APPRAISAL OF RESULTS 21 



which are characterized by a high (tropical- 

 type) tropopause showed larger errors at 16 

 km than at 8 km, the reverse of the usual trend 



in table 2. It is apparent from inspection of 

 table 2 that errors in recovering N (z) at alti- 

 tudes of 3 km or more are likely to be small. 



10 



POINTS AFFECTED BY LOW-LEVEL SUBSIDENCE 



REGRESSION LINE 



AN M - 14 54 + 06437 AN f ± 640 



r = 795 



J L 



20 30 40 50 60 70 80 00 100 110 120 



AN[ FROM EXPONENTIAL MODEL 



Figure 7. Correlation of AN; monthly mean ^-profiles versus monthly 

 mean exponential model. 



Table 2. Absolute errors in recovering mean N from map 

 contours for 32 stations (in N-units). 



The total variance, o- T 2 , in using the maps of 

 N(z) given in appendix A can be estimated in 

 terms of the following error model : 



V = <V + "m 2 , (7) 



where * s ' is the variance of 5-year mean values 

 (as compared to long-term means) , °m 2 is the 

 variance of errors in mapping N (z) . Random 



instrumental errors are included in ° & 2 . The 

 value of a 5 can be estimated at about 2.5 iV-units 

 (table 1), while a M may be estimated at 1.5 X 

 P.E., where the probable error, P.E., is given 

 approximately by the mean absolute errors in 

 table 2. These would yield 2.1 N at 3 km, 2.4 N 

 at 8 km, and 1.4 N at 16 km, for °m. A reason- 

 able estimate for °m at the surface (0 km) 

 would be 1.0 iV-unit. These estimates may be 

 combined to yield approximate "t values, as 

 shown in table 3. Minimum and maximum val- 

 ues were obtained by permutations of the. values 

 in tables 1 and 2. The standard errors of the 

 5-year means have been assumed to be a con- 

 stant percent of mean N(z), independent of 

 altitude. 



