Abstract
Urban flooding has increased in response to impervious surface intensification, the loss of green areas, and high-intensity rainfall associated with climate change. Sustainable urban drainage systems (SUDS) are an appealing option for stormwater management; however, their hydraulic control capabilities have received little attention. We developed a comparative model-based approach with 24 scenarios to contrast the hydrologic and hydraulic response of a highly discretized (HD) 1D model and a coupled 1D–2D model, considering the impact of rainwater harvesting systems and tree pits. An additional scenario was modeled including attenuation storage tanks, green roofs, and pervious pavements. A heavily urbanized flood-prone catchment with severe land-use constraints in Bogotá, Colombia, was selected for analysis. The findings revealed that SUDS can contribute to reducing the number of flooded junctions, overloaded conduits' length, overloading time, nodal inundation depth, and waterlogging extent. Furthermore, the HD 1D model can reproduce the coupled 1D–2D model results in terms of hydrologic response and some hydraulic control indicators. Further research is needed for an accurate description of the internal hydraulic mechanisms of SUDS interacting with overland flow. The key findings of this study provide model-based evidence to support urban stormwater management decision-making in data-scarce environments.
HIGHLIGHTS
24 scenarios contrasting SUDS performance using a HD 1D and a coupled 1D–2D model.
Discussion on data requirements, limitations, and hydrologic–hydraulic response.
HD 1D models can reproduce the hydrologic response of 1D–2D models.
HD 1D models can partially reproduce the hydraulic 1D–2D model results.
SUDS provide hydraulic control in heavily urbanized areas.
INTRODUCTION
Rapid urbanization has induced abrupt changes in land use, affecting natural hydrologic processes and flood patterns in cities (Akhter & Hewa 2016). The loss of green areas and the intensification of impervious surfaces have produced an adverse increase in runoff volumes and peak discharges (Qin 2020). This is aggravated by high-intensity rains associated with climate change (IPCC 2022). Furthermore, the limited capacity of the existing drainage network increases the risk of waterlogging and pluvial flooding (Maksimović et al. 2009). In some cases, expansion of the conventional pipe-ended drainage system is not always feasible due to space restrictions or financial constraints (Taji & Regulwar 2019). Therefore, past studies have stressed the importance of studying sustainable urban drainage management (SUDM) strategies such as best management practices, green infrastructure, low impact development (LID) techniques, and sustainable urban drainage systems (SUDS) (Vogel et al. 2015).
SUDS have gained significant attention as urban flood risk management strategies (Gimenez-Maranges et al. 2020) since they attempt to replicate the natural predevelopment drainage conditions. The Ministry of Housing, City, and Territory of Colombia introduced the concept of SUDS in 2017 as a strategy to mitigate the effect of soil sealing caused by new urban developments (Ministry of Housing of Colombia 2021). Bogotá, the capital city of Colombia, has championed SUDS implementation through local design and construction guidelines since 2018 (EAAB 2018). Despite this, urban water decision makers are still hesitant regarding SUDS' hydraulic control capabilities and the impact on the existing drainage system due to efficiency uncertainties and a lack of quantitative evidence (Ortega et al. 2023).
The most widely used hydrodynamic models for assessing the performance of SUDM strategies include the US Environmental Protection Agency Storm Water Management Model (EPA SWMM5 or SWMM5) (Rossman & Simon 2022), the Personal Computer Storm Water Management Model (PCSWMM) (CHI 2023), MIKE URBAN (DHI 2019), the Model for Urban Stormwater Improvement Conceptualization (MUSIC) (eWater 2015), and InfoWorks ICM model (Innovyze 2019). The application of these models in the study of SUDM has primarily focused on the hydrologic response (e.g., reduction of runoff volume and peak discharge), often neglecting or generalizing the interaction with the pipe-ended drainage network (Cui et al. 2019). Nevertheless, the complexity of urban systems requires more advanced approaches that consider the interaction between one-dimensional (1D) sewer flows and two-dimensional (2D) overland flows (Chang et al. 2015), which are altered by flow path-modifying elements such as buildings, streets, and pavement curbs (Blanc et al. 2012).
The increase in high-quality information availability, advanced computer capacity, the inclusion of geographic information systems (GIS), and new mathematical approaches have enabled the development of coupled 1D–2D models (Pina et al. 2016). Coupled 1D–2D models simulate flood propagation on the ground surface through 2D grid cells (Chang et al. 2015) and can represent the bidirectional discharge between the 1D and 2D domains using direct links or weir/orifice-type elements (Maksimović et al. 2009). Few studies have assessed SUDM alternatives' performance through coupled 1D–2D models, mainly focusing on their flood mitigation capabilities. For instance, Ellis & Viavattene (2014) reported an improved visualization of the flooding mechanisms at any location by adopting a 1D–2D flow model to investigate the most suitable SUDS options to reduce the flood extent and depth. Moreover, Haghighatafshar et al. (2018) investigated the collective impact of blue-green retrofits in reducing flood volumes and depths through a coupled 1D–2D model, which facilitated the analysis of the interaction between the existing pipe-ended drainage system and the overland flow at a catchment/neighborhood scale. These studies highlighted the complexity of urbanized watersheds due to the topographic surface, interactions with drainage infrastructure, and land-use conflicts. Nevertheless, research assessing the hydraulic control capabilities of SUDM strategies, i.e., the impact on the surcharging and flooding dynamics of the existing drainage network elements such as manholes and pipes (Cui et al. 2019; Mu et al. 2022), is lacking.
Coupled 1D–2D models enable a more detailed representation of surface geometry and flow physics (Vojinovic & Tutulic 2009) while avoiding simplifications of the hydrologic processes and the hydraulic network (Pina et al. 2016). Nonetheless, these robust modeling approaches require high-resolution terrain data and can be computationally prohibitive, limiting their reproducibility (Maksimović et al. 2009; Blanc et al. 2012). Therefore, there is a need for feasible modeling alternatives that can be employed in data-scarce places to foster the implementation of SUDM.
Considering the aforementioned gaps, the present study aims to develop a comparative model-based approach composed of a highly discretized (HD) 1D sewer model and a coupled 1D–2D hydrodynamic model. The objectives are to contrast (i) data requirements, model construction process, and computational costs; (ii) the ability of the HD 1D model to reproduce the hydrologic and hydraulic response of the 1D–2D model; (iii) SUDS hydrologic performance in terms of peak discharge and total outflow volume reduction; (iv) SUDS hydraulic control capabilities according to the number of surcharged junctions, the number of flooded junctions, the overloaded conduits' length, and the average overloading time; and (v) SUDS flood control potential in terms of maximum nodal inundation depth and waterlogging extent. The study site is a highly urbanized catchment in Bogotá that frames an interesting case study for improved sustainable stormwater management given its particular condition as a waterlogging- and flood-prone area, with limited potential for SUDS implementation on both private and public lands.
METHODOLOGY
Study area
The study site is located in Bosa, 1 of the 20 administrative divisions (localidades) of Bogotá, the capital city of Colombia. Bosa is located to the south-west of Bogotá with a predominant residential land use (87%) (Galindo 2013). The mean annual precipitation in Bosa ranges between 620 and 700 mm with a bimodal distribution: the first rainy period corresponds to April and May, and the second falls from October to December. Although Bosa is served by a separate storm sewer system, by 2012, about 68% of its total area registered some level of flood exposure. The low-lying nature of the land and large social housing projects that have reduced the soil's natural permeability have contributed to Bosa's flood vulnerability (Rojas 2018).
