the sea surface. If each observation had a little extra height added 

 to it as the observer looked along the crest to the highest part of the 

 crest, then the average of these heights would tend to be higher than 

 the average of the heights of the waves passing a fixed point. 



11. Other Errors 



There is finally the question of the reliability of the height estimates 

 as made by visual observations. Can an observer estimate the wave 

 height of a wave thirty feet high within plus or minus two or three 

 feet? Any such error, if consistent, in the estimation of the individual 

 wave heights would introduce errors in the reported values. Very 

 little is known about the nature of such errors, but there does seem 

 to be a tendency to overestimate wave heights when visual observations 

 are made. A cheap easily used instrumental aid for the measurement 

 of wave heights would be a very useful device to be supplied to ship's 

 personnel taking wave observations if such an instrument could be 

 devised. 



When the possibility of observer error, in addition to statistical 

 error, is considered, it is seen that the results of table 7 are a good 

 test of the theories given above and of the forecasting methods which 

 were verified against the height observations. 



12. Summary 



In summary, based on the above results, the following rules can 

 be given for the visual observations of wave heights: 



(1) The heights of the waves passing a fixed point should be 

 observed. (The point could also be fixed relative to a moving 

 ship,) 



(2) All heights should be recorded (or if this is too difficult, 

 all heights in excess of a fixed lower bound should be observed 

 and the theory of the truncated distribution then used). 



(3) At least fifty values, preferably one hundred values, should 

 be recorded, 



(4) Table 3 then gives values for the confidence limits to be 

 placed on the observations. The value is more exact theo- 

 retically if all waves are observed, and it is approximately 

 correct when a truncated distribution is used. 



30 



