Application of Support Vector Machine (SVM) and Boosted Regression Tree (BRT) to Model the Sensitivity of Gully Erosion in the Watershed of Shore River Moher City

Document Type : Full length article

Authors

1 Department of Geomorphology, Faculty of Social Sciences, University of Mohaghegh Ardabili, Ardebil, Iran

2 Department of Geomorphology, Faculty of Economics, Management and Social Sciences, Shiraz University, Shiraz, Iran

Abstract

ABSTRACT
The aim of this study is to develop sensitive gully erosion models by implementing a machine learning algorithm (Support Vector Machine and Boosted Regression Tree) in the Moher basin. First, gully areas are identified, and then 13 variables predisposing to gully erosion (Slope, Slope Direction, Topographic Wetness Index, Streem Power Index, Terrain Ruggedness Index, Distance from Waterway, Drainage Density, Distance from Road, Land use, NDVI, Avera annual Rainfall, Geology, and Soil Texture) were selected. The variance inflation coefficient was used to evaluate multicollinearity between variables. Finally, a gully erosion sensitivity map was prepared in the environment (R). Also, the effect of physical and chemical characteristics of soil on gully erosion was investigated using Multivariate Regression. Regarding the importance of variables, Geology has the most significant effect on gully erosion in the SVM model, Land use, and the BRT model. The predicted sensitivity map was validated with the help of the receiver operating characteristic (ROC) curve. The results showed that the area under the curve (AUC) in the Support Vector Machine and Boosted Regression Tree models were calculated as 0.92 and 0.94, respectively, which led to accurate prediction. Also, the results showed that the sand variable (9.299), sodium absorption ratio (7.967), and TNV (6.185) have the most significant effect on gully erosion
Extended abstract
Introduction
Currently, many countries are facing severe land degradation and soil erosion. Soil erosion caused by water is a major environmental concern, affecting approximately one billion hectares of land worldwide. Gully erosion is one of the most essential factors in land degradation in dry and semi-arid regions, causing significant soil losses and the transfer of sediments to low-altitude areas. Many researchers investigated gully erosion sensitivity using remote sensing (RS) and geographic information systems (GIS) techniques. Traditional data mining methods cannot establish the relationship between geoenvironmental factors and gully erosion processes. Therefore, machine learning models are highly efficient for evaluating areas sensitive to gully erosion. Support vector machines (SVM) and boosted regression trees (BRT) are machine-learning techniques to model gully erosion. In the lower reaches of a watershed or near agricultural lands, gully erosion caused sediment production and desertification of the region. Using SVM and BRT models, it is possible to prepare a map of gully erosion sensitivity and use zoning to reduce potential damage and manage crises.
 
Materials and methods
The study area covers approximately 101350 hectares and is located in southern Iran, with an elevation range of 387 to 1672 meters above sea level. The basin is situated between longitude 52°24'52" to 52°59'52"E and latitude 27°22'27" to 41°49'27"N. Initially, a gully erosion points map was created based on 200 gully and non-gully points, identified through field visits and data obtained from the Natural Resources Department of Fars province. Random non-gully points were also selected throughout the basin using a Geographic Information System (GIS). Furthermore, 13 variables were chosen for modeling based on the approach utilized by previous researchers. These variables encompassed slope, slope aspect, drainage density, distance from the stream, land use, geology, soil texture, stream power index (SPI), topographic wetness index (TWI), topographic roughness index (TRI), average precipitation, distance from the road, and vegetation cover. Multicollinearity analysis employing the variance inflation factor (VIF) was conducted to determine the linear correlation between variables. Subsequently, machine learning algorithms, including Support Vector Machine (SVM) and Boosted Regression Trees (BRT), were employed for modeling. Finally, the accuracy of the models was evaluated using the ROC curve. Moreover, the effect of physical and chemical characteristics of soil on gully erosion was investigated using multivariate regression.
 
Results and discussion
Investigating the collinearity between the selected variables is crucial in creating gully erosion sensitivity maps. Among the 13 selected variables, no significant collinearity was observed in the SVM and BRT models. The areas with low sensitivity to gully erosion are mainly concentrated in the region's northern, northwestern, and southwestern parts. The moderately susceptible areas are in the middle of the basin, while the highly susceptible areas are in the southern and southeastern parts. The SVM model had an AUC value of 92.0% for the training dataset and 93.0% for the test dataset. The BRT model had an AUC value of 96.0% for the training dataset and 94.0% for the test dataset. Both models demonstrated high accuracy.
Soil erosion can cause severe environmental and human damage through natural or human-induced gully formation. In dry and semi-dry areas, the formation and development of gullies are mainly problematic and lead to a decrease in soil quantity and quality and a reduction in agricultural productivity. Additionally, the sediment resulting from soil erosion can accumulate at the basin outlet and create problems for the environment and humans. Therefore, it is essential to investigate the formation and development of gullies and determine their importance in environmental modeling and management.
In the Mohr watershed, gully erosion has become a problem, and identifying vulnerable areas using SVM and BRT models is useful for sustainable land management. Topography, slope, surface runoff, drainage density, soil erosion, soil moisture, vegetation cover, and geomorphological processes affect the formation and expansion of gullies. Most gullies are formed and expanded in the north, northeast, east, and northwest directions. Gully erosion has a positive relationship with drainage density, and gullies are formed in areas with low slopes and high moisture. Soil texture and land use are also important for controlling gully erosion. The type of relationship between erosion factors and surface cover of the region is determined based on land use. The presence of vegetation cover in this area is highly important and effective in controlling erosion. The absence of vegetation cover creates runoff and causes erosion and sedimentation on a large scale. Additionally, the impact of road networks is crucial in soil erosion. The results of geological variables show that Quaternary sediments play an important role in the formation of gully erosion. Quaternary sediments are effective in gully erosion formation due to high levels of gypsum and salt, fine-grained sediment deposits, loess instability, soil structure weakening, and lack of organic matter. Soil texture is also used as a controlling mechanism in gully erosion. The results of physical and chemical characteristics showed that the sand variable (9.299), sodium absorption ratio (7.967), and TNV (6.185) have the most significant effect on gully erosion.
 
