Digital mapping of soil erodibility factor in response to land use change using machine learning models
Abstract Understanding the spatial variability of soil erodibility and its associated indices across different land uses is critical for sustainable land use planning and management. Traditional methods for measuring these variables are often time-consuming and costly. To address this, the study emp...
Saved in:
| Main Authors: | , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
SpringerOpen
2025-06-01
|
| Series: | Environmental Systems Research |
| Subjects: | |
| Online Access: | https://doi.org/10.1186/s40068-025-00402-w |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1850111529068789760 |
|---|---|
| author | Wudu Abiye Orhan Dengiz |
| author_facet | Wudu Abiye Orhan Dengiz |
| author_sort | Wudu Abiye |
| collection | DOAJ |
| description | Abstract Understanding the spatial variability of soil erodibility and its associated indices across different land uses is critical for sustainable land use planning and management. Traditional methods for measuring these variables are often time-consuming and costly. To address this, the study employed digital soil mapping (DSM) and machine learning (ML) models as efficient and cost-effective alternatives to predict soil erodibility and its indices, including clay ratio, critical level of organic matter, crust formation, dispersion ratio, and soil aggregate stability. 50 soil surface samples (0–20 cm depth) were collected from forest, agricultural, and pasture land uses. Soil physicochemical properties were determined through laboratory analyses. The study utilized Multiple Linear Regression (MLR) and machine learning models, including Random Forest (RF), Support Vector Machine (SVM), Artificial Neural Network (ANN), and an ensemble of the four single models. These models were trained using the repeated tenfold cross-validation method and evaluated based on root mean square error (RMSE), mean absolute error (MAE), and coefficient of determination (R2). The results demonstrated that the ANN model outperformed others in predicting soil erodibility (R2 = 0.98, MAE = 0.00341, RMSE = 0.0031. The SVM and RF models also performed well, with SVM achieving R2 = 0.93, MAE = 0.00541, RMSE = 0.0038, and RF achieving R2 = 0.87, MAE = 0.0037, RMSE = 0.00557 for soil erodibility prediction. The superior performance of ANN is attributed to its ability to model complex, non-linear interactions among variables influencing soil erodibility. Nonetheless, challenges such as data quality requirements and the risk of overfitting highlight the need for careful model calibration. The spatial prediction of soil erodibility across land uses revealed distinct patterns. Forest soils exhibited the lowest mean erodibility values (0.0313 t ha⁻1 h MJ⁻1 mm⁻1), reflecting their higher resistance to erosion due to better soil structure and organic matter content. In contrast, agricultural land uses recorded the highest mean erodibility values (0.0320 t ha⁻1 h MJ⁻1 mm⁻1), likely due to frequent tillage and reduced vegetation cover, which increase erosion susceptibility. Among soil types, Calcaric Cambisols were identified as the most erosion-prone, while Lithic Leptosols were the least susceptible, attributed to differences in soil texture, structure, and organic matter content. Finally, the basin was classified based on soil erodibility classes. The analysis showed that 81.18% of the basin (covering 546.6 km2) falls under the less erodible class, highlighting the basin’s overall resilience to erosion. In conclusion, the study demonstrates that machine learning-based models can accurately predict soil erodibility and its indices. The resulting maps provide a valuable baseline for land use planning, natural resource management, and decision-making processes. |
| format | Article |
| id | doaj-art-6261e2dc70814153aef644e332c1e2e1 |
| institution | OA Journals |
| issn | 2193-2697 |
| language | English |
| publishDate | 2025-06-01 |
| publisher | SpringerOpen |
| record_format | Article |
| series | Environmental Systems Research |
