Data recorded by the SMART 1 array in Taiwan are used to make the first direct measurements of two-dimensional earthquake rupture propagation. Using frequency-wavenumber techniques and ray theory, we obtain estimates of the spatial extent, duration, and directions of rupture propagation during two earthquakes located offshore Taiwan. Our estimates include uncertainties due errors in fault plane location and orientation, velocity structure, and slowness (ray parameter) measured at the array. We find that the January 29, 1981, ML=6.3 Taiwan earthquake ruptured unilaterally updip and toward the west on a 60¿ dipping, 109¿ striking reverse fault. Using P waves, we find its fault length and duration to be 25¿18 km and 7.4¿3.4 s, respectively. S waves indicate a fault length and duration of 27¿15 km and 9.4¿3.6 s, respectively.

We suggest that the November 14, 1986, ML=7.0 Hualien, Taiwan earthquake was triggered by a shallow (h≈ 14 km) foreshock ML=5.4 which occurred on a subparallel or splay fault approximately 5 s before the mainshock. Based on P waves, rupture propagation during the foreshock is unilateral towards the northeast and slightly downdip on a 58¿ dipping, 38¿ striking reverse fault. Its spatial extent and duration are approximately 23 km and 5.4 s. The mainshock initiated at greater depth (h≈34 km) than the foreshock with its primary direction of rupture unilateral toward the northeast and slightly updip on a 57¿ dipping, 43¿ striking reverse fault. Using P waves, we found a spatial extent and duration of 68¿15 km and 12.4¿2.2 s. S waves indicate a similar extent, duration, and direction of propagation for the mainshock. However, we found it difficult to distinguish the foreshock and mainshock S waves. We show, by example, that these results represent an improvement over what can be obtained using long-period (T>5 s) teleseismic methods.

Comparison of our results with those obtained using teleseismic data suggests either a short duration slip-time function or that most slip occurs near the rupture front. Our results indicate that much of the complexity in the seismograms recorded during these events, by the SMART 1 array, is due to spatial and temporal variations of the source. This suggests that, at least in some cases, predictions of both the phase and amplitude of strong ground motion, as a function of time, would require a priori knowledge of source properties. Finally, we recommend siting small (≈10 station) dense arrays on lang above offshore subduction zones, which are considered seismic gaps, because they can provide more detailed measurements of rupture propagation than is presently possible. ¿ American Geophysical Union 1991