BAY OF BENGAL

CYCLONE THREAT MODELS

FINAL REPORT

Contract AID/SOD/PDC-C-0294' SAI Project 1-425-00-813,

SCIENCE APPLICATIONS, INC. Post Office Box 1303, 1710 Goodridge Drive, McLean, Virginia 22102, (703) 821-4300

BAY OF BENGAL

Final Report

Contract AID/SOD/PDC-C-0294

SAI Project 1-425-00-813

June 1984

Prepared -or

Agnyfor 'Interntional Development

,Washington,; DC 205213

Prepared by

Jerry D. Jarrell

Science Applications, Inc.

205 Montecito Avenue

Monterey, ZA 93940

TABLE OF CONTENTS

Page

SECTION * . . . 1 1.0, Introduction ......... 1.1 Background . . . . . . . . . . . .

1.2 The Tropical Cyclone Problem. . . . ,. 3

1.3 The Use of Probability in the Choice

Between Two Alternatives .... . 5

1.4 Choice Among Three or More Alternatives 8,

1.5 Basis for Probabilities . . . .... . Ji1

SECTION 2 THE WIND PROBABILITY MODEL . . . 13,

2.0 Statistical Model Basis . . . . . .. . . 13

2.1 Model Operation . . . . . ....... 16 2 ,1 .A ' Inpu t . . . . ... .. . . 1 +. . ..

2,1,2 Computations and Associated Assumptions 17

21.2.1 Validity of Assumptions ' 9 -:. . . 20 .,0; +":;i,.+.:. . +:23:::21 3 Output. .. 0; .0 < 0+'

201..1 Darkness, ... .... .0 " 0 0 23

2.1.3o2 Area Threat o o o e..... .. 25'

2.1.3.3 Point Threat.... ...... . 26,

SECTION.3 STORM SURGE PROBABILITY-MODEL . + 27

30 Introduction. o . . . . . .. . . 27

3.1 Background a' a - ... 280

3.2- Storm Surge Model . . . . .. .o.. . .32'

303 The Probability Model . . . . 3..32

3.4 Useof Information. . . . . . 34.

3.5 Other Applications. . . o "3

3.6 Accuracy of the Storm'Surge Threat

Model 0 0 0 0 0 0 0 9, 0 0. 0 . I 37

i ' .

TABLE OF CONTENTS continued

-°Page

SECTION 4 : CYCLONE/HURRICANE ACCEPTABLE RISK

MODEL (CHARM © ) .. . . - 0-.*' 0. 39

.0 Introduction .. * * * * * * 39

4.1 The Threat ....... .... ,,.40

4 An Idealized Warning System .... . .. ,4

4 .3 Decision Analysis in Warning. . . . . 44

4.4 Modular Design of CHARM© . .. . . 53

4.5 Decision Analysis Subsystem'Description'55

SECTION 5 WORLDWIDE TROPICAL CYCLONE THREAT

ESTIMATION SYSTEM. " " . . .1. . . 65

5.0 Introduction .... ... .. 0 ..... - .65

5.1 The Threat ..... .. ... .. . .0 . 66,

5.2 Meaning of Threat Indices . .0 0 0 . 6 5.3 Information Flow . ...... . . . . . 70

5.4 No Threat Message........... 73

SECTION6- TECHNOLOGY TRANSFER .... 0 0 . .. . 74

6.0 Goals and Accomplis.kents . . . . .. .74

SECTION 7 REFERENCES. . .. ... . . . 75

APPENDIX.. .7.6.. .,76

1.0

SECTION 1

Introduction.

This is the final report under contract AID/SOD/ PDC-C-0294 (SAI Project #1-425-00-813) but it also discusses related work begun under contract # AID/SOD/PDC-C-0110. The total funding under these two contracts was $361,931.

This report is divided into 6 main sections, begin­ning with an introduction which outlines the program with respect to U.S. government objectives. The introductory sec­tion also describes the problems to be solved and provides

background insight as to how programs and models defined in subsequent sections can be used.

Section 2 describes a tropical cyclone wind proba­bility model which was developed for the Bay of Bengal. Sec­tion 3 describes a tropical cyclone storm surge probability model developed for the same area.

Section 4 develops the Cyclone/Hurricane Acceptable Risk Model (CHARM0 , SAI 1982) to the systems level. Section 5 describes an Office of Foreign Disaster Assistance (OFDA) worldwide tropical cyclone threat model. This real time op­erational model grew out of research described in sectons 2 and 3 as well as related projects which were joint Navy-OFDA. funded.

Section 6 describes technology transfer efforts whereby the models have been explained in the scientific lit­

erature and in briefings at workshops, symposia and scientif-1, ic professional meetings.

"-­

1.1 Background.

It is the policy of the U.S. government, through the Agency for International Development to:

1. Render emergency relief to victims of natural and man-made foreign disasters, in coordination with other

goverAments, international agencies, and voluntary organiza­tions. Such assistance can be provided to the people of any

nation affected by disasters and must, to the greatest extent possible, reach those areas most in need of relief and reha­

bilitation.

2. Monitor all potential and actual disaster situ­ations.

3.,_ AssIst in -rehabilitation when such rehabilita­

tion is beyond the capacity of local rescurces.

4. Encourage and participate in foreign .disaster preparedness through the provision of technical assistance

and international training programs.

5,. Consider on a case-by-case basis longer term reconstruction assistance and implement the program as a de­velopment tool in areas where there has been severe social

and economic disruption.

6. Support the efforts of international organiza­

tions and voluntary agencies involved. in foreign disaster

assistance.

-2.­

1.2

7. Increase'.U.S. technical capacity to define di­saster-prone conditions, and to recommend disaster-avoidance,

measures.

8. Initiate, within international forums,efforts to increase other donor participation in disaster prepared­ness and disaster relief activities.

With regard to cyclones in the Bay of Bengal, the research reported herein supports the U.S. policy objectives above. To some extent all the above policy objectives are supported, however the primary emphasis is on monitoring (objective #2), disaster preparedness (#4), and increasing technical capacity to define disaster prone conditions and to recommend disaster-avoidance measures (#7).

The Tropical Cyclone Problem

Tropical cyclones include the hurricane (in oceans bordering North America); typhoons (north western Pacific ocean); and cyclones of the South Pacific ocean and the Indian ocean, including the Bay of Bengal and Arabian Sea. Historically the great disasters have been in the Bay of Bengal. The November 1970 Bangledesh cyclone is estimated to have killed over 200,000 people (ESCAP, LRCS, WMO.1977) and the November 1977 Andhra Pradesh (India) cyclone killed per­

haps 20,000.

The disaster potential in the Bay of Bengal is partly attributable to the distribution of population in low lying and unprotected areas along the Bay, but more important is the configuration of the Bay which results in large cy­clone induced storm surges (often incorrectly referred to as tidal waves). Surges are an elevation of the sea surface

-3­

because of the "barometer effect" (pushing up the water into the cyclone's central low pressure) combined with the buildup of wind driven water along the coast. These effects are superimposed on the astronomical tides which have a 12-hour period (two highs and two lows per 24-hour day) and a large tidal range, at least with some combinations of the solar­lunar cycles. An unfortunate coincidence of a large storm surge and high tide can increase the mean water level by sev­eral meters, thus inundating large coastal areas with devas­tating effects.

The great killer, then, is the storm surge. The work discussed herein both with storm surge and with cyclone winds, which independently account for another substantial contribution to the damage/casualty total. This report also describes the introduction of probability to interpret cy­clone forecasts in time of threat.

Forecasts tell, in very specific language, where the cyclone will be, at what time, and describe its configu­ration (size and wind distribution). If it were not for in­accuracies in these forecasts, specifying the impact of a cy­clone on a locale would be rather straightforward; however, inaccuracies are a fact of life even with advanced forecast­ing capability. While forecast errors have gradually been reduced over the past few decades, errors still persist and their reduction seems to have leveled off in the decade of the seventies (Jarrell et al., 1978; Neumann, 1978). The forecast improvement was most pronounced in regions of the world where observation improved. Enhanced observation was mostly due to post-WWII aircraft reconnaissance, supplemented increasingly after 1964 by meteorological satellites. In the Bay of Bengal the satellite era is just emerging, thus we may expect to see further improvement in forecasting capability there.

-4­

Within a few years forecast accuracies in the Bay of Bengal may approach those of the Atlantic and Eastern Pacific hurricanes and the Western Pacific typhoons. The ob­jective of the work reported here is to make current fore­casts with their inherent inaccuracies more usable. Barring unanticipated dramatic improvement in forecast accuracy, there will be a continuing need for this kind of analysis.

The aim of this research was to develop a reliable estimate of the probability of 40, 65, and 100 kt winds and storm surge over 2 meters for points under,threat of a trop7'. cal cyclone.

1.3 The Use of Probability in the Choice Between Two' Alternatives

We use probabilty as a means :of quantifying the risk of an event occurring. Often events can be expressed as binary outcomes, a choice of two: the event will or will not

occur. In this context, for a given case the real probabili­ty (presently unknown) is either 0 (the event will not occur) or 1 (the event will occur). Any other outcome is impossi­

ble.

Over the longer term, specific events have a cer­tain likelihood of occurring. This probability of occurrence can be represented by a number between 0 and 1. It is this point of view that is usually advocated in decision theory. The techniques used are those which maximize the overall,

long term outcome, not necessarily the immediate decision &t hand. For example, if evacuation is considered, all the costs, both monetary and human, must be weighed against the potential savings in terms of life and property.

The following loss table illustrates the tradeoff between evacuation and non-evacuation.

Preparation Outcome

Actions Hit Miss

Costs of Costs of Evacuate Evacuation (C) Evacuation (C)

Do not ,eiacuate Avoidable None

Losses (L)

CONTINGENCY LOSS TABLE

If we want to minimize probable losses, we will want to evacuate only when the probable loss with evacuation is less than the probable loss without evacuation. Note that unavoidable losses are not a relevant part of the problem.

If P is the probability of a "hit" of sufficient magnitude to necessitate evacuation, then the expected or probable losses with evacuation would be C and those without evacuation P x L. We want to order evacuation only if C < P x L or, rearranging, we evacuate when P > C/L.

To illustrate this principle, suppose a boatowner is considering moving his boat to an inland shelter. Let's say the boat is worth $10,000 and its costs $100 to move it. With C= 100, L = 10,000, he would not want to move it unless the probability (P) of its being 'lost exceeded the ratio of these numbers:

P > C/L = :%

Obviously in real world situations life is not that simple. We have great difficulty putting dollar values on both evacuation costs and losses because of a host of com­plexities. Note that it is necessary that both C and L be expressed in the same units, but not necessarily monetary units. They could, for example, be human lives although it requires some innovation to express costs of preparation in those terms. One way which has been suggested is through loss of credibility in the warning systems. That is, if evacuation is ordered unnecessarily and repeatedly ("crying wolf"), it becomes increasingly difficult to effect an evacu­ation because of an unconvinced, reluctant public response. Unnecessary evacuation uses up credibility with the price measured in a future loss of lives (when the real wolf ap­

pears).

The condition (P > C/L) is offered as a necessary condition, that is, don't act unless this condition is met. This condition, however, is not uufficient, i.e. even when

this condition has been met, there are still other factors to consider before acting. For example, suppose the boatowner's criteria was met, such that there was a greater than 1% prob­ability of destructive conditions occurring three days from now. If it only takes a few hours to move the boat, he should wait. Severe tropical cyclone damage is a rare event: if we have a probability of occurrence of 1%, then we have a

probability of non-occurence of 99%. If the boatowner waits a day or so, in most cases the threat will subside. Simply stated, this principle is that we should put off action until we have just enough time to complete the action before the event occurs. In this connection, one must not overlook the reality of darkness and/or the onset of inclement weather

ahead of the approaching cyclone as factors in slowing any preparedness action.

-7­

1.4 Choice Among Three or More Alternatives

The preceding example of two choices of action ex­tends easily to choosing among three alternate courses of action. Consider the following example of a medium sized, and hypothetical, police force which will man various emer­gency posts dependent upon which of 3 hurricane events is forecast. We assume the most likely (the mode) will be fore­cast.

