Document Type : Research Paper
Authors
1 Ph.D. Student, Department of Range and Watershed Management, Faculty of Natural Resources, Lorestan University, Khorramabad, Lorestan Province, Iran
2 Associate Professor, Department of Range and Watershed Management Engineering, Faculty of Natural Resources, Lorestan University, Khorramabad, Lorestan Province, Iran
3 Assistant Professor, Soil Conservation and Watershed Management Research Department, Agricultural and Natural Resources Research and Education Center, Kurdistan Agricultural Research, Education and Extension Organization
Abstract
This research was conducted with the aim of mapping the susceptibility to gully erosion using artificial intelligence models in the Alashtar watershed, Lorestan Province. The study area, covering 797.64 km2, is part of the Karkheh watershed. In this study, 12 effective factors were used as input data, including slope, aspect, precipitation, distance from road, distance from river, distance from fault, soil type, land use, geological formation, Topographic Wetness Index (TWI), Topographic Position Index (TPI), and Normalized Difference Vegetation Index (NDVI). Out of a total of 151 observation points (89 gully points and 62 non-gully points), 70% were used for training stage and 30% for testing stage. The performance of three AI models—Multilayer Perceptron Artificial Neural Network (MLP), Maximum Entropy (MaxEnt), and Flexible Discriminant Analysis (FDA)—was evaluated using the ROC curve and the Area Under the Curve (AUC) index. The results showed that the MLP model, with AUC values of 0.98 in the training phase and 0.92 in the validation phase, had the best performance in predicting gully erosion susceptibility. This was followed by the FDA (AUC = 0.87) and MaxEnt (AUC = 0.5) models, respectively. Analysis of the influencing factors revealed that most gullies were located in precipitation classes of 600-700 mm, distances greater than 300 meters from faults, roads, and rivers, slope classes of 0-5% and 5-15%, northern aspects, dry farming land use, and geological formations of old alluvium and marls. Furthermore, a direct relationship was observed between the TWI index and gully occurrence, while an inverse relationship was found for the NDVI index. Finally, the gully erosion map was prepared using the MLP model. Result shown that artificial neural network, is an effective tool for identifying areas susceptible to gully erosion and helps to planning and management to control this phenomenon in similar regions.
Keywords
)