Study site (a) location and (b) land-use distribution. Please refer to the online version of this paper to see this figure in color: http://dx.doi.org/10.2166/wst.2023.173.
Study site (a) location and (b) land-use distribution. Please refer to the online version of this paper to see this figure in color: http://dx.doi.org/10.2166/wst.2023.173.
Data collection and preprocessing
Input data used to create the HD 1D sewer model and the coupled 1D–2D model included terrain conditions, land-use distribution, and drainage network information. Catchment delineation and data preprocessing were performed using QGIS Desktop software version 3.16.0-Hannover.
A digital elevation model (DEM) was developed from 1-m contour line information provided by the local water utility (LWU) from a terrestrial LiDAR survey. Land-use discretization was performed with vector layer data freely available in Bogotá's Spatial Data Network (www.ideca.gov.co). In the case of unclear information, field observations and Google Earth Pro version 7.3.4.8248 were used to contrast the polygon vector information. Raster data for infiltration capacity (50-m cell size) and water table depth (250-m cell size) were obtained from a city-scale project led by the Secretary of the Environment and the LWU to develop SUDS planning and design guidelines (Universidad de los Andes 2015).
Drainage system information was provided as vector layer data by the LWU. Duplicated elements (manholes, pipes, and outfalls), unconnected manholes, and dead-end pipes were removed to achieve model accuracy (Hua et al. 2020). Then, manholes' and pipes' invert elevations were inspected to detect anomalies. During this process, it was found that none of these elevations agreed with realistic values according to the study site's ground elevation as extracted from DEM (on average 30 m above the ground level). For this reason, manholes' rim and invert elevations, as well as pipes' inlet and outlet elevations, were adjusted using DEM elevations and drainage system geometry information.
Observed data to perform a calibration process were not available as drainage sewer flow is not routinely monitored by local authorities in Bogotá. Therefore, flow observation campaigns were conducted between 29 April and 14 May 2022, at one of the open channel's outlets; however, the results only reflected the contributions to baseflow, and an adequate representation of the local drainage system's response to a rainfall event was unsatisfactory.
Data scarcity for calibrating flood models is a technical and financial challenge often present in urban areas (Schmitt et al. 2004); therefore, finding flexible solutions to provide significant results within the framework of urban water decision-making is a key. Alternative approaches include the use of information from citizen observations, news, surveillance cameras, and social media images (Assumpção et al. 2018; Moy de Vitry & Leitão 2020). In the case of Bosa, information from an official report of the local risk management authority (IDIGER, by its acronym in Spanish) was used in a validation process, which described a sudden rise of 0.95 m in the open channel's water level (Santa Isabel Channel) after a 30.4 mm rainfall event on 14 December 2011 (FOPAE 2011). Rainfall data of 5-min resolution was retrieved from an online platform (www.sire.gov.co) managed by IDIGER.
Model theory
The present study employed PCSWMM to develop the HD 1D model and the coupled 1D–2D model. PCSWMM uses the EPA SWMM5′s hydrology and hydraulic engine and additional group-decision support tools such as GIS technology (James et al. 2010) to provide strong and powerful simulation capacities in urban environments. SWMM5 is a physically based, discrete-time simulation model in which surface runoff outflow is given by Manning's equation and flow routing within a conduit obeys Saint-Venant's continuity and conservation of momentum equations (James et al. 2010).
In the 1D domain, each sub-catchment is treated as a nonlinear reservoir. Inflow comes from rainfall or any user-defined upstream flow, and surface runoff occurs when the depth of water in the reservoir exceeds the maximum depression storage (James et al. 2010). On the other hand, the 2D domain is discretized as a set of nodes (2D junctions) and open conduits without walls (2D conduits), constituting a 2D mesh that captures the topography extracted from DEM (Leitão et al. 2010). Free surface flow between the 2D junctions and the 2D conduits is solved by using the 1D Saint-Venant flow equations. Moreover, a simultaneous transition between the 1D and 2D domains is possible according to the connectivity of the link-node system, i.e., inlet control bottom orifices or direct connections.
Model setup
Schematic representation of the model setup considering SUDS assessment in (a) the highly discretized 1D model and (b) the coupled 1D–2D model.
Schematic representation of the model setup considering SUDS assessment in (a) the highly discretized 1D model and (b) the coupled 1D–2D model.
Highly discretized 1D model parametrization and setup
A 1D rainfall–runoff model was developed considering the drainage network information (manholes or junctions, pipes or conduits, and outfalls) and the HD land-use distribution of the study site. An inflow boundary condition was defined by delineating the corresponding upstream land-use distribution (buildings, roads, and pervious surfaces) and the drainage network.
The conduits' roughness attribute, i.e., Manning's roughness coefficient, was assigned according to the material type, being 0.015 for concrete pipes and 0.01 for PVC pipes. The trapezoidal open channel (Figure 1) was delineated through multiple conduits with concrete lining, i.e., Manning's roughness coefficient of 0.015. An entrance loss coefficient of 0.05 was assigned to all conduits, whereas an exit loss coefficient of 1 was assigned to those elements discharging into the receiving water body (Butler et al. 2018), i.e., the open channel. Excess volume was assumed to pond over the manhole, preventing water from being lost from the system. Therefore, all 1D junctions were assigned a ponding area of 400 m2, whereas this parameter for the open channel junctions was expressed according to the upstream and downstream conduits' length. The ponding area value is irrelevant within the framework of analysis in this study, since it is not intended to compare the flooding volume produced by the two modeling approaches. Nonetheless, using a nonzero ponding area instead of surcharge depths enables flow excess to accumulate atop the node and reenter the system when downstream capacity becomes available.
To have a realistic approximation to the overland flow generated by the 2D approach, each polygon of all land-use vector layers (buildings, corridors, roads, pervious areas, plazas, sidewalks, and others) was treated as a unique sub-catchment. Parameters of slope (%), area (m2), width (m), impervious percentage area (%), Manning's n overland roughness values, and depression storage (mm) are summarized in Table 1.
