Tables VM and VIII give a breakdown of the statistical analysis for each 

 forecast interval and for each OWS location and Argus Island. In some examples, 

 the number of observations is not large enough to be statistically valid, hence 

 conclusions cannot be drawn from these tables for all categories. Table IX 

 presents a side by side comparison of each station's statistical analysis for the 

 USWB and FNWF inputs, together with various combinations and a final summa- 

 tion of all the observations. The combined statistical analysis, showing the bias 

 for significant wave heights to be -1 .2 feet and the RMS error to ±5.8 feet, 

 comes within an acceptable value when consideration is given to the fact that a 

 sampling variability in the average observed significant height of about 10% is 

 built into the statistical analysis as noted by Pierson and Tick (1965). 



The concluding evaluation data are shown in Figures 16 to 37. The 

 machine predicted frequency-amplitude wave-spectra histograms are super- 

 imposed on the computed wave meter or wave staff spectra, as given by the 

 equipment on the OWS station or at Argus Island. If there is good correlation 

 between these two spectral curves, then certainly the predicted wave spectra 

 are acceptable. In choosing samples for this report, the procedure was to select 

 examples by visual inspection of: the greatest correlation for each station loca- 

 tion without regard to the input or the forecast interval (Figures 16 to 20); the 

 greatest correlation for each forecast interval without regard to the input or the 

 station (Figures 21 to 26); and (Figures 27 to 31 and 32 to 37) the least correlation 

 in each of the same two categories. In choosing the greatest and least correlation 

 samples, the significant height correlation was considered with other factors being 

 equal in order to select the most appropriate examples. The chosen representative 

 set of samples has observed wave heights from 3.4 to 36.4 feet with the majority 

 between 10 and 30 feet. The observed and predicted wave heights are shown on 

 each figure as well as the date, time, forecast interval, and location. 



An inspection of Figures 16 to 20 shows that even the greatest correla- 

 tion samples for station A and Argus Island could be improved considerably; whereas, 

 the correlation samples for stations I, J, and K coincide quite closely. The poor 

 correlations for station A and Argus Island may be the result of the station locations. 

 Both stations are relatively near the boundaries of the gridpoint network with 

 respect to the prevailing winds. 



The numerical procedures normally would tend to be least accurate in 

 these regions since the propagational effects would not be so accurately projected 

 there as at stations I, J, and K which have a much wider expanse of water in the 

 western semicircle. Furthermore, at station A the location is such that the paths 

 of the low pressure centers are to the south of or possibly very close to the station. 

 The winds would be very much more irregular there than for the other three OWS 

 locations so that a slight discrepancy in the predicted path and speed of the low 



