The need to enhance the resilience of urban drainage systems (UDSs) in view of emerging global climate change and urbanisation threats is well recognised. Blue-Green Infrastructure (BGI) provides a suitable strategy for building the resilience of existing UDSs. However, there are limited quantitative studies that provide evidence of their effectiveness for increased uptake in cities. In this research, coupled one-dimensional–two-dimensional (1D–2D) modelling is applied to assess the effectiveness of BGI that include rainwater harvesting systems, infiltration trenches, bioretention cells, and detention ponds using two case study UDSs located in Kampala that experience catastrophic pluvial flooding caused by extreme rainfall. The resulting flooding impacts are quantified considering ‘failed’ and ‘non-failed’ UDS initial states, using total flood volume and average flood duration as system performance indicators. The study results suggest that spatially distributed rainwater harvesting systems singularly lead to a reduction in total flood volume and average flood duration of 16–45% and 18–24% in the case study UDSs, respectively. Furthermore, the study results suggest that BGIs are more effective during moderate rainfall (T < 10 years). Based on the study findings, city scale implementation of multifunctional rainwater harvesting systems is recommended as a suitable strategy for enhancing UDSs’ resilience.

  • Coupled 1D–2D modelling considering system failures in Urban Drainage Systems (UDSs) applied to assess effectiveness of Blue-Green Infrastructure (BGI).

  • Flooding impacts quantified for ‘failed’ and ‘non-failed’ UDSs’ initial states.

  • Results confirmed that BGIs are more effective during moderate rainfall.

  • Multifunctional rainwater harvesting systems provide a promising resilience enhancing option.

Enhancing the resilience of urban drainage systems (UDSs) to flooding in view of emerging climate change and urbanisation threats is a subject of current research (Mugume et al. 2017; Webber et al. 2020; Rodriguez et al. 2023). UDSs comprise two distinct but interconnected systems that include the minor system, a complex network of pipes (sewers), open channels, culverts, inlets, manholes, storage tanks, overflows, and equipment such as pumping stations that are constructed for collection, transport of flows, and disposal of stormwater, waste water, or combined flows to waste water treatment plants or receiving waters, and the major system which comprises preferential pathways such as roads, foot paths, natural ground depressions, and small water courses that create a surface flow network during flooding conditions (Maksimović et al. 2009; Digman et al. 2014; Butler et al. 2018).

The general resilience of an urban drainage system can potentially be enhanced by increasing the flexibility or redundancy attributes of the system during design or retrofit (Mugume et al. 2015a). In this research, the definition and interpretation of resilience in engineering systems is considered and is defined as the ‘the degree to which the system minimises the level of service failure magnitude and duration over its design life when subject to exceptional loading conditions’ (Butler et al. 2016). Specifically, resilience is interpreted as a measure of how the system performs when subjected to unexpected threats that exceed design conditions that curtail its ability to meet the required level of service (Minsker et al. 2015; Butler et al. 2016). On this basis, its postulated that by quantifying the resulting flood reduction benefits of a given adaptation strategy, when the urban drainage system is subjected to exceptional loading conditions that lead to failure, the specific resilience can be quantified.

Urban drainage system failures that lead to flooding are categorised in this research as either functional or structural (component) failure. Functional failures are caused by hydraulic overloading of the system and can be caused by occurrence of extreme rainfall, urban creep and expansion, increased dry weather flows, or excessive infiltration (Mugume & Butler 2017; Guptha et al. 2022; Mcdonnell & Motta 2022). Structural failures conversely are caused by malfunction of system components and include gully pot and sewer blockages, sewer collapse, pump and sensor failures, deposited sediments, and solid waste (Mugume et al. 2015b; Pleau et al. 2022; Rosin et al. 2022).

Mugume et al. (2015b) developed and applied the middle-based Global Resilience Analysis (GRA) approach to systematically evaluate the resilience of an existing UDS in Kampala when subjected to a wide range of structural failure scenarios resulting from random cumulative link (pipe) failure. A resilience index that combined the total flood volume and average flood duration at each link failure level into a single matrix was developed and applied to quantify system residual functionality. In a related study, the GRA approach was applied to evaluate the performance of the UDS when subjected to a wide range of random functional failure scenarios caused by extreme rainfall (Mugume & Butler 2017). The study results suggested that UDSs’ functional resilience is significantly influenced by extreme rainfall intensity and spatial variability of rainfall. More recently, Panos et al. (2021) applied a one-dimensional–one-dimensional (1D–1D) dual drainage modelling and ‘tipping point’ resilience assessment approach to quantify the increase in flows (caused by climate and land-use change) that would be required to cause a UDS to exceed regulatory flooding standards and to assess the number and size of bioretention cells that would be required to increase the system's ability to convey the increased flows.

In another recent study, Rodriguez et al. (2023) applied the GRA approach coupled with the use of continuous simulations to assess the performance of combined sewer systems (CSSs) and to assess the effectiveness of BGI options in reducing combined sewer over flows (CSOs) using a case study of Fehraltorf, Switzerland. Furthermore, Guptha et al. (2022) extended the GRA approach to evaluate the effect of implementing BGI options (infiltration trenches and retention ponds) on the enhancement of resilience of an existing UDS in Gurugram City, India, when subjected to combinations of structural and functional failures. However, these studies did not utilise coupled one-dimensional–two-dimensional (1D–2D) modelling approaches for improved simulation of surface flows during flooding conditions.

Blue-Green Infrastructure (BGI) provide promising strategies for adaptation of existing grey UDSs to enhance their resilience to flooding (Almaaitah et al. 2021). BGIs are conceptualised, designed, and implemented not only to reduce urban runoff volumes and peak flows (leading to general increase in the redundant capacity in existing UDSs), but also to provide multiple benefits such as provision of alternative water supplies, urban greening, mitigation of urban heat island effect, recreation, urban agriculture, and improvement of liveability in cities (Sörensen & Emilsson 2019; Green et al. 2021).

Various studies have been conducted to assess the resilience benefits of BGIs. Webber et al. (2020) applied a cellular automata-based rapid scenario screening framework to assess the flood reduction potential of Green Infrastructure (GI) strategies using an urban catchment in Melbourne City Centre (Australia). Ortega Sandoval et al. (2023) applied highly discretised 1D and 1D–2D models to assess the performance of rainwater harvesting (RWH) systems, tree pits, storage tanks, green roofs, and pervious pavements in reducing flooding in Bogotá, Colombia. Sörensen & Emilsson (2019) used empirical insurance data to evaluate the effectiveness of an existing BGI that was retrofitted in the Augustenborg area in Sweden. Rodriguez et al. (2021) investigated the effect of spatial location of GI within a catchment on the resilience of UDSs using exploratory spatial data analysis (ESDA). Conversely, Goncalves et al. (2018) evaluated the performance of BGI in reducing flood risk in a coastal region of Southern Brazil that experiences high-intensity rainfall. The study results suggested reductions of total flood volume of between 30 and 75%, which were achieved through combination of centralised and decentralised options of detention ponds, infiltration trenches, and rain gardens. These studies concluded that BGI can reduce flooding caused by less extreme rainfall with low return periods, leaving residual flood risk from more extreme rainfall events.

However, very few studies have assessed the performance of BGI options using coupled 1D–2D modelling approaches, thereby neglecting critical interactions between the existing minor drainage system, manholes, and surcharging and the overland flow during flooding conditions (Haghighatafshar et al. 2018; Ortega Sandoval et al. 2023). Furthermore, the effectiveness of BGI in enhancing the global resilience of existing UDSs when subjected to structural failures that lead to urban flooding has not been extensively studied.

In addition, most computer modelling, experimental, or sensor monitoring studies based on existing BGI retrofit projects have been implemented in cities in Europe, Australia, North America, and China (Jefferson et al. 2017; Almaaitah et al. 2021; Fu et al. 2023). However, their adoption in developing country cities has been limited by insufficient evidence of their performance, lack of clear policies, guidelines, and standards, and limited benefits to the urban poor that reside in most low-lying flood prone areas (Kalantari et al. 2018; Wijesinghe & Thorn 2021; Thoms & Köster 2022). Because of these limitations, BGIs have only been implemented or extensively studied in very few cities of developing countries, such as Beira, Mozambique, and Windhoek, Namibia, with varying results (Simon & Duchhart 2015; Wijesinghe & Thorn 2021; Thoms & Köster 2022).

In this research, a methodology for the comparative evaluation of the effectiveness of various BGI options for reducing urban flooding impacts has been developed. It is demonstrated using two case study UDSs that drain the Kinawataka and Nalukolongo catchments in Kampala city, Uganda. The methodology entails development of a coupled 1D–2D model for simulation and characterisation of flooding in the two catchments considering ‘failed’ and ‘non-failed’ UDS initial system states under current climate change conditions. The failed initial UDS state represents internal system failures such as blockages, sediment deposition, and deposition of solid waste in existing UDSs, while the non-failed initial state represents a desired UDS condition, in which all the system components are well maintained and clean (Mugume et al. 2015b). The characterisation of flooding impacts was based on model simulation outputs comprising flooding extent, total flood volume, and average flood duration.

In this research, a new methodology that combined coupled one-dimensional–two-dimensional (1D–2D) urban drainage modelling, and consideration of structural and functional failures in UDSs was adopted. Structural failures were investigated through consideration of failed and non-failed UDS initial state conditions, while functional failures were investigated through simulation of UDS performance under varying rainfall intensities with return periods, T, of 2, 5, 10, 25, 50, and 100 years.

Coupled 1D–2D modelling approach

Coupled 1D–2D models provide a more accurate and realistic option for simulation of surface flooding through a network of 2D grid cells and can represent bidirectional flow between the 1D and the 2D domains using direct links or weir/orifice type elements (Leandro et al. 2009; Maksimović et al. 2009; Digman et al. 2014; Ortega Sandoval et al. 2023).

