~ Pergamon

PIT: S0273-1223(99)00340-6

Wat. Sci. Tech. Vol. 39. No. 12,pp. 241-248.1999CI999IAWQ

Publishedby ElsevierScience LtdPrinted in Great Britain.All rights reserved

0273-1223199520.00 + 0.00

MODELLING RUNOFF AND DIFFUSEPOLLUTION LOADS IN URBAN AREAS

F. H. S. Chiew and T. A. McMahon

Cooperative Research Centre/or Catchment Hydrology. Department ofCivil andEnvironmental Engineering. University 0/Melbourne. Parkville. Victoria 3052.Australia

ABSTRACT

This paper presents a simple approach for estimating long-term runoff and diffuse pollution loads in urbancatchments. and discusses conceptual modelling methods for simulating daily runoff and pollution loads.The modelling results for several catchments in Australian capital cities are presented. The study indicatesthat long-term and daily runoff can be estimated reasonably accurately using simple approaches. However,the water quality characteristic can vary considerably between catchments, and in the absence of data, themodels can only provide a guide to the probable range of diffuse pollution load generated from a catchment.C> 1999 IAWQ Published by Elsevier Science Ltd. AUrights reserved

KEYWORDS

Catchment;diffusepollution; model; runoff; stormwater;urban.

INTRODUCTION

Runoff volumes and pollution loads increase considerably when a catchment is urbanised. Methods forestimating runoff and pollution loads are required to investigate the impacts of urbanisation and to evaluatemanagement and design options for controlling water quality in urban waterways and receiving waters.Estimates over a range of time scales are used for different managementstudies and data availability. Thispaper presents a simple approach for estimating long-term runoff volumes and diffuse pollution loads, anddescribes two conceptualmodels for simulatingdaily runoff and pollution loads.

ESTIMATIONOF AVERAGEANNUALRUNOFFAND POLLUTIONLOADS

The average annual runoff can be estimatedas

Runoff = rei A, Rainfall + rep (1 - Ai)(Rainfall+ Outdoor Water Use) (1)

where A, is the fraction of effective impervious area in the catchment, and rei and rep arc runoff coefficients(proportionofrainfall that becomes runofi) in the imperviousarea and pervious area. respectively.

241

242 F. H. S. ClIIEW and T. A. MCMAHON

Most of the runoff in urban areas come from impervious surfaces. The key variable in estimating runoff istherefore the fraction of effective impervious area. All the rain falling onto effective impervious surfacesbecomes runoff after an initial loss (due to water filling the surface depressions and pores) is satisfied. Thefraction of effective impervious area can be estimated from a runoff-rainfall plot of small events (see Figure1), but there can be a large scatter in the data (see Figure 1b). As runoff from small events is generated onlyfrom effective impervious surfaces, the slope gives an estimate of the fraction of effective impervious area,and the intercept ofthe rainfall axis is an estimate of the initial loss.

The fraction of effective impervious area can also be estimated from aerial photographs and knowledgeabout the drainage system. The aerial photograph can be used to estimate the fraction of directly connectedimpervious area (impervious surfaces that are directly connected to the drainage system). This is oftensimilar to the fraction of total impervious area, except in catchments where roofs are not connected directlyto the drainage system. The fraction effective impervious area is typically about 80 to 90% of the fraction ofdirectly connected impervious area (see Boyd et al., 1993), because not all impervious surfaces that appearto be directly connected are directly connected (for example, blockages in gutters can result in water flowingonto pervious surfaces).

The impervious area runoff coefficient, rei, in equation (I) is typically about 0.8 to 0.95, depending on theamount of initial loss and the number of raindays. The pervious area runoff coefficient, rep, is dependent onthe climate and physical catchment characteristics. In Australian capital cities, it ranges from 0.05 in drierareas to 0.4 in the catchments.

(a) Dee Why Creek170 ha catchment in Sydney

fraction effective impervious = 0.35initial loss = Imm •

(b) Blackburn Lake200 ha catchment in Melbourne

40 .-::--:---;:::--:-:---:-__--:---:::-::----,fraction effective impervious = 0.2initial loss = Imm

30

~.....l::I 20 •~

•10

10

20 40 60 80 100 20 40 60Rainfall (mm) Rainfall (mm)

Plots show all storm events. Points above the line indicate events where surface runoff is alsogenerated from the pervious surfaces.

Figure 1. Event rainfall-runoffplot to estimate fraction ofeffective impervious area in the catchment.

