Abstract
Murphy et al. present a methodology to provide updated and improved estimates of the upper tail of induced earthquake magnitude distributions for the Groningen gas field, the Netherlands. In particular, they propose regression models to characterize the evolving nature of both the magnitude of completion (which corresponds to the threshold in the classical peaks over threshold extreme value model) and the exceedance magnitude. These regression approaches are routinely used to analyze environmental extremes. Our discussion focuses on some aspects of their suitability and applicability to investigate changes in extremes. We consider the risk of confounding physical and observational non‐stationarity, the difficulty of attributing changes in high quantiles to the threshold model or the exceedance model when both vary, and the role of threshold stability in estimating and interpreting the upper endpoint of the Generalized Pareto distribution.