|USFDC Home | USF Electronic Theses and Dissertations||| RSS|
This item is only available as the following downloads:
xml version 1.0 encoding UTF-8 standalone no
record xmlns http:www.loc.govMARC21slim xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.loc.govstandardsmarcxmlschemaMARC21slim.xsd
leader nam Ka
controlfield tag 001 001789759
007 cr mnu|||uuuuu
008 070607s2006 flu sbm 000 0 eng d
datafield ind1 8 ind2 024
subfield code a E14-SFE0001471
Henson, Joshua I.
Strategic geographic positioning of sea level gauges to aid in early detection of tsunamis in the Intra-Americas sea
h [electronic resource] /
by Joshua I. Henson.
[Tampa, Fla] :
b University of South Florida,
ABSTRACT: A tsunami is a series of large amplitude, shallow water waves generated by an event capable of displacing a massive volume of water. The displaced water propagates at speeds in excess of 800 kph until it dissipates or impacts a shoreline where it slows to 30 --^ 50 kph [NOAA and USGS Fact Sheet, 2005]. Earthquakes are the predominant tsunamigenic event, however, landslides, avalanches, submarine slumps or slides, volcanic eruptions, volcano flank failure, and meteor impact into an ocean can also cause a tsunami [McCann, 2004; O'Loughlin and Lander, 2003; Pararas-Carayannis, 2004]. This study includes past Caribbean tsunamigenic events assumed to be regionally destructive and generated by earthquakes and/or massive submarine slides/slumps. The approximate study area is from 7¨N, 59¨W to 36¨N, 98¨ W. Caribbean tsunami data suggests that a tsunami will occur in this region once every three years, and destructively once every 21 years [O'Loughlin and Lander, 2003]. Excluding the December 2004 Indian Ocean tsunami, approximately 13.8% of all tsunamis and 83% of all tsunami fatalities worldwide have occurred in the Caribbean [O'Loughlin and Lander, 2003]. In the past 150 years, 2,590 victims died from tsunamis in the Caribbean. As a result of these^ recorded fatalities and the rise of Caribbean population by almost 300% from 1950 to 2000 [CIAT et al., 2005], protection of human life is a primary reason for establishing a tsunami warning system in this region. The goal of this study is to identify the minimum number of sea level gauge locations to aid in tsunami detection in order to provide the most warning time to the largest number of people. This study defines which historical tsunamis were likely to have been regionally destructive, analyzes the tsunamigenic potential and population distribution of the Intra-Americas Sea (IAS), models 42 historical tsunamis with the United States Navy Coastal Ocean Model (NCOM), and recommends 12 prioritized locations for coastal sea level gauge installation. The results of this systematic approach to assess priority locations for coastal sea level gauges will assist in developing a tsunami warning system for the IAS and are currently being used by NOAA and IOCARIBE-GOOS.
Thesis (M.A.)--University of South Florida, 2006.
Includes bibliographical references.
Text (Electronic thesis) in PDF format.
System requirements: World Wide Web browser and PDF reader.
Mode of access: World Wide Web.
Title from PDF of title page.
Document formatted into pages; contains 60 pages.
Adviser: Frank Muller-Karger, Ph.D.
Tsunamigenic risk analysis.
x Marine Science
t USF Electronic Theses and Dissertations.
Strategic geographic positioning of sea level gauges to aid in early detection of tsunamis in the Intra-Americas Sea by Joshua I. Henson A thesis submitted in partial fulfillment of the requirements for the degree of Master of Science College of Marine Science University of South Florida Major Professor: Fra nk Muller-Karger, Ph.D. Doug Wilson, Ph.D. Mark Luther, Ph.D. George Maul, Ph.D. Date of Approval: April 6, 2006 Keywords: iocaribe-goos, tsunamigenic ri sk analysis, numerical modeling, ncom, caribbean Copyright 2006 Joshua I. Henson
Acknowledgements Many people have contributed their time and experience to th is work. I thank them for their dedication and especially thei r patience. They ar e Christine Kranenburg, Carrie Wall, Jesse Lewis, Judd Taylor, Brock Murch, Remy Luerssen, Dr. Chuanmin Hu, Dr. Luis Garcia-Rubio, Dr. Chris Moses, Zhiqiang Chen, Digna Rueda, Dr. Paul Zandbergen, Damaris Torres-Pulliza, Inia So to, and Laura Lorenzoni. I also thank Vembu Subramanian, Jeff Scudder, Cliff Merz, Rick Cole, and Jay Law for their support and expertise with ocean/met systems. I w ould especially like to thank Dr. Steve Morey for sharing his knowledge and experience of the NCOM, without which, much of this work would not have been possible. I must also thank my fiance, Susan, for her support and understanding over the past two years. This work was funded by NOAA grant number EA133R-05-SE-5280.
Note to the reader: The orig inal document contains color which is necessary to fully understand some figures. The original manuscript is on file with the USF library in Tampa, FL.
i Table of Contents List of Tables iii List of Figures iv Abstract v Introduction 1 Historical Tsunamis in the IAS Region 2 The Caribbean and Surrounding Tectonic Plates 3 Tsunamigenic Earthquakes 4 Tsunamigenic Submarine Slides, Slumps and Landslides 5 Tsunamigenic Volcanic Events 6 Sea Level Gauges in the Caribbean and Adjacent Regions 8 Methods 9 Creation of Tsunamigenic Events List 9 Determination of IAS Tsunamigenic Potential 16 Modeling 18 Bathymetry 19 Initial conditions 19 Determination of Coastal Grid Po ints (CGP), Population Data Incorporation, and Time Series Analysis 21 Sea Level Gauge Location Determination 25 Location Priority for Coastal Sea Level Gauges 26 Results and Discussion 28 Modeling Validity 28 Tsunami Travel Time and IAS Coastline Vulnerability 29 Sea level Gauge Location Priority 31 Operational Sea Level Gauges in the Caribbean 33 Conclusions and Recommendations 36 References 38 Bibliography 44 Appendices 46 Appendix A: A Note on Dry Cell Issues and Bathymetry Alterations 47
ii Appendix B: Initial Condition and NC OM Parameter Options Experiments 48 Introduction 48 Methods 48 Results and Discussion 49 Surface field output interval 49 Integration time step 49 Initial amplitude and e-folding radius 50 Seafloor roughness 51 Total run time 51 Appendix C: Travel Time Post Processing Experiments 52 Introduction 52 Methods and Discussion 52 Appendix D: Isochron / NCOM Travel Time Comparison Tests 56 Introduction 56 Methods 56 Results and Discussion 57 Appendix E: Travel Time Verification Study 58 Introduction 58 Results and Discussion 58 Aguadilla 58 Mayagez 58 Boqueron 59 Conclusion 60
iii List of Tables Table 1. List of modeled even ts, ordered chronologically 11 Table 2. Weight assignments to the tsunamigenic event source map 16 Table 3. Sensitivity test results summary 20 Table 4. List of populati on and tourist centers 23 Table 5. Decision rank matrix 32 Table 6. List of initial sea level ga uge locations recommend for a tsunami warning system 32 Table B1. Time series analysis locations 48 Table B2. Values used in se nsitivity experiments 1-10 48
iv List of Figures Figure 1. Plate boundaries and ts unamigenic source regions 4 Figure 2. The locations of the 42 historic al tsunamis simulated in this study 10 Figure 3. 1-degree resolution grid, map of the IAS, hist orical tsunami origins, and tsunamigenic source regions 17 Figure 4. Sector total weights 17 Figure 5a. All 10,623 coastal grid points used in the initial time series analysis study 22 Figure 5b. Inset of figure 5A; Close-up view of CGPs around Puerto Rico 22 Figure 6. 8,009 CGPs with at tributed population data 23 Figure 7. Population a nd tourist centers 24 Figure 8. Top 5% of risk sectors 26 Figure 9. Impact frequency 30 Figure 10. Mean travel time 30 Figure 11. Median travel time 31 Figure 12a. Operational and recommended s ea level gauge stations in the IAS 34 Figure 12b. Inset of figure 12a; Close-up vi ew of stations around PR, the USVI, and the Dominican Republic 35 Figure C1. Surface elevation time series at Krum Bay, USVI resulting from the 1918 Puerto Rico tsunami 52 Figure C2. Surface elevation time series at a generic point resulting from the 1918 Puerto Rico tsunami 54 Figure D1. Sector I30 isochrones 56 Figure E1. Sea level time series at Agua dilla resulting from the 1918 Puerto Rico Tsunami 58 Figure E2. Sea level time series at Maya gez resulting from the 1918 Puerto Rico Tsunami 59 Figure E3. Sea level time series at just east of Mayagez, in the bay of Mayagez resulting from the 1918 Puerto Rico Tsunami 59 Figure E4. Sea level time series at Boque ron resulting from the 1918 Puerto Rico Tsunami 60
v Strategic geographic positioning of sea level gauges to aid in early detection of tsunamis in the Intra-Americas Sea Joshua I. Henson ABSTRACT A tsunami is a series of large amplitude, shallow water waves generated by an event capable of displacing a massive volume of water. The displaced water propagates at speeds in excess of 800 kph until it dissipates or impacts a shoreline where it slows to 30 50 kph [ NOAA and USGS Fact Sheet 2005]. Earthquakes are the predominant tsunamigenic event, however, landslides, avalanches, submarine slumps or slides, volcanic eruptions, volcano flank failure, and me teor impact into an ocean can also cause a tsunami [ McCann, 2004; OLoughlin and Lander, 2003; Pararas-Carayannis 2004]. This study includes past Caribbean tsunami genic events assumed to be regionally destructive and genera ted by earthquakes and/or massive submarine slides/slumps. The approximate study area is from 7N, 59W to 36N, 98 W. Caribbean tsunami data suggests that a tsunami will occur in this region once every three year s, and destructively once every 21 years [ OLoughlin and Lander, 2003]. Excluding the December 2004 Indian Ocean tsunami, approximately 13.8% of all tsunamis and 83% of all tsunami fatalities worldwide have occurred in the Caribbean [ OLoughlin and Lander 2003]. In the past 150 years, 2,590 victims died from ts unamis in the Caribbean. As a result of these recorded fatalities a nd the rise of Caribbean populat ion by almost 300% from 1950 to 2000 [ CIAT et al ., 2005], protection of human life is a primary reason for establishing a tsunami warning system in this region. The goal of this study is to identify the minimum number of sea level gauge locations to aid in tsunami de tection in order to provide the most warning time to the largest number of people. This study defines which historical tsunamis were likely to have b een regionally destructive, analyzes the tsunamigenic potential and popul ation distribution of the Intra-Americas Sea (IAS), models 42 historical tsunamis with the United States Navy Coastal Ocean Model (NCOM), and recommends 12 prioritized locations for coastal sea level gauge installation. The results of th is systematic approach to assess priority locations for coastal sea level gauges will assist in deve loping a tsunami warning system for the IAS and are currently being used by NOAA and IOCARIBE-GOOS.
1 INTRODUCTION A tsunami is a series of large amplitude, shallow water waves generated by an event capable of displacing a huge volume of wate r. Whether a wave is considered to be a shallow or deep water wave depends on its wavelength and the depth of water. Deep and shallow-water waves are defined by the ra tio of their wavelength to the water depth. Deep-water wave: < 2 z Shallow-water wave: > 20 z where, = wavelength and z = water depth While tsunamis are usually generated in deep water, they are considered shallow-water waves because a typical wavelength of a tsuna mi is 220 km and the average depth of the Caribbean is approximately 2.6 km. Tsunamis can propagate at phase speeds in excess of 800 kph until they dissipate or encounter shallow water where they slow to 30 50 kph [ NOAA and USGS Fact Sheet, 2005]. Tsunami dissipation primarily depe nds on the magnitude and character of the tsunamigenic event, alt hough bathymetry and bottom type must also be considered. Eventually, the tsunami is likely to impact a shoreline where life and property are in harm's way. This study seeks to understand how and where tsunamis are generated, how they travel throughout the Caribbean and ad jacent regions, and where a minimum number of sensors should be located to most efficien tly warn the public of an impending tsunami. In order to produce a warning, a system must be in place to first detect the wave and predict potential impact lo cations and severity. Differe nt types of tsunami warning systems/networks are current ly being successfully employe d to measure, record, and telemeter both oceanographic and meteorological data. Standard means of telemetry include satellite, radio, cellu lar, telephone line, or Inte rnet. One type of tsunami monitoring system involves Real-Time Ki nematic Global Positioning System (RTKGPS) technology [ Kato, et al., 2001]. Curtis  suggests a multi-sensor approach. The Pacific Tsunami Warning System utilizes a combination of coastal sea level gauges and Deep-ocean Assessment and Reporting of Tsunamis (DART) buoys to acquire data for tsunami detection and propagation/run-up prediction. Wave data is captured and telemetered to a base station and input to a model. The model then predicts locations the tsunami is likely to impact. The predominant tsunamigenic events are earthquakes; however landslides, avalanches, submarine slumps or slides, vol canic eruptions, volca no flank failure, and oceanic meteor impact can also cause a tsunami [ Lander, et al., 2002; McCann, 2004; Pararas-Carayannis 2004]. Often, a tsunami is the re sult of coinciding events, thus it can be difficult to identify the tsunamigenic source. Seismic and/or volcanic activity can produce a submarine landslide, which can in turn generate a tsuna mi. When analyzing events from pre-instrument periods it can be difficult to determine if a submarine slump
2 or slide occurred and the actua l direct tsunamigenic event, such as this, may have gone undetected. The manner in which a tsunami is generated will affect the warning time available [ Lander, et al., 1999]. This warning time can be maximized by predicting how and where the next Intra-Americas Sea (IAS) tsunami is most likely to occur. In general, the closer a sea level gauge is to a tsunami origin the more warning time available. When designing a tsunami warning system it is critical to unde rstand the types of tsunamigenic mechanisms, the coastlines that are more likely to be affected by a tsunami, tsunami travel time to those coasts, and th e resulting effects from historical tsunamis [ Lander, et al., 1999]. However, the historical record is incomplete. Therefore, this study aims to define what hi storical tsunamis were likely to have been regionally destructive by simulating tsunamigenic events with the potential to have far-field (greater than 1000 km) destructive consequences and i llustrating where impacts were possible. These efforts are used, in conjunction with IAS population data, to determine the most critical and advantageous locations for the installation of coastal sea level gauges. Discussed later, most sub-aerial landslid es and volcanic tsunami origins are only locally destructive and are therefore not consider ed in this study. In order to determine if a tsunami is truly destructive at a location, high resolution bathymetry and a model with run-up capability is required to predict the extent of i nundation. Wave height along the coast is not analyzed in this study because lo cal effects dictate the necessity of very high bathymetric and model grid resolution to determine wave amplitude at the seashore. Run-up results along a coastline can vary by a factor of 10 [Hwang and Lin, 1969; Smith and Shepherd 1994]. Historical Tsunamis in the IAS Region Excluding the December 2004 Indian Ocean tsunami, approximately 13.8% of all tsunamis and 83% of all tsunami fatalities wo rldwide have occurred in the Caribbean [ O'Loughlin and Lander 2003]. Caribbean tsunami data over the past 100 years suggests that a tsunami will occur in this region once every three years and destructively every 21 years [ O'Loughlin and Lander 2003]. Shallow earthquakes, magnitude 6.5 or greater, cause the majority of Caribbean tsunamis [ McCann, 2004]. OLoughlin and Lander  describe 127 reported tsunamis in the Caribbean basin over approxi mately the past 500 years. Of those reported, the authors find that 53 are almost certainly true tsunamis and another 8 are most likely true. These tsunami events were generated by various sources including but not limited to earthquakes, submarine slides/s lumps, volcanic eruptions, and more likely a combination of those three. Understanding how past tsunamis have affected the region will help determine how future tsunami disasters can be mitigated. The historical record of tsunami origins and affected areas is sparse. The data used in this study is taken from both O Loughlin and Lander  and the National Geodetic Data Center [ NGDC 2005]. These original tsunami origin data have 0.1 degree precision [ Dunbar 2005, personal correspondence], and while there are historical records of areas affected by some of these events, for others there is no information on effects or arrival location.
