Resilience effect of decentralized detention system to extreme flooding events

Flooding is one of the most recurring and destructive natural disasters worldwide that have caused substantial social and economic losses, particularly in urban watershed due to the concentration of critical infrastructure, economics, and populations (Wannous & Velasquez 2017). The severity and frequency of flooding events have been increasing in recent years, with 223 flood events in 2021, surpassing the 2001–2020 average of 163 floods, due to the synergetic effects of climate variability and socioeconomic changes (Giorgi et al. 2019; Delforge et al. 2022). More specifically, urban flooding results in damages amounting to approximately $96 billion yearly, with an estimated loss of $9 billion in the United States alone (World Bank 2016). The impacts of climate change and urbanization are expected to exacerbate the frequency and severity of urban watershed flooding events (Hettiarachchi et al. 2018; Zhou et al. 2019).

There are various contributors to flooding, such as structural failure (Cheng et al. 2010), aging infrastructure (Braud et al. 2013), rising sea levels (Vitousek et al. 2017), and increasing temperature (Wasko & Sharma 2017). However, urbanization-induced impervious surfaces and intensified rainfall due to climate change are the primary causes of flooding (Zhang et al. 2018). High-intensity rainfall produces rapid and extreme discharge, resulting in flooding and high peaks downstream. Land cover changes to impervious surfaces cause a significant increase in runoff volume and peak discharge, leading to flooding. The compounded effects of climate and land cover changes have led to a synergistic increase in flooding volume and peak discharge, contributing to more extreme flooding events (Li & Burian 2022).

Detention pond systems are a commonly used measure to mitigate urban runoff and flood discharge during high-intensity rainfall events. A detention pond temporarily stores storm runoff and releases it with a time delay, reducing discharge within the flow capacity of receiving rivers and mitigating peak discharges at the basin outlet (Goorden et al. 2021). The flood mitigation effects of detention pond systems depend on the design parameters such as location, pond size, outlet type and size, and outlet control (Park et al. 2012). The conventional approach to designing detention ponds has determined optimal design parameters for specific design storm events (e.g., 2-, 10-, and 20-year return periods) with single or multiple objectives related to costs, hydraulics (e.g., discharge, optimal location, and outlet type and size) (Ngo et al. 2016). The approach has generally offered a huge size of detention pond in a centralized manner by the economics of scale (Ngo et al. 2016).

The current detention system design considers historical data or average predictions; however, changing conditions from climate and socioeconomic changes create more significant uncertainty (Giorgi et al. 2019; Shen et al. 2019). The detention system, especially based on a centralized approach with a large single detention pond, can make the watersheds more vulnerable to flooding events that exceed the expected and in turn, introduce more catastrophic losses in society and local economics – which is described by the concept of ‘risk transference’ (Etkin 1999). Thus, the current decision-making approach needs to incorporate the extremity and uncertainty of flooding events into the planning of detention pond systems.

The synergetic impacts of climate and land cover changes lead to more extreme flooding events. Planning infrastructure to address these disruptions may require significant system capacity expansion, which can aid in preventing flooding. However, this approach entails high costs, environmental impacts, and significantly large space which may not be available in densely populated urban watersheds, and even if such measures are undertaken, they cannot guarantee the complete prevention of all extreme, uncertain flooding events (Wang et al. 2013). In this regard, incorporating the resilience concept into infrastructure decision-making – i.e., the transition of the approaches from prevention/protection to mitigation/restoration – has gained prominence in coping with extreme disruptions and uncertainties (Mugume et al. 2015; Shin et al. 2018). Resilience is the comprehensive capability of the system to withstand or absorb system disruptions and quickly recover to a normal state (Shin et al. 2018, 2020). Thus, resilience-based infrastructure strategies focus on minimizing the system's functional losses and rapidly recovering the disrupted performance to the pre-disrupted or better state (Shin et al. 2020).

