Modelling of rainwater reduction and hydrological performance of selected green infrastructure (GI) facilities in urban catchments
Publicado en línea: 11 mar 2025
Páginas: 1 - 20
DOI: https://doi.org/10.2478/oszn-2025-0005
Palabras clave
© 2025 Krzysztof Muszyński et al., published by Sciendo
This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 3.0 License.
According to a memorandum by the United States Environmental Protection Agency (USEPA) [Grumbles 2007], green infrastructure (GI) refers to objects that mimic natural processes of infiltration and evapotranspiration while enabling the proper recharge of groundwater and maintaining appropriate flow, as well as hydrological and hydraulic stability. Global warming and ongoing climate change increase the risk of flooding [Muller 2007], increasing the need for proper rainwater management. However, this is a major logistical and technical challenge, especially in urban areas, where the existing and often dense development makes it difficult or completely impossible to use GI.
Rainwater drainage systems effectively manage water runoff in urbanized areas; however, under certain conditions, such as heavy rainfall, they are the infrastructural component that contributes most to hazards, such as flash floods. Such phenomena have been successfully modelled using the Storm Water Management Model (SWMM) engine [Nur Oktalinov Fikri et al. 2022], which helps understand the impact of the increase in rainfall frequency on the formation of hazards and in the process improving the operation of drainage systems. It is important that spatial databases be integrated with hydrological and meteorological information, as this enables the identification of relationships between individual processes both directly and indirectly linked to flash flooding [Tarek et al. 2017]. One of the options for mitigating the effects of flooding in urban spaces is retention reservoirs, which may be used even in such difficult places as airport areas [Barszcz 2017]. The location of these retention reservoirs may also be optimized in an urban rainwater drainage system [Wang et al. 2017]. SWMM modelling is used for detailed analysis of the structure of such networked reservoirs in terms of minimizing and preventing overloading of rainwater collectors [Dąbrowski et al. 2022]. This is influenced by the location of the rainwater supply and drainage channels, among other factors.
Drainage-equipped rain gardens, also known as bio-retention cells (BC), are relatively simple in their design and operation, and their functionality has been researched in a variety of applications, including planning methods and urban agriculture; assessment approaches; sustainability and climate adaptation; and ecosystem services and their benefits [Korkou et al. 2023]. The well-documented positive outcomes of these uses [Choi et al. 2021] stem from employing natural processes to mitigate the impacts of climate change and supporting other socially sustainable development goals, including both quantitative and qualitative protection of receiving waters; runoff reduction; minimization of flooding; the prevention of groundwater pollution in urban ecosystems; increased biodiversity; and improving microclimate control [Kasprzyk et al. 2022] by lowering air temperatures and increasing ambient humidity. It has also been observed [Wang et al. 2018] that the quality and quantity of peak surface runoff are more significantly influenced by urbanization factors (especially by land-use intensity and population density) than by those related to climate change. Therefore, the implementation of GI and the reduction of impervious surface areas should be given serious consideration in both existing and newly-developed urban areas.
GI has been modelled in the SWMM software by researchers and engineers worldwide, creating an example of Low Impact Development (LID). GI is recognized as a popular tool for reducing the quantity and improving the quality of surface runoff from impervious surfaces [Roy et al. 2014], [Huang et al. 2020]. It has an important role in addressing problems that have arisen from the interaction between urbanization, climate change, and demography.
It is well known that, due to impervious surfaces’ concomitant limitation of biologically active areas, surface runoff is increased and accelerated [Li et al. 2018], [Zhang et al. 2018], and stormwater pollutant concentrations are increased [Wang et al. 2013], [Das et al. 2018], [Godyń et al. 2023]. GI has been demonstrated [Rahman et al. 2020] to increase transpiration and reduce ambient temperatures, contributing to the minimization of urban heat islands while effectively reducing pollutants such as total suspended solids [Tiveron et al. 2018]. Godyń et al. [2024] estimated the impact of GI at the urban-catchment scale and assessed its ability to mitigate environmental risks. Another research project evaluated the benefits and effectiveness of 165 rain barrels and 81 rain gardens installed on 30% of properties in four experimental catchments [Roy et al. 2014]. Measures taken to mimic natural ecosystems have been proven to have a variety of ecosystem benefits.
The runoff of rainwater from green roofs and bio-retention cells was successfully modelled using SWMM [Jeffers et al. 2022], demonstrating the accuracy of applying SWMM LID for such calculations (NSE > 0.81) while considering reasonable assumptions for individual parameters. Hamouz et al. [2020] found that the implementation of a green roof on just 11% of a roof’s area can significantly reduce maximum runoff. McCutcheon et al. [2013] modelled GI using SWMM for both long-term simulations and single precipitation events and found that it generates reasonable and relevant results. On the other hand, modelling the hydrological response of green roofs requires the adaptation of many parameters, and the results of the simulations have not always been satisfactory [Burszta-Adamiak et al. 2013]. For most of the tested precipitation events, the model had limited capabilities in accurately simulating the hydrograph of rainwater runoff from green roofs, with Nash coefficients showing negative values. Therefore, the quality of green roof simulation results remains unproven.
Rezaei et al. [2019] performed an impact assessment using SWMM for vegetated swales and rain gardens to measure quantitative runoff reduction and the improvement of surface runoff quality, using field data and synthetic rainfall at high intensities of 70–90 mm. Their results showed that the peak runoff reduction decreased as rainfall intensity increased from 27% to 19%. The model was calibrated on a single actual rainfall event with an intensity of 12.5 mm; implementation of GI was followed by a peak runoff reduction of 23%.
