Geology Reference
In-Depth Information
Fig. 7.23 Variation of Gamma statistic (
ʓ
) for the data corresponding to different combination of
input data sets
elimination of each input). Daily wind speed (W) is observed as the most signi
cant
input in evaporation modeling, of which (
scenario) resulted in very high
values of Gamma statistics. The embedding 1011 model [W, RH, Ed] was identi
“
No W
”
ed
as one of the best structures because of its low noise level (
value) (Fig.
7.20
) and
low V-ratio value (indicating the existence of a reasonably accurate smooth model).
Two other models are the 1111 model [W, T, RH, Ed] and the 1101 model [W, T, Ed].
As per the results obtained from GT, explained in Table
7.5
and Figs.
7.20
,
7.21
,
7.22
and
7.23
, training data length for evaporation modeling was identi
ʓ
ed as 2,413
with the lowest Gamma value 0.02116 and SE 0.00103 (Figs.
7.20
and
7.22
) for the
combination [W, RH, Ed] (
'
No T
'
scenario). Even though the Gamma value is a bit
high for [W, T, RH, Ed] (
scenario), both
models can make a model with less complexity and error (less gradient and SE in
Table
7.5
). If the
'
All
'
scenario) and [W, T, Ed] (
“
No RH
”
'
'
scenario data set is selected for modeling, it can make a
model with the training data length equal to 1325, with a Gamma value 0.02118
and SE value 0.00103 (Fig.
7.21
). Another close combination with less complexity
and error is [W, T, Ed]. The M-test analysis on input combination [W, T, Ed] (
“
No
RH
All
scenario) has shown that it can make an optimum model with optimum
training data length equal to 2327, with the Gamma value 0.02401 and SE value
0.00088. However, in this section, the GT analysis suggested three different and
close performing models with different input structure and training data lengths.
Further analysis using entropy theory, AIC, and BIC explained in the following
sections helped to make a concrete decision on the model input structure and
training data length.
”