Minor event requires 2 man-days Moderate event requires 10 man-days Major event requires 50 man-days

There are "penalty" costs associated with either underestimating or overestimating the requirements given in the following cost table. The elements on the diagonal rep­resent the cost of correct manning, and those off the diago­nal represent tize cost of either over or under manning.

ACTUAL OCCURRENCE

ACTION OPTIONS MINOR MODERATE MAJOR

1. Man for minor event 2 17 97 2. Man for moderate event 11 10 85 3. Man for major event 60 55 50

COST CONTIGENCY TABLE FOR HURRICANE MANNING (man-days)

A rational decision is one which minimizes the ex-' pected cost. If we can estimated the probabilities of each' outcome occurring PI, P2, P3, then the expected cost'

for:

-_8­

Option 1 is C1 = 2xPi .+ 17xP2 .. 97xP3,

Option 2 is C2 =i llxP1 + l0xP2 + 85xP", and

Option 3 is C3 60xP..+55xP +5xP3

For example, suppose P1 = .45, 35 andP2 . P3 .20. Note that, aside from these, three options, we assume no other outcome is possible (here, a minor event also

'includes no event), i.e, P1 + P2 + P3 - 1.00.

C1 = 2x.,45 + 17x.35:+ 97x.20 = 26.25

C2"= llx,45 +lOx.35 + 8,5x.20 =25.45

C3 = 60x.45 + 55x.35 + 50x.20 = 56.25

course of action with minimum ex-Since C2 is the

pected costs, it should be selected. Notice that the minor event is most likely to occur (with a probability of 0.45), but in this case the best bet, is slight over preparation. The difference in C1 and C2 may be trivial, but certainly

either is preferable to C3.

Figure 1.1 summarizes this problem in three graphi-.

cal depictions. Figure 1.la shows a lxl graph of P1 vs:

P2" Note for any PI' P2 combination, a value of,, P3 is determined since P3 1 - P1 - P2. Thus, in the lower left corner where P1 and P2 are small, P3 is near

1. The space blocked off in the lower left is the region

where event 3 (major hurricane event) is more likely than either events 1 or 2; similarly, the upper left and lower right corners represent the combinations wherein events 1 and 2, respectively are the most likely outcome. Notice that the combination of P1 ' P2 in the upper right triangle do not

exist, since P1 + P2 < 1.

-9­

1.00

Figure l.la. Illustrates sec­tions of a lxl probability square wherein events 1, 2 or 3 are the most likely of the three mutually exclusive events

1 0 -5C -which are also exhaustive(P1+P 2 +P3 = 1.0)

0 0.50 1.00

1.00 P

Figure l.lb. Illustrates sec­tions of a lxl probability square wherein the expected cost is minimized by planning for

. ... events 1, 2 or 3.

10.50

mi-3j____ _ 0

,.00005 1 Figure 1.1c. Combines figures

2_ 71.la and l.lb. Illustrates con­1.00 ditions under which over-prepara­

tion (vertical hatching) and under-preparation (black) mini­

liii mizes expected cost. Number pairs (N1 , N2 ) are minimum expected cost event, most likely event.

0.50 "

2,3 3,3

0 - 10­0 0.50 1.00

1.5

Figure l.lb is a graphical representation of all

possible combinations of Pl. P2 and the vrinimum expected

cost of each for the police department prob.em. Note that

there is a similarity between figures la and lb. Near the

corners where the outcome is fairly certain, the most likely

outcome is the best one to plan on; however, in the middle

region that is not true.

Figure l.lc combines figures l.la and l.lb. In the

area with vertical cross-hatching, the most likely event to

occur is event 1, a minor event; however, the expected cost

is less in this region if we provide more manpower than that

which is most likely required. In the shaded regions, the

least expected cost occurs when we underman.

This type of argument can be extended to any number

of possible courses of action, although the computations be­

come laborious. A graphical solution (of the type illustrat­

ed here) will work for up to 3 courses of action. A program­

mable calculator can easily handle up to 5 or 6 options.

The difficult part of applying such methods is ob­

taining the probabilities. Determining the probabilities is

the objective of this research.

Basis for Probabilities

In the Bay of Bengal, probabilities are available

based on climatology. For example, Neumann and Mandal (1978)

describe an analog scheme which models such a probability.

This probability is solely based on climatology biased to

persistence. To the extent that the forecaster can improve

on climatology, probabilities based on such a forecast will

be an improvement over climatological probabilities. By

-11­

"improvement" we mean higher probabilites will be followed by the event's actual occurrence while lower probabilities will be followed by actual non-occurrence. This means the user's critical probability (PC = C/L, see Section 2.0) will be exceeded less often in non-hit cases. Since the forecaster can take into account not only climatology and persistence, but also synoptic factors and the analog scheme itself, he

should usually beat the analog.

2.0

SECTION 2

THE WIND PROBABILITY MODEL

Statistical Model Basis

The probability estimation model is based on a study of tropical cyclone forecasts issued by the Joint Typhoon Warning Center (JTWC) on Guam during the 7 years be­ginning with 1971, the year when the JTWC began forecasting in the Bay of Bengal.

The particular forecasts were 205 motion and maxi­mum wind forecasts for cyclones in the north Indian Ocean; however, there were only 137 forecasts where the nowcast po­sition was actually in the Bay of Bengal. This number of forecasts is barely adequate to support a statistical study. In a similar motion forecast study for western Pacific ty­phoons, Nicklin (1977) had over 5000 cases. Jarrell (1979) also studied wind forecast errors in the western Pacific and had a sample of over 2000 cases. The large difference in sample size is somewhat reduced because the Bay of Bengal fore;asts were iss-ed at 12-hour intervals vs. an interval of 6 hours in the western Pacific, such that consecutive fore­

casts are more nearly independent. Additionally, the general characteristics of the error distributions are known to a good approximation from these and other earlier studies. This study was intended more to confirm similarity rather

than break new ground.

An attempt was made to stratify forecasts into dif­ficulty classes. The most successful stratification was on the basis of forecast direction of motion. Errors are usual­ly smaller for cyclones moving west and larger for cyclonesi

-13­

moving northeast. These directions represent typical tracks before and after recurvature, respectively. "Recurvature" is a term signifying a transfer from control by the near­equator easterly air currents to the westerly currents of the mid-latitudes. Pre-recurvature tracks in the easterlies are typically west-northwest while post recurvature tracks are typically toward the northeast. The westerlies are much stronger, and hence cyclone forward speeds are much greater after recurvature than before. Larger errors are also asso­ciated with higher forward speeds. The best direttional sep­aration occurred at about 3400 (north-northwest), i.e., di­rections from 3400 to 1800 appeared to behave differently from the remainder of the cyclones. Some statistical sununary results are shown in Table 2.1. The group with headings 340w clockwise to 1800 consists mainly of post recurvature cases and is shown as Sector 1 in Table 2.1. Sector 2 forecasts (pre-recurvature headings 180* to 3400) are the most common forecasts. In Sector 1 the mean W-E error is somewhat posi­tive (east) while the mean S-N error is significantly posi­tive (north), since these forecasts are typically for north to northeast motion, this represents a speed over-forecast. For Sector 2, the W-E mean error is significantly negative (west), while the mean S-N error is slightly positive (north). Again these are over forecasts in speed along the track. These speed errors may in part be attributable to the lack of a good climatology for the area and hence the Guam forecaster relies on his experience with faster moving Pacific typhoons. For this sample, the average forward speed was 5.4 kts; if we 'add the average 24 hour speed error (2.9 kts) to that, we arrive very close to the average typhoon forward speed of 9 kts. This appears t.o be a correctable bias. Without this bias, there appears to be no significant difference in the difficulty classes, hence an artificial set of statistics based upon the removal of this bias is used in the model and the directional discrimination is eliminated.

14­

Sector I- Sector,2 Combined

ERROR MAGNITUDE 0 HR 24 11R 48 HR 0 HR N'HR A HR 0 HR, 24 HR 48 HR Mean (n4i) 54 173 251 5 151' 280 54 160 271 Std Dev (nmi) 41 99 114 48 98 148 45 99 138

S-N ERROR (En)

Mean (nmi) 17 21 72 11 15 10 134 17* 30* Std Dev (nmi) 45 124 197 38 107 41 '94 114 195

W-E ERROR (Ee)

Mean (nmi) 6 -41. 2, 27 -85 14*-175 -34*'! -117* Std Vev (nmi) 47 153 187 55 A 119 184 52 ]34 185

WIND ERROR (EW) .

Mea~n (kt) 0 8 12 -.1 -2, 2 2 5 Dev (kt) 9 16 22 8 13 22 8 15 23

CORRELATIONS , "

Ee to En 0.40 'o.37- 0,.22 -0.06 0 17 .26 0.20.40: 7' 0.2.6.. .. 2

Ee to Exw -0.16 -0.25 -;-51 -0.25 -0.00 0.19 .0.02* 0.00* 0.04*

En to Ew 0.03 -0.18 -0.23 0.09 -0.16 -0.'05A 0,1 -0.15* -0.10*

CASE COUNT 56 55 26 81 851 54 13613 7 80o,*assumed to equal zero on'the model

Table 2.1. Summary of results of a statistical study ofU.' errors in JTWC forecasts for .the Bay of Bengal.

-15­

2.1

This permits the pooling of cases for maximum statistical stability. The correlation coefficients are of some passing interest. The correlation between the error components is small but comparable to those found in the western Pacific by Nicklin (1977) and in various stucves (see for example Neumann 1975, and 1978). Notice the small correlation between the error in maximum winds and displacement error components. Since none of these in th~e pooled sample are significantly non-zero, these errors will be treated as independent in the model.

While we are satisfied with the stability of the forecast error statistics, we would have preferred a longer period of record. We have also seen a rather severe bias in the Guam forecasts which we assume will be corrected. The presence of this bias serves to flag the risk in applying statistics derived from forecasts from one source to fore­casts of another source. For this reason we recommend that the statistical package be derived from and tailored for the driving forecasts. For example, if one of the nations on the Bay chooses to adapt this model relative to their own fore­casts, a statistical package derived from their forecast should be substituted for the Guam statistics. In some cases this may also provide a more ixtensive basis.

Model Operation

Like most models the wind threat model can 'be thought of in three stages: Input, computation and output. Input and output are perhaps of most concern to the user

because of their visibility, but the important work goes on in the computation stage. It is there that the underlying mathematical relationships are expressed and exercised and it is there that any simplifying assumptions are made. The lat­ter, together with the input, determine the validity of the output. The following description provides information which the user needs to understand in order to fully appreciate the output information.

2.1.1 Input

The input is taken exclusively from the JTWC cy-. clone warnings. It consists of cyclone identification infor­mation, used for output labelling, and forecasts of latitude, longitude and maximum wind at 0, 24 and 48 hours after fore­cast valid time. Figure 2.1 illustrates an actual forecast of tropical cyclone 17-79 originated by the JTWC at 0800 GMT on 11 May 1979. The bottom line in Figure 2.1 is the neces­sary input information extracted from the warning. The in­formation (except the second entry, month: 5 = May) is under­lined in the warning. This cyclone warning will be used, again in an output example.

2.1.2 Computations and Associated Assumptions

The actual computations are carried out on a grid of points spaced at 60 nmi intervals along the periphery of the Bay of Bengal and along the Andaman Island Chain. These­computations are executed at 3-hour time steps from forecast, valid (nowcast) time to 48 hours hence. The following as­sumptions are implicit in the computations.