The available coupled 1D–2D modelling software that allow for simulation of BGI options include InfoWorks ICM and CS, MUSIC, WINDES, KalypsoHydrology, PCSWMM, and MIKEURBAN among others (Hellmers et al. 2018). In this research, urban drainage modelling was undertaken using PCSWMM v7.5.3406, a physically based coupled 1D–2D model developed by Computation Hydraulics International Ontario, Canada (Chitwatkulsiri et al. 2022; Manchikatla & Umamahesh 2022; Ortega Sandoval et al. 2023). PCSWMM utilises the US EPA SWMM 5.0.015 hydrology and hydraulics engine that is primarily developed for modelling of water quantity and quality in urban areas for both short- and long-term simulation runs (Rossman 2016). The key components and features that require representation in the model include urban hydrological modelling, 1D-hydraulic modelling, and the coupled 1D–2D modelling.

In the 1D model, each sub-catchment is represented as a nonlinear reservoir. Inflows are generated by rainfall or any user-defined upstream flow, while losses are caused by infiltration or evapotranspiration, resulting into a net surface runoff when the depth of water in the reservoir exceeds the maximum depression storage (Rossman 2016; Ortega Sandoval et al. 2023). The conveyance portion of the 1D drainage system is modelled with a network of nodes and links and is based on a solution of the full dynamic wave (Saint-Venant) equations (Butler et al. 2018). Nodes are points that represent simple junctions, flow dividers, storage units, or outfalls. Links connect nodes to one another with conduits (pipes and channels), pumps, or flow regulators (orifices, weirs, or outlets) (Rossman 2016).

By contrast, the 2D domain is discretised as a set of nodes (2D junctions) and open conduits without walls (2D conduits), constituting a 2D mesh that captures the ‘height characteristics’ of urban catchments based on topographical information extracted from a Digital Elevation Model (DEM) (Leandro et al. 2009; Maksimović et al. 2009; Butler et al. 2018). The free surface flow between the 2D junctions and the 2D conduits is based on the solution of 1D Saint-Venant flow equations using the finite-element method to compute water level and velocity values (Butler et al. 2018). Furthermore, a simultaneous transition (coupling) between the 1D model and 2D surface flow model is enabled by the connectivity of the link-node system (inlet control bottom orifices or direct connections) (Leandro et al. 2009; Maksimović et al. 2009).

Modelling of failed and non-failed UDS initial states

Two UDSs’ baseline asset condition states were investigated: failed and non-failed UDSs. The failed UDS initial state represents an existing UDS that is not well maintained and in which all the links have experienced partial or full structural failure, which could be caused by structural collapse of concrete or stone pitched channel sections, blockages, deposited sediments, or solid waste, that collectively constrain the hydraulic conveyance capacity of the existing UDSs. In this UDS initial state, Manning's n value of 100 was used for all conduits (links) (Mugume et al. 2015b). This UDS initial state condition represents the asset condition of UDSs in most cities in developing countries with poor structural conditions and which are fully or partially filled with sediments and solid waste. In this study, we considered full (complete) failure of all conduits (links) in the UDS to provide an upper limit of the maximum possible urban drainage flooding caused by structural failure of all conduits. The non-failed UDS initial state represents the desired UDS condition, in which all the links are well maintained and clean, in good structural condition, devoid of sediments or solid waste, and with the design hydraulic conveyance capacity. Manning's n value of 0.020 was used to model all the links in the non-failed system (Mugume et al. 2015b). Using the approach of modelling, the failed and non-failed UDS initial states enable quantitative evaluation of the effectiveness of UDS cleaning, repair, and maintenance strategies, which can effectively complement the benefits of BGI. The resulting flooding impacts were quantified using parameters that included flooded area, total flood volume, and average nodal flood duration.

Case study 1D–2D model setup

Data sets used in the study

The models of the existing UDSs that drain the Kinawataka and Nalukolongo catchments were built using PCSWMM. The key data sets such as observed daily rainfall data, Digital Elevation Models, satellite imagery, and urban drainage network data that were used to build the urban drainage models are described in Table 1.

Table 1

Details of various types of data used for developing the 1D–2D urban drainage models

Data setData qualityData period or year of publicationSourcePurpose
77-year daily rainfall for Makerere University rain gauge station Daily rainfall (mm) 1943–2019 Uganda National Meteorological Authority Rainfall database Extreme rainfall analysis, development of IDF curves and rainstorms 
Shuttle Radar Topography Mission (SRTM) Digital Elevation Model 30 m × 30 m spatial resolution 1st August 2022 United States Geological Surveys (USGS) SRTM Digital Elevation data for identifying the areas prone to flooding 
Sentinel-2 satellite image, Satellite image 10 m × 10 m spatial resolution 2nd August 2023 Base map in ArcGIS GIS-based land-use and land-cover analysis imperviousness (PIMP) Visualisation of flooding extent 
Existing urban drainage network data (open channels, cross culverts, storage units, pipes, inlets) Urban drainage network parameters (bottom width, top width, side slopes), lengths of channel sections, invert levels, slopes, cross culvert diameters, outfall dimensions) 2016 and 2022 Kampala Capital City Authority (KCCA) reports and as-built drawings 1D urban drainage model build 
Data setData qualityData period or year of publicationSourcePurpose
77-year daily rainfall for Makerere University rain gauge station Daily rainfall (mm) 1943–2019 Uganda National Meteorological Authority Rainfall database Extreme rainfall analysis, development of IDF curves and rainstorms 
Shuttle Radar Topography Mission (SRTM) Digital Elevation Model 30 m × 30 m spatial resolution 1st August 2022 United States Geological Surveys (USGS) SRTM Digital Elevation data for identifying the areas prone to flooding 
Sentinel-2 satellite image, Satellite image 10 m × 10 m spatial resolution 2nd August 2023 Base map in ArcGIS GIS-based land-use and land-cover analysis imperviousness (PIMP) Visualisation of flooding extent 
Existing urban drainage network data (open channels, cross culverts, storage units, pipes, inlets) Urban drainage network parameters (bottom width, top width, side slopes), lengths of channel sections, invert levels, slopes, cross culvert diameters, outfall dimensions) 2016 and 2022 Kampala Capital City Authority (KCCA) reports and as-built drawings 1D urban drainage model build 

Extreme rainfall analysis

In this research, a 77-year observed daily rainfall record for Makerere University for the period 1943–2019 was obtained and utilised. The maximum daily rainfall for each month and year was extracted from the record. The maximum daily rainfall ranges from 54.4 mm during the driest month of January to 107.5 mm during the wettest month of May (Figure 1).
Figure 1

Observed maximum daily rainfall for Kampala City for the period 1943–2019.

Figure 1

Observed maximum daily rainfall for Kampala City for the period 1943–2019.

Close modal

The Intensity–Duration–Frequency (IDF) curves were derived from the observed daily rainfall data using the Annual Maximum Series (AMS) method. Based on the analysis, the 2-year 24-h rainfall total for Kampala was determined as 64.8 mm. To account for the effect of the current climate change on the observed extreme rainfall, a change factor of 1.2 was applied to upscale the 2-year 24-h total (Mugume et al. 2013). A detailed description of the methodology for the derivation of IDF curves is presented in Mugume & Butler (2017).

Based on the IDF curves, T2, T5, T10, T25, T50, and T100 design storms for the Nalukolongo (Figure 2) and Kinawataka (Figure 3) catchments were constructed using the alternating block method (Balbastre-Soldevila et al. 2019). The design storms formed the basis for the hydraulic loading of the coupled 1D–2D models developed using PCSWMM.
Figure 2

Extracted design storm profiles for Nalukolongo catchment.

Figure 2

Extracted design storm profiles for Nalukolongo catchment.

Close modal
Figure 3

Extracted design storm profiles for Kinawataka catchment.

Figure 3

Extracted design storm profiles for Kinawataka catchment.

Close modal

Rainstorms with a fixed duration of 30 h and return periods, T, of 2, 5, 10, 25, 50, and 100 years were used in the model (hereafter referred to as T2, T5, T10, T25, T50, and T100, respectively). The rainstorms were applied uniformly across all the sub-catchments within each catchment.

1D Description of existing Kinawataka and Nalukolongo UDSs

The existing UDSs in Kampala currently experience frequent pluvial flooding events and therefore require urgent rehabilitation. More so, the existing Kinawataka and Nalukolongo UDSs (Figure 4) have experienced multiple structural failures caused by the collapse of the lined open channel sections, scour holes, sediment, and solid waste deposition (GIZ 2022a). The Kinawataka catchment is highly urbanised with mixed developments that include high-income residential areas on Naguru Ntinda and Kyambogo Industrial Areas, Kyambogo University, low- to medium-income areas of Bweyogerere and Kireka, and low urban poor areas located downstream of the catchment. Conversely, the Nalukolongo catchment is highly urbanised with mixed developments that include the industrial areas of Nalukolongo, highly urbanised commercial areas of Kibuye, low- and medium-income residential areas of Rubaga, Ndejje, and Bunamwaya, and the downstream low-income areas of Masanafu, Natete, and Kyengera. The two catchments were selected for comparative analysis because Nalukolongo catchment (7.5% [3.4–17.3%]) is steeper than Kinawataka (5.9% [0.5–14.7%]). Furthermore, Nalukolongo catchment has many low- to medium-income households with smaller plot sizes, while Kinawataka has considerable medium- to high-income households and institutions with relatively large plot sizes in the upstream catchment areas of Naguru, Ntinda, and Kyambogo and which could be suitable for the retrofitting of the BGI.
Figure 4

Location of Nalukolongo and Kinawataka catchments in Kampala.

Figure 4

Location of Nalukolongo and Kinawataka catchments in Kampala.

Close modal
Figure 5 shows the selected flood hotspot locations in the Nalukolongo and Kinawataka catchments during the scoping study. The flooding hotpots were identified through a review of the Kampala Capital City Authority (KCCA) flood reports for the period 2014–2022, field visits to the specific locations specified in the flood reports after occurrence of rainfall events, and geolocation of each flooding hotspot using GPS mobile app and mapping (GIZ 2022a, 2022b).
Figure 5

Location of selected flooding hotspots in (a) Kinawataka and (b) Nalukolongo catchments in Kampala.

