Chapter 5 : Design calculations

Transcription

1 Chapter 5 : Design calculations 5. Purpose of design calculations For design calculations, it is important to assess the maximum flow, maximum water level and maximum velocity which occur in every node or pipe with a certain return period. Therefore, a detailed model including all (relevant) pipes is necessary. However, the simulation of long time series using detailed models is extremely time consuming. Even if the pipe network is simplified to only the relevant pipes, a maximum reduction in calculation time by maybe a factor of 2 can be achieved. The simplification of the rainfall input is the remaining alternative. The simplification of the long rainfall series to single storm events with a certain return period leads to (relatively) fast calculations. In that case it is generally assumed that the return period of an effect (water depth, flow, etc...) is equal to the return period of the storm which causes this effect. This assumption of linearity is most often acceptable for design calculations because only peak discharges and maximum levels are important. The possible non-linear behaviour of sewer systems will not have a large influence for high return periods and short storm durations. For these extreme and short events the hydraulic performance will not be affected very much by the antecedent rainfall conditions and the variability of the rainfall. However, if the antecedent conditions become more important and the storm durations increase, the temporal variability should be incorporated. Furthermore, the spatial variation of the rainfall may also be important. Finally, it is important to note that design calculations do not only involve simulation software and rainfall input. Other aspects have to be taken into account such as boundary conditions, design criteria, accuracy of input data and the interpretation by the design engineer. In this chapter the different aspects involved in design calculations will be investigated with great emphasis on the influence of the rainfall and the correlations with the design criteria. Chapter 5 : Design calculations 5.

2 5.2 Design calculations using single storm events 5.2. Rainfall input used in Flanders before 996 Before 987, a rainfall intensity of 60 l/s/ha (for flat areas) or 90 l/s/ha (for steep areas) for a duration of 5 minutes was used for design calculations [De Backer, 978]. In this guideline of De Backer it is stated that the return period in this case is year and that for problem areas a return period of 2 years should be used. As compared with the new IDF-relationships, a rainfall intensity of 60 l/s/ha over 5 minutes has a return period of about 0.3 years and a rainfall intensity of 90 l/s/ha over 5 minutes has a return period of about 0.8 years. For longer storm durations (i.e. longer concentration times) the use of this rainfall volume will lead to even lower return periods. In 987, new guidelines were written, in which the design return period was fixed at 2 years (5 years for problem areas) and the use of a critical storm duration was generalised [Berlamont, 987]. The rainfall intensities in these guidelines were based on the IDF-relationships by Demarée [985]. As compared with the new IDF-relationships (see paragraph 2..5) the rainfall intensities in these guidelines still lead to a small overestimation of the return period (underestimation of the flow). The resulting return periods are shown in figure 5.. For every storm duration and the corresponding rainfall intensity, the return period following the new IDF-relationships is presented in this figure. In these guidelines only rainfall intensities up to a storm duration of 60 minutes were given. For sewer systems with larger critical storm durations (i.e. concentration times) the use of a maximum storm duration of 60 minutes will lead to a larger overestimation of the return period. This is shown by the small dashed lines in figure 5.. These small dashed lines are obtained using the rainfall volume over the largest used storm duration of 60 minutes and spreading them out over the considered storm duration. In 992, with the use of Sphyda version 2.02, English storms were introduced (rainfall generator version.42 [WS, 992]) (see paragraph.4.3). When this rainfall was compared with the IDF-relationships of Demarée [985], a significant disagreement was found (however the disagreement with the new IDF-relationships is limited for high return periods). Therefore, these English synthetic storms were slightly calibrated to the IDF-relationships of Demarée and added to Spida version 2. (rainfall generator version.44, [WS, 994]). The use of different storm types with different origin has an impact on the design return period. Again, the return period of previously used rainfall can be determined based on the new IDF-relationships. In figure 5.2, a comparison of the return period is made for the different design storms, which have been in use in Flanders since The influence of rainfall and model simplification on combined sewer system design

3 7 return period (years) guidelines 987, T = 5 years composite storm, T = 5 years 3 2 composite storm, T = 2 years guidelines 987, T = 2 years storm duration (min) Figure 5. : Return periods for the rainfall input corresponding to the guidelines of 987 (storm durations up to 60 minutes) as compared with the composite storms for 2 and 5 years (the effect for larger concentration times (i.e. storm durations longer than 60 minutes) on the return period is illustrated by the small dashed lines) return period T (year) Sphyda 2.02 T = 5 years Spida 2. T = 5 years composite storm T = 5 years Sphyda 2.02 T = 2 years Spida 2. T = 2 years composite storm T = 2 years storm duration (min) Figure 5.2 : Return periods for the different rainfall input used for combined sewer system design in Flanders since 992 as compared with the composite storms for 2 and 5 years. Chapter 5 : Design calculations 5.3