3 Here, a numerical model is used to simulate historical tsunamis. The criteria used to select events that are simulated ar e discussed under Methods ("Creation of tsunamigenic events list"). The simulations are performed with the U.S. Navy Coastal Ocean Model (NCOM), discussed under Methods ("Modeling"). The Caribbean and Surrounding Tectonic Plates Tectonic activity due to plate movement is the principal cause of earthquakes, 80% of which occur along the plate boundaries in the oceanic crust [ Woods Hole 2005]. In order to fully understand the nature of the earthquakes that may generate tsunamis, the plate boundaries and their movement must also be understood. Figure 1 shows the plates in the region, their boundaries, and summarizes their interact ions. The Caribbean (CA) plate is bordered to the north and east by th e North American (NA) and South American (SA) plates, to the south by the SA, North Andes (ND), Panama (PM), and Cocos (CO) plates, and to the west by the CO plate [ Bird 2003; Lander, et al., 2002; McCann, 2004; O'Loughlin and Lander 2003; Pararas-Carayannis 2004]. Sitting on the CA plate are the islands of Hispaniola, Puerto Rico, and Ja maica to the north, the Lesser Antilles to the east, and to the west is Central America. The South American continent boarders the CA plate to the south [ Bird 2003; McCann 2004]. The CA plate is moving eastward approximately 20 3 mm/yr relative to the NA and SA plates [ Demets 1993; Grindlay, et al. 2005; Lander, et al., 2002; McCann, 2004; O'Loughlin and Lander 2003; Pararas-Carayannis 2004; ten Brink, et al. 2004]. Some estimates are as high as 37mm/yr [ Mercado and McCann, 1998; Sykes, et al., 1982]. The NA and SA plates are subducting un der the eastern margin of th e CA plate, leading to the formation of the Lesser Antilles volcanic arc. At the northern boundaries, the CA plate is sliding past the NA plate leading to transp ressional motion (compressive loading as a result of shear stresses) and uneven or oblique subduction near Puerto Rico [ Lander, et al. 2002; McCann 2004; O'Loughlin and Lander, 2003]. The southern boundary is characterized by a complex convergent margin near Venezuela and strike-slip faults on land [ McCann, 2004]. The CO plate is subducting under the CA plate on the western boundary, which also forms a chain of volcanic activity [ Lander, et al., 2002]. Further explanation on the tectonic regime of the CA and adjacent plates can be found in McCann  and Grindlay et al. .
90W 90W 80W 80W 70W 70W 60W 60W 5N 5N 10N 10N 15N 15N 20N 20N 25N 25N 04008001,200200kmCANASACANASANDPMCACONZ Active Fault Plate Bending Slow Earthquake Platform Deformation Yuc-Cuba Belt; Holcombe Ext Belt; Beata Ridge High Risk Med Risk Low Risk CA-NA slide PM-ND slide CA-SA slide NA-SA slide NZ-PM slide PM-CA slide ND-SA slide CO subducting under NA SA subducting under CA CO subducting under CA NZ subducting under ND CA subducting under ND CO subductin g under PM Figure 1 Plate boundaries and tsunamigenic source regions. This figure is a composite of two datasets. Bird  consists of the plate boundaries and between plate interaction represented by the plate labels and colored dots as shown in the legend. McCann  illustrates areas of tsunamigenic sources. Tsunamigenic Earthquakes The nature in which a tsunamigenic earthquake occurs will dictate the attributes of a resulting tsunami. There is a range of possible outcomes due to seismic activity in the Caribbean, some of which are more likely to produce a tsunami [Grindlay, et al., 2005; McCann, 2004; Mercado and McCann, 1998]. Typically, significant vertical deformation of the sea floor (i.e. a dip/slip earthquake) is required for tsunami generation. This deformation can be due to either isostatic rebound of an accretionary prism near a subduction zone or a change in crustal elevation [McCann, 2004; Okal, et al., 2003]. The direction of movement, depth of deformation, length and width of the deforming fault or plate boundary, deformation dip and slip angles, and focal depth will determine the size of the tsunami [McCann, 2004; Polet and Kanamori, 2000; Zahibo, et al., 2003a]. For example, a shallow subduction zone earthquake or an earthquake with a more vertical angle of deformation will usually displace a larger volume of water and consequently 4
5 generate a larger tsunami [ Bilek and Lay 2002; Polet and Kanamori, 2000]. The overlying geology also determines whether a tsunami will result from an earthquake [ Bilek and Lay 2002; Kanamori 1972]. There may be stronger motion at the sea floor than the measured seismic moment would typica lly represent if a rupture occurs within a sedimentary wedge or the r upture velocity is slow [Okal, et al. 2003; Polet and Kanamori 2000]. Regions where there is potenti al for an earthquake with a slow rupture velocity, or slow earthquake, to occur have a higher pot ential to produce tsuna mi larger than a seismometer would otherwise indicate [ Polet and Kanamori, 2000; Todorovska and Trifunac 2001]. When the sea floor deformatio n velocity is on the same order as tsunami velocity (i.e. a slow earthquake, slid e, or slump) the tsunami may be amplified by an order of magnitude [ Todorovska and Trifunac 2001]. The amplification may be caused by constructive interfer ence as the tsunami is produ ced since a slow rupture velocity will yield a long er duration earthquake [ Bilek and Lay 2002]. McCann  defines seismic tsunamigenic threats in the Caribbean (see Figure 1) into the following categories: platform deformation, plate bendi ng, slow earthquake, belts and ridges, active faults, and low to high tsunamigenic risk. These regions are base d on the geologic and tectonic regime of the IAS. Note how the plate boundaries/interactions [ Bird 2003] coincide with the tsunamigenic zones [ McCann, 2004]. Tsunamigenic Submarine Slides, Slumps and Landslides A landslide is very similar to a submarine slide or slump, except that falling debris from a landslide begins above the su rface of the water. A submarine slide or slump is a gravity driven mass movement of ma rine sediment and rocks. Sediments that have accumulated on a slope may become unstable and slide down. As the debris flows down it displaces the water in front of it. Th e volume of water displ aced is equal to the volume of sliding debris. Therefore, if the volume of land is known, the amount of displaced water can be determined and fr om this information wave weight, period, wavelength, and velocity can be calculated. These types of tsunamigenic events ar e typically initiated by an earthquake, hurricane or volcanic event such as an erupti on or flank failure but may also be initiated without an apparent catalyst [ Jiang and LeBlond 1992; Lander, et al., 1999; McCann 2004; O'Loughlin and Lander, 2003; Pararas-Carayannis 2004; von Huene, et al., 1989; Watts and Grilli 2004 (Submitted)]. Therefore, it may be difficult to determine whether a slide or an earthquake is the source of a tsunami. For example, the tsunami can be caused by a slide or slump that may or may not be related to an eart hquake. In the former case the slide or slump is the secondary tsunamigenic event, while in the latter it is the primary tsunamigenic event. A dip/slip earthquake, as described in the previous section, can produce a tsunami whether or not a slide or slump occurs. Many tsunamis have been generated in areas of the Caribbean where strike/slip plate movement dominates the tectonic activity [ McCann 2004]. This suggests a slide or slump as either the primary or secondary tsunamigenic mechanism because vertical deformation of the sea floor is not typically associated with strike/slip plate movement. Through mapping, Grindlay et al.  shows historic evidence of massive sl umps or slides along the northern Puerto
6 Rico margin which most likely generated tsunamis and cracking on the eastern edge of the Mona rift that may lead to the same mass failures as have happened in the past. Understanding how a tsunami forms he lps determine their propagation and destructive potential. Another characteristic of a slide or slump tsunami is their shorter period [ Fryer and Watts, 2000; Fryer, et al., 2001; Watts, et al. 2003]. In many of these cases, using evidence such as severed or damaged cables, wh ich typically result from a submarine slide or slump, will help de termine exactly how a tsunami formed [ Grilli and Watts 2005; McCann, 2004; Watts, et al. 2003]. Other parameters that influence the occurrence of a slide or slump include bottom type and slope steepness [ McCann, 2004]. Watts and Grilli [2004 (Submitted)] developed an equation to determine the potential amplitude of a tsunami based on initial slide thickness, initial slide length, mean slide depth, and the mean incline angle. From the Watts and Grilli [2004 (Submitted)] equation, McCann  produced a map illustrating the tsunamigenic potential for every possible slump or slide in the Caribb ean. This map shows that every coastline along the strike/slip margin in the west is a potential site for a s lide or slump induced tsunami. However, since slide or slump tsuna mi-like waves have a much shorter period than a typical tsunami, they dissipate faster and are typically only locally dangerous [ Pararas-Carayannis 2004]. Therefore, unless the sl ump or slide is massive, it is unlikely to be regionally destructive. Without detailed ocean bottom mapping and analysis it is difficult to determine the potenti al for a massive slide or slump. Hence, the slide or slump tsunamigenic potential map of the IAS is not considered in this work. Tsunamigenic Volcanic Events The subducting NA and SA plates melt pr ogressively under the CA plate with increasing depth as pressure and temperature rises. As this material heats up, density decreases and the material will tend to rise toward the surface again. This causes an increase in pressure on the crust that lies above, leading to the fo rmation of the Lesser Antilles volcanic arc [ Martin-Kaye, 1969; Pararas-Carayannis 2004]. Volcanoes along the Lesser Antilles chain are the most likely source for volcanic tsunamigenic events in the Caribbean Sea. Overall, approximately 5% of tsunamis are volcanic in origin [ O'Loughlin and Lander 2003; Sigurdsson, 1996]. There are many different volcanic tsunamigenic mechanisms from eruption to structural failure. OLoughlin and Lander  and Pararas-Carayannis  review case studie s of such events, which generated tsunamis that affected Montserrat, Martinique, St. Vincent, and Grenada. These waves originated from the Soufriere Hills volcano on Mont serrat Island, the Mt. Pele volcano on Martinique, the La Soufrire volcano on St Vincent, and Kickem Jenny, a submarine volcano north of Grenada, respectively. Energy from volcanoes can be transferred to the sea via three main mechanisms: explosions (rapid expansion of gas or other fluids by therma l energy), seawater displaced due to pyroclastic flow reach ing the sea (kinetic energy) and collapse of a caldera (potential energy) [ O'Loughlin and Lander 2003; Sigurdsson, 1996]. Each volcano has individual characteristics that can produce a tsunami and dict ates the attributes of the generated wave. The parameters discu ssed by Pararas-Carayannis  include
7 geochemical, volcanic explosivity and blast geometry factors, blast orientation and mechanism, and growth and collapses of lava domes. The tsunami of 26 December 1997 was most likely caused by a landslide triggered by an eruption of the Souf riere Hills volcano on Montserrat [ Heinrich, et al. 2001; Heinrich, et al. 1999a; Heinrich, et al. 1998; Heinrich, et al. 1999b; PararasCarayannis 2004]. This event resulted in a wave that impacted regions approximately 10 km away, flooding areas at distances of about 80 m inland with a run-up of approximately 3 m [ Heinrich, et al. 2001; Heinrich, et al. 1999a; Heinrich, et al. 1998; Heinrich, et al. 1999b; Pararas-Carayannis 2004]. Pararas-Carayannis  also discu sses Kickem Jenny, an active submarine volcano located approximately 8 km north of Gr enada. He reports that one tsunami-type wave generated by this volcano had approxi mately 2 m wave hei ghts observed at the northern coast of Grenada but its effects were not felt in Barbados. He goes on to explain in detail what factors predis pose a volcano (both sub-aerial and submarine) to tsunami generation and illustrates the resulting tsunami characteristics. His findings indicate that: i. Landslides usually only create local tsunamis. ii. An eruption as large as 20 kilotons w ill not cause a destructive tsunami since this energy does not translate efficiently to water waves [ Gisler, et al. 2004]. iii. Small tsunami-like waves may be generated by a submarine dome collapse. iv. Currently, the most destructive tsunami that could be created by Kickem Jenny would cause a maximum run-up of 3 m on the northern coast of Grenada and 1 to 2 m along the western coasts of Barbados, Trinidad, and St. Vincent. These waves would have a period of 1 4 min. v. Tsunamis of volcanic origin can be pr edicted in advan ce because volcanic activity is understood, mon itored, and renders warning signs prior to eruption or flank failure. vi. Volcanic tsunamigenic events are not a significant basin-wide threat; however, they are locally dangerous. Specifically this applies to the Soufriere Hills, Mt. Pele, La Soufrire, and Kickem Jenny volcano. Smith and Shepherd  also modeled a Kickem Jenny eruption and the results were different from those of Gisler et al.  and Pararas-Carayannis . Smith and Shepherd  found that the ther mal energy reservoir can yield large bubble expansion with an efficient energy yiel d. Based on this work, Sigurdsson  observed that a Krakatoa magnitude erupti on occurring at Kickem Jenny can lead to a run-up height of 40 m at Grenada and 7 m in the Virgin Islands [ O'Loughlin and Lander 2003; Smith and Shepherd 1994]. However, it was later foun d that more likely to occur is an eruption that would yiel d a run-up of 8 m at Grenada an d 1 m in the Virgin Islands [ O'Loughlin and Lander 2003; Smith and Shepherd 1995]. In the event that a destructive tsunami is generated by Kicke m Jenny, its path and dispersal through the Caribbean region is more predictable than th e behavior of a seismic tsunamigenic event because the volcano is well instrumented, th e number of tsunamigenic parameters is limited, and the point of origin is relatively fixed. The Seismic Research Unit of the University of the West Indies in Trinidad currently monitors the volcano for activity (http://www.uwiseismic.com/KeJ/kejhome.html), but to date there is no tool that
8 emergency response managers can use to predict the generation or dispersal of a tsunami. Nonetheless, most tsunamis of volcanic origin have relativ ely local destructive effects and/or are predictable. This limits how useful a basin wide tsunami warning system will be to protect the public from volcanic tsuna migenic events. The best defense against local tsunamis is public educati on. Therefore, these events wi ll not be considered in this study. Sea Level Gauges in the Caribbean and Adjacent Regions Approximately 60 sea level gauge stations were installed in the Caribbean and surrounding countries by NOAA (National O ceanic and Atmospheric Administration), programs such as RONMAC (Water Level Ob servation Network for Latin America) and CPACC (Caribbean Planning for Adaptation to Global Climate Change), and other locally and internationally funded program s to examine local sea level changes. Government organizations, educational instit utions, and independent companies maintain these stations. As of February 2006, the stations were in various stat es of disrepair, the majority no longer collecting data, and in many cases installations are missing equipment. To contribute to a tsunami warning network, mo st stations will need to be replaced, while others simply need additional hardware such as a GPS card and/or GOES transmitter [ Henson and Wilson 2005]. The IAS Tsunami Warning System (TWS) (IOC UNESCO, 2005) proposal recommends integration of an infrastructure including 31 upgraded sea level stations throughout the wider Caribbean Sea. As of Fe bruary 2006, out of the 60 stations that had been deployed historically throughout the IAS region, 17 are fully operational and transmitting data, 16 are not operational but th e equipment is accounted for, and 10 are questionably operational. The remaining are either no longer operational or not physically there [ Air-Sea Monitoring Systems, 2006; Henson and Wilson 2005]. Puerto Rico has been aggressively pursuing the development of a tsunami-ready sea level gauge network. The Puerto Rico Seismic Network (PRSN) have begun installing 10 sea level gauge stations around the island and on e base station in Mayagez [ von Hillebrandt-Andrade 2006, personal correspondence]. The base station will be capable of processing data from these and other sea level stations throughout the IAS.
