sustainability by altering agricultural yield, biodiversity, landscape characteristics, and hydrology of a region (Fayas et al. 2019; Jodhani et al. 2023b, 2024d). The dissociation, transfer and subsequent accumulation of soil particles by geomorphologic operatives (water or wind) are causing soil erosion. These depends on the degree of association between various factors, such as soil properties, topography, climatic conditions, ground cover and human-induced agricultural practices, deforestation, demographic growth, etc. (Jodhani et al. 2021, 2023a; Bhatt et al. 2024a).

There are two methods to estimate the soil loss, namely modeling and experiment (Bhatt et al. 2024b). The experiment-based methods such as fixed-point observation (Blanco-Sepúlveda 2018; Boardman & Evans 2020), laser scanning (Yermolaev et al. 2018; Li et al. 2020), experimental plots (Loughran 1989; Boix-Fayos et al. 2006; Giambastiani et al. 2022), and radioisotope tracers (Sac et al. 2008; Zhang et al. 2018) are limited in use due to high cost and suitability for small scale areas. The process-based models, i.e. Water Erosion Prediction Project (WEPP; Nearing et al. 1989), European Soil Erosion Model (Morgan et al. 1998), Chemicals Runoff Erosion in Agricultural Management Systems (Line & Meyer 1988) models have complex structure, and intensive input data driven model and hence difficult to apply to large-scale regions and are not mature (Renard et al. 1997; Jetten et al. 2003; Panagos et al. 2015). Empirical models were developed in the 1980s namely Universal Soil Loss Equation (USLE) (Wischmeier & Smith 1978), Revised Universal Soil Loss Equation (RUSLE) (Renard et al. 2017; Jodhani et al. 2023c; Dzwairo et al. 2024; Patel et al. 2024), and Modified Universal Soil Loss Equation (MUSLE) (Williams 1975)) are relatively less complex and required limited input data for assessing the soil loss in large-scale regions. Many scholars have applied RUSLE model and suggested it has a great potential to estimate soil loss at small- and large-scale regions and improve the agriculture productivity (Kulimushi et al. 2021; Achu & Thomas 2023; Guduru & Jilo 2023; Sartori et al. 2024). Few scholars have applied RUSLE to predict soil erosion under both combined and isolated global climate and land use and land cover (LULC) scenarios (Sun et al. 2014; Biddoccu et al. 2020; Borrelli et al. 2020; Kumar & Singh 2021; De Silva et al. 2023). Plant biomass-covered lands are less prone to erosion (Šamonil et al. 2023). These two agents are becoming more pronounced due to the changing LULC (Panagos et al. 2015; Fashae et al. 2017, 2020). The estimated the slope length factor (LS-factor) using a high-resolution (25 m) digital elevation model (DEM) for the whole European Union, for identifying the areas at risk of soil erosion. The LS-factor modification experiment was carried out in Swiss alpine grasslands improved the prediction of soil erosion risk (Schmidt et al. 2019; Jodhani et al. 2024a).

In India, many climatic regions face extensive productive soil loss (Ravi et al. 2010; MoEF 2011; FAO and ITPS 2015) with an average soil loss of 16.4 t/ha (CSWCRTI 2011; Bhattacharyya et al. 2015). Trivedi et al. (2010) reported that 29% of India's geographical area is under severe soil erosion. Sehgal & Abrol (1994) found 187.80 Mha of land in the degraded category due to erosion. Sharda et al. (2013) noticed that around 91% of the country is subjected to soil erosion. In India, two agents are primarily responsible for soil erosion; water and wind (Kumar et al. 1995). Surprisingly, soil erosion has been a persistent concern in central India's marginal alluvial plains, particularly in and around the Yamuna River Basin, wreaking havoc on crop output and quality of water (Singh & Phadke 2006; Bhatt et al. 2024b; Jodhani et al. 2024b, 2024c).