(a) That the forecast represents the mean of all •

possible outcomes, that the deviation of the actual position

,17­

WTXX31 PGTW 111000 TROPICAL CYCLONE 17-79 WRNG NR 20 POSIT 13.2N6 082.3E3 at 110800Z ACCURATE-WITHIN 4-0 NM BASED ON EYE FIXED AT 13.3N7 082.7E7 AT 110615Z BY SATELLITE PRESENT MOVEMENT: WEST-NORTHWEST AT 05 KTS PRESENT WIND DISTRIBUTION: MAX SUSTAINED WINDS 90 KTS NEAR CENTER WITH GUSTS TO 110 KTS RADIUS OF OVER 50 KT WINDS 75 NM RADIUS OF OVER 30 KT WINDS 150 NM OVER WATER REPEAT POSIT 13.2N6 082.3E3 at 11080UZ FORECASTS: 12 HRS VALID 112000Z 13.8N2 081.5E4 MAX WINDS 100 KTS WITH GUSTS-'25 KTS RADIUS OF OVER 50 KT WINDS 80 NM 24 HRS VALID 120610Z 14.2N7 080.5E3 MAX WINDS 105 KTS ,'ITH GUSTST'- 30 KTS RADIUS OF OVER 50 KT WINDS 90 NM OVER WATER RADIUS OF OVER 30 KT WINDS 175 NM OVER WATER EXTENDED OUTLOOK: 4.HRS VALID 130800Z 14.7N2 078.9E4 MAX WINDS 30 KTS WITH-=TS TO 45 KTS DISSIPATING OVER LAND NEXT.WARNINGS AT 111600Z, 112200Z, 120400Z and 121000Z REMARKS: LATEST SATELLITE DATA INDICATES TC 17-79 HAS CONTINUED TO INTENSIFY OVER THE PAST 12-24 HOURS, WITH SLOW INTENSIFICATION EXPECTED UNTIL LANDFALL. TC 17-79 HAS SLOWED TO CS KTS, HENCE', AN ADDITIONAL 06 HOURS WAS ADDED TO THE WARNINGS VALID PERIOD. TC17-79 CONTINUES TO OSCILLATE ABOUT AN OVERALL WEST-NORTHWEST TRACK, HENCE, THE FIX POSITION WAS NOT USED VERBATIM. BT

"17-79", 5, 1108, 132, 823, 90, 138, 815, 100, 147, 789,

30, 48

Figure 2.1. An actual cyclone warning issued by the Joint Typhoon Warning Center on Guam for Cyclone 17-79 on 110800 GMT May 1979. The added line across the bottom is the extracted model input.

from this mean is a random variable pair described by a bi­variate normal fiequency distribution whose parameters are given in Table 2.1.

(b) That the actual maximum wind is a normally dis­tributed random variable about the forecast, with parameters

given .n Table 2.1.

(c) When landfall is forecast, the accompanying re­duction in the maximum wind forecast is solely attributable to land influence, otherwise, a trend established prior to landfall would have continued. If no trend can be inferred from the forecast, it is assumed the nowcast wind would have continued in the absence of landfall.

(d) That probabilities can be adequately interpo­lated in space between representative points spaced 60 nmi

apart.

(e) That the forecast positions and the statistical parameters valid for 0, 24, and 48 hours can be interpolated to 3-hour time steps and, further, that linear interpolation

of probabilities between 3-hour time steps is valid.

(f) That probabilities can be summed over time, i.e., the probability of an event occurring within a three­hour timestep is the sum of the probabilities at the two end points less the probability of an occurrence at both times. That the probability of an occurrence at both times can be determined from a geometric parameterization of probability

along the forecast track.

(g) That wind errors and forecast position errors, notwithstanding the landfall case in (c) above, are indepen­

dent.

(h) That the shape of the mean wind radial profile is similar to that of western Pacific typhoons and is related

to maximum wind speed.

2.1.2.1 Validity of Assumptions

Although the validity of the foregoing assumptions has not been established herein, there Is considerable evi­dence to cupport most of them.

The concepts expressed in assumptions (a) and (b) have been firmly established in the development of the U.S. Navy Pacific typhoon STRIKP (strike probability) and WINDP (wind probability) programs (Jarrell, 1979). The actual val­ues of the statistical parameters in Table 2.1 are subject to error. This error can reasonably be as great as 10% in the important standard deviations. As an illustrative exam­ple of the impact of a 10% error, for comparison, we first estimated the probability that after 24 hours a cyclone would actually be within circles of radius 50 and 100 nmi centered on a 24-hour forecast.

The probabilities of the cyclone actually being within these 50 and 100 nmi circles (where it is predicted) are 11 and 36%, respectively. To demonstrate that these probabilities are reasonable, we examined the 24-hour fore­cast errors for 1978 (independent data) for comp-rison. There were only 28 such forecasts, but of those, 14% and 39% actually verified within 50 and 100 nmi, respectively, of the 24-hour forecast point. We estimate that any point 100 miles.

,20­

removed from the forecast point has 2/3 the chance of being

struck as does the forecast point, while a point 200 miles

away has 1/5 as great a chance of being struck. It is not true (as has been stated) that the forecast point is the saf­

est place to be, and it is also clear that a great many other

points are also threatened. Being able to quantify that

threat is the unique capability of this model.

To simulate the effect of a 10% error in the stan­

dard deviations we computed several probabilities using stan­

dard deviations of 90%, 100% and 110% of those in Table 2.1.

The probabilities estimated are for a cyclone being within circles of radius 50 and 100 nmi centered on the forecast

point (as before) and removed to the northwest distances of

100, 200 and 300 nmi. These estimates are given in Table

2.2.

DISTANCE OF CIRCLE CENTER NW OF FORECAST POINT CIRCLE RADIUS 0 nmi 100 nmi 200 nmi 300 nmi

50 nmi 13-11-9.3 7.8-7.1-6.5 1.8-2.1-2.4 .15-.29-.43

100 nmi 42-36-31 28-25-23 8.2-9.0-9.6 1.0-1.6-2.1

Table 2.2 Probabilities (%) of cyclone being within various circles in the vicinity of the 24-hour forecast

The first number under each column is based on the smaller

standard deviations, the second number uses standard devia­

tions given in Table 2.1, and the third number is associated

with the larger standard deviations. This range should pro­

vide some sense of the limits on the accuracy of our proba­

bility estimates. It is doubtful that the uncertainty intro­duced by imperfect knowledge of the statistical parameters is

great enough to be important for most purposes.

-21­

The above simulation also provides some insight into what happens to probabilities when forecast accuracy improves (and the standard deviation of the errors decreases). Those probabilities near the track (distance zero) increase while those far from the track (distance large) decrease. For perfect forecasting, those probabilities along the forecast track would be 1.0 while those far from the track would be zero.

Assumption (c) is based upon the author's knowledge of the practicalities of forecasting and is a rather straightforward mechanism for removing a foreseeable bias.

Assumptions (d) and (e) are the result of testing and are considered to represent the minimum time and space model resolution without significant distortion of the re­sults. Finer resolution would not materially affect the out­put, but for purposes other than the present (e.g., storm surge) a finer resolution in both time and space may be re­quired.

The validity of assumption (f), (time summation) has been thoroughly demonstrated in the U.S. Navy models, over a wide range of probabilities and circumstances.

As previously mentioned, the low correlation be­tween errors in maximum wind forecasts and position forecasts (Table 2.1) supports an assumption of independence between these errors (assumption (g)).

-22­

Assumption (h) is perhaps the weakest of the above assumptions. Unlike the Atlantic and North Pacific, there are virtually no real wind measurements available in and around the Bay of Bengal cyclones. For this reason there is little choice but to assume a maximum wind determined pro­file. Maximum wind itself is likewise not measured, but is estimated from satellite imagery. Since maximum wind is treated as a random variable (about the forecast) so, too, is the wind profile. There is little doubt that real data could improve this aspect, and hence the wind probability estimates in general. The absence of this data, however, is a more serious problem with deterministic forecasts since they do not anticipate inaccuracies in this or any other forecasts.

An appropriate measure of the impact of all of these assumptions would be to test the model on independent data. A meaningful test would require a few hundred cases which would take years to acquire. There appears to be suf­ficient substantiation of the validity of the probabilities. The utility of this information was demonstrated in a case study of Cyclone 17-79.

2.1.3 Output

The wind threat output is in two forms, an area­threat form and a point-threat form. These are illustrated for cyclone 17-79 from 0800 GMT 11 May 1979 in Figure 2.2.

2.1.3.1 Darkness

For planning purposes, a simple darkness scale has been included in the output (see Figure 2.2). This defines

-23­

FOR CYCLONE 17-71 rROM 1101307 PR0qABtLITT7S Or EVENTS dIT4N

LOIJRS OF OI0.APESS CHPS 12HOS 244RS 16HRS 4 ',,.* . 4. +

44HRS + .

STRIK-. WITHIN SqI LANKA 0,03C WINDS or Ar LEAST 4CKNOTS G,.C0 WINDS OF AT LEAST 69NOTS ,03C WINOS OF AT LEAST C'0KNOTS G.50

.030

.0,

.03 ,000

.331 ,00G4 .302 .0%1

.005

.014.

.0a ,003

,014 .032 .010 90G8

STRIKE. WTTHIN TAMIL NADU WINOS OF AT LEAST 40KNOTS WINOS Cc AT LEAST 6;YNOTS WINOS OF AT LEAST 1CIKNOTS

.003

.30O

.034

.03C0

.062 .125

.L82 .336 #0R s171 *0q 053

.172

.431

.228

.091

.209

.504 o273 .1±6

STRIKE WTTHIN AN PPAO.5SW WINOS OF 4T LEAST 41KNOTS WINOS Oc AT LEAST 6;KNITS wINOS OF AT LEAST 13:KNOTS

I,F ,Gjq .*3Ca

.031

.132

.196

.169

.042

.289 ,414 .34q .138

.311

.553

.463 e209

.446

.642

.528

.248

STRIKE WITHIN OPSA-W.-IEN 0."3 a00oO WIMOS OF AT LEAST 41KNOTS 0.4i0 3.000 WINOS *IF AT LAST 65KNOTS 0.000 3•.0J WI"OS OF AT LEAST 10,2qNOTS 0.300 0.000

STqrI( WITHIN EAST INrI ' g .309 ,1q3 WINOS OF AT LEAST 4(0KNOTS .009 .193 WINOS OF AT LEAST 65<JOTS .3)1q *.192 WINOS OF AT LEAST 10KNOTS .001 .061

a001 .908 .319 .52 s032 .011 .001 .005

.4L5 .567

.416 .573

.43 e565 s195 .302

.017

.079

.023'

.01G

.671

.687 .669 .370

•4W.

4

4.

+

4.

+

.

+

'

4

* . ' L " : , :. . IT . 4 . .:

+ ... +. + +

...+' +, . ++ + + + + + +

* + 4.",+

.. .4 :"G +

. ,4. 4

4 +

4 4. + +

4. 9

* ' 4' . . 4 44

+ S 4. " 4.+ 4. 4. .W 4.+ 4. 4 4' 4. .. 4

0­

+ 4 4'. .. 4,u + 4 4 4. . 4, 4. 4.4. 0800

* 04

+ + + * + +4

• +

+ +

+ "

+ +..4 +4

Fiue 4.. Win Thea Moe uptfrCyln 77,00

,24­

darkness as 1800 to 0600 LST and uses a -6 time zone. Each print position represents two hours.

2.1.3.2 Area Threat

The area threat is provided to serve the alerting function of government where certain preparatory actions can be taken regardless of the actual location of a disaster

within a political jurisdiction. Here, the threat of 40, 65 and 100 kt winds now and within 12, 24, 36 and 48 hours is provided. The rationale for the selection of 40, 65, and 100 kt winds is provided below:

40 kts: This is about the point where it becomes extremely difficult to perform outdoor tasks, hence when 40 kt winds arrive any physical preparations that will be re­

quired should already be completed.

65 kts: This is about the wind force where signif­icant wind damage begins to occur and where storm surge first

begins to be a problem.

100 kts: Under ordinary circumstances, winds in excess of 100 kts can signal a major disaster. The intent here is to treat the probability of winds in excess of 100 kts as the probability of a major disaster occurring. Al­though there are a great many other contributing factors,

this is a satisfactory first approximation.

For example, in Figure 2.2 the 48-hour threat of 65

kt winds to Andhra Pradesh is 0.528. This can be read as follows: "The probability of at least 65 kt winds being ob­served at some point on the Andhra Pradesh coast between 0800 GMT 11 May and 0800 13 May is estimated to be 53%.

-25­

2.1.3.3 Point Threat

The wind threat to a point is handled somewhat ar­

bitrarily in a way which combines the level of the threat

with the urgency of that threat. The threat used here is the

probability of at least 65 kt winds, T(65). To avoid confu­

sion with established warning/watch/readiness conditions in

use in parts of the world, a system of color codes is used.

