Geoscience Reference
In-Depth Information
on the importance of the proper operation of the sea level stations, especially the open-ocean
DART stations, whose sea level observations of tsunami waves are not distorted by bathymetric
irregularities and local harbor resonances that affect the coastal sea level observations.
Japanese scientists have been leading in tsunami forecast modeling, have had forecast
models in operation for a while (including for near-ield events), and are able to draw from a
very sophisticated, densely covered observation network. They are also very active in develop-
ing new methods for real-time forecasting (e.g., using the inversion method; Koike et al. 2003).
In brief, the SIFT model identiies an interim wave ield from its database based on the seis-
mic data (inferred source parameters and epicentral location) once an earthquake is triggered.
As the tsunami arrives at sea level stations along its propagation path, tsunami amplitude data
are used to improve the forecast by scaling the pre-computed free-surface distribution. Finally,
the resultant scaled surface is used to initialize a boundary value problem and determine, at
high resolution, the wave ield, including inundation at the locations of interest. The three steps,
in more detail, are as follows:
1.
A pre-computed database of wave ields from unit earthquake sources is consulted
: NOAA/
PMEL built a database of 1,299 unit earthquakes. The sealoor displacement is com-
puted by the linear-elastic dislocation theory and is applied for each unit earthquake,
each representing a magnitude 7.5 earthquake with a deformation area 100 km long
by 50 km wide. Because the NOAA system was initially developed to produce fore-
casts for U.S. coastlines, the current database includes only events in the Paciic Ocean
and the Caribbean Sea, although efforts are under way to extend the database to the
Indian Ocean and the Mediterranean Sea. In addition, the database was developed for
thrust events only and is now being updated for other types of earthquakes, particu-
larly for the Caribbean region. By linearly combining the wave ields from adjacent unit
sources, the most plausible and realistic tsunami scenarios are roughly inferred from
the earthquake parameters. For example, a magnitude 8.7 earthquake with an ap-
proximate 400 km by 50 km deformation area requires superimposing the results from
four adjacent segments. Because the unit sources are arranged in a pair of parallel
rows, larger events with widths on the order of 100 km can also be represented. Each
archive includes data on the spatial distribution of wave heights and luid velocities;
this information is needed to initialize the boundary conditions, which is then used to
calculate in near-real time the inundation in speciic locales.
2.
Data assimilation from DART station data is performed
: In this step, near-real-time mea-
surements of the tsunami are used to scale the combined wave ield constructed from
the database. Once the tsunami is recorded by the DART sensor, the pre-computed
wave time series (wave heights and arrival times) are compared to and scaled using
the observed wave time series by minimizing a least square it. This scaling process
can achieve results as soon as the full wavelength of the leading wave is observed and
is updated with observations of the full wave time series. When the wave arrives at the
next buoy, the tsunami wave heights are corrected again, although the experience to