Summary of sub-catchment attributes in the highly discretized 1D model
Type of sub-catchment . | Slope (%) . | Area (m2) . | Width (m) . | Imperviousness (%) . | Manning's n – overland flow . | Depression storage (mm) . |
---|---|---|---|---|---|---|
Buildings | 2–30 | 10–5,200 | 0.09–140 | 100 | 0.01 (I) | 0.128–0.495 |
Corridors | 0–1 | 670–1,000 | 0.80 | 40 | 0.014 (I); 0.15 (P) | 0.747 (I); 2.285 (P) |
Others | 0–8 | 10–2,300 | 0.80 | 100 | 0.05 (I) | 0.873a |
Pervious surfaces | 0–9 | 80–14,900 | 0.80 | 0 | 0.15 (P) | 2.285a |
Plazas | 0–3 | 160–12,000 | 0.80 | 100 | 0.014 (I) | 0.875a |
Roads | 0–5 | 100–1,400 | 0.80 | 100 | 0.1 (I) | 0.836a |
Sidewalks | 0–11 | 30–900 | 0.80 | 100 | 0.12 (I) | 0.747a |
Type of sub-catchment . | Slope (%) . | Area (m2) . | Width (m) . | Imperviousness (%) . | Manning's n – overland flow . | Depression storage (mm) . |
---|---|---|---|---|---|---|
Buildings | 2–30 | 10–5,200 | 0.09–140 | 100 | 0.01 (I) | 0.128–0.495 |
Corridors | 0–1 | 670–1,000 | 0.80 | 40 | 0.014 (I); 0.15 (P) | 0.747 (I); 2.285 (P) |
Others | 0–8 | 10–2,300 | 0.80 | 100 | 0.05 (I) | 0.873a |
Pervious surfaces | 0–9 | 80–14,900 | 0.80 | 0 | 0.15 (P) | 2.285a |
Plazas | 0–3 | 160–12,000 | 0.80 | 100 | 0.014 (I) | 0.875a |
Roads | 0–5 | 100–1,400 | 0.80 | 100 | 0.1 (I) | 0.836a |
Sidewalks | 0–11 | 30–900 | 0.80 | 100 | 0.12 (I) | 0.747a |
aCalculated with the average slope.
I, impervious surface; P, pervious surface.
The Horton equation was employed to estimate infiltration losses, considering its practical utility, simplified data requirements, and ability to reflect in situ conditions (Rasool et al. 2021). The semi-empirical Horton model has proven to be efficient in predicting infiltration processes in lawn soils (Duan et al. 2011) with low infiltration rates (Rasool et al. 2021), both characteristics of the present case study. The values used for initial and final (soil's saturated hydraulic conductivity) infiltration rates were 1.80 mm/h and 0.05 mm/h, respectively, according to infiltration raster data (Section 2.1.1). The decay constant and drying time values were 2 (1/h) and 7 (days), respectively (Rossman 2015; Butler et al. 2018), considering Sandy Clay Loam is predominant in the study area. Finally, the HD 1D model consisted of 95 1D conduits, 95 1D junctions, 1 outfall, and 804 sub-catchments, excluding the elements from the upstream boundary condition.
Coupled 1D–2D model parametrization and setup
An integrated 1D–2D model allows for the representation of the interaction between the minor and major drainage systems (1D and 2D domains, respectively), considering the multiple flow pathways around and through obstructions such as buildings and walls. Due to the high computational load, the 1D–2D approach did not model the upstream boundary condition as in the HD 1D model (Section 2.3.1). Instead, inflow hydrographs were assigned to the most upstream open channel junction.
The 1D domain of the coupled 1D–2D model replicated the drainage system parametrization employed in the HD 1D model. Nevertheless, instead of using a ponding area at the network junctions, a surcharge depth of 30 m was assigned to avoid double accounting of overland flow. Furthermore, only building and sidewalk polygons were assigned as sub-catchments, leaving the rest of the land uses within the 2D domain. This is done to reduce the computational load involved in creating the 2D mesh with multiple (buildings) and narrow (sidewalks) polygons. The routing strategy considered building runoff to sidewalks and sidewalk runoff to the nearest 2D junction. The 1D domain of the 1D–2D model consisted of 95 1D conduits, 95 1D junctions, 1 outfall, and 424 sub-catchments.
The 2D domain construction in PCSWMM comprises three main steps, i.e., bounding layer delineation, 2D node generation, and 2D mesh creation. The bounding layer defines the extent of the 2D model and can consist of multiple polygons to represent subareas' roughness and infiltration parameters. In the present study, the bounding layer included pervious surfaces, roads, and the overall bounding polygons. Types of mesh, resolution, roughness, and seepage rates (to represent infiltration within the 2D cells) are described in Table 2. Mesh types in PCSWMM include hexagonal, rectangular, directional, and adaptive. Hexagonal and rectangular mesh types, which generate uniform cells specified by the cell resolution, provide a better representation of open areas and roadways, respectively. Moreover, the PCSWMM's 2D interface allows for adding an obstructions layer, which in this study included the building and sidewalk polygons.
Attributes of the bounding layer for 2D nodes generation
Bounding layer polygons . | Mesh type . | Resolution (m) . | Roughness . | Seepage rate (mm/h) . |
---|---|---|---|---|
Overall bounding polygon | Hexagonal | 5 | 0.05 (urban surfaces)a | n/a |
Pervious surfaces | Hexagonal | 5 | 0.15 (grass, short)a | 0.351b |
Roads | Rectangular | 3 | 0.011 (smooth asphalt)a | n/a |
Bounding layer polygons . | Mesh type . | Resolution (m) . | Roughness . | Seepage rate (mm/h) . |
---|---|---|---|---|
Overall bounding polygon | Hexagonal | 5 | 0.05 (urban surfaces)a | n/a |
Pervious surfaces | Hexagonal | 5 | 0.15 (grass, short)a | 0.351b |
Roads | Rectangular | 3 | 0.011 (smooth asphalt)a | n/a |
After parametrizing the bounding layer, 2D nodes are generated, allowing the 2D cells' approximate locations and their elevations from the DEM to be defined. In this step, elevations of 2D nodes within the roads and pervious surfaces bounding polygons were lowered by 1 and 2 mm, respectively, to represent the depression storage parameter. Afterward, the 2D mesh is created, allowing 2D nodes to be connected through 2D conduits (open rectangular channels). The 2D model was composed of 11 475 2D junctions, 22 875 2D conduits, and 11 474 2D cells.
The present study adopted bottom orifices to connect the 1D and 2D domains. Orifices can represent the catchbasins into which overland flow enters and is then conveyed by the pipe-ended system. Therefore, bottom orifices were defined by a rectangular cross section with 1.20 m of width (to represent two 0.60 m catchbasins per manhole) and 0.12 m of height (according to field observations). The dynamic wave routing method selected to solve the 2D approach allows two inlet capacity constraints to be modeled using bottom orifices, i.e., inflow or upwelling. This is useful to represent surface flooding from a surcharged manhole.
Model validation
Calibration of urban drainage models is a challenging task, constrained by the availability of detailed monitoring data that consider major and minor drainage system interactions (Schmitt et al. 2004; Hunter et al. 2008; Moy de Vitry & Leitão 2020). Nevertheless, accurate flow path description (Schmitt et al. 2004) and the use of plausible ranges of friction parameters (Hunter et al. 2008) can reduce the uncertainty of model predictions. This is the case for the HD 1D model and the coupled 1D–2D model developed in the present study. Considering the aforementioned data and the absence of monitored data from the local drainage network, a validation strategy was established based on (i) open channel water depth information and (ii) a hydrologic–hydraulic performance comparison between the two modeling approaches.
