Can the implementation of Low Impact Development reduce basin runoff?

Low impact development (LID) was promoted as an alternative to conventional urban drainage 10 methods. The effects of LID at site or urban scales have been widely evaluated. This project aims to investigate the impact of LID implementation on basin runoff at regional scale in a half urbanized catchment; especially the overlap of urban and rural sub-flows at peak times is concerned. A SUPERFLEX conceptual model framework was adapted as a semi-distributed model to simulate the rainfall-runoff relationship in the catchment for San Antonio, Texas as a case study. Scenario analyses of both urban development and LID implementation were 15 conducted. Results show that (1) the infill urban development strategy benefits more from runoff control than the sprawl urban development strategy; (2) in non-flood season permeable pavements, bioretention cells, and vegetated swales decrease peak runoff forcefully and permeable pavements, bioretention cells, and green roofs are good at runoff volume retention; (3) contrary to the general opinion about the peak reduction effect of LID, for partly urbanized, partly rural basins and extremely wet conditions, the implementation of LID practices 20 delays urban peak runoff and may cause stacking of rural and urban sub-flows, leading to larger basin peaks.


Introduction
Urbanization brought numerous environmental and hydrological changes to river basins and led to severe disturbance to the natural water processes. Unwanted vegetation is removed for urban development, diminishing the vegetation interception and transpiration. Large areas of pervious native soil are replaced by impervious 25 concrete and asphalt for human convenience, impeding runoff infiltration and subsurface water retention.
Without sufficient and continuous groundwater recharge, more environmental issues occur, such as land subsidence, groundwater shortage, and water quality degradation (Ahiablame and Shakya, 2016). These human activities modify catchments from a relatively robust natural condition to a sensitive and unstable urbanized status, resulting in water scarcity in dry seasons and waterlogging or urban flooding in rain seasons (Gilroy and 30 Mccuen, 2009;Ahiablame et al., 2012).
To solve flooding problems, the Conventional Drainage approach (CD) is widely exploited in urban areas, which adopts rapid and centralized water transfer strategy: Drainage systems are built to rapidly collect and convey the storm and wastewater from urban impervious areas to centralized municipal facilities, nearby water bodies, or downstream rural areas. The CD approach does not solve water problems such as peak flows and 35 water quality issues, which only shifts the problems to another place to some extent. Low Impact Development (LID) is promoted as an alternative to CD, seeking environmentally friendly solutions for current urban water https://doi.org/10.5194/hess-2021-32 Preprint. Discussion started: 29 January 2021 c Author(s) 2021. CC BY 4.0 License.
To investigate the problem of peak stacking of the rural and urban peaks, case study is conducted. The SUPERFLEX conceptual model framework is used to develop a simulation model of the rainfall-runoff relationship of this partly urbanized catchment. Further urbanization of this catchment is foreseen, and to deal with the uncertainty of future urban development, scenario analysis of both LID implementation and 80 urbanization are used to give a reliable answer to the research question. The specific objectives were to (1) investigate the different rainfall-runoff relationships of urban and rural sub-areas; (2) examine the influence of urbanization on the basin runoff; and (3) assess the influence of LID implementation on the basin runoff, especially on the overlap of urban and rural sub-flows at peak times.
The structure of this paper is organized as follows: The study area and data are introduced in Section 2. The 85 methodology about the model setup and scenario design is illustrated in Section 3. The main research results regarding the effects of urbanization and LID implementation on catchment scale are shown in Section 4. Some discussions concerning the limitation of this research and recommendations for future urban development are mentioned in Section 5. And finally, the conclusions are drawn in Section 6.

