Geology Reference
In-Depth Information
Fig. 6.10 Cross-correlations of Q(t) with different time lags of precipitation (P(t-i)) and runoff
data (Q(t-i))
First cross correlation is suited in linear systems and rainfall runoff process is
nonlinear and the result from cross correlation may not always match the Gamma
Test. Secondly the cut off points based cross correlation is more subjective than the
Gamma Test and Entropy Theory.
To verify the reliability of the above results in GT and Entropy Theory in
identifying the suited number of data points in model construction, a data parti-
tioning approach was adopted [
22
]. Different scenarios of data partitioning into
training and testing periods were tried in order to discover the optimal length of
training data required for modelling without over
tting
occurs because of a large number of parameters and training data length, which lead
to a precise
tting during training. Over
fit memorising the set of training data and thereby loose generalization
and poor validation results. Figure
6.11
shows different partitioning scenarios and
the related CORR and RMSE values for each scenario during training and vali-
dation using an ANN model. As per Fig.
6.11
, the best RMSE is obtained for
Fig. 6.11 Different data partitioning scenarios into training and testing periods and their
modelling results