An important study carried out in the city of Parma, Ohio analysed a 3-year period of GI operation in urban conditions [Jarden et al. 2015]. The analysis included street-connected bio-retention cells, rain gardens, and rain barrels serving about 45.7% of the residential catchment area (mainly impervious surfaces). The installation and use of GI on selected streets in a residential area of the city resulted in a peak flow reduction of approximately 33% and a total storm runoff reduction of 40%.
Reductions in peak flow and runoff volume to relieve the load on existing rainwater drainage systems can have synergistic benefits by minimizing negative impacts on receiver waters through reducing the frequency of flash floods, improving the infiltration process, increasing the recharge of groundwater aquifers [Zhang et al. 2019], and improving the habitats of regional species of flora and fauna [Granados-Olivas et al. 2016].
Actual studies of the impact of GI on receivers are not easy, but are critically needed. Urban-catchment-scale modelling is essential, and the simulation results should provide guidance for implementation of GI. A significant amount of research has been carried out on the calibration of models built with the use of continuous long-term data (e.g., [D’Ambrosio et al. 2022], [Epps et al. 2019], [Cipolla et al. 2016], and [Del Giudice et al. 2016]), as well as models based on single precipitation events (e.g., [Arjenaki et al. 2020], [Yim et al. 2016]). There were also some studies where the authors did not use a calibration process to assess model quality due to a lack of data (e.g., [Agarwal et al. 2019]) or calibrated their models using the comprehensive runoff coefficient (CRC) method [Bai et al. 2019].
The results of the research mentioned above typically focus on the percentage reduction in rainwater quantity. Exceptionally, the performance of bioretention systems has been presented as the temporal behaviour of the variables such as inflow, outflow, storage capacity and infiltration in a bioretention box experiment [Batalini de Macedo et al. 2018]. These studies were carried out under laboratory conditions; the most valuable practical studies are those carried out in the field [Burszta-Adamiak et al. 2023], where, among other things, selected rainfall events were observed during which water was ponding in a rain garden. The amount of retention in the rain garden was directly influenced by the thickness of the soil layers and the infiltration process, which are described in detail by Kravchenko et al. [2024]. Field studies, laboratory experiments and mathematical modelling integrate specific indicators, and their incorporation contribute to the development of more scientific and practical systems for GI in urban environments [Mao et al. 2024].
Therefore, unit results are important because they characterise the operation of equipment under specific given conditions. Such results are difficult to generalise, but, they can be of great importance in engineering practice locally, and can serve as models for other cases with similar conditions. Considering the knowledge gaps in the field of GI efficiency and hydrological performance [Jiang et al. 2015], this study focused on the important operating parameters of BC in single-family residential buildings in the northern part of Cracow (Poland). The analysis of the existing rainwater drainage system operation, and the size and location of surface runoff routes and depression storage sites in a simulation without GI, helped identify places that are potentially more suitable for the construction of GI, along with their individual unit loads and operating parameters.
The aims of this study were to: (1) verify the possibility of implementing GI to reduce the amount of surface runoff of rainwater in an existing urbanised area; (2) show the relationship between the average reductions in total volume and peak flows for individual BC under different rainfall durations and intensities; and (3) estimate the rainwater inflow limit for an area with GI facilities under different rainfall durations and intensities at which there is a 100% quantitative reduction of inflow.
A hydrodynamic model built in the SWMM environment was used to assess the impact of GI on the reduction in rainwater runoff to the receiver from the residential catchment area under study. Simulations were performed for the existing conditions using the current development conditions of the study area. The course of the rainwater drainage system was then mapped, and the rainfall and flow volumes were measured at the W1 outfall (the location is shown in Figure 1) to the receiver on August 26, 2023.
Location of the measuring point in the study area with catchment for the existing outfall W1
A rainfall model created on the basis of the distribution of the phase series of rainfall maxima from 33 years (1986–2019), which is considered the most reliable tool for the dimensioning of rainwater drainage systems in Cracow, was used [Precipitation model 2024]. Calculations were performed for design rainfall with the DVWK method for the probability of occurrence of p = 10% and a duration of 15 to 180 minutes, in accordance with the recommendations of the Cracow model for dimensioning rainwater drainage systems, based on PN-EN 752:2017-06 [2017].
The prospective conditions included eight scenarios: S0 for real precipitation, and S1–S7, as shown in Table 1. For scenarios S0–S7, selected drainage-equipped rain gardens – bio-retention cells (BC) – were integrated into the existing settlement layout.
Analyzed scenarios
| 90 min (4.2 mm) | 15 min | 30 min | 45 min | 60 min | 90 min | 120 min | 180 min | |
| 7.8 | 273.32 | 175.33 | 134.52 | 105.37 | 74.81 | 59.55 | 42.46 | |
| 4.2 | 24.60 | 31.56 | 36.32 | 37.93 | 40.40 | 42.88 | 45.86 | |
| 2.80 | 98.40 | 63.12 | 48.43 | 37.93 | 26.93 | 21.44 | 15.29 |
LPS – litres per second
The research approach used at the initial stage was to integrate the available databases. These were the Digital Terrain Model 1m; the Database of Topographical Objects; and the National Integration of Utilities Networks [GUGiK 2023], along with the GI Concept. The GI Concept incorporated drainage-equipped rain gardens – represented in SWMM as bio-retention cells (BC) – with the parameters for BC sourced from the literature. The quality of the input data determined all further spatial analysis activities, which were carried out using Geographic Information System (GIS) and Geographic Resources Analysis Support System (GRASS) software.
