- Email: [email protected]
Modelling sedimentation processes in a constructed stormwater wetland
The Science of the Total Environment 266 Ž2001. 61᎐68
Modelling sedimentation processes in a constructed stormwater wetland D.J. Walker U Department of Ci¨ il & En¨ ironmental Engineering, The Uni¨ ersity of Adelaide, Adelaide 5005, Australia Received 17 September 1999; accepted 28 June 2000
Abstract The design and operation of constructed wetlands for the treatment of stormwater relies heavily on promoting sedimentation, and being able to predict accurately the expected effectiveness of the pond in removing material from the inflows. A study of sediment behaviour has been carried out in a stormwater wetland in Adelaide, Australia where computer predictions, based on solving the hydrodynamic equations and the transport equation, have been compared to deposition patterns observed in the field. The long-term residence time distribution has been shown to be useful in predicting overall sediment removal rates. Comparisons between the model and field observations indicate generally good agreement. Sources of potential error identified include the variable nature of the runoff concentrations entering the pond and the sediment size distribution. The importance of the transient nature of the flow events was highlighted by the spread of sediment throughout the whole pond. 䊚 2001 Elsevier Science B.V. All rights reserved. Keywords: Numerical model; Sedimentation; Stormwater; Constructed wetland
U
Tel.: q61-8-8303-4319; fax: q61-8-8303-4359. E-mail address: [email protected] ŽD.J. Walker.. 0048-9697r01r$ - see front matter 䊚 2001 Elsevier Science B.V. All rights reserved. PII: S 0 0 4 8 - 9 6 9 7 Ž 0 0 . 0 0 7 3 0 - 0
62
D.J. Walker r The Science of the Total En¨ ironment 266 (2001) 61᎐68
1. Introduction Sedimentation in ponds and wetlands is important, not only for removing the sediment itself, but also nutrients and contaminants which are readily attached to fine particles ŽFennessy et al., 1994; Raisin et al., 1997.. There have been a number of studies undertaken of sedimentation in wetlands both in situations where there was an essentially constant flow through ŽBrueske and Barrett, 1994; Fennessy et al., 1994; Hey et al., 1994; Prescott and Tsanis, 1997; Tsanis et al., 1998. and where the main flows were due to runoff from storms ŽReinelt and Horner, 1995; Raisin et al., 1997; Lloyd, 1997; Murphy et al., 1998.. For the latter, the intermittent nature of stormwater wetlands leads to a fundamentally different situation where sediment is more likely to be distributed around the entire basin ŽFennessy et al., 1994; Walker et al., 1994.. There have been a number of attempts in the past to model wetland sedimentation. Some have related outlet concentration to inlet concentration Že.g. Reinelt and Horner, 1995.. A simple approach that has been taken is to relate sedimentation rates to the distance from the inflow point ŽFennessy et al., 1994. but this ignores the two-dimensional nature of the flow patterns. The most comprehensive model to date is that reported by Tsanis et al. Ž1998. who solved the
hydrodynamic and sediment transport equations to predict water quality parameters in a natural wetland. The model was used to predict average sediment concentrations in the wetland and, after adjustment of settling velocities and external loadings, gave results that agreed reasonably well with field observations. In the present work a field study investigating sedimentation in a stormwater wetland has allowed verification of a computer model that solved the hydrodynamic and sediment transport equations to enable the overall removal efficiency and the resulting sedimentation patterns to be predicted. A particular focus of the work concerned the residence time distribution ŽRTD. of the pond.
2. Study site The site of the study was the Minkara stormwater wetland located approximately 20-km south of Adelaide in South Australia. The wetland was formed in 1992 with the construction of a concrete weir across a natural drainage channel at a point where it passed through a twin pipe culvert. No artificial planting was undertaken but after 7 years the pond has been infiltrated naturally by several species of macrophytes with the dominant being T. Orientalis. As is evident from Fig. 1 the majority of the plants have developed around the
Fig. 1. Minkara wetland, showing areas of macrophyte that have established naturally around the perimeter of the pond, and the location of sediment traps used in the study.
D.J. Walker r The Science of the Total En¨ ironment 266 (2001) 61᎐68
periphery of the wetland where the water depths are most favourable. The wetland has a volume of 2700 m3 and covers an area of 3500 m2 . The catchment has an area of 130 ha. leading to an area ratio of approximately 0.25%. A survey of the wetland has indicated an average depth of 0.77 m and a maximum depth of 1.40 m. For modelling purposes the wetland has been considered flat bottomed since the maximum depth never exceeds twice the mean depth ŽHamilton and Mitchell, 1997.. An analysis of the sediment deposited on the bottom of the wetland since construction ŽJacobi and Murphy, 1996. indicated a mean particle size of 15 m.