Study area
The research catchment is a sub-basin of San Antonio River with 4544 km 2 . The City of San Antonio takes 27 % of the research catchment (1209.5 km 2 ) and is located near the basin outlet as shown in Fig. 1. Several rivers and creeks, including San Antonio River, flow through San Antonio downtown and then join with Medina

River. 95
San Antonio has a transitional humid subtropical climate featuring hot and humid summers and mild to cool winters. The average annual precipitation is 737 mm. The soil in San Antonio City mainly belongs to moderately permeable clayey soils. Edwards Aquifer is the most prolific groundwater aquifer in the study area, which provides the water for people in San Antonio. To release the stress of the Edwards Aquifer, a groundwater recharge project is developed in the north part of San Antonio by holding back storm runoff in 100 recharge zones. Except for this official groundwater recharge project, managed stormwater infiltration is not allowed in other places, to avoid groundwater pollution (The Edwards Aquifer Website, 2020). Natural stormwater infiltration will inevitably take place in rural and unpaved areas and in LID practices like permeable pavements and bioretention cells. This water will recharge the local groundwater and drains slowly to the river system. 105 San Antonio has separate foul sewer and stormwater systems. The precipitation collected by the stormwater pipeline system in urban areas is discharged directly to nearby water bodies without treatment. For the wastewater, three major wastewater treatment centers provide water treatment to people in San Antonio and neighboring cities, and the treated water is discharged to nearby rivers (San Antonio Water System, 2020). https://doi.org/10.5194/hess-2021-32 Preprint. Discussion started: 29 January 2021 c Author(s) 2021. CC BY 4.0 License. As for the social condition, San Antonio city is the seventh most populous city in the U.S. with more than 1.5 million residents (Ready and Montoya, 2019). It is also the fastest-growing of the top ten largest cities in the United States (The City of San Antonio -Official City Website, 2020). From 2010 to 2017, San Antonio 115 experienced a population growth rate between 1.5 % and 2.0 % and the city still keeps a stable demographic expansion. With this stable population increase, the urban land use of San Antonio is expected to grow at a more or less equal pace.

Hydrological data
Retrieved from the USGS website (https://www.usgs.gov/), the precipitation, evaporation, and runoff data from 120 study catchment were collected for hydrological modeling of 600 research days, from 2017-04-12 00:00 to 2018-12-02 23:30. The first 365 days are the calibration period and the last 235 for verification. The time scale is 30 minutes to reflect the fast water response character of urban areas.
Precipitation data from 10 monitoring stations are available, and Thiessen polygons method is used to calculate the total precipitation in this catchment. Evaporation data come from a meteorological station in the research 125 area. The discharge data from three catchments (study catchment and two sub-catchments) are collected from three streamflow monitoring stations as shown in Fig. 1. Among 600 research days, the precipitation and evaporation amounts are 1335 and 1054 mm, and the discharge is 166 mm.

Methodology
To address the research problem, three urbanization and five LID implementation scenarios are designed to deal 130 with the prediction uncertainty and alternative LID practices. SUPERFLEX conceptual model framework is adapted to a semi-distributed model with urban and rural surfaces to simulate the rainfall-runoff relationships of https://doi.org/10.5194/hess-2021-32 Preprint. Discussion started: 29 January 2021 c Author(s) 2021. CC BY 4.0 License. the current situation and the eight scenarios. The scenario design is introduced in this section, following the model setup.

Urbanization scenarios
According to the projected urbanization information provided by "City of San Antonio: Comprehensive Plan" (2010), there will be 1.1 million new residents in San Antonio by 2040. Since in this research current time is defined as 2017, expected population growth between 2017 and 2040 is estimated as 0.9 million new residents.
The government of San Antonio planned to terminate the unconstrained sprawl of the city but adopt the infill 140 strategy and retrofit existing urban and suburban areas to attract more investment in the urban core and save the high cost of infrastructure and utility services.
Based on the information above, three urban development scenarios for 2040 were designed as shown in Table   1. Scenario A offers an extreme infill urban development situation, in which city size in 2040 will be the same as it is now. Scenario B presents a partial-infill, partial-sprawl urban development situation. In this scenario, 70 145 % of the new residents will live in current urban areas, while 30 % of new residents will be living in new suburban areas. In scenario C, 50 % of the new residents are assumed to stay in urban expansion areas, while the other 50 % will infill current vacant and underutilized urban areas. Since the infill development strategy may lead to compact living space, per capita living space for scenario A is assumed to be 0.85 times the current areas, and this ratio is 0.9 for scenario B, while no compact living space is assumed for scenario C. 150

