]\gt ; Time Factor in Correlation between Barornetric Heights .
403 ' On InflueYlce of the Time Factor on the Correlation between the Barometric Heights at Stations more than 1000 Miles apart By F. E. -BROWNE-CAVE , Girton , Cam .
CoInmunicated by Professor KARL , F.R.S. Received June 3 , and in revised form , December 20 , 1904 , \mdash ; Bead June 16 , 1904 .
1 .
An investigation of the relationship of the daily barometric }heights on both sides of the Atlantic Ocean has been in progress for years , and in a preliminary note by Professor Pearson and myself* some account was given of the contemporaneous ; of a chain of stations from the extremes north of Norway down the west coasts of Europe and Africa .
Observations for this east side of the Atlantic have now been copied for twenty years , as far as stations available from Norway to the Cape , and the only need here is more aid in the very laborious reductions necessary before any inferences can be drawn .
A similar chain of stations from Nova Scotia to ths Falkland Isles has been completed , with the exception of Brazil , from which , so far , we have been able to obtain no data whatever .
2 .
It seemsd desirable , before entering on the cross Atlantic correlations , to obtain a better conception of how the time factor influences the intensity of correlation .
For this purpose two stations , lVilmington and Halifax , seemed specially appropriate .
They are both on the East Atlantic coast , the one in North Carolina and the other in Nova Scotia .
The results obtained for these stations.afford an illustration of the manner in which the theory of correlation can be used for the purposes of meteorological prediction .
In a paper published in 1898 , Professor Pearson and Dr. Lee indicated that the proper way to deal with the pressure relationship between two or more stations was to proceed by the method of correlation ; and they illustrated this by the simultaneous correlations of various stations in the British Isles .
The corresponding simultaneous correlations for stations from Norway to Sierra Leon were given by Professor Pearson and myself in the note already referred to .
The object of the present paper is to illustrate this method of correlation as applied to prediction , the fundamental points being ( n ) the choice of two stations which , although far apart , have a considerable correlation with at least 12 , and if possible , 24 hours ' the determination for any two stations of the interval for which the correlation is a maximum .
If the practical meteorologist is accustomed to predict roughly the barometric condition at a station , from stations and , then , if he has vations for twenty years or more for these stations , modorn statistical methods place him in a position to choose * ' Boy . .
Proc vol. 70 , pp. 465\mdash ; 470 .
' Phil. Trans , vol. 190 , pp. 423\mdash ; 169 .
404 Miss .
]-Cave .
Influence of the the best i1ltervals ) etween the readings in order to find , from anterior observations at the four stations , the most able barometric height at A a given day , and the most probable error of this esti- : mate .
They thus s1lpply a means of giving to his predictions a greater .
precision than is at present attainable .
It should also be remembered that , although the present paper only deals with barometric observa .
tions , the same methods could be applied to temperature , or to any other of the quantities with which meteorology is concerned .
In order to show that the relations between the two stations inly considered in this paper are not due to purely local conditions , corresponding ations have been made for two stations in the Southern Hemisphere , St. Helena and Cape Town .
Although the shortness of the period ( 1893-98 ) for which ) servations taken at St. Helena tilable , prevents these resuIts from being as fully reliable as those tained for the North American stations , yet they sufficient to show the general nature of the pressure relationship ) the two places , and the influence of the time factor .
The latitudes and longitudes , and hours of observation as follows Halifax , N. , W. ; 7Vilmington , W. to June 30 , 1888 , Cape Town , S. , E. ; 8 St. Helerla , S. , W. ; 9 This gives arc of about , or ) 1090 miles ) etween Halifax and Wilmington , , or about miles betwcen Town and St. Helena .
Roughly speaking , Halifax is north-east of Wilmington , and Cape Town south-east of St. Helena .
The twenty years of the data for Halifax and Wilmington divided into two decades , 1879-88 and 1889-98 , and the summer winter months ( equinox to equinox ) were dealt with separately .
As was pointed out in the previous note , there is a marked differencs between the summer and winter resuIts , and two ten-year periods desirable for mutual control .
It soon became obvious that the summer winter time factors were different in character , the for ximum correlation being considerably shorter in summer .
The lnmbers obtained indicntc that change of pressure passcs yyilmington to Halifax , i.e. , from south-west to it Pearson ; nsti tovided tingtou aifax anadi nayed bnadian MOffice kerintended topying o United States ; the readings for Cape Town } owe to the medialion of Dr. W. .
Shaw , who procured from the Cape Meteoroiogical cost of copying them being defrayed by a grant from the Govel.nu ) GrtlIit Commiltee of the Roynl Society . .
For the southern pair of stations there is same kind of \amp ; fference between the summer and vinter results ; this scems to indithat the phenomenon is really connected with the season , and does depend on inliuences external to the Earth , since the summer for the first pair is the for the second .
Ioreover , it is found for $he two southern stations the change of pressure passes from north-west south-east , so that in predicting we ought to St. Helena earlier than Cape Town .
The tables give the tions obtained :\mdash ; Table I.\mdash ; Correlation of rometric Heights at Wilmington .