9 METHODS This study seeks to determine where the minimum number of sea level gauges should be located to maximize the warning time to the largest amount of people. This is achieved by analyzing how and where regiona lly destructive tsunamis form, propagate, and impact a coastline as well as evaluating the coastal populati on distribution. It is also essential to know where coasta l sea level gauges are operation al so monitoring efforts are not duplicated. Although this study does not intend to pinpoi nt an origin location, it examines areas where a tsunami is more likely to occur by using a tsunamigenic event source map [ McCann, 2004] and the origins of 42 historical tsunamis. This risk analysis is critical to maximizing warning time because a sea leve l gauge should be installed closest to a tsunami origin. Propagation, travel time, and impact analysis is accomplished through the simulation of the histori cal tsunamis with the NCOM. There are several sub-studies involved in using the NCOM including para meter sensitivity a nd initial condition analyses and travel time calculations. The amount of warning time available is derived from a combination of mode ling with the NCOM, developing isochrones, and estimating travel time to coastal population centers throughout the region. The isochrones are developed independently and then te sted against the NCOM results. The tsunamigenic risk analysis uses a one-degree resolution grid whereas the NCOM uses a 2 arc-min resolution grid. Thes e are two independent studies in that the former is used to determine where the next tsunami is most likely to occur and the latter is used to understand tsunami propagation a nd travel time. Accuracy, resolution, and precision are not transferable from one grid to the other. Creation of Tsunamigenic Events List A total of 61 tsunamis have affected the IAS region in the past 500 years. Event data is taken from both OLoughlin and La nder  and the NGDC tsunami database . Since most volcanic and shore based landslide tsunamigenic events have localized effects, they are omitted from this study [ O'Loughlin and Lander 2003; Pararas-Carayannis 2004; Smith and Shepherd 1995]. Events are also discarded if the origin is located inland, the or igin latitude and longitude ca nnot be found, or the event did not originate in the IAS. Each event is rated on a scale of 0 4 by the source authors [ NGDC 2005; O'Loughlin and Lander 2003] according to the validity of the historical observations. These ratings are derived in different ways, ar e qualitative, and can be subject to opinion. The validity rating from the two datasets are co mpared and the higher rating is used. In an effort to create the largest list of proba ble events, the 42 simulated historical events have a validity rating of 3 or higher. All simulated tsunamigenic sources are assumed to be regionally destructive such as a dip/slip earthquake or massive underwater slump/slide whether caused by an earthquake or not.
It is necessary to adjust some of the historical origin coordinates to properly initialize the NCOM. Where possible, the origin is moved closer to or along a plate boundary, but in some cases they are moved perpendicular to isobaths. Figure 2 shows the origins of the simulated tsunamis based on historical information. The locations reflect adjusted origins. Table 1 provides a list of the events that are modeled and notes which origins are adjusted. DDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDD 90W 90W 80W 80W 70W 70W 60W 60W 5N 5N 10N 10N 15N 15N 20N 20N 25N 25N 04008001,200200km Figure 2 The locations of the 42 historical tsunamis simulated in this study. The origin of the tsunamigenic events are represented by an X (see Table 1). Note that some events have originated in the same location. 10
Table 1 List of modeled events, ordered chronologically (see also Figure 2). Shaded cells denote events whose origin is adjusted. The latitude and longitude listed reflect the adjusted location. Where applicable, original coordinates are in parenthesis. Sources: OLoughlin and Lander  and the NGDC Tsunami Database . No information was found for cells that are blank. Tsunami Origin Latitude (N) Longitude (E) Date Time Validity rating Earthquake magnitude and corresponding scale Source type and brief description Venezuela 10.80 (10.70) -64.20 (-64.20) 9/1/1530 1430 UT 4 Earthquake S. Belize 16.00 (16.20) -88.20 (-88.50) 11/24/1539 2300 LT 4 Earthquake Venezuela 10.80 (10.70) -64.10 (-64.10) 9/1/1543 2300 LT 4 Earthquake Leeward Is. 17.50 -61.50 4/16/1690 4 Ms 8.0 Earthquake; dispute regarding exact day, found 4/06/1690 as well Jamaica 17.70 (17.90) -76.80 (-76.90) 6/7/1692 1643 UT 4 Ms 7.5 Earthquake induced submarine landslide Venezuela 10.60 (10.60) -64.50 (-64.30) 1726 3 Earthquake Venezuela 10.50 (10.50) -64.50 (-64.30) 1750 3 Earthquake Hispaniola 18.10 (18.30) -70.70 (-70.70) 10/18/1751 1900 UT 4 Ms 7.3 Earthquake Haiti 18.00 (18.40) -72.20 (-72.80) 11/21/1751 0750 LT 3 Earthquake Martinique and Barbados 14.40 -61.00 4/24/1767 0600 UT 3 Shocks 11
Table 1 (Continued) Tsunami Origin Latitude (N) Longitude (E) Date Time Validity rating Earthquake magnitude and corresponding scale Source type and brief description Haiti 18.70 (18.60) -72.63 (-72.80) 6/3/1770 1915 LT 4 Earthquake Costa Rica 10.20 -82.90 2/22/1798 4 Earthquake Venezuela 11.50 -66.90 3/26/1812 3 Earthquake Jamaica 17.70 (18.00) -76.30 (-76.50) 11/11/1812 1818 UT 3 Earthquake Costa Rica, Nicaragua, and Panama 9.60 (9.50) -82.20 (-83.00) 5/8/1822 0500 UT 4 Ms 7.6 Earthquake Martinique 14.40 -61.00 11/30/1823 1130LT 4 Earthquake Martinique 14.20 -61.10 11/30/1824 0330LT 3 Earthquake Trinidad and St. Christopher 12.40 (12.40) -61.60 (-61.50) 12/3/1831 1140 UT 4 Earthquake Hispaniola and Cuba 19.97 (19.50) -72.10 (-72.10) 5/7/1842 2200 UT 4 Ms 8.1 Earthquake (No effect in PR) Guadelope 16.10 -62.20 2/8/1843 1435 UT 4 Mw 8.3 Earthquake induced landslide Cumana, Venezuela 12.10 -63.60 7/15/1853 1415 LT 3 Ms 6.7 Earthquake Honduras 16.00 (16.20) -88.20 (-88.50) 8/9/1856 4 Ms 7.5 Earthquake 12
Table 1 (Continued) Tsunami Origin Latitude (N) Longitude (E) Date Time Validity rating Earthquake magnitude and corresponding scale Source type and brief description St. Thomas, St. Croix, Puerto Rico, Dominica 18.10 -65.10 11/18/1867 1850 UT 4 Ms 7.5 Earthquake; along the north scarp of the Anegada Trough; 15 to 20 km SW of St. Thomas; St. Croix, St. Thomas, and Isla de Vieques formed a triangle around the epicenter; others believe it may have been of volcanic origin on Little Saba Puerto Rico 18.10 -65.10 3/17/1868 1045 UT 4 Earthquake Venezuela 10.80 (10.70) -63.80 (-63.80) 8/13/1868 1137 LT 4 Earthquake Lesser Antilles 15.50 -61.50 3/11/1874 0430 LT 4 Earthquake Jamaica 19.60 -75.50 8/12/1881 0520LT 4 Earthquake Panama 10.00 -79.00 9/7/1882 1418 UT 4 Ms 8.0 Earthquake (Landslide?) Haiti 19.70 -74.40 9/23/1887 1200UT 4 Earthquake Venezuela 11.00 -66.40 10/29/1900 0842 UT 4 Ms 8.4 Earthquake 13
Table 1 (Continued) Tsunami Origin Latitude (N) Longitude (E) Date Time Validity rating Earthquake magnitude and corresponding scale Source type and brief description Jamaica 18.50 (18.20) -76.60 (-76.70) 1/14/1907 2030 UT 4 Ms 6.5 Earthquake induced submarine landslide Puerto Rico 18.50 -67.50 10/11/1918 1414 UT 4 Ms 8.25 Earthquake induced submarine landslide (subduction near the Brownson deep [Mona Canyon]; cables cut in several places) Puerto Rico 18.50 -67.50 10/24/1918 2343 LT 4 After shock from the 10/11/1918 earthquake Cumana, Venezuela 10.60 -65.60 1/17/1929 1152 UT 4 Ms 6.9 Earthquake (fault activity; slides and collapses) Cuba 19.50 -75.50 2/3/1932 0616 UT 3 Ms 6.7 Earthquake Hispanola 19.30 -68.90 8/4/1946 1751 UT 4 Ms 8.1 Earthquake Puerto Rico 19.50 -69.50 8/8/1946 1328 UT 4 Ms 7.9 2nd shock from 8/4/46 earthquake; this one located 100 km to the NW Barbados, Antigua, Dominica 15.80 -59.70 12/25/1969 2132 UT 4 Ms 7.7 Earthquake 14
Table 1 (Continued) Tsunami Origin Latitude (N) Longitude (E) Date Time Validity rating Earthquake magnitude and corresponding scale Source type and brief description Leeward Is. 17.00 -62.40 3/16/1985 1454 UT 4 Ms 6.3 Earthquake (Possible Landslide) Puerto Rico 19.23 (18.90) -68.77 (-63.80) 11/1/1989 1025 UT 3 Ms 5.2 Earthquake Costa Rica, Panama 9.90 (9.60) -82.60 (-83.20) 4/22/1991 2156 UT 4 Ms 7.6 Earthquake Venezuela 10.90 (10.60) -63.50 (-63.50) 7/9/1997 1924 UT 3 Mw 7.0 Earthquake 15
16 Determination of IAS Tsunamigenic Potential This study simulates events with the potential to have far-field (greater than 1000 km) destructive consequences and illustrates where impacts are possible. The proximity of the islands to each other makes it difficult for tsunami energy to propagate out of the region (or, in the case of orig ins outside of the region, to mo ve into the Caribbean Sea). In order to determine if a tsunami is tr uly destructive at a location, high resolution bathymetry and a model with run-up capability are needed to predict the extent of inundation. The tsunamigenic potential is an index that considers both the spatial frequency of tsunamigenic events and the geologic and tect onic regime of the region. This index helps understand where the next tsunamigenic even t is likely to occur. In order to quantitatively measure the tsunamigenic potentia l of events it is necessary to place the data into bins. Through experimentation it was determined that 1-degree resolution is optimal because it is large enough to encompass more than one event but small enough to discern distinct regions of tsunami source areas. The McCann  tsunamigenic event source map (see Figure 1) is used to incorporate the geologic and tectonic regime of the region. Assigning a weighting system (Table 2) to the event source map, based on sour ce type, allows it to be used as a relative tsunamigenic risk map. The weights, although subjective, allow for a quantification of the tsunamigenic event potential. High, medium and low risk can be directly translated into weights (3, 2, 1 respectively) but sl ow earthquake potential, plate bending, or platform deformation regions as well as active faults, geologic belts and ridges also increase the potential for a region to produce a tsunami and are therefore assigned a weight of 1.5. This tends to be more important where areas of high, medium, and low risk overlap these regions. Table 2 Weight assignments to the tsunamigenic event source map [ McCann 2004]. High risk Medium risk Low risk Slow earthquake, belt or ridge, plate bending, platform deformation, active fault 3 2 1 1.5 The weight attributes of each source type are applied to the 1-degree resolution grid (Figure 3) and when a grid cell or bin is not completely covered by a source type, the fractional area each source type encompasses is calculated. This is multiplied by the weight of the source type to determine the weight of the bin. Multiple weight types in a single bin are combined in superadditive proce ss. For example, if a bin contains 1/3 high risk, 1/5 slow earthquake, and 1/ 3 platform deformation the resulting weight is: (1/3 3) + (1/5 1.5) + (1/3 1.5) = 1.6. The fractional ar eas can be both greater than or less than 1 since source types overlap. The final valu e of each bin is calculated by adding the spatial frequency to the poten tial bin weights (Figure 4).
DDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDD 90W 90W 80W 80W 70W 70W 60W 60W 5N 5N 10N 10N 15N 15N 20N 20N 25N 25N 04008001,200200km51525AGM 1020300DJP Active Fault Plate Bending Slow Earthquake Platform Deformation Yuc-Cuba Belt; Holcombe Ext Belt; Beata Ridge High Risk Med Risk Low Risk Figure 3 1-degree resolution grid, map of the IAS, historical tsunami origins, and tsunamigenic source regions. X represents the location of the historical origins. Historical origin data from OLoughlin and Lander  and NGDC . Tsunamigenic source data from McCann . 90W 90W 80W 80W 70W 70W 60W 60W 5N 5N 10N 10N 15N 15N 20N 20N 25N 25N 04008001,200200km5 1020300DJP 1525AGM > 0 1.0 1.1 2.0 2.1 3.0 3.1 4.0 4.1 5.0 5.1 6.0 17 Figure 4 Sector total weights. Result of binned historical tsunami origins and weight assignments to 1-degree resolution grid. Bins without coloring have a value of zero.
Modeling The purpose of modeling historic tsunamis in this study is to understand tsunami propagation throughout the region, determine which coastlines are likely to be affected, and measure the travel time to those locations. The conditions used to initiate the simulations (Table 3) are the same for every historical tsunami simulation due to a lack of specific historical data. The tsunami origin coordinates were originally accurate to 0.1 degrees [Dunbar, 2005, personal correspondence], however 19 had to be adjusted as described in Methods ("Creation of tsunamigenic events list"). Two different models are considered for simulating the historical tsunamis. They are the US Navy Coastal Ocean Model (NCOM) and Tsunami Travel Time software developed by the Russian Tsunami Laboratory. The Tsunami Laboratory [Gusiakov, 2000] developed a tsunami travel time (TTT) program to calculate and plot isochrones in both the North Atlantic and IAS. The user-defined inputs are source type (point or ellipse), point of origin, and run time as well as different isochron display options. This program has the ability to calculate the travel time to 70 locations in the IAS but does not allow them to be user defined. The isochrones could be used to estimate travel time to other locations but they can be difficult to interpret. Therefore, this model/program has limited utility for this study. The NCOM is a three dimensional model that can use a full-explicit, semi-implicit, or split-explicit time stepping integration scheme [Martin, 2000; Morey, et al., 2003b; Morey, et al., 2003a]. Others who have simulated historical tsunamis in the Caribbean use different models but these are based on the same basic equations used by the NCOM [Mader, 2001; Mercado and McCann, 1998]. Some of the basic equations of motion that NCOM solves are listed here in Cartesian coordinates from Morey et al. [2003b] (Equations 1 4). Although the Coriolis term is accounted for in the NCOM, its contribution is relatively small given the simulation duration (6 hr). zuKzFxpfvQuutuMu01V* (1) zvKzFypfuQvvtvMv01V* (2) gzp (3) Qzwyvxuv* (4) z S T ,, 18
19 ratr arameter 3) ity ) = gratational acceleration (m/s2) M = vertical eddy coefficient for momentum be The CFL condition is discuss cell. sed and metry and o id because the wave speed and hence travel time depends on the bathymetric solution. Bathym ion. n a difference in bathym etric resolution, only ETOPO2 bathymetry is used in this udy. Initial c tions where, u, v = velocity vector terms (m/s) = del opeoV = unit vector Q = a volume source or sink term (m 3 /s) t = time (s) f = coriolis p 0 = reference water density (kg/m p = pressure (Pa) S = salin F u F v = friction vector terms (N x, y, z = coordinate directions T = potential temperature ( o C) g vi K The 42 historical tsunamis are simulated with the NCOM using a leap-frog, semiimplicit time stepping integration scheme. This allows the use of larger time steps while maintaining stability and accuracy [ Morey, et al. 2003b; Rueda and Schladow 2002]. However, if too large a time step is used and the Courant, Friedrichs, and Lewy (CFL) condition is violated, gravity waves (such as those modeled for this research) may slowed down [ Bartello and Thomas, 1996; Dupont 2001]. ed later under the subsection "Initial Conditions". The NCOM is run in a barotropic mode with one depth averaged vertical grid Tidal components are not included in any of the simulations and the temperature and salinity features are not utili zed. All tsunamis are assumed to be shallow water waves. Sea boundaries are open and allow uninhibited passage. Land boundaries are clo act as a vertical wall. La nd is set to 20 m above sea level, and to avoid dry cell conditions as a wave reaches the coast, th e minimum water depth is set to 4 m (see Appendix A A note on dry cell issues and ba thymetry alterations). Wave run-up on land is outside the scope of this study due to a lack of high resolution bathy coastal topography for the study area, a nd a lack of high quality historical observations/measurements to ground truth mode l results. The grid resolution is set t match the bathymetry resolution (2 arc-min). It is not necessary to develop a higher resolution gr re etry ETOPO2 [ NGDC 2001] is a global, 2 arc-mi n resolution bathymetric and topographic dataset created by the NGDC. It is the highest resolution global bathymetry publicly available. Higher reso lution is available for select areas of the Caribbean reg The Puerto Rico Tsunami Warning and Mitigation Program used the ETOPO2 and a National Ocean Service (NOS) multibeam dataset [ NGDC 2004] to create a higher resolution bathymetry data set [ Mercado-Irizarry 2005, personal correspondence]. The NOS dataset covers the U.S. Exclusive Economic Zone (EEZ) around Puerto Rico at 15second resolution. However, to avoid artifici ally inducing a differe nce in model results based o st onditions Known as an inverse tsunami problem, a method of determining some initial conditions for a tsunamigenic event is to back calculate them from historical observa
20d to dix B describes the seninterval of 800 kph, two time steps will pass as a wave moves from one grid point to another. of tsunami impacts [Mader, 2001; Murty, 1977]. However, the historical record for tsunamis in the Caribbean region is poor and it is difficult to reconstruct such events with any accuracy. Some works have used a seismic or initial condition model [Mercado anMcCann, 1998; Meyer and Caicedo O., 1998] to determine the initial wave parameters while other models such as NCOM and MOST (Method of Splitting Tsunamis) can also run with user-defined initial conditions. For this study, several sensitivity tests are rundetermine initial wave amplitude and e-folding radius, bottom roughness coefficient, model time step, surface field output interval, and total run time. Appen sitivity tests in detail and the results are summarized in Table 3. The surface field output interval depends on the temporal resolution required to consistently identify the exact moment of tsunami impact. A surface field output of 45 sec is sufficient to obtain adequate resolution. The sensitivity experiments converged on a model time-step of 7.5 sec and a grid spacing of 2 arc-min, which also satisfies the CFL condition. The CFL condition states that the time step must be smallerthan the time it takes for a wave to propagate from one grid point to the next (Equation 5). Based on a celerity x tc C C1 (5) = wave celerity (m/s) delta t = time step (s) delta x = grid space (m) y tltsamplitude (m) e radius coefficientstep output interval run time (hr) where, c Table 3-folding SensitivitBottom roughness est resuTime summary Surface field Initial (m) (s) (s) Total 4 10,000 0.003 7.5 45 6 The shape of the initial wave adds the most uncertainty to the results of the simulations presented here. However, too little is known about the initial conditions of all of the events simulated. Therefore, in order to compare the output from each mrun, the same initial conditions are used to initialize all of the historical tsunamis simulations. Zahibo et al. [2003b] has also used the same initial conditions for 19 odel historic h as Kowalik and Whitmore , Shuto , and Mercado and McCann . al events, and a time step and grid spacing of 6 sec and 3 km, respectively. Each tsunami is modeled as a point source using a normalized Gaussian dome with an amplitude of 4 m and an e-folding radius of 10 km (Table 3; Equations 6 11).This assumes that the entire water column is composed of an incompressible fluid and that this generation process is instantaneous [Okada, 1985]. This assumption is based onprevious works suc
21he initial wave is produced by the formula, T 22*2*,Rreampji w ave (6) n sea level in meters; R is the -folding radius in meters; and i and j are grid coordinates. (7) jidxiix,*,*00 (8) where amp is the initial height of the sea surface above mea e r 22yx jidyjj y 30 1000*,cos*1950.111,jialatdjidx (9) 30 1000*1950.111,jidy (10) 3017,jjialat (11) Coastal Grid Points (CGP), Population Data Incorporation, and Time Determination of Series A nalysis Once the simulations are performed, custom programs, written in Research Systems, Inc Interactive Data Language (RSI IDL), are used to determine impact andcalculate the travel time to the coastlines throughout the IAS. One program uses the ETOPO2 bathymetry data to identify the first grid point just off shore. This produces 10,623 grid points in an area approximately from 7N, 59W to 36N, 98 W (Figure 5A close up of CGPs around Puerto Rico illustrates their resolution (Figure 5B). The A). ther custom program performs an automated time series analysis and is discussed later. o
22 100W 100W 90W 90W 80W 80W 70W 70W 60W 60W 10N 10N 15N 15N 20N 20N 25N 25N 30N 30N 35N 35N 04008001,200200km Coastal Grid Points Figure 5A All 10,623 coastal grid points used in the initial time series analysis study. 67W 67W 66W 66W 18N 18N 1830'N 1830'N 01020305km Coastal Grid Points Figure 5B Inset of figure 5A; Close-up view of CGPs around Puerto Rico Population data is obtained from the Latin American and Caribbean Population Database [CIAT, et al., 2005]. This database encompasses the Caribbean and South and Central American regions at a mean resolution of 33 km. The resolution varies from country to country and is generally 9 53 km. This data is incorporated into each coastal grid point using an euclidian allocation technique (with ESRI ArcGIS ), where each CGP is assigned the value of the population cell closest to it. The CGPs bordering the
23st 4 6). Sea level gauges throughout the Caribbean are capable of warning this coastline. continental United States are not used because it is shown later in the results that the travel time to where the continental US is impacted by the simulated tsunamis is at leahr (Figure 90W 90W 80W 80W 70W 70W 60W 60W 10N 10N 15N 15N 20N 20N 25N 25N 30N 30N 35N 35N 04008001,200200km he CGPs boardering the continental United States seen in 5A are not shown here. he high y ave nters. The resulting dataset is summarized in table 4 and displayed in figure 7. Table 4 List of pCoordinates from .com andJamaica Figure 6 8,009 CGPs with attributed population data. T Efficient use of a limited number of sea level gauges requires that each gauge warn the greatest number of people possible. To accomplish this, population centers are identified using the population data just described. A population center, due to tand variable resolution of the population data set, is defined as a CGP having a population of over 500. Once these points are identified, the dataset is edited to eliminatereplicates and points in close proximity to each other. Using this method, it is necessarto supplement this list with major tourist locations since these do not necessarily hhigh populations. These locations are also referred to as population ce opulation centers. denotes added tourist location; www.fallingrain adjusted to nearest CGP. St. Johns, Antigua and Barbuda* Near Old Harbour, Basseterre, Saint Kitts and Nevis* Kingston, Jamaica Basse-Terre, Guadeloupe (France)* Ponce, Puerto Ric o Christiansted, St. Croix (Virgin Island s)* Les Cayes, Haiti M arigot, Sint Maarten (Neth. Ant.)* Mayaguez, Puerto Rico Coastal Grid Points
Table 4 (Continued) Roseau, Dominica* Fajardo, Puerto Rico Fort-de-France, Martinique (France)* Santo Domingo, Dominican Republic Castries, St. Lucia* Near Jeremie, Haiti Bridgetown, Barbados* Near St. Marc, Haiti Kingstown, St. Vincent and the Grenadines* Cap-Haitien, Haiti St. George's, Grenada* Santiago De Cuba, Cuba Puerto Limon, Costa Rica* South Beach,Bahamas (New Providence) Portobelo, Panama* Near Barcelona, Venezuela Cancun, Mexico* Near PuertoCabello, Venezuela Playa del Carmen, Mexico* Near Carupano, Venezuela Willemstad, Curacao* Pampatar, Venezuela Cartagena, Colombia La Ceiba, Honduras Barranquilla, Colombia San Juan, Puerto Rico Santa Marta, Colombia Port-of-Spain, Trinidad and Tobago near Oranjestad, Aruba Havana, Cuba Puerto Cortes, Honduras Manzanillo, Cuba "/"/"/"/"/"/"/"/"/"/"/"/"/"/"/"/"/"/"/"/"/"/"/"/"/"/"/"/"/"/"/"/"/"/"/"/"/"/"/"/"/"/ 90W 90W 80W 80W 70W 70W 60W 60W 10N 10N 15N 15N 20N 20N 25N 25N 04008001,200200km Figure 7 Population centers (represented by the squares). 24 The travel time and associated data corresponding to each grid point are extracted from the model output and written to files in a two step process also using a RSI IDL program. Step one reads in data for all grid points, one record at a time. After this is
completed, the second step writes a sea surface elevation time series file for every point and one event file containing the travel time and associated data for each GCP. This threshold is used to determine if the CGP is impacted by the simulated tsunami. The CGP is considered to be impacted only if the threshold is met. Travel time is defined as when the first peak or trough, above a threshold (Equation 12), reaches the CGP. 00001.022nnHH (12) Both peaks and troughs are considered to determine travel time because, due to the initial condition uncertainty, phase error may be present. A peak or trough is identified when the time series meets the criteria set forth in both equations 12 and 13. These equations, in conjunction with the program just described, can accurately identify impacted locations and the first peak or trough in a surface elevation time series (Appendix C). 011nnnnHHHH (13) where, H = sea surface height and n = record number Sea Level Gauge Location Determination A sea level gauge for a tsunami warning system should be positioned to maximize warning time. Several factors are considered to calculate warning time. These include population centers, locations where a tsunami may occur, travel time or propagation speed, and wave dissipation. The Pacific Tsunami Warning System is designed to detect a tsunami within 30 min after the generating earthquake [Bernard, et al., 2001]. The IAS TWS proposal, accepted by the IOC (Intergovernmental Oceanographic Commission), recommends at least 15 min of warning time [IOC-UNESCO, 2005]. This study calculates the warning time by subtracting the travel time to the population center from the travel time to a sea level gauge. A population center is considered to be warned if it can be notified within 30 min after tsunami generation. In general, the closer the gauge is to the tsunami origin, the more warning time available to population centers. Knowing where a tsunami will originate is essential to determining where a gauge should be installed. In Methods (Determination of IAS Tsunamigenic Potential), the relative risk of where tsunamigenic events will occur is shown (see Figure 4). The McCann  tsunamigenic source map, used in part to create the tsunamigenic risk map, appears to have a gap in tsunami risk just north of Venezuela in sectors N25 and N26 (see Figure 3). This is assumed to be a gap because, based on his methodology for classifying risk or source areas and the frequency of historical tsunamis occurring in those sectors, they should be covered by a region of low risk. This additional low risk value is added to the value of sectors N25 and N26 as if completely covered by a low risk area. The bins or sectors without a value are discarded and the values of the remaining sectors, ranging from 1.00E-4 5.05, are relatively evenly distributed. The upper ~ 5%, or 15 of these sectors, are considered to be where tsunami-genesis risk is relatively highest (Figure 8). 25
26 90W 90W 80W 80W 70W 70W 60W 60W 5N 5N 10N 10N 15N 15N 20N 20N 25N 25N 04008001,200200km51525AGM 1020300DJP > 3 3.5 3.6 4.0 4.1 4.5 4.6 5.0 5.1 5.5 Figure 8 Top 5% of risk sectors. Note that the color bar shown here is different than that shown in figure 4. The NCOM is not used to model events from the center of the risk sectors because a tsunami many not necessarily originate there. The NCOM is specific in this regard where as an isochron system is general and renders conservative estimates. Isochron development and validation using the NCOM are discussed in Appendix D. Travel time is measured from the center of the shaded sectors in Figure 8 to the nearest point of land and to the population centers using a series of isochrones. The recommended gauge location corresponds to the point of land nearest to the center of the relatively higher risk tsunamigenic sectors. With this strategy, each point closest to a high-risk sector should receive a sea level gauge resulting in 15 locations. However, some sectors are closest to the same point of land and the final number of locations identified is discussed later. For simplicity, gauge locations are referred to as the sector they correspond to. Location Priority for Coastal Sea Level Gauges Through an iterative experimental process a simple decision matrix is developed to evaluate the relatively highest risk sectors in the following categories: i. Sector risk value ii. Number of population centers the sector gauge can warn in time iii. Number of population centers less than 1000 km away iv. Number of sectors closest to one potential gauge location v. Number of sectors sharing a border
27 Each sector is assigned a rank in all categor ies, the ranks are a dded together, and the sector with the lowest number is assigned an overall rank of 1, the second lowest a rank of 2, etc. The final priority list includes all aspects with equal consideration since all ranks are simply added together. The sector risk values are ranked so th e sector with the highest relative risk receives first priority. This means that, to a first order, a sea level gauge is most useful within or nearest to a sector that is the mo st likely to generate a tsunami. This location, though, may not be able to warn as many popul ation centers as a nother, reducing its effectiveness. According to the warning time criteria of 30 min, each location has the potential to warn a certain number of population centers. Some locations can warn all but 1 while others do not have time to warn up to 8 cente rs. However, in the Caribbean, the risk to population centers is low if th ey are at least 1000 km away from the tsunami origin [ Zahibo, et al., 2003b]. A direct line distance is used in this study, si nce the resulting complex island reflections and refractions s oon after tsunami generation make it difficult to perform accurate ray tracing. The list of population centers each gauge can warn is reduced to those less than or equal to approximately 1000 km aw ay from the center of the sector. The sector and corre sponding gauge that warns th e most population centers less than or equal to approximately 1000 km away is given higher priority. In some cases, different risk sectors are closest to the same point of land (Figure 10). It is more efficient to install a sea leve l gauge on a point of land closest to more than one sector. This gives the gauge the ability to warn of a tsunami originating from multiple sectors. Higher prio rity is allocated to sector s that share a gauge location. Population centers near multiple higher risk sectors have increased potential to be impacted by a tsunami. To account for this se ctor density or clus ters of higher risk sectors, the number of borders each sector shares with anothe r sector is counted. In this manner, higher priority is skewed towards the clusters of risk centers.
