Prospective Earthquake Forecasts at the Himalayan Front after the 25 April 2015M 7.8 Gorkha Mainshock

Abstract

When a major earthquake strikes, the resulting devastation can be compounded or even exceeded by the subsequent cascade of triggered seismicity. As the Nepalese recover from the 25 April 2015 shock, knowledge of what comes next is essential. We calculate the redistribution of crustal stresses and implied earthquake probabilities for different periods from daily to 30 years into the future. An initial forecast was completed before a M=7.3 earthquake struck on May 12, 2015 that enables a preliminary assessment; post-forecast seismicity has so far occurred within a zone of 5-fold probability gain. Evaluation of the forecast performance, using 2 months of seismic data, reveals that stress-based approaches present improved skill in higher magnitude triggered seismicity. Our results suggest that considering the total stress field, rather than only the co-seismic one, improves the spatial performance of the model based on the estimation of a wide range of potential triggered faults following a mainshock. 2 IntroductionThe Himalayas rise by absorbing 18-22 mm/yr of Indian-Eurasian plate convergence (Ader et al., 2012) (Fig. 1). On the 25 th of April, 2015, a M=7.8 earthquake ruptured the low-angle (10˚) fault contact between the two plates, leading to more than 8000 confirmed fatalities, 18,000 people injured, and a million families affected (Nepal Red Cross Society 1 ). Eight UNESCO World Heritage sites were damaged or destroyed. The 25 April shock struck the eastern edge of a 500-km-wide gap between historic earthquakes along the Himalayan front (Fig. 1). The potential of triggered earthquakes beneath highly populated basins in the central Himalaya could compound this catastrophe akin to tragedies in Turkey , China , and New Zealand . Recent studies, based on retrospective experiments, investigate the predictive power of short-term earthquake forecasts within different distance ranges and periods (Segou et al., 2013;Strader and Jackson, 2015), but prospective forecasts still face challenges such as the quality of real-time data, the availability of credible historical/modern earthquake catalogs ,and very short research windows.In this study, we calculate the expected redistribution of stress in the Himalayan crust and develop a method, especially applicable to frontier regions, to determine the probability of triggered earthquakes in both space and time. We make prospective earthquake forecasts using four methods incorporating physics-based and statistical approaches for varying time horizons, and formally evaluate the shortest-term calculations.The results reveal the efficiency of physics-based forecasts in estimating short-term earthquake probabilities and their critical contribution in modeling off-fault triggered events, highlighted by the M=5.6 Xegar (3 hours after the mainshock) and the M=7.3 Kodari (May 12) earthquakes. Development of Short-Term Earthquake ForecastsShort-term earthquake forecasts can use empirical statistics to anticipate cascades of triggered events, or simulations of stress redistribution fol...

Improving Physics‐Based Aftershock Forecasts During the 2016–2017 Central Italy Earthquake Cascade

“…Although the focus of our study is on the improvements of CRS models, we acknowledge that other ETAS parametrizations may perform differently. Previous experiments show a good performance of ETAS both in retrospective (Cattania et al, ; Helmstetter et al, ; Marzocchi, Murru, et al, ; Werner et al, ; Woessner et al, ) and during unfolding sequences (e.g., Marzocchi et al, ; Marzocchi & Lombardi, ), but some weaknesses related with early catalog incompleteness (Omi et al, ; Segou & Parsons, ) and the absence of fault interaction effects (e.g., Marzocchi & Lombardi, ) have been reported.…”

“…The receiver planes formulation follows two approaches: the optimally oriented planes (King et al, ) that assumes that earthquakes nucleate on hypothetical faults in favorable orientation with respect to the regional stress field and the geological receiver plane (GRP) approach. Both approaches received extensive criticism in recent years: the optimally oriented planes for relying on the knowledge of the largely unknown regional stress tensor to resolve stress changes on hypothetical planes that might not even exist (Segou & Parsons, ), whereas GRP may miss unmapped faults, when even in well‐studied regions such as California ~30% of seismicity occurs on previously unidentified structures (Field et al, ). Here we adopt the Segou et al () approach where GRP is informed by presequence focal mechanisms and active fault maps to achieve a representation of the spatial structural heterogeneity.…”

“…Nowadays, the implementation of laboratory-derived friction laws describing the seismicity response to an earthquake perturbation (Dieterich, 1994) is the basis of many physics-based forecasts known as Coulomb rate-state (CRS) models (Cattania et al, 2018;Cocco et al, 2010;Parsons et al, 2012Parsons et al, , 2014Segou, 2016;Toda & Enescu, 2011;Toda et al, 2005, among others). Recent advances in CRS modeling demonstrated that addressing the uncertainties behind physics-based models leads to a predictive power comparable to empirical models (Cattania et al, 2018;Segou & Parsons, 2016). However, the limitation for an operational use of CRS models is that their parametrization goes beyond a mere earthquake catalog and requires representations of earthquake sources and nearby receiver faults that are unlikely to be available immediately after a major event.…”

What Controls Variations in Aftershock Productivity?

“…Many of these forecasts are empirical statistic models and, notably, rely heavily on high-quality seismic data both before and during the mainshock (Gerstenberger et al, 2005;Omi et al, 2015). Other models embed physics-based rules that rely on local information about long-term seismicity, the state of stress, and the geometry of the faults (e.g., Segou & Parsons, 2016;Field et al, 2017). All of these approaches raise the critical issue of calibrating hazard in poorly instrumented areas.…”

Machine learning improves forecasts of aftershock locations

“…Until a few years ago, most statistical forecasts of aftershocks were more accurate than were physicsbased forecasts, such as that of the authors. But there are now cases in which physics-based forecasting performs as well as purely statistical approaches 12,13 . The time would seem ripe for methods based on artificial intelligence to enter the fray, and the work of DeVries et al has established this beachhead.…”