the relative prevalence of winds from cold quarters (see Diagram 5,
fig. IX combined with the fact that " warm " winds and " cold " winds
(Diagram 4, figs. 2 and 3) are relatively colder at that time than at
other times of the year; whereas the cooling effect in October
November may be attributed to the prevalence of the winds we have
called " temperate " (i.e., N.W. and S.E. winds), which at that time of
the year become especially cold.
The July maximum corresponds to exceptional warmth of the generally cold or temperate groups of winds (figs. 1 arid 3 of Diagram 4), and the January maximum corresponds especially to the frequency of occurrence of warm winds.
Effect of the Length of Period Analysed on the Position and Amplitude of the Curve.
As has been stated above, these effects have all been studied for Kew, and reference has always been made to the analysis of the tioenty-fiv&year mean curve of temperature for Kew. It will, however, be seen that a possibility of error is here introduced, since it cannot be assumed that the mean curve for the nine years on which our results- have been mainly based would have the same harmonic coefficients as the curve for the twenty-five years. Analyses were therefore made of the curves of mean temperature for various periods of years at Kew (Table I). It will be seen that the first-order curves for the twenty- five years arid nine years are practically identical, and the positions of the maxima of both the first- and second-order curves are also almost identical for the two periods of years. The third- and fourth-order curves are larger for the shorter period, but are still relatively small, so that the curves of difference from the twenty-five-year first-order curve, which we have been discussing above, fairly represent (as has been assumed) the nine-years second order curve.
It may here be of interest to make a further study of Table I. It will be seen that not only the twenty-five and nine-year curves have been analysed, but also the curves for the five years and four years which made up the nine years, and again for four single individual years. Of these individual years, the year 1884 was chosen at random, the year 1873 on account of its extremely warm winter, and the year 1895 on account of its extremely cold and late winter. The results of the analysis for 1898 a year with exceptionally hot weather in August and September have been added to the table. It shows a retarded first-order curve, with a second-order curve of mean amplitude and rather early maximum. It will be noticed that the coefficients of the first-order curve are practically the same in all the analyses, except those of the very abnormal years. A still more striking phenomenon is the persistence of the second-order effect
