The results of this research were generated from the application of the proposed methodology to a case study consisting of a real urban catchment located in Espoo, southern Finland. The study catchment experienced rapid development and transitioned from being a coniferous forest in 2001 to a residential area in 2006 [55
]. Figure 4
shows the location of the catchment and the spatial distribution of its sewer network, which was provided by the Helsinki Region Environmental Services Authority HSY.
also depicts the aerial photography of the study catchment in 2007 [56
], when it was at full development. A Digital Terrain Model (DTM) with a cell size of 2 m was acquired from the National Land Survey of Finland [57
], whilst the Geological Survey of Finland revealed that the study catchment laid on a layer of sandy till with bedrock below it [23
], which most likely corresponds to a HSG of B. A set of 60 points with values of Water Table Depth near the catchment area was extracted from the global grid created by Fan et al. [58
] at 1,603,781 sites worldwide from government archives and published literature.
The study catchment was delineated and simulated in SWMM by Jato-Espino et al. [59
], who determined that it was formed of 79 subcatchments covering 10.535 ha and optimized its stormwater modelling using Design of Experiments (DOE) with three monitored calibration (CAL 1, CAL 2 and CAL 3) and validation (VAL 1, VAL 2 and VAL 3) rainfall events (see Table 3
). The calibration of the simulations revealed that six parameters had a statistically significant impact on the discharge of the study catchment. Below are specified the calibrated values for each of them, whose combination maximized the fit between observed and predicted outflow: percentage of imperviousness (80% of the initial value obtained using GIS tools), width (80.4% of the initial value obtained using GIS tools), slope (115.15% of the initial value obtained using GIS tools), Manning’s roughness for impervious area (0.0135), Depth of depression storage on impervious areas (0.379 mm) and Manning’s roughness for conduits (0.015). The simulation of the validation events with these calibrated parameters reproduced the real hydrographs monitored at the outlet of the catchment with high accuracy, as demonstrated by the goodness-of-fit measures used to test them (see Table 3
): Root-Sum Squared Error (RSSE), coefficient of determination (R2
) and Nash–Sutcliffe model efficiency coefficient (E).
Once the accuracy of the model was validated, the study catchment was re-simulated with the calibrated parameters by these same authors using synthetic storms designed for different return periods and climate scenarios [59
], in order to assess its response to storm events caused by Climate Change. Hence, in addition to the stationary scenario in which precipitation was assumed to remain constant over time, two different greenhouse gas concentration trajectories were considered: RCP4.5 and RCP8.5 [60
]. Table 4
lists the values of Annual Maximum Daily Precipitation (AMDP) associated with each combination of return period and climate scenario. These values of AMDP were used to design synthetic storms through the combination of Intensity–Duration–Frequency (IDF) curves and the Alternating Block Method. Their duration was 106 min in all cases, according to the lag time of the total catchment. These data were used as the basis for carrying out the prioritization of flood-sensitive areas in the study catchment.
3.1. Search for Feasible Locations for the Implementation of Sustainable Drainage Systems (SuDS)
The search for feasible locations for the implementation of SuDS in the study catchment started with the preparation of maps related to the geometric and hydrologic criteria to be met by each system according to Table 1
: HSG, slope, building buffer, road buffer, stream buffer, Water Table Depth and area. Since the hydrological condition of the soil below the study catchment corresponded to an HSG of B, there was no restriction in these terms for any type of SuDS, which required at most a type B soil.
Slope in the catchment area was determined from the DTM and classified according to the three thresholds defined in Table 1
: 4%, 5% and 15% (see Figure 5
a). The area corresponding to these thresholds covered 14.40%, 20.78% and 55.49% of the whole study catchment, which provided multiple opportunities to install different types of SuDS. Since the location of the downspouts in the buildings was unknown and there was no stream close enough (<30 m) to the study catchment to be considered, the buffer-related calculations were limited to roads (see Figure 5
b), whose presence only restricted the implementation of bio-retention cells (see Table 1
). In this case, the ratio of road buffer areas to the whole catchment area was 70.91%.
An exploratory analysis of the dataset with values of Water Table Depth was carried out in the first place to model this criterion. Normality of this dataset was ensured by the p
-value reached according to the Shapiro–Wilk test (0.092), whilst the shape of the semivariogram cloud suggested that there was no spatial autocorrelation between measurements, since the average squared difference of values for all pairs of points increased as the distance between the pairs of points increased. The best combination of coefficient of determination and R2
and Root-Mean Squared Error (RMSE) was provided by Ordinary Kriging, which yielded values of 0.797 and 3.482, respectively. The interpolation surface obtained shown in Figure 5
c using this method demonstrated that Water Table Depth was not an issue, since the groundwater level was at least 3.50 m below the ground (see Table 1
). The last criterion to check was the maximum area to be covered by SuDS. The combination of feasible zones according to the four other criteria (HSG, buffers, slope and Water Table Depth) demonstrated that no limit in terms of surface area was exceeded.