Figure 5

Location of selected flooding hotspots in (a) Kinawataka and (b) Nalukolongo catchments in Kampala.

Close modal

1D Model setup

The existing primary and secondary drainage systems in both catchments consists of a network of pipes, trapezoidal open channel sections, and culvert crossings constructed using stone pitching and reinforced concrete.

The 1D urban drainage model was setup using the case study specific hydrological and hydraulic data sets (Table 1). GIS-based spatial analysis was applied to delineate the catchment into sub-catchments and to pre-process key sub-catchment-specific information that included sub-catchment area, slopes, and widths. Furthermore, GIS spatial analysis of satellite imagery was applied to compute the percentage imperviousness (PIMP) for both catchments using the iterative supervised classification method (Han & Burian 2009). The average sub-catchment slope was computed as 5.9% [0.5–14.7%] and 7.5% [3.4–17.3%] for the Kinawataka and Nalukolongo catchments, respectively. The PIMP was computed at 69.6 and 69.5% for Kinawataka and Nalukolongo catchments, respectively (Figure 6) using GIS spatial analysis of satellite imagery.
Figure 6

Land-use and land-cover in Nalukolongo and Kinawataka catchments.

Figure 6

Land-use and land-cover in Nalukolongo and Kinawataka catchments.

Close modal

In addition, the hydraulic data were obtained from existing KCCA drainage reports as-built drawings and the recently concluded Greater Kampala Integrated Flood Resilience study (KCCA 2016; GIZ 2022a). The key data sets that were collected included open channel cross-section dimensions (bottom widths, top widths, side slopes), lengths of channel sections, and invert levels. Manning's roughness coefficient was assigned based on two UDS initial states, that is: failed (n = 100) and non-failed (n = 0.020). Additional information was also obtained on culvert crossings and outfalls. The modelled 30.8 km long Kinawataka UDS comprises 54 conduits and 44 nodes, and drains into Lake Victoria. The gradients of the open channel sections range from 0.063 to 9.673%. The modelled UDS drains a total area of 32.8 km2, which was delineated into 37 sub-catchments. Conversely, the modelled 7.1 km long Nalukolongo UDS comprises 22 conduits and 22 nodes, and discharges into the River Mayanja, a tributary of River Kafu, which subsequently feeds into the Victoria Nile. The gradients of the open channel sections range from 0.088 to 0.991%. The modelled UDS drains a total area of 33.3 km2 delineated into 22 sub-catchments.

Coupled 1D–2D model setup

In this research, the 1D–2D model was created through a series of steps that entailed delineation of a bounding layer, 2D node generation, and 2D mesh creation. The bounding layer defines the extent of the 2D model and consists of multiple polygons to represent sub-area roughness and infiltration parameters (Ortega Sandoval et al. 2023). The 1D model was connected to the created 2D model, and boundary conditions were assigned (Figure 7).
Figure 7

1D–2D model setup.

Figure 7

1D–2D model setup.

Close modal

The bounding layer was setup with a 30 m resolution hexagonal mesh and a uniform Manning's roughness coefficient of 0.030. Considering the large urban catchment areas (33.1 and 32.7 km2 for the Nalukolongo and Kinawataka catchments, respectively) with relatively high PIMP levels (approximately 70%), a uniform roughness coefficient was applied to simplify the 2D modelling process and to reduce the computational resource requirements for the 2D modelling. In addition, obstructions were excluded from the 2D model (buildings, roads, pavements, and sidewalks). The 2D node layer was generated using the DEM elevation data to represent the floodplain topography.

For 1D–2D simulation, each junction in the 1D model was connected directly to the closest 2D junction point, using the connecting tool in PCSWMM and thereby allowing the 2D nodes to be connected through the 2D conduits. For the Kinawataka catchment, the 2D model was composed of 83,297 2D junctions and 83,297 2D cells. For the Nalukolongo catchment, the 2D model was composed of 127,332 2D junctions and 127,332 of 2D cells. This connection allows for the free transfer of flow from the 1D drainage model to the 2D model for flood extents in areas. However, this approach excluded the flow from the 2D areas back to the 1D network (Leandro & Martins 2016). In addition, flow modelling of obstructions was excluded, because of the large size of the discretised 2D urban catchment area and the limitation of the available computational resources. Despite these limitations, the proposed approach presents an efficient approach for modelling the resulting flooded area and depths with moderate computational requirements. In this research, we used a computer with 1 TB Hard disk space, Intel(R) Core (TM) i5–9500 CPU @ 3.00 GHz 3.00, 16.0 GB RAM, 64-bit operating system, x64-based processor, and Windows 11 Pro. Figure 8 illustrates the adopted coupled 1D–2D modelling approach.
Figure 8

Adopted coupled 1D–2D modelling approach.

Figure 8

Adopted coupled 1D–2D modelling approach.

Close modal

Local sensitivity analysis

The calibration of urban drainage models is constrained by the lack of simultaneously measured rainfall, discharge, and data from actual flood events (de Vitry & Leitão 2020). As a result, alternative sources of data such as citizen science, social media, and surveillance cameras have been proposed (de Vitry & Leitão 2020). Furthermore, in the absence of such information, emphasis can be placed on improving model accuracy through improved description of flow paths, accurate estimation of land-use and land-cover parameters, and use of plausible friction factors so as to minimise uncertainties in flood model predictions (Hunter et al. 2008). In this research, the One factor at a Time (OAT) local sensitivity analysis was undertaken to identify the most sensitive model parameters and to ensure that they are computed accurately (Sweetapple et al. 2014).

Model validation

In addition, model validation (de Vitry & Leitão 2020) was undertaken using field measurement of the observed maximum flood depths at selected flood hotspot locations in the case study catchments. A total of 30 critical flood locations were selected based on the flood extent maps generated using the coupled 1D–2D modelling approach. The measurement of observed maximum flood depths based on identified high water (flood) marks on existing buildings was undertaken during field surveys. Secondly, georeferencing of the selected flood affected properties was undertaken. Such an approach was applied in a recent study in which post-storm surveys on reported flooding impacts were undertaken and used to validate the flood modelling results (Perini et al. 2016; Mugume et al. 2024).

The maximum flood depth was determined as the distance between the existing ground level and the highest water mark on a given flood affected building. Furthermore, the coordinates of each surveyed building were obtained using a mobile phone GPS application (Mugume et al. 2024). A total of 30 flood affected properties/locations were georeferenced and the observed maximum flood depth measured (Figure 5).

The measured maximum flood depth was compared to the modelled flood depth at each of the locations. The corresponding simulated flood depth at each of the locations was extracted from the flood depth map and the results compared. The Nash–Sutcliffe Efficiency (NSE) and Root Mean Square Error (RMSE) parameters were applied to evaluate the goodness of fit between the observed and simulated flood depths.

The NSE Index values range from negative infinity to 1 (Equation (1)). Values between 0 and 1 are generally acceptable, with NSE = 0.65 considered as unsatisfactory; NSE = 0.80 as acceptable; NSE = 0.90 as good; and NSE > 0.90 as very good (Gupta & Kling 2011; Ritter & Muñoz-Carpena 2013):
formula
(1)
where is the mean of the observed flood depths; dm is the simulated flood depth; and , the observed flood depth at time t.
Conversely, for the RMSE parameter (Equation (2)), the lower the value of the RMSE, the higher the accuracy of the prediction (Gupta & Kling 2011; Seenath et al. 2016).
formula
(2)
where do is the observed flood depth, ds is the simulated flood depth, and N is the number of observations.

Modelling of BGI options

BGI option selection

Stakeholder consultations were undertaken between June and December 2022 as part of the Greater Kampala Integrated Flood Resilience Partnership Project (GIZ 2022a) to determine the most context-appropriate BGI options for Kampala. The key stakeholders were mainly drawn from various agencies involved in urban drainage and flood risk management, such as the KCCA, the Ministry of Water and Environment (MWE), Ministry of Lands, Housing, and Urban Development, Uganda National Meteorological Authority (UNMA), National Environmental Management Authority (NEMA), and the Uganda Manufacturer's Association (UMA). As part of the stakeholder engagement process, two in-person workshops were held in Kampala from 31 March 2022 to 1 April 2022 and 4 August 2022.

During the first workshop, the stakeholders undertook a preliminary selection of potential BGIs from a long list of well-described and graphically illustrated options. The BGI options that were presented to stakeholders mainly included green roofs, urban forests, urban wetlands, bioswales, infiltration strips, infiltration trenches, permeable pavements, rain gardens, detention ponds, bioretention cells, water squares, urban green spaces, and RWH systems. Furthermore, during the second workshop, stakeholders further reviewed and prioritised the BGI options for specific flooding hotspots using criteria that included technical efficiency, investment costs, land availability, ease of maintenance, and social acceptability. Based on the outcomes of the stakeholder consultations, four BGI options were selected for detailed modelling and analysis. These included RWH tanks, detention ponds, infiltration trenches, and bioretention cells (Mugume et al. 2023).

Modelling of BGI options in PCSWMM

The various BGI options modelled, such as RWH, detention ponds, infiltration trenches, and bioretention cells, represented a spatial coverage of between 0.4 and 1.5% of the catchment area.

The modelled BGI options were implemented in the 1D model using the Low-Impact Development (LID) control tool in PCSWMM. The LID controls were represented by a combination of vertical layers whose properties were defined for each single unit. Bioretention cells and infiltration trenches were placed in the sub-catchments and set to treat 60 and 20% of impervious (pavements, roads, roofs, and paths) and pervious areas, respectively. Conversely, the RWH tanks were modelled to capture runoff primarily from roofs.

