p020090625754952011548.pdf

Risk Analysis DOI: 10.1111/j.1539-6924.2008.01112.x

Improving Tsunami Warning Systems with Remote Sensingand Geographical Information System Input

Jin-Feng Wang1 and Lian-Fa Li1

An optimal and integrative tsunami warning system is introduced that takes full advantageof remote sensing and geographical information systems (GIS) in monitoring, forecasting,detection, loss evaluation, and relief management for tsunamis. Using the primary impactzone in Banda Aceh, Indonesia as the pilot area, we conducted three simulations that showedthat while the December 26, 2004 Indian Ocean tsunami claimed about 300,000 lives becausethere was no tsunami warning system at all, it is possible that only about 15,000 lives couldhave been lost if the area had used a tsunami warning system like that currently in use in thePacific Ocean. The simulations further calculated that the death toll could have been about3,000 deaths if there had been a disaster system further optimized with full use of remotesensing and GIS, although the number of badly damaged or destroyed houses (29,545) couldhave likely remained unchanged.

KEY WORDS: Disaster system; loss reduction; optimal design; scenario simulation

At 7:58 a.m. (local time) on December 26, 2004,beneath the Indian Ocean west of Sumatra, Indone-sia, pent-up energy from the compression forces ofone tectonic plate grinding under another found aweak spot in the overlying rock. The rock was thrustupward, and shook as a 9.0 magnitude earthquakesent its vibrations out into the ocean. The earthquakecaused vertical deformations of the ocean floor. Thedeformation displaced a volume of water, which gen-erated the tsunami. Tsunamis spread out in all di-rections; the massive waves washed over islands andcrashed against coastlines in Indonesia, Sri Lanka,Thailand, southern India, and even the east coast ofAfrica. Tens of thousands of people were killed, mil-lions made homeless.

1 Institute of Geographic Sciences and Natural Resources Re-search, Chinese Academy of Sciences, Anwai, Beijing, China.

Address correspondence to Jin-Feng Wang, Institute of Ge-ographic Sciences and Natural Resources Research, ChineseAcademy of Sciences, A11, Datun Road, Anwai, Beijing 100101,China; tel: + 86 10 4888965; fax: + 86 10 64889630; [email protected].

To improve the efficiency of remote sensing indisaster management, we combined remote sensingand geographical information system (GIS). If weknow nothing about an event, remote sensing mayprovide some spectrum information of objects. If wehave some data in advance, we can choose suitablesensors to target the object in due place and time; re-mote sensing and GIS can be much more effective ifcombined with the components of a disaster system(He et al., 1993; Wang, 1993).

This article first reviews the existing tsunamiwarning system and remote sensing and GIS appli-cations for the December 26, 2004 Indian Oceantsunami. We analyzed the disaster system and inves-tigated the potential roles of remote sensing and GISin each of the components. Next, we simulated threescenarios investigating the disaster, first recreatingthe extent of the disaster without a warning system(current situation), second investigating the disasterif there had been a warning system such as the oneused in the Pacific Ocean, and third simulating thedisaster if an optimally designed warning system hadbeen installed. The simulations show clearly how

1 0272-4332/08/0100-0001$22.00/1 C 2008 Society for Risk Analysis

2 Wang and Li

disaster relief and prevention could be improved ifremote sensing and GIS were incorporated into adisaster system effectively.

1. EXISTING TSUNAMI WARNING SYSTEMS

1.1. United States

The U.S. National Tsunami Hazard MitigationProgram (NTHMP) (Bernard, 2005) is a state/federalpartnership created to reduce tsunami hazards alongU.S. coastlines. Established in 1996, the NTHMP co-ordinates the efforts of five Pacific states, namely,Alaska, California, Hawaii, Oregon, and Washing-ton, with the three federal agencies of NOAA (Na-tional Ocean and Atmosphere Agency), FEMA(Federal Emergency Management Agency), andUSGS (US Geological Survey Bureau). In theprogram:

(1) Deep ocean tsunami data and numerical mod-els are established for tsunami forecasting;

(2) A seismic network enables the tsunami warn-ing centers to locate and size earthquakesfaster and more accurately;

(3) Twenty-two tsunami inundation maps havebeen produced, which cover 113 coastal com-munities with a population at risk of over amillion people; and

(4) A tsunami-resilient communities program hasbeen initiated through awareness, educa-tion, warning dissemination, mitigation in-centives, coastal planning, and constructionguidelines.

The significant impact of the program wasdemonstrated on November 17, 2003, when anAlaskan tsunami warning was canceled because real-time, deep ocean tsunami data (amplitude as small as1 cm under 6000 m of water) indicated the tsunamiwould be nondamaging (Bernard, 2005). Cancelingthis warning averted an evacuation in Hawaii, avoid-ing a loss in productivity valued at $68 M.

1.2. Japan

Since 1952, Japan has been developing its Pa-cific tsunami warning system. Around Japan there isa network of 300 seismic sensors, including 80 on theseabed, which record seismic activity and water pres-sure around the clock. Tsunamis are predicted basedon the location, strength, and depth below the sur-face of a quake. A tsunami alert is issued on radio

channels and TV screens within 3 minutes for themost at-risk areas, giving a 10-minute warning to al-low residents to evacuate these areas.

1.3. International Tsunami Information Center

Under the auspices of the IntergovernmentalOceanographic Commission, an International Coor-dination Group for the Tsunami Warning System inthe Pacific was established in 1968, and the Interna-tional Tsunami Information Center (ITIC) was es-tablished in 1965. Both the ITIC and the U.S. Pa-cific Tsunami Warning Center are located in Hawaiiand are hosted by the National Weather Service. TheITICs responsibilities include:

(1) Providing information about tsunami warningsystems;

(2) Assisting the establishment of national warn-ing systems and improving preparedness fortsunamis for all nations throughout the PacificOcean; and

(3) Fostering tsunami research and its applicationto prevent loss of life and damage to property.

1.4. Improvement

The current tsunami warning systems have beeninstalled quite well, but there are still several impor-tant issues that need to be addressed.

1.4.1. Understanding Nature

We still do not understand nature sufficientlywell for absolutely reliable prediction of tsunamis.Therefore, we need to continue to modernize warn-ing systems and continually check all data. For exam-ple, we know that some small earthquakes could gen-erate very powerful tsunamis. However, on March29, 2005, the Sumendala Indonesia quake, which reg-istered at 8.7 Ms, caused a small tsunami and failedto trigger a destructive disaster like that on Decem-ber 26, 2004 in the Indian Ocean.

1.4.2. Short Response Time

Although a 10-minute warning is the hoped-forresponse time, a quake closer to the shore couldcut that time dramatically. In a worst-case scenario,according to projections in 2003 by Japans CentralDisaster Management Council, three simultane-ous quakes could generate a magnitude of 8.7,

Improving Tsunami Warning Systems 3

killing 28,000 people, including nearly 13,000 whoselives would be claimed by the resulting tsunami(http://english.aljazeera.net/NR/exeres/4C4B0D4C-8598-4000-94F2-4961857C619E.htm).

