Groundwater Level Prediction Using Supervised Committee Fuzzy Logic (Case Study: Meshginshahr Plain)

Groundwater is the main supply of drinking and agriculture demands in Meshginshahr plain located on Northwest of Iran in the Province of Ardebil. The investigation of groundwater level fluctuations is necessary for effective groundwater management in this plain. For this purpose, artificial intelligence methods are interested due to high ability, cost effectiveness, needing less data, and fast running. This study presents a supervised committee fuzzy logic (SCFL) model to predict groundwater level at three piezometers in the study area. For implementing SCFL model, first fuzzy logic models such as Mamdani fuzzy logic (MFL), Larsen fuzzy logic (LSL) and Sugeno fuzzy logic (SFL) were applied to predict groundwater level using precipitation, temperature, discharge of abstraction wells and groundwater level with one month lag data. Then a supervised committee fuzzy logic as a non-linear model was used to combine the outputs of individual fuzzy models to reap the advantages of all three models simultaneously. Three different criteria RMSE, MAE and R2 were used to assess the prediction efficiency and accuracy of models. Based on results, MAE values of SCFL model are 0.12, 0.04 and 0.03 for piezometer 1, 2, and 3 respectively for training step. It presents the superiority of SCFL model over the individual fuzzy models. Also SCFL model could reduce prediction RMSE to 6% for piezometer 1 and 8%, 14% for piezometers number 2 and 3 respectively.