Flood Hazard and Risk Assessment in the Mpanga River Catchment Using Integrated Hydrological Modeling and Decision Support Tools

Abstract

Floods increasingly threaten communities and infrastructure in Uganda due to climate variability and land use changes. This study assessed flood hazard, vulnerability, and risk in the Mpanga River Catchment using the Rainfall–Runoff–Inundation (RRI) model integrated with the Analytical Hierarchy Process (AHP). The RRI model showed good performance during calibration (NSE = 0.83) and validation (NSE = 0.71), enabling the generation of hazard maps for different return periods. Results revealed a clear escalation in flood extent with increasing return period, where inundation expanded from about 120.5 km² in the 5-year event to nearly 348.4 km² under the 100-year flood scenario. Vulnerability was evaluated through AHP using nine indicators (Land use, population density, distance to river, elevation, rainfall, slope, drainage density, Total Wetness Index, and soil type); however, only Land Use and population density were retained in the final mapping due to data relevance and weight dominance. Combining hazard and vulnerability produced risk maps that revealed most of the catchment falls under low to moderate risk, with high-risk areas concentrated in upstream urbanized zones. Validation with satellite-derived flood maps confirmed model reliability. Evaluation of mitigation strategies showed dams and channel improvements to be the most effective in reducing flood extent. The study provides a practical framework for flood risk management in data-scarce environments, supporting evidence-based planning and interventions.

1. Introduction

Flooding is among the most frequent and damaging natural disasters globally, with significant human, economic, and environmental impacts [1,2]. In sub-Saharan Africa, climate variability, land degradation, and rapid urban expansion have increased flood susceptibility [3,4]. Uganda faces growing flood risks, affecting over 45,000 people annually and causing losses exceeding $62 million [5], and between 1993–2018, floods had destroyed 65,458 houses [6]. The Mpanga River catchment, located in southwestern Uganda, has seen recurring flood events, including a severe incident in November 2012 that damaged infrastructure and disrupted livelihoods [7]. Studies attribute the increasing frequency and intensity of these events to climate change, deforestation, and unregulated land use practices [8,9,10]. However, despite these threats, local authorities lack spatially detailed flood risk information needed for effective planning, and affected communities remain vulnerable and underprepared.

A wide range of methods have been used to assess flood hazard, from physical hydrodynamic models to semi-quantitative and multi-criteria approaches [4,11]. Hydrological models such as SWAT, HEC-HMS, and RRI are often used to simulate runoff and inundation, while statistical techniques like flood frequency analysis support infrastructure design [12,13]. In parallel, the integration of geospatial data, remote sensing products (e.g., CHIRPS, ERA5), and socio-environmental variables has improved flood mapping in data-scarce regions [14,15]. However, there is limited application of integrated hydrological modeling, spatial vulnerability analysis, and mitigation scenario evaluation in Ugandan river basins such as Mpanga, which constrains evidence-based flood risk management and long-term planning. The Rainfall–Runoff–Inundation (RRI) model, in particular, has shown potential in simulating catchment-wide flood dynamics, including lateral flow and surface–river interactions, across different terrains [16,17]. However, its application at daily resolution often suffers from data uncertainty and model sensitivity, making monthly-scale simulations a viable alternative for long-term hazard zoning and strategic planning [18]. Recent studies also include flood-inundation mapping using satellite imagery, particularly Sentinel-1 SAR data processed via change detection or machine learning methods, which has proven effective for validating hydrological model outputs. For instance, Suab et al. [19] generated flood extent maps in Sabah, Borneo region and Malaysia, using Sentinel-1 and Sentinel-2 imagery and demonstrated strong spatial agreement with modeled flood outputs in low-lying plains. Similarly, several studies, including those in the Brahmaputra and Ganga basins, have applied Sentinel-1 change detection techniques to map inundation and cross-validate model simulation results, showing good agreement in areas with limited data [20,21].

Despite these advances, a key gap remains in the integration of hydrological simulations with spatial vulnerability assessments to produce actionable flood risk maps, especially in East African basins like Mpanga. While studies have applied models such as HEC-RAS and SWAT for flood hazard mapping in Uganda [8,22], few have combined these outputs with systematic spatial vulnerability analysis, such as population exposure or land use pressures [23]. Moreover, most existing works emphasize event-based or annual flood simulations, which are often inadequate for medium- to long-term flood planning in basins with high climatic and topographic variability [24,25]. Schoppa et al. [26] highlight the need to capture long-term flood risk dynamics and changing hydrological patterns beyond individual flood events to support sustainable basin planning and risk-informed decision making. The limited use of monthly scale hydrological modeling further constrains the ability of planners to identify persistent risk zones or assess the long-term benefits of mitigation scenarios. As a result, decision-makers lack the tools to prioritize interventions or evaluate trade-offs across competing strategies, particularly in data-constrained and hazard-prone regions [27,28].

To address this research gap, the present study develops an integrated flood risk assessment for the Mpanga River catchment by combining hydrological simulations from the RRI model with vulnerability analysis using the Analytical Hierarchy Process (AHP). The study (i) simulates flood hazard patterns for multiple return periods; (ii) quantifies vulnerability using land use and population data; (iii) generates composite flood risk maps; and (iv) evaluates mitigation scenarios, including dam construction, channel improvement, and infiltration enhancement. Unlike previous studies, this research uniquely demonstrates the added value of coupling RRI-based hydrological modeling with socio-spatial vulnerability indicators, thereby providing more robust and scalable insights into flood risk hotspots. This study offers a practical framework for flood risk management in data-limited regions exposed to intensifying climate extremes.

2. Materials and Methods

2.1. Study Area

The Mpanga River catchment is situated in southwestern Uganda and forms part of the larger Upper Nile Basin. It spans an estimated area of 5202 km² and is hydrologically subdivided into four major subcatchments: Upper Mpanga (7.4%), Middle Mpanga (22.6%), Lower Mpanga (9.2%), and Rushango (60.9%) [29]. This study focuses on the Upper and Middle Mpanga subcatchments, which together cover approximately 1600 km². These subcatchments include the districts of Kyenjojo, Bunyangabu, and Kabarole in the Upper section, and Kamwenge and Kitagwenda in the Middle section. Geographically, the study area lies between 0°29′06″–0°37′26″ N latitude and 30°25′26″–30°28′48″ E longitude (see Figure 1). The terrain is characterized by pronounced elevation gradients, ranging from 1157 m to 2980 m above sea level. These elevation differences influence runoff dynamics and contribute to varying flood susceptibility across the catchment. Climatically, the region experiences a humid tropical climate with bimodal rainfall patterns. Annual precipitation ranges from 1200 mm to over 3000 mm in wetter years, while average monthly temperatures fluctuate between 27 °C and 31 °C. Relative humidity levels generally vary from 60% to 80%, with higher values occurring during the rainy seasons [8,30]. These climatic and topographic conditions make the Mpanga catchment particularly sensitive to intense rainfall events, which often result in surface runoff, riverine flooding, and land degradation.

2.2. Data

To support the modeling and analysis, a variety of datasets, as shown in Table 1, were obtained from several sources, spanning different spatial and temporal scales. Observed rainfall data for 19 stations with varying periods were available. However, due to significant data inconsistencies that limited their suitability for direct application, they were not considered in this study. Particularly, these datasets exhibited substantial data gaps and missing records across multiple stations, temporal discontinuities with non-overlapping observation periods, inconsistent measurement intervals, and the presence of anomalous values and outliers likely associated with recording errors. The possibility of incorporating partial station records was also considered. However, due to the fragmented nature of the datasets and the lack of sufficient temporal overlap between stations, it was not feasible to construct a consistent and spatially representative rainfall input through interpolation without introducing significant uncertainty. Consequently, ERA5 reanalysis rainfall data were adopted as the primary input for this study due to their spatial consistency, temporal completeness, and proven reliability in hydrological applications within data-scarce regions of East Africa [25,31,32].