The results delineated the watersheds, identifying the direction and structure of the runoff flow path, the land use, and a complete and detailed rainwater drainage system. GI planning consisted of incorporating selected forms of GI into the existing urban layout of the housing estate in such a way as to maintain the entire existing development of the housing estate while changing the elevation shape as little as possible. The above analyses and data aggregation provided the foundation for creating a hydrodynamic model of the studied area in the SMWW environment. All sub-catchments were delineated and characterized in detail, including a description of their fraction of impervious area [Porębska et al. 2023].
In addition, the existing rainwater inflows to the junctions were analysed, and the drainage system supply was carefully assessed. The model, based on the precipitation and flow measurement data obtained at the W1 outfall, was calibrated and evaluated. Finally, the model was used to run GI scenarios, and the model responses were compared and subjected to a detailed analysis.
The correctly calibrated model was fed design rainfall hyetographs for scenarios S0–S7. Simulations were carried out for both the existing and prospective conditions, enabling comparisons of peak flow, total flow reduction and other operating parameters of the GI facilities.
The analysed area, covering a total of 4.17 ha, is located in the northern part of the city of Cracow (Poland) and lies entirely within the hydrological catchment of the Sudół River, a right-bank tributary of the Prądnik (Białucha) River. The area is built up with housing estates of single-family residential buildings largely built around 1995 and arranged mostly in rows along a north-south axis, as shown in Figure 1. The results of spatial analyses conducted using data from the Database of Topographical Objects with the use of GIS tools are presented in Table 2.
Spatial analyses based on the Database of Topographical Objects of the study area
| Roofs | 1.12 | 26.9 |
| Roadways, pavement, alleys | 0.40 | 9.6 |
| Single- and multi-family residential land | 2.17 | 52.0 |
| Biologically active areas | 0.48 | 11.5 |
The predominant land use class in the study area was vegetated (typically trees and shrubs on lawns between housing blocks) with a small portion of the area comprising pedestrian and bicycle routes, as well as single-family residential land. Multi-family residential areas were a fragment of the adjacent catchment and comprised a negligible amount of the study area (0.03 ha). Grassland areas receive low runoff rates due to the significant presence of trees with broad, spreading crowns. The pedestrian and bicycle routes, separated as alleys, had mostly permeable surfaces. In addition, the roadways, including local and other roads, were made of paving blocks or hardened gravel. All the above indicates that 63.5% of the catchment area under investigation was biologically active.
According to the maps, the roofs of all existing buildings (119 units) were directly connected to the rainwater drainage system by gutter systems. The roadways, mainly covered with prefabricated paving blocks in poor condition, were equipped with inlet grates. Rainwater that does not accumulate in depression storage or manage to infiltrate joints thus predominantly enters the rainwater drainage system directly. The entire area, as shown in Figure 1, discharges into the receiver — a ditch that is a right-bank tributary of the Sudół River.
The gravity-based rainwater drainage system consists of 116 junctions, with an average depth of 2 m below ground level, and pipes totalling 1.38 km in length, with diameters of 0.15–0.5m. Outfall W1 was measured to investigate flows during precipitation. At a distance of 350 m from the centroid of the W1 catchment, there was a rain gauge belonging to the research network of Cracow University of Technology, Faculty of Environmental Engineering and Energy, Department of Geoengineering and Water Management.
For the purpose of determining the height and slopes of the land surface, a detailed Digital Terrain Model with a 1 m grid was used. Using this data and GIS-GRASS tools, the directions of surface runoff, subcatchments, and the slope of the land in the northern direction (towards the receiver) were determined at an average of 1.4%. The maximum elevation within the entire study area is 228.4 meters above sea level (m a.s.l.), while the minimum is 223.8 m a.s.l.
The analysis of ground conditions revealed that the soils in the study area belong to the coarse B type, characterised by above-average permeability (infiltration coefficient value: 3.8–7.6 mm∙h−1). The soils consist of sandy medium-deep soils, loess, and sandy loams.
With the use of aggregated data, a visualization of the imperviousness of the studied area was created. The results of the analysis are shown in Figure 2, which illustrates the studied area in terms of: (A) the layout of the entire drainage infrastructure, (B) land development, (C) altitude, location, and directions of surface water runoff determined on the basis of terrain slopes, and (D) subcatchments and imperviousness implemented in the SWMM.
Results of detailed analyses of the study area using available databases: A — route of the drainage system; B — land use; C — elevation and directions of surface water runoff; D — subcatchment imperviousness
Advanced modelling requires a calibration process in order to fine-tune many input parameters that are important for the quality of the model’s response [Khan et al. 2022]. In this study, meteorological measurements that had been collected continuously at a measuring station 350 m from the centroid of catchment W1 were used as climatic data. A total rainfall of 4.4 mm was recorded on August 26, 2023, which followed a 7-day dry period. Precipitation occurred in two stages – the first lasted 10 minutes (from 4:48 to 4:58 a.m.) with 0.2 mm of rain, and the second lasted 90 minutes (from 2:21 to 3:51 p.m.) with 4.2 mm of rain (Figure 3). The instantaneous flow during the entire 4.2 mm precipitation event was measured at outfall W1 with an accuracy of 1 minute.
