8:45 AM - 9:00 AM
[S13-1-02] Uncertainties in moment tensor estimation for induced earthquakes illustrated at the example of the Groningen gas field, The Netherlands
The Groningen gas field is one of the largest gas fields in the world, in production since 1963. Since the early 1990's, induced seismicity started to occur in its vicinity. Approximately 220 earthquakes with M ≥ 1.5 were registered from 1991 to July 2013 in the vicinity of the Groningen field area. The strongest event recorded so far (ML = 3.6) occurred on 17th August 2012 near the village of Huizinge, raising huge damage claims as well as public concerns.
Starting in 1991, a geophone monitoring network operated by KNMI was installed. This network has been extended several times, providing a detection threshold of M=1.5 since 1995. We will use a newly developed python-based probabilistic moment tensor inversion tool named“Grond" to re-analyse four events that occurred from 2006 to 2009, for which focal mechanisms have been computed earlier (Kraaijpoel and Dost, 2012).
Each of the four events has been recorded on up to 6 accelerometers at distances of less than 10 km and on up to 8 shallow borehole strings at distances of 15 to 90 km. Due to the inherent capabilities of probabilistic methods, we are able to analyse trade-offs between inversion parameters and to assess uncertainties. We illustrate the influence of parameters of the forward modelling and the choice of the objective function on the resulting moment tensor.
“Grond" adapts an optimization scheme using elements of simulated annealing and the bootstrap technique. Rather than optimizing a single objective function, it tries to find volumes in parameter space that satisfy low misfit values over an ensemble of N variations of the objective function. In every optimization method, there is a trade-off between number of tested models and ability to find the global minimum without getting trapped in local minima. Usually a compromise is made depending on the characteristics of the problem to be solved. In our optimization, however, the degree to which the parameter space is explored can be tuned by the user.
Starting in 1991, a geophone monitoring network operated by KNMI was installed. This network has been extended several times, providing a detection threshold of M=1.5 since 1995. We will use a newly developed python-based probabilistic moment tensor inversion tool named“Grond" to re-analyse four events that occurred from 2006 to 2009, for which focal mechanisms have been computed earlier (Kraaijpoel and Dost, 2012).
Each of the four events has been recorded on up to 6 accelerometers at distances of less than 10 km and on up to 8 shallow borehole strings at distances of 15 to 90 km. Due to the inherent capabilities of probabilistic methods, we are able to analyse trade-offs between inversion parameters and to assess uncertainties. We illustrate the influence of parameters of the forward modelling and the choice of the objective function on the resulting moment tensor.
“Grond" adapts an optimization scheme using elements of simulated annealing and the bootstrap technique. Rather than optimizing a single objective function, it tries to find volumes in parameter space that satisfy low misfit values over an ensemble of N variations of the objective function. In every optimization method, there is a trade-off between number of tested models and ability to find the global minimum without getting trapped in local minima. Usually a compromise is made depending on the characteristics of the problem to be solved. In our optimization, however, the degree to which the parameter space is explored can be tuned by the user.