There is a growing interest in urban runoff processes, since more than half of the world’s population live in urban areas. As a result, urban water researchers and practitioners are increasingly concerned with how to manage urban runoff with the effort to maintain more water on-site and to replicate natural hydrological processes. Urban runoff is generated when rainfall flows over land or impervious surfaces, such as paved streets, parking lots, and rooftops and does not seep into the ground. Apart from direct damage, heavy rainfall can also lead to a sequence of cascading events, such as power interruptions, traffic congestion, business interruptions, and pollution of water bodies [1
In the past, the urban runoff control was focused on efficient surface drainage and flood control for a given return period rainfall event that was often of a larger magnitude [2
]. However, researchers and practitioners are becoming increasingly concerned with the runoff resulting from smaller and more frequently occurring rainfall events that can cause a sequence of negative effects in urban areas and receiving waters [3
The main pollutants found in runoff come from fertilizers (nutrients), humans and animals (bacteria), chemicals (pesticides), roofs and roads (metals), and from vehicles (hydrocarbons) [4
]. Suspended sediments constitute the largest mass of pollutant loadings to receiving waters from urban areas and is generally conveyed by urban drainage as non-point pollution [5
]. Polluted urban drainage runoff can be harmful to plants, animals, and people, and its quality was largely ignored in the design of urban drainage systems until approximately 1980 [6
Green Infrastructure (GI) is an attractive option for urban water managers as it has the potential to provide a range of benefits and co-benefits. If carefully designed and implemented, GI can be effective in dealing with problems associated with floods and droughts as well as with poor urban runoff quality [7
]. However, retrofitting GIs in long-established urban areas can be a technically very challenging and costly task [9
]. Previous studies aiming to evaluate the performance of GI in urban water systems have been carried out by: (i) modelling tools for stormwater management and the economics of GI practices [10
], (ii) evaluating the importance of GI in small and medium-sized towns [11
], (iii) examining the performance of vegetative swales to improve runoff in an urban area with moderate traffic [12
], (iv) proposing a flexible modelling framework for hydraulic and water quality performance assessment of stormwater GI [13
], (v) combining ecosystem services with a cost-benefit analysis for the selection of green and grey infrastructure for flood protection in a cultural heritage [14
], and (vi) combining co-benefits and stakeholders’ perceptions into the GI selection for flood risk reduction [15
]. The results from these studies have shown good potential for application of GI in urban water management.
Numerical models have proved to be invaluable for modelling flows in urban areas [16
], while multi-objective optimisation can provide useful support in decision-making processes. In addition, the combination of numerical models and optimisation tools, such as the NSGA-II optimiser has proved to be particularly useful for dealing with stormwater-related issues [18
]. The objective of the NSGA-II algorithm is to improve the adaptive fit of a population of candidate solutions to a Pareto front that is constrained by a set of objective functions. The algorithm uses an evolutionary process with surrogates for evolutionary operators, including selection, genetic crossover, and genetic mutation. The population is sorted into a hierarchy of sub-populations that is based on the ordering of Pareto dominance. Similarity between members of each sub-group is evaluated on the Pareto front, and the resulting groups and similarity measures are used to promote a diverse front of non-dominated solutions [21
Investigations where NSGA-II have been used to optimise GI have focused on: (i) multiobjective optimisation for combined quality-quantity urban runoff control [22
], (ii) selecting an optimal sustainable drainage design for urban runoff reduction [23
], and (iii) proposing an evolutionary and holistic assessment of green-grey infrastructure for CSO reduction [24
]. Other multiobjective evolutionary algorithms are also used to optimise GI, as in the case of (iv) minimization of cost, sediment load, and sensitivity to climate change in a watershed management application [25
], (v) optimal selection and placement of green infrastructure to reduce the impacts of land use change and climate change on hydrology and water quality [26
], (vi) optimal sizing of GI treatment trains (i.e., a sequence of multiple stormwater treatments) for stormwater management [27
], and (vii) a quantitative modelling framework to support decision making in Sustainable urban Drainage Systems (SuDS) design alternatives [28
]. The above approaches have produced promising results and may become a useful tool for planning and decision making of drainage systems.
Based on the previous studies, the benefits of applying GI measures (or practices) are well known. However, currently available methodologies are more focused on the optimal coverage area of GI instead of a GI-type preference. The key advantage of the present approach is that the number of equal GI-size units redistributed within the subcatchment has been taken into account within the optimisation process and this is presented in more detail in the subsequent sections.
