<p>Coastal bathymetry measurement is tedious, time-consuming, and expensive task, making it difficult for large-scale bathymetric surveys. To overcome the limitations of manual bathymetric surveys, this research study estimates the coastal bathymetry along the west coast of Mumbai using single-band reflectance in the visible-nir spectra of Landsat imagery and chart datum (CD) depths. Water depths were predicted using machine learning (ML) models such as Random Forest (RF), Support Vector Regressor (SVR), and a Neural Network (NN). The impact of tidal fluctuations and sewage discharges through drains, creeks, and marine outfall on depth prediction has also been assessed and compared through depth cross-sections. The blue band emerged as the most informative for depth prediction showcasing consistently lower error metrics across models. RF demonstrates satisfactory accuracy across tidal conditions, whereas the NN model exhibits comparable performance to RF in low tide (LT) conditions but is less accurate in high tide (HT) scenarios. SVR captures bathymetry patterns effectively but faces challenges in replicating the full depth range. These findings highlight the importance of band selection and tidal considerations for accurate coastal bathymetry assessment, which is crucial for informed coastal management and environmental protection.</p>