4 5.2.2 Combination of previously used software and rainfall Since each new type of design storm was introduced together with a change in simulation software, the global effect on the design must be studied. It was found that the return periods of previous designs were overestimated as compared with the current technology (i.e. the sewer pipes were underdesigned). For the Flemish design practice, it can roughly be concluded that a design which was made between 992 and 994 (with the idea to design it) for a return period of 5 years, only has a return period of 3 years. The main cause of this is the underestimation of the piezometric level by the software when the pipes get pressurised as described in paragraph.2.. These results were obtained by comparing the maximum water levels and peak flows in all nodes and pipes of two complex sewer systems for which the different combinations of software and rainfall input were applied [Vaes & Berlamont, 998a]. For a design between 994 and 996 this return period is about 4 years (for an original return period of 5 years), while the main cause in this period is the rainfall input, as described in paragraph.4.3. and illustrated in figure 5.2. This overestimation of the return period appears to be important, but it is not significantly larger than the uncertainty on the design. In figures 5.3 and 5.4 the influence of the water input into the sewer system (caused by rainfall, impervious area or runoff coefficient) on the return period is shown. In figure 5.3 the effect of an underestimation of the rainfall input into the sewer system with 0 % is shown. The relationship between the underestimation of the rainfall intensity and the return period is based on the new IDF-relationships which are described in paragraph return period T (year) T = 0 years T = 5 years T = 2 years storm duration (min) Figure 5.3 : Effect on the return period of an underestimation with 0 % of the water input (impervious area, rainfall intensity, losses, ) into the combined sewer system(for the composite storms with a return period of 2, 5 and 0 years). 5.4 The influence of rainfall and model simplification on combined sewer system design

5 An underestimation of the water input of 0 % already has a significant effect on the design. Moreover, an increase in water input into the sewer system often results in a more than linear increase of the piezometric levels, certainly when pressurised flow occurs in many pipes (head losses increase quadratically with the discharge in pressurised pipes). The piezometric level rises especially in these nodes where the level is critical already. This also means that in the previous designs, several critical nodes may not have been discovered. This uncertainty, which is inherent to the sewer system data (as well on rainfall input as on the parameters of the runoff model, e.g. impervious area and runoff coefficient), can lead to an extra overestimation of the return period in specific pipes. In figure 5.4, the effect of an overestimation of the water input into the sewer system of 0 % on the return period is shown return period T (year) T = 5 years T = 2 years storm duration (min) Figure 5.4 : Effect on the return period of an overestimation with 0 % of the water input (impervious area, rainfall intensity, losses, ) into the combined sewer system (for the composite storms with a return period of 2 and 5 years) Design calculations using composite storms The composite storms, as they have been derived in this work (see paragraph 2.3), include the mean rainfall characteristics, but neglect the intrinsic temporal rainfall variability. In order to assess the ability of these composite storms to predict the real discharges and piezometric levels in a sewer system, a comparison has been made between (full hydrodynamic) simulations with the composite storms on the one hand and long rainfall series on the other hand. To limit the calculation time this has only been done for a few small sewer systems. The rainfall input used for the continuous Chapter 5 : Design calculations 5.5

6 simulations is the rainfall series (of Uccle) of 27 years, which has also been used for the development of the composite storms (see paragraph 2.3). In figures 5.5 till 5.7, the simulation results are shown, which have been obtained for different pipes in three small sewer systems (i.e. 50 to 200 nodes). The lines show the relationship between flow and return period obtained with the composite storms, while the markers show these relationships for the continuous simulations. The relationships obtained with the composite storms show a very smooth line, because these composite storms are based on IDF-relationships which have been obtained by an extreme value analysis (see paragraph 2..5). The results obtained with the continuous simulations are the simulation results without any extreme value estimation or regression. For the figures 5.5 till 5.7 no independency criterion has been used to distinguish two effects (i.e. flow) above a certain threshold, which occur shortly after each other. However, the analysis for the IDF-relationships showed that the influence of dependent effects is very small for high return period. This has been confirmed by analysing the results of the continuous simulations using an independency criterion of a minimum of 2 hours between the events, which showed only small differences for the lower flows. 00 return period [year] 0 0. pipe ; continuous simulation pipe 2; continuous simulation pipe 3; continuous simulation corresponding simulation with composite storms flow [m 3 /s] Figure 5.5 : Comparison between the results of continuous simulations with a rainfall series of 27 years (markers) and the results of simulations with composite storms (lines) for three downstream pipes in a small sewer system with a gravitary outflow. 5.6 The influence of rainfall and model simplification on combined sewer system design

