- Email: [email protected]
Modelling of overflow emissions in flanders
~
Perganlon
Wal. Sci. T~clt. Vol. 38. No. 10. pp. 41-48.1998.
IAWQ C 1998 Published byElsevier Science LId. Printed in Great Britain. Allrights reserved
PI!: S0273-1223(98)OO731-8
0273-1223198 $19'00 + ()OOO
MODELLING OF OVERFLOW EMISSIONS IN FLANDERS G. Vaes and J. Berlamont Hydraulics Laboratory, University ofLeuven; de Croylaan 2. B-3001 Heverlee, Belg ium
ABSTRACT Ideally,for emissioncalculations long term hydrodynamic simulations should be performed. but this requires long calculationtimes. Simplifications are consequently necessary. Due to the non-linearbehaviourof sewer systems. hydrodynamic simulations using single storm evenIS often will not lead 10 a good probability estimation of the overflow emissions. Simplified models using long time simulations give better results if they are well calibrated. To increase the accuracy hydrodynamic simulations with short time series can be used. The short time series are selectedfrom the long time historical rainfallseries using a simplifiedmodel. To test the accuracyof these three methods,hydrodynamic long tenn simulations wereperformedfor several (small) sewer systems with differentcharacteristics to comparewith. ~ 1998Publishedby Elsevier Science Ltd. All rights reserved
KEYWORDS Overflow emissions; probability estimation; reservoir model; single storm event; short time series. INTRODUCTION Since more and more trunk sewers and treatment plants are built in Flanders. the water quality in the receiving waters is steadily improving. A lot of investments are in progress to improve the situation further. There still is a long way to go, but in the future the water quality in the Flemish brooks and rivers will (hopefully) be good again. Most of the sewer systems in Flanders are combined sewer systems. where rain water is transported together with sewage in the same pipes. Because treatment plants can not be designed for very high flow rates. combined sewer overflows are built in into the sewer system. Because of the accidental occurrence of overflow events. these events will have a large effect on the water quality and consequently on the biological life in the receiving water. A sudden disturbance of a good water quality will be more noticed than a continuous moderate water quality and will possibly also create more damage to the aquatic life (e.g, sudden fish kill). Not only the physical reality is important for this, but also the psychological perception of it. For this reason it is important to know the emissions at the overflows and their effect on the receiving water. It is obvious that it is impossible to measure the emissions on the thousands of overflow structures in Flanders . Besides. one wants to predict the emissions in advance to be able to optimize the design and one wants to know the long term performance of the overflows. The only way to assess the emissions is thus to model the sewer systems numerically. Besides the emissions, also other parameters influence the water quality in the receiving water, such as the characteristics of the 41
42
G. VAES and J. BERLAMONT
receiving water itself, the direct runoff, the effluent discharges of the treatment plant, etc... Even if we can estimate the water quality in the receiving water, we still do not know the effect on the biological life in the water. The assessment of overflow emissions is thus only one step in the whole process of integrated modelling, but a very important one. EMISSION MODELLING The ideal emission calculation is a detailed hydrodynamic routing (full implementation of the 'de SaintVenant' equations) using long time historical rainfall series and performing a statistical analysis on the emissions afterwards. This kind of simulations is well-known, but very time consuming. Therefore simplifications are necessary. Besides, also sediment transport and water quality aspects should be incorporated. However, the physics of these water quality phenomena are still not completely understood and water quality calculations are even more time consuming. Moreover, more extensive and accurate data are necessary for water quality simulations and more extensive calibration is required. The more calibration is needed the more unreliable models become; not everyone will calibrate a model in the same (right) way. Besides, how reliable are the data the model is calibrated with? This does not mean that detailed water quality models are useless; on the contrary: they can make the behaviour of the systems more comprehensible and give more insight in the processes which are going on in the sewers. When one proceeds to integrated and stochastic modelling the calculation time requirements become even more stringent; e.g. performing Monte Carlo simulations with a detailed integrated model using several decades of rainfall input would take years/decades to run for only one sewer system. Besides reducing the system to the principal aspects, gives a better view on the behaviour of the system. Too detailed models give a false feeling of accuracy, because of the uncertainty on the input data and the model parameters. These difficulties in emission modelling direct modellers towards a pragmatic solution. Somewhere there is an optimum between the accuracy of the model results and the efforts that have to be made to reach these results. This optimum can be different according to the circumstances and influenced by climate, land use, topography. type of sewer system, etc... One of the major principles in modelling is that all model parts must have the appropriate accuracy, taking into account the sensitivity for the parameters, to have an optimal model. This means that a detailed routing model for the flow in the sewers