Table rela ion of St. Helena .
itive v hen reading is taken : hen is taken later at Capc these maxima .
After some consideration , the curve was selected as likely to give a reasonable interpolation resul axis of being that of the correlation , and being the interval , days , between the readings , counted as positive when the obssrvation tions wating eurve , those given imagnitude taximum ordinates.eason oahare nntended tepresent tation daken 1ater atation ionvenient emethods tosition a to 9 of Table I , and rows 4 to 8 of Table II .
The ving curvee , were determined:\mdash ; WILMINGTON AND HALIFAX .
Maximum ordinate : at , i.e. , hours .
: \mdash ; lmum ordinate : at , i.e. , hours .
Maximum ordinate : at , i.e. , hours .
Wint , 1889\mdash ; 1898 :\mdash ; Maximum ordinate : at , i.e. , hours .
: \mdash ; Maximum ordinate : at , i.e. , hours .
Winter , 1893\mdash ; 1898:\mdash ; Maximum ordinate : at , i.e. , hours .
uggcstion i Time Factor on Correlation between Barometric Heights .
407 therefore appears that the maximum correlation between \amp ; alifax and Wilmington in the summer occurs when the barometer at is read about 16 hours later than that at Wilmington , and that corresponding interval for the winter is about 23 hours .
The )intervals for St. Helena and Cape Town are about 11 hours and hours respectively .
Some caution may be necessary in accepting he values obtained for the summer intervals , since it is possible that diurnal variation may affect the correlation ; this point could only settled by means of observations taken at other hours of the day .
iThis effect of the diurnal variation , if it really exists , may conceivably account for some part of the difference between the intervals obtained the summer months of the two decades for the American stations .
Nevertheless , it seems reasonable to conelude that , in order to predict the barometric height at Halifax at 9 , it is desirable to use the reading taken at lVilmington at 10 .
on the previous morning in winter , while in the summer the yilmington reading should be taken at about 5 Diagram I ( next page ) shows the interpolation curves used .
5 .
The physical interpretation of these intervals for which the correlation is a maximum , is not ) vious to me , although it may possibly be so to practical meteorologi.sts .
Of course , the intervals , 11 to 24 hours , are too short to be looked upon as corresponding to the transfer of an actual disturbance between the stations .
Even if this were not the case , they ] not give the average intervals between the arrival of the same isobar at the two places considered , since none of the pressures correlatea may be alike .
I should be glad to see the point adequately treated by one with practical meteorological experience , and .
will merely remark here that the correlation depends not upon equality of pressure , but upon proportionality of deviations from the local means .
A full explanation must also account for the fact that the wi11ter and summer relations differ in a very marked way .
6 .
Another point which seemed to deserve investigation was the correlation between the daily rise or fall of the barometer at Halifax and Wilmington .
This investigation would have involved considerable labour if it had been necessary to proceed by the direct method of preparing tables of the daily rise or fall , and then calculating the correlations from them .
The work has been considerably lessened by the use of the following method , hich is due to Professor Pearson .
Let be the reading on one day and that on the following day at the first station , and the corresponding readings at the second station after any interval .
Let be the means , the standard deviations , the correlation of and , that of and , that of and , and so on , and that of and The mean values of and will VOL. LXXIV .
408 Miss F. R Cave-Browne-Cave .
of .
the DIAGRAM I. clearly be and respectively .
Hence , if be number of pairs correlated , and if denotes summation , we have 1904 .
] on Correlation between Barometric Heights .
409 It may be noted that this formula would apply to any case in which was desired to correlate the difference of one pair of quantities with the difference of another pair .
If we were not dealing with the summer and winter months separately , we should have , with sufficient approxination , since the quantities included in the and the groups would be the same , with the exception of a single reading at each end of the period 'considered .
The formula would then take the simple form ( ii ) .
But in our case the and groups differ by a reading at each equinox of each year considered , and it was therefore decided to use the accurate equation ( i ) , although ( ii ) would have been sufficient to give a good idea of the correlation .
In order to use this formula , it was necessary to calculate the correlations between the barometric heights on consecutive days at Halifax and at Wilmington .
Table III gives these correlations , and Table gives the correlations between the daily rise at the two stations for various intervals .
Table III.\mdash ; Correlation of Barometric Heights on Consecutive Days .
The most noticeable fact which appears from resuIts is the analogy between them and the space relations considered in the previous note .
In that case we found that with increasing dista1lce the correlation diminished to a negative minimum , and then began to increase ; and the negativs correlations obtained vere of considerable magnitude .
This does not seem to occur for the time relations when the actual readings are considered ; for , although the values given in Tables I and II seem to indicate that a minimum has been reached , the only negative correIations which have been found are so small as to be inter of very little signifioance .
But it does occur in a very marked way for the time relations when we correlate the daily rise at the stations .
This analogy strengthens our belief that the intervals of maximum correlation are of great importance from the standpoint of atmospherical physics , and our desire to ses the practical meteorologist 's explanation of the matter is correspondingly increased .
7 .