Rainfall data were obtained from ERA5 reanalysis rainfall products, which provide hourly gridded records. For this study, daily time step data was used. As shown in Figure 1, the Kampala–Fort Portal Road Station (No. 84212), which lies within the catchment boundary, was selected for this study. Discharge data from this station for the period 2000–2015 (Table 2) was used. Land use/land cover (LULC) information for 2008 was obtained to enable characterization of catchment surface conditions, while a 30-m resolution Digital Elevation Model (DEM) supported terrain analysis. River cross-sectional dimensions (depth and width) were obtained during a field survey conducted in May 2023. Channel width was measured manually using a tape measure, while water depth was determined by inserting a rod into the river and recording the point where the water level marked the rod. Measurements were taken at 30 representative points along the river reach, including straight sections, bends, and locations exhibiting noticeable morphological variation, in order to capture the local range of channel conditions and ensure that potential spatial variability was adequately assessed. However, the observed differences in channel width and depth did not exhibit a consistent downstream trend to support the development of a reliable hydraulic geometry for distributed parameterization within the study catchment. Given the relatively uniform geomorphological characteristics of the modeled river reaches and the absence of established hydraulic geometry, drainage area relationships for the study catchment, the river channel was parameterized as spatially uniform within each modeled reach. The mean values of the measured width and depth were adopted as representative cross-sectional parameters and applied consistently across all corresponding river grid cells in the model. Additional hydroclimatic variables, such as temperature and solar radiation (2000–2015), were obtained from the Climate Forecast System Reanalysis (CFSR) dataset, while the population density data (30 m × 30 m, year 2020) were obtained from the Humanitarian Data Exchange (HDX) [33] portal because it represented the most recent high-resolution, gridded population dataset available for the study area.

Table 1. Summary of the data used in this study.

Date Type	Scale (Period)	Date Source	Description
Daily precipitation (mm)	9.6 km (2000–2015)	ERA5 https://app.climateengine.org/climateEngine	Reanalysis data Obtained on 23 June 2023
Daily observed Discharge	2000–2015	Directorate of Water Resources Management (DWRM)	Stream Discharge values (m3/s)
Digital Elevation Model (DEM)	30 m × 30 m resolution	https://earthexplorer.usgs.gov/	Obtained on 23 June 2023
Land use Land cover (LULC)	30 m resolution (2008)	National Forestry Authority (NFA)	Derived from Landsat 8 OLI imagery (USGS GLOVIS), 180 × 185 km coverage [34]
Soil Data	Scale of 1:5,000,000	Food and Agriculture Organization	Obtained on 23 June 2023.
CFSR	10° × 10° (2000–2015)	https://power.larc.nasa.gov/data-access-viewer/	Temperature, Solar Radiation Obtained on 23 June 2023
Population Density	2020 30 m × 30 m resolution	https://data.humdata.org/dataset/highresolutionpopulationdensitymaps-uga	Obtained on 23 June 2023

Table 1. Summary of the data used in this study.

Date Type	Scale (Period)	Date Source	Description
Daily precipitation (mm)	9.6 km (2000–2015)	ERA5 https://app.climateengine.org/climateEngine	Reanalysis data Obtained on 23 June 2023
Daily observed Discharge	2000–2015	Directorate of Water Resources Management (DWRM)	Stream Discharge values (m3/s)
Digital Elevation Model (DEM)	30 m × 30 m resolution	https://earthexplorer.usgs.gov/	Obtained on 23 June 2023
Land use Land cover (LULC)	30 m resolution (2008)	National Forestry Authority (NFA)	Derived from Landsat 8 OLI imagery (USGS GLOVIS), 180 × 185 km coverage [34]
Soil Data	Scale of 1:5,000,000	Food and Agriculture Organization	Obtained on 23 June 2023.
CFSR	10° × 10° (2000–2015)	https://power.larc.nasa.gov/data-access-viewer/	Temperature, Solar Radiation Obtained on 23 June 2023
Population Density	2020 30 m × 30 m resolution	https://data.humdata.org/dataset/highresolutionpopulationdensitymaps-uga	Obtained on 23 June 2023

Table 2. Discharge data details obtained from DWRM.

Sn	Lat	Long	Stn Code	Name	Start Date	End Date	%Missing
1	0.64611	30.39306	84212	R. Mpanga at Kampala—Fort Portal Road	01-Jan-00	31-Dec-15	0

Table 2. Discharge data details obtained from DWRM.

Sn	Lat	Long	Stn Code	Name	Start Date	End Date	%Missing
1	0.64611	30.39306	84212	R. Mpanga at Kampala—Fort Portal Road	01-Jan-00	31-Dec-15	0

2.2.1. Rainfall Data

ERA5 reanalysis data were utilized to provide comprehensive, consistent, and spatially distributed precipitation inputs for the modeling process. In the study area, a total of 16 ERA5 grid points covering the Mpanga River catchment were selected for the period from 2000 to 2015. This timeframe was chosen to ensure robust and reliable inputs for hydrological modeling, coinciding with the availability of high-quality discharge data. The 16-year period offers sufficient temporal coverage to capture interannual and seasonal variability in precipitation, which is essential for accurately modeling the hydrological processes within the catchment. Developed by the European Centre for Medium-Range Weather Forecasts (ECMWF), ERA5 provides global daily aggregated estimates of atmospheric variables at a resolution of 0.25° × 0.25° [31]. It is extensively used in hydrological applications across East Africa due to its temporal consistency and effective representation of regional precipitation dynamics [25,32]. The daily rainfall data were used as input to the hydrological model to simulate catchment runoff and capture peak flow dynamics. However, due to gaps and inconsistencies in the observed discharge records, the simulated daily outputs were aggregated to monthly time steps for calibration and validation of the model. Aggregating simulated daily discharge to monthly values reduces uncertainties while preserving seasonal hydrological trends, thereby improving model stability and reliability. Due to its relatively coarse spatial resolution, ERA5 rainfall data may smooth localized rainfall extremes, which can affect the accuracy of daily scale hydrological simulations in heterogeneous tropical catchments. The decision to incorporate daily rainfall data enabled the evaluation of peak flow events, while the utilization of monthly data reflects standard practices in data-scarce catchments, where daily scale calibration is often impractical [17,18,35,36].

2.2.2. Discharge

Observed river discharge data were obtained from the Directorate of Water Resources Management (DWRM) under Uganda’s Ministry of Water and Environment for the period 2000–2015, as summarized in Table 2. The RRI model was tested using both daily and monthly time steps to evaluate its performance under different temporal resolutions. At the daily scale, the raw discharge data were first subjected to quality control procedures, including the detection and removal of abrupt spikes and statistical outliers, to ensure reliability in simulating short-term flood dynamics. In addition, the same dataset was aggregated to monthly averages to support simulations at a coarser temporal resolution, which helped smooth out high-frequency fluctuations and better capture long-term hydrological variability. This dual-scale approach allowed for a comparison of model behavior at different timesteps and improved the robustness of the flood hazard assessment. Monthly aggregation, in particular, has been widely applied in data-limited catchments as it reduces short-term noise and enhances stability in calibration and validation. While the RRI model simulations were conducted at a daily time step, calibration at this resolution proved challenging due to uncertainties in both input and observed data. In particular, the coarse spatial resolution of ERA5 rainfall (~0.25°) limits its ability to capture localized, high-intensity precipitation events, which are critical for accurate daily flood simulation. These inconsistencies and potential measurement errors in the observed discharge data reduce the reliability of daily scale calibration. As a result, daily calibration was associated with higher uncertainty, and simulated discharge outputs were aggregated to monthly values to improve robustness and stability in model evaluation [25,37].