Pollution load

The average annual pollution load can be estimated as

Pollution Load = EMC x Runoff (2)

The event mean concentration (EMC) is defined as the pollutant load washed off by a storm event dividedby the event runoff volume. It can be estimated by monitoring water quality concentration and dischargeover a storm event. As the EMC can vary considerably between storms, monitoring should be carried outover several events, and the EMCs of the different storms averaged to provide the EMC value for thecatchment. The EMC, rather than dry weather concentration, is used because most of the loads aretransported by the big events.

Modelling runoff and diffuse pollution loads in urban areas 243

The EMC depends on catchment and climate characteristics, and can vary by more than an order ofmagnitude between catchments. A good event monitoring program is thus essential where accurateestimates of pollution loads are required . However, in the absence of water quality data, EMC valuesreported in the literature can be used as a guide to estimate the likely range of pollution load. Athayde et al.(1983) and Torno (1984) provide a summary of EMC data, mainly from the US Nationwide Urban RunoffProgram, and more recently, Duncan (1999) (see also Chiew et al., 1997; Duncan, 1997; and Mudgway etal., 1997) gives summaries of EMC for 21 water quality parameters from data reported in over 400 separatestudies worldwide (mainly North America, Europe, Australia and Japan). Figure 2 shows examples ofEMCdata summarised by Duncan (1999).

Plots give mean ± one standard deviation of values reported in the literature

Total Suspended Solids Tota l Phosphorus

Roads

Roofs

Residential

Industrial

Commercial

Agricultura l

Forest

10 100Concentration (mgfL)

1000 0.01 0.1 I

Concentration (rng/L)

Figure 2. Compressed examples ofEMC data summaries given in Duncan (1998).

Average annual estimates

Table 1 compares the annual runoff estimated using equation (I) with the recorded runoff in II catchmentsin capital cities in eastern Australia. The comparison indicates that long-term runoff volumes can beestimated reasonably accurately (estimates in nine of the II catchments are within 20% of the recordedvolumes), if the fraction of effective impervious area is known. However, the pollution load can varyconsiderably between catchments, and in the absence of local event water quality data, the EMCs in Figure 2can only be used as a guide to estimate the probable range of diffuse pollution load generated from acatchment (see also Table 1).

MODELLING DAILY RUNOFF

Daily estimates of runoff and pollution loads are often required to investigate shorter term impacts, todetermine seasonal characteristics of urban runoff, to study alternative water quality management options ,and as inputs to water quality models . Runoff from effective impervious surfaces can easily be modelledbecause all the rainfall becomes runoff after an initial loss has been satisfied. Runoff from the pervious area('pervious' refers to all surfaces that are not effective impervious area) can be simulated using conceptualmodels that mimic the catchment processes. These models cons ist of one or more storages and equationsthat describe the movement of water between the storages. Chiew et al. (1995) and Chiew and McMahon(1996) showed that there is little difference between the better conceptual approaches for modelling dailyrunoff.

244 F. H. S. CHIEW and T. A. MCMAHON

Table 1. Estimates of average annual runoff and pollution loads in 11 catchments

Catchment Area Frac. Data Values(ha) eff. period Recorded values Values estimated using equation ( I) Estimated

imp. using model(Ai) in Figure 3

Ann. Out. Ann. Ann. Total Total Ann.rain water runoff runoff suspended phosphorus runoff

(mm) use (mm) (mm) solids (kglha) (mm)(mm) (kg/ha)

SydneyPowells Creek 287 0.40 8/92-6/94 740 75 470 380 200 - 2000 0.6 - 3 400Salt Pan Creek 668 0.27 8/92-6/94 650 75 220 280 100 - 1000 0.4 - 2 270Cup & Saucer Creek 478 0.35 8/92-6/94 780 65 430 370 200 - 2000 0.6 - 3 390Greendale Creek 178 0.47 6/92-7/94 970 45 630 510 300 - 2000 0.8 - 4 560Bumtbridge Creek 372 0.12 6/92-7/94 970 70 200 290 200 - 1000 0.4 - 3 380Dee Why Creek 171 0.35 6/92-7/94 970 55 520 430 200 - 2000 0.6 - 4 490

CanberraYarralumla Creek 445 0.25 1172-12/95 640 75 290 250 100 - 1000 0.4 - 2 300Long Gullv Creek 490 0.18 1/92-12/95 740 70 220 230 100 - 1000 0.4 - 2 300

BrisbaneSandy Creek 227 0.20 8/94-6/97 1130 70 600 540 300 - 2000 0.8 - 5 750Cressev Street 207 0.17 8/94-6/97 1310 70 670 590 300 - 2000 0.9 - 5 870

MelbourneBlackburn Lake 202 0.28 1/96-12/97 730 40 340 280 100 - 1000 0.4·2 370

• The fraction of effective impervious area is estimated from plots of storm runoff versus storm rainfall.

