9

Click here to load reader

Storm water quality modelling, an ambitious objective?

Embed Size (px)

Citation preview

Page 1: Storm water quality modelling, an ambitious objective?

E) Pergamon

PH: S0273-1223(97)00771-3

War. Sci Tech. Vol. 37. No. I. pp. 2OS-213.1998.C 1998 IAwQ. Published by Elsevier Sciencelid

Printed in Great Britain.0273-[223198 $[9-00 +0'00

STORM WATER QUALITY MODELLING,AN AMBITIOUS OBJECTIVE?

M. Ahyerre, G. Chebbo, B. Tassin and E. Gaume

Centre d'Ense ignement et de Recherche pour la Gestion des Ressources Naturelles etde l'Environnement, Ecole Nationale des Ponts et Chaussees , 6 et 8 Avenue BlaisePascal, Cite Descartes - Champs-sur-Marne, F-77455 Mame-La-Yallee Cedex 2,France

ABSTRACf

As a consequence of the awareness of the pollution impact of storm sewer overflows, managers need tools toevaluate and control stormwaters according to water quality criteria. After an experience of 25 years in stormwater quality modelling, very few models are widely and regularly used. According to managers this is dueto their cost and their low level of accuracy.

The generation and the transport of the pollution in urban systems during a storm event are very complexbecause they concern many media and many space and time scales. Nevertheless. a typology of the existingmodels shows that this complexity has been inscribed into the models. This tendency towards complexitymakes sewer quality models difficult to put into operation and three main difficulties can be underlined :doubtful mathemat ical formulation of processes, uncertainties on input and calibration data, difficulties andcost of calibration.

Further research is needed to improve the modelling approach and basic knowledge. and we think that a cleardistinction should be made between management tools and research models. ~ 1998 IAWQ. Published byElsevier Science Ltd

KEYWORDS

Sewer flow quality modelling ; urban pollution management; model calibration; sampling uncertainties.

INTRODUcrION

Since the late sixties, an important number of research programs (National Urban Runoff Program, in theUSA (1978-1983) , French campaign (1980-1982), ...) have shown the importance of storm water pollutionand its impact on ecosystems and on the different uses of water. As a consequence, many managers haveincluded the problem of storm water pollution in their procedures.

Managers need tools to evaluate and control storm waters according to quantitative and qualitative criteria .In order to help stormwater quantitative management, researchers have built models that simulate the courseof a raindrop within a catchment, entering the drainage network and finally reaching the point of outletToday, those models are efficient and commonly used by managers. Concerning stormwater qualitymanagement, researchers reasoned in the same way and built complex models whose structure correspondsto the course of the pollution. Many computer models have been proposed since 1971 (first version ofSWMM by the US·EPA). The evolution of those models raises two main questions:

205

Page 2: Storm water quality modelling, an ambitious objective?

206 M. AHYERRE et al.

Do such quality models meet managers' needs 7How much confidence can we have in those models?

The aim of this article is to answer these two questions. Firstly, we will describe the manager's point of viewconcerning storm water quality models. Then, through an analysis of the complexity of the system, we willexamine how it has been described in the existing models. We will finally underline the present difficultiesof storm water quality modelling.

THE MANAGERS' POINT OF VIEW

A recent enquiry (Bailly, 1996) gives indications about the motivations leading to new storm water projectsin France. Based on the study of 63 projects between 1989-1996, the most often expressed motivation wasthe protection of the environment (84% of the projects), while limitation of flooding concerned only 8% ofthe projects. Since sewer overflows during storm events have various types of impacts on receiving waters ­shock, cumulative, stress effects, each with different representative time scales - managers need specifictools to make diagnoses and investigate a certain number of upgradings in order to limit the mass and thefrequency of storm sewer overflows at the point of outlet (see Table I).

