The median of the 10 estimates was used as a reference (except for skewness 

 and kurtosis, for which the reference values were zero). Any estimate that 

 deviated from the reference value by a preset amount was flagged by the 

 computer on a printed output summary. The preset deviation comparators were 

 found empirically by examining time-series records of malfunctioning gages for 

 comparison with records from functioning gages. The comparators are not 

 listed here because they apply to the set of gages used in this experimental 

 setting and so are somewhat unique (however, the principle can be applied 

 elsewhere) . 



102. To test for temporal stationarity, the standard deviations were 

 used. The square of the standard deviation is the variance (pressure variance 

 is related to sea surface variance) which is proportional to wave energy. So, 

 if the pressure standard deviations were stationary, wave energy was expected 

 to be likewise. To check for stationarity, the median standard deviations of 

 the four records were intercompared. The median of these was used as refer- 

 ence in this test and as a scale of the collection as a whole. The criterion 

 for stationarity was that no median standard deviation depart from this 

 reference by more than 15 percent. 



103 . The integral parameters are good for information about a record or 

 collection as a whole. However, if the data have a few isolated spikes or 

 short duration offsets, these properties may not be obvious because of the 

 smoothing effect of integration. Sharp changes in time -series data result in 

 increased variances over broad regions in frequency spectra and so must be 

 avoided if not real. To check for sudden changes at any point in collected 

 data, the maximum and minimum values, maximum and minimum first derivatives, 

 and maximum and minimum second derivatives were isolated from the full 16,384- 

 point time series for each gage. Both dimensional and nondimensional forms 

 were computed. Normalizing parameters were the time-step (0.5 sec in all 

 cases) and the reference standard deviation from the stationarity check above. 

 The medians of these point parameters were found from the set of 10 gages, and 

 deviations were flagged as for the integral parameters above, also using 

 empirically determined comparison criteria. 



104. If one or more properties of a gage were flagged, that gage was 

 not used in analysis. The analysis programs were written so that data from a 

 subset of gages could be analyzed. The trade-off for fewer gages is a loss of 

 directional resolution. Loss of some gages is more critical than others. For 



40 



