USING NEURAL NETWORKS TO PREDICT PH CHANGES IN ACIDIFIED EASTERN CANADIAN LAKES

Predicting changes in the pH of lakes with changes in acid precipitati on is a difficult endeavor, especially in areas with high concentratio ns of natural dissolved organic acids, such as are found in much of ea stern Canada. Statistical techniques have been used in a number of stu dies to try to predict changes in large areas, but cannot be used with confidence because of autocorrelation between variables. Because of t he difficulties encountered with chemical and statistical techniques, we used a neural network approach to see if patterns could be detected in water chemistry data. We produced a model that was able to accurat ely predict the current pH of 164 lakes within 2% of measured values. We then varied lake sulphate concentration, the main acidification inp ut to this region, to see how the lake pH would change. The neural net approach seemed more sensitive than statistical approaches in making predictions.