ISSN: 2321-8819 (Online) 2348-7186 (Print) Impact Factor: 1.498 Vol.

7, Issue 9, September, 2019

A Model to Predict Turbidity Levels of Water in Kelani River, Sri Lanka and to
Determine the Required Alum Dosage For Settling Suspended Solids

J. M. T. Y. Jayaweera1*, S. K. L. S. Rupasinghe2, R. P. K. C. M. Ranasinghe1, J. A. D. Tharaka1

1
Department of Mathematics, University of Sri Jayewardenenpura, Sri Lanka.
2
Research and Development Division, National Water Supply and Drainage Board (NWSDB),
Ratmalana, Sri Lanka
*
E-mail: [email protected]
*
Tel. Number: 071-0811234

Abstract by industrial, domestic and other wastes, runoff

from urban areas such as roads, rooftops,
Water turbidity or the murkiness of water caused by
constructions, land clearing and other industrial
suspended solids (SS) is the most common issue related
to water sources in Sri Lanka. Murkiness of water is activities [3]. Conductivity is identified as a reason
usually treated with alum (Aluminium sulphate). Jar test for the decrease of turbidity [4].
is used to select the optimum dose of alum, and there is Primary production in an aquatic ecosystem is
no automated system for dosage calculation.
affected by turbidity having an inverse relationship.
For the present study, data was collected from the water This in turn affects the food and feeding of other
treatment plant in Biyagama, the right bank water organisms of the aquatic ecosystem. High turbidity
treatment plant of Kelani river. We constructed statistical of water is likely to increase the water temperature
models for predicting the turbidity levels in Kelani river. which in turn can bring adverse effects to dissolved
Using these models we calculated the required alum
oxygen levels of water that may interfere with life
dosage for settling SS in raw water. All predictors were
significant for these models. We created an efficient and cycles of various aquatic organisms. The danger of
user-friendly automated computer program to calculate clogging of the gills of fish and burying of bottom
the optimum alum dosage. dwelling organisms due to suspended solids in
water with high turbidity is yet another issue.
Keywords: Water Turbidity, Kelani River, Disease-causing pathogens or toxic pollutants are
Biyagama Water Treatment Plant, Multiple linear also carried by particles in turbid water [5, 6, 7].
regression model, Excel VBA userform
Removing turbidity is an important task in drinking
Introduction water treatment process as the suspended solids in
There are very important natural resources, and turbid water lead to bad tastes and odours of water
water is one of them. Although almost three fourths and there may be toxic pollutants and disease-
of the surface is filled with water 1% is portable. causing pathogens. These suspended solids are
This 1% potable water sources are becoming minute particles and they cannot be removed
increasingly contaminated due to erosion, domestic through just filtration [5, 6, 8, 9]. In Sri Lanka,
waste and industrial waste, hazardous and toxic Alum (Aluminium Sulphate) is the most commonly
substances etc. In addition, the rapid increase in used coagulant to remove turbidity by coagulation.
world population and industrialization has There are alternative coagulant methods used in
attributed to increased water pollution, which has other parts of the world [10]. The maximum alum
now become a severe global problem [1]. amount in potable water (residual alum dose)
Therefore, well-operated and efficient treatment should be 0.2 mgl-1 to avoid decaying of bones [11]
processes are a main necessity for the production of while adding less amounts of alum doses result
safe potable water. inefficient floc formation [5]. Also, alum is one of
the cost factors related to the water treatment
There are several issues related to water resources process. Therefore, using an optimal dosage of
in the world and in the Sri Lankan context, alum for raw water is important.
turbidity in raw water is identified as the most
common issue. Turbidity is the murkiness of water Jar test is the universally recognized test in use to
due to suspended solids such as silt, sand, germs examine the coagulant dosage which produces the
and other small particles in water [2]. The sources best floccule. New chemicals are also tested using
responsible for water turbidity in Sri Lanka are the jar test, to find out more effective, affordable
identified as increased soil erosion in upstream coagulants which produce less sludge. Presently,
areas due to development activities, heavy rainfall, Jar test is widely used for selecting the optimum
landslides, climate changes, inappropriate dose of Alum and there is no proper automated
vegetation etc. In addition, turbidity is also caused system as an alternative [8]. Therefore,
establishment of a statistical model for predicting

