There is often a need for making a high-resolution or a complete bathymetric map based on sparse point measurements of water depth. The common practice of previous studies has been to calibrate a single global depth regression model for an entire image. The performance of conventional global models is limited when the bottom type and water quality vary spatially within the scene .For a more accurate and robust water-depth mapping , this study proposes a regression model for a geographical region or local area rather than using a global regression model. The global regression model and Geographical Weighted Regression (GWR) model are applied to Landsat 8 and RapidEye satellite images. The entire data analysis workflow was carried out using GRASS GIS Version 7.0.0. Comparison of results indicates that the GWR model improves the depth estimation significantly, irrespective of the spatial resolution of the data processed. GWR is also seen to be effective in addressing the problem introduced by heterogeneity of the bottom type and provide better bathymetric estimates in near coastal waters. The study was carried out at Pureto Rico, northeastern Caribbean sea. Two different satellite data were collected in order to test the algorithm with high and moderate resolution data. RapidEye data has 12-bit radiometric resolution and 5 meter spatial resolution. Even though Landsat 8 data also has 12-bit radiometric resolution, it provides 30 m spatial resolution. In order to calibrate and evaluate the estimated depth, high accuracy LiDAR depth data (4 m resolution) provided by NOAA is used. The study was demonstrating GWR model to estimate depth, evaluate and compare the results with a global conventional regression model. The comparative study between conventional global model and GWR model shows that GWR model significantly increases the accuracy of the depth estimates and addresses spatial heterogeneity issue of the bottom type and water quality. The GWR model provide better accuracy at both Landsat 8 (R-squared=0.96 and RMSE=1.37m) and RapidEye (R-squared=0.95 and RMSE=1.63m) than global model at Landsat 8 (R-squared=0.71 and RMSE=3.71m) and RapidEye (R-squared=0.71 and RMSE=4.04m).