ABSTRACT
Pluvial flooding is a significant threat in many Sub-Saharan African cities, driven by rapid urbanisation, climate change, failures, and insufficient hydraulic capacity of existing urban drainage systems (UDSs). However, limited modelling studies have investigated the effect of sediment and solid waste accumulation on the performance of UDSs during extreme loading conditions. In this research, the influence of combined scenarios of sediment and solid waste deposition and extreme rainfall on urban flooding in the upper Lubigi catchment in Kampala was investigated. The collected high-resolution rainfall data and water level observations in combination with field observations were used for model calibration and validation. The results show three locations that account for over 50% of the flooded volume in the catchment, based on the existing conditions. A near-complete blockage of sediment and solid waste at three culvert crossings increased the flood volumes by up to 40% for higher (≥50 years) return periods and by up to 105% for low return periods (≤10 years). This research underscores the importance of using high-resolution rainfall data in flood modelling, as well as the necessity of improved UDS asset management, and effective solid waste and sediment management in order to achieve resilient and sustainable water management in cities.
HIGHLIGHTS
Utilised high-resolution rainfall and water level measurements for model calibration and validation.
Modelled scenarios combining extreme rainfall, sediments, and solid waste deposition.
Identified three locations accounting for more than 50% of the flooded volume in the Lubigi upstream catchment.
Approximately 105% increase in the total flood volume for rainfall with a 2-year return period due to increased sediment and solid waste.
INTRODUCTION
Many rapidly urbanising Sub-Saharan African cities are grappling with increased risk of urban flooding caused by urban creep, peri-urban expansion, increasing dry weather flows, inadequate capacity of existing urban drainage systems (UDSs) and insufficient solid waste management practices (Owusu & Oteng-Ababio 2015; Sohn et al. 2020; OECD/UN/ECA/AfDB 2022; IPCC 2023; Møller-Jensen et al. 2023; MacAfee & Löhr 2024; Mugume et al. 2024b).
Furthermore, it has been projected in recent studies that climate change will exacerbate the risk of pluvial, fluvial, and coastal flooding in many parts of Africa due to increased frequency and intensity of extreme rainfall, rising sea and inland lake water levels, expansion of urban areas in low-lying flood-prone areas, and low adaptive capacity (Owusu & Oteng-Ababio 2015; Trisos et al. 2022; IPCC 2023; Møller-Jensen et al. 2023; Mugume et al. 2024a).
Urban pluvial flooding occurs when runoff caused by extreme rainfall exceeds the capacity of an existing UDS (Maksimović et al. 2009). In many developing countries, sediment and solid waste deposition in the existing UDSs can further reduce the hydraulic conveyance capacities, leading to increased risk of pluvial flooding (Barthelémy et al. 2016). In addition, uncontrolled discharge of untreated sewage into separate stormwater systems, sewer overflows, and cross-connections often lead to the pollution of receiving waters and the occurrence of water-borne diseases (Chen et al. 2022; Wu et al. 2024).
The conventional UDS design is focused on minimising the probability of occurrence of failures using pre-defined average recurrence intervals (Mugume & Butler 2017; Haghighatafshar et al. 2020). However, to enhance the resilience of UDSs, a risk-based paradigm shift that considers all possible failure scenarios is required (Mugume et al. 2015; Haghighatafshar et al. 2020).
Kampala, the capital of Uganda has experienced increasing frequency and magnitude of pluvial flooding events in recent years, particularly in low-lying and low-income areas, due to rapid urbanisation, sediment and solid waste accumulation, inadequate capacity of the existing UDSs and insufficient UDS asset management practices (Sliuzas et al. 2013; Mugume et al. 2015; Pérez-Molina et al. 2017; Richmond et al. 2018; Arinabo 2022; Adeke & Mugume 2025).
A number of recent modelling studies have been conducted to investigate the causes of flooding and quantify the resulting impacts in Kampala. Kiyengo et al. (2019) applied geographic information system (GIS)-based spatial analysis and two-dimensional (2D) modelling using the Hydrologic Engineering Center's River Analysis Software (HEC–RAS) to assess the spatial and temporal flooding patterns in Kampala's Lubigi catchment, considering data spanning the period from 1983 to 2014. The study findings suggested that extreme floods occurred once every 3 years during the study period.
Habonimana (2014) used high-resolution rainfall data to calibrate a Weather Research and Forecasting model and applied the data to develop a flood model for the Lubigi upper catchment. The study findings suggested that high-intensity, short-duration rainfall results in more severe flooding impacts when compared to long-duration, low-intensity rainfall events.
In another study, Pérez-Molina et al. (2017) developed a cellular automata model to simulate the effect of urban growth on flooding in the upper Lubigi catchment. The findings suggested that due to rapid urbanisation, policies focused on the preservation of urban wetlands are insufficient in minimising the projected urban flooding impacts, thus necessitating the implementation of combinations of flood reduction strategies.
In a recent study, Mugume et al. (2024b) developed a coupled 1D–2D model to evaluate the effectiveness of blue-green infrastructure (BGI) on flooding considering scenario combinations of ‘failed’ and ‘non-failed’ UDS initial states and extreme rainfall. The study results suggested that catchment-scale implementation of multifunctional rainwater harvesting systems could lead to a reduction in the total flood volume of 16–45% and contribute to the enhancement of existing UDS resilience to flooding.