However, the accurate estimate of the slope length factor (LS) was often not compared for the region specific and when applying the models to assess soil erosion. LS-factor represents the sensitivity of soil to denudation and transport by external erosional forces, steeper and longer the slope poses higher the risk for erosion in the basin. Further, majority of the study lacking the future role of LULC change on the soil loss. In this study we aimed to assess the DEM sensitivity and estimated future, P-factor and C-factor, further the LS-factor was estimated for different DEMs (SRTM, ASTER, MERIT, AW3D30, NASADEM, CARTOSAT, and TanDEM-X). Previously the LS-factor was not evaluated for different DEMs in Central India. Based on this the soil loss was estimated for different LULC scenarios (2030 and 2040).

The southwest monsoon brings rain and the average annual rainfall is 1,138 mm. The summer is hot, and the winter is cool, with an average minimum and maximum temperature of 6.7 and 44.2 °C, respectively. In the higher Vindhyan reaches (southern basin), medium black soils predominate, followed by mixed red and black soils in the intermediate reaches (central basin), and alluvial soils in the lower plains (northern basin). Agriculture is the most common land use practice, and due to a lack of irrigation systems, it is mostly rainfed. The principal crops are millet, wheat, and gram grown on the majority of the basin's land. The basin's forest is a tropical dry deciduous type, degraded in the BRB's northern section but abundant in the southeast.

Scientists have developed various DEM data, which is freely available for many applications. DEM data are commonly used in soil loss, hydrological, infrastructural, planning and topographic analysis. For soil loss, the LS- factor can be derived from these DEM and LS-factor together constitute the topographic factor, reflecting the surface topography's effect on soil erosion. It has been demonstrated that alterations in slope steepness are far more susceptible to differences in slope length than variations in slope steepness. In this research, we employed eight DEM sources, including TanDEM-X DEM, MERIT DEM, CARTOSAT DEM, SRTM DEM (V3 and V4-1), ALOS DEM, AW3D30 and NASADEM (as shown in Table 1).

This factor indicates soil's potential sensitivity to erosive agents that separate and transfer loose particles (Bhatt et al. 2024a). It suggests that intrinsic soil characteristics and profile qualities have an impact. K-factor ranges from 0.10 (soils with a lot of clay) to 0.45 (soils with a lot of sand).

LS-factor can be easily calculated based on DEM (Kumar & Singh 2021). However, DEMs resolution and accuracy can vary depending on the production technique, which impacts on the LS computation. For the LS-factor calculation, we have used Desmet & Govers (1996) method, which is available in SAGA software. This method has been chosen due to its suitability for landscape-scale soil erosion modeling of complicated topography.

This factor was calculated by examining the LULC and allocating an arbitrary fixed value to each LULC category (Patel et al. 2024). The types of vegetation, density, cropping and other management practices influence soil erosion (Šamonil et al. 2023). C-factor is the most critical factor in predicting future soil loss (Achu & Thomas 2023). It is a dimensional less factor with a range of 0–1 (total lack of cover), resulting into a severe erosion risk (Kumar & Singh 2021).

It reflects techniques such as strip cropping, terraces, contouring, and silt fences that are used to minimize soil erosion (Kundu et al. 2024; Pandey et al. 2024). The P-factor value ranges between 0 and 1, with P equal 1 indicating poor conservation practices, P around 0 indicating poor conservation practices, and P around 0 indicating a good human-made erosion-resistant facility (Bhatt et al. 2024b).

The earth observation datasets were pre-processed after applying atmospheric and geometric corrections, hence these operations increase image interpretability. A total of six satellite images were used for each period of investigation. The supervised classification approach was applied to classify the earth observation data using maximum likelihood algorithm and prepared maps of 2005, 2010 and 2015. These classified images are considered as input data in the LULC change modeling package developed by Clark Labs at Clark University. The land change modeler (LCM) has two algorithms namely Markov chain (MC) and Multi-Layer Perceptron neural network (MLPNN) (Singh et al. 2015; Singh et al. 2022). The LCM assesses the trend of variation trend from one LULC class to another and predicts the land use pattern with respect to the previous variation trend considering components such as slope, aspects, roads, and water bodies. The LCM model was applied to predict the 2015 image based on the previous year (2005) and final year (2010), further the predicted image of year 2015 was compared with the classified image (reference image) of 2015. Afterwards the year 2030 and 2040 image was predicted.

