Predicting conductivity and acidity for small streams using neural networks

Acidity and conductivity are parameters which must be studied to assist in the management and understanding of watercourses. Studies of these parameters can require continuous series of some length which can be difficult to obtain. For instance, data series may have gaps because of problems with data acquisition. These gaps, which can interfere with analysis, can be filled in with model-generated data. The purpose of this study is to model Moose Pit Brook and Pine Marten Brook pH and conductivity with neural networks. These streams are in the region of Kejimkujik National Park in Nova Scotia, Canada. Daily flow values and the time of year were used as inputs for the networks. The coefficients of determination for the networks chosen to predict the output variables varied from 0.802 to 0.976 for the training series and from 0.716 to 0.967 for the evaluation series.