Geology Reference
In-Depth Information
couple DWT with nonlinear models like ANN, ANFIS and SVM and apply these
novel hybrid schemes to three case studies. In some cases, a well calibrated data
based model may not provide faultless forecast results over a longer time. To
improve and make such dynamic conceptual models suitable for operational and
long term real time predictions, integration of on-line or sequential data assimilation
techniques could be used. Partial Recurrent Neural Networks (PRNN) is good
example for such model with data assimilation principle. Kalman
filtering is a most
common data assumption exercise widely used in environmental application along
with data based models with de
nite state space architecture. Application of data
assimilation approaches and related case studies are beyond the scope of this topic.
The relevant questions in data based modelling in hydrology are how useful is a
model for predicting a particular component within the hydrological cycle and does
a complex model work better? Though such debates are prominent in physics based
modelling, related literature is almost non existant in the case of data based mod-
elling. The usefulness of any model depends ultimately on the directional accuracy
of its estimates, not on its ability to generate unassailably correct numerical values
[
76
]. Critical evaluation on the usefulness of models based on sensitivity modelling
error and complexity is essential in data based modelling. This study introduces an
index of utility for critical evaluation of models in different modelling situations
which utilises information like model sensitivity (response to changes in training
data set), model complexity (changes in training time) and model error (closeness of
simulation to measurement) for all used models in this topic. Extreme value
modelling is a challenging
field in hydrology. This made an attempt to use state-of-
art Statistical Blockade in extreme value modelling and compare the capabilities
with other data driven approaches.
In short, this topic aims to address the above mentioned issues in data based
modelling by the following means:
1. The application of novel approaches in data and model selection to avoid the
aforementioned dif
culty associated with data based modelling;
2. A reliability check of the novel data selection approaches with conventional
methods;
3. To propose and use new wavelet hybrid schemes with traditional data based
intelligent systems;
4. To investigate the capabilities of popular and widely used arti
cial intelligent
models with newly proposed hybrid schemes.
5. To introduce statistical blockade to hydrology and compare the capabilities with
other models.
five objectives are accomplished through four case studies
broadly dealing with data based modelling in respective
The above mentioned
fields.
Chapter
2
describes the modelling issues associated with data based modelling in
hydrology. The chapter gives a detailed description on puzzling questions in
hydrology like model selection, selection of model input architecture, selection of
training data length, selection of best data interval etc. Another main goal of this
chapter is to suggest an approach to identify and characterize modelling quality as a