The hypsometric integral and curves were generated using the CalHypso tool available in QGIS software as a separate plug-in. Hypsometry aims to find a dimensionless ratio of the basin's cross-sectional area to altitude (Dowling et al. 1998). According to Strahler (1952), hypsometric analysis can help determine erosion status at different levels. In addition, the convex upward curve denotes the basin's young stage, whereas the concave upward curve denotes the basin's old stage (peneplain/monadnock). Strahler (1952) suggested that an S-shaped curve with an upper portion concavity and a lower portion convexity represents the mature stage. According to Ritter et al. (2002), mature stage sub-watersheds will experience slight erosion, but it may be severe during high runoff periods or in the sense of entrenched meandering. As a result, the BRB's, hypsometric curve (HC) and hypsometric integral (HI) were evaluated in order to understand sub-watersheds health better.

LULC has a major significant influence on soil erosion. After classification, seven classes are obtained: agricultural land, built-up areas, dense forest, open/degraded forest, shrub land, and wasteland and water body. The overall accuracies for 2005, 2010, and 2015 were 91, 93, and 95%, respectively. The findings indicate that in 2015, the BRB is dominated by agriculture (76.40%), open forest (10.23%), dense forest (6.72%), water body (3.26%), wasteland (1.36%), shrub land (1.2%) and followed by built-up areas (0.83%), respectively. Consequently, LULC of 2030 and 2040 have revealed evidence that three LULC classes would experience a growing trend. In 2030 a slight growth was observed in the built-up area, which covers 1.03%, agricultural land shares 76.79% and dense forest will cover 8.55%, respectively. Built-up areal extent will be doubled in 2040 of 2015 and will occupy 1.66%. Meanwhile, agricultural land will also slightly increase to 76.97%, whereas dense forest will increase to 11.17%. The result indicates that dense forest and built-up areas will increase due to adoption for improved forest management plans and population growth.

The BRB's R-factor was varied between 1,734.86 and 3,640.77 MJ mm/ha/h/year. The R-factor spatial pattern showed a high value occupied by southwest and southeast portion followed by the middle part. The northern portion of the basin relatively showed low to medium value, which might accompany relatively low to medium value, which might accompany the lower rainfall received by the southwest monsoon pattern. The K-factor varied widely approximately 0.02–0.015. The BRB's topmost southwest segment had a lesser K-factor, whereas it was higher in the mid and north portions. The area has a high percentage of loam, sandy loam, and silt loam.

The dynamic parameters (C and P-factor) were actively detected for LULC classes (2015, 2030, and 2040), because changes in these input variables might affect erosion rate. The spatio-temporal variation in C-factor was observed in the drainage basins as a direct consequence of LULC change. This is because of large intensity of the values that were widespread along the agricultural plains in 2015, and then compelled to disperse all around areas modified primarily to use for establishment and farming in the years 2030 and 2040.

Since these LULC categories of dense forest, open forest, and shrub land lack any other kind of support practices, a P-value of 1.0 was allocated to such land use types. The upper basin is characterized by heavy industrialization and urbanization. On the middle section of the basin is driven by farming practices, especially the harvesting of pulses (beans, chickpeas and lentils). Stone crushing and thermal power plants are involved in the lower reaches.

The LS-factor was calculated using different resolution DEMs such as ASTER 30-m, AW3D 30-m, NASA30-m, SRTM 30-m, CARTOSAT 30-m, MERIT 90-m, SRTM 90-m and TanDEM-X 90-m, which depict the spatial variability in length and slope. The minimum and maximum LS-factor values obtained from DEM as ASTER 30-m, ALOS 30-m, NASA 30-m, SRTM 30-m, CARTOSAT 32m, MERIT 90-m, SRTM 90-m, and TanDEM-X 90-m are as follows 0.03–19.96; 0.03–16.38; 0.03–18.61; 0.03–13.11; 0.03–12.67; 0.03–15.49; 1.46–25.46 and 1.61–51.21, respectively (Table 2). The highest LS-factor value is attached to the study region's upper southeast and middle sections, while the lowest LS-factor is found in the bottom part of the basin for all the DEMs (Supplementary material, Appendix, Figures 2–5(a)–(b), Figures 3(a) and 3(b), 4(a) and 4(b), and 5(a) and 5(b)). Meanwhile, variation in erosion factors suggest high soil loss from basin's lower and middle portions, that are characterized by high R-factor, high C-factor and high LS-factor and lower value of K-factor (Figures 3(a) and 3(b), 4(a) and 4(b) and 5(a) and 5(b)). Meanwhile, variation in erosion factors suggests high soil loss from the lower and middle portions of the basin that are characterized by high R-factor, high C-factor and high LS-factor and lower value of K-factor.