28 RESUL the lation lo cation priority, and currently operational sea le vel gauges within the IAS. Modeli TS and DISCUSSION This project is built on a se ries of sub-studies whose results have been used to develop methodology. The results of this system atic approach to assess sea level gauge location and priority should assist in deve loping a tsunami warning system for the IntraAmericas Sea. Here we review the modeli ng decisions and results, vulnerability of IAS coastline to tsunami impact, sea leve l gauge instal ng Validity Major aspects of modeling include choosi ng the correct model, the accuracy of the initial conditions, and the validity of assumptions. De pending on the model used fo both propagation and initial disp lacement there may be differences in calculated wave amplitudes. However, previous studies have not evaluated whethe r the choice of model affects travel time estimates [ Mercado and McCann 1998; Whitmore, 2003; Zahibo, et al. 2003a]. Travel times estimated here, in ge neral agree with those calculated in bo Weissert r th  and Mercado and McCann  and observed by Reid and Taber . lt of ee Table m McCann  report generally agree with th ty n is nami al ong a multi-segment fault line whereas it is considered a point source here. Weissert  developed an isochron time chart for the 1867 Virgin Islands tsunami (see Table 1). Travel times are in reasonable agreement for open areas, but less in regions of more complicated bathymetry. For example, he estimates a travel time of 100 120 min to the Northeast co ast of Cuba, but the NCOM travel time calculation was approximately 250 350 min. Here, there is a significant difference between the travel times to the coast and between th e ranges of travel times. This may have been a resu a coarser bathymetry used in Weisserts study (ETOPO5) or the breakdown of that models ability to simulate a tsunami in sh allow water, as explained by the author. Mercado and McCann  simulated th e 1918 Puerto Rico tsunami (s 1) and show a sea level time series for three Puerto Rico locations: Aguadilla, Mayagez, and Boqueron. These three time seri es are compared to those generated fro the NCOM output. As in this study, travel ti me to these locations is taken as the time corresponding to the first peak or trough on the Mercado and McCann  sea surface elevation time series. Reid and Taber (1919) report observations of the 1918 Puerto Rico tsunami. The travel times they and Mercado and ose produced in this study (Appendix E). Any discrepancies with Mercado and Mc Cann  may be because they use a higher bathymetric and grid re solution, more accurate bathym etry, and run-up capabili (Mercado and McCann use a 3 arc-sec grid resolution where a 2 arc-min resolutio used in this study). In addition, the locati on and shape of the initial wave is also different. They generate the tsu
29 Tsunami Travel Time and IAS Coastline Vulnerability Based on the temporal frequency of histor ical tsunamigenic events, this region is due for another destructive tsunami [ O'Loughlin and Lander 2003; Pararas-Carayannis 2004; Zahibo, et al., 2003b; Zahibo, et al., 2003a]. This work attempts to point out where the next tsunami is likely to occur and where a sea level gauge should be located to give the largest number of people the greatest warning time. Several works have discussed the local na ture of devastating effects from many historical tsunamis [ Mercado and McCann, 1998; Meyer and Caicedo O. 1998; PararasCarayannis 2004; Zahibo, et al., 2003a]. It has also been shown that tsunamis generated in the Caribbean can be destructive as far away as 2 3 hr [ Zahibo, et al., 2003b]. In order to determine the IAS coastline vulnerabil ity, here it is assume d that these tsunamis can be destructive up to 6 hr away. Figure 9 displays where 42 historical ts unamis have impacted and indicates the frequency of impact at those locations. To show where the continental United States has had the potential to be impacted, all 10,623 CGP s are included in figures 9 11. Some areas are never hit and some are hit by ever y tsunami modeled. The two main factors controlling this are the origin location and ba thymetry. To incorporate travel time with impact frequency and travel time, the mean tr avel time is displayed in figure 10. It can be inferred that where the mean travel time is low ( 30 min), the majority of tsunamis impacting that location originated close to it. The opposite can be inferred where the mean travel time is high (> 1.5 hr). The median travel time helps understand what locations may be more vulnerable to a regional tsunami regardless of impact frequency (Figure 11). Compared to mean travel time, the median tends to be lower at locations that are hit more frequently. The mean travel time is longer than the median 64% of the time, which means that there are more locations that are hit more often from tsunamis that travel long distances. This is an indication of their vulnerability to regional tsunami impact.
30 Figure 9 Impact frequency. Locations where a CGP was impacted by at least one of the 42 historical tsunamis. Colors denote frequency of impact at that location. 90W 90W 80W 80W 70W 70W 60W 60W 10N 10N 15N 15N 20N 20N 25N 25N 30N 30N 35N 35N 04008001,200200km 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 90W 90W 80W 80W 70W 70W 60W 60W 10N 10N 15N 15N 20N 20N 25N 25N 30N 30N 35N 35N 04008001,200200km 0.473 0.500 0.501 0.750 0.751 1.000 1.001 1.250 1.251 1.500 1.501 1.750 1.751 2.000 2.001 2.250 2.251 2.500 2.501 2.750 2.751 3.000 3.001 3.250 3.251 3.500 3.501 3.750 3.751 4.000 4.001 4.250 4.251 4.500 4.501 4.750 4.751 5.000 5.001 5.250 5.251 5.500 5.501 5.750 5.751 6.000 Figure 10 Mean travel time. Similar to figure 9 but here colors denote mean travel time in hours to that location.
0.375 0.500 0.501 0.750 0.751 1.000 1.001 1.250 1.251 1.500 1.501 1.750 1.751 2.000 2.001 2.250 2.251 2.500 2.501 2.750 2.751 3.000 3.001 3.250 3.251 3.500 3.501 3.750 3.751 4.000 4.001 4.250 4.251 4.500 4.501 4.750 4.751 5.000 5.001 5.250 5.251 5.500 5.501 5.750 5.751 6.000 90W 90W 80W 80W 70W 70W 60W 60W 10N 10N 15N 15N 20N 20N 25N 25N 30N 30N 35N 35N 04008001,200200km Figure 11 Median travel time. Similar to figure 9 but here colors denote median travel time in hours to that location. Sea Level Gauge Location Priority This study uses a two-pronged approach to determine the IAS regional tsunami risk. One assumes that a tsunami impact has the potential to be destructive up to 6 hr from the origin and the other assumes that a tsunami will only be destructive within approximately 1000 km from the origin. The former is important when determining what locations have historically had the potential for impact and the latter is considered when optimizing and prioritizing gauge locations. Table 5 summarizes the rank of the higher risk sectors by the factors dictating the installation location priority. These ranks were combined in a linear fashion to determine an overall rank (Table 6). In the event two sectors have the same value, they are assigned the same rank. The gauge corresponding to the sector with the highest overall rank should be installed first. The insertion the low risk area over sectors N25 and N26 described in Methods (Sea Level Gauge Location Determination) led to the addition of sector N25 to the list of relatively higher risk sectors. Table 6 shows the prioritized list of initial locations for sea level gauges recommended to provide an efficient warning system. When two sectors share the same potential gauge location and have a different priority, the higher priority rank is applied to both sectors. Several sectors share priority and two different locations are recommended for sector G22. Priority sharing can be resolved in a number of ways. The importance of one factor can be increased or decreased, a multiplier can be applied to a factor, or other factors can be included in the decision matrix such as site infrastructure, security of a site, and maintainability. As explained earlier, this study assesses regional 31
32 tsunami risk of impact based on historical tsunamigenic events, the geologic and tectonic regime of the region, wave propagation dynamics, and the location of major population centers within a range of 1,000 km from the center of the higher risk sectors. Nonetheless, a complete warning system s hould also consider exactly where run-up and inundation would occur and to what extent. Table 5 Decision rank matrix. The sector s are arranged in alphabetical order. Sector Risk value # of sectors with same closest land # of sectors sharing a border with sector of interest # warned < 1000 km away Total F21 5 3 1 2 11 F22 8 3 1 5 17 G19 15 3 2 4 24 G20 13 3 1 1 18 G21 11 3 1 3 18 G22 3 3 2 4 12 G24 14 3 3 6 26 G28 1 2 2 8 13 G29 12 2 1 9 24 H29 9 1 1 10 21 I29 4 1 1 8 14 I30 2 1 2 11 16 N25 6 3 3 7 19 O7 10 3 3 12 28 O10 7 3 3 12 25 Table 6 List of initial s ea level gauge locations recommended for a tsunami warning system. Locations are listed in order of hi ghest to lowest priority groups. Location coordinates should only be used as a guideline. Sector Approximate location fo r gauge installation Priority F21 Arena Gorda, Dominican Republic (-68.52, 18.78) 1 G22 Isla Mona, Puerto Rico (-67.89, 18.09) or Boqueron, Puerto Rico (-67.17,18.02) 2 G28, G29 Barbuda (-61.80,17.64) 3 H29, I29, I30 La Desirade, Guadeloupe (-61.05, 16.32) 4 F22 Aquadilla, Puerto Rico (-67.15, 18.50) 5 G20 G21 Boca Chica, Dominican Republic (-69.61, 18.45) Isla Saona, Dominican Republic (-68.57, 18.11) 6 N25 Punta Arenas, Venezuela (10.9667, -64.4) 7 G19 Las Calderas, Dominican Republic (-70.5, 18.20) 8 O10 Portobelo, Panama (-79.65, 9.55) 9 G24 Isla de Vieques, Puerto Rico (-65.45, 18.10) 10 O7 Punta Manzanillo, Costa Rica (-82.64, 9.63) 11
33 Changing the number and location of populat ion centers, as well as the decision criteria, may affect the suggested gauge prio rity. The population centers were selected based on population and tourism alone and ma y not need to be warned if they are protected by a wide continental shelf or other wave energy dissipation medium. In addition, the number of warnable populati on centers will increase if tsunamis have destructive capability at dist ances greater than 1000 km. Answers to these possibilities require higher resolution ba thymetry, modeling more origin s (including those that are hypothetical in areas of higher tsunamigenic pot ential), as well as calculating run-up and inundation. The installation location c oordinates are dependant on where the center of the higher risk sectors are and s hould therefore only be used as a guideline. The number of initial gauges recommended for installation ma y change if the definition of a high risk sector changes. The locations selected are based on the top 5% of the relatively higher risk sectors and do not constitute a finite list. Additional ar eas should be considered for sea level gauge installations, specifically Ve nezuela near Margarita Island, the southeast coast of Jamaica, and the southeast coast of Cuba. Although table 6 lists only one location per sector, in some cases 2 or 3 sensors may be more effective. It may take only one gauge to determine if the seismic event caused a tsunami, but this is a binary appro ach. It may not give enough information as to where else and to what extent the tsunami ma y impact on a larger scale. More sea level gauges can be used to detect a tsunami originat ing on either side of an island, and/or also improve travel time and wave height predictions. A more general approach to a warning sy stem is the installation of DART buoys. They have the potential to yield better pr edictions because, unlike a coastal sea level gauge, they receive a tsunami signal without being compromised by local effects or coastal noise. Although a DART buoy may prove more useful in propagation and wave height prediction as well as cover a larger origin area, they may not provide as much warning time. This approach cannot warn lo cations that are the same distance from the tsunami origin as the buoy, because a tsunami w ill reach both locations at about the same time. This reduces their usefulness and re quires that a robust warning system employ a combination of both coastal and open ocean sea level gauges. Operational Sea Level Gauges in the Caribbean Figures 12a and b show the locations of the fully operational and proposed gauges as well as the recommended locations in table 6. Approximately 60 sea level gauges have been installed in the Caribbean and ad jacent regions over the last 10 years and were thought to be operational. Of these, only 17 are currently operational and transmitting data, 16 are not operational but the equipment is accounted for, 10 are questionably operational, and the remaining are ei ther no longer operational or gone [ Air-Sea Monitoring Systems 2006; Henson and Wilson 2005]. The IAS TWS proposal [ IOCUNESCO 2005] recommends that 31 sea level stat ions become tsunami ready to operate within the IAS TWS. The PRSN begun installing 10 sea level gauges [ von Hillebrandt-Andrade 2006, personal correspondence] and the NOAA Nationa l Ocean Service (NOS) has 7 sea level gauges installed throughout Puerto Rico and the US Virgin Islands Two of the PRSN
tsunami ready gauges (Aguadilla and Isla Mona) and one of the NOAA NOS gauges (9752695) coincide with locations recommended by this study. Any sea level gauges used for tsunami warning must be supported as a part of an operational system and regularly maintained. Support can come from a variety of sources because coastal sea level gauges are typically a component of a larger station capable of collecting various other data including wind speed and direction, barometric pressure, precipitation, salinity, dissolved oxygen, water clarity, solar radiation, and current flow. These stations therefore have many applications, such as storm surge warnings and studies, hurricane forecasting, geostrophic current analysis, land subsidence, plate tectonics, commercial and recreational fishing and diving, search and rescue operations, and commercial shipping. [`[`[`[`[` 34 [`[`[`[`[`[`[`AAAAAAAAA A iDiiiiiiiDiiii DiDiDDDDDDDiDiDDDD 90W 90W 80W 80W 70W 70W 60W 60W 10N 10N 15N 15N 20N 20N 25N 25N 30N 30N 04008001,200200km [ ` NOAA NOS stations COMPS stations IAS TWS Proposed stationsPRSN Proposed stations A i DMy Recommended Locations Figure 12a Operational and recommended sea level gauge stations in the IAS. There are 12 operational sea level gauges sponsored by the NOAA NOS, 11 recommended locations for sea level gauges, 31 IAS TWS proposal locations, 10 PRSN locations proposed for the Puerto Rico Tsunami Ready Tide Gauge Network, as well as 11 Coastal Ocean Monitoring and Prediction System (COMPS) gauges shown in the figure. The alternate location for sector G22 is also shown. Box in northern Caribbean is enlarged in figure 12b.
