مدلسازی هدررفت خاک ناشی از فرسایش خندقی در مناطق فاقد آمار

نوع مقاله : مقاله پژوهشی

نویسندگان

• بهرام چوبین ¹
• امید رحمتی ²
• سید مسعود سلیمان‌پور ³
• صمد شادفر ⁴
• احمد نجفی ایگدیر ⁵

¹ بخش تحقیقات حفاظت خاک و آبخیزداری، مرکز تحقیقات و آموزش کشاورزی و منابع ‌طبیعی اصفهان، سازمان تحقیقات، آموزش و ترویج کشاورزی،

² بخش تحقیقات حفاظت خاک و آبخیزداری، مرکز تحقیقات و آموزش کشاورزی و منابع ‌طبیعی کردستان، سازمان تحقیقات، آموزش و ترویج کشاورزی،

³ بخش تحقیقات حفاظت خاک و آبخیزداری، مرکز تحقیقات و آموزش کشاورزی و منابع ‌طبیعی فارس، سازمان تحقیقات، آموزش و ترویج کشاورزی، شیراز،

⁴ پژوهشکده حفاظت خاک و آبخیزداری، سازمان تحقیقات، آموزش و ترویج کشاورزی، تهران، ایران

⁵ بخش تحقیقات حفاظت خاک و آبخیزداری، مرکز تحقیقات و آموزش کشاورزی و منابع ‌طبیعی آذربایجان غربی، سازمان تحقیقات، آموزش و ترویج کشاورزی،

چکیده

کلیدواژه‌ها

• حوزه آبخیز چوپانلو
• فرسایش خندقی
• مدلسازی
• هدررفت خاک
• یادگیری ماشین

موضوعات

• فرسایش و حفاظت خاک

عنوان مقاله [English]

Modeling soil loss due to gully erosion in the data-scarce regions

نویسندگان [English]

• Bahram Choubin ¹
• Omid Rahmati ²
• Seyed Masoud Soleimanpour ³
• Samad Shadfar ⁴
• Ahmad Najafi Eigdir ⁵

کلیدواژه‌ها [English]

• Choopanlu watershed
• Gully erosion
• Machine learning
• Modeling
• Soil loss

EXTENDED ABSTRACT

Introduction

Gully erosion is a critical and detrimental form of land degradation and soil loss affecting various regions worldwide. Its significant impact necessitates comprehensive assessments and monitoring efforts to mitigate its effects. However, traditional field measurement techniques for gully erosion can be time-consuming and costly, posing challenges in accurately quantifying soil loss from this process. To overcome these limitations, this study aimed to estimate soil loss resulting from gully erosion through the application of machine learning models. The research was conducted in the Choopanlu watershed, situated in West Azarbaijan province, Iran. By leveraging the capabilities of machine learning, the study sought to establish a more efficient and reliable method for assessing soil loss caused by gully erosion. This, in turn, would aid in formulating effective strategies to manage and control land degradation in the region.

Material and methods

The study commenced with field monitoring to identify gullies in the area to achieve this objective. Following this, digital layers of factors influencing gully expansion were prepared to facilitate gully clustering and selection. These factors included topographical characteristics (elevation, slope, aspect, surface curvature, and relative slope position index), vegetation, land use, soil, lithology, and hydroclimate indicators (distance from stream, drainage density, topographic wetness index, annual precipitation, and frequency of heavy rainfall events). Subsequently, the volume of soil lost due to gully erosion during the three-year period (2021-2023) was measured as the dependent variable for the selected gullies through field observations. In this study, three machine learning models were employed for modeling: Random Forest (RF), Support Vector Machine SVM), and Artificial Neural Network (ANN). The cross-validation approach was utilized for modeling process.

Results and discussion

Cochran's formula results indicated that among the 67 identified gullies in the field, a minimum sample size of 58 gullies was required. Following clustering, this number of selected gullies was chosen from the three identified clusters. The annual soil erosion caused by the selected gullies was estimated to be 172 tons in 2021, 196 tons in 2022, and 208 tons in 2023. The results of the multicollinearity analysis indicated that the elevation exhibited collinearity and thus was excluded from the modeling process. Based on the RMSE values and the coefficient of determination, it can be inferred that the RF model demonstrated the best performance, followed by the SVM model with moderate performance, and the ANN model exhibiting the poorest performance in modeling soil loss due to gully erosion. Consequently, the Random Forest model emerged as the most suitable model among the tested methods.

Conclusion

In conclusion, this study successfully applied machine learning models to estimate soil loss from gully erosion in the Choopanlu watershed. The analysis of 67 identified gullies, with a minimum sample size of 58, revealed an increasing trend in annual soil erosion over the three-year period. Multicollinearity analysis led to the exclusion of elevation, and subsequent model comparison indicated that the RF model outperformed the SVN and ANN models in predicting soil loss due to gully erosion. These findings demonstrate the potential of utilizing machine learning, particularly the RF model, for efficient and accurate assessment of soil loss from gully erosion in similar contexts. Further research could explore the application of these models in other regions and investigate additional factors influencing gully erosion processes.

Authors’ contribution:

Choubin B: Conceptualization, data curation, methodology, software, writing - original draft preparation.

Rahmati O: Conceptualization, methodology, writing—review and editing.

Soleimanpour SM: Conceptualization, supervision, validation.

Shadfar S: Conceptualization, project administration.

Najafi Eigdir A: Data curation, writing - original draft preparation.

Data Availability Statement:

The datasets are available upon a reasonable request to the corresponding author.

Ethical considerations:

The study was approved by the Ethics Committee of the Agricultural Research, Education and Extension Organization (AREEO). The authors avoided from data fabrication and falsification.

Funding:

The study was funded by the Agricultural Research, Education and Extension Organization (AREEO), Iran, and Grant No. 0-50-29-024-990514.

Conflict of interest:

The authors declare no conflict of interest.