1D–2D models are a solid tool for modeling urban floods, being able to provide enough detailed data in terms of discharge through inlets and water levels (Leandro et al. 2011). The use of both a fine grid resolution and the same drainage network boundary conditions (Leandro et al. 2009) enables meaningful comparisons between sewer-based model and sewer/surface model results. The present study employed a 1 m × 1 m grid cell resolution and the same drainage network parametrization in the two developed models. Therefore, the 1D–2D model's outfall discharge time series was employed as a benchmark for the HD 1D model. The fitness of the model was evaluated through the Nash–Sutcliffe efficiency (NSE) coefficient, where NSE < 0.65 is unsatisfactory; 0.65 < NSE < 0.80 is acceptable; 0.80 < NSE < 0.90 is good; and NSE ≥ 0.90 is very good (Ritter & Muñoz-Carpena 2013). Furthermore, the hydraulic performance agreement was assessed according to the number of surcharged nodes, the number of flooded nodes, the overloaded conduits' length, and the average overloading time. RE (Equation (2)) was employed as evaluation criteria for this assessment, considering the 1D–2D results as observed data (xo).
Comparison of the two modeling approaches
Scenario analysis
Selection, location, and parametrization of SUDS
The present study assessed different SUDS typologies for implementation on public and private land. For private land, three SUDS typologies (green roofs (GRs), rainwater harvesting systems (RWHS), and attenuation storage tanks (ASTs)) were considered and evaluated using the transdisciplinary approach for assessing the SUDS potential feasibility proposed by Ortega et al., under review. In this approach, the GIS-based results of a physical restrictions assessment (soil characteristics, size constraints, and land-use suitability) are constrained by the evaluation of six local context barriers (cultural/behavioral, financial, institutional/organizational, technical, political, and urban form), providing a percentage of feasible implementation. After this assessment, RWHS were selected and distributed according to a potential feasibility rate of 48%.
Furthermore, tree pits (TPs) were selected as SUDS typologies for public land, obeying current conditions of SUDS implementation in the city of Bogotá. Bogotá's LWU issued design and construction guidelines in 2018 to foster the implementation of ASTs, bio-retention systems, extended dry basins, pervious pavements (PP), infiltration trenches, and TP. Nevertheless, operation and maintenance liabilities have affected the assessment of large-scale strategies, turning most of the efforts toward TP pilot projects. Therefore, after the evaluation of physical restrictions, the present study used the current tree distribution of the study area for the TP location. This approach is an attempt to determine if the conversion of tree units to TP units could have a hydrologic–hydraulic impact on the study site. In summary, alongside considering physical constraints, RWHS and TP were selected based on the assessment of local context barriers and current pilot projects, respectively. Moreover, a combination of these two SUDS measures was tested as a representation of a complete scheme operating on both public and private lands.
SUDS can be modeled in PCSWMM as LID controls. They are considered properties of a given sub-catchment (new or existing) and are designed to capture surface runoff via infiltration, detention, and/or evapotranspiration processes through different vertical layers whose properties are defined on a per-unit-area basis (James et al. 2010). Therefore, in both the HD 1D model and the coupled 1D–2D model, RWHS were assigned as units to building polygons, whereas TPs were assigned as units to sidewalk polygons. Table 3 shows the parameters employed in this study to describe the different layers of RWHS and TP based on specialized literature and detailed construction drawings utilized by the urban development institute (IDU, by its acronym in Spanish) in Bogotá.
Parameter values for SUDS design
Layer . | Parameter . | TP . | RWHS . |
---|---|---|---|
Surface | Berm height (mm) | 200a | n/a |
Vegetation volume (fraction) | 0.1b | n/a | |
Surface roughness (Manning's n) | 0b | n/a | |
Surface slope (%) | 0b | n/a | |
Swale side slope (run/rise) | n/a | n/a | |
Soil | Thickness (mm) | 800a | n/a |
Porosity (volume fraction) | 0.6a | n/a | |
Field capacity (volume fraction) | 0.11b | n/a | |
Wilting point (volume fraction) | 0.05b | n/a | |
Conductivity (mm/h) | 70c | n/a | |
Conductivity slope | 5b | n/a | |
Suction head (mm) | 61b | n/a | |
Storage | Thickness (mm) | 700a | 900d |
Void ratio (voids/solids) | 0.5b | n/a | |
Seepage rate (mm/h) | 500b | n/a | |
Clogging factor | 0 | n/a | |
Underdrain | Drain coefficient (mm/h) | 3b | 2.5b |
Drain exponent | 0.5b | 0.5b | |
Drain offset height (mm) | 200a | 100 | |
Drain delay (h) | n/a | 24 |
Layer . | Parameter . | TP . | RWHS . |
---|---|---|---|
Surface | Berm height (mm) | 200a | n/a |
Vegetation volume (fraction) | 0.1b | n/a | |
Surface roughness (Manning's n) | 0b | n/a | |
Surface slope (%) | 0b | n/a | |
Swale side slope (run/rise) | n/a | n/a | |
Soil | Thickness (mm) | 800a | n/a |
Porosity (volume fraction) | 0.6a | n/a | |
Field capacity (volume fraction) | 0.11b | n/a | |
Wilting point (volume fraction) | 0.05b | n/a | |
Conductivity (mm/h) | 70c | n/a | |
Conductivity slope | 5b | n/a | |
Suction head (mm) | 61b | n/a | |
Storage | Thickness (mm) | 700a | 900d |
Void ratio (voids/solids) | 0.5b | n/a | |
Seepage rate (mm/h) | 500b | n/a | |
Clogging factor | 0 | n/a | |
Underdrain | Drain coefficient (mm/h) | 3b | 2.5b |
Drain exponent | 0.5b | 0.5b | |
Drain offset height (mm) | 200a | 100 | |
Drain delay (h) | n/a | 24 |
n/a, not applicable.
Design storms
Rainfall assignation was different according to the modeling approach. In the HD 1D model and in the 1D domain of the coupled 1D–2D model, rainfall data were assigned to all sub-catchments as intensity time series (mm/h). In the 2D domain of the coupled 1D–2D model, all the rainfall volume is assumed to be available for hydraulic routing, and rainfall time series are assigned as a baseline inflow time series (m/s) to the 2D junctions. A representative rainfall volume is obtained by multiplying the cell area (m2) by the converted rainfall.
Scenarios
This study employed a comparative model-based and scenario approach to assess the hydrologic–hydraulic impact of SUDS. Twelve scenarios were created based on the three design storms (Section 2.6.2) and four different SUDS implementation settings: (1) business-as-usual (BAU), to represent the current conditions in the absence of SUDS; (2) RWHS, as the sole implementation of RWHS; (3) TP, as the sole implementation of TPs; and (4) MIX, as a combined implementation of both RWHS and TP. For the remainder of this article, each scenario will be addressed in a coded form based on its setting (BAU, RWHS, TP, or MIX) and design storm (5, 10, or 100), e.g., BAU5. The 12 scenarios were replicated in each modeling approach (HD 1D and coupled 1D–2D), for a total of 24 simulations.
Furthermore, we modeled an additional scenario (MULT100) considering extreme rainfall conditions (100-year T design storm) and the collective impact of five SUDS typologies: (1) RWHS under a 100% implementation scheme; (2) TP, doubling the number of units used previously; (3) GRs and ASTs in residential sectors, where RWHS were not suitable according to slope and tributary area; and (4) PP in corridors, plazas, and other land uses. Design parameters were assigned according to Núñez Collado et al. (2019) and City of Toronto (2021).