2.2.3. Digital Elevation Model (DEM)

Topographic data were derived from the Shuttle Radar Topography Mission (SRTM) DEM with a spatial resolution of 30 m. The SRTM DEM is widely used in hydrological and flood modeling studies due to its global availability, fine resolution, and compatibility with hydrodynamic models [38]. Although higher-resolution local DEMs can improve terrain representation, the SRTM 30 m data have been shown to be reliable for large-scale watershed modeling and topographic corrections in East African environments [17,24]. Furthermore, the DEM was processed to correct sink areas and ensure hydrological consistency using ArcGIS tools (v10.7.1). Flow direction and flow accumulation grids were generated, which facilitated delineation of the subcatchments and stream networks used in the model. The DEM provides the elevation input needed by the Rainfall–Runoff–Inundation (RRI) model to simulate overland and channel flows across the Mpanga catchment.

2.2.4. Land Use/Landcover

Land use and land cover (LULC) information for the Mpanga River catchment was obtained from the National Forestry Authority (NFA), Uganda, to assess spatial and temporal changes relevant to hydrological modeling. The dataset was a classified national LULC map for 2008, with a spatial resolution of 30 m and a cartographic scale of 1:250,000. This dataset included detailed land cover categories such as broadleaved and coniferous plantations, bushland, commercial and subsistence farmland, grasslands, wetlands, water bodies, built-up areas, tropical high forest (both well-stocked and depleted), and woodland. It was selected as the baseline due to its high spatial detail, consistency with local land use definitions, and credibility stemming from national-level field classification. It was also deemed suitable for hydrological modeling given its inclusion of hydrologically relevant classes such as wetlands, forests, croplands, and built-up areas, which influence infiltration and runoff behavior. The dataset was reclassified into a legend consisting of 19 harmonized thematic classes: forest, woodland, grassland, bushland, wetland, open water, seasonal swamp, subsistence farmland, commercial farmland, built-up area, bare ground, rocky outcrops, plantations, industrial area, road infrastructure, urban parkland, institutional area, mining area, and airport. This process involved cross-referencing land cover definitions, grouping similar classes, and aligning categories that influence hydrological processes. The dataset was reprojected to the WGS_1984_UTM_Zone_36N coordinate system and clipped to the Mpanga catchment boundary. Field verification was conducted in May 2023 using GPS-referenced ground truth points and land cover observation data, with a total of 30 ground control points considered for validation. While direct verification of 2008 conditions was not possible due to time constraints, recent observations were compared with areas exhibiting minimal land use change to validate classification stability. Furthermore, several unchanged land cover types were used as benchmarks to verify spatial consistency. Based on the field verification results, the NFA 2008 dataset was found to represent on-the-ground conditions more accurately and consistently, and was therefore adopted for subsequent hydrological modeling and analysis due to its strong alignment with observed land use patterns and relevance to local environmental conditions.

2.2.5. Soil Data

Soil data for the Mpanga catchment were obtained from the FAO–UNESCO Soil Map of the World, accessed via the FAO geospatial data portal on 23 June 2023 [39]. This dataset provides generalized soil classification at a spatial scale of 1:5,000,000 and is available in geographic (lat/lon) projection. The soil data were reprojected to WGS_1984_UTM_Zone_36N and converted to a raster format compatible with the RRI model. To ensure consistency with other input datasets, the soil raster was resampled and reassigned to a 30 m resolution using the bilinear interpolation method in ArcGIS. Although coarse in resolution, the FAO dataset is widely used in regional and continental-scale hydrological studies due to its broad coverage and compatibility with remote sensing and distributed modeling frameworks [40,41]. The soil classification provided essential input for defining hydrological response units, particularly with respect to infiltration capacity and lateral subsurface flow parameters within the catchment.

2.2.6. CFSR

Temperature and solar radiation data for the Mpanga catchment were obtained from the Climate Forecast System Reanalysis (CFSR) dataset, accessed via NASA’s POWER Data Access Viewer on 23 June 2023 [42]. The dataset provides hydroclimatic variables at a 10° × 10° resolution for the period 2000–2015. Although coarse in resolution, CFSR is widely applied in hydrological and climate-related studies due to its global coverage, temporal continuity, and suitability for distributed modeling frameworks [43]. These variables were used for capturing evapotranspiration dynamics and energy balance processes within the catchment.

2.2.7. Population Density

High-resolution population density data (30 m × 30 m) for the year 2020 were obtained from the Humanitarian Data Exchange (HDX) portal. The dataset provides gridded population estimates at national and subnational levels, derived from census and satellite inputs. Despite inherent uncertainties in disaggregated estimates, such high-resolution population data are widely used in vulnerability and risk assessments due to their fine spatial detail and compatibility with GIS-based analysis [44]. These data were particularly useful in identifying exposure patterns and population distribution across the Mpanga catchment.

2.3. Methodology

2.3.1. Rainfall–Runoff–Inundation Model

The Rainfall–Runoff–Inundation (RRI) model is a distributed, two-dimensional hydrological model developed by the International Centre for Water Hazard and Risk Management (ICHARM) in Japan. It was designed to simulate both rainfall–runoff processes and flood inundation simultaneously within a single modeling framework [45]. Unlike traditional hydrological models that treat runoff generation and flood routing as separate processes, RRI integrates surface and channel flow by solving diffusive wave equations on a grid-based mesh. This structure makes the model well-suited for simulating overland flow, channel storage, lateral subsurface flow, and surface–river interactions at the basin scale [16,46]. Figure 2 shows the methodological flow diagram adopted for this study.

The RRI model (Figure 3) was selected for this study due to its suitability for large, data-scarce catchments, its capacity to utilize remotely sensed inputs, and its proven application in tropical and mountainous basins where land use and topography significantly influence runoff dynamics [17,18]. Compared to other models such as SWAT, HEC-HMS, or LISFLOOD, RRI offers an advantage in coupling flood depth estimation with hydrological routing without requiring highly detailed hydraulic geometry. Moreover, it can simulate inundation depth and extent directly from rainfall and land surface conditions, enabling risk mapping with fewer assumptions [15].