Table I. Comparison of the nature and location of the impact of storm sewer flows according tothe type of management

Impact location

Quantitative management the whole catchment

Qualitative management the overflow point

Impact caracterisation

volume of water overflowed

mass of pollution thrown outfrequency of overflows

In this context, many computer models have been proposed since 197I. After an experience of 25 years,very few water quality models are widely and regularly used in France. The same study (Bailly, 1996)showed that less than 20% of the projects concerning the management of storm sewer pollution used a waterquality model. Managers often have a subjective answer to explain this situation. Storm water qualitymodels appear not to be cost effective, because of the cost of the calibration campaigns, especially if oneconsiders their level of accuracy compared to hydraulic models.

THE URBAN SYSTEM'S COMPLEXITY AND ITS INSCRIPTION INMODELS

The &enerat;on and the transport of the pollutjon in the system

Phenomena leading to storm sewer pollution involve various media, various time and space scales andvarious pollutants .

The atmosphere, the watershed and finally the sewer system are successively passed through during thecourse of a raindrop. Physical, physico-chemical and biological phenomena occur at each stage of theprocess. The structure and the behaviour of the physical system itself are very complex. For example, sewersystems are composed of various elements such as pipes and special works with different physicalcharacteristics (slopes, diameters), interacting with each other.

Space scales vary greatly if one considers the watershed (from some hectares to thousands of hectares), thepipes of the sewer system (from tens to hundreds of metres), the special works (from hundreds to thousandsof cubic metres) and the particles (from tens of micros to some centimetres).

Page 3: Storm water quality modelling, an ambitious objective?

Stonn water quality modelling 207

Concurrently time scales vary from several days. that correspond to the dry weather periods. to someminutes to characterize the rain or even less if one considers the kinetics of heavy metals (exchangebetween particulate and dissolved matter).

Moreover. many variables are used to describe the pollution. Suspended solids are the major vector ofpollutants, but other parameters such as COD, BOD, heavy metals, hydrocarbons and bacteria types andconcentrations are of a great interest in describing the impact on receiving waters.

~polo&y of the models

The general structure of storm water quality models is not very different from one model to another, it iscopied from the course of the pollution on the urban system. It is possible to define a typology based on themodelling of the two sub-systems : the water cycle and water quality (see Figure I).

Models are spaced all over the table. this shows the variety of the current situation. We emphasize that mostactual models belong to the same category (hydrological-hydrodynamic model). The present tendency istowards making the models more complex.

FLOW MODELING

Hydrological models Hydrological-Hydrodynamic models

MOUSE·TRAPDHI.94

CONVEC I

HYPOCRASB-KraJewski.92

Constantconcentration

Solid transport

Q AccumulationU on theA watershedL --======~ _ITY

Figure I. Typology of existing softwares .

It is often assumed that the observed inaccuracy of oldest models can be reduced by making them morecomplex, despite the fact that data concerning the phenomena are rarely available. Till now. many stormsewer models were directly produced in research centres with an academic culture and with the purpose oftaking each phenomenon into account. So. those models should be considered more as research tools thanmanagement tools.

THE PRESENT DIFFICULTIES

In general, complex models are regarded as better or more accurate than simple ones. We are now going toanalyse the consequences of this tendency towards complexity and see how difficult it is to put complexsewer quality models into operation. Three difficulties can be underlined: a sparse knowledge concerning theprocesses involved, the uncertainties and the lack of data, and difficulty in calibration.

Page 4: Storm water quality modelling, an ambitious objective?

208 M. AHYERRB et al.

A sparse knowled~e about the processes inyolyed

The dynamics of accumulation, mobilization and transport of pollutants during wet weather is not wellknown and knowledge is often sparse, especially concerning processes in the sewer. For instance, manyresearch programs have shown that the contribution of sediments to combined sewer overflows is great.

Indeed, according to results of mass balances (Chebbo, 1992, Bachoc, 1992, Krejci et al., 1987), 30 to 50%of suspended solids come from in-sewer sources. But these authors also show that the particles sampled atthe outlet of the sewers are fine and organic while in-sewer sediments are mainly coarse and inorganic. So,we have to look for some other sources than the sediments to explain the in-sewer contribution to combinedsewer overflows. Different sources and mechanisms can be suggested:

hydraulic singularities that allow fine particles to settle (Bachoc, 1992),the bacterial film that grows during dry weather in combined systems and that can be pulled upduring wet weather (Krejci et al., 1987) is nearly never taken into consideration,a highly concentrated largely organic 'layer' in combined sewers identified by numbers of researcherscould be resuspended during storms (Ashley et al., 1994).