The present paper provides a novel approach that aims to configure GI for urban runoff and pollutant reduction using the optimal number of units. The research includes an assessment of the performance of GI measures when dealing with two main objectives: environmental (i.e., pollution load, peak runoff, flood volume) and economic (i.e., investment costs). This proposed framework has been implemented in the coding environment LAZARUS (a free source Delphi compatible with cross-platform IDE for rapid application development) in order to couple a hydrodynamic model with an optimisation algorithm. This coupled model searches for optimal GI units that can achieve runoff reduction, better runoff quality, and least investment costs. The potential of this method has been demonstrated in the real-life case study of Cali, Colombia, where different GI units were evaluated while considering the environmental and economic objectives.
4. Results and Discussion
4.1. Initial Performance of the Drainage System
The hydrodynamic model was run for the selected return period events of 2, 5, 10, 20, and 50 years without implementing any of the GI measures (i.e., present state). Simulation results indicate a peak runoff of 147, 171, 185, 200, and 227 m3/s, respectively. The TSS loading at the outfall of the system was found to be 37,348; 40,388; 42,635; 45,184; and, 49,113 kg/day, respectively.
4.2. GI Placement
Potential suitable location/areas for different types of GI were identified from the analysis of urban land use, stream location, soil classification, land ownership, and impervious layers. The placement of GIs was carried out by finding the available space for GI in each subcatchment. The minimum percentage of available area (ha) was found to be 2.8% and the maximum 32%. Figure 4
presents the maximum number of GI units for each subcatchment. With the criteria presented in Section 3.2.2
, the maximum number of GI units found was 468 divided, as follows: 116 units of BR, 116 units of IT, 164 units of vs. and 72 units of PP. Figure 5
depicts an example of the TSS loading in each subcatchment for a five-year return period event before and after GI placement (applying the maximum number of units). On average, a reduction of 40% of TSS could be potentially obtained after implementing GI measures.
The second hydrodynamic model run was carried out for 2, 5, 10, 20, and 50-year return period events with the maximum number of GI units (i.e., 468 units). Simulation results indicate a peak runoff reduction to 114, 131, 140, 153, and 172 m3/s, respectively. Also, the reduction of TSS at the outfall of the system was found to be 22,737; 24,955; 26,717; 28,585; and, 31,471 kg/day, respectively.
4.3. Assessing Optimal GI Measures
In order to obtain optimal GI solutions distributed within the catchment, a trade-off between each objective reduction (i.e., pollution load, peak runoff, flooding volume) and investment costs was introduced as an optimisation problem. As described above, four different GI measures were evaluated (BR, IT, vs. and PP) and the selection of these measures were described in Section 3.2
. These measures were evaluated by running simulations for 2, 5, 10, 20, and 50-year return period events. The maximum number of GI units was a subject of the optimisation process. Figure 6
shows the non-dominated solutions obtained for the mentioned objectives.
a shows that for smaller events, with solution s-2 an investment of $
7.5 million can achieve a pollution load reduction of 43%. For larger return period events (up to 50 years) a $
7 million investment suggests a pollution load reduction of 40% (solution s-10). In terms of peak runoff reduction, Figure 6
b presents solution s-12 with a peak runoff reduction of 30% by investing $
6.4 million for a two-year event. With solution s-20 it is possible to reduce peak runoff by 27% for a 50-year event and $
6 million investment. Figure 6
c presents optimal solutions for flood volume reduction. Solution s-22 shows a reduction level of 80% for a two-year event with an investment cost of $
7.3 million. Solution s-30 demonstrates a flooding volume reduction of 68% by investing $
7.4 million (up to 50 years).
presents the comparison of the selected solutions for each objective reduction. The catchment points (letters in red colour) that are presented in Figure 4
have been used for comparison purposes between the present state (no GI placement) and the computed optimal solutions.
Solutions s-2 and s-10 indicate an important pollution decrease, especially where the water quality deterioration in points G, H, I, and J of the Meléndez catchment is very significant. Solution s-12 and solution s-20 are able to regulate the flow of the river when a rainfall event occurs in the upper part of the catchment (points A, B, and C), and thus reduce the river flow at the entrance of the city (between points G and H). Solution s-22 and solution s-30 indicate the possibility of reducing the risk of flooding, particularly in points H and I where the highest flood volumes occur in a mostly residential area. Figure 6
presents the optimal number of units with the aim of identifying the GI type that better reduce the three objectives for the study area.
As it can be seen from Figure 7
, GI types, such as infiltration trench (IT02) and vegetative swale (VS01) for small and large events present the largest number of GI units deployed in the catchment reducing the three objectives. Comparing these numbers with the results that were obtained in Figure 6
and Table 2
, a larger number of IT02 and VS01 would have an effect on improving runoff quality and quantity when compared to bio-retention cells (BR) and porous pavement (PP), despite the small differences in the investment costs. According to the characteristics of the GI design shown in Table 1
, IT02 could have been more placed due to the rate value at which water seeps into the native soil below the layer (greater than IT01).
