3.1. Remotely sensed and gridded data

3.1.1. Digital Elevation Model (DEM).

For topographical analysis of the study area, a Shuttle Radar Topography Mission (SRTM) DEM with 30-m spatial resolution was obtained from the data portal of the United States Geological Survey (USGS) http://earthexplorer.usgs.gov/ [40]. It was used to delineate the study area for hydrodynamic modelling of river discharge. It was further treated to extract layers of AHP parameters such as elevation zones, slope, Topographic Wetness Index (TWI), drainage density, distance from river, etc. [11]. The following paragraphs provide a brief overview of all the parameters obtained from the DEM.

Elevation is regarded as one of the most important parameters in flood risk assessment since it influences water runoff velocity, flow depth, and drainage patterns, and hence flood distribution [41]. For this study, the hydrologically corrected DEM was used for geospatial analyses to evaluate flood hazards based on topographic features. Since lower altitudes are often more vulnerable to flooding, elevation was first extracted and divided into five different elevation zones to assess the difference in terrain elevation [42] across the study area.

Slope was derived from the study area DEM to assess terrain steepness, which controls the direction and speed of surface runoff. Steep slopes accelerate runoff, reduce infiltration, and quickly route water towards downstream floodplains [41]. In contrast, gentle slopes retain water longer, enhance infiltration, and are more prone to flooding. Thus, slope strongly influences both flood likelihood and flow dynamics, making it a key flood hazard parameter [42].

Drainage density () was derived from the stream network of the catchment area by calculating the total stream length per unit area (Equation 1), reflecting how closely spaced the rivers and streams are within the catchment or basin [9]. Areas with higher generally have quicker surface runoff, shorter lag times and concentrated flooding in the downstream areas or floodplains [3]. Hence, it is considered a significant indicator in flood risk assessment.

(1)

where = drainage density (km/km²), L = total length of streams (km), and A = basin area (km²).

The Topographic Wetness Index (TWI) was derived from the DEM, which determined the spatial distribution of terrain-driven soil moisture [43] by integrating the slope and the upstream contributing area of the SRB. Areas with higher TWI values indicate greater water saturation due to poor drainage, whereas lower values indicate well-drained areas. Equation 2 was used by ArcGIS Pro for determining TWI as given below.

(2)

where A_s = upslope contributing area (m² per unit contour length), and β = slope angle (radians) [42].

The stream polyline shapefile obtained after treating the DEM was used to calculate the distance from the river using the Euclidean Distance tool of ArcGIS Pro. This helped in categorizing the areas based on their degree of closeness to the main river channel and other small streams, which is a crucial component of flood hazard [3,9].

3.2. Demographic and infrastructure data

4.1. AHP–GIS integrated conceptual framework for flood risk assessment

4.1.1. Analytical Hierarchy Process (AHP) for flood risk assessment.

In this study, AHP was employed to systematically evaluate the relative importance of factors contributing to flood risk by adopting the hierarchical steps (i to v as follows) for flood risk assessment mentioned in different studies [9,11].

1. (i) The goal was defined (i.e., flood risk map), and the decision hierarchy was structured for organizing the decision elements into levels as shown in Fig 2, i.e., Level 1: Goal (flood risk map), Level 2: Criteria (hazard and vulnerability maps), and Level 3: Sub-criteria (parameters such as precipitation, slope, drainage density, LULC, soil, etc.).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2.

Decision hierarchy for flood risk assessment using AHP in the study area [9,11,41].

https://doi.org/10.1371/journal.pclm.0000915.g002

1. (ii) To achieve the goal (Level 1) and to meet the criteria (Level 2), the significance of each parameter (Level 3) was determined by using Saaty’s 1–9 scale (S1 Table). The matrix-based pairwise comparisons of parameters of hazard and vulnerability (S2 Table and S3 Table) were achieved using Equation 3. This helped in the hierarchical arrangement of parameters after being assigned a weight based on the relative influence and importance of one over another as per the expert opinion and recommendations from the literature [9,41,49].

(3)

Where is the priority weight and represents the relative importance of parameter over j.

As per this systematic hierarchy process, the variation in the significance of each parameter in the context of flood hazard and vulnerability is shown by the derived weights in S2 and S3 Tables. It indicates that precipitation with 27% weightage is the most significant hazard parameter, followed by elevation (23%), distance from the river (19%), slope (11%), TWI (8%), drainage density (8%), and soil (4%). On the other hand, when considering vulnerability, population density appeared to be the dominant factor with a weightage of 58%, followed by LULC at 31% and distance to road at 11%, indicating a strong relationship between the demographic situation and flood vulnerability.

1. (iii) The calculation of AHP weights was carried out in a single integrated process. The eigenvector (Vp) for each parameter was first obtained through Equation 4, and these values were then used to compute the weightage coefficient (Cp) using Equation 5, making sure that the sum of all coefficients remained equal to 1 [9]. After preparing the pairwise comparison matrices for both hazard and vulnerability, each matrix was normalized (S4 and S5 Tables), according to Equation 6, where is the number of parameters being compared (i.e., 7 for hazard and 3 for vulnerability in this study). Further, the priority weights for each parameter were then derived by averaging the normalized values across each row, given by Equation 7, where represents the general formula for calculating the weight of the i^th factor. Once computed, these become the individual weights of parameters (w₁, w₂, ….. w_n), and by averaging each row of the normalized matrix, the complete priority vector (C) was calculated using Equation 8.