LID implementation scenarios
According to local regulations, the implementation of LID is not strictly mandatory for every development or redevelopment project, and there is great flexibility in the selection of LID practices. Therefore, this research 155 will adopt four most common and typical LID practices to design five LID implementation scenarios based on the conventional urban development scenario C. The first four scenarios assume moderate LID implementations, as 15 % of the precipitation on urban impervious (grey) surfaces will be collected by a single type of LID practice -bioretention cells, vegetated swales, extensive green roofs, and permeable pavement. This will allow us to compare the different hydrological performances of these LID practices. 160 The last scenario assumes a wide scale LID implementation, as 50 % of the precipitation on urban impervious (grey) surfaces will be conveyed by mixed LID practices (bioretention cells, 15 %; vegetated swales, 15 %; extensive green roof, 5 %; permeable pavement, 15 %), to provide an optimistic and flexible LID implementation plan. Green roofs and permeable pavements serve the area where they are constructed; the ratio https://doi.org/10.5194/hess-2021-32 Preprint. Discussion started: 29 January 2021 c Author(s) 2021. CC BY 4.0 License. of drainage and construction areas of bioretention cells and vegetated swales are 1.5 and 3, respectively, since 165 both collect and retain stormwater from a larger contributing area. Besides, the cascading connections among these LID practices were designed based on realistic construction considerations as shown in Fig. 2.

Hydrological model 170
To avoid the drawbacks of too complex models including high-data requirement, equifinality, and model uncertainty, and to distinguish rural and urban areas in the same catchment, a conceptual flexible SUPERFLEX (Fenicia et al., 2011) hydrological modeling framework is adapted in a semi-distributed model with the urban or rural surfaces. While SUPERFLEX has been used to simulate the rainfall-runoff relationships in different natural landscapes, it has not yet been employed in a highly urbanized catchment. That is why, several urban 175 water processes, as well as four LID modules (bioretention cells, green roofs, bioswales, and permeable pavements) had to be formulated and added to the SUPERFLEX framework.

Semi-distributed model setup
The hydrological model starts from two simple lumped pre-models, one for a rural and one for an urban subcatchment, respectively. The dominant water processes will be identified from lumped models and inherited by 180 semi-distributed models for the simulation of the whole study catchment.
For parameter calibration, the initial range of each parameter was given based on empirical values (Gharari et al., 2014). Then, random parameter sets were sampled between the maximum and minimum limitations with the Monte Carlo method, and more complex models with more parameters were tested with larger numbers of parameter sample sets to ensure the calibration scale as "fair" as possible. For semi-distributed models, two 185 constraints are exploited to reduce the risk of unrealistic combinations of parameters (Hrachowitz et al., 2014): (1) the maximum percolation velocity in urban areas is assumed to be smaller than it in rural areas (Pmax,R > Pmax,U). (2) the water storage depth in the unsaturated zone in urban areas is assumed to be smaller than it in rural areas (Sumax,R > Sumax,U).

The expression of urban development in the model 207
In the semi-distributed model, three urbanization scenarios are expressed with two parameters, 1) the proportion 208 of urban areas in the whole catchment and 2) the proportion of urban grey areas in urban areas (1-D), under the 209 assumption that the degree of urban construction (including water drainage system) and population density are 210 assumed to be consistent in the city. The numerical features of the three urbanization scenarios are shown in 211 Table 3. 212 Compact factor of per capita urban grey areas (-) The shaded numbers are used to adapt the model for urbanization scenarios A, B, and C 214