However, most of the studies were undertaken using the historical daily rainfall data for Kampala, which was disaggregated using the TRRL East African Flood Model (Fiddes et al. 1974). Furthermore, there are limited studies that have investigated the effect of sediment deposition and solid waste deposition on pluvial flooding in East African cities.
Min & Tashiro (2024) developed a coupled 1D–2D model using an InfoWorks ICM model in Yangon, Myanmar, utilising high-temporal resolution rainfall data and measured water level data from loggers as model inputs. The study findings suggested that sediment and solid waste deposition and receiving water level variations, such as tidal effects, contribute more to pluvial flooding when compared to extreme rainfall. In a similar study, Pervin et al. (2020) modelled the effect of solid waste on flood risk using case studies of UDSs in Sylhet, Bangladesh, and Bharatpur, Nepal using calibrated and validated hydraulic models. The study results suggested that solid waste deposition contributed to 18.5 and 7.6% increase in flood risk in Sylhet and Bharatpur, respectively. Finally, a similar study by Quang et al. (2022) developed and applied a numerical model to investigate the effect of sediment deposition on urban drainage performance. The study findings suggested that a 40% reduction of the conduit cross-section area resulted in lower reduction (60–70%) of the hydraulic performance of the conduit when compared to a corresponding increase in Manning's roughness coefficient which led to 80–90% reduction in the hydraulic performance of the conduit.
Based on the critical review of recent studies, there exists a critical need to undertake research focused on modeling the effect of sediment and solid waste deposition on pluvial flooding in cities. This is largely attributed to the complex interplay of urbanization that leads to abrupt changes in urban form, increased frequency of occurrence of extreme rainfall, inadequate hydraulic capacity of the existing UDSs, and sediment and solid waste accumulation, which in combination lead to more severe urban flooding impacts. It is, therefore, crucial to investigate these dynamics in order to develop appropriate strategies for enhancing flood resilience in cities. In this respect, therefore, the main objective of this study is to investigate the effect of sediment and solid waste deposition in UDSs on urban flooding. More specifically, this study aims to identify locations vulnerable to flooding in the Lubigi catchment, considering scenarios of combinations of sediment and solid waste deposition rates and extreme rainfall. The following research questions formed the basis for the study:
Can in situ high-resolution rainfall and water level measurements facilitate accurate modelling of flooding impacts in failed UDSs?
How do variations in the rates of sediment and solid waste deposition and the occurrence of extreme rainfall influence flooding in the existing Lubigi UDS?
To address the aforementioned research questions, a new methodology was developed, combining a collection of in situ high-resolution rainfall and water level measurements with 1D hydraulic modelling of scenarios that combine extreme rainfall, sediment, and solid waste deposition rates to quantify the resulting flooding impacts in the existing Lubigi UDS in Kampala. The findings of this study underscore the critical importance of consideration of sediment and waste deposition for more accurate modelling of pluvial flooding impacts in developing cities.
PROPOSED APPROACH AND METHODOLOGY
Adopted study approach for field data collection, GIS analysis, and 1D hydraulic modelling.
Adopted study approach for field data collection, GIS analysis, and 1D hydraulic modelling.
modelling approach
PCSWMM, developed by Computational Hydraulics Inc, was used to create the 1D model. PCSWMM is a widely used modelling program for urban drainage and flooding studies all over the world (Chitwatkulsiri et al. 2022; Ortega Sandoval et al. 2023; Mugume et al. 2024b). PCSWMM uses the United States Environmental Protection Agency (US EPA) EPA SWMM5.2 dynamic rainfall–runoff simulation engine (Rossman & Simon 2022) along with support tools like GIS technology. PCSWMM offers the option to choose from different methods for flow routing. Dynamic wave routing is the most comprehensive routing method, involving computation of the full Saint-Venant flow equations (Rossman & Simon 2022). This approach can represent pressurised flow, with flooding occurring when the water depth at a node surpasses the maximum available depth. The flooded volume is either lost from the system or ponded atop of the node. The ponded volume on top of the node is, subsequently, allowed to re-enter the system when hydraulic capacity is available. The dynamic wave routing approach can account for backwater effects, flow reversal, and local head losses (Rossman & Simon 2022).
Rainfall observations
Kampala has two major rain seasons occurring from March to May and from September to November and receives an average annual rainfall of 1,292 mm (Mugume 2015; Nimusiima et al. 2021). April is, in general, the wettest month, with approximately 180 mm of average monthly rainfall, while July is the driest month, receiving around 60 mm of the average monthly rainfall (Mugume 2015).
Water level observations
Water level staff gauge located in the primary Lubigi channel at the Hoima Road box culvert crossing.
Water level staff gauge located in the primary Lubigi channel at the Hoima Road box culvert crossing.
Case study 1D-model setup
Lubigi catchment with the upstream part of the Lubigi catchment, which is included in the model, shown with elevations from the DEM.
Lubigi catchment with the upstream part of the Lubigi catchment, which is included in the model, shown with elevations from the DEM.
