Artificial Neural Networks and Support Vector Machines: Contrast Study for Groundwater Level Prediction

Abstract

Abstract. The evaluation of relationships between irrigation demand and groundwater availability is important to effectively manage water sources for agricultural production. In the field of groundwater modeling, Artificial Neural Networks (ANN) has been widely used in the forecasting of level fluctuations, because of its performance and capacity to reproduce the groundwater seasonal variability. Meanwhile, Support Vector Regression (SVR) is another machine learning technique that has gained popularity in hydrologic sciences because it provides global predictions, accurate estimates and is based on statistical learning theory. These types of machine learning algorithms are useful especially when field measurement of daily groundwater levels is unsustainable. In addition to the limited number of studies that evaluates the advantages and disadvantages of ANN and SVR in groundwater level prediction, there is a need to develop more research for the analysis of new models in forecasting groundwater levels. The objective of this study was to compare two machine learning algorithms for the prediction of groundwater variations at a local level in a shallow alluvial aquifer. ANN and SVR were evaluated in terms of optimization efficiency and runtime speed for an individual well located in Sunflower Mississippi, to identify the best model for simulating changes in groundwater levels, providing estimations up to three months ahead. The analysis of performance between the two models was based on the mean square error (MSE) results and the model’s ability to represent the groundwater withdrawal periods. Results from ANN and SVR simulations showed that the models can efficiently predict daily groundwater levels. However, for the recharge (winter) season, ANN performed better than SVR with MSE values of 0.005612 and 0.09369 m, respectively. A robust and suitable groundwater forecasting model is a useful decision support tool for farmers and stakeholders to develop groundwater conservation plans and improve crop production with an efficient use of water.