Greater temporal and spatial sampling allows environmental processes and the well- being of our waterways to be monitored and characterised from previously unobtainable perspectives. It allows us to create models, make predictions and better manage our environments. New technologies are emerging in order to enable remote autonomous sensing of our water systems and subsequently meet the demands for high temporal and spatial monitoring. In particular, advances in communication and sensor technology has provided a catalyst for progress in remote monitoring of our water systems. However despite continuous improvements there are limitations with the use of this technology in marine environmental monitoring applications. We summarise these limitations in terms of scalability and reliability. In order to address these two main issues, our research proposes that environmental monitoring applications would strongly benefit from the use of a multi-modal sensor network utilising visual sensors, modelled outputs and context information alongside the more conventional in-situ wireless sensor networks. However each of these addi- tional data streams are unreliable. Hence we adapt a trust and reputation model for optimising their use to the network. For our research we use two test sites - the River Lee, Cork and Galway Bay each with a diverse range of multi-modal data sources. Firstly we investigate the coordination of multiple heterogenous information sources to allow more efficient operation of the more sophisticated in-situ analytical instrument in the network, to render the deployment of such devices more scalable. Secondly we address the issue of reliability. We investigate the ability of a multi-modal network to compensate for failure of in-situ nodes in the network, where there is no redundant identical node in the network to replace its operation. We adapt a model from the literature for dealing with the unreliability associated with each of the alternative sensor streams in order to monitor their behaviour over time and choose the most reliable output at a particular point in time in the network. We find that each of the alternative data streams demonstrates themselves to be useful tools in the network. The addition of the use of the trust and reputation model reflects their behaviour over time and demonstrates itself as a useful tool in optimising their use in the network.