• The annual values are averaged over the period of data (i.e., they are not long-term averages).

• Outdoor water use is estimated to be 75 mm/year for fully residential areas and 65 mm/year for parks and playgrounds (seeMitchell, 1998). Values in the table are averaged over the entire catchment, but all outdoor water use is attributed only to thepervious areas.

• In using equation (1), rc;=0.85 is used, for Sydney, Canberra and Melbourne, r,,=0.2, and Brisbane, r,p··0.3.

• In calculating pollution load, TSS EMCs of 50 - 400 mg/L and TP EMCs of 0.15 - 0.85 mg/L are used (see Figure 2).

In the absence of data, the simple model in Figure 3 can be used to characterise daily runoff. Surface runoffis generated from the pervious area when saturation occurs, and baseflow is simulated using a linearrecession. Evapotranspiration is dependent on the amount of water in the soil store, but cannot exceed thepotential rate (see Figure 3).

The annual runoff volumes estimated by this model for 11 catchments are given in Table 1. The reliabilityof the estimates are similar to those estimated using equation (I), because no model calibration wasperformed (the storage capacity and baseflow factor were set at 80 mm and 0.03 respectively). However,the model can provide an indication of the daily flow characteristics, as illustrated in Figure 5 for twocatchments.

lOT · min ( io ' 10, PET)

(\CC ChiC'" and McMahun. Illt)..)

outdoorwater usc

R! !am

surface runoff

Ai

80 mm

51.-,.j~j; buse flow

= 003, S

4 •

1 111111

Figure 3, Structure of a simple conceptual model of daily runoff..

Modelling runoff and diffuse pollution loads in urban areas

Cooperative Research Centre for Catchment Hydrology (CRCCH>daily urban runotTmodel

245

Where there is runotT data for model calibration, better estimates of daily runoff can be obtained bysimulating the catchment processes in more detail.

Figure 4 shows the CRCCH daily urban runoff model which retains the simplicity of the earlier model, butprovides a better conceptual representation of the processes. Like the earlier model, all the daily rainfall inthe effective impervious area becomes runoff once the daily initial loss is satisfied. The remaining area ismodelled as two separate parts with ditTerent storage capacities (related to effective soil depth). The firsthas a smaller storage capacity and represents parts of the catchment that saturates easily. The secondrepresents the remainder of the catchment with a greater soil storage capacity. Surface runotToccurs whenthe storage capacities are exceeded (when saturation occurs). This pervious area runoff model is based onthe partial area saturation excess runotTgeneration, a concept favoured by the newer hydrologic models (seeZhao, 1992; Robinson, 1993). The model also simulates infiltration excess runotT, but this is not shown inFigure 4.

Water from the soil stores recharges a groundwater store when the storage exceeds a certain amount ('fieldcapacity'). Recharge is calculated as a parameter (which mimics the hydraulic conductivity) times theamount the storage exceeds the 'field capacity'. Baseflow from the groundwater store is simulated using alinear recession.

rain!ETl

! Outdoorwater USe

surface runoff•effective

impervious.....

I::'!"i"'""~ '--

effectiveno" ... in pervious... eo~r!" 0

~- r... ~~ .. ~ .~: ..Igroundwater store (GW) I baseflow

- k GW •Need to spec ify effective fraction imperviousnessand the initial loss in the impervious area. the twofractions of the remai ning area (A l and Al) andtheir storage capacities (S / cap and Slcap ).

Model can also rout flows and simulate infi ltrationexcess runoff(two parame ters for each process)

Figure 4. Structure of the CRCCH daily urban runoff model (modet parameters are highlighted in bold and italics) ,

Evapotranspiration is calculated using the algorithm in Figure 3. The evapotranspiration demand is satisfiedfirst from the larger store, therefore allowing for some redistribution of water between the two stores. Therouting of flows to the catchment outlet can also be simulated by the model.