The expression of LID practices in the model 215
The expression of LID in the semi-distributed model follows two procedures. First, the hydrological routes of 216 LID practices were designed and fit in the urban module. And then, reasonable values were assumed for those 217 involved parameters based on relevant literatures, realistic field test results, and data from local government files 218 (Carter and Jackson, 2007;Carter and Rasmussen, 2006;Collins et al., 2008;Hunt et al., 2008;Li et al., 2009;219 San Antonio River Authority, 2015; San Antonio Water System, 2020;Van Seters et al., 2006;). The schematic 220 model figure of four LID practices in the urban SUPERFLEX module is shown in Fig. 4. The mathematical 221 expressions of hydrological routes and the quantitative comparison of the parameter values are shown in Table  222 5. 223

Different hydrological responses of urban and rural areas 235
The simulated rainfall-runoff relationship under the current condition is shown in Fig. 5, with NSE as 0.68 and R 2 as 0.90 during the calibration period and 0.69 and 0.84 during the verification period. The observed total basin runoff is 166 mm in 600 research days. And it is 166 mm for the model result, in which urban districts (27 % of the catchment areas) produce 63 % of total runoff; only 37 % of total runoff is discharged from rural areas (73 % of the catchment areas). In general, rural runoff is more stable than urban runoff. Urban areas generate 240 peak runoff frequently, no matter in dry or rainy seasons, while in rural areas the peak runoffs appear less often with lower summits.
With the parameter calibration results as shown in Table 6, the different hydrological characters of rural and urban areas can be further explained: First, the precipitation distribution factor for the unsaturated zone (D) and the maximum unsaturated storage depth (Sumax) in the rural module are larger than them in the urban module, 245 which creates larger water retention capacity in rural areas; Second, the larger evaporation coefficient (Ce) in rural areas leads to massive water evaporation, which can be explained by the favorable vegetation condition in rural areas; Third, although the percolation capacity is similar in urban and rural areas with close the maximum percolation velocities (Pcmax), because of the dense vegetation the capillary rise capacity in rural areas is far larger than it in urban areas with higher maximum capillary rise velocity (Cmax); Finally, the value for 250 parameter Ks, indicating the flow rate of deep groundwater, in rural areas is larger than it in urban areas, which indicates a stable and fluent groundwater flow in rural areas. All these four hydrological differences lead to a more stable rural runoff than urban runoff.

Urbanization influences on basin runoff
The simulated total runoff volumes and the maximum peak values in 600 research days for the three 260 urbanization scenarios are shown in Table 7. From the view of total runoff volume, all three urbanization plans increased the total basin runoff at different levels. Scenario C, with the highest level of urban sprawl development (50 % of new residents following the infill development) without the compact of per capita living space, brought 14 % additional total basin runoff compared to the current situation. For scenarios B and A, in which 70 % and 100 % of the new residents following the infill development and with the compact factors 0.9 265 and 0.85, these growth rates are 8.7 % and 2.7 % respectively.  Fig. 6 shows the simulated total basin runoffs for three urbanization scenarios and the current situation. It can be found that, for most peak runoffs in the dry season, all three urbanization plans 270 brought obvious increases, among which scenario C always brought the largest peak runoffs followed by scenarios B and A consecutively.   Table 7.

275
For the maximum peak runoff happened in the rainy season (10 th Sep. 2018), the peak value increased by 16 % and 7.5 % for scenario C and B, as shown in Table 7. However, in scenario A, as a full infill urbanization scenario without urban expansion, the maximum peak runoff unexpectedly declines by 4.3 %. This is because the intensive rainfall events in the rainy season filled up the water retention capacity in rural and urban green areas, and the maximum peak is not only generated from the urban grey areas but also from large areas of urban 280 green surfaces. As shown in Fig. 7, the maximum peak runoff experienced two summits in succession from 9 th Sep to 11 th Sep. The small lower peak I was mainly contributed by the urban grey surface with fast hydrological response, and in this time, urban green areas were getting saturated and generating stable and rising outflow.
When it came to summit II, the flow from urban green surfaces reached the peak and became the main contribution of the summit II. In urbanization scenario A, part of urban green areas was replaced by urban grey 285 areas to meet the needs of population growth, and therefore the runoff of the latter summit II is partially moved forward and superimposed on the previous summit I, which causes the decrease of the latter summit II. This phenomenon also occurred on the next peak happening on 17 th Sep and caused a less peak in urbanization scenario A than the current situation, but the only difference is that the runoff generated from the almost saturated rural places played a more significant role for this peak event. 290 https://doi.org/10.5194/hess-2021-32 Preprint. Discussion started: 29 January 2021 c Author(s) 2021. CC BY 4.0 License.