38
Asian Journal of Multidisciplinary Studies, 7(9) September, 2019
 ISSN: 2321-8819 (Online) 2348-7186 (Print) Impact Factor: 1.498 Vol. 7, Issue 9, September, 2019

turbidity levels is identified as the significance of 2. Build up a model to predict alum dosage
the study. for settling suspended solids (SS).
pH, conductivity, colour, alkalinity,
Biyagama water treatment plant of the Kelani river
turbidity were considered as the
and Hanwella rainfall data collection centre were
independent variables and alum dosage
used as data sources in this research. Kelani river is
was considered as the dependent variable.
by far Sri Lanka’s second longest river [12] and
according to the Environmental Foundation
3. Study of rainfall with turbidity and alum
Limited (EFL), it is the most polluted contaminated
dosage – To find out the correlations
and in high risk. Kelani river system provides
between rainfall and turbidity, rainfall and
goods and services including potable water and
alum dosage and build up models with
other industrial and domestic usage [13]. Biyagama
rainfall as a variable to predict turbidity
water treatment plant, which is fed by the Kelani
levels and alum dosage for settling
river, is the first ISO 9001:2015 QMS certified
suspended solids. Pearson correlation
water treatment plant in Sri Lanka [14]. It has the
coefficient was used to find out the
capacity to purify 160 million litres of water per
correlations between rainfall and turbidity
day. Hanwella rainfall data collection centre is the
and rainfall and alum dosage.
nearest data collection centre to the Biyagama
water treatment plant.
4. Create an automated program to find the
The study was focussed on the construction of required alum dosage for settling
statistical models for predicting the turbidity levels suspended solids (SS) through the model
and calculating the required Alum dosage for to predict the alum dosage without rainfall
settling SS. Through the turbidity model, factors as a variable - Excel VBA (Excel Visual
affecting turbidity can be identified and it enables Basic for Applications) was used to create
predicting the turbidity based on the available the automation program.
independent variables. To calculate the required
For the models we constructed, autocolinearity and
alum dosage, it is important to build up a statistical
multicolinearity were tested and assumptions of a
model for predicting alum dosages [15]. Therefore,
linear regression model were checked. Data had
through the alum model, factors affecting alum
been recorded over the time. Then, stationarity was
dosage were identified. It may be difficult without
also tested to check whether a trend is visible.
a scientific knowledge to calculate the alum dosage
SPSS was used to analyse the data. Mean Absolute
through the alum model manually. A user-friendly
Percentage Error (MAPE) was used to find out the
automated programwas created to calculate the
prediction accuracy of models [17].
required alum dosage for settling suspended solids
[16], so that it can be utilized by any person to Mean Absolute Percentage Error (MAPE) can be
obtain the optimum alum dose. expressed as follows:
Material & Methods  1 n Oi   E i  
MAPE      100%
Data of Biyagama water treatment plant for a
period of 3 years from 2015 August to 2018 August
n
 i 1 O i  

Oi  is the i th actual value and E i  is
was considered as the sample for this research. 652
data points were included in the sample and where
purposive sampling method was applied. All th
the i predicted value.
variables were identified as continuous variables.
Hence, the multiple linear regression model was Results
identified as the most appropriate model for the
dataset. Table 1 suggest that Durbin–Watson statistics for
both the models did not lie within 1.5-2.5 range,
This research is divided into four main parts as; depicting the models were affected by
1. Build up a model to predict turbidity autocollinearity and hence residuals were not
levels independent and identically distributed. The data
pH, conductivity, colour, alkalinity were resource, Kelani river is identified as a continuous
considered as the independent variables flow, resulting in continuous generation of data,
and turbidity was considered as the values for the considered variables had been
dependent variable. recorded once a day. Therefore, slight correlations
could be found in between consecutive values of
the same variables. Nevertheless, a regression

39
Asian Journal of Multidisciplinary Studies, 7(9) September, 2019
 ISSN: 2321-8819 (Online) 2348-7186 (Print) Impact Factor: 1.498 Vol. 7, Issue 9, September, 2019

approach was used in the study, as in the initial tested using Augmented Dickey-Fuller test. There
analysis, time plots of turbidity and alum did not were enough evidence to conclude that the series is
show any sign of a possible trend as show in Fig. 1. stationary as the null hypothesis was rejected at
This was further confirmed when stationarity was 0.05 level of significance (𝑝 value  0.00)
Turbidity

Alum
Time
Time

Figure 1: Time series plots of turbidity and alum dosage

For the models that were considered (Table 1), all the 𝑝 values from the individual t tests to check the
significance of predictors were less than 0.05. Hence all the predictors were significant at 0.05 level of
significance. All VIF (Variance Inflation Factor) values were less than 10, indicating that the models were not
affected by multicollinearity. Eventhough, it was observed that normality in distribitions of residuals were very
close to expected normality, the assumptions of normality were violated according to the results of Shapiro-
Wilk test (𝑝 value  0.00). No significant patterns were envisarged in the scatterplots of predicted value versus
residuals.
Table 1: Details of multiple linear regression models of alum and turbidity
Durbin
P value
VIF Sum of squares Correlation Watson
(t test)
Statistic
Log Turbidity model Without Rainfall
(Constant) 0.000 Regression 52.342
R 0.780
Inverse of pH 0.000 1.195 Residual 33.626
Log color 0.000 1.091

