Unlike the situation with hurricanes, tornadoes, or floods, there are no storm clouds or rising river levels to foretell an earthquake. Because they hit without advance warning, earthquakes are particularly terrifying. When earthquakes strike, they can cause massive human casualties and large amounts of damage. The January 1994 earthquake in Northridge, California, killed 57 people and injured almost 12,000 others while causing more than $ 25 billion in damages. But these numbers pale in comparison to what happened in Kobe, Japan, one year later. The Great Hanshin Earthquake there killed more than 5,000 people, left more than 300,000 homeless, and resulted in more than $ 300 billion in damages. Because earthquakes have the potential to greatly impact society, the US government embarked on an ambitious programme in the 1970s to develop methods for predicting earthquakes. The National Earthquake Hazards Reduction Programme sought to develop technologies that would allow for earthquake prediction on time scales of hours to days. Such predictions would not necessarily lead to reduced damage, but the hope was that they could reduce injuries and the loss of life suffered in a large quake. Scientists were optimistic in the beginning, in part due to a number of apparent successes in anticipating some earthquakes in the United States and China. However, earthquake prediction has proved more difficult than expected. One method of earthquake prediction involves studying the geologic history and noting when previous quakes have occurred. One study of a particular segment of the San Andreas Fault near Parkfield, California, noted that it had experienced four earthquakes over the previous 100 years at intervals of roughly 22 years. Based on this information, scientists predicted in 1984 that the area had a 95 percent likelihood of experiencing a moderate earthquake sometime between 1985 and 1993. As part of the Parkfield experiment, steps were taken to prepare for the expected event, including the development of warning strategies and studies of public response. Through November 1998, however, no earthquake had occurred in Parkfield, leading many people to conclude that the experiment had been a failure. Joanne Nigg, a sociologist who has studied the Parkfield experiment, concluded that the project was at least somewhat successful in forging links between scientific procedures and policy concerns. Much was learned about publicly issuing earthquake predictions in particular, that earthquake predictions themselves have important impacts on society. If an earthquake does occur in Parkfield, scientists will be prepared with a dense network of scientific instruments to record the quake and improve knowledge about how and why earthquakes occur. From the perspective of the late 1990s, it is evident that expecting timely and accurate earthquake predictions was too ambitious. In the mid-1980s the National Earthquake Hazards Reduction Programme reported to the US Congress that earthquake prediction was more problematic than had been anticipated. Today scientists are more focused on developing improved estimates of long-term earthquake probabilities, measured in decades or centuries. The programme is also working on early warning systems that detect ground motion after an earthquake has started. This information can be used to warn people farther from the epicenter (the point where the earthquake originates). The goal is to create early warning systems to notify people that a large earthquake has begun, from a few seconds to minutes in advance. This warning could allow some useful actions, such as shutting down or backing up systems in a nuclear power plant. In the early 1990s this type of warning system provided Mexico City about 75 seconds of notice that an earthquake had occurred off the coast. The ambitious Parkfield programme was successful at least in one of th