63
bottles were changed. The data demonstrated the rapid build-up of turbidity during a storm event, but no attempt was made to relate this to the sediment in the traps. The sediment in the bottles was filtered onto pre-weighed Whatman GFrC filter papers within 1 h of collection and dried in an oven at 90⬚C overnight. Long-term sedimentation rates were converted to cmrday by dividing them by the product of the bulk density Ž1.5 grcm3 . multiplied by the sediment trap opening area Ž8.0 cm2 ., ŽFaas and Carson, 1988..
4. Numerical models 3. Methods and techniques The field study was undertaken in two parts. At the start of winter 1998, coarse sand horizon markers were installed at five locations Žsee Fig. 1. throughout the pond. One year later, sediment cores were taken using a Perspex corer with a diameter of 40 mm. The depth of buildup was measured, and the samples were subsequently dried and analysed for organic content by heating to 500⬚C for an hour after being milled and dried at 110⬚C for 2 h. The average organic content was found to be 15%. During June to September 1999, sediment traps were installed at the same locations and monitored on a regular basis. The sediment collection bottles were made from high density polyethylene ŽHDPE. with an opening of 32 mm, a depth of 128 mm and a volume of 250 ml. The aspect ratio of 4 was selected based on the recommendations of Hakanson et al. Ž1989., ˚ Fennessy et al. Ž1994.. The traps were located on poles in the wetland and were retrieved at least fortnightly by canoe so as not to disturb the area. Replacement bottles were filled with distilled water and frozen prior to installation in the wetland to reduce the chance of any disturbance caused by manipulating the traps ŽBrueske and Barrett, 1994; Fennessy et al., 1994.. Greenspan turbidity sensors operating at 860 nm wavelength and recording data every 5 min were located with three of the traps Ž2, 4 and 7.. Data were downloaded using a laptop computer when sediment
The numerical simulation of sedimentation in the wetland was undertaken using a suite of numerical models, the first to solve the hydrodynamics of the pond during inflow events, the second to solve the sediment transport and deposition, and the third to predict the runoff events over the period under study. The relatively shallow depth and the assumption that stratification would not be significant, especially during flow events, led to the selection of a two-dimensional depth-averaged model for the flow simulations. The transient form of the equations was chosen so that the development of flow patterns in time could be described. Details of the model can be found in Walker Ž1998., although many other researchers Že.g. Tsanis et al., 1998. have used similar models. Rather than try to model 5 years of continuous records, a selection of 11 typical events of varying size Žfrom 10 to 400% of pond volume. and duration Žfrom 5 to 13 h. were run and the predicted flow patterns were stored for later use with the transport model. The sediment transport was modelled using a two-dimensional horizontal approximation to the transport equation where a source term included the process of deposition. A full three-dimensional model would have been preferred but the fine nature of the sediment Ž d50 s 15 m. meant that concentration variations through the depth could be ignored. Deposition at each time-step was assumed to be based on the product of particle fall velocity and the mean concentration. The
64
D.J. Walker r The Science of the Total En¨ ironment 266 (2001) 61᎐68
Fig. 2. Flow chart for numerical models used in study.
low flow velocities in the pond meant that erosion of deposited material was assumed not to occur. The model is based on those reported by Tsanis et al. Ž1998., Teisson Ž1991.. The model used a fixed inflow concentration and was run for each of the flow simulations outlined above. The inflow concentration was selected based on observations over a number of years. Although a simplification of the actual behaviour, it was not possible to derive a simple relationship that could be used to predict inflow concentration based, for example, on the inflow rate. The output from the model was a predicted rate of sedimentation over the entire solution grid. Once again, results for each of the events were stored for later use. The runoff from the catchment was modelled using a computer simulation package AWBM ŽCRCCH, 2000. based on original work by Boughton Ž1984.. The model represents the catchment processes using a number of storages that must be calibrated. The calibration was carried out using 2 months of data collected during the study and then tested on a separate 5-week section. The root mean square error was found to be 1.33 mm of runoff per day. The final prediction of sedimentation rates was determined by running the runoff model for a 5 year simulation and summing the number of events based on size, using the 11 predetermined sizes developed for the flow modelling. Over the 5
years there were, for example, 78 where the volume was 10% of pond volume, 42 where the volume was 20%, 18 of 40%, gradually reducing to five events where the inflow volume was 300% of pond volume. Using the results of the transport model runs the total sediment accumulation was calculated by summing the accumulation for each size event multiplied by the number of events predicted for the simulation. A flow chart for the programs is shown in Fig. 2.
