Floods are one of the most prevalent and devastating natural disasters, and their impact on various societies worldwide has always been significant. While preventing floods is nearly impossible, obtaining sufficient information about flood situations and timely detection can mitigate potential damages. With advancements in remote sensing satellite technology and progress in deep learning and machine learning, the capability to flood detection with higher precision and efficiency has been achieved. In this regard, this study aims to develop ensemble-optimized models for flood detection utilizing Sentinel-1 satellite data and compare the performance of these models. The first step involved feature extraction from the images using the pre-trained deep neural network model, VGG-16. Subsequently, machine learning algorithms including Random Forest (RF) and Gradient Boosting (GB) were employed as classifiers, and Genetic Algorithm (GA) and Harris Hawks Optimization (HHO) were utilized for hyperparameter optimization of these classifiers. The prediction accuracy of the four ensemble flood detection models RF-GA, RF-HHO, GB-GA, and GB-HHO were 90.97%, 90.37%, 91.45%, and 91.61%, respectively. Model GB-HHO exhibited the lowest error rate and the highest prediction accuracy. The findings of this study indicate that all four models offer acceptable performance and accuracy rates. Moreover, Gradient Boosting-based classifiers, GB-HHO and GB-GA, exhibit superior prediction accuracy compared to Random Forest and demand significantly lower computational resources for model training processes.