Auto-DCA: An automated approach to decline curve analysis
- 1Department of Computer Science, I. K. Barber Faculty of Science, University of British Columbia, 3333 University Way, Kelowna, V1V 1V7, BC, Canada; Corresponding author.
- 2Department of Computer Science, I. K. Barber Faculty of Science, University of British Columbia, 3333 University Way, Kelowna, V1V 1V7, BC, Canada
- 3Faculty of Engineering, University of Victoria, 3800 Finnerty Road, Victoria, V8P 5C2, BC, Canada
Abstract
Decline Curve Analysis (DCA) remains a fundamental technique for forecasting production rates and estimating reserves in the oil and gas industry. However, traditional DCA workflows often depend on subjective analyst decisions, particularly in selecting decline start points, leading to inconsistencies and bias. This paper presents an automated DCA framework designed to reduce subjectivity, enhance reproducibility, and optimize the balance between error minimization and data coverage. The methodology integrates statistical models, automated start-date detection using an error tolerance approach, and multiple error metrics including Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), and Symmetric Mean Absolute Percentage Error (SMAPE) for validation. A quantitative evaluation against manual analysis using industry-standard software on 40 horizontal Montney wells demonstrates that the automated approach achieves lower error when fitting the complete production history. Using the full evaluation window (100%), RMSE decreases from 11.90 to 3.62 (absolute reduction of 8.28), and SMAPE improves from 34.67% to 18.58% (absolute reduction of 16.09 percentage points). The automated method outperforms manual analysis for the 100%, 75%, and 50% evaluation windows, while manual analysis achieves a better fit to the most recent production (15% window) and the two methods are close at the 25% window, consistent with standard industry practice of optimizing forecasts against recent data for reserve estimation. Even for recent data windows, the automated approach remains competitive (RMSE of 2.10 vs. 1.62 for the 15% window), while offering the advantages of a reproducible framework capable of processing thousands of wells without human intervention. On production withheld from all methods, three machine learning models trained on thousands of other wells had lower errors for forecasts of up to 5.75 years, while for forecasts of 6.5 to 8.75 years the differences were not statistically significant; unlike these models, the automated approach produces forecasts that are defined by a few decline parameters and can be reviewed and reproduced. The framework successfully processes wells automatically, with horizontal wells exhibiting median initial decline rates of 18.7%/year and b-factors ranging from 0.00 to 1.00.
Related
- ADEPTS: An auto-differentiable framework for time-dependent nonlinear thermo-chemical mantle convection inversion
- A Prototype Probability Based Warning System for a Strong AMOC Decline
- Large estimated size of the Australian Indigenous population before its extensive decline following colonial invasion
- Alarming decline in the carbon sink of European forests driven by disturbances
- Dust Decline Amplifies High‐Cloud Ice‐to‐Liquid Transition and Buffers the Radiative Feedback Under Warming
BibTeX
@article{abraha2026autodca,
title = {{Auto-DCA: An automated approach to decline curve analysis}},
author = {Temesgen Mikael Abraha and Afra Antara Anjum and Homayoun Najjaran and Patricia Lasserre and Yves Lucet},
journal = {Applied Computing and Geosciences},
year = {2026},
month = oct,
doi = {10.1016/j.acags.2026.100419},
url = {https://doi.org/10.1016/j.acags.2026.100419},
note = {CC BY},
}
RIS
TY - JOUR
TI - Auto-DCA: An automated approach to decline curve analysis
AU - Temesgen Mikael Abraha
AU - Afra Antara Anjum
AU - Homayoun Najjaran
AU - Patricia Lasserre
AU - Yves Lucet
PY - 2026
DA - 2026/10/30
JO - Applied Computing and Geosciences
DO - 10.1016/j.acags.2026.100419
UR - https://doi.org/10.1016/j.acags.2026.100419
AB - Decline Curve Analysis (DCA) remains a fundamental technique for forecasting production rates and estimating reserves in the oil and gas industry. However, traditional DCA workflows often depend on subjective analyst decisions, particularly in selecting decline start points, leading to inconsistencies and bias. This paper presents an automated DCA framework designed to reduce subjectivity, enhance reproducibility, and optimize the balance between error minimization and data coverage. The methodology integrates statistical models, automated start-date detection using an error tolerance approach, and multiple error metrics including Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), and Symmetric Mean Absolute Percentage Error (SMAPE) for validation. A quantitative evaluation against manual analysis using industry-standard software on 40 horizontal Montney wells demonstrates that the automated approach achieves lower error when fitting the complete production history. Using the full evaluation window (100%), RMSE decreases from 11.90 to 3.62 (absolute reduction of 8.28), and SMAPE improves from 34.67% to 18.58% (absolute reduction of 16.09 percentage points). The automated method outperforms manual analysis for the 100%, 75%, and 50% evaluation windows, while manual analysis achieves a better fit to the most recent production (15% window) and the two methods are close at the 25% window, consistent with standard industry practice of optimizing forecasts against recent data for reserve estimation. Even for recent data windows, the automated approach remains competitive (RMSE of 2.10 vs. 1.62 for the 15% window), while offering the advantages of a reproducible framework capable of processing thousands of wells without human intervention. On production withheld from all methods, three machine learning models trained on thousands of other wells had lower errors for forecasts of up to 5.75 years, while for forecasts of 6.5 to 8.75 years the differences were not statistically significant; unlike these models, the automated approach produces forecasts that are defined by a few decline parameters and can be reviewed and reproduced. The framework successfully processes wells automatically, with horizontal wells exhibiting median initial decline rates of 18.7%/year and b-factors ranging from 0.00 to 1.00.
SN - 2590-1974
ER -