200

MONTHLY WEATHER REVIEW

JUNE1967

A PRELIMINARY REPORT O N NUMERICAL SEA CONDITION FORECASTS

ABSTRACT
Since July 1956 the Joint Numerical Weather Prediction Unit a t Suitland, Rld., has been makingmachine forecasts of seaconditions on an operational basis. These prognoses are based on the 1,000-mb. wind forecasts derived from the two-level, thermotropic model currently in use at JNWP. Two different sea-forecast models have been tested to date. The first utilized only the forecast winds at theend of the forecast period and therefore yielded fully developed waves. A model incorporating duration in a crude manner is now indailyoperation. This paper describes both methods, compares the numerical results with observed conditions, and outlines future plans.

sea. The segment which is customarily chosenincludes the top third of the height spectrum, and the average or In describing and forecasting sea conditions, one significant height of these waves is what one usually generally deals with two more or less distinct problemsreports and attempts to forecast. In this paper, it is to wind-driven waves, often called sea, and another class be understood thatthis convention has been followed is defined as waves which of waves called swell. Swell and all references to wave height really mean sign& have moved outside of the locale in which they were cant wave height. generated, whereas wind waves are those presently in a One of the products of JNWP which is available on a spec& generating area. daily basis is a numerical forecast of the 1,000-mb. In the &stmachine forecasts of sea conditions atpressure heights at points on a 30x34 grid covering aptempted a t the Joint Numerical Weather Prediction Unit proximately two-thirds of the Northern Hemisphere. (JNWP), only wind waveshave been considered. The The grid interval is 381 k m . a t 60 N. From these forereasons for neglecting swell are primarily two; first, the castheights one can readily obtainthe surface winds total problem is thereby greatly simplified, and second, in which serve as the basis for the sea-condition prognostimost cases swell is a less important factor in the preparacation. A two-levelmodel of the atmosphere developed does not tion of prognostic sea-condition charts.This by Thompson [l] is currently being used at JNWP to mean that swell is always negligible. As a matter of fact, make the 1,000-mb. forecast. in some areas (e. g., subtropical high pressure cells) swell Since the basic forecasts are an operational requirementis frequently the only contributing factor, more and placed on JNWP and are therefore run on a daily schedadvanced forecast models will undoubtedly to have ule, it was felt that sea-condition forecasts might turn incorporate this feature. outto be a relatively inexpensive byproduct.For this Considering wind-driven waves alone, then, the princireason it wasdecided to undertake an investigation of pal parameters whichdefine the height to which they numerical sea forecasting models wherein relevant paramwill Snally grow are windspeed, duration, and fetch. eters would be treated in decreasing order of importance. Obviously, a 50-knot wind blowing for only 10 minutes The machine forecasts described in this preliminary will not generate waves as high as the same wind blowing report represent the first phase of this investigation and for a period of 10 hours. Similarly, a 50-knot wind are essentially an attempt to apply numerical methods flowingacross a puddle 100 yards in diameter will not to the subjective, prognostic technique developed in the n generate waves as high as it would on a lake 100 miles i Division of Oceanography, U. S. Navy Hydrographic diameter. Office. The sea at any particular location is actually composed of waves covering a wide band of heights andwave2. FULLY DEVELOPEDWAVES lengths. I t has become more or less standard practice Probably the simplest model which one can use for the among oceanographers and forecasters to deal with a numerical prediction of sea conditions assumes that both certain segment of the entire spectrum wave when infinite. Under these conditions attempting to describe and/or forecast the state of the durationandfetchare the sea is said to be fully arisen, The height to which a *Any opinions expressed by the writer are his own and do not necessarily reflect the wave will grow is assumed to be dependent only upon the view of the Navy Department at large.

