Integrated Flood Susceptibility and Multi-Temporal Flood Risk Prioritization in Pakistan Using Hydro-Climatic and Geospatial Indicators

Abstract

Flood susceptibility in Pakistan is strongly influenced by hydro-climatic variability, land-surface conditions, topography, and recurrent floodplain exposure; however, national-scale studies often lack a comprehensive assessment that captures both spatial patterns and temporal flood-risk dynamics within a single framework. This study is one of Pakistan’s first national efforts to address the gap between flood risk assessment and prioritization through a unified geospatial assessment. This study assesses flood susceptibility across Pakistan for 2002, 2012, and 2022 using a GIS-based AHP approach by integrating climatic, environmental, topographic, hydrological, soil, LULC, and anthropogenic indicators. The study results were further analyzed through district-level assessments, risk change analysis, persistence mapping, LULC exposure assessments, and the Comprehensive Flood Risk Priority Index (FRPI). The results show that high and very high flood susceptibility zones are primarily concentrated along the Indus River corridor, lower floodplains, and coastal Sindh, accounting for more than 7% of the total land area of Pakistan. Persistent flood hotspots are identified in Rann of Kutch (66.6%), Jacobabad (65.0%), and Jafarabad (61.1%), indicating strong temporal stability of flood-prone conditions. LULC exposure analysis reveals that cropland is the dominant exposed class, with the highest district-level exposure observed in Badin (17.1%) and Larkana (10.1%). The FRPI further identifies priority flood-risk zones where susceptibility, persistence, risk change, and exposure converge, with the highest FRPI values observed in Jacobabad (0.742), Rann of Kutch (0.738), and Badin (0.711). Model validation demonstrates strong predictive performance, with susceptibility ROC-AUC values ranging from 0.85 to 0.87 and FRPI AUC reaching 0.85. The proposed framework provides a robust decision-support tool for targeted flood-risk management and climate-resilient land-use planning in Pakistan.

1. Introduction

Flooding remains one of the most destructive hydro-meteorological hazards worldwide, causing severe losses of human life, infrastructure, agriculture, ecosystem, and socio-economic system [1,2]. Its impacts are intensifying under changing rainfall regimes, rapid land-use conversion, urban expansion, and increasing human occupation of the floodplain [3,4]. Major Pakistani cities, including Karachi, Lahore, and Rawalpindi, have experienced extreme flooding events that exposed weaknesses in drainage systems, land-use planning, and urban resilience, while affecting millions of people and economic activities [5,6].

Pakistan is highly susceptible to flooding due to its monsoon climate, complex terrain, extensive Indus River system, and large areas of agricultural and residential areas located within floodplain environments. The Indus River corridor, the lower floodplains, and the coastal delta have been repeatedly hit by major flood events, exposing farmland, settlements, transportation networks, and critical infrastructure to repeated flood disasters [7,8]. The floods of 2010 and 2022, in particular, demonstrate that flood risk in Pakistan is not only a hydrological issue but also a land-use and planning challenge that requires an assessment approach that can capture both physical vulnerability and exposure to human environmental systems [7,9]. As highlighted in previous studies, the consequences of climate change extend beyond direct inundation and require improved drainage design, resilient infrastructure, and flood-sensitive land use planning [10,11]. Therefore, Pakistan was selected as the study area because its recurrent large-scale floods, diverse flood-generating mechanisms, and rapid land-use changes make it an appropriate case for evaluating spatial and temporal flood-risk patterns at the national scale [12].

Geospatial technologies, including Geographic Information Systems (GIS), remote sensing, and multi-criteria decision analysis, have been widely applied to flood susceptibility mapping [13,14]. Past research has employed the Analytic Hierarchy Process (AHP), frequency ratios, logistic regression, random forests, and other machine learning methods to integrate flood-influencing factors such as rainfall, elevation, slope, drainage density, topographic moisture index (TWI), soil, land use/land cover (LULC), normalized difference vegetation index (NDVI), and distance from rivers [15,16]. In Pakistan, several regional studies have assessed flood susceptibility and vulnerability in areas such as Sindh, Karachi, Hunza Nagar, and Swat using rainfall, land use, drainage, slope, vegetation, and urban-exposure indicators [17]. These studies have improved the identification of flood-prone areas and demonstrated the usefulness of geospatial modelling for flood-risk assessment [18]. However, much of the existing work remains focused on local or regional case studies and single-period susceptibility or hazard maps [19,20]. Even recent high-resolution studies in Pakistan have mainly emphasized susceptibility and exposure mapping, while temporal flood behaviour, flood persistence, flood-risk change, LULC-based exposure, and district-level priority integration remain less fully developed [21,22].