5. Results The main result of the study is shown in Fig. 3 where the numerical predictions of annual sedimentation are shown. Superimposed on the contours are the observed sedimentation rates from both the long-term and short-term study. The concentration of material near the inlet is evident, but the spread into areas some distance from the inlet is also predicted to occur. Based on the results of the field study the estimates are considered reasonable despite the general overprediction of the rates by the model. The most likely explanation is an overestimate of the inflow concentrations. The relative sedimentation rates between the stations are in fair comparison with the observed field rates. The even pattern of sedimentation is partly due
D.J. Walker r The Science of the Total En¨ ironment 266 (2001) 61᎐68
65
Fig. 3. Contours of predicted annual sedimentation in Minkara wetland based on 5 year simulation. The figures in the boxes are from the field study where the first number is the sediment buildup over a full year, and the second is based on a 3-month short-term study. All figures are in millimeters.
to the fact that the variations in the bottom contours and the effect of vegetation have been ignored. Despite this the results are useful. It should be remembered when comparing the two estimates of sedimentation rate that, as pointed out by Faas and Carson Ž1988., sedimentation traps give estimates of the potential sedimentation rate rather than the actual. In this case it would be expected that the short-term rates would give an overestimate of the true rate.
As an additional aspect to the present study a consideration of the use of the residence time distribution ŽRTD. in predicting overall sediment removal rates was undertaken. Following the work of Walker Ž1998. a RTD for Minkara wetland was calculated taking account of the interval between events and the short-circuiting in the pond due to the less-than-ideal flow patterns during events ŽFig. 4.. If the sedimentation process is assumed to depend purely on the time available for set-
Fig. 4. Long-term residence time distribution for Minkara wetland based on 5 year simulation Ž1994᎐1998.. The mean residence time is 12 days, but a significant portion is resident for less.
D.J. Walker r The Science of the Total En¨ ironment 266 (2001) 61᎐68
66
tling, the removal of sediment is a function of the residence time alone. In this case the removal efficiency can be predicted directly. Given that all residence times longer than the fall velocity will result in complete removal, and shorter times will result in a proportional loss, the removal Ž E . can be determined as:
Es
t max
tfall
Ý
f Žt. q Ý
tfall
0
t f Žt. t fall
tmax
= 100
Ž1.
Ý f Žt. 0
where E is the percentage removal, f Ž t . is an element of the RTD, t fall is the time in seconds for a particle to fall through the depth of the pond Ž0.77 m. and t max is the maximum residence time in seconds from the RTD. The efficiency could be calculated based on a single particle size or by summing over a number of size fractions. For the present wetland both gave a predicted removal efficiency of 50%. This is of the same order as the observed figure ŽWalker et al., 1997. of 60᎐70%. The underestimation is likely due to the problems associated
with using a daily timestep to represent flow events that are often of much shorter duration.
6. Discussion Predictions made on the basis of numerical modelling of stormwater ponds have a number of uses. In the design of these facilities it is possible at present to estimate overall removal efficiencies using such factors as the area of the pond relative to the catchment and the incoming sediment concentrations ŽDuncan, 1997.. The present methodology is unique in that it can be used to optimise the shape of the ponds, the location and size of islands designed to improve flow patterns, and to assess the effect of the hydrological inflow characteristics on pond performance. An example of this is shown in Fig. 5 where islands have been incorporated into the design to reduce short-circuiting and improve retention. When this pond layout was assessed using the same procedures described above the predicted removal efficiency was 52% Žcompared to 50% for the initial design.. The reduced pond volume caused by the inclusion of the islands was more than offset by the superior flow patterns Ži.e. islands reduce the flow rate
Fig. 5. Predicted flow patterns for modified stormwater pond system incorporating islands to optimise retention.
D.J. Walker r The Science of the Total En¨ ironment 266 (2001) 61᎐68
and increase retention time.. While the predicted difference is quite small, the important aspect of the exercise is the ability to make a quantitative assessment of the behaviour and comparisons based on the behaviour. It is noted that the present study has made a number of simplifying assumptions. The constant depth model allowed two-dimensional models to be used that are easily developed and efficiently run. The fine nature of the sediment was also crucial to allowing a two-dimensional analysis of the transport of sediment in the pond. A daily timestep in the rainfall runoff model was able to use relatively crude hydrological input, and this is certainly the first area that should be upgraded by employing an hourly timestep. A detailed sensitivity analysis was not undertaken as part of the study but experience with the models show that the characteristics of the inflow sediment concentration and particle sizes would be the most critical in terms of accuracy. Field data were available to allow estimates to be made, but these were crude and did not take account of the variations that could be expected due to inter-event times, rainfall intensity and catchment characteristics.
