3. Results and discussion

3.1 Single Event Simulation

Hypothetical extensive green roofs (as mentioned in Section 2) are used for Bi (i = 1, 2, …, 5). When choosing different Green roofs or Bioretention cells as LID type in SWMM, the simulation shows slight differences in the amount of lid drainage and final storage obtained by but the surface runoff is consistent (Tables S1 and S2). We used the predicted runoff from 2-yr, 10-yr, and 100-yr precipitation events to evaluate the potential for surface runoff reduction.
3.1.1 Sensitivity Analysis
The results of sensitivity analysis show that the saturated hydraulic conductivity is the most sensitive parameter for evaporation and surface runoff; wilting point is the most sensitive parameter affecting the initial LID storage; and soil field capacity is the most sensitive parameter affecting the final storage capacity (Table 4). Based on the sensitivities of the 5 parameters and main objective of this study to mitigate urban flooding by GRS, a combination that minimizes surface runoff was selected (Table 5). The following analyses use these substrate properties, since these parameters produced the best overall result from the sensitivity test.
3.1.2 Runoff Retention of GRS
Fig. (5) shows the comparison of simulated runoff at the condition of the TRS and GRS. The key features are summarized in Table 6. Results indicated that the GRS significantly performs better than the TRS in reducing surface runoff. The GRS has a most significant influence on reducing urban runoff for the 2-year precipitation event. The reduction in total runoff volume by the GRS decreases as precipitation intensity increases: 42%, 34%, and 27% reductions for 2-yr, 10-yr, and 100-yr precipitation events, respectively. Meanwhile, we find that runoff coefficient (Q/P) of the GRS increases from 57.41% to 72.19% when the precipitation intensity changes from 2-yr storm to 100-yr storm (Table 6), which indicated that the retention capacity of the GRS decreases with the increase of the precipitation intensity.
It is of great importance to effectively reduce the peak surface runoff since it often accompanied by maximum erosions and local flooding (Li and Babcock, 2014). Our simulation results show that the GRS could give a much higher reduction for the peak flow than the TRS in the 2-yr precipitation event. For larger precipitation events (i.e., 10-year and 100-year precipitation events), however, this effect is reduced significantly (Table 6). Additionally, we found that the GRS has little ability to postpone runoff peak timing (less than 1 minute) in all the assumed precipitation events (Fig. 5).
Because ET is negligible in the single event simulation, the volume of retained stormwater is the difference between precipitation and discharged water. For the GRS, the ratios of retained to precipitation volumes under 2-yr, 10-yr, and 100-yr precipitation events are 75%, 55%, and 41%, respectively. This result is consistent with some previous studies (Carpenter and Kaluvakolanu, 2011; Carter and Rasmussen, 2006; Fassman-Beck et al., 2013; Getter et al., 2007; VanWoert et al., 2005) and indicates that there are limitations for runoff reduction by green roofs in heavy precipitation events.
3.1.3 Mitigation Conditions of Flooding Nodes and Overloaded Conduits of CSS / SWS in GRS
The ability of the GRS to mitigate urban flooding can be showed from the amount of floods discharge generated after storm events of different intensities (Fig. 6). In the 2-yr precipitation event, neither the GRS nor the TRS generate flooding (thus results were not shown in Fig. 6). In the 10- and 100-yr precipitation events, however, the flooding volumes of the GRS are 82% and 28% less than those of the TRS, respectively. In addition, the peak flooding flow in the case of the GRS is significantly less than that of the TRS (Fig. 6) and the peak flooding flow is 72%, and 19% lower than those of the TRS in the 10-yr and 100-yr precipitation events, respectively. Moreover, the drainage condition of CSS / SWS in GRS is much better than that of TRS. According to the simulations from the SWMM, in the 10-yr precipitation event, there are one flooding node (J4) and one overload conduit (C4) in the CSS / SWS of TRS (Figs. 7a and c). As shown in the Fig.7c, the GRS reduces the capacity of conduit (C4) to less than 1 (1 means overloaded) and the duration and volume of flooding node (J4) in the GRS are much less than that of the TRS. In the 100-yr precipitation event, there are three flooding nodes (J2, J3, J4) and two overloaded conduits (C3, C4) in the TRS (Figs. 7b and d). The GRS, even in such a high intensity, still reduces the flood risk well with only one overloaded pipeline (C4) and three flooding nodes. Although there are still three flooding nodes (J2, J3, J4) in the GRS case in the 100-year events, they are significantly mitigated.

