Floods rank among the most destructive natural disasters worldwide, with their frequency and severity escalating due to climate change, rapid urbanization, and land-use changes. Accurate and timely flood detection is critical for effective disaster response and mitigation, especially in vulnerable arid and semi-arid regions frequently affected by flash floods. This study develops a multi-stage fusion framework for flood mapping by integrating Sentinel-1 SAR and Sentinel-2 multispectral data within the Google Earth Engine (GEE) cloud platform. The proposed approach was applied to the extreme flash flood event that struck Konarak County, southeastern Iran, in early January 2022. In the first stage, five supervised machine learning classifiers, including Random Forest (RF), Gradient Boosting Decision Tree (GBDT), Classification and Regression Tree (CART), Support Vector Machine (SVM), and K-Nearest Neighbors (KNN), were independently trained using the fused SAR and optical dataset. In the next stage, Initial flood maps generated by the classifiers were fused at the decision level using the majority voting approach to produce the final flood map. The results show that the decision-level fusion achieved superior performance in terms of statistical metrics. It yielded an overall accuracy of 97.50%, F1-score of 97.42%, and a Kappa coefficient of 0.939, outperforming all individual classifiers considered in this study. In contrast, the RF classifier, which demonstrated the best performance among the classifiers, achieved an overall accuracy of 96.67%, an F1-score of 96.61%, and a Kappa coefficient of 0.933. The proposed method jointly exploits SAR and optical features and combines multiple classifier outputs to improve the accuracy and reliability of flood detection.