Data-Driven Approaches for Understanding and Forecasting Well Leakage

In the transition to a low-carbon energy economy, oil and gas wells previously drilled for the extraction of fossil fuels will either need to be abandoned or converted to support novel subsurface energy operations (e.g., carbon storage, hydrogen storage). Abandoned legacy wells overlying such operations will also need to be monitored to ensure their integrity. To date, millions of wells have been drilled across the United States, and resources to abandon and monitor these wells are limited. This highlights the need to develop methodologies that prioritize their plugging, abandonment, and monitoring. The integrity of a well and its propensity for leakage are critical elements to be considered in prioritization schemes. This study uses publicly available regulatory data from the Wattenberg field in Colorado to train machine learning models, including gradient-boosted decision trees and spatial regression that forecast well integrity issues. This study will demonstrate the capability of models to make accurate predictions, identify drivers of future well leakage, and provide a framework for developing data-driven methodologies that determine wells and locations with a high risk of integrity loss and leakage.