Journal of Watershed Management Research

Extended Abstract
Background: Soil erosion is one of the most serious environmental threats to agricultural lands and natural areas. Among the various forms of soil erosion, gully erosion is the most dominant process of land degradation and an advanced form of soil erosion. Therefore, identifying effective factors and predicting areas susceptible to gully erosion are the key factors for land management and planning.
Spatial forecasting of gully erosion using models that are based on land sensitivity to gully erosion and their output leads to the preparation of gully erosion risk maps in the form of gully erosion is the most suitable strategy for land management planning in watersheds to prevent gully erosion. If there is a zoning map of gully erosion in watersheds, socioeconomic damage could be controlled through appropriate measures and strategies, in addition to preventing high costs to control gully erosion. According to the available models and relationships, it can be concluded that most efforts in gully erosion modeling have been focused on the location of gullies, as well as predicting the possibility or non-possibility of gully occurrence. As such, only a part of the effective factors have been investigated due to the variability of factors affecting the gully. The present research aims to determine the most important factors affecting the gully erosion and prioritize the factors, with final zoning of the gully points and sensitivity to the gully in the Maroon watershed.
Methods: Maroon watershed is located in the southern and southwestern slopes of the Middle Zagros. The area of the watershed is 3837 km2, the maximum height from the sea level is 3420.3 m, and the minimum is 372.8 m. The average annual rainfall is 815 mm, and the long-term average annual temperature in the representative stations of the plains and the stations located in the highlands are 19 and 16 ºC, respectively. According to the Dumarten system, the climate type is Mediterranean to semi-arid. In summary, the present research was carried out as follows: 1) Selecting the area, preparation of the distribution map of gully occurrence (dependent variable), and its random division into two training or calibration (70 percent) and experimental or forecasting (30 percent) categories, 2) preparation of the maps of 23 effective factors (independent variables), 3) selecting the affecting factors using alignment test between the affecting factors and the occurrence of gullies, 4) running the model, 5) validation and evaluation of the model, and 6) preparation of a zoning map of susceptibility to gully erosion.
In this study, the location of the gullies created in the Maroon Dam watershed was first recorded using a global positioning system and Google Earth images, which were converted to points based on gully polygons. Then, recorded points were divided into two groups of training data (70%) and experimental data (30%) for the purpose of calibration and validation of the models, respectively. Next, 23 factors affecting the occurrence of gully erosion were identified based on scientific sources and watershed conditions, followed by preparing a map of all factors. The weights of these factors were determined based on the frequency ratio (FR). Correlation relationships among factors were examined using the collinearity test. In the next step, three models of maximum entropy, Dempster-Schaffer, and the weight of evidence were calibrated and validated using the weighted data of the affecting factors and the training and experimental spatial data of ditch distribution. A ditch erosion zoning map was prepared for each model, and then the optimal model was selected based on the validation data. The jackknife test and Operating Characteristic (ROC) curve were used to determine the thresholds of the factors affecting the occurrence of gully erosion and to evaluate the efficiency, respectively.
Results: The results showed that the ratio of gully frequency has the highest amount in the very high class based on the cell area index. The validation results in maximum entropy, Dempster-Shaffer, and the weight of evidence models based on the ROC index were 0.85, 0.81, and 0.78, respectively, indicating the higher accuracy of the maximum entropy model than the Dempster-Shaffer and the weight of evidence methods. Moreover, the weight of evidence method presented the lowest accuracy among the studied methods. In general, all three models had an acceptable percentage of ROC, which indicates the high performance of all three models. The results showed that land use and geological parameters had the highest contribution in the occurrence of gully erosion, with contributions of 24 and 18 percent, respectively.
Conclusion: Gully erosion is one of the most important natural hazards in the Maroon Dam watershed, which requires attention to soil conservation and gully erosion control issues. In this regard, one of the initial and required measures to determine sensitive areas and control and implement appropriate management measures is to prepare a gully erosion map. For this purpose, different methods and models, i.e., the weight of evidence, Dempster-Shaffer model, and entropy method, were used to prepare a gully erosion sensitivity map. In total, 23 parameters affecting gully erosion at the watershed level were identified and used as independent variables to predict gully erosion. The results of the factors affecting gully erosion showed that the lithological factor had the greatest impact on the ROC. Different parameters have different levels of participation in the occurrence of gully erosion, so that land use and geology parameters, with contributions of 24 and 18 percent, had the highest contribution to the occurrence of gully erosion. In general, the high accuracy of the gully erosion sensitivity map obtained from the maximum entropy model can be used to take appropriate management measures to prevent gully erosion in very sensitive areas.