Offshore Freshened Groundwater prospecting using Machine Learning

Offshore freshened groundwater (OFG) represents a significant potential resource, with global volumes estimated at 10⁵–10⁶ km³. However, the scarcity of subsurface data on continental shelves poses challenges to understanding OFG systems' offshore extent, depth, and freshwater volume. Addressing these gaps, the OPTIMAL project leverages global geomorphological and sea-level datasets to develop machine learning models for OFG prediction and characterization. We present the results of the first stage of the project, including surrogate model design and parameter space definition. A suite of surrogate models was developed to capture key geological and geomorphological parameters influencing OFG systems. These 2D continental shelf profiles were defined by five parameters derived from open-source global datasets including shelf width, shelf-break depth, coastal unconsolidated sediment thickness and offshore aquifer properties. Numerical modeling of marine transgressive and regressive cycles was applied to these models to generate a training dataset encompassing OFG system realizations and associated parameter spaces. Initial ML models trained on this dataset demonstrate the feasibility of using surrogate models to overcome data scarcity issues in OFG characterization. Future work will refine these models, with a binary classification system to identify OFG presence and a multi-output regression for resource feasibility ranking. These results highlight the potential of integrating data-driven approaches to improve our understanding of OFG systems, providing a scalable framework for predicting OFG distribution and characteristics at both global and local scales.