The proposed BGI options were modelled individually, while maintaining the same total storage volume to enable a proof of concept of the comparative evaluation of their effectiveness with respect to reducing resulting flooding impacts. The spatial coverage of RWH was based on the analysis of the number of households within the catchment. The analysis of the population within each catchment was based on the published official population data (UBOS 2016). Using the assessed population data and by assuming an average household size of four persons in Kampala (Mugume et al. 2017) and considering that 70% of the households install a RWH tank, the total volume allocated to RWH was determined as 274,670 and 300,804 m3 for the Kinawataka and Nalukolongo catchments, respectively (Supplementary material, Table A1). Various model simulations were run for both baseline and the proposed BGI options to test their effectiveness in reducing flooding. To enable a comparative evaluation of the performance of each BGI option, the same retained volume was specified for each BGI option. The total number of units of each BGI option is presented in Supplementary material, Tables A2 and A3 for the Kinawataka and Nalukolongo catchments, respectively.

For the Kinawataka catchment, a total of 53,934 RWH units with a total storage volume of 274,670 m3 were modelled in PCSWMM as ‘rain barrels,’ which were spatially distributed in all the sub-catchments based on the area covered by each sub-catchment. Conversely, a total of 60,161 RWH units were considered for the Nalukolongo catchment. The RWH units were assumed to be empty prior to the start of the rainfall event. A drain delay of 6 h was adopted for modelling purposes.

Furthermore, a total of 10 centralised detention ponds were modelled in PCWSMM, while maintaining the same total storage volume for each catchment. The inlet of each pond was modelled as an open channel that is connected to the link within the existing UDS immediately upstream of the location of the pond (and with the same channel cross-section dimensions). In addition, the outlet of each detention pond was modelled as an open channel that connects to a link within the existing UDS immediately downstream of the detention pond location (with the same channel cross-section dimensions as the upstream link).

Bioretention cells were modelled in PCSWWM, each with an area of 120 m2 and a volume of 72 m3. The hydraulic conductivity and porosity values of 70.8 mm/hour and 0.3 were applied in the model (Havik 2012). A total of 3,815 and 4,178 bioretention cells were spatially distributed in all sub-catchments in both the Kinawataka and Nalukolongo catchments, respectively.

Infiltration trenches were modelled in PCSWWM at a sub-catchment level. The hydraulic void ratio and seepage rates of 0.4 and 70.8 mm/hour were applied in the model (Havik 2012). A total of 3,052 and 3,342 infiltration trenches were spatially distributed in all sub-catchments in the Kinawataka and Nalukolongo catchments, respectively. Supplementary material, Appendix Tables A4, A5, A6, and A7 present the parameters and properties for the different BGI options that were modelled. A total of 100 simulations were run and the results were compared to determine the percentage reduction in the resulting flooding impacts. The average computational time for undertaking each simulation was between 3 and 4 h.

Cost–benefit analysis

The enhancement of resilience in UDSs by implementing various adaptation strategies cannot be achieved at any cost. It is therefore vital to investigate the trade-offs between operational performance benefits of implementing a given adaptation strategy and the attendant whole-life costs over the design life of the UDS. By taking into account the cost of failure that could arise from the inability of the system to deliver its expected or prescribed level of service, a more complete view of cost-effectiveness of the proposed adaptation strategies is gained (Alderson et al. 2015). In this research, discounted cost analysis is applied to evaluate net benefits achieved by implementing the proposed BGI options considering a design (service) life of 50 years. The discounted total cost, PVCT,y, of each option, y, is computed using Equations (3) and (4) (Mugume et al. 2015a). The approach of discounting enables the comparison of costs and benefits that occur in the different time periods (HM Treasury 2011).
formula
(3)
formula
(4)
where CT is the total cost (US$); r is the discount rate (r = 3.5% for initial 30 years, then r = 3.0%); ts is a given time period (service years) during a UDS's design life, j (years); Co is the Capital cost (US$); COM is the operation and maintenance (O&M) cost (US$); and CTF is the cost of failure (which was estimated by computing the direct tangible flooding costs in US$). The costs were estimated considering four options as follows:
  • (a)

    Non-failed UDS option: This option entails rehabilitation of the existing UDS through implementation of specific cleaning and maintenance activities that would change the existing system from a failed to the non-failed state

  • (b)

    BGI option 1: Implementation of RWH systems

  • (c)

    BGI option 2: Implementation of infiltration trenches

  • (d)

    BGI option 3: Implementation of bioretention cells

  • (e)

    BGI option 4: Implementation of centralised storage tanks

Capital costs

For each of the considered options, bills of quantities were prepared for each single unit. The unit construction cost for each BGI option was estimated using prevailing construction cost unit rates in Kampala for excavation, surface preparation, ground improvement, backfilling, false work, formwork and concrete finishes, steel reinforcement, concrete works, water proofing, landscaping, grassing, purchase of plastic RWH tanks, and plumbing, among others. Thereafter the cost of wide-scale implementation of each option was computed by taking into consideration the specific number of units specified in Supplementary material, Tables A2 and A3.

Operation and maintenance costs

The operation and maintenance (O&M) costs of UDSs range from 0.5 to 10% of the total construction costs (HR Wallingford 2004; Houle et al. 2013). In this study, O&M costs for the Baseline UDS option was computed assuming 5% of the total construction costs (HR Wallingford 2004; Houle et al. 2013). Conversely, the O&M costs for the BGI options 1, 2, 3, 4 were computed considering 2% of the total construction costs, because BGIs require reduced frequency of on-site inspections resulting from involvement of local communities in their management and have lower O&M costs when compared with the conventional grey UDSs (Green et al. 2021; Li et al. 2021). For conversion of costs from Uganda shillings (UGX) to United States dollars (US$), the Jan–July 2023 average exchange rate (1US$ = 3,700 UGX) is applied.

Direct tangible flooding costs

The direct tangible flooding costs associated with a given option, CTF, are computed as a function of total flood volume, flood damage cost per unit volume of flooding, and the probability of occurrence of flooding () during the considered UDS design life using Equation (5) and (6) (Mugume et al. 2015a). The flood volume is the total volume of flooding from the 1D urban drainage system, that is the total of the flood volumes simulated at each of the 1D nodes in the model.
formula
(5)
formula
(5)
where j is the design life of the UDS (years); VTF is the total flood volume (m3); fc is the unit direct tangible flooding cost (US$/m3 of flooding); is the probability of flood occurrence during a service life of t : tj (years); and T is the rainfall return period (years).
The unit direct tangible flooding cost, fc, for the Nalukolongo and Kinawataka catchments were based on the computed expected annual flood damages (EADs) for the Kinawataka and Nalukolongo catchments (GIZ 2022a) and included damage to building structure, contents, and damage to roads and railways. The EAD were computed based on the flood depth–damage curves developed for Kampala (World Bank 2020). In the study, the estimated EAD for 2020 were computed as US$ 8,688,772 and US$ 9,386,934 for the Kinawataka and Nalukolongo catchments, respectively. The unit direct tangible flooding cost, fc, was computed as
formula
(7)
where fc is the unit expected annual flood damage (US$/m3); EAD is the expected annual flood damage; af is the simulated flooded area, and df is the simulated average flood depth. In this research, the computed average flood depths, df,p, range from 0.64 to 0.67 m and the computed values of fc range from 4.1 to 4.9 US$/m3 of flooding.

Computation of net benefits

To convert all costs to ‘present values’, discount rates of 3.5 and 3.0% are applied to the first 30 years and the subsequent 20 years, respectively, as recommended in HM Treasury (2011). The difference between the computed total discounted costs for the ‘Non-failed’ option (PVCT,NF) and each tested BGI option, (PVCT,y) represents the net benefit, NB,y, attributed to the option. The net benefit NB,y (%), is expressed as percentage of PVCT,NF using Equation (8).
formula
(8)

Coupled 1D–2D modelling results for the existing UDSs

Local sensitivity analysis

The results of the One-Factor at a Time (OAT) sensitivity analysis suggested that the simulated total flood volume is less sensitive to changes in the considered model parameters for the failed UDS when compared with the non-failed UDS. In addition, the study results suggested that for the failed UDS initial state, the simulated total flooded area is less sensitive to changes in Manning's n value for the 2D cells [−8.1 to 9.7%]. Conversely, the corresponding results for the non-failed UDS initial state suggest a slight reduction in model sensitivity [−6.2 to 9.0%] when compared with the failed state

Furthermore, Manning's roughness coefficient, n, for the conduits and Percentage Imperviousness (PIMP) were identified as the most sensitive model parameters. The results of the OAT sensitivity analysis for the non-failed UDS suggested that PIMP [−30.9 to 19.4%] was more sensitive when compared with Manning's n [−13.8 to 10.3%]. In this research, therefore, emphasis was placed on the accurate estimation of the PIMP in the two catchments using GIS-based spatial analysis of the most recent satellite imagery. (Refer to Supplementary information, Section 3.1 for detailed description of OAT sensitivity analysis.)

Model validation

The computed NSE values for the Kinwataka and Nalukolongo UDSs were greater than 0.793 and 0.800, respectively, considering the T25 and T50 rainfall return periods. The results suggested that the measured maximum flood depths at the 30 locations were attributed to pluvial flooding events caused by extreme rainfall with return periods of between 25 and 50 years (Table 2 and Table 3). This result is also confirmed by the results of extreme rainfall analysis, which indicate that the observed maximum annual rainfall (107.5 mm recorded in 1991) was associated with a return period of 37 years. Conversely, the computed RMSE values were 0.263 and 0.142 (for T50 return period) for the Kinawataka and Nalukolongo UDSs, respectively. This suggests that the model prediction accuracy for Nalukolongo was slightly higher than that of the Kinawataka UDSs.

