Geoscience Reference
In-Depth Information
TABLE 8.5
Input Variables and Notation Used in Eureqa Rainfall-Runoff
Modelling for Annapolis River at Wilmot
Input Predictor
Lag 1
Symbol
Output
Lag 2
Daily discharge (m
3
/s)
Q
Q
(
t
)
Q
(
t
− 1)
Q
(
t
− 2)
Total daily rainfall (excluding snowfall) (mm)
RF
RF
(
t
− 1)
RF
(
t
− 2)
Minimum daily temperature (°C)
LT
LT
(
t
− 1)
LT
(
t
− 2)
Maximum daily temperature (°C)
UT
UT
(
t
− 1)
UT
(
t
− 2)
Total daily snowfall (mm)
SF
SF
(
t
− 1)
SF
(
t
− 2)
Thickness of daily snow cover (cm)
SC
SC
(
t
− 1)
SC
(
t
− 2)
TABLE 8.6
Mathematical Functions Used in
Eureqa Rainfall-Runoff Modelling
for Annapolis River at Wilmot
Symbol
Description
+
Addition
−
Subtraction
*
Multiplication
/
Division
sin
Sine
cos
Cosine
C
Constant
TABLE 8.7
Other Settings Used in Eureqa Rainfall-Runoff Modelling for Annapolis River at Wilmot
Setting
Description
Error metric
Squared error
Relative weight assigned to variables
None
Data splitting
Treat all points equally (50:50, training/testing)
Data pre-processing options (e.g. removal of outliers)
None
Stopping point
Not defined
Our knowledge of the climatic conditions in the catchment area supports the logical
assumption that snow melt in a hydrological model for river flow is potentially important.
More detailed analysis of the error using a variety of goodness-of-fit statistics is presented in
Figure 8.9. That particular graphic shows that the model with a complexity value of 25 is one of
the most accurate solutions. It also confirms that similar levels of fitness can be achieved, using
even simpler models, perhaps ones that were evolved at earlier stages in the model development
process. However, simpler solutions could have been developed at any point during the overall
search process. The variables used in such models did not include snow melt and so could be
rejected. By using multiple lines of evidence (complexity, goodness-of-fit, utilised variables), we