Soil loss data from gauge observation were used to evaluate the RUSLE results for all DEMs and C- and P-factors. The baseline data was considered for 2015 and significant variation was reported for 2030–2040. The average soil loss for 2015 based on LS-factor of DEMs (ASTER 30-m, AW3D 30-m, NASA30-m, SRTM 30-m, CARTOSAT 30-m, MERIT 90-m, SRTM 90-m and TanDEM-X 90-m) is found to be 22.59; 19; 20.87; 16.55; 16.21; 4.2; 27.76, and 40.48 t/ha/year, respectively. Further, the RUSLE outputs were evaluated with the ground-based gauge observation data of the Central Water Commission (CWC), India for the same period. The erosion estimate based on SRTM 30-m is closer to the observation followed by the CARTOSAT 30-m DEM (Table 3).

The difference between the estimated and observed soil erosion rate is 5.4% for SRTM 30-m DEM and 7.3% for CARTOSAT 30-m DEM. The estimated soil erosion rate based on SRTM 30-m DEM ranged from 0.03 to 67.57 t/ha/year, while CARTOSAT 30-m DEM-based soil erosion rate ranged from 0.03 to 65.45 t/ha/year for year 2015. Maximum value of soil erosion rate was obtained in the upper southeast and southwest portion of the basin followed by the middle portion where degraded forest and shrub land are abundant in the year 2015. The northern part occupied the lower soil erosion rate, which closely reflected the R-factor in the basin.

A decrease in the estimated average soil erosion could be observed from 2015 LULC to 2040 LULC conditions. This decline is attributed to better forest management practices and other such steps that have been implemented by the local administration and communities. The reduction in soil loss was especially noticeable for SRTM 30-m DEM, followed by the CARTOSAT 30-m DEM for the near future 2040. The BRB has comparatively higher soil erosion rates than other sub-arid areas in India which can lower soil erosion rates than other sub-arid areas in India, ascribed to disparities in precipitation, vegetative cover, and basin relief.

RUSLE an empirical model was used to estimate the future soil loss through selecting the different DEMs. Kumar & Singh (2021) have studied the sensitivity of LS-factor for the Himalayan River basin. The slope steepness factor played a critical role in determining the soil loss from a study area. The LS-factor was determined for ASTER 30-m, AW3D 30-m, NASA30-m, SRTM 30-m, CARTOSAT 30-m, MERIT 90-m, SRTM 90-m and TanDEM-X 90-m.

The enhanced erosion rate in shrub land and open/degraded forest areas, especially when compared to LULC types with human signature verification (agriculture land and settlements), can be directly linked to topographic effects. These effects persist as greater LS-factors in these LULC types. However, a significant variation in soil loss was noted not just in farmlands but in shrub land and open/degraded forest areas, which can be perceived as having more severe impacts on tracking total soil loss using the RUSLE. The shrub land and open/degraded forest are primarily dispersed across the highly undulating topography, particularly in the BRB's upriver parts, which helps to explain why these LULC types have rising LS-factor values.

Greater slope lengths show a greater LS-factor, according to Hoyos (2005). The region with concave topography, where overland flow converges, has a high LS-factor. According to Nizar et al. (2024), the region where flow diverges (convex topography) is where the lower slope length exhibits the lower LS value. According to Nizar et al. (2024), the regions with greater LS factors also have higher steepness values.