1.4.3. No System

Some poor Asian countries have no tsunamiwarning system. The sudden and enormous loss of300,000 lives and property in the December 26, 2004tsunami affected numerous countries across differentcontinents.

GIS and remote sensing have already beenwidely used in monitoring, predicting, early warning,loss assessment, rescue activities, and aid crisis man-agement for natural disasters. The effectiveness andusefulness of the technologies could be enhanced ifthey were incorporated into an optimally designeddisaster management system.

2. ROLE OF REMOTE SENSING IN THEDECEMBER 26, 2004 INDIANOCEAN TSUNAMI

Remote sensing has been widely used in the De-cember 26, 2004 tsunamis detection, mapping, mon-itoring, damage assessment, and crisis management,and would play a large role in an effective warningsystem due to its wide and insightful eyesight inobserving similar catastrophic disasters. Examples ofremote sensing data and software for the December26, 2004 tsunami are presented below.

2.1. Remote Sensing Data

(1) SPOT (Systeme Pour lObservation de laTerre) and IKONOS image were used to de-tect the progression of tsunami waves and sed-iment change, and to obtain damage details.

(2) MODIS (moderate resolution imaging spec-troradiometer) band 1 (0.620.67 mm) andband 2 (0.8410.876 mm) (250-m spatial res-olution) were used to map coastal flooding.

(3) MODIS band 2 was resampled to 500 m forstudying sediment load.

(4) MODIS-Terra data were acquired less thanthree hours after the earthquake, and aspecial emphasis was made to understandtrails during propagation and after impact oncoastlines.

(5) NDVI (normalized differential vegetation in-dex) ratios were used to map tsunami-affectedareas.

(6) Landsat, RS-1C, 1D, Oceansat-1, and Re-sourcesat in India were used for analysis anddamage assessment.

(7) Aerial photography was used by ISRO (In-dian Space Research Organisation) to assessthe state of agricultural areas, coastal vegeta-tion, and beaches.

(8) Orbimage, DigitalGlobe, and Space Imag-ing provided the U.S. military and NationalGeospatial-Intelligence Agency with imageryof affected areas for the U.S. Agency of In-ternational Developments Office of ForeignDisaster Assistance on a daily basis.

2.2. Software

Two examples of software used in the analysisof remote sensing data for the December 26, 2004tsunami are presented and we suggest here that GISand remote sensing software tools are used togetherto enhance the analysis and efficiency of informationprocessing.

(1) ENVI (Environment for Visualizing Images)is a software tool that was used to processand analyze various remote sensing data rele-vant to the tsunami for a survey of the precur-sor environment before the tsunami and lossassessment. ENVI provides a flexible frame-work to integrate the functionality of a GISsuch as ARCGIS.

(2) ARCGIS is a GIS software that was used togenerate a vector layer of the inundated areaand the inundated area statistics were com-puted from this layer. It was also used to helpmonitor the hazards and make temporal as-sessments as a guide for logistical services, andhelp model the hazards as an important datasource.

3. DISASTER SYSTEM AND MODELING

To investigate quantitatively the potential of re-mote sensing and GIS in tsunami impact relief, thedisaster system (Section 3.1) is decomposed into sev-eral components. There are many studies of the com-ponents (Petak & Atkinson, 1982; Shi, 1996; Haimes,2008) and the equations of the essential compo-nents are presented in what follows: magnitude of the

4 Wang and Li

Human activities

Physical processes MS Wave I

ocean shoreFig. 1. Hazard propagation (twohorizontal arrows) in space and time andimpacts on humans (two vertical arrows).The two vertical arrows indicate thelinkage between physical processes andhuman activities, which, like valves, allowthe physical impacts on humans to bemodified by hard and soft interventions(see Fig. 2). MS stands for the magnitudeof the earthquake and I refers to theintensity of the wave after landing.

earthquake (the source of tsunami, Section 3.2), in-tensity of the tsunami (Section 3.3), human society(Section 3.4), and the loss caused by the interactions(bonds) between the hazard and humans (Sections3.5 and 3.6). The relationships presented are prob-ability models and they are spatially explicit or im-plicit. Section 4 highlights the nodes of the disastersystem where remote sensing and GIS can be em-bedded to improve the performance of the system.Because our objective is not to provide an integra-tive software tool or package for modeling the dis-aster system but, rather, to assess GIS and remotesensing techniques used in a tsunami warning system,the above theories are simplified (Section 5) and thenused in calculations.

3.1. Disaster System

Natural hazards impact human society throughthe bond between humans and nature. Humans, na-ture, and the bond form a disaster system. Fig. 1shows the disaster system: the top layer representshumans and the bottom layer represents nature. Inthe case of a tsunami, an earthquake causes wavepropagation that forms a destructive tsunami thatpotentially affects humans on coastlines. As can beseen in this figure, there is a link between humansand the natural hazard (tsunami). Logically, losses

System: Society Input: Hazard Output: Loss

Dissipate energy

Monitor

Early warning

Monitor

Forecasting

Monitor

Evaluation

Robustness Relief

Soft intervention

Surveillance

Hard intervention

Chain

Fig. 2. Disaster system.

could be reduced or eliminated if the link is bro-ken, by displacing humans permanently from the riskarea, which could be identified by remote sensing andGIS analysis and records of hazards in history, or bytemporal retreat based on professional hazard fore-casting, which could be facilitated by remote sens-ing and GIS, or by expedient escape according to atsunami warning implemented with remote sensingsurveillance.

Fig. 2 shows the components of a disaster sys-tem and their relationship. Natural hazards causinghuman and economic losses could be remarkablymitigated by human intervention. Dissipation mea-sures (hard intervention) refers to planning floodingzones, releasing rock stress, planting slopes, etc. Asociety becomes more robust by enhancing buildingsto resist earthquakes, relocating houses to noninun-dation zones to avoid flooding, providing disaster ed-ucation, and practicing preparation exercises. Reliefactions include rescuing buried victims and providingfood, water, tents, and medical and sanitary aid. Inaddition to these concrete and hard measures, lossescould also be significantly reduced by soft interven-tion, i.e., monitoring, forecasting, early warning, andevaluating the state and processes of the system com-ponents.

We modeled the disaster system to evaluate theefficiency of remote sensing and GIS as components


of the system in disaster relief through their roles inforecasting an event in advance, tracing the progres-sion of the disaster and evaluating the hazards spa-tial impact. Their roles could be further enhanced bythe design of an integrated disaster system.

3.2. Magnitude of Hazard: MS

As much information and as many indicators aspossible are gathered to improve the accuracy andprecision of the hazard forecast. These factors, or rel-evant phenomena, are integrated into a probabilistic,summarized as:

MS0(k) P1(X(i1,1)) P2(Z1(i2,3)),where MS0 = the magnitude of the earthquake at itsepicenter k, X() are the factors in the location set{i1} that induce MS0 with a probability P1, and Z1()is a detectable phenomenon at a location set {i2} re-lated with X or MS0 by a probability P2, such asthe abnormally shaped cloud that was found hover-ing over Sumatra before the quake, the epicenter ofthe December 26, 2004 Indian Ocean tsunami earth-quake, animal evacuation prior to the hazard, andclusters of rupture movements prior to a major earth-quake. The time of earthquake occurrence is givenas zero, and 1 > 0 and 3 > 0 are times ahead ofthe major quake. The symbol denotes probabilityforecasting and denotes conjunction. Remote sens-ing may sense some of X(i1, 1) and Z(i2, 3) atplaces {i1} at 1 and {i2} at 3, respectively.