One of the resilience strategies is an increase in the decentralization of a system. Various systems, such as water supply (Rabaey et al. 2020), transportation (Mohebbi et al. 2020), and energy microgrid (Wang et al. 2022), have introduced the decentralized system design approach to cope with uncertain disruptions. In the context of flood management, an example of such resilience strategies would be the implementation of spatially distributed systems with multiple detention ponds in their decentralized portions to detain storm runoffs (van Duin et al. 2021). Distributed detention ponds could be a resilience strategy for flood management that mitigate peak discharge and offer distributed flood mitigation benefits. They control the conveyance of peak discharge through a watershed and decrease the flow rate regionally (Myers & Pezzaniti 2019). Previous studies have explored the effects of distributed detention ponds on flood control, especially from the aspect of optimizing the design, vulnerability, or risk of a watershed during flooding events (Mobley & Culver 2012; Pereira Souza et al. 2019). A few studies have been conducted for decision-making on resilience-based flood control strategies. For example, Miguez & Veról (2017), Bertilsson et al. (2019), and Rezende et al. (2019) explored the design alternatives for stormwater management against future climate change. Wang et al. (2019) investigated the adaptation strategies for enhancing flood resilience of an urban watershed during flooding events. Mugume et al. (2015) determined the effectiveness of distributed detention ponds using a functionality-based global approach.

However, to the best of the author's knowledge, no studies have been made to investigate the resilience effects of decentralized, distributed detention systems to extreme flooding events driven by the combined impacts of high-intensity rainfall and land cover change, depending on the decentralization levels of the system. In addition, extreme flooding can produce the inundation of multiple sites in upstream and midstream as well as downstream regions, which can increase social and local economic losses in the entire watersheds (Ngo et al. 2017). The optimal options to reduce overall downstream flood may not satisfy for mitigating all inundations in upstream and midstream regions (Pereira Souza et al. 2019). Thus, the decision-making to address extreme flooding events needs to incorporate the evaluation of spatial performance and the effectiveness of the flood control options, which have not been covered in previous studies.

With this motivation, this study presents the resilience effects of a decentralized detention system with distributed detention ponds to extreme flooding events. The flood resilience of a watershed will be compared depending on the decentralization levels of the detention system and the combination of rainfall intensities and land cover changes. This study also proposes an advanced flood resilience measure that integrates information on the regional flood mitigation performance of distributed detention systems in an urban watershed. The contributions of this study are (1) proposing a new integrated flood resilience measure based on the regional performance of spatially distributed, decentralized detention systems; (2) identifying how flood resilience is characterized along with the decentralization levels of detention system and the combined impacts of rainfall and land cover changes; and (3) providing insights into the implementation of decentralized detention systems against extreme flooding events.

The definition of system resilience has evolved in engineering fields, focusing on maintaining non-disrupted performance and quickly returning to normal or desired states following disruptive events (Shin et al. 2018). In the context of flooding, resilience is defined as the ability of the system to resist, adapt, and withstand extreme rainfall storms (Joyce et al. 2017; Bertilsson et al. 2019).

Inspired by the resilience definitions, various resilience measures in water engineering fields have been released for incorporating resilience-based strategies into decision-making. For example, Miguez & Veról (2017), Chen & Leandro (2019), and Bertilsson et al. (2019) used conceptual frameworks and indicator-based metrics to quantify flood resilience. Likewise, Mugume et al. (2015) used functionality-based global resilience to evaluate the performance of urban drainage networks during flooding events. However, the resilience metrics proposed in the previous studies provide a lack of information on the spatially regional performance of the flood control systems to extreme events. Wang et al. (2019) proposed grid cell-based metrics to determine flood resilience; however, the resilience measure has a lack of considering the variation of functional performance during flooding events. In addition, the proposed resilience measure focused on adaptation strategies rather than the resilience performance of decentralized systems. In this regard, this study proposes an advanced flood resilience measure that integrates the resilience of sub-basin flood control systems in spatial domains.

