Satellite imagery along with image processing techniques are essential tools for bathymetry retrieval since they are relatively inexpensive in term of time and budget in comparison to the conventional bathymetry survey. Approximately, 1126 bathymetric maps in various scales were produced by Badan Informasi Geospasial (BIG) covering Indonesia’s oceans and coastal waters. Those maps actually only cover 24,9% of total maps required to cover all Indonesia waters. Moreover, gaps in survey data, particularly in shallow water area, are also problematic. Only 12% of Indonesia’s waters are covered by a detailed survey. Hence, BIG is exploring remote sensing techniques to help with the improvement of Indonesia’s bathymetric maps including Satellite-Derived Bathymetry (SDB). SDB is promising due to its ability to fill the gap of depths derived from conventional hydrographic surveys. This is important for BIG since achieving wide-area depth coverage via the hydrographic surveys is costly in terms of time and money including for the safety of surveyors. This paper focuses on modelling bathymetry from multispectral images along Tanjung Kelayang, Tanjung Priok and Cilamaya shallow water area, Indonesia. A nonlinear machine learning technique was used to derive shallow water bathymetry by combining single beam echosounding measurements and the reflectance of red, green, blue, and near infrared bands of remotely sensed imagery. SVM can work better in a clear water area, for i.e. Tanjung Kelayang with RMSE less than 0.6. Furthermore, higher accuracies were obtained at depth range 0-5 m and 10-15 m for Tanjung Kelayang and Tanjung Priok, respectively. Highest accuracy was given in the depth range 0-5 m when applying SVM using four bands, but when using only three bands lowest RMSE value was obtained in the depth range of 5-10 m. SVM are able to estimate depth information in the shallow water area, especially in the water depth of <6 m (for i.e., Tanjung Kelayang and Cilamaya). However, the SDB results are noisy in the deeper water area >6 m. The inclusion of NIR band to the datasets results in a better accuracy. From results, we can see a correlation between the accuracy of bathymetric model and water clarity. High turbidity impacted the sensitivity of the depth algorithm. A complete data set containing water quality and benthic data is needed for further analysis to determine specific source of error.