Using ensemble decision trees to understand and forecast well integrity issues in the Wattenberg Field of Colorado

Oil and gas well integrity is critically important to maintain to prevent potential negative environmental and human health impacts and ensure the future use of reservoirs for other subsurface energy operations. Well integrity testing information are not widely available in the United States (US) as only a few states have regional well integrity testing programs, and most companies keep well testing information private. This creates a challenge for prospective operators and regulatory agencies who must characterize the leakage risks of wells to prioritize remediation and abandonment efforts or optimize site monitoring designs. In this study, we focus on the Wattenberg Field of Colorado a heavily developed natural gas and condensate field in the Denver Basin. Over 25% of wells in the Wattenberg Field have experienced sustained casing pressure (SCP), an indicator of an internal well integrity issue. We use publicly available regulatory information to build random forest and gradient boosted decision tree models that predict the occurrence of SCP in wells. Our approach represents a new data-driven methodology for categorizing oil and gas wells by their risk for integrity issues.