7 00 return period [year] 0 pipe ; continuous simulation 0. pipe 2; continuous simulation corresponding simulation with composite storms flow [m 3 /s] Figure 5.6 : Comparison between the results of continuous simulations with a rainfall series of 27 years (markers) and the results of simulations with composite storms (lines) for two downstream pipes in a small sewer system with a single pump at the outflow. 00 return period (year) 0 0. continuous simulation simulation with composite storms flow (m 3 /s) Figure 5.7 : Comparison between the results of continuous simulations with a rainfall series of 27 years (marker) and the results of simulations with composite storms (line) for a downstream pipe in a small sewer system with a single pump at the outflow. Chapter 5 : Design calculations 5.7

8 The results in figure 5.5 are those for a small sewer system with a gravitary outflow, which behaves linearly. The results obtained with the composite storms are very close to those obtained with the continuous simulations. Figures 5.6 and 5.7 show the results for two small sewer systems with a single pump at the downstream end. These sewer systems have a very non-linear relationship between storage in the sewer system and outflow. It is clear from figures 5.6 and 5.7 that the results obtained with the composite storms are not as close in agreement as those obtained with the continuous simulation as compared with the results in figure 5.5. From the large fluctuation on the results obtained with the continuous simulations, it is clear that the large uncertainty on the rainfall input also leads to a large uncertainty on the results, certainly for high return periods. Using the composite storms, this uncertainty will partially be covered up. For the three tested (small) combined sewer systems the results obtained with the composite storms were always on the safe side, i.e. the flow is slightly overestimated. However it is not clear from the limited number of test cases if this is a coincidence or a systematic difference The influence of the storm duration Increasing the storm duration (for design calculations in Flanders) from 20 minutes (before 996 [Aquafin, 994a]) to 360 minutes (using the composite storms) has only an effect on the piezometric levels at the downstream end of a large system. The piezometric level will not rise in the upstream critical nodes, but at the downstream end of the system the piezometric level can create a more critical situation in some nodes or some new critical nodes may be observed. Before 996, for each storm duration a specific storm had to be used. This led to a practice where only one critical storm duration was taken into account in order to limit the results processing and to cope with the problem of simultaneous visualisation of the results from different simulations. Mostly the critical storm duration for the most downstream part was therefore used, but sometimes even a mean critical storm duration was used. However, the critical storm duration is different for each part of the sewer system. Upstream parts can be associated with small critical storm durations, while downstream parts can be associated with longer critical storm durations. When using a different storm for every storm duration, the maximum water levels and peak flows in every node or pipe must be taken as the maximum that occurs in one of the simulations for the whole range of storm durations. This problem has been solved since the composite storms were introduced in 996 [VMM, 996]. 5.8 The influence of rainfall and model simplification on combined sewer system design

9 5.3 Including temporal rainfall variability 5.3. Need for temporal rainfall variability The comparison between the simulation results for continuous simulations and simulations with composite storms indicates that significant differences may be found when the intrinsic variability of the rainfall is neglected. The differences will be small for sewer systems which behave linearly, because the immediate rainfall determines the peak flow and maximum water levels. When the systems start to behave more as capacitive systems, i.e. where the storage becomes an important parameter, the differences will be larger. A capacitive system has a memory which remembers the antecedent rainfall. Often sewer systems have an emptying time, which tends towards 2 hours. If a severe storm occurs within a short period after an earlier storm, the antecedent rainfall may still occupy a large amount of the storage capacity in the combined sewer system. The larger the influence of the memory is, the larger the intrinsic variability of the rainfall will influence the simulation results. For example, for a sewer system in a flat region with one pump at the downstream end, the throughflow is almost independent of the storage volume. The storage volume in the system is therefore mainly dependent on the inflow. More and more capacitive sewer systems have been built recently and will still be built in future. The best combined sewer system, in order to obtain an optimal load into the treatment plant, has a throughflow which is as small and as constant as possible. Therefore, large storage volumes are necessary to retain the rainfall and to attenuate the flow. These storage volumes can be built in the sewer system (on-line storage) or at the combined sewer overflow (off-line storage). However, more and more attention is going to source control, certainly in recent years [VMM, 996, 999]. This means that storage is provided in rain water tanks, infiltration trenches, etc... For these source control implementations the influencing antecedent rainfall period is even larger. This requires larger storage volumes (relative to the contributing area) than for downstream storage [Vaes & Berlamont, 998c, d, e, 999]. This all amplifies the need to take into account the intrinsic variability of the rainfall for specific design calculations Modified single storm events When storage is built in upstream of the combined sewer system, i.e. before the rain water enters into the sewer pipes, the rainfall input can be preprocessed in order to take into account this storage. This can be done as described in paragraph 2.3.5, where the effect of rain water storage tanks has been incorporated in modified (composite) design storms in the program Rewaput. These local source control implementations are easy to model with a simple reservoir model, which can handle Chapter 5 : Design calculations 5.9

