1902 .
] Correlation between Barometric Heights .
\lt ; \#171 ; On the Correlation between the Barometric Height at Stations on the Eastern Side of the Atlantic .
" By Miss E. E. Cave-Browne-Cave , Research Student of Girton College , Cambridge , with some assistance from Karl Pearson , F.R.S. , University College , London .
Received June 3 , \#151 ; Read June 19 , 1902 .
( 1 ) In a memoir on the correlation and variation of the barometric height at divers stations in the British Isles* by Professor Karl Pearson and Dr. Alice Lee , it is suggested ( i ) that interesting results might be obtained by correlating the barometer at stations on the east and west sides of the Atlantic , allowing an interval of time between the observations ( see p. 459 ) , and ( ii ) that with a certain distance between stations , the correlation !
; would be found to be negative , i.e. , a high barometer at the one station corresponding to a low barometer at the second ( see p. 467 ) .
, " Q(V_ ( 2 .
) In order to deal with these points , steps were taken m 1897 to collect the necessary material .
Twenty years , 1879 1898 inclusive , were selected for consideration , and the early morning barometric observations for these years , copied from material provided by the kindness of the British Meteorological Office for the following East Atlantic stations B0d0 , Flore , Skudesnaes , Valencia , Lisbon , and Funchal .
These give a very fair chain of stations from the north of Norway to Madeira .
On the west side of the Atlantic we obtained data for the same years for Halifax and Toronto by aid of the Director of the Canadian Meteorological Service .
So far as we can judge the Canadian returns .appear satisfactory and satisfactorily copied , and we have heartily to thank the Director for nominating an efficient and careful copyist .
An appeal to the American Weather Bureau in Washington , and later to the Smithsonian Institute , led the latter institution to kindly offer a copy of the American data for New York , Wilmington , and Key West , free of expense to us .
We regret , however , that the copy thus procured has been discovered after some work upon it to be unreliable .
We still trust , however , that we may be able to proceed to work on accurate copies of the American .observations .
( 3 .
) Meanwhile a preliminary study has been made of the East * \#163 ; Phil. Trans. , ' A , vol. 190 ( 1897 ) , pp. 423\#151 ; 469 .
t The first application of the mathematical theory of correlation to meteorology occurs in the paper just cited .
Samples of barometric correlation tables aie there printed , and it is shown how the mean height of the barometer at Station for a given height at Station B may be expressed in terms of standard deviations and correlation coefficients .
An elementary account of correlation will be found in the * Journal of the Boy .
Statis .
Soc. , ' vol. 60 , pp. 1 44 , by Gh TJ .
Yule .
466 Miss Cave-Browne-Cave and Prof. K. Pearson .
[ June 3 , Atlantic stations , and this has impressed us with the desirability of continuing , if possible , our chain of stations right down the west coast of Africa , even to the Cape .
The great mass of material to be dealt with , and the many new problems which arise in an almost entirely novel investigation of this kind , have meant of course very slow progress , and while publication of the final conclusions must be delayed for some time yet , it seems desirable to draw attention to some of the results already reached for the East Atlantic stations .
( 4 .
) In the first place it was soon discovered that the winter and summer months ( equinox to equinox ) must be treated separately .
It was already known that the average height varied considerably in the summer and winter months , but there are also very significant differences in the variability , and , in what we are most concerned with , the correlation .
For example , there is hardly any correlation ( 0'04 ) between Lisbon and Valencia in the summer , but in the winter it is quite considerable ( 0*22 ) .
Further , the results worked out in two groups of 10 years each , show that very sensible differences in mean , variation , and correlation can exist between one decade and the next , so that at least 20 and probably more years are desirable if we are to obtain steady values for the barometric constants .
In the next place while we have found a small but sensible cross Atlantic barometric correlation after a definite interval of time , we must wait for more complete American data , and for still closer investigation of the best interval for different stations before results on this point are published .
The second suggestion , however , has been amply verified , and to draw attention to this is the principal object of the present preliminary notice .
As we go generally south from any station , we reach a point at which for readings on the same day there is no correlation at all .
For stations beyond this 'point the correlation becomes negative , reaches a negative maximum , and then begins to decrease .
Clearly it must reach a second zero .
What happens after this ?
Does the correlation remain zero for all greater distances ?
To fully answer this problem we must obtain data south of Sierra Leone\#151 ; in fact , we want data for St. Helena 's , Ascension , and the Cape , and have taken steps to obtain them .
But our confident belief is that the negative zone will be found to be followed by another zone of positive correlation , but of far less intensity .
In fact , starting with any northern station and going south across the equator , we expect to reach alternate zones of positive and negative correlation , each zone marked by a successively smaller maximum .
These zones are not peculiar to the earth , but rather to the station from which we start .
