Fuzzy C-Mean Clustering Based Inference System for Saltwater Intrusion Processes Prediction in Coastal Aquifers

Abstract

Flow as well as salt transport processes of coastal aquifers are dependent on the density variation of water, and are complicated and extremely non-linear in nature. Simulation of these complicated systems to an acceptable degree of accuracy using a suitable surrogate model can be useful to achieve computational efficiency in modelling. In this study, a Sugeno type Fuzzy Inference System (FIS) is developed to predict salinity concentrations at specified monitoring locations as a result of groundwater pumping. Fuzzy c-mean clustering (FCM) algorithm is used to develop the FIS. Heterogeneity of aquifer is incorporated by using different hydraulic conductivities at different layers of the aquifer. A numerical simulation model, FEMWATER is adopted to generate the required patterns of inputs and outputs for initial training of the developed FIS. The FIS analyzed different combinations of antecedent transient pumping values, and returns concentration values at different monitoring locations. The performance of the FIS for training and validation datasets is evaluated by comparing the outputs obtained from the numerical simulation model. The trained and validated FIS is utilized as an approximate simulator of the coupled flow and salt transport processes. It is then used to forecast salt concentrations at different monitoring locations. Performance evaluation results indicate that the developed FIS can be applied to forecast salt concentrations at specified monitoring locations. The FIS is able to simulate the complex physical processes of heterogeneous coastal aquifers, and can be suitable for incorporation in a coupled simulation-optimization technique to develop optimum pumping strategy.