218
Wolffetal.
training data set: even
38 sample data set: odd
R" = 0.73
ANN
~ 37 f - _ S _ O -+-=_O_.54_-+-_ _ -+--.-_*---I
i c:
~ 36
Q)
c:
~
o
~ 35~··----~~--~~--~4------+-----I
...
E
,g
Z. 34
:5
m
34
35
36
observed salinity
reference data set: even
38 sample data set: odd
37
38
MAT
R" = 0.69
SO = 0.49 .._. ____________ .1 _____ _
. . . . . . _ . . . . _ . . _+_ . . . . --. . _----F ... 36
c:
...
E Q)
I
------t---~.~_,~~w~~L-_+----~
~ 35~----~--~~~--~f-----+_----~
E
E
,g
34
35
36
37
38
observed salinity
training data set: odd
38 sample data set even
R2 = 0.58
'"
SO = 0.63
~ 37 f---+---+_---+------~
t c:
~ 36
Q)
c:
'iii
:§
ANN
~ 35r----A-r--~~~--·~~------_+-----__4
...
E
,g
~34
.5
'iii
'"
34
35
36
observed salinity
reference data set odd
38 sample data set even
34
35
36
observed salinity
37
38
MAT
37
38
Fig. 8. Salinities obtained by ANN and MAT vs. annual mean salinities taken from Levitus (1982) for each half of
the CLIMAP core-top data set. Stipled lines represent bivariate regression lines. Solid lines show expected 1: 1
relationships. R2 is goodness-of-fit of regression line. Standard deviation (SD) refers to deviations of estimated
from observed salinities. MAT shows slightly lower standard deviations but straight regression lines deviate considerably from 1: 1 relationship.