to spread out the confidence limits, since the effective number of 

 observations is decreased by the correlation between values. Another 

 model could be constructed in which two successive complete waves 

 are correlated with the value unity and all parts are independent. 

 Such a record (again an impossible one) would be like figure 4. 



Then N wave height observations are really N/2 independent obser- 

 vations. With the use of N/ 2 instead of N in table 3, the confidence 

 limits given in table 3 with a safety factor were obtained. This safety 

 factor is thus a very crude attennpt to estimate the effect of stronger 

 correlation between successive waves. 



Figure4. An Artificial Wave Record. 



Trough Amplitude Equals Preceding Crest Amplitude. 



Heights Correlated with the Value ot Unity, Two by Two 



5. Practical Conclusions 



From these results, it is evident that at least 50 wave heights 

 must be observed and tabulated before a reliable estimate of the 

 average wave height (or v^) ^^^ t)e found. Very little confidence 

 can be placed in an estimate based on 25 values; 100 values would 

 be much more reliable. If 100 values are observed, if every wave 



18 