2nd power of inverse R Square 0.609 0.989

0.000 1.583
of conductivity Total 85.968

Inverse of Adjusted
0.000 1.527 0.606
Alkalinity R Square
Alum model Without Rainfall
(Constant) 0.000 Regression 2291.207
2nd power of Log R 0.749
0.000 2.052 Residual 1795.568
turbidity
Log color 0.000 1.215
R Square 0.561
Conductivity 0.000 2.260 0.884

Alkalinity 0.011 1.885 Total Adjusted

4086.775
R Square
0.557
pH 0.000 1.382

40
Asian Journal of Multidisciplinary Studies, 7(9) September, 2019
 ISSN: 2321-8819 (Online) 2348-7186 (Print) Impact Factor: 1.498 Vol. 7, Issue 9, September, 2019

Durbin
P value
VIF Sum of squares Correlation Watson
(t test)
Statistic
Log Turbidity model With Rainfall
(Constant) 0.000 Regression 53.593
R 0.790
Inverse of pH 0.000 1.215 Residual 32.375
Log color 0.000 1.108

2nd power of inverse R Square 0.623

0.000 1.609 1.052
of conductivity
Total 85.968
Inverse of alkalinity 0.000 1.548
Adjusted
2 days lag of 0.620
0.000 R Square
rainfall 1.161
Alum model With Rainfall
(Constant) 0.000 Regression 2329.309
2nd power of log R 0.755
0.000 2.159 Residual 1757.466
turbidity
Log color 0.000 1.220
R Square 0.570
Conductivity 0.000 2.261 0.909
Alkalinity 0.021 1.894
Total 4086.775
pH 0.000 1.387 Adjusted
0.566
2 days lag of R Square
0.000 1.186
rainfall

Model to predict turbidity levels

Based on alkalinity, conductivity, colour, pH, a model for predicting turbidity was constructed.
1 1
log 𝑇𝑢𝑟𝑏𝑖𝑑𝑖𝑡𝑦 = −2.849 + 5.064 ( ) + 658.116 ( )2
𝐴𝑙𝑘𝑎𝑙𝑖𝑛𝑖𝑡𝑦 𝐶𝑜𝑛𝑑𝑢𝑐𝑡𝑖𝑣𝑖𝑡𝑦
1
+ 0.177 log 𝐶𝑜𝑙𝑜𝑢𝑟 + 22.050 ( ) (1)
𝑝𝐻
As depicted in Table 1, 𝑅2 = 0.609 and adjusted 𝑅2 =0.606indicating approximately 61% of the variation of the
dependent variable could be explained by the model.
Model to predict alum dosage for settling suspended solids (SS)
Based on turbidity, alkalinity, conductivity, colour, pH,the model for predicting alum dose was built up.
Alum = 14.524 + 1.795(log 𝑇𝑢𝑟𝑏𝑖𝑑𝑖𝑡𝑦)2 − 0.050 𝐴𝑙𝑘𝑎𝑙𝑖𝑛𝑖𝑡𝑦 + 0.045 𝐶𝑜𝑛𝑑𝑢𝑐𝑡𝑖𝑣𝑖𝑡𝑦
+ 1.337 log 𝐶𝑜𝑙𝑜𝑢𝑟 − 1.335 𝑝𝐻 (2)
2 2
As shown in Table 1, 𝑅 and adjusted 𝑅 values were 0.561 and 0.557 respectively.
Rainfall with turbidity and alum dosage
Correlations
As shown in below Table 2, values of Pearson correlation coefficient between turbidity and rainfall in the
considered day with previous day, two days before and there days before (the first three lags) were 0.172, 0.233,
0.269 and 0.127, respectively. It was observed that the 2 nd lag of rainfall (𝑅𝑎𝑖𝑛𝑓𝑎𝑙𝑙𝑡−2 ) had the highest
correlation with the turbidity of the river. It was discovered that the 2nd lag of rainfall was the most influential in
the alum dosage as well.