The study area has a history of chronic flooding and experiences severe flooding every year. On 12 July 2017, most parts of the study area were flooded due to heavy rainfall. The total rainfall during the flood event was reported to be 116 mm, compared to the mean annual rainfall of 821 mm. This rainfall event corresponds to a return period of approximately 100 years, which exceeds the design frequency for the river drainage capacity (Sadin 2017). It is reported that the rainfall in the study area will be more intensified with high fluctuation due to climate change (Markus et al. 2018). Furthermore, the study area – partially included in the city of Chicago – has a rapid increase in impervious surfaces compared to the second half of the 21st century due to rapid urbanization. Thus, using the regional and integrated flood resilience measures (described in the previous section), this study evaluated the resilience of distributed detention systems in the proposed study area to extreme flooding events that are generated by the combination of climate and land cover changes.

The regional resilience measure in Equation (1) quantifies the resilience based on the watershed's hydrologic functionality (performance of interest). This study defined functionality as the ratio of a threshold discharge to the discharge under the flooding conditions for a given time step. Here, the threshold discharge indicates a maximum river discharge that produces no flooding. When the river discharge during a flooding event exceeds the threshold discharge, the functionality level drops below unity (or 100%), implying a failure (flooding) state of sub-basins. By the definition of functionality, the functionality can be estimated at a value of unity when the river discharge during the flood event is less than the threshold discharge. The functionality values equal to unity imply no failure of the detention system – i.e., a normal state of a basin with no occurrence of flooding. In this context, the maximum value of the functionality is constrained to unity.

The first step is the hydrologic analysis to evaluate the discharge hydrograph for the study area during flooding events. This study employed the HEC-HMS model for hydrologic analysis in the proposed study area. The HEC-HMS model is widely used to analyze rainfall-runoff and routing processes with topographic and meteorological properties of the basin (Archer & Fowler 2015). Each component of the rainfall-runoff process, including loss (runoff volume), transformation (direct runoff), and routing, of a watershed, is represented by a separate module in the HEC-HMS model. Soil Conservation Service Curve Number, SCS Unit hydrograph, and Muskingum routing methods are chosen for loss, transformation, and routing methods, respectively. These approaches are selected based on their applicability and limitations, data availability, suitability for the same hydrologic situation, stability, widespread acceptance, and well-established researcher recommendations. HEC-HMS consists of the following components: basin model, meteorological model, control specification, and input data.

The meteorological model consists of the parameters such as rainfall and discharge. The rainfall data for the study area were obtained using Precipitation Estimation from Remotely Sensed Information Using Artificial Neural Networks-Cloud Classification System (PERSIANN-CCS), which provides a real-time global high-resolution (0.04° × 0.04° pixel) satellite rainfall dataset (Nguyen et al. 2019) and PERSIANN-CCS are considered effective in rainfall-runoff simulation (Bhusal et al. 2022). The discharge data for the studying river segment was obtained from the USGS database. The study area has one USGS gauging station (USGS 05534500) at the most downstream location. The discharge data is available for the observations from 1986. The daily time-series rainfall data and discharge data were extracted for a time period of 2 years (2020–2021).

Control specification is a critical HEC-HMS component used for rainfall-runoff simulation (Scharffenberg 2016). Control specifications are used to regulate the starting and ending time of the simulation with the constant time step between starting and ending periods.

First, the geometry data (e.g., river centerlines, bank lines, flow paths, and cross-sections) for the target river segment was digitized using the RAS-Mapper tool in GIS. RAS-Mapper is a geospatial tool available in the HEC-RAS model, which is used to create geometric input files and visualize output flood extent in the study reach (Brunner 2016). LIDAR 1 m DEM data was used in the RAS-Mapper tool to create the geometry data with the alignment of the river line. Then, as a parameter to calculate the open channel flow in HEC-RAS, manning's n values – the degree of flow resistance – at each cross-section were assigned based on the land use and land cover classification, which follows a standard reference method by Chow (1986).

HEC-RAS also requires the data of discharge flowing into the river segment from upstream as a boundary condition. The discharge data for HEC-RAS was considered as the discharge at the outlet of each sub-basin in HEC-HMS. Finally, the model was calibrated by adjusting manning's n value as suggested by HEC-RAS User Manual (Brunner 2016). The simulation's water depth was compared to the water depth at the downstream gauging station for the calibration of the model. Finally, by analyzing the discharge range using HEC-RAS, the threshold discharge that produced no overflow on riverbanks was found for each river reach in sub-basins in the study area.