(4)(5)(6)(7)(8)

1. (iv) The overall priority (D) was calculated by multiplying each column of the original matrix by the corresponding value in the priority vector (C). Each obtained D was then divided by the corresponding value in the priority vector to determine the rational priority (E).
2. (v) The degree of consistency in the comparison matrices for hazard and vulnerability parameters was evaluated by obtaining the maximum eigenvalue (λ_max), Consistency Index (CI) and Consistency Ratio (CR) as shown in S6 Table, using Equations 9, 10 and 11, respectively.

(9)

These tests helped ensure the reliability of judgements [52] with acceptable CR ≤ 0.1 (S6 Table). As per rule, the comparisons were reconsidered and adjusted to detect the discrepancies in some cases.

(10)(11)

where λ_max is the maximum eigenvalue of the comparison matrix.

The Random Index (RI), shown in S7 Table, represents the average consistency index obtained from pairwise comparison matrices of different sizes [14]. The consistency assessment parameters (eigenvalue, CI and CR) given in S6 Table demonstrate that AHP-based pairwise comparisons made for both flood hazard and vulnerability parameters in the study area are logically sound and statistically valid. The CR value of 0.07 for hazard assessment and 0.01 for vulnerability assessment demonstrate the strong agreement among the selected parameters [9,11,14]. In the case of vulnerability, the near-perfect consistency indicates that population density, settlements, extensive agricultural lands, and compromised or destroyed evacuation routes exacerbate flood vulnerability in the study area. These statistics also clearly demonstrate the reliability and credibility of the hazard, vulnerability, and risk maps developed using AHP-generated weights of parameters for this study [41].

4.1.2. Application of GIS for spatial analysis and mapping.

The hazard and vulnerability parameters were selected based on their relevance to flood generation and exposure in the Swat River Basin. Hazard parameters (precipitation, elevation, slope, distance from river, drainage density, TWI, and soil type directly) influence flood magnitude and spatial distribution. On the other hand, vulnerability parameters (population density, land use/land cover, and distance to roads) capture human exposure and susceptibility. Class thresholds for each parameter were determined using a combination of statistical methods (using Jenks natural breaks in ArcGIS Pro), literature references, and expert judgement, ensuring that each class should reflect the spatial variability that is meaningful and reproducible for flood risk mapping.

The raster layers of each parameter of hazard and vulnerability were obtained as presented in Fig 3 showing the spatial distribution of parameters in the study area. These were then reclassified into five distinct classes ranging from very low to very high as shown in Table 1. Fig 3 and Table 1 present the classification of each parameter with class range, flood probability and class rating/ranking for the AHP–GIS framework. For precipitation (July-August), the 134–159 mm range represents the very high hazard class, whereas the 82–100 mm range corresponds to the very low hazard class. In terms of elevation, the 575–1168 m range represents the very high hazard class, while 3228–4851 m corresponds to the very low hazard class. Similarly, for the ‘distance from river’ parameter, the 0–1037.2 m range is very high hazard class, whereas 6223.4–12755 m range is classified as a very low hazard. For slope, values between 0° and 9.47° represent the very high hazard class, while slopes of 38.52°–71.72° fall under the very low hazard class.

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Class-wise ranking of the hazard and vulnerability parameters.

https://doi.org/10.1371/journal.pclm.0000915.t001

Download:

• PNG
  larger image
• TIFF
  original image

Fig 3. Spatial distribution of flood hazard (a-g) and vulnerability (h-j) parameters for the study area.

https://doi.org/10.1371/journal.pclm.0000915.g003

To identify saturated land surfaces capable of generating overland flow after rainfall, the TWI was classified into five ranges (Fig 3), where the range 14.02–24.46 represents a very high hazard class and 2.27–5.93 indicates the very low hazard class. Likewise, drainage density was categorized into five classes (Fig 3), with the 0–0.08 km/km² class corresponding to the very low hazard class and 0.522–0.82 km/km² representing the very high hazard class. Furthermore, soil texture of the study area was classified (Fig 3) into four types: i.e., clay, loam, silt loam, and sandy loam, ranked according to their flood vulnerability from highest to lowest, respectively.

Furthermore, considering the flood vulnerability, population density in the study area was also categorized into five classes, as shown in Table 1. The class with very low population density was ranked as 1 (32–598 people/km²), while very high population density (6823–14464 people/km²) class was ranked as 5. Similarly, for the distance to the road, the range 0–2049 m was ranked as very high hazard class and 14343–24082 m corresponds to the very low hazard class. Considering the significance of land management and surface characteristics in altering the runoff dynamics and shaping the flood vulnerability, the LULC of the study area was divided into five categories, such as built-up area, cropland/grassland, forest, snow/ice, and water body, and were ranked as 5, 4, 3, 2, and 1, respectively.

