Anda di halaman 1dari 10

TR CM 0002

CSIR Centre for Mathematical


Modelling and Computer
Simulation

CALIBRATION AND
VALIDATION OF WATER
QUALITY MODEL
(Cae 1 River)

S Himesh, Rao C V C*, Mahajan A U*

Technical Report CM 0002

May 2000
Bangalore 560 037, India
Calibration and Validation of Water Quality Model
Case 1. River
1 2 3
By Himesh S ., Rao C.V.C. and Mahajan A.U.

Abstract
One dimensional steady state water quality model is calibrated and validated. Model coefficients are
estimated through field and laboratory studies. Performance of the calibrated model was statistically
evaluated for its predictive ability of water quality. Among different water quality parameters predicted (DO,
BOD, TKN, NO3-N, NO2-N), BOD results are found be in best agreement with the measured results with the
coefficient of regression of 0.9, coefficient of variation 7.2 %, Root Mean Square Error 19 % and relative
error 8%.

Key words: Calibration, Validation, Verification, Dispersion Coefficient, Biochemical Oxygen Demand, Dissolved
Oxygen, Model Coefficient, Kinetic Coefficient, Modelling, Environmental Impact Assessment

Introduction
Since prediction is the most important component of the Environmental Impact
Assessment study, accuracy of the predictive models and hence the uncertainties
involved in the predicted results must be known before hand to make any meaningful
judgement with regards to the impacts of the proposed project. Uncertainty is an
inevitable component of all predictions. The extent of uncertainty that invariably creeps
into prediction depends upon the quality of data and the nature of model used.
Information on impacts based on predictions is one of the most scientific and objective
components of EIA study [Peter 1998]. The extent of decision maker being wrong or
right with regards to the implementation of the proposed activity is greatly influenced by
the accuracy of the model prediction in addition to many other subjective elements
included in EIA (methods of evaluation of impacts, nature of impacts, socio-economic
component etc.). Because of its sensitive nature and ability to influence the final results
of EIA, prediction of impacts has been recognised as one of the major sources of
uncertainty in EIA studies. Since it is an accepted truth that, uncertainty is

1. Scientist, CSIR Centre for Mathematical Modelling and Computer Simulation (C-MMACS), NAL,
Bangalore –560 037, Karnataka, India. E-mail: himesh@cmmacs.ernet.in
2 and 3. Scientist EIRA Division, NEERI, Nagpur- 440 020, Maharastra, India.
an inherent and inevitable component of all predictions, it must be minimised to a level
that is acceptable from decision-makers point of view. Since Mathematical models are
extensively used in EIA studies for the prediction of the quality of the major components
of environment (air, water, land), and also in the evaluation of various pollution control
alternatives, these models must be calibrated and validated in order to minimise errors
involved in the prediction. The usefulness of an appropriately calibrated model ni
evaluating water pollution control strategies is demonstrated in one of the case studies
of water quality modelling (for the control of Phosphorous) of upper Mississippi River and
lake Peptone, [Seng Lung et al., 1995].

Purpose of the Study


Identification of sensitive model parameters through sensitivity analysis and their
estimation through laboratory and field studies, subsequent calibration and validation of
the one dimensional river water quality model under steady state condition were the
main objectives of the study. Performance evaluation of the calibrated model was also a
part of the study.

Calibration
Model calibration is the first stage testing or tuning of the model to a set of field data not
used in the original construction of the model. Such tuning is to include consistent and
rational set of theoretically defensible parameters and inputs (Thomann 1982). Model
calibration is actually the process by which one obtains estimates for the model
parameters through the comparison of field observations and model predictions. Even if
the steady state condition is assumed, the environmental parameters can still vary due
to random changes of temp, stream discharge, time of day, and general weather
conditions. Due to this inherent dynamic nature of the environment, discrepancies
between the predicted and observed results are bound to occur. How credible the model
is, or what is the level of confidence that can be placed on the model predictions? is all
depending on the range of discrepancies mentioned above. Such discrepancies must be
minimised to the extent possible by identifying and minimising sources of error
(measurement errors, conceptual error in the model, approximation errors due to nature
of model being calibrated). The effect of measurement errors can be minimised by
optimising data collection procedures like, collecting data in most sensitive locations and
by collecting optimum number of replicates. A first order error analysis or sensitivity
analysis can be used to identify critical measurements and sampling locations. It is also
desirable to estimate the variance of measurements between replicate samples at select
locations during study.

Verification and Validation


It is the testing of the calibrated model against the additional set of field data preferably
under different environmental conditions (river flow, waste load etc.), to further examine
the range of validity of the calibrated model. Collection of data for validation is such that,
calibration parameters are fully independent of the validation data. The model so verified
can be used for forecasting of water quality under a variety of perturbed environmental
conditions. However, it is apparently rare that, following a forecast and subsequent
implementation of environmental control program, an analysis is made of the actual
ability of the model to predict water quality responses, such an exercise is regarded as
“post audit.” of the model. It is nothing but the subsequent examination and verification
of the models predictive ability following the implementation of environmental control
program. The same model that is used herein for the present study was also used in the
prediction of water quality during EIA studies conducted earlier, for the same river but
with the assumed waste loads of the proposed industry.

