Journal of Computer and Communications, 2014, 2, 8-13
Published Online January 2014 (http://www.scirp.org/journal/jcc)
http://dx.doi.org/10.4236/jcc.2014.22002
OPEN ACCESS JCC
Monitoring the Sewage Draining in Shenzhen Reservoirs
Using Hyperspectral Data
Yin Li1, Wei Pan1, Xiaomao Yang2, Qinglin Tian1
1CNNC Beijing Research Institute of Uranium Geology (BRIUG), Beijing, China; 2College of Environmental Sciences and Engi-
neering, Peking University, Beijing, China.
Received September 2013
ABSTRACT
Freshwater resources are regarded as the foundation of urban development and assure the sustainable prosper-
ity of the city. The contaminations of fresh water in reservoirs can threaten safety of people directly and force
the Industrial processes to be suspended. Therefore, developing a method to detect the potential locations where
contaminated water drains off into the reservoirs efficiently and precisely is a challenging task but in urgent
need. In this research, we used the air-borne sensor Hymap to get the hyperspectral data of Shenzhen. Finally we
find a way to invert five water quality parameters (Suspended Solid (SS), Total Nitrogen (TN), Total Phosphorus
(TP), Chemical Oxygen Demand (COD), Biochemical Oxygen Demand (BOD5)) from the Hymap image and we
distinguish the clear water and polluted water on the image successfully.
KEYWORDS
Hyperspectral; Water; Hymap; Shenzhen Reservoirs
1. Introduction
Shenzhen is a modern international metropolis with de-
veloped industrial and densely population. Freshwater
resources are regarded as the foundation of Urban de-
velopment and assure the sustainable prosperity of the
city. The contaminations of fresh water in reservoirs can
threaten safety of people directly and force the Industrial
processes to be suspended. There are 171 reservoirs and
396 pound s in Shenzhen, 17 of the reservoirs are middle
size. Total capacity is 611 million cubic meters, each
year to provide 350 million cubic meters of drinking wa-
ter. Although the enormous amounts of water in the re-
servoirs, the per capita is less than 200 cubic meters,
about 1/12 of the national rate. It’s an urgent task to pre-
vent contamination of existing freshwater resources.
There are so many reservoirs in Shenzhen that it’s too
difficult to monitor the sewage draining in Shenzhen
reservoirs by manual effort. The reservoirs in Shenzhen
are surrounded by a large number of farms and industrial
parks; a way to monitor the sewage draining quickly and
effectively is in urgent need [1-3].
Monitoring water quality by remote sensing has grad-
ually developed from qualitative to quantitative and the
number of parameters is increasing. We can invert Chlo-
rophyll a, suspended solids, yellow substance, transpa-
rency, turbidity, water temperature, etc., and the retrieval
accuracy continues to improve [4].
2. Data and Process
2.1. Data
We get 34 water samples for this paper, 14 of them are
clear water and the other 20 of them are contaminated.
The locations of the samples are widespread in Shenzhen
and typically in the inlet, centre or outlet of the reservoirs.
Collecting time is October 31 to November 7, a period of
eight days.
Recode the locations of the 34 water samples by GPS
(the detail is shown in Figure 1), take note of water
transparency, water depth, surface temperature and other
related data. We had performed a chemical analysis on
those water samples (Including SS, TP, TN, BOD5 and
COD etc.). See Table 1 below for details.
We used ASD Fieldspce (R) Pro to measure the spec-
trums of the sample when we collected the samples. ASD
Fieldspce (R) Pro can measure continually the spectrums
from 350 nm to 2500 nm, the spectral resolution is 1 nm
and the number of output bands is 2501. The process
measurement was in clear, calm days. We take ten spec-
trums for each sample, then take the average as the orig-

