734 
best single technique. If the techniques are essentially 
different, then when their prognostic indications are 
inconsistent, the forecaster’s judgment must enter into 
any selection between them. That judgment is likely to 
be prejudiced in favor of the one or two techniques 
which are believed best on the basis of past perform- 
ance. 
In the following classification of forecasting tech- 
niques all of those classified as extrapolation in type 
employ exclusively the current and past weather pat- 
terns, including usually trend and in a few cases acceler- 
ation patterns (first- and second-order time derivatives 
of the weather elements). Since most of these extrapola- 
tion techniques are employed for forecasts of more than 
one type (time range) they are classified by their 
essential nature, rather than by the type of forecast for 
which they are used, as statistical (dealing with the time 
sequences of weather in limited areas), synoptic (dealing 
with geographical weather patterns), or mathematical 
(dealing with the fundamental equations of motion, 
continuity, and state). 
In contrast to the extrapolation techniques of fore- 
casting, those techniques classified as physical are con- 
cerned primarily with conditions which are expected to 
impose modifying influences on the future development 
of the current weather pattern, influences which are not 
in evidence in the past development, hence cannot be 
considered in any procedure involving extrapolation 
only. These physical processes are further subdivided 
between those whose effects are specifically related only 
to the current weather pattern, hence are of significance 
only to the short-range or daily type of forecast, and 
those which are more permanent in their influence, 
affecting any weather pattern for some time to come, 
and which are therefore of primary significance for the 
extended or long-range type of forecast. 
1. Extrapolation Techniques—based on the existing. 
state and trend of the weather, exclusive of modifying 
physical factors. 
a. Statistical techniques of extrapolation—based on 
the time sequence of change and association of the 
weather elements individually in restricted localities. 
The purely statistical, as opposed to synoptic, forecast 
aids are usually applied to the long-range forecasting of 
the mean weekly, monthly, seasonal, or even annual 
anomalies of pressure, temperature, or precipitation. 
Statistical methods are occasionally applied to daily 
forecasts, usually to supplement the empirical climato- 
logical information of a forecaster faced with the prob- 
lem of forecasting for a region with which he has had 
relatively little experience. For this purpose statistics - 
can be prepared, on the basis of past records, to indicate 
the probability of any meteorological occurrence, or 
combination of occurrences, to assist the forecaster in 
choosing between alternative possibilities, or to decide 
the degree of severity which an anticipated condition 
is likely to attain. However, by virtue of long experi- 
ence in forecasting for a given region the forecaster 
usually develops a strong sense of the probability of any 
weather occurrence in his district, and his judgment of 
the potentialities of a given synoptic situation is likely 
WHATHER FORECASTING 
to stand him in far better stead in estimating its 
probable development than are the statistical proba- 
bilities. Consequently, purely statistical methods can 
make at best only a very dubious contribution to fore- 
casting of the short-range or daily type, particularly in 
middle and higher latitudes where the weather is so 
erratically variable from day to day. Although cli- 
matology and the statistical probability of both time 
and space sequences of weather change may well be 
developed further for the guidance of the forecaster, it 
can be assumed with reasonable certainty that no 
amount of effort expended in this direction will lead to 
any radical improvement of the shorter-range types of 
forecast. 
For the extended and long-range types of weather 
forecasting the statistical extrapolation techniques as- 
sume greater relative importance with increasing range, 
as their applicability and performance are relatively 
independent of the time interval to which they are 
applied, whereas the usual short-range synoptic extra- 
polation techniques rapidly become ineffective as the 
time range is extended. The statistical extrapolation 
techniques which are applied to the longer-range fore- 
casting of mean pressure, temperature, and precipita- 
tion anomalies at selected points are quite simple in 
principle, being based essentially on persistence and 
trend, linear lag-correlation (regression), and periodic 
analysis. It is quite obvious that none of those statistical 
extrapolation techniques can be expected to forecast the 
exceptional or unusual weather occurrences that are of 
greatest practical importance, because they are all keyed 
to the most probable occurrence. 
Pure statistical extrapolation techniques are identical, 
whether the length of the unit period involved is the day, 
the week, the month, the season, or the year. Linear 
regression equations, based on linear correlation, may 
be computed from past records to give the most prob- 
able numerical anomaly of any meteorological element 
for any selected station or district, on the basis of the 
values of the preceding one, two, or any arbitrarily 
selected number of periods. The simplest example of 
this technique is to compute the probability of simple 
persistence or reversal of the sign of an anomaly from 
one calendar month to the next. Usually such lag 
correlations are low, but occasionally in restricted areas 
and for certain seasons the correlation is high enough to 
have real forecasting significance. This method of sta- 
tistical extrapolation is assisted by the fact that in 
many regions there exists high contemporary seasonal 
correlation between pressure, temperature, and rainfall, 
so that the forecast anomalies of these elements must be 
mutually interdependent in a meteorologically con- 
sistent manner. 
An elaboration of the lag-correlation technique of 
statistical extrapolation is furnished by the use of lag 
correlation between current or past weather anomalies 
in one region, or in one branch of the general circulation, 
and subsequent anomalies in another region or another 
branch. This technique, which is capable of almost 
unlimited experimental application, has been used by 
many investigators, but it is only occasionally that 
