Water pollution in the Bolívar department of Colombia is a critical environmental problem. This problem affects aquatic ecosystems, water quality, and access to clean water for domestic, agricultural, and industrial use. Conventional water contamination monitoring methods are often costly, laborious, and limited in spatial and temporal coverage. This makes accurate and timely assessments, particularly in large or difficult to access regions. Moreover, the dependence on spot sampling and laboratory analysis limits the availability of the results. This limits the ability to respond to unexpected contamination events. This limits the ability to respond to unexpected contamination events. These limitations have prompted the adoption of alternative approaches. These include the use of real-time sensors, remote sensing, and satellite imagery. Predictive models based on artificial intelligence have also been developed. These tools allow optimizing water body monitoring and extend the spatial and temporal coverage of water contamination monitoring. In this context, this study aims to address the challenges of conventional monitoring. This study proposes a deep learning-based model to estimate and predict contamination in strategic wetlands, with a particular focus on \textit{Ciénaga de la Virgen}. The methodology integrates Sentinel-2 remote sensing data with \textit{in-situ} measurements of key indicators of water contamination. A dataset was constructed by combining spectral information and water contamination parameters from monitoring stations. The results achieved a promising prediction of the contamination variables. The proposed framework facilitates sustainable management of water resources and contributes to understanding the dynamics of contamination.