Comment on “Earthquake magnitude estimation from peak amplitudes of very early seismic signals on strong motion records” by Aldo Zollo, Maria Lancieri, and Stefan Nielsen

Abstract

[1] An early warning system should determine the location and magnitude of an earthquake as rapidly as possible in order to broadcast an alarm to regions that will undergo severe ground shaking. It was recently claimed by Zollo et al. [2006] that earthquake size could be determined from only the first 2-seconds of P- or S-wave strong-motion data; this represents a fraction of the rupture time for larger M > 7 events. Using this relatively short amount of data, the method of analysis was to find the peak ground displacement (PGD), which was reported to scale with earthquake magnitude; such rapid information would play an important and much needed role in an earthquake early warning (EEW) system. Here we perform a similar analysis on strong-motion data from the KiK-net and K-NET arrays in Japan and find no compelling evidence that the peak ground displacement during the first couple of seconds of P-wave is related to the eventual size of a large earthquake. [2] KiK-net is a network of accelerometers that are co-located with the Hi-net seismic network in Japan [Horiuchi et al., 2005], all of which were installed after the 1995 Kobe earthquake. KiK-net consists of ∼700 stations with sensors set in 100 m deep boreholes and have a range of ±2000 gals with a sensitivity of 0.23 mgal/count, a flat response up to 20 Hz and are recorded using 24 bit a/d converters with a sample rate of 200 Hz. KiK-net is a triggered system with about 15 seconds of pre-trigger waveform data available for each event. [3] The Kyoshin network (K-NET) is another strong-motion array deployed throughout Japan. K-NET consists of about 1000 stations of tri-axial accelerometers with an interstation spacing of about 25 km. The K-NET sensors are similar to KiK-net but they are installed free field and have a somewhat lower recording precision (18 bit a/d) and a sample rate of 100 Hz. The K-NET database predates the KiK-net data by several years and it is included here to increase the number of records in our analysis for larger magnitude earthquakes, which are the main concern of any EEW system. Starting in 2004, the newer K-NET sensors have a range of ±4000 gals with a sensitivity of 0.63 mgal/count and 24 bit resolution. [4] The P-wave data from Hi-net is basic to Japan’s EEW system that can locate earthquakes in real-time [Horiuchi et al., 2005; Rydelek and Pujol, 2004]. Therefore, the further ability to estimate an earthquake's size (M > ∼6) from just a couple of seconds of P-wave displacement data from the K- and Kik-net arrays would be an invaluable addition to this EEW since then both earthquake size and location could be quickly estimated. [5] We analyzed K- and KiK-net data from 1363 earthquakes M ≥ 3.5 as determined by the Japanese Meteorological Agency [Tsuboi, 1954]. The records from these two networks are treated independently in the following analysis, which is comparable to that of ZLN except with differences specific to the seismotectonics of Japan and the strong-motion arrays. Instead of limiting the maximum hypocentral distance to 50 km, we used a distance cutoff that depended on the size of the earthquake: 50 km for events M ≤ 5.5 with a linear increase to 75 km for larger events (see Figure 1). The low-background noise of the KiK-net stations allowed for an accurate determination of the P-arrival at distances greater than 50 km for the larger events; indeed, we only used those events for which the P-arrival was clearly identified by our automated P-picker [Horiuchi et al., 1992]. In Japan, 75 km is below the distance at which other seismic phases (e.g. Pn) would arrive before the direct P-wave. As done by ZLN, we then used 2 seconds of the vertical component of calibrated P-wave data after the initial arrival and determined the PGD in this relatively short time window. In addition, travel times of P- and S-waves were estimated from the average velocity profile in Japan and based on these times we used only hypocenters >16 km to avoid contamination by the S-wave in this 2 second window. However, unlike ZLN where both the P- and S-wave data were used, the results here are only for P-waves since we believe that an effective EEW would be best served with rapid analysis of the P-wave seismic energy, i.e. before the arrival of the destructive S-waves. [6] The histograms in Figure 1 are the number of usable records as functions of hypocentral distance and earthquake magnitude. As expected, a drop occurs for small events at the 50 km limit but as this limit is increased we start to include the important records from the larger events since the P-waves for these events are well recorded. The significantly larger dataset used here affords an adequate sampling of earthquakes at all magnitudes; therefore we did not average over magnitude bins as done by ZLN. [8] A plot of the data used in this equation is shown in Figure 2, which also shows the linear fit for the station-average values for earthquakes M ≤ 5.5. The fit to the smaller events is excellent and a clear trend with a very high correlation coefficient (R = 0.99) is observed but the data for the larger events departs significantly from this trend and has large scatter. Therefore, we believe that fitting a single line to all the data in Figure 2 is unjustified. The data is better modeled by a linear relation between the logarithm of PGD and magnitude up to M ∼ 5.5 and then a mostly flat line above this magnitude. Below M5.5, however, a couple of seconds of P-wave data would contain nearly the entire history of fault rupture and a scaling relation, as observed, is consistent with conventional seismic source theory. Our analysis of the K- and KiK-net data reveals the lack of a significant relation between PGD and magnitude for events larger than about magnitude 5.5; therefore we cannot support the claim that eventual size of large earthquakes is predictable from just 2 seconds of the initial P-wave data. We also note that the large scatter at all magnitudes would raise fundamental concerns regarding the usefulness of this prediction method, if possible, in a real-time EEW system. Using a much smaller dataset, Wu and Zhao [2006] suggested that the PGD measured from three seconds of P-wave data can be used to predict the magnitude up to M6.5 and noticed that a larger earthquake (M6.7) seemed to violate this relationship. [9] Other investigations [Allen and Kanamori, 2003; Olson and Allen, 2005] of the seismogram from the initial portion of rupture have resulted in the claim that the size of an earthquake may be predicted. Unlike the present study, the method of analysis in these investigations was to determine a velocity spectral peak in the first several seconds of P-wave seismic data, which was reported to scale with earthquake magnitude. We performed a similar method of analysis on a larger set of high-quality waveform data from the Japanese Hi-net array [Rydelek and Horiuchi, 2006] and could not support the claim that earthquake rupture is a deterministic process based on our analysis. [10] It should be noted that the interpretation of geological data does not lend support to the claim that fault rupture is deterministic. Wesnousky [Wesnousky, 2006] studied surface traces of historical earthquakes and concluded that variations in rupture lengths (>15 km) are apparently controlled by fault complexity and the preexisting stress field from past earthquakes and not necessarily by the initial slip pulse or stress drop at the onset of rupture. [11] It would be very beneficial if the magnitude of large earthquakes could be predicted from just several seconds of P-wave data since a reliable real-time warning could be broadcast before the arrival of the highly destructive S-waves. The results of our analysis of the KiK-net and K-NET data indicate that the (eventual) sizes of larger earthquakes are difficult to estimate from this short amount of data. It may prove useful, however, to issue a warning that a larger earthquake has occurred if the ground displacement in this initial p-wave data exceeds some pre-established threshold criteria. [12] PAR thanks the Humboldt Foundation for support.