<![CDATA[Google Earth Engine (GEE) has emerged as one of the most powerful cloud-based platforms for processing and analyzing remote sensing imagery. By integrating vast Earth observation archives with scalable computational resources, it provides an accessible environment for researchers, practitioners, and decision-makers. In 2025, Google’s AlphaEarth Foundation introduced a novel embedding model trained on diverse Earth observation datasets available on the GEE server. This model, generated from annual time-series imagery and offered in an analysis-ready format, enables general-purpose applications such as classification, clustering, regression and change detection. Despite its potential, the performance and capabilities of this embedding model remain largely underexplored. This study evaluates the effectiveness of the embedding datasets in GEE for supervised classification method. Comparative experiments were conducted against widely used remote sensing imagery, including Sentinel-2 and Landsat 9 imagery, using multiple algorithms such as K-Neural Network (KNN), Support Vector Machine (SVM), Random Forest (RF), Classification and Regression Trees (CART), and Object-Based Image Analysis (OBIA). In addition, a case study was carried out to examine the use of embedding datasets for mangrove classification. Validation using overall accuracy demonstrates that embedding datasets achieve superior results compared to conventional imagery. Classification using the embedding dataset achieved an average overall accuracy of 94%, outperforming Landsat 9 (83.1%) and Sentinel-2 (82.5%). Moreover, the embedding dataset produced a classification pattern similar to OBIA, even without the need for image segmentation. The findings highlight the potential of embedding datasets to enhance classification accuracy and broaden the scope of remote sensing applications, suggesting new opportunities for leveraging advanced machine learning representations in geospatial analysis.]]>