1. INTRODUCTION

JUNE 1957

MONTHLY WATHER REVIEW

201

windspeed and the maximum allowable height for any speed is often called the fully developed wave height. A number of empirical and semi-empirical relationships between significant wave height andwindspeed have been found in the case of fully arisen seas (e. g., Cornish [2], Rossby and Montgomery [3], Pierson, Neumann,and James [4]). These authors have found that the heights vary with different powers of the windspeed ranging from the first to the fifth. Since the numerical forecast yields 1,000-mb. pressure heights at grid points, the windis customarily computed over an interval of two mesh lengths. I n order to obviate the use of a square-root routine (thetotal wind mustbe determinedfromtwo components) theequation of Rossby and Montgomery was selected for computing the fully developed seaheights, namely :

A H=-V
9

(1)

where H is the fully developed wave height, A is a nondimensional constant taken to be 0.3, g is gravitational acceleration, and V is windspeed. In the application of equation (l), 70 percent of the geostrophic wind at 1,000 mb. was used for V. (This was done to take into account the various elements that produce subgeostrophic winds at the sea surface.) Using the 36-hour prognoses derived from the thermotropic model, numerical forecasts of fully developed wave heights were made at JNWP for several months. As one might expect, the results were far from good when compared with actual conditions. On the other hand, they were not completely discouraging either; in areas where the assumptions of unlimited fetchandduration were reasonably true, the predicted and observed wave heights agreed remarkably well, indicating that equation (1) is basically correct. The tendency was to predict high waves tobe too high and low waves too low. Theso discrepancies are primarily due, it appears, to the assumption of infinite fetch and duration in case of high waves, the and to the neglect of swell in areas of weak winds where the waves were predicted to be too low. Analyses prepared at the U. S. Fleet Weather Central (FWC), Washington, D. C., following the methods developed at the U. S. Navy Hydrographic Office (Schule and Ropek[5]) have beenused to test the accuracy of the numerical computations. The FWC analyses are based primarily on ship observations; however, in areas of sparse data, continuity, and computations utilizing the prediction curves of Bretschneider [6] and others are used tofill in. In figure l a is shown a chart of wave heights determined from equation (1) using the observed 1,000mb. height field a t 1500 GMT January 8, 1957, as input data. By comparingwith the sea-condition analysis a t 1.-Wave-height comparisons for January 8, 1957: (a) 1230 GMT for the same day, figure IC,one can readily see FIGURE JNWP analysis of fully developed waves at 1500 OMT; JNWP (b) that in general the patterns are quite similar; however, analysis at 1500 GMT using the four-class,durationmodel;and the waves obtained from the numerical computation are (c) Fleet WeatherCentral analysis for 1230 QMT. Wave heights much too high in the areasof strong winds. i feet. n

202
Test a t all sea pointsr Does W Direction (t-12) d i

MONTHLY WEATHER REVIEW

JUNE 1957

Wind Direction ( t )

25O7

I
I

I
and duration of 6 hours t o

Test a t a l l aea polnt8.uhere no wave height yet obtained: Does Wind Mrection (t-24) Wind Direction ( t ) 2 257

T e s t a t a l l sea points &era_ no wave height yet obtained: Does Wind Direction (t-36) Wind Direction (t) f 257

yr
4

Use mean wind last 21, hours and duration of 18 hours t o

no

FIQURE

2.-Flow diagram for four-class, duration model outlining the method of computation.

45

401
35
1-j

$ I

IO
6 hours

01
0

I O

20

Wind
FIGURE 3.-Graph

Speed In Knots

30

40

so

of wave-height versus wind speed curves used in the four-class, duration Duration test. in hours. (After Pierson, Neumann, and James [4].)

FIGURE 4.-(a) Example of a numerical, 24-hour wave forecast (two-class, duration) verifying at 1500 QMT March 28, 1957, and (b) Fleet Weather Central analysis of observed heights a t 1230 GMT the same day.