After classifying each parameter layer, the final weights obtained for each parameter through AHP (S2 and S3 Tables) were then integrated within a GIS framework using ArcGIS Pro to obtain hazard and vulnerability raster/maps (Fig 3) with different levels (very low to very high) of hazard and vulnerability (Table 1) [10]. These maps were then combined to produce risk maps for flood risk assessment using an index-based model, where flood risk is defined as a function of hazard and vulnerability, as expressed in Equation 12 [13]. In this study, the vulnerability component incorporated indicators of exposure, including population density, LULC, and distance to road. Therefore, flood risk was assessed as the interaction between flood hazard and vulnerability of exposed elements.

(12)

By adopting this approach, a composite flood risk map was produced that spatially represented the combined influence of hazard and vulnerability parameters. The final risk values were then reclassified into distinct classes ranging from very low to very high, enabling the identification of the areas at high to very high risk to prioritize them for flood mitigation and management. Hence, the AHP guaranteed scientific accuracy was used to assess the significance of each criterion, while GIS enabled spatial display and overlay of the criteria [9,41].

4.2. River flood inundation assessment in the high risk zone of the study area using HEC-RAS model

For river flood inundation assessment, a 2D unsteady flow analysis was performed on the high risk region of the study area (identified through the GIS–AHP approach) using HEC-RAS 6.7 Beta 2 hydraulic modelling software, developed by the U.S. Army Corps of Engineers (USACE) obtained from the official website https://www.hec.usace.army.mil/software/hec-ras/download.aspx [20]. Fig 4 presents a general layout of the working principle for the HEC-RAS flood model.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 4. General layout of the working principle for the HEC-RAS flood model.

https://doi.org/10.1371/journal.pclm.0000915.g004

For hydraulic [54] modelling, the delineated and hydrologically corrected SRTM-DEM (30-m) was used for terrain construction, while the projection file for UTM Zone 43N [55] was assigned to RAS Mapper to ensure spatial correctness and compatibility for smooth geospatial data representation and processing. For intensive flood mapping, a 2D flow area was delineated covering the Chakdara downstream reach up to the confluence of the Panjkora and Swat Rivers, defining the simulation domain (area of interest, AOI), as shown in Fig 5. The upstream and downstream conditions are explicitly defined at the domain spans ~20 km along the river corridor. Breaklines were incorporated to represent the main channel alignment and guide flow distribution within the 2D domain.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 5. Spatial extent of the 2D HEC-RAS model domain, showing the upstream and downstream boundary condition (BC) lines (red), breaklines (black), and the delineated 2D flow area (shaded), exported and visualized in a GIS environment over a georeferenced satellite basemap.

https://doi.org/10.1371/journal.pclm.0000915.g005

Boundary conditions were assigned upstream by creating a flow hydrograph using the mean flow of July 2007 (the base-year) for calibration and the peak flow of the flood year 2010 for validation. The boundary condition for downstream was assigned using the ‘Normal Depth’ approach, with a slope value of 0.001 to ensure a consistent water level downstream. This approach is appropriate because the downstream reach of the SRB is characterized by relatively steep terrain, where flow is mainly governed by channel slope and gravity. The model reach is ~ 20 km long, and no additional inflows occur near downstream boundary. Under such conditions, the flow is generally free flowing with minimal influence from downstream controls. Therefore, the normal depth assumption provides a reasonable and physically consistent representation of flow conditions at the model outlet. Riverbank lines were established to outline the channel limits for confining the water flow [18,56]. These boundary conditions enabled the model to accurately replicate water flow velocity, calculate water surface elevations, and determine flow depth [24]. For the 2D simulation procedure, a mesh with a 100 × 100 cell size was established to accurately simulate dynamic flow patterns [19].

HEC-RAS computes flow hydraulics based on conservation of mass and momentum, represented by the Saint-Venant’s (SV) equations, while the flow resistance exerted by the channel and floodplain is characterized by using Manning’s roughness coefficient. Detailed descriptions of governing equations and computational procedures are available in the HEC-RAS Technical Reference Manual and related literature [19,57–59].

The HEC-RAS model was successfully calibrated using river flow data from the Chakdara [44] gauge station for the year 2007, i.e., the base-year for our previous study on SRB [54], in which the snowmelt runoff model (SRM) was used for modelling the historical and future streamflow dynamics between 2005 and 2015. Manning’s roughness coefficients (n-values) were applied to different river sections [58,59]. During calibration, the lowest volume accounting error (0.103804 x 10^-4%) was achieved at n = 0.12 after several trial simulations using different Manning’s n values (Table 2). This value was adopted for further modelling because it provided the most stable performance, showing the fewest variations between the observed and simulated flow volumes. This calibration step was crucial for enhancing the model’s reliability and ensuring an accurate representation of the river’s hydraulic behavior [16,19].

