Forecasting Probabilistic Seismic Shaking for Greater Tokyo Based on 400 Years of Intensity Observations

The long recorded history of earthquakes in the Kanto region affords a rare opportunity to forecast the distribution of shaking almost exclusively from intensity observations rather then from source, path and site models. We calculate the Poisson (or, time-averaged) probability of severe damage (JMA shaking intensity 6) by using 10,347 felt or higher (JMA intensity≥3) observations recorded during 1600-2000 in a 390 x 545 km area centered on Tokyo. To analyze this rich dataset we divided the area into 5x5 km cells and combined historical (pre-1926) and modern (post-1926) intensity datasets. As the cell size grows, the number of observations increases but the geologic units that influence shaking (eg. alluvial channels, exposed bedrock) would be mixed, so we would forfeit the ability to correlate shaking and site effects. Most of the pre-1926 observations were digitized from Usami (2003) and are incomplete at low intensity levels, yet this period includes several destructive earthquakes for the region such as 1703, 1855 and 1923, and is thus rich in high intensities. Since 1926, JMA has systematically recorded intensity observations, but few large shocks have struck. After merging these datasets, we calculate the annual frequency at each intensity levels 3-7 to produce a regional frequency-intensity relation, which we fit by a power-law. The intercept to the regional curve was then adjusted for each cell to forecast JMA shaking frequencies and Poisson probabilities. This approach permits robust shaking likelihoods even for the many cells with few historical observations. The resulting probability of severe shaking (JMA intensity≥6) over an average 30-yr period is, for example, ~35% in the Tokyo-Kawasaki-Yokohama corridor, ~30% in Tateyama, ~25% in Nagano, ~15% in Choshi, and ~10% in Chiba. Correlations with surface geology and plate boundary faults are both quite evident.