The model of the existing UDS that drains the upstream Lubigi catchment was built as a 1D hydraulic model in PCSWMM. The primary and secondary channels consist mainly of open trapezoidal channels of concrete, stone pitching, and earth material. The open channels connect with concrete box culverts and pipe culverts located at the existing road crossings. In 2020–2021, the primary channel from Gulu Highway at Bwaise to Hoima Road and the secondary Nakamiro channel were relined using cable concrete lining (KCCA 2020). The 1D-model setup was applied using the case study specific data described in Table 1.
Information on data used in the 1D model
Data set . | Quality and resolution . | Data period or date of publication . | Source . | Purpose . |
---|---|---|---|---|
Tipping bucket rain gauge | High-resolution rainfall (mm) | 1st March–30th April 2024 | Primary source | Model calibration and extreme rainfall analysis |
Staff water level gauge | Water level (m) | 1st March–April 2024 | Primary source | Model calibration and validation |
77-year daily rainfall for Makerere University rain gauge station | Daily rainfall (mm) | 1943–2019 | Uganda National Meteorological database | Extreme rainfall analysis |
Shuttle Radar Topography Mission (SRTM) Digital Elevation Model | 30 m × 30 m spatial resolution | 23-September-2014 | USGS SRTM | Digital elevation data for dividing into catchment and sub-catchments |
Sentinel-2 satellite image | 5 m × 5 m spatial resolution | 22-August-2023 | Copernicus | Land-use and land cover analysis imperviousness (PIMP) from spectral imagery in ArcGIS |
Base map World Imagery ArcGIS Pro | 0.31 m (1), 0.46 m (2), 0.31 m (3) spatial resolution | 15-October-2022 (1), 20-February-2023 (2), 16-April-2023 (3) | Esri, ArcGIS Pro | Land-use and land cover analysis imperviousness (PIMP) from spectral imagery in ArcGIS |
Existing and designed urban drainage network data (open channels, cross culverts) | UDS parameters (bottom width, side slopes), lengths of channel sections, invert levels, slopes, cross culvert diameters | 2016 and 2020 | KCCA reports and design drawings | Channel morphology for the 1D urban drainage model |
Data set . | Quality and resolution . | Data period or date of publication . | Source . | Purpose . |
---|---|---|---|---|
Tipping bucket rain gauge | High-resolution rainfall (mm) | 1st March–30th April 2024 | Primary source | Model calibration and extreme rainfall analysis |
Staff water level gauge | Water level (m) | 1st March–April 2024 | Primary source | Model calibration and validation |
77-year daily rainfall for Makerere University rain gauge station | Daily rainfall (mm) | 1943–2019 | Uganda National Meteorological database | Extreme rainfall analysis |
Shuttle Radar Topography Mission (SRTM) Digital Elevation Model | 30 m × 30 m spatial resolution | 23-September-2014 | USGS SRTM | Digital elevation data for dividing into catchment and sub-catchments |
Sentinel-2 satellite image | 5 m × 5 m spatial resolution | 22-August-2023 | Copernicus | Land-use and land cover analysis imperviousness (PIMP) from spectral imagery in ArcGIS |
Base map World Imagery ArcGIS Pro | 0.31 m (1), 0.46 m (2), 0.31 m (3) spatial resolution | 15-October-2022 (1), 20-February-2023 (2), 16-April-2023 (3) | Esri, ArcGIS Pro | Land-use and land cover analysis imperviousness (PIMP) from spectral imagery in ArcGIS |
Existing and designed urban drainage network data (open channels, cross culverts) | UDS parameters (bottom width, side slopes), lengths of channel sections, invert levels, slopes, cross culvert diameters | 2016 and 2020 | KCCA reports and design drawings | Channel morphology for the 1D urban drainage model |
Surface types analysed from spectral imagery using ArcGIS spatial analysis.
Hydraulic data for the primary and secondary channels was obtained from KCCA (2016). The data used included open channel cross-section dimensions (depth, bottom widths, and side slopes), material, and length. In addition, information on culvert crossings, including cross-sectional dimensions (height and widths), type of culvert, number of barrels, material, and length, was used. For the relined primary channel and the secondary Nakamiro channel, design maps were obtained from KCCA (2020), which included more detailed drawings than KCCA (2016), as well as invert levels. Unfortunately, as-built maps were not available. The invert levels in KCCA (2020) served as a basis for calculating the invert levels for the rest of the catchment area using the given slopes and open channel lengths from KCCA (2016). In addition, site visits were conducted to verify the data obtained from KCCA (2016). Data for open channel and culvert dimensions, material, and culvert type were corrected for some sections after the site visits.
Overview of the modelling setup in PCSWMM showing the primary open channel (1) and eight secondary channels (2–9). The catchment area (green) is delineated into 35 sub-catchments and extended with eight more additional sub-catchments (blue) to account for water downstream that could affect the upstream catchment area.
Overview of the modelling setup in PCSWMM showing the primary open channel (1) and eight secondary channels (2–9). The catchment area (green) is delineated into 35 sub-catchments and extended with eight more additional sub-catchments (blue) to account for water downstream that could affect the upstream catchment area.
Extreme rainfall analysis
Largest recorded rainfalls during the period 1st March to 30th April, 2024. (a) RE1 with a total volume of 52.6 mm and (b) RE2 with a total volume of 43.4 mm. Each point represents the volume grouped into 5-min intervals.
Largest recorded rainfalls during the period 1st March to 30th April, 2024. (a) RE1 with a total volume of 52.6 mm and (b) RE2 with a total volume of 43.4 mm. Each point represents the volume grouped into 5-min intervals.