Download:

• PNG
  larger image
• TIFF
  original image

Table 2. Estimation of Manning’s n values during the calibration process of HEC-RAS over year-2007 for the study area.

https://doi.org/10.1371/journal.pclm.0000915.t002

The model was validated by simulating the observed peak flow of July 2010, using the same Manning’s n value (0.12), with a 0.122920 x 10% volume error. Simulation results, such as flow depth, velocity, and water surface elevation (WSE) were accessed via the RAS Mapper.

Since the observed hydrograph at the Chakdara gauge station was used as the upstream boundary condition, direct comparison between observed and simulated discharge at this location is not meaningful. Furthermore, the absence of downstream gauge data limits the possibility of hydrograph-based validation within the model domain. Therefore, model performance was evaluated using volume conservation and sensitivity analysis of Manning’s roughness coefficient. The low volume error indicates good numerical stability and mass conservation rather than implying perfect agreement with observed discharge.

The calibrated and validated HEC-RAS model was used to simulate river discharge and map future inundation under the mid- and late-21st century seasonal RCP scenarios, specifically 2.6, 4.5, and 8.5, which were acquired from our previous study [44]. Su et al. (2016) [60] statistically downscaled these scenarios for the Indus River Basin, considering the expected increases in temperature and precipitation during the mid (2046–2065) and late (2081–2100) 21st century (S8 Table). Rising temperatures can accelerate snow and glaciers melt, which may increase flood risk, especially with intensified monsoon rainfall. Moreover, a predicted decline in winter and spring precipitation in such mountainous areas could further put strain on water resources [19,60]. In the RAS Mapper, the simulation plans were made for the base-year, flood year, and RCP scenarios, followed by the visualization of the model’s outputs (i.e., velocity, depth, and water surface elevation) under each simulation.

To assess the potential inundated area, two methods were employed by adopting the different approaches of flood inundation analysis from various studies [19,61,62].

Method 1: Spatial Analysis using HEC-RAS generated Depth Raster: For each RCP scenario, the HEC-RAS generated depth raster (representing the inundation) was subsequently imported into ArcGIS Pro for spatial analysis [61]. It was overlaid on the designated 2D Flow Area (i.e., Area of Interest-AOI), and then by using the ‘Raster Calculator’, the ‘total inundated area’ (i.e., area of depth raster) was calculated using Equation 13, while the ‘percentage of AOI inundated’ was calculated using Equation 14.

(13)

(14)

Method 2: Spatial analysis using River channel Polygon: A representative polygon (~31.60 km²) of the Swat River was created by observing channel width between 2005 and 2010 through Google Earth Pro. It was then overlaid on a depth raster using ArcGIS Pro to separate the in-channel flow from overbank flooding. This helped in identifying the ‘specific inundated area’ extending beyond the representative polygon of the river and was calculated using Equation 15 and 16.

(15)

(16)

This approach distinguished normal flow from overbank flooding, offering a clearer picture of flood hazard under climate change scenarios [25].

5.1. Spatial distribution and assessment of flood hazard, vulnerability, and risk using GIS–AHP approach

The spatial distribution of flood hazard, vulnerability, and risk in the Swat River Basin was assessed using an integrated GIS–AHP approach. Fig 6 shows the spatial distribution of the flood hazard and vulnerability across SRB under five distinct classes, while Table 3 shows the spatial extent distribution of flood hazard, vulnerability, and risk in the study area.

Download:

• PNG
  larger image
• TIFF
  original image

Table 3. Spatial distribution of flood hazard, vulnerability and risk in the study area derived from GIS–AHP analysis.

https://doi.org/10.1371/journal.pclm.0000915.t003

Download:

• PNG
  larger image
• TIFF
  original image

Fig 6. Spatial distribution of the (a) flood hazard and (b) flood vulnerability in the study area.

https://doi.org/10.1371/journal.pclm.0000915.g006

It is found that moderate hazard dominates the basin (~47.9%, ~ 1641.8 km²), which is widely distributed across mid-elevation valleys and foothill regions. Moreover, high hazard zones ~34.4% (~1180.3 km²) are concentrated along main Swat River channel, where low elevation and flat floodplain topography increases susceptibility to inundation. The prominent areas identified in high hazard zones were Chakdara, Syedabad, Fatehpur, Madyan, Bahrain, Matta, Khawazakhela, Amandara, Batkhela, Darmai, Dadahara, and Barikot. In contrast, low to very low hazard zones (~19.9%) are mainly confined to the upland and high-elevation mountainous areas in the north of the study area, where steep slopes limit the water accumulation; hence, they are prevented from flooding.

Although very high hazard zones cover only ~2.9% (~98 km²) of the study area, they include highly populated areas such as Charbagh, Saidu Sharif, Mingora, Ouch, Allabad, Dakorak Kalay, Mula Khel, Deolai, and Kanju. These areas are aligned with historical flood hotspots, particularly in 2010, 2022, and 2025 [35,37].

