The two purposes of this work are to better understand the correlates of reservoir-induced seismicity (RIS) and to develop a model for predicting RIS from reservoir characteristics. Data from 29 reservoirs associated with RIS and 205 reservoirs not associated are analyzed using statistical discriminant analysis. These analyses show significant correlations between RIS and reservoir depth and between RIS and reservoir volume, but lesser correlations of RIS with stress and with geology. Data on fault activity are too few to allow inferences of correlation.
A prediction model based on discriminant analysis and calibrated by the data provides a rough estimate of the probability of RIS for specific sites. This model does not allow precise estimates, but it does distinguish between probabilities of RIS in the range of, say 30 per cent from those of 5 per cent. The base-rate frequency of RIS, knowing nothing more than that reservoir depth exceeds 92 m (deep and very deep reservoirs), is about 14 per cent. Given the most favorable set of reservoir attributes, the present model would reduce this probability to about 3 per cent. Given the least favorable, the model would increase this probability to almost 70 per cent. The standard deviation of the larger estimates is on the order of 12 per cent (absolute).