42-23 Evaluating Artificial Intelligence as an Aid to Electrical Resistivity Imaging Interpretation at a Fuel Impacted Site

Abstract:

Electrical Resistivity Imaging (ERI) produces large datasets that can be difficult to interpret. The objective of this research was to examine whether AI could be useful in identifying zones of interest at a light non-aqueous phase liquid (LNAPL) impacted site and improve subsurface characterization. Rather than using only manual or statistical analysis, the goal was to evaluate how effectively AI could process the large datasets and if it could accurately make decisions about anomalies of interest. AI demonstrated a very clear advantage in efficiency. The AI models excelled at completing repetitive or time-consuming tasks such as isolating the highest and lowest resistivity values and locating them along the ERI line. Automating this step of interpreting the data reduces the need for manual coding and accelerates an initial interpretation. However, several limitations to the AI method became apparent. Many AI platforms withhold their full capabilities behind a paywall by imposing file-size restrictions, thus preventing the full ingestion of the datasets and reducing the accuracy of the interpretation. Additionally, the AI was more efficient with images of datasets that utilize the raw pixel data rather than files of raw data. The most reliable results occurred when the AI outputs were using paid AI engines and integrating results with the known hydrogeologic model. While AI could prove to be a powerful tool in geophysics it should not replace human judgement and interpretation but can be a useful aid in processing large amounts of data across sites.

© Copyright 2026 The Geological Society of America (GSA), all rights reserved.