in combination with a very rough surface runoff model for example, will not be an optimal model for emission calculations. For these reasons different types of simplifications were investigated for the use in Flanders. SIMPLIFICATIONS To reduce the calculation times one can either simplify the rainfall input or the sewer system model. However, emission calculations are most sensitive to rainfall input. If one tries to simplify the rainfall into single storm events. the amount of information will decrease strongly and the accuracy of emission predictions using this approach will often decrease likewise. This is especially true for non-linear systems. Sewer systems behave in a non-linear way when they have no gravitary outflow (pumps, vortex controls), when they are looped, when the pipes get pressurized or when there is a backwater effect. This is often the case in flat regions as in Flanders. For non-linear systems the frequency of the overflow event will not necessarily equal the frequency of the rainfall event which leads to this overflow event. In that sense, emission calculations using single storm events will not give an accurate estimation of the probability of an overflow event. This effect increases as the frequency of the event and the non-linearity increase. This means that for design purposes one can use single storm events with a high return period without any problem. For these reasons it is often better to use simplified (conceptual) models for emission predictions than simplified rainfall. The statistical information about overflow events is much more important than the accuracy of the individual emission predictions. Detailed simulations with single storm events will make the behaviour of the system more comprehensible, but will never allow us to assess the emission probabilities. The variability of the rainfall in Flanders is so high that water quantities dominate the water quality. For this reason it is not necessary to incorporate very detailed water quality aspects, certainly not when a conceptual
Overflowemissions
43
model is used for the water quantities. Besides, water quality simulations are strongly influenced by local and time varying (stochastic) conditions, which makes it mostly impossible to model it in detail in an economical way. The use of constant pollution concentrations in a first stage or a very simplified transport model is thus acceptable; they are giving often as good (as bad) results as very detailed water quality models. MULTI STAGE APPROACH IN FLANDERS Because of the important investments in sewer systems and overflow constructions which have to be made during the coming years in Flanders, a considerable number of emission calculations has to be performed. Practically manageable computer tools for sewer system calculations are only available since about one decade; whilst only since the last few years commercially available computers are fast enough to perform realistic emission calculations. In Flanders, this has led to an approach in several stages when designing overflows, which is implemented in the new guidelines (Vlaamse Milieumaatschappij, 1996). The first stage in assessing the emissions of the overflow exists in performing hydrodynamic simulations using high frequency composite storms (Vaes et al., 1996). These are single storm events, one for each frequency or return period, which include the mean rainfall volumes for every duration between 10 and 720 minutes. For storm durations up to 240 minutes also the mean antecedent and posteriori rainfall is included. The IntensitylDurationlFrequency-relationships (IOF-relationships) which are used, are based on historical time series of 27 years from Uccle, which is approximately representative for the whole of Flanders (Vaes et al., 1996). For linear systems and overflows with a low overflow frequency this approach is acceptable in order to estimate the probability of the overflow events. For sensitive receiving waters (with regard to the effect of overflows) this first stage is not considered sufficient . To obtain a higher accuracy one has to move to another stage. For the second stage in assessing the emissions of the overflow a conceptual (reservoir) model of the sewer system is used. For this reservoir model the calibration is of major importance in order to obtain good results. For this calibration hydrodynamic single storm simulations can be used. The major parameters are the "storage" and the "throughflow" . Not only the maximum values for these parameters are important, also the variation of the throughflow with the storage is significant information. Graphs have been determined to perform a quick estimation of overflow frequency and mean overflow volume in function of the maximum storage and the maximum throughflow capacity for two extreme cases: a constant throughflow and a throughflow which varies linearly with the storage (Vlaamse Milieumaatschappij, 1996). For the third stage dynamic simulations are performed using short time series, which are selected from the long historical rainfall series, using a reservoir model (Vaes and Berlamont, 1996). When desired, the second stage (reservoir model) can be skipped and a dynamic simulation with a general set of short time series can be performed. For this general approach a set of short time series has been selected which include all rainfall events which will lead to an overflow event for a wide range of (Flemish) sewer systems. This reduces the rainfall input already with more than 90% (Vaes and Berlamont, 1996). These three stages in the emission assessment include water quantity aspects only. It is however possible to include a simple pollution transport model in the reservoir model. Ideally, it is necessary to include also