[`[`[`[` 35 [`AAAAAAAAA [`[`A iiiiii DDDDDD 68W 67W 66W 68W 67W 66W 65W 65W 18N 18N 19N 19N 05010025km Figure 12b Inset of figure 12a; Close up view of stations around PR, the USVI, and the Dominican Republic. Illustrates the proximity of the locations recommended in this study with those already installed by NOS and those recommended by the PRSN. Note where the locations recommended in this study overlap the NOAA NOS and PRSN proposed locations. [`NOAA NOS stations i D My Recommended Locations IAS TWS Proposed stations A PRSN Proposed stations
36 CONCLUSIONS and RECOMMENDATIONS The goal of a tsunami warning system is to mitigate loss of life and property caused by a tsunami. Different types of systems/networks are currently being successfully employed to measure, reco rd, and telemeter both oceanographic and meteorological data for tsunami warning. This study determined prioritized locations for coastal sea level gauges in th e IAS based on tsunami generation risk factors, tsunami propagation throughout the region, population distribution, a nd tsunami travel time to population centers. These locations will give the maximum warning time to the largest number of people in the most efficient manner. A database of all sea level gauges inst alled or thought to be installed was compiled and used to coordinate the recomm ended locations. The expansion of the IAS regional tsunamigenic event risk analysis was accomplished by combining the spatial frequency of 42 historical tsunamis with a modified tsunami source map from McCann . This study assumes that the 42 tsunamis were generated by either a dip/slip earthquake or massive slide/slump and were regionally destructive. Each historical tsunami was modeled with the NCOM enabling estimations of where historical tsunamis have had the potential to a ffect and the travel time to 10,623 coastal locations. An animation of each simulation is available fr om the author upon request. Throughout this work a GIS database was created which will also be useful to those planning the IAS tsunami warning system. This study established that initially, 12 sea level ga uges are recommended, and 3 of these locations already have or are pl anned to have a gauge. These locations correspond to the land closest to the center of the relatively hi gher risk sectors and should serve as a guide for installa tion location. The list provided in Table 6 is not allencompassing, but represents a start and will primarily warn against tsunamis that originate in the higher risk sectors. To de termine exactly where a sea level gauge should be installed a thorough site evaluation is n ecessary. During the site evaluation, factors that need to be considered are those such as access to open water, proximity to a reef or other shoaling feature, infrastruc ture and security of site, and ease of station maintenance. It is difficult to predict where a tsunami will occur and how much damage it will do. Quantifying damage prediction for affected areas requires a bett er understanding of tsunamigenic event origins, higher resoluti on bathymetry, propagation modeling in the littoral zone, and inundation mapping. Run-up and/or inundation calculations must be performed for areas most susceptible to ts unami impact (Figures 9 11). Mercado and McCann  have begun doing this for Puerto Rico and this is already a viable product for the Pacific at the P acific Tsunami Warning Center [ Titov, et al. 2001]. Sea level gauges are a part of a larger system that records, processes, and telemeters data. These stations can provi de meteorological and oceanographic data to support other projects such as hurricane and storm surge monitoring and prediction, climate change monitoring, and assist in improving numerical models [ Alverson 2005].
37 These types of systems in other areas around the US are already used by harbor pilots, ship captains, the Coast Guard, recreational and commercial divers and fishermen, the surfing and sailing industry, sc ientists, and the general public Therefore, to guarantee continued existence and viability, these sta tions must have a multi-mission purpose to garner multifaceted support because thankfully, tsunamis do not occur very often. [ Baptista, et al. 2003]
38 References Air-Sea Monitoring Systems, A.-S. M. S. (2006), Assessment and restoration of the MACC sea-level monitoring st ations travel report. Alverson, K. (2005), Watching over the worlds oceans, Nature 434, 19-20. Baptista, M. A., et al. (2003), New study of the 1755 earthquake source based on multichannel seismic survey data and tsunami modeling, Natural Hazards and Earth System Sciences 3, 333-340. Bartello, P., and S. J. Thomas (1996), Th e cost-effectiveness of semi-Lagrangian advection, Monthly Weather Review 124, 2883-2897. Bernard, E. N., et al. (2001), Early detec tion and real-time repor ting of deep-ocean tsunamis, paper presented at International Tsunami Symposium, Seattle Washington. Bilek, S. L., and T. Lay (2002), Tsunami earthquakes possibly widespread manifestations of frictional conditional stability, Geophysical Research Letters 29, 1673. Bird, P. (2003), An updated di gital model of plate boundaries, Geochemistry, Geophysics, Geosystems 4, 1027-1079. CIAT, et al. (2005), Latin Am erican and Caribbean Populat ion Database, Version 3, edited, Centro Internacional de Agricultura Tropical (CIAT). Curtis, G. D. (2001), A multi-sensor resear ch program to improve tsunami forecasting, paper presented at International Tsunami Symposium, Seattle, Washington. Demets, C. (1993), Earthquake slip vectors and estimates of present-day plate motions, Journal of Geophysical Research 98 6703-6714. Dunbar, P. (2005), National Geophysical Data Center (NGDC). National Oceanic and Atmospheric Administration (NOAA)/ Nati onal Environmental Satellite, Data, and Information Service (NESDIS), edited. Dupont, F. (2001), Comparison of numerical methods for modeling ocean circulation in basins with irregular coasts, Dissertation thesis, McGill University, Montreal. Fryer, G. J., and P. Watts (2000), The 1946 Unimak tsunami: Near-source modeling confirms a landslide, EOS, Transactions, American Geophysical Union 81, 748.
39 Fryer, G. J., et al. (2001), S ource of the tsunami of 1 April 1946: A landslide in the upper Aleutian Forearc, in Prediction of underwater landslide hazards, edited by P. Watts, et al., Balkema, Rotterdam. Gisler, G., et al. (2004), Tw o-Dimensional Simulations of Explosive Eruptions of Kickem Jenny and other Submarine Volcanoes, paper presented at NSF Caribbean Tsunami Workshop, Nationa l Science Foundation; Los Alamos National Laboratory, Puerto Rico, 30-31 March. Grilli, S. T., and P. Watts (2005), Tsunam i generation by submarine mass failure I: Modeling, experimental validati on, and sensitivity analyses, Journal of Waterway, Port, Coastal, and Ocean Engineering 131, 283-297. Grindlay, N. R., et al. (2005), High risk of tsunami in the northern Caribbean, EOS, Transactions, American Geophysical Union 86, 121, 126. Gusiakov, V. (2000), Tsunami travel time cal culation program for the Caribbean region, edited, p. Program is on a CD, Tsunami Laboratory, Institute of Computational Mathematics and Mathematical Geophysics, Siberian Division of the Russian Academy of Sciences, Novosibirsk, Russia. Heinrich, F., et al. (2001), Numerical simulation of the December 1997 debris avalanche in Montserrat, Geophysical Research Letters 28 2529-2532. Heinrich, F., et al. (1999a), Numerical modeling of a landslide generated tsunami following a potential explosi on of the Montserrat Volcano, Physics, Chemistry and Earth Sciences 24, 163-168. Heinrich, F., et al. (1998), Simulation of water waves generated by a potential debris avalanche in Montserrat, Lesser Antilles, Geophysical Research Letters 25 3697. Heinrich, F., et al. (1999 b), Modeliser un raz de maree cree par un volcan, La Recherche, 318, 67-71. Henson, J., and D. Wilson (2005), Preliminary st atus report on tide gauges and observing stations in the Caribbean and adjacent waters, paper presented at The Group of Experts on the Global Sea Level Observ ing System (GLOSS) ninth session, IOC UNESCO, Paris, France, 24 25 February. Hwang, L., and A. C. Lin (1969), Experimental investigations of wave run-up under the influence of local geometry., in Tsunamis in the Pacific Ocean edited by W. M. Adams, pp. 407-425, East-West Centre Press, Honolulu. IOC-UNESCO. (2005), An Intra-Americas Sea Tsunami Warning System Project Proposal, edited, UNESCO Intergovernmental Oceanographic Commission.
40 Jiang, L., and P. H. LeBlond (1992), The coup ling of a submarine s lide and the surface waves which it generates, Journal of Geophysical Research 97, 12,731,744. Kanamori, H. (1972), Mechanism of Tsunami Earthquakes, Physics of the earth and planetary interiors 6 346-359. Kato, T., et al. (2001), A new tsunami m onitoring system using RTK-GPS, paper presented at International Tsunami Symposium, Seattle, Washington. Kowalik, Z., and P. M. Whitmore (1991), An i nvestigation of two tsunamis recorded at Adak, Alaska, Science of Tsunami Hazards 9, 67-83. Lander, J. F., et al. (2002), A Brief Hi story of Tsunamis in the Caribbean Sea, Science of Tsunami Hazards 20, 57. Lander, J. F., et al. (1999), Use of tsunami hi stories to define the local Caribbean hazard, paper presented at Intern ational Symposium on Marine Positioning, Melbourne, Florida, December. Mader, C. L. (2001), Modeling the 1755 Lisbon tsunami generation and propagation across the Atlantic Ocean to the Caribbean, Science of Tsunami Hazards 19, 9398. Martin-Kaye, P. H. A. (1969), Summary of the geology of the Lesser Antilles, Overseas Geology & Mineral Resources 10, 172-206. Martin, P. J. (2000), A description of th e Navy Coastal Ocean Model Version 1.0, 39 pp, Naval Research Laboratory, Stennis Space Center, MS. McCann, W. R. (2004), Estimating the thre at of tsunamigenic earthquakes and earthquake induced-landslide tsunami in the Caribbean, Earth Scientific Consultants, Westminster, CO. Mercado-Irizarry, A. (2005), Professor of Physical Oceanogr aphy at the Department of Marine Sciences of the University of Pu erto Rico; Coastal Hazards Specialist of the Sea Grant Program, edited, Puerto Rico. Mercado, A., and W. McCann (1998), Numerical simulation of the 1918 Puerto Rico tsunami, Natural Hazards 18, 57-76. Meyer, H., and J. H. Caicedo O. (1998), Ev aluation of tsunami wo rse scenarios in the Caribbean Sea and simulation of wave we ights a TIME project activity (Oral Poster), paper presented at Okushiri Tsunami/UJNR Workshop. Morey, S. L., et al. (2003b), Export pathways for river discharged fresh water in the northern Gulf of Mexico, Journal of Geophysical Research 108 3303-3318.