The RRI model was configured and run at both daily and monthly time steps to evaluate its performance under different temporal scales over the 2000–2015 period. This approach improves model stability; however, it may smooth peak flows and reduce sensitivity to short-duration extreme events. The short simulation period introduces uncertainty when extrapolating to higher return periods such as 50- and 100-year events and thus should be interpreted as indicative scenarios rather than precise predictions. The spatial resolution of the model was set to 30 m, matching the resolution of the input DEM and land cover datasets. While sensitivity to grid resolution was not explicitly tested, the selected 30 m resolution represents a balance between computational efficiency and spatial detail, consistent with similar studies [48,49,50]. Model inputs included: rainfall data from ERA5, SRTM-based DEM, LULC data, and Soil parameters. Initial model parameters, including Manning’s roughness coefficient, soil hydraulic conductivity, and soil depth, were derived from literature and adapted to local conditions based on catchment characteristics [51,52]. For river channels, Manning’s roughness coefficient values typically ranged between 0.03–0.05, representing natural streams with earth beds and moderate vegetation cover [53,54]. Soil hydraulic conductivity values were extracted from regional soil classifications and prior hydrological studies in similar East African catchments, with estimates ranging between 1 × 10⁻⁶ and 1 × 10⁻⁴ m/s depending on soil texture and structure [40,55]. These parameters provided critical input for simulating infiltration, overland flow, and channel resistance in the Mpanga catchment. The RRI model domain was built in accordance with hydrological flow direction, and river cross-sections were approximated based on DEM-derived stream width and slope corresponding to the modeling period of 2000–2015. Based on the calibrated model outputs, a flood hazard layer was generated to represent the spatial extent and depth of inundation of each month between 2000 and 2015. To derive the design discharges associated with different flood magnitudes, a flood frequency analysis was carried out on the simulated annual maximum peak discharges obtained from the RRI model. First, the annual maxima series was extracted by identifying the highest discharge in each year of the simulation period. These values were then ranked in descending order and assigned plotting positions using the Weibull formula:

$$P={\displaystyle \frac{m}{n+1}},$$

(1)

where P is the exceedance probability, m is the rank of the event, and n is the total number of years of record. The recurrence interval or return period (T) was computed as the inverse of the exceedance probability, i.e.,

$$T={\displaystyle \frac{1}{P}}$$

(2)

To model the statistical distribution of extreme floods, the Gumbel Extreme Value Type I distribution was fitted to the annual maxima discharge series, as it is widely applied in hydrological frequency analysis [53]. The distribution parameters (location and scale) were estimated using the method of moments, and goodness-of-fit was assessed using the Kolmogorov–Smirnov test to ensure reliability of the fitted curve. The test yielded a K–S statistic of D = 0.167 with a corresponding p-value = 0.746, indicating that the null hypothesis (that the data follow a Gumbel distribution) cannot be rejected at the 5% significance level and the results are shown in

Table 3. Kolmogorov–Smirnov Test Results for Gumbel Distribution (Discharge Data).

Parameter	Value
Sample Size (n)	16
K–S Statistic (D)	0.167
p-value	0.746
Significance Level (α)	0.05
Decision	Accept H0

below.

These results suggest that the Gumbel distribution provided an acceptable and statistically consistent fit for the simulated annual maximum discharge series, and is therefore appropriate for flood frequency analysis in this study. Once the probability distribution was established, discharges corresponding to return periods of 5-, 10-, 25-, 50-, and 100-years were extrapolated directly from the fitted distribution curve. These design discharges represent the magnitude of flood events expected to occur, on average, once in the specified recurrence interval.

2.3.2. Model Calibration and Parameter Optimization

The RRI model was calibrated to ensure an accurate representation of hydrological processes in the Mpanga River catchment. Calibration involved adjusting sensitive hydrological and hydraulic parameters to minimize differences between simulated and observed discharge. The calibration process was conducted using observed discharge data from the Kampala–Fort Portal Road station for the period 1 May 2000 to 1 January 2001, while validation was performed for the period 1 May 2002 to 1 July 2003. The calibration followed a manual trial-and-error approach combined with sensitivity analysis, which is commonly used in distributed hydrological modeling in data-scarce environments [17,18,56]. Key parameters influencing runoff generation, infiltration, and channel routing were systematically adjusted until acceptable agreement between simulated and observed discharge was achieved. The most sensitive parameters included Manning’s roughness coefficient for river and slope, soil hydraulic conductivity, soil depth, and lateral subsurface flow parameters. Initial parameter values were obtained from literature and regional studies, and were iteratively refined during calibration. The final calibrated parameter values used in the RRI model are presented in

Table 4. Calibrated RRI model parameters.

Parameter	Description	Initial Range	Calibrated Value	Unit
Manning’s n (River)	Channel roughness coefficient	0.03–0.05	0.035	–
Manning’s n (Slope)	Surface roughness	0.2–0.4	0.30	–
Soil depth	Effective soil depth	0.5–1.5	1.0	m
Hydraulic conductivity	Soil infiltration capacity	1 × 10−6–1 × 10−4	5 × 10−5	m/s
Lateral subsurface coefficient	Subsurface flow control	0.01–0.1	0.05	–
Infiltration parameter	Surface infiltration rate	0.1–0.5	0.25	–

.

2.3.3. Evaluation of Model Performance

The performance of the RRI model was evaluated by comparing the simulated discharge against observed discharge data through both visual and statistical methods. A graphical inspection of the hydrographs was used to assess the overall shape, timing, and magnitude of the flows. In addition, three widely accepted statistical metrics, summarized into

Table 5. Performance metrics of model simulation [58,59].

Evaluation Metrics	Mathematical Equations	Interval	Ideal Value
Percent Bias (PBIAS)	$PBIAS={\displaystyle \frac{{\sum }_{i=1}^n(Xi-Yi)}{{\sum }_{i=1}^n\left(Xi\right)}}\ast 100$	$[-\infty ,+\infty ]$	0
Nash-Sutcliffe Efficiency coefficient (NSE)	$NSE=1-{\displaystyle \frac{{\sum }_{i=1}^n{(Xi-Yi)}^2}{{\sum }_{i=1}^n{(Xi-\overline{X})}^2}}$	$[-\infty ,1]$	1
Coefficient of Determination (R2)	$R^2={\{{\displaystyle \frac{{\sum }_{i=1}^n(Xi-\overline{X})(Yi-\overline{Y})}{{\left[{\sum }_{i=1}^n{(Xi-\overline{O})}^2\right]}^{1/2}{\left[{\sum }_{i=0}^n{(Yi-\overline{Y})}^2\right]}^{1/2}}}\}}^2$	$[0,1]$	1

, were applied to quantify model accuracy for the calibration period (1 May 2000–1 January 2001) and the validation period (1 May 2002–1 July 2003), these periods were chosen to align with the earliest continuous, quality streamflow records available on River Mpanga [57,58,59]. The statistical metrics applied are the Nash–Sutcliffe Efficiency (NSE), the coefficient of determination (R²), and Percent Bias (PBIAS). NSE measures how well the model predictions match the observed data, with values closer to 1 indicating higher predictive accuracy. R² quantifies the strength of the linear relationship between observed and simulated values. PBIAS evaluates the average tendency of the model to systematically overestimate or underestimate flow values. These metrics are commonly used in hydrological modeling to assess performance across different time steps and hydrological conditions [56,58,59,60]. Thresholds for interpreting these metrics were based on established modeling guidelines to ensure a consistent and objective evaluation of model performance.

2.3.4. Flood Inundation Mapping Based on Satellite Images

To validate the RRI model outputs, flood extents derived from Sentinel-2 NDWI composites for the high-flow months of September, October, and November were compared for the years 2012 and 2015. These years were specifically chosen due to their significant flood events in the Mpanga catchment, as indicated by historical records and local reports. For instance, in 2012, River Mpanga experienced substantial flooding, leading to the inundation of Mpanga Market in Fort Portal and the washing away of bridges and other infrastructure. Similarly, in 2015, heavy rainfall resulted in the river bursting its banks, causing widespread flooding and disruption to local businesses. These events underscore the occurrence of extreme hydrological conditions during these years, making them suitable for assessing the model’s performance under high-flow scenarios. GEE offers free access to multi-agency geospatial datasets and enables cloud-based processing, making it suitable for large-scale, time-efficient analysis of surface dynamics [61,62,63]. To estimate surface water extent, the Normalized Difference Water Index (NDWI) was applied to selected cloud-free Sentinel-2 composites. NDWI is a spectral index that enhances water features by contrasting green and near-infrared reflectance values, enabling identification of flooded areas in vegetated or urban landscapes [15]. The NDWI was computed using the following equation:

$$NDWI={\displaystyle \frac{({Green}_{band}-{NIR}_{band})}{({Green}_{band}+{NIR}_{band})}}$$

(3)

where “Green” and “NIR” represent the reflectance values in the green (Band 3) and near-infrared (Band 8) wavelengths, respectively.

