Enhanced recovery

Analysis of Athabasca Oil Sands Investigates SAGD Performance Variability

Steam-assisted gravity drainage (SAGD) performance in bitumen-recovery projects in Alberta is affected by geological deposits, reservoir quality, and operational experience.


Steam-assisted gravity drainage (SAGD) performance in bitumen-recovery projects in Alberta is affected by geological deposits, reservoir quality, and operational experience. The authors reviewed and analyzed actual field production and injection data for 28 Athabasca oil-sands-deposit SAGD well pairs (WPs). Based on analysis of field-production data, a numerical model was built and calibrated against production data from two of the poorer-performing WPs among the 28 studied. Agreement between simulated and actual cumulative oil and steam/oil ratio (SOR) was within 10% after 6 years of operations on a first-iteration basis.

Problem and Investigation Methodology

A survey of the literature reveals that some aspects of the SAGD process, particularly with regard to the behavior of gas, still largely are not understood. While successful history-matching simulations of SAGD WPs exist for several different projects, these rely on dead-oil pressure/volume/temperature treatment and exclude any findings and discussion on gas production. Such history-matched models likely would need to be modified significantly to be useful for modeling the wind-down stage of SAGD operations. Recent numerical studies modeling the generation of aquathermolysis gases focused on the injection of noncondensable gas into a mature SAGD steam chamber.

In preparation for a series of numerical studies on aspects of SAGD performance, a reservoir model simple enough to be relatively adaptable for different geological settings was built. Such a model had to be calibrated rigorously with actual production data to provide a high degree of confidence in its results and predictive capability in appropriate contexts. The investigation in the complete paper is aimed at addressing several fundamental aspects of SAGD operations, including the effects of reservoir heterogeneities and the behavior of gas. The first part of the complete paper is focused on reviewing and analyzing actual field production data spanning more than 1,700 days from 28 SAGD WPs from four different pads (A, B, C, and D) of the Jackfish 1 project; this synopsis will not include that extensive data.

Jackfish 1 Project. Jackfish 1 consists of 42 WPs divided into six pads (including the 28 WPs in Pads A, B, C, and D analyzed for the authors’ study) and is part of the geological oil-sands trend of the Athabasca oil-sands deposit. Jackfish 1 has a nameplate capacity of 35,000 BOPD, with a designed SOR of 2.7. Consistently exceeding 90% of its nameplate capacity since first steam, it is commonly considered to be a successful SAGD project.

The results of data analysis for the WPs studied indicated the following:

  • Considerable variance exists in the recovery performance of SAGD WPs on the same pad. Considering the small drainage area of the 7-WP pad (800×800 m), such variances bear important implications for the planning and execution of SAGD projects (particularly greenfield development but also with regard to brownfield expansion). The authors stress that forecasting project productivity should not be based on a single-WP, deterministic simulation model without any consideration for distributions in reservoir geology and parameters affecting SAGD performance. This assessment is equally applicable in the estimation of reserves.
  • Geological features large and small and most likely undetected will affect SAGD performance in ways not easily explainable. Advances in both hardware and software modeling (static as well as dynamic flow) in the last 10 years allow for relatively quick evaluation of sensitivities to geological variations on recovery performance, measure of scoping for uncertainties, and risks for a project. Such work could be conducted in straightforward fashion for many single-WP models for different geological settings; they would take more time for pad (multiple-WP) models, especially if discretized wellbore simulation is used.
  • The finding of constancy and consistency in the gas/oil ratio (GOR) value (of 3.9 m3/m3) for all 28 WPs studied, irrespective of reservoir settings, steaming duration, and most- vs. least-productive WP, was unexpected and fortuitous. There was, however, no clear trend discernible from the gas/liquid ratio data.
  • Applying material-balance consideration to the overall gas phase led to the discovery of extra gas production as a result of aquathermolysis after certain steam-injection periods within the first 15–27 months of first steam for the 28 WPs studied.