The model has six parameters (if the parameters for the impervious area are determined from event rainfall­runotTplot, and routing is not required). Application of the model to the 11 catchments in Table I, and othercatchments in Australia, indicated that the model consistently provides a satisfactory simulation of dailyrunoff. Testing of the model using independent data sets (data that are not used for the model calibration)showed that the total estimated runoff volumes were generally within 10% of the recorded runotT volumes,

246 F. H. S. CHIEW and T. A. MCMAHON

and the correlations (coefficient of efficiency) between the daily simulated and recorded runoff weregenerally greater than 0.8.

The characteristics of the daily flows in two catchments, and the daily flows simulated by the CRCCRmodel and the simple conceptual model in Figure 3, are compared in Figure 5. As expected, the calibratedmodel performed better, particularly in the simulation of high flows, because the model parameters werecalibrated to minimise an objective function that reflects the simulation of high flows. The simple model,without any calibration, can often provide a satisfactory simulation of daily flows (see Figure 5b), althoughin some cases, the runoff estimates can be very poor (see Figure Sa). Thus, where adequate rainfall-runoffdata are available, the CRCCR model should be used to estimate daily runoff, but where there is little data,the use ofa simpler model is sufficient to provide an indication ofthe daily flow characteristics.

-- Estimated using model in Figure 3 (no calibration)

- - - - Estimated using CRCCR model (model was calibrated)

100 ...--------------.

- Recorded runoff

100 ...--------------,(b) Yarralumla Creek

(Canberra)

10

(a) Sandy Creek(Brisbane)

0.112510 30 50 70 90959899 12 510 30 SO 70 90959899

Percentage of time daily flow is exceeded

0.1

Figure 5 Flow duration plots comparing daily runoff simulated by the models with the recorded runoff

MODELLING DAILY POLLUTANT LOAD

There are various water quality models that attempt to simulate dry weather pollutant accumulation andwashoff over storm events in urban catchments. These models may be useful in studying pollutant buildupand transport processes, and estimating pollutant loads generated over storm events (see Chiew et al., 1997).However, where daily loads are required, the available data rarely justify the use of these models. In moststudies, there is only sufficient data to estimate daily diffuse pollutant load as

Load+

surface runoff(from impervious and pervious surfaces) x EMCbaseflow x dry weather concentration (3)

The daily conceptual models described in the previous section allow for the modelling of the different waterquality characteristics associated with the different runoff components. The literature can provide a guide tothe typical range of EMC values (see Figure 2), but where accurate estimates of diffuse pollution loads areneeded, event monitoring should be undertaken to determine the EMC for the catchment. The dry weatherwater quality concentration is usually lower than the EMC, and it can be determined from several dryweather baseflow samplings. There is usually little reason to use a more complex model than the linear onedescribed by equation (3) because it is difficult to define a clear relationship between runoff and EMC (seeFigure 6).

Modelling runoff and diffuse pollution loads in urban areas 247

Nevertheless, in some catchments, a power relationship between pollutant load and runoff may provide abetter description of the data

LOAD = a RUNOFF b (4)

Here, the parameters, a and b, should be obtained by optimising an objective function that reflects the needsof the study, rather than by simply minimising the surn of square of errors (SSE) between the predicted and'recorded' loads in the log scale, as is conventionally done because of mathematical simplicity (see Figure7). This is because the log scale gives similar weight when minimising the errors in the big and smallpredicted loads, while in practice, it is usually more important to estimate the bigger loads accurately.

Blackburn Lake (Melbourne)0.8 r------....:----~--.,

Salt Pan Creek (Sydney)600 r------~~...::..:;-----,

10

• •

••

•• •

246 8

Storm runoff (mm)

•.-..i 400'-'o::EUl 200tI) •tI)

E-< •0 •

8 02 4 6

Storm runoff (mm)

0.0 L..-__.a...-__.a...-__-'--_---'

o

~0.6..~

g •U 0.4 ••• •• •~ - •!: 0.2 1-.- • ••..... ••

Figure 6. Event mean concentrationversus storm runoff in two catchments.

15105

Dotted line shows power relationshipwith a and b optimised to minimisethe SSE between the predictedandactual loads in the linear scale.

oo

Full line is the same regression .'20000 in the log plot on the left. ",,,,,10000 ",

30000

40000 ,-----------_---,

10010

Interceptand gradient of the linearregression in this log plot gives aand b directly•

1000

10000

100000r-----------~___,

~]:9~

JStorm runoff (mm)

Figure 7. Total suspendedsolids versus runoff in BurntbridgeCreek (Sydney)plotted on linear and log scales(showing that minimisingSSE of Load, rather than 10g(Load) puts more weighton estimatingthe large loads

accurately).

