Extended Abstract
Introduction: Flood events are among the most significant natural disasters, causing substantial damage each year in Iran. One effective approach in watersheds for mitigating the consequences of flooding is the implementation of watershed management operations. Moreover, evaluating the effects of each operation is crucial. In natural disaster management, particularly in the case of floods, timing is the most critical factor. This underscores the need for a rapid response model to facilitate precautionary actions and early warnings. Numerous studies have been conducted to measure and classify the effects of floods from various perspectives. Generally, the damages are evaluated both directly and indirectly in flood impact assessments. Advances in various fields of artificial intelligence, especially in water resources, have made these technologies viable options for modeling hydrological and hydraulic processes. Consequently, the present research employs machine learning methods to predict the impact of water management measures on floods in the Behesht Abad watershed.
Materials and methods: The main focus of this research is to simulate the effects of watershed operations using machine learning methods in the Behesht Abad watershed. Discharge and rainfall data from 1999 to 2020 were used. Effective factors in flood occurrence, including canopy cover, soil, and topography, were analyzed in ArcGIS using satellite imagery, GEE, and field surveys. Furthermore, the effects of watershed management actions, particularly biological operations, were simulated using Support Vector Machine (SVM) and Random Forest (RF) models. In summary, discharge flow predictions were made based on a dataset comprising 7,850 records, with 70% (5,495 records) used for model training and 30% (2,355 records) for testing. Biological measures such as mounding, sowing, seeding, and seedling were simulated to assess their impact on flow rates. By implementing these predicted biological plans in the studied watershed, changes in vegetation and land use parameters were modeled. The predicted layers were then used to update numerical values related to vegetation cover, including the NDVI index and land use, which were recalculated and integrated into the modeling process.
Results and Discussion: In this watershed, the SVM simulation indicated that the highest discharge flow occurred in 2003, reaching approximately 500 m³/s. According to the RF simulation, this value increased to 520 m³/s in 2016. The results demonstrate that biological operations reduce discharge flow and have the least impact on peak discharge flow. A comparison between SVM and RF revealed that SVM performed better in discharge flow prediction. Based on the results, the R² values for the training and testing phases were 0.96 and 0.89, respectively, while the NTS values for the training and testing phases were 0.95 and 0.86, respectively.
Conclusion: The role of biological watershed management measures in reducing surface runoff and their effect on flood variables, particularly through their influence on the watershed's concentration time and curve number, is undeniable. It is also essential to examine the impact of these measures on the watershed's hydrological processes. The results indicate that applying machine learning models is a cost- and time-effective approach for discharge flow estimation, flood management, and flood control in planning and projects. By leveraging these methods, communities and governments can enhance flood preparedness, improve management strategies, and ultimately reduce the impact of flooding events on human lives and infrastructure.