Faster Than Real Time Tsunami Warning with Associated Hazard Uncertainties

Tsunamis are unpredictable events and catastrophic in their potential for destruction of human lives economy. The unpredictability occurrence poses a challenge to the tsunami community, as it is difficult obtain from tsunamigenic records estimates recurrence rates severity. Accurate efficient mathematical/computational modeling thus called upon provide forecasts hazard assessments. Compounding this warning centres physical nature tsunamis, which can travel at extremely high speeds open ocean or be generated close shoreline. Thus, must not only accurate but also delivered under severe time constraints. In immediate aftermath earthquake event, there uncertainties source such location, rupture geometry, depth, magnitude. Ideally, these should represented warning. However practice, quantifying intensity ( i.e., maximum amplitude) due feasible, since requires large number resolution simulations. We approximate functionally complex computationally expensive simulations with simple cheap statistical emulator. A workflow integrating entire chain components quantification developed here - sources, simulation scenarios using GPU version Volna-OP2 on non-uniform mesh an ensemble construction emulator training data, prediction intensities associated massively parallelized finite volume code heart workflow, we use emulation compute locations interest. Such integration balances trade-off between desired accuracy uncertainties, within given fully generic independent (1945 Makran earthquake) studied here.