Only some types of earthquakes () causetsunamis:

MS(k) = MS0(k) m,where m is a switch function, equal to 1 if the MS0is a tsunami earthquake, 0 otherwise, based on ob-servations and mechanistic assumptions (some rele-vant phenomena). MS is what we are interested in.The significance of distinguishing MS0 and MS is thattsunami prediction based on MS0 sometimes leads toa false warning. Unfortunately, there is currently lit-tle knowledge about (McCloskey et al., 2005); con-sequently, it is not known if should trigger the warn-ing system to switch on or off once an earthquake isdetected, although there is a statistical relationship(Kulikov et al., 2005). This situation may be over-come if more is understood about the mechanism bywhich earthquakes trigger tsunamis and if we havemore sensors on both the bottom and the surface ofoceans.

3.3. Intensity of Hazard: I

IS( j, 5) = g(MS(k), d(k, j)) Z2(5);5( j) = d(k, j)/v;

where IS(j, 5) is the intensity of a tsunami at theshore site j, i.e., the height and speed of the waveof the tsunami, d(k, j) is the distance between thequake epicenter and the front of the wave, g is afunction mapping the quake magnitude, distance,and geomorphical features to the height and speedof the wave, Z2( 5) is a natural phenomenon thataccompanies IS(t + 5), such as suspended sedimentconcentration in the ocean, 5 is the time elapsedfrom the quake (MS) outbreak, and v is the spreadvelocity of the tsunami wave. Koshimura et al.(2001), Pelinovsky et al. (2002), and Sato et al.(2003) used a deterministic approach and Geistand Parsons (2006) employed probabilistic analy-sis to specify the formula and parameters. Therewas remote sensing of Z2 after the wave landed(http://www.ias.ac.in/currsci/mar102005/709.pdf), anda 10-m-resolution SPOT4 image shows un-usually large waves off the coast of Thailand(http:/ /www.spotimage.fr/html/ 167 240 241 781 .php). Water from all inlets J along the shore floodedand accumulated at site i on land, IS in the sea trans-ferred its form to IL, which cost lives and damagedproperty:

IL(i, 5) =

Jh(IS( j, 5), w(i, j)) dj Z3(5),

in which IL(i, 5) is the depth, area, and durationof flooding. The expression Z3( 5) is a phenomenonthat accompanies IL(i, 5) such as moisture changeson land. The time delay from a tsunamis arrival ata coastline j to anywhere i on land is very short, soit is omitted here. The term w(i, j) is the pathwayfrom the shoreline inlet j to a site i on land. Thereis a precise relationship between wave height h onthe shoreline and hazard IL on land in hydrology(Matsutomi et al., 2001; Choi et al., 2002), which couldbe easily acquired from remote sensing and GIS andis a prerequisite of a detailed digital elevation model,and land cover over area {i} just before the arrival ofa tsunami.

3.4. Human Society: H

Human society, composed of movable and im-movable objects, under a hazard risk, is denoted bya human risk matrix H:

6 Wang and Li

H(z, m + n | )

=

h1,1 . . . h1,m h1,m+1 . . . h1,m+n

h2,1 . . . h2,m h2,m+1 . . . h2,m+n

. . . . . . . . . . . . . . . . . .

hz,1 . . . hz,m hz,m+1 . . . hz,m+n

= H(M; N | ),

where hz ,k refers to the number of the kth type hu-mans or properties distributed over z-level intensityof a risk zone, and z = 1, 2, . . . , z. There are m typesof movable objects and n types of immovable prop-erties, and the matrix H accordingly is partitionedinto two parts: movable M (left part of the matrix)and immovable N objects (right part of the matrix),which have different vulnerabilities and calculations.The movable objects M (humans, animals) and im-movable property N (houses, crops) react differentlyto condition , i.e., the knowledge of hazards, gov-ernment regulations, and warnings. The various re-actions to lead to various distributions and adop-tions of human society H(M, N) just before a hazarddescends.

3.5. Vulnerability: V

Vulnerability of human society (H) to hazard in-tensity (I) is an empirical data matrix V:

V(z, m + n)

=

v1,1 . . . v1,m v1,m+1 . . . v1,m+n

v2,1 . . . v2,m v2,m+1 . . . v2,m+n

. . . . . . . . . . . . . . . . . .

vz,1 . . . vz,m vz,m+1 . . . vz,m+n

= V(M; N),

where vz ,k is the percentage loss of a kth type of ob-ject under intensity z. V is obtained by empirical re-gression (Sonmez et al., 2005; Badal et al., 2005; An-derson, 2003; Van Der Voet & Slob, 2007) or en-gineering experiment. We used observed losses toregress the estimated height of the wave over spaceto evaluate the values in this article.

3.6. First Loss from Hazard Arrival to Its PhysicalEnd: L0

By multiplying the vulnerability matrix and hu-man risk matrix, we obtain the first loss matrix:

L0(M; N | ) = V(M; N) H(M; N | )

(0; 0) 6 : H(0; 0),prediction,knowledge, andregulation

(v(M)h(M); v(N)h(N)) 5 : H(M; N),optimalwarning,knowledge,no regulation

(v(M)h(M); v(N)h(N)) 4 : H(M; N),early warning,no knowledge

(v(M)h(M); 0) 3 : H(M; 0),no warning,knowledge,regulation

(v(M)h(M); v(N)h(N)) 2 : H(M; N),no warning,knowledge,no regulation

(v(M)h(M); v(N)h(N)) 1 : H(M; N),no warning,no knowledge

in which L0(M; N | ) is the loss of movable objectsM and immovable property N. Knowledge of risk induces hazard-resistant infrastructure and build-ings, N < N, where is the percentage of dam-age; government regulations on hazard managementlead to both human and property displacement awayfrom the risk zone; risk warning promotes the re-treat of movable objects. In condition 6, there isa prediction of hazard. Therefore, the movable ob-jects M should have enough time for retreat so thatnone are lost, and the knowledge of hazards and gov-ernment regulations guarantee that immovable prop-erty will be built outside of risk zones so that no im-movable property will be lost either. Under the 5condition, there are no government regulations, andsome real estate is built inside risk zones, althoughbuildings may be enhanced to resist hazard to some


degree (0 < < 1) according to hazard knowledge. Athorough use of remote sensing and GIS and a high-efficiency disaster management system enables themaximum retreat of moveable objects, and the lossis M, where 0 < < 1. In condition 4, there isno knowledge of hazards or regulations concerninghazards, and the real estate in risk zones is vulnera-ble to hazards. However, early warning gives peoplesome time to escape the risk zones, and the loss isM, 0 < < 1 and < . In condition 3, there isno early warning of hazards, people are completelyunaware of the coming catastrophe, and are vulnera-ble to the hazard v(M)h(M), whereas the knowledgeand regulation of hazards is such that the real estatehas been built outside the risk zones (Robert et al.,2003). In condition 2, there is no warning, so peopleare fully vulnerable v(M)h(M) to the coming hazard.There are also no government regulations, so real es-tate has been built within the risk zones, although theloss may be modified by the owners knowledge of acoming hazard v(N)h(N). The worst case is 1, inwhich there is no warning and no foreknowledge, sothere are no regulations governing hazards, and bothhumans and their property are completely vulnerableto disaster.