10 continuous long time simulations in a very short computation time. This methodology can also be used to incorporate the effect of a non-linear surface runoff model or to simulate the effect of infiltration facilities, even if they are influenced by the ground water table. Currently, for sewer system design a fixed runoff coefficient of 0.8 is used for the impervious area. If (long term) measurements were available, a more realistic runoff model could be calibrated and used, including a more capacitive (depression storage based) runoff model (e.g. [Vaes et al., 998c]). For this runoff model the long term behaviour could then be included in the design calculations by routing long rainfall series through the simple conceptual runoff model and incorporate the effect in the design storms. The same is true for infiltration facilities and runoff from pervious areas. Simple conceptual models can be used to reshape the design storms, so that simple design storms are obtained without neglecting the effect of the rainfall variability on the infiltration facilities. Infiltration and retention facilities often behave non-linearly, because the outflow is often very strictly limited. Continuous long term simulations are thus necessary. The implementation of a simple conceptual model for the upstream retention facilities is simple and the simulation of long time series in this conceptual model does not require long calculation times. One set of parameters for the Rewaput model requires only about 5 seconds of calculation time on a Pentium II 233 MHz computer. If the parameters vary over a specific catchment, the parameter distribution can be discretised and several sets of parameters can be taken into account. In that case the discretisation step, the deviation on the parameters and the number of varying parameters determine the number of calculations which have to be performed. In the model Rewaput, a triangular distribution is implemented to approximate the stochastic character of storage volume and water consumption (i.e. variation over a catchment) [Vaes & Berlamont, 998b]. For each variation within this triangular distribution the effect is multiplied by the weight corresponding to the parameter distribution in order to calculate the global effect. Using two stochastic parameters the calculation time quadratically increases. To reduce the calculation time the discretisation has to be chosen taking into account the deviation on the parameters. Although the storage in rain water tanks and infiltration facilities is not always completely available during severe rainfall, i.e. because the storage is already filled with the antecedent rainfall, this kind of upstream retention facility still can have a large influence on the rainfall runoff to the sewer system. It has been shown that well-designed rain water tanks can even significantly reduce the peak flow in sewer systems, if they are installed on a large scale [Vaes & Berlamont, 998b]. In figure 5.8, an example is shown of what the possible effect of rain water tanks on a design storm can be. In this case (figure 5.8), it is assumed that rain water tanks of 5000 l per 00 m 2 roof area are built for 30 % of the total impervious area and that 00 l per day and per 00 m 2 roof area rain water is consumed. This almost reduces the peak of the composite storm for 5 years to the value of the composite storm for year (figure 5.8). 5.0 The influence of rainfall and model simplification on combined sewer system design

11 rainfall intensity (mm/h) T = 2 year : 52.4 mm/h T = year : 4.9 mm/h design storm T = 5 year tank volume = 5000 l/00m 2 rain water use = 00 l/day/00m 2 30 % connected area modified composite storm original composite storm time (min) Figure 5.8 : Effect of the installation of rain water tanks on a large scale on the composite storm for a return period of 5 years. This example shows the effect if 30 % of the impervious areas is connected to rain water tanks with a volume of 5000 l per 00 m 2 roof area and a rain water use of 00 l per day and per 00 m 2 roof area Design calculations using short selected rainfall series If retention or infiltration facilities are built into the combined sewer system, the approach using modified single storm events (as described in paragraph 5.3.2) is not applicable anymore. In recent years the revaluation of ditches is promoted for rain water runoff in Flanders [VMM, 996, 999]. This means that the transport function at one side and the retention and infiltration function at the other side are combined. To obtain a good evaluation of these types of storm water runoff, the intrinsic rainfall variability must be retained. The same is valid for non-linearly behaving sewer systems, which is often the case when large in-line storage is used and the throughflow is strongly limited. This can also be the case when in-line storage has to be designed in order to prevent flooding from the system. In this case the rainfall variability can only be retained by selecting the specific short rainfall series (SSRS) out of the original long rainfall series as has been described in paragraph 4.2 (based on IDF-relationships). Besides the fact that the appropriate SSRS have to be selected, the reduction of the calculation time is the most important criterion in the judgement on the applicability of the SSRS. The design return period, the maximum storm duration and the necessary antecedent rainfall period have a large Chapter 5 : Design calculations 5.