Conclusion
A study was conducted to prepare a gully erosion sensitivity prediction map using machine learning algorithms (SVM and BRT) in the Shoor River watershed (Mohr county). This research investigated 13 gully erosion-sensitive variables and their performance in gully development. The entire dataset was randomly divided into 70/30 for training and validation. The area under the curve (AUC) values for SVM and BRT were 0.92 and 0.94, respectively. In both models, more than half of the watershed was classified as low to very low sensitivity (SVM: 89.22%, BRT: 92.84%), while 2.62% and 3.58% of the area were classified as highly sensitive in SVM and BRT models, respectively. Also, the results of the physical and chemical characteristics of the soil showed that sand (9.299), sodium absorption ratio (7.967), and TNV (6.185) have the most significant effect on the expansion of gully erosion in the study area. Finally, the results showed that machine learning models can determine gully erosion boundaries, and the resulting susceptibility maps can be used as essential tools for protecting and sustainably managing gully erosion-prone areas in the study area.
 
Funding
There is no funding support.
 
Authors’ Contribution
All of the authors approved the content of the manuscript and agreed on all aspects of the work.
 
Conflict of Interest
Authors declared no conflict of interest.
 
Acknowledgments
We are grateful to all the scientific consultants of this paper.

Keywords

Main Subjects

References

  1. Amiri, M., Pourqasmi, H., Ghanbarian, Gh-A., & Afzali, S. F. (2019). Spatial modeling of gully erosion using different scenarios and event weighting algorithm. Watershed Engineering and Management, 11(4), 1016-1032. [In Persian]
  2. Angileri, S.E., Conoscenti, Ch., Hochschild, V., Märker, M., Rotigliano, E., & Agnesi, V. (2016). Water erosion susceptibility mapping by applying stochastic gradient treeboost to the Imera Meridionale river basin (Sicily,Italy), Geomorphology, 262, 61-76. https://doi.org/10.1016/j.geo morph.2016.03.018
  3. Arabameri, A., Pradhan, B., Rezaei, K., & Conoscenti, C. (2019). Gully erosion susceptibility mapping using GIS-based multi-criteria decision analysis techniques. Catena, 180, 282-297. https://doi.org/10.1016/j.catena.2019.04.032
  4. Avand, M., S. Janizadeh, S.A. Naghibi, H.R. Pourghasemi, S. Khosrobeigi Bozchaloei, and Th Blaschke, (2019). A comparative assessment of random forest and k-nearest neighbor classifiers for gully erosion susceptibility mapping. Water, 11, 2076. https://doi.org/10.3390/w11102076
  5. Azedou, A., Lahssini, S., Khattabi, A., Meliho, M., & Rifai, N. (2021). A methodologicalcomparison of three models for gully erosion susceptibility mapping in the rural municipality of El Faid (Morocco). Sustainability, 13, 682. https://doi.org/10.3390/su13020682
  6. Bui, D. T., Pradhan, B., Lofman, O., Revhaug, I., & Dick, O. B. (2012). Landslide susceptibility assessment in the Hoa Binh province of Vietnam: a comparison of the Levenberg–Marquardt and Bayesian regularized neural networks. Geomorphology, 171, 12-29. https ://doi.org/10.1016/j.geomorph.2012.04.023
  7. Davoudi Rad, A. A., & Mohammadi, M. (2022). Checking the accuracy of watershed erosion sensitivity map using SVM and MARS methods in Shazand watershed. Iran Watershed Science and Engineering, 16(59), 12-22. [In Persian].
  8. Grohmann, C. H., & Riccomini, C. (2009). Comparison of roving-window and search-window techniques for characterising landscape morphometry. Computers & Geosciences, 35(10), 2164-2169. https://doi.org/10.1016/j.cageo.2008.12.014
  9. Hosseinalizadeh, M., Kariminejad, N., Rahmati, O., Keesstra, S., Alinejad, M., & Behbahani, A.  M. (2019). How can statistical and artificial intelligence approaches predict piping erosion susceptibility?. Science of the Total Environment, 646, 1554-1566. https://doi. org/10.1016/j.scitotenv.2018.07.396
  10. Kirkby, M., & Bracken, L. (2009). Gully processes and gully dynamics. Earth Surface Processes and Landforms: The Journal of the British Geomorphological Research Group, 34(14), 1841-1851. https://doi.org/10.1002/esp.1866
  11. Kutner, M. H., Nachtsheim, C. J., Neter, J., & Li, W. (2004). Applied linear statistical models. McGraw-hill.
