Geology Reference
In-Depth Information
Fig. 6.4 The variation of gamma static and standard error with unique daily data points
rapid fall off of the M-test SE graph, relatively low V-ratio value (indicating the
existence of a reasonably accurate and smooth model), the regression line
t with
slope A = 0.04139 (low enough as a simple non-linear model with less complexity)
and good
it with SE 0.01009. These values altogether give a good indication that it
is quite adequate to construct a nonlinear predictive model using around 1,056 data
points and remaining data out of total 2,240 data points were used as the validation
data set for modelling.
6.4.1.2 Gamma Test with Other Data Intervals
Brue catchment observations from 1994 to 1995 were used for the study to analyse
the working GT with different data frequencies. The Gamma Test analysis was
performed for different data sets with the data collection frequencies of 15, 30, 60
and 120 min. The different combination of inputs evaluated in this study is shown
in Fig.
6.5
. The least Gamma value was observed when we used the input data
combination equivalent to [4, 4] for 15 min data set (high frequency data), i.e. four
antecedent runoff and four antecedent rainfall as model input. The network input
include information Q (t),Q(t
1),Q(t
2),Q(t
3),P(t),P(t
1),P(t
2) and
−
−
−
−
−
P(t
3) for multi-step
—
ahead forecasting when we used 15 min data. The Gamma
−
statistic (
) and Standard Error (SE) variation with unique data points of 15 min
data, obtained from the M-Test analysis are shown in Fig.
6.5
. The test produced an
asymptotic convergence of the Gamma statistic to a value of 0.00273 at around
16,004 data points (i.e. M = 16,004). The standard error (SE) corresponding to
M = 16,004 was relatively small at
Г
0.00041 [
18
]. As a result, M = 16,004 could
be used effectively to construct a reliable model, and we used 16,004 data points for
the training. Similarly, the Gamma Test based analysis has identi
*
ed [2, 2], [3, 3]
and [4, 3] input combinations as the best ones with the least gamma statistic value
for the data sets with time intervals of 30, 60 and 120 min. The M- Test details can
be found in Fig.
6.5
.