Through the analysis of these results, it can be observed that BR02 has been mainly used with the purpose of reducing pollution load (solutions s-1 to s-10) as compared to the other two objectives. From the three types of BR, BR02 differs from the other two due to the thickness of the soil layer, the rate value at which water seeps into the native soil below the layer (both greater than BR01 and BR03) and the draining type (it uses infiltration rather than underdrains). In line with the design characteristics, for larger events the use of PP01 in the catchment increases in solutions s-8 and s-10 for pollution load reduction and solutions s-18 and s-30 for peak runoff and flooding volume reduction, respectively.
The main constraint of these optimal solutions is the amount of financial resources that is required to the initial GI implementation, which is $
19.9 million dollars. Within this framework important investment cost reductions were obtained in terms of pollution load, peak runoff, and flooding volume to $
7 million, $
6 million, and $
7.4 million, respectively, for events up to 50 years. In terms of the effect of a rapid increase in rainfall intensity and similar to the work that is presented in [41
], different return periods (2, 5, 10, 20, and 50-year) enable not only a better understanding to achieve optimal solutions but also an effective GI placement minimizing erroneous and costly intervention for urban runoff reduction.
According to other researches, although we manually determined subcatchment parameters and GI process layers, much of this work could be automated. However, it is important to consider the stakeholders input into the final design decisions for each subcatchment, possibly iteratively applying this framework with additional constrains until a satisfactory solution is found see [22
]. The methodological framework that is presented here demonstrates a possible way to select one solution from different alternatives. For the case study area of Cali, these solutions that can maximise environmental and economic objectives for up to a 50-year return period event could be considered as preferred. However, since different GI type units can produce a similar performance, the preferred combination of GIs would depend on the objectives that need to be achieved. In our future work, we will address such issues by incorporating a preference-based multi-objective model within the present methodological framework.
The present paper describes a novel methodological framework that aims to configure GI for urban runoff and pollutant reduction using the optimal number of units. The present work addressed the assessment of the performance of GI measures dealing with environmental and economic objectives. The proposed methodological framework has been implemented in the coding environment LAZARUS, which is a free-source Delphi compatible cross-platform. The code combines a hydrodynamic model and an optimisation algorithm. Simulations of hydraulic, hydrologic and quality aspects were performed within the SWMM package, while the NSGA-II model was used for process optimisation. The mass of a pollutant transported during a storm event has been modelled as a coupled build-up and wash-off process, providing the stormwater pollutant load that is generated from the urban catchment. The work was demonstrated in a real-life case study of Cali (Colombia) where bio-retention-cells (BR), infiltration trenches (IT), vegetative swales (VS), and porous pavement (PP) were evaluated while considering pollution load, peak runoff, and flood volume objectives at the lowest possible investment cost.
There are currently actions that are aiming to reduce the pollution load in the Meléndez river, but its water quality is still continuing to decline. Similarly, in spite of substantial investment in flood control structures in the catchment, there is still the risk of flooding as the investments are not executed according to their priority and their true impact in the catchment.
The results of this study show that by investing an amount of $7.7 million with a higher number of BR units (up to 83 units) within a specific configuration, a pollution load reduction for larger events can be obtained with solution s-10. The solutions also show that an increase in the number of vs. units (up to 76 units) with the same investment can yield a reduction in peak runoff for both smaller and larger events (s-16 and s-18). Similarly, with the same level of investment and with a larger number of PP units (up to 22 units), solution s-30 would help to reduce flood volume for shorter and larger events.
The application of multi-objective optimisation processes for GI configuration may become a good choice in terms of reducing investment cost without compromising the efficiency of the drainage system. The results show an advantage of having an optimal number of GIs, as the GI types mainly reflect the impact on the reduction of the three objectives. This suggests that if the type of GI measure and its number of units are taken into account within the optimisation process, it is possible to achieve optimal solutions to reduce the proposed reduction objectives with a lower investment cost. In terms of disadvantages, one of the key disadvantages is that this approach is not able to incorporate surface water infiltration process into the hydrodynamic model based on a given infiltration equation (i.e., modified Horton method) in a 1D-2D modelling approach while taking into account the expensive computational time which limits its application for real-life purposes.
The present work also demonstrates how different performance can be used to address different objectives and to identify a solution that can be suitable for the study area. In our future research, we plan to extend the present methodological framework by taking into consideration a preference-based multi-objective model that can reflect different preferences for different performance measures.