41 Morey, S. L., et al. (2003a), The annual cycle of riverine influence in the eastern Gulf of Mexico basin, Geophysical Research Letters 30 1867-1871. Murty, T. S. (1977), Seismic Sea Waves: Tsunamis, Bulletin of the Fisheries Research Board of Canada, Bull. 198 NGDC (2001), 2-Minute Gridded Global Re lief Data (ETOPO2), edited, National Geophysical Data Center, NOAA Sate llite and Information Service. NGDC (2004), NOS EEZ Bathymetry Data, edited, National Geophysical Data Center, NOAA Satellite and Information Service. NGDC (2005), NGDC Tsunami Database, edit ed, National Geophysical Data Center, NOAA Satellite and Information Service. NOAA and USGS Fact Sheet (2005), Tsunami Detection and Warnings, edited, United States Department of Commerce, United States Department of the Interior. O'Loughlin, K. F., and J. F. Lander (Eds.) (2003), Caribbean Tsunamis: A 500-Year History from 1498-1998 263 pp., Kluwer Academic Publishers, Dordrecht. Okada, Y. (1985), Surface deformation due to shear and tensile faults in a half space, Bulletin of the Seismological Society of America 75, 1135-1154. Okal, E. A., et al. (2003), The defici ent T waves of tsunami earthquakes, Geophysical Journal International 152 416-432. Pararas-Carayannis, G. (2004), Volcanic tsunami generating source mechanisms in the eastern Caribbean region, Science of Tsunami Hazards 22, 74-114. Polet, J., and H. Kanamori (2000), Sha llow subduction zone earthquakes and their tsunamigenic potential, Geophysical Journal International 142, 684-702. Reid, H. F., and S. Taber (1919), The Porto Rico earthquak e of 1918, House of Representatives, 66th Congress, 1st Session, Washington D.C., United States. Rueda, F. J., and S. G. Schladow (2002), Quantitative comparison of models for barotropic response of homogeneous basins, Journal of Hydraulic Engineering 128, 201. Shuto, N. (1991), Numerical si mulations of tsunamis its present and near future, Natural Hazards 4, 171-191. Sigurdsson, H. R. (1996), Volcanic Tsunami s, Summary Report for the UNESCO IOC IOCARIBE Tsunami Warning Works hop, UNESCO IOC IOCARIBE, Paris.
42 Smith, M. S., and J. B. Shepherd (1993), Preliminary investigation of the tsunami hazard of Kickem Jenny submarine Volcano, Natural Hazards, 7, 257-277. Smith, M. S., and J. B. Shepherd (1994), Explosive submarine eruptions of Kickem Jenny volcano: Preliminary investigations of the potential tsunami hazard in the eastern Caribbean region, paper presen ted at Caribbean Conference on Natural Hazards; volcanoes, earthquakes, windstorms, floods, St. Anns Trinidad and Tobago, October 11-15. Smith, M. S., and J. B. Shepherd (1995), Potential Cauchy-Poisson waves generated by submarine eruptions of Kick 'em Jenny volcano, Natural Hazards, 11, 75-94. Sykes, L. R., et al. (1982), Motion of Cari bbean plate during the last 7 million years and implications for earlier cenozonic movements, Journal of Geophysical Research 70, 5065-5074. ten Brink, U. S., et al. (2004), New seafloor map of the Puerto Rico Trench helps assess earthquake and tsunami hazards, EOS, Transactions, American Geophysical Union, 85, 349-360. Titov, V. V., et al. (2001), Project SIFT (Short-term Inundation Forecasting for Tsunamis), paper presented at International Tsunami Symposium, Seattle, Washington. Todorovska, M. I., and M. D. Trifunac (2001), Generation of tsunamis by slowly spreading uplift of the sea floor, Soil Dynamics and Earthquake Engineering 21, 151-167. von Hillebrandt-Andrade, C. (2006), Puerto Ri co Tsunami Ready Tide Gauge Network; Puerto Rico Seismic Network, edited. von Huene, R., et al. (1989), A large tsuna migenic landslide and debris flow along the Peru trench, Journal of Geophysical Research 94 1703-1714. Watts, P., and S. T. Grilli (2004 (Submitted)), Tsunami generation by submarine mass faliure: Part I wavemaker models, J. Waterway, Port, Coastal, and Ocean Engineering Watts, P., et al. (2003), Landslide tsunami cas e studies using a Boussinesq model and a fully nonlinear tsunam i generation model, Natural Hazards and Earth System Sciences 3, 391-402. Weissert, T. P. (1990), Tsunami trav el time charts for the Caribbean, Science of Tsunami Hazards 8, 67-78. Whitmore, P. M. (2003), Tsunami amplitude prediction during events: A test based on previous tsunamis, Science of Tsunami Hazards 21, 135-143.
43 Woods Hole, O. I. (2005), Major Caribbean ea rthquakes and tsunamis a real risk, edited, SCI/TECH. YubaNet.com. Zahibo, N., et al. (2003b), Estimation of far-field tsunami potential for the Caribbean coast based on numerical simulation, Science of Tsunami Hazards 21, 202-222. Zahibo, N., et al. (2003a), The 1867 Virgin Island tsunami: observations and modeling, Oceanologica Acta 26, 609-621.
44 Bibliography Anderson, J. (1784), An account of Morne Garo u, a mountain in the island of St. Vincent with a description of the volcano on its summit, Philosophical Transactions of the Royal Society 125, 32. Anderson, T. (1903), Recent volcanic eruptions in the West Indies, The Geographical Journal Anderson, T. (1908), Report on the eruptions of the Soufrire in St. Vincent in 1902, and on a visit to Montagne Pele in Martinique The changes in the district and the subsequent history of the volcanoes, Philosophical Transactions of the Royal Society Series A-208 (Part II) 275-352. Anderson, T., and S. J. Flett (1903), Report on the eruption of th e Soufrire of St. Vincent in 1902 and on a visit to Montagne Pele in Martinique, Royal Society Philosophical Transactions Series A-200 (Part I) 353-553. Carey, S. N., and H. R. Sigurdsson (1978), D eep-sea evidence for distribution of tephra from the mixed magma eruption of th e Soufrire on St. Vincent, 1902: Ash turbidites and air fall, Geology 6, 271-274. Choi, B. H., et al. (2003), Simulation of the trans-oceanic tsunami propagation due to the 1883 Krakatau volcanic eruption, Natural Hazards and Earth System Sciences 3, 321-332. Earthquake Research Committee, (2002), Seismic activity in Japan -regional perspectives on the characteristics of de structive earthquakes Headquarters for Earthquake Research Promotion, Japan. Feuillet, N., et al. (2002), Ar c parallel extension and localiz ation of volcanic complexes in Guadeloupe, Lesser Antilles, Journal of Geophysical Research 107, 2331. Hooper, D. M., and G. S. Mattioli (2001) Kinematic modeling of pyroclastic flows produced by gravitational dome collapse at Soufriere Hills Volcano, Montserrat, Natural Hazards, 23, 65-86. Mader, C. L. (2001), Modeling the La Palma Landslide Tsunami, Science of Tsunami Hazards 19, 150-170.
45 Mercado, A., and W. McCann (2001), Evalua tion of the tsunami hazard for Eastern Hispaniola and Western Puerto Rico in the Caribbean Sea Region (Oral Poster), paper presented at International Tsunami Symposium, Seattle, Washington. Pelinovsky, E., et al. (2004), Tsunami gene rated by the volcano eruption on July 12-13, 2003 at Montserrat, Lesser Antilles, Science of Tsunami Hazards 22, 44-57. Perret, F. A. (1937), The eruption of Mt. Pelee, 1929-1932, Carnegie Institute of Washington Publication 458, 126. ten Brink, U. S., et al. (2002), The nature of the crust under Cayman Trough from gravity, Marine and Petroleum Geology 19, 971-987. USGS (1999), Seismic and Tsunami Hazard in Puerto Rico and the Virgin Islands, paper presented at USGS Workshop, Colegio de Ingenieros y Agrimensores, San Juan, 23-24 March. Yalciner, A. C. (2004), Modeling of tsuna mis in the eastern Mediterranean and comparison with Caribbean (Oral Poster ), paper presented at NSF Caribbean Tsunami Workshop, National Science Foundation; Los Alamos National Laboratory, Puerto Rico, 30-31 March.
47 APPENDIX A: A NOTE ON DRY CELL ISSUES AND BATHYMETRY ALTERATIONS The NCOM is running on a 2 arc-min reso lution grid with on e depth averaged vertical grid cell. Some of the historical ts unami simulations will cause all of the water in a shallow grid cell to slosh out leaving the cell dry. The NCOM will crash when this occurs because it is not able to solve the equations describe d in Methods (Modeling). To solve this problem, the model is configured to modify the ETOPO2 bathymetry dataset to eliminate cells that are shallower than 2 m. This converts cells less than 2 m to 2 m. However, this is not sufficient for some of the simulations and the ETOPO2 bathymetry is changed to be no shal lower than 4 m. It is assumed that differences observed between a 2 and 4 m m odification are equivale nt to differences between a 2 m alteration and no adjustment. This assumption is valid because the number of cells affected from each modification is equivalent. Wave propagation, group speed, and therefor e travel time are dependant on water depth. Changing the bathymetry has the po ssibility of affecting the results of the simulation. In order to determine if thes e changes are significant, the 1918 tsunami originating off of Puerto Rico is simulated with both a 2 and 4 m bathymetry adjustment and the same locations listed in Appendi x B (see Table B1) are evaluated. These experiments are run for 8 hr. A surface elevat ion time series is plotted for each location and the data from one experiment is regre ssed against the other. Two locations are specifically discussed. The linear regression of surface elevation ti me series at Punto Higuero results in a correlation coefficient of 0.9779. The Sa nto Domingo site has coefficient of 0.9942 up to record 171 (approximately 2 hr), but th is value declines to 0.8965 once the full 8 hr (641 records) is used. When comparing the 2 and 4 m simulations to each other, the majority of the sites (see Table B1) have very similar amplitudes for approximately the first 100 (74.25 min) to 200 (149.25 min) record s but diverge therea fter. Based on the correlation and amplitude similarities between each simulation for all of the sites, the difference between bathymetry filters is not significant.
48 APPENDIX B: INITIAL CONDITIO N AND NCOM PARAMETER OPTIONS EXPERIMENTS Introduction Sensitivity tests are run to identify initi al wave amplitude and e-folding radius, bottom roughness coefficient, model time step, surface field output interval, and total run time. These parameters are tested using the 1918 tsunami originating off of Puerto Rico. The final results of these te sts are discussed in the main body of the manuscript under Methods (Modeling). These choices are used to simulate all 42 historical tsunamis. Methods After each 1918 simulation a time series is extracted for 7 locations around Puerto Rico, the United States Virgin Islands, and the Dominican Republic (Table B1). The grid resolution and bathymetry data used for these te sts is identical to that used in the main study. Table B2 lists the initia l condition and model parameter experiment values tested. The e-folding radius, bottom roughness coefficien t, and total run time values are finalized in later trials. These later experiments use an initial amplitude, model time step, and a surface field output interval of 4 m, 7.5 sec, and 45 sec respectively. The bottom roughness experiment uses al l 10,623 CGPs for analysis. A surface elevation time series is plotted for each location and the data from one experiment is regressed against the other. A linear regression correlation coefficient of 1 should result if the change in the paramete r(s) does not affect the model output. This result means that the NCOM is not sensitive to those changes. Table B1 Time series analysis locations. The latitude and longitude is rounded to the nearest CGP and the site elevation is taken from the ETOPO2 bathymetry. Name Long (E) Lat (N) Site elev. (m) Caja de Muertos, PR -66.50 18.60 -885 Isabella, PR -67.00 18.53 -301 Punta Higuero, PR -67.23 18.40 -131 Rio Grande, PR -65.73 18.47 -44 Tortola, USVI -64.60 18.50 -18 Krum Bay, USVI -64.90 18.37 -1 Santo Domingo (Rio Ozama), DR -69.87 18.47 -13 Table B2 Values used in sensitivity experiments 1-10. Exp # Initial amp (m) Integration time step (s) Surface field output interval (min) [surface field output interval (s)] / [time step (s)] 1 2 12.00 6.00 30.00 2 4 12.00 6.00 30.00 3 4 12.00 3.00 15.00 4 4 12.00 1.50 7.50 5 4 31.00 1.50 2.90 6 4 6.00 1.50 15.00
49 APPENDIX B (Continued) Table B2 (Continued) Exp # Initial amp (m) Integration time step (s) Surface field output interval (min) [surface field output interval (s)] / [time step (s)] 7 4 7.50 1.50 12.00 8 4 15.00 1.50 6.00 9 4 3.75 1.50 24.00 10 4 7.50 0.75 6.00 Results and Discussion Surface field output interval The parameter that dictates temporal re solution of a surface elevation time series is the surface field output inte rval. The limiting factor for th is interval is the amount of space available for data storage. Decr easing the surface field output interval proportionally increases the memory required to store the output data. Temporal resolution is important to identify the first peak or trough reaching the CGP. When the output interval is 6 or 3 min the surface elevation time series resolution is not high enough to discern the exact moment of impact. Based on the Rio Grande, PR and Caja de Muertos, PR locations, the output interval needs to be 1.5 min. However, the location closest to the tsunami origin, Punta Hi guero, has a travel time of 1.5 min faster in experiment 9 than in expe riments 7 and 8. To resolve th is discrepancy, the surface field output interval was decr eased from 1.5 min in experiment 9 to 45 sec in experiment 10. Experiment 10 uses the same integration time step as experiment 7 and a surface field output interval of 45sec. Two locations had the same travel time, one location was 45 sec slower and the remaining 4 locations were 45 sec faster. This discrepancy is within the accuracy of the model output considering its depe ndence on the initial conditions and bathymetry. It should be noted that the integration time step and surface field output interval must be a multiple of each other in order to compare the effect of changing the integration time step. The data from each surface field output interval will be interpolated by the NCOM if the integrati on time step is not a multiple of the surface field output interval (see Tabl e B2). Therefore, experiment s 4 and 5 data are not used. Integration time step An integration time step of 12 sec will run a simulation at 50% real time. The limiting factor for the time step is the amount of time required to run the simulation. This must be considered due to the time constraints of this project. Increasing the model time step proportionally reduc es the processing time. Based on the results from experiments 4 through 8, an integration time step of at least 7.5 sec is necessary. The model result s have significant differences between a 24 and 6 sec and between a 6 and 12 sec integrat ion time step. The differences between integration time steps of 7.5 sec (experimen t 7) and 15 sec (experiment 8) are not as evident. The travel time calculated to all of the locations (see Table B1) in experiment 7
50 APPENDIX B (Continued) is the same as in experiment 8. However, the phases of the surf ace elevation time series at the sample locations produced by experiments 7 and 8 are not in agreement after initial impact and wave magnitudes are never in agre ement. When comparing the locations, the wave height difference between an integration time step of 7.5 and 15 sec decreases with distance from origin but this relationship is not linear. There is a stronger correlation between experiments 7 and 8 with decreasing water depth after normalizing it with either distance from origin or travel time. The av erage linear regression correlation coefficient between the two experiments for all locations is just above 0.5. This correlation is not sufficient to conclude that an integration time step of 7.5 sec yi elds the same results as an integration time step of 15sec. Therefore, an integration time step of 15 sec is too long. Experiment 9 tests an integration time step of 3.75 sec to determine if an integration time step of 7.5 sec is sufficient. The comparison between experiments 7 and 9 (7 and 3.75 sec) gives similar results to the comparison of 7 and 8 (7 a nd 15 sec). The general trend of increased correlation with distance away fr om the origin and increasing travel time is stronger. For all but one location, Punta Higuero, PR, the travel time is the same for experiments 7, 8, and 9. The variation between experiments 7 and 9 is acceptable and a time step 7.5 sec is used. For an average tsunami celerity in the Caribbean of approximately 450 kph, roughly 3.5 time integration steps pass as a tsunami moves from one grid point to another. Almost 2 time steps pass if the celerity is 800 kph (a mo re typical speed in deeper water). In either case, the CFL condition is met as long as the celerity is less than 1584 kph (Mach 1.29) which is only possible if the ocean is ~ 255 km deep. The CFL condition is always be satisfied. CFL is described in more detail in the main body of the manuscript under Methods (Modeling). Initial amplitude and e-folding radius Differences in initial amplitude and e-fo lding radius will change which locations are impacted. This is because a larger initial amplitude and/or e-folding radius impart more energy to the ocean a nd the resulting tsunami travels farther over a shelf or other shoaling feature. The travel time and phase re sults appear to be the same when the initial amplitude is either 2 or 4 m. Based on this and previous works, an initial amplitude of 4 m is an accurate representation of the tsunamigenic events simulated in this study [ Mercado and McCann, 1998; Meyer and Caicedo O. 1998; Zahibo, et al., 2003b; Zahibo, et al., 2003a]. The travel time is affected by a change in e-folding radius from 10 to 40 km. The travel time is affected in the short term since the origin with the larger e-folding radius is closer to a coastline by 30 km. This travel time difference is on the order of 6 10 min. Travel time should not be affected in the long term because celerity is only a function of bathymetry and gravity. In addition, an efolding radius of 40 km produces a larger amplitude wave train, creating dry cells. An e-folding radius of 10 km is used.
