Estimation of Daily Groundwater Table Using Backpropagation Neural Network Model by Assessing Training Algorithm

Abstract

The Onça stream is a Brazilian waterway, a canal on the left bank of the Rio das Velhas and a tributary of the São Francisco River. Although it plays an important role in the management of groundwater levels in the Ribeirao Preto region, the final groundwater capacity is declining in many parts of Brazil in response to abstraction. Predicting and forecasting stream water level using simple but effective methods can provide a reliable tool for future management of groundwater. In the present study, the daily water level of a particular well from the year 2004 to 2014 was considered as the dataset. The adopted database was segregated as 70% for training the above-mentioned models and the remaining 30% data was utilized for model testing. One of the soft computing methods, Backpropagation Neural Network was utilized for the prediction of daily groundwater table. In this, BPNN model, various training algorithm were utilized and compared based on the value of coefficient of correlation (R)and the time taken. The model also assessed by various statistical parameters like root means square error (RMSE), mean total error (MAE), determination coefficient (R²), Normalized mean biased error (NMBE). The results depicted that the developed model have good forecasting ability.