IDF curves for the Lubigi upstream catchment using a change factor of 1.3 to account for the current climate change.
IDF curves for the Lubigi upstream catchment using a change factor of 1.3 to account for the current climate change.
To account for the larger rainfall events in the model, design storms were created using the alternating block method (Chow et al. 1988; Balbastre-Soldevila et al. 2019) In the alternating block method, the design hyetograph produced specifies the rainfall depth occurring in n successive time intervals of duration Δt over a total duration Td=nΔt. After selecting the design return period, the rainfall intensity is read from the IDF curve for each duration and the corresponding rainfall depth computed as a product of intensity and duration. Thereafter, the amount of rainfall to be added for each additional time step (block) is computed as the difference between successive rainfall depths. Subsequently, the blocks are reordered in a time sequence, centring the maximum intensity at the required duration Td, and arranging the rest alternately on either side of the central block (Chow et al. 1988).
A similar approach is described in the widely adopted Wallingford Procedure, which is based on a comparison of point and areal rainfall data for areas where several rain gauges exist (Butler et al. 2024).
Design storms for the Lubigi catchment area for four return periods (T2, T10, T50, and T100 years) and a fixed duration of 30 h. The rainfall peak occurs at t = 120 min.
Design storms for the Lubigi catchment area for four return periods (T2, T10, T50, and T100 years) and a fixed duration of 30 h. The rainfall peak occurs at t = 120 min.
Sensitivity analysis and calibration
A one-at-a-time sensitivity analysis was conducted to identify the most sensitive model parameters. A total of 15 parameters in PCSWMM were included. The parameters were changed one at a time by increasing and decreasing its value by 50% while the rest of the parameters remained unchanged. Manning's roughness coefficient for channel flow, average value baseflow, and percentage imperviousness were determined to be the most sensitive parameters. As described above, the percentage imperviousness had been calculated by using ArcGIS spatial analysis of spectral imagery; therefore, this value remained unchanged and Manning's roughness coefficient for channel flow and average value base flow were adjusted during calibration.
The model was calibrated using two recently observed high-resolution temporal rainfalls from March: rain event RE1 (shown in Figure 7) and a rain event of 13.4 mm on 27th March, along with the corresponding observed water levels at the Hoima box culvert crossing. The calibration entailed manual adjustment of Manning's n value for channel flow and average base flow to match the observed water level peaks before and after rainfall, while ensuring that the model parameters remained within reasonable bounds.
Because of calibration, the primary and secondary channels containing sediment and solid waste, and the wetland were assigned a Manning's roughness coefficient of 0.05. Secondary channels without sediment and solid waste were assigned a Manning's roughness coefficient of 0.0200–0.0325 depending on their specific characteristics. Culvert crossings currently containing sediment and solid waste were assigned a Manning's roughness coefficient of 0.10. The primary and secondary channels were assigned average base flows of 4 dm3/s and 1 dm3/s, respectively.
Validation
The model was validated using rainfall and water level observations ranging from 5th April to 23rd April 2024. A total of 40 water level observations, covering 11 rainfall events, were included. The validation was evaluated with the Nash–Sutcliffe efficiency (NSE) and the root mean square error (RMSE) (Equations 4 and 5).





Experiment design and scenario description
Scenario 1: Current sediment and solid waste deposition: The existing UDS under the current conditions of sediment and solid waste deposition. These conditions were calibrated and validated.
Scenario 2: Moderate blockage: Culvert cross-sectional area reduced by 50% at Kawaala, Nakamiro, and Bombo Road crossings, reflecting significant, but not complete blockages.
Scenario 3: Severe blockage: Culvert cross-sectional area reduced by 80% at the Kawaala, Nakamiro, and Bombo Road crossings, representing a near-complete blockage.
Scenario 4: ‘Non-failed’ UDS: Manning's roughness coefficient of all the links in the modelled UDS is set to a uniform value of 0.020, representing the desired UDS initial state condition, in which all the system components are well maintained and clean.
Scenario 5: Single component failure: Manning's roughness coefficient of every single component (culvert crossing) was increased to 100, representing the complete failure of the culvert crossing.
Culvert crossings with increased sediment and solid waste blockage, used for Scenarios 2 and 3. The numbers next to the channels are the same as in the overview, shown in Figure 6. Kawaala and Bombo Road crossings are located on the primary channel (1), and Nakamiro crossing is located on a secondary channel (3).
Culvert crossings with increased sediment and solid waste blockage, used for Scenarios 2 and 3. The numbers next to the channels are the same as in the overview, shown in Figure 6. Kawaala and Bombo Road crossings are located on the primary channel (1), and Nakamiro crossing is located on a secondary channel (3).
RESULTS
Model performance
The validation was executed for 40 water level observations from April 2024. The NSE resulted in 0.81, which was considered acceptable. The RMSE was calculated to be 0.18 m, compared to the standard deviation, which was 0.41 m.
Calibration and validation results. The modelled water level is plotted against the observed water level. The circular symbols represent a measured water level at a specific time, while the triangular symbols represent an estimated peak water level based on the observed high-level flood marks on the Hoima Road box culvert wall.
