Dynamic forecasting of the next largest earthquake during hydraulic fracturing stimulations

Abstract

Most traffic-light systems (TLS) related hydraulic fracturing stimulations still adopt the maximum observed magnitude as the decisive metric to aid decision making by stakeholders. However, waiting for the red-light magnitude to be observed is not a proactive stance, especially given that jumps of up to two magnitude units are evidently common enough between events. Clearly there is a need to actively forecast rather than to passively record the size of the next largest earthquake (NLE). In this study, we demonstrate that we can do just that using an ensemble of 6 existing models from the literature designed with similar purposes in mind (Shapiro et al. 2013; McGarr 2014; Mendecki 2016; van der Elst et al. 2016; Galis et al. 2017; Cao et al. 2020). Following a logic-tree approach, these 6 models are calibrated and dynamically weighted in near real-time using as sole inputs the initial parts of the earthquake catalogs and the reported injection rates. The proposed forecasting tool is tested against 18 past stimulations from 9 different Enhanced Geothermal Systems around the world (Helsinki, Basel, Soultz, Cooper Basin, Basel, Pohang, FORGE, Paralana, Newberry). Overall, the results indicate a consistent (across sites and time) and accurate estimation of the next largest magnitude with a tight uncertainty range (1σ) of less than 0.5 magnitude units. Our proposed framework underestimated the next largest magnitude only in one occasion (out of the 18 stimulations), while reliably maintaining a tight safety margin of less than 1 magnitude unit. We recommend that the forecasted NLE replaces the largest observed magnitude as the default metric adopted by future TLS governing any type of fluid-injection operation.