Extended Abstract
Introduction: Soil erosion and water crisis are among the most significant threats that endanger the security of water and soil in the country. In order to address these problems, measures are being taken to soil and water conservation. The most important stage in implementing these plans is the structural measures of watershed management, which involves identifying the correct locations for implementing these plans. Proper site selection of check dams has a significant impact on reducing the cost of watershed activities and increasing their effectiveness. Traditional methods of locating check dams based on more information layers and the need for their integration and analysis are difficult and may result in errors, while modern and highly efficient data mining methods based on the quality and quantity of data and machine learning have been introduced. In this study, the maximum entropy (MaxEnt) model was used to locate check dams in the Dehdar Taleghan watershed.
 
Materials and methods: The Dehdar Taleghan watershed is located in the north of Alborz province, with variable elevation from the highest point at 4050 meters above sea level to the outlet of the watershed at 2248 meters. There are two villages located within the basin, which covers an area of 4780 hectares. The locations of existing and proposed check dams were extracted based on the database of implemented check dams in the Natural Resources and Watershed Management head office of Alborz province and the mechanical check dams’ manuals of detailed-executive studies. In this study, 14 factors that affect the placement of watershed check dams including DEM, slope, distance from stream, roads, and faults, density of stream, roads, and faults, lithology, land use, stream power, stream rank, flow accumulation, and precipitation were used to determine suitable locations. The multicollinearity was checked using the variance inflation factor (VIF) and tolerance index. After confirming the absence of multicollinearity between variables, the existing check dam’s points were randomly divided into training data (70%) and validation data (30%). The importance of each variable in explaining the model was determined using the MaxEnt model and the Jackknife plot, which was performed using MaxEnt software. In this study, the performance of the model in the training and validation stages was evaluated using the receiver operating characteristic (ROC) curve and its area under the curve (AUC).
 
Results and Discussion: The results showed that there is no linear relationship between the factors, and therefore all factors were used in the modeling process. The results of the Jackknife plot showed that distance from streams, slope, flow accumulation, stream order, elevation, mean precipitation, and lithology were the most important factors affecting the visibility of check dams, and they had a significant impact on predicting areas with potential for check dam construction. The accuracy of the model prediction was excellent in both the training (0.959) and validation (0.961) stages. Field surveys confirmed that the model accurately identified critical streams in terms of flooding and sedimentation, with a total of 30.3 kilometers of critical and supercritical stream identified. During various field visits, 11 check dams were identified in the studied streams. It is worth noting that the final map (critical areas) and the visibility of check dams, as assessed by consulting companies in detailed executive studies, had a 92% level of agreement, demonstrating the high accuracy of machine learning models for predicting check dam’s visibility.
 
Conclusions: In this study, a map of suitable areas for check dam’s construction in the Dehdar Taleghan watershed was prepared using the MaxEnt model, taking into account influential environmental variables. The ROC curve showed that the model's accuracy in estimating areas with potential for check dam’s construction was excellent in both the training and validation stages, indicating excellent model performance. Based on the results, it can be said that the MaxEnt model has a high ability to determine areas with potential for check dam’s construction. Due to its speed and high accuracy, the model is recommended for use in similar studies, especially in developing countries facing a shortage of facilities and financial resources. Combining geographic information systems with modern machine learning models to determine areas with potential for check dam’s construction, especially in developing countries like Iranis recommended.