Geology Reference
In-Depth Information
are still in use, even in a plethora of new sophisticated mathematical tools [
42
,
44
,
102
]. Over the last few decades, data mining techniques have been introduced and
widely applied in hydrological studies as powerful alternative modeling tools, such
as Arti
cial NeuralNetworks (ANN) [
14
,
21
,
29
,
31
,
66
,
68
], fuzzy inference
system (FIS) [
32
,
66
68
,
80
,
100
,
105
], and Data-Based Mechanistic (DBM)
models [
101
,
102
]. A comprehensive review by the ASCE Task Committee on
Application of Arti
−
cial Neural Networks (ANN) in Hydrology [
4
,
5
] shows the
acceptance of ANN technique among hydrologists.
A major criticism of ANN models concerns their limited ability to account for
any physics of the hydrologic processes in a catchment [
1
,
2
,
51
]. That concern was
partially ruled out through a study by Jain et al. [
38
], which proved that the
distributed structure of the ANN is able to capture certain hydrological processes
such as in
ow, etc. These AI
techniques exhibit many advantages over conventional modeling techniques,
including the ability to handle enormous amounts of noisy data from dynamic and
nonlinear systems. However, despite the good performance of these techniques
when used on their own, there is still room for further improving their accuracy and
reducing their uncertainties [
29
,
31
]. One new trend is to combine these AI tech-
niques so that individual strengths of each approach can be exploited in a syner-
gistic manner [
67
]. In this respect, parallel to ANN application, some researchers
came up with neural networks coupled with linear dynamic models such as ARX
and ARMAX to form NNARX (neural network autoregressive with exogenous
input) and NNARMAX (neural network ARMA with exogenous input) [
27
,
47
].
Integration of neural networks with fuzzy rules has introduced a model type called
the neuro-fuzzy system. Neuro-fuzzy models make use of potential abilities of both
these intelligent techniques in a single framework, effectively using the learning
ability of ANN to construct a best fuzzy set of IF
ltration, base
flow, delayed and quick surface
THEN rules. Another hybrid
approach which recently appeared in hydrology is the integration of discrete
wavelets transformed with neural nets. In recent years it has proved its abilities as a
strong mathematical tool in analyzing time series properties such as variations,
periodicities, trends, etc. [
7
,
8
,
54
,
55
,
91
,
99
,
104
]. Some studies show that the
wavelet transform is suited for predictions in hydrology and water resources. Wang
and Ding [
94
] have effectively applied neuro-wavelet (NW) models to perform
short- and long-term prediction of hydrological time series [
3
,
49
,
70
,
106
] per-
formed the application of a NW technique for modeling monthly stream
-
ows and
compared the results with the conventional ANN, regression, and transfer function
models. A detailed literature review of Support Vector Machines (SVMs)-based
hydrologic modeling and forecasting can be found in Yu et al. [
103
]. Similar
research advances could be found in solar radiation modeling as well. Over the last
few decades, many empirical and physical radiation models have been proposed to
estimate solar radiation from other meteorological variables including ARMA and
Fourier analysis [
28
,
65
]. Recently, approaches for predicting solar radiation series
have been developed using arti
cial neural networks (ANNs) reported from dif-
ferent parts of the world [
69
,
82
].