Thus Valencia is positively correlated with Bed# .
Lisbon , however , is negatively correlated with Bede , but positively with Valencia .
We require to go as far south as Funchal to find a negative correlation 1902 .
] Correlation between Barometric Heights .
467 with Valencia .
To get a negative correlation with Lisbon we must go as far as Sierra Leon , which has become positive again for both B0d0 and Valencia .
At St. Helena 's we have our second negative correlation zone for both Bodfr and Valencia , while we are only in the second positive zone for Lisbon .
In other words , the curve of barometric correlation with distance from a station appears to give roughly the form:\#151 ; We do not find with increasing distance a diminishing correlation , as of a curve rapidly asymptoting to ox , but as it were a wave-curve of diminishing amplitude .
There is not apparently an area of positive correlation surrounded by a field of zero correlation , but going south there are only points of zero correlation , not of zero correlation .
Probably if the area of investigation can be extended we shall find lines not zones of zero correlation round each station , separating districts of positive and negative correlation .
Now the results which we have copied for Sierra Leon and St. Helena 's are not at present complete enough to enable us to be absolutely definite with regard to them .
In particular , Skudesnaes and Sierra Leon appear to give a discordant result .
But our object in the present note is merely a preliminary one .
We wish to call attention to a rather remarkable relation between the barometric height of stations at increasing distance , and by so doing emphasise the importance of complete returns for stations in the southern hemisphere .
( 5 .
) The following will illustrate our point:\#151 ; Table I. Station .
Bodo . ! !
Skudesnaes .
Valencia .
Lisbon .
Remarks .
Bodo i 1 0 -6905 0 -0940 -0 -2815 20 years .
J Skudesnaes 0 -6905 1 1 I 0 -4631 -0-1606 99 Valencia . .
0 -0940 1 0 *4631 1 0 -1547 99 Lisbon .
, . .
-0-2815 -0-1606 | 0 -1547 1 19 i Funchal * .
-0-1881 -0*2305 - 0 *0723 0 -7429 99 Sierra Leon + 0-0079 [ -0-0505 ?
] ] +0*0188 -0 0495 10 years* St. Helena .
| -0-0841 i -0-0231 i -0*0224 1 + 0 0135 6 years ( winter only ) .
The results for the first five stations are based on far more reliable material than those for the last two .
But the latter suffice to suggest 468 Miss Cave-Browne-Cave and Prof. K. Pearson .
[ June 3 , such a conclusion as that the barometer in the north of Norway may be as closely related to a station\#151 ; St. Helena in the southern hemisphere\#151 ; as to Valencia in its own ; the correlation for the months of Valencia and Bod0 is 00590 , while its correlation with St. Helena for the winter months is - 0-0841 .
Valencia , far nearer to St. Helena , is less highly correlated with it than B0do is , while Bpd0 itself is nearer related to St. Helena than to Sierra Leone .
The relationship here is not , be it remembered , a relationship of average heights , but of the daily fluctuations of the barometer at the two stations\#151 ; i.e. , it would be safer to predict the barometric height in the winter at B0d0 on any day from the height on the same day at St. Helena , than to predict it from the height at Valencia .
Results of this kind seem of sufficient interest to deserve notice , even before the whole material has been collected and reduced .
What we are certain about is , that a zone of positive correlation is followed by a zone of negative correlation .
What we are less sure about is , that this negative zone is again followed by a positive zone of much less intensity , but our rather meagre results certainly suggest it .
( 6 .
) The accompanying tables give the more complete results for the five more northerly stations .
Table II.\#151 ; Height and Variation of Barometer at East Atlantic Stations .
Means in mm. Period .
Bodo .
Skudesnaes .
V alencia .
Lisbon .
Funchal .
1879-88 .
Summer " Winter ... * . .
" Whole year .
, 757-006 753 -315 755 -194 758 324 757 *866 758-099 760 -285 760 013 760-151 763 -328 765 -516 764 -402 764-558 765 -030 761 -790 1889-1898 .
Summer. . .
, , Winter ... .
" Whole year 758-174 754 -171 756 -210 759 -008 757 -927 758-478 761 -160 760-278 760 -727 763- 199 764- 895 764 -031 764 -721 764-929 764-823 j 1879-1898 .
Summer. . .
" Winter ... , , Whole year .
757 -590 753 -743 , 755*702 758 -666 757-896 758 -288 760 -722 760145 760-439 763-264 765 *206 764 -217 764-639 1 764-979 764-806 For brevity , the probable errors are not at present given , but the differences between the two decades are as a rule sensibly larger than the probable errors .
Several interesting results flow from this table .
Thus , north of Valencia , the summer barometer is higher than the winter ; south of Valencia , the winter higher than the summer .