7. Conclusion Models have been developed to allow the prediction of overall sediment removal rate as well as the likely sedimentation patterns in a stormwater wetland. The models solve the transient flow field that develops in the wetland during events, the resulting transport and deposition of sediment in the wetland and also predict the time history of events based on daily rainfall records. Comparison with results of a complementary field study has indicated that the model predictions are reasonable and that the modelling would be useful in the design of stormwater facilities. The models allow a more detailed analysis of pond shape and features to be undertaken than is possible at present where crude parameters such as relative area or inflow concentrations are unre-
67
sponsive to pond features where the designer could have some input. References Boughton WC. A simple model for estimating the water yield of ungauged catchments. Institution of Engineers, Australia. Civil Eng Trans; C 1984;E26:83᎐88. Brueske CC, Barrett GW. Effects of vegetation and hydrologic load on sedimentation patterns in experimental wetland ecosystems. Ecol Eng 1994;3:429᎐447. CRCCH Ž2000.. AWBM Water Balance Model. Cooperative Research Centre for Catchment Hydrology, Melbourne, Australia. http:rrwww.catchment.crc.org.aurproductsrindex.html Duncan H.P. Urban stormwater treatment by storage: a statistical overview. Cooperative Research Centre for Catchment Hydrology Report 97r1, Australia 1997, 85pp. Faas RW, Carson B. Short-term deposition and long-term accumulation of lagoonal sediment, Great Sound, New Jersey. Mar Geol 1988;82:97᎐112. Fennessy MS, Brueske CC, Mitsch WJ. Sediment deposition patterns in restored freshwater wetlands using sediment traps. Ecol Eng 1994;3:409᎐428. Hamilton DP, Mitchell SF. Wave-induced shear stresses, plant nutrients and chlorophyll in seven shallow lakes. Freshwater Biol 1997;38:101᎐110. Hakanson L, Floderus S, Wallin M. Sediment trap assem˚ blages ᎏ a methodological description. In: Sly PG, Hart BT, editors. Hydrobiologia. Sedimentrwater interactions IV. Kluwer Academic, Belgium, 1989:481᎐490. Hey DL, Kenimer AL, Barrett KR. Water quality improvement by four experimental wetlands. Ecol. Eng. 1994; 3:381᎐397. Jacobi D and Murphy S. Evaluation of an urban stormwater wetland. Unpublished Honours Research Report, Department of Civil & Environmental Engineering, The University of Adelaide, 1996, 167pp. Lloyd S. Influence of macrophytes on sediment deposition and flow patterns within a stormwater pollution control pond. Thesis for M.Eng.rSc., Dept Civil Engineering, Monash University, 1997, 161pp. Murphy SE, Daniell TM and Walker DJ. The dynamic effectiveness of a primary pond in a stormwater wetland. Hydrastorm ’98, Joint International Symposium on Stormwater Management and International Conference on Hydraulics in Civil Engineering, Adelaide, 1998: 445-450. Prescott KL, Tsanis IK. Mass balance modelling and wetland restoration. Ecol. Eng. 1997;9:1᎐18. Raisin GW, Mitchell DS, Croome RL. The effectiveness of a small constructed wetland in ameliorating diffuse nutrient loadings from an Australian rural catchment. Ecol. Eng. 1997;9:19᎐35. Reinelt LE, Horner RR. Pollutant removal from stormwater runoff by palustrine wetlands based on comprehensive budgets. Ecol. Eng. 1995;4:77᎐97.
68
D.J. Walker r The Science of the Total En¨ ironment 266 (2001) 61᎐68
Teisson C. Cohesive suspended sediment transport: feasibility and limitations of numerical modelling. J Hydraul Res 1991;29:755᎐769. Tsanis IK, Prescott KL, Shen H. Modelling of phosphorous and suspended solids in Cootes paradise marsh. Ecol Eng 1998;114:1᎐17. Walker D, Murphy S and Gamble S. Minkara wetland. Report
prepared for The City of Happy Valley, South Australia, 1997, 44pp. Walker D, Cadzow S, Watkins B. Flow and sediment dynamics in an artificial wetland, water down under 94. Joint IAH, IEAust, Vol 1994;3:429᎐432. Walker DJ. Modelling residence time in stormwater ponds. Ecol Eng 1998;10:247᎐262.
