Generating ElarmS AlertMaps

Figure 2.33: AlertMap frame for the 20 July 2007 $M_{w}$ 4.2 Oakland earthquake, at 4 seconds after event origin. Color scale represents predicted MMI according to the scale at bottom. Circular contours represent time until onset of peak ground motion in seconds, based on a moveout of 3.75 km/s. Triangles represent strong-motion accelerometer stations, inverted triangles represent broadband velocity stations, and diamonds represent collocated installations. Larger grey stations have reported triggers at this time.

Since February of 2006 we have been automatically executing the ElarmS algorithm for every event of $M \geq 3$ in northern California (Wurman et al., 2007). As this processing is performed without human input, we refer to the procedure as Non-Interactive (NI) processing. We currently produce AlertMap output as part of the NI processing, allowing us to evaluate the performance of ElarmS for every event, a task that was much more difficult before the development of the AlertMaps.

In the NI processing procedure, ElarmS produces plain-text output once per second for 60 seconds beginning at the event origin time. The output contains all the data available to ElarmS in that second, including peak ground motion observations for each station, as well as station triggers and associated parameters such as maximum predominant period ( $\tau_{p}^{max}$) and peak velocity or displacement ($P_{d/v}$) for that station and channel. The output also includes the event information (location, origin time, magnitude and other metadata) for each event associated from the individual station data.

The AlertMap output is entirely derived from this plain-text timeslice data, which is converted to AlertMap output via a series of perl scripts. The first of these digests the ElarmS text output at each second and converts it to XML files formatted for use by ShakeMap (Wald et al., 2005). A ShakeMap subroutine then executes for each second of output, producing a raster of ground motion predictions on a regular grid of approximately 2.5$^{\circ}$lat by 4.2$^{\circ}$lon around the estimated epicenter of the event. In addition, ShakeMap produces an XML file of predicted ground motion at just under 400 broadband stations of the NCSN and BDSN networks. All these ground motion predictions are based on the event location and magnitude, as well as any observations of actual peak ground motion available in each second. Using present computing resources, this step takes approximately 10 seconds to produce a one second timeslice. Reducing this processing time is a matter both of dedicating more computing resources to the problem, and of simplifying the ShakeMap procedure and optimizing it to run in the minimum possible time.

When the ground motion grid files are available for all 60 seconds (less than 60 seconds if the event was not detected immediately) a script generates a Postscript map for each second, representing all the data and event parameters available in that second. Sixty of these maps are combined to produce an animation which presents the performance of ElarmS in an intuitive geospatial and temporal manner.

Figure 2.34: AlertMap for the Oakland event at 8 seconds after origin time. Black stations are those currently experiencing peak ground motions, and colored stations have reported peak ground motions according to the scale at bottom.

Figure 2.35: ShakeMap produced by USGS for the Oakland event. Color scale represents observations of MMI at stations (triangles) and is interpolated for all points between stations.