ABSTRACT Earthquakes induced by injecting or extracting gas from underground reservoirs can pose a significant hazard to surrounding infrastructure and populations. Safeguarding against future seismic hazards requires accurate models for the upper tail of the earthquake magnitude distribution that are able to represent various intervention strategies. For these models, we need efficient inference methods and reliable estimates of uncertainty. We propose a novel extreme value modelling procedure, which uses known changes in the earthquake measurement network, to automatically select a parametric spatio‐temporal extreme value threshold which accounts for undetected earthquake values. We introduce methods to propagate the uncertainties in the extreme value model parameters, the threshold parameters and the threshold functional formulation through to future hazard estimates. We apply our methodology to the earthquake catalogue from the Groningen gas field in the Netherlands, delivering clear improvements over existing analyses and providing the first quantification of the different sources of uncertainty in such estimates. The procedure has the potential to be useful for a broad range of extreme value contexts to account for threshold uncertainty when parametric threshold models are used, or where data are missing due to limitations in measurement equipment.
Murphy et al. (Wed,) studied this question.