Consideration of the temporal and spatial variation of earthquake ground motion is important in the design of large structures (such as dams and containment structures for nuclear power plants), long structures with widely-spaced multiple supports (such as bridges and surface pipelines), and long buried structures (such as gas and oil pipelines, etc.). Recent data available from closely-spaced arrays of seismographs have been indispensable in the study of the space-time variation of ground motion. One such digital seismograph array is the SMART-1 array located in Lotung, Taiwan, which has recorded a large number of events. Some studies have been initiated based on the data obtained from this array (Bolt et al. 1982; Loh et al. 1982; Loh 1985; Harichandran & Vanmarcke 1986).
An initial model based on the preliminary study of one far-field event (Event 20) recorded by the SMART-1 array, which characterized the ground motion as a random field in space and time, was proposed by Harichandran and Vanmarcke (1986). In the present work, the results of additional extensive analyses on two far-field events (Events 20 and 24) are resented. Only 15-second long strong motion parts of the accelerograms dominated by S-waves were used in the analyses. The main characteristics of these events are given in Table 1. The model proposed earlier is verified, and a simpler model which may suffice for some applications is suggested. Kanai-Tajimi and "double-filter" models for the point spectral density function of ground acceleration are compared.
The regularity between the gross time delays for different pairs of stations (also observed in the earlier study) are used to derive relative arrival times at each station through a multiple linear regression model. These arrival times are then used to estimate the direction of gross wave propagation. It is found that for both events the direction of propagation is away from the epicenter.