Selection of the Model


One-dimensional steady state river water quality model, Qual-2E version developed by
US –EPA [Brown et al., 1987] was selected for the present study because, the same
model was employed in the earlier EIA studies at the same site. Since it was observed
that, water quality parameters in the stream vary predominantly in the longitudinal
direction, one-dimensional approximation was assumed. The relevant processes
considered in the present study can be simulated with this model. The model is one of
the widely accepted water quality simulation tools for waste load allocation purpose and
water quality impact analysis studies. For details regarding data requirements, sampling,
and model coefficients for Qual-2E, the reader may refer to the handbook on stream
sampling for waste load allocation [William et al., 1992].
Identification of Pertinent State Variable
Sensitivity analysis was performed to identify the critical model parameters and
sensitive sampling locations. Headwater BOD was found to be the most sensitive
parameter followed by de-oxygenation and reaeration rate. Dispersion coefficient was
found to be least sensitive parameter of the model. This parameter is seems to have
been included only to take care of numerical dispersion.

Hydrodynamic Data
River flows, velocity, and cross sectional details at different locations of the river were
collected. Since the velocity of a river varies with width and depth due to functional
effects, the mean velocity was estimated over 0.8 and 0.2 times the total depths of water
in the stream. Average of the two velocities was considered as the representative
velocity. However, in shallow rivers, even a single velocity of 0.6H would be sufficient.
Flow was obtained by the estimation of a velocity over the reach of the river using dye.
The time of travel (t) between the locations of the river was estimated using the tracer
(Rodamine). The average velocity of the river was thus found to be 0.05m / Sec. Since
the river cross section closely resembled trapezoidal shape, assumption of trapezoidal
cross section was considered appropriate. Bottom width of the river varied from 15 to 50
mt., whereas the top width varied from 20 to 60 mt. Side slopes varied from 0.015 to 0.1
and 0.03 to 0.1 for both sides of the canal (left and right bank w.r.t flow direction).
Longitudinal bed slope of the canal varied from 6.6E-07 to 1.3E-06.

Water Quality Data


Water quality data was collected at 12 sampling locations over a stretch of 10.4 Km.
Details of sampling locations and model network are shown in Fig.1. Average values of
DO, BOD, TKN, NO 2 and NO 3 and PO 4 varied from 9.1 - 12.8, 2-3, 3-8, 0.05-0.1, 1-1.5,
and 0.1-0.4 mg/L respectively.

Waste Load
The only wastewater out-fall being discharged into the river was from the National
Fertiliser Limited. Average value of wastewater flow was found to be 0.12 m3/Sec. The
values of BOD, DO, NO 2, NO3, TKN and PO4 were found to be 12, 7.6, 2.2, 2.6, 9 and
2.1 respectively.
Kinetic Coefficients
Based on sensitivity analysis, the nature of wastewater and the process to be simulated,
it was decided to estimate the following model coefficient as shown in Table.1. First
order error analysis (in-built program in Qual-2e) was performed for three locations;
upstream and downstream of outfall, and at the location of minimum DO. The results
indicated that, sensitivity coefficient of CBOD was relatively more for wastewater flow
and CBOD.

Table. 1: Model Coefficients

Model Coefficient Estimated Literature Calibrated Units

Kd 0.1 1/day
SOD 6.0 1-10 gm/m2/day
KL 0.001 0.002-0.004 Kcal/m2/sec
Kn 0.015 0.003-0.900 mg/L
Kp 0.030 0.01-0.1630 mg/L
β3 1.210 0.02-0.4000 0.2 1/day
ka 4.000 0.00-100 1/day
β1 0.10-1.00 1.0 1/day

β2 0.20-2.00 2.0 1/day


α5 3.00-3.50 3.0 mgo/mg-N
α6 1.00−1.14 1.0 mgo/mg-N
αο 1.00 1.00-100 µg-chla/mg-A
α1 0.09 0.07-0.09 mg-N/mg-A
α2 0.015 0.01-0.02 mg-p/mg -A
Field Estimation of Dispersion Coefficient
Dispersion coefficient is regarded as one of the most important parameters, which
characterises the ability of the stream to disperse pollutants. This coefficient combines
the effect of diffusion with dispersion and generally known as hydrodynamic dispersion.
In the present study, this parameter was estimated through tracer study conducted in the
field. Rodamine was used as a tracer based on the traits of an ideal tracer viz., water
solubility, ease of detection, minimum background interference, stable, non toxic to
human beings, harmless to aquatic life and inexpensive (McCutcheon 1989). Based on
field studies, longitudinal dispersion coefficient was estimated to be 1m 2 / Sec and
dispersion constant was found to be K= 1223. Refer to Thomann et al. (1987) for
dispersion coefficient calculations.

