Geology Reference
In-Depth Information
Table 6.3 Variation of information criterions AIC and BIC with changes in training data length
Scenarios
Training data length
AIC
BIC
Case 1
250
17182.14
10843.5
Case 2
500
15219.36
8880.709
Case 3
750
13661.33
7322.688
Case 4
1000
12574.73
6236.081
Case 5
1100
12878.32
6539.673
Case 6
1250
13594.4
7255.757
Case 7
1500
15422
9083.357
Case 8
1750
11420.54
5081.889
Case 9
2000
13488.32
7149.678
The AIC and BIC values have shown that (t
1) step of runoff information
−
(Q(t
1)) is the most in
uencing data series followed by other data series like
−
Q(t
3). Even though this study
uses the only eight inputs for modelling, the positive results of the study highlights
its potential of application to a wide range of applications in the
2),Q(t
3),P(t
1),Q(t
4),P(t
2) and P(t
−
−
−
−
−
−
field of hydrology.
To study the variation of AIC and BIC with an increase in training data points, we
have performed several LLR model constructions with different training data lengths
(training data points 250, 500, 750, 1,000, 1,100, 1,250, 1,500, 1,750 and 2,000),
including all available input data series (i.e. four antecedent rainfall data series and
four antecedent runoff data series). The analysis results of variation of AIC and BIC
with training data lengths are given in Table
6.3
. The corresponding variation is
pictorially represented in Fig.
6.12
. It shows that there a dip in the criterion
values (both AIC and BIC) when we used 1,000 data points for the training.