Modelling sedimentation processes in a constructed stormwater wetland D.J. Walker U Department of Ci¨ il & En¨ ironmental Engineering, The Uni¨ ersity of Adelaide, Adelaide 5005, Australia Received 17 September 1999; accepted 28 June 2000
Abstract The design and operation of constructed wetlands for the treatment of stormwater relies heavily on promoting sedimentation, and being able to predict accurately the expected effectiveness of the pond in removing material from the inflows. A study of sediment behaviour has been carried out in a stormwater wetland in Adelaide, Australia where computer predictions, based on solving the hydrodynamic equations and the transport equation, have been compared to deposition patterns observed in the field. The long-term residence time distribution has been shown to be useful in predicting overall sediment removal rates. Comparisons between the model and field observations indicate generally good agreement. Sources of potential error identified include the variable nature of the runoff concentrations entering the pond and the sediment size distribution. The importance of the transient nature of the flow events was highlighted by the spread of sediment throughout the whole pond. 䊚 2001 Elsevier Science B.V. All rights reserved. Keywords: Numerical model; Sedimentation; Stormwater; Constructed wetland
U
Tel.: q61-8-8303-4319; fax: q61-8-8303-4359. E-mail address: [email protected] ŽD.J. Walker.. 0048-9697r01r$ - see front matter 䊚 2001 Elsevier Science B.V. All rights reserved. PII: S 0 0 4 8 - 9 6 9 7 Ž 0 0 . 0 0 7 3 0 - 0
62
D.J. Walker r The Science of the Total En¨ ironment 266 (2001) 61᎐68
1. Introduction Sedimentation in ponds and wetlands is important, not only for removing the sediment itself, but also nutrients and contaminants which are readily attached to fine particles ŽFennessy et al., 1994; Raisin et al., 1997.. There have been a number of studies undertaken of sedimentation in wetlands both in situations where there was an essentially constant flow through ŽBrueske and Barrett, 1994; Fennessy et al., 1994; Hey et al., 1994; Prescott and Tsanis, 1997; Tsanis et al., 1998. and where the main flows were due to runoff from storms ŽReinelt and Horner, 1995; Raisin et al., 1997; Lloyd, 1997; Murphy et al., 1998.. For the latter, the intermittent nature of stormwater wetlands leads to a fundamentally different situation where sediment is more likely to be distributed around the entire basin ŽFennessy et al., 1994; Walker et al., 1994.. There have been a number of attempts in the past to model wetland sedimentation. Some have related outlet concentration to inlet concentration Že.g. Reinelt and Horner, 1995.. A simple approach that has been taken is to relate sedimentation rates to the distance from the inflow point ŽFennessy et al., 1994. but this ignores the two-dimensional nature of the flow patterns. The most comprehensive model to date is that reported by Tsanis et al. Ž1998. who solved the
hydrodynamic and sediment transport equations to predict water quality parameters in a natural wetland. The model was used to predict average sediment concentrations in the wetland and, after adjustment of settling velocities and external loadings, gave results that agreed reasonably well with field observations. In the present work a field study investigating sedimentation in a stormwater wetland has allowed verification of a computer model that solved the hydrodynamic and sediment transport equations to enable the overall removal efficiency and the resulting sedimentation patterns to be predicted. A particular focus of the work concerned the residence time distribution ŽRTD. of the pond.
2. Study site The site of the study was the Minkara stormwater wetland located approximately 20-km south of Adelaide in South Australia. The wetland was formed in 1992 with the construction of a concrete weir across a natural drainage channel at a point where it passed through a twin pipe culvert. No artificial planting was undertaken but after 7 years the pond has been infiltrated naturally by several species of macrophytes with the dominant being T. Orientalis. As is evident from Fig. 1 the majority of the plants have developed around the
Fig. 1. Minkara wetland, showing areas of macrophyte that have established naturally around the perimeter of the pond, and the location of sediment traps used in the study.
D.J. Walker r The Science of the Total En¨ ironment 266 (2001) 61᎐68
periphery of the wetland where the water depths are most favourable. The wetland has a volume of 2700 m3 and covers an area of 3500 m2 . The catchment has an area of 130 ha. leading to an area ratio of approximately 0.25%. A survey of the wetland has indicated an average depth of 0.77 m and a maximum depth of 1.40 m. For modelling purposes the wetland has been considered flat bottomed since the maximum depth never exceeds twice the mean depth ŽHamilton and Mitchell, 1997.. An analysis of the sediment deposited on the bottom of the wetland since construction ŽJacobi and Murphy, 1996. indicated a mean particle size of 15 m.