The hazard map (Fig 6a) shows that high to very high flood hazard zones are concentrated in valley bottoms and downstream settlements, mainly due to the convergence of geomorphology, human settlements, and agriculture. On the other hand, moderate hazard exists in the relatively elevated areas of the catchment mainly due to the slope oriented flash flooding. The snow- and glacier-fed SRB experiences ample rainfall during monsoon season (July-August), along with the accelerated snow melting process, both collectively trigger the major flash and riverine floods [44]. The basin’s steep topography accelerates runoff generation by dominating the infiltration process and intensifying flash flood risks [17]. In addition, the distance from the river is another significant factor that highlights the susceptibility of the settlements and agricultural lands in the floodplain area to flood hazards due to direct exposure to channel overflow [60]. Overall, the results highlight that precipitation and elevation jointly control flood hazard, where the intense rainfall acts as the primary trigger, while topography governs the spatial distribution of flooding.

These findings are in parallel with other studies assessing the flood hazard in the mountainous basins. For instance, Khan et al. [33] reported that flood hazards are high along the downstream Swat River floodplain, including Mingora and Khwazakhela. Similarly, Ghosh and Kar [11] assessed flood hazard by applying AHP in West Bengal, India, and found that spatial clustering of flood hazard is concentrated in low-lying floodplains due to the overlapping of geomorphology, intensive agriculture and dense human settlement. This supports the validity and broader applicability of results of the current study. Such spatial distinction of high and very high flood hazard zones underlines the importance of incorporating hazard zonation into local planning and flood preparedness and mitigation [41].

Flood vulnerability assessment (Fig 6b) shows that low vulnerability dominates (~52.8%, 1812.3 km²), followed by moderate (~25%, 890.3 km²), while high and very high vulnerability zones are limited but concentrated in urban areas. Only ~11.8% (~405 km²) of the study area falls under high vulnerability, primarily concentrated in urban and peri-urban settlements such as Chakdara, Syedabad, Fatehpur, Madyan, Bahrain, Matta, Khawazakhela, Amandara, Batkhela Darmai, Dadahara, and Barikot. A very small fraction (~0.2%, ~ 5.6 km²) corresponds to very high vulnerability concentrated in the Mingora city, which is the commercial hub and densely populated city of the study area. This reflects the dominant influence of population density, where dense settlements, built-up infrastructure, and proximity to the river coincide to significantly increase flood susceptibility. These results highlight that population and urbanization exposure are the key determinants of flood vulnerability, particularly in the floodplain [41]. Ullah and Zhang [63] also found that urban growth onto natural floodplains increased susceptibility in the Panjkora River Basin. Many other studies [8,17,24] also reported that unplanned urban growth and dense infrastructure intensify flood exposure. Ekmekcioğlu et al. [46] found that the highly populated areas of Istanbul are at high risk of flood. Hussain et al. [64] demonstrated that 25% of the western-middle to the northern part of the Shangla district comprises high to very high flood vulnerability because of the proximity to streams and rivers, high precipitation, elevation, and population density. Aydin and Sevgi [10] concluded that densely populated areas in the mountainous Baltis province of Turkey are at high risk, especially the areas which coincide with high precipitation and drainage density.

Spatial distribution of the flood risk is presented in a flood risk map (Fig 7) generated by the integration of hazard and vulnerability maps. This risk map is classified into five risk zones, and the area under each risk zone is quantified and presented in Fig 7 and Table 3.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 7. Flood risk map of the study area in the Swat River Basin.

https://doi.org/10.1371/journal.pclm.0000915.g007

The risk map indicates that low to very low risk zones cover more than half (~53.2%) of the study area, mainly in elevated and sparsely populated areas with ample distance from the river. Moreover, approximately ~18% (~562.2 km²) of the study area falls under moderate risk, especially in the mid-valley foothills, where high rainfall, medium population density performing highland agricultural activity, and proximity to stream convergence gradually increase the flood risk [11].

In contrast, high (~16.4%, ~ 562.2 km²) and very high (~12.4%) risk zones are concentrated along the floodplain area of Swat River, where high flood hazard coincides with high population density and infrastructure presence. These are the areas with low elevation experiencing high precipitation and accommodating dense population and agricultural activities for being close to the Swat River. The most significant urban settlements indicated in these zones include Mingora, Saidu Sharif Airport, Peshawar High Court-Mingora Bench, Kanju, Chakdara, Amandara, Ouch, Batkhela, Kabal, Charbagh, and Khawazakhela. This zone also includes important road networks such as N95 (Swat Motorway, Mingora-Kalam Road), M16 (Swat Motorway), N45 (Chakdara-Dir Road), which are typically located along river corridors and floodplains for easy accessibility. The results indicate that precipitation, elevation and population density are the primary factors governing flood risk in this zone. Intense precipitation acts as the main trigger of flooding, while low elevation and valley-bottom topography facilitate water accumulation and downstream runoff, increasing flood. At the same time, high population density significantly amplifies vulnerability, as dense settlements and infrastructure are concentrated within the floodplain. Additional factors such as slope-driven runoff, high topographic wetness, relatively impermeable soil, proximity to the river, and intensive agriculture activities further contribute to the overall risk. It also emphasizes that flood risk is not only process-driven but also exposure-driven, as most settlements and infrastructure in the basin are located along the main river [3]. Hence, such zones are the most critical flood risk hotspots as reported by previous studies [3,4]. For instance, Manandhar et al. [65] observed that densely populated areas of Lahore, like Johar Town and Gulshan-e-Ravi, were noticeably more vulnerable to flood due to fast urbanization and inadequate land planning. Mukhtar et al. [66] highlighted that population density in villages along the riverbanks and the changes in land use, including deforestation and agricultural development on steep terrain are the key flood susceptibility parameters in the Hunza-Nagar region, thus it possesses low capacity to deal with glacial flood hazards. A flood modelling research in the Chenab River Basin by Tariq and Van DE Giesen [30] found that encroachment on natural drainage channels caused by growing population in floodplain areas directly amplified the flood risk. Another study on flood zones in Bangladesh by Sakib et al. [55] showed that the conversion of wetlands to towns decreased natural flood absorption and enhanced the flood risk. Therefore, the findings of this study also highlight that flood risk is shaped by the combined influence of hydro-climatic drivers, topographic controls, and human exposure.