Rainfall mass curve and hyetograph of 4.2 mm precipitation (a) and flow observed at outfall W1 (b) used for the calibration process
The detailed subdivision of the area enabled precise delineation of the surface runoff paths, allowing the subcatchments to be properly connected along these paths and according to the slope of the land, as long as there were no drainage inlets on the surface (such as roads or parking places). The polygons generated through geoprocessing allowed for a high level of mapping accuracy in the surface description of the study area.
The empirical modified Horton model was used to calculate the infiltration process. According to Rossman and Huber [2016], this is one of the best-known computational methods for the infiltration process, and it is widely used in rainfall-runoff process modelling [Sofijanic et al. 2022; Li et al. 2015]. Horton’s method is considered better suited for single-event simulations compared to the Green-Ampt method [Parnas et al. 2021], while its modified formula provides a more accurate estimation of the infiltration process under low rainfall intensities [Rossman, Simon 2022]. The model is based on calculations consistent with the dynamic wave equations for a 0.5-second routing step for both rainy and dry weather. Time-step restriction was calculated using the formula (1) from the dynamic flow routing model EXTRAN [Roesner et al. 1992]. The shortest sections of gutter pipes determined the routing step value (Δ
The parameters describing the subcatchments were subjected to the calibration process. For each separated subcatchment, the currently-existing development (according to the Database of Topographical Objects) was extracted using GIS geospatial analysis tools. Separate parameters characterizing the existing surface, such as slope, the proportion of permeable and impermeable surfaces, Manning’s roughness coefficients, and retention capacity, were assigned to the prepared polygons. Table 2 lists the values of the basic parameters obtained in the calibration process of the single precipitation phenomena under study. The hydraulic width of the outflow stream [m], which is not a physical parameter, was estimated using formula (2) [Nowogoński et al. 2019].
Imperv: the percentage of impervious surface within the catchment area; N Imperv: factor N Perv: factor Dstore Imperv: retention capacity of the impervious surface; Dstore Perv: retention capacity of the pervious surface; Zero Imperv: percentage of the impervious surface with no retention capacity.
Values of the basic parameters for the land use groups in the study area obtained during the SWMM calibration process
| Building roofs-BUBD | 87.47 | 0.010 | 0.045 | 3.69 | 3.69 | 36.00 | 0.7 | |
| Other roadway-SKJZ08 | 13.1 | 36.0 | 0.015 | 0.300 | 1.20 | 5.08 | 40.00 | 0.3 |
| Local roadway-SKJZ06 | 34.6 | 49.5 | 0.010 | 0.300 | 1.20 | 5.08 | 40.00 | |
| Alley-SKRP01 | 2.0 | 3.2 | 0.011 | 0.280 | 1.20 | 5.08 | 40.00 | |
| Grass vegetation-PTTR01 | 1.0 | 2.5 | 0.140 | 0.280 | 5.08 | 7.62 | 20.00 | 0.3 |
| Orchard-PTUT03 | 1.1 | 2.88 | 0.210 | 0.280 | 5.08 | 7.62 | 20.00 | |
| Single-family dwelling-PTZB02 | 2.0 | 4.02 | 0.105 | 0.280 | 5.08 | 7.62 | 20.00 | |
The impervious fractions of the subcatchments were not subjected to the calibration process. The values were determined using a detailed analysis of land cover according to Database of Topographical Objects, along with high-resolution orthophotos and field inspections.
The parameter values were used to determine the correct model response for single precipitation events under the existing conditions. The aim of the calibration was to align the simulated hydrograph with the measured hydrograph. It was important that the total outflow volume, peak discharge, peak time, and rising/falling line matched the measured values as closely as possible. For this purpose, the values of N Imperv, N Perv, Dstore Imperv, Dstore Perv, Zero Imperv, and Parameter A from formula (2) were updated for the subcatchments representing each type of land use. These values for each type of development were selected by gradually adjusting them up or down. The entire process was carried out using a trial-and-error method, with a simulation performed after each parameter change. Based on the results, parameters were calculated for the qualitative assessment of the model, along with a hydrograph plot (Figure 6 for the parameters based on Table 3). Of particular importance in the calibration process were the parameters describing depression Zero Imperv and Dstore Imperv for the impervious parts. The Dstore Imperv and Perv values obtained were close to the range suggested by Endreny [2006].
The hydrodynamic model generated continuity errors of 0.0% for surface runoff and 0.0% for flow routing. These values clearly indicated that the model quality was excellent [Rossman et al. 2022]. In the next step, the basic measures of fitness coefficients were calculated to assess the quality of the constructed model. These were used to compare the model’s response with the measured data. Table 4 summarises the calculated indicators, along with short comments for each. The symbols used in the formulas are as follows:
The basic measures of fitness coefficients for built SWMM model
| NSEC | Nash-Sutcliffe Efficiency Coefficient | Compare the results of model’s response and measurement data |
|
0.885 | -∞ - 1; very good | [Titterington et al. 2017], [Moriasi et al. 2007], [Lin et al. 2017] |
| ISE | Integral Square Error | The accuracy of the matching between simulated and observed data |
|
0.02 | 0 – 3; excellent | [Shamsi et al. 2017] |
| RMSE | Root Mean Square Error | Differences between observed and simulated values |
|
0.442; < 0.5*1.307 | less than half of the standard deviation; good | [Singh et al. 2004], [Moriasi et al. 2007] |
When above basic indicators of the model’s effectiveness were found to be completely satisfactory, the constructed model for a single precipitation event was considered correctly calibrated.
