A neural network approach to smarter sensor networks for water quality monitoring

Environmental monitoring is evolving towards large-scale and low-cost sensor networks operating reliability and autonomously over extended periods of time. Sophisticated analytical instrumentation such as chemo-bio sensors present inherent limitations because of the number of samples that they can take. In order to maximize their deployment lifetime, we propose the coordination of multiple heterogeneous information sources. We use rainfall radar images and information from a water depth sensor as input to a neural network (NN) to dictate the sampling frequency of a phosphate analyzer at the River Lee in Cork, Ireland. This approach shows varied performance for different times of the year but overall produces output that is very satisfactory for the application context in question. Our study demonstrates that even with limited training data, a system for controlling the sampling rate of the nutrient sensor can be set up and can improve the efficiency of the more sophisticated nodes of the sensor network.