probability from the NED Boston stage-frequency curve, selecting events to 

 model, and fitting a curve to the raw modeling results. 



108. It is beyond the scope of this report to assign error bounds 



to the NED stage-frequency curve. However, a simple investigation of the pos- 

 sible error in the curve would be as follows. The curve was based upon 131 

 years of record, 57 of which were from a continuous record at the NOS tide 

 gage. Because of the relatively long record, the bottom portion of the curve 

 (i.e. return periods of less than 15 years) should be very accurate. The 

 middle portion of the curve (i.e. return periods between 15 and 100 years) is 

 within the length of record and should be accurate to within a few tenths of a 

 foot. The portion of the curve above the 100-year return period would be more 

 uncertain with, of course, the uncertainty increasing with return period. 

 However, because of the extremely flat nature of the curve (there is only a 

 1-1/4 ft difference between the 50- and 500-year levels), it seems safe to 

 predict that the curve should be accurate to within a half foot even at the 

 500-year return period. 



109. The potential error from the curve fitting process can be best 

 seen in plots of raw versus regressed output. For Fox Hill Drawbridge, the 

 raw and the regressed still-water level stage-frequency curves, previously 

 presented in Figure 35, had a linear regression correlation coefficient of r 



= 0.997 . Figure 56 shows the raw versus regressed output for the "Wide Berm 

 + 1-ft Cap" alternative at Roughans Point which had a correlation coefficient 

 of r = 0.994 . These correlation coefficients are representative of those 

 occurring at all locations. The regression was highly accurate and poten- 

 tially introduced only minor error into the total process. The lowest corre- 

 lation coefficient was greater than 0.98 for both the still-water level loca- 

 tions and the interior of Roughans Point. 



Determining Error Bands for the Selection Process 



110. The selection process that determined which events were selected 

 for modeling was designed specifically for this project. As a result of 

 limited experience with this technique, it is much more difficult to determine 

 the potential error of the selection process as compared to the potential 

 error of the more familiar processes of data collection, data analysis, and 

 numerical modeling. In order to estimate the variability of the selection 



85 