The verification of the HEC-HMS and HEC-RAS models was conducted by comparing the simulated data with observations (discharge for HEC-HMS and water depth for HEC-RAS) from the gauging station (USGS 07019317) in the study area for the period from 1 January 2020 to 31 December 2021. The performance of model verification was determined through multiple statistical indices – i.e., root mean square error (RMSE), Nash–Sutcliffe efficiency coefficient (NSE), coefficient of determination (R²), standard deviation ratio (RSR), and percentage bias (PBIAS) (Moriasi et al. 2015). Table 1 summarizes the mathematical representation of the indices.

Table 1

Indices for HEC-HMS model verification

Indices .	Mathematical expression .	Acceptable range .
Root mean square error
Nash–Sutcliffe efficiency coefficient		0.5 < NSE ≤ 1
Coefficient of determination		>0.5
Standard deviation ratio		0 < RSR < 0.7
Percentage bias		±25%

Indices .	Mathematical expression .	Acceptable range .
Root mean square error
Nash–Sutcliffe efficiency coefficient		0.5 < NSE ≤ 1
Coefficient of determination		>0.5
Standard deviation ratio		0 < RSR < 0.7
Percentage bias		±25%

where denotes the observed data; denotes the simulated data from the model; denotes the average value of the total number of observed data; and n denotes the total amount of data.

View Large

To investigate the resilience effects of decentralized detention systems to extreme flooding events, five decentralization options were defined. The first option is zero decentralization level with a single large detention pond basin in the study watershed. The detention volume was assumed to be 1,200,000 ⁠. The other four options are the decentralization levels with two, three, four, and five detention ponds with the same total detention volume as the single detention pond – e.g., for the option of two detention ponds, each detention pond has the detention volume of 600,000 ⁠. The best location for the single and decentralized detention pond (i.e., 2-, 3-, 4-, and 5-detention ponds) was determined by a trial-and-error method to find a location where the detention ponds maximize the reduction of peak flow during the 100-year return period rainfall event of the study area. The best locations for the distributed ponds were also determined by the same approach as the single detention pond option.

In this study, the occurrence of flooding events was hypothetically considered by the combination of heavy storm events and land cover changes. For storm events, the rainfall corresponding to 2-, 5-, 10-, 25-, 50-, 100-, and 200-year return periods were considered. The rainfall data, including the rainfall frequency and 24-h rainfall intensities for the study watershed, was obtained from the National Oceanic and Atmospheric Administration, National Weather Service (NOAA, NWS) rainfall frequency Atlas 14 (Bonnin et al. 2006), which is referred as NOAA-14. Similarly, four imperviousness scenarios were created hypothetically to demonstrate the impact of land use change. For the land use change scenarios, the impervious area of each sub-basin was considered as 20, 40, 60, and 80% of the sub-basin.

The effectiveness of decentralized detention systems in mitigating floods is highly dependent on the optimal location and combination of individual detention ponds within the watershed. Table 2 shows the best locations of detention ponds for the decentralization levels, which produced the maximum flood mitigation effects – i.e., the reduction of peak discharge at the watershed outlets for a flooding event by a 100-year return period rainfall to the study area. For the option of a single detention pond, its installation at the most downstream location in the study area offered a less effective option to control the peak discharge at the outlet of the watershed. The result shows that for the centralized detention system, placing a single detention pond at sub-basin 4 (along the river stream) provided the best performance in the reduction of peak flow for the 100-year return period rainfall event. Similarly, in the case of decentralized detention systems, i.e., the options of 2-detention ponds, 3-detention ponds, and 4-detention ponds, installing the detention ponds farther from the outlet of the watershed provides more effective reduction of peak flow. For the 5-detention ponds option, the detention ponds are distributed throughout each sub-basin.