12 influence on the reduction of the rainfall input. Therefore, these parameters have to be limited. This leads to a certain domain of applicability of these SSRS. As combined sewer system design calculations are always carried out for return periods starting from 2 years (minimum year for storm water systems), it is no problem to limit the return period (i.e. selecting only the events with a return period higher than e.g. year). The number of SSRS must however be large enough to be able to process the results in a statistically acceptable way. Also the maximum storm duration can be limited without hindering the applicability, e.g. to 360 or 720 minutes, because rarely these high concentration times occur. The remaining parameter is the antecedent period, which can increase to a very long period. In the Flemish guidelines it is stated that the maximum emptying time of a combined sewer system should be 2 hours [VMM, 996], but often this can be higher. Certainly for storm water sewers (where there is no strict limit for the emptying time in order to prevent septic reactions) and when more retention capacity is built in into the sewer system, the emptying time can increase to periods longer than 2 hours. However, the current applicability in practice imposes an upper limit for the antecedent period of 2 to 24 hours. The combination of IDF-relationships for the selection of short rainfall series for design applications and the emptying time for the corresponding antecedent period, does not guarantee that the selection is perfect. As explained in paragraph 4.4, the use of a reservoir model for the selection of the antecedent period could take into account larger antecedent periods without increasing significantly the length of the rainfall series or even further reducing (the antecedent periods for) the SSRS and thus the calculation time. A specific reduction of the rainfall input for a sewer system simulation will not lead to a proportional reduction in calculation time, because more iterations are necessary when the flow in the system becomes more complicated. To get an idea about the calculation times, tests were performed on three sewer systems using a set of SSRS for a minimum return period of 2 years, a maximum storm duration and a maximum antecedent period of 2 hours. The results, obtained with Hydroworks version 4.0 [WS, 998a] on a Pentium II 233 MHz computer, are shown in table 5.. It is clear from the results in the table that the complexity and the magnitude of the sewer system have a large influence on the calculation times. However, for normal and small sewer systems the methodology, using SSRS, reaches a stage of practical applicability. In the future, computers and software will become faster, which offers good perspectives for this methodology. However, this methodology will never replace design storms, because of the trial and error nature of design calculations, but will be useful to perform a check simulation following the design stage. As compared with the use of the composite storms, the use of SSRS results in a relative reduction in calculation time with a factor 2 to 3 per day of rainfall input. 5.2 The influence of rainfall and model simplification on combined sewer system design

13 sewer system Table 5. : Calculation times for a set of selected short time series with a total duration of 5 days for three different sewer systems. number of nodes total calculation time calculation time per day rainfall input calculation time per day rainfall input and per 000 nodes complex, different pumping stations including treatment plant simple, single pump hours 8.0 minutes 6.8 minutes hours 3.3 minutes 9.7 minutes 57 8 minutes 0.36 minutes 6.3 minutes A major limitation at this moment for the use of selected short time series (and continuous time series simulations) is the statistical processing of the results. It is not possible to perform a statistical analysis on a long time series of results or on a set of results obtained from simulations using a set of short time series with the (in Flanders) currently used software Hydroworks [WS, 998a]. It is not only necessary to determine the distributions of different hydraulic parameters, but also to perform an extreme value analysis on the results, because of the large variability of the phenomena for high return periods. As discussed in paragraph the system behaviour can change the type of extreme value distribution. This means that the type of extreme value distribution for the effect may be different from the one for the rainfall input. The new possibilities in Infoworks [WS, 998c] make the statistical processing easier. In Infoworks the percentage of exceedance can be determined for a certain threshold or the threshold for a certain percentile can be extracted. However, it is not possible to obtain a full distribution of the results or to apply an extreme value analysis. The second (longer) time step that can be chosen during dry weather flow when the simulations are not critical and no results are required, can boost continuous simulations. However, only instantaneous parameters can be chosen to control the use of this second time step during the insignificant periods. Parameters which describe cumulative effects (e.g. volume) are however often more important to select the significant parts from long time series. Therefore the selection of short time series based on IDF-relationships, as described in paragraph 4.2, is preferred. Chapter 5 : Design calculations 5.3

14 5.4 Spatial rainfall variation 5.4. Areal reduction coefficient In the past, often an areal distribution coefficient has been applied in Flanders for sewer system design in order to incorporate the spatial variation of the rainfall within a storm [Berlamont, 987; VMM, 996]. This areal reduction coefficient was based on the work of Frühling, who measured at Breslau (actually Wroclaw, Poland) that the rainfall intensity at 3 km from the centre of the storm was halved and was negligible at 2 km [Mennes, 90]. Frühling proposed a parabolic spatial distribution of the rainfall within a storm (figure 5.9) : i x i 0 = x (5.) in which i x is the rainfall intensity at a distance x of the centre of the storm, where i 0 is the maximum rainfall intensity. Figure 5.9 : Spatial rainfall distribution proposed by Frühling [Mennes, 90]. If this spatial distribution is integrated over a circular area, an areal reduction coefficient R A is obtained (table 5.2) : ψ A = L (5.2) in which L is the diameter of a circular area (in [m]) over which the rainfall is integrated. Although Frühling based his work on (a limited number of) experiments, the (rather arbitrary) assumption of a parabolic distribution is not in accordance with 5.4 The influence of rainfall and model simplification on combined sewer system design

