Density of Water Masses in Ocean, Vertical and Horizontal Density Distribution 199 



Observations of more recent expeditions have shown that negative stabilities extend- 

 ing down at the most to about 250 m are of such a frequent occurrence, that they 

 are difficult to account for by observational errors alone. For example, in ninety-five 

 cases with E greater than —100 the observational errors must be 0-04%o in S or 1°C 

 in temperature. There is, however, further confirmation of the reality of this pheno- 

 menon. This comes from the occurrence of negative values throughout the entire 

 layer, and the fact that mostly a pronounced regional distribution of stations with 

 negative values of E is found which would scarcely be possible if random observa- 

 tional errors would have been made. In the Atlantic, for example, there is an extended 

 area with negative values of E in the entire open ocean from 50° S. to 20° N. The 

 highest negative values (< —200) fall within a latitudinal zone between 15° and 20° S. 

 and there is probably a corresponding zone also in the North Atlantic approximately 

 between 20° and 30° N. 



This instability in the top layer in tropical and subtropical areas must be due to the 

 eff"ectiveness of evaporation. The increase in salinity and the decrease of the temper- 

 ature at the surface leads to an increase in density and to a reduction in stability. 

 Solely incoming radiation during day time works in the opposite direction, which 

 compensates the density increase by a corresponding rise in temperature, but during 

 night time when incoming radiation is missing and evaporation continues, the density 

 increase will predominate and negative stability values can persist for a considerable 

 time as long as the intensity of evaporation is sufficient. It is, however, a rather pe- 

 culiar phenomenon that a vertically unstable stratification can be maintained for a 

 longer time over such an extended area in the top layer in spite of convection and 

 mixing. 



Fig. 92. Circulation in a convection cell according to Benard. 



Perhaps a possible explanation lies in the "convection cells", first observed and 

 investigated experimentally by Benard (1901). He was able to show that when a rela- 

 tively thin layer of a liquid with volatile components was cooled by evaporation, the 

 entire mass of the liquid divided into a number of cells. In each of these the liquid 

 rises in the centre, diverges in the upper part of the cell and descends again in the 

 outer parts as shown schematically in Fig. 92. The diameter of the cells corresponds to 

 about three or four times that of the thickness of the liquid layer. Instability in the 



