نوع مقاله : پژوهشی
نویسندگان
1 دانشیار بخش تحقیقات حفاظت خاک و آبخیزداری، مرکز تحقیقات و آموزش کشاورزی و منابع طبیعی استان فارس، سازمان تحقیقات، آموزش و ترویج کشاورزی، شیراز، ایران
2 استادیار بخش تحقیقات حفاظت خاک و آبخیزداری، مرکز تحقیقات و آموزش کشاورزی و منابع طبیعی استان کردستان، سازمان تحقیقات، آموزش و ترویج کشاورزی، سنندج، ایران
3 دانشیار پژوهشکده حفاظت خاک و آبخیزداری، سازمان تحقیقات، آموزش و ترویج کشاورزی، تهران ایران
4 کارشناس ارشد بخش تحقیقات حفاظت خاک و آبخیزداری، مرکز تحقیقات و آموزش کشاورزی و منابع طبیعی استان فارس، سازمان تحقیقات، آموزش و ترویج کشاورزی، شیراز، ایران
چکیده
کلیدواژه‌ها
موضوعات
عنوان مقاله [English]
نویسندگان [English]
Introduction and Goal
Water erosion is one of the most important factors in land degradation. Among the different types of water erosion, gully erosion is a very obvious and prominent type of soil erosion and is one of the most important challenges threatening food production, human health, and the ecosystem. Since the amount of soil loss due to gully erosion is directly related to environmental factors, it is possible to model the amount of soil loss due to gullies based on environmental conditions. On the other hand, field measurement of the amount of soil loss due to gully erosion is very time-consuming and costly, and direct measurement of gully erosion on a large scale is not possible. In this regard, this study aimed to evaluate the effectiveness of support vector machine (SVM) and artificial neural network (ANN) models in modeling soil loss due to gully erosion in the Mahurmilati watershed located in the southwest of Fars province.
Materials and Methods
In field visits, the geographical location of all gullies located in this watershed was recorded using a GPS device, and after matching with Google Earth satellite images, these points were transferred to the GIS and a gully distribution map was drawn. Field measurements of the dimensional parameters of 70 gullies were carried out over four years (2021-2024), including: gully length, upper width, lower width, and gully depth, and the volume and weight of soil lost due to gully erosion were also calculated. For the modeling process, 15 environmental factors were selected as independent variables (predictor variables). In the modeling process, environmental factors were considered as independent variables, and the rate of soil loss in gullies was considered as a dependent variable. The gullies were randomly divided into two groups: training (70%) and validation (30%). Modeling was performed using two models: support vector machine and artificial neural network, with a cross-validation approach. The accuracy of the models was evaluated using quantitative criteria such as root mean square error (RMSE), determination coefficient (R2), RSR error index, and goodness of fit (d).
Results and Discussion
The results of the assessment of the prediction accuracy of the models showed that the smallest error size in terms of the root mean square error (RMSE) criterion was related to the support vector machine model, followed by the artificial neural network model in second place. Based on the evaluation criterion of the coefficient of determination (R2), the support vector machine model (R2=0.41-0.59) was in first place and the artificial neural network model (R2=0.21-0.34) was in second place. In terms of the evaluation criterion of the RSR error index, the artificial neural network model was in first place and the support vector machine model was in second place. The highest degree of agreement between observational and forecast data in terms of agreement evaluation index (d) was for the support vector machine model (0.81), and the artificial neural network model with an agreement index size of 0.63 was in second place in performance. Therefore, based on the results of this study, the performance of the support vector machine model was better in terms of evaluation criteria of root mean square error (RMSE), explanation index, and agreement index, compared to the artificial neural network model, and it was introduced as the superior model for predicting the amount of soil loss due to gully erosion in the Mahurmilati watershed of Fars province.
Conclusion and Suggestions
In evaluating the performance of a model for simulating a phenomenon, several factors play a role in this issue that must be considered. Therefore, in this study, 15 environmental variables were used to predict soil loss due to gully erosion. In addition, using variables provided the model with very important information in terms of spatial and temporal changes and watershed characteristics. Also, in this study, the monitoring period was not long-term; but the performance of the two models used for predicting soil loss due to gully erosion was appropriate. One of the important advantages of this research was that the two models used were selected from among the most efficient artificial intelligence models, and by considering time-varying characteristics, future forecasting conditions were provided, and therefore, there is no need for repeated field measurements in the coming years. In fact, the model can predict the amount of soil loss caused by gullies in subsequent years based on the learnings gained during the training phase and by considering new rainfall and vegetation cover information. This finding was the most important distinguishing feature of this research and indicates that modeling can save time and money and provide valuable services to the country's water and soil conservation management. In this regard, it is suggested that models based on artificial intelligence and machine learning structures be used in future research.
کلیدواژه‌ها [English]
)