There is a nonlinear relationship between rainfall and well water levels. This is one of the most complicated hydrologic phenomena to understand due to the existence of spatial and temporal incoherent geomorphic and climatic factors. The aim of this study is to predict the well water level by the artificial neural networks. A well is located on the campus area of Izmir Institute of Technology, Izmir, Turkey. While precipitation, outside temperature, and evaporation formed the input vector, the water levels were the target outputs. Precipitation and evaporation data were also recorded on the same campus area. The feed forward back propagation neural network is employed using the package program, called NeuroSolutions for Excel, due to its success in learning process, creating graphics for the results and sensitivity analysis. The findings of this study show that the model can successfully predict water levels in the well, with mean absolute error (MAE) of 37 cm and correlation coefficient (R) of 0.91 in the training stage and MAE = 0.40 cm and R = 0.80 in the testing stage. The sensitivity analysis results revealed that the outside temperature is the most effective parameter and theevaporation was least.

Key words: Artificial neural network, well water, prediction, precipitation, evaporation, temperature.