estimate which represents a random variable which has a probability distribu- 

 tion. If this probability distribution can be determined, confidence intervals 

 could be calculated by specifying the probability that the true flood level 

 lies between a range of heights about the estimated value. Confidence inter- 

 vals are relatively easy to determine when dealing with a single data set, for 

 example, confidence intervals about the mean value of a set of data. However, 

 the calculation of stage-frequency curves as done in this study involves mul- 

 tiple data sets and multiple modeling systems. Even if it were possible to 

 determine confidence intervals about each of the processes separately, there 

 would still be the problem of combining separate intervals into one interval 

 for the final stage-frequency curve. The total 90 percent confidence interval 

 would not be the sum of the 90 percent confidence intervals of all the pro- 

 cesses. For example, the storm surge model may overpredict, the wave model 

 underpredict, and the probability model assign too low a probability. Conse- 

 quently, no attempt will be made to place error bounds on the final curves. 

 Instead, a verbal description of the types and, where possible, the magnitudes 

 of the various sources for error will be given. A method has been developed 

 to show curves for the error associated with the process of selecting a 

 limited number of events to be modeled from the infinite number of possible 

 events. Since the physical modeling was not a part of this report, no attempt 

 will be made to determine the potential for error from the physical model- 

 ing. The reader will have to analyze the following paragraphs and determine 

 how the possible error will influence any engineering decisions. 



104. The modeling of still-water level involved three main parts: data 

 collection and analysis, numerical model calibration, and simulation. The 

 tide gage data used in the project were carefully screened to remove spurious 

 data points; therefore, this information was probably corrected to about 

 0.1 ft. Calculating accurate tide time-histories was difficult. Five sets of 

 tidal constituents, each based on an analysis done for a different time 

 period, were tested. Due to the large tidal range at Boston, slight errors in 

 the phase of the predicted tide can cause significant errors when calculating 

 the storm surge time-histories. The storm surge time-histories used for 

 combination with tide were edited by eye to remove any errors caused by poor 

 tide prediction. The numerical grid, as shown by the calibration results, had 

 sufficient resolution to accurately model tide in the areas where calibration 

 data were available. WTFM has performed well in numerous studies, and the 



83 