Figure 7: (a) Two successive total basin peaks in flood season as the maximum peak in 600 research days and its following peak, in which the maximum basin peak experienced two summits (I and II), and (b) their rural and urban sub-flows in three urbanization scenarios and current situation.
Overall, the infill urban development strategy is more helpful in basin runoff control for both total volumes and 295 peak values than the sprawl urban development strategy. And secondly, even though urbanization inevitably brings the growth of basin runoff volume, the peak value of total basin runoffs can be reduced by adjusting the areas of permeable and impermeable surfaces, as the faster runoff can help to spread the peak over a longer period of time, hence reduce peaks in the total runoff.

LID performance in the non-flood season 300
The time series of forecast runoff in five LID scenarios and the Conventional urban Development scenario (CD scenario C, following a half-infill and half-expansion urban development strategy) are shown in Fig. 8.   Table 8. Figure 9 shows a zoom in on the typical peak event happening on 29 th Mar. 2018.
The third maximum peak runoff in 600 research days happening on 29 th Mar. 2018, non-flood season is selected as the typical peak runoff to further reveal the LID performance on peak runoff reduction in the non-flood 310 season as shown in Fig. 9. Sharing a similar feature with other peak runoffs, the typical peak experiences two times of summit I and II. The LID practices always reduce the first summit more significantly than the second one. This is because the first summit I is mainly generated by urban grey areas with rapid hydrological response, which is the domain of LID practices. However, the second summit II is mainly generated by large areas of urban green surfaces with slow hydrological response, and therefore the LID practices have limited influence on 315 summit II.  Table 8 https://doi.org/10.5194/hess-2021-32 Preprint. Discussion started: 29 January 2021 c Author(s) 2021. CC BY 4.0 License.
The modelled total basin runoff volume and two summit values of the typical total basin peak runoff in five LID 320 and one CD scenarios are shown in Table 8.

The performance of bioretention cell scenario
Bioretention cells have significant reduction effects on both total runoff volume and peak runoff values, second only to the permeable pavements. The total basin runoff was reduced by 2.4 % from 182 mm to 178 mm in 600 research days. As for peak values, bioretention cells produced considerable reduction on the summit I of the 330 typical peak with a removal proportion as 8.8 % from 7.3 mm to 6.7 mm , and the robustness of bioretention cells is also satisfactory resulting in a 1.4 % reduction ratio for the summit II.
The strong runoff reduction ability of bioretention cells ascribes to the rapid water infiltration between soil granules and a large volume of water transpiration by lush vegetation. According to the simulation results shown in Table 9, 55 % of the precipitation falling on bioretention cells infiltrated into the underground, and 22 % of 335 the rainwater evaporated. Then, 8 % of rainwater was retained in bioretention cells, and finally, the overflow from bioretention cells was only 16 %.

The performance of permeable pavement scenario
Permeable pavements show the best hydrological performance on basin runoff reduction among the four LID practices. In the permeable pavement scenario, the total runoff volume declined 2.5 % from 182 mm to 178 mm, 340 and the two summits of typical peak runoff were reduced by 9.5 % and 2.2 % respectively. Even though sharing a similar total runoff volume reduction with bioretention cells, permeable pavements better reduced peak runoffs.
In 600 research days, permeable pavements generated the least overflow, only 8.6 % of the total input rainwater, as shown in Table 9. Almost 90 % of the rainwater consumption of permeable pavements depends on the 345 infiltration because additional stormwater retention space is available in the subbase, the base and between the permeable pavers or in the porous asphalt pores. Since the large water retention capacity and forceful peak runoff reduction ability, bioretention cells and permeable pavements can be seen as the most effective LID practices for urban flood control and for releasing pressure on the urban drainage system.