15 what was recently found in the analysis of the rainfall from the dense pluviometric network of Antwerp [Luyckx & Herftijd, 996; Houbrechts & Truyen, 997; Luyckx et al., 998d; Willems, 998a, 999; Willems & Berlamont, 998a]. From this rainfall data of Antwerp, it could be concluded that the spatial distribution is more likely to be Gaussian and that the mean value for the standard deviation F using this Gaussian distribution is about 2.5 km : ψ A i x i 0 = x 2 exp 2 2 σ Integrated over a circular area, this leads to an areal reduction factor R A of (table 5.2) : (5.3) 8 σ 2 L 2 = exp L 2 8 σ 2 (5.4) Table 5.2 : Areal reduction coefficients R A as a function of the diameter L and the area of the circular region over which the rainfall is integrated. diameter L (m) circular area (ha) R A Frühling R A Gaussian Chapter 5 : Design calculations 5.5

16 When this Gaussian spatial rainfall distribution is compared with the proposed parabolic distribution (figure 5.0), it can be concluded that the influence of the spatial variation of the rainfall on the total rainfall volumes becomes important only starting from larger areas than assumed by Frühling. The results of the integration of the Gaussian distribution are shown in table 5.2. The comparison between the parabolic and the Gaussian variation may lead to the conclusion that for small catchments the rainfall input will be underestimated using the Frühling coefficient, while for large catchments the influence of the spatial variation of the rainfall is even larger. However, there is a large uncertainty on the shape of the storms. Furthermore, it is remarkable that this Gaussian distribution also fits to the experiments of Frühling (figure 5.9). i x /i Frühling Gaussian distance x (km) Figure 5.0 : Spatial variation of the rainfall in a storm using a Gaussian distribution. There is an additional important parameter to estimate, namely the peak intensity. The above discussed implementation of the spatial variability is only valid if the peak intensity is known. Based on point rainfall measurements the peak rainfall intensities are rarely observed. Because spatial distributed rainfall for a long measurement period (several decades) is not available in Flanders, all sewer system design is based on point rainfall measurements. Using an areal reduction factor in this case may lead to a significant underestimation of the rainfall volumes. If point rainfall measurements are taken for a long period, one can assume that a mean sample is obtained of the rainfall at one place. The measured peak intensities are however smaller than the real peak intensities and it is difficult to estimate the difference 5.6 The influence of rainfall and model simplification on combined sewer system design

17 between them. In figure 5. the spatial rainfall distribution is shown in a case where the peak intensity would be underestimated by the rain gauge measurement with 20 %. This figure also shows that in this case an areal reduction factor is not necessary for areas with a diameter smaller than 6 km (or a circular area of about 3000 ha). Taking into account the underestimation of the measured peak rainfall intensities, for small areas even an areal augmentation factor should be used instead. This leads to the conclusion that it is not recommended to use an areal distribution coefficient in sewer system design, unless the probability estimation of the peak rainfall measurements is taken into account i x /i distance x (km) Figure 5. : Spatial rainfall distribution when the peak intensity is underestimated by the rain gauge measurement with 20 % and the real rainfall intensity is thus 20 % higher. Chapter 5 : Design calculations 5.7

18 5.4.2 Dynamic influence of the spatial rainfall variation In the previous paragraph the spatial variation of the rainfall was assumed to be static. However, storms usually move over the catchment, which gives a dynamic character to the spatial variation of the rainfall. For large sewer systems the concentration time can be several hours. In this time a storm can move. This can have different effects depending on the direction of the storm as compared with the main flow direction : S If the main flow direction in the sewer system is the same as the dominant direction of the storms moving over a catchment, the use of an areal reduction coefficient will lead to an underestimation of the flow. In this case the discharges might be even larger than in the case that no reduction coefficient is used at all. S If the main flow direction in the sewer system is the opposite of the dominant direction of the storms moving over a catchment, the use of an areal reduction coefficient may still lead to an overestimation of the flow. S If the main flow direction in the sewer system is perpendicular to the dominant direction of the storms moving over a catchment, the incorporation of a static areal rainfall distribution will on average lead to a good flow prediction. All intermediate possibilities can occur. Tests performed by Luyckx & Herftijd [996] show that this dynamic effect of a storm moving over a catchment can result in a significant increase in flow and water depth. However, this study also showed that the mean result from four perpendicular storm directions (moving over the centre of the catchment) approaches the results of a static storm in the centre of the catchment. This means that there would be no influence of this dynamic storm behaviour if there is no preferential storm direction. However, the rainfall data from the dense pluviometric network of Antwerp show that there is a preferential storm direction, i.e. from south-west to north-east [Willems, 998a, 999; Willems & Berlamont, 998a]. Although this might be the general case, for thunderstorms the storm direction probably is less preferential. Because of the dynamic behaviour of the spatial variation of the rainfall, in theory, it could be preferable to avoid main sewers directed from south-west to north-east. The distribution of the storm directions is however broad and in practice other parameters (e.g. topography, etc...) play a larger role in the determination of the direction of the main sewers. In these cases extra safety could be included in the design using simulations with storms moving over the catchment. This is relatively easy to implement and does not require longer calculation time, although the choice of the storms must be considered carefully in terms of corresponding return period. 5.8 The influence of rainfall and model simplification on combined sewer system design

