Simulation-optimization framework in urban flood management for historic and climate change scenarios

Rapid urbanization has a significant effect on urban floods. This situation has further deteriorated due to climate change, which results in an increment in the frequency of short-duration heavy intensity rainfall (Zhuang et al. 2019). Many studies proposed Best Management Practices (BMPs) for storm-water runoff quantity and quality control to mitigate urban floods (Lucas & Sample 2015; Esmail & Suleiman 2020). BMPs offer one of the best solutions for flood preparedness in an urban catchment (Li et al. 2019). With more rainfall intensity, the runoff volume captured by the BMPs increases (Lewellyn et al. 2016; Hou et al. 2020). The placement of BMPs can reduce flooding, pollutant load, economic loss, human loss, and health hazards (Boguniewicz-Zabłocka & Capodaglio 2020). It also improves ecological conditions while providing opportunities for water conservation, infiltration, and potable water saving (Antolini et al. 2020; Sarma & Rajkhowa 2021). BMPs include rain gardens, bioretention, rain barrels, vegetated rooftops, and permeable pavements (Braune & Wood 1999; Gülbaz & Kazezyilmaz-Alhan 2015). However, determining the optimal location, placement, and selection of cost-effective BMPs to obtain maximum runoff reduction is a challenge in the case of an urban watershed with distributed rainfall (Qin et al. 2013). These necessitate efficient modeling approaches (Adham et al. 2016) that assess BMP effectiveness (Yao et al. 2015). Relevant literature related to runoff-based study of BMPs, siting tool, screening, and simulation-optimization are as follows:

Many studies have used runoff models, namely the Storm Water Management Model (SWMM) and Hydrologic Engineering Center Hydraulic Modeling System (HEC-HMS) for estimation of runoff (Zoppou 2001). Oleyiblo & Li (2010) utilized HEC-HMS and HEC-GeoHMS for Misai and Wan'an catchments in China. They confirmed the model's appropriateness, capacity, and reasonableness for flood forecasting in catchments. Ramly & Tahir (2016) used HEC-GeoHMS and HEC-HMS to simulate flood at the upper Klang–Ampang River Basin, Malaysia, and processed the Digital Elevation Model (DEM) in the ArcGIS 10.2 environment. The simulated and observed discharges were 308.7 and 298.8 m³/s, respectively. Sarminingsih et al. (2019) analyzed the rainfall-runoff relationship for Garang watershed, Semarang, Indonesia, using HEC-HMS. They found that HEC-HMS was suitable enough to model the rainfall-runoff process. Karunanidhi et al. (2020) performed rainfall-runoff analysis for the Lower Bhavani river, India, and found that the annual runoff varied from 102.04 to 463.02 mm. The basin's average surface runoff volume would be 81 × 10⁶ m³, respectively, according to the Soil Conservation Service-Curve Number (SCS-CN) model. The number of researchers applied HEC-HMS for India (Islam et al. 2014; Ramachandran et al. 2019) and elsewhere (Ramly & Tahir 2016; Sarminingsih et al. 2019). However, very few investigated the impact of BMPs for runoff reduction in association with HEC-HMS. Hence, a study on the applicability of BMPs and their location identification for urban runoff computation is of utmost significance. Location of BMPs can be facilitated with a siting tool: EPA System for Urban Storm-water Treatment and Analysis INtegration (EPA-SUSTAIN 2014).

A limited number of researchers used SUSTAIN, which has SWMM as an inbuilt runoff estimation model, to locate BMPs. Kurkalova (2014) discussed BMP selection and placement aspects and concluded that the cost-effectiveness of BMPs and placement played a significant role in financial strategy. Similar views are echoed on the placement of BMPs by Sun et al. (2016).

Lee et al. (2012) used SUSTAIN for placing porous pavement and bioretention for 23 sub-watersheds in Kansas City, USA. They concluded that 2.2 and 1.6% of an entire watershed could be changed to bioretention and porous pavement. Jia et al. (2015) implemented bioretention, rain barrel, porous pavement, grassed swale, and green roof using SUSTAIN for cost-effective placement at a college campus in Foshan, China. Total peak flow, runoff volume, and pollutant loads decreased by 13.8, 14.5, and 17–21%, respectively. Warganda & Sutjiningsih (2017) studied BMPs in an urban watershed at Universitas Indonesia Campus, Indonesia. They assessed the performance of ten types of BMPs and justified the applicability of SUSTAIN for placement of BMPs at strategic locations. You et al. (2019) implemented Low Impact Development (LID)/BMPs in a residential community of China using SUSTAIN, and achieved a runoff reduction of 38.8% with a cost of $ 1.7 million.

Beck (2014) employed SUSTAIN to quantify BMP implementation impacts in Ballona Creek, Dominguez Channel, and Los Angeles River watersheds. Seven BMP types were optimized using Non-dominated Sorting Genetic Algorithm-II (NSGA-II) for the cost-effectiveness curve. Results indicated that dry weather total daily load (TMDL) could reduce by 80–99%, and 10–50% in wet weather TMDL. The reduction of 20–50% in peak runoff is noticed along with groundwater recharge of 12,000–30,000 ac-ft per year.