The performance of vegetated swale scenario 350
Vegetated swales achieved an appreciable peak runoff reduction similar to bioretention cells and permeable pavements. As shown in Table 8, vegetated swales decreased 7.6 % of the summit I of typical peak runoff. But the sustainability of this peak runoff reduction ability is weak: rather than exhaustively consuming, vegetated swales delayed the runoff of the first summit I till the second one, which caused a larger summit II.
As for the retention of total runoff volume in the long term, the performance of vegetated swales was not 355 outstanding. During 600 research days, only 1 % of the total runoff volume was preserved. For the 437 mm stormwater conveyed by vegetated swales, 75 % of the rainwater was discharged to the urban drainage system, and only 25 % of the total rainfall was absorbed by the soil layer. It can be explained by the fast water transportation mechanism of vegetated swales. Without sufficient water retention capacity, vegetated swales do not support stable and continuous infiltration. Therefore, the total runoff reduction volume of vegetated swales 360 is distinctly smaller than the other three test LID practices.

The performance of extensive green roof scenario
The extensive green roofs brought about the least peak runoff reduction among four test LID practices. Two summits of typical peak runoff were reduced by 4.4 % and 0.2 %, which are far less than in other scenarios.
However, the reduction on total runoff volume of green roof scenario is more satisfactory with the reduction 365 ratio as 2.3 %, which is close to this ratio in permeable pavement (2.5 %) and bioretention cell (2.4 %) scenarios. According to the specific runoff retention amount, the water consumption of green roofs relied on evaporation (43 %) and water storage (20 %), while 37 % of the rainwater overflowed.
The significant difference between runoff volume and peak value reductions ascribes to the small water retention capacity of green roofs: Although the green roof shares a similar model structure to bioretention cells 370 with vegetation and soil, the soil thickness of extensive green roofs is small and no infiltration process happen on rooftops. The small water retention capacity of green roofs leads to a sensitive hydrological performance to the predecessor rains: If there are no or fewer predecessor rains, the green roof can still play a role in peak runoff reduction; however, when it comes to rainy seasons, the green roof will be easily filled up by the frequent storm events and lose its peak runoff reduction function. 375

The performance of mixed LID scenario
The mixed LID scenario is the most forceful LID scenario to reduce both the peak runoff and the total runoff volume. The typical peak runoff was decreased considerably in the mixed LID scenario, as 28 % for the submit I and 3.5 % for the submit II. As for the total runoff volume in 600 research days, the mixed LID practices restricted the generation of total basin runoff volume from 182mm to 170mm with a 6.9 % reduction ratio. 380 https://doi.org/10.5194/hess-2021-32 Preprint. Discussion started: 29 January 2021 c Author(s) 2021. CC BY 4.0 License.
Except for the large contribution area of LID practices, another advantage of the mixed LID scenario attributes to the cascade connection among LID practices, which adjusts the unbalanced water capture capacities of different LID practices, reinforcing the robustness of the LID system. Table 10 shows the specific water retention amounts of 4 LID practices in the mixed LID scenario. The evaporation of green roofs and bioretention cells, and the infiltration of bioretention cells and permeable 385 pavements realized abundant water uptakes. In the mixed LID scenario, bioretention cells became the most effective and efficient LID practices among four test LID practices with the largest runoff volume consumption and fewer construction areas. Comparing the hydrological performances of bioretention cells in single bioretention cells and mixed LID scenarios, with more water input and less construction area, the water retention ability of bioretention cells, especially the evaporation, was better developed with almost the same 390 proportion of overflow.