  12. LeRoux, J.J., & Sumner, P.D. (2012). Factors controlling gully development: comparing continuous and discontinuous gullies. Land Degradation & Development, 23, 440-49. https doi.org/10.1002/ldr.1083
  13. Lucà, F., Conforti, M., & Robustelli, G. (2011). Comparison of GIS-based gullying susceptibility mapping using bivariate and multivariate statistics: Northern Calabria, South Italy. Geomorphology, 134(3-4), 297-308. https://doi.org/10.1016/j.geomorph.2011.07.006
  14. Mohammady, M., & Davudirad, A. (2023). Gully Erosion Susceptibility Assessment Using Different Machine Learning Algorithms: A Case Study of Shazand Watershed in Iran. Environmental Modeling & Assessment, 1-13. https://doi.org/10.1007/s10666-023-09910-4
  15. Morgan, R. P. C. (2009). Soil erosion and conservation: John Wiley & Sons.
  16. Naghibi, S. A., Pourghasemi, H. R., & Dixon, B. (2016). GIS-based groundwater potential mapping using boosted regression tree, classification and regression tree, and random forest machine learning models in Iran. Environmental monitoring and assessment, 188, 1-27. https://doi.org/10.1007/s10661-015-5049-6
  17. Pourghasemi, H. R., Gayen, A., Haque, S. M., & Bai, S. (2020). Gully erosion susceptibility assessment through the SVM machine learning algorithm (SVM-MLA). Gully Erosion Studies from India and Surrounding Regions, 415-425. https://doi.org/10.1007/978-3-030-23243-6_28
  18. Pradhan, B. (2013). A comparative study on the predictive ability of the decision tree, support vector machine and neuro-fuzzy models in landslide susceptibility mapping using GIS. Computers & Geosciences, 51, 350-365. https://doi.org/10.1016/j.cageo.2012.08.023
  19. Rahmati, O., Tahmasebipour, N., Haghizadeh, A., Pourghasemi, H. R., & Feizizadeh, B. (2017). Evaluation of different machine learning models for predicting and mapping the susceptibility of gully erosion. Geomorphology, 298, 118-137. https://doi.org/10.1016/j.geomorph.2017.09.006
  20. Safari, A. Karam, A., Shadfar, S., & Ahmadi, M. (2019). The influence of soil characteristics on the morphology and spread of gully erosion (case study: Mehran Lamard River Basin, Fars). Quantitative Geomorphology Research, 8 (1), 130-146. [In Persian].
  21. Saha, A., S.Ch. Pal, A. Arabameri, I. Chowdhuri, F. Rezaie, R. Chakrabortty, P. Roy, & M. Shit, (2021). Optimization modelling to establish false measures implemented with ex-situ plant species to control gully erosion in a monsoon-dominated region with novel in-situ measurements, Journal of Environmental Management, 287, 112284. https://doi.org/10.1016/j.jenvman.2021.112284
  22. Saha, S., Roy, J., Arabameri, A., Blaschke, T., & Tien Bui, D. (2020). Machine learning-based gully erosion susceptibility mapping: A case study of Eastern India. Sensors, 20(5), 1313. https://doi.org/10.3390/s20051313
  23. Shahbazi, A., Vakili Tejareh, F., Alwandi, E., Bayat, A., & Asadi Nelivan, O. (2021). Evaluation of artificial neural network models and maximum entropy in the zoning of hydrological erosion sensitivity of Golestan dam watershed. Iranian Watershed Science and Engineering Journal, 12-23, 15. [In Persian].
  24. Wang, F., Sahana, M., Pahlevanzadeh, B., Pal, S. C., Shit, P. K., Piran, M. J., . . . Mosavi, A. (2021). Applying different resampling strategies in machine learning models to predict head-cut gully erosion susceptibility. Alexandria Engineering Journal, 60(6), 5813-5829. https://doi.org/10.1016/j.aej.2021.04.026
  25. Wang, G., Chen, X., & Chen, W. (2020). Spatial prediction of landslide susceptibility based on GIS and discriminant functions. ISPRS International Journal of Geo-Information, 9(3), 144. https://doi.org/10.3390/ijgi9030144
  26. Zucca, C., A. Canu, and R. D. Peruta, (2006). Effects of land use and landscape on spatial distribution and morphological features of gullies in an agropastoral area in Sardinia (Italy), Catena, 68, 87-95. https://doi.org/10.1016/j.catena.2006.03.01
Physical Geography Research
Volume 55, Issue 4
December 2024
Pages 83-101

Files

History

  • Receive Date: 02 August 2023
  • Revise Date: 27 October 2023
  • Accept Date: 29 November 2023

Share

How to cite

Statistics