LONG-TERM EARTHQUAKE FORECASTING
There is no early guest more unwanted than a large earthquake. While guests that arrive too early for a dinner party are merely a nuisance, the recent rash of unexpected large earthquakes represents serious hazards to people in affected areas, and these temblors have also caused vexing scientific problems for those seismologists who try to anticipate the arrival of large earthquakes.
After a relatively quiet period of nearly 15 years, large earthquakes have returned to the global stage in the last few years. As we now realize, the decades of the 1950's, 1960's, and 1970's were times of numerous great earthquakes throughout the world, with a considerable fraction of the world's plate boundaries rupturing in great earthquakes. This era included the three "truly great" earthquakes (Mw of 9 or larger) of the 20th century, in addition to many other of the largest events of our century. We are fortunate that such giant events have not occurred in recent decades, though every year brings us unfortunate reminders that small earthquakes in the wrong place can still cause terrible disasters. Nevertheless, here I am concerned with the largest earthquakes that have occurred, and therefore I shall be mostly concerned with plate interface events in subduction zones.
Seismologists thought that we had made progress in the long-term forecasting of large plate boundary earthquakes. Long-term forecasting, based on the seismic gap hypothesis and its successors, had a surprising string of successful forecasts on the intermediate time-scale in the decade of the 1970's. But starting in 1980, several events have occurred that violate the well known seismic potential map published by McCann et al. in 1979 (PAGEOPH 117, pp. 1,082-1,147, 1979). We shall briefly review the basis of long-term forecasting, then return to the recent violations.
Plate tectonics provides the underlying framework for deterministic long-term forecasting. If a plate boundary segment breaks as a large earthquake, then the recurrence time to the next "characteristic" large earthquake for that segment should be the time required to build up the stress due to steady tectonic motions. Global surveys showed that recurrence times between characteristic large earthquakes are variable, but most of them fall in the range of between several decades and one or two hundred years. The early seismic potential maps used the same recurrence time for all subduction zone segments, e.g., the map of McCann et al. in 1979. Subsequent work on recurrence times showed that different subduction zones have recurrence times that systematically differ from a global average. For example, the Mexico subduction zone has shorter recurrence times, just a few decades, and the "characteristic" large earthquakes are smaller than those in the Chile subduction zone. Thus, more refined estimates of future earthquake occurrence are obtained by characterizing the regional variations in recurrence time, and ultimately using a different recurrence time for each segment of the plate boundary. In detail, we can even allow the recurrence time to vary within each plate boundary segment; observed sequences would seem to demand this variability. The most recent global map of long-term forecasting (Nishenko, PAGEOPH 135, pp. 169-259, 1991) combines deterministic and statistical estimates of recurrence times to provide probabilities of "characteristic" large earthquake occurrence over the 1989-1999 ten-year period for a total of 96 plate boundary segments.
To focus on one ambiguous element in this forecasting methodology: Exactly what is the "characteristic" large earthquake for a segment? The ideal case is where a plate boundary segment reruptures with an event that has exactly the same fault length and magnitude. If a segment reruptures with a smaller earthquake, how do we decide when it qualifies as the "characteristic" large event for that cycle? Associated with the above ambiguity is the fact that subduction zones can change their "rupture mode" from one cycle to the next. The classic example is the Ecuador-Colombia subduction zone, where the entire zone ruptured in a great event in 1906 with Mw of 8.8, and has since reruptured with a sequence of three events in 1942 (Ms 7.9), 1958 (Ms 7.8), and 1979 (Mw 8.2). The 1906 rupture zone clearly consists of three segments that can rupture separately. Obviously, our definition of "characteristic" earthquakes must be flexible enough to accommodate the possibility of a switch in rupture mode from one earthquake cycle to the next.
Let us now return to the recent violations of long-term forecasting. The famous map of McCann et al. (1979) used a simple color code to portray earthquake potential: "Red zones" are plate boundary segments where large earthquakes are expected, and "green zones" are segments where we should be safe from large earthquakes for a few decades. In a controversial test of the McCann et al. (1979) map, Kagan and Jackson (J. Geophys. Res. 96, pp. 21,419-21,431, 1991) argued that the overall tendency of large earthquake occurrence in the 1980's was to occur outside the "red zones", not in them. On the other hand, it must be remembered that the 1980's was a decade of small earthquakes compared to the previous decades. It seems that large earthquakes have returned in the 1990's, but it is odd that the largest earthquake thus far (the 4 October 1994 Kurile event, Mw 8.3) was an intraplate event within the subducted slab, just beneath the seismogenic interface. We have also seen large events occur in the previously quiet Marianas subduction zone. Furthermore, other recent large events have either been intraplate events (the 9 June 1994 Bolivia deep event, Mw 8.2), or back-arc thrust events (the 12 July 1993 Hokkaido, Mw 7.8; 12 December 1992 Flores Island, Mw 7.9). Hence, there are several unusual, large events in the 1990's. But even if we focus on plate boundary earthquakes in mature subduction zones, there are still some serious problems with the recent large events.
To avoid long discussions on whether events are "characteristic" earthquakes or not, let us agree that any earthquake with magnitude larger than 7.5 poses serious hazards, and thus we should strive to anticipate their arrival. Clearly, a serious violation of earthquake forecasting occurs when a large earthquake arrives too soon. This happens when the recurrence time for the next event in a plate boundary segment is much shorter than the forecast recurrence time. In particular, a large interplate earthquake that occurs in a "green zone" of the McCann et al. (1979) map is a dangerous early "guest," and we now have four examples of this bad behavior. Now, let us consider the four violations one-by-one:
What to do about the early arrival of the above events? Do these events signal the end to any kind of deterministic long-term forecasting? No. On the contrary, I think that once we accept the fact that rupture mode can change from one earthquake cycle to the next, then a pattern emerges from these early-arriving events. The simplest evidence for a change in rupture mode is that in all four cases the magnitude of the recent event is less than that of the previous large earthquake. In detail, the epicentral "segment" of the preceding large event reruptures with a recurrence time less than the previous time. While there are likely to be several acceptable explanations for this observed pattern amongst these early-arriving events, I suggest a simple one here: rapid "healing" of the epicentral segment during great earthquake rupture causes it to be reloaded to a high stress level due to slip in adjacent segments. The key component to this idea is that during the great earthquake, the epicentral fault segment is able to "heal" and stop slipping within ten seconds, thus slip in adjacent segments elastically reloads this epicentral segment. These high residual stress levels in the epicentral segment explain why the next recurrence time is anomalously short and why only the epicentral segment reruptures. Thus, these early arriving earthquakes tell us that the rupture mode has changed for all these subduction zones, and that we should prepare for other early arriving "guests."
If we accept this explanation that recent large events arrived too soon due to a change in rupture mode from one great event to a few large events, then we have the appearance of a global conspiracy to change rupture modes in the last decade. There are two views on this apparent conspiracy: One, it is just a coincidence, which is allowed by the small number of cases; two, it is not a coincidence; indeed, we should have expected this change given the frequent great earthquakes in the 1950's, 1960's, and 1970's. To complete our somewhat strained analogy, we can view our early arriving "guests" of the 1990's as the large "consequences" that soon follow the great seismic "parties" that occurred in the decades of the 1950's through 1970's.
Larry J. Ruff
To send a letter to the editor regarding this opinion or to write your own opinion, contact Editor John Ebel by email or telephone him at (617) 552-8300.
Posted: 11 February 1999