Geology Reference
In-Depth Information
Table 4.1 Some applications of SVM models in hydrology
Research area
Application
Bray and Han
[
14
]
Rainfall-runoff
dynamics
Applied to Bird Creek catchment in UK to simulate
runoff with varying units and magnitude
Moghaddamnia
et al. [
61
]
Surface water
hydrology
Improved prediction of daily evaporation using
other measured variables as inputs
Remesan et al.
[
74
]
Hybrid models
Applied SVM in conjunction with wavelets to
predict stream flow in Brue catchment in UK
Hossein et al.
[
37
]
Evapotranspiration
modeling
Estimation of ET
0
values using other daily weather
variables from IRIMO, Iran as inputs
Jian et al. [
40
]
Stream
ow
modeling
Effective modeling and site-speci
c forecasting of
stream
flow and reservoir in
flow for reservoir
management (in Manwan Reservoir)
Mohsen et al.
[
62
]
Groundwater level
prediction
Study has used pumping, weather, and
groundwater level data for modeling purpose
Seyyed et al.
[
81
]
Ground water
quality
Probabilistic support vector machines (PSVMs)
was used for the purpose
Wei [
95
]
Hybrid model
They have used an advanced wavelet kernel SVMs
(WSVMs) for forecasting hourly precipitation dur-
ing tropical cyclone (typhoon) events
Ganguli and
Reddy [
26
]
Drought modeling
Drought modeling in Western Rajasthan, India
Wu et al. [
98
]
River stage
modeling
Next day river water height prediction at Han-Kou
Station in Yangtze River
all sets of available four kernels were used. To narrow down the kernel functions, the
parameters of each kernel were kept at their default values and, consequently, a fair
comparison of each SVM could be made. In the next stage, we have tried SVM
modeling with different kernel functions and different SVR types (
ʽ
-SV regression
ʵ
ʵ
and
-SV regression and linear kernel
performed quite a lot better than the remaining models, as per some of our case
studies. In this
-SV regression). The results from SVMS with
value 0.01; these values
are usually evaluated through trial and error experiments which are most economical
on computing processing time and have the lowest RMSE (root mean squared error).
Some of the studies adopted SVM alone, and the DWT-based hybrid SVM scheme in
environmental science and applied engineering is covered in Table
4.1
.
figure, the cost value was chosen to be 2 and
ʵ
References
1. Aksoy H, Guven A, Aytek A, Yuce MI, Unal NE et al (2007) Discussion of generalized
regression neural networks for evapotranspiration modelling by O. Kişi
i (2006). Hydrol Sci J
ş
52(4):825
-
828
2. Aksoy H, Guven A, Aytek A, Yuce MI, Unal NE et al (2008) Comment on Kisi O (2007)
Evapotranspiration modelling from climatic data using a neural computing technique. Hydrol
Process 21(14):1925
-
1934. Hydrol Process 22(14):2715
-
2717