The intersection of areas in which these criteria were met separately resulted in Figure 6
a. Bio-retention cells, which shared locations with either vegetative swales or infiltration trenches, were not included in the map because their area was smaller than that corresponding to any other option. The two remaining types of SuDS listed in Table 1
, rain barrels and rooftop disconnection, were not considered either because the location of the downspouts was unknown. This is a refined map, excluding marginal and disconnected feasible areas whose consideration was irrelevant in practical terms. As for overlap, the area associated with vegetative swales in Figure 6
a was also valid for infiltration trenches. PPS were the type of SuDS that involved larger feasible area, covering 16.62% of the study catchment, followed by green roofs and infiltration trenches with 9.82% and 1.44%, respectively. This fact, which was consistent with the wide applicability of PPS introduced in previous sections, supported the focus of the methodology on this specific system from now on.
3.3. Hydrological Simulation of Permeable Pavement Systems (PPS)
The information contained in Figure 6
was used to model the influence of PPS on the hydrological response of the study catchment. Firstly, the catchment was simulated without PPS, in order to identify which areas were more sensitive to flooding under the return periods and climate scenarios listed in Table 4
. Simulation duration was set at 150 min, since this value proved to be enough for runoff to cease. The time step for reporting was 2 min, in order to match the frequency with which flow rate was originally monitored in the study catchment [59
], whereas routing was set at 3 s to minimize surface runoff and flow routing continuity errors. The results yielded by these simulations were then used for locating PPS in strategic areas to avoid node flooding and conduit surcharge along the sewer network. Figure 7
is a representation of the minimum area of PPS required to avoid flooding under different combinations of climate scenario and return period. The location of PPS was limited to parking areas exclusively, in order to reproduce the most feasible and easiest to integrate solutions in practical terms. Therefore, no isolated and/or difficult to connect pavement reach was considered in subsequent calculations.
The simulations proved that the storms associated with a return period of two years were enough to produce floods in the study catchment for the RCP scenarios. In contrast, a value of 10 years was required to cause the same impact under the stationarity assumption, which highlights the increase in drainage capability required by Climate Change. The inclusion of different PPS configurations was found to avoid any flooding problem up to the following return periods: 2, 10 and 50 years for RCP8.5, RCP4.5 and stationary scenarios, respectively. These figures demonstrate the capability of this type of SuDS to mitigate the effects of heavy rainfall events beyond the common magnitudes used to design urban drainage systems (stationary scenarios with return periods of 2, 5 or at most 10 years). In quantitative terms, the presence of PPS involved an average volume reduction at the outlet of the study catchment of 40%–50%.
The stormwater simulation of the catchment configurations illustrated in Figure 7
resulted in the hydrographs represented in Figure 8
. Each plot includes the four following hydrographs: without PPS, with Porous Asphalt (PA), with Porous Concrete (PC) and with Permeable Interlocking Concrete Pavement (PICP). Figure 8
a,c,e was obtained from the simulation of minimum PPS areas required to avoid flooding, whereas Figure 8
b,d,f corresponds to the location of PPS in all available parking areas (see Figure 7
d). The comparison of minimum (Min) and maximum (All) PPS areas was introduced to demonstrate the capability of these systems to not only reduce runoff volumes, but also delay peak flows. Figure 8
g does not include this distinction, because the minimum and maximum coincided for the mitigation of the 50-year storm in the stationary scenario. As for the combination of RCP4.5 and a return period of 10 years, the minimum PPS area used to avoid flooding was enough to produce a delay in peak flow (see Figure 8
e), because only the parking area in the south of the catchment was omitted in comparison with the scheme depicted in Figure 7
These hydrographs were analysed using statistical techniques to verify that the hydrological impact of PPS was significant. The p
-value obtained for the four hydrographs (without PPS, with PA, with PC and with PICP) for the three return periods (2, 10 and 50 years) and climate scenarios (stationary, RCP4.5 and RCP8.5) using the Shapiro–Wilk test was 0.000 for all these combinations, which suggested that the samples under analysis were not normally distributed and had to be evaluated through non-parametric tests. Hence, the Kruskal–Wallis test was applied to confirm the absence of differences in the hydrographs associated with the PPS types. The p
-values, which were 1.000 in all cases, enabled the acceptance of this hypothesis. These results showed that any of the three different types of PPS could be used to compare the results obtained with and without them installed, which demonstrated that the hydrological impact of these systems at a catchment scale was extremely similar. Under these circumstances, the Mann–Whitney U test proved that the differences in hydrological response of the catchment with and without PPS (PA, PC or PICP) were statistically significant in all cases (p
-values < 0.05), except for the situation represented in Figure 7
a, which involved the inclusion of a PPS area of only 0.031 ha in the east of the study catchment. In overall terms, these results proved that PPS can make a significant difference to the amount of excess stormwater generated in an urban catchment due to intense rainfall events.