H 

 II »* -T— , for deep-water conditions 



and p 



H mo L 2 

 II—*- ^— j , Ursell's parameter for shallow-water wave conditions 



( 2 " d s) 



Figure 10 shows little or no systematic error for the prediction method based 



on L D . Figure 10 also shows that the percent error ranges from -33 to 



+44 percent, but that for most tests the error is within about ±10 percent. 



18. Approximately 25 percent of the tests had a percent error greater 

 than ±10 percent. Because of this, it may be useful in some critical or life- 

 threatening situations to use a value of ^ max greater than the expected 

 value produced by Equation 5 when using the recommended coefficients. Fig- 

 ure 11 shows how the percent error which has been normalized by the standard 

 deviation of the data set a seems to have the shape of a normal distribu- 

 tion. To test this hypothesis, namely, that the percent error has a normal 

 distribution, a Kolmogorov-Smirnov (K&S) test was performed. This test is 

 used to determine whether or not the data deviate a statistically significant 

 amount from the assumed normal distribution model (Cornell and Benjamin 1970). 



19. The K&S test indicates that the normal distribution for error should 

 be accepted at the 20-percent significance level. A 20-percent level is a 

 more severe criterion than a 10-percent level as it indicates there is a 

 20-percent chance of rejecting a model which is in fact true, a Type I error. 

 The 20-percent significance level is the most severe criterion commonly tabu- 

 lated for the K&S tests. Recognizing that errors have a normal distribution 

 provides an easy way to give more conservative estimates of ^ max than is 

 provided by a regression equation. Generally, about half the errors are above 

 the regression curve and about half are below, so the curve represents a 50_ 

 percent exceedance level. In Figure 12 a more conservative trend is shown 

 above the regression curve. The conservative curve was generated by increas- 

 ing the runup regression coefficient a by two standard deviations of the 

 percent error, i.e., 



a Q (conservative a) = a ( 1 + 2o) = 1 . 143 ( 1 .0 + 2 x 0.1286) = 1.437 



The value of the runup coefficient b remains the same, i.e., b = 0.202 . 



15 