Table 2

Coupled 1D–2D flood model validation results for the Kinawataka catchment

 
 
Table 3

Coupled 1D–2D flood model validation results for the Nalukolongo catchment

 
 

Kinawataka UDS model results

The results of the coupled 1D–2D modelling for the existing Kinawataka UDS considering a failed state are presented in Figure 9. The flood extent maps show that flooding is concentrated around the primary channels. The simulated flooded area considering the T = 2 year rainstorm was 3.03 km2, and covers parts of the Ntinda Industrial area in Nakawa, Upper Kinawataka industrial area, Kyambogo (Banda), and Katogo slum downstream of the Kireka road railway crossing. The simulated flooding extent considering the T5, T10, and T25 rainstorms resulted into a flooding extent of 3.12, 3.44, and 3.66 km2, respectively. This represented increases of 2.9, 13.5, and 20.8%, respectively, when compared with the T2 event. The T50 and T100 events contributed to an increase in flooding extent of 23.1 and 38.3%, respectively, when compared with the T2 event. In all cases, the flooding is concentrated in the lower parts of the catchment. The corresponding simulation results considering a the non-failed state are presented in Figure 10. The results show that flooding extent minimally reduces by between 1.7 and 10.4% when compared with the corresponding results considering the failed UDS initial system state.
Figure 9

Simulated flooding extent (flooded area) for the existing Kinawataka UDS considering the failed initial state. Blue colour is the inundated area and the red mark is the outlet. Top row, Left to Right – T2, T10, Middle row, Left to Right – T25, T50, and Bottom row, Left to Right – T50, T100.

Figure 9

Simulated flooding extent (flooded area) for the existing Kinawataka UDS considering the failed initial state. Blue colour is the inundated area and the red mark is the outlet. Top row, Left to Right – T2, T10, Middle row, Left to Right – T25, T50, and Bottom row, Left to Right – T50, T100.

Close modal
Figure 10

Simulated flooding extent (flooded area) the existing Kinawataka UDS considering the non-failed initial state. Blue colour is the inundated area and the red mark is the outlet. Top row, Left to Right – T2, T10, Middle row, Left to Right – T25, T50, and Bottom row, Left to Right – T50, T100.

Figure 10

Simulated flooding extent (flooded area) the existing Kinawataka UDS considering the non-failed initial state. Blue colour is the inundated area and the red mark is the outlet. Top row, Left to Right – T2, T10, Middle row, Left to Right – T25, T50, and Bottom row, Left to Right – T50, T100.

Close modal

Nalukolongo UDS model results

The results of the coupled 1D–2D modelling for the existing Nalukolongo UDS considering a failed UDS initial state are presented in Figure 11. The flood extent maps show that flooding is concentrated around the primary channels and in the downstream parts of the catchment that include Wankulukuku, Sembule, Nalukolongo road, and Natete areas. The simulated flood area considering the T2 rainstorm was 3.02 km2. The T5, T10, and T25 rainstorms resulted in a flooding extent of 3.25, 4.21, and 4.54 km2, representing increases of 7.6, 39.4, and 50.3%, respectively, when compared with the T2 year event. The simulated flood extents considering the T50 and T100 events were 4.62 and 4.73 km2, which presented increases of 53.0 and 56.0%, respectively, when compared with the T2 event. The corresponding simulation results considering a non-failed state are presented in Figure 12. The results show that flooding extent minimally reduces by between 1.6 and 3.2% when compared with the corresponding results considering the failed UDS initial system state.
Figure 11

Simulated flooding extent (flooded area) for the existing Nalukolongo UDS considering a failed UDS initial state. Blue colour is the inundated area and the red mark is the outlet. Top row, Left to Right – T2, T10, Middle row, Left to Right – T25, T50, and Bottom row, Left to Right – T50, T100.

Figure 11

Simulated flooding extent (flooded area) for the existing Nalukolongo UDS considering a failed UDS initial state. Blue colour is the inundated area and the red mark is the outlet. Top row, Left to Right – T2, T10, Middle row, Left to Right – T25, T50, and Bottom row, Left to Right – T50, T100.

Close modal
Figure 12

Simulated flooding extent (flooded area) for the existing Nalukolongo UDS considering a non-failed UDS initial state. Blue colour is the inundated area and the red mark is the outlet. Top row, Left to Right – T2, T10, Middle row, Left to Right – T25, T50, and Bottom row, Left to Right – T50, T100.

Figure 12

Simulated flooding extent (flooded area) for the existing Nalukolongo UDS considering a non-failed UDS initial state. Blue colour is the inundated area and the red mark is the outlet. Top row, Left to Right – T2, T10, Middle row, Left to Right – T25, T50, and Bottom row, Left to Right – T50, T100.

Close modal

Effectiveness of modelled BGI options considering a failed UDS initial state

Kinawataka UDS

The results suggest that use of detention ponds led to a reduction in total flood volume 19% [6–28%] considering a failed UDS initial state. The use of bioretention cells led to the reduction of total flood volume of 22% [3–35%]. And the use of infiltration trenches and spatially distributed RWH tanks led to the highest reduction in total flood volume of 33% [7–56%] and 45% [15–57%], respectively (Figure 13). The results further indicated that RWH, infiltration trenches, and bioretention cells led to more than 13% reduction in the average flood duration. However, the use of detention ponds led to minimal reduction in total flood volume of 17% [6–28%] and average flood duration of 5% [1–8%].
Figure 13

BGI effect on reduction of (a) total flood volume (top) and average flood duration (bottom) for Kinawataka failed UDS.

Figure 13

BGI effect on reduction of (a) total flood volume (top) and average flood duration (bottom) for Kinawataka failed UDS.

Close modal

Nalukolongo UDS

The results suggest that the use of detention ponds led to a reduction in total flood volume of 7% [2–17%] considering a failed UDS initial state. Use of bioretention cells led to the reduction of total flood volume of 9% [6–13%] (Figure 14). The results further suggest that the use of infiltration trenches and spatially distributed RWH tanks led to the highest reduction in total flood volume of 14% [9–23%] and 16% [9–23%], respectively. Considering the duration of flooding, the results indicate that RWH, infiltration trenches, and bioretention cells led to more than 20% reduction in the average flood duration. However, the use of detention ponds led to minimal reduction in the average flood duration of 7% [3–12%] when compared with all considered BGI options. In addition, the study results indicate that the flood reduction benefits of bioretention and detention ponds considering the failed UDS initial state condition are less promising.
Figure 14

BGI effect on reduction of (a) total flood volume (top) and (b) average flood duration for Nalukolongo failed UDS.

Figure 14

BGI effect on reduction of (a) total flood volume (top) and (b) average flood duration for Nalukolongo failed UDS.

Close modal

Effectiveness of modelled BGI options considering a non-failed initial state

Model simulations considering a non-failed UDS initial state condition were undertaken and the results compared. The modelling of the non-failed UDS initial state condition enabled the quantification of the combined effect of implementing BGIs and the implementation of effective cleaning and maintenance of the existing grey UDSs.

Kinawataka UDS

The results indicate that detention ponds under non-failed UDS conditions led to marginal reduction in total flood volume of 20% [4–30%] when compared with the failed UDS initial state condition of 19% [6–28%]. Similar to detention ponds, the performance of bioretention cells improved by 4% under the non-failed UDS initial state condition. Furthermore, the use of detention ponds and infiltration trenches under non-failed UDS initial conditions led to a moderate reduction in the average flood duration of 10–13%, when compared with the failed UDS initial state condition.

The results further suggest that use of spatially distributed RWH systems leads to the highest reduction in total flood volume and average flood duration of 50% [16–65%] and 23% [3–23%] (Figure 15), respectively. More so, the results further indicate the average reduction in total flood volume for RWH systems and infiltration trenches increased under the non-failed UDS initial state by 5 and 4%, respectively. The results also indicate reduction in average flood duration for RWH systems and infiltration trenches increased under the non-failed UDS initial state by 5 and 11%, respectively.
Figure 15

BGI effect on reduction of (a) total flood volume (top) and (b) average flood duration for Kinawataka failed UDS.

Figure 15

BGI effect on reduction of (a) total flood volume (top) and (b) average flood duration for Kinawataka failed UDS.

Close modal

Nalukolongo UDS

The results indicate that detention ponds under the non-failed UDS condition led to slightly higher reduction in total flood volume of 12% [8–17%] when compared with the failed UDS initial state reduction of 7% [2–17%] (Figure 16). On the contrary, the performance of bioretention cells improved by 9% under the non-failed UDS initial state when compared with the failed UDS initial state condition. Furthermore, the use of detention ponds and infiltration trenches under the non-failed UDS initial condition led to moderate reduction in the average flood duration of 5–9%, when compared with the failed UDS initial state condition. The results further suggest that the use of spatially distributed RWH systems led to the highest reduction in total flood volume and average flood duration of 23% [9–34%] and 43% [36–54%]. More so, the results further indicate the average reduction in total flood volume for RWH systems and infiltration trenches increased under the non-failed UDS initial state by 7 and 3%, respectively. The results also indicate the reduction in average flood duration for RWH systems and Infiltration trenches increased under the non-failed UDS initial state significantly increased by 21%.
Figure 16

BGI effect on reduction of (a) total flood volume (top) and (b) average flood duration for Nalukolongo non-failed UDS.

Figure 16

BGI effect on reduction of (a) total flood volume (top) and (b) average flood duration for Nalukolongo non-failed UDS.

Close modal

Cost–benefit analysis

The results of the cost–benefit analysis are presented in Figure 17. Plausibly, the computed capital costs of implementing the detention pond, infiltration trench, and bioretention cell options are higher than the non-failed option by at least 342, 283, and 185%, respectively, which is attributed to the costs related to the attendant civil works (excavation, ground improvement, reinforced concrete works, water proofing, and landscaping among others). When the costs of failure (direct tangible flooding costs) are taken into consideration, the RWH option leads to net benefits of 13 and 38% for the Kinawataka and Nalukolongo UDSs, respectively, considering a system's service life of 50 years. The use of bioretention cells led to a net benefit of 0.04% for Nalukolongo UDS only, while trenches and centralised detention ponds led to net losses in the considered case study UDSs.
Figure 17

Computed discounted total costs for the baseline and tested BGI options for (a) Kinawataka and (b) Nalukolongo UDSs.

Figure 17

Computed discounted total costs for the baseline and tested BGI options for (a) Kinawataka and (b) Nalukolongo UDSs.

