Pressure/flow modeling and induced seismicity resulting from two decades of high-pressure deep-well brine injection, Paradox Valley, Colorado

The Bureau of Reclamation operates a deep injection well near Paradox Valley, Colorado. Intermittent injection testing began in 1991, followed by near-continuous injection since 1996. Daily average injection flow rates and surface injection pressures have been recorded since 1991, and seismicity has been monitored since 1985. Before the injection flow rate was decreased in 2013 in response to the largest induced earthquake to date, observed wellhead pressures were increasing and approaching the maximum permitted pressure. Potential solutions were needed to provide long-term reductions in wellhead pressures and to minimize the occurrence of future large-magnitude induced earthquakes. A key step in this process is to establish whether the trend of increasing pressures is more likely caused by far-field reservoir pressurization or near-well flow impairment. The spatiotemporal occurrence of induced seismicity is fit relatively well by a 1D diffusive triggering front relationship, suggesting that the reservoir can be modeled as a porous medium. Simple 1D, uncoupled porous models are therefore used to analyze the wellhead pressure response for the period of long-term injection. These simple models are found to provide a reasonable fit to the pressure/flow data. We did not observe the expected changes in the model parameters that would indicate significant near-well flow impairment, and thus we have concluded that the observed pressure increase is more likely related to far-field pressurization. Somewhat surprisingly, given the complex geologic structure and the heterogeneity in the locations of induced seismicity, the analysis demonstrates the ability of a simple radially symmetric porous model to fit the pressure/flow data in the near-well area (within approximately 2 km). Although the induced seismicity data suggest significant heterogeneity, the daily average wellhead pressure/flow data are insensitive to these features. In the absence of additional constraints, use of more complicated models is unlikely to produce substantial additional benefit for modeling in the near-well area.