other aspects than the emissions; an integrated water quality model of the river basin is then required. For this the rules for simplifications are quite similar as for emission predictions. Mostly, detailed water quality modelling is not necessary. because for the most sensitive brooks and rivers in Flanders there may not be an overflow structure at all. Also for discharges to receiving waters with lower standards, it may be more effective to choose the location of the overflow structure in function of the environmental effect than to limit the overflow events . Because of the high uncertainties on many input and model parameters, one will have to switch in the future to stochastic models instead of deterministic models. Assuming distributions for the most uncertain input and model parameters and performing Monte-Carlo simulations. will enable one to estimate the accuracy of
G. VAES and J. BERLAMONT
44
the results. But, this will lead to much longer calculations times, so that simplified models are certainly required. SINGLE STORM EVENTS
When one assumes that the probability of the overflow event is the same as the probability of the rainfall event which causes it, one can determine the frequency distribution for the overflow parameters using single storm events. The first requirement for this is that the probability of the rainfall events is well estimated. The composite storms which are now used in Flanders are based on IDF-relationships, but IDF-relationships can be determined in many different ways. The rainfall time step, the aggregation method for higher durations, the criteria to remove dependent events, the definition of an event and the extreme values method used for the higher return periods can influence the results significantly. Therefore IDF-relationships should be developed in function of their usage (Vaes et al., 1996). overflow discharge (m3/s) -long time series
0.8
-- composite storms
0.6 0.4
0.21_~~~_~=::::====::::==::i;;;;;:::=:::;::::::=_--l o o
2
4
6
8
10
12
frequency (#/year) Figure I. frequency distribution forthe twooverflows (. and +) in the sewersystemof Geel-Larum,
overflow volume (m3)
3S000
dynamic simulation + composite storms
30000
reservoir model + composite storms
-
2S000 20000
-reservoir model + long time series
ISOOO 10000 SOO:
l---'-~::::::~:;;;~=~==:::~===;=="i""'--'---ro
S
10
IS
20
2S
30
3S
40
4S
SO
frequency (#/year)
Figure 2. Probability of overflow volumes fromthe sewersystem of Dessel,
To check the consequences of the assumption of linearity between rainfall event and overflow event, long time hydrodynamic simulations (27 years) have been performed and the results were compared with single storm simulations. Several sewer systems were used for this with very different characteristics : linear as well as very non-linear systems and overflows with low as well as high overflow frequencies. This
Overflow emissions
45
comparison showed that the higher the overflow frequency and the higher the non-linearity. the less accurate the probability estimation is for the overflow events. In figure I the overflow discharge frequency distributions are shown for the long time series and the composite storms for two overflows with different overflow frequencies. In figure 2 the overflow volume frequency distributions are shown for one overflow, obtained with composite storms in combination with a dynamic simulation as well as with a reservoir model. Also the results of the reservoir model using the long time rainfall series are shown : the influence of the rainfall data used is much more important than the influence of the sewer system model. RESERVOIR MODELS In a flat region like Flanders often a (static) reservoir model (Figure 3) can lead to quite accurate emission predictions (Vaes, 1996a; Vaes and Berlamont, 1997) . This means that only a continuity equation and a simple throughflow equation have to be solved. The throughflow equation for the reservoir model must be well calibrated. Often a constant throughflow or a linear relat ionship between the throughflow and the storage in the system is assumed. In many real cases this relationship is much more complex (Figure 4). However it is very important to model this relationship as correctly as possible, because it affects the accuracy of the emission predictions strongly. The relation between storage and throughflow up to the moment the overflow starts spilling is used (Figure 6), as the storage at the beginning of the overflow event is a good estimation for the mean storage in the system (Figure 5). Only some extra "dynamic" storage is not incorporated for more severe storms. but this will hardly influence the overflow volume.
o
.
\\\\\\\ \\\\\\\taJn \\\\\\\
\\\ \ \ \\
evaporation
depression storage ~overflow
L
storage infiltration
in the sewer system
throughflow to the
treatment - . . plant Figure 3. schematicrepresentation of a reservoir model.
throughflow (mm/h)
10
8 6
-e-f= 1 -o-f=3
--f=2 _ . f=4 -~
4
--f=5 --f=7
f=6 --f=8
2
o
o
2
4
6
8
10
12 storage (rom)
Figure4. Storage/throughflow-relationship forthe second spilling overflow in the sewersystemof Geel-Larum for a rangeof composite storms withdifferent frequencies f.
46
G. VAES and J. BERLAMONT
12
storage in the sewer system (mm)
10 maXimum
i
8
beginning
6
4 2
end
~~BflII~HIIIHHIIHIHHHHH
! I II I I
0+---+---+---+--+--+--+--+--+--+--+--+----1
o
1
2
3
5
4
6 7 8 9 10 11 12 maximum overflow discharge (mmlh)
Figure S. Varialion of the storageduringthe overflowevent in two sewersystems(. and .) for a wide range of rainfallevents(composite stormsas wellas constantintensities overdifferentdurations).
throughflow Q (lis)
200 175 150 125 100 75 50 25
• real relationship -
'static' approach
.QI. . . .. . . . . . . . .. . . . . .••• .. - •••
• ••
••
.
