In this study, bathymetric models were developed in coastal waters of the Mexican Caribbean, using an empirical algorithm and satellite images from Landsat 9, Sentinel-2 and SuperDove CubeSats. The model used is based on the relationship between the blue and green spectral bands and was adjusted with depth data taken <em>in situ</em> in the coastal area of Mahahual, Quintana Roo using linear regression, regularized regression and random forest approaches. The study showed that the model enables the estimation of depth relatively accurately, up to 20 meters. Sentinel-2 and random forest presented the best performance, with an RMSE error of 0.79 meters, followed by Landsat (0.88 m) and SuperDove (1.02 m). The most significant errors occurred at shallow depths of less than 5 m or greater than 20 m. Preprocessing of the images, particularly sunlight correction and spatial filtering, was crucial to improving the results. Remote sensing offers a very economical alternative for mapping bathymetry in shallow, low-turbid coastal areas.