Close modal

Model setup and validation

In this research, coupled 1D–2D modelling of two large urbanised catchments located in Kampala, Uganda, was undertaken, considering failed and non-failed initial UDS states. The adopted coupled 1D–2D model implemented in PCSWMM software considers direct rainfall (assuming uniform spatial distribution) on each sub-catchment surface in the hydrology module in the 1D model domain using the linear reservoir model for transformation of rainfall into runoff (Rossman 2016; Ortega Sandoval et al. 2023) as opposed to 2D direct rainfall modelling approaches in which rainfall is directly applied to the 2D grid cells. The adopted modelling approach allows for simultaneous transition between the 1D and 2D domains based on a direct connection of the link-node system (Ortega Sandoval et al. 2023). However, the 2D model is not affected if no sewer flooding occurs. The adopted 2D model structure could introduce an additional source of uncertainty.

Furthermore, the results of the OAT sensitivity analysis suggested that the 2D modelling was less sensitive to changes in Manning's n value for the 2D cells. This was largely attributed to the topography of the case study catchments, which consists of various hills and valley bottoms where the primary drainage channels are located (as opposed to flat terrains). During flooding conditions, the flooding extent is limited to the valley bottoms. Changes in model parameters therefore led to limited increase in the flooding extent as opposed to in catchments with flat terrains where a slight change in the model parameters would lead to a significant increase in the total flooded area.

The developed models were validated through comparison of measured flood depths at selected key flooding hot stop locations and the corresponding simulated flood depths at those locations. It is always critical to compare the observed flood depths (which are simultaneously measured with the rainfall that caused flooding) with the corresponding simulated results to effectively validate a physical-based model. However, because of the absence of simultaneous rainfall and flood depth measures for the case study UDSs, the adopted validation approach compared the observed maximum flood depths with simulated flood depths corresponding to various return periods at the selection locations. Based on the comparison of these results, the return period of rainfall event that led to the observed maximum flood depths at the selected hotspot locations was determined as 37 years. The results therefore suggest that if only the daily precipitation is available, the corresponding design rainfall event (with the same total rainfall and duration as the recorded observed maximum rainfall) should be used for model validation purposes.

Furthermore, the study has demonstrated potential approaches for improved urban drainage modelling in cities with limited or no model calibration data sets. In addition, the study has demonstrated the need and importance of identifying sensitive model parameters using OAT sensitivity analysis. The study has also demonstrated that by the accurate determination of PIMP and catchment slopes, model uncertainties can be minimised, therefore leading to improved estimation of pluvial flooding extent and depth in urban areas (e.g. Reinstaller et al. 2022). In future research, the reduction of coupled 1D–2D urban drainage model uncertainties through use of 2D direct rainfall (rain-on-grid) modelling approaches and spatially varied Manning roughness coefficients should be investigated (David & Schmalz 2020; Godara et al. 2023).

1D–2D Flood modelling

The 1D–2D modelling results suggest that the occurrence of moderate rainfall events with return periods T ≤ 2 years led to extensive flooding in both catchments. Although secondary and primary UDSs in Kampala are designed using T2 and T10 return periods, respectively (KCCA 2016), the coupled 1D–2D modelling results confirm that the UDSs are insufficient to convey resulting flows (caused by T2 and T10 rainstorms) confirming the insufficient hydraulic conveyance capacity within the existing UDSs.

Furthermore, the study results suggest that flooding extent within the Nalukongo UDS is less sensitive to the increases in the magnitude of the extreme rainfall events when the return period exceeds T10 (i.e. events with return periods less than T10 would have already led to significant flooding of the flood plain areas). This could be largely attributed to the topography of the case study area, which comprises gentle hills that drain in the narrow valleys where the existing UDSs are constructed. Conversely, the results suggest that the flooding extent within the Kinawataka catchment is sensitive to increases in the magnitude of the extreme rainfall events. For example, the occurrence of the T50 and T100 extreme rainfall events led to the highest increase in flooding extents of 23 and 38%, respectively, when compared with the flooding extent resulting from the T2 event.

The results therefore suggest that use of T10 as a standard for re-design for rehabilitation and expansion of the Nalukolongo UDS could lead to substantial reduction of the flooding impacts in the Nalukolongo catchment. However, for Kinawataka, any future design or rehabilitation of the existing UDS should be based on T25, to achieve a similar reduction in the resulting flooding impacts.

Last but not least, comparison of flooding extents obtained considering the failed and non-failed UDS initial states suggests that the cleaning and maintenance actions for the UDS system would only lead to a minimal reduction of the flooding impacts, which further underscores the need for the implementation of the proposed BGI.

Effectiveness of BGI

In this study, consideration of the failed UDS initial states favourably represented the prevailing UDS conditions in Kampala city, and indeed in most cities in developing countries. The study results showed that the use of spatially distributed RWH tanks and infiltration trenches in the Kinawataka catchment led to the highest average reduction in total flood volume of 33–45%. Relatively lower reductions in the average total flood volume of 14–16% were obtained when the same BGI options were implemented in the Nalukolongo catchment.

Conversely, the two BGI options also led to at least 18 and 22% reduction in the average flood duration in the Kinawataka and Nalukolongo catchments under a failed UDS initial state condition, respectively. The study results therefore suggest that the use of spatially distributed RWH systems leads to the highest average reduction in total flood volume (16–45%) and average flood duration (18–24%), followed by infiltration trenches, which resulted in reduction in total flood volume and average flood duration of 14–33% and 18–22%, respectively. The study results also confirmed that the highest reduction in total flood volume and average flood duration was obtained for rainfall events with low return periods (<T10). The use of RWH systems also provides multiple benefits such as urban flood reduction and provision of alternative water supply, especially in cities with intermittent water supplies. However, the use of RWH systems for multiple uses would require tanks equipped passively or more advanced active control systems for regulation of flow of stored rainwater into the UDS (Mugume et al. 2017) and a parallel plumbing system that may increase capital costs. In addition, part of the tank would not be available for storm water control purposes.

The study results also confirmed that the use of large centralised detention ponds in the two case study catchments led to the lowest average reduction in total flood volume (7–19%) and average flood duration (5–7%) under the failed UDS initial conditions. The study results confirm that BGIs need to be spatially distributed in upstream parts of the catchments so as to reduce the hydraulic loads that enter the existing UDS. This will increase the redundancy and flexibility attributes of the UDS system, which in turn increases the general resilience attributes of the UDS (Mugume et al. 2015a).

Furthermore, the results suggest that the effectiveness of bioretention cells significantly improved by up to 6.3% under the non-failed initial UDS condition, hence suggesting the need to combine the use of an infiltration-based BGI infrastructure with regular maintenance of both the BGI infrastructure and the existing grey UDSs. Lastly, the results suggested that wide-scale implementation of spatially distributed RWH systems provided the highest reduction in the flooding impacts in both catchments, under both the non-failed and failed initial UDS conditions. This study's findings agrees with Webber et al. (2020) who concluded that catchment-wide implementation of decentralised RWH systems provided the most effective strategy for an urban catchment in Melbourne, Australia. RWH systems also provide additional co-benefits such as provision of alternative water supplies, particularly in developing country cities that grapple with intermittent water supplies. Achieving the multiple benefits of urban flood reduction and provision of alternative water supplies could be operationalised in practice through the use of rainfall threshold-based pluvial flood early warning systems to alert households to empty their tanks prior to the start of a rainstorm (Young et al. 2021) or through use of dual-purpose RWH systems (Mugume et al. 2017).

Cost–benefit analysis

In this research, the computation of the net benefit of implementing each of the BGI options considered the costs of the UDS system failure that was quantified through computation of the direct tangible flooding costs. The results suggest that implementation of RWH systems at a catchment scale provides the most cost-effective strategy over the design life of the UDS. Our study findings, which were based on detailed flood modelling in two large urban catchments in Kampala (>32 km2), validate the conclusions of Webber et al. (2019) who investigated reduction in flood damages, comparing RWH systems with green roofs, water butts, and permeable pavements using a relatively small 0.49 km2 residential suburb in a UK city.

By contrast, the implementation of the bioretention cells in the Nalukolongo catchment resulted in net benefits that were almost equal to the baseline option of cleaning and maintenance of the existing UDS. The rest of the BGI options (i.e. infiltration trenches and centralised detention ponds) that are associated with higher capital investment costs were not cost-effective in the considered case study UDSs.

Based on the study, it is concluded that catchment-scale implementation of RWH systems increases the spatial distribution of control strategies in upstream parts of the catchment and is therefore more cost-effective due to the relatively low unit capital and operation and maintenance costs. This in turn provides a cost-effective strategy that could potentially increase the resilience of UDSs to unexpected failures especially when combined with rainfall threshold-based flood early warning systems or the use of dual-purpose RWH systems.

There are limited studies that have rigorously investigated the effectiveness of BGI options in enhancing the resilience in existing UDSs during failure conditions that lead to urban flooding. In this research, a methodology that combines coupled 1D–2D modelling with the consideration of functional and structural failures has been developed and applied to evaluate the effectiveness of BGI options in enhancing the resilience of existing UDSs. The methodology has been demonstrated using the UDSs of Nalukolongo and Kinawataka in Kampala city, Uganda.

OAT local sensitivity analysis enabled identification of PIMP and Manning's n as the most sensitive model parameters. The GIS-based spatial analysis of high-resolution satellite imagery enhanced the estimation of urban land cover parameters and improved the accuracy of the developed UDS models. The developed coupled 1D–2D models were validated using an approach that employed field surveys for the collection of georeferenced maximum flood depths data and comparison with corresponding simulated flood depth values at selected hotspot locations. The computed NSE Index and RMSE values suggested that the coupled 1D–2D modelling approach provided a good fit when compared with the observed maximum flood depths in the two catchments. The proposed model validation approach provides a promising approach for overcoming urban drainage model calibration challenges in cities that lack measured or proxy calibration data sets. The proposed methodology can be further enhanced through targeted urban flood data collection campaigns in which flooding parameters and the rainfall events (sub-daily rainfall duration and magnitude) are measured simultaneously and utilised for model calibration and validation purposes.