Obviously,

6 : L0(0; 0) < 5 : L0(M; N) < 4 : L0(M; N);

3 : L0(M; 0) < 2 : L0(M; N) < 1 : L0(M; N),

where denotes 6: L0(0, 0) < 3: L0(M; 0), et al.In more detail,

L0(z, m + n | ) = V(z, m + n) H(z, m + n | )

=

v1,1h1,1 . . . v1,mh1,m v1,m+1h1,m+1 . . . v1,m+nh1,m+1

v2,1h2,1 . . . v2,mh2,m v2,m+1h2,m+1 . . . v2,m+nh2,m+1

. . . . . . . . . . . . . . . . . .

vz,1hz,1 . . . vz,mhz,m vz,m+1hz,m+1 . . . vz,m+nhz,m+n

m+nj=1

v1, j h1, j

m+nj=1

v2, j h2, j

. . .m+nj=1

vz, j hz, j

zi=1

vi,1hi,1 . . .z

i=1vi,mhi,m

zi=1

vi,m+1hi,m+1 . . .z

i=1vi,m+nhi,m+n

zi=1

m+nj=1

vi, j hi, j

.

According to this matrix, loss from a disaster can beevaluated on three scales:

Scale 1 :vz,mhz,m, loss of mth type object in zone z;Scale 2 :

zi=1 vi,mhi,m, loss of mth type object in entire

hazard area;m+n

j=1 vz, j hz, j , loss over zth levelrisk zone; and

Scale 3 :z

i=1m+n

j=1 vi j hi j ,total loss within thedisaster area.

If loss on scale 1 is found, the losses on lev-els 2 and 3 are calculated by summing losses ofscale 1. Losses on levels 3 or 2 may be approx-imately estimated by sampling techniques (Wanget al., 2002) without data at lower levels. Further-more, with downscaling techniques and prior data,losses on levels 2 or 1 may be estimated once weknow the losses on levels 3 or 2, respectively.

3.7. Secondary Loss from the Physical Endof a Hazard to the End of Relief Action: L0

The losses of some movable and unmovable ob-jects increase even after the physical hazard processreaches an end if the relief actions are not immedi-ate and efficient (Sever et al., 2006; Slotta-Bachmayr,2005; Kuwata & Takada, 2004). Fig. 3 illustratesschematically an empirical relationship, where 2 de-notes the delay from hazard arrival to the beginningof relief actions, i.e., the time to identify the locationand seriousness of a disaster, and the time to accessthe target area from the resource site. Remote sens-ing speeds the identification of the target, thus short-ening 2. The expression 4 denotes the delay fromthe time relief efforts begin to the time of rescue. Thevalue (>0) is an empirical parameter, reflecting theefficiency of relief, and is related to the equipment,

8 Wang and Li

Relief delay (2 + 4)

L

L0

L0

Lmax

Fig. 3. Schematic relationship between disaster loss and reliefdelay (Fu & Chen, 1995).

experience, size of the rescue action, and food andmedical aid. The higher the efficiency, the smaller is, which is expressed by:

L(M; N | 2, 4; ) = L0(M; N | ) [1 + (1 e(2+4))].

If the relief actions are totally inefficient, taking 2 +4 in the above equation, the maximum lossesafter the physical end of a hazard are:

Lmax(M; N | ) = L0(M; N | )(1 + )= L(M; N | , 2, 4; ).

The above general model chain from MS to I to Hand finally to L was specified by mechanism, sys-tem dynamo language, stochastic process, or Tay-lor expansion, then calibrated by regression if therewere corresponding observed data. A quick estima-tion was made by a simplification of the modeling andusing empirical data, as described in Section 5 in thisarticle.

time, t

loss, L

space, j

landocean spread, 2h (5)

attack, 10min

reaction delay, 3d (2)

relief, 1mon (4)

prediction (1 or 3)

recovery, 1y

Fig. 4. Schematic time periods oftsunami process.

4. DISASTER SYSTEM OPTIMIZATIONAND REMOTE SENSING AND GIS INPUT

4.1. Efficient Time Interval

Time is the most crucial factor in disaster re-duction: earlier warning allows enough time for re-treat, and timely relief reduces secondary loss. Fig. 4shows the timing of a tsunami process schematically:an earthquake broke out and is denoted by the ori-gin point of a coordinate system composed by spacej, time t ( 1 or 3 denotes the times ahead of thequake of predictions using relevant factors or relatedphenomena, see Section 3.2), and loss L. The quaketriggered a tsunami that took about two hours ( 5)to spread from the origin to shorelines and about afurther 10 minutes to hit humans and the propertydistributed along the coast; it might take about threedays ( 2) to identify and approximately assess thelosses distributed over the widely impacted areas. Inlight of this information, massive and efficient disas-ter relief actions were implemented that lasted aboutone month ( 4); then recovering actions began, in-cluding reconstructing houses, lifelines, and other fa-cilities, and these actions lasted about one year ormore. The daily losses caused by the tsunami startedfrom the time of the quake, increased dramaticallyafter the waves hit the shore, and reached a max-imum; the daily losses continued to the periods ofthe instant relief actions and recovering actions. Thelonger the time odd (=1 1 + 3 3 + 5 5), theless the loss L; the shorter the time even (=2 2 +4 4), the less the loss L. The various values reflectthat the times in different disaster stages have differ-ent importance to the loss L(M, N), and the weightscan be estimated from past disaster records.


Table I. Optimization of Disaster System

Disaster SystemComponents Current Theory Optimization

Monitoring Prior knowledge Sandwich sampling (Wang et al., 2002)Classic statistics Early alert (Ramirez & Carlos, 2004))

Hazard prediction Data-driven single Data-driven mega model (Wang et al., 1996)model

Disaster forecast Vulnerability Scenarios simulation(Li et al., 2005; Wyss, 2005)

Loss evaluation Single hazard Regional integration (Wang et al., 1997)Relief actions Contingency Resources allocation, computer-based planning

(Wang et al., 2001; Simonovic & Ahmad, 2005)

We define a value called the efficient time in-terval, , to represent the efficiency of a disastersystem:

= odd even.Obviously, the larger the values are, the less thelosses. Remote sensing and GIS play crucial roles inproviding timely detailed and correct information ofthe hazard and disaster states. Furthermore, a highlyefficient disaster system would fully utilize remotesensing and GIS functionalities.

4.2. Optimization of Disaster System

We investigated the possibilities of lengtheningthe efficient time interval by optimizing the disastersystem using state-of-the-art theories.

