262 THE ATLANTIC. [chap. v. 



Ill order to eliminate as far as possible from tlie results of 

 our serial temperature soundings errors depending upon irreg- 

 ularities in the action of the thermometers, it has been found 

 necessary in all cases, instead of trusting to their individual 

 indications, to construct a free-hand curve for each series, and 

 to take the indications from the curve. If the readings of the 

 thermometer are plotted to scale, and if we attempt to con- 

 struct such a symmetrical curve as that represented in Fig. 53, 

 a page of the Curve-book, selected at random as an example, 

 the curve naturally passes through the greater number of the 

 plotted points, leaving out one or two at a greater or less dis- 

 tance at either side. Where many of the thermometers are 

 astray, as not unfrequently happens when the serial sounding 

 is taken in heavy weather, this process requires to be performed 

 with some judgment, and is liable to a certain amount of error ; 

 but it is wonderful in a series of such curves how strong the 

 internal evidence is of their accuracy. A certain marked tem- 

 perature phenomenon, for example, is indicated in a certain 

 locality by an irregularity in the curve ; and as we recede from 

 the cause of disturbance, the irregularity gradually dies out, to 

 be replaced very probably by an irregularity due to some other 

 cause. This is well shown in the curves representing the grad- 

 ual change of temperature from west to east in the North At- 

 lantic and the North Pacific (vol. i., Figs. 100 and 101, pp. 362 

 and 366). The temperatures used in the text and in the tem- 

 perature sections are taken from such curves. In the tables in 

 the Appendices the temperatures are given as they were read 

 from the thermometers, after applying the known corrections 

 for pressure and error of zero-point, in order that the actual 

 data from which the curves w^ere constructed might be sup- 

 plied. This will explain the discrepancies which frequently 

 occur between the temperatures referred to in the text and 

 those given in the tables. 



Eeferring, in the first place, to the distribution of tempera- 