Although there are fewer forest areas in the basin's bottom part, the influential forest types are dry deciduous open/degraded forest, which have a higher C-factor (Asempah et al. 2024). Consequently, a greater LS and C-factor value of open/degraded forest and shrub land resulted into higher soil erosion rate. In contrast, lower LS and C-factor value in agricultural land contribute to lower erosion rate. Nizar et al. (2024) suggested that lower LS- and C-factor is reduces soil loss. Furthermore, soil conservation strategies adopted in agricultural fields and farms (i.e., low P-factor value) for lower soil erosion rates in these LULC types. The future LULC and projected rainfall impact on soil erosion for Sri Lanka was explained by De Silva et al. (2023). Yu et al. (2024) assessed how human-induced road cutting/development affects the C and P factors and they enhanced the erosion rate. Pandey et al. (2024) outlined that the soil loss affects the carbon budget in the forested land. The soils were formulated from Vindhyan rocks rich in gneiss and granites from the Deccan trap, as well as highly ferruginous beds and occasionally limestone (Pandey et al. 2023). The basin has a high precipitation density (30–50 mm/h) triggers soil loss. Groundwater recharge is poor because groundwater in the catchment area is mainly constrained to secondary porosity zones formed in rocks (Pandey et al. 2022). The basin is surrounded by sandstone hills and is blanketed by categories of rigid rocks of varying ages. Several studies have reported the presence of the Bundelkhand craton in the basin, which consists of highly resistant rocks that generally resist degradation (Singh et al. 2023; Bhatt et al. 2025).

According to this study's findings, a lower resolution DEM may lead to a decrease in slope gradient. It directly affects soil erosion; thus, the DEM grid size should be adequately chosen to accurately represent the landscape topographic features. Thus, assessment of the impact of different resolutions open-source DEMs on soil erosion modeling is quite significant. However, fewer studies have evaluated their effects on soil erosion using open-source DEMs. Further, the erosion rate is expected to increase in the 21st century due to an ever-increasing population, accelerated land use and changes in the climate (Asempah et al. 2024) will deteriorate the soil quality and agricultural yield in the Himalayan region (Mandal et al. 2023). de Souza Batista et al. (2024) explains that the nature-based and bioengineering approaches will reduce the soil loss from the tropical environment.

The RUSLE is an empirical model and more reliable to estimate soil loss from plot. The satellite-based estimation of LS factors needs to be validated. There is changing climate which usually affects the accurate estimation of rainfall erosivity factors. The different pixels need to be resampled at same pixel size for better estimation (Kumar & Singh 2021; Kumar et al. 2022). The RUSLE model is reliable for sheet and rill erosion and it does not predict the sediment yield (Dzwairo et al. 2024). Under the extreme rainfall events (storms) it is less accurate for soil loss estimation. The model has limitations in terms of estimation of sediment export, sediment deposition, total soil loss (USLE) and soil erosion potential (Patel et al. 2024). However, InVEST model estimates sediment export, sediment deposition, total soil loss (USLE) and soil erosion potential (Patel et al. 2024). The input data accuracy severely affects the model output. The integration of RUSLE and machine learning strategy will enhance the soil loss accuracy in tropical plateau region (Kundu et al. 2024). The future rainfall erosivity was not considered the dynamic which may reduce the results of the study. The future impact of climate change on soil erosion rate in a tropical Indian catchment was studied by Anbazhagan et al. (2024).

The RUSLE was used in this study to investigate the effectiveness of LULC and DEM resolution in the BRB for soil loss. The outputs of RUSLE based on different DEMs have been validated with in-situ data obtained from the Central Water Commission (CWC), India. DEMs, which have provided the closest soil erosion rate to the observed value was selected to estimate soil loss under different year LULC. The conservation and management practices for the years 2015, 2030, and 2040 were evaluated and soil loss was estimated. The soil loss for SRTM_30_FILLED DEM was 16.55, 14.46, and 14.51 t/ha/year, respectively, for 2015, 2030 and 2040. This is due to loss of agricultural land and vegetation coverage at the expense of built-up areas. The SW_6 is in its old stage as per HC/HI and less prone to future erosion. Furthermore, the analysis suggests that the southeast and middle BRB require critical intervention for sustainable agriculture by adopting bioengineering strategy to reduce soil loss. Because these areas have important perennial sources of running water in the basin and high moisture content leads to rapid rock weathering. We recommend that nature-based solutions should be applied throughout the study region to reduce soil losses. Finally, this will help to make plans and projects for effective conservation of erosion areas.

The authors sincerely thank different agencies for providing the DEM data at no cost.

All relevant data are included in the paper or its Supplementary Information.

The authors declare there is no conflict.