Geoscience Reference
In-Depth Information
TABLE 8.2
Input-Output Variables Used in GEP
Estimating of Pan Evaporation
Symbol
Units
Description
T
max
°C
Daily maximum temperature
T
min
°C
Daily minimum temperature
T
avg
%
Average daily temperature
RH
max
%
Maximum daily relative humidity
RH
min
%
Minimum daily relative humidity
RH
avg
%
Average daily relative humidity
Rs
MJ m
−2
Daily solar radiation
U
avg
m s
−1
Average daily wind speed
E
PA N
mm day
−1
Daily measured pan evaporation
automatically using a Class A Pan. Further details on our data set are provided in Table 8.2. NLWS
records are used principally because the data that were available for this particular monitoring
station covered a longer period of observation. In addition,
E
PA N
data were logged automatically at
NLWS, reducing the potential error that sometimes occurs with manual logging.
The GEP algorithm implemented in
GeneXproTools
4.0 was used to predict
E
PA N
from daily
meteorological data (Beriro
et al., 2013). Software settings are shown in Table 8.3. Ten independent
solutions were evolved. The preferred model was selected using testing values for
R
-squared (0.77),
mean absolute error (1.44), RMSE (2.00) and a one-at-a-time response function sensitivity analysis
of the evolved solutions (Beriro
et al., 2013). The goodness-of-fit statistics are summarised in
Table 8.4. The preferred GEP model is shown in Equation 8.5 and scatterplots comparing observed
E
PA N
with modelled values are provided in Figure 8.6.
Of particular interest in this study is that only 40% of models passed their respective sensitivity
analysis - meaning there was a 60% model redundancy rate (see Section 8.5.4 for a description of
how the sensitivity analysis was undertaken). Model redundancy is therefore a real issue - assuming
that is - that it is important for an evolved model to accurately reflect the conceptual underpin-
nings of its associated natural system. Results show that the performance of the preferred model
compared well to other published studies (Abudu
et al., 2011; Moreno
et al., 2010) supporting the
conclusion that GEP is able to produce a good predictive model for
E
PA N
:
Ê
PAN
=
Sub-ET
1
+
Sub-ET
2
+
Sub-ET
3
(8.5)
2
(
)
2
(
)
−
()
TRHT
−
Rs
max
min
max
Sub-ET1
=
A
tan
(8.5a)
(
)
−
7 488647
.
RH
min
Sub-ET
2 =
U
(8.5b)
avg
2
2
(
)
(
)
(
)
(
)
UT
0 286712
8 177459
.
T
avg vg
min
A
Sub-ET
3
=
tan
(8.5c)
RH
+
.
avg
in which
Ê
PA N
represents our calculated estimate of measured
E
PA N
.