Like earthquakes, an underground nuclear explosion creates seismic signals that travel through the Earth. For use as a monitoring tool, a seismic monitoring network must be able to both detect and identify the source of seismic signals. Detection consists of recognizing that a seismic event has occurred and locating the source of the seismic signals. Identification involves determining whether the source was a nuclear explosion, as distinct from an earthquake or a chemical explosion used in industries such as mining. In addition, the monitoring system must be able to perform these tasks despite any plausible attempts to evade the monitoring system.
If Earth motions were caused only by small earthquakes and underground explosions, then the recording of such events would be easy. In principle, seismometers are sensitive enough to detect even the smallest explosion.4 However, processes such as winds, ocean waves, and even automobile traffic, generate motion which seismometers faithfully record. It is the naturally occurring background noise level, not the inherent sensitivity of the instruments, that is usually the limiting factor in detecting underground nuclear explosions.5 The extent to which a seismic network can distinguish earthquakes and explosions from background noise depends on several factors including the types of seismic station used, the number and distribution of stations, the local background noise, and the efficiency with which seismic waves travel through the region.
In practical terms, detecting a seismic event means more than observing a signal above the noise level at one station. There must be enough observations to estimate the location of the event that created the detected signal. This task requires determining four numbers: latitude, longitude, depth, and origin time. All of these numbers have some uncertainty associated with them. Seismic waves travel from their source (the explosion or earthquake) to individual recording stations along different pathways through the Earth. Consequently, no two signals recorded at separate stations will be identical, and calculations of location never will have perfect precision. Determining the location of a seismic event is an iterative process - one compares the calculated arrival times of the various seismic waves with their observed arrival times. By making such comparisons at several stations, we can determine the most probable location along with an ellipse within which lies the true location at some level of confidence. Typically, the ellipse is described in such a way that there is a 95% probability that the seismic event occurred within its area.
Once a seismic event has been detected, the next task is to determine if it was created by an underground nuclear explosion. Other seismic events include natural earthquakes, rockbursts in mines, and chemical explosions conducted for mining, quarry blasting, and construction. On a global basis, over 90% of all seismic events can be identified as earthquakes simply because they are too deep or not in a plausible location for a nuclear explosion.6 For seismic events that cannot be distinguished by depth and location, other methods of discrimination are used. The other methods are based on physical differences between earthquakes and explosions.
Explosions happen instantaneously, in one spot, sending out seismic waves of approximately the same strength in all directions. Earthquakes typically occur when rocks slide against each other over a relatively wide area for a longer time, and, as a consequence, will send out different seismic signals in different directions.7 The differences between these processes often can be seen in the observed seismic waves and used to distinguish earthquakes from explosions.
As the seismic signal gets smaller, nuclear explosions must be distinguished not only from earthquakes, but also from other kinds of explosions. Industrial chemical explosions pose a difficult problem because their seismic signals have physical characteristics similar to those of nuclear explosions. Consequently, methods routinely used to differentiate earthquakes from explosions may not be able to distinguish all legitimate chemical explosions from clandestine nuclear tests, and in some cases on-site inspections may be required. Fortunately, industrial explosions in the range of one to 10 kilotons are extremely rare. In addition, the vast majority of industrial explosions are "ripple-fired,"with bursts extending the total duration of the blast to a second or more.8 Ripple-firing frequently produces an identifiable signature in the observed seismic signal over a broad frequency band.9 It is therefore important for seismic stations to record data over a broad bandwidth that extends to high frequencies (above 10 Hz).
The Yunokummunarsk ExplosionThe challenges that face seismic monitoring can be partially illustrated through the following occurrence: On September 16, 1979, Soviet scientists detonated a nuclear device in a Ukrainian coal mine complex near the town of Yunokommunarsk ("Young Communist"). The explosion, which had a yield of 0.3 kilotons, was intended to fracture rock and thus release hazardous methane gas from the mine.10 The test was not declared internationally, and was discovered by the West only through a 1992 Izvestiya newspaper report.11 A seismic array operating 2,200 kilometers away in Norway recorded the explosion, but the signal was considered not to be of high enough quality to be included in the catalog of seismic events.12
To a certain extent, the explosion provides an example of a possible evasion scenario - conducting a small explosion under the guise of a legitimate mining operation - in which the signal was recorded but not identified as an underground nuclear weapon test. Such a scenario, however, should not be applied directly to a non-nuclear state. In conducting the test, the Soviet Union 1) was a sophisticated nuclear- weapon state with a highly developed nuclear arsenal and weapons of precisely known yield, 2) had experience conducting hundreds of underground nuclear explosions, and 3) probably was not particularly fearful of being caught. The 8,000 inhabitants of the town were evacuated under a "civil defense drill" (perhaps as a precaution against injury from ground motion or an accidental venting of radioactive material), and the explosion was felt over a radius of nearly four miles.13,14 It was reported, in fact, that the explosion knocked a portrait of V.I. Lenin off the wall at the mine's party headquarters, thus suggesting that explosions of this size were not commonplace at the site.
