Simulation of Water Table in Tubular Well Located in Unconfined Aquifer: a Comparison of Predictive Techniques

Abstract

The use of autoregressive models for the prediction and gap filling from historical time-series has grown notably in the study of hydrological data, especially for activities that involves the study of water balance and hydric demand management in watersheds. This work aimed to examine and compare the potential use of different predictive techniques of autoregressive models for groundwater level simulation in tubular well located in unconfined aquifer in the State of Rio Grande do Sul, Brazil. Autoregressive Moving Averages (ARMA), Support Vector Machines (SVM) and Artificial Neural Networks (ANN) methods were used to analyze the performance of these methodologies for the prediction of groundwater levels on hourly and daily scale. Static water level and rainfall series were collected every hour by means of an automatic data logger monitoring station. The methodology based on Artificial Neural Networks presented the best performance, evidenced by the Nash-Sutcliffe Coefficient (NSC) in the order of 0,99 in hourly scale and, in order of 0,84 in daily scale. The residue analysis stage demonstrated the small margin of error achieved, allowing to validate the model for practices and studies in time series of groundwater levels.

Keywords

• Groundwater
• Hydrologic modeling
• Autoregressive Moving Averages
• Support Vector Machines
• Artificial Neural Networks