The hydrodynamic model allowed for the evaluation of the rainwater drainage system’s performance in terms of the amount of rainwater runoff from the studied area in the current conditions (accurate mapping). The model was then modified to simulate the operation of the rainwater drainage network under hypothetical conditions in which the study area incorporates GI facilities, represented in the SWMM as LID. LID objects are typically represented by a combination of correctly defined and characterized layers filled with an assumed type of material.
Following Bond et al. [2021], identical parameters were assumed for the BC facilities embedded in the study area, as shown in Table 5, which remain within the ranges of the parameters presented in Hörnschemeyer et al. [2023], with the exception of the drainage coefficient parameter. The LID facilities were differentiated only by the location of the installation and the individual area. In addition, it was assumed that the entire newly inserted subcatchment was occupied by the indicated LID object (the SWMM allows the separation of the LID surface within the selected subcatchment).
Summary of the LID parameters used in the calibrated SWMM model according to [Bond et al. 2021]
| Surface layer | Berm height | mm | 250 |
| Vegetation volume | share | 0.1 | |
| Surface roughness | – | 0.3 | |
| Surface slope | % | 1 | |
| Swale Side Slope | run/rise | – | |
| Soil layer | Thickness | mm | 600 |
| Porosity | share vol. | 0.45 | |
| Field capacity | share vol. | 0.121 | |
| Wilting point | share vol. | 0.057 | |
| Conductivity | mm/hr | 91 | |
| Conductivity slope | – | 44 | |
| Suction head | mm | 50 | |
| Storage layer | Thickness | mm | 400 |
| Void ratio | voids/solids | 0.54 | |
| Seepage rate | mm/hr | 2.6 | |
| Clogging factor | – | 0 | |
| Drain | Flow coefficient | mm/hr | 5.4* |
| Drain exponent | – | 0.5 | |
| Offset height | mm | 200 |
The parameter was calculated using formula (3)
The drainage coefficient parameter is defined as the rate of water outflow through the drainage system and described as a function of the water height in the retention layer above the drainage outlet [USEPA 2022]. Therefore, formula (3), used in Lee et al. [2018] and described in the Drain Advisor, was applied. For a time of
Figure 4 presents a proposal for the implementation of selected LID devices for the existing housing estate under consideration. The cadastral division of land (boundaries of cadastral plots) and the layout of the existing underground infrastructure, excluding rainwater drainage, were omitted, as was any street furniture. When conceptualizing the layout of the LID infrastructure, the provisions of a technical catalogue [Iwaszuk et al. 2019] regarding the area requirements of selected LID structures were taken into account. For BC, the minimum required area was 1% of the drained catchment area. The selected grassy green areas offer opportunities for practical LID implementation, which, combined with underground and above-ground drainage systems, can increase the residents’ living comfort [Liu et al. 2022] and improve rainwater management.
Illustration of the implementation of green infrastructure (BC) in the study area of an existing housing estate.
Rain gardens equipped with a drainage system (represented in SWMM as “bio-retention cells BC”) connected to the existing rainwater drainage network were selected. A total of 32 LID facilities covering 0.474 ha were applied. Line BC systems (16 pcs, with a total area of 0.459 ha) formed the main surface runoff routes approximately 3.8 m wide, and were located in existing biologically active areas. The largest was 836 m2, while the smallest was 70 m2. Street BC (16 pcs, with a total area of 0.015 ha) were built along roads with a total area of 9.2 m2. Where conditions allowed, excess overflows from the LID systems were connected to the downstream and adjacent LID systems, or alternatively, to the existing rainwater drainage network.
The total rainwater runoff at outfall W1 was 14.4 m3 (measured) and 15.2 m3 (simulated). Pearson’s correlation coefficient (
Linear regression plot of R2 for a dataset of the measured rainwater flow (Qo) at outfall W1 and the simulated value (Qs) based on the SWMM
The results of the single-event modelling of the observed real precipitation showed a significant reduction in runoff at the tested outfall W1 with the implementation of LID. The average value of simulated instantaneous flows was 2.21 LPS (litres per second), compared to the measured flows of 2.09 LPS, with standard deviations of 1.516 and 1.307 LPS, respectively. The measured peak flow was 4.54 LPS, recorded at 3:00 p.m., while, according to the SWMM, the maximum flow occurred at 3:07 p.m. and amounted to 4.35 LPS. The flow rate simulated for the existing conditions, measured on August 26, 2023, is illustrated in Figure 6, which also shows the model’s response in the case of LID implementation in the housing estate under study, assuming no outflow.
Flow observed at outfall W1 on August 26, 2023, along with the modelling results for the existing conditions, and for the scenario with the selected LID equipment implemented
The simulation results for scenarios S0 – S7 are summarized in Table 6. The findings demonstrated a significantly positive impact of LID infrastructure on the total volume of rainwater in the drainage system in the studied area, achieving an average reduction of 86% across scenarios S1 – S7. The rate of total runoff reduction (LID vs. non-LID) decreased as the duration and total rainfall increased, with a minimum reduction value of 78% in scenario S7 and a maximum of 95% in scenario S1.