3.2 Continuous simulations

In the continuous simulations (31-year), the change in surface runoff is consistent with that of rainfall with June as the peak month of surface runoff (Fig. 8). There is a relatively high average evaporation between April and September in the GRS case. In August and September, the rainfall is relatively low (Fig. 9). In addition to reducing runoff, ET can reduce urban temperatures, especially during warm months (Tabares-Velasco, et al., 2012). The surface runoff in the GRS case accounts for 47% of annual precipitation, while that in the TRS case accounts for 90% (Table 7).
Green roofs can be beneficial to urban environment by reducing surface runoff through ET (Li and Babcock, 2014). The pore space in the soil layer of green roofs, which provides space for retention, can hold water by capillary forces until water is lost via ET. Therefore, ET amount can be quantified by analyzing the moisture loss in green roofs (Li and Babcock, 2014). The evaporation amounts in the GRS and TRS cases are very close (Fig. 10) and account for 11.6% and 10.2% of annual precipitation, respectively. The reason could be that the evaporation values are calculated based on water balance in the SWMM, without considering the actual dynamic process. The output of PET (Potential Evapotranspiration) in the SWMM, which account for 25.2% of the annual precipitation, is calculated by Hargreaves method (Fig. 10). The RET calculated based on the Penman-Monteith method and Pan Evaporation method, account for 38.6%, 37.7% total precipitation respectively. The AET calculated by the Advection-Aridity method and Granger-Gray method, however, account for 44.9% and 36.7% of total precipitation, respectively (Fig. 10). The ET values calculated based on the Penman-Monteith method, Pan Evaporation method, and Granger-Gray method are very close. They are a little smaller than the ET obtained from the Advection-Aridity method and much higher than the PET result of the Hargreaves method. Therefore, the ratio of annual ET value in the GRS to total precipitation should range from 36.7% to 38.6%, which represents the rainwater retention potential of green roofs in MLRYR.
Besides the rainwater retention potential of the GRS, the evaporation curves in Fig. 10 also illustrate the difference in evaporation patterns between the GRS and TRS. Although the ratio of evaporation in the GRS to that in the TRS is just 1.14:1, the timings of evaporation from GRS and TRS are quite different. In the TRS, evaporation and precipitation are concurrent. The evaporation curve is intermittent consistent with rain event timing. In the GRS, however, the rain water could be retained and evaporated between precipitation events. This result has a flaw because the evaporation result of SWMM is based on the principle of water balance. Since the TRS cannot retain water, precipitation and evaporation occur simultaneously in the time series outputs of SWMM to achieve a balance. However, in real-world the relative humidity should be very high and limit evaporation during precipitation periods. Although there exists a drawback of the output data, the general trend of the continuous evaporation curve of the GRS implies more continuous, and likely larger, cumulative heat dissipation from the building. It thereby has potential to mitigate the urban heat island (UHI) effect. Similar results were also shown in previous researches that have evaluated green roofs’ thermal benefits and suggested that GRS can reduce the UHI effect through changing roof albedo, decreasing heat transfer into buildings, decreasing long-wave radiation from leaves, evapotranspiration, and heat storage by plants (Kwok, Wong, and Lau 2013; Tabares-Velasco, et al., 2012; Coutts, et al., 2013). Thus, thermal benefits could be another attractive function of green roofs if applied at the urban scale.

