Evaluation of the efficiency of three data mining models in zoning areas prone to gully erosion (Case study: Upper Watershed of Boustan Dam)

Introduction

Gully erosion is one of the most destructive forms of water erosion, leading to significant soil loss in dry and semi-arid regions and considered as one of the most important environmental problems. Anyway, explaining and developing a suitable model for predicting the temporal and spatial formation, growth, and expansion of gully erosion, considering the importance of this type of erosion, has been of great interest to many soil conservation experts and researchers. In recent years, remote sensing and data mining techniques have been increasingly used to identify and map areas susceptible to gully erosion, providing valuable information for land managers and policymakers. Regarding the validation and training capability of models for performing reasonable and accurate modeling, usually, the more advanced the models are, the more learning ability they will have, and the relationship between different variables will be better identified. Producing a sensitivity map is a fundamental tool for land use planning with the aim of reducing land degradation, which is essential for preserving natural resources and evaluating the relationship between gully occurrence and influential factors

Materials and Methods

The aim of this research is to investigate the sensitivity of the upper basin of the Boustan Dam to gully erosion using object-based techniques and data mining algorithms. To this end, field visits were conducted to select 81 gullies in the area. Some of factors that are examined in the present research are mentioned below: Slope, aspect, slope length index (LS), elevation, planCurvature, distance from the river, drainage density, topographic wetness index (TWI), Height Above the Nearest Drainage (HAND), average annual rainfall, distance from the road, distance from the fault, land use, geomorphon, soil texture, B7, B5, B3, Normalized Difference Vegetation Index (NDVI), The Normalized Difference Built-up Index (NDBI), The Normalized Difference Water Index (NDWI), and geology. QuickBird satellite images from 2021 and Orfeo software were used for monitoring and identifying gullies in the area by segmenting the desired image. First, using collinear analysis on 23 effective erosion occurrence indices the distance from the fault, DEM, NDWI, NDBI, B3, B5 and B7 were removed due to collinearity above 5. After the linear operation, all the indices were integrated with the segmentation map obtained from the Orfeo environment. Finelly with three models, Random Forest, Maximum Entropy, and Support Vector Machine and The selected indices were modeled in Python (Colab).

Results and Discussion

The results obtained from utilizing the object-oriented method in the Orfeo software further demonstrated its effectiveness in accurately identifying gullies. With an impressive accuracy rate of 91.3%, this method has proven to be highly reliable in generating machine learning maps with high precision. Also, research findings indicate that the influential factors contributing to gully erosion in the study area were identified as the rainfall index, distance from the river, hand index, distance from the road, and valley index. Torrential rain emerged as a significant driver of gully erosion. also, the distance from the river played a crucial role due to the concentration of surface and subsurface flows towards waterways. The HAND index played a prominent role in modelling the sensitivity of the study area compared to other sub-indices derived from digital elevation models (DEM). While other sub-indices typically encompass specific topographic or hydrological information. The HAND index exhibited promising applications in assessing natural hazards. Locations close to roads were found to be more vulnerable to water erosion, and valleys were identified as especially susceptible to gully erosion due to their conducive conditions for rapid water flow and erosion. This observation is substantiated by extensive field studies. Furthermore, the zoning results using these indices showed that in the Random Forest model, 544.23 hectares of the area are at high or very high risk of erosion, which is better than the Maximum Entropy and Support Vector Machine models in predicting erosion-prone areas. Finally, the ROC curve was used to validate the model, and AUC values of 0.95 and 0.94 were obtained in the Random Forest model during the training and validation stages, indicating the high accuracy of this model in predicting areas with high sensitivity to gully erosion.

Conclusion

This study successfully employed object-based image analysis algorithms and data mining techniques to create a sensitivity map of the region. The object-based method proved effective in identifying the gully of the region with the mean shift algorithm, and the Random Forest algorithm demonstrated the best performance in predicting gully erosion-prone areas. Important factors such as rainfall, distance from the river, hand index, and distance from the road and valley were identified as key factors contributing to gully erosion. The findings of this study provide valuable information for basin resource management and preservation. Implementing the recommendations from this study can help mitigate the effects of gully erosion in the future and help ensure the sustainability of the Boustan dam and its surrounding ecosystem.