The simulation results of the calibrated model for the entire study area at the sewer outfall for scenarios S0–S7
| Max flow [LPS] | 4.6 | 449.2 | 464.0 | 454.8 | 393.1 | 359.3 | 349.9 | 266.2 |
| Total volume [m3] | 17 | 539 | 752 | 916 | 953 | 1008 | 1072 | 1139 |
| Max flow [LPS] | 0.0 | 17.5 | 31.6 | 31.7 | 25.2 | 22.6 | 22.1 | 17.6 |
| Total volume [m3] | 0 | 29 | 56 | 111 | 134 | 167 | 204 | 246 |
| Max flow [LPS] | 100% | 96% | 93% | 93% | 94% | 94% | 94% | 93% |
| Total volume [m3] | 100% | 95% | 93% | 88% | 86% | 83% | 81% | 78% |
Meanwhile, the implemented LID facilities reduced peak flows by an average of 94%, with a maximum of 96% in scenario S1 and a minimum of 93–94% for scenarios S2–S7. In this case, there was no significant increase in reduction relative to the duration of the rainfall. The above results exclude the real precipitation in scenario S0, where the reduction in both total runoff and peak flow was 100%.
The graph in Figure 7 presents the average total volume and peak flow reduction as a function of precipitation duration. Here, the effectiveness of LID facilities in reducing rainwater runoff at the proposed locations in the area under study decreased relative to the entire existing drainage system and depended on the average intensity and duration of precipitation. As indicated by high R2 values, the reduction followed the polynomial trend lines that had been determined.
Average reductions in the total volume and peak flow for scenarios S0–S7
Considering the volume of rainfall in the studied catchment in relation to the calculated retention volume at all LID facilities, the percentage of LID storage was in the range of 28–37% (32% on average) of the total synthetic rainfall volume, as shown in Figure 8.
Percentage distribution of the rainwater volume and storage volume of LID facility as a function of rainfall depth
Considering the rainfall volume of 365 m3 that fell on the surface of the study catchment on the day of the measurement, and the estimated LID retention volume of 41 m3, the percentage of rainfall volume retained by LID was 11%. It should be noted that SWMM separates initial and final storage in its summary report, with the difference between these volumes representing the potential rainwater retention volume. In this study, LID storage was absolutely dependent on the adopted LID parameters (Table 5), and the initial layer was approximately 187 m3.
The inflow values shown in Table 4 must be reduced by the amount of infiltration (average 0.033 m3/m2) and evaporation (average 0.0036 m3/m2), which depend on the local soil and atmospheric conditions, as well as the precipitation duration. The difference between rainfall depth and LID storage reflected the total volume of rainwater that either flowed through the LID (surface and drain outflow) or bypassed it. This figure was influenced by many factors, including as depression storage, LID structure, evaporation and infiltration. The maximum calculated LID retention volume was approximately 1100 m3 (for scenario S7).
Table 7 presents the results of quantitative calculations illustrating the performance of individual LID facilities. Key values, such as total inflow, surface outflow and drain outflow, are given in m3. These represent the total rainwater flowing into the LID, and the total rainwater flowing out through the drainage system or overflowing to the drainage system, respectively. The indicated LID areas represent the minimum and maximum facility sizes, while the sum indicates the overall area of installed LID.
Summary of the simulation results of the calibrated SWMM for individual LID facilities implemented in the study area for scenarios S0–S7
| Total Inflow (m3) | 37.9 | 658.2 | 987.1 | 1234.8 | 1304.7 | 1404.0 | 1512.8 | 1629.2 |
| Surface and Drain Outflow (m3) | 0.0 | 32.7 | 133.6 | 273.6 | 318.7 | 385.4 | 458.9 | 538.8 |
| Average inflow (L/m2) | 0.008 | 0.139 | 0.208 | 0.260 | 0.275 | 0.296 | 0.319 | 0.344 |
| Average outflow (L/m2) | 0.000 | 0.007 | 0.028 | 0.058 | 0.067 | 0.081 | 0.097 | 0.114 |
The average inflow and average outflow were calculated based on individual inflows and outflows per unit area (L/m2). The calculated average reductions in surface runoff indicated the high efficiency of the applied LID solutions. The efficiency of rainwater surface runoff reduction ranged from 67% to 95% for scenarios S1–S7, with an average of 78% with a standard deviation of 9%.
The total inflow in in Table 7 represents the volume of water flowing into the LID. The difference between the rainwater volume and the total inflow into the LID accounts for all types of losses, and takes into account the surface characteristics; pervious and impervious depression storage; the impervious fraction of subcatchments; and infiltration and evaporation along the surface runoff path.
The graphs presented below show the effectiveness of reducing the total inflow (L/m2) of rainwater using LID in the studied housing estate for all scenarios, considering all 32 LID units (Figure 9). These values were calculated by assuming the total inflow of rainwater from all surface types into a single LID, as well as the
Dependence of the total inflow per 1 m2 and reduction in rainwater for BC cells in the residential area under study for scenarios S0–S7
Figure 9 illustrates the unit inflow reductions for BC (red points) for which there were exponential trend lines with a statistically significant coefficient of determination (R2) for α ≤ 0.05, demonstrating the relationship between reduction and total inflow. The flattening of this line was clearly noticeable for shorter precipitation periods (S1, S2). For longer precipitation events, characterized by lower intensities, there was a systematic decrease in rainfall-runoff reduction.