Table I shows current theories used for disastercomponents and possible new optimization technolo-gies to improve them. Nowadays, the surveillancenetwork is distributed by prior knowledge (Office ofScience & Technology of U.S. President, 2005) or byclassical/spatial sampling theories (Cochran, 1977).However, a new spatial sampling theory called thesandwich spatial sampling technique, which quanti-tatively combines both prior knowledge and priorprobability, was proven to be much more efficientthan others in designing a surveillance network, andalso offered several advantages: (1) it combined re-mote sensing surface information (a spectrum) fromthe air and sampling point detailed information onthe ground (confirmation) to improve the surveil-lance efficiency; and (2) it combined fine spatialresolution, but expensive and less frequent images(IKNOIS or QUICKBIRD), and coarse, but fre-quent (SPOT or MODIS), images to produce a newimage with higher resolutions in space and time andspectrum at a lower cost. Some hazards such asearthquakes are too complicated to be modeled, but

thanks to the accumulation of observation, we canuse data-driven mega models to filter out several dis-tinct patterns from the observed data that are of-ten a mixture of various mechanisms and factors un-derlying earthquakes (Wang et al., 1996). With theaccumulation of observed and test data and the im-provement of computer virtual reality techniques,vulnerability analysis (Petak & Atkinson, 1982) isready to be implemented by scenario simulations (Liet al., 2005; Wyss, 2005). Departmental and regionalstudies benefit different groups. If a region is at riskfrom various hazards, people there are interested inan integrative risk assessment when they choose loca-tions for homes, and are also concerned with individ-ual hazards when buying life and property insurance.A technique based on GIS is now available to mapout an integrated zonation that shows the risk at bothindividual and integrated levels (Wang et al., 1997;Peduzzi et al., 2005). The change in secondary lossesto the delay of relief actions varies for living beings,lifeline engineering, real estate, etc., and the compo-sition of the hazard susceptibility varies over space;in addition, the first loss from the first attack of ahazard varies in space; therefore, an optimal and dy-namic arrangement of relief resources and computer-based evacuation emergency planning will maximizethe reduction of secondary losses (Wang et al., 2001;Simonovic & Ahmad, 2005).

4.3. Remote Sensing and GIS Input

Remote sensing and GIS play a critical role inimproving the performance of disaster systems inforecasting risk, tracing hazards, evaluating loss, andguiding relief actions. To improve the efficiency ofdisaster systems, it is necessary to enhance the per-formance of all of the components, with each compo-nent corresponding to more specific remote sensors.

10 Wang and Li

By investigating the link of every component of adisaster system with remote sensing and GIS (Leung,1997) (Fig. 2) and then running a model chain (Sec-tion 3), we can compare the difference in system per-formance between cases with and without remotesensing and GIS data as input, and compare the util-ity of various resolutions in space and time and thespectrum of images in disaster reduction.

Although the task of specifying and calibratingall models in a disaster system remains a challengefor scientists in many fields, it is possible to evalu-ate quantitatively the role of remote sensing and GISdata in disaster reduction through the efficient timeinterval , which directly reflects the essential char-acteristics of remote sensing and GIS in providingtimely correct and detailed spatial information of adisaster before, during, and after an event.

Remote sensing and GIS lengthen odd andshorten even, and hence expand , which leads toa rise in evacuation rate of movable objects and,

Fig. 5. Pilot area and risk zones of Banda Aceh, Indonesia (map source: QuickBird, Landsat 7, ETM and SRTM; partial data source:ECJRC, 2005; AIT, 2005; CGI, 2005).

accordingly, a reduction of losses. The argument iscalibrated in the following section.

5. THREE SCENARIOS

Fig. 5 shows the pilot area of the primary im-pact zone (PIZ) in Banda Aceh, Indonesia, the placethat received the greatest impact of the December 26,2004 tsunami. About 30,000 people died and 29,545houses collapsed or were badly damaged inside thePIZ according to a later survey and statistics (CGI,2005). The PIZ can be further divided into threezones with different risk levels of both intensity ofwave (I in Section 3.3) and population density (hu-man society H in Section 3.4), ranging from high riskclose to the seashore to low risk farther away. Therisk zones boundary is comprehensively determinedby the texture and color of the remote sensing im-ages after the disaster (L0 in Sections 3.6 and 3.7,but accounted for by post hoc survey) (QuickBird,


0 1 2 3 4 5

0

1

2

3

4

5

6

distance from the beach (km)

wav

e he

ight

(m

)

wave height

Fig. 6. Tsunami wave height in Banda Aceh, Indonesia, December26, 2004. (Data sources: UNEP, 2005; Ahmet et al., 2004; Tierney,2005; Keith, 2005; Helen, 2005; Quirin, 2005; ADB, 2005.)

Landsat 7, ETM and SRTM; partial data source forthis maps productionECJRC, 2005; AIT, 2005;CGI, 2005).

We simulated three damage scenarios for theDecember 26, 2004 Indian Ocean tsunami disasterusing three assumptions: (1) scenario 1, without anearly warning system, i.e., the current situation, 1 inL0(M; N | ) described in Section 3.5; (2) scenario 2,with a system like the existing Pacific Ocean tsunamiwarning system, 4 in L0(M; N | ) in Section 3.5;and (3) scenario 3, with an optimized disaster systemfully using remote sensing and GIS as indicated inTable I, 5 in L0(M; N | ) in Section 3.5.

Fig. 6 shows the intensity of the tsunami as it pro-gresses (intensity I in Section 3.3). The intensity ismeasured by wave height: the higher the wave, themore destructive power it has. The wave height de-creases as the distance from the shoreline increases.It took 10 minutes for the tsunami to travel from theearthquake epicenter to the PIZ shoreline, and 20minutes to reach the entire PIZ. The curve is basedon the actual sample measurement of the tsunamithat occurred in Banda Aceh, Indonesia, December26, 2004 by a field survey and reports (UNEP, 2005;Ahmet et al., 2004; Tierney, 2005; Keith, 2005; Helen,2005; Quirin, 2005; ADB, 2005) and spatial interpo-lation (Atkinson, 2005; Penning-Rowsell et al., 2005).

Fig. 7 establishes the relationship of numbers ofdeaths to the wave height of the tsunami (vulnera-bility V in Section 3.5) through regression using ob-servations. Due to the decrease in wave height withincreasing distance from the shore, the number ofdeaths is positively correlated to the wave height andnegatively correlated to the shore distance. After hit-

0 1 2 3 4 5 60

2000

4000

6000

8000

10000

12000

14000

16000

18000

Dea

ths

(per

sons

)

Wave height(m)

Fig. 7. Tsunami wave height vs. observed deaths in Banda Aceh,Indonesia, December 26, 2004. (Data sources: UNEP, 2005;Ahmet et al., 2004; Tierney, 2005; Keith, 2005; Helen, 2005; Quirin,2005; ADB, 2005.)