LID performance in flood season 395
The analysis above concerns the general peak runoffs in non-flood seasons. However, for two successive peak runoffs happening between 15 th Sep. 2018 and 24 th Sep. 2018 in flood season, all the five LID scenarios lost the peak reduction ability, as shown in Fig. 10. Table 8 records the specific total basin peak values. It is noticed that except for the peak happening on 23 rd Sep. 2018 in permeable pavements scenario, all the five LID scenarios brought bigger peak values than CD scenario C. And further, the mixed LID scenario, which was supposed to be 400 the most powerful runoff reduction plan, led to peak increases of 2 % from 6.35 mm to 6.47 mm and from 3.57 mm to 3.65 mm, which is 2.3 %.
https://doi.org/10.5194/hess-2021-32 Preprint. Discussion started: 29 January 2021 c Author(s) 2021. CC BY 4.0 License.  Table 11. 3 % This anomalous condition is triggered by the intensive rainfalls during flood season have exhausted the water retention capacity in rural areas, and therefore these peak runoffs are not only generated from urban areas but also from rural areas. The runoff delay function of LID practices slowed down part of urban peak runoff, which 410 caused stacking of urban and rural peaks and in consequence increased the total basin peak runoff. As the second peak event on 23 rd Sep. 2018 shown in Fig. 10, the urban sub-flow occurred two summits generating from urban grey and green surfaces, respectively. The summit I of urban sub-flow in CD scenario C was almost erased by LID practices in mixed LID scenario, but the decreased summit I was partly delayed and superimposed on the summit II, which brought a larger stack of urban and rural peaks and increased the total 415 basin runoff by 2.3 %.

405
Even though the increase is small, it is to be concluded that during extremely wet conditions, the effect of implementing LID measures on peak flow reduction is negligible, if not negative in basins with combined urban and rural land use. https://doi.org/10.5194/hess-2021-32 Preprint. Discussion started: 29 January 2021 c Author(s) 2021. CC BY 4.0 License.

Transferability of research results
Since the specific geographical conditions of study catchment, the research results about the growing total basin peaks after LID implementations are limited to restricted regions and specific weather conditions. The transferability of the result presented here needs to consider the following characters in other catchments: First, the basin should have a substantial portion of urban area to make the effects relevant for downstream areas. 425 Second, the location of urban areas ought to be close to the outlet of a (sub-)watershed to make the risk of stacking of faster urban peaks and slower rural peaks relevant, in particular in relation to LID implementation.
Finally, the extent of LID implementation will quantitatively influence the basin peaks, as a higher degree of LID implementation may bring a larger stack of urban and rural sub-flow in flood seasons.

Limitations 430
To decrease the model uncertainty caused by over-complex models and to determine a suitable level of model complexity (Hrachowitz et al., 2014), this research used a relatively simple semi-distributed model to simulate the rainfall-runoff relationship on the catchment scale. Heterogeneity within the rural and urban areas is not represented in our semi-distributed model.
Then, important assumptions are used in the urban development and LID implementation scenarios. First, 2.4 435 million residents are supposed to live in San Antonio City with three compact factors (0.85, 0.9, and 1) for living space in 2040 in three urbanization scenarios. Uncertainty in these figures is high. Next, the extent of urban construction and LID implementation is assumed to be consistent throughout urban areas. However, the construction density of urban core areas might be larger than the new-developed suburban areas. Finally, five LID implementation scenarios presume optimistic LID implementation conditions by using favorable LID 440 parameters, hence overlooking practical implementation, operation, and maintenance problems such as the damage of LID practices and the blockage in soil media. All such limitations lead to discrepancy between the model results and reality. The results however show that the answer to our research question, "[what is] the influence of LID implementation on the basin peaks at a catchment scale" remains valid for different urban development and LID implementation scenarios. 445