Red represents the greatest, most imminent threat while green

represents the most remote threat; orange and yellow repre­

sent intermediate threats in between red and green. The def­

initions used herein are as follows:

Sybol Color T(65) Time

R RED > 20 % 24 hours

0 ORANGE > 10 % 36 hours

Y YELLOW > 5 % 48 hours

G GREEN > 2.5% 48 hours * NONE > 2.5% No limitation

These are illustrated in Figure 2.2 for points on the Bay of

Bengal for Cyclone 17-79 on 11 May 1979 at 0800 GMT.

-26-'

3.0

SECTION 3 STORM SURGE PROBABILITY MODEL

INTRODUCTION

The concept of cyclone threat analysis has been de­veloped over th-. past few years. This type of analysis is an effort to enumerate the weather events that can occur, to in­terpret each in terms of human impact, and to determine the likelihood of each occurring.

An analogy between the cyclone threat analysis and a combat situation can be made. The enemy is a cyclone (typhoon, hurricane or tropical storm). His primary weapons are winds and rain, although winds may cause more devastation through storm surge (abnormally high tides and waves along coastal areas) while river flooding may be the major damage agent in inland areas. Forecasts can be thought of as intel­ligence estimates. The difference between a threat analysis and a forecast is that the former considers all potential threats and ascribes a likelihood or probability to each, where the latter presents only the most likely of all those

possibilities.

The prudent commander will seek to devise a plan which will satisfactorily meet a wide range of enemy capabil­ities. In this analogy it is readily apparent that providing for the single most likel, tactic is a perilous course, par­ticularly when the enemy may choose one or more of several nearly equivalent alternate courses. Statisticians illus­trate this point by reminding us that the most likely sum on the throw of two dice is seven, yet the odds against seven occurring on a single roll are 5:1. When a commander's limited assets require choices between competing courses of action, he will want to cover the most likely and, at the

-27­

same time, the potentially most damaging contingencies.- To evaluate such factors he needs an estimate of the likelihood of the enemy employing all possible tactics and the likely outcome (losses) under various opposing postures.

The simplicity frequently ascribed to nature that it is non-conniving overlooks the endless variety of modes' nature can use to configure its forces for the attack.

The cyclone threat concept turns forecasts into estimates of the point probability of a cyclone strike winds above a certain level, or storm surge above a certain level. The potential for site-specific models to evaluate the local impact (damage and casualties) given the meteorological con­ditions and the state of preparedness lies in the future, while even more sophisticated models may be able to evaluate the pros and cons of taking (or not taking) certain actions.

This section documents the development of a surge probability model for the Bay of Bengal. The previous sec­tion documented the strike and wind threat estimates.

3.1 Background

Intense tropical and extratropical cyclones occur worldwide. Generally the more severe tropical cyclones (with winds over 64 kt) are called hurricanes in the Atlantic and northeast Pacific, typhoons in the northwest Pacific and cy­clones in south Pacific ocean and the Indian ocean. Table 3.1 gives fatalities associated with some of the most notori­ous tropical cyclones. Clearly, in terms of human suffering and the loss oflife, citizens of countries bordering the Bay

Year Location Deaths

1970 East Pakistan 300,000 1737 India 300,000 1881 China 300,000 1923 Japan 250,000 1897 East Pakistan 175,000 1876 East Pakistan 100,000 1864 India 50,000 1833 India 50,000 1822 East Pakistan 40,000 1780 Antilles 22,000 1839 India 20,000 1789 India 20,000 1965 East Pakistan,.. 19,279 1963 East Pakistan, 11,468 1963 Cuba-Haiti 7,196 1900 Texas 6,000 1960 East Pakistan 5,149 1960 Japan 5,000

Table 3.1 Deaths resulting from noteworthy tropical disasters (from Frank and Husain, 1971).

Population affected 4.7 million Crop loss $63 million, Loss of cattle 280,000 Loss of poultry 500,000 Houses damaged 400,000 Schools damaged 3,500 Fishing boats destroyed 9,000

(marine) Fishing boats destroyed 90,000

(inland water)

Table 3.2 Estimated damage from the November."1970 cvclone­

(from Frank and Husain, 19'

-29­

of Bengal face the most severe threat from tropical cyclones compared with other regions of the world.

Although the immediate effect of a tropical cyclone can be devastating, the survivors may face chronic adverse conditions long after the passage of the storm. This would be particularly true for those communities with economies based almost entirely on a single product. For example, the damage estimates given in Table 3.2 for the 12 November 1970 cyclone which impacted Bangladesh reflect extensive devasta­tion but do not directly illustrate the human suffering that followed. Frank and Husain (1971) estimated that 46,000 of the 77,000 inland fishermen were killed outright, in addition to severe damage to the operations of the remaining 40%. This meant that approximately 65% of the fishing capacity of Bangladesh was lost. The extent of human suffering can be imagined when one realizes that 80% of the protein intake by the Bangladesh population is in the form of fish (Frank and

Husain, 1971).

Clearly, agencies charged with minimizing human loss and suffering from the effects of Bay of Bengal cyclones can use a reliable risk assessment. Figure 3.1 depicts the problem faced by decision makers in using a forecast,. The cyclone forecast in this figure simulates conditions about two days prior to the devastating November 1970 cyclone strike in Bangladesh. The small inset graphs represent the time evolution of strike probability for four different points. Notice that the probability of the cyclone striking points near the track is not appreciably different than the probability of striking points removed from the track a con­siderable distance. This implies a risk for points removed

-30­

+ 30 3(YO

° 1100

0hours

A% -+

0

0 D0

0 00 bous 48

51 m/s p (50 m/s)

Figure 3.1. Simulation of a forecast for the November 1970 Bangledesh cyclone. Insert graphs depict the evolution overtime of the probability of the cyclone striking one of four points. The inset graph at the lower right depicts the pro aiiyof the cyclone having winds of at least 51ms0(100 kt).

from the track which is comparable to those along the track. The decision maker, then, must allow for error.

The existing wind and strike probability models can provide the decision maker with risk information relative to wind and, to some extent, other disaster agents (e.g., flood­ing and storm surge). Extracting all but the crudest storm surge risk information is beyond even the most sophisticated user because of extremely complex relationships between storm surge and controlling factors such as local bathymetry, cy­clone strength, cyclone motion and astronomical tides. Many of these factors are also constantly changing. The storm surge risk assessment model evaluates the myriad combinations of these factors and calculates a risk or threat level for stretches of coastline over periods of time.

3.2 Storm Surge Model

The model used was the Texas A&M Gulf of Mexico tide model (Reid and Whitaker, 1981). This model was set up on six different computational spaces along the Bay of Bengal. Cyclones were simulated to have struck 16 different points wi.th combinations of three wind speeds, three track headings, and two forward speeds (see Figure 3.2). Combina­tions totaling 258 were selected to cover climatologically realistic possibilities. The surge which resulted at various neighboring points was cataloged for each landfalling cy­clone.

3.3 The Probability Model

The probability model seeks to determine the prob­ability that storm surge will exceed the local high tide by two meters or more. Two meters'is treated here as a disaster

threshold.

-32­

ChlVmTTA M

%• ,,ff

A Y 10t

PA N _ _ _ _ _ _ _ _ _ _ _ _

1 " 0 -

Ulu,

.I E R G '' L ~ il l'

F r 2g ner .

Tloat~ Bul

A w di

too

,I II 1

tO

Figure 3.2. Impact points and coastline segment numbering. Arrows indicate typhoon direction toward impact points.

The model first calculates the joint probabilities (P) of the cyclone making landfall at each of the 62 target points at each of 48 hours during the forecast and having wind speeds of 13 values + 5 kt (30, 40 ... 150 kt). For each time/target combination, a most likely approach direc­tion and speed are calculated. The storm surge model output is then extracted for the most representative landfall point (usually the closest), the closest speed and direction and the output is interpolated for wind speed. The implied surge at all points neighboring the target point is adjusted for tides. If the adjusted surge is at least two meters above high tide, then P is added to a running sum of probability for the neighboring points at the current landfall time.

Figure 3.3 depicts the time evolution of storm surge probability for the same forecasts as shown in Figure 3.1. Notice the strong influence of astronomical tides on the probabilities. In comparison to Figure 3.1, the proba­bilities range up to 10% compared to 1% for a strike (Figure 3.1). This is because storm surge at a point at a particular time can be caused by a cyclone striking at other points (particularly those to the left) and at either earlier or later times.

The resulting probabilities are time integrated and summarized in message form fo: delivery to OFDA. An example. of such a message will be illustrated in Section 5.

Use of Information

In addition to incountry use, the wind and storm surge threat analysis can, be useful in its present form to alert the State Department Operations Center of a pending disaster.

-34­

3.4

+ 30- d-300° 100080

10

110

51 L/ 07 I0

0 51 1s 0(

hours 48

0 0

Figure 3.3. Similar to figure 3.1 except the upper insert graphs show the evolution over time of the probability of 2 m (or greater) storm surge above high tide. A strongastronomical tide dependence is apparent.

The color codes given in the OFDA message have the

following interpretation relative to the occurrence of 65 kt (or greater) winds or 2 meter (or greater) surge above high

tide:

COLOR MEANING

Red The probability of the event (over 65 kt winds or over 2 meter storm surge) is at least 20% within the next 18 hours. Disaster potential is high and

imminent. Preparations should be complete or near­

ing completion. These actions are expensive, and will not be done unnecessarily more than four times

for each time they prove to have been necessary.

Orange The probability of the event (see above) is at

least 10% within the next 24 hours. Disaster po­tential is high and approaching. This is the time

to order preparations of major cost: closing down

facilities, evacuating, moving, etc. These actions

are relatively expensive; they will prove necessary

one of every 5 to 10 times they are ordered.

Yellow, The probability of the event is at least 5% within the next 36 hours. Disaster potential over a broad

area is moderate but not imminent. This is the time to plan future actions and to preposition per­

sonnel and materials. These actions are relatively

inexpensive and will prove necessary once for every

10 to 20 times they are ordered.

Green The probability of the event is at least 2.5% with­in the next 48 hours. The disaster, potential is'

low and distant in time. This is the time to re­view plans and establish communications with re­sponsible parties. These actions are very inexpen­sive and will be performed 20 to 40 times for every one occasion of necessity. Stated another way, ac­tions will prove necessary for only 2.5 to 5% of the area in which it is ordered.

3.5 Other Applications

An equally useful application of threat data is for input to an acceptable risk model. The SAI Cyclone/Hurricane Acceptable Risk Model (CHARM0 ) is the only such model design­ed for tropical cyclone problems.

In these applications, specific actions (more usu­ally groups of actions) are considered in view of the threat from winds, storm surge and/or flooding. Based on cost/ ben­efit ratios, optimum sets of actions are recommended. One of the unique features of CHARM0 is that it is adaptive, that is, it adapts to an existing disaster control model, rather than requiring what may be a very familiar and effective lo­cal methodology to conform to some arbitrary standards.

3.6 Accuracy of the Storm Surge Threat Model

With statistical models it is customary to perform reliability testing on a large sample of independent tests. In the Bay of Bengal, cyclonea are relatively rare so that it would require upwards of ten years to accumulate a data base of sufficient size to properly address this problem. The wind and strike sections of the model are very similar to models in the western Pacific, eastern Pacific and Atlantic Ocean basins where adequate data have permitted extensive

-37­

testing; the storm surge portion has not been i. tested. However, the surge simulations have been compared with obser­vations of actual surge events, and appear to agree quite

well.

Probably the greatest question in the methodology is treating surge output from a landfall at one point as ap­plying for landfall at adjacent points. The limits over which this is permitted were carefully selected and the prob­abilities likely suffer much less than would corresponding deterministic forecasts; nevertheless, this remains an unre­

solved area of doubt.

4.0

SECTION 4

CYCLONE/HURRICANE ACCEPTABLE RISK MODEL (CHARM© )

Introduction

In the past century, man has made significant ad­vancement in science and technology. However, the effective

application of this progress has, at times, lagged behind the science. In no field of endeavor is this more evident than in the area of mitigation of natural disasters. Despite suc­cess in certain areas, we are continually faced with our

failure to prevent the preventable.