1950 1955 1960 1965 1970 1975 19800

200

400

600

800

1000

1200

1400

1600

0

10

20

30

40

50

60

70

80

90

100

Dam

age

(bill

ion

yen

)D

eath

(per

son

s)

Timeline (year)

Death (persons) by flooding Percentrate rate o

f TV

(%)

TV population rate

Typhoon movement forecast started in 1953

Computer-based forescast started in 1959

Usage of GIS and RS started in 80s

Fig. 8. Technology development and disaster reduction (statisticson flooding, Hirok, 2005; Ramirez & Carlos, 2004).

ting the shore, the power of the wave decreases withincreasing distance from the shore due to the ham-pering effect of the shore and reduction of the kineticenergy that results in less destructivity and thus fewerdeaths. Fig. 7 also shows the number of deaths in-creases gradually with increasing wave height up to3 m, above which a rapid increase in deaths occurs.

Fig. 8 is adapted from Hiroki (2005). Based onstatistics, the figure indicates that each use of newtechnology (from the 1953 typhoon forecast systemto 1959 computer-based forecasting to the 1980s GISand RS data) were critical points at which the loss(deaths caused by flooding) dropped drastically. Thehistoric assessment of the contributions of the tech-niques to rescue efficiency (Fig. 8), together with

12 Wang and Li

0 10 20 30 40 500.0

0.2

0.4

0.6

0.8

1.0

0.0

0.2

0.4

0.6

0.8

1.00 2 4 6 8 10

Eva

cuat

ion

Rat

e

Scenario1

Scenario2

Scenario3

Efficient Time Interval (minutes)

Fig. 9. Evacuation rates increase withincrease of efficient time interval inthree scenarios (partial sources: Kenzo,2005; AIT, 2005; Yamazaki, 2001;ECJRC, 2005).

other simulated results (AIT, 2005; ECJRC, 2005)and the extrapolation of this trend, demonstratesthe effectiveness and contribution of GIS and re-mote sensing technologies, and also indicates the po-tential contribution of optimal technology proposedby this article. From Fig. 8, the timing efficiencycould be improved by a factor of about 12 with thewarning system and by a factor of about 34 withthe warning system optimized with GIS and remotesensing techniques. The values of these improve-ments (Fig. 8) were used to simulate the evacuationrates corresponding to the critical timing points un-der scenarios 2 and 3, whereas scenario 1 is the fac-tual situation of the 2004 Indonesian tsunami fromsurveys.

The loss (L0 in Section 3.6) is negatively pro-portional to the evacuation rates of objects, whereasthe latter is positively proportional to . As for mov-able humans, the evacuation rates (including time co-efficients) in Fig. 9 were firstly acquired at criticalpoints from the arrival of the tsunami to 45 minuteslater from survey reports (Kenzo, 2005; AIT, 2005;Yamazaki, 2001; ECJRC, 2005), then the points wereregressed and extrapolated to produce curves, con-sidering Fig. 8, GIS and remote sensing applicationexperience, human physical potentials, and the localtransportation situation. We acknowledge that thereexists some arbitrariness in the estimation of the dis-aster parameters due to the nonrepeatability of theevent that cannot be experimentally investigated, butthe uncertainty is mitigated by Fig. 8 and studies

by Kenzo (2005), AIT (2005), Yamazaki (2001), andECJRC (2005).

From Fig. 9, it can be seen that without anywarning system (scenario 1), the evacuation is slowand ineffective, and there were many people whohad still not been evacuated 50 minutes after thetsunami caused the most damage (as the scenario1 plot indicates). This was the actual condition inthe Banda, Aceh disaster. With a warning systemsimilar to the existing Pacific tsunami system, theevacuation is effective. Under this scenario, after 15minutes, the evacuation rate starts to increase until35 minutes, at which time all people have been evac-uated. The damage from the tsunami is assumed tobe partial and moderate in this scenario. Further-more, the plot for scenario 3 indicates the evacuationis timely and effective and all components relevantto before, during, and after the disaster in the disas-ter system were optimized as shown in Table I usingGIS and remote sensing as inputs. Ten minutes afterthe tsunamis occurrence, the rescue activity wouldbe initiated (Ramirez & Carlos, 2004). We assumethat people can be completely evacuated, with fewlosses, within 20 minutes under this scenario. In sum-mary, the proportions of evacuation rates under thethree scenarios 1, 2, and 3 corresponding to the effi-cient time interval are roughly 0, 0.5, and 1, respec-tively.

With the above parameters, we projected disas-ter scenarios in three distinct technical conditions.The results are summarized in Table II. The loss rate


is determined by the vulnerability and the amountof property (including humans and houses). The hu-man loss rate and loss amount decreases from sce-narios 1 to 3 and from A (high risk), to B (midrisk) to C (low risk). The housing loss rate has slightdifferences in that the effects of are not consid-ered. Therefore, the housing loss risk is the samein each risk zone, but decreases from A to B toC under each scenario. Each risk zone correspondsto z in the equation H(z, m + n | ) in Section3.5. In summary, there are four risk levels and soz = 4. Here, only humans are considered as mov-able objects and hence m = 1. Similarly houses areincluded in the immovable object group, and so n= 1. corresponds to three different conditions(i.e., the scenarios in the first rows of Table III,scenarios 13): no knowledge and no warning, knowl-edge and warning, and knowledge and warning withoptimal technologies using GIS and remote sens-ing as inputs; then, the losses were estimated by theequation L0(M; N | ) = v(M; N) h(M; N | ) inSection 3.5.

5.1. Scenario 1

We first estimated the rates of the number ofdead and missing persons among the total populationin the PIZ.

Let Lz be the number of deaths in a certain zonez, hz be the number of residents in z, and Iz the in-tensity of the disaster in z. For simplification of thesimulation, we assume the vulnerability of human so-ciety to a tsunami is proportional to the hazard in-tensity. This assumption is rational because a very in-tense hazard results in much greater loss (Shi, 1996).Therefore, we use the following equation to simplifythat in Section 3.5 for our simulation:

Lz = vz hz = g Iz hz,where vz is the vulnerability of human society to atsunami under various circumstances of (Section3.6), and is proportional to Iz by a coefficient .The vulnerability coefficient involves the intensity(wave height) and the vulnerability of human societyunder different technical conditions (scenarios) andhelps build the relationships among intensity (Iz),death (the human loss, Lz), and population (hz).

For hz, an even distribution of the populationover the PIZ is assumed:

hA = h (1.8/4.7), hB = h (1.5/4.7),hC = h (1.4/4.7),

where h is the total number of residents in the PIZ,and 1.8 + 1.5 + 1.4 = 4.7 km are the mean widths ofA, B, and C and the PIZ measured in Fig. 5.

For Iz, the intensity of a tsunami is described byits wave height, and is obtained from Fig. 6:

IA = 4.5 6m, IB = 2 4.5m, IC = 0 2m.Consequently,

LA : LB : LC = 18 (4.5 6) : 15 (2 4.5) : 14(0 2) 55% : 30% : 15%.

For houses, the proportion of collapsed struc-tures between the zones is assumed to be similar tothe human death proportion.

The proportions and total deaths M and thenumber of collapsed and seriously damaged houses Nin the PIZ were estimated for each of the risk zonesin scenario 1, as listed in Table II.

