79 



i. If no solution, or a spurious solution, is found the solution parameters 



are adjusted, and the optimization repeated. 

 ii. If a good solution is found, progress to the next window. 



5.3.1 Pre-Processing of Record 



Accommodating Measurement Error 



Measurement error can be a major source of difficulty with any high order data 

 interpretation method. Local methods can be especially sensitive, as each window 

 solution relies on the detail contained in a small segment of the record. In applying 

 the LFI method to a single pressure trace, Sobey (1992) found it necessary to apply 

 a simple moving average filter to field and laboratory data. In his work, the primitive 

 kinematic free surface boundary condition was used, requiring an estimate for the 

 local gradients in the water surface. In the current work, a modified version of the 

 boundary condition (Eq. 5.3) is used which does not require these gradients. This 

 makes the method less sensitive to noise, so it was not necessary to apply any filtering 

 for the results presented here. 



If there is substantial noise in the measured record, the system of equations can be 

 substantially overspecified, allowing the least squares optimization to accommodate 

 the noise in the record. When this is possible, it is preferable to applying a smoothing 

 filter to the record, as it does not impose any assumptions on the nature of the record. 

 However, if the error bands are very large on the data, it may still be necessary to 

 apply filtering to the raw measurements. 



The Mean Water Level 



The mean water depth, /^, must be specified as part of the potential function. 

 As a time series of the water surface is provided, it is a simple matter to compute a 

 mean of the measured records. The mean should be taken over a period much longer 

 than a typical wave period, but short enough to accommodate changes in the local 

 water level due to astronomical and storm tides. In keeping with the local nature 



