Hawaiian Rainfall Estimates — Stidd 
273 
Hawaiian rainfall. This rest (Aubert, MS.) 
covered the period February, 1948, through 
January, 1950. An experienced long-range 
forecaster and a meteorologist without much 
long-range experience worked independently, 
using observed mean monthly sea-level maps 
to estimate the rainfall for various Hawaiian 
stations. They relied heavily on Solot’s find- 
ings. In order to compare the present method 
of objective estimates with this test, the 2- 
year period involved in the test was deleted 
from the data of the present study, and all the 
necessary correlations were recomputed. New 
key points were chosen and the regression 
constants were altered so that the resulting 
regression equations were entirely independ- 
ent of any of the data from the test period. 
In that portion of the original test which 
dealt with winter months, the experienced 
forecaster had a skill score of 0.25 as against 
a score of 0.20 for the inexperienced fore- 
caster. The objective method, computed by 
a statistical clerk, had a skill score of 0.29. 
These scores may be interpreted in the 
usual sense, namely, that they are the correct 
portion of the group of forecasts which would 
have been in error had no skill been involved. 
Without more detailed information they have 
little meaning in the absolute sense, but they 
are stated here to indicate relative abilities. 
Thus, all tests to date seem to indicate that 
objective methods based on correlation field 
patterns can do at least as well as the sub- 
jective methods of trained forecasters. 
LIMITATIONS OF METHOD 
The chief objection to the correlation field 
method is the fact that only one or two points 
on the map are considered and all the rest of 
the map is completely disregarded. As has 
previously been stated, it does not seem feas- 
ible to overcome this objection simply by the 
addition of more points, because each added 
point boosts the multiple correlation only 
very slightly. An alternative approach to the 
problem would be to attempt to express all 
the salient features of the map in a few simple 
Fig. 7. Correlation field patterns of 700-millibar 
height versus Oahu dry-index (average of four low- 
rainfall stations on Island of Oahu), a. Monthly 
amounts, winter, 102 cases; b, five-day amounts, winter, 
255 cases. 
terms, using the method of orthogonal poly- 
nomials as described by Wadsworth (1948) 
or the method of spherical harmonics as de- 
scribed by Haurwitz and Craig (1952). This 
relatively small number of parameters could 
then be correlated with rainfall and the best 
ones chosen as predictors. Such an approach 
tested on data from the United States 
mainland, showed little improvement over 
the correlation field method where the 
key-point correlations were =±=0.6 or higher. 
Where the correlation fields were weaker than 
this, however, improvement was indicated. 
