Evaluating the Performance of Dez Dam Reservoir Operation in Flood Control Using Machine Learning Algorithms and Remote Sensing

Abstract Floods are considered one of the most destructive natural phenomena, causing extensive damage to the various sectors, including agriculture, infrastructure, housing, and socio-economic activities. The watersheds of the Dez and Karun rivers in Khuzestan Province require comprehensive management and precise analyses due to the frequent occurrence of the floods and their economic consequences. In this study, machine learning algorithms and Landsat 5, 7, and 8 satellite imagery were employed to identify flood-prone areas and evaluate the extent of flood-induced damages. Flood hazard zonation maps were generated with satisfactory accuracy and validated against field observation data, achieving an overall accuracy of 75%. By integrating hazard maps with land-use maps, vulnerable assets including agricultural lands, orchards, and rural settlements were identified. Also, regression models analyzed the relationship between river discharge and flood extent with 82.6% accuracy, revealing that outflow discharge is a key determinant of flood severity and spatial distribution. The findings demonstrate that combining remote sensing technologies with machine learning methods provides a robust tool for flood risk assessment and effective crisis management. Based on the results, can be proposed mitigation strategies for flood-prone areas, flood-related insurance policy frameworks, and optimized natural resource management. Finally, we strongly recommend avoiding development in high-risk flood zones and implementing watershed-scale risk reduction measures.