| spelling | doaj-art-6261e2dc70814153aef644e332c1e2e12025-08-20T02:37:36ZengSpringerOpenEnvironmental Systems Research2193-26972025-06-0114112910.1186/s40068-025-00402-wDigital mapping of soil erodibility factor in response to land use change using machine learning modelsWudu Abiye0Orhan Dengiz1Faculty of Agriculture, Department of Soil Science and Plant Nutrition, Ondokuz Mayıs UniversityFaculty of Agriculture, Department of Soil Science and Plant Nutrition, Ondokuz Mayıs UniversityAbstract Understanding the spatial variability of soil erodibility and its associated indices across different land uses is critical for sustainable land use planning and management. Traditional methods for measuring these variables are often time-consuming and costly. To address this, the study employed digital soil mapping (DSM) and machine learning (ML) models as efficient and cost-effective alternatives to predict soil erodibility and its indices, including clay ratio, critical level of organic matter, crust formation, dispersion ratio, and soil aggregate stability. 50 soil surface samples (0–20 cm depth) were collected from forest, agricultural, and pasture land uses. Soil physicochemical properties were determined through laboratory analyses. The study utilized Multiple Linear Regression (MLR) and machine learning models, including Random Forest (RF), Support Vector Machine (SVM), Artificial Neural Network (ANN), and an ensemble of the four single models. These models were trained using the repeated tenfold cross-validation method and evaluated based on root mean square error (RMSE), mean absolute error (MAE), and coefficient of determination (R2). The results demonstrated that the ANN model outperformed others in predicting soil erodibility (R2 = 0.98, MAE = 0.00341, RMSE = 0.0031. The SVM and RF models also performed well, with SVM achieving R2 = 0.93, MAE = 0.00541, RMSE = 0.0038, and RF achieving R2 = 0.87, MAE = 0.0037, RMSE = 0.00557 for soil erodibility prediction. The superior performance of ANN is attributed to its ability to model complex, non-linear interactions among variables influencing soil erodibility. Nonetheless, challenges such as data quality requirements and the risk of overfitting highlight the need for careful model calibration. The spatial prediction of soil erodibility across land uses revealed distinct patterns. Forest soils exhibited the lowest mean erodibility values (0.0313 t ha⁻1 h MJ⁻1 mm⁻1), reflecting their higher resistance to erosion due to better soil structure and organic matter content. In contrast, agricultural land uses recorded the highest mean erodibility values (0.0320 t ha⁻1 h MJ⁻1 mm⁻1), likely due to frequent tillage and reduced vegetation cover, which increase erosion susceptibility. Among soil types, Calcaric Cambisols were identified as the most erosion-prone, while Lithic Leptosols were the least susceptible, attributed to differences in soil texture, structure, and organic matter content. Finally, the basin was classified based on soil erodibility classes. The analysis showed that 81.18% of the basin (covering 546.6 km2) falls under the less erodible class, highlighting the basin’s overall resilience to erosion. In conclusion, the study demonstrates that machine learning-based models can accurately predict soil erodibility and its indices. The resulting maps provide a valuable baseline for land use planning, natural resource management, and decision-making processes.https://doi.org/10.1186/s40068-025-00402-wDigital soil mappingErosion susceptibilityLand use planningMachine learning and Soil erodibilityRemote sensing |
| spellingShingle | Wudu Abiye Orhan Dengiz Digital mapping of soil erodibility factor in response to land use change using machine learning models Environmental Systems Research Digital soil mapping Erosion susceptibility Land use planning Machine learning and Soil erodibility Remote sensing |
| title | Digital mapping of soil erodibility factor in response to land use change using machine learning models |
| title_full | Digital mapping of soil erodibility factor in response to land use change using machine learning models |
| title_fullStr | Digital mapping of soil erodibility factor in response to land use change using machine learning models |
| title_full_unstemmed | Digital mapping of soil erodibility factor in response to land use change using machine learning models |
| title_short | Digital mapping of soil erodibility factor in response to land use change using machine learning models |
| title_sort | digital mapping of soil erodibility factor in response to land use change using machine learning models |
| topic | Digital soil mapping Erosion susceptibility Land use planning Machine learning and Soil erodibility Remote sensing |
| url | https://doi.org/10.1186/s40068-025-00402-w |
| work_keys_str_mv | AT wuduabiye digitalmappingofsoilerodibilityfactorinresponsetolandusechangeusingmachinelearningmodels AT orhandengiz digitalmappingofsoilerodibilityfactorinresponsetolandusechangeusingmachinelearningmodels |