Near real-time updating of pager loss estimates

Initial alerts by PAGER (Prompt Assessment of Global Earthquakes for Response) within minutes following an earthquake include several uncertainties, mainly due to potential inaccuracies in location, depth, fault delineation, and shaking estimates. We enhance an updating framework by incorporating early reports of fatalities within the first 24 hours, or so, of an earthquake to update PAGER’s overall fatality estimates and its resulting alert level. Though initial loss reports by officials or the media are uncertain and often undercount the eventual reported impacts, their temporal evolution provides predictive constraints for the PAGER model. The proposed framework helps capture these in a systematic way to minimize potential large fluctuations in PAGER alerts as ShakeMap (the USGS product which estimates how an area is affected by an earthquake) gets updated in the early hours after an earthquake. The new framework also accounts for uncertainties associated with early fatality reports as well as PAGER model-related uncertainties in order to improve the overall impact forecast. This updating framework improves the loss estimate and alert level to the correct level within the first 24 hours even when the initial estimation from PAGER is assumed to be off by two levels of alert, which is plausible due to potential over- or under-estimation of the PAGER model. While test results are very encouraging, our future work aims at implementation of operational PAGER model updating, which entails additional challenges in acquiring useful data, estimating their credibility, and developing rigorously tested operational code and protocols