Geology Reference
In-Depth Information
Chapter 6
Data Based Rainfall-Runoff Modelling
Abstract This chapter explores the data selection and modelling approaches in the
context of Rainfall-Runoff modelling. The main goals of the chapter are (i) to
present the data driven models and data driven models in conjunction with data
selection modelling approaches like the Gamma Test, Entropy Theory, AIC and
BIC using daily information from the Brue catchment. (ii) to explore the effect of
data frequency (data interval for modelling) on data based real time
flood fore-
casting using different data sets. (iii) to perform a detailed investigation of hybrid
wavelet based intelligent models in the context of runoff prediction. (iv) to make
comparative investigation of the model performance of ANNs, ANFISs, SVMs,
W-ANN, W-ANFIS and W-SVMs. The last section of the chapter suggests a simple
procedure to estimate the utility of different models considering different attributes
like uncertainty (in terms of model sensitivity and error) and complexity (in terms
of modelling time) and applied to rainfall runoff modelling.
6.1 Introduction
Rainfall-runoff dynamics is a complex phenomenon due to its nonlinear and mul-
tidimensional interactions in the hydrological system. From the late 1990s many
research have extensively applied neural networks in rainfall and runoff modelling
[
1
,
7
,
20
]. The merits and demerits of arti
cial neural networks (ANN) are clearly
discussed and reviewed in the ASCE task committee on application of ANNs in
hydrology [
2
,
3
]. Many studies have discussed resemblances and disparities
between ANN and other statistical models [
6
] in rainfall runoff simulations. Even
though ANNs are considered as
'
black-box
'
models, a study by Wilby et al. [
24
]
showed that speci
c architectural features of ANNs can be interpreted with respect
to the quasi-physical dynamics of a parsimonious conceptual rainfall-runoff model.
In the early twenty
field of
hydrology. Maier et al. [
13
] provided a good review of neural network models used
since the year 2000. Models such as Adaptive Network-based Fuzzy Inference
first century, different types of ANNs are applied in the