The magnitude of the correlations which have been found to exist between the readings on consecutive days at Wilmington and Halifax , suggests the possibility of predicting barometric heights at the latter station .
with a fair degree of accuracy from the heights observed at Wilmington on the previous day .
In the paper by Professor Pearson and Dr. Lee already referred to , a table is given showing the barometric heights at Stonyhurst on 50 days , calculated from the simultaneous readings Southampton and Laudale .
The difference between the calculated and observed heights at Stonyhurst was small in every case , the mean being about .
At that time no correlations had been calculated between observations taken on different days ; but the values now obtained make it possible to apply the theory to actual prediction .
In predicting for Halifax from a single station more than 1000 miles away , we are naturally unable to obtain such good results , but , on the other hand , we have the advantage of calculating the probable heights at Halifax a day beforehand .
In order to see how far the prediction is verified , forty dates were taken at random , one in summer , and one in winter for each year .
It was impossible to allow the best intervals , 16 hours in summer , and 23 in winter , as we do not possess readings for ilmington taken at the required times , and the predictions have , therefore , been made with the interval of 26 hours for the first 10 years , and 25 for the second .
the readings are taken later at Halifax .
1904 .
] Time Factor on Barometric Heights .
Table gives the means , standard deviations , and regression coefficients which were employed in the caIculations .
Table for Halifax and Wilmington .
The corresponding prediction equations are as follows , if denote the probable height at Halifax corresponding to a height observed at Wilmington on the previous day .
Summer 1879\mdash ; 1888 Summer 1889\mdash ; 1898 YVinter 1879\mdash ; 1888 Winter 1889\mdash ; 1898 These results bring out clearly the marked difference betweell summer and vinter , which has already been shown in the case of the correlations .
But in both cases there is also a difference between the two decades reduced ; this is partly due , no doubt , to the fact that the Wilmington observations were taken at different hours in the two decades , but this would not account for the whole difference , which has been found also in dealing with other stations .
It may be due to variations corresponding to those of random sampling ; or it may indicate a gradual change , whether periodic or progressive , in the physical constants involved ; and this question can only be settled by dealing with observations extending over a longer period .
The results of the predictions are shown in Table VI .
The theoretical mean error for and for winter , taking the mean summer and mean inter values of and for the two decades .
These theoretical errors are slightly larger than the actual mean errors for the forty random dates here considered .
should presumably improve the predictions by taking the interval for which the correIation is a maximum ; but even without this improvement the degree of accuracy attained , though not very great , might be 412 Miss F. E. Cave-Browne-Cave .
Influence of the Table .\mdash ; Comparison of Predicted and Observed Heights at Halifax .
A. Summer B. Wiinier Nonths .
sufficient to be of service in practical meteorology .
Still better resulta could probably be obtained by making the prediction for one station depend upon observations at two or more other stations , so situated as to give suitable intervals for maximum correlation ; and it may also be desirable to take into account the height observed on the day at the station for which the prediction is required , for the results given in Table III show that in some cases , at any rate , there is a * It would be of little use to predict for intervals of or ours ; at hours would be desirable , if this does not make it necessary to take great as seriously to weaken the correlation .
Time Correlation betnoeen Barornetric Heights .
413 iderable correlation between the barometric readings on successive ilays at the same station .
Moreover , it is possible that the correlation between the daily rise or fall at different stations might give better esults for prediction than the correlation between the actual heights bserved ; but what has already been done is sufficient to indicate the manner in which modern statistical methods may aid the meteorologist : this part of his work .
As regards the more fundamental , though immediately useful , task of meteorological science , the intervals of maximum correlation , a few examples of which have been considered .
paper , will , I believe , be found ultimately to be of importance in relation to the physics of the atmosphere .
Although considerable : labour is involved in the calculation of these intervals , even for a pair of stations , yet the increase of theoretical knowledge which vould result from an adequate investigation of the manner in they depend upon local conditions , and upon the positions and .
distances of the places considered , ] probably be amply sufficient to justify the labour expended upon the matter .
8 .
The conclusions to be drawn from the results given in this paper are as follows :\mdash ; ( i ) The correlation between the barometric readings at two stations upwards of 1000 miles apart depends upon the interval between the readings .
In the case of Halifax and Wilmington , the correlation is sensible for at least 9 days , and it reaches a maximum for an interval of about 16 hours in summer and 23 in winter .
For these stations , and also for St. Helena and Cape Town , the observation at the more easterly station should be taken later for maximum ( ii ) There is a considerable correlation between the daily rise at Halifax and WilmingtoIl , and this correlation changes with the interval in a manner somewhat analogous to that in which the correlation between simultaneous heights at two stations approximately on the same meridian depends upon the distance between them .
( iii ) There are considerable diffe.rences between the summer and winter correlations , and these differenoes are of the same general nature for both pairs of stations considered .
( iv ) It is possible to predict the ) arometric height at one station from an earlier height at a second tion more than 1000 miles away , with a fair degree of accuracy , the mean observed error for forty dates , taken at random , for Halifax and gton , being