•• ••
••••
••••
o
o
20
40
60
80
100
120
140
Figure6. Bi-Iinear'static'approach for the storage/throughflow-relationship of the sewersystemof Bilzen-Tabaartwijk.
A linear relationship is easy to implement and leads to a very fast calculation (104 to 106 times faster than a hydrodynamic simulation). To keep the advantages of a linear model and still include more complex relationships, a piecewise linear relationship can be implemented (Figure 6) (Vaes, 1996b). This is easy to fit to any existing relationship and a linear model is obtained by a coordinate transformation. When the simulations with a hydrodynamic model show a lot of hysteresis in the storage/throughflowrelationship , the "dynamic" storage can be taken into account as a piecewise linear relationship of the inflow (Figure 7). To calibrate this a few more hydrodynamic simulations with single storm events for different return periods (different inflow rates) are necessary. Other important parameters are the concentration time over which the inflow is averaged and the runoff model used. When the dynamic storage is incorporated. it can be useful to vary the concentration time with the inflow so that high inflows are less averaged than low
Overflow emissions
47
inflows and thus reach the overflow faster. The runoff model can have a large influence on the emission results, but it is often very difficult to calibrate.
outflow QOUI (m3/s) V stal = kstal QOUI
200
175 150 125 100
• real relationship
75 -- dynamic storage Vdyn - static storage VSlat 50 -- total storage Viol 25 0 ......~--+---~---+----+----4----4-----+-o 20 40 80 100 120 60 140 Figure7. 'Dynamic' approachfor the storageahroughflow-relationship of the sewersystemof Bilzen-Tabaanwijk.
A comparison of the results of reservoir models for different sewer systems with the above mentioned long term hydrodynamic simulations shows that the emission predictions are very good if the storage/throughflow. relationship is well calibrated. Although the error on some individual emission events can be significant (Figure 8), the probability distributions of the overflow parameters are well predicted (Figure 9). overflow discharge (m3/s)
•
0.6
0.5
- - dynamic simulation •• - bi-linealr 'static' reservoir model
0.4 0.3
0.2 0.1
o
5
10
15
20
25
30
35
40
time (hours with overflow) Figure8. Compressed timeseriesfor the overflow discharges fromthe sewersystemof Gecl-Larum usingthe rainfallserieso( Uccle(or 1967.
48
G. VAES and J. BERLAMONT
overflow volume (mm)
40 35 30
dynamic simulation
25
hi-linear 'static'reservoir model
20
15 10 5
o 4-+--t--+--+-~:::::+:=+~~==-t""=--+--1
o
1
2
3
4
5
6
7
8
9
10
11
12
13
overflow frequency (#/year)
Figure 9. Probability distribution of the overflow volumes for the sewer system of Geel-Larum using the rainfall series for Uccle from 1967 to 1993
CONCLUSIONS
Hydrodynamic simulations with single storm events will often not lead to an accurate probability estimation of the overflow events. This is especially true for non-linear systems and overflows with a high overflow frequency. Ideally, for emission predictions water quality models or even immision models for the receiving waters should be used. However, it makes no sense to use a sophisticated water quality model in combination with single storm events when the probability of these events can not be accurately predicted. Because of the variability of the rainfall. it is very important that the statistics for the quantitative overflow parameters are maintained. Thus. long term simulations with simplified models seem to be the optimal choice between calculation time and accuracy of the calculation results. When a higher accuracy of the emission predictions is required hydrodynamic simulations with short selected rainfall series can be used. These short rainfall series have to be selected based on a simplified model. This approach leads to a very high accuracy. but the calculation time will also increase. REFERENCES Vaes. G. (I 996a) . Multi-lin~ar reservoir model for emission calculations. 9 1h European Junior Workshop on impact of urban runoff on waste water treatment plants and receiving waters. Kilve, UK. Vaes, G. (l996b). Remuli : reservoir model with mult! linear throughflow relationshtp, Hydraulics Laboratory, University of Leuven, Belgium. Vaes. G. and Berlamont.l. (1996). Short rainfall series as input for emlssio« calculations, 7 th International conference on urban storm drainage, Hannover, Germany. Vaes. G. and Berlamont. J. (1997). Characterization of sewer rystems with storage/througflow-relationships, conference on Mathematical Modelling in Environmental Pollution. Madrid, Spain. Vael, G., Willems, P . and Berlamont, J. (1996). IDF·relationships and composite storms for the design of s~wer systems (in Dutch), Hydraulici Laboratory. University of Leuven, Belgium. Vlaamse Milieumaalschappij (1996) . Flemish guidelines for an integrated managemen: in urban drainage (in Dutch), V1aamse Milieumaatschappij. Erembodegem, Belgium.