Monitoring the Sewage Draining in Shenzhen Reservoirs Using Hyperspectral Data
OPEN ACCESS JCC
9
Figure 1. The locations of the 34 samples.
Table 1. Detail of the Biochemical parameters of samples.
Biochemical
parameters Clear water Polluted water
mean max min me a n ma x min
CODcr
Cr6+
Cu
Zn
Ni
Cyanide
SS
AN
TN
BOD5
TP
24.01
_ _ _
_ _ _
_ _ _
_ _ _
_ _ _
17.64
1.88
5.47
5.59
0.17
43.3
_ _ _
_ _ _
_ _ _
_ _ _
_ _ _
31
12.1
18.4
10.7
0.73
11.6
_ _ _
_ _ _
_ _ _
_ _ _
_ _ _
8
0.25
1.67
2.01
0.025
188.71
0.022
0.52
0.27
0.47
0.052
108.85
23.27
29.93
41.09
3.38
419
0.051
3.09
0.78
3.95
0.211
767
43
50.2
90.2
6.1
71.5
0.007
0.061
0.025
0.08
0.005
29
7
10.6
16.7
0.539
inal reflective spectrum of the sample. The data got by
ASD is synchronous Hymap data, which has 128 bands,
continue from 400 nm to 2500 nm. The spectral resolu-
tion of Hymap is 15 ~ 18 nm.
2.2. Processing of Data
Due to the spectrometer response differently in each
band, there is some noise in the original reflective spec-
trums. The original reflective spectrums were smoothed
by Mean Filter and the spectrum in 350 nm to 1300 nm
has been selected to be regarded the spectrum of the
samples. The details of the spectrum are show in the
Figure 2 below.
2.3. Spectrum Analysis
The most important influencing factors on inland water
spectrum are chlorophyll a and suspended solids, other
parameters are difficult to find independent spectral cha-
racteristics, we need extra information of the relationship
between different elements and invert them by remote
sensing indirectly.
Derivative spectrum can be mathematically calculated
from reflective spectrum. It shows the positions of min-
imum and the maximum bending point in reflection
spectrums. Research shows that derivative spectrum is
very sensitive to signal to noise ratio of the spectrum, but
it’s less sensitive to signal to noise ratio in lower differ-
ence, and it widely used in practical application. The re-
sult of correlation analysis is showed in the figures (Fig-
ures 3 and 4) below .
The figures show that the all six parameters (SS, TN,
AN, BOD5, TP, COD) have low correlations with all the
original reflective spectrum bands, but the some of them
have a high correlation with the derivative spectrum
bands. The derivative spectrum has a strong ability to
invert water parameters, but the derivative spectrum must
be calculated from high spectral resolution spectrum data.
To monitor the sewage draining quickly and efficiently,
we need to use the airborne or satellitic remote sensing,
but the spectrums get from aircrafts can’t have a high
spectral resolution like ASD, so the idea that getting in-
formation from derivative spectrum doesn’t work.
The figure also shows five parameters (SS, TN, BOD5,
TP, COD) have some consistent correlation with the de-
rivative spectrum bands, In band 150, band 200 and some
other bands, the correlation is significantly higher than
others. It indicates the five parameters have comprehen-
sive representations in some range of the spectrum. In

Monitoring the Sewage Draining in Shenzhen Reservoirs Using Hyperspectral Data
OPEN ACCESS JCC
10
Figure 2. Spectrum of the samples.
Figure 3. Correlations with original reflective spectrum bands.
this study, we choose six obvious reflective (or absorp-
tive) peaks. The detail is shown below in Table 2.
Count all the area of the six peaks in the 34 original
reflective spectrums, choose ten spectrums randomly,
build simultaneous equation with the peaks area and the
all six parameters, then use the model to approach the
parameters of the 34 samples. The result is given in the
Figure 5 below.
The figures show that the approach works well except
AN. This way can be used in invert the parameters of the

Monitoring the Sewage Draining in Shenzhen Reservoirs Using Hyperspectral Data
OPEN ACCESS JCC
11
Figure 4. Correlations with original reflective spectrum bands.
Figure 5. Result of approach the parameters. a. Sample of a Table footnote.