Calibration and validation results. The modelled water level is plotted against the observed water level. The circular symbols represent a measured water level at a specific time, while the triangular symbols represent an estimated peak water level based on the observed high-level flood marks on the Hoima Road box culvert wall.
Locations vulnerable to flooding
Locations vulnerable to flooding. The numbers next to the channels are the same as in the overview, shown in Figure 6.
Locations vulnerable to flooding. The numbers next to the channels are the same as in the overview, shown in Figure 6.
Table 2 shows the flooding loss at these three locations. In the model, flooding loss refers to water that has overflowed the drainage system and is no longer part of the drainage network. This could be described as flooded water being spread out on the surface, effectively being ‘lost’ from the drainage system and not returning to the source. For the recorded rain event RE1, described in Figure 7, flooding at Bwaise, Alice Kaggwa, and Kamwokya Kisalosalo accounted for 57% of the total flooding in the catchment. For the synthetic T10 and T100 rain events, flooding was approximately 20% compared to the entire catchment area. It should be noted that the recorded rain event RE1 was 52.6 mm over 5.5 h, while the synthetic T10 and T100 rain events were 113.7 and 156 mm, respectively, both over 30 h.
Flooding losses at Bwaise, Alice Kaggwa Road, and Kamwokya Kisalosalo Road
Rainfall . | Bwaise flooding loss (m3) . | Alice Kaggwa Road flooding loss (m3) . | Kamwokya Kisalosalo Road flooding loss (m3) . | Aggregate (m3)/% of total flood loss . |
---|---|---|---|---|
RE1 (52.6 mm) | 149,000 | 64,000 | 78,000 | 291,000/57 |
T10 (113.7 mm) | 209,000 | 94,000 | 140,000 | 443,000/19 |
T100 (156.4 mm) | 252,000 | 115,000 | 176,000 | 543,000/20 |
Rainfall . | Bwaise flooding loss (m3) . | Alice Kaggwa Road flooding loss (m3) . | Kamwokya Kisalosalo Road flooding loss (m3) . | Aggregate (m3)/% of total flood loss . |
---|---|---|---|---|
RE1 (52.6 mm) | 149,000 | 64,000 | 78,000 | 291,000/57 |
T10 (113.7 mm) | 209,000 | 94,000 | 140,000 | 443,000/19 |
T100 (156.4 mm) | 252,000 | 115,000 | 176,000 | 543,000/20 |
Note. The fifth column shows the percentage of flooding in these three locations compared to the total flooding in the catchment. RE1 is the largest recorded rainfall and T10 and T100 are synthetic rainfalls.
Scenario analysis
The following section presents the results from the scenario analysis, summarised in Table 3. Scenario 3 resulted in a large percentage increase in the total flood volume when compared to Scenarios 1 and 2, particularly for the rainfall events with lower return periods (RE1 and T2). In addition, Scenarios 2 and 3 led to an increase in upstream flooding when compared to Scenario 1, which was attributed to backwater effects caused by increased sediment and waste deposition in downstream sections of the existing Lubigi UDS.
Results from the scenario analysis including the current scenario (1), 50% blockage of sediment and solid waste at three locations (2) and 80% blockage of sediment and solid waste at three locations (3)
![]() |
![]() |
Note. The flooding indicators are highlighted in light grey. RE1 is the largest recorded rainfall and T2–T100 are synthetic rainfalls.
The study results suggest that flooding increases with higher rainfall return periods, sediment, and solid waste deposition. The computed flooding loss to rainfall runoff ratio also increased with the increasing rates of sediment and solid waste deposition. Furthermore, the observed rain event RE1 resulted in a higher flooding loss ratio when compared to design storm T2 and was comparable to the corresponding Scenario 3, T10 results. In addition, the study results indicated that low return period rainfall events (RE1 and T2) led to a high increase in the total flood volume.
1D-model simulation results considering the ‘non-failed’ UDS scenario considering the observed RE1 rainfall scenario.
1D-model simulation results considering the ‘non-failed’ UDS scenario considering the observed RE1 rainfall scenario.
The study results also suggest that the most significant residual flooding would occur in the Nakamiro secondary channel, at the locations of major culvert crossings along the main Lubigi primary channel, and in the downstream nodes located after the Hoima Road crossing.
Percentage increase in the total flood volume resulting from single component (culvert crossing) failures.
Percentage increase in the total flood volume resulting from single component (culvert crossing) failures.
Nakamiro crossing where (1) is located upstream of the crossing on the Lubigi channel, (2) is located upstream of the crossing on the Nakamiro channel, and (3) downstream of the crossing.
Nakamiro crossing where (1) is located upstream of the crossing on the Lubigi channel, (2) is located upstream of the crossing on the Nakamiro channel, and (3) downstream of the crossing.
DISCUSSION
Extreme rainfall analysis
The tipping bucket rain gauge used in this research records point measurements, and all sub-catchments were assigned to the same rain gauge, regardless of the distance from it. However, spatial variability of rainfall is a significant source of uncertainty in flood modelling, regardless of the catchment size (Arnaud et al. 2011). In addition, the high-resolution rainfall time series collected during the study period covered a relatively short period of 2 months. Using a longer period would enhance the understanding of the temporal variability of rainfalls in Kampala and increase the likelihood of observing extreme rainfalls.