Numerical Simulation and Results

The model is calibrated with field data from two central WPs of the worst-performing Pad B (specifically, B4 and B5). There are several reasons for this. First, matching data for WPs draining clean, high-quality reservoirs are relatively straightforward. Also, it is well-known that the top-quality SAGD reservoirs are already exploited; in the future, operators most likely will be forced to make do with much-lower-quality reservoirs. In such a context, learnings derived from analysis and modeling of performance-challenge SAGD WPs are of more value.

Several instances of operational interruption are seen, even within the first year; two interruptions of considerable durations occurred around 730 and 1,095 days. SAGD works best as a continuous process; frequent interruptions in the early stage would affect steam chamber formation and fluid production negatively.

Analysis revealed that performance for the B4 and B5 WPs were quite poor: recovery of less than 140,000 m3, with cumulative steam/oil ratio (CSOR) greater than 3 after 6 years of operation. Their steam-chamber development and growth were challenged by the presence of flow baffles and barriers.

Numerical Reservoir Model. The model is not built or extracted from any stochastic (static) model. The authors believe that a combination of thorough understanding of the process and reasonable interpretation of historical data would compensate for the lack of a stochastic model.

The model is based on simplified shoebox geometry of 800-m length and 100-m width; its thickness allows for the presence of a bottomwater zone and a top interval having different petrophysical (porosity and permeability) fluid-saturation values from those in the main SAGD-able column. A discretized wellbore is used, with injectors and producers being divided into eight segments of 100 m. Both wells have dual-tubing completions.

The model has four fluid components: Water, bitumen, methane, and carbon dioxide (CO2). Monthly steam-injection data for the whole history (including instances of interruptions, as well as the initial bullheading of steam into the injector and producer) of the B5 WP were input into the model. A 50/50 split was assumed for steam injection through the short and long tubing; fluid production only took place through the long tubing in the producer. Constraints on the producer were total liquid rate and minimum flowing bottomhole pressure.

Fig. 1a provides a comparison of simulation results with the actual history of the B5 WP. The simple-geology simulation model was able to replicate, mostly, the general profiles and associated features of cumulative steam injected, cumulative oil and cumulative water produced (on the left axis), and total gas produced (on the right axis). The model matched cumulative water closely but underinjected steam (5%). Good agreement was achieved for cumulative oil produced (within 7%). Additionally, the simulated cumulative (total) gas produced tracked quite closely the data for the B5 WP (end-date agreement within 8%), even though divergences existed between them around 730 and 1,095 days.

Fig. 1—(a) Comparison of simulation results with field data for cumulative steam, oil, water, and gas produced for the B5 WP. (b) Comparison of simulation results with field data for cumulative oil and CSOR for the B5 WP.


In summary, the reservoir model was calibrated successfully with actual injection and production history spanning almost 6 years of SAGD operations for the B5 WP.


  • SAGD project planning should not be based on single-WP deterministic simulation, as had been done with several projects currently in operation.
  • Reported production data showed a constant and consistent trend for the produced GOR of 3.9 m3/m3 for Jackfish 1. This finding enabled material-balance calculation for the gas phase and led to the determination of extra gas production caused by aquathermolysis.
  • Findings regarding gas-production behavior in Jackfish 1 enabled the construction and calibration of a simulation model to compare with the B5 WP. This model incorporated a desorption model to describe the generation of CO2 from aquathermolysis and came within 8% of the reported gas produced for B5 after approximately 6 years of operation. The model is judged to be ready for use in simulation studies of different aspects of SAGD performance.

This article, written by JPT Technology Editor Chris Carpenter, contains highlights of paper SPE 195348, “Steam-Assisted Gravity Drainage Performance Variability—Analysis of Actual Production Data for 28 Athabasca Oil Sands Well Pairs,” by Quang T. Doan, Vincano; S.M. Farouq Ali, SPE, University of Houston; and Thomas B. Tan, SPE, Petrostudies Consultants, prepared for the 2019 SPE Western Regional Meeting, San Jose, California, 23–26 April. The paper has not been peer reviewed.