5.2. Sensitivity analysis

GIS–AHP flood model sensitivity was analyzed through the single parameter sensitivity analysis (SPSA) and map removal sensitivity analysis (MRSA) [53], demonstrated in Table 4 and Table 5, respectively.

Download:

• PNG
  larger image
• TIFF
  original image

Table 4. Impact of parametric weights on the flood risk assessment using Single Parameter Sensitivity Analysis (SPSA).

https://doi.org/10.1371/journal.pclm.0000915.t004

Download:

• PNG
  larger image
• TIFF
  original image

Table 5. Impact of parametric weights on the flood risk assessment using Map Removal Sensitivity Analysis (MRSA).

https://doi.org/10.1371/journal.pclm.0000915.t005

Table 4 presents the consistency of flood risk results under varying parameter weights and shows clear differences between empirical weights and effective weights derived through SPSA. It is found that among hazard parameters, precipitation was assigned empirical weight of 27%, while its effective weights range is 6.7%–72% (mean 32.1%, SD = 11.2), marking it as most significant parameter in this study. Likewise, among vulnerability parameters, the population density (58% empirical weight) also demonstrated the high sensitivity with effective range 14.7–77.3%, mean 31.4%, and SD = 9.6. With these results, it is confirmed that precipitation and population density are the most influential determinants of flood risk in SRB, aligning with findings of previous studies [3,41].

In addition, a critical role is played by another hazard parameter, i.e., elevation (23%) and vulnerability parameter, i.e., LULC (31%), by showing strong variability in effective weights with ranges of 7.7–76.7% and 6.9–77.5%, and SD = 7.83 and 7.92, respectively, in shaping flood risk patterns in the study area. However, parameters with low empirical weights, such as slope, drainage density, topographic wetness index, and soil type, also displayed relatively lower ranges of effective weights and smaller standard deviations. This outcome highlighted their secondary but stable contribution to determining flood risk zones in the study area. Similar patterns have been reported by sensitivity analyses performed by Rahmati et al. [52] for the Mashhad Plain Basin in Iran, Waseem et al. [3] for urban Swat in Pakistan, and Kazakis et al. [67] for the Rhodope–Evros region in Greece, where precipitation, population density, infrastructure and geomorphology parameters were confirmed as the most sensitive parameters for flood hazard and risk zonation achieved through integration of AHP and GIS.

In the context of MRSA, Table 5 illustrates minimum and maximum values representing the lower and upper limits of the variation index observed on excluding each layer, while the average effect of the parameter’s removal on the risk map is presented by the mean value, where standard deviation (SD) indicates the sensitivity and stability of this effect. It was found that removal of population density (SD = 0.64) and precipitation (SD = ~0.51) caused greater fluctuations, hence determining the major role in shaping flood risk in particular zones where these hazard and vulnerability parameters overlapped. Moreover, the moderate shifts were produced by the removal of elevation (mean = ~4.81%, SD = ~0.48) and LULC (mean = ~4.59%, SD = ~0.49), reflecting their significance in defining the valleys with agricultural landscape. On the other hand, parameters such as topographic wetness index, slope, soil type and drainage density exhibit comparatively lower mean (≈4.53-4.65) and smaller SDs (≈0.39–0.42), indicating their supportive role in assessing flood risk with stability. The findings of SPSA were reinforced by MRSA by showing that while some variations in effective weights were caused by the removal of a particular layer, the model was stable, as the overall changes were minor with SD < 0.65 for all layers. Lastly, these results suggest that although the flood risk map is most sensitive to population density and precipitation, the overall robustness is still preserved and maintained since the exclusion of no single parameter led to extreme variations in the output. In addition to the sensitivity analysis, the direct and indirect factors influencing flood frequency and risk were clearly distinguished. It is reflected that direct factors include the hazard parameters, such as precipitation and elevation, that primarily govern the probability and magnitude of flooding. On the other hand, the indirect factors are the vulnerability parameters, where population density, LULC, and proximity to infrastructure significantly influence the distribution of flood risk by determining exposure and susceptibility. Together, the interplay of these direct and indirect factors ensures that the resulting flood risk maps reliably represent both the probability of flooding and the vulnerability of affected areas. Together, the SPSA and MRSA results confirm that the spatial patterns of high and very high vulnerability zones are reliable. This provides confidence that the flood risk map accurately represents areas of greatest hazard and exposure, supporting its use for planning and mitigation purposes. This robustness is consistent with observations by Kazakis et al. [67] for the Rhodope–Evros region in Greece, and Ullah et al. [41] for Quetta, Pakistan, where rainfall- and settlement-related parameters also produced the largest fluctuations under sensitivity testing, while geomorphological parameters maintained a supportive and stable contribution.