These analyses also revealed the limit values at which BC reduced the entire inflow, clearly determined by the duration and intensity of precipitation. Figure 10 details these limit values, including the minimum, maximum and average fully reduced rainwater inflows per 1 m2 of the LID facility area. For precipitation in scenario S2, with an average intensity of 175.33 LPS/ha, the values of average fully reduced inflows at LID were revealed to be the lowest at 2.30 L/m2 (excluding real rainfall). In contrast, in scenario S1, BC were loaded with the largest tributary inflows, which were fully reduced.
Maximum, minimum and average values of the rainwater inflow limit [L/m2] per 1 m2 of surface area of LID facilities for scenarios S0–S7 for which there was a 100% quantitative reduction of inflow
The highest and most fully-reduced values of inflow per unit area of LID occurred for heavy rainfall in scenario S1, amounting to 29.72 L/m2 (with an average of 5.79 L/m2). During the first phase of long-term precipitation, the LID layers saturated slowly, thus increasing the volume of retained rainwater with slight simultaneous infiltration into the native soil. Over time, the water storage capacity decreased, but infiltration into the ground increased. When the retention volume of the LID was exceeded, excess water was diverted to the rainwater drainage system, which also reached its maximum hydraulic capacity, followed by surface outflow of rainwater.
This shows that LID were more effective in reducing the total inflow during short and heavy precipitation events. However, for scenario S7, a slow and systematic increase in the maximum and average total inflow was observed, reaching a value similar to scenario S1 (29.64 L/m2). This ability to accept more inflow water and fully reduce it during long-term rainfall can be explained by the increase in infiltration.
The results of the study are summarised in Table 8 below.
Summary of the results obtained for the considered precipitation scenarios S0–S7
| 1A | Rainwater volume [m3] | 365 | 2137 | 2742 | 3156 | 3296 | 3510 | 3725 | 3984 |
| 1B | Final LID storage [m3] | 41 | 788 | 970 | 1024 | 1036 | 1057 | 1081 | 1108 |
| 2A | Average inflow [L/m2] | 0.008 | 0.139 | 0.208 | 0.260 | 0.275 | 0.296 | 0.319 | 0.344 |
| 2B | Average outflow [L/m2] | 0.000 | 0.007 | 0.028 | 0.058 | 0.067 | 0.081 | 0.097 | 0.114 |
| 4A | Total volume [m3] | 17 | 539 | 752 | 916 | 953 | 1,008 | 1,072 | 1139 |
| 4B | Reduced volume [m3] | 0 | 29 | 56 | 111 | 134 | 167 | 204 | 246 |
| 5A | Peak flow [LPS] | 4.60 | 449.16 | 464.03 | 454.78 | 393.10 | 359.34 | 349.85 | 266.20 |
| 5B | Reduced peak flow [LPS] | 0.00 | 17.51 | 31.63 | 31.66 | 25.19 | 22.57 | 22.05 | 17.58 |
Pearson’s linear correlation coefficients were calculated for the results that were statistically significant. The strongest direct proportional correlation (0.98) was found between the results of no. 2 and no. 4, while a weaker relationship (0.80) was observed between no. 2 and no. 5. A significant correlation (0.83) was observed for no. 1 and no. 5B. The weakest relationships were found for no. 3, although there was a significant correlation (0.69) between no. 3 and no. 1. An inversely proportional relationship (−0.76) was found between no. 1 and no. 5.
Proper model calibration, based on the measured data from a single real precipitation event, allowed for the accurate reproduction of the conditions prevailing in the study area. The runoff-run-on process was modelled for a single actual precipitation event (S0), as well as for conditions prevailing in hypothetical scenarios (S1–S7) and for alternative rainwater management using GI. Due to the lack of sufficient verification data, single-event calibration and basic measures of model response fitness coefficients were used. This approach ensured an accurate reproduction of the peak value and shape of the hydrograph, while continuous calibration provided a more accurate estimate of runoff volume over time [Rezaei et al. 2019].
The simulations carried out as part of this study proved that the implementation of GI results in a significant reduction in both peak flow and runoff volume of rainwater. According to [Golden et al. 2018], LID can help achieve many hydrological objectives, such as reducing peak flow and restoring natural surface flow and infiltration to the original conditions (i.e., those existing before an area underwent human development).
In this paper, it has been demonstrated that for the outfall, W1, of the entire drainage system, an average total volume reduction of 86% and a peak flow reduction of 94% can be achieved with GI covering approximately 10% of the catchment drainage area. The inclusion of this 10% GI area in the structure of the housing estate did not change its functionality or spatial arrangement. The analyses showed that the efficiency of reducing rainwater surface runoff in BC ranged between 67% and 95%. Most importantly, the efficiency of BC is highest for intense and short (15 min) rainfall events, reaching a 95% reduction in runoff volume. Similarly high reductions, exceeding 97%, were reported in a seven-year study of a rain garden in Xi’an, China [Guo et al. 2019], and in a one-year observation of a rain garden in North Carolina, USA, where 78% of the water either exfiltrated the cell or left via evapotranspiration [Hunt et al. 2006]. Other studies have also reported high volume reductions, ranging from 43 to 94% [Mai et al. 2019], from 37 to 61% [Jiang et al. 2015], and reaching 50% [Christensen et al. 2012]. However, these studies mainly involved field sampling and laboratory analysis.
