Determination of groundwater level change using M5 decision tree and artificial neural networks methods

In this study, groundwater behavior was investigated with a stochastic model using groundwater level (GWL) as dependent variables and precipitation, temperature and previous day's groundwater level as independent variables. Observation well station PI98-14 near Prairie Island in Minnesota, USA, where the groundwater level is high, was selected. It is thought that the models to be created for this observation station will be useful in monitoring and controlling groundwater in Minnesota, monitoring possible changes in the future and taking precautions against possible problems. In the study, the data obtained from the PI98-14 observation well of the reservoir near the Prairie Island in the city of Minnesota in the United States was used. These data, which include 2025 days of precipitation (Y), temperature (S) and the groundwater level of the previous day (GWLt-1), were obtained from the United States Geological Survey (USGS). These obtained data were used as inputs in Artificial Neural Networks (ANN), Multiple Linear Regression (MLR) and M5 Decision Tree (M5T) methods and new groundwater level (GWL) values were estimated. In each of the three models used, one evaluation was made. The estimation results of the models were compared according to the Coefficient of Determination (R2), Mean Absolute Error (MSE) and Root Mean Square Error (RMSE) criteria. According to this comparison, all models showed close results and were successful in groundwater level estimation.