Shishegar et al. (2018) expressed that mathematical modeling is an effective way of optimizing the performance of BMPs and suggested categorization of different techniques such as Linear Programming (LP) and Non-Linear Programming (NLP). Loáiciga et al. (2015) applied Binary Linear Integer Programming (BLIP) for selecting two types of BMPs for a case study of Boulevard, USA. They found BLIP to be advantageous for deciding whether to place or not to place the specified BMPs developed with standardized designs. Sadeghi et al. (2018) used the Mixed Integer Linear Programming (MILP) model with La Recarga runoff model for placing BMPs in Los Angeles city, CA. Wu et al. (2020) studied optimal combination of LIDs for Gongming, Shenzhen, South China.

Increments in urban floods are anticipated due to climate change in the years to come. Therefore, similar studies to that of the historic scenario to gauge the impact of placing BMPs on urban flood mitigation need to be conducted. In this regard, the following section discusses the literature review related to impacts of climate change.

Coupled Model Intercomparison Project 5 (CMIP5) initiated the mechanism of Global Climate Models (GCMs), and Representative Concentration Pathways (RCPs) to assess future rainfall in climate change situations (Taylor et al. 2012). Khaing (2015) studied the impact of climate change (ICC) on hydropower generation and streamflow in the Myitnge river basin, Republic of the Union of Myanmar. He applied HEC-HMS to study the changes in hydrological regimes of the future, and projected a decrease in long-term average annual discharge using 2 GCMs. There was a decreasing discharge trend in RCP 8.5 of GFDL-CM3 with a range of 18–23% and 13–25% in Upper Yeywa and Yeywa, respectively. Rukundo & Doğan (2016) studied the occurrence of floods for Kigali city, Rwanda, Africa. They used HadCM3 with A2 and B2 scenarios. An increase in rainfall is expected in June 2020 (30.2%), May 2070 (27%), and November 2070 (23.6%) for HadCM3-A2. Das et al. (2018) focused on scenario uncertainty and modeling, using the possibility theory and reliability ensemble averaging for projecting stream flows over the Wainganga river basin, India. They found that MPI-ESM-LR in the RCP 8.5 scenario has the highest possible value compared to RCP 4.5. A similar increasing trend is found for HadCM3-B2. Effects of climate change on urban floods in northern China are studied by Zhou et al. (2018). The flood volume increase is noted under RCP 8.5. Zhou et al. (2019) found that changes caused by urbanization on runoff were significantly higher than climate change. Nilawar & Waikar (2019) quantified climate change impacts on streamflow and sediment concentration in India's Purna river basin, under RCPs 4.5 and 8.5 for four future periods. An increment in both was found as compared to the baseline condition. Ramachandran et al. (2019) performed integrated flood modeling with HEC-HMS and HEC-RAS over Adyar sub-basin, India using data of 5 GCMs and RCP 4.5. There is an increase of 34.3–91.9% in the peak discharge and 12.6–26.4% in the flooded area compared to the present climate.

Kuntiyawichai et al. (2020) used HEC-HMS to generate hydrographs based on three GCMs, RCPs 4.5 and 8.5 for the lower Nam Phong River Basin, Thailand. Flood damage under RCP 4.5 was more significant than that of the baseline. Avashia & Garg (2020) found that the number of flood events in 42 Indian cities is considerably lower in RCP 2.6 than RCPs 4.5 and 8.5. An increase in mean annual precipitation under RCP 8.5 is noticed. Lee et al. (2021) evaluated hydrological risks of n-year floods in Korea. An increase in hydrologic flood risk was observed as compared to the present level. Lompi et al. (2021) simulated the rainfall-runoff process and the associated flood risks of future scenarios for a case study of Pamplona. Design peak discharge is higher in RCP 8.5 than in RCP 4.5. Mandal et al. (2021) employed SWAT for the Subarnerekha river basin for computing river discharges and other hydrological fluxes. An increase in rainfall for all RCPs is reported. Javadinejad et al. (2021) studied climate change impact on storm-water and the probability of maximum flood for the Zayandeh rud river basin, Iran. Maximum probable precipitation in the basin for RCPs 2.6, 4.5, and 8.5 can change by +5, −5, and −10% for the period of 2006–2040 as compared to that of 1970–2005.

As inferred from the literature review, limited studies have discussed various aspects of BMP placement, runoff estimation, and optimization procedures for historic and climate change scenarios (Ramachandran et al. 2019; Kuntiyawichai et al. 2020). Accordingly, research gaps identified were as follows:

• 1.

  Limited studies on the analysis of BMP siting and placement for historic and climate change scenarios.

• 2.

  No study is reported on the Ratio of Runoff Reduction to Cost (RRRC) as a screening process before optimizing the location of BMPs.

• 3.

  Limited studies are reported on utilizing BLIP for placement or non-placement of BMPs.

• 4.

  Limited studies are noticed on considering existing policies of government and infrastructure creation.

• 5.

  No extensive studies are reported for Indian conditions.

Accordingly, a simulation-optimization approach was proposed including runoff simulation, BMP siting, and screening based on RRRC, followed by optimal placement of BMPs using BLIP. Hence, the objectives both in historic and climate change scenarios after considering research gaps are as follows:

• 1.

  To compute basin-wise runoff using HEC-HMS.

• 2.

  To site locations of potential BMPs using EPA-SUSTAIN BMP.

• 3.

  To screen multiple BMPs through a two-phase screening process, i.e., satellite imagery (primary screening) and then based on RRRC (secondary screening).

