APLICAÇÃO DE REDES NEURAIS ARTIFICIAIS PARA PREVISÃO DO NÍVEL DE ÁGUA SUBTERRÂNEA EM POÇO DE MONITORAMENTO NA BACIA SEDIMENTAR DO ARARIPE, CEARÁ

Application of artificial neural networks for groundwater level forcasting on monitoring well in Araripe Sedimentary Basin, Ceara

Abstract

Groundwater level forecasting is essential for water availability. Formalisms such as Artificial Neural Networks (ANN) are broadly used to time series modeling and forecasting. The aim of this paper was to evaluate the application of ANN models for forecasting groundwater level of one well installed at Medium Aquifer System in Araripe sedimentary basin, Ceará. Feedforward ANN models with one hidden layer were applied. By using the time series past values as model inputs, the optimal network architecture was researched, with respect to the number of nodes on input and hidden layers. The ANN models were applied according to single and combined modeling approaches, through linear combination of forecasts, such as Simple Average and Simple Median. The model performances were measured and compared by well-known statistics metrics, such as RMSE and R². Results highlights the ANN good performance, with RMSE = 0,032 m and R² = 0,99. Single models outperformed combinations. This research showed how ANN are efficient models for forecasting groundwater level, even on complex systems and with a few input variables, representing a tool with large applicability on groundwater resources management.