The saturation of rainwater storage layers leads to a decreasing reduction in total volume as rainfall duration increases, reaching a minimum value of 67% for a three-hour rainfall. Taking into account rainfall intensity, the runoff reduction efficiency from the simulations decreases as total rainfall increases. The study also indicated that GI is more effective in reducing total inflow and peak flow for high-intensity heavy rainfall events that are shorter than 30 minutes.
It should be noted that the layered structure of the applied LID, which determined the course of the infiltration, retention, and evaporation processes, was assumed and constant in the simulations. The materials and plants used in the LID can also influence the rate and effectiveness of rainwater inflow reduction. The results obtained were a consequence of the adopted structural parameters of individual LID layers. These parameters determined the LID final storage capacity, which was approximately 1100 m3 and accounted for an average of 28% of the total rainfall volume. For such values, a 67% reduction in total inflows to the BC was achieved. The value of the reduction increases with the increasing LID storage in the rainfall volume, reaching a 95% reduction for 37%.
As a result of the analyses, the relationship curves of the total inflow per 1 m2 and the outflow reduction were developed, which could help in the selection of BC facilities and in the estimative calculation of the amount of rainwater that does not enter the drainage system. Detailed analyses of individual LID facilities revealed a strong relationship (R2 = 0.5983–0.8185) between the total inflow per 1 m2 and the reduction in rainwater, with the highest value occurring for 15 minutes of rainfall. The relationship weakened as the duration of rainfall increased. The minimal fully quantitative reduction of the average inflow to LID of 2.30 L/m2 was noted for 30 min of precipitation (S2), whereas the maximum reduction of 5.79 L/m2 was observed for 15 min of precipitation (S1).
This could be explained by the analysis of the percentage distribution between rainfall volume and retention volume. Under the assumed parameters, individual LID sites reached the maximum storage layer charge after approximately 30 min of rainfall. Therefore, once this time threshold was exceeded, rapid surface and drain runoff occurred, resulting in a more than a two-fold decrease in inflow reduction.
Detailed databases, high-precision and accurate subcatchment delineation, and calibration using a single flow measurement at outflow W1 allowed for the matching of the hydrographs of measured and simulated flows. Obviously, the inclusion of more measurement data from additional control sites would reduce this model’s uncertainty. The construction of LID in an existing development is a complex task due to numerous constraints. This type of solution should be developed at the concept, design and construction stages. These efforts will contribute to more effective protection of rainwater recipients on both qualitative [Smith et al. 2023] and quantitative [Guo et al. 2013] scales. It has been pointed out that methods for measuring and optimising the performance of integrated multifunctional green infrastructure have not been fully developed [Wang et al. 2020]. This study has attempted to estimate the hydrological performance of the LID at one selected site.
The complex interdependency of the individual LID performance parameters in the system under study highlights opportunities for future studies that could introduce additional factors, such as an altered LID structure. The hydrological performance of the LID discussed here, and rainwater reduction values of the entire system, should also be verified using other precipitation phenomena occurring in the study area and the introduction of new calibration points. In addition, future modifications could include scenarios involving varying LID locations. Further research is necessary due to the lack of detailed technical loading parameters for the relationship between LID and precipitation.
The parameters and indicators discussed above highlight the dependencies that are crucial for the effective operation of the BC equipment. Their interpretation enables a comprehensive assessment of the functionality of the GI solutions implemented in the area under study. It is evident that both the average inflow to the LID, and the total volume discharged at the outfall, significantly depend on rainwater volume. The outfall of total volume reduction at the discharge point of the sewer system is most significantly determined by the efficiency of the LID operation, specifically the average outflow from the LID. However, reduction in peak flow at the outfall is also most closely linked to the percentage of LID storage in rainfall volume. Thus, the structure of individual BC unit layers determines the reduction of peak flow, while the amount of inflow, and its reduction in individual LID units, define the rainwater volume reduction at the outfall. There is also a relationship between the reduction in total inflows in BC devices and the LID volume, but it is not sufficiently strong.
The aforementioned interdependencies determine the effectiveness of total volume and peak flow reduction at the drainage outlet. At the outset of the study, it was believed that the ultimate efficiency of using distributed GI facilities, including the rain gardens under analysis that were equipped with drainage systems, should be the overriding criterion for their selection, considering the structure of individual layers and their location in the area to be drained. Obviously, while the location (which was not analysed in this study) in areas of existing development is very limited, the structure and interconnection of these facilities perform a very important function in the overall system. The analyses indicated that the reductions in both total volume (78–95%) and peak flow (93–96%) at the system outlet were greater than the inflow-outflow reductions for the individual LID devices (67–95%). This can be explained by the fact that the devices are connected within a cohesive system, where the outflow from an overflowing LID is often intercepted by the next device downstream. This relationship is particularly evident for long-duration rainfall events.
Correct conclusions based on simple relationships between individual phenomena can have a significant impact on the correct management of rainwater in urbanised areas. Such summaries and databases could support the development of guidelines for engineers and urban planners, while also contributing to environmental benefits both locally and on a broader scale.
Verified hydrological performance parameters need to be incorporated into planning, design and strategy documents [Muszyński et al. 2022]. In the context of a changing climate, these documents must strive to achieve the targets set out in the provisions of the European Green Deal [Tache et al. 2023] and the Sustainable Development Goals (SDGs). Therefore, it is considered that a portion of the currently required minimum of 40–60% of biologically active area should be allocated to the development of GI [Resolution… 2018]. In addition, it is proposed that GI should be designed with the goal of minimizing both the total volume and peak flow of rainwater.