Many coastal areas, especially in developing countries or those with limited marine activity, lack detailed depth measurements. Past data in these areas may be incomplete or outdated, making it difficult to create accurate seafloor maps. This is important for the preliminary design of coastal structures. This study aims to explore the best way to use satellite images and open-source software to create Satellite-Derived Bathymetry (SDB) models. Our approach uses three machine learning algorithms (KNN, RF, MLR) to analyze satellite images of different areas. The images come from open-source databases. We use the closest truth data to the targeted area to train the algorithms to predict the unseen data. Our research shows that using satellite data to measure water depth can accurately determine depths of up to 27 meters. Furthermore, our assessment reveals mean absolute errors averaging 0.72 meters and root mean square errors averaging 1.0 meter, with accuracies around 94.6% for both samples. Random Forest (RF) performed better than KNN and MLR. In Sample El-Dabaa, RF performed well with Landsat-08 Single Scene image. The area has rocky cliffs with seagrass, steep slopes, and strong wave movement. In Sample El-Arish, RF's best results came with single scene image from Landsat-08. This area has sandy soil, gentle slopes, and gentle wave movement. Generally, usage of Single Scene image or Median Value image with ML algorithm depends on the seafloor dynamics.