Recommendations
With regards to urbanization, the infill urban development strategy is recommended for flood control rather than the sprawl urban development strategy. Secondly, although urbanization may inevitably result in the rising of total runoff volume, extreme peak runoffs could be controlled by adjusting the ratio of urban grey and green areas and creating the time differences between the peak runoffs from these sub-areas. However, stacking of 450 peak flows from urban and rural parts of the basin should be avoided by making use of the faster urban runoff versus the slower rural one.
To improve our understanding on total basin peak control, future research can study the hydrological response times of different landscapes. The sub-areas dominated by different land use categories, soil types, topographic conditions, urbanization extents, and their positions in one catchment can be studied. This research assumed 455 homogeneous rural and urban hydrological patterns. Future research could further analyze this problem https://doi.org/10.5194/hess-2021-32 Preprint. Discussion started: 29 January 2021 c Author(s) 2021. CC BY 4.0 License.
considering spatially heterogeneous areas, such as partial urban development and uneven LID implementation condition, using distributed models with more precise data supports.

Conclusions
In this research, a case study for the catchment of San Antonio, Texas was conducted to investigate the 460 influence of LID implementation on the basin runoff at a catchment scale. Scenario analyses of both urban development and LID implementation were adopted to give a reliable answer to the research question. A SUPERFLEX conceptual model was adapted as a semi-distributed model to simulate the rainfall-runoff relationships of study catchment's hydrological behavior under different scenarios. It was found that: 1. The urban surface, taking 27 % of study catchment, generated 63 % of total basin runoff as 101 mm in 600 465 research days, while the last 73 % rural areas only produced 37 % of total runoff as 58 mm. And with less water retention capacity urban areas yielded peak runoffs more frequently than rural areas.
2. The infill urban development strategy benefits more from runoff control than the sprawl urban development strategy. All three urban development scenarios brought growth of total runoff volume with increase ratios as high as 14.3 % for scenario C (half-infill and half-expansion urban development plan), 8.7 % for scenario B (70 470 %-infill and 30 %-expansion), 2.7 % for scenarios A (fully infill development). Fortunately, however, by converting the urban development strategy from urban expansion to infill development, the extreme peak runoff can be reduced from 8.6 mm (scenario C) to 8.0 mm (scenario B) and 7.1 mm (scenario A) that even smaller than 7.4 mm in the current situation.
3. All the five LID implementation scenarios performed powerful runoff reduction ability on the peak runoffs in 475 non-flood seasons. Bioretention cells and permeable pavements can significantly decrease both total basin runoff volume (with the reduction ratios as 2.4 % and 2.5 %) and peak values (8.8 % and 9.5 %). With more water input and less construction area (as the ratio of drainage to the construction area increases from 1 to 1.5), the water retention ability of bioretention cells, especially evaporation, can be better developed with almost the same proportion of overflow. Vegetated swales, without substantial water retention capacity, seem to perform a 480 limited runoff volume reduction (1 %) than the other three LID practices but have satisfactory peak runoff reducing ability (7.6 %). On the contrary, green roofs have the worst peak runoff removal ability (4.4 %) and normal runoff volume reduction capability (2.3 %). The mixed LID scenario provides a forceful solution on runoff reduction as the typical peak runoff was decreased by 28.3 %, and the total basin runoff volume in 600 research days was restricted from 182 mm to 170 mm with a 6.9 % reduction ratio. 485 4. However, when it comes to flood season, all the five LID scenarios lost the peak reduction ability, and a higher degree of LID implementation leads to larger total basin peak runoffs. This is because the runoff delay function of LID practices slowed urban peaks, which caused more stack of urban and rural sub-flows and thus increased the total basin runoff. Two consecutive peak runoffs happening in flood season rose by over 2 % in the mixed LID scenario. 490 Data availability: Precipitation, evaporation, and runoff data are available on the USGS website (https://www.usgs.gov/). The impervious surface data related to Fig. 1 were retrieved from NASA Socioeconomic Data and Applications Center (SEDAC, https://sedac.ciesin.columbia.edu/).