Predicting catastrophic events by assimilating geoscientific data
Data assimilation is a highly sophisticated process for incorporating observed data into a reference model in order to make predictions. For example, you probably already use the weather predictions from such a data assimilation system to make decisions about your schedule or to plan your next vacation. Similarly in geosciences, being able to simulate the evolution of soils or rock using a model is a considerable advantage with direct applications such as the prediction of reservoir behavior, the updating of hydraulic properties or the distribution of geomechanical properties.
But what exactly is a model? A model is simply a representation of reality by a set of parameters that describe a phenomenon of interest. In addition, a model is said to be dynamic when its parameters change over time. For example, an aquifer could be represented by a simple 2-layer initial model with 2 distinct resistivity values (t0 diagram). The injection of water at the surface would subsequently make this model dynamic, while the resistivity of the medium would evolve over time with the infiltration of water into the ground as illustrated on images t1 to t4.
Another concrete, but more complex example of a dynamic environment is that of a mine which evolves as its mining activity progresses. The extraction of minerals involves the construction of galleries, which generates voids within the rock and modifies the properties and internal stresses of the rock mass. The return to equilibrium of the bedrock is often associated with a phenomenon called "rockburst", which corresponds to a spontaneous and violent rupture of the rock that can lead to potentially serious repercussions, such as the cessation of operations, the breakage of mining equipment or significant risks to the safety of miners. As a result, it becomes extremely important to monitor mining activity and the best way to do this is through microseismic monitoring. Although mitigation techniques related to engineering and rock mechanics are generally implemented in mines, these techniques are not perfect and unforeseen events can still occur.
As part of the microseismic studies, the seismic waves generated by normal activity in the mine and the blasting activities required for the development of galleries are recorded by the microseismic monitoring systems in place.
Using an initial medium velocity model, it is possible to identify the locations where the different seismic events were measured. As mining progresses, seismic velocities vary and the initial velocity model becomes inaccurate. The microseismic travel times measured and assimilated during the exploitation work then make it possible to gradually update the velocity model of the area of interest.
The analysis of the modifications made to the initial velocity model has a very interesting potential to predict the future behavior of the rock mass. In concrete terms, this monitoring method consists of identifying the locations where the changes in velocity occur in order to locate the hypocentres and highlight the sectors where the rock mass is under stress and where a potentially dangerous event is more likely to occur. occur. This analysis tool can therefore give mine operators an additional tool to identify areas that could contain risks to workers and equipment, before a disaster occurs.
A demonstration of microseismic monitoring is presented in the article by Dip et al. (2021) from the journal Near Surface Geophysics. In this paper, it is proposed to use the Ensemble Kalman Filter to update a 3D seismic velocity model by measuring the effect of changing rock mass stresses, which translates into velocity changes of seismic waves.
The results of this article are shown in the animation above. On this animation, the evolution of seismic velocities obtained following the update of the dynamic model is illustrated by the color scale changing from blue to red. These seismic velocities represent areas where the internal stresses of the rock mass were the highest. The location where the rockburst actually occurred is represented by the purple circle, which coincides fairly closely with the area containing the greatest concentration of velocity change in the rock mass. The event occurs at time t6 of the animation, when the model update predicted the largest speed increase, followed by stress relief. In a real situation for this case study, if the predictive model had been in place at the time of the catastrophic event, the latter could have been signaled four hours before its occurrence.
Dip, A. C., Giroux, B., & Gloaguen, E. (2021). Microseismic monitoring of rockbursts with the ensemble Kalman filter. Near Surface Geophysics, 19(4), 429-445.