Metrics for multi-scenario assessment
The RWHS, TP, and MIX scenarios were assessed by contrasting their results with the BAU scenario conditions. The hydrologic performance was measured in terms of peak discharge (m3/s) and total outflow volume (m3). The impact on the conventional pipe-ended drainage system was assessed considering the number of surcharged junctions, the number of flooded junctions, the overloaded conduits' length (km), and the average overloading time (min).
Surcharge and overloading conditions are useful indicators for predicting localized flooding in urban areas (Schmitt et al. 2004) and preventive maintenance of the drainage network. A surcharged condition at a junction occurs when the water surface elevation (WSE) is above the crown of its highest connected conduit. Alternatively, a junction is flooded when the computed WSE exceeds its rim elevation. The overloading condition in a conduit occurs when its upstream end is full and the hydraulic grade line slope is greater than the conduit slope. Furthermore, the flood control potential was assessed in terms of maximum nodal inundation depth (cm) and waterlogging extent (m2) in moderate (15 cm ≤ overland water depth < 30 cm) and severe (overland water depth ≥30 cm) conditions.
RESULTS
Validation
The validation strategy included (i) the comparison of the water depth levels reached in the open channel against official observations from a rainfall event on 14 December 2011, between 20:00 and 23:00 (Section 2.1.1) and (ii) a hydrologic–hydraulic comparison of the HD 1D model performance with the coupled 1D–2D model results.
Preliminary validation results: water depth in open channel conduits at several places using the (a) highly discretized 1D model and (b) the 1D–2D model, and (c) peak discharge hydrograph agreement between the two modeling approaches.
Preliminary validation results: water depth in open channel conduits at several places using the (a) highly discretized 1D model and (b) the 1D–2D model, and (c) peak discharge hydrograph agreement between the two modeling approaches.
Despite the satisfactory agreement in the open channel water depths and the peak discharge hydrographs, there was a wide discrepancy in the number of both surcharged and flooded nodes. Therefore, a sensitivity analysis was performed to evaluate the parameters that most affect the hydraulic performance of the HD 1D model, with the 1D–2D model results serving as a benchmark (Leandro et al. 2011). The chosen parameters were slope, width, Manning's n coefficient (pervious and impervious surfaces), impervious percentage area, depression storage (pervious and impervious surfaces), maximum infiltration rate, and minimum infiltration rate. These parameters were selected according to previous studies on SUDM (Baek et al. 2015; Ahiablame & Shakya 2016; Akhter & Hewa 2016), and the uncertainty ranges were assigned according to specifications found in the study by Rossman (2015).
The width parameter showed a significant influence on the number of flooded junctions. Therefore, a uniform width value of 0.05–1 m was tested in all the sub-catchments corresponding to corridors, plazas, sidewalks, roads, pervious areas, and others. A value of 0.80 m was finally chosen, reducing the number of flooded junctions from 25 to 6 (Table 4). Although there was no significant improvement in the number of surcharged nodes, the behavior of the remaining hydraulic parameters was satisfactory. Furthermore, the hydrographs' agreement between the HD 1D model and the coupled 1D–2D model reached an NSE coefficient of 1 (not displayed here to avoid redundancy).
Summary of compared values for hydraulic component validation
Validation parameter . | Before validation . | After validation (widtha = 0.80 m) . | ||||
---|---|---|---|---|---|---|
HD 1D . | 1D–2D . | RE (%) . | HD 1D . | 1D–2D . | RE (%) . | |
Number of surcharged junctions | 28 | 3 | −833 | 22 | 2 | −1,000 |
Number of flooded junctions | 25 | 2 | −1,150 | 6 | 2 | −200 |
Overloaded conduits’ length (km) | 3.5 | 1.9 | −82 | 2.1 | 1.7 | −21 |
Average overloading time (min) | 5.86 | 2.00 | −193 | 2.33 | 2.1 | −11 |
Validation parameter . | Before validation . | After validation (widtha = 0.80 m) . | ||||
---|---|---|---|---|---|---|
HD 1D . | 1D–2D . | RE (%) . | HD 1D . | 1D–2D . | RE (%) . | |
Number of surcharged junctions | 28 | 3 | −833 | 22 | 2 | −1,000 |
Number of flooded junctions | 25 | 2 | −1,150 | 6 | 2 | −200 |
Overloaded conduits’ length (km) | 3.5 | 1.9 | −82 | 2.1 | 1.7 | −21 |
Average overloading time (min) | 5.86 | 2.00 | −193 | 2.33 | 2.1 | −11 |
aWidth parameter adjustment only in corridors, plazas, sidewalks, roads, pervious areas, and others sub-catchments.
Comparison of the two modeling approaches
Model development
As described in Sections 2.3.1 and 2.3.2, the two modeling approaches employed the same drainage network. The biggest difference in terms of model setup was found in the effective description of the rainfall–runoff response. In the case of the HD 1D model, the sub-catchments' parametrization was a time-consuming task since each land-use polygon was considered a unique sub-catchment, avoiding attribute averaging. Moreover, a flow routing strategy had to be defined to accurately represent the overland flow dynamics. Therefore, additional tasks of geometry simplification and tag identification were necessary to facilitate the parametrization of each sub-catchment in terms of slope, flow length, and impervious percentage area.
Modeling environment in PCSWMM to develop (a) the highly discretized 1D model and (b) the coupled 1D–2D model. Please refer to the online version of this paper to see this figure in color: http://dx.doi.org/10.2166/wst.2023.173.
Modeling environment in PCSWMM to develop (a) the highly discretized 1D model and (b) the coupled 1D–2D model. Please refer to the online version of this paper to see this figure in color: http://dx.doi.org/10.2166/wst.2023.173.
The simulation run time of all 1D scenarios was 1 s, while the average time in the 1D–2D simulations was 8 min. The average size of the output file following a 1D model approach was 0.01 GB, while in the case of the coupled 1D–2D model, it was 0.27 GB.
The runoff continuity and flow routing continuity (FRC) errors were negligible (ranging 0.03–0.58%) and did not exceed the reasonable threshold of 10% (James et al. 2010) in the two modeling approaches. Nevertheless, the simulations with the 100-year T rainfall event through the 1D–2D model posed an additional computational challenge with an extremely high FRC error (−7 856 430.2%). This condition is likely related to the 2D junction's rainfall assignation and the 2D conduits' routing capabilities, which are affected by the low-lying nature of the study area. Therefore, the dynamic wave's normal flow criterion (which determines when supercritical flow occurs in a conduit) was changed from ‘Slope & Froude’ to ‘Froude number (FN)’ only, i.e., FN > 1.0. After this substitution, the simulations with a design storm of 100-year T achieved an average FRC error of 0.45%.
Reproducibility of 1D–2D model results via the HD 1D model
Table 5 displays the average RE and RMSE values for each hydrologic and hydraulic performance metric. Flood control potential indicators were not included given the reduced capacity of simpler models to reproduce the overland flow response, which may lead to a biased comparison with the coupled 1D–2D model results.