Perganlon
Wal. Sci. T~clt. Vol. 38. No. 10. pp. 41-48.1998.
IAWQ C 1998 Published byElsevier Science LId. Printed in Great Britain. Allrights reserved
PI!: S0273-1223(98)OO731-8
0273-1223198 $19'00 + ()OOO
MODELLING OF OVERFLOW EMISSIONS IN FLANDERS G. Vaes and J. Berlamont Hydraulics Laboratory, University ofLeuven; de Croylaan 2. B-3001 Heverlee, Belg ium
ABSTRACT Ideally,for emissioncalculations long term hydrodynamic simulations should be performed. but this requires long calculationtimes. Simplifications are consequently necessary. Due to the non-linearbehaviourof sewer systems. hydrodynamic simulations using single storm evenIS often will not lead 10 a good probability estimation of the overflow emissions. Simplified models using long time simulations give better results if they are well calibrated. To increase the accuracy hydrodynamic simulations with short time series can be used. The short time series are selectedfrom the long time historical rainfallseries using a simplifiedmodel. To test the accuracyof these three methods,hydrodynamic long tenn simulations wereperformedfor several (small) sewer systems with differentcharacteristics to comparewith. ~ 1998Publishedby Elsevier Science Ltd. All rights reserved
KEYWORDS Overflow emissions; probability estimation; reservoir model; single storm event; short time series. INTRODUCTION Since more and more trunk sewers and treatment plants are built in Flanders. the water quality in the receiving waters is steadily improving. A lot of investments are in progress to improve the situation further. There still is a long way to go, but in the future the water quality in the Flemish brooks and rivers will (hopefully) be good again. Most of the sewer systems in Flanders are combined sewer systems. where rain water is transported together with sewage in the same pipes. Because treatment plants can not be designed for very high flow rates. combined sewer overflows are built in into the sewer system. Because of the accidental occurrence of overflow events. these events will have a large effect on the water quality and consequently on the biological life in the receiving water. A sudden disturbance of a good water quality will be more noticed than a continuous moderate water quality and will possibly also create more damage to the aquatic life (e.g, sudden fish kill). Not only the physical reality is important for this, but also the psychological perception of it. For this reason it is important to know the emissions at the overflows and their effect on the receiving water. It is obvious that it is impossible to measure the emissions on the thousands of overflow structures in Flanders . Besides. one wants to predict the emissions in advance to be able to optimize the design and one wants to know the long term performance of the overflows. The only way to assess the emissions is thus to model the sewer systems numerically. Besides the emissions, also other parameters influence the water quality in the receiving water, such as the characteristics of the 41
42
G. VAES and J. BERLAMONT
receiving water itself, the direct runoff, the effluent discharges of the treatment plant, etc... Even if we can estimate the water quality in the receiving water, we still do not know the effect on the biological life in the water. The assessment of overflow emissions is thus only one step in the whole process of integrated modelling, but a very important one. EMISSION MODELLING The ideal emission calculation is a detailed hydrodynamic routing (full implementation of the 'de SaintVenant' equations) using long time historical rainfall series and performing a statistical analysis on the emissions afterwards. This kind of simulations is well-known, but very time consuming. Therefore simplifications are necessary. Besides, also sediment transport and water quality aspects should be incorporated. However, the physics of these water quality phenomena are still not completely understood and water quality calculations are even more time consuming. Moreover, more extensive and accurate data are necessary for water quality simulations and more extensive calibration is required. The more calibration is needed the more unreliable models become; not everyone will calibrate a model in the same (right) way. Besides, how reliable are the data the model is calibrated with? This does not mean that detailed water quality models are useless; on the contrary: they can make the behaviour of the systems more comprehensible and give more insight in the processes which are going on in the sewers. When one proceeds to integrated and stochastic modelling the calculation time requirements become even more stringent; e.g. performing Monte Carlo simulations with a detailed integrated model using several decades of rainfall input would take years/decades to run for only one sewer system. Besides reducing the system to the principal aspects, gives a better view on the behaviour of the system. Too detailed models give a false feeling of accuracy, because of the uncertainty on the input data and the model parameters. These difficulties in emission modelling direct modellers towards a pragmatic solution. Somewhere there is an optimum between the accuracy of the model results and the efforts that have to be made to reach these results. This optimum can be different according to the circumstances and influenced by climate, land use, topography. type of sewer system, etc... One of the major principles in modelling is that all model parts must have the appropriate accuracy, taking into account the sensitivity for the parameters, to have an optimal model. This means that a detailed routing model for the flow in the sewers in combination with a very rough surface runoff