Furthermore, in a recent study based on a case study of the Nakivubo catchment in Kampala City, Mugume & Butler (2017) suggested that the use of ARFs for spatial averaging of rainfall can lead to overestimation of the magnitude of flooding resulting from spatially varied extreme rainfall events. This, therefore, underscores the need for further research focused on the derivation of area reduction factors for Kampala using relatively long high spatial and temporal resolution rainfall data collected using a dense network of automated tipping bucket rain gauges or through the use of recent gridded re-analysis satellite rainfall data products (Nakkazi et al. 2022; Lutz et al. 2024). In addition, future research could also focus on a comparative evaluation of alternative methods for creating design storms in Kampala (Guo & Zhuge 2008; Markolf et al. 2021).
Model performance
The model generally overestimates the observed water levels at low levels that range from 0.25 to 0.8 m and underestimates for higher water levels greater than 0.8 m (Figure 11). This discrepancy is likely due to several key limitations. First, the adopted modelling approach considered bed load sediment and solid waste deposition but did not account for the attendant transport processes. During severe flooding events, large amounts of sediment and solid waste are transported downstream through the catchment. KCCA oversees the maintenance of the UDSs in Kampala. Collecting data on the amount and frequency of the removal of deposited sediments and solid waste in the Lubigi catchment area would enhance the understanding of sediment and solid waste transport processes, which could be used for future refinement of the developed model.
Additionally, there are uncertainties in the hydraulic data used to build the 1D model. The invert levels are based on design drawings from KCCA (2020) for a section of the primary channel, rather than as-built drawings, and the invert levels for the rest of the catchment were calculated using these invert levels along with the information from KCCA (2016). As mentioned previously, the KCCA (2016) report contains uncertainties regarding the open channels and culverts, thereby introducing uncertainty to the model.
The estimated peak water level values used in calibration and validation are uncertain as they are based on water level indications on the box culvert wall. Using this method, the time of the peak is not accurately captured. An automated water level metre would better capture the water level variations, especially when there are rapid changes after a severe storm. This enables continuous water level monitoring instead of the 2–3 daily measurements taken during the study.
In most 1D modelling studies, flood occurrence is represented using virtual cones or reservoirs on the top of each node in which floodwater is temporarily stored and allowed to flow back into the minor system when the hydraulic capacity allows (Butler et al. 2024). This assumption is called ‘allow ponding’ in PCSWMM. In this research, flooding is simulated as the water lost through overtopping of the nodes in the drainage system onto the surrounding land. The ‘allow ponding’ feature was tested during calibration of the model to simulate the resulting flooding impacts. However, this 1D modelling approach resulted in unreasonably high and prolonged water levels compared to the observed water levels measured at the Hoima Road box culvert crossing. On this basis, the 1D modelling option that allowed the exceedance flows to be ‘lost’ from the system was used to minimise the resulting errors in the simulation of water levels in the 1D minor system.
Lastly, there are uncertainties related to using a uniform PIMP value for all sub-catchments in the model. However, given that the catchment is highly urbanised, and with imperviousness levels greater than 70%, this assumption was considered plausible. Despite these uncertainties, the model validation resulted in acceptable NSE and RMSE values, indicating that the model can reliably predict real-world conditions.
Scenario analysis
The study results obtained using a calibrated and validated 1D urban drainage model have demonstrated that partial failures caused by sediment and solid waste deposition in the existing UDSs can be reliably modelled using three options. Option 1 entails the reduction of the cross-sectional areas and adjustment of the invert levels of the modelled culvert crossings (Scenarios 1, 2, and 3), Option 2 entails the adjustment of Manning's roughness coefficient for open channels and culvert crossings (Scenarios 4 and 5), while Option 3 focuses on the identification of critical culvert crossings (Scenario 5) using a critical component analysis approach (Johansson & Hassel 2012).
For Option 1, while sediment and solid waste in pipes can be straightforwardly modelled with SWMM software, there is no similar provision for modelling sediments and solid waste in open channel systems, suggesting the need to include this option in future updates of the SWMM software.
Furthermore, modelling the effect of sediment and solid waste deposition in the existing UDSs using Option 2 has been undertaken in a recent study (Mugume et al. 2024b). In the study, sediment and solid waste conditions were modelled using a lower bound of Manning's roughness coefficient of 0.02, representing a clean and well-maintained existing UDS (‘non-failed UDS’), and an upper bound of Manning's roughness coefficient equal to 100, representing partial or full structural failure (‘failed UDS’) due to deposited sediment or solid waste. In Table 4, Scenario 1 in this study is compared with ‘non-failed UDS’ and Scenario 3 with ‘failed UDS’ for the Nalukolongo catchment, with similar catchment characteristics and which drains into the same receiving water body (River Mayanja) as the Lubigi catchment. In the study by Mugume et al. (2024b), flooding was modelled with a coupled 1D–2D model using the same IDF curves (Figure 8), but with a climate change factor of 1.2 compared to 1.3 used in this study. While this current study also included high-resolution rain data with the corresponding water level observations for calibration and validation, no such data was available for Nalukolongo.