Deriving flood inundation maps during the high-rainfall season was challenging due to persistent cloud cover in optical Sentinel-2 imagery. To address this, monthly median composites of Sentinel-2 images were generated in GEE, minimizing noise from cloud shadows and outliers while ensuring reliable temporal coverage, even during peak wet-season months [64]. A threshold value (NDWI ≥ 0.2) was applied to mask out non-water pixels, based on prior studies in similar environments and visual inspection of the Mpanga catchment imagery [15]. The masked NDWI image was exported and converted into binary water/non-water masks, which were further processed to calculate flood extent in hectares across the Mpanga catchment. These NDWI-derived flood maps, while not representing specific events, offered a spatial reference to qualitatively assess model-simulated flood extents in the absence of earlier high-resolution flood records. Flooded zones identified using NDWI were imported into GIS and visually compared with simulated flood layers for plausibility checks [65,66].

2.4. Vulnerability Mapping Method

This study applied Analytical Hierarchy Process (AHP) for developing the flood vulnerability maps for the study area. The AHP methodology is structured in steps: Section 2.4.1, Section 2.4.2 and Section 2.4.3. AHP, developed by Saaty [67], is a widely applied multi-criteria decision-making (MCDM) technique that facilitates structured evaluation of complex decisions through pairwise comparisons and hierarchical structuring. It enables the integration of both qualitative and quantitative criteria and has proven effective in spatial risk assessment, especially where diverse environmental, social, and physical variables must be considered simultaneously [68,69]. AHP was considered an appropriate approach because of its interpretability, transparency, and proven reliability in disaster risk modeling compared to other MCDM approaches such as TOPSIS or weighted linear combination (WLC) [12,70]. AHP is particularly suitable for flood risk assessment because it allows domain experts or literature-based reasoning to assign weights reflecting the relative importance of contributing factors, such as land use, population density, and infrastructure exposure [71,72]. The AHP methodology follows the following steps.

2.4.1. Selection of Indicators

To capture the multi-dimensional nature of flood vulnerability, nine indicators were identified through a literature review and expert consultation. These represent the three key dimensions of vulnerability: Social exposure (population density), Environmental sensitivity (land use/land cover (LULC), soil type, rainfall), and Physical susceptibility (elevation, slope, distance to river, drainage density, and topographic wetness index (TWI)). Each indicator reflects different flood-related processes. For example, population density expresses potential human exposure to flooding, while LULC influences infiltration, runoff, and resilience (e.g., wetlands vs. settlements). Elevation, slope, and distance to river determine flood propagation and inundation potential. Soil type, TWI, and drainage density regulate infiltration and runoff generation. Rainfall represents the intensity of the triggering hazard. Together, these indicators provide a comprehensive basis for assessing vulnerability by linking human exposure with catchment hydrological response. All layers were standardized to a 30m resolution for consistency in GIS analysis.

2.4.2. Weight Assignment Through Pair-Wise Comparisons

The AHP involved constructing a pairwise comparison matrix to evaluate the relative weights of each criterion. Each criterion was rated using Saaty’s 1–9 scale based on its perceived influence on flood risk, where 1 represents equal importance and 9 represents extreme importance of one factor over another, as shown in

Table 6. Prioritization of parameters.

Weight	Description
1	Equal importance
3	Moderate importance
5	Strong importance
7	Very strong importance
9	Extreme importance
2, 4, 6, 8	Intermediate values

. The resulting pairwise comparison matrix for the vulnerability indicators is presented in

Table 7. Matrix for the vulnerability indicators.

Matrix	LULC	Population Density	Distance to River	Elevation	Rainfall	Slope	Drainage Density	TWI	Soil Type
LULC	1	2	9	9	9	9	9	9	9
Population density	0.5	1	8	9	9	9	9	9	9
Distance to river	0.11	0.13	1	2	2	3	7	9	9
Elevation	0.11	0.11	0.5	1	1	2	4	7	8
Rainfall	0.11	0.11	0.5	1	1	2	4	7	8
Slope	0.11	0.11	0.33	0.50	0.50	1	2	3	4
Drainage density	0.11	0.11	0.14	0.25	0.25	0.5	1	2	2
TWI	0.11	0.11	0.11	0.14	0.14	0.33	0.5	1	1
Soil type	0.11	0.11	0.11	0.13	0.13	0.25	0.5	1	1
Total	2.28	3.79	19.70	23.02	23.02	27.08	37.00	48.00	51.00

. Expert knowledge and literature references guided the scoring, giving higher priority to LULC and population density, as they directly influence human exposure and surface susceptibility [72,73].

The pairwise comparison matrix was then normalized, and the principal eigenvector was computed to obtain the final weight vector. A consistency check was conducted by calculating the Consistency Ratio (CR), which evaluates the logical coherence of the judgments. A CR value less than 0.10 was considered acceptable, ensuring the reliability of the resulting weightings [67,74].

$$CR={\displaystyle \frac{CI}{RI}},$$

(4)

where RI = random index varies according to the number of factors used in the pairwise matrix and

$$CI\left(Consistency Index\right)={\displaystyle \frac{{\delta }_{max}-n}{n-1}}$$

(5)

where n represents the number of factors being compared in the matrix and ${\delta }_{max}$ is the highest eigenvalue of the pairwise comparison matrix.

2.4.3. Composite Vulnerability Index Generation

Once weights were established, each indicator map was reclassified and standardized to a 0–1 scale to allow comparison across heterogeneous data types. The weighted overlay analysis was implemented in GIS by applying the AHP-derived weights to each standardized layer.

The final composite vulnerability index was generated as:

$$V={\displaystyle {\sum }_{i=1}^n}(W_i\times X_i)$$

(6)

where W_i is the normalized weight of indicator i, and X_i is the standardized score of indicator i.

This produced a spatially explicit vulnerability map highlighting areas with high human exposure and biophysical susceptibility. The integration of AHP and GIS follows best practices in recent flood risk studies [71,75], demonstrating reliability in data-scarce yet hazard-prone environments.

2.5. Flood Risk Mapping

Following the development of the composite vulnerability index, flood risk mapping was carried out by integrating flood hazard and vulnerability layers within a Geographic Information System (GIS) environment. The process followed a structured overlay approach based on the weighted outputs of the Analytical Hierarchy Process (AHP). The flood hazard layer, generated from RRI model outputs, represented the spatial extent and depth of inundation at a monthly resolution for five return periods: 5-, 10-, 25-, 50-, and 100-year events. These scenarios were used to account for the varying likelihood of flood occurrence and support long-term strategic planning. The flood risk map was then produced by performing a raster overlay of the flood hazard and vulnerability indices using a weighted linear combination (WLC) approach, a method widely adopted in integrated flood risk assessments [69,75]. The flood risk mapping approach adopted in this study is based on the widely recognized formulation:

$$\mathrm{F}\mathrm{l}\mathrm{o}\mathrm{o}\mathrm{d} \mathrm{R}\mathrm{i}\mathrm{s}\mathrm{k}=Flood \mathrm{H}\mathrm{a}\mathrm{z}\mathrm{a}\mathrm{r}\mathrm{d}\times Flood \mathrm{V}\mathrm{u}\mathrm{l}\mathrm{n}\mathrm{e}\mathrm{r}\mathrm{a}\mathrm{b}\mathrm{i}\mathrm{l}\mathrm{i}\mathrm{t}\mathrm{y}$$

(7)

In this context, “hazard” refers to the spatial extent and intensity of flooding derived from the RRI model across multiple return periods. The flood hazard maps were classified into hazard levels of very low, low, medium, high, and very high based on depth thresholds of <1, 1–4, 4–9, 9–16, >16 respectively. Meanwhile, the “vulnerability” layer represents the relative susceptibility of areas to flood impacts and was developed through the Analytical Hierarchy Process (AHP) using two spatial indicators: LULC and population density, due to data relevance and weight dominance. These layers were reclassified into vulnerability levels, weighted based on previous studies in hydrology and GIS-based multi-criteria analysis using AHP pairwise comparisons as shown in

Table 8. Population Density Reclassification.

