Prediction of lake eutrophication using ANN and ANFIS by artificial simulation of lake ecosystem

Data-driven models are getting increased attention in recent times for management and prediction of eutrophication in surface water bodies. The present study has used artificial neural network (ANN) and adaptive neuro-fuzzy inference system (ANFIS) techniques to predict eutrophication indicators dissolved oxygen (DO), Secchi depth (SD) and chlorophyll-a (chl-a) in water bodies in Assam, India. The dataset for the model development was gathered from two artificial prototype lakes (Lake-I and Lake-II) where eutrophication scenarios were replicated in controlled environment by periodic addition of waste water. Eleven physio-chemical water quality parameters were monitored for a span of 9 and 11 months' time for Lake-I and Lake-II, respectively. Two sets of DO, SD and chl-a models were trained under both ANN and ANFIS technique from the dataset of two lakes. All the models trained under ANN and ANFIS showed very good correlation between observed and model predicted values. The trained models were subsequently tested with natural water body data from samples collected from six different locations in Assam, India and reasonable predicting accuracy was obtained. Comparing the results of the study, it was found that the ANFIS models from Lake-II data were able to predict the values of DO, SD and chl-a more accurately where coefficient of determination values of 0.99 has been observed during both training and testing stage. Results of this study revealed suitability of the data-driven modelling approach for lake eutrophication management under circumstances where prolonged water quality data are not available for the waterbody.