The integrated flood resilience evaluation in Figure 6 reveals the resilience effects of a decentralized detention system compared to a single large-volume (centralized) detention system. The result indicates that the resilience increased by 5.8% when a single large-volume detention pond was replaced with five distributed detention ponds during a 2-year return period of rainfall intensity. In addition, it is noted that the flood resilience for the high intensity of rainfall events increased as the decentralization levels of the detention system increased. For instance, the result illustrates that, for a 200-year return period of rainfall intensity, the flood resilience of a watershed increased by 8, 9, 15, and 16.4% in the case of 2-, 3-, 4-, and 5-detention ponds options, respectively, compared to a single large-volume detention pond. The result illustrated a synergetic increase in the system's resilience as the number of detention ponds increased.

Interestingly, for the rainfall events of the 2-, 5-, and 10-year return periods, the detention system with the four-ponds option produced higher resilience than the 5-detention ponds option. This happened because, for the 5-detention ponds option, the peak discharges of the downstream reach (Figure 2) and outflow from the detention pond at sub-basin 5 occurred at the same time at the watershed outlet and, in turn, the peak discharge and flood duration at the watershed outlet were more increased. Detention system options 1, 2, 3, and 4 have no detention pond at sub-basin 5. Thus, the peak discharge of downstream reach converged with the discharge of sub-basin 5 at the time period for the recession limbs of sub-basin 5's hydrograph. On the other hand, the 5-detention ponds option includes a detention pond in sub-basin 5. It was analyzed that the discharge of the downstream reach and the outflow discharge of the detention pond at sub-basin 5 contributed to the peak discharge at the watershed outlet simultaneously. Finally, the more increase in the peak discharge and flood duration resulted in the lower flood resilience for the 5-detention ponds option, compared to option 4 for the 2-, 5-, and 10-year return periods of rainfall events. It is noted that option 5 was analyzed to have higher flood resilience than option 4 for the higher return periods of rainfall events because it produced more runoff storage effects at sub-basin 5.

Overall, a significant increase in flood resilience can be observed as the detention system is more decentralized with distributed detention ponds throughout the watershed. It is considered that the decentralized detention system can provide synergistic runoff storage effects that reduce the peak discharge and flooding duration at the upstream and midstream regions and consequently at the watershed's outlet during high-intensity rainfall events, which, in turn, increases the flood resilience of the watershed.

Similarly, the variation of failure duration results in Figure 7(b), depending on the detention system's decentralization levels and the rainfall events' return period, shows consistent results with peak discharge. The decentralized detention system with distributed ponds was found to be more effective in reducing the failure duration during extreme flooding events. This is because, when the detention ponds are distributed throughout the watershed, outflow discharge from each detention pond reaches the outlet of the watershed at different lag times due to their storage effects, which produces lower peak discharge and, consequently, shorter flood duration.

However, it was evaluated that the decentralized detention system can attenuate the decrease in flood resilience from the combined impacts of high rainfall intensity and imperviousness. For example, in comparison to the base scenario (i.e., 2-year return period rainfall intensity, 20% imperviousness, and no detention system), installing a single detention pond in sub-basin 4 (option1) produced a reduction of the flood resilience by 26% for the scenario of 200-year return period rainfall intensity and 80% imperviousness. However, the flood resilience decreased by 20% for the option of 2-detention ponds, which shows better performance in flood mitigation under extreme flooding events.

In addition, it was analyzed that the effect of attenuating the decrease in flood resilience to extreme events increased as the decentralization level of the detention system increased. For example, the options of 3-, 4-, and 5-detention ponds showed a reduction of flood resilience by 17, 16, and 14%, respectively, compared to the base scenario. Therefore, it is noted that the decentralized detention system can contribute to enhancing the flood resilience of a watershed to extreme flooding events that are produced by heavy rainfall and urbanization.

As seen in Figure 8, the flood resilience increases as the decentralization level of the detention system increases. This effect is more apparent in extreme event scenarios with higher rainfall intensity and imperviousness. For instance, from Figure 8(b)–8(f), as the decentralization levels of the detention system increase, the increase in the flood resilience is observed from 0.97 to 1.00 for the scenario of rainfall intensity of the 2-year return period and 20% imperviousness. However, the flood resilience increases from 0.72 to 0.85 for the scenario of rainfall intensity of the 200-year return period and 80% imperviousness. Thus, the result notes that the decentralized detention system can be a more effective option to enhance the flood resilience of a watershed to cope with extreme flooding events than a single large detention system.