Average RE and RMSE results from comparing the two modeling approaches
Parameter . | Average RE (%) . | RMSE . |
---|---|---|
Hydrologic control | ||
Peak discharge | 1 | 0.15 m3/s |
Total outflow volume | 1 | 556 m3 |
Hydraulic control | ||
Number of surcharged nodes | −1,260 | 27 junctions |
Number of flooded nodes | −103 | 2 junctions |
Overloaded conduits’ length | −14 | 0.26 km |
Average overloading time | −32 | 0.95 min |
Parameter . | Average RE (%) . | RMSE . |
---|---|---|
Hydrologic control | ||
Peak discharge | 1 | 0.15 m3/s |
Total outflow volume | 1 | 556 m3 |
Hydraulic control | ||
Number of surcharged nodes | −1,260 | 27 junctions |
Number of flooded nodes | −103 | 2 junctions |
Overloaded conduits’ length | −14 | 0.26 km |
Average overloading time | −32 | 0.95 min |
The HD 1D model's ability to reproduce the hydrologic response simulated by the coupled 1D–2D model was satisfactory, in line with the validation results. In terms of hydraulic response, the general tendency of the HD 1D model is to overestimate the results of the dual drainage model. The best agreement was obtained in predicting the overloaded conduits' length and the average overloading time, with RMSE values of 0.26 km and 0.95 min, respectively. Although the RE results for the number of flooded nodes were above 50%, the RMSE value of two junctions indicates acceptable model performance. The results of both RE and RMSE for the number of surcharged junctions revealed an unsatisfactory performance of the HD 1D model in this regard.
SUDS performance
This section shows the performance results obtained using the coupled 1D–2D model, as dual drainage models provide more accurate results due to the detailed representation of surface geometry, representation of 1D and 2D flow interactions, and overland flow simulation capacity (Leandro et al. 2009; Vojinovic & Tutulic 2009; Blanc et al. 2012; Chang et al. 2015).
Peak discharge and total outflow volume
SUDS hydrologic control capacity: (a) peak discharge and (b) total outflow volume in each of the tested scenarios, following the coupled 1D–2D modeling approach. Please refer to the online version of this paper to see this figure in color: http://dx.doi.org/10.2166/wst.2023.173.
SUDS hydrologic control capacity: (a) peak discharge and (b) total outflow volume in each of the tested scenarios, following the coupled 1D–2D modeling approach. Please refer to the online version of this paper to see this figure in color: http://dx.doi.org/10.2166/wst.2023.173.
The MIX5 scenario yielded the highest reduction efficiencies in both peak discharge (5.2%) and total volume (5.4%). It is worth noting that the results of the two hydrologic control capacity indicators in the MIX scenarios were consistent with the individual RWHS and TP performance results, suggesting a cumulative effect. For instance, although RWHS potential to reduce peak discharge ranged between 2.2 and 4.8% and TP efficiencies were quite reduced in all cases (<1%), the two SUDS strategies worked in tandem, resulting in better performance under the MIX scenarios.
Surcharged and flooded nodes
The maximum number of surcharged and flooded nodes, i.e., 3 and 4, respectively, occurred in the 100-year storm scenarios, which represents only 3 and 5% of the total number of nodes in the study area. This indicates an acceptable design capacity of the drainage network under mild and heavy rain conditions.
RWHS proved to be efficient in reducing the number of flooded nodes under 5 and 100-year T design storms: from 2 to 0 and from 4 to 3, respectively. The MIX scenarios achieved the same efficiencies as the RWHS simulations, most likely due to the null efficiency demonstrated in all TP scenarios, i.e., no additional hydraulic control contribution for the MIX scenarios. Moreover, none of the SUDS scenarios were effective in reducing the number of surcharged nodes.
Overloaded conduits' length and average overloading time
SUDS hydraulic control capacity: (a) overloaded conduits’ length and (b) average overloading time in each of the tested scenarios, following the coupled 1D–2D modeling approach. Please refer to the online version of this paper to see this figure in color: http://dx.doi.org/10.2166/wst.2023.173.
SUDS hydraulic control capacity: (a) overloaded conduits’ length and (b) average overloading time in each of the tested scenarios, following the coupled 1D–2D modeling approach. Please refer to the online version of this paper to see this figure in color: http://dx.doi.org/10.2166/wst.2023.173.
The reduction of overloading time ranged between 11 and 63 and 12 and 63% through the RWHS and MIX scenarios, respectively. The best performance was achieved through the RWHS5 scenario. According to the time reduction efficiencies achieved via TP scenarios ranging between 1 and 8%, the MIX scenarios reflected a cumulative effect of RWHS and TP capabilities. Furthermore, unlike the overloading length results, the overloading time reduction potential tends to decrease when rainfall T increases.
Flood control potential
Table 6 summarizes each modeled scenario's flood control potential in terms of nodal inundation depth, moderate waterlogging condition, and severe waterlogging condition reductions. The maximum nodal inundation depth was 7 cm (100-year T design storm), which contrasts with the presence of waterlogged areas (water depth ≥15 cm). This indicates (i) an acceptable drainage network performance under extreme rainfall conditions and (ii) the significant influence of the study area's low-land nature on overland flows.
Flood control potential in the coupled 1D–2D modeling approach
Design storm T (years) . | Scenario . | Nodal inundation depth . | Area; moderate waterlogging conditiona . | Area; severe waterlogging conditionb . | |||
---|---|---|---|---|---|---|---|
Max. (cm) . | Reduction (%) . | Total (m2) . | Reduction (%) . | Total (m2) . | Reduction (%) . | ||
5 | BAU | 2 | 11,842 | 1,045 | |||
RWHS | 0 | 100 | 11,507 | 2.8 | 997 | 4.6 | |
TP | 1 | 50 | 11,764 | 0.7 | 1,045 | 0.0 | |
MIX | 0 | 100 | 11,507 | 2.8 | 997 | 4.6 | |
10 | BAU | 3 | 15,154 | 1,126 | |||
RWHS | 1 | 67 | 14,839 | 2.1 | 1,107 | 1.7 | |
TP | 3 | 0 | 15,079 | 0.5 | 1,126 | 0.0 | |
MIX | 1 | 67 | 14,828 | 2.2 | 1,107 | 1.7 | |
100 | BAU | 7 | 20,286 | 1,493 | |||
RWHS | 6 | 14 | 19,683 | 3.0 | 1,492 | 0.1 | |
TP | 7 | 0 | 20,250 | 0.2 | 1,493 | 0.0 | |
MIX | 6 | 14 | 19,610 | 3.3 | 1,492 | 0.1 |
Design storm T (years) . | Scenario . | Nodal inundation depth . | Area; moderate waterlogging conditiona . | Area; severe waterlogging conditionb . | |||
---|---|---|---|---|---|---|---|
Max. (cm) . | Reduction (%) . | Total (m2) . | Reduction (%) . | Total (m2) . | Reduction (%) . | ||
5 | BAU | 2 | 11,842 | 1,045 | |||
RWHS | 0 | 100 | 11,507 | 2.8 | 997 | 4.6 | |
TP | 1 | 50 | 11,764 | 0.7 | 1,045 | 0.0 | |
MIX | 0 | 100 | 11,507 | 2.8 | 997 | 4.6 | |
10 | BAU | 3 | 15,154 | 1,126 | |||
RWHS | 1 | 67 | 14,839 | 2.1 | 1,107 | 1.7 | |
TP | 3 | 0 | 15,079 | 0.5 | 1,126 | 0.0 | |
MIX | 1 | 67 | 14,828 | 2.2 | 1,107 | 1.7 | |
100 | BAU | 7 | 20,286 | 1,493 | |||
RWHS | 6 | 14 | 19,683 | 3.0 | 1,492 | 0.1 | |
TP | 7 | 0 | 20,250 | 0.2 | 1,493 | 0.0 | |
MIX | 6 | 14 | 19,610 | 3.3 | 1,492 | 0.1 |
aModerate waterlogging condition: 15 cm ≤ overland water depth < 30 cm.
bSevere waterlogging condition: overland water depth ≥30 cm.