5.2. Scenarios 2 and 3

From the curves in Fig. 9, the proportions of theloss rates under the three scenarios are found to be10:5:1 or 1:0.5:0.1 (the evacuation rate is 1:5:10) formovable humans. For immovable real estate, the re-lationship is simply 1:1:1 due to s limited effectupon the loss of houses; i.e., there are similar lossrates under different scenarios. Using the models de-scribed in Section 3.6, we estimated the deaths in PIZin scenarios 2 and 3 to be 15,000 and 3,000, respec-tively, based on the survey data of 30,000 deaths inscenario 1. By two largely rational assumptions thatthe proportions of losses between A, B, and C remainunchanged in 1, 4, and 5, and that the losses staythe same for immovable objects under the three tech-nological conditions, we easily obtain the remainingfigures in Table II.

6. CONCLUSIONS AND DISCUSSION

The merits of GIS and remote sensing in disas-ter relief could be enhanced in an optimized disas-ter system. We measured their potential contributionin disaster reduction through an indicator called theefficient time interval by modeling a disaster systemand simulating its performance under three condi-tions. Under the assumption of a tsunami attack withthe same intensity as that of the December 26, 2004Indian Ocean tsunami, the simulations in the PIZin Banda Aceh, Indonesia, disclosed quite differentscenarios for the losses of human life and houses in

14 Wang and Li

Table II. Summaries of Three Damage Scenarios of the Tsunami Under Three Technological Conditions

Scenario 1 Scenario 2 Scenario 3Zones L(M, N | 1) L(M, N | 4) L(M, N | 5)

A Deaths (persons) Rate >55% >55% >55% 0.5 = 27.5% 0.1 = 5.5%

Number >16,500 >8,250 >1,650Houses collapsed and damaged badly Rate >55% 55% 55%

Number >16,249 >16,249 >16,249

B Deaths (persons) Rate 3055% 3055% 3055% 0.5 = 1527.5% 0.1 = 35.5%

Number 9,00016,500 4,5008,250 9001,650Houses collapsed and damaged badly Rate 3055% 3055% 3055%

Number 8,86316,249 8,86316,249 8,86316,249

C Deaths (persons) Rate 1530% 1530% 1530% 0.5% = 7.515% 0.1% = 1.53%

Number 4,5009,000 2,2504,500 450900Houses collapsed and damaged badly Rate 1530% 1530% 1530%

Number 4,4318,863 4,4318,863 4,4318,863

Sum Deaths (persons) Rate 100% 100% 0.5 = 50% 100% 0.1 = 10%Number 30,000 15,000 3,000

Houses collapsed and damaged badly Rate 100% 100% 100%Number 29,545 (80% of BA) 29,545 (80% of BA) 29,545 (80% of BA)

Note: BA refers to Banda Aceh; Rate = L(M; N) in zone z/H(M; N) in zone z.

different conditions: scenario 1 with no warning sys-tem at all, which is the same as the current situation,scenario 2 with a warning system like that currentlyserving in the Pacific Ocean, and scenario 3 under theproposed optimized warning system fully using GISand remote sensing.

The idea of disaster reduction that implementsremote sensing and GIS into an optimized tsunamiwarning system is that the technology shortens thedelay of an early warning and assessment of therisk while lengthening the period for evacuation, andmore accurate information of the location and sever-ity of the disaster is provided. Remote sensing is usedto improve the accuracy of the data and to speed datagathering and information processing. GIS providesa platform to deal with and integrate various geospa-tial data and is used to implement its functionalitieswith other modules. ENVI is used to deal with re-mote sensing data to extract relevant information forstatistical analysis. The probabilistic or mathematicalmodels (e.g., magnitude of earthquake, intensity oftsunami, assessment of primary, and secondary lossescaused by a tsunami and resulting disasters) are im-plemented by Matlab, SPSS, and SAS. In our simula-tion, we obtained Fig. 5 using GIS and RS materialsand maps and estimated the timing efficiency using

GIS and RS (scenario 3). The efficiency of GIS andRS techniques in relief timing can be improved cor-responding to scenario 3 and L(M, N | 5).

The parameters of the procedure were cali-brated, which gave an estimation of the losses forthree different levels of technologies. Using the dataand materials of the Indian tsunami on December 26,2004, a simulation was performed. The factual situa-tion without a tsunami warning system caused a greatnumber of deaths (30,000). The simulation showedthat a warning system similar to that in the Pacificcould halve the number of deaths (15,000; see TableII). However, if equipped with sophisticated GIS, re-mote sensing, and optimization techniques, the warn-ing system, as the simulation shows, could reducenumber of deaths to 3,000.

Our comparison and conclusion have significantimplications for planning, building, and optimizationof a tsunami warning system. The proposed opti-mized disaster system should be equipped with state-of-the-art instruments and advanced theories rele-vant to disasters. GIS and remote sensing could beimplemented into every node of the system fromforecasting, monitoring, and early warning to assess-ing, regulating, and intervening, and this would con-siderably relieve the impact of a tsunami hazard.


ACKNOWLEDGEMENTS

This study is supported by the NSFC (70571076,40471111, 40601077), MOST (2006AA12Z215,2007AA12Z233, 2007DFC20180), and CAS(KZCX2-YW-308).

REFERENCES

ADB (Asian Development Bank). (2005). An Initial Assessmentof the Impact of the Earthquake and Tsunami of December26, 2004 on South and Southeast Asia (report). Availableat http://www.adb.org/Documents/Others/Tsunami/impactearthquake tsunami.pdf.

Ahmet, C., Dogan, P., Sukru, E., et al. (2004). December 26,2004 Indian Ocean Tsunami Field Survey at North Suma-tra Island (a survey report). Available at http://www.ioc.unesco.org/iosurveys/Indonesia/yalciner/htm.

AIT (Asian Institute of Technology). (2005). Working Paperson AITs Response to the Earthquake and Tsunami inSouth and Southeast Asia (a collection of papers). Availableat http://www.tsunami.ait.ac.th/Documents/AIT TsunamiPaper.pdf.

Anderson, L. (2003). Community vulnerability to tropical cy-clones: Cairns, 19962000. Natural Hazards, 30, 209232.

Atkinson, P. (2005). Spatial prediction and surface modeling. Ge-ographical Analysis, 37, 113123.

Badal, J., Vazquez-Prada, M., & Gonzalez, A. (2005). Preliminaryquantitative assessment of earthquake casualties and dam-ages. Natural Hazards, 34, 353374.

Bernard, E. (2005). The U.S. National Tsunami Hazard Mitiga-tion Program: A successful statefederal partnership. NaturalHazards, 35, 524.

CGI (the Consultative Group on Indonesia). (2005). Indonesia:Preliminary Damage and Loss Assessment (The December26, 2004 Natural Disaster) (a survey report). Available athttp://www.undp.org/cpr/disred/documents/tsunami/indonesia/reports/damage assessment0105 eng.pdf.

Choi, B., Pelinovsky, E., Ryabov, I., & Jin, S. (2002). Distribu-tion functions of tsunami wave heights. Natural Hazards, 25,121.

Cochran, W. (1977). Sampling Techniques, 2nd ed. New York:John Wiley & Sons Inc.

