Exposing Critical Data Reliability Issues: How Public Water Production Records Fail to Capture Reality in Tight Oil Wells

The unconventional oil and gas industry relies heavily on publicly available production data for environmental impact assessments, economic evaluations, and regulatory decision-making. However, a comprehensive analysis across over 10,000 tight oil wells in four states reveals alarming discrepancies between public and proprietary water production data that fundamentally undermine the reliability of models used for critical industry and policy decisions. These findings challenge the validity of numerous published forecasts and highlight the urgent need for improved data collection and reporting standards.

This blog post is based on the technical paper “The Reliability of Public Water Production Data for Tight Oil Wells”, co-authored by Frank Male¹˒², Ian Duncan², and Kiran Sathaye. The paper was presented at URTeC 2025. Download the full paper here.

The Hidden Crisis in Water Production Data

Hydraulically fractured wells consistently generate substantial water production that significantly impacts per-well economics, environmental considerations, and operational planning. This produced water rarely provides economic value while imposing substantial costs for pumping, processing, and disposal. State regulatory agencies have historically focused primarily on hydrocarbon production, often treating water production reporting as a secondary consideration despite its critical importance for environmental impact assessment and economic viability evaluation.

The accuracy of publicly available water production estimates has remained controversial, creating uncertainty in the robustness of models based on production data. Researchers have attempted to forecast water production using publicly available datasets, but the fundamental reliability of these data sources has not been systematically examined across different regulatory jurisdictions. This knowledge gap presents significant challenges for environmental impact assessment, economic modeling, and policy development in unconventional resource development.

The potential for data quality variation across states stems from differing regulatory standards and reporting requirements. Arkansas requires no water production reporting, while Texas mandates yearly well tests and New Mexico expects monthly reports. These regulatory inconsistencies create a patchwork of data quality that may significantly impact the reliability of basin-wide or national-scale analyses that depend on public production records.

Comprehensive Multi-State Data Validation Methodology

The study addresses these critical data reliability questions through systematic comparison of public and proprietary water production data across 10,000 tight oil wells spanning New Mexico (1,200 wells), North Dakota (1,300 wells), Texas (7,700 wells), and Utah (100 wells). This extensive dataset enables robust statistical analysis across different regulatory environments and geological settings.

Public data sources varied by state, reflecting different regulatory approaches and reporting standards. Texas data originated from the Railroad Commission, comprising lease-wide monthly production reports and yearly well tests that include water and hydrocarbon production over 4-to-24 hour periods. Proprietary data came from the Novi Labs database, where members of the Novi Data Network contributed high-quality proprietary production records, supplemented by additional company contributions provided under strict confidentiality agreements.

The analytical methodology involved multiple comparison approaches including raw data analysis, transformed and normalized data evaluation, and assessment of both water production volumes and water cut percentages. Comparisons utilized Pearson’s correlation coefficients and Spearman’s rank-correlation metrics across different time intervals including first six months, first year, and first five years of production. One thousand bootstrapped samples were analyzed to determine statistical uncertainty in correlation measurements.

Shocking State-by-State Data Quality Revelations

The analysis revealed dramatic differences in data quality across states that fundamentally challenge assumptions about public data reliability. Utah consistently demonstrated exceptional data quality with R-squared values above 0.9 across all time periods, indicating that public production records accurately reflected proprietary operational data throughout the well lifecycle.

New Mexico and North Dakota exhibited intermediate data quality characteristics with early production showing R-squared values of approximately 0.6, improving to above 0.9 for later production periods. This improvement pattern suggests that regulatory reporting systems in these states achieve greater accuracy as wells mature and production stabilizes, potentially reflecting improved operator reporting practices or enhanced regulatory oversight over time.

Texas presented the most concerning data quality issues with R-squared values never exceeding 0.6 and falling to 0.1 for early water production periods. These extraordinarily poor correlations between public and proprietary data indicate fundamental problems with the state’s water production reporting system that persist throughout the well lifecycle. Even after five years of production, Texas public data failed to achieve the accuracy levels demonstrated by other states during early production periods.

The temporal patterns revealed consistent improvement in correlation metrics as production time increased across all states, suggesting that longer production histories enable more accurate data compilation. However, the magnitude of improvement varied significantly between states, with Texas showing minimal improvement compared to the substantial accuracy gains observed in New Mexico and North Dakota.

Advanced Machine Learning Solutions for Data Correction

Recognizing the critical limitations of Texas public data, the study implemented gradient boosting machine learning techniques to improve water production predictions for Midland Basin tight oil wells. The machine learning approach achieved significant improvement, increasing R-squared values from 0.44 for raw public data to 0.62 for corrected predictions, representing a 41% improvement in predictive accuracy.

The machine learning model identified key predictors of accurate water production including public water cut data, public one-year cumulative oil production, and well age. Surprisingly, water volumes injected during hydraulic fracturing proved to be poor predictors of subsequent water production, challenging conventional assumptions about flowback and formation water relationships.