• 4.

  To employ BLIP for achieving an optimal combination of BMPs for maximum runoff reduction in terms of placement or non-placement of BMPs.

The developed methodology (refer to Figure 1) was applied to Greater Hyderabad Municipal Corporation (GHMC). The workflow is as follows: EPA-SUSTAIN BMP siting tool has been used to determine the potential location of BMPs. Here, BMP siting and screening procedures were performed in sequence. Runoff values have been simulated using HEC-HMS independently. Later, BLIP optimization was performed for optimal placement of BMPs by using simulated runoffs and calculated pollutant loads. In addition, impacts of BMP placement on runoff for various combinations of budget availability (BA) and pollutant load are studied. This process is repeated for historic RCPs 6.0 and 8.5. The developed mechanism ensures efficient utilization of available open spaces in an urbanized catchment and rooftops for reducing the severity of floods and damage caused due to excess runoff. A case study, data collection and processing details, which are part of the methodology, are presented next.

GHMC has an area of 625 km² that falls in the catchment of the Musi river (11,000 km²). It is divided into 16 zones based on a storm-water network (Figure 2) and gets an annual average rainfall of 840 mm. Due to its undulating topography in some areas and increase in imperviousness, rainwater tends to gather in the low-lying regions resulting in loss of property (Vemula et al. 2019). Hence, there is a need for runoff reduction. BMPs can facilitate the same in the available open space of 125 km², approximately 20% of the urban catchment.

Salient data collected and processed are rainfall, runoff, land use land cover (LULC), population, and reduction efficiencies. Brief but relevant information is as follows:

Historic rainfall data is collected from GHMC and the Directorate of Economics and Statistics (DES) for 2016. The temporal distribution was assumed to be uniform over 24 h. In the climate change scenario, the future extreme rainfall events analyzed are from 2006–2100. They are based on GCM, Geophysical Fluid Dynamics Laboratory Coupled Physical Model (GFDL-CM3), and the non-linear regression-based statistical downscaling technique was used to downscale the RCP scenarios to the catchment scale (Vemula et al. 2019; Swathi 2020). Event-based extreme events are considered as follows:

• 440.35 mm (0.30, 7.21, 79.42, 117.61, 181.69, 35.39, 3.40, 2.33, 3.30, 2.71, 1.26, 0.88, 1.33, 1.99, 0.78, 0.57, and 0.18 mm, mid-century extreme event, 2050) for RCP 6.0.

• 624.2 mm (0.21, 0.10, 1.29, 1.39, 0.83, 0.89, 2.82, 3.93, 5.27, 10.55, 17.78, 17.42, 15.70, 33.04, 48.59, 324.14, 135.86, 3.45, and 0.94 mm, mid-century extreme event, 2064) for RCP 8.5.

More details about the event-based approach are available in the Supplementary material (refer to Section S1). In the present study, only RCPs 6.0 and 8.5 were chosen for demonstration (Vemula et al. 2019; Swathi 2020). However, detailed results related to the remaining two RCPs, 2.6 and 4.5, are available from the corresponding author on request.

Runoff data at Hussain Sagar corresponding to rainfall for September 1–19, 2016 was obtained from Hussain Sagar Lake Development Authority (HSLDA 2016). This runoff data was used for calibration as discussed in the calibration and runoff estimation section. The input data used is Advanced Space-borne Thermal Emission and Reflection Radiometer (ASTER)-based digital elevation model (DEM) of spatial resolution 30 m × 30 m (USGS 2016). Details of storm-water drainage network, soil data, water quality data of Hussain Sagar Lake, and guidelines for placing rainwater harvesting structures are available from the corresponding author on request. Percentage impervious land use was based on the supervised classification for 1995, 2006, and 2016. Imperviousness for 1991, 2001, and 2015 was obtained from Sannigrahi et al. (2018) and Nayan et al. (2020).

LULC data has been acquired from the open street maps (Open Street Maps 2016) of spatial resolution of 20 m. In addition, some of the building data were downloaded as shapefiles having 2.62 m multi-spectral spatial resolution (Google Earth 2016). LULC and imperviousness percentage of the year 2031 was prepared from the available Master plan of Hyderabad Metropolitan Development Authority (HMDA 2019). Curve fitting is used to represent the available data by assigning a best-fit curve. The sigmoidal curve was found to be the best practical fit (with R² value of 0.9567) as it was adhering to the constraint of imperviousness percentage not exceeding 100%. Land imperviousness and CNs for future periods corresponding to RCPs 6.0 and 8.5 were extrapolated using the sigmoidal curve. Imperviousness is expected to increase to 84.85% by 2050 and 87.04% by 2064. Corresponding CNs were 88 and 89. The assessment of population for various time zones is needed. The polynomial fit was identified due to its higher R² value. The forecasted population for 2050 and 2064 was 1.902 × 10⁷ and 2.420 × 10⁷, respectively.

Table 1 presents depth, runoff, and pollutant load reduction efficiencies for different BMPs recommended for GHMC (Rao & Surinaidu (2012) for bioretention, Battiata et al. (2010) for vegetated filter strip, and HMWSSB (2012) for the remaining BMPs). Pollutant load reduction efficiency of each type of BMP varies from 0 to 75%, while runoff reduction efficiency varies from 10 to 80%. In addition, runoff reduction efficiencies for each BMP may not vary significantly across GHMC as the catchment is highly urbanized with predominantly sandy loamy soil (Rangari et al. 2021). Here, infiltration basin, infiltration trench, and rain barrel are termed as rooftop BMPs, whereas the remaining are non-rooftop BMPs (refer to Table 1).