Most of the flood control indicators performed better under the mildest design storm (5-year T), although the best moderate waterlogging control efficiencies were achieved under the 100-year scenario. RWHS demonstrated flood control capacity in all tested scenarios to some extent, although efficiencies achieved under more severe rainfall conditions (10 and 100-year design storms) were quite low. TP implementation showed a positive impact on nodal inundation reduction only under the 5-year rainfall condition and a negligible or null impact in all scenarios of moderate and severe waterlogging conditions. This most likely influenced the efficiencies achieved by the MIX scenarios, which were very similar to the impact produced by the RWHS implementation.
SUDS implementation: local constraints vs full potential
Hydrologic and hydraulic performance of SUDS in BAU100, MIX100, and MULT100 scenarios
Performance indicator . | BAU100 . | MIX100 . | MULT100 . |
---|---|---|---|
Peak discharge (m3/s) | 8.46 | 8.25 | 7.51 |
Total outflow volume (m3) | 27,530 | 26,640 | 23,270 |
Number of surcharged nodes | 3 | 3 | 3 |
Number of flooded nodes | 4 | 3 | 2 |
Overloaded conduits’ length (km) | 2.7 | 2.6 | 2.1 |
Average overloading time (min) | 5.81 | 5.13 | 0.97 |
Maximum nodal inundation depth (cm) | 7 | 6 | 3 |
Area, moderate waterlogging condition (m2)a | 20,286 | 19,610 | 12,589 |
Area, severe waterlogging condition (m2)b | 1,493 | 1,492 | 1,171 |
Performance indicator . | BAU100 . | MIX100 . | MULT100 . |
---|---|---|---|
Peak discharge (m3/s) | 8.46 | 8.25 | 7.51 |
Total outflow volume (m3) | 27,530 | 26,640 | 23,270 |
Number of surcharged nodes | 3 | 3 | 3 |
Number of flooded nodes | 4 | 3 | 2 |
Overloaded conduits’ length (km) | 2.7 | 2.6 | 2.1 |
Average overloading time (min) | 5.81 | 5.13 | 0.97 |
Maximum nodal inundation depth (cm) | 7 | 6 | 3 |
Area, moderate waterlogging condition (m2)a | 20,286 | 19,610 | 12,589 |
Area, severe waterlogging condition (m2)b | 1,493 | 1,492 | 1,171 |
aModerate waterlogging condition: 15 cm ≤ overland water depth < 30 cm.
bSevere waterlogging condition: overland water depth ≥30 cm.
SUDS setting in (a) MIX scenario and (b) MULT100 scenario. Color gradients represent the number of units of RWHS and TP, per building and sidewalk, respectively. RWHS, rainwater harvesting systems; TP, tree pits; AST, attenuation storage tanks; GR, green roofs; PP, pervious pavements. Please refer to the online version of this paper to see this figure in color: http://dx.doi.org/10.2166/wst.2023.173.
SUDS setting in (a) MIX scenario and (b) MULT100 scenario. Color gradients represent the number of units of RWHS and TP, per building and sidewalk, respectively. RWHS, rainwater harvesting systems; TP, tree pits; AST, attenuation storage tanks; GR, green roofs; PP, pervious pavements. Please refer to the online version of this paper to see this figure in color: http://dx.doi.org/10.2166/wst.2023.173.
Waterlogged areas (Aw), i.e., water depth ≥15 cm in (a) BAU100 scenario, (b) MIX100 scenario, and (c) MULT100 scenario. Please refer to the online version of this paper to see this figure in color: http://dx.doi.org/10.2166/wst.2023.173.
Waterlogged areas (Aw), i.e., water depth ≥15 cm in (a) BAU100 scenario, (b) MIX100 scenario, and (c) MULT100 scenario. Please refer to the online version of this paper to see this figure in color: http://dx.doi.org/10.2166/wst.2023.173.
DISCUSSION
Model development
The model development process confirmed the observations of several authors stressing the high computational cost in terms of time, memory, and power (Leandro et al. 2009; Leitão et al. 2010) and the propensity for instability of 1D–2D models (Vojinovic & Tutulic 2009). Depending on the case study, this can restrict the number of iterations and/or simulations, affecting the significance of the results. However, the final application, the study area size, and the available computational tools (processing and modeling) might aid in the reduction of these limitations.
The relatively easy setup and parametrization of sewer-based models have been highlighted as advantages compared to sewer/surface approaches (Vojinovic & Tutulic 2009; Chang et al. 2015; Pina et al. 2016). Nevertheless, time-consuming tasks under the HD 1D model, such as sub-catchments' attribute assignation and flow routing definition, can be simplified using coupled 1D–2D models when a high-resolution DEM is available. This suggests that, in addition to the well-known benefits of the more complex models in terms of results accuracy and visualization, model construction may be an additional advantage depending on the size of the study area and the model's final application.
The accurate representation of the flow interactions between TP and the drainage system posed a challenge in the present study. The current option to represent any type of SUDS in stormwater management model (and therefore, in PCSWMM) is as an attribute of a given sub-catchment. This approach is useful for the hydrologic control analysis; however, the hydraulic control assessment may be affected. This limitation could be overcome through a ‘virtual node-orifice-flap valve’ scheme, as suggested by Pina et al. (2016) for linking the major and minor drainage systems in simpler models. In this method, virtual nodes allow overland flow (from sub-catchments) to be discharged to the pipe-ended system, orifices represent sewer inlets (with limited capacity), and flap valves allow the transition of sewer flow to the surface. Nevertheless, given the resolution of both the HD 1D model and the coupled 1D–2D model, this would imply a time-consuming task for each of the implemented TP units. Furthermore, the representation of TP's internal hydraulic processes (e.g., backflow effect) cannot be fully represented. Further development is required to improve the hydrodynamic assessment of SUDS at the micro (on-site) scale, with either 1D or 1D–2D modeling approaches.
Reproducibility of 1D–2D model results via the HD 1D model
The majority of comparative studies addressing the performance of sewer-based models and sewer/surface approaches have focused on flood impact assessment in terms of water depth and extent (Vojinovic & Tutulic 2009; Chang et al. 2015). The present study aimed to broaden the existent literature by assessing the hydrologic–hydraulic performance of SUDS under a sewer-based HD 1D model and a sewer/surface coupled 1D–2D model. The outcomes of these two modeling approaches revealed both similarities and differences, depending on the aspect evaluated. For instance, peak discharge and total outflow volume reductions were overestimated by the 1D approach by a maximum of 3 and 4%, respectively. This contrasts with a previous study comparing a semi-distributed (sewer-based) and a fully distributed (sewer/surface) model in which the former overestimated the volume discharged by up to 100% (Pina et al. 2016). The consistency of the results from the two modeling approaches in the present work could be attributed to the level of discretization of the simplest model, which avoided parameter averaging. This suggests that a HD 1D model can provide comparable results to a coupled 1D–2D model when assessing the hydrologic control capacity of SUDS strategies such as RWHS and TP.