The detection and evaluation of the threat from natural hazards has been significantly advanced by science.

However, this is but the first in a series of steps which must be taken to effectively minimize losses from the fore­

cast danger. Many of the steps in any warning system are closely tied to the political and socio-economic structures

of society and, therefore, are often made complex beyond the task. However, any effective warning system must recognize

and reflect these facts.

The Cyclone/Hurricane Acceptable Risk Model

(CHARM0 ; SAI, 1982) utilizes a decision analysis system to integrate the best science available with an effective warn­ing system. It is the goal of CHARM* to reduce those factors in the decision making process which hinder or delay effec­tive warning, to simple decision matrices which can be effec­tively utilized by a computer, and thus speed the decision

making process. This section will describe the rationale of CHARM0 and provide a system description.

-39­

4.1 The Threat

The tropical cyclone has historically been a great killer of man. During 1970 alone, over 250,000 people died in the exceptionally high storm tides which swept the low ly­ing northern coast of the Bay of Bengal under the force of a tropical cyclone. A tropical cyclone can ravage an area with high winds, flooding from torrential rains, and abnormally high coastal water due to storm surge. High winds have the greatest impact within about 100 miles of the coast and lose strength as the storm moves inland. However, the heavy rains associated with the tropical cyclone can cause flooding hun­dreds of miles inland as the storm moves away from the coast­al area. Storm surge can cause the greatest impact in terms of death and destruction, but it is limited to the coastal

areas and tidal plains.

Loss of life and prevention of maximum destruction by tropical cyclones can be achieved by only a limited number of means - these include depopulation of threatened areas, construction of permanent defences, or the prediction of and adequate warning against individual storms. The first two solutions are impractical for most areas affected by tropical cyclones, and current warning systems unableare to provide reliable warning early enough to justify mass evacuation in most cases. Consequently, a combination of some staged tem­porary evacuation, limited barrier or shelter construction, when tied together by a sound disaster plan with well-defined and coordinated alternative action sets, may be the only re­alistic way to attack this problem. CHARM© integrates the prediction and warning phases of tropical cyclone mitigation efforts into a diaster plan, and aids on-scene decision mak­ing as well as optimizing action sets during plan design.

-40­

4.2 An Idealized Warning System

Providing warning of an impending natural disaster to the public is a complex process. It usually involves the interaction of physical, techno'Igical and social systems and requires careful coordination if the desired result--the avoidance of disaster or reduction of the scale of devasta­tion--is to be achieved. The successful operation of a warn­ing system usually requires the active participation of both local and regional government.

A valid warning consists of two discrete compo­nents, one of which highlights the existence of danger while the other evaluates and directs courses of action which pre­vent or minimize risk. Although warning systems may vary in size, scope, and effectiveness in reflecting their setting and the potential hazard, they increasingly involve the use of sophisticated technology and the cooperation of a variety of organizations and individuals.

An idealized warning system is illustrated in Figure 4.1 (from Foster, 1980). A system such as CHARM0 may not actively incorporate all the steps illustrated in Figure 4.1 but all of the steps must certainly be considered in the development of a CHARM0 system.

In the design and implementation of any warning system the initial step must be recognition of the danger. The recognition of the threat must be at the official level in order to deal effectively with the danger to a population

Official Official Recognition of Recognition of

Threat New Threat

Design of

Warin

Wrnistml av Sytm- L

lndsight

eiwAll-Clear

Transmission

of

Testing the.

arning System . :

a

Installation of

Warning RelatedTechnology and

and Action iTransis

-if f Wnv­

ducation User

D by E

DCto of Change, Collation

of Data, Evalua-ion of ThreatWanaig

Deiso

DeiinTassof

toaFedbango

Figure 4.1-. Idealized Warning System (from Foster, 1980)

from a natural hazard such as a tropical cyclone. Once re­cognized, the threat must be addressed by the design of a system to monitor and evaluate the hazard and issue warnings if danger increases beyond certain thresholds. (This is par­ticularly difficult in the case of tropical cyclones in the North Pacific or Atlantic where the average forecast error for 24-, 48- and 72-hour forecasts are approximately 120 n mi, 240 n mi and 360 n mi, respectively.)

Installation and operation of the system is follow­ed by education of the user groups. This includes both those within the infrastructure who will be a part of the system and the general public who must interpret the warnings and take appropriate action. Testing the system is important to insure that it is technically sound and that public interpre­tation and actions are correct. Modifications may be neces­sary at this stage. Such testing with modification is obvi­ously appropriate at system initiation, but should also be a regular part of system maintenance, perhaps eventually on an annual, pre-season basis.

The detection of change, collation of data, and evaluation of the threat is primarily a technical function. However, geophysical, demographic and socioeconomic consider­ations may impact the final threat evaluation. Decisions must be made as to who should be warned, when they should be warned, and how they should be warned. Transmission of the warning message or messages must then be made.

The receipt and interpretation of the messages and the action taken by the recipients form the final link of the warning chain but not the final step in the warning system. Information about the response elicited by warning messages

-43­

4.3

forms a feedback loop to initiate transmission of corrective or follow-on messages to the user groups. Transmission of an, all-clear message is made when danger has passed. A hind­sight review of the operation of the warning system during potential disaster situations is performed, and implementa­tion of any necessary improvements is then made.

Decision Analysis in:Warning

Once there has been recognition and evaluation of a threat, the magnitude of the threat must then be measured against an array of vulnerabilities, response requirements, and resources available for response, to determine: (1) what the specific hazard is, (2) who should be warned, and (3) how and when they should be warned. For example, elements of the military, civil defense, police, etc., will require notice at lesser threat levels and at longer lead times than the gener­al population. Similarly, a low-lying area which may require evacuation due to threat from storm surge may need early no­tification in order to complete the evacuation. There are many such considerations and decisions which must be made, during the period of time between recognition of the threat' and the arrival of the hazard (provided of course that time

is available).

In many ways this decision analysis lends itself to automation in that hazard levels can be pre-defined; vulnera­bilities, mobilization levels, and capabilities pre-determin­

ed; and requirements pre-assessed. Automation of the process, where possible, also improves the warning system. Valuable time is gained. A comprehensive and viable warning scheme is designed prior to actual threat, thereby reducing the proba­bility of error under stress during the event. All necessary participants can be programmed into the scheme, allowing them

,44­

ample time to accomplish their necessary tasks. Coordination between responsible parties is much more assured if warning is timely and keyed to the actual hazard. (Although the CHARM* system is keyed to tropical cyclones, other hazards such as tsunami warning and severe extra-tropical weather could be incorporated into the system.)

The automation (or semi-automation) of the decision analysis process must be based on consideration of all of the following elements of a disaster cycle:

a. Hazard Analysis

b. Vulnerability Analysis

*c. State of Preparedness

d. Prediction and Warning Systems'

e. Mobilization Capabilities

f. Assessment Analysis

g. Requirements Analysis h. Resource Analysis

i. Rescue and Evacuation Capabilities j. Emergency Assistance Preparedness

k. Risk Analysis

Each of the above must be evaluated for each geographic loca- ! tion and type of threat.

Hazard analysis considers the various aspects of the particular threat. This analysis would include the type of hazard (e.g., flood, wind, etc.) under initial considera­tion (primary hazard) and secondary hazards likely to occur as a result of the initial threat. The duration of the threat, the intensity or level of threat, and the geographic extent of the threat are a part of this analysis. This

information is basic to an evaluation of the potential hazard to the threatened area.

Vulnerability analysis is the evaluation of a spe­cific threat to a specific target. Vulnerability analysis may be based on threat to:

o Geographic areas o Population (demographic areas)

o Industries

o Types of structures o Low topographic elevations

o Etc.

Hazard analysis and vulnerability analysis are sim­ilar types of assessments and a valuable way to depict either is by mapping the hazard to show probable areal extent. This may take the form of a simple single hazard - one purpose mapping, single hazard - multiple purpose mapping, or multi­ple hazard - multiple purpose mapping, depending upon the complexity of the problem.

The state of preparedness of key organizations in­volved in disaster mitigation is a major factor in the suc­cess of minimizing the impact of the hazard. The pre-event training of personnel, the organization structure and cooper­ation of the local civil defense, Red Cross, local civil au­thority, police, and other key organizations, are all neces­sary elements in successfully handling the threat of a major hazard. Roles should be clearly defined prior to the actual event. Delineati11 of lines of authority in advance are also important to help prevent confusion. Pre-defined roles could be keyed to time steps during the approach of a tropical cy­clone to assure sufficient time to accomplish tasks.

-46­

The prediction and warning systems are the key ele­

ments in preventing a major disaster from occurring. All other aspects of preparation, coordination, mobilization, and response, are of little use without proper warning. It ie of paramount importance that forecast and warning systems func­tion at a high level of technical efficiency - all else de­pends upon this. The capability to identify the hazard, mon­itor changes in the hazard, and predict its target, is the primary purpose of the prediction system. Effective mitiga­

tion, however, requires not only technical capabilities but a highly effective warning system to warn the target areas. This entire sequence, from initial identification of the threat to the final all-clear message, is most likely to ben­efit from an automated, preprogrammed system such as CHARM when there can be adequate warning time for such events as a tropical cyclone or a tsunami.

The mobilization capabilities of a threatened com­munity can be an important element in minimizing hazard im­

pact. Speed of mobilization of key personnel, organizations and even the public itself, and the availability of the re­sources needed to achieve mobilization, are critical factors in determining the extent to which the hazard can be con­

trolled.

The capability to perform ongoing assessment analy­sis--prior, during, and after the hazardous event--provides important feedback into the system. The ability to assess preparedness, mobilization, etc., prior to the event allows

an opportunity to shore up weak areas. Assessment during the event is critical in evaluating immediate needs in crippled areas where unpredicted impacts have greatly increased the danger. After the hazard has passed, the assessment analysis is valuable for the recovery process, for identifying lessons

,47,

learned, and for subsequent feedback into the system to cor­rect problem areas.

Requirements analysis is an outgrowth of the as­sessment analysis. It determines the needed follow-up steps to correct those weak areas in the system which were outlined in the various assessments. Resource analysis is the identi­fication of the resources needed to meet requirements. Com­promise may be necessary in this analysis an risk analysis may aid in striking compromises.

Rescue and evacuation capabilities are usually those of the civil and military authorities plus any outside help available to meet the projected or actual needs. Pre­positioning of some of these capabilities may be critical for their effective use. Emergency assistance preparedness is the ability of the local government, civil or military, Red Cross, etc., to provide emergency medical, food and other necessary aid.

CHARM© , as its name implies, incorporates risk analysis as a part of the decision making process. Why risk analysis? First, because there is risk involved in the deci­sions made and second, because risk analysis provides a means to assess the problem in a more quantitative manner. Evalua­tions and decisions result in either no action or the imple­mentation of action. Each action (or non-action) carries with it a certain element of danger or risk of success versus failure. Even the act of issuing a warning concerning im­pending danger carries risk.

A forecast is a prediction that an event of speci­fied magnitude will take place at a certain time and loca­tion. It does not necessarily imply that any preparatory

-48­

or precautionary steps be taken as a result of this informa­tion. A warning, however, is a recommendation (or an order), based on the prediction, to take preparatory, protective or defensive action. The decision to warn therefore carries with it a great deal of responsibility. Once a warning is issued, particularly if it is for a severe event requiring much preparation, public confidence ir the agency issuing the warning is either enhanced or reduced according to the out­come. This places public response to future warnings at risk. For this reason, the decision to warn cannot be taken lightly and is generally accompanied by indecision and stress on the part of the issuing agency. CHARM0 is an attempt to reduce this indecision and stress through analysis of the preceding elements, and a logical, stepped-flow analysis pro­gram incorporating those essential elements with a pre­defined risk analysis program.