3.3 Performance of the IGRS

We also evaluated whether the IGRS, a combined system of green roof and rooftop disconnection, is more effective at mitigating urban flooding and groundwater depletion. In the single event simulation, flooding only occurs in the cases of 10-yr and 100-yr precipitation events. The flooding volumes in the cases of GRS and IGRS are very close (Figs. 11a - 11f, Table 8). In the 10-yr precipitation event, with the increase of greenbelt from 1m to 3m, the flood volume in the IGRS case is 2.09% - 2.77% less than that in the GRS. As in the 100-yr precipitation event, with the increase of greenbelt, the flood volume in the IGRS is 0.8% - 1.0% less than that in the GRS. When compared with the TRS, the flooding volume in the GRS and IGRS are 67.69% - 71.10% and 30.13% - 33.22% (depending on green belt width) less than those of the TRS for the 10- and 100-yr precipitation events, respectively. In terms of the drainage condition of CSS / SWS, the flooding nodes (Figs. 12a - 12f) and overloaded conduits (Figs. 13a - 13f) in the IGRS are also consistent with the conditions in the GRS. As shown in the Figs. 12a-12f and Figs. 13a-13f, both the IGRS and the GRS can mitigate the flooding nodes and reduce the overloaded time of conduits when compared with the TRS, but the capacity of flood detention is slightly affected by the width of greenbelt. Results of runoff volume indicated that the differences in runoff volume and runoff coefficient (Q/P) between GRS and IGRS (Table 9) are insignificant. However, the simulated peak flows in the GRS case are 16.18% - 16.95%, 25.07% - 25.90% and 28.83% - 29.46% (depending on green belt width) lower than those in the IGRS case for the 2-yr, 10-yr, and 100-yr precipitation events, respectively (Table 9).
In the continuous simulation, evaporation, surface runoff, and infiltration are three key parameters to evaluate the effect of the systems on retaining stormwater. In the same precipitation event, evaporation in the IGRS is 0.4% more than that in the GRS (Table 10). Although the surface runoff in the case of GRS is 20-27% less than that in the IGRS case, the overflowing water from GRS is drained into the CSS / SWS, which increases the burden on the CSS / SWS and does not recharge groundwater. In the IGRS case, however, the overflowing water was diverted from green roofs to infiltrate in greenbelts and recharge groundwater, which could thereby reduce the burden on the CSS / SWS. As a result, infiltration is the most important measure of the IGRS effectiveness. The proportions of infiltrated water from annual precipitation are shown in 3 categories based on the belt widths (i.e., 1 m, 2 m, and 3 m; See Fig. S2). For 1 m, 2 m, and 3 m greenbelts, the infiltration of IGRS accounted for 10%, 16%, and 19% of annual precipitation respectively, while that of GRS was only 0.36%, 0.73%, and 1.09%. With the 3 different green belt widths, IGRS would recharge to groundwater with total runoff volumes of 2126, 3273, and 4033 m3 ha-1 yr-1 more than with GRS.
The XIFS is a representative of the whole city because of the homogeneous climate condition (precipitation, solar radiation, wind, etc.) and the consistent planning standard for building density and floor area ratio according to the Urban Planning of Nanchang City. Therefore, we can project the simulated results from this study site to most areas of the city if not all. Assuming that the IGRS is widely used in the city, the total estimated runoff that can recharge groundwater would be beneficial to mitigate urban drought and groundwater depletion problems in the city. According to the comparison of infiltration ratio between GRS and IGRS (Fig. S3), although not as good as GRS in controlling surface runoff, the IGRS can substantially enhance groundwater recharge. In the whole city, the GRS and IGRS can be used in different areas according to their own characteristics. Particularly, the GRS can be adopted in the areas where CSS / SWS capacity is insufficient to cope with large peak runoff because the peak runoff of GRS is 16-29% less than that of IGRS and can be used as an effective way to reduce surface runoff. In places where the capacity of CSS / SWS is large enough to deal with surface runoff, the IGRS can be used as an innovative way to make full use of the rich rainwater resources in Nanchang to recharge groundwater since urban flooding is not a threat in these places. The cities in the MLRYR are generally densely built areas that have a large potential to be retrofitted with GRS and IGRS.