5.3. Assessment of riverine flood inundation in the very high flood risk zone of the study area using hydrodynamic model

The HEC-RAS model was used to simulate the flow conditions during 2007 (base-year of our previous SRM study [44]), 2010 (flood year) and under future climate change scenarios in the very high risk zone of SRB. Hence, this section presents the simulated variations (Table 6) in significant river dynamic parameters, i.e., flood depth, velocity, and water surface elevation under the mentioned flow conditions.

Download:

• PNG
  larger image
• TIFF
  original image

Table 6. Variations in river flow dynamics under the flow conditions of base year, flood year and mid- and late-21st century RCP scenarios.

https://doi.org/10.1371/journal.pclm.0000915.t006

The depth of the river flow varied across different scenarios, as shown in Table 6. The results indicate a clear increasing trend in maximum flow depth across the scenarios analyzed and offer valuable information about the possible effects of climate change on hydrodynamic patterns. The base year (2007) showed a maximum depth of ~25 m, which increased to ~29 m during the flood year (2010), emphasizing the severity of historically significant floods. Projections for the mid-21^st century illustrate depths ranging from ~29 m (RCP 2.6) to ~30 m (RCP 8.5), consistently surpassing the 2010 flood depth. This escalating trend continues into the late-21^st century, ranging from 29 m (RCP 2.6) to 31 m maximum depth under the RCP 8.5 scenario. It should be noted that the simulated flood depth represents a range of values between 0 and maximum across the AOI.

The HEC-RAS generated flow velocity maps of the study area are presented in S1 Fig. It is found that the simulated flow velocity also demonstrated an overall increase from historical conditions to future projections with some notable distinctions. As shown in Table 6, the base year maximum velocity was ~ 3.7 m/s, which rose to ~5 m/s during the 2010 extreme flood. In the mid-21^st century, velocities ranged from ~5.1 m/s (RCP 4.5) to ~6.4 m/s (RCP 8.5), all significantly higher than the 2010 flood. In addition, the velocity of ~5.75 m/s under RCP 2.6 is notably higher than RCP 4.5, suggesting complex flow dynamics or specific hydrograph shapes influencing peak velocities in the model. By the late-21^st century, RCP 2.6 and 4.5 both showed high maximum velocities of ~6.3 m/s, closely matching the highest mid-21^st century velocity. Under the RCP 8.5 late-21^st century scenario, the maximum velocity decreased slightly to 5.34 m/s despite the highest depth. This suggests a complex interaction between depth and velocity, where higher flows spread more widely across the floodplain, consequently, peak discharge may propagate differently through the channel and overbank areas, reducing concentrated velocities at specific points [19].

The HEC-RAS generated flow WSE maps of the study area are presented in S2 Fig. Table 6 shows that WSE consistently increased from the base year through the historical flood event and into the future RCP scenarios, indicating higher flood levels. The results indicate that the base-year WSE had a mean value of 642 m ASL, which rose to 645 m ASL during the 2010 extreme flood as well as under mid- and late-21^st century RCP scenarios, further confirming an overall increase in baseline flood levels. This consistent upsurge in WSE can be attributed to increasing precipitation and snowmelt due to temperature rise under all RCPs, resulting in high river discharge due to climate change. These findings support previous studies [19,60,68,69] that flood magnitudes and frequencies will be exacerbated under future climatic shifts, hence posing extreme risks to downstream communities and infrastructure in the Indus Basin.

Similarly, assessment of the inundated area is crucial for flood risk assessment, mitigation and management as well as for land use planning. To determine the flood extent and depth of inundation in the Area of Interest (AOI ~ 103.66 km²), HEC-RAS and ArcGIS Pro were used in conjunction. Table 7 presents the summary of inundation assessment under climate change scenarios, i.e., mid-21^st and late-21^st century RCPs, using two distinct approaches, i.e., HEC-RAS and ArcGIS.

Download:

• PNG
  larger image
• TIFF
  original image

Table 7. Flood inundation comparison under 21st century RCP scenarios using HEC-RAS and ArcGIS approaches.

https://doi.org/10.1371/journal.pclm.0000915.t007

The flood inundation assessment through HEC-RAS and ArcGIS shows that the inundated area increased under mid- and late-21st century RCP scenarios, as depicted in Table 7. By using HEC-RAS and ArcGIS techniques, the ‘Total Inundated Area’ (Specific Inundated Area) under mid-21st century RCPs 2.6 and 4.5 are found to be ~ 54% (~23.3%) and ~54.3% (~24%) of the AOI, respectively. It is also found that the maximum total (specific) inundated areas, i.e., ~ 55% (~24.5%) and ~56.5% (~26.01%) of AOI, are under RCP 8.5 of mid- and late-21st RCP scenarios, respectively. It projects an increase in flood risk in the study area due to the adverse impact of climate change based on uncontrolled greenhouse emissions and consequent rises in temperature and precipitation.

