Synthetic Time Series Data in Groundwater Analytics: Challenges, Insights, and Applications

Abstract

This study presents ‘Synthetic Wells’, a method for generating synthetic groundwater level time series data using machine learning (ML) aimed at improving groundwater management in contexts where real data are scarce. Utilizing data from the National Water Information System of the US Geological Survey, this research employs the Synthetic Data Vault (SDV) framework’s Probabilistic AutoRegressive (PAR) synthesizer model to simulate real-world groundwater fluctuations. The synthetic data generated for approximately 100 wells align closely with the real data, achieving a quality score of 70.94%, indicating a reasonable replication of groundwater dynamics. A Streamlit-based web application was also developed, enabling users to generate custom synthetic datasets. A case study in Mississippi, USA, demonstrated the utility of synthetic data in enhancing the accuracy of time series forecasting models. This unique approach represents an innovative first-of-its-kind tool in the realm of groundwater research, providing new avenues for data-driven decision-making and management in hydrological studies.