Thursday 2 July 2015

Simulating induced seismicity using geomechanics

The best possible words in the life of an academic are undoubtably "paper accepted". Since I've recently had a paper accepted in EPSL. I thought I would add a layman's summary of it here. 

The paper is available here, and is "open access", so you shouldn't need a subscription to read it. 

Our motivation is to try to understand and model why subsurface processes cause induced earthquakes. Induced seismicity has become a controversial issue in relation to fracking, but in fact the risk of inducing an earthquake during fracking are much lower than the risk of inducing an earthquake by other subsurface activities, such as geothermal energy, waste water injection and carbon capture and storage. 

The main reason for this is simply a matter of volume - the more volume you inject, the more likely you are to trigger an earthquake. While much has been made of the water volumes used for fracking, they are actually quite small in the grand scheme of things. The volumes injected for waste-water disposal and for CCS are much larger than the volumes used for fracking. This is why we've seen such increases in seismicity in places like Oklahoma in recent years (it's got very little to do with fracking). 

We've also seen induced seismicity - albeit of small magnitude, less than mag 3 - at two pilot CCS projects, the Decatur project in Illinois, and at the In Salah project in Algeria, which is the subject of our study. 

Firstly, a brief introduction to the In Salah site. It's a gas field in the middle of the Sahara desert. 

Due to natural geological processes, the natural gas that is produced contains a relatively high percentage of CO2. This must be stripped off before the gas can be sold - there are minimum CO2 content requirements. Usually, the CO2 would just be vented to the atmosphere. However, the operators of the site, BP, Statoil and Sonatrach, decided to use the site as a pilot project for CO2. So they instead re-injected the CO2 into the water-leg of the reservoir (part of the reservoir unit that is filled with water rather than gas). The image below shows the basic principles in cartoon form.  

In total nearly 4 million tonnes of CO2 were injected between 2004 - 2011. The average car emits about 4 tonnes of CO2 per year, so that's the equivalent of the annual emissions of 1 million cars.

The site was monitored using a number of methods, but it was clear from relatively early on than the CO2 injection was producing geomechanical deformation. As a result, microseismic monitoring was used to image any small earthquakes. You can read more about the results of the microseismic monitoring here, but the main conclusions were that thousands of small-magnitude (mostly around magnitude 0.0) events had been induced. The largest event was magnitude 1.7, which is probably too small to be felt by humans at the surface (we can detect them with seismometers though of course), and definitely too small to cause damage. Fortunately, all the events were confined to the reservoir unit, so there was no evidence that the seismicity was providing a pathway for CO2 to escape.  

So, what's this latest paper all about?

The basic premise of our study was that induced events occur on pre-existing fractures. They occur because industrial activities change the state of stress in the subsurface, moving a fault from a stable to an unstable state, which allows it to move, triggering an earthquake. So in theory, if we can predict or model where the faults and fractures are, and we can predict or model the changes in stress generated by our activities, we can resolve the stress changes onto the faults, and work out when and where faults might trigger seismicity. The purpose of our paper was to assess how well this approach works in practice. 

To model the size, orientation and positions of faults and fractures I am indebted to my colleague Dr. Clare Bond at Aberdeen, who build a structural model of the reservoir, which simulates how the reservoir geometry we observe today could have formed from the originally-flat sedimentary layers. This produces a strain map, which is then converted into a discrete fracture network to account for how fractures would have accommodated the modelled strain. The resulting fracture map is shown below: you can see that fractures are not uniformly distributed across the reservoir, but there are bands of intense fracturing running through the reservoir, and zones with much fewer fractures. 
In order to simulate the stress changes induced by injection, I am indebted to another colleague, Rob Bissell, from Carbon Fluids Ltd., who built a geomechanical simulation of the injection process. More details about this model are available here. The model provides a map of stress and pore-pressure changes at monthly intervals through the injection period. 

In order to work out whether the modelled stress changes would be sufficient to induce seismicity, for each modelled fracture we resolved the modelled stress from the nearest node of the geomechanical model into normal and shear stresses on the fracture face. If the shear stress exceeded the Mohr-Couloumb criteria, then an event will occur. The size of the event will be determined by the stress drop generated by the event, which will be a function of the shear stress, and the size of the fracture, which is pre-determined in the model provided by Dr Bond. 

Therefore we have a method to simulate when and where an earthquake may occur, and how big it will be. We tested our model simulation results against the microseismic observations made by my colleague Dr. Anna Stork in this paper

The figure below shows that the relative rates of seismicity predicted by the model matches that observed at In Salah. CO2 injection re-starts in late 2009. However, only a small amount of seismicity is observed. Injection rates increase in summer 2010, and for 4 months the rate of induced seismicity also increases. Once injection rates are reduced, the number of events decays away as well. This behaviour is well captured by our model. 

In terms of magnitudes, our modelled largest event matched very well the observed largest magnitude of M=1.7. Magnitudes are determined by the size of the fault and the stress drop, so this indicates that Dr Bond's model did a good job of simulating the fault/fracture sizes, and that Rob Bissell's model did a good job of simulating the stress changes induced by injection. 

Overall, our model does a good job of simulating induced events at In Salah, which is encouraging in terms of our future ability to mitigate induced seismicity at future projects. We have outlined a workflow that can be followed at sensitive sites where induced seismicity may be an issue. For example, the modelling approach can be used to assess whether alternative injection strategies may lower the risk of inducing an event.