ECJRC (European Commission Joint Research Centre). (2005).Preliminary Damage Assessment and Information ReportingDuring the Earthquake and Subsequent Tsunami Disasterin the Indian Ocean (Version of Debriefing Note) (report).Available at http://tsunami.jrc.it/Reports/Tsunami11 Jan10vWebJRC.pdf.

Fu, C., & Chen, X. (1995). Seismic risk zonation of China. In J.Wang (Ed.), Regionalization of Hazards in China. Beijing:China Science & Technology Press.

Geist, E., & Parsons, T. (2006). Probabilistic analysis of tsunamihazards. Natural Hazards, 37, 277314.

Haimes, Y. Y. (2008). Risk Modeling, Assessment, and Manage-ment, 3rd ed. New Jersey: John Wiley & Sons.

He, J., Tian, G., & Wang, J. (1993). Progress of Monitoring andEvaluating Natural Disasters Using Remote Sensing. Beijing:China Sciences & Technology Press.

Helen, P. (2005). Scientists seek action to fix Asias ravagedecosystems. Nature, 433, 94.

Hiroki, H. (2006). Early Warning. Presented at the World Con-ference on Disaster Reduction. Kobe, Japan. Available athttp://www.unisdr.org/wcdr/thematic sessions/cluster2.htm.

Keith, A. (2005). Watching over the worlds oceans. Nature, 434,1920.

Kenzo, H. (2005). Early Warning, Ministry of Land, Infrastructureand Transport, Japan (report).

Koshimura, S., Imamura, F., & Shuto, N. (2001). Characteristics oftsunamis propagating over oceanic ridges: Numerical simula-tion of the 1996 Irian Jaya earthquake tsunami. Natural Haz-ards, 24, 213229.

Kulikov, E., Rabinovich, A., & Thomson, R. (2005). Estimation oftsunami risk for the coasts of Peru and northern Chile. NaturalHazards, 35, 185209.

Kuwata, Y., & Takada, S. (2004). Effective emergency transporta-tion for saving human lives. Natural Hazards, 33, 2346.

Leung, Y. (1997). Intelligent Spatial Decision Support System.Berlin and New York: Springer Verlag.

Li, L., Wang, J., & Wang, C. (2005). Typhoon insurance pricingwith spatial decision support tools. International Journal ofGeographical Information Science, 19(3), 363384.

Matsutomi, H., Shuto, N., Imamura, F., & Takahasgi, T. (2001).Field survey of the 1996 Irian Jaya earthquake tsunami inBiak Island. Natural Hazards, 24, 199212.

McCloskey, J., Nalbant, S., & Steacy, S. (2005). Earthquake riskfrom co-seismic stress. Nature, 434, 291.

Office of Science & Technology of U.S. President, Execu-tive Office of the President. (2005). Strategic Plan forthe U.S. Integrated Earth Observation System. Available athttp://www.ostp.gov/html/EOCStrategicPlan.pdf.

Peduzzi, P., Dao, H., & Herpld, C. (2005). Mapping disastrous nat-ural hazards using global datasets. Natural Hazards, 35, 265289.

Pelinovsky, E., Kharif, C., Riabov, I., & Francius, M. (2002).Modelling of tsunami propagation in the vicinity of theFrench coast of the Mediterranean. Natural Hazards, 25, 135159.

Penning-Rowsell, E., Floyd, P., Ramsbottom, D., & Surendran, S.(2005). Estimating injury and loss of life in floods: A deter-ministic framework. Natural Hazards, 36, 4364.

Petak, W., & Atkinson, A. (1982). Natural Hazard Risk Assess-ment and Public Policy Anticipating the Unexpected. Berlin:Springer-Verlag.

Quirin, S. (2005). On the trail of destruction. Nature, 433, 350353.Ramirez, F., & Carlos, P. (2004). The local tsunami alert sys-

tem [SLAT]: A computational tool for the integral man-agement of a tsunami emergency. Natural Hazards, 31, 129142.

Robert, B., Forget, S., & Rousselle, R. (2003). The effectivenessof flood damage reduction measures in the Montreal region.Natural Hazards, 28, 367385.

Sato, H., Murakami, H., & Kozuki, Y. (2003). Study on a simplifiedmethod of tsunami risk assessment. Natural Hazards, 29, 325340.

Sever, M., Vanholder, R., & Lameire, N. (2006). Management ofcrush-related injuries after disasters. New England Journal ofMedicine, 254, 10521063.

Shi, P. J. (1996). Theory and practice of disaster study. Journal ofNatural Disasters (in Chinese), 5(4), 617.

Simonovic, S., & Ahmad, S. (2005). Computer-based model forflood evacuation emergency planning. Natural Hazards, 34,2551.

Slotta-Bachmayr, L. (2005). How burial time of avalanche victimsis influenced by rescue method: An analysis of search reportsfrom the Alps. Natural Hazards, 34, 341352.

Sonmez, F., Komuscu, A., Erkan, A., & Turgu, E. (2005). An anal-ysis of spatial and temporal dimension of drought vulnerabil-ity in Turkey using the standardized precipitation index. Nat-ural Hazards, 35, 243264.

Tierney, K. (2005). Effective Strategies for Hazard Assessmentand Loss Reduction: The Importance of Multidisciplinary and

16 Wang and Li

Interdisciplinary Approaches (Report from Internet).Available at http://www.ncdr.nat.gov.tw/chinese/download/paperTierney/pdf.

UNER (United Nations Environment Programme). (2005). Afterthe Tsunami: Rapid Environmental Assessment. Availableat http://www.unep.org/tsunami/reports/TSUNAMI reportcomplete.pdf.

Van Der Voet, H., & Slob, W. (2007). Integration of probabilis-tic exposure assessment and probabilistic hazard characte-rization. Risk Analysis, 27, 351371.

Wang, J. F. (Ed.). (1993). Methodology for Assessing Natural Dis-aster Risk in China. Beijing: China Science & TechnologyPress.

Wang, J. F., & Li, Q. L. (1996). Adaptive structure model of earth-quake trend regionalization. Earthquake Research in China,12, 7888.

Wang, J. F., Liu, C. M., Wang, Z. Y., & Yu, J. J. (2001). A marginalbenefit model for distributing water resources in space. Sci-ence in China (Series D), 31(5), 422427.

Wang, J. F., Liu, J. Y., Zhuan, D. F., Li, L. F., & Ge, Y. (2002).Spatial sampling design for monitoring cultivated land. Inter-national Journal of Remote Sensing, 23(2), 263284.

Wang, J. F., Wise, S., & Haining, R. (1997). An integrated region-alization of earthquake, flood and drought hazards in China.Transactions in GIS, 2(1), 2544.

Wyss, M. (2005). Human losses expected in Himalayan earth-quakes. Natural Hazards, 34, 305314.

Yamazaki, F. (2001). Applications of remote sensing and GISfor damage assessment. In R. B. Corotis, G. I. Schueller,& M. Shinozuka (Eds.), Structural Safety and Reliability(report). Lisse/Abingdon/Exton(PA)/Tokyo: A. A. BalkemaPublishers.