Risk analysis provides some quantitative results (cost/benefit) to aid the decision maker. It is only when expectations of anticipated loss of life, injuries and damage reach unacceptable limits that individuals are willing to disrupt their normal patterns of activity to turn to precau­tionary actions. Any disaster warning system must prevent more loss than it causes. This presupposes that the impact of a hazard, both with and without preparation, can be known with as much accuracy as the losses incurred in preparation. These costs cannot be known with certainty, however, they can be addressed in a systematic way with CHARM0 . This is the basis for an acceptable decision making process utilizing risk analysis. (It should be noted here that, although risk analysis may not be stated in an explicit manner, it is al­ways an inherent part of the decision making process.)

-49­

Two elements are introduced by CHARM© to reduce the uncertainty associated with issuing warnings. Probability is utilized in the prediction and risk analysis is performed using the probabilistic forecasts. The former tends to quan­tify the forecasts thereby increasing their utility to deci­sion-makers, while the latter quantifies the risks at each time step of the decision making process.

The function of the Decision Analysis Module may be more properly described as a decision-aiding process. The steps outlined in Figure 4.2 illustrate the stages involved in the decision making process. The decision analysis pro­gram will incorporate this process to provide the alternative solutions from which the decision maker may choose.

Step One is a stated objective by the decision mak­er from a pre-selected set of menu objectives. These objec­tives must be carefully chosen and worded and should utilize input from the appropriate elements of the disaster cycle listed and discussed earlier in this section. An example of a stated objective may be: When should X (specified) area be issued an evacuation warning?

Step Two is problem detection, diagnosis, and defi­nition. Problem detection could be the output on the SURGEP (surge probability) program forecasting a high probability of storm surge of unacceptable level for the area of concern. Diagnosis and definition would be accomplished utilizing stored data, current data from external sources, and output data from system programs. This data would serve as input to Step Three, alternative determinations.

In Step Three all fixed data are entered into each of a number of alternative responses. Then variable data,

-50­

I Objective of decision maker STEP ONE (menu select)

tProblemdetect­ion, diagnosis, STEP TWO and definition

Alternativ Alternativ Alternative STEP THREE

Outcome Outcome Outcome STEP FOUR 1 2 N " ' ,. :/i

! !I I Selection of • :'. STEP, FIVE Alternative S FIVE

Implementation 1E STEP'SIIX of Alternativej

Evaluation and Audit of STEP SEVEN Outcome

Figure 4.2. The Decision Process

-51­

i.e., that data which will drive the responses to different outcomes, are entered into each alternative response. Fixed

data could be entries such as:

1) the number of people in the area which must be evacuated;

2) the rate and mode of evacuation; 3) loss of evacuation routes as water level reach­

es X meters;

4) etc.

Variable data'.could be:.

1) hours for water depth to reach X meters; 2) reaction time of population in hours;

3) percent of population requiring forceable evic­

tion;

4) etc.

The results of each .of the alternative responses will be presented in Step Four, outcome projection. From these projections the decision maker can evaluate the prob­able outcomes and make a selection of one of these outcomes (Step Five) to implement (Step Six). An audit or evaluation (Step Seven) would be the final step. This may be both an ongoing evaluation and a final audit after the crisis has

passed.

It must be realized that in the process of problem definition if enough data are self-evident, then the use of N-alternative solutions is not necessary. In this case, the decision analysis program may be used in a simple analysis or predictive mode to aid the decision maker.

-52­

4.4 Modular Design of CHARM0

CHARM0 is designed to be a modular system incorpo­rating a group of sub-system level programs. These programs

as shown in Figure 4.3 are:

o WINDP/STRIKP - gives the probability (instanta­

neous and summed) of 30 and 50 kt (or other

specific) winds at a geographic point. o SURGEP - gives the probability that the storm

surge level will exceed X meters at a geographic

point.

o FLOODP - gives the probability that the flood

level will crest at X meters or will exceed X

meters.

o DECISION ANALYSIS - extracts, compares, and

evaluates analyzed data and provides alterna­tive, as well as pre-selected, solutions to de­

cision analysis situations.

o LOCAL DISASTER PLAN - a full-fledged disaster

plan or an integration program connecting CHARM

to an existing disaster plan.

CHARM0 is designed to be operated at the official decision making level of the government which issues disaster

warnings to both government agencies and the public at large.

CHARM© is an analysis and decision-aiding system which uti­lizes forecasts, assimilates local preparedness data and aids the decision maker in the vital task of mitigating the poten­

tial hazards from a tropical cyclone. In its simplest con­figuration, CHARM0 would consist of the WINDP/STRIKP Program

and the Decision Analysis Program. SURGEP and FLOODP pro­grams could be added as appropriate for the geographic loca­tion. The Local Disaster Plan Program is either a full­

-53­

INI SUI LOi WINDP SURGEP DECISION FLOODP LCTER ST DISASTER

S_____ ANLSI

Figure 4.3. Components 'of CHARM*.:

4.5

fledged disaster plan (if needed) or an integration type pro­gram connecting CHARM0 to the local disaster network. All of the programs of CHARMO are designed to be fully independent modules, with the possible exception of the Disaster Plan Program.

Certain components of CHARM0 are fully developed while others are under development or have not yet been fund­ed for development. The WINDP/STRIKP Program is fully devel­oped and is currently in use worldwide in areas threatened by tropical cyclones. This program is site- independent, there­fore it can be programmed easily to any geographic area for use. SURGEP is a site-specific program, that is, the program must be reprogrammed for each site depending upon the local bathymetric, geophysical and other conditions. It is cur­rently in use for the Bay of Bengal area only. FLOODP is a necessary site-specific module for those areas subject to deqtructive flooding from heavy rains associated with the tropical cyclone. It has not been developed to date. The Local Disaster Plan module can be very site-specific, de­pending upon the degree of development and effectiveness of the existing local disaster plan. The Decision Analysis mod­ule is the key element of the CHARMO concept and is described in this report in more detail, down to the component level.

Decision Analysis Subsystem Description

A good or "idealized" warning system would produce an expected or desired output from forecast data or predicted behavior of a threatening hazard. An increase in danger de­tected by a warning network, for example, should lead auto­matically to the target group response required to minimize losses. However, due to the complexity of the entire system required to produce this desired result, it is not often achieved. Failure to respond to warning may occur for a

-55­

variety of technical or social reasons which can arise during any one of the 14 steps outlined in Section 4.3. The Decision Analysis module of CHARM* utilizes the capabilities of the modern computer to monitor and analyze, in real time, much of the threat detection and warning system, as well as the capabilities of governments and communities to act.

In a typical warning system the decision maker is besieged with incoming data from many sources; the success or failure of the warning system may hinge on the ability of the decision maker to handle this data with discrimination. He must organize incoming data, extract that which is impor­tant, and prioritize tasks to be performed. He must assess the potential for disaster and anticipate public reaction to the forthcoming warning message, basing his judgment on the impacts of, and public responses to, similar past threats. This thought process forms the basis for determining, and then issuing, appropriate precautionary orders. The decision maker's dilemma may be illustrated by the diagram in Figure 4.4.

MOBILIZATION

~Figure 4.4 Decision Analysis or Decision Paralysis?

-56­

There is no magic in the process of good decision analysis. It requires a lot of hard work, proper preparp­tion, good organization, considerable pre-analysis and the best available data for input. It is possible to assemble a finely trained and well coordinated team to bring together all of the elements required. However, it is difficult to maintain a high degree of efficiency and skill over a period of time, due to human nature and changes in team personnel. The use of computer analysis aids in decision making can pro­vide a tested, systematic approach to the problem by outlin­ing the component steps of an effective response; this is in­valuable in mitigating the potential for disaster.

It is not possible, nor was it the intent of this section, to outline a complete data flow description of the Decision Analysis Module without a complete work-up of the problem leading to the actual programming. However, a de­scription of the program elements, probable information links, and typical examples of data contained in files, rou­tines, etc., will set the foundation for the module construc­

tion.

Figure 4.5 depicts basic data flow for CHARMO. Initial information on the threat, as well as changes in po­sition and intensity, flow from the various information sources directly into the hazard analysis programs and also into the decision analysis monitor and storage files. Output from the hazard analysis programs is sent to the decision analysis program, as well as to the vulnerability analysis progam to update vulnerability files. Output from this

'57.

Informatio Ge n Sources

WeatherCil Service DETECTION,Di A... o... . .:. ... ARNING A

Satellites MONITERTNG . Private

Radar SYSTEMSSYTM nepi

WMO Puli

etc. ec

F_I ' -- TIME

I00 PSTRIKE P SURGE PS FLOODL~ MATRICES

P~n(-gA?,tPROGRAM PROGRAM A A Y I R P R D

L HAZARD "" e-L

ANALYSIS MBmasic

ASSESSMENT

ANALYSIS

ANALYS IS ANA LY S IS EO

,re 4.i,5: CHARMoD -Basia.Data Flow­

program is fed to both the risk analysis files and the deci­sion analysis files. The decision analysis program is the decision maker's operating system to receive data, query other programs, check various files for status, and perform the various sub-routines of the decision analysis program in objective decision making.

As the initial and updated data flow into the sys­tem, the decision analysis program is used to check and up­date the status of the various elements which are critical in preparing for and handling the threat. The time matrices provide lead time necessary for the various agencies, groups and services to effectively perform their functions. Expect­ed Times of Arrival (ETAs) of various threats or threatening levels are fed into the time matrices and updated as new data arrive. Flags are set if ETAs drop below preparation time for critical actions, and a warning is sent back to the deci­sion maker through the decision analysis program. Continuous checks on preparedness, requirements, and resources are made via the assessment analysis program. As field data are re­ceived concerning these areas, the files are updated. Areas where initial preparedness is downgraded, for one reason or another, may require a diversion of resources prior to the threat levels reaching critical levels (i.e., unable to re­position). Mobilization is also monitored through the assessment analysis program and continually updated through­out the threat period.

Assessment analysis is also fed into the risk anal­ysis program, as this may directly affect the risk associated with certain decisions. The decision maker has the option to utilize risk analysis as appropriate, or as is deemed neces-, sary to aid the decision making process.

-59­

At appropriate times, dictated by required actions or by standard forecast times, decisions are prompted, re­quired checkpoints and procedures to follow are provided, ap-.

propriate files checked, necessary programs (if any) are prompted, etc., all leading the decision maker to the next necessary decision point in the local disaster plan.

A generalized outline of the principal elements in­volved in CHARM© is provided below in Table 4.1. The ele­ments are briefly described, probable system linkage is list­ed (this, however, is design-dependent), and typical types of data or programs within the element are identified.

