Application of K-star Algorithm for Groundwater Level Forecasting (Case study: Aspas plain)

In this study, the groundwater level of Aspas Plain, located in the northwest of the catchment area of Tashk and Bakhtegan Lakes in Fars province, was simulated using the data of 39 piezometers in the aquifer plain in the period of 2001-2017. In this regard, K-Star and ANN models with 1, 2 and 3 lags were used in two phases of training and testing. ANN model in this study was also investigated with the number of hidden layers and different neurons in addition to the mentioned lags. According to the drop and fluctuations of the groundwater level in the mentioned statistical period, the studied aquifer was divided into 4 regions a, b, c and d. The results of the simulation of the groundwater level in area “a” showed that the K-Star model with 2 lags performed better than the ANN model in different lags. But in the other three regions, the ANN model with a lag-1 has provided the best performance. ANN model with 2 hidden layers and 4 neurons in region b, with 2 hidden layers and 6 neurons in region c, and with 4 hidden layers and 5 neurons in region d were introduced as the best model for simulating groundwater level values. On average, the best models introduced in regions b, c and d were able to improve the error rate by 57, 42 and 81%, respectively, compared to the K-Star model. The poor performance of the K-Star model in two regions b and d is also clearly visible. In general, the results showed that lag-1 of the ANN model had the best performance in simulating the values of the groundwater level in the three regions b, c and d with the number of hidden layers and different neurons.