With improvements in signal analysis, it is likely that had the blast occurred today, it would have been detected, located accurately, and reported as an unusual explosion (but probably not identified as a nuclear explosion based solely on seismic data).15 Futhermore, several other sources of seismic data, including open stations operating within the former Soviet Union, are now available and - by providing valuable data at regional distances - improve the identification of low-magnitude events.
However, it is not enough simply to detect and identify seismic events. The monitoring system also must demonstrate a capability to perform these tasks despite any credible scenario that attempts to evade the system. Through the years, many creative evasion scenarios have been imagined, including 1) detonating a test during a naturally occurring earthquake to hide the test's seismic signal in that of an earthquake; 2) detonating a series of tests to imitate an earthquake; 3) decoupling or muffling the seismic signal of a test by detonating the explosion in a large underground cavity; and 4) masking a test with a large legitimate industrial explosion.16 Decoupling appears to be potentially the most effective of all evasion possibilities. In this scenario, a country would secretly conduct an explosion deep below the surface of the Earth in a large underground cavity, perhaps under the guise of a mining operation. If the hole is sufficiently large, the stresses imparted by the explosion to the wall of the cavity will not exceed the elastic limit of the rock, and therefore seismic waves will not be generated efficiently. Even if a small decoupled explosion was detected, it might be difficult to distinguish the test from a legitimate industrial explosion.
As the explosion becomes smaller, the monitoring task grows increasingly difficult. At lower yields it is easier to decouple small explosions, there are more earthquakes and industrial explosions of comparable size, and there are more factors that can strongly influence a seismic signal. When monitoring yields below 1-2 kilotons, seismic signals from small decoupled explosions must be differentiated from the many small earthquakes and industrial explosions of comparable magnitude. Although it might prove possible to detect such explosions, there is as yet no seismological discriminant that can reliably identify nuclear from non-nuclear explosions at small magnitudes.
Throughout the 1970s and 1980s, the "verification limit" for monitoring the Soviet Union was established by determining the yield at which it could not be demonstrated that a decoupled nuclear explosion would be detected and identified with high confidence (see box "1970's & 1980's 'How-Low-Can-We-Go' Algorithm "). Although it is tempting to apply the same analysis to the new problem of monitoring a global test ban treaty in the context of non-proliferation, such an approach may lead to an inappropriate answer. As we shall see in the following half of this chapter, assumptions appropriate for monitoring the Soviet Union in the context of a bilateral superpower treaty may not be appropriate for monitoring a global comprehensive test ban treaty in the context of nonproliferation.
1970's & 1980's "How-Low-Can-We-Go" Algorithm Step 1 - Determine threshold for detection Detecting an event does not mean simply that a seismic signal is seen at a station. Traditionally, it means that there is a 90% probability that an identifiable P-wave signal will be observed above the noise level at 4 or more seismic stations. By postulating a specific distribution of stations, noise levels, and propagation characteristics, the threshold for a hypothetical network can be estimated. Step 2 - Determine identification threshold Because a seismic signal must be clearly detected before it is identified, the identification threshold is higher than the detection threshold. The standard rule-of-thumb to account for this difference is to add 0.5 magnitude units to the detection threshold. Step 3 - Factor in possible evasion scenarios The most credible scenario for evading a seismic monitoring system is "decoupling". In such a scenario, the seismic signal is muffled by detonating the explosion in a large underground cavity. The reduction in the size of the seismic signal produced by such muffling is considered to be approximately a factor of 70. Step 4 - Define as treaty threshold After the detection threshold has been increased by 0.5 magnitude units for identification, translated into a yield based on extrapolating from U.S. experience, and multiplied by 70 to account for decoupling, the resulting yield is considered to be the monitoring threshold.
Return to:Table of Contents
Return to: Chapter II:The New Challenge
Continue to:Chapter II: Rethinking the Problem