Effectiveness of Real-Time Near-Field Tsunami Inundation Forecasts for Tsunami Evacuation in Kushiro City, Hokkaido, Japan - Post-Tsunami Hazard: Reconstruction and Restoration - page 157

Goto et al. 1997 ) on the spherical coordinate system. A nested grid system with
resolutions of 30, 10, 3.33, and 1.11 arc-sec is used.
A homogeneous Manning's roughness coeffi cient of 0.025 is assumed on the
grid system, a value widely used in tsunami inundation modeling (Imamura 2009 ).
The maximum tsunami inundations over 3 h of simulation are stored in the database
(Fig. 11.4 ).

11.2.2 Tsunami Numerical Model

During a real tsunami event, the fi rst step for NearTIF is to simulate tsunami propa-
gation by use of a tsunami numerical model in real time. After information about the
tsunami source is obtained, we can use it as an input for the tsunami numerical
model. The input can be earthquake source parameters (i.e., moment magnitude,
epicenter, strike, dip, rake, and depth), an earthquake fault model, such as slip dis-
tribution, or initial sea surface deformation. The tsunami numerical model solves
the linear shallow water equations to simulate the tsunami waveforms at the near-
shore points. The model parameters and the bathymetry data used should be the
same as those used for the tsunami waveforms database. To simulate 3 h of tsunami
propagation and obtain simulated tsunami waveforms at points near shore to Kushiro
takes approximately 2 min.

11.2.3 Tsunami Database Search Engine

The tsunami database search engine is designed to search for a fault model in the
database that gives the precomputed tsunami waveforms most similar to those obtained
by the tsunami numerical model. The search engine is programmed to rank the fault
models based on the similarity between precomputed tsunami waveforms in the data-
base and the tsunami waveforms from the tsunami numerical model. To evaluate the
similarity, we used root mean square (rms) misfi t between the precomputed tsunami
waveforms and simulated tsunami waveforms. For a specifi c event and a specifi c site,
a fault model with the highest rank will be selected as the site-specifi c best fault model.
The steps for obtaining the site-specifi c best fault model are listed below:

￿ The simulated tsunami waveforms within a time window are selected for the rms
analysis. The time window is based on wave cycles of the tsunami waveforms
that are automatically detected by the zero up/down crossing method.
￿ The NearTIF algorithm analyzed only the sets of tsunami waveforms with the
mean of maximum heights that is within a threshold of 30 % from the tsunami
waveforms from the tsunami numerical model.
￿ The tsunami waveforms are shifted by an optimal time shift ( o) that minimizes

the rms misfi t of the waveforms. Every scenario will have a root mean square

˄