Groundwater Level Prediction: Analyzing the Performance of LSTM and QLSTM Model

This study uses a multi-core, high-performance computer system to evaluate the performance of Quantum Long Short-Term Memory (QLSTM) and classical LSTM models for predicting groundwater levels. Leveraging historical groundwater data from different monitoring sites, this study conducts extensive numerical experiments in high-performance workstations. Groundwater level data collected by the United States Geological Survey (USGS) at two monitoring stations, i.e., Champion Well referred as site 1 and Pisgah Forest referred as site 2, located in North Carolina, has been used to train the QLSTM and LSTM models leveraging features such as evapotranspiration, precipitation, temperature, surface pressure, and wind speed. The outcome of this study provides distinct benefits to the meteorological community by enabling rapid prediction of groundwater levels at different locations. Besides model accuracy, this study also provides insights into the optimization of computational resources and provides an effective platform for deploying QLSTM and LSTM algorithms. Our finding suggests that the QLSTM model performs comparatively better than the LSTM model in predicting the groundwater level for rapidly changing environmental conditions.