This study describes the method and initial testing of a new forecast system built on years of effort by the whole tsunami research community. The goal of tsunami research has always been toward practical applications that will reduce the impact of this natural disaster and save lives. Real-time forecasting is an important but not the only component of this effort, which includes tsunami warning, education and community planning. This work demonstrates that a forecast, if based on direct tsunami observations and carefully designed numerical models, can finally provide accurate and timely community- and source-specific tsunami amplitude estimates for real-time tsunami assessment. This has been the goal of the tsunami research community since the first tsunami warning system was established in Japan in 1933.
The described method is only the first step in providing accurate, timely, robust and global real-time tsunami forecast. All these forecast goals (accuracy, speed, reliability, global coverage), remain formidable challenges. Our forecast research and development efforts highlighted the following outstanding gaps in the tsunami research that should be resolved in order to meet our forecasting goals.
As in several other publications [Titov, 2009; Tang et al. 2006, 2008a, 2008b; Wei et al., 2008; Titov et al., 2005], the tsunami source events discussed in this paper are far from the forecast locations. The Hawaii forecast models were chosen for demonstration because of their abundant tsunami records with relatively high signal-to-noise ratios, which is essential to assess forecast accuracy. Coastal wave amplitudes at Alaska and the U.S. West Coast for recent tsunamis (since the mid-1980s when corresponding deep ocean observations became available) are generally small and overwhelmed by noise from the continental shelf and local resonance; forecast results outside Hawaii have been published elsewhere [Wei et al., 2008; Tang et al., 2008a]. Several studies indicate, however, that the described method will be effective for local tsunami forecasting as well. The local American Samoa modeling results for the 29 September 2009 Samoa tsunami conducted at NCTR (http://nctr.pmel.noaa.gov/samoa20090929-local.html) and several recent papers [Bernard and Titov, 2007; Borrero et al., 2009] show that forecast accuracy for local events may be sufficient for critical decisions during event response. Forecast accuracy is, of course, only one component of the local tsunami problem, along with warning timing, response measures, planning, education and other factors. A thorough discussion of local forecast issues is outside of the scope of this study but will be included in several forthcoming papers at NCTR. More research is needed to evaluate this technique for local forecasting. Nevertheless, the initial results demonstrate that the method is applicable to the local tsunami problem, where not only initial tsunami arrival forecast is important, but the forecast of hours of tsunami impact is essential for critical emergency decisions during the event.
While establishing the NOAA DART network has been a huge leap in tsunami observations, and is key to tsunami forecast accuracy, questions of the optimum network size and placement of individual DARTs for best forecast accuracy require substantial additional research. DART data assimilation into forecast modeling (inversion methods) has only recently started to be researched [Percival et al., 2009] and requires substantial development. New tsunami observation methods may complement gaps in DART coverage and improve the forecast accuracy.
Assessing accuracy of the high-resolution forecast models is not easy due to limited number of observations for a particular site. New methods of assessing forecast model accuracy are needed, since there will be many forecast locations with no historical tsunami data. Model assessment of the tsunami current velocities has uncertain accuracy, due to lack of observation data. Additional research on modeling tsunami flow velocities is needed. New improved models may be required to forecast nonseismically induced tsunamis. While obviously incomplete, this list provides the immediate research needs for further forecast development.
The new forecast system not only summarizes years of research but also provides unparalleled opportunities to further tsunami science and understanding of the tsunami phenomenon. The forecast tools provide access to models, data, and a platform for testing new methods of tsunami research for future forecast application. The paper describes several research studies completed using the forecast tools with the discussion of new scientific results, in addition to the description of the development and tests of operational forecast models.
In the present study, sensitivity tests of nearshore tsunami wave characteristics were conducted for ranges of model grid setups, resolutions and parameters. Four forecast models (standby inundation models) were described for the coastal communities of Hilo, Kahului, Honolulu and Nawiliwili in Hawaii. The computational grids for the forecast models were derived from the best available bathymetric and topographic data sources. The models were tested with fourteen historical tsunamis and 6197 scenarios of simulated TMw 7.5, 8.2, 8.7 and 9.3 tsunamis based on subduction zone earthquakes in the Pacific. The outputs of the forecast models are compared to both historical water level data and numerical results from reference inundation models of higher resolution to ensure numerical consistency.
The accuracy of the maximum wave height computed by the four forecast models of 2 arc sec (60 m) resolution is greater than 80% when the observed maximum wave height is greater than 0.5 m, and the error is less than 0.3 m while the observation is less than 0.5 m. The error of the modeled first arrival is within 3% of the travel time. Wavelet analysis of the tsunami time series indicates that the peak wave period often coincides with the one of the resonant periods of the harbor where the tide gauge is located. This peak frequency may partially depend on the geographic location of the tsunami source. The optimized forecast models can accurately provide a site-specific forecast of first wave arrival time, wave amplitudes, and inundation limit within minutes of receiving tsunami source information constrained by deep ocean DART measurements. It is also capable of reproducing later tsunami waves reflected or scattered by far-field bathymetry that may arrive hours after the first arrival.
The tsunami hazard assessment study shows tsunami waves nearshore can vary significantly from the same magnitude tsunamis from different subduction zones or different locations on the same subduction zone. Furthermore, the offshore maximum wave amplitude is not a good indicator for the amplitude at a tide gauge. Therefore, earthquake magnitude alone is ambiguous and insufficient to provide information for accurate coastal tsunami amplitude forecasting. By including local bathymetry and topography and utilizing deep ocean tsunami measurement data via DART-constrained propagation scenario, which reflects the event-specific tsunami magnitude, the high-resolution forecast models are able to quickly provide accurate site-specific coastal predictions.
The authors thank the reviewers for their comments and suggestions; Elena Tolkova and Jean Newman for their extensive assistances; Mick Spillane for providing information on Kahului offshore; Marie Eble, Nazila Merati, Lindsey Waller, Clyde Kakazu and Allison Allen for assistance with tide gauge and DART data; Eddie Bernard, Yong Wei, Robert Weiss, Diego Arcas, Burak Uslu and Arun Chawla for discussions; Edison Gica for assistance with the TSF database; Christopher Moore for assistance with MOST I/O; David Borg-Breen and Doug Jongeward for hardware and software support; and Ryan L. Whitney for comments and editing. This research is funded by the NOAA Center for Tsunami Research (NCTR), PMEL contribution 3332. This publication is partially funded by the Joint Institute for the Study of the Atmosphere and Ocean (JISAO) under NOAA Cooperative agreement NA17RJ1232, contribution 1757.
Return to Previous section
Go to References