The best locations of single or distributed detention ponds were analyzed to be upstream or midstream sub-basins rather than downstream sub-basins. These results can be explained in the following two ways. First, the detention systems with the ponds in the downstream sub-basin contributed somewhat to the time delay in flood runoff. However, the delayed runoff synchronized with the peak discharge of the mainstream that flowed less attenuated from upstream and midstream sub-basins and, in turn, contributed to the higher peak discharge at the watershed outlet (Antolini & Tate 2021). On the other hand, the detention system in the upstream and midstream sub-basins could avoid the synchronization of the downstream sub-basin runoff with the peak discharge of the mainstream – which flowed significantly attenuated from upstream and midstream sub-basins. Second, this result is well aligned with source control, in which mitigating regional floods will enable the suspension of flooding augmentation in watershed outlets (Kundzewicz & Takeuchi 1999). Thus, the decision-making on distributed detention systems needs to consider the incorporation of the decentralized detention ponds into upstream and midstream basins, which is contrary to the existing centralized approach of detention systems installing a single detention pond at the downstream region of a watershed.

The results demonstrated that the decentralized detention system produced higher resilience of the study watershed to flooding events. In addition, the resilience effects of the decentralized system increased as the decentralization levels (the number of detention ponds) and the extremity of the flooding events increased. The decision-making on detention systems can suggest an increase in the size (or capacity) of detention ponds to cope with extreme flooding events (Sahoo & Pekkat 2018). However, the decentralized detention system provides synergetic runoff storage effects that reduce peak discharge and flooding durations regionally and consequently in the watershed's outlet. Thus, within the budget constraints, maximizing the decentralization levels of the detention system is recommended to resiliently cope with extreme flooding events, rather than retrofitting or increasing the size of the existing detention ponds.

The result also showed that the increase in surface imperviousness interacted with higher rainfall events synergistically to decrease the system's resilience. Thus, incorporating the land cover change toward more perviousness into the decision-making with the implementation of decentralized detention ponds will contribute more effectively to coping with extreme flooding events in a watershed. This is also aligned with the green infrastructure (or low-impact development) strategies that aim to recover the hydrologic cycle to a pre-developed condition (Sohn et al. 2017).

However, compared to a centralized system, the implementation of a decentralized system generally accompanies substantial challenges from, e.g., substantial cost requirements, complexity in designing and operating the decentralized systems, and social adoption (Restemeyer et al. 2018). In particular, a decentralization system is an unattractive option under budget constraints in infrastructure investment in terms of the economy of scale (Sousa 2021). However, in the long-term perspective, the decentralized system can suggest beneficial effects such as reducing the flood damage and recovery costs from extreme flooding events; minimizing the environmental impacts; and reducing the chances of unexpected system failure as a failure of one control structure can be supported by other control structure in the system (Mugume et al. 2015; Hesarkazzazi et al. 2022). In addition to the budget constraint, spatial and environmental constraints (e.g., ecological impacts) also pose challenges in the planning of optimal and specific locations of detention ponds and their construction in a basin (Liu et al. 2016; Hosseinzadeh et al. 2023). Thus, the practical implementation of a decentralized detention system needs to be planned with the cost-effective and eco-friendly design and operation of distributed detention ponds to maximize the long-term effects.