Flood volumes from this study, compared with Mugume et al. (2024b), which is highlighted in light grey
Design storm . | Scenario 1 Existing UDS (m3) . | Non-failed UDS (m3) . | Difference in flood volume (%) . | Scenario 3 80% sediment (m3) . | Failed UDS (m3) . | Difference in flood volume (%) . |
---|---|---|---|---|---|---|
T2 | 599,000 | 1,000,000 | −40% | 1,229,000 | 1,200,000 | +2% |
T10 | 1,597,000 | 2,100,000 | −24% | 2,309,000 | 2,500,000 | −8% |
T50 | 2,273,000 | 2,750,000 | −17% | 3,176,000 | 3,100,000 | +2% |
T100 | 2,674,000 | 3,250,000 | −18% | 3,612,000 | 3,500,000 | +3% |
Design storm . | Scenario 1 Existing UDS (m3) . | Non-failed UDS (m3) . | Difference in flood volume (%) . | Scenario 3 80% sediment (m3) . | Failed UDS (m3) . | Difference in flood volume (%) . |
---|---|---|---|---|---|---|
T2 | 599,000 | 1,000,000 | −40% | 1,229,000 | 1,200,000 | +2% |
T10 | 1,597,000 | 2,100,000 | −24% | 2,309,000 | 2,500,000 | −8% |
T50 | 2,273,000 | 2,750,000 | −17% | 3,176,000 | 3,100,000 | +2% |
T100 | 2,674,000 | 3,250,000 | −18% | 3,612,000 | 3,500,000 | +3% |
Note.The non-failed UDS and failed UDS are based on Manning's roughness coefficients of 0.02 and 100, respectively, for all conduits.
In Table 4, a comparison between the two studies is shown. The model in this study simulates smaller flood volumes for Scenario 1 compared to the non-failed scenario in the Nalukolongo UDS, especially for smaller return periods, despite the use of larger Manning's roughness coefficients in this study. Scenario 3 shows similar flood volumes to those reported for the failed Nalukolongo UDS. However, the ratio of flooding loss to surface runoff, as highlighted in this study and presented in Table 4 indicates that the majority of runoff contributes to flooding. This could be attributed to the recent rehabilitation and expansion of Lubigi and Nakamiro drainage channels that resulted in increased hydraulic conveyance capacity in the Lubigi UDS when compared to the Nalukolongo UDS which is yet to be reconstructed. As a result of this, design storms with the same characteristics result in lower flooding impacts in the Lubigi UDS (Scenario 1) when compared to the Nalukolongo UDS (Non-failed UDS Scenario).
In Option 3 (Scenario 5), critical culvert crossings, whose failure contributes to the most significant increase in flooding impacts, were identified. The results suggest that in instances of limited operation and maintenance (O&M) budgets, KCCA could focus the limited resources on targeted cleaning and maintenance of the identified critical culvert crossings and through classification and ranking of the criticality of open channel sections and pipes in the existing urban drainage network (e.g. Simone 2023).
Lastly, the study findings obtained using all the considered scenarios have demonstrated that for the modelled sediment and solid waste deposition scenarios, less water will flow through the culverts, thereby enhancing the resulting backwater effect. The study has also demonstrated that enhancing UDS asset management through improved system cleaning and maintenance initiatives (Scenario 4) results in a considerable reduction of the total flood volume resulting from the high frequency, low-magnitude rainfall events. However, the location of flooding would be shifted to more downstream areas of the catchment.
The study results also point toward the need to combine grey UDS interventions with the installation of innovative BGI options in upstream parts of the catchment, where the rainfall runoff is generated, which could delay the runoff and minimise the residual flooding that occurs during extreme events (Sörensen & Emilsson 2019; Alves et al. 2020; Wang et al. 2023; Mugume et al. 2024b). Additionally, restoring the natural state of the catchment by re-establishing wetlands in the highly developed upstream areas of the catchment and preventing development and informal settlements in the less developed downstream areas could further minimise future flood risk.
Locations vulnerable to flooding
Flooding is more dispersed across the catchment during long-duration rainfall, whereas shorter-duration rainfall results in more concentrated flooding at the three locations: Bwaise, Alice Kaggwa Road, and Kamwokya Kisalosalo Road. In addition, the results of the critical component analysis highlighted three additional locations, i.e. the Kalerwe-Northern bypass, Kawaala, and Hoima Road crossings, as potential flood-prone areas. The modelling study findings agree with a recent rapid flood vulnerability assessment study undertaken in Bwaise and Kalerwe zones located in the Lubigi catchment in which flood-prone areas as well as locations with flow obstructions were geo-located and mapped (Kintu 2021). The study results also agree with a recent news article published by the Uganda Radio Network, which reported an extreme flooding event that led to the loss of lives in Bwaise (Kimumbasa zone) and Kalerwe-Northern bypass (Mukwaya zone) in the Lubigi catchment (Kamali 2024).
Flooded location: Bwaise
The highest volume of flooding occurs at Bwaise (Table 2), which is located downstream in the catchment, receiving water from 4 secondary channels. One possible explanation for the extensive flooding could be a backwater effect (Cappato et al. 2022), caused by high flow velocities upstream of the box culvert crossing on the Nakamiro Channel (location 2, Figure 16) and the sharp angle where the secondary channel enters the main channel.
