| Title | Modeling and Forecasting Induced Seismicity in Deep Geothermal Energy Projects |
|---|---|
| Authors | Eszter KIRÁLY, J.Douglas ZECHAR, Valentin GISCHIG, Dimitrios KARVOUNIS, Lukas HEINIGER, Stefan WIEMER |
| Year | 2015 |
| Conference | World Geothermal Congress |
| Keywords | , CSEP testing, deep geothermal systems, forecasting seismicity, Soultz-sous-Forêts, St. Gallen |
| Abstract | Over the past decade deep geothermal projects in Switzerland have demonstrated our limited understanding of induced seismicity: unexpected earthquakes occurred during stimulation periods and led to increases in seismic hazard. Such events may influence public opinion of potential future geothermal projects. Monitoring and controlling induced seismicity with warning systems requires models that are updated as new data arrive and that are cast in probabilistic terms (to represent uncertainties in the physical processes and model parameters). Our goal is to improve the forecasting skill owing to validated physical constraints. As a first step, we seek to answer the question: is it possible to reliably forecast the seismic response of the geothermal site during and after stimulation based on observed seismicity and hydraulic data? To answer this question, we are exploring the predictive performance of various stochastic and hybrid models. The aim is to balance model prediction performance and model complexity: which parameters are necessary to forecast seismicity well, and which increase model complexity but do not give better results? The long-term aim of this research is to develop an on-site decision-making tool for geothermal projects to jointly maximize operational safety and economic output during all project phases. In this preliminary study, we consider four variants of a 3D model that combines seismogenic index calculations with spatial density estimates based on kernel-smoothed seismicity. We apply the model to the Basel 2006-2007, Soultz-sous-Forêts 2004 and St. Gallen 2013 datasets and generate a series of six-hour forecasts. We assess the models using metrics developed by the Collaboratory for the Study of Earthquake Predictability. We check the overall consistency of six-hour forecasts with the observations, comparing the number and spatial distribution of forecast events with the observed induced earthquakes. We also compare models with each other in terms of information gain, allowing a pairwise ranking of the models. Our results indicate that the performance of the model depends on the dataset: the number of forecast events are consistent with the observations in Basel, systematically overestimated for Soultz-sous-Forêts (stimulation of 2004), and underestimated for St. Gallen. Smoothed seismicity models with or without temporal weighting show significantly better results than a model with a spatially uniform distribution in terms of joint log-likelihood, cumulative information gain per earthquake, and average probability gain. In this study, we have not considered the distribution of earthquake magnitudes: each model forecasts only the number of earthquakes above a minimum magnitude. |