f
e
1 :
c
.a E
. .
-.2 I .6 1.4 1 .2 . .
1840 1060 1880 19W 1920 1940 1960
FIG.
2. Trends of mean temperature in Northern and Southern Hemispheres, by pentads, within indicated latitude limits.Annual average data above and winter-season data below. All curves are area-weighted averages of 100-latitude band data in Table 5.
to the early
1940s,
and a marked tendency for cooling since the early1940s.
Average temperaturesin
recent years have apparently returned to about the levels of the1920s.
Turning next to a comparison of the zonally integrated trends for the Northern and Southern Hemispheres, shown in
Fig. 2,
w efind
that the trends have beenFIG. 3. Trend of annual mean temperature in tropics (30°
N.-
300S.)
by pentads, based on numbers of stations indicated.162
qualitatively similar
in
both. Inasmuch as the mag&tudes of temperature changes are
known
to be a strong function of latitude (at least in the Northern Hemis- phere), the trend for the Northern Hemisphere has been shown in Fig.2
for00-600 N.
as well as for00-800 N.,
to compare with the trend for
Oo-600 S.
(Climatological data for higher southern latitudes are virtually lacking.) Inclusion of data for600-800 N. in
the Northern Hemis- pheric averages are seenin
Fig.2
(dotted curves) to increase the magnitude of the net secular trend of that hemisphereby
as m u c h as half although the added data represent only a one-seventh increase of surface area.Of
special interest for this symposium, perhaps, is the fact that the tropics have sharedin
the secular warming of the past century, as well asin
a tendency for cooling since the1940s.
Fig.3
shows the pentadal variation of annual mean temperature since1880 in
the zone between300N.
and300s.
The net warmingin
this zone had reached nearlylo F.
from about1880
to1940,
and the cooling since about1940
has thus far totalled about-0.30F.
Callendar
(1961)
has also published zonally integrated values of m e a n annual temperature changein
different increments of latitude. For the six decades ending with1950,
when Callendar's data terminated, these changes have been found to agree closely with the changes derived frommy
o w n and Willett's data (see Landsberg and Mitchell,1961).
This is reassuring in view of dissimilar data selection and analysis procedures in the two sludies. In terms of Merence between consecutive 30-year periods, this comparison is shown in Table 1.lREPRESENTATIVENESS OF ZONALLY INTEGRATED TRENDS
Fig. 4
indicates the geographical locations of all stations that contributed at least20
years of data to the trends illustratedin
the previous figures. The most casual glance at Fig.4
serves to remind us thatin
speaking of average trends over large parts of the world w e are really speaking primarily of trends over the continents and a few relatively well populated archipelagoa. Large expanses of ocean, especiallyin
the Southern Hemis- phere, are completely devoid of historical data (areas outlinedin Fig. 4). With
this fact in mind, let us next consider the representativeness of the available data from t w o different points of view. First, to what extent are the temperature trends shown in Figs.1-3
repre- sentative of the principal areas of civilization-i.e., primarily the continental land masses and archipe- lagos-for which data are comparatively abundant?
Second, to what extent are the calculated trends repre- sentative of the earth as a whole, and therefore indicative 1. Inasmuch aa C d e n d a r did not treat global data available since 1950, it iß impossible to compare notes with him concerning evidence of the recent culmination of thr warming.
World-wide pattern of secular temperature change
T A B L E
1. Thirty-year change of annual mean temperature(OF.),
1890-1920 to 1920-501 -After Callendor After Willett and Mitchell
Zone
Inclusive latitudes A Inclusive latitudes A
World 60° N.-50”
S.
+0.41 600 N.-500S.
$0.37North temperate Tropical South temperate
600-250
N.
250 N.-250
S.
250-500
S.
+0.70
$0.31 t0.25
600-30°
N.
60°-200N.
300 N.-300 S.
1
200 N.-200S.
1
300-50oS. 200-50’ S.1. Callendar’s data are 1891-1920 to 1921-50; Willett and Mitchell’s data are 1890-1919 to 1920-49.
$0.64 +O37 +0.35 +0.39 + O J O +O.OS
of variations of the net planetary atmospheric heat budget
?