Population Density (Persons/km2)	Vulnerability Class	Score
<50	Very Low	1
50–150	Low	2
150–300	Moderate	3
300–600	High	4
>600	Very High	5

and

Table 9. LULC Reclassification.

LULC Class	Vulnerability Level	Score
Tropical High Forest	Very Low	1
Woodland, Wetland, Open Water, Coniferous Plantations, Broadleaved Tree Plantations	Low	2
Grassland, Bush, Subsistence Farmland	Moderate	3
Depleted Tropical High Forest, Commercial Farmland	High	4
Built-up Areas	Very High	5

below, and combined using a weighted overlay in ArcGIS to generate a composite vulnerability map.

Finally, flood risk mapping was conducted by raster multiplication of the hazard and vulnerability layers within a GIS environment. This integrated approach allowed for spatially explicit identification of flood risk zones, consistent with recent practices in flood risk modeling [69,75]. This formulation enables the identification of areas where high flood hazard coincides with high vulnerability, thereby prioritizing regions for targeted mitigation and planning.

2.6. Proposed Mitigation Scenarios

To evaluate flood reduction strategies, three mitigation scenarios were simulated using the calibrated RRI model: (i) dam construction, (ii) channel improvement and (iii) infiltration measures. These scenarios were selected due to their suitability for the Mpanga catchment’s high seasonal rainfall, steep topography, and flood-prone river system, and their alignment with flood management practices in similar East African river basins, such as the Nyando and Tana Rivers in Kenya [76,77]. Dams address peak flows from the upper catchment, channel improvements reduce overflow in populated areas, and infiltration measures leverage the catchment’s soil permeability to reduce runoff and support groundwater recharge. Assessing these scenarios individually allows for a clear understanding of their relative contributions to flood mitigation and helps prioritize interventions that are technically feasible and socially acceptable within the Mpanga catchment context.

2.6.1. Dam Construction

A flow control structure was introduced at the designated upstream river grid cells corresponding to the proposed dam location. This was represented in the model as a partial flow obstruction, which reduces downstream discharge transmission and increases upstream water storage. The dam effect was simulated by modifying the flow routing conditions to reflect regulated outflow behavior, thereby reproducing attenuation of flood peaks through reservoir storage dynamics [50].

2.6.2. Channel Improvement

Modifications to channel roughness and cross-sectional geometry were made to represent river training and conveyance enhancement measures. Manning’s roughness coefficient (n) was adjusted within the ranges recommended in the RRI model manual [47], with typical values ranging from approximately 0.01 to 0.80 for overland and floodplain surfaces and from 0.025 to 0.10 for natural channel conditions, depending on land cover and channel characteristics. In this study, n values were reduced within the channelized reaches to reflect improved hydraulic efficiency resulting from channel lining, vegetation clearing, and reduced flow resistance. In addition, channel geometry parameters were modified by increasing effective channel depth and width within the model grid to represent channel widening and deepening, thereby enhancing flood conveyance capacity and reducing overbank flooding [47].

2.6.3. Enhanced Infiltration

The soil infiltration parameter in the RRI model was modified to represent improved catchment infiltration capacity due to land management interventions such as soil and water conservation practices. The infiltration rate was increased from 2.000 to 5.000 d⁻¹, where d⁻¹ denotes per day. In the RRI model formulation, this parameter represents the maximum potential infiltration capacity as a lumped conceptual representation of soil hydrological behavior, rather than a directly measured saturated hydraulic conductivity. The increase, therefore, reflects improved soil structure, reduced surface sealing, and enhanced infiltration potential under conservation practices [47].

3. Results and Discussion

3.1. Hydrological Model Performance

The initial RRI modeling in simulating discharge yielded unsatisfactory performance, with NSE values of only 0.10 during calibration (1 May 2000–1 January 2001) and 0.28 during validation (1 May 2002–1 July 2003), and R² values of 0.44 for both periods (see Figure 4). Correlation coefficients (CC) were also low at 0.66 (calibration) and 0.73 (validation). These weak results reflect the coarse spatial resolution of ERA5 reanalysis data and the sensitivity of the RRI model to short-term rainfall input inaccuracies, which limited its ability to capture peak flows and daily variability accurately. This mismatch between daily rainfall estimates and observed flows was identified as both a limitation of the input dataset and of the modeling framework, forming an important outcome of this study.

Given these limitations, the analysis adopted a monthly scale simulation approach, which showed a substantial improvement in model performance (see Figure 5a). During the calibration period (1 May 2000–1 January 2001), the monthly model achieved an NSE of 0.83, R² of 0.97, and a PBIAS of +8.3%. For the validation period (1 May 2002–1 July 2003), NSE declined to 0.71, R² to 0.89, and PBIAS increased slightly to +12.4% as shown in Figure 5b below. These results fall within the “good” to “very good” performance range for monthly scale simulations according to established criteria [56], and are consistent with current regional modeling practices in data-limited basins [25,28].

The higher performance during calibration reflects the model’s sensitivity to parameter tuning and input consistency. The moderate drop during validation is typical in distributed hydrological modeling, especially where changes in climate or land cover are not explicitly accounted for [37,60]. Nonetheless, all indicators remained within acceptable bounds for monthly runoff simulation.

Recent applications of the RRI model in similar tropical and sub-Saharan African contexts support these findings. For instance, Abdelmoneim et al. [15] reported NSE values of 0.72–0.79 when modeling Nile Basin floods using RRI, while Savitri et al. [46] documented performance ranges of 0.70–0.80 across different rainfall inputs in Southeast Asia. These comparisons confirm the suitability of RRI for flood hazard assessment under limited data availability. Overall, the RRI model performed robustly under observed Mpanga catchment conditions and provides a defensible basis for downstream flood hazard and risk analysis.

3.2. Flood Inundation Extent Across Return Periods

Following hydrological simulation using the RRI model, flood inundation extents were generated for five return periods: 5-, 10-, 25-, 50-, and 100-year flood events (Figure 6a,b,c,d respectively). As expected, the inundated area increased progressively with return period, expanding from 120.5 km² under the 5-year scenario to approximately 348.4 km² (Figure 6a) for the 100-year flood (Figure 6d). This trend is consistent with established hydrological principles, where longer recurrence intervals are associated with more intense precipitation and runoff volumes, resulting in broader flood extents [46,78].

Flood-prone areas were primarily located in low-lying zones, floodplain corridors, and near river confluences, especially in Kamwenge, Kitagwenda, and parts of Bunyangabu districts. These regions share topographic features that increase flood susceptibility, such as gentle slopes, poorly drained soils, and saturated land cover [17,30]. The spatial extent and shape of inundation reflect the combined effects of elevation, stream network density, and land use, emphasizing the advantage of using a distributed model like RRI for spatially explicit floodplain analysis.