SUMMARY AND CONCLUSIONS

This paper presents a simple approach for estimating long-term runoff and diffuse pollution loads in urbancatchments, and describes two conceptual models for simulating daily runoff.and pollution loads. Testing ofthe models on several catchments in capital cities in eastern Australia indicates that long-term runoff can beestimated reasonably accurately, and daily runoff can be characterised adequately using simple conceptualmodels.

248 F. H. S. CHIEW and T. A. MCMAHON

In the absence oflocal data, the runoffmodels and water quality data reported in the literature can be used toestimate the likely range of diffuse pollution load generated from a catchment. However, because waterquality characteristics can vary considerably between catchments, water quality monitoring over severalstorm events should be undertaken where accurate estimates ofpollution loads are required.

ACKNOWLEDGEMENTS

The data have been provided by Sydney Water, AWT-Ensight, ACT Electricity & Water, Brisbane CityCouncil and Melbourne Water. The authors would like to thank Sharyn Ross and Jai Vaze for compilingand analysing some ofthe data.

REFERENCES

Athayde, D. N., Shelley, P. E., Driscoll, E. D., Gaboury, D. and Boyd, G. (1983). Results of the Nationwide Urban RunoffProgram. U.S. Environmental Protection Agency, Washington D.C., PB84-185537.

Boyd, M. J., Bufill, M. C. and Knee, R. M. (1993). Pervious and impervious runoff in urban catchments. Hydrological Sciences,38(6), 463-478.

Chiew, F. H. S., Duncan, H. P. and Smith, W. (1997). Modelling pollutant buildup and washoff: keep It simple. Proceedings ofthe 24,h International Hydrology and Water Resources Symposium, November 1997, Auckland, New ZealandHydrological Society, pp. 131-136.

Chiew, F. H. S. and McMahon, T. A. (1994). Application of the daily rainfall-runoff model MODHYDROLOG to 28 Australiacatchments. Journal ofHydrology, 153,383-416.

Chiew, F. H. S. and McMahon, T. A. (1996). Conceptual modelling of daily runoff in urban catchments. Proceedings of the .,4International Conference on Urban Storm Drainage, September 1996, Hannover, Germany, IAHRJIAWQ JointCommittee on Urban Storm Drainage, Seeliger Sofort-Druck, Hannover, 1,323-328.

Chiew, F. H. S., Mudgway, L. B., Duncan, H. P. and McMahon, T. A. (1997). Urban Stormwater Pollution. CooperativeResearch Centre for Catchment Hydrology, Melbourne, Australia, Industry Report 97/5, 18 pp.

Chiew, F. H. S., Osman, E. H. and McMahon, T. A. (1995). Modelling daily and monthly runoff in urban catchments.Proceedings of the r Intemauonal Symposium on Urban Stormwater Management, July 1995, Melbourne, Australia,Institution of Engineers, National Conference Publication, 95/3( I), 255-260.

Duncan, H. P. (1997). An overview of urban stormwater quality. Proceedings of the 24'· International Hydrology and WaterResources Symposium, November 1997, Auckland, New Zealand Hydrological Society, pp. 143-148.

Duncan, H. P. (1999). Urban Stormwater Quality: A Statistical Overview. Cooperative Research Centre for CatchmentHydrology, In Press.

Mitchell, V. G. (1998). Development of an Urban Water Balance Model to Assess the Re-Use Potential of Stormwater andWastewater. Ph.D. Thesis, Department of Civil and Environmental Engineering, University of Melbourne, Australia, 273pp.

Mudgway. L. B., Duncan, H. P., McMahon, T. A. and Chiew, F. H. S. (1997). Best Practice Environmental ManagementGuidelinesfor Urban Stormwater. Cooperative Research Centre for Catchment Hydrology, Report 97n, 125 pp.

Robmson, M. (1993) Changing ideas regarding storm runoff processes in small basins. In: Flow Regimes from International andExperimental Network Data (FRIEND) (Ed: Mark Robinson), Institute of Hydrology, United Kingdom, Volume 3,pp.3-16.

Torno, H. C. (1984). The nationwide urban runoff program. Proceedings of the 3'" International Conference on Urban StormDrainage (Editors: P.Balmer, P. Malmquist and A.Sjoberg), Goteborg, Sweden, pp. 1465-1474.

Zhao, R. J. (1992). The Xinanjiang model applied in China. Journal ofHydrology, 135, 371-381.