63
bottles were changed. The data demonstrated the rapid build-up of turbidity during a storm event, but no attempt was made to relate this to the sediment in the traps. The sediment in the bottles was filtered onto pre-weighed Whatman GFrC filter papers within 1 h of collection and dried in an oven at 90⬚C overnight. Long-term sedimentation rates were converted to cmrday by dividing them by the product of the bulk density Ž1.5 grcm3 . multiplied by the sediment trap opening area Ž8.0 cm2 ., ŽFaas and Carson, 1988..
4. Numerical models 3. Methods and techniques The field study was undertaken in two parts. At the start of winter 1998, coarse sand horizon markers were installed at five locations Žsee Fig. 1. throughout the pond. One year later, sediment cores were taken using a Perspex corer with a diameter of 40 mm. The depth of buildup was measured, and the samples were subsequently dried and analysed for organic content by heating to 500⬚C for an hour after being milled and dried at 110⬚C for 2 h. The average organic content was found to be 15%. During June to September 1999, sediment traps were installed at the same locations and monitored on a regular basis. The sediment collection bottles were made from high density polyethylene ŽHDPE. with an opening of 32 mm, a depth of 128 mm and a volume of 250 ml. The aspect ratio of 4 was selected based on the recommendations of Hakanson et al. Ž1989., ˚ Fennessy et al. Ž1994.. The traps were located on poles in the wetland and were retrieved at least fortnightly by canoe so as not to disturb the area. Replacement bottles were filled with distilled water and frozen prior to installation in the wetland to reduce the chance of any disturbance caused by manipulating the traps ŽBrueske and Barrett, 1994; Fennessy et al., 1994.. Greenspan turbidity sensors operating at 860 nm wavelength and recording data every 5 min were located with three of the traps Ž2, 4 and 7.. Data were downloaded using a laptop computer when sediment
The numerical simulation of sedimentation in the wetland was undertaken using a suite of numerical models, the first to solve the hydrodynamics of the pond during inflow events, the second to solve the sediment transport and deposition, and the third to predict the runoff events over the period under study. The relatively shallow depth and the assumption that stratification would not be significant, especially during flow events, led to the selection of a two-dimensional depth-averaged model for the flow simulations. The transient form of the equations was chosen so that the development of flow patterns in time could be described. Details of the model can be found in Walker Ž1998., although many other researchers Že.g. Tsanis et al., 1998. have used similar models. Rather than try to model 5 years of continuous records, a selection of 11 typical events of varying size Žfrom 10 to 400% of pond volume. and duration Žfrom 5 to 13 h. were run and the predicted flow patterns were stored for later use with the transport model. The sediment transport was modelled using a two-dimensional horizontal approximation to the transport equation where a source term included the process of deposition. A full three-dimensional model would have been preferred but the fine nature of the sediment Ž d50 s 15 m. meant that concentration variations through the depth could be ignored. Deposition at each time-step was assumed to be based on the product of particle fall velocity and the mean concentration. The
64
D.J. Walker r The Science of the Total En¨ ironment 266 (2001) 61᎐68
Fig. 2. Flow chart for numerical models used in study.
low flow velocities in the pond meant that erosion of deposited material was assumed not to occur. The model is based on those reported by Tsanis et al. Ž1998., Teisson Ž1991.. The model used a fixed inflow concentration and was run for each of the flow simulations outlined above. The inflow concentration was selected based on observations over a number of years. Although a simplification of the actual behaviour, it was not possible to derive a simple relationship that could be used to predict inflow concentration based, for example, on the inflow rate. The output from the model was a predicted rate of sedimentation over the entire solution grid. Once again, results for each of the events were stored for later use. The runoff from the catchment was modelled using a computer simulation package AWBM ŽCRCCH, 2000. based on original work by Boughton Ž1984.. The model represents the catchment processes using a number of storages that must be calibrated. The calibration was carried out using 2 months of data collected during the study and then tested on a separate 5-week section. The root mean square error was found to be 1.33 mm of runoff per day. The final prediction of sedimentation rates was determined by running the runoff model for a 5 year simulation and summing the number of events based on size, using the 11 predetermined sizes developed for the flow modelling. Over the 5
years there were, for example, 78 where the volume was 10% of pond volume, 42 where the volume was 20%, 18 of 40%, gradually reducing to five events where the inflow volume was 300% of pond volume. Using the results of the transport model runs the total sediment accumulation was calculated by summing the accumulation for each size event multiplied by the number of events predicted for the simulation. A flow chart for the programs is shown in Fig. 2.
