Abstract: Current methods for extracting Ulva prolifera information based on remote sensing imagery face significant challenges, such as low accuracy and universality of fixed threshold methods, and the need for massive training samples in supervised classification methods (e.g., support vector machine (SVM) and random forest (RF)). To address these challenges, this study proposed an automatic IForest-RF method for extracting Ulva prolifera information, which combines the isolation forest (IForest) for unsupervised anomaly detection and the RF for supervised classification. The IForest-RF method was tested using the Sentinel-2 L2A satellite imagery of two regions with significantly different Ulva prolifera distributions in the Yellow Sea on June 26, 2024. Its extraction accuracy was compared with those of the other five methods: IForest combined with Otsu's method (IForest-OTSU), IForest combined with SVM (IForest-SVM), traditional Otsu's method, SVM, and RF. Furthermore, the effectiveness of the IForest-RF method in identifying large-scale Ulva prolifera areas was tested. The results demonstrate that the IForest-RF method achieved overall accuracies of above 90% and Kappa coefficients exceeding 95% in both test regions, significantly outperforming the other five methods. It yielded an overall accuracy of 98% and a Kappa coefficient of 95% in the large-scale test. Overall, the IForest-RF method exhibits high accuracy, excellent adaptability, and a high degree of automation, providing reliable technical support for real-time monitoring of Ulva prolifera disasters and ecological assessment.