The above tabulation considered in conjunction with an inherent error 

 of ±.02 in °"t(h) and a very probable ^rror of ±.05 in S ,h, when de- 

 termined from a noncontinuous salinity sounding (in depth), means that 

 the end points of the density inequality in 15 may easily be in error 

 by ±.04 and also that the correct value of ^(h) ma y actually lie out- 

 side the range of the inequality, especially when 15<S<25. The like- 

 lihood of salinities being lower than 15°/°° i s small unless a sounding 

 is taken in an inshore region of intensive fresh-water runoff or unless 

 there is am influx of ice from colder regions which is melting at the site 

 of the sounding. The Zubov-Defant model does not attempt to account for 

 these nonstatic, nonconvective effects upon the water mass. Therefore, 

 under the conditions of 0-1), (12), and. (15), the reference line for possible 

 ice formation is°f (S) just as for the case of S< 24.70 °/oo. The equation 

 of the new reference line is 



(16) 

 0" f (S)=0.8075S-0.087. 



Table 2 will reveal that 0-6) has a maximum error of 1.01, or at most half 

 the experimental error in G t(h). Maximum density can be represented to 

 the same degree of accuracy by 



o- m (S)=0.8029S-0.030, (1?) 



while the temperature of maximum density (Sverdrup, 1942) is 



T m (S) = 4.00°C-0.2l50 S. (18) 



From the arguments used in deriving (10^ with the additional restric- 

 tion that 



T o,h ST h<^ 5 o,h] (19) 



when 



a f< S o,h> <cr t (h) - "m< S o l h>» (20) 



it can be concluded that if a column of water (o,h) has coordinates 

 L S ,h; °"-t,(h)] which plot above the freezing density line (0.000) in 

 Figure 2, i.e., if 



cr t (h)> 0.8075 S 0jh - 0.087, (21) 



