obtained in table 7 with the reservation that the resxilts are only- 

 approximate. In table 7, the 90 percent confidence limits are given 

 on the assumption that the observations consisted of 50 independent 

 height values. For example, for the observations made by the USCGC 

 MENDOTA on 2 September 1953 at 1200Z, the Observed Significant 

 Height (corrected) was 28 feet, and 90 percent of the time (under the 

 assum^ptions which were made) the true significant height would be 

 between 25 and 32 feet on the basis of many more observations. 



The forecast error as a departure of the forecast value from the 

 closest value of the 90 percent confidence limits is then entered as 

 the band error for 50 independent observations. 



The band error is a better measure of the discrepancy between the 

 forecast ajid observed values because it does not penalize the forecast 

 value for the unavoidable observational error which is due to the small 

 sample size. 



Three of the twelve forecasts are within the 90 percent confidence 

 of the observations. Four more are within three feet of the 90 percent 

 confidence limits. When the band error for an assumed 50 independent 

 height observations is studied, it is seen that the forecasts are quite 

 accurate. 



The last two entries of the table show the 90 percent confidence 

 limits on the assunnption that the heights are really only 25 independent 

 observations. This permits a spread of ten feet between the upper and 

 lower bounds of some of the limits. For more precision it is evident 

 that visual observations should consist of 100 observations at least, 

 in order that it would be possible to be sure of somewhere near 50 

 independent values. 



Under these conditions, four forecasts fall within the 90 percent 

 confidence limits. Five more fall within five feet of the 90 percent 

 confidence linaits. Under these conditions, though, the confidence limits 

 are so broad that the observations are of little use in saying anything 

 about the wave properties. One of the purposes of this paper is to show 

 that reliable observations are needed and that they cannot be reliable if 

 enough individual values are not observed. 



There is a consistent bias running through the data. The observed 

 values consistently run higher than the forecast values. Much more 

 data need to be collected before this bias can be established as real 

 or false. There is, though, a possible explanation for this bias. It 

 is that the observers did not keep an eye exactly at a fixed point on 



29 