-60­

Table 4.1 Relationship of Data Requirements and Information Links to Principal Elements of CHARM0

Element o Prediction/Forecast System

(official weather forecasts)

Information Links o Hazard Analysis o Decision Analysis o Source Systems

Typical Data o Forecasts o System Status o Source Data

(for data)

o Warning System (all \,arning systems utilized)

o Decision Analysis o Warning Systems

o Advisories, Warnings, All-Clears

o Official Orders o Feedback

o Hazard Analysis (WINDP/STRIKP, SURGEP and FLOODP Programs)

o Decision Analysis o Vulnerability

Analysis

o Outputs from: - WINDP/STRIKP - SURGEP - FLOODP (i.e., instantaneous and

zsummed probabilities of: - strike within predefined

radius

- wind exceeding predefined levels (e.g., 30 & 50 kt)

- storm surge exceeding specific levels (e.g., lm, 2m, ... , 8m)

- flood levels exceeding specified crest levels (e.g., lm, ... , 8m); these probabilities at 6-hr (or other) intervals and for specified lat/lon points

o Hazard Mapping - single hazard - one purpose - single hazard - multiple

purpose - multiple hazard ­ multiple purpose

Table 4.1 (Cont'd)

Element o Vulnerability Analysis

(files and program to define vulnerabilities to hazards)

Information Links o Decision Analysis o Risk Analysis o Assessment Analysis

Typical Data o Demographic Data o Geographic Data o Civil Engineering Data o Industrial Data o Communications o Transportation Routes o Psychological & Sociological Factors

o Vulnerability Mapping

o Time Matrices (time matrices to insure adequate warning &

o Decision Analysis o ETAs for specified threats/ threat levels

control timing of decision o Lead times for: points) - advisories to government,

groups, civil, military,

W-industry, etc. control times for vulnerability files

- control times for mobilization

- all advisories, warnings to public sectors

o Assessment Analysis o Decision Analysis o State of preparedness (e.g., O Vulnerability Analysis

o Risk Analysis

by geography, organization, industry, etc.)

o Requirements status o Preparedness File o Requirements File o Resources File

o State of mobilization (e.g., by geography, organization, etc.)

o Mobilization File o Threat vs readiness matrix o Time vs readiness matrix o Communications status o Assessment update o Analysis program

Table 4.1 (Cont'd)

Element o Preparedness Files

(set of files (or file)depicting current readiness)

Information Links o Assessment Analysis

Typical Data o State of readiness table (by

organization) o Notification status matrix. o Emergency assistance

preparedness o Requirements File

(,able of all requirementsneeded to combat hazard)

o Assessment Analysis o Requirements vs resource status table

o Requirements priority table o Resources File

(file of available resources)

aAssessment Analysis o Resources status table o Resources location/relocation

table

I

o Mobilization (files on-current status c4 mobilization)

o Assessment,Analysis

o Resource prioritization table

o State of mobilization (b"geographic area), i.e., - standby - preparation - mobilize - dismiss

o Geographic location of equipment

o Notification of key personnel o Mobilization lead time'tablI o Rescue and evacuation

capabilities

Table 4.1 (end)

Element o Risk Analysis (program

evaluates risk involved

Information Links o Decision Analysis

Typical Data o Cost/benefit analysis o Cost vs benefit tables

in action in a single or multiple outcome choice)

o Decision Analysis (program prompts, provides menues, etc., and guides decision maker to decision)

o Time Matrices o Assessment Analysis o Risk Analysis o Warning Systems o Detection/Monitor

o Decision analysis program (guides decision maker through process)

o Menu of advisories, warnings, official orders, etc., with

Systems specific wording o Advisory, warning status table

(e.g., by geographic region, agency, industry, etc.)

o Action priorities table & checklist

o Contingency tables o Historical event file (effects

of similar past events)

SECTION 5 WORLDWIDE TROPICAL CYCLONE THREAT ESTIMATION'svYqqRM

5.0 Introduction

The tropical cyclone, a severe storm 'of tropical origin, represents one of the most destructive forces on earth. Referred to as typhoons, hurricanes, tropical storms or often just cyclones, severe storms of this type are known worldwide. They form in tropical waters a few degrees awayfrom the equator in both hemispheres and generally move west­ward, then turnoften poleward. Many times they either strike land where they dissipate or they die on an eventual eastward track in the mid-latitudes. This tropical cyclonelife cycle may take up to 2 weeks to complete. Their threat is great to shipping, but it is greatest to populated coastal areas. Because of the threat to shipping, the Department of Defense (DoD), primarily the Navy, cooperating with the Department of Commerce, has developed a worldwide capability to track and forecast these storms. Various other government agencies make use of this information for various purposes. In particular, the Department of State uses DoD forecasts to­gether with other information to describe the general world situation.

The DoD forecasts commonly used predict the track and maximum winds over the next three-day period. These de­terministic forecasts represent the best estimate of the fu­ture, but by no means represent the only possible future for the storm. For this reason it is necessary to allow for other reasonable outcomes. This is beyond the capability of all but a handful of trained tropical meteorologists. The Navy, with the cooperation of the Office of Foreign Disaster Assistance, has developed a series of computer models which

-65­

interpret the forecasts in terms of threat indices. The Navy will be running these models on a routine operational basis and will provide a summary of the threat indices to the State Department Operations Center and affected foreign service posts. This section will describe this modeling effort.

5.1 The Threat

The usual ways in which tropical cyclones impact populated areas are with winds, torrential 'rainscausing flooding, and wind-induced abnormal tides called storm surge. Of these three impacts, storm surge is responsible for per­haps as high as 90-95% of cyclone-related deaths. High winds and storm surge are a problem in coastal (including island) areas, while storm surge is confined to the tidal plains. Winds are a serious problem only for a zone within 50 to 100 miles of a coast. Heavy rains, on the other hand, can bring disastrous consequences such as landslides and flooded streams and lakes hundreds of. miles inland along the narrow path of the storm as it dies overland.

All these impacts are highly localized so that a small deviation from the forecast would translate to an all or nothing difference in the local impact. The threat models quantify the risk of disastrous phenomena by using probabili­ty. The messages which will be provided consider the threat of hurricane force winds (winds over 74 mph or 63 kt) world­wide and the threat of storm surge of at least 2 meters above high tide in the Bay of Bengal. The accompanying threat of fresh water flooding is not addressed because the technology to do so with reasonable reliability has not yet been devel­oped. This, as well as the storm surge threat for the re­mainder of the world, may be added piecemeal to the existing program when possible.

-66­

5.2 Meaning of Threat Indices

The threat indices are color codes: green, yellow, orange and red for increasing levels of threat (higher proba­bility) and decreasing lead time. These are most meaningful to officials responsible for taking actions which would re­duce damage and loss of llfe. Since the threat will general­ly be to foreign soil, this will apply to the State Department only to tic extent that U.S. embassies, missions and consulates become involved. Other State Department ac­tions may include advising potential relief agencies of an imminent disaster or to some extent planning coordinated ac­tions with other interested governments.

When actions are taken based on either a probabili­ty or a forecast, after the fact they may prove to have been either unnecessary or inadequate. For example, when a hurri­cane threatens, the U.S. National Weather Service typically covers a coastal area of about 300 miles with a hurricane warning when only about 100 miles of coast will actually re­ceive hurricane force winds. Thus the probability of experi­encing hurricane force winds for a single point within the warning area is about one-third (33%). Coastal points adja­cent to the; warning area are threatened but the probability of hurricane force winds is less. People in those areas face an acceptable risk from the Weather Service point of view. The concept of overwarning (300 miles were warned for 100 miles to be affected) is a necessary precaution in the face of uncertainty to insure adequate warning for those who will be affected.

Table 5.1 describes the four color codes in terms of lead time, probability level, overwarning factors and key words which describe types of actions generally appropriate

-67­

Table 5.1 Description of color codes.

a. RED The probability of the event (65 kt winds or greater; 2 meter or greater storm surge*) is at least 20% within the next 18 hours. Disas­ter potential is high and imminent. Prepara­tions should be complete or nearing comple­tion. These actions are expensive, and will not be done unnecessarily more than four times for each time they prove to have been neces­sary. (Overwarning factor is less than 5).

b. ORANGE The probability of the event (65 kt winds or greater; 2 meter or greater storm surge*) is at least 10% within the next 24 hours. This is the time to order preparations of major cost: closing facilities, evacuating, moving, etc. These acions are relatively expensive; they will prove necessary once for every 5 to 10 times they are ordered. (Overwarning fac­tor is 5-10.)

c. YELLOW The probability of the event (65 kt winds or greater; 2 meter or greater storm surge*) is at least 5% within the next 36 hours. Disas­ter potential over a broad area is moderate, but not imminent. This is the time to plan future actions and to preposition personnel and materials. These actions are relatively inexpensive and will prove necessary once for every 10 to 20 times they are ordered. (Over­warning factor is 10-20.)

d. GREEN The probability of the event (65 kt winds or greater; 2 meter or greater storm surge*) is at least 2.5% within the next 48 hours. The disaster potential is low and distant in time. This is the time to review plans and establish communications with responsible parties. These actions are very inexpensive and will be performed 20 to 40 times for every one occa­sion of necessity. Stated another way, ac­tions will prove necessary for only 2.5 to 5% of the area in which it is ordered. (Over­warning factor is 20 to 40.)

* Storm surge probabilities apply to the Bay of Bengal only.,

-68­

Table 5.2. Schedule issue times for warnings in, various. ocean basins. Special warnings may be issued at' other' than scheduled times as circumstances warrant.

AGENCY TIMES OF ISSUE

N.E. Pacific NWOC1 Pearl Harbor 00,06,12,1800 GMT (+WSFO2 San Fran)*

N. Cent. Pacific NWOC1 Pearl Harbor 00,06,12,1800.GMT

(+WSFO2 Honolulu)* N.W. Pacific NOCC/JTWC3 Guam 00,06,12,1800GMT S.E. Pacific NWOC1 Pearl Harbor 03 and 1500 GMT. S.W. Pacific NOCC/JTWC3 Guam 07 and 1900 GMT Atlantic NEOC4 Norfolk 04,10,16,2200 GMT

(+NHC- Miami)*

S. Indian NOCC/JTWC3 Guam 02 and 1400 GMT N. Indian NOCC/JTWC3 Guam 04,10,16,2200 GMT

NWOC1 (Naval!: Western Oceanography Center), WSFO2 (Weather Service Forecast Office, NOCC/JTWC 3 (Naval Oceanography Command Center/Joint Typhoon Warning Center), NEOC4 (Naval Eastern Oceanography Center), and NHC 5 (National Hurricane Center). *Where a Department of Commmerce (DoC) agency is involved, warnings are issued by DoC with DoD coordination.

-69­

5.3

to each threat level. The values selected for lead time and threat levels are based on an evolving body of knowledge in­cluding U.S. military experience with readiness conditions, relating the color codes to National Weather Service hurri­cane watches and warnings, and limited independent research. It is reasonable to expect that some adjustment of these values will be required over time.

InformationFlow

During the summer and fall monthS in both the nort.­hern and southern hemisphere (i.e., all- year) tropical cy­clone formation is possible. Most of the time the cyclones are far out at sea and of little concern to interests other than shipping. The Navy's Fleet Numerical Oceanography Center (FNOC) in Monterey, CA, routinely receives tropical cyclone warnings from units in Norfolk, VA; Pearl Harbor, HI; and Guam which form a world-wide tropical cyclone forecasting network. These warnings are received at six or twelve hourly intervals during the life of the cyclone. Table 5.2 shows the normal issue times for warnings in various ocean basins.

These warnings are received at FNOC and manually entered into the FNOC computer system. The entry of the warning triggers the execution of a series of programs. Within several of these programs is a call to a routine which creates the State Department message. These *products are shown in Figure 5.1 (Bay of Bengal only) and Figure 5.2 (world oceans except Bay of Bengal).

Appendix A lists the cities for which threats are evaluated. For the most part these cities are those in trop­ical cyclone prone areas with populations greater, than

"70"­

FROM: FLENUMOCEANCEN MONTEREY CA

TO: SECSTATE WASHINGTON DC //OPS CENTER//

(others to be determined)

INFO: (Various Defense or Navy staff and Navy

Oceanography Command activIties some of

which will be included only during or

after checkout period)

(Various State Department or other

agencies with interest in the Bay of Bengal)

BT

UNCLAS //N03140//

SUBJ: BAY OF BENGAL CYCLONE THREAT

1. CYCLONE 31-38 LOCATED AT 18.4N 86.5E AT 12092000Z WITH 70KT CENTER WINDS CONSTITUTES THE FOLLOWING THREAT AS

DETERMINED BY AN OFDA AUTOMATED THREAT MODEL USING A DOD FORECAST FROM JTWC GUAM (REFER TO USERS MAN NO. XX).

A. THREAT OF HURRICANE FORCE WINDS ON COASTAL AND ISLAND

LOCATIONS

THREAT LEVEL SECTION OF COASTLINE

DANGER (RED) 19.2N 84.7E (INDIA) TO 20.5N 86.3E (INDIA)

ALERT (ORANGE) 18.5N 83.9E (INDIA) TO 22.ON 89.8E (B DESH)

CAUTION (YELLOW) 18.5N 83.9E (INDIA) TO 22.3N 91.7E (B DESH) WATCH (GREEN) 17.8N 83.1E (INDIA) TO 21.5N 92.3E (B DESH)

B. THREAT OF STORM SURGE OF AT LEAST 2 METERS ABOVE HIGH TIDE THREAT LEVEL SECTION OF COASTLINE

DANGER (RED) THRESHOLD NOT EXCEEDED AT ANY POINT

ALERT (ORANGE) 21.8N 87.8E (INDIA) TO 22.4N 90.8E (B DESH)

CAUTION (YELLOW) 21.2N 87.0E (INDIA) TO 22.4N 90.8E (B DESH) WATCH (GREEN) 21.2N 87.0E (INDIA) TO 22.3N 91.7E (B DESH)

BT

Figure 5.iSample message providing'threat,estimates for the Bay of Bengal.