3. EFFECT OF DURATION
The curves of Sverdrup and Munk [7] and Neumann effect of duration is usually more important than that of fetch except in particular situations (e. g., strong offshore winds). I n ordertodetermine just how much the incorporationof duration would modify the fully developed wave patterns shown in figure la, a duration model was tested on the same case. Four observed 1,000-mb. fields a t 12-hour intervals prior to and including 1500 GMT January 8, 1957, wereused as
[8] indicate that the

input data. In other words, it was assumed a priori that a perfect wind forecast was available in order to eliminate this source of error from the comparison. The computationalprocedure for the four-class duration model is shown in theform of a flow diagram in figure 2. Time (t) represents the end of the forecast period; (t-12) is 12 hours earlier, etc. The wave height versus windspeed relationships for the four allowable duration times of 6, 18, 30, and 42 hours (from Pierson, Neumann, andJames [4]) were storedinthe computer. Figure 3 shows t,hese same data plotted as four duration curves on

JUIFE 1957

MONTHLY WEATHER REVIEW

203

a chart having windspeed as theabscissa and wave height as the ordinate. The results of the four-class duration computation are shown in figure lb. All-in-all, the wave heights agree more closely with those observed than in the case of fully developed waves. I n particular, the maxima are lower than those obtained from equation (1). The 21-foot maximum to the rear of the cold front off Boston is almost exactly reproduced bythe numerical method. The maximum ahead of the occluded front is still computed to be too high; there is some doubt, however, about the analysis here, for no ship reports were received from the area of strongest southerly winds. The maximum west of Ireland split into two centers in the duration test; nevertheless, the overall correspondence is good. The improvement in the Pacific area (not shown here) was even greater, especially around a severe storm in midocean. Throughout March 1957, 24-hour forecasts of wave heights based on an extremely simple, two-class duration model were madeon an operational schedule. I n this series of tests the 12- and 24-hour,1,000-mb. forecasts were used to determine the wind speed and direction. If thewind shifted more than 25' in direction during the 12-hour interval, a duration of 6 hours was assumed to apply; otherwise, the duration was arbitrarily taken to be 24 hours. Figure 4 shows an example of a numerical forecast and the verifying analysis. A summary of the results obtained duringMarch 1957 a t three locations inthe North Atlantic is shown in figure5. The forecast and observed wave heights transmitted via facsimile from the U. S. Fleet Weather Centralare presented for comparison. Monthly mean values of forecast minus observed heights (3'-0) divided by observed heights (0)have been determined for the three points. At latitude 55' N., longitude 15' W., the objective forecasts had a mean error of 43 percent compared with 19 percent for the subjective forecasts made at the Fleet Weather Central. This point is closest to the edge of the numerical forecast grid, and the predicted 1,000-mb. heights are apt to in error here dueto be boundary influences. This might help to explain this difference; however, whatever the cause, the "hand" forecasts were clearly superior a t this point. At the othertwo locations the numerical method (in spite of the crudeness of the model) wasclose to being competitive. Errors of 28 and 36 percent obtained for the JNWP forecasts compared with 25 and 29 percent for those from the weather central. It should be pointedout that there is a 2S-hour time difference between the end of the numerical forecast period and the observation time used in making the comparisons. Wave patterns can change appreciably in this numerical method is truly time, and it may well be that the competitive here.
4. CONCLUSIONS

1 Ai:
I
I

1 0

1 5

2 25 0
1957

5
1230 GMT

IO

1 25 20 5

FLEET WEATHER ANALYSIS MARCH

(FEET)

FIGURE 5.-Scatter diagram of predicted versus observed wave heights for J N W P and FWC forecasts (24 hour) at three Atlantic points during March 1957. Location of point and mean of forecast minus observed over observed heights, ( F - O ) / O , (percent) given in upper left of each diagram.

