Page:Proceedings of the Royal Society of London Vol 69.djvu/77

Rh

A second classification was then made in which, while the types of weather were still distinguished, curves of the difference between first order curve values and mean daily temperature observed during the prevalence of barometric gradients for each of eight points of the compass, were separately drawn. Each of these curves showed certain characteristics which corresponded with some of those of the second- order curve ; either the February maximum or the May minimum or some other characteristics were prominent, but no single one of the curves for the separate gradients showed a complete curve with all the characteristics of the second-order curve. It was, however, very remarkable that the curves for the same winds during different types of weather were closely parallel a result which confirmed the pre- vious observation that neither the mean difference in temperature between cyclonic and anticyclonic days, nor the relative frequency of their occurrence, could give rise to the second-order variation in atmospheric temperature, or indeed to any other simple periodic effect. It was therefore decided to abandon the separate classification of cyclonic and anticyclonic days.

Examination of the curves above described showed that the number of days on which any particular wind occurred during a ten-day interval in five successive years, was not sufficient to eliminate the effect of exceptional days from the mean value obtained. A further period, from 1880-1884, was therefore examined, and the curves were constructed from monthly means, instead of from ten-day means, for the nine years. Thus sufficient observations were obtained to give mean values which might be regarded as independent of single exceptional variations from the mean. The observations which were retained for the investigation were those relating to the days which had been previously classed as cyclonic or anticyclonic ; the days which were not classified, on account of the want of any definite indication, were omitted. That this omission did not appreciably affect the results arrived at, may be inferred from the fact that one of the curves (Diagram 6, fig. 1, p. 75) could be obtained in two ways first, by taking the temperature difference and frequency of wind for the included days, and, secondly, by taking the temperature difference from the actual means for all days. The results were quite identical in the two cases.

The effect of wind-direction on temperature was investigated by means of curves whose co-ordinates are respectively proportional to