Monitoring the Sewage Draining in Shenzhen Reservoirs Using Hyperspectral Data
OPEN ACCESS JCC
12
water. Analyze the relationship between with the area of
the six peaks, the result indicates the peak 2 (from 684
nm to 706 nm) and the peak 4 (from 783 nm to 834 nm)
can be used to distinguish the polluted water and the
clear water. As the Figure 6 show s that the clear water
and polluted water are located separately, almost in two
lines. It means that we can use an index calculated from
the airborne or satellitic data.
2.4. Image Analysis
Use the GPS information of the 34 samples to find the
corresponding Hymap data. Because of the lower spec-
tral resolution and the influence of the atmosphere, the
spectrums of the samples are different with the data get
by ASD at some wavelength. The peak from 757 nm to
767 nm and the peak from 988 nm to 1155 nm do show
in the spectrums get from the Hymap image, but the peak
2 and peak 4 are still obvious in the Hymap spectrums.
Using the band 25, band 26, band 27, band 18, band 19
and band 20 to calculate the area of peak 2 and peak 4
and using the ratio of area of peak 2 to area of peak 4 as
the index to distinguish the clear water and polluted wa-
ter. The results are shown below. The red points indicate
the locations of polluted water.
As the Figure 7 sh ow s, we can identify the inlet of the
polluted water quickly and efficiently and it is also useful
in detecting the small inlet which is too small to find by
eyes.
3. Conclusion
Because of the complex aquatic environment of Shenz-
hen, SS, TN, AN, BOD5, TP and COD are not composed
of certain particles, they don’t have any certain spectral
characteristics. For all six parameters have low correla-
tions with all the original reflective spectrum bands, we
can’t use any single bands or ratio of two bands to invert
those parameters quantitatively. Derivative spectrum is
sensitive to the changes in the proportion of SS, TN, AN,
BOD5, TP and COD, some bands have high correlations
with the five parameters, it can work well with data that
have high signal to noise ratio and high spectral resolu-
tion, but we need to further study how to use this tech-
nique on Hymap image. The area of the six peaks can be
used for inverting the proportion of SS, TN, AN, BOD 5,
TP and COD accurately; it can work effectively in mon-
itoring the sewage draining in Shenzhen reservoirs; the
physical significance of this technique needs further
study. The ratio of area of peak 2 to area of peak 4 is
effective in distinguishing the polluted water and clear
Figure 6. The relationship between peak 2 and peak 4. a. Clear water are shown by blue points and polluted water are shown
by red points.

Monitoring the Sewage Draining in Shenzhen Reservoirs Using Hyperspectral Data
OPEN ACCESS JCC
13
(a)
(b)
(c)
(d)
(e)
Figure 7. The result of image analysis.
Table 2. Detail of the peaks.
Reflective
peaks Centre point (nm) Begin point (nm) End point (nm)
1
561 510 617
706 684 730
760 757 767
807 783 834
1068 988 1155
675 648 700
water, it can be used on Hymap image successfully, and
it’s a quick and effective way to monitor the sewage
draining in Shenzhen reservoirs.
REFERENCES
[1] H. T. Duan, L. Yu and B. Zhang, “Hyperspectral Data
Applied in Monitoring and Evaluating the Water Trophic
State of Lake Chagan,” Acta Scientiae Circumstantiae,
Vol. 26, No. 7, 2006, pp. 1219-1226.
[2] S. J. Li and X. J. Wang, “The Spectral Features Analysis
and Quantitative Remote Sensing Advances of Inland
Water Quality Parameters,” Geography and Territorial
Research, Vol. 18, No. 2, 2003, pp. 66-68.
[3] S. Wang, F. Yan and Y. Zhou, “Quality Monitoring Us-
ing Hyperspectral Remote Sensing Data in Taihu Lake
China,” IEEE International Proceedings of Geoscience
and Remote Sensing Symposium, 2005, pp. 4553-4556.
[4] E. Wit and J. McClure, “Estimating Total Suspended
Sediments Concentrations and Transparency with Hyper-
Spectral Reflectance in Shitoukoumen Reservoir, Jilin
Prvoince,” Acta Tournal of Lake Sciences, Vol. 19, No. 3,
2007, pp. 269-274.