The model demonstrated reasonable accuracy between negative one and positive two standard deviations from the mean, with low water cut predictions becoming progressively more challenging to correct. This performance limitation suggests that while machine learning can significantly improve data quality, fundamental reporting deficiencies may limit the achievable accuracy improvements.

These machine learning results provide proof-of-concept that advanced analytical techniques can partially compensate for regulatory data quality deficiencies. However, the computational complexity and proprietary data requirements for model training limit the practical applicability of this approach for routine industry or regulatory use.

Critical Industry and Environmental Implications

The documented data quality deficiencies present serious implications for environmental impact assessment, economic evaluation, and regulatory oversight in unconventional resource development. Texas represents the largest unconventional oil producing state in the United States, making the poor quality of its water production data particularly concerning for national-scale environmental and economic analyses.

Previously published forecasts of water balance in the Permian Basin and Eagle Ford tight oil plays may contain significant errors due to reliance on unreliable public data sources. These forecasting errors could lead to inadequate infrastructure planning for water handling and disposal, incorrect environmental impact assessments, and flawed economic projections that influence investment decisions and regulatory policies.

The inability to accurately predict water production affects critical operational decisions including when wells become uneconomical due to water lifting and disposal costs, impacting reserves estimates and field development planning. Incorrectly estimated water production can lead to premature well shut-ins or continued operation of uneconomic wells, both of which result in suboptimal resource recovery and economic performance.

Environmental stakeholders relying on public data for impact assessment may significantly underestimate or mischaracterize the water management challenges associated with unconventional development. This could lead to inadequate environmental protection measures or inappropriate regulatory responses that fail to address actual environmental risks.

Transformative Recommendations for Industry Practice

The research findings necessitate immediate attention to data quality improvement across multiple dimensions of unconventional resource development. Operators should consider partnering with data providers and researchers to improve model fidelity through proprietary data sharing arrangements that maintain competitive confidentiality while enabling more accurate analytical results.

Alternative data sources such as saltwater disposal volumes may provide more reliable estimates of actual water production than direct regulatory reporting, particularly in states with demonstrated data quality deficiencies. This approach could enable cross-validation of reported production volumes and identification of systematic reporting errors.

Regulatory authorities, particularly in Texas, should consider implementing enhanced reporting requirements and validation procedures to improve public data accuracy. This could include monthly reporting mandates, automated data validation systems, and penalties for inaccurate reporting that create stronger incentives for accurate data submission.

The machine learning methodology developed in this study provides a framework for data quality improvement that could be implemented by operators, data service companies, and regulatory agencies. However, the success of such approaches depends on access to high-quality proprietary data for model training and validation.

Revolutionary Data Quality Assessment Framework

The study establishes a systematic methodology for evaluating production data quality that can be applied across different regulatory jurisdictions and geological settings. The bootstrap sampling approach provides robust uncertainty quantification that enables more informed decision-making about data reliability in specific applications.

The comparison of water production volumes and water cut percentages reveals important insights about the relative reliability of different data transformation approaches. Water cut data generally demonstrated superior reliability compared to raw water production volumes, suggesting that normalization approaches may partially mitigate absolute volume reporting errors.

The temporal analysis demonstrating improving data quality over well lifecycle provides guidance for analytical applications that can adjust confidence levels based on well maturity. This framework enables more sophisticated uncertainty handling in production forecasting and economic evaluation applications.

The multi-state comparison approach reveals the critical importance of regulatory environment in determining data quality, providing valuable insights for policy development and regulatory reform initiatives. This methodology could be extended to evaluate data quality for other production parameters and across additional regulatory jurisdictions.

Uncover the Complete Data Quality Analysis

This analysis represents only the beginning of a comprehensive data reliability assessment that fundamentally challenges industry assumptions about public production data quality. The complete 11-page technical study includes detailed statistical validation procedures, comprehensive machine learning methodologies, state-by-state regulatory analysis, and step-by-step implementation guidelines for data quality improvement that can be immediately applied across diverse operational settings.

Want access to the full research along with our library of resources? [Get Full Access to Our Content Hub →]

Key Takeaways

• A multi-state validation compared public versus proprietary water production data across 10,000 tight oil wells in NM, ND, TX, and UT.
• Utah public water production data matched proprietary records with R-squared above 0.9 across all evaluated time periods.
• New Mexico and North Dakota showed R-squared near 0.6 in early production, improving to above 0.9 in later periods.
• Texas correlations never exceeded R-squared 0.6 and fell to 0.1 for early water production, indicating persistent reporting issues.
• A gradient boosting model for Midland Basin wells increased R-squared from 0.44 using raw Texas public data to 0.62.
• Public water cut, public one-year cumulative oil, and well age were top predictors, while hydraulic fracturing injected volumes were weak predictors.
• Water cut generally showed higher reliability than raw water volumes, indicating normalization can partially reduce absolute reporting error impacts.
• Published water-balance forecasts for the Permian Basin and Eagle Ford may be materially affected when they rely on Texas public water data.