Information about cost/m³ was provided by HMWSSB (2018) for both rooftop and non-rooftop BMPs which are as follows:

• 1.

  Rooftop BMPs: Rs. 4,000 (Rs. 3,500 for placement and Rs. 500 for silt traps, gutters, or downspouts used for transporting the collected rainwater, leaf screens, first-flush diverters, and roof washers).

• 2.

  Non-rooftop BMPs: Rs. 3,850 (Rs. 3,500 for placement and Rs.350 for silt trap).

BMPs with specific dimensions and efficiencies were placed for historic scenario and will be assumed to be same for future scenarios by providing regular maintenance. Accordingly, 10% of construction and placement cost were considered, which is prevailing at present as the operation and maintenance (O & M) costs in the budget. O & M is assumed to be constant in the modeling framework and may vary according to change in the budget as communicated by HMWSSB authorities.

Details of HEC-GeoHMS/HEC-HMS project setup, EPA-SUSTAIN siting tool, screening procedures, and BLIP are as follows:

HEC-GeoHMS is an extension of HEC-HMS, which works on the Geographical Information System (GIS) platform. It helps users to visualize and analyze spatial data, delineate sub-basins and streams that provide inputs to HEC-HMS. All the GIS data layers representing sub-catchments, conduits, junctions, and outfall were imported to HEC-HMS to develop the model structure for runoff calculation. HEC-HMS consists of five modules as follows:

• In the basin module, each sub-basin is provided with a specific CN and types of losses.

• In the meteorological module, each sub-basin is assigned with rain gauge information. The rain gauges were set to various catchments based on the Thiessen polygon approach. Rainfall data were transferred to respective sub-catchments based on a specified hyetograph method.

• The start date and end date of rainfall data were given as temporal input for each sub-basin in the control specifications module.

• Time-series data were given as rainfall input based on rain gauge information in the time-series module.

• Storage discharge functions were mentioned as necessary reservoir conditions to calculate runoff in the paired data manager module.

The runoff computation was dependent on CN employed for calibration of each sub-basin, and accordingly, final runoff could be obtained at the outlet. The shapefiles created in HEC-GeoHMS are supported as background maps in HEC-HMS and flow connectivity operations. The peak discharges, runoff volume, and runoff depth for each sub-catchment, junction, and outlet were calculated. This information was required for calibrating the model.

EPA-SUSTAIN (2014) BMP siting tool suggests multiple feasible BMPs for a given location based on its topographical features employing a suitability criteria matrix. Input/criteria include slope, drainage area, hydrologic soil group, imperviousness, depth of water table, building buffer, road and stream buffer, land ownership, and land use suitability (Cheng et al. 2009). The siting tool uses a concept of buffer, where a strip of land around or from the feature (i.e., roads, railways, and streams) is available for placement of BMPs. In this tool, volume-dependent BMPs (bioretention, rain barrel, constructed wetland, grassed swales, infiltration basin, infiltration trench, sand filter surface, and vegetated filter strip) and area-dependent (porous pavement) BMPs were considered.

However, we can place only one BMP in a given location. This constraint is due to land use, location specifications, materials used, and the constitution of BMPs. In addition, the efficiencies of BMPs are not the same (Table 1). It prevents the use of two BMPs of different categories over the same area. For this purpose, a two-stage screening procedure consisting of primary screening and secondary screening was developed. Primary screening was established on background satellite imagery used for ground-truthing. Here, ground-truthing implies interpretation of remotely sensed satellite images in the form of aerial photographs (raster images or vectorial map representation in GIS), making decisions on on-site supervision and compatibility of BMPs with surface characteristics. This ensures that incompatible BMPs in terms of land use are not considered. Accordingly, more suitable BMPs were prioritized over other overlapping BMPs. Suitability criteria applied for primary screening is that a vegetated filter strip is preferred on the circumference of the lakes or water bodies. Similarly, sand filter surface, porous pavement beside roads or pavements, sand filter surface near buildings, grassed swales along the railway lines, bioretention in open areas, infiltration trench, infiltration basin, and rain barrel for rooftop harvesting. A secondary screening module filters BMPs located at the same place in such a way that one BMP per location is achieved. The secondary screening was based on the ratio of runoff reduction volume capability of each BMP to the cost of respective BMP.

Figure 3 displays a visualization of the effects of screening on overlapping BMPs at a given location. BMPs are first sited using the siting tool (Figure 3(a)) followed by a primary screening where BMPs are screened based on their suitability with their respective location(s) (Figure 3(b)). The primary screened BMPs then go through a secondary screening process where a decision is made based on RRRC for placement of cost-effective BMP (Figure 3(c)).

Accordingly, RRRC is ⁠. Here, = 1, 2,…, n are the number of BMPs; is the runoff reduction volume capability of ith BMP (m³); is the cross-sectional area occupied by ith BMP (m²); is rainfall at a location where the ith BMP is placed (m); is the runoff reduction efficiency of ith BMP (%) (row 1 of Table 1); is the cost of ith BMP (Rupees); is the cost of ith BMP (m³) (Rupees); ‘1’ or ‘2’ where ‘1’ = Rooftop BMP and ‘2’ = Non-rooftop BMP, and is the volume occupied by ith BMP (m³).

