Summarization of Historical Data 



Regression techniques could also supply a new approach to 

 summarizing historical sea-surface temperature data. It is 

 assumed that it is possible to develop a realistic regression model. 

 In this study the model was derived by a pseudo-objective method 

 involving a trial -and-error approach to determining the inter- 

 action terms. Before regression modeling can become completely 

 satisfactory as a method of sea-surface temperature summariza- 

 tion, it will be necessary to develop objective methods of deter- 

 mining the main effects and their interactions. If it is assumed 

 that a physically acceptable regression model can be developed, 

 it might still be asked how such a model can yield estimates of the 

 day-to-day and location-to-location sea-surface temperature. An 

 unpublished NEL study suggests that an 8-to-lO-year time-series 

 record of sea-surface temperatures is long enough to produce 

 reliable long-term estimates that are Independent of the time 

 period of observation. Thus, if there is available a 10-year 

 record of sea-surface temperatures covering the area for which 

 the regression model was developed, the regression equation can 

 be fitted to each year of data to provide 10 yearly sets of regres- 

 sion coefficients. A sample could then be drawn from each yearly 

 distribution and combined into a composite sample which could be 

 considered a sample drawn from the 10-year time period. The 

 regression equation could then be fitted to this composite sample 



75 



