Artificial Neural Network Approach for Predicting Transient Water Levels in a Multilayered Groundwater System under Variable State, Pumping, and Climate Conditions

Abstract

The feasibility of training an artificial neural network (ANN) for accurately predicting transient water levels in a complex multilayered ground-water system under variable state, pumping, and climate conditions is demonstrated. Using real-world data, an ANN was developed for a public supply wellfield and ground-water monitoring network located near Tampa Bay, Florida. The ANN was trained to predict transient water levels at 12 monitoring well locations screened in different aquifers in response to changing pumping and climate conditions. The trained ANN was then validated with ten sequential seven-day periods, and the results were compared against both measured and numerically simulated ground-water levels. The absolute mean error between the ANN predicted and the measured water levels is 0.16 m, compared to the 0.85 m absolute mean error achieved with the calibrated numerical model at the same locations over the same time period. The ANN also more closely reproduced the dynamic water level responses to pumping and climate conditions. The practical implication is that if ANN technology can achieve superior ground-water level predictions, it can be used to improve management strategies for a wide range of ground-water problems, from water quantity to water quality issues. It can also serve as a powerful sensitivity analysis tool for quantifying interrelationships between different variables, fostering a better understanding of the hydrogeologic system, and improving future modeling endeavors. And while physical-based numerical modeling retains some advantages over the ANN technology, both approaches may be used in a complementary fashion to achieve sound decision-making for complicated ground-water management problems.