Upon completion of the screening process, BLIP-based optimization was performed to ensure optimal placement of BMPs. It was assumed during modeling that (a) only one BMP can be placed in the available space, i.e., the placement of a BMP was assumed as 1, and non-placement was considered zero, (b) the depth of the BMP will not vary, and (c) the location of BMPs are fixed for both RCPs while harvesting rainwater.

The following section presents results and discussion related to historic and climate change scenarios where individual zones of GHMC are analyzed.

This section discusses computation of runoff, siting of BMPs, screening, and application of BLIP for the historic rainfall experienced by GHMC during September 20–28, 2016.

HEC-HMS was calibrated for the rainfall event of 176.9 mm (occurred during September 1–19, 2016) at zone 12 of GHMC. This event resulted in a flood depth of 0.5–3.9 m. It was found that areas in zone 5 are very highly vulnerable whereas areas of zones 1, 4, 10, and 12–15 are highly vulnerable (Surwase & Manjusree 2019; Vemula et al. 2019; Swathi et al. 2020). Calibration was performed for an observed runoff of 163 mm, and the resulting simulated value was 160.9 mm.

The spatial variation was taken into account by considering the runoff computation by the HEC-HMS model on a basin level using HEC-Geo HMS (USACE 2009). HEC-Geo HMS serves as a geospatial hydrologic modeling extension of HEC-HMS, which spatially generates all sub-basins based on catchment delineation of all streams. Significant inputs to HEC-GeoHMS were DEM and stream network. The stream network for the area was obtained from DEM by processing the Arc Hydro tool, which was a plugin in ArcGIS software. One hundred and twenty-eight sub-catchments were delineated based on stream order 3. Rainfall data from 12 rain gauge stations were available. The Thiessen polygon approach was employed due to its ability to spatially map the rainfall weights of each station with areal coverage (Ridwan et al. 2021). In addition, a sparse rain gauge network and less variability in rainfall are additional features that encouraged the use of the Thiessen polygon approach (Malik & Kumar 2020). The sub-watersheds were further grouped into ten different partitions for ease of evaluation. Based on this partition, ten junctions were identified where runoff could be calculated. Junctions were named based on the sub-catchments. Employed parameters, their ranges, obtained values, and remarks are presented in Table 2.

These calibrated values are obtained by comparing the simulated flows at the Hussain Sagar outlet for zone 12 with observed flows available from Hussain Sagar Lake Development Authority (HSLDA 2016). The calibration of HEC-HMS with observed flows resulted in an R² of 0.99 for zone 12 of GHMC. This enabled us to employ the same for future scenarios. Accordingly, all the 16 zones of GHMC were simulated for runoff for the extreme event rainfall, i.e., 215.9 mm for September 20–28, 2016. The summation of all these runoffs at the sub-basin outlets is the total runoff. The simulated runoff at the outlet for GHMC was determined to be 198.76 mm. Inference from the results was that zones 15, 12, 13, and 14 have the highest runoff values of 272.89, 252.75, 219.90, and 209.89 mm compared to other zones. Upon simulating the runoff values using HEC-HMS, we move ahead to discuss the results related to BMP siting.

BMPs were placed based on the feasibility constraints and GHMC guidelines. For example, buffer varies from 15 to 20 m for arterial roads. Here, buffer refers to the strip of land in the periphery of roads or streams, i.e., the maximum distance within which BMP should be placed for roads and the minimum distance beyond BMP should be established for streams. Depth to water table from the ground affects the choice of specific BMPs. For example, drainage area >1.01 × 10⁵ m² favors constructed wetland or wet pond; road buffer >30.48 m favors all non-rooftop BMPs except bioretention, grassed swales, and vegetated filter strip. A high-water table (>0.61 and <1.22 m below ground level) will favor the choice of bioretention, grassed swales, porous pavement, sand filter surface, and vegetated filter strip over a constructed wetland or wet pond. As a note, only one type of rooftop BMP can be placed on any given building. However, competition existing among non-rooftop BMPs is mainly due to the terrain characteristics (EPA-SUSTAIN 2014). Table 3 (columns 2 and 3) presents results related to siting analysis as follows:

Out of 81.70 km², rooftop harvesting covers 75.04 km², more than non-rooftop BMPs, i.e., 6.66 km², thus significantly reducing runoff. Rooftop BMPs, namely infiltration basin, infiltration trench, and rain barrel, occupied 34.82, 38.29, and 1.93 km², respectively, totaling 75.04 km². The number of BMPs in this category was 88,415; 349,417; 54,440 totaling 492,272 out of 505,383 BMPs. The number and type of BMPs give an idea of the budget required for installation of BMPs. Since each BMP has specific efficiency, the number of BMPs will help to understand the amount of water thus saved. It also gives a quantitative idea of locations where BMPs can be placed. Reported results confirm that rooftop BMPs are more prominent than non-rooftop. It implies that the buildings can be used as a source of flood mitigation, and thereby proving itself an efficient BMP.