-71­

FROM: FLENUMOCEANCEN MONTEREY CA TO: SECSTATE WASHINGTON DC // OPS CENTER //

(Others to be determined)

INFO: (Various Defense or Navy staff and Navy

Oceanography Command activities some of which will be included only during or after

checkout period)

(Various State Department or other agencies with interest on an area basis)

BT

UNCLAS //N03140//

SUBJ: TROPICAL CYCLONE WIND THREAT 1. TYPHOON CLARA SEP 18, 1981 1200 GMT LOCATED IN NORTH WEST,

PACIFIC AT 15.2N 127.1E WITH CENTER WINDS OF 75 KTS.-ESTIMATE OF THREAT OF HURRICANE FORCE WINDS FOLLOW:

OFDA SELECTED POINTS THREAT LEVEL

MANILA ROP CAUTION (YELLOW)

HONGKONG BCC CAUTION (YELLOW)

TAINAN ROC DANGER (RED)

KAOSIUNG ROC DANGER (RED)

TAIPEI ROC ALERT (ORANGE)

HANGZHOW PRC CAUTION (YELLOW)

SHANGHAI PRC WATCH (GREEN) 2. THREAT TO ALL OTHER OFDA POINTS BELOW WATCH ((RERN RTRR;A '

THRESHOLD FROM THIS. STORM.

BT

Figure 5.2 Sample message providing threat estimates for areas other than the Bay of Bengal.

-72­

5.4

250,000. Several Aesser cities are included if they are key cities politically, for example, a capital city or the main city of aa island group. With the exception of U.S. Territories, all the cities are foreign.

No Threat Messaqe i

Because of the confined space in the Bay of Bengal, the existenice of a cyclone is sufficient to indicate a sig­nificant threat to life. In the broad expanse of other ocean areas this is not the case. Consequently, a threat message will be generated for each Bay of Bengal warning received, but this is not a reasonable procedure for other world oceans. When none of the cities listad in Appendix A are significantly threatened, no message whatsoever will be gen­erated. Under this system ambiguous situations can arise, because it is possible that one or more of the following con­ditions could be true even if no warning message is generat­ed:

a) No storms exist; a) One or more storms exist but listed cities are

not threatened; or c) There is a breakdown in the sy!tem at some

point.

once the procedures are well established, a breakdown should, rarely occur. With the virtual elimination of breakdOwn~. the ambiguity is of little consequence.

-73­

6.0

SECTION 6

TECHNOLOGY TRANSFER

Goals and Accomplishments

The goal of technology transfer was to encourage the use of systematic decision aids in tropical cyclone di­saster preparation actions. Of course, this goil underlies the larger objective of more effective preparedness actions and, consequently, mitigation of disastrous ofthe effects tropical cyclones. Surprisingly, the greatest positive im­pact of this program has been at home, where the U.S. National Weather Service (NWS) has initiated a primitive ver­sion of the probability models described above; this was well received by the public in the 1983 hurricane season. Incoz­poration of these models into the NWS would not have occurred had not OFDA and the Navy begun issuing probabilities in the Caribbean (OFDA) and the southeast states (Navy). However, virtually no OFDA warnings were actually transmitted to for­eign service ports in the Caribbean because of an inactive hurricane season; as a result, few countries have actually been exposed to these data products in real time. Despite this lack of experience, Mexico, Jamaica and Venezuela have expressed interest in the products. There has been a consid­erable amount of interest in these products in other areas, such as the Philippines (Dr. Kintanar coauthored with SAI scientists a paper on this subject), Fiji, Burma, Bangladesh, India, China and Australia.

-74­

Section 7

REFERENCES

ESCAP, LRCS, WMO, 1977. Guidelines for disaster prevention and preparedness in tropical cyclone areas. Geneva, 1977.

Foster, H.D., 1990. Disaster Planning - The Preservation of Life and Property. Springer-Verlag, Wew York, 275 pp.

Frank, N.L. and S. A. Husain, 1971. The Deadliest Tropical Cyclone in History. Bul. Amer. Met. Soc., 59, pp. 438-444.

Jarrell, J.D., S. Brand and D. S. Nicklin, 1978. An analysis of Western North Pacific Tropical Cyclone Forecast Errors, Mon. Wea. Rev., 106, 7.

Jarrell, J.D., 1981. Tropical Cyclone Wind ProbabilityForecasting, Naval Environmental Prediction Research Facility, Contractor Report, CR 81-03.

Neumann, C.J., 1975. A statistical study of tropical cyclone positioning errors with economic applications. NOAA technical memorandum NWS SR-82.

Neumann, C.J., 1978. Verification of Tropical Cyclone Forecasts (excerpt from proposed manual on tropical cyclone forecasting for the WMO).

Neumann, C.J. and G. S. Mandel, 1978. Statistical predictionof tropical storm motion over the Bay of Bengal and Arabian Sea, Indian J. Met. Hydrol. Geophys., 29, 3, pp. 487-500.

Nicklin, D.S., 1977. A statistical analysis of western North Pacific tropical cyclone forecast errors, Naval Postgraduate School MS Thesis.

Reid, R.O., and R. E. Whitaker, 1981. Numerical model for astronomical tides in the Gulf of Mexico. Report to U.S. Army Engineer Waterways Experiment Station. Dept.of Ocn., Texas A&M University. Contract DA-CW39-79-C-0074. 115 pp.

-75­

APPENDIX A

Office of Foreign Disaster Assistance

List of Cities

for Tropical Cyclone Threat Analysis

-76­

ARABIAN SEA (14)

LAT LONG

INDIA (8)

Trivandrum 8:28N 76:57E

Calicut 11:15N 75:46E

Bombay 18:58N 72:50E

Surat 21:1ON 72:50E

Baroda 22:18N 73:12E

Mangalore 12:52N 74:52E

Bhaunagar 21:46N 72:09E

Jamnagar 22:28N 70:04E

PAKISTAN (1)

Karachi 24:52N 67:03E,

IRAN (1)

Abadan .0:2.N48:16E

IRAQ (1)

Al-Basrah.: 3000 4:,7

S. YEMEN (1)

Aden 12:45N 45:12E

tJAE (1)

Dubayy 25:18N 55:18E

OMAN (1)-

Masqat 23:.37N 58: 35E

A-I

NORTH PACIFIC (53)

THAILAND (1)

Bangkok

VIETNAM (4)

Ho Chi Minh

Da Nang

Hue

Haiphong

HONG KONG (1)

CHINA (9)

Shantou

Xiamen

Fuzhou

Ning-Bo

Hangzhou

Shanghai

Tsingtao/Qingdao

Tientsin/Tianjin

Luda

TAIWAN (3)

Kaohsiung

Tainan

Taipei

S. KOREA (3)

Inchon/Seoul

Kwangju

Pusan

LAT LONG

13:45N 100:31E

10:45N 106:40E

16:04N 108:13E

16:28N 107:36E

20:52N; 106:41E

22:15N" 114:10E

23:23N 116:41E

24:28N 118:07E

26:06N 119:17E

29:52N 121:31E

30:15N 120:10E

31:14N 121:28E

36:06N 120:19E

39:08N 117:12E

38:53N 121:35E

22:38N i20:17E

23:00N 120:11E

25:03N 121:30E

37:28N 126:38E

35:09N 126:54E

35:06N 129:03E

-2

JAPAN (22)

LAT LONG

Naha, Okinawa 26:13N 127:40E

Kagoshima 31:36N 130:33E

Kumamoto 32:48N 130:43E

Nagasaki 32:48N 129:55E

Fukuoka 33:35N 130:24E

Kitakyushu 33:53N 130:50E

Hiroshima 34:24N 132:27E

Matsuyama 33:50N 132:45E

Kurashiki 34:35N 133:46E

Okayama 34:39N 133:55E

Himeji 34:49N 134:42E

Kobe 34:41N 135:10E

Osaka 34:40N 135:30E

Nagoya 35:1ON 136:55E

Hamamatsu 34:42N 137:44E

Shizuoka 34:58N 138:23E

Yokosuka 35:18N 139:40E

Yokchama 35:27N 139:39E

Tokyo 35:42N 139:46E

Chiba 35:36N 140:07E

Niigata 37:55N 139:03E

Sendai 38:15N, 140:53E

PHILIPPINES (4)

Manila .14:35N 121:OOE

Cebu 10:20N 123:40E

Davao 7:04N 173:36E

Bacolod/Lloilo 10:40N 122:57E

PACIFIC ISLANDS (6)

Agana, Guam

Truk

Palau

Saipan/Tinian

Yap

Ponape

SOUTH PACIFIC (24)

AUSTRALIA (7)

Brisbane

Darwin

Perth

Cairns

Newcastle

Sydney

Norfolk Island

NEW ZEALAND (4)

Auckland

Manukau

Wellington

Whangarei

AFRICA (2)

Durban, S. Africa

East London,

S. Africa

LAT LONG

13:28N 144:45E

7:25N 151:47E

7:30N 134:30E

15:12N 145:45E

9:31N 138:06E

6:55N 158:15E

27:28S 153:02E

12:28S 130:50E

31:56S 115:50E

16:55S 145:46E

32:56S 151:46E

33:52S 151:13E

29:02S 167:57E

36:52S 174:46E

37:02S 174:54E

41:18S 174:47E

35:43S 174:19E

29:55S 30:56E

33:OOS 27:55E

A-4

ISLANDS (13)

LAT LONG

Madagascar

Tulear 23:21S 43:40E

Diego Suarez 12:16S 49:17E

Fiji

Nandi 17:48S 177:25E

Suva 18:08S 178:25E

Port Louis 20:1OS 57:30E

Mauritius

Noumea,

New Caledonia 22:16S 166:27E

Nukualofa,

Tonga Is. 21:08S 175:12W

Easte.i Is. 27:07S 109:22W Niue Is. 19:02S 169:52W

Apia, Somoa 13:50S 171:44W

Page Page,

Am Somoa 14:16S 170:42W

Rarotonca is. 21:14S 159:46W

Saint-Denis,

Reunion 20:52S 55:28E

NORTH ATLANTIC (29)

Nassau, Bahamas 25:05N 77:21W Hamilton, Bernuda ' 32:17N 64:46W

St. Johns, Antigua 17:06N 61:51W

Roseau, Dominica 15:18N 61:24W

Fort-de-France,

Martinique 14:36N 61:05W

Castries,

St. Lucia 14:01N 61:01W

A-5

NORTH ATLANTIC (continued)

Bridgetown,

Barbados

Kingstown,

St. Vincent

St. Georges,

Grenada

PUERTO RICO (2)

San Juan

Ponce

VIRGIN ISLANDS (2)'"

St. Thomas

St. Croix

Pointe-a-Pitre,

Guadalupe

Port of Spain,

Trinidad & Tobago

Willemstad,

Netherlands Antilles

CUBA (2)

Santiago de Cuba,

Havana

Kingston, Jamaica

Santo Domingo

Dominican Republic,

San Salvador,

El Salvador

Barranquilla,

Colombia

LAT LONG

i3:06N 59:37W

13:09N 61:14W

12:03N 61:45W

.8.28N 66:07W

1: 1 N .6.6:37W'

18:21N _64: 5 5W

17:45N 64:45W

16.14N 61':32W'

.­

10:39N 61:31W

12:06N 68:5W

20 i:0N 75:49W:

23.-08N 82:22W

18:,'N 76:48W

18:28N 69:54W'

13: 42N 89:12W

30:59N 74:.48W,

A-6

VENEZUELA (2)

LAT IONG

Maracaibo 10:40N 71:37W

Caracas 10:30N 66:56W

MEXICO (3)

Tampico 22:13N 97:51W

Vera Cruz 19:12N 96:08W

Matamoros 25:53N 97:30W

Belize, Belize 17:30N 88:12W

Port-au-Prince, Haiti 18:32N 72:20W

EASTERN PACIFIC (2)

MEXICO

Mazatlan 23:13N 106:25W

Acanulco 16:51N 99:55W.

A-7