In addition, the performance of four BGI options that included spatially distributed RWH systems, infiltration trenches, bioretention cells, and detention pond systems was investigated using the developed case study UDS models. For the considered BGI options, two UDS initial states were considered, that is the failed and non-failed states. The failed UDS state was applied to model UDSs that are characterised by sustained inadequate system cleaning and maintenance that leads to structural failures such as sewer collapse, blockages, deposited sediments, or blockages caused by solid waste, a common occurrence in many developing country cities. It is noted that the BGIs were modelled in PCSWMM using the LID module, based on the principle of effective rainfall, which does not consider hydraulic interactions between the modelled BGI options and the 2D surface flows during flooding conditions, thus necessitating further research.

Based on the study, the following conclusions specific to the case study UDSs in Kampala can be made.

  • Wide-scale implementation of spatially distributed RWH systems and infiltration trenches at an urban catchment level led to the highest reduction in the flooding impacts and consequences, when compared with bioretention cells and detention pond systems even when the existing UDSs experienced system failures. It is therefore concluded that RWH systems when implemented at a wider urban catchment scale increase redundant capacity and flexibility attributes of existing grey UDSs and thus provide a promising and cost-effective option for enhancing UDSs’ resilience to flooding caused by a combination of moderate rainfall and structural failures in UDSs in rapidly urbanising cities;

  • The developed methodology has facilitated improved understanding of the effect of BGI on the hydraulic performance behaviour of existing UDSs during failure conditions that lead to urban flooding.

Future research should focus on (a) improved modelling of bidirectional flow interactions between the 1D and 2D urban drainage model domains at a microcatchment level, (b) investigation of the effectiveness of 2D direct rainfall modelling approaches in reducing coupled 1D–2D modelling uncertainties, (c) development of new flood depth–velocity–damage relationships to facilitate improved estimation of EADs in developing country cities, (d) quantification of multiple benefits of BGI to give a more complete view of their cost-effectiveness, and (e) investigation of key barriers to implementation of BGI in developing country cities.

Acknowledgement is given to the Directorate of Water Resources Management (DWRM); Ministry of Water and Environment, Uganda; Kampala Capital City Authority (KCCA); and the German Development Cooperation (GIZ) for the provision of the data sets used in the study. The insights of two anonymous reviewers are also gratefully acknowledged. We also confirm that there are no relevant financial or non-financial competing interests to report.

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

The authors declare there is no conflict.

