Development of the Earthquake Information System for Local Regions Using Minimum Seismic Network and Neural Network Technology

Abstract

To reduce the earthquake disaster, it is very important to evaluate strong ground motions as soon as possible after an earthquake occurs. In Japan, high-density seismograph networks were installed after the Hanshin-Awaji earthquake. The typical spacing between each seismograph is approximately 20 km. The seismograph density in Japan is enough to evaluate the seismic intensity distribution throughout Japan. However it is not enough to evaluate ground motions in local area that is important not only for the government of local area but also for the civilian self-disaster prevention. In this article, therefore, we have developed a new method to estimate high-resolution PGA (Peak Ground Acceleration) distribution from earthquake data observed at only two seismic stations using neural network techniques. The neural network learns relationship between the PGA of two seismic stations, H/V resonant frequency data and the soil zoning data. After the training period the neural network becomes to be able to estimate PGA at any given points. The new method has been applied to estimate PGA distribution in Zushi City, Kanagawa, Japan. As the results, high-resolution PGA distribution map has been estimated immediately with good accuracy.