41
Asian Journal of Multidisciplinary Studies, 7(9) September, 2019
 ISSN: 2321-8819 (Online) 2348-7186 (Print) Impact Factor: 1.498 Vol. 7, Issue 9, September, 2019

Table 2: Pearson correlation coefficients of rainfall data

Correlation Coefficient
Rainfall Turbidity Alum
0.172 0.219
Considered day
0.233 0.308
1 day before considered day
0.269 0.359
2 days before considered day
0.127 0.220
3 days before considered day

Model to predict turbidity with rainfall

Alkalinity, conductivity, colour, pH, rainfall were the response variables and turbidity was the predictor variable
for the model to predict turbidity.
2
1 1
log 𝑇𝑢𝑟𝑏𝑖𝑑𝑖𝑡𝑦 = −2.639 + 4.576 ( ) + 627.118 ( )
𝐴𝑙𝑘𝑎𝑙𝑖𝑛𝑖𝑡𝑦 𝐶𝑜𝑛𝑑𝑢𝑐𝑡𝑖𝑣𝑖𝑡𝑦
1
+ 0.162 log 𝐶𝑜𝑙𝑜𝑢𝑟 + 20.839 ( ) + 0.003 𝑅𝑎𝑖𝑛𝑓𝑎𝑙𝑙𝑡−2 (3)
𝑝𝐻
Related with table 1, 𝑅2 and adjusted 𝑅2 values were 0.623 and 0.620 respectively.

Model to predict alum dosage for settling suspended solids (SS) with rainfall
Turbidity, alkalinity, conductivity, colour, pH and rainfall were considered as independent variables and alum
dosage was considered as the dependent variable for this model.
Alum = 14.000 + 1.711(log 𝑇𝑢𝑟𝑏𝑖𝑑𝑖𝑡𝑦)2 − 0.045 𝐴𝑙𝑘𝑎𝑙𝑖𝑛𝑖𝑡𝑦 + 0.044 𝐶𝑜𝑛𝑑𝑢𝑐𝑡𝑖𝑣𝑖𝑡𝑦
+ 1.290 log 𝐶𝑜𝑙𝑜𝑢𝑟 − 1.262 𝑝𝐻 + 0.017 𝑅𝑎𝑖𝑛𝑓𝑎𝑙𝑙𝑡−2 (4)

According to table 1, 𝑅2 and adjusted 𝑅2 values were 0.570 and 0.566 respectively.

Error measures of models

As shown in Table 3, Mean Absolute Percentage (MAPE) values of alum models (2 and 4) were 11% and
MAPE values for turbidity models (1 and 3) were 17%, approximately. Therefore, the accuracy of the
predictions obtained out of the four models were relatively high.
Table 3: MAPE values for four models

Turbidity Models Alum Models

With rainfall Without rainfall With rainfall Without rainfall

16.70% 17.23% 10.81% 10.98%

Automated program to calculate the required entering data for date, pH, alkalinity, turbidity,
Alum dosage for settling suspended solids (SS) colour, conductivity to the database through the
through the model to predict the alum dosage data entering userform. User also can search, delete
without rainfall as a variable. and edit data rows in the database by entering the
date. Data can be saved in the database by the data
Excel VBA was used to create the automated
entering userform.
program. This program was based on two
userforms; login userform and data entering
Discussion
userform and an excel database.
Data about rainfall are not recorded in the
In the login userform, user is able to enter the
Biyagama water treatment plant and therefore, data
system by entering a valid username and a
of Hanwella rainfall data collection centre were
password and receive the required alum dosage by
used for the study since it is the nearest data

42
Asian Journal of Multidisciplinary Studies, 7(9) September, 2019
 ISSN: 2321-8819 (Online) 2348-7186 (Print) Impact Factor: 1.498 Vol. 7, Issue 9, September, 2019

collection centre to the Biyagama water treatment longitudinal through time series analysis) could be
plant. Hence, separate models with rainfall factor considered for the analysis.
were built, to enhance the accuracy by reducing the
Conclusion
effect of distance. It can be assumed that the
second lag of rainfall had the highest correlation In this study, four models were constructed in order
with turbidity of the river since it takes some time to predict turbidity levels, alum dosage for settling
for rainwater to accumulate with river. Our SS, turbidity levels with rainfall as a variable and
findings agree with other researches, Coulliette and alum dosage for settling SS with rainfall as a
Noble (2008) and Goransson, Larson and Bendz variable. All predictors were significant for these
(2013), supporting the claim that a typical delay in models. Having a high colour index and rainfall are
turbidity response to rainfall [18, 19]. identified as the reasons for the increase of
turbidity and the colour index, rainfall, and
The study was carried out with the available data
turbidity factors have a positive effect on the alum
records and known factors in the laboratory and
dose as well.
more unexplained errors could have been
accounted by the regression model, if more factors The pH, conductivity and alkalinity factors cause
such as soil erosion, land use, surface runoff, speed the reduction of the turbidity and as a result, the
and time of coagulation and flocculation were required alum dose as well. The accuracy of the
considered [20]. predictions obtained out of the models were
relatively high as depicted by the mean absolute
Alum is used as a coagulant in coagulation in water
percentage errors. Based on the model using known
treatment process. Therefore, speed and time of
factors, an automated system was created to
coagulation may have significant correlation with
calculate the optimum alum dosage to settle
alum dose [21, 22]. Then it is suggested to improve
suspended solids in raw water.
alum models by introducing such factors to mimic
the physical conditions. Acknowledgement
Panel data analysis approach could be used as an Authors would like to thank the contribution of
advanced analysis method, considering the multi- National Water Supply and Drainage Board
dimensional (cross- sectional and longitudinal) (NWSDB), Sri Lanka and Department of
effect and irregularity of time intervals for future Meteorology, Sri Lanka for providing data for the
researches. Then both dimensions of the dataset study.
(cross-sectional through regression analysis and