Alderson
D. L.
,
Brown
G. G.
&
Carlyle
W. M.
2015
Operational models of infrastructure resilience
.
Risk Anal.
35
,
562
586
.
https://doi.org/10.1111/risa.12333
.
Almaaitah
T.
,
Appleby
M.
,
Rosenblat
H.
,
Drake
J.
&
Joksimovic
D.
2021
The potential of Blue-Green infrastructure as a climate change adaptation strategy: A systematic literature review
.
Blue-Green Syst.
3
,
223
248
.
https://doi.org/10.2166/bgs.2021.016
.
Balbastre-Soldevila
R.
,
García-Bartual
R.
&
Andrés-Doménech
I.
2019
A comparison of design storms for urban drainage system applications
.
Water
11
,
1
15
.
https://doi.org/10.3390/w11040757
.
Butler
D.
,
Ward
S.
,
Sweetapple
C.
,
Astaraie-imani
M.
,
Diao
K.
,
Farmani
R.
&
Fu
G.
2016
Reliable, resilient and sustainable water management: The Safe & SuRe approach
.
Global Challenges
63
77
.
https://doi.org/10.1002/gch2.1010
.
Butler
D.
,
Digman
C.
,
Makropoulos
C.
&
Davies
J. W.
2018
Urban Drainage
, 4th edn..
CRC Press, Taylor and Francis Group
,
London and New York
.
Chitwatkulsiri
D.
,
Miyamoto
H.
,
Irvine
K. N.
,
Pilailar
S.
&
Loc
H. H.
2022
Development and application of a real-time flood forecasting system (RTFlood system) in a tropical urban area: A case study of Ramkhamhaeng Polder, Bangkok, Thailand
.
Water
14
.
https://doi.org/10.3390/w14101641
de Vitry
M. M.
&
Leitão
J. P.
2020
The potential of proxy water level measurements for calibrating urban pluvial flood models
.
Water Res
.
https://doi.org/https://doi.org/10.1016/j.watres.2020.115669
.
Digman
C. J.
,
Anderson
N.
,
Rhodes
G.
,
Balmforth
D. J.
&
Kenney
S.
2014
Realising the benefits of integrated urban drainage models
.
Water Manage.
167
,
30
37
.
https://doi.org/10.1680/wama.12.00083
.
Fu
G.
,
Zhang
C.
,
Hall
J. W.
&
Butler
D.
2023
Are sponge cities the solution to China's growing urban flooding problems?
Wiley Interdiscip. Rev. Water
10
,
1
9
.
https://doi.org/10.1002/wat2.1613
.
GIZ
2022a
Situational Analysis Report. Technical support for the implementation of the Greater Kampala Integrated Flood Resilience Partnership. Kampala
.
GIZ
2022b
Flooding Situation and Hotspot Analysis Report. Technical support on the scoping for the Greater Kampala Integrated Flood Resilience Partnership
.
Godara
N.
,
Bruland
O.
&
Alfredsen
K.
2023
Simulation of flash flood peaks in a small and steep catchment using rain-on-grid technique
.
J. Flood Risk Manage.
16
,
1
14
.
https://doi.org/10.1111/jfr3.12898
.
Goncalves
M. L. R.
,
Zischg
J.
,
Rau
S.
,
Sitzmann
M.
,
Rauch
W.
&
Kleidorfer
M.
2018
Modeling the effects of introducing low impact development in a tropical city: A case study from Joinville, Brazil
.
Sustainability
10
.
https://doi.org/10.3390/su10030728
.
Green
D.
,
O'Donnell
E.
,
Johnson
M.
,
Slater
L.
,
Thorne
C.
,
Zheng
S.
,
Stirling
R.
,
Chan
F. K.
,
Li
L.
&
Boothroyd
R.
2021
Green infrastructure: The future of urban flood risk management
.
Wiley Interdiscip. Rev. Water
8
,
e1560
.
Gupta
H. V.
&
Kling
H.
2011
On typical range, sensitivity, and normalization of mean squared error and nash-sutcliffe efficiency type metrics
.
Water Resour. Res.
47
,
2
4
.
https://doi.org/10.1029/2011WR010962
.
Guptha
G. C.
,
Swain
S.
,
Al-Ansari
N.
,
Taloor
A. K.
&
Dayal
D.
2022
Assessing the role of SuDS in resilience enhancement of urban drainage system: A case study of Gurugram City, India
.
Urban Clim.
41
,
101075
.
https://doi.org/10.1016/j.uclim.2021.101075
.
Haghighatafshar
S.
,
Nordlöf
B.
,
Roldin
M.
,
Gustafsson
L.-G.
,
la Jansen
J. C.
&
Jönsson
K.
2018
Efficiency of blue-green stormwater retrofits for flood mitigation – conclusions drawn from a case study in Malmö, Sweden
.
J. Environ. Manage.
207
,
60
69
.
https://doi.org/10.1016/j.jenvman.2017.11.018
.
Han
W. S.
&
Burian
S. J.
2009
Determining effective impervious area for urban hydrologic modeling
.
J. Hydrol. Eng.
14
,
111
120
.
https://doi.org/10.1061/(ASCE)1084-0699(2009)14 : 2(111)
.
Havik
J. C.
2012
The Fate and Transport of Nutrients in Shallow Groundwater and Soil of an Urban Slum Area in the City of Kampala
.
Utrecht University
,
Uganda
.
Hellmers
S.
,
Manojlovic
N.
,
Palmarlcclottl
G.
,
Kurzbach
S.
&
Frohle
P.
2018
Multiple linked sustainable drainage systems in hydrological modelling for urban drainage and flood risk management
.
J. Flood RIsk Manage.
11
,
S5
S18
.
HM Treasury
2011
The Green Book: Appraisal and Evaluation in Central Government
.
London, UK
.
Houle
J. J.
,
Roseen
R. M.
,
Ballestero
T. P.
,
Puls
T. A.
&
Sherrard
J.
2013
Comparison of maintenance cost, labor demands, and system performance for LID and conventional stormwater management
.
J. Environ. Eng.
139
,
932
938
.
https://doi.org/10.1061/(asce)ee.1943-7870.0000698
.
HR Wallingford
2004
Whole Life Costing for Sustainable Drainage
.
Howbery Park
,
Wallingford
.
Hunter
N. M.
,
Bates
P. D.
,
Neelz
S.
,
Pender
G.
,
Villanueva
I.
,
Wright
N. G.
,
Liang
D.
,
Falconer
R. A.
,
Lin
B.
,
Waller
S.
,
Crossley
A. J.
&
Mason
D. C.
2008
Benchmarking 2D hydraulic models for urban flooding
.
Proc. Inst. Civ. Eng. Water Manage.
161
,
13
30
.
https://doi.org/10.1680/wama.2008.161.1.13
.
Jefferson
A. J.
,
Bhaskar
A. S.
,
Hopkins
K. G.
,
Fanelli
R.
,
Avellaneda
P. M.
&
McMillan
S. K.
2017
Stormwater management network effectiveness and implications for urban watershed function: A critical review
.
Hydrol. Processes
31
,
4056
4080
.
https://doi.org/10.1002/hyp.11347
.
Kalantari
Z.
,
Ferreira
C. S. S.
,
Keesstra
S.
&
Destouni
G.
2018
Nature-based solutions for flood-Nature-based solutions for flood-drought risk mitigation in vulnerable urbanizing parts of East-Africa
.
Curr. Opin. Environ. Sci. Health
5
,
73
78
.
https://doi.org/10.1016/j.coesh.2018.06.003
.
KCCA
2016
Kampala Drainage Master Plan 2016: Final Technial Report
.
Kampala
.
Leandro
J.
,
Chen
A. S.
,
Djordjević
S.
&
Savić
D. A.
2009
Comparison of 1D-1D and 1D
–2D
coupled (Sewer/Surface) hydraulic models for urban flood simulation
.
Hydraul. Eng.
135
,
495
504
.
Li
F.
,
Chen
J.
,
Engel
B. A.
,
Liu
Y.
,
Wang
S.
&
Sun
H.
2021
Assessing the effectiveness and cost efficiency of green infrastructure practices on surface runoff reduction at an urban watershed in China
.
Water
13
,
1
19
.
https://doi.org/10.3390/w13010024
.
Maksimović
Č.
,
Prodanović
D.
,
Djordjević
S.
,
Boonya-Aroonnet
S.
,
Leitão
J. P.
&
Allitt
R.
2009
Overland flow and pathway analysis for modelling of urban pluvial flooding
.
Hydraul. Res.
47
,
512
523
.
https://doi.org/10.3826/jhr.2009.3361
.
Mcdonnell
J.
&
Motta
D.
2022
Changing rate of urban creep and urban expansion over time and its impact upon the hydrologic response of a catchment
.
Water Sci. Technol.
85
,
383
397
.
https://doi.org/10.2166/wst.2021.629
.
Minsker
B.
,
Baldwin
L.
,
Crittenden
J.
,
Kabbes
K.
,
Karamouz
M.
,
Lansey
K.
,
Malinowski
P.
,
Nzewi
E.
,
Pandit
A.
,
Parker
J.
,
Rivera
S.
,
Surbeck
C.
,
Wallace
W. A.
&
Williams
J.
2015
Progress and recommendations for advancing performance-based sustainable and resilient infrastructure design
.
J. Water Resour. Plann. Manage.
141
.
https://doi.org/10.1061/(asce)wr.1943-5452.0000521
.
Mugume
S. N.
&
Butler
D.
2017
Evaluation of functional resilience in urban drainage and flood management systems using a global analysis approach
.
Urban Water J.
14
,
727
736
.
http://dx.doi.org/10.1080/1573062X.2016.1253754
.
Mugume
S. N.
,
Gomez
D.
, &
Butler
D.
2013
Statistical methods for climate change impact assessment on urban rainfall extremes for cities in tropical developing countries – A review
. In:
International Conference on Flood Resilience: Experiences in Asia and Europe
(
Butler
D.
,
Chen
A. S.
,
Djordjevic
S.
&
Hammond
M. J.
, eds).
University of Exeter
,
Exeter, United Kingdom
.
Mugume
S. N.
,
Diao
K.
,
Astaraie-Imani
M.
,
Fu
G.
,
Farmani
R.
&
Butler
D.
2015a
Enhancing resilience in urban water systems for future cities
.
Water Sci. Technol. Water Supply
15
,
1343
1352
.
https://doi.org/10.2166/ws.2015.098
.
Mugume
S. N.
,
Gomez
D. E.
,
Fu
G.
,
Farmani
R.
&
Butler
D.
2015b
A global analysis approach for investigating structural resilience in urban drainage systems
.
Water Res.
81
,
15
26
.
https://doi.org/10.1016/j.watres.2015.05.030
.
Mugume
S. N.
,
Gomez
D.
,
Melville-Shreeve
P.
&
Butler
D.
2017
Multifunctional urban flood resilience enhancement strategies
.
Proc. Inst. Civ. Eng. - Water Manage.
170
,
115
127
.
https://doi.org/10.1680/jwama.15.00078
.
Mugume
S. N.
,
Abasabyoona
G.
,
Engwau
J.
,
Sempewo
J. I.
,
Van de Sande
B.
&
Butler
D.
2024
Assessing the impact of the rise in Lake Victoria water levels on urban flooding using a GIS-based spatial flood modelling approach
.
Urban Water J.
21
,
219
233
.
https://doi.org/10.1080/1573062X.2023.2284960
.
Mugume
S. N.
,
Kibibi
H.
&
Wakhata
M.
2023
Effectiveness of blue-green infrastructure for enhancing urban drainage system resilience
. In:
Proceedings of the 40th IAHR World Congress. Rivers – Connecting Mountains and Coasts
(
Habersack
H.
,
Tritthart
M.
&
Waldenberger
L.
, eds).
IAHR
,
Vienna
, pp.
1488
1497
.
doi: 10.3850/978-90-833476-1-5_iahr40wc-p0876-cd
Ortega Sandoval
A. D.
,
Sörensen
J.
,
Rodríguez
J. P.
&
Bharati
L.
2023
Hydrologic–hydraulic assessment of SUDS control capacity using different modeling approaches: A case study in Bogotá, Colombia
.
Water Sci. Technol.
87
,
3124
3145
.
https://doi.org/10.2166/wst.2023.173
.
Panos
C. L.
,
Wolfand
J. M.
&
Hogue
T. S.
2021
Assessing resilience of a dual drainage urban system to redevelopment and climate change
.
J. Hydrol.
596
.
https://doi.org/10.1016/j.jhydrol.2021.126101
.
Perini
L.
,
Calabrese
L.
,
Salerno
G.
,
Ciavola
P.
&
Armaroli
C.
2016
Evaluation of coastal vulnerability to flooding: Comparison of two different methodologies adopted by the Emilia-Romagna region (Italy)
.
Nat. Hazards Earth Syst. Sci.
16
,
181
194
.
https://doi.org/10.5194/nhess-16-181-2016
.
Pleau
M.
,
Tao
D. Q.
,
Grondin
F.
,
Shishegar
S.
&
Fradet
O.
2022
Sensor fault detection for urban drainage systems using redundant measurements
.
Urban Water J
19
(
10
),
1038
1046
.
Reinstaller
S.
,
Krebs
G.
,
Pichler
M.
&
Muschalla
D.
2022
Identification of high-impact uncertainty sources for urban flood models in hillside peri-urban catchments
.
Water
14
,
1
25
.
https://doi.org/10.3390/w14121973
.
Ritter
A.
&
Muñoz-Carpena
R.
2013
Performance evaluation of hydrological models: Statistical significance for reducing subjectivity in goodness-of-fit assessments
.
J. Hydrol.
480
,
33
45
.
https://doi.org/https://doi.org/10.1016/j.jhydrol.2012.12.004
.
Rodriguez
M.
,
Fu
G.
,
Butler
D.
,
Yuan
Z.
&
Sharma
K.
2021
Exploring the spatial impact of green infrastructure on urban drainage resilience
.
Water
13
,
1
21
.
https://doi.org/10.3390/w13131789
.
Rodriguez
M.
,
Fu
G.
,
Butler
D.
,
Yuan
Z.
&
Cook
L.
2023
Global resilience analysis of combined sewer systems under continuous hydrologic simulation
.
J. Environ. Manage.
344
,
118607
.
https://doi.org/10.1016/j.jenvman.2023.118607
.
Rosin
T. R.
,
Kapelan
Z.
,
Keedwell
E.
&
Romanoc
M.
2022
Near real-time detection of blockages in the proximity of combined sewer over fl ows using evolutionary ANNs and statistical process control
.
Hydroinformatics
24
,
259
273
.
https://doi.org/10.2166/hydro.2022.036
.
Rossman
L. A.
2016
Storm Water Management Model Reference Manual Volume I – Hydrology (Revised). Report No. EPA/600/R-15/162A
.
Cincinnati, Ohio
.
Seenath
A.
,
Wilson
M.
&
Miller
K.
2016
Hydrodynamic versus GIS modelling for coastal flood vulnerability assessment: Which is better for guiding coastal management?
Ocean Coast. Manage.
120
,
99
109
.
https://doi.org/10.1016/j.ocecoaman.2015.11.019
.
Simon
C.
&
Duchhart
I.
2015
Green Infrastructure as A Tool to Reduce Flooding: The Case in Chota, Mozambique
.
Wageningen University
,
The Netherlands
, pp.
1
31
.
Sörensen
J.
&
Emilsson
T.
2019
Evaluating flood risk reduction by urban blue-green infrastructure using insurance data
.
J. Water Resour. Plann. Manage.
145
.
https://doi.org/10.1061/(asce)wr.1943-5452.0001037
.
Sweetapple
C.
,
Fu
G.
&
Butler
D.
2014
Identifying sensitive sources and key control handles for the reduction of greenhouse gas emissions from wastewater treatment
.
Water Res.
62
,
249
259
.
https://doi.org/10.1016/j.watres.2014.06.002
.
Thoms
A.
&
Köster
S.
2022
Potentials for Sponge City implementation in Sub-Saharan Africa
.
Sustainability
14
.
https://doi.org/10.3390/su141811726
.
UBOS
2016
National Population and Housing Census 2014 - Main Report
.
Kampala
.
Webber
J. L.
,
Fletcher
T. D.
,
Cunningham
L.
,
Fu
G.
,
Butler
D.
&
Burns
M. J.
2020
Is green infrastructure a viable strategy for managing urban surface water flooding?
Urban Water J.
17
,
598
608
.
https://doi.org/10.1080/1573062X.2019.1700286
.
Wijesinghe
A.
&
Thorn
J. P.
2021
Governance of urban green infrastructure in informal settlements of windhoek. Namibia
.
Sustainability
13
,
1
25
.
https://doi.org/10.3390/su13168937
.
World Bank
2020
Kampala Disaster Risk and Climate Change Resilience Strategy Report
.
Kampala
.
Young
A.
,
Bhattacharya
B.
&
Zevenbergen
C.
2021
A rainfall threshold-based approach to early warnings in urban data-scarce regions: A case study of pluvial flooding in Alexandria. Egypt
.
J. Flood Risk Manage.
14
,
1
16
.
https://doi.org/10.1111/jfr3.12702
.
This is an Open Access article distributed under the terms of the Creative Commons Attribution Licence (CC BY 4.0), which permits copying, adaptation and redistribution, provided the original work is properly cited (http://creativecommons.org/licenses/by/4.0/).

Supplementary data