Geology Reference
In-Depth Information
Fig. 5.20 Gamma statistic (
ʓ
), Gradient, Standard Error and V-ratio for the normalised daily data
used as inputs for daily solar radiation modelling (M = 1,098)
corresponding to the data points is also shown in Fig.
5.20
. In the
gure we can
note that, the SE corresponding to M = 770 is very small at
0.0019, which shows
the precision and accuracy of the Gamma statistic. We also performed M-tests in
different dimensions varying the number of inputs to the model (Fig.
5.19
), which
clearly presented the response of the data model to different combination of inputs
data sets. From the Fig.
5.19
we can also deduce that the combination of precipi-
tation, daily maximum temperature, daily mean temperature and extraterrestrial
radiation (ETR) can make a good model comparable to the combination which
composes all inputs. The signi
*
cance of the wind velocity and daily mean tem-
perature data sets was relatively small when compared to other input sets since the
elimination of these inputs made less variation in the Gamma statistic value. The
M-Test analysis results in different scenarios are shown in Fig.
5.21
which are
“
All
”
,
“
No ETR
”
,
“
No T
max
”
,
“
No T
mean
”
,
“
No T
min
”
,
“
No P
”
and
“
No U
”
(the scenario
corresponding to all six inputs, and other scenarios corre-
sponding to elimination of each input). Extraterrestrial radiation is observed as the
most signi
“
All
”
cant input in solar radiation modelling, of which (
“
No ETR
”
scenario)
resulted in very high value of Gamma statistic.
The embedding 111111 model (a six input and one output set of I/O pairs) was
identi
value), the rapid
decline of the M-test SE graph, low V-ratio value (indicating the existence of a
reasonably accurate smooth model), the regression line
ed as the best structure because of its low noise level (
ʓ
fit with slope A = 0.1108
(low enough as a simple non-linear model with a minimum complexity) and good
it with SE 0.0019. These values altogether can give a clear indication that it is quite
adequate to construct a nonlinear predictive model using around 770 data points
with an expected MSE around 0.0354.