Porous pavement, sand filter surface, bioretention, grassed swales, vegetated filter strip, wet pond, and constructed wetland occupy respectively 0.58, 2.21, 2.16, 0.68, 0.85, 0.09, and 0.09 km², totaling 6.66 km². The number of bioretention BMPs was 2,882 out of total non-rooftop BMPs of 13,111, as bioretention has the highest RRRC among the non-roof top BMPs; and hence, it is prioritized over others. However, the sand filter surface with the highest number of BMPs of 3,661 is more economical and easily implementable. It was deduced from the results of zones 1 and 14 that a greater number of BMPs did not always necessarily mean occupying more area. Effects of screening were predominant in zone 8 implying that its terrain was the most compatible (satisfying the constraints mentioned) with the placement of different types of BMPs.

498,908 BMPs were filtered, and the area covered by them was 78.72 km² (75.04 km² occupied by rooftop BMPs and 3.68 km² by non-rooftop BMPs). Reduction in non-rooftop BMPs area after primary screening is 2.98 km² as per siting analysis, and no visible effect on rooftop BMPs is found.

Analysis of data shows 496,956 BMPs (492,272 rooftop and 4,684 non-rooftop) covering an area of 77.57 km² (respectively 75.05 and 2.52 km²). The reduction in BMPs from primary screening to secondary was 1,952, with a corresponding reduction in the area of 1.15 km². Zones 2 and 16 show no screening effect due to the absence of non-rooftop BMPs. The calculated cost of 496,956 BMPs was Rs. 25.68 × 10⁹ (Rs. 18.01 × 10⁹ for rooftop and Rs. 7.67 × 10⁹ for non-rooftop). These are used further for estimating RRRC and for filtering BMPs for BLIP purposes. Post-screening locations of BMPs are presented in Figure 4. It is observed that the number of infiltration trenches is more, compared to other BMPs. Non-rooftop BMPs are fewer in number as the suitable open space available for BMP placement is comparatively fewer than the rooftop area. The total area of buildings in GHMC is around 84.91 km² which occupies 13.58% of the catchment area. In addition, zones 12 and 15 are presented in the Supplementary material (refer to Section S2). These two zones are presented due to their higher catchment areas of zone 12 (138 km²) and zone 15 (67 km²) and being more prone to flooding. However, other zones are not presented to minimize repetition.

As a note, EPA-SUSTAIN is used for determining the potential location of BMPs under the feasible constraints and screening procedures for removal of any overlap among BMPs. On the other hand, BLIP facilitates optimal placement or non-placement of BMPs.

BLIP is performed on the outputs of secondary screening adhering to its constraints of budget and pollutant load reduction. There is no minimum runoff reduction potential targeted in this study. This is because the runoff reduction capacity of BMPs varies with BA and pollutant reduction target. Accordingly, runoff reduction potential has been used only as an objective function in this paper and not as a constraint, thereby preventing the establishment of a condition on minimum runoff reduction. Higher budgets facilitate placement of more BMPs which, in turn, capture more surface runoff and vice-versa.

For this purpose, the total runoff was estimated by summing the individual runoffs at outlets of each zone, i.e., 142.21 × 10⁶ m³. Total pollutant load (TPL) was the product of total runoff value and BOD₃ of 14 mg/l (The Times of India 2018), resulting in 1,990.9 tonnes. Although reasonable estimates of water quality concerning BOD values were not available, as an attempt, this average value was introduced to the BMP placement mechanism by taking into consideration water quality as well. This process ensured a holistic approach toward evaluating the impacts of BMPs on runoff (water quantity and water quality). The perceived constraints were 100% BA and 3.38% TPL, i.e., Rs. 25.68 × 10⁹ and 67.2 tonnes. With this information, BLIP was performed individually for an optimized placement combination of BMPs to reap benefits in maximizing runoff reduction.

BLIP resulted in 495,607 BMPs covering an area of 76.99 km². Infiltration basin and infiltration trench occupied 34.65 and 38.28 km², resulting in runoff reduction of 5.95 and 4.42%, respectively. Runoff at the outlet before placement of BMPs was 198.76 mm, whereas it is 177.22 mm after placement, showing a reduction of 21.54 mm. The budget utilized is Rs. 24.82 × 10⁹, and the pollutant load reduction achieved was 67.38 tonnes. Runoff reduced was 15.41 × 10⁶ m³. It can be stored and utilized for urban catchment requirements. The population of GHMC was 8.99 × 10⁶ in the year 2016 (GHMC 2011a, 2011b). Accordingly, the harvested water serves the need of the urban population for 11.43 days based on the requirement of 150 liters per capita per day (CPHEEO 1999). The analysis conducted here indicates that BMPs may be used for flood control and other conservation measures. The remaining possible combinations are discussed later in the sensitivity analysis section.

In an attempt, results related to two RCPs, 6.0 and 8.5, were presented here. Like the historic scenario, the simulation-optimization approach comprises HEC-HMS for simulating runoff, screening for filtering BMPs, and BLIP for optimization (refer to Figure 1). Calibrated parameters and projected population, LULC, and CN are used (refer to Data collection and processing section). The siting tool and screening procedure results will remain the same as the historic scenario, as these depend on the topographical constraints of land use and available space. Related results of RCPs 6.0 and 8.5 are as follows:

Since the secondary screening results remain the same, the number of BMPs does not vary. But, since the BMPs need to be maintained to ensure their runoff and pollutant load reduction efficiencies, the budget increases due to these additional maintenance costs.