Calibration Methodology
Calibration process is mainly based on the field measurements that helps in choosing
the empirical coefficients in water quality models and also in the verification of the
consistency of the model’s initial and boundary condition with that of the in-stream
measurements. Calibration of the hydrodynamic part of the model was first carried out
by comparing simulated hydrodynamic variables (depth and velocity) with the measured
ones.

Next, the calibration of the process compartment of the model was carried out
sequentially by using transformation kinetic parameters. The order of calibration is;
temperature, BOD, DO, nitrification and phosphorous related parameters respectively.
While calibrating water quality process part of the model, estimated parameters are
fixed, few parameters are extracted from standard modelling literature, remaining
parameters are obtained by tuning them till the observed and predicted results closely
matched. Refer Table.1 for details regarding model coefficients; estimated, calibrated
and extracted from literature.
Performance Evaluation of the Model
The calibrated model was used to predict the water quality with an independent set of
data as a part of validation exercise. Results of model predictions were fairly good, and
performance of the model was further confirmed through statistical evaluation of the
results. Statistical analysis of the model results indicate that, simulated CBOD results
were relatively in best agreement with the measured values with coefficient of regression
0.9, coefficient of variation 7.2%, RMS 19.2% and relative error 8%. Few typical plots of
observed and simulated results obtained during different phases of calibration are shown
in Fig 2 to 6.

Summary and Conclusions


One dimensional steady state water quality model was calibrated and validated. Model
coefficients are estimated through field and laboratory studies. Performance of the
calibrated model was evaluated through statistical techniques. Model output was found
to be very sensitive to headwater quality. Sensitivity of the Carbonaceous Biochemical
Oxygen Demand decay rate was found to be increasing towards downstream of the
outfall. Model coefficients related to settling rates of CBOD, Nitrogen and Phosphorous
fraction are found to be moderately sensitive. Among all the model coefficients of the
Qual-2e, dispersion constant was found to be least sensitive and seems to have been
included only to take care of numerical dispersion. Based on statistical evaluation of the
calibrated model it is evident that, the model has been calibrated reasonably well.
Among the state variables presented here, prediction of CBOD results found to be better
with relative error of 8%, considering the fact that 20 % deviation is normally acceptable
in standard river water quality modelling practices.

Acknowledgements
This paper is a part of the research work carried out at Environmental Impact and Risk
Assessment Division of National Environmental Engineering Research Institute–Nagpur.
Authors gratefully acknowledge the support and express their gratitude to the Scientist-
in-Charge & Head, C-MMACS –Bangalore, Director NEERI-Nagpur and project team
members of EIRA Division.
APPENDIX: I. REFERENCES
Brown, L.C., and Barnwell, T.O., Jr. (1987). “ The Enhanced Stream Water
Quality Models, Qual-2E and Qual-2E UNCAS: Documentation and Users
Manual.” EPA /600/3-87/007, Envir. Res. Lab., Envir. Protection Agency (EPA),
Athens, Ga.

Mc Cutcheon., S.C. (1989). Water Quality Modelling. Vol.-I. Transport and


Surface Exchanges in Rivers. CRC press. Inc., 103-110.

Peter., W. (1998). Environmental Impact Assessment: Theory &Practice .


Academic Division of Unwin Hyman Ltd., 62-97.

Seng Lung, W.U., and Larson C.E. (1995). “ Water Quality Modelling of Upper
Mississippi River and Lake Peppin.” Journal of Environmental Engineering,
Vol.121, No.EE10, 691-699

Thomann., R.V., and Muller., J.A. (1982) “ Verification of Water Quality Models.”
Journal of Environmental Engineering, Vol.108, No.EE 5, 923-940.

Thomann, R.V., and Muller, J.A. (1987). Principles of Surface Water Quality
Modelling and Control. Harper & Rowe, Publishers, Inc., New York, N.Y., 75-
80.

William, B.M., George, L. B., Thomas, M.G., and Kay, M. (1992). Hand Book, “
Stream Sampling for Waste Load Allocation Application.” US EPA, Office of
Research and Development, Washington, D.C 20460.
APPENDIX: II.NOTATIONS
Kd = De-oxygenation rate
SOD = Sediment Oxygen Demand
K L = Michaelis-Menton half saturation coefficient for light
Kn = Michaelis-Menton half saturation coefficient for nitrogen
Kp = Michaelis-Menton half saturation coefficient for phosphorous
Ka = Reaeratioin coefficient
β1 = Rate constant for biological oxidation of NH3 to NO2
β2 = Rate constant for biological oxidation of NO2 to NO3
β3 = Rate constant for the hydrolysis of organic nitrogen to ammonia
α1 = Fraction of algal biomass that is nitrogen
α 2 = Fraction of algal biomass that is phosphorous
α5 = Oxygen uptake per unit of NH3 oxidation
α6 = Oxygen uptake per unit of NO2 oxidation
α ο = Ratio of Chlorophyll-a to algal biomass
DO = Dissolve d oxygen
BOD = Biochemical Oxygen Demand
TKN = Total Kjeldahl Nitrogen
K = Dispersion Constant
D = Longitudinal dispersion coefficient