5. Results The main result of the study is shown in Fig. 3 where the numerical predictions of annual sedimentation are shown. Superimposed on the contours are the observed sedimentation rates from both the long-term and short-term study. The concentration of material near the inlet is evident, but the spread into areas some distance from the inlet is also predicted to occur. Based on the results of the field study the estimates are considered reasonable despite the general overprediction of the rates by the model. The most likely explanation is an overestimate of the inflow concentrations. The relative sedimentation rates between the stations are in fair comparison with the observed field rates. The even pattern of sedimentation is partly due
D.J. Walker r The Science of the Total En¨ ironment 266 (2001) 61᎐68
65
Fig. 3. Contours of predicted annual sedimentation in Minkara wetland based on 5 year simulation. The figures in the boxes are from the field study where the first number is the sediment buildup over a full year, and the second is based on a 3-month short-term study. All figures are in millimeters.
to the fact that the variations in the bottom contours and the effect of vegetation have been ignored. Despite this the results are useful. It should be remembered when comparing the two estimates of sedimentation rate that, as pointed out by Faas and Carson Ž1988., sedimentation traps give estimates of the potential sedimentation rate rather than the actual. In this case it would be expected that the short-term rates would give an overestimate of the true rate.
As an additional aspect to the present study a consideration of the use of the residence time distribution ŽRTD. in predicting overall sediment removal rates was undertaken. Following the work of Walker Ž1998. a RTD for Minkara wetland was calculated taking account of the interval between events and the short-circuiting in the pond due to the less-than-ideal flow patterns during events ŽFig. 4.. If the sedimentation process is assumed to depend purely on the time available for set-
Fig. 4. Long-term residence time distribution for Minkara wetland based on 5 year simulation Ž1994᎐1998.. The mean residence time is 12 days, but a significant portion is resident for less.
D.J. Walker r The Science of the Total En¨ ironment 266 (2001) 61᎐68
66
tling, the removal of sediment is a function of the residence time alone. In this case the removal efficiency can be predicted directly. Given that all residence times longer than the fall velocity will result in complete removal, and shorter times will result in a proportional loss, the removal Ž E . can be determined as:
Es
t max
tfall
Ý
f Žt. q Ý
tfall
0
t f Žt. t fall
tmax
= 100
Ž1.
Ý f Žt. 0
where E is the percentage removal, f Ž t . is an element of the RTD, t fall is the time in seconds for a particle to fall through the depth of the pond Ž0.77 m. and t max is the maximum residence time in seconds from the RTD. The efficiency could be calculated based on a single particle size or by summing over a number of size fractions. For the present wetland both gave a predicted removal efficiency of 50%. This is of the same order as the observed figure ŽWalker et al., 1997. of 60᎐70%. The underestimation is likely due to the problems associated
with using a daily timestep to represent flow events that are often of much shorter duration.
6. Discussion Predictions made on the basis of numerical modelling of stormwater ponds have a number of uses. In the design of these facilities it is possible at present to estimate overall removal efficiencies using such factors as the area of the pond relative to the catchment and the incoming sediment concentrations ŽDuncan, 1997.. The present methodology is unique in that it can be used to optimise the shape of the ponds, the location and size of islands designed to improve flow patterns, and to assess the effect of the hydrological inflow characteristics on pond performance. An example of this is shown in Fig. 5 where islands have been incorporated into the design to reduce short-circuiting and improve retention. When this pond layout was assessed using the same procedures described above the predicted removal efficiency was 52% Žcompared to 50% for the initial design.. The reduced pond volume caused by the inclusion of the islands was more than offset by the superior flow patterns Ži.e. islands reduce the flow rate
Fig. 5. Predicted flow patterns for modified stormwater pond system incorporating islands to optimise retention.
D.J. Walker r The Science of the Total En¨ ironment 266 (2001) 61᎐68
and increase retention time.. While the predicted difference is quite small, the important aspect of the exercise is the ability to make a quantitative assessment of the behaviour and comparisons based on the behaviour. It is noted that the present study has made a number of simplifying assumptions. The constant depth model allowed two-dimensional models to be used that are easily developed and efficiently run. The fine nature of the sediment was also crucial to allowing a two-dimensional analysis of the transport of sediment in the pond. A daily timestep in the rainfall runoff model was able to use relatively crude hydrological input, and this is certainly the first area that should be upgraded by employing an hourly timestep. A detailed sensitivity analysis was not undertaken as part of the study but experience with the models show that the characteristics of the inflow sediment concentration and particle sizes would be the most critical in terms of accuracy. Field data were available to allow estimates to be made, but these were crude and did not take account of the variations that could be expected due to inter-event times, rainfall intensity and catchment characteristics.