model for example, will not be an optimal model for emission calculations. For these reasons different types of simplifications were investigated for the use in Flanders. SIMPLIFICATIONS To reduce the calculation times one can either simplify the rainfall input or the sewer system model. However, emission calculations are most sensitive to rainfall input. If one tries to simplify the rainfall into single storm events. the amount of information will decrease strongly and the accuracy of emission predictions using this approach will often decrease likewise. This is especially true for non-linear systems. Sewer systems behave in a non-linear way when they have no gravitary outflow (pumps, vortex controls), when they are looped, when the pipes get pressurized or when there is a backwater effect. This is often the case in flat regions as in Flanders. For non-linear systems the frequency of the overflow event will not necessarily equal the frequency of the rainfall event which leads to this overflow event. In that sense, emission calculations using single storm events will not give an accurate estimation of the probability of an overflow event. This effect increases as the frequency of the event and the non-linearity increase. This means that for design purposes one can use single storm events with a high return period without any problem. For these reasons it is often better to use simplified (conceptual) models for emission predictions than simplified rainfall. The statistical information about overflow events is much more important than the accuracy of the individual emission predictions. Detailed simulations with single storm events will make the behaviour of the system more comprehensible, but will never allow us to assess the emission probabilities. The variability of the rainfall in Flanders is so high that water quantities dominate the water quality. For this reason it is not necessary to incorporate very detailed water quality aspects, certainly not when a conceptual
Overflowemissions
43
model is used for the water quantities. Besides, water quality simulations are strongly influenced by local and time varying (stochastic) conditions, which makes it mostly impossible to model it in detail in an economical way. The use of constant pollution concentrations in a first stage or a very simplified transport model is thus acceptable; they are giving often as good (as bad) results as very detailed water quality models. MULTI STAGE APPROACH IN FLANDERS Because of the important investments in sewer systems and overflow constructions which have to be made during the coming years in Flanders, a considerable number of emission calculations has to be performed. Practically manageable computer tools for sewer system calculations are only available since about one decade; whilst only since the last few years commercially available computers are fast enough to perform realistic emission calculations. In Flanders, this has led to an approach in several stages when designing overflows, which is implemented in the new guidelines (Vlaamse Milieumaatschappij, 1996). The first stage in assessing the emissions of the overflow exists in performing hydrodynamic simulations using high frequency composite storms (Vaes et al., 1996). These are single storm events, one for each frequency or return period, which include the mean rainfall volumes for every duration between 10 and 720 minutes. For storm durations up to 240 minutes also the mean antecedent and posteriori rainfall is included. The IntensitylDurationlFrequency-relationships (IOF-relationships) which are used, are based on historical time series of 27 years from Uccle, which is approximately representative for the whole of Flanders (Vaes et al., 1996). For linear systems and overflows with a low overflow frequency this approach is acceptable in order to estimate the probability of the overflow events. For sensitive receiving waters (with regard to the effect of overflows) this first stage is not considered sufficient . To obtain a higher accuracy one has to move to another stage. For the second stage in assessing the emissions of the overflow a conceptual (reservoir) model of the sewer system is used. For this reservoir model the calibration is of major importance in order to obtain good results. For this calibration hydrodynamic single storm simulations can be used. The major parameters are the "storage" and the "throughflow" . Not only the maximum values for these parameters are important, also the variation of the throughflow with the storage is significant information. Graphs have been determined to perform a quick estimation of overflow frequency and mean overflow volume in function of the maximum storage and the maximum throughflow capacity for two extreme cases: a constant throughflow and a throughflow which varies linearly with the storage (Vlaamse Milieumaatschappij, 1996). For the third stage dynamic simulations are performed using short time series, which are selected from the long historical rainfall series, using a reservoir model (Vaes and Berlamont, 1996). When desired, the second stage (reservoir model) can be skipped and a dynamic simulation with a general set of short time series can be performed. For this general approach a set of short time series has been selected which include all rainfall events which will lead to an overflow event for a wide range of (Flemish) sewer systems. This reduces the rainfall input already with more than 90% (Vaes and Berlamont, 1996). These three stages in the emission assessment include water quantity aspects only. It is however possible to include a simple pollution transport model in the reservoir model. Ideally, it is necessary to include also other aspects than the