From the 9 months of caseswhich have been run at

JNWP, the following somewhat preliminary conclusions

1. Computations of fully developed seas yield fair results in general, but tend to overforecast high waves and underforecast weak ones. The results a t 24 and 36 hours are not competitive with subjective techniques. 2. Incorporation of ('duration'' intothe numerical forecasts led to an appreciable improvement in quality. Over the Atlantic during March 1957, the results a t 24 hours were almost as good as those obtained by experienced subjective forecasters. The extremely simple, twoclass, duration model used during March 1957 appears to be capable of forecasting the gross pattern rather well; however, it is also clear that a closer determination of the duration time is desirable. 3. Any numerical forecast of sea conditions will be only as good as the numerical forecast of surface winds upon which it is based. To date, themachine forecasts of low-level flow patterns have not been as good as those prepared by the experienced synoptic meteorologist(even though the reverse may be true already a t upper levels).

204

MONTHLY WEATHER REVIEW

JUNE1957

This need not discourage further research, however, for improvementin low-level numerical prediction canbe expected to continue. 4. Assuming that fetch is i n h i t e does not appear to be too restrictive in most cases. It would probably suffice to correct this approximation in regions of strong, offshore flow. 5. Neglecting swell can lead to significant errors in , limited areas under special conditions. Any refined forecast technique should attempt toinclude the movement of at least the largest waves outside the generating area and treat their decay for a limited period of t,ime.

of Oceanography, U. S. Navy HydrographicOffice, during the past year, and their suggestions for future lines of attack are mostwelcome.

REFERENCES
1. P. D. Thompson, On the Theory of Large-Scale a Two-Dimensional Baroclinic Disturbances in Equivalent of the Atmosphere, Quarterly Journal of the Royal Meteorological Society, vol. 79, No. 339, Jan. 1953, pp. 51-69. 2. V. Cornish, Ocean Waves and Kindred Geophysical Phenomena, Cambridge University Press, 1934, 164 pp. 3. C.-G. Rossby andR.B. Montgomery, The Layer of Frictional Influence in Wind andOcean Currents,

5. FUTURE PLANS
The IBM model 701 electronic computer which was used in all of these tests is being replaced at JNWP by the IBM 704. On the 701, a numerical forecast of wave Hemispheres ocean heights for two-thirds of the Northern area required only 5 minutes of machine time (once the 1,000-mb. fields were available). When low-level wind forecasts are forthcoming from the IBM 704, it should bepossible to cover the entire Northern Hemisphere in 5 minutes and at the same time use a more refined seaforecast model. With these considerations inmind, the present plan is f i s t to program a model which will determine durationtothenearesthourand correct for fetch in the vicinity of the coasts. Because of the changeover to a new machine, wave forecasting on an operational a two-level basis will be discontinued at JNWP until atmospheric model is running on the IBM 704.

Papers in Physical Oceanography and Meteorology,

Practical Methods for Observing and Forecasting Ocean Waves by Means Wave Spectra and Statistics, of U. S. NavyHydrographic Office, Publication No.
603, 1955. 5. J. J. Schule, Jr., and J . F. Ropek, U. S. Navy Hydrographic Office Synoptic and Prognostic Wave Charts, U. S. Navy Hydrographic Office Technical Report No. TR-16, 1955. 6. C. L. Bretschneider, Revised Wave Forecasting Curvesand Procedure, University of California, Institute of Engineering Research, Technical Report No. 155-47, series 29, issue 47, 1951. 7. H. U. Sverdrupand W. H. Munk, W i n d , Sea, and Swell: Theory of Relations for Forecasting, U. S. Navy Hydrographic Office Publication No. 601,1947. 8. G. Newmann, On Ocean Wave Spectra and a New Method of Forecasting Wind-Generated Sea, U. S. Beach Erosion Board, Technical Memorandum No. 43, 1953, 42 pp.

ACKNOWLEDGMENTS
The author would like to express his appreciation to Capt. W. E. Oberholtzer, Jr., USN, for his continuous encouragement duringthe progress of this study.The subjective forecasts used inthe comparisons were prepared by the duty aerologists at the U. S. Fleet Weather Central, Washington, D. C. Anumber of veryfruitful discussions were held with staff members of the Division