7. Conclusion Models have been developed to allow the prediction of overall sediment removal rate as well as the likely sedimentation patterns in a stormwater wetland. The models solve the transient flow field that develops in the wetland during events, the resulting transport and deposition of sediment in the wetland and also predict the time history of events based on daily rainfall records. Comparison with results of a complementary field study has indicated that the model predictions are reasonable and that the modelling would be useful in the design of stormwater facilities. The models allow a more detailed analysis of pond shape and features to be undertaken than is possible at present where crude parameters such as relative area or inflow concentrations are unre-
67
sponsive to pond features where the designer could have some input. References Boughton WC. A simple model for estimating the water yield of ungauged catchments. Institution of Engineers, Australia. Civil Eng Trans; C 1984;E26:83᎐88. Brueske CC, Barrett GW. Effects of vegetation and hydrologic load on sedimentation patterns in experimental wetland ecosystems. Ecol Eng 1994;3:429᎐447. CRCCH Ž2000.. AWBM Water Balance Model. Cooperative Research Centre for Catchment Hydrology, Melbourne, Australia. http:rrwww.catchment.crc.org.aurproductsrindex.html Duncan H.P. Urban stormwater treatment by storage: a statistical overview. Cooperative Research Centre for Catchment Hydrology Report 97r1, Australia 1997, 85pp. Faas RW, Carson B. Short-term deposition and long-term accumulation of lagoonal sediment, Great Sound, New Jersey. Mar Geol 1988;82:97᎐112. Fennessy MS, Brueske CC, Mitsch WJ. Sediment deposition patterns in restored freshwater wetlands using sediment traps. Ecol Eng 1994;3:409᎐428. Hamilton DP, Mitchell SF. Wave-induced shear stresses, plant nutrients and chlorophyll in seven shallow lakes. Freshwater Biol 1997;38:101᎐110. Hakanson L, Floderus S, Wallin M. Sediment trap assem˚ blages ᎏ a methodological description. In: Sly PG, Hart BT, editors. Hydrobiologia. Sedimentrwater interactions IV. Kluwer Academic, Belgium, 1989:481᎐490. Hey DL, Kenimer AL, Barrett KR. Water quality improvement by four experimental wetlands. Ecol. Eng. 1994; 3:381᎐397. Jacobi D and Murphy S. Evaluation of an urban stormwater wetland. Unpublished Honours Research Report, Department of Civil & Environmental Engineering, The University of Adelaide, 1996, 167pp. Lloyd S. Influence of macrophytes on sediment deposition and flow patterns within a stormwater pollution control pond. Thesis for M.Eng.rSc., Dept Civil Engineering, Monash University, 1997, 161pp. Murphy SE, Daniell TM and Walker DJ. The dynamic effectiveness of a primary pond in a stormwater wetland. Hydrastorm ’98, Joint International Symposium on Stormwater Management and International Conference on Hydraulics in Civil Engineering, Adelaide, 1998: 445-450. Prescott KL, Tsanis IK. Mass balance modelling and wetland restoration. Ecol. Eng. 1997;9:1᎐18. Raisin GW, Mitchell DS, Croome RL. The effectiveness of a small constructed wetland in ameliorating diffuse nutrient loadings from an Australian rural catchment. Ecol. Eng. 1997;9:19᎐35. Reinelt LE, Horner RR. Pollutant removal from stormwater runoff by palustrine wetlands based on comprehensive budgets. Ecol. Eng. 1995;4:77᎐97.
68
D.J. Walker r The Science of the Total En¨ ironment 266 (2001) 61᎐68
Teisson C. Cohesive suspended sediment transport: feasibility and limitations of numerical modelling. J Hydraul Res 1991;29:755᎐769. Tsanis IK, Prescott KL, Shen H. Modelling of phosphorous and suspended solids in Cootes paradise marsh. Ecol Eng 1998;114:1᎐17. Walker D, Murphy S and Gamble S. Minkara wetland. Report
prepared for The City of Happy Valley, South Australia, 1997, 44pp. Walker D, Cadzow S, Watkins B. Flow and sediment dynamics in an artificial wetland, water down under 94. Joint IAH, IEAust, Vol 1994;3:429᎐432. Walker DJ. Modelling residence time in stormwater ponds. Ecol Eng 1998;10:247᎐262.