References
1. https://www.nationalgeographic.com/environment/freshwater/freshwater-crisis/, Retrieved on
19/7/2018

2. H˚akanson, L. (2006). Suspended particulate matter in lakes, rivers, and marine systems, The
Blackburn Press, New Jersey, United States, 2006.

3. Wang, X. M., Hao, R., Huo, J., & Zhang, J. F. (2008). Modeling sediment transport in river networks,
Physica A, 387, 6421–6430.

4. Winston, W. E., & Criss, R. E. (2002). Geochemical variations during flash flooding, Meramec River
basin, May 2000, J. Hydrol., 265, 149–163.

5. Hamilton, J.L., & Luffman, I. (2009). Precipitation, pathogens, and turbidity trends in the Little River,
Tennessee, Phys. Geogr., 30, 236–248.

6. https://dec.alaska.gov/water/wqsar/wqs/pdfs/Attachment_F_FAQ_Turbidity_in_Surface_Waters_1108
13.pdf+&cd=19&hl=en&ct=clnk&gl=lk, Retrieved on 19/7/2018

7. Bhutiani, R., & Khanna, D. (2007). Ecological study of river Suswa: modeling DO and BOD, Environ.
Monitor. Assess., 125, 183–195.

8. Introduction to Water Treatment, American Water Works Association, Colorado, 1984, vol. 2.

43
Asian Journal of Multidisciplinary Studies, 7(9) September, 2019
 ISSN: 2321-8819 (Online) 2348-7186 (Print) Impact Factor: 1.498 Vol. 7, Issue 9, September, 2019

9. Standard methods for the examination of water and wastewater, American Public Health Association,
American Water Works Association and Water Environment Federation, Maryland, 1995, ed. 19.

10. https://www.snf.co.uk/chemical-coagulants-used-water-treatment/, Retrieved on 19/7/2018.

11. Sri Lanka Standard (SLS) 614:2013, Sri Lanka Standards Institution, 2013.

12. https://www.researchgate.net/publication/237054480_An_Artificial_Neural_Network_Model_for_Rive
r_Flow_Forecasting, Retrieved on 10/09/2018.

13. http://crip.lk/?p=3221, Retrieved on 12/09/2018.

14. http://bwtp.yolasite.com, Retrieved on 13/09/2018.

15. https://www.statisticssolutions.com/the-multiple-linear-regression-analysis-in-spss, Retrieved on

22/08/2018.

16. https://www.tutorialspoint.com/vba, Retrieved on 25/10/2018.

17. https://www.forecastpro.com/Trends/forecasting101August2011.html, Retrieved on 30/11/2018.

18. Coulliette, A.D., & Noble., R.T. (2008). Impacts of rainfall on the water quality of the Newport River
Estuary (Eastern North Carolina, USA), Journal of Water and Health, 6.4, 473- 482.

19. Goransson., G., Larson., M., & Bendz., D. (2013). Variation in turbidity with precipitation and flow in
a regulated river system – river Gota Alv, SW Sweden, Hydrol. Earth Syst. Sci., 17, 2529–2542, 2013

20. https://www.fondriest.com/environmental-measurements/parameters/water-quality/turbidity-total-
suspended-solids-water-clarity/, Retrieved on 19/7/2018.

21. https://iaspub.epa.gov/tdb/pages/treatment/treatmentOverview.do?treatmentProcessId=1934681921,
Retrieved on 19/7/2018.

22. Saritha, V., Srinivas, N., & SrikanthVuppala, N.V. (2017). Appl Water Sci, 7: 451.

44
Asian Journal of Multidisciplinary Studies, 7(9) September, 2019