Geology Reference
In-Depth Information
8.3 Case Study in Hydrology
8.3.1 Study Area
The study was performed with discharge data from the Beas River which originates
in the Himalayas and
flows for an
approximate length of 470 km and it has a basin area of around 12,561 km
2
. The
major two dams in the basin are Pong dam and Pandoh dam, both of which are
predominantly used for irrigation, hydroelectric, and
flows into the Sutlej River. The Beas River
flood control purposes. The
Pandoh dam is a diversion dam and diverts nearly 4,716 million m
3
of Beas waters
into the Sutlej River. The daily meteorological data used for the study are from
National Centres for Environmental Prediction (NCEP) Climate Forecast System
Reanalysis (CFSR) gridded data. The available data from the above-mentioned data
sets are daily precipitation (mm/day) [Precip], daily maximum temperature (
°
C)
[Tmax], daily minimum temperature (
C) [Tmin], daily average solar radiation (MJ/
m
2
daily [Solar], daily average wind velocity (m/s) [Wind], and relative humidity
(%) [RH]. The study has used river discharge time series (daily in
°
ow data to the
Pong dam) obtained from Bhakra Beas Management Board (BBMB) [
7
]. The study
has used 5 years
'
data for the analysis from 1 January 1998 to 31 December 2002.
8.3.2 Application of Statistical Blockade
The capability of SB is checked in this case study in comparison to state-of-the-art
ANNs and support vector repressors. The methodology adopted is shown in
Fig.
8.7
. The histograms (with bin size of 100) for different meteorological data sets
for 1 January 1998 to 31 December 2002 are given in Fig.
8.8
. This data was used
for synthetic generation of peak events and corresponding input space points to
train the support vector regressor. As one can see Fig.
8.7
, we have applied SB for
identi
flood over a given threshold value.
After inputting the input space and output values to the model, as per the
threshold value given by the modeler, the SB uses a classifier (support vector
classi
cation of peak
filter out candidate MC points which will not generate
values of interest to us in the tail. After proper training, the next time the SB sees a
tail value (e.g., extremely high
er in our case) to
flood value), the model would be able to give a
corresponding combination of inputs which could possibly make high
ood sce-
narios. So, in a way, a properly trained SB is more suitable to
find input space or
range or input space which produces extremes tail values. Thus, it has huge sig-
ni
cance in hydrological terms to identify quickly possible multi-dimensional input
space causing major
floods. The SB can make synthetic data from unblocked tail
values
fit the points to a GPD, so this method is very useful in data sparse situations
and ungaged stations.