T R E N D S AS R E P R E S E N T A T I V E O F T H E D A T A A R E A S
( R E G I O N S O F A B U N D A N T D A T A )
Let us postulate that the stations whose data have been used in this trend study are randomly distributed over the region of available data outlined
in
Fig.4
(hereafter denoted for simplicity asthe
“data areas”, which roughly coincidewith
the continents).By
the methods givenin my
research paper (Etchell,1961~)
w e can then obtain an estimate ofthe
standard error of the zonally averaged trends as representative of the data areas.An
outline of these methods and some statistics necessary for their application here are givenin
a n appendix at the end of the paper.Trends of mean annual and winter temperature since
1890
are summarized alongwith
their associated stan- dard errors for various latitude zones of the worldin
Tables2
and3. The trends
are shown as 30-year changes from1890-1919
to1920-49
(the same intervals asin
Table1) in
Table2
and as 10-year changes from1940-49
to1950-59 in
Table3.
T h e ñrst period coincideswith
the great “world-wide” warming, andthe
second coincides with the cooling evidentin
Figs1-3
which followed. It is clear from these tables that boththe
warming and subsequent cooling were shared by (the data areas of) the Northern and Southern Hemispheres and the tropics.T h e statistical significance of the warming w a s moder- ately
high in the
Southern Hemisphere (trend equal to2-3
standard errors) and extremelyhigh
elsewhere as well as for the world as a whole (trend equal to4-7
standard errors).
T h e significance of the cooling has also been moder- ately high, but has varied substantially
with
latitude and season. (The trend for winterin
the Northern Hemis- phere was comparable to its standard error; all othertrends shown in Table
3
are about two or three timestheir
standard errors.) T h e recent coolingin
the tropics, as well as the annual average cooling for the world as a whole, has beenhighly
significant (exceeding about3
standard errors, corresponding to significance levels of99
per cent and higher).T R E N D S A S R E P R E S E N T A T I V E O F T H E W O R L D AS A W H O L E
In view of the sizeable oceanic regions of the world totally devoid of data, it is only
with
certain reservations that the trends noted above can be taken as representa- tive of planetary average conditions. Until such time as these data gaps have been ñlled, eitherby
conven- tional ocean weather stations,by
automatic weather buoys, or possiblyby
indirect sensing techniques involving artificial earth satellites, secular fluctuations of temperaturesin
these areas can never be determinedwith
m u c h c0nfidence.lIn
order to evaluate the available trend data as a measure of planetary average conditions, w ewill
adopt approaches described below.In the ñrst approach, w e assume that the u n k n o w n trends averaged for the no-data areas are identically zero. That is to say, the no-data areas (open oceans) are presumed not to have participated to an appreciable extent
in
any of the net temperature changes measured elsewhere (the data regions). Inasmuch as there is little justification-either empirical or theoretical-for sup- posing that average trends over the oceans are normally opposed in sign to those over the continents, this assumption provides the basis for a conservative test of the statistical significance of the observed trends as1. Climatological analyses of synoptic ship reports of air and sen temper- ature have been m a d e which have Borne value in this regard. Except for certain well-travelled shipping lanes in the northern oceans, however, irrebdar timing and spacing of the observation8 and various sources of inhomogeneity greatly complicate the interpretation of long-term trends evident in 8uch data.
163
E.
_.--
..e-.- ....--.
___
164
wor&wide pattern of secutar:temperature change
TABLE
2. Thirty-year temperature change (OF.), 1890-1919 to 1920-49,A,
selected latitude belts; standard error of estimateSE(A)
for data areas; significance ratioIA]
toSE(A)
applicable to areas of good data coverage (t) and whole earth (to), assuming zero net temperature change in areas of no dataAnnual Winter
Latitude zone
r 1
A f SE(A) x' A 1 S E M tx
World
800 N.-600
S.
1-0.49 f0.07 6.7 4.3 3-0.71 10.11 6.2 4.5600 N.-600
S.
1-0.39 rfr0.07 5.9 3.4 +0.44 fO.10 4.6 3.0800
N.-Oo
+0.64 fO.09 6.8 5.1 +1.14 10.18 6.5 5.0600
N.-Oo
+0.43 f0.07 6.3 4.4 1-0.61 10.13 4.9 3.4300 N.-300
S.
+0.35 kO.09 3.8 2.0 1-0.49 rfr0.12 4.0 2.200-600
S.
+0.34 f0.11 3.0 1.3 +0.28 10.15 1.9 1.0Northen Hemisphere Tropics1
Southern Hemisphere
1. Winter value for tropics is average in Jnnuary, February, and December north of Equntor and in June, July, nnd August south of Equator.
