Estuaries and coastal areas contain increasingly exploited resources that need to be monitored, managed and protected efficiently and effectively. This requires access to eliable and timely data and management decisions must be based on analysis of collected data to avoid or limit negative impacts. Visually supported multi-modal sensing and data fusion offer attractive possibilities for such arduous tasks. In this paper, we demonstrate how an in-situ sensor network can be enhanced with the use of contextual image data. We assimilate and alter a state-of-the-art background modelling technique from the image processing domain in order to detect turbidity spikes in water quality sensor measurements automatically. We then combine this with visual sensing to identify abnormal events that are not caused by local activities. The system can potentially assist those charged with monitoring large scale ecosystems, combining real-time analytics with improved efficiency and effectiveness.