51 APPENDIX B (Continued) Seafloor roughness The travel time is also affected by a difference in seafloor roughness. Two coefficients, 0.01 and 0.003, are tested. Out of 10,623 coastal grid points analyzed, approximately 2% have a difference in trav el time. The average difference is 10 min with a maximum and minimum difference of 78 and 0.75 min, respectively. Of the points with a different travel time (2%), approximately 43% of them have a faster travel time associated with a roughness coefficient of 0.01. This may be to due to wave interactions and shore reflecti ons. Baptista et al.  showed that travel time and wave heights change proportionally w ith seafloor roughness. A more common coefficient is 0.003 and is used in this study [ Mercado and McCann 1998]. Total run time Although these experiments are run for 8 hr it is only necessary to run a simulation for 6 hr. This is determined by observing propagation throughout the IAS via animations created from the NCOM output. More information regarding the physical parameters and numerical options used is available upon reques t of the authors.
APPENDIX C: TRAVEL TIME POST PROCESSING EXPERIMENTS Introduction Using the initial conditions and NCOM parameters identified in the sub-study described in Appendix B, the 1918 tsunami generated off of Puerto Rico is simulated to develop a travel time post processing method. A surface elevation time series is extracted for each location listed in Appendix B (see Table B1). These time series are used to determine a signal to noise threshold and a method for peak and trough identification. The travel time to each CGP is defined as the time corresponding to the first peak or trough. It is easy to manually distinguish between noise and the tsunami signal (Figure C1). However, in order to automate the travel time calculation process, a threshold criterion is necessary. The hypothesis is that the arrival of a tsunami should be associated with a rate of change greater than some value found in the numerical noise. The requirement is to determine a threshold that is surpassed after the last peak or trough in the noise but before the first peak or trough in the tsunami signal. Krum Bay, USVI-0.08-0.06-0.04-0.0200.020.040.06012345Time (hr)Sea Level (m) 6 Figure C1 Surface elevation time series at Krum Bay, USVI resulting from the 1918 Puerto Rico tsunami. It is important to note that, as explained in the main body of this manuscript, the elevation is not necessarily representative of realistic amplitude. Methods and Discussion The time series output is used to develop and test both the threshold and peak/trough identification equations. The 9 locations used in the experiment are those seen in table B1 as well as two other locations along a coastline where an impact is not expected. This expectation is visually derived from animation created from the NCOM output. Data is recorded at every grid point in the model whether a tsunami signal is present or not. Since the amplitudes calculated by the NCOM may not be accurate, small amplitude signals (less than 0.25 m) are considered for a tsunami signal. Therefore, relative changes must be used identify the presence of a tsunami. 52
APPENDIX C (Continued) Using the hypothesis noted in the introduction a simple rate of change formula is tested (Equation C1). 1nn H H (C1) where, H = sea level and n = record number This equation describes the elevation at the time or record in question minus the record prior to it. A tsunami signal is considered present if the resulting value is greater than some number yet to be identified. This number or threshold is derived by applying the equation to a time series where the tsunami signal can be readily observed as in figure C1. This formula acts as a criterion to be met before a second formula is applied to identify the peak or trough. However, equation C1 discriminates between increasing and decreasing sea level. To eliminate sign bias this equation is squared (Equation C2). 21nnHH (C2) Equation C2 fails to consistently distinguish the beginning of the tsunami signal from the noise and at some locations produces larger values in the noise than at the beginning of the tsunami signal. Therefore, no limiting value will work and another solution is required. Rate of change is defined here, as a change in elevation over a constant time period. The next hypothesis tests if an increased time period yields smaller values in the noise than at the beginning of the tsunami signal. The next equation tested is 22nnHH (C3) This equation does not produce higher values in the noise than at the beginning of the tsunami signal and thus is acceptable. The next step is to determine what the threshold value should be. Equation C3 is applied to the test locations time series' and since the time at the first peak is already established for these locations, a value that marks the beginning of the tsunami signal is selected. The result of this iterative process is a value of 0.00001 (Equation C4). This consistently marks the beginning of the tsunami signal when it is present and does not when no signal is present. Figure C2 is an example of a location that did not record a tsunami signal. 00001.022nnHH (C4) 53
APPENDIX C (Continued) -0.0002-0.00015-0.0001-0.0000500.000050.00010.000150.00020123456Time (hr)Sea Level (m) Figure C2 Surface elevation time series at a generic point resulting from the 1918 Puerto Rico tsunami. Note that a tsunami signal is not present and a tsunami would not be considered to have impacted this location. Once the threshold is reached, another equation determines the exact record or time at the peak or trough. Equation C5, 1nn H H (C5) results in a positive number if the sea level is going up and a negative number if it is going down. The same is true for equation C6, nn H H 1 (C6) Like sign numbers divided will always yield a positive number and opposite sign numbers divided will always yield a negative number. As the signal continues in the same direction the result to equation C7 will be positive, but if it changes direction over 3 records the result will be negative. Therefore, if equation C7 is satisfied it can be said that a peak or trough is present at record n. 011nnnnHHHH (C7) 54
55 APPENDIX C (Continued) This only holds true if the temporal re solution is high enough to eliminate the possibility of two data points ha ving the exact same elevation at the same peak or trough. Based on the 1918 simulation and locations test ed, the temporal resolution (45 sec) is high enough. However, during the post proce ssing of the NCOM data, it was discovered that in some cases this resolution is no t high enough. Post processing consists of evaluating a time series at 10,623 points for 42 s imulations, to determine the first peak or trough by applying the criteria shown in equati ons C4 and C7. Out of these 446,166 time series, 5 had two points at the same peak, and 2 had two points in the first peak. When a time series has two points at the same peak or trough, equation C7 will result in a divide by zero error. The program written for post processing notes this error and returns the point at which this occurs. This is theref ore easily corrected by manually analyzing the surface elevation time series fo r those locations and selec ting the first peak or trough value.
APPENDIX D: ISOCHRON / NCOM TTT COMPARISON TESTS Introduction The isochrones are developed using an average tsunami celerity of approximately 450 kph (Figure D1). This value is calculated using historical observations, the NCOM results, Mercado and McCann  results, and the average depth of the Caribbean. DDDDDDDDDD DDI30I29H29G28G29G24198519691874186818531843183118241690 60W 60W 15N 15N 20N 20N 010020050km 1767 & 1823 Figure D1 Sector I30 isochrones. The isochrones start at 5 min and increase in 5 min intervals out to 30 min. The largest isochron is 60 min. Note the 1767 and 1823 tsunamis have the same origin coordinates. Methods Out of the 15 relatively higher risk sectors (see Figure 8 in Methods Sea Level Gauge Location Determination), 10 encompass an area where at least 1 historical tsunami originated. Travel time from the tsunami origins within 9 of the sectors is compared to travel time calculated from the center of the corresponding sectors. For example, in figure D1, travel time estimates from the 1969 tsunami are compared to travel time estimates from the center of sector I30. Note that the tsunami origins being compared are not in the same location. In the event more than 1 tsunami origin is with in a sector the tsunami origin closest to the center of the sector is used. Within these 9 sectors, 37 locations are not protected based on the isochron analysis. These 9 sectors are used to compare the NCOM and isochron warning times. It is not possible to estimate NCOM travel time to some of the locations selected for a sea level gauge due to the limitations of ETOPO2 bathymetry. In these cases the point is adjusted to the nearest CGP. 56
57 APPENDIX D (Continued) Results and Discussion In general, the NCOM travel times are sl ower than the isochron travel times but they are also equal to and faster than th e isochron times. The difference between the isochron and NCOM travel times is larger at locations close to the center of a sector and at those that have deep water close to the co ast. This difference is most significant for population centers close to or at th e 30 min warning time criteria. The mean travel time difference, where th e NCOM travel time is shorter than the isochron travel time, is 16.5 19.45 min. The st atus of 4 locations could be changed to warned if this is considered a significant difference. However, as stated above, the NCOM tsunamis used for comparison originate in slightly different locations and warning time estimates must be conservative Changing the status of a few locations based on this sub-study will eliminate some conservation.
APPENDIX E: Travel Time Verification Study Introduction Travel time is defined as the record or time corresponding to the first peak or trough seen at a location. A simulation of the 1918 tsunami generated off of Puerto Rico is used for verification. Although the surface elevation time series shown here only display the first 75 min, this simulation was run for 8 hr. Reid and Taber  report travel times to a variety of locations, but only Mayagez, Aguadilla, and Boqueron are discussed here for the sake of brevity. These locations are used to compare results from this study to results from Mercado and McCann  and historical observations from Reid and Taber . Travel times reported from both studies are in general agreement with those calculated here. Results and Discussion Aguadilla The Mercado and McCann  travel time to Aquadilla is very similar to that found in this study. Both works estimate a travel time of ~ 6 min (Figure 1E). Reid and Taber  report a travel time from different observers of 4 7 min. Mercado and McCann  show a 3 m trough after the initial crest (0.7 m) and the second peak at twice the amplitude of the first. Although there are phase and wave height differences, the relative amplitude of the peaks and troughs shown here are also similar to those found in Mercado and McCann . -1-0.8-0.6-0.4-0.200.20.40.60.81051015202530354045505560657075Time (min)SSH (m) Figure E1 Sea level time series at Aguadilla resulting from the 1918 Puerto Rico Tsunami. Mayagez This study as well as Reid and Taber  show a travel time of approximately 23 min to Mayagez. Mercado and McCann  show large trough arriving first at 23 min and the following crest at 30 min. The amplitude seen here (Figure 2E) is smaller than that seen in the Mercado and McCann  time series (~ 0.8 m). Interestingly, another time series from this study at a grid point just east of Mayagez, in the bay of Mayagez, has a maximum amplitude of ~ 0.9 m (Figure 3E). This time series looks very similar to that shown in Mercado and McCann . The offshore grid point may 58
APPENDIX E (Continued) be more appropriate to use for comparison because here, the behavior of the wave closer to shore may not be properly resolved due to the bathymetry and/or model resolution. -0.2-0.15-0.1-0.0500.050.10.150.20.25051015202530354045505560657075Time (min)SSH (m) Figure E2 Sea level time series at Mayagez resulting from the 1918 Puerto Rico Tsunami. -1.4-1.2-1-0.8-0.6-0.4-0.200.20.40.60.811.2051015202530354045505560657075Time (min)SSH (m) Figure E3 Sea level time series just east of Mayagez, in the bay of Mayagez resulting from the 1918 Puerto Rico Tsunami. Boqueron Reid and Taber  report a travel time of 45 min, the Mercado and McCann  time series shows a travel time of approximately 47 min to the first trough (arrives first), and this study finds a travel time of approximately 45 min to the first crest (Figure 4E). The phase difference between this work and Mercado and McCann  seen at Mayagez is present here as well. In addition, the Mercado and McCann  temporal resolution appears to be higher than that used in this study (45 sec) which may also effect 59
APPENDIX E (Continued) the differences in results. The amplitude seen here (Figure 4E) is smaller than that published in Mercado and McCann  (0.9 m). -0.2-0.15-0.1-0.0500.050.10.150.2051015202530354045505560657075Time (min)SSH (m) Figure E4 Sea level time series at Boqueron resulting from the 1918 Puerto Rico Tsunami. Conclusion Reasons for the discrepancy with Mercado and McCann  may be because they use a higher bathymetric and horizontal grid resolution, more accurate bathymetry, and run-up capability. They use a 3 arc-second grid resolution and this study uses a 2 arc-min grid resolution. In addition, the location and shape of the initial wave is also different. They generate the tsunami along a multi-segment fault line whereas it is considered a point source here. Run up and/or local bathymetric effects are not considered and the resulting wave amplitudes are therefore, in some cases, smaller. This can be taken into account by adding a multiplier to all of the wave heights but this requires further analysis since the local bathymetric effects can also decrease wave height. In general, based on comparisons with Reid and Tabers  historical observations and the Mercado and McCann  modeling, this studys reported relative amplitudes and travel times appear to be accurate. 60