In al cases the winter height is more variable than the summer , but the 1902 .
] Correlation between Barometric Heights .
460 Standard Deviations in mm. Period .
Bode .
] Skudesnaes .
Yalencia .
j Lisbon .
Funchal .
| 1 1879-88 .
Summer M Winter. . . . . .
" Whole year. .
7*391 11 056 9*551 7*573 11*299 9 -588 7 *746 10 *508 9 -207 1 3 -812 6 *079 5 171 3 064 1 4 -781 !
4 -007 1 1889-1898 .
Summer ... } ) Winter. . .
" Whole year 7*837 10-999 9 *729 7*501 11*146 9-482 7- 227 10 *228 8- 839 3 -393 6*253 5*076 2 -964 5 *053 4 -125 1879-1898 .
Summer ... " Winter " Whole year 7 *640 11036 9*653 | 7 -545 | 11 -223 9 -537 7 -504 10-370 y *030 3-609 6*175 5*127 3 015 | 4-919 i 4-066 winter variability tends to approach the summer variability the further south we go* ; the absolute variability decreasing , however , as we go south .
Table III.\#151 ; Correlation of the Barometric Heights at the East Atlantic Stations .
( See attached sheet .
) Now it will be seen that from these stations , for which our results are very reliable , that _ ( i. ) The correlation between certain stations is sensibly different for summer and winter .
( ii .
) The correlation varies somewhat for the same pair of stations from decade to decade .
, ( iii .
) In every case beyond a certain distance from a station the-correlation becomes negative , reaches a negative maximum , and then begins to decline . .
Now we refrain from either plotting curves for these changes in correlation or from drawing inferences as to the nature of the instantaneous distribution of barometric height along a meridian until wo have been able to deal with more complete data from equatorial stations , and from stations south of the equator .
But we believe this alternation of instantaneous high and low barometer !
along a meridian , and the discovery of uncorrelated stations , are of sufficient physical interest to deserve immediate notice .
We propose to call a station , uncorrelated with a given station , a station nodal to it .
Thus to every station will exist a first nodal line of stations , the barometric height at which is uncorrelated with that at the given station ' .
Within this * There is an apparent exception to this in Bodo and Skudesnaes .
+ " High " and " low " barometer at any station mean here above and below the* mean for that station .
470 Correlation between Barometric .
first nodal line there will be positive correlation , beyond it negative correlation , until we reach the second nodal line ; beyond this the correlation may become again positive , and so on .
All nodal lines appear to be relative to a given station , and are not to be considered as curves peculiar to the earth 's surface .
One important result of this system of places nodal to a given station is , that it probably depends on the interval of time between the- observations at the two stations .
Accordingly , in endeavouring to predict the height of the barometer on the east side of the Atlantic from an antecedent series of heights on the west side , we might easily fail to get satisfactory results if the stations on the west side were approximately nodal to one or more stations on the east side .
In this case we might do better to take our stations further to the west , or nearer the maxima of the next internodal zone .
We hope shortly to complete our calculations to the Cape , and then to finish the work already begun on the American stations .
Meanwhile , we think that the correlation of a series of stations following roughly a parallel of latitude across Europe and Russian Asia would throw a flood of light on whether a chain of roughly north and south stations differs wholly in character from a chain of east and west stations .
The magnitude of the computations , however , almost precludes the idea that any individual worker or workers can hope to complete such a task within a reasonable period .
Place .
Period .
Lisbon .
Funchal .
Winter .
Whole year .
Summer .
Winter .
Whole year .
Bed 1879\#151 ; 88 1889\#151 ; 98 1879\#151 ; 98 -0-2919 -0-3299 -0-3124 -0-2697 -0-2906 -0-2815 -0-0349 -0-0346 -0-0326 -0-2291 -0-2707 -0-2505 -0-1772 -0-1993 -0-1881 Skudesnaes 1879\#151 ; 88 1889\#151 ; 98 1879\#151 ; 98 -0-1430 -0-1681 -0-1556 -0-1513 -0-1691 -0-1606 -0-1461 -0-1527 -0-1479 -0-2787 -0-2525 -0-2642 -0-2387 -0-2227 -0-2305 Yalencia. .
1879\#151 ; 88 + 0-1862 1889\#151 ; 98+ o -2514 1879\#151 ; 98 + 0-2178 + 0-1301 + 0 1837 + 0-1547 -0-0926 -0-0630 !
-0-0768 -0-1315 -0-0077 -0 -0688 -0-1191 -0 -0255 -0-0723 Lisbon . .
1879\#151 ; 81 ] \#151 ; \#151 ; ; +0-7792 + 0-7213 + 0-7194 + 0-7768 + 0-7315 + 0-7559