In the present study, the entire Indus River Basin was not explicitly modelled, and flood inundation was simulated for SRB, i.e., the sub-basin of IRB, using the projected river discharge under RCP 2.6, 4.5, and 8.5 scenarios obtained from our previous work. The projected discharge under all scenarios exceeds the 2010 flood event, indicating the potential for high-intensity flooding in the future. Thus, our HEC-RAS simulation results provide an indirect assessment of future flood risk in the significant parts of the Indus River system, consistent with approaches employed in other basins. A study by Hasson et al. [70] also indicates increased seasonality and intensity of monsoon precipitation and a westward expansion of the monsoon into northwest Pakistan, both contributing to higher runoff and flood potential. Similar approaches have been applied in other river basins to link climate variability with flood occurrence. For instance, in the Mahanadi River Basin, India, flood frequency and extreme streamflow events were analyzed in relation to climate variability modes, such as the Indian Monsoon, El Nino Southern Oscillation, and Indian Ocean Dipole, to assess potential future flooding [71].

Considering the estimates presented in Table 7, especially under RCP 8.5, there is an urgent need for adaptive flood control measures such as long-term strategies like land-use management in flood-prone areas, implementation of early warning systems, and reinforcing infrastructure to prepare for larger flood events. The results showed that flood-inundated areas can be defined and calculated by adopting different modelling approaches. It also advocates that comprehensive flood modelling approaches are significant for assessment of overbank flood risk assessment, which is usually not possible by simple channel depth studies alone.

Lastly, in this study, RCP scenarios were incorporated within a process-based modelling framework, where climate projections of temperature and precipitation were used to simulate seasonal discharge using the Snowmelt Runoff Model (SRM) in our previous study [44] and subsequently applied in HEC-RAS to assess flood inundation. Therefore, the results should be interpreted as a scenario-based assessment of potential flood magnitudes, rather than a deterministic prediction of individual flood events. This approach is consistent with established practices where climate projections are coupled with hydrological models to assess future flood inundation, particularly in data-scarce mountainous basins [19,72,73]. In snow- and glacier-fed basins such as the Swat River Basin, increasing temperatures and changing monsoonal patterns are expected to enhance snowmelt-driven runoff and extreme precipitation, thereby increasing flood potential [44].

The significance of this study lies in the integration of basin-scale flood risk assessment with detailed flood inundation modeling. The GIS-AHP analysis enabled the identification high- and very-high-risk zones across the entire Swat River Basin by considering flood hazard and vulnerability parameters, whereas the HEC-RAS simulations provided a detailed flood inundation assessment of depth, velocity and water surface elevation in the very-high risk zone at the Chakdara Downstream under historical and future climate scenarios. This approach is particularly relevant for the SRB due to its steep mountainous terrain, snowmelt-monsoon influenced hydrology, and concentration of settlements in the floodplain, which collectively increase flood risk under changing climatic conditions. Moreover, the study provides important insights for a data-scarce and relatively underexplored mountainous region of Pakistan, where basin scale flood risk assessments remain limited. The findings of the study also provide a spatially targeted basis for flood mitigation and climate adaptation planning in mountainous river systems.

Future research could extend this methodology to other tributaries of the UIB and incorporate real-time hydrological forecasts, high-resolution climate projections, or additional demographic factors to improve predictive capability. Additionally, evaluating the nature-based solutions for flood mitigation within this framework could further enhance the effectiveness of studies on flood risk assessment and management. Moreover, the flood risk maps and the flood inundation maps generated offer practical guidance for policymakers, local planners, and disaster management authorities. This would enable targeted interventions, improved preparedness, and effective allocation of resources to mitigate flood impacts.

A limitation of this study is the scarcity of observational data for evaluating the hydrodynamic model. The observed discharge at the Chakdara gauge station was used as the upstream boundary condition, whereas the simulation domain is the downstream reach. Consequently, the upstream hydrograph was a model input, not a simulated output, and no downstream discharge or water-level observations were available to independently validate the simulated hydraulic response at the downstream end of the model domain. Therefore, model performance was evaluated using volume accounting error and sensitivity analysis of Manning’s roughness coefficient. The low volume accounting error shows good numerical stability and mass conservation. However, it does not prove that the simulated hydraulic conditions match the observed ones. As a result, some uncertainty remains in the simulated flood inundation dynamics. While the model provides a robust framework for the flood hazard assessment under the available data conditions, the findings should be interpreted considering this limitation. Future studies should incorporate downstream discharge, water-level observations, or observed flood extent data to enable more comprehensive validation and improve confidence in the applicability of the results to other flood events and river reaches.