Moreover, while flood extent increased with each return period (

Table 10. Percentage of risk levels for the different return periods.

Level	5-Year (%)	10-Year (%)	25-Year (%)	50-Year (%)	100-Year (%)
Very low	9.6	5.8	4.8	4.2	3.7
Low	63.5	49.2	44.7	41.9	39.3
Moderate	19.3	31.6	33.2	34.2	35.7
High	7.4	13.1	16.9	19.2	20.7
Very high	0.1	0.3	0.4	0.5	0.5

and Figure 7), the core flood zones, primarily along the Mpanga River’s mid-reach, remained relatively stable in their geographic footprint. This indicates the persistence of natural flood corridors, which should be prioritized for protection, structural interventions, or land use zoning. As shown in Table 10, the area under “very high” flood risk grows marginally, while areas under “moderate” and “high” risk rise steadily with increasing return period.

3.3. Flood Inundation Extent and Spatial Distribution from Satellite Observations

A visual comparison was made between the RRI-simulated flood extents and NDWI-derived flood surfaces from Sentinel-2 imagery in a GIS environment for September, October, and November for the years 2012 and 2015. The two layers were overlaid using semi-transparent symbology, allowing spatial alignment and differences to be examined, particularly in low-lying floodplains (Figure 8). This qualitative assessment revealed a good spatial agreement, although certain discrepancies were noted in densely vegetated or terrain-shadowed areas, which are well-documented limitations of optical satellite imagery in flood mapping. To complement the visual analysis, a pixel-based comparison was carried out by converting the two datasets into binary flood, non-flood rasters and assessing the degree of spatial overlap. The results indicated that in 2015, the overlap between the RRI-simulated and Sentinel-2 NDWI flood extents was approximately 17.6%, while in 2019 it was about 11.5%. These relatively modest percentages reflect the broader and more continuous flood coverage predicted by the RRI model compared to the more fragmented, channel-focused detection of Sentinel-2 imagery. Nonetheless, the areas of overlap primarily along river corridors and low-lying floodplains suggest that the model is able to capture the main flood-prone zones observed in satellite data, even though differences remain due to methodological and sensor-related factors [19]. The relatively low spatial overlap (11.5–17.6%) between NDWI-derived flood extents and RRI-simulated inundation reflects limitations in using optical remote sensing for validation rather than poor model performance. NDWI from Sentinel-2 is affected by cloud cover, vegetation masking, and mixed pixels, which can lead to underestimation of flooded areas, especially in densely vegetated parts of the Mpanga catchment. In addition, the use of monthly median composites may smooth short-duration flood peaks captured by the RRI model. These differences arise because NDWI captures only surface water at the time of satellite overpass, while the RRI model simulates potential inundation based on hydrological and terrain processes. Consequently, shallow, transient, or vegetation-covered flooding may not be detected in the NDWI results. Despite the limited agreement, NDWI still provides a useful indication of general flood-prone zones, but validation results should be interpreted cautiously. Improved robustness could be achieved by integrating Synthetic Aperture Radar (SAR) data, such as Sentinel-1, which can detect inundation under cloud and vegetation. Studies have shown that incorporating improved hydrological process representation, including infiltration and low-impact development (LID) measures, can further enhance flood simulation accuracy and reduce uncertainty [79,80].

These findings are consistent with recent applications of satellite-enhanced hydrological modeling. For example, Eini et al. [81] demonstrated the effectiveness of satellite soil moisture assimilation in improving flood extent predictions, particularly in mountainous and tropical regions. Likewise, high-resolution DEM evaluations by Zhao et al. [82] confirm that terrain quality is a major determinant of inundation accuracy, a factor carefully addressed in this study using SRTM-DEM preprocessing. Furthermore, Buchanan et al. [83] emphasize that persistent flood corridors tend to show robust spatial patterns over time, even under climate change scenarios. The RRI model’s ability to reproduce these spatially stable high-risk zones enhances confidence in its predictive power for long-term flood zoning. Taken together, these results underscore the suitability of the RRI model and Sentinel-2 validation approach for strategic flood hazard mapping in Uganda and similar data-limited environments. The evidence provides actionable insights for local planners, including zoning flood-prone areas for controlled development or prioritizing them for structural interventions like dykes or retention basins.

3.4. Flood Vulnerability Assessment

The pairwise comparison judgments used in the AHP framework were derived from a synthesis of expert knowledge reported in previous peer-reviewed studies in hydrology and GIS-based multi-criteria analysis. These studies were conducted by domain experts who have extensively worked on flood vulnerability and risk assessment, and they consistently provide convergent recommendations on the relative importance of socio-environmental exposure factors in flood vulnerability modeling [84,85,86,87]. The analysis produced a Consistency Ratio (CR) of 0.098, which is within the acceptable threshold of 0.1 [67], confirming that the judgments were logically coherent. Among the indicators, land use/land cover (weight = 0.6) and population density (weight = 0.3) (see

Table 11. AHP-derived weights for flood vulnerability indicators.

Factor	Weight
Land use/land cover (LULC)	0.600
Population density	0.300
Distance to river	0.0364
Elevation	0.0210
Rainfall	0.0210
Slope	0.0105
Drainage density	0.0053
Topographic Wetness Index (TWI)	0.0031
Soil type	0.0027

) dominated the weighting scheme, indicating their critical role in shaping flood vulnerability across the study area, consistent with findings from earlier East African studies [30]. Although additional indicators such as distance to river, elevation, rainfall, slope, drainage density, TWI, and soil type were assessed during the AHP, to preserve the multidimensional structure of vulnerability, their normalized weights were collectively less than 10%. A sensitivity inspection revealed that including these very low-weight variables in the weighted linear combination altered the composite vulnerability index by less than 3% at the pixel level and did not significantly modify the spatial ranking of high- and low-vulnerability zones. This indicates that the vulnerability model is strongly driven by socio-environmental exposure factors rather than secondary geomorphological controls within this catchment. Accordingly, the GIS-based overlay analysis incorporated only LULC and population density, producing a vulnerability surface (Figure 9) that highlights spatial variations in exposure to flood hazards.

Figure 9 shows that the middle subcatchments, especially in Kamwenge and Kitagwenda districts, had the highest vulnerability levels. These areas are characterized by dense settlements, unregulated development in riparian zones, and extensive agricultural encroachment. Conversely, the upper catchments, such as Kabarole District, exhibit lower vulnerability due to their forest cover, steeper slopes, and lower population density. These patterns are consistent with findings from other regions in sub-Saharan Africa. For instance, Pimenta et al. [88] conducted a flood susceptibility mapping study in an urban Amazon context, where the AHP-GIS results showed that areas with high impermeable surfaces and urban development characteristics were concentrated in the high-susceptibility zones. This supports the idea that urban expansion and human-environment interactions influence flood vulnerability patterns, not just physical terrain factors such as slope or elevation. Similarly, Fox et al. [73] emphasized the growing importance of vulnerability-adjusted flood risk mapping, showing that social exposure can outweigh physical hazard in shaping community-level flood outcomes.

The composite flood risk map, produced by overlaying the vulnerability surface with RRI-derived hazard layers (5–100-year return periods), classified areas into five risk categories: very low, low, moderate, high, and very high. Overall, 53.1% of the catchment falls under moderate risk, while 16.3% and 3.8% of the area were classified as high and very high risk, respectively. The most exposed zones align with flood corridors in Kamwenge and Kitagwenda, where flood hazard intersects with demographic pressure and unregulated development. These results are not only theoretically consistent but also practically important. Studies such as Madi et al. [70] have demonstrated how spatially explicit vulnerability-weighted risk maps can inform targeted disaster preparedness programs in similar semi-rural African watersheds. Moreover, Harshasimha & Bhatt [72] argue that AHP-based risk assessments are particularly well-suited for regions lacking detailed hydrological records but facing mounting socio-economic exposure to flooding.

