Geoscience Reference
In-Depth Information
60
Key
Frontier
Solutions
50
40
Preferred solution
30
20
10
40
0
5
10
15
20
25
30
35
Complexity
FIGURE 8.8
Eureqa
: accuracy-complexity trade-off for rainfall-runoff modelling of Annapolis River.
Preferred solution
120
Key
AME (m
3
s
-1
)
100
MSDE (dimensionless)
PDIF (m
3
s
-1
)
RMSE (m
3
s
-1
)
80
60
40
20
0
-20
0
5
10
15
20
25
30
35
40
Complexity
FIGURE 8.9
Eureqa
: model complexity plotted against different measures of fit for rainfall-runoff modelling
of Annapolis River. Notes: AME, absolute maximum error; PDIF, peak difference; RMSE, root mean squared
derivative error; MSDE, mean squared derivative error.
can determine a prferrred model, that is, one that is a selected over others. The least complex
solution is, as might indeed be expected, a naïve/persistence model that has a complexity of 1
(Equation 8.6). The preferred complexity-accuracy solution is presented in Equation 8.7 (com-
plexity = 25). This represents the simplest model that included a snow-related factor, in which
depth of snow cover, not daily snow fall or maximum daily temperature, appears to be a primary
indicator of snow melt.
Simplest model:
Q
(
t
) =
Q
(
t
− 1)
(8.6)