738 
applied to daily forecasting, and to extended or long- 
range forecasting based on mean maps. 
Analogue forecasting has not proved to be notably 
successful. It is better suited to daily than to extended 
forecasting, but even in the daily range the extensive 
effort in the United States to supply analogue informa- 
tion to all forecasters during the war years does not 
appear to have won many supporters for their use. For 
the more extended forecast ranges the method suffers 
the inevitable disadvantage of any routine technique of 
synoptic extrapolation—that of complete inelasticity 
in meeting rapidly and erratically fluctuating conditions. 
Essentially the method fails at two points: 
(1) Our knowledge of the mechanics of the general 
circulation is completely madequate to the task of 
establishing any analogue classification of the large- 
scale circulation patterns which is of basic prognostic 
significance. In the absence of such knowledge the 
establishment of sufficient analogy between two synop- 
tic weather patterns to justify the use of the subsequent 
weather sequence in one case to forecast that of the 
second case in detail requires a basic similarity not only 
of the large-scale sea-level synoptic patterns, but also 
of the upper-level patterns and of the preceding se- 
quence of development by which the analogous patterns 
come into existence. Needless to say such strict specifi- 
cations can only rarely be satisfied in the selection of 
an analogue. 
(2) In the absence of a prognostically determinative 
analogue classification, the selection of an analogue is of 
little or no assistance to the experienced forecaster. As 
with the use of statistical aids in short-range forecasting, 
the experienced forecaster will usually do better by 
judging the synoptic indications of each individual case 
in the light of his past experience. In extended or long- 
range forecasting, the case for analogues is even weaker. 
In other words, the use of analogues has even less to 
offer towards any real improvement of weather fore- 
casting in the present state of our physical knowledge 
of the problem than do the other techniques of synoptic 
extrapolation. 
c. Mathematical techniques of extrapolation—based 
on various manipulations of the equations of motion 
and continuity. Accurate weather forecasting by mathe- 
matical computation is an ultimate objective for the 
attamment of which nearly every meteorologist hopes, 
but as a practical reality it appears today to be quite as 
distant as when Richardson [8] made his classical con- 
tribution to the problem in 1922. Richardson failed 
completely to derive, from the theoretical equations, 
satisfactory forecasts even of the short-range (6-hr) 
changes of the meteorological elements. This failure was 
doubtless caused in part by his efforts to deal with all 
of the variables at once, which complicated his calcula- 
tions to a point where he was unable to identify the 
sources of his errors, but also by the further fact, since 
proved by Charney [4], that his unit time interval (6 
hr) was far too large, relative to his space grid, to permit 
a reasonable (convergent) solution of the equations. 
Interest in the numerical prediction of weather has 
been greatly stimulated by the recent development of 
WEATHER FORECASTING 
high-speed computing devices, which places the feasi- 
bility of lengthy numerical reckoning on an entirely 
new basis. In spite of initial optimism, however, as to 
the potentialities of these computing techniques, it is 
generally recognized at present by those who have been 
working on these methods that no radical advance in 
practical weather forecasting in the near future is prob- 
able. One difficulty lies in the great mass of observa- 
tional data that must be treated in order to provide in 
time and space a sufficiently extensive and dense net- 
work of observations to compute the future state of the 
atmosphere a day or more in advance. However, that is 
a technical problem that doubtless can be overcome. A 
second difficulty lies in the magnitude of random local 
variations of the weather elements, variations which are 
large compared with the permissible tolerance of observ- 
ational errors. Possibly this difficulty also can largely be 
overcome by smoothing techniques. But the principal 
difficulty lies in the fact that computing devices are not 
brains. They must be told what do to. At present our 
understanding of the mechanics of the general circula- 
tion is quite unequal to this task; there exists no practi- 
cal conception of how the large-scale circulation 
processes work; to serve as a physical or theoretical 
basis of computation. Furthermore, as pointed out 
above, it is not even known to what extent the future 
state of motion of the atmosphere is determined by the 
preceding state, or to what extent internal or external 
energy sources need be taken into consideration. Hence 
the mathematical extrapolation techniques run up 
against the same obstacle as do the less rigorous extra- 
polation techniques—need of a better understanding of 
the mechanics of the general circulation. At present, 
high-speed computing machines may be very useful to 
test the applicability of physical or theoretical models to 
the large-scale atmospheric processes in nature, but 
certainly they do not in themselves offer any particular 
hope of solving the basic problem of weather forecasting, 
which is to acquire a better understanding of the irregu- 
larly fluctuating circulation patterns. Their practical 
usefulness will probably increase as this knowledge is 
gained. 
A number of computational extrapolation techniques 
have been applied to practical weather forecasting, 
techniques which are based on various manipulations 
of the equations of motion and the equation of con- 
tinuity in which certain simplifying assumptions are 
made in their practical application to the atmosphere. 
The most promising of these forecast aids are those 
based essentially on the tendency equation of Bjerknes 
[3], and on Rossby’s use of the vorticity principle [9] 
to compute upper-level wave motions and air-particle 
trajectories. 
Numerous efforts, involving the tendency equation, 
have been made to compute the fields of horizontal 
convergence and vertical motion in the middle tropo- 
sphere from the observed wind field and to compare the 
thermal advective effects on the observed temperature 
field with the dynamic effects. Synoptic patterns com- 
puted in this manner are correlated with observed pres- 
sure changes, that is, eyclogenesis and anticyclogenesis, 