emissions; an integrated water quality model of the river basin is then required. For this the rules for simplifications are quite similar as for emission predictions. Mostly, detailed water quality modelling is not necessary. because for the most sensitive brooks and rivers in Flanders there may not be an overflow structure at all. Also for discharges to receiving waters with lower standards, it may be more effective to choose the location of the overflow structure in function of the environmental effect than to limit the overflow events . Because of the high uncertainties on many input and model parameters, one will have to switch in the future to stochastic models instead of deterministic models. Assuming distributions for the most uncertain input and model parameters and performing Monte-Carlo simulations. will enable one to estimate the accuracy of
G. VAES and J. BERLAMONT
44
the results. But, this will lead to much longer calculations times, so that simplified models are certainly required. SINGLE STORM EVENTS
When one assumes that the probability of the overflow event is the same as the probability of the rainfall event which causes it, one can determine the frequency distribution for the overflow parameters using single storm events. The first requirement for this is that the probability of the rainfall events is well estimated. The composite storms which are now used in Flanders are based on IDF-relationships, but IDF-relationships can be determined in many different ways. The rainfall time step, the aggregation method for higher durations, the criteria to remove dependent events, the definition of an event and the extreme values method used for the higher return periods can influence the results significantly. Therefore IDF-relationships should be developed in function of their usage (Vaes et al., 1996). overflow discharge (m3/s) -long time series
0.8
-- composite storms
0.6 0.4
0.21_~~~_~=::::====::::==::i;;;;;:::=:::;::::::=_--l o o
2
4
6
8
10
12
frequency (#/year) Figure I. frequency distribution forthe twooverflows (. and +) in the sewersystemof Geel-Larum,
overflow volume (m3)
3S000
dynamic simulation + composite storms
30000
reservoir model + composite storms
-
2S000 20000
-reservoir model + long time series
ISOOO 10000 SOO:
l---'-~::::::~:;;;~=~==:::~===;=="i""'--'---ro
S
10
IS
20
2S
30
3S
40
4S
SO
frequency (#/year)
Figure 2. Probability of overflow volumes fromthe sewersystem of Dessel,
To check the consequences of the assumption of linearity between rainfall event and overflow event, long time hydrodynamic simulations (27 years) have been performed and the results were compared with single storm simulations. Several sewer systems were used for this with very different characteristics : linear as well as very non-linear systems and overflows with low as well as high overflow frequencies. This
Overflow emissions
45
comparison showed that the higher the overflow frequency and the higher the non-linearity. the less accurate the probability estimation is for the overflow events. In figure I the overflow discharge frequency distributions are shown for the long time series and the composite storms for two overflows with different overflow frequencies. In figure 2 the overflow volume frequency distributions are shown for one overflow, obtained with composite storms in combination with a dynamic simulation as well as with a reservoir model. Also the results of the reservoir model using the long time rainfall series are shown : the influence of the rainfall data used is much more important than the influence of the sewer system model. RESERVOIR MODELS In a flat region like Flanders often a (static) reservoir model (Figure 3) can lead to quite accurate emission predictions (Vaes, 1996a; Vaes and Berlamont, 1997) . This means that only a continuity equation and a simple throughflow equation have to be solved. The throughflow equation for the reservoir model must be well calibrated. Often a constant throughflow or a linear relat ionship between the throughflow and the storage in the system is assumed. In many real cases this relationship is much more complex (Figure 4). However it is very important to model this relationship as correctly as possible, because it affects the accuracy of the emission predictions strongly. The relation between storage and throughflow up to the moment the overflow starts spilling is used (Figure 6), as the storage at the beginning of the overflow event is a good estimation for the mean storage in the system (Figure 5). Only some extra "dynamic" storage is not incorporated for more severe storms. but this will hardly influence the overflow volume.
o
.
\\\\\\\ \\\\\\\taJn \\\\\\\
\\\ \ \ \\
evaporation
depression storage ~overflow
L
storage infiltration
in the sewer system
throughflow to the
treatment - . . plant Figure 3. schematicrepresentation of a reservoir model.
throughflow (mm/h)
10
8 6
-e-f= 1 -o-f=3
--f=2 _ . f=4 -~
4
--f=5 --f=7
f=6 --f=8
2
o
o
2
4
6
8
10
12 storage (rom)
Figure4. Storage/throughflow-relationship forthe second spilling overflow in the sewersystemof Geel-Larum for a rangeof composite storms withdifferent frequencies f.
46
G. VAES and J. BERLAMONT
12
storage in the sewer system (mm)
10 maXimum
i
8
beginning
6
4 2
end
~~BflII~HIIIHHIIHIHHHHH
! I II I I
0+---+---+---+--+--+--+--+--+--+--+--+----1
o
1
2
3
5
4
6 7 8 9 10 11 12 maximum overflow discharge (mmlh)
Figure S. Varialion of the storageduringthe overflowevent in two sewersystems(. and .) for a wide range of rainfallevents(composite stormsas wellas constantintensities overdifferentdurations).
throughflow Q (lis)
200 175 150 125 100 75 50 25
• real relationship -
'static' approach
.QI. . . .. . . . . . . . .. . . . . .••• .. - •••
• ••
••
.