Meanwhile, this study introduced a new flood resilience measure (Equations (1) and (2)) to evaluate the resilience of a watershed, which integrates information on the spatially regional performance of sub-basins with decentralized detention systems against extreme flooding events. A common approach to evaluate flood resilience has considered the runoff (or peak discharge) at the outlet of the most downstream sub-basin to understand the overall hydrologic response of a watershed with flood control options (Sahoo & Pekkat 2018). The proposed resilience measure evaluates the hydrologic response of not only the entire watershed but also the regional sub-basins, which allows for consideration of local inundations in upstream and midstream basins and the regional performance of decentralized detention systems to extreme flooding. In this context, the proposed resilience measure was discussed by additional comparison with the common approach. Table 3 summarizes the comparison of normalized quantiles resilience values of two resilience evaluation approaches (i.e., a common approach and a proposed approach) for the study watershed. It is noted that the proposed flood resilience measure amplifies the scale to better differentiate the resilience levels for the study watershed. For example, the range (i.e., the difference between the maximum and minimum values) of resilience values obtained from the proposed resilience measure from non-flooding events to the extreme flooding events was 0.61, while it was only 0.39 for the common resilience evaluation approach. While the 5th and 10th quantiles of the common resilience value are 0.65 and 0.71, the proposed resilience measure evaluates the values of 0.4 and 0.46; however, the 95th quantile of both the resilience measure is 0.98. Thus, it is considered that the proposed resilience measure can distinguish the resilience effects of decentralized flood control systems for the broad range of flooding conditions (e.g., rainfall intensity and imperviousness) and decentralization levels.

The flooding events are expected to become more severe and extreme in the coming years, which can result in catastrophic consequences in the social, economic, and infrastructure sectors. It is essential to transform current flood control systems into more resilient systems for coping with extreme flooding events. In this context, this study quantitatively investigated the questions associated with the resilience effects of a decentralized system approach to extreme flooding events from combined impacts of climate and land cover changes. The findings are summarized as follows:

• The detention system had more significant effects on flood mitigation when single or distributed detention ponds were located at upstream or midstream sub-basins, rather than downstream sub-basins. Thus, the planning of a decentralized detention system needs to consider the detention ponds distributed at upstream and midstream regions.

• Decentralized detention systems were a more resilient option to mitigate the impacts of flooding events by the synergistic storage effects on the overall watershed, compared to the centralized detention system. The resilience effects of the decentralized system increased as the extremity of flooding events increased. Thus, enhancing the decentralization levels of detention systems with cost-effectiveness is required to resiliently cope with extreme flooding events.

• The combined impacts of increasing rainfall intensities and impervious surfaces in the study area resulted in a significant reduction in flood resilience. However, the reduction rate of flood resilience decreased as the decentralization levels of detention systems increased. The implementation of a decentralized detention system with improving surface perviousness will suggest more effective flood mitigation, especially in addressing extreme flooding events.

• The proposed measure for integrated resilience of a watershed was useful in evaluating the regional hydrologic responses of the watershed (i.e., up-, mid-, and downstream sub-basins) and the resilience effects of decentralized detention systems to extreme flooding events. The proposed measure can be used in decision-making to evaluate the performance of decentralized flood control systems.

This study quantitatively demonstrated the contributions of decentralized detention systems to enhancing flood resilience to extreme flooding events. However, further research is required for the practical implementation of decentralized detention systems. First, as discussed above, the decentralized system may require higher construction, operation, and maintenance costs, in the context of economies of scale. Thus, further investigation needs to explore a systems approach to design the configurations and operation of decentralized detention systems to maximize cost-effectiveness and flood resilience. Second, this study investigated the resilience effects of decentralized detention systems in an urban watershed at a systems level. The practical construction of a detention pond in a basin can be limited due to local constraints in spatial, social, and environmental domains. Thus, there is a need to investigate the engineering insights into optimal locations and specific design of distributed detention ponds with the objectives related to spatial accessibility and social, economic, and environmental impacts. Third, some sub-basins can be unavailable to install more distributed detention ponds for a decentralization strategy due to limited space, budget constraints, technology uncertainties, and public acceptance. Various types of flood control measures, such as permeable surface paving, rain gardens, green roof, and rainwater harvesting, have been introduced and applied to mitigate the flooding impacts. Thus, the decentralization strategy can be extended to the combination of diverse flood control options, including detention ponds. There is a need to investigate the synergistic effects of diverse flood control options on improving flood resilience to extreme events.

All relevant data are included in the paper or its Supplementary Information.

The authors declare there is no conflict.