The Sentinel-2 mission is a key Copernicus program, providing consistent coverage from two satellites with polar-orbiting, sun-synchronous orbits with a revisit time of 5 days at the equator. Each satellite delivers consistent, high-resolution (10-60 m) optical imagery, suitable for monitoring water bodies, especially for high-latitude, cloud- and snow-prone regions such as Iceland. This study evaluates the effectiveness of three cloud-snow masking approaches -(a) the MSK_CLDPRB and MSK_SNWPRB, which are probability-based masks, (b) the Scene Classification Layer (SCL) codes: 1,2,3,8,9,10,11 - in improving image usability for subsequent remote sensing analyses, and (c) a machine learning method which is using the CART algorithm trained on six spectral bands and SCL cloud-snow code labels from one image per year, applied through the entire collection. To focus on Iceland’s challenging conditions, we investigated the Landeyjahöfn Harbour during February from 2019 to 2024. We assessed cloud-snow pixels’ removal by each approach by computing the percentage of them. Results indicate that our SCL-based method outperformed either the machine learning or probability-based methods in percentage and number of removal pixels and has the advantages of being implemented into preprocessing workflows for studies in cloud- and snow-prone regions such as the Landeyjahöfn Harbour. This reiterates the strength and reliability of the SCL product based on the ESA multi-source classification algorithm, specific to Sentinel-2 Level-2A imagery. Our work represents a contribution to establishing effective cloud-snow masking over optical remote sensing images, as well as a perspective on AI infusion in the context of satellite imagery for water-related studies.