THE FORECAST PROBLEM 
is based on the expectation of a continued uniform trend 
of development of the synoptic weather pattern. In this 
character it is particularly susceptible to the basic 
weakness of all of the extrapolation techniques of fore- 
casting—eross inaccuracy if extended beyond a very 
limited time range under rapidly or erratically changing 
conditions. 
All of these synoptic extrapolation techniques, which 
in the past have constituted almost the sole basis of 
short-range and daily forecasting practice, and which 
even today are supplanted only occasionally and to a 
minor degree by techniques of mathematical extrapola- 
tion or physical reasoning, are limited by a ceiling of 
potential performance, which probably is reflected very 
closely by the best current forecasting. This ceiling is 
imposed by the failure, and presumable inability of this 
synoptic extrapolation procedure, even when supple- 
mented by the best synoptic experience, to anticipate 
systematically or evaluate reliably the indecisively er- 
ratic or exceptionally abnormal changes of the synoptic 
weather patterns. The question may legitimately be 
raised, To what extent and over what periods of time 
is the succeeding weather pattern determined by the 
preceding pattern? or To what extent may factors 
quite external to the lower atmosphere play a determin- 
ing role? But it appears certain that if either synoptic 
or mathematical extrapolation techniques are to lead 
to any basically substantial improvement of weather 
forecasting, it will be only on the basis of a much better 
physical understanding of the mechanics of the atmos- 
pherie circulation processes than is represented by the 
present-day models. 
Tn recent years there has rapidly developed an exten- 
sive application of synoptic extrapolation techniques to 
extended and long-range forecasting, for periods of five 
or ten days, a month, or even a season in advance. 
Instead of applying the extrapolation techniques to the 
weather patterns on a sequence of daily synoptic charts, 
making use of 24- or 12-hr changes and the 3-hr tend- 
encies, five-day, weekly, or monthly mean charts pre- 
sent the weather patterns which are extrapolated by 
means of weekly or half-weekly and monthly or half- 
monthly changes, and 24-hr tendencies. The techniques 
of extrapolation are not essentially different; basically 
only the time scale is changed. The maps which are 
used must be geographically extensive, covering at 
least half (meridionally) and preferably the whole of the 
Northern Hemisphere, thus making possible the analy- 
sis and characterization by circulation indices, such as 
the zonal westerly index, of the state of the general 
circulation as an integrated whole [14]. The synoptic 
extrapolation techniques are more effectively supple- 
mented by statistical aids in this extended type of fore- 
casting than in the daily forecasting, notably by lag-cor- 
relation statistics and by the comparison of prevailing 
with normal tracks of cyclones and anticyclones. 
This type of extended synoptic extrapolation (by 
use of mean charts), supplemented by statistical and in 
some cases by physical considerations, forms the basis 
of the five-day and the experimental monthly forecasts 
disseminated by the Extended Forecast Section of the 
737 
U. S. Weather Bureau [6], of the 5- and 10-day and 
longer-range forecasts of Baur [1] in Germany, of the 
composite-chart technique of the Multanovski-Pagava 
school in Russia, and of many others. 
The statistical and synoptic extrapolation techniques 
as applied to extended and long-range forecasting are 
definitely limited in their potentialities. These tech- 
niques perform best for short-range forecasting, show- 
ing decreasing skill with increasing range. However, 
their performance in extended and long-range forecast- 
ing of mean weather conditions over extended periods 
goes far beyond anything that is attainable by the 
extrapolation of daily weather patterns. Probably 75 
per cent for temperature and 65 per cent for rainfall 
represent approximately the top skill performance 
which can be hoped for as an average in the forecasting 
of mean conditions by these methods for one week in 
advance on the basis of our present limited under- 
standing of the general circulation. Doubtless the skill 
margin of these potential ceiling performances should be 
cut in half for monthly forecasts, and possibly to one- 
third for seasonal forecasts. 
Frequently the extended forecasts of the mean weath- 
er patterns are used as a guide by which, and frame- 
work within which, to extend the prognosis of the 
day-to-day sequence of weather patterns beyond the 
two- or three-day limit which is justifiable by the daily 
forecasting techniques. There is certainly some justifica- 
tion for making such an extension, possibly to five or 
six days beyond the current date, but beyond that, even 
though mean conditions can be forecast with some 
degree of skill, no detectable skill in forecasting the 
day-to-day sequence of change has been demonstrated, 
nor does it seem likely to be in the absence of some 
radical advance in the science of weather forecasting. 
This advance can come only from an increased physical 
understanding of the mechanics of the general circula- 
tion, not from additional statistical or synoptic manipu- 
lation. 
The use of analogues represents a routine method of 
forecasting by synoptic extrapolation which has been 
rather extensively employed and which can be, but is 
not necessarily, applied quite independently of other 
techniques. This method consists essentially of classify- 
ing the large-scale synoptic weather patterns by their 
significant features in some manner such that past maps 
or conditions which are essentially similar to the current 
pattern may be extracted readily from files. The weather 
sequences which followed in the past under similar 
conditions, at the same season of the year, are then used 
as a guide to the prognostication of the current 
sequence. This extrapolation can be made by applying 
the statistical analysis of a number of similar synoptic 
patterns in the past (Baur’s method), or by selecting 
from past patterns the single closest fit to the current 
pattern, and following in detail the weather sequence 
which resulted in that case (IKvick’s method). Usually 
analogue classification and selection is based primarily 
on sea-level maps, but upper-level charts may be given 
any desired weight. The method has been extensively 