TABLE
3. Ten-year temperature change (OF.) 1940-49 to 1950-59;
standard error of estimate for data areas;
and significance ratios for data areas for whole earth (see heading to Table 2)Annual Winter
Latitude zone
t 1
A
f
S E W tZ A f W A ) rgWorld
800 N.-600
S.
-0.19 f0.06 3.3 1.8 -0.23 hO.09 2.4 1.1600 N.-60O
S.
-0.17 f0.05 3.3 1.7 -0.17 f0.08 2.1 0.6800
N.-Oo
-0.26 kO.09 2.9 1.7 -0.22 10.17 1.3 0.7600 N.-OO -0.23 h0.08 3.0 1.6 -0.10 10.14 0.8 0.1
300 N.-300
S.
-0.24 f0.06 3.6 1.9 -0.26 hO.09 2.9 1.500-600
S.
-0.11 f0.07 1.6 0.8 -0.23 fO.09 2.6 0.9Northern Hemisphere Tropics1
Southern Hemisphere
1. Winter vnlue for tropics is average in January, Febrnav, and December north of Equator and in June, Suly, and August south of Equntor.
representative of planetary conditions. Accordingly, let the change of planetary m e a n temperature
in
a given latitudinal band i, between any two time intervals, be givenby
(1)
-
8. 1- - c. Jr
.+
(1-
Ci) Fi,where
&
is the average trend for the data areas,8'i
is that for the no-data areas (here assumed identically equal to zero), and Ci is the percentage total area of the band at least1,000
miles distant from any historical- record station within the z0ne.lIn
this w a y w e m a y calculate globally representative trends corresponding to the data-area trendsin
Tables2
and3 by
the formulaA, =
ZCiwiSi, (2)where
the
zui are proportional to the areas of the100
bands involved(Cwi = 1).
These trends, in turn, m a y-
be expressed as ratios to the same standard errors as before,
by
which w e prescribe that the error of estima- tion of the (zero) trendsin
the no-data areas ishypo-
thetically equal to that of the changes in the data areas.
T h e resulting ratios (i.e., values of tg, also shown
in
Tables2
and3)
are suitable for estimating the statis- tical significance of the trends from the planetary point of view. Itw i l l
be seen in Table2
that the observed warming trends between1890
and19-49,
with the exceptionof
thosein
the Southern Hemisphere, remainhighly
significant(2-5
standard errors). According to Table3,
however, the cooling since1940
loses much of its significance w h e n viewedin
this manner.In
a second approach to this problem, let us consider 1. This fiyre represents the shortest geographical distance nt which thespatial eorrelntion coefficient of mean iemperature changes typically diminishes to zero. It is found from cmpiricd evidence that this distance is nearly independent of latitude, at least in the Northern Hemisphere.
Values of C for each 100-latitude band are s h o w n in Tnble 7.
165
Changes of climate Les changements
de
climatthe
€allowing
question: W h a t range of hypothetical average temperature changein
the no-data areas (i.e., what values ofSr,)
can apply such that, w h e n these are averagedwith
the measured trends for the data areas, the resulting trends are not significantly different from zero?If
such values of8'z,
denoted asArc,
can be declared unreasonable, then w e are justified in accepting the measured trends as evidence not only of redistri- butions of heatwithin
the atmosphere, but also of changes of the heat budget of the whole planetary atmosphere.For this purpose, w e m a y set the global trend equal to its expected value
Ãc = Ão +
[Z(l- Ci)wi]Arc, (3) withÁ o
givenby
(2), and define to such that to>
tg= Ã/SE(Ã). If
for convenience the sign ofho
is always taken as positive, and that of
Aro
adjusted to agree, w e arethen led
to the inequality(4)
For to
=2.58
(corresponding to the99
per cent significance level), and for the stipulation thatSE@) = SE(A),
ranges ofA',
are shown in Table4
for the s a m e periods of record and latitude zones asin
Tables2
and3. With
the help of the maps of secular change to be presented in the next section, these values for the no-data areas have been tentatively evaluated as definitely reasonable, marginally reasonable, or unrea- sonable. Although such an evaluation must at this stage be taken as provisional, it has been extremelyto
SE@) - Ão
"c< ~ ( 1 -
cihi
difñcult