statistics of the chosen parameters are given in tables 1 through 

 10 (appendix B) which indicate maximiim variations of the temporal means 

 (over a period of years) and spatial means (from a given 1-degree quad- 

 rangle to the next). The period of time covered hy the data is too short 

 to determine any possible cycle or trend. Assuming normal distribution, 

 confidence limits of the observations about the mean can be computed from 

 X ± 2Sx = 95^ confidence limits where Sx is the standard deviation. Like- 

 wise the confidence limits of the mean can be computed from X ± t'SE, 

 where t is the t distribution and SE is the standard error. 



The ranges are indicative of year-to-year variation of the thermal 

 structure dioring any given month. Unfortunately, a correlation exists 

 between the magnitude of the ranges and the number of years of observa- 

 tions. For example, the mean of the ranges of all groups of 6 or more 

 years of surface temperature observations is 4.i|-°F, and the mean of the 

 ranges of all groups of 2 to 5 years of observations is 2.5°F, 



Standard deviations and standard errors are more variable and are 

 larger than expected in some cases, probably because of slide processing, 

 reference temperature, instrumental, key punch, and coding errors. All 

 data were carefully screened to minimize key punch and coding errors. 

 The remainder of the standard deviation value after elimination of errors 

 can be ascribed to variation in time and space. In mathematical terms if 

 o"s denotes the standard deviation of the observations, its components 

 can be subdivided as follows: 



^s'-°-o' + ^' + a,l + ai} + <r^l +a,%+c;^, (i) 



where the subscripts denote components due to field operator, instru- 

 mental, key punch, coding, slide processing, and reference temperature 

 errors, and space and time variations, respectively. 



Population means and standard deviations of the various parameters 

 for the area were computed by the LGP-30 by sorting the SERC deck by 

 months and years for the 20-minute quadrangle centered at 35°N,48°W. 

 The statistics from this analysis are shown for the various thermal 

 structure parameters in appendix B, 



Instead of computing and then examining the data for significant 

 spatial or temporal differences between the means of a given thermal 

 structure parameter, a more powerful tool known as the 2-way analysis of 

 varian'ce (unequal numbers in subclasses) was employed. The technique, as 

 outlined in Kendall (2), was programmed for the LGP-30. 



The 2-way analysis of variance was considered to be only a gross tech- 

 nique, for example, checking computations for significant difference between 

 1-degree quadrangles over a number of years for a given month for a given 

 thermal structure parameter (historical data). When no significant differ- 

 ence existed between means, a probability of 100 percent was assigned to it. 