19 5.5 Other design aspects 5.5. Design safety The current return period for the design of sewer systems in Flanders is 2 years (5 years in problem areas) [VMM, 996]. For this return period the piezometric level must remain 0.5 m below ground level. An extra check simulation for flooding is required with a return period of 5 years (0 years in problem areas). The European guidelines (EN 752 [BIN, 996]) suggest higher return periods. They even state that the pipes may not be pressurised for the design return period. This is practically not feasible in flat regions as in Flanders. Moreover, there is little reason to restrict pressurised flow as long as the pipe capacity is higher than the flow and the pressurised flow is thus only caused by the downstream boundary condition, e.g. the combined sewer overflows. The capacity of a pipe is the maximum discharge that can be transported in a steady state with free outflow. The capacity of a pipe is large enough if the slope of the piezometric line is smaller than or equal to the slope of the pipe. Historically, the design return period has periodically fluctuated (see paragraph 5.2.). In 987, the design return period of 2 years (5 years for problem areas) was prescribed and the use of the critical storm duration was generalised [Berlamont, 987]. But at that time most sewer design was still carried out using the rational method or a variant of it. It is generally known that the rational method overestimates pipe sizes, although the dynamic effects are not taken into account. By only rounding off the diameters, the return period can rise from 2 years to even 5 years [Chow, 98]. The return period for flooding can be even 4 to 5 times higher than the design return period [Jensen and Prisum, 982]. These safety margins were taken into account in the design criteria. When hydrodynamic simulations were generally applied, the criteria were not changed. However, the safety margins on the design decreased considerably; for instance, because the real diameters are used for the calculations. If a sewer system is well designed and the return period for flooding is 5 years, the system will indeed flood on average every 5 years. The use of hydrodynamic simulations itself can also have a negative effect on the design quality. This software is accessible to more people and the stimulus to reflect on the design diminishes. In the end, the design will only comply with the formulated rules as there are : a piezometric level of 0.5 m below ground level for a return period of 2 years and no flooding for a return period of 5 years. This kind of designing 'up to the limit' increases the sensitivity of the system for flooding and is often not the most economical solution. This is illustrated in figure 5.2, were the longitudinal section of a few pipes is shown for a design up to the limit in the upper graph as compared with a normal design in the lower graph. The calculation for a return period of 5 years becomes a design calculation with respect to flooding, which leads to less safe situations, while it should be a check simulation in order to increase the safety. Chapter 5 : Design calculations 5.9

20 A return period of 5 years for flooding does not necessarily means that this water also causes damage (e.g. water on the street versus when it enters the houses). This difference is not made in the Flemish guidelines and design practice. Figure 5.2 : Longitudinal profile of a few pipes to show an example of a design up to the limit (upper graph) as compared with a normal design (lower graph) Boundary conditions System components such as overflows, pumps, discharge control devices, flap valves, etc... have a major influence on the flow through the sewer system. Therefore, they have to be modelled carefully. The relationships between piezometric level (or head) and the discharge must be determined accurately for these components. In practice, often default parameter values are used due to a lack of information. The modelling accuracy for these components often determines the accuracy of the water levels in the whole sewer system. Even more important is the operational uncertainty. Pumps may fail, flap valves may block, etc... The simulation of these failures can lead to a better understanding of the system and an increased safety of the design. The most important boundary conditions are the connections between the sewer systems and the receiving waters : the discharge capacities of pumps and overflows. Often the receiving waters influence the flow in the sewer system : flap valves close, 5.20 The influence of rainfall and model simplification on combined sewer system design

21 overflows submerge, etc... In addition, boundary conditions are often not constant. When the water level in the receiving water changes independently of the flow in the sewer system (e.g. tide rivers), superposition can be applied to determine the mean return period T : T = n n (5.5) i= Ti in which n different return periods T i are considered for the same phenomenon, but using n different boundary conditions (with the same probability). This means that the frequency (inverse of the return period) of a specific phenomenon (discharge or piezometric level) is the mean of the frequencies of the individual scenarios (if each scenario has the same probability; otherwise weight functions equal to the probability have to be used). In figure 5.3 this is illustrated for the piezometric level using two different scenario's for the boundary conditions. This is a quite laborious methodology, because for each scenario the same discharge or piezometric level must be assessed with the accompanying return period (mostly by interpolation). 4 piezometric level h (m) h scenario scenario 2 0 T T return period T (year) Figure 5.3 : Visualisation of the calculation of the return period T (i.e. the inverse of the frequency f) for a combination of two different boundary conditions : T = 4.3 year and T 2 = 8.3 year Y f = 0.23 / year and f 2 = 0.2 / year Y f mean = 0.8 / year Y T mean = 5.7 year. However, in most cases the variability of the water level in the receiving water is not independent of the flow in the sewer system. The rainfall will lead to an increase in discharge in both the combined sewer system and the receiving water; sometimes at the same time, but usually with a time shift. This was shown for example by Dahl et al. [996] for a specific case study. Often only rough estimations are made on the interaction between sewer systems and receiving waters. The integrated modelling Chapter 5 : Design calculations 5.2