Nevertheless, the relative contribution of each of these sources has to be assessed and validated, as existingresults exhibit broad differences from one sewer system to another (Krejci et al., 1987, Ristenpart et al.,1995). These examples point out the lack of knowledge on the sources of pollution and processes. Thus,models can only be a poor description of the real behaviour of a watershed during a rain event.

Furthermore, mathematical formulation for in-sewer solid transport quality models. are derived from alluvialhydrodynamics and do not take into account the peculiar phenomena described above (presence of a biofilm,of a concentrated layer, of local pollution sources). Current solid transport models used are Acker White'smodel, Velikanov's model (based on a turbulent energy analysis), and Macke's model (based on a 'nodeposition' criterion). In those models, the solids are described inexactly as far as physical characteristics ortheir settling velocities are concerned. All above mentioned equations were established and are valid incontinuous flow. Nevertheless, it is clear that during a rain event the flow is transitory. In rivers, one candiscount the effect of the walls which is not the case in sewer systems (Perrusqufa, 1992). Furthermore,depending on the shape of the pipes rapid variations of the hydraulic radius may occur particularly in thecase where a walkway exists (Verbanck, 1996). This causes discontinuities in the processes which are onlypoorly taken into account in the models.

So, there is a discrepancy between the sparse knowledge concerning phenomena and mathematicalformulation.

Uncertainties and Jack of data

Inputdata. Efficiency of a model is completely dependent on the quality of the input data. As modelers say,'Garbage in, Garbage out'. If the data is insufficient or of poor quality, then, the reliability of the tool isweak, and the results are unusable.

Input data can be divided in two categories : initial conditions (geometry of the system, initial pollutionloads) and boundary conditions (rain, dry weather flow).

Concerning the initial conditions the following remarks can be made. Information on the specificities of thevarious parts of the watershed (land use, surface slopes, ...) is needed, in order to know their contribution tothe runoff. These data are uncertain, especially for large watersheds. Moreover, if the horizontalcharacteristics of the sewer system are generally well known, the vertical ones which affect strongly thehydraulic behaviour of the water are often different from the projected ones.

Page 5: Storm water quality modelling, an ambitious objective?

Stenn water quality modelling 209

Finally, the description of the initial pollution loads in the network and on the watersheds, and of itscharacteristics is impossible at a lower cost Furthermore. sediment sewer deposits are prone not only tospacial but also to temporal variations (Jack et al., 1996). This indicates that all sediment data should beobtained from data collected at the same time, which is impossible.

The most important boundary condition that has to be taken into account is the rain. Most rain events withan environmental impact are convective ones characterised by an important variability in time and space.

Those events are only well measured through dense rain gauge networks or radar data which are seldomavailable. In addition, using current equipments, rain intensity measurement errors is in the range of20% (Bertrand-Krajewski, 1991).

Calibration data. Calibration data are usually time series of flow and pollutants concentrations in the sewersystem and at least at its outlet. Concerning flow, depending on the measuring method used, uncertainty is inthe range of 5 to 25% (Maksimovic et al., 1986). Considering the quality parameters, uncertainty iscumulated at each stage of the measurement: sampling (problems of representativeness), conservation andanalysis. A first cause of uncertainty comes from the representativeness of the samples due to spatial andtemporal variability of the pollution parameters . This can be illustrated by a successive (each 15 seconds)sampling at different depths in the 'VieiIle du Temple' pipe in 'Le Marais' catchment, Paris (Ahyerre, 1996).The obtained profile (see Figure 2) shows the heterogeneity of the vertical concentrations in the sewer.

2S

20

~ lS

~'u 10:t

5

800o+--~;:=====------<

o 200 400 600

Suspended Solids (mgll)

Figure 2. Multi-depth suspendedsolids 'Le Marais', Vieille du Temple Pipe (20lO4I96).