Simulated runoff was 602.65 mm. The equivalent runoff volume was 46.78 × 10⁷m³, and the pollutant load was 6,549.82 tonnes. Similarly, the achievable pollutant load reduction was 202.39 tonnes. The budget for placing all secondary screened BMPs and maintaining them was projected to increase to Rs. 148.95 × 10⁹ for RCP 8.5. The budget utilized is Rs. 143.14 × 10⁹, and the final pollutant load reduction achieved is 194.67 tonnes. The number of BMPs utilized for this purpose is 492,916. Classifications of BMPs are infiltration trench (349,228), infiltration basin (88,369), rain barrel (51,179), sand filter surface (1,623), vegetated filter strip (1,107), bioretention (603), grassed swales (570), porous pavement (196), wet pond (40), and constructed wetlands (1). The resultant runoff is 547.63 mm, and the runoff reduced is 55.03 mm. Equivalent runoff volume reduced is m³. Water thus saved serves GHMC for 11.53 days.

In total, the simulated runoff was higher for RCP 8.5 than for RCP 6.0. The simulation-optimization mechanism resulted in a 9.04 and 9.13% runoff reduction by placing 492,149 BMPs and 492,916 BMPs, respectively, for RCP 6.0 and RCP 8.5 (difference of 767 BMPs concerning RCP 6.0). This difference in the runoff reduction percentages may be attributed to their duration, affecting the rainfall-runoff simulations in HEC-HMS. The runoff reduced under RCP 8.5 (46.78 × 10⁷m³) is significantly higher than RCP 6.0 (33.29 × 10⁷m³). However, the number of days this stored runoff can be used increases slightly (10.33 to 11.53 days). This is due to the increased (projected) population of 2.420 × 10⁷ in 2064 (RCP 8.5) as compared to the projected population of 1.902 × 10⁷ in 2050 (RCP 6.0), which demands more water consumption per day. Tables S1–S3 are presented in the Supplementary material showing the number and percentage of different types of BMPs for 100% BA for historic, RCPs 6.0, and 8.5 scenarios in respect of individual zones and GHMC. This provides enormous information about the number of rooftop and non-rooftop BMPs for these scenarios.

Impacts of different combinations of cost and pollutant load reductions on runoff volume reduction are studied. Related parameters were analyzed both for historic and climate change scenarios, which are as follows:

Ten strategies were formulated for this purpose, namely A to J. A (10% BA, 0.34% TPL), B (20% BA, 0.68% TPL), C (30% BA, 1.02% TPL), D (40% BA, 1.36% TPL), E (50% BA, 1.7% TPL), F (60% BA, 2.05% TPL), G (70% BA, 2.39% TPL), H (80% BA, 2.73% TPL), I (90% BA, 3.07% TPL), J (95% BA, 3.24% TPL). Table 4 presents the related results.

Runoff reduction increased about seven times from 1.59 to 10.77% from strategies A to J (column 5 of Table 4). The number of BMPs increased from 99,278 (a total of 99,278 rooftop BMPs and 0 non-rooftop BMPs) to 495,165 (491,368 rooftop BMPs and 3,797 non-rooftop BMPs) (columns 8–10 of Table 4). Water demands of the entire population of GHMC are met for 1.68–11.36 days. 14.57 × 10⁶ and 15.31 × 10⁶ m³ of the total runoff reduction can be achieved, and 472,974 out of 495,165 of the BMPs get placed with strategy G (70% BA and 2.39% TPL). This demonstrates the advantage of using BLIP to place BMPs where budgets can be efficiently utilized.

The rain barrel was in the order of minimum preference among rooftop BMP placements. It is due to the inability of a rain barrel for pollutant load reduction, which gives it the minimum weightage. This enlightens us by indicating that the more the pollutant reduction required in the constraints, the lower the number of rain barrels. The Supplementary material section mentions the detailed classification of BMPs from strategy A to J (refer to Section S3). Optimization results tend to follow the law of diminishing marginal utility from the moment the non-rooftop BMPs get placed. This is due to the impact of runoff reduction efficiency, pollutant reduction efficiency, depth, and a number of the BMPs, which play a significant role during optimization.

Effects of stabilization become evident from strategy ‘F’ (60% BA and 2.05% TPL). This is because non-rooftop BMPs, which are relatively costly, come into existence. Even among the non-rooftop BMPs, the ones with lower cost are preferred. This is why about 94.5% of the total runoff reduction and 95.4% of the BMPs placement can be achieved with strategy G. Zones 12, 13, and 15 cover the significant proportion of the BMPs. The range of areas from A to J are 3.31–18.55 km² (zone 12), 1.96–12.86 km² (zone 13), and 1.93–11.8 km² (zone 15). This is due to their respective zone areas, which constrained the total open space available for BMP placement. Non-rooftop BMPs confirm the existence of stabilization. BMPs with a higher proportion of non-rooftop BMPs tend to show stabilization sooner than zones with a lesser ratio.