Running the model with the observed 53.4 mm rain event results in a maximum flow of 20, 16, and 25 m3/s at locations (1), (2), and (3), respectively, (Figure 16). The small increase in flow in location (3) indicates that there could be turbulence and a loss of energy in the junction, causing the water level to rise at location (1), leading to flooding. To further analyse this, developing a coupled 1D–2D model would help capture the water level after flooding occurs. A 2D model – whether a standalone 2D rain-on-grid model or a coupled 1D–2D model – would also represent the flood extent and depths across the catchment area, which a 1D model alone is unable to provide. In addition, the 2D model could be used to assess the effectiveness of BGI or a flood barrier in the crossing in preventing water from flooding the surrounding areas. However, the only available digital elevation model (DEM) for this size of the catchment had a resolution of only 30 × 30 m, and there were high vertical differences compared to the invert levels of the conduits as documented by KCCA. This discrepancy poses challenges in creating accurate 2D models. Mark et al. (2004) recommend a 1 × 1 m–5 × 5 m resolution for urban flood modelling, as roads, sidewalks, and buildings can be captured with this level of detail. Lastly, more complex numerical modelling approaches, such as computational fluid dynamics modelling, could be applied in future research to investigate the influence of channel confluence design on flow conditions and flood risk in the existing UDSs.
Flooded locations: Alice Kaggwa Road and Kamwokya Kisalosalo Road
Alice Kaggwa Road is a culvert crossing consisting of four circular pipes with a head and wing wall at the entrance. In the field, the diameter of each pipe was approximated to be 1 m. Kamwokya Kisalosalo Road has a similar culvert design but with three pipes, instead of four. One reason for the significant flooding in both places could be undersized culvert crossings. Running the model with the rain event RE1 results in an entrance flow of 16 and 25 m3/s and an exit flow of 10 and 7.5 m3/s for Alice Kaggwa and Kamwokya Kisalosalo, respectively. This indicates that both culverts act as bottlenecks, leading to flooding at the entrance of each culvert. This conclusion is strengthened by the fact that Kamwokya Kisalosalo Road has fewer pipes and is experiencing a larger volume of flooding (Table 2) along with a greater reduction in flow compared to Alice Kaggwa Road. It is noteworthy that the KCCA (2016) report describes both crossings as having box culverts. However, on-site observations confirmed they are circular culverts with smaller dimensions. Changing to box culverts with larger dimensions at both locations would potentially reduce flooding at these locations.
CONCLUSIONS AND RECOMMENDATIONS
This research has investigated the impact of sediment and solid waste deposition, combined with extreme rainfall, on urban flooding in the upper Lubigi catchment in Kampala. Using PCSWMM software, a 1D hydraulic model was developed to simulate various scenarios of sediment and solid waste deposition in combination with extreme rainfall events. The collected high-resolution rainfall data and water level observations in combination with field observations were used for model calibration and validation. This is, to the authors' knowledge, the first systematic research attempt to develop a calibrated and validated urban drainage model, considering sediment and solid waste deposition scenarios and using high-resolution rain data and the corresponding water level observations in Uganda.
The study results, obtained using a calibrated and validated 1D urban drainage model, have demonstrated that partial failures caused by sediment and solid waste deposition in the existing UDSs can be reliably modelled with three main approaches: (a) reduction of the cross-sectional areas and adjustment of the invert levels of the modelled culvert crossings, (b) adjustment of Manning's roughness coefficient for open channels and culvert crossings, and (c) use of the critical component analysis methodology. The study findings have also demonstrated that for all modelled sediment and solid waste deposition scenarios, less water will flow through the open channel and culvert cross-sections, thereby enhancing the backwater effects and increasing the risk of flooding in upstream areas.
The study has also demonstrated that enhancing UDS asset management through improved system cleaning and maintenance initiatives can result in a considerable reduction of flooding impacts caused by high-frequency and low-magnitude rainfall events. In addition, the study has underscored the need to combine grey UDS interventions with the installation of innovative BGI options to enhance the general resilience of the existing UDSs.
Based on the study findings, future research focused on (a) Improved estimation of area reduction factors and design storms in cities using relatively high-resolution rainfall and water level measurements or gridded re-analysis satellite rainfall data products, (b) More comprehensive modelling of sedimentation in open channel drainage systems through comparison of the combinations of reduced cross culvert and open channel cross-sections with the corresponding results obtained using adjustment of Manning's roughness coefficients, and (c) Modelling of multi-phase water and sediment flow conditions in open channel drainage systems using computational fluid dynamics modelling approaches, is recommended.
This research addresses a critical research gap and presents a new approach for the evaluation of the effect of scenario combinations of extreme rainfall and sediment and solid waste deposition on pluvial flood risk in data-scarce cities. The presented research can inform decision-makers in the selection and prioritisation of strategies for enhancing urban flood resilience and sustainable water management in cities.
ACKNOWLEDGEMENTS
It is given to the Directorate of Water Resources Management (DWRM), Ministry of Water and Environment, Uganda, and Kampala Capital City Authority (KCCA) for provisioning the data sets used in the study. The insights of three anonymous reviewers are also gratefully acknowledged.
FUNDING
This study was made possible through grants from Åforsk, Miljöfonden, and Sveriges Ingenjörer in Sweden and from the ARUA/U21 Early Career Researcher Collaboration Award 2023.
DATA AVAILABILITY STATEMENT
All relevant data are included in the paper or its Supplementary Information.
CONFLICT OF INTEREST
The authors declare there is no conflict.