A second source of uncertainty comes from the sampling procedures. In 'Le Marais' catchment, during wetweather. the difference between the concentration of samples taken with two different samplers workingsimultaneously (Buhler PP92 and Vergamon 94, that have a same sucking velocity of 0.8 mls). was around15% (Ahyerre, 1996).

The last source of uncertainty is the analysis technique, in the laboratory. For instance, the repetitiveness isaround 10% for the suspended solids and COD and 30% for the BOD (Saad et al., 1996). Furthermore, theanalytical techniques may vary from one laboratory to another. Differences of around 40% for themeasurement of suspended solids on a single sample between different laboratories have been reported(Paitry et al..• 1986). Thus, available data for calibration of models are uncertain and it makes calibration aparticular problem.

Calibration of models

First, we will see that simulation results are highly dependent on the value of the parameters chosen. whichmakes calibration indispensable. Then, we will highlight its difficulty and its cost.

JWST J7••-H

Page 6: Storm water quality modelling, an ambitious objective?

210 M. AHYERRE et al.

The importance of calibration. In general the information in the models on the behaviour of the system ispoor. Some models have to be considered as 'tool boxes' in which one can choose between different types ofbehaviours of the system.

This is for instance the case of the well known SWMM model (Huber et al., 198I). because it allows theuser to choose between various mathematical formulations for the build-up and transfer of pollutant loads.but also because the values of the parameters used in these formulations are not weJIdefined. Let us considerthe exponential pollutant wash off formulation proposed in the model (equations I and 2).

C(t)=R q(t)Wa.f~",,-I.DD(t) (I)(.'o~r

DD(t+!:.t) = DD(t) - C(t) . q(t) . t:.t (2)

Crt): water suspended concentration downstream the watershed,q(t): water discharge downstream the watershed.t:.r: computation time step,DD(t): dust and dirt load still available on the surface of the watershed,Rcoef ' Washpo: parameters.

If Washpo =I and Rcoef =0.77 mm-I (values proposed in the SWMM user's manual). 90% of the pollutantload available at the beginning of a rain event on the surface of the watershed is washed off with 13 mm ofrain. and a kind of 'first-flush' effect will be reproduced. On the other hand, if Rcoef =0.005 mrrr l which isnot a an absurd value, 460 mm of rain will be necessary to wash off 90% of the poJIutant stock. In this casethe pollutant concentration will be almost constant if Washpo =I, or proportional to the discharge ifWashpo = 2 during most rain events. This shows that a great diversity of behaviours can be taken intoaccount with such a model depending on the values of the parameter, and the calibration on the site ofinterest is therefore a major phase of its implementation.

The difficulty ofcalibration. The small amount of data usually available for calibration makes it a particularproblem and even in the case of very simple models, calibration can fail : similarly good fit betweenrecorded data and simulated results can be obtained with quite different vectors of model parameters. Inother words, a great uncertainty in the value of the adequate parameters can exist.

Table 3. Automatic calibration of the SWMM3 model, results of various trials

Model Trial I Trail 2 Trial 3 Trial 4 Trial 5coef

Initial Final Initial Final Initial Final Initial Final Initial Final

ddlim 10 66 60 95 10 120 10 50 10 39

ddpow 0.8 0.42 0.6 0.52 0.4 0.8 0.4 0.4 0.8 0.78

washpo 2.5 1.6 1.5 1.54 0.5 2.441 0.5 1.68 2.5 13.78

Rcoef 0.5 0.019 0.014 0.014 0.001 0.001 0.1 0.026 0.05 0.03

criteria 409921 430201 759239 429347 459793

Several trials run with a very simple version of SWMM (model with 4 parameters. the two parameterspreviously described for the wash off formulation, and two more parameters for the pollutant build upformulation during dry weather periods) illustrate this fact. The data used to calibrate the model have beenmeasured in Quebec City's sewer system during the 1982 summer period (Lavallee et al., 1984). The dataavailable for calibration represent a total amount of 80 concomitant measurements of flow and suspendedsolid concentration concerning five heavy rain events. Several model calibration trials have been run. Thesimple least squares has been selected as criterion. and the Powell Method (Press et al., 1988) has been usedfor its optimisation. The results are summarised in Table 3.