Values of these constraints (columns 2 and 3 of Table 5) are chosen at random based on the maximum budget values and pollutant load reduction for the optimization process. Table 5 presents results for extreme future rainfall events under different BA and TPL constraints. From Table 5, it is observed (respectively, for RCPs 6.0 and 8.5 and 10–70% BA and different pollutant load targets) that:

It was found from Table 5 that in case for RCP 6.0, runoff reduced is 36.71, 22.28, and 5.71 mm for 70, 40, and 10% BA and 2.16,1.23, and 0.31% pollutant load reduction. The number of BMPs employed for this purpose are 470,049; 288,807; and 107,417. Water saved in this process serves the catchment for 9.8, 5.95, and 1.52 days, respectively. Similarly for RCP 8.5, runoff reduced is 52.16, 31.67, and 8.14 mm for 70, 40, and 10% BA and 2.19,1.25, and 0.31% pollutant load reduction. The number of BMPs employed for this purpose are 470,381; 291,652; and 109,403. Water saved in this process serves the catchment for 10.93, 6.64, and 1.71 days.

The number of BMPs placed for a given budget showed much less variation. Pollutant load reduction achieved can be found to increase from 29.01 to 125.97 tonnes and 43.26 to 153.79 tonnes (column 5 of Table 5). This reduction in pollutant load by placing BMPs helped preserve various lake bodies and reduce the load on Sewage Treatment Plants.

Urban density (the number of people inhabiting a given urbanized area) will increase in the coming years, increasing building, road, and railway line densities. It indicates that infiltration trenches, basins, rain barrels, porous pavements, and grassed swales will increase. On the other hand, open spaces and lakes may tend to reduce. Accordingly, vegetated filter strips, constructed wetland, wet pond, and bioretention may reduce. These urban sprawl results may affect the already placed BMPs of vegetated filter strips, constructed wetland, wet ponds, and bioretention, necessitating discarding them.

Hydrological model HEC-HMS, siting tool EPA-SUSTAIN, a two-stage screening procedure, and optimization model BLIP are coupled to develop a simulation-optimization framework that facilitates runoff reduction. Optimal utilization of all buildings and open spaces in an urban catchment that facilitates conservation of water following the guidelines and policies of the relevant agencies is also considered. Greater Hyderabad Municipal Corporation, India, is chosen as the case study for demonstration. Three extreme rainfall events considered are historic, RCP 6.0 and RCP 8.5.

We studied the impact of two situations. In the first situation, it is pre-BMP implementation (only HEC-HMS for computation of runoff). The second situation is post-BMP implementation (coupling of HEC-HMS, EPA-SUSTAIN, and BLIP). Runoff reduction is achieved because of the fundamental tendency of BMPs (having relatively high permeability) to allow water to permeate through it into the ground. The extent of permeability is determined by the runoff reduction efficiency, while pollutant trapping potential is a function of the pollutant reduction efficiency. The comparison of post- and pre-BMP implementation helped to understand the impact of BMPs on runoff.

In the historic extreme event, rainfall is 215.9 mm, and runoff is 198.76 mm (pre-placement of BMPs). However, after the placement of BMPs (after EPA-SUSTAIN and BLIP), runoff at the outlet is 177.22 mm, showing a reduction of 21.54 mm. 495,607 BMPs occupying 76.99 km² are facilitated for this purpose. Similarly, for RCPs 6.0 and 8.5, extreme rainfalls are 440.35 and 624.2 mm, and consequently, the runoff produced is 428.35 and 602.65 mm. Runoff after placement of 492,149 and 492,916 BMPs (occupying 76.73 and 77.09 km²) in RCPs 6.0 and 8.5 is 389.63 and 547.62 mm, respectively, for the situation of 100% BA and 3.05% (for RCP 6.0) and 3.09% (for RCP 8.5) pollutant load reduction. Corresponding runoff reduced due to these BMPs is 38.72 and 55.03 mm and water saved in this process could meet the water demands of GHMC for 10.33 and 11.53 days.

The research work presented in this paper is helpful to society as maximum runoff reduction and pollutant load reduction are needed. Runoff captured by implementing BMPs will help satisfy the drinking water requirements of the community though partially. In addition, pollutant reduction can minimize health hazards in this process. We believe that this simultaneous management of both quality and quantity of water and related further research undoubtedly improve quality of life. Another potential of the work is its interdisciplinary nature where researchers from chemistry, social sciences, and biology can also participate. In addition, potential areas researchers can explore are sizing of BMPs, potable water efficiency, considering many other pollutants and future land availability projections into the existing modeling framework.

Keeping in view the advantages of BMP placement, an immediate action plan can be implemented for betterment of Hyderabad. Rainwater harvesting structures in a building can be made a mandatory requirement while planning the construction of buildings. It can be followed by using available open spaces. We also hope it will help the policymakers in reliable and easy decision-making.

This work is supported by Information Technology Research Academy (ITRA), Government of India under ITRA-water grant ITRA/15(68)/water/IUFM/01. The third author would like to acknowledge the funding support provided by the Council of Scientific and Industrial Research (CSIR), New Delhi, through a project with reference number 22(0782)/19/EMR-II dated 24.7.19. Special acknowledgments go to Prof D. Nagesh Kumar, Indian Institute of Science, Bangalore, for providing valuable guidance during the revision of the manuscript and critical suggestions. We also thank the editor and two anonymous reviewers for their valuable inputs. The authors thank GHMC officials and other agencies for providing all the support.

The authors declare that they have no conflict of interests.

Data cannot be made publicly available; readers should contact the corresponding author for details.