The Mw 7.0 Kumamoto, Japan, earthquake occurred on 15 April 2016 at 16:25 UTC. Using ground accelerations recorded by 104 near‐field stations, we investigate spatial variability of observed ground motions, apparent period dependence, and azimuthal variation, as well as rupture directivity effects on various intensity measures. We develop a simplified ground‐motion model that includes both geometric and anelastic attenuation terms. Comparisons of observed and predicted ground motions suggest that predictions from the Next Generation Attenuation‐West2 models provide good fits for the overall observation. Analysis of spatial distribution of the residuals shows that observed peak ground velocity (PGV) and long‐period spectral accelerations (SAs) in the 150°–180° azimuth range along the rupture backward direction (southwest of the fault) can be as low as 0.3–0.8 times the average observation of this event. Long‐period ground motions on the northeast side of the fault in the forward direction are much higher than average, with PGV and long‐period SAs ranging from 1.2 to 1.5 times the average. There is clear period dependence of the strong ground motion variation. The biases due to directivity generally decrease with decreasing period for all azimuth ranges. On the distance dependence of directivity effects, our study shows that directivity effects can be considered practically nonsignificant for stations close to the hypocenter. We also perform a log–linear regression of the residuals, using a new directivity predictor. Our results show that for the 2016 Mw 7.0 Kumamoto earthquake, rupture directivity produces significant amplifications in the rupture forward direction, whereas deamplification effects are observed in the rupture backward region. Directivity effects are particularly relevant for PGV and long‐period SA (i.e., SA at periods ≥2.0 s). Such effects do not have systematic influence on peak ground acceleration and short‐period ground motions (i.e., SA at periods <2.0 s).
Electronic Supplement:Figures of variation of regression residuals with Rrup for observed peak ground acceleration (PGA), peak ground velocity (PGV), and spectral accelerations (SAs) and of regression residuals versus VS30.