Page 7: Storm water quality modelling, an ambitious objective?

Stonn water quality modelling 211

The simulation results obtained in trials 2 and 4 are presented in Figure 3. The standard error for the five rainevents is about 40 rng/l, Various sets of parameters give acceptable results and it seems difficult to identifythe 'best' set.

400 .,.----- - - - - - - - - - - - - - - - - - - -r50

109.5.4.3. 6. 7.time (hours)

Figure 3. Example of stimulation result (SWMM), june 16th 1982, Quebec City .

OlJl.--lll---.JlL= ----=_ ..:J1L _ ..:::== =i.O2

c libration results ot trial 2 40300

calibration results 01trial 4

~..measured data 30 £

~ z-£200 . . . . . . . . . . . . . . . . . hyetograph 'iii

cen CIIen 20 .S

c.~

10010

Finally, let us notice that, as in many implementations of storm water quality models, we had not enoughdata to save some of them for a validation phase. The good fit between measured and simulated dataobtained during calibration does not say much about the predictive power of the model : its capacity toreproduce well the pollutographs which could be measured during new rain events. As a conclusion we cansay that we had not enough data to implement the model correctly on the Quebec City's sewer system.

CONCLUSION

Storm water quality models are not widely and regularly used by managers. The generation and the transportof the pollution are very complex phenomena and the complexity of the system has been inscribed in themodels. This trend has led to various difficulties :

a discrepancy between the knowledge and the equations of the models,an uncertainty on the input and calibration data,a difficulty in calibration.

Further research is needed to improve modelling approach and basic knowledge . Concerning the modellingapproach. we think that a clear distinction should be made between management tools (that could be verysimple) and research models .

For management tools, we think that, following an approach widely used in hydrology, simple models withjust few parameters are interesting. Studies on various places (Chebbo, 92) show that suspended solidcharacterization is very similar for the same kind of networks and rain characteristics. In such a case, itseems that the sewer system acts like a filter which reduces the variability and the heterogeneity of theoutput variables. If the 'filter action' of the city can be verified, models with few parameters could be used ina predictive mode with a significant reliability . This hypothesis has to be examined on the basis of the resultsof various measurement campaigns. Concurrently. it would be interesting to put together all the measured

Page 8: Storm water quality modelling, an ambitious objective?

212 M. AHYERRE et al.

data available in Europe in a single database to compare the different results obtained on various sites andmaybe try to generalise some of them. The French database QASTOR (Saget, 1994) could be the frameworkfor such a project. Indeed, the French database QASTOR with all the French data available since 1970gathers in a single tool data on pollution discharged during wet weather at the outlet of storm sewers orcombined sewers. It has been enriched with other European data (Dutch and German) and this effort shouldgoon.

To improve the knowledge on the processes, it also seems important to make an effort on the quality of thedata (precision, frequency of the samplings, and knowledge of the uncertainties at each stage of themeasurement procedures). Since collecting data is very expensive, it seems that experimental watershedsshould not be too numerous, but intensely fitted out. Research models have to be developed for thesewatersheds, acting as 'numerical laboratories', that would test various conceptual descriptions of the samesystem.

As an example, a research program, co-ordinated by the CERGRENE since 1994, on a small urbancatchment of the district 'Le Marais' in Paris aims to fit out an experimental catchment, allowing wet weatherpollution to be studied at different levels of the water cycle (Gromaire et al., 1997). This program is going tobe included in a research program which begins in Paris on 'nested' watersheds with surfaces ranging fromsome tens to some thousands of hectares. This new program will be focused on the study of the dynamics ofthe pollutants of wash off, the relation between upstream and downstream pollution characteristics and theimpact of the structure and the dimension of the sewer network. Such programs should give the modeler agood deal of good quality data.

REFERENCES

Ahyerre , M. (1996). Analyse des profits verticaux de matieres en suspension par temps sec dans Ie bassin versant urbain duMarais . DiplOmed'etudes apprcfondies, Univers ite Paris XU. Paris. France .