•• ••
••••
••••
o
o
20
40
60
80
100
120
140
Figure6. Bi-Iinear'static'approach for the storage/throughflow-relationship of the sewersystemof Bilzen-Tabaartwijk.
A linear relationship is easy to implement and leads to a very fast calculation (104 to 106 times faster than a hydrodynamic simulation). To keep the advantages of a linear model and still include more complex relationships, a piecewise linear relationship can be implemented (Figure 6) (Vaes, 1996b). This is easy to fit to any existing relationship and a linear model is obtained by a coordinate transformation. When the simulations with a hydrodynamic model show a lot of hysteresis in the storage/throughflowrelationship , the "dynamic" storage can be taken into account as a piecewise linear relationship of the inflow (Figure 7). To calibrate this a few more hydrodynamic simulations with single storm events for different return periods (different inflow rates) are necessary. Other important parameters are the concentration time over which the inflow is averaged and the runoff model used. When the dynamic storage is incorporated. it can be useful to vary the concentration time with the inflow so that high inflows are less averaged than low
Overflow emissions
47
inflows and thus reach the overflow faster. The runoff model can have a large influence on the emission results, but it is often very difficult to calibrate.
outflow QOUI (m3/s) V stal = kstal QOUI
200
175 150 125 100
• real relationship
75 -- dynamic storage Vdyn - static storage VSlat 50 -- total storage Viol 25 0 ......~--+---~---+----+----4----4-----+-o 20 40 80 100 120 60 140 Figure7. 'Dynamic' approachfor the storageahroughflow-relationship of the sewersystemof Bilzen-Tabaanwijk.
A comparison of the results of reservoir models for different sewer systems with the above mentioned long term hydrodynamic simulations shows that the emission predictions are very good if the storage/throughflow. relationship is well calibrated. Although the error on some individual emission events can be significant (Figure 8), the probability distributions of the overflow parameters are well predicted (Figure 9). overflow discharge (m3/s)
•
0.6
0.5
- - dynamic simulation •• - bi-linealr 'static' reservoir model
0.4 0.3
0.2 0.1
o
5
10
15
20
25
30
35
40
time (hours with overflow) Figure8. Compressed timeseriesfor the overflow discharges fromthe sewersystemof Gecl-Larum usingthe rainfallserieso( Uccle(or 1967.
48
G. VAES and J. BERLAMONT
overflow volume (mm)
40 35 30
dynamic simulation
25
hi-linear 'static'reservoir model
20
15 10 5
o 4-+--t--+--+-~:::::+:=+~~==-t""=--+--1
o
1
2
3
4
5
6
7
8
9
10
11
12
13
overflow frequency (#/year)
Figure 9. Probability distribution of the overflow volumes for the sewer system of Geel-Larum using the rainfall series for Uccle from 1967 to 1993
CONCLUSIONS
Hydrodynamic simulations with single storm events will often not lead to an accurate probability estimation of the overflow events. This is especially true for non-linear systems and overflows with a high overflow frequency. Ideally, for emission predictions water quality models or even immision models for the receiving waters should be used. However, it makes no sense to use a sophisticated water quality model in combination with single storm events when the probability of these events can not be accurately predicted. Because of the variability of the rainfall. it is very important that the statistics for the quantitative overflow parameters are maintained. Thus. long term simulations with simplified models seem to be the optimal choice between calculation time and accuracy of the calculation results. When a higher accuracy of the emission predictions is required hydrodynamic simulations with short selected rainfall series can be used. These short rainfall series have to be selected based on a simplified model. This approach leads to a very high accuracy. but the calculation time will also increase. REFERENCES Vaes. G. (I 996a) . Multi-lin~ar reservoir model for emission calculations. 9 1h European Junior Workshop on impact of urban runoff on waste water treatment plants and receiving waters. Kilve, UK. Vaes, G. (l996b). Remuli : reservoir model with mult! linear throughflow relationshtp, Hydraulics Laboratory, University of Leuven, Belgium. Vaes. G. and Berlamont.l. (1996). Short rainfall series as input for emlssio« calculations, 7 th International conference on urban storm drainage, Hannover, Germany. Vaes. G. and Berlamont. J. (1997). Characterization of sewer rystems with storage/througflow-relationships, conference on Mathematical Modelling in Environmental Pollution. Madrid, Spain. Vael, G., Willems, P . and Berlamont, J. (1996). IDF·relationships and composite storms for the design of s~wer systems (in Dutch), Hydraulici Laboratory. University of Leuven, Belgium. Vlaamse Milieumaalschappij (1996) . Flemish guidelines for an integrated managemen: in urban drainage (in Dutch), V1aamse Milieumaatschappij. Erembodegem, Belgium.