22 with time series is the only possibility for a correct incorporation of the time differences in peak flow in sewers and receiving waters. However, this will not likely be standard practice in the near future Data validation It has been standard practice for some years in Flanders to perform rainfall and flow measurements to validate the hydrodynamic models [Aquafin, 992, 994a, 996, 997]. This is a crucial phase in the design process. The measurements are not meant to be used for model calibration, but to check that no large errors have been made. The necessity of these measurements rises from the uncertainty on the input data for the models and the large impact of it is shown in figure 5.3. In Flanders, a lot of sewer system data originates from years or even decades ago and changes might not have been accounted for. However, if all sewer data should be redrawn, this would require much time and money. Using the flow measurements, a more purposeful data correction can be performed. Still, data collection remains one of the most sensitive phases in the design process. Some years of short term flow measurement campaigns have improved the models, but also have shown the large uncertainties that still exist on the model input [Willems, 995, 998b; Willems & Berlamont, 998b, c, 999]. In the future, special attention should also be paid to the spatial variability of the rainfall as discussed in paragraph 5.4. Because of the high influence of the water input into the sewer system, flow measurements should in the future be extended to long term measurements. This can be most easily done at the system outflows (e.g. pumps, combined sewer overflows, etc ). As a result, this can lead to a real calibration of the models (dry weather flow, impervious area, losses, contributing pervious area, ) so that they become operational models. It is clear that the data acquisition is a major bottle neck in the current sewer system design. This is a time consuming work, which is however very important. In Flanders, a lot of sewer system design are not yet performed using hydrodynamic simulations. Moreover, many sewer system designs were made decades ago. The data are only available on plans and design notes. There is thus a technological discrepancy in time, but also from one sewer system to another. This discrepancy can only be eliminated by a general review of the municipal sewer system plans as has already been stressed several times in the past [Berlamont, 994; Vlario 994; Vaes & Berlamont, 998a, 999b]. The main task in this is certainly the inventorying of the existing sewer systems. Once the sewer system data are available, the implementation of hydrodynamic simulations will then only be a small step in the process. The inventorying of the sewer system data is not only necessary for the design simulations, but also for the management of the municipal sewer systems The influence of rainfall and model simplification on combined sewer system design

23 5.6 Conclusions Every hydrological design is driven by the rainfall input. The extreme variability of the rainfall makes the design calculations complicated. However, a good design is in most cases obtained using single storms with only the mean rainfall characteristics included. In the past (before 996) some underdesign has taken place partially due to the underestimation of the rainfall input. Since the use of the composite storms, the calculated hydraulic parameters are much closer to those calculated with continuous long term simulations. It is however important to emphasize that there is a large uncertainty on the storm water input, because of the rainfall variability, but also because of the uncertainty on the contributing area and the runoff model. It has been shown that a limited error on the rainfall input can lead to significant differences in design return periods. The composite storms lead to an accurate prediction of the hydraulic parameters if the sewer system does not behave non-linearly. When the system behaviour becomes too far from linear, the temporal variability of the rainfall should be incorporated. If the non-linearity is situated upstream of the sewer system (e.g. in rain water tanks, infiltration facilities, runoff model, etc...) the rainfall variability can be incorporated using a simple model and long term simulations. The results of these simulations can then be incorporated in modified composite design storms. One of the weakest points in the current design methodology is the runoff model. Long term measurements and calibration of the runoff model (contributing area, losses, etc...) are necessary to obtain a good estimation of the rainwater input. For the design simulation the effect of this runoff model then could be incorporated in modified composite design storms. With this kind of simple model it is even possible to incorporate the uncertainty on the runoff parameters as has been implemented for the rain water tanks. If the non-linearity is situated in the sewer system itself, it is necessary to perform calculations with the original rainfall series. The calculation time then can be reduced using only a selection of short rainfall series. This still requires long calculation times and will therefore not replace the single storm simulations in the near future. However, it might serve as a check simulation. Up to now, the spatial variability has most often been neglected, because little knowledge and data were available to lead to a practical incorporation of the spatial variability in the design. This certainly needs more research. The first steps in this direction show that it can have a large influence on the flow in the sewer system. The design of a sewer system is more than the use of a simulation software and some design storms. It involves catchment data input, design methodology, design criteria and boundary conditions, which are certainly of equal importance as the software and the rainfall. All these parameters cannot be discussed separately. Furthermore, all Chapter 5 : Design calculations 5.23