Ashley . R. M., Arthur ,S., Coghlan. B. and McGregor. 1.(1994) . Fluid sediment in combined sewers. Wat. Sci. Tech. 29(1-2). 113­123.

Bachoc, A. (1992) . Le transfert des solides en reseau d'assainissement unitaire . These de doctoral, Institut National Poly techniquede Toulouse. Toulouse, France.

Bailly . C. (1996). Identification des strat~gies de prise en compte du temps de pluie dans "elaboration des sys~mes

d'assa inissemenl. Ecole Nationale des Ponts et Chaussees, Paris. France.Bertrand-Krajewski, J. L. (1991) . Model isation des debits et du transport solide en reseau d'assainissement, Etude bibliographique.

Ecole nauonale des ingenieurs des travaux ruraux et des techniques sanitaires, Lyonnaise des Eaux, Strasbourg. France .Chebbo , G. (1992) . Solides des rejets pluviaux urbains, Caracterisation et traitabil ite. These de doctorat , Ecole Nationale des Ponts

et Chaussees, Paris. France .Gromaire-Mertz, M. C.• Chebbo, G. and Saad, M. (1998) . Origins and characterisation of urban wet weather pollut ion in

comb ined sewer systems : the experimental urban catchment 'Le Marais ' in Paris. Wat. Sci. Tech.• 37(1) (this issue) .Huber. W. C.• Heany, J. P.• Nix. S. J.• Dickinson. R. E. and Polmann, D. J. (1981). Storm Water Management Model user's

manual. version III. University of Flor ida. Gainsville , Florida.Jack. A. G.• Pertie, M. M. and Ashley. R. M. (1996) . The diversity of sewer sediments and the consequences for sewer flow

quality modelling. Wat. ScL Tech.• 33(9).207-214.Krejci. V,. Dauber. L., Novak. and Cuger, W. (1987) . Contribution of different sources to polluant loads. Proc.Fourth Int. Con];

Storm Urban Drainage, Lausanne, SwitzerlandLavallee . P.• Lessard. P. and Villeneuve. J·P. (1984), Water quality variation in running waters due to combined sewer

overflowing. Evaluation of negative influence . Proc. Third Int. Conf. Urban Storm Drainage. Goteborg, Sweden, 761·769.

Maksimovic, Radojkovic (1986) . Urban drainage catchments: selected worldwide rainfall-runoff data from experimentalcatchments. Pergamon Press. Belgrade . Yougoslavia.

Paltry, A.• Doubreres (1986) . Echantillonnage des maueres en suspension en reseau d'assain issement. Quelques reflexions.Serv ice departemental de I'assainisscment de Seine-st-Denis, Rosny-sous-bois, France.

Perrusqufa, G. S. (1992) . An experimental study on the transport of sediment in sewer pipes with a permanent deposit.War. Sci.Tech. 25(8). 115-122.

Press. W. H.. Flannery, B. P.• Tenkolsky, S. A. and Venerling, W. T. (1988). Numerical reciepes in C. The art of scientificcomputing. Cambridge University press, Cambridge. England.

Ristenpart , E.• Ashley . R. M. and Uhl, M. (1995) . Organ ic near-bed fluid and particulate transport in sewers. Wat. Sci. Tuh.31(7).61 -68.

Page 9: Storm water quality modelling, an ambitious objective?

Stormwaterqualitymodelling 213

Saad,M., Mertz.M. C. and Chebbo,G. (1996).Protocole de mesuredes MES,des MVS,de la DCOet de la DBO.bassinversanturbainexperimental du Marais.rapportd'avancement. CERGRENE, Paris,France.

Saget, A. (1994). Base de donnees sur la qualit6 des rejets urbains de temps de plie : distribution de la pollution rejetee,dimensionsdes ouvragesd'interception. Thesede Doctoral, Ecolenationale des pontset chaussees, Paris.France.

Verbanck, M. A. (1996).Assessment of sedimentbehaviour in a cunette-shaped sewersection.Waf. Sci. Tech. 33(9),49-59.