Remote sensing data encompasses multiple modalities, including multispectral images, Synthetic Aperture Radar (SAR), Digital Elevation Model (DEM), and more. Each type of data possesses unique properties and characteristics. Preparing a dataset for machine learning that includes these modalities is challenging and requires specific geospatial data expertise. In this paper, we introduce a novel service, MuSS, for multimodal remote sensing data processing and analysis. MuSS dynamically prepares datasets based on user specifications, leveraging Google Earth Engine (GEE) to seamlessly access and process diverse satellite data. Using the MuSS-prepared multimodal satellite dataset, we propose an unsupervised deep learning model for pixel-level land-cover classification. Unlike traditional methods that require labeled data, which is difficult to gather in remote sensing, our model utilizes derived indicators to accurately distinguish water and other surfaces without labeled training data. Our service is particularly beneficial for agriculture and assessing climate change impacts, yet its configurable ETL pipeline ensures broad applicability. Experimental results demonstrate that our approach significantly outperforms traditional baseline methods, highlighting the service’s efficacy and versatility.