Predicting ground motion from induced earthquakes in geothermal areas

Induced seismicity from anthropogenic sources can be a significant nuisance to a local
population and, in extreme cases, can lead to damage to vulnerable structures. One
type of induced seismicity that is of particular recent concern, and which in some cases
is a limit on development of a potentially-important clean energy source, is that
associated with geothermal power production. A key requirement for the accurate
assessment of seismic hazard (and eventually risk) is a ground-motion prediction
equation (GMPE), predicting the level of earthquake shaking (in terms of, for example,
peak ground acceleration) given an earthquake of a certain magnitude at a particular
distance. For geothermal-related seismicity, practically no such models currently exist
and consequently the evaluation of seismic hazard in the vicinity of geothermal power
plants is associated with high uncertainty.
Various ground-motion datasets (from Basel, Geysers, Hengill, Roswinkel, Soultz and
Voerendaal) were compiled and processed and the moment magnitudes for all events
recomputed homogeneously. These data are used to statistically test the similarity
between ground motions from induced and natural earthquakes. In addition, the ability
to extrapolate recent GMPEs derived for natural seismicity to make them applicable for
ground-motion prediction close to the source of (generally small) induced shocks is
examined. Subsequently stochastic models are developed based on a single corner
frequency and with parameters constrained by the available data. Predicted ground
motions from these models are fitted with functional forms to obtain easy-to-use
GMPEs that account for epistemic uncertainties as well as being associated with
standard deviations characterizing the aleatory variability. As an example, we
demonstrate the potential use of these models using data from Campi Flegrei. In
addition, we study correlations between instrumental ground-motion parameters and
macroseismic intensities and recommend that intensities are predicted directly rather
than through such correlations, which underpredict observed intensities of induced
events.