Nonetheless, the current vulnerability layer does not yet incorporate critical socioeconomic dimensions, such as income levels, education, housing quality, or access to emergency services. As emphasized by Fox et al. [73] and Beshir & Song [89], these attributes significantly shape community resilience and recovery capacity, and their absence may underestimate vulnerability in informal or marginalized settlements. Future work should aim to incorporate household survey data, census information, or participatory vulnerability scoring to capture these more nuanced dimensions. In summary, the AHP-GIS vulnerability mapping approach has successfully identified spatial patterns of flood exposure in the Mpanga catchment. By combining it with model-derived hazard data, the study offers a rigorous, policy-relevant framework for prioritizing flood risk reduction across districts. The maps generated can support strategic planning for flood-resilient infrastructure, land use regulation, and early warning systems in the most affected areas.

3.5. Evaluation of the Proposed Mitigation Measures

To explore viable flood mitigation strategies, three distinct intervention scenarios were simulated using the calibrated RRI model under the 100-year return period condition. These represent a range of structural and nature-based solutions aimed at reducing flood extent and depth, particularly in the vulnerable upper and middle Mpanga subcatchments.

3.5.1. Scenario 1: Channel Section Improvement

This structural intervention involved widening and deepening selected river channels to enhance flow conveyance. The modification was applied to known hydraulic bottlenecks between upstream and midstream reaches. Model simulation results for the 100-year return period (Figure 10a) showed a 79.0% reduction in flood extent under this scenario, particularly within upstream valleys (Figure 10b). This significant decline demonstrates the effectiveness of strategic channel rehabilitation in steep, confined catchments, echoing findings from similar East African contexts [25].

Despite the strong performance, such hard-engineering approaches often require significant capital investment, routine maintenance, and careful environmental consideration. The benefits of channelization can diminish over time if upstream sediment loads or encroaching land use are not concurrently managed.

3.5.2. Scenario 2: Enhanced Infiltration as a Nature-Based Solution

This scenario modeled a nature-based intervention aimed at promoting infiltration and reducing runoff. The infiltration rate for the 100-year return period (Figure 11a) was increased from 2.000 d⁻¹ to 5.000 d⁻¹ to mimic the effects of reforestation, contour farming, or infiltration trenches. The model showed a modest 18.0% reduction in flood extent under this scenario (Figure 11b). Although less impactful in reducing peak inundation, the approach offers co-benefits for groundwater recharge, soil conservation, and ecosystem restoration.

This aligns with findings by Calow et al. [90], who demonstrated that nature-based solutions (NBS) offer long-term resilience when integrated into watershed management frameworks. However, the lower efficacy here may be attributed to shallow soils and steep slopes, which limit infiltration capacity, highlighted as a recurring constraint in other tropical highland basins [17]. Furthermore, recent studies highlight that infiltration-based mitigation measures, such as low-impact development (LID) and permeable surfaces, can significantly reduce surface runoff and flood peaks while improving baseflow [91]. These approaches enhance the natural hydrological processes by increasing soil water storage and delaying runoff generation, which ultimately reduces flood intensity and improves watershed resilience. The enhanced infiltration scenario adopted in this study aligns with these findings and provides a sustainable approach for long-term flood risk reduction.

3.5.3. Scenario 3: Dam Construction in Strategic Upstream Locations

This scenario simulated the placement of retention dams at two high-risk upstream junctions for the 100-year return period (Figure 12a). The goal was to regulate peak flows, reduce downstream inundation, and temporarily store stormwater during extreme rainfall events. The model showed the highest flood reduction, with a decline of 80.3% in inundated area and flood depths in midstream zones dropping from 0.3–0.7 m to below 0.1 m (Figure 12b). These results confirm the hydraulic efficiency of upstream retention in moderating flood propagation, especially in multi-tributary catchments. These findings are supported by Abdelmoneim et al. [15], who found that multi-purpose reservoirs substantially mitigate peak discharge and reduce downstream hazard exposure, especially under increasing climate variability.

As summarized in

Table 12. Percentage reduction in floods in the proposed scenarios.

Scenario	Description	Flood Reduction (%)
Scenario 1	Channel improvement	79.0
Scenario 2	Enhanced infiltration	18.0
Scenario 3	Dam construction	80.3

, channel improvement and dam construction yielded the highest performance, both reducing flood extents by nearly 80%. Nature-based infiltration strategies, though less impactful in isolation, offer critical complementary value and should not be disregarded, particularly in low-gradient zones and as part of integrated catchment management plans. Interestingly, the upstream sub-catchments yielded the largest benefits across all scenarios, reinforcing the idea that intervening early in the flood path has the greatest potential for downstream protection [18,25].

While the technical effectiveness is clear, the feasibility and sustainability of each scenario must be carefully considered. Dam construction, for instance, may raise issues of land acquisition, ecosystem disruption, or sedimentation, while channel improvements often face high costs and infrastructure durability challenges. Moreover, social acceptability and institutional capacity are critical to implementation. As Fox et al. [71] and Beshir & Song [89] highlight, community involvement and alignment with local governance structures are essential to ensuring that mitigation infrastructure is not only installed but also maintained and integrated into long-term planning. Finally, the lack of socioeconomic variables in the vulnerability index remains a limitation. Integration of equity-based considerations, such as housing quality or poverty levels, would further sharpen the prioritization of mitigation strategies, especially in informal or underserved communities.

4. Conclusions

This study presented an integrated flood risk assessment for the Mpanga River catchment by combining hydrological simulations from the RRI model with AHP-based vulnerability mapping. Daily scale modeling remains valuable for representing hydrological processes; however, its application in calibration is constrained by data limitations, and results should be interpreted with caution. The RRI model performed reliably under data-limited conditions, with NSE values of 0.83 (calibration) and 0.71 (validation), validating its suitability for long-term hazard mapping. Flood hazard maps across multiple return periods revealed persistent risk zones along floodplains and river confluences, especially in Kamwenge and Kitagwenda. The vulnerability assessment identified socio-environmental factors, particularly land use and population density, as key drivers of risk. When overlaid, these layers revealed that over 20% of the catchment faces high or very high flood risk. Three mitigation scenarios were tested. Dam construction and channel improvements each reduced flood extents by nearly 80%, while enhanced infiltration achieved an 18% reduction. The results highlight the effectiveness of structural measures and the complementary value of nature-based solutions, especially when targeted in upstream zones. Beyond hazard reduction, these findings have direct implications for climate change adaptation and local policy. As hydroclimatic variability is expected to intensify under future climate projections, the identified high-risk zones should guide land use zoning, floodplain protection, and climate-resilient infrastructure design. Integrating these spatial outputs into district development plans and disaster risk reduction strategies can help limit settlement expansion in flood-prone areas and prioritize investments in drainage upgrades, ecosystem restoration, and early warning systems. Although the study faced data constraints, particularly in socioeconomic variables and daily event validation, it provides a practical, evidence-based framework for risk-informed planning in flood-prone catchments. Future work should integrate finer-scale social data, explore cost–benefit dimensions of interventions, and support the development of early warning systems. Ultimately, a layered flood management strategy that balances structural, ecological, and social approaches will be key to building resilience in the Mpanga River catchment and similar contexts.