Regarding the SUDS hydraulic control capabilities, the differences between the results provided by the HD 1D and the 1D–2D modeling approaches were not as uniform as the hydrologic performance values. This outcome is reasonable considering the detailed flow path representation offered by the 2D mesh, whereas a routing strategy was followed in the HD 1D model to recreate this effect. However, as observed during the validation process, the width parameter had a great influence on the hydraulic control capabilities. Although the 1D model continued to overestimate the results of the 1D–2D model, the best agreement between the two modeling approaches was found in the number of flooded nodes and the overloaded conduits' length. It is worth noting that the hydraulic control results are subject to the quality of the terrain information and the urban configuration of the study area (Leandro et al. 2009; Vojinovic & Tutulic 2009). Nevertheless, the results of the present study suggest that a HD 1D model can provide meaningful results for the vulnerability assessment of the drainage network, enabling a more efficient response to avoid waterlogging and urban flood inundation in less extreme rainfall events.
The ability to reproduce the hydrologic and hydraulic control results of a coupled 1D–2D model through a HD 1D model represents an advantage in terms of computational cost, data acquisition and processing, and model development. This is useful for urban water decision-making when large areas and multiple scenarios need to be assessed. However, for the results to be conclusive, a proper calibration process with observed data must be performed.
SUDS performance
The study of SUDS' hydrologic control capabilities has been widely covered in the literature of RWHS (Aceves & Fuamba 2016; Hua et al. 2020) and TP (Grey et al. 2018). RWHS demonstrated greater capacity than TP in terms of peak discharge and total outflow volume reductions. However, better efficiencies were achieved under the MIX scenarios. This confirmed the findings of previous studies in which greater hydrologic control effectiveness is reached when multiple typologies are implemented (Ahiablame & Shakya 2016; Cui et al. 2019). The percentage of implementation (Hua et al. 2020) and the land-use distribution (Pina et al. 2016) play an important role in achieving better efficiencies, either individually or in combination. Furthermore, the hydrologic control capacity of both RWHS and TP was reduced with the most extreme rainfall event (100-year T), which has been described in previous studies, indicating better efficiencies for mild events (Grey et al. 2018; Huang et al. 2018; Mu et al. 2022).
The hydraulic control provided by RWHS and TP, assessed using two different modeling approaches, represents a novelty in the study of SUDS. The present research revealed that, in addition to hydrologic control capabilities, RWHS can aid in the reduction of the number of flooded junctions, the overloaded conduits' length, the overloading time, the nodal inundation depth, and the waterlogging extent (moderate and severe conditions). Furthermore, although to a much lesser extent, TP demonstrated efficiency in reducing average overloading time and nodal inundation depth. Previous studies assessing the impact of rain gardens, whose water balance is theoretically similar to that of TP, have found better hydraulic control efficiencies (Cui et al. 2019; Movahedinia et al. 2019). This suggests that higher implementation densities or TP management trains might enhance TP's hydraulic control efficiency. Nevertheless, as with hydrologic control capabilities, the hydraulic control effectiveness is subject to the impact of rainfall volume and intensity (Cui et al. 2019; Hua et al. 2020; Mu et al. 2022).
The SUDS implementation strategy adopted in the present study was an attempt to represent a realistic implementation in both public and private land: while RWHS implementation potential was reduced by more than 50% due to physical restrictions and local context constraints, the final distribution of TP considered the current tree distribution, capturing the preferences and operation and maintenance capabilities of the local administration. The present study demonstrated that, under these limitations, TPs do not provide substantial hydrologic–hydraulic control effectiveness. Nevertheless, previous research has demonstrated TP's significant contribution to reducing urban runoff under careful planning (Grey et al. 2018). Other benefits such as temperature regulation, improved stormwater runoff quality, increased biodiversity, and esthetic value (Lamond et al. 2015) are relevant for assessing TP impact given the study area's high urbanization rate (81%).
The scenario approach proved to be a useful tool for assessing the role of SUDS in supplementing the existing drainage network at the catchment scale. Greater efficiencies in the reduction of all hydrologic and hydraulic control indicators were seen with the implementation of multiple SUDS typologies and higher percentages of installation (MULT100 scenario). Earlier studies have discussed that ponds and natural low-lying areas are important for flood control as they provide a large volume for the detention of excess water (Villarreal et al. 2004; Sörensen & Emilsson 2019). However, these measures were not tested in the MULT100 scenario, considering the high limitations of the built environment. Despite this, the reduction of the moderate and severe waterlogging extents was efficient to a certain extent. Further discussion is needed to assess the cost-effectiveness of an integrated scheme implementing RWHS, TP, GR, AST, and PP.
CONCLUSION
This study tests and discusses data requirements (quality and quantity), methodological differences, computational limitations, and the hydrologic–hydraulic performance of SUDS using a HD 1D model and a coupled 1D–2D model using PCSWMM. After 24 simulations, it was evident that HD 1D models are able to reproduce the hydrologic and hydraulic response produced by coupled 1D–2D models. This represents a significant advantage for analyzing the vulnerability of the existing drainage network and SUDS' complementary performance in data-scarce environments while avoiding the high computational costs of sewer/surface models.
Furthermore, in addition to the well-known hydrologic control capabilities (peak discharge and total outflow volume control), findings revealed that SUDS can aid in the reduction of the number of flooded junctions, overloaded conduits' length, overloading time, nodal inundation depth, and waterlogging extent. The use of multiple SUDS typologies and/or higher percentages of installation, aiming to improve on-site and catchment scale stormwater control, is a key to achieving better efficiencies. Moreover, performance results across all evaluated criteria demonstrated that, even in highly urbanized, waterlogging- and flood-prone catchments with limited implementation potential on both public and private lands, SUDS can provide hydrologic–hydraulic control to some extent, mainly during less extreme precipitation events.
The main limitation of the present study was the lack of observed data to perform a proper calibration process. Nevertheless, the validation strategy, along with the comparative model-based multi-scenario assessment, enables the present study to provide significant results for supporting sustainable stormwater decision-making and SUDS planning at the urban scale in flood-prone areas with highly restricted land use. Further development is needed for an accurate description of the internal hydraulic mechanisms of SUDS typologies that interact with overland flow to improve the understanding of particular hydraulic phenomena such as the backflow effect and flow aggregation from multiple sources.
ACKNOWLEDGEMENTS
The authors would like to acknowledge Bogotá's Local Water Utility (EAAB) for providing the drainage network and LiDAR terrain survey data. The authors would also like to acknowledge Computational Hydraulics International (CHI) for providing a university grant to use PCSWMM in the initial stage of this study.
DATA AVAILABILITY STATEMENT
All relevant data are included in the paper or its Supplementary Information.
CONFLICT OF INTEREST
The authors declare there is no conflict.