This limitation is crucial because flood risk is not static. The vulnerability of certain areas to flooding can change over time due to changes in hydroclimate, LULC, vegetation and water dynamics, and exposed land-use systems. Similarly, some areas may experience only occasional flooding, while others may be repeatedly affected by floods. Therefore, flood risk assessment requires not only mapping flood-prone areas but also assessing how susceptibility changes, in which areas it persists, and which land-use classes are exposed within high-risk areas.

To address these gaps, the present study develops an integrated national-scale geospatial assessment of flood risk in Pakistan for three benchmark years: 2002, 2012, and 2022. In this study, flood susceptibility is assessed using hydro-climatic, land-surface, topographic, hydrological, soil, LULC, and anthropogenic indicators, and the analysis is further extended to examine how susceptibility patterns change and persist over time. This allows the identification of areas where flood susceptibility has increased, decreased, remained stable, or repeatedly occurred under high-risk conditions. The assessment is further linked to LULC information to examine how different land-cover classes are distributed across susceptibility zones and to quantify the exposure of cropland and built-up areas within high- and very high-flood-prone regions. These analyses are subsequently integrated into a district-level Flood Risk Priority Index (FRPI), which converts complex spatial and temporal information into a clear priority ranking for flood risk management. The FRPI highlights districts where multiple risk conditions overlap, making it a practical decision-support output for targeted mitigation, agricultural protection, land-use regulation, and climate-resilient planning. This integrated approach provides a clearer and more practically meaningful understanding of flood-risk patterns compared to conventional single-period mapping approaches.

Unlike previous national-scale flood studies in Pakistan, which primarily focused on single-period susceptibility or exposure mapping, the present study integrates multi-temporal flood susceptibility assessment, flood persistence analysis, flood risk change evaluation, LULC-based exposure assessment, and district-level flood risk prioritization within a unified framework. This combination represents a key novelty of the study and provides a more comprehensive basis for flood-risk management and climate-resilient planning.

2. Materials and Methods

2.1. Study Area

Pakistan lies in South Asia between roughly 23–37° N and 60–77° E, covering about 0.88 million km² of very diverse landscapes [23,24,25]. In the north, the country includes the high mountain ranges of the Hindu Kush, Karakoram and Himalaya [26,27]. These give way southwards to the broad Indus alluvial plains in the central and southeastern regions, the arid plateaus of Balochistan in the west, and a narrow coastal belt along the Arabian Sea in the south [28,29,30] Figure 1. The hydrology of Pakistan is dominated by the Indus River and its major tributaries (Jhelum, Chenab, Ravi, Sutlej, Kabul and Swat), which together form the main riverine flooding system [31,32]. Most rainfall occurs during the summer monsoon (June-September), when moist southwesterly winds from the Arabian Sea bring heavy precipitation to northern, northeastern and central parts of the country [33]. In the northern mountains, snow and glacier melt further increase flows in the Indus and its tributaries, while in the western highlands, steep slopes and sparse vegetation create favourable conditions for flash floods [34,35].

2.2. Datasets

This study uses a combination of gridded climate products, satellite-derived indices, and ancillary geospatial datasets to represent the main factors that control flood susceptibility across Pakistan. The variables are grouped into three categories: hydro climatic conditions, land-surface and urban characteristics, and terrain, hydrological and soil properties. Dynamic variables were compiled for three benchmark years, 2002, 2012 and 2022, so that long-term changes in climate, land cover and urbanization can be examined in a consistent way, while DEM, slope, TWI, distance to river, distance to road, and soil texture were treated as static variables throughout the study period. An overview of all datasets is provided in

Table 1. Multi-source climatic, spectral and topographic datasets used for flood susceptibility assessment in Pakistan.

Data Product	Data Source	Spatial Resolution	Temporal Coverage
Air temperature	ERA5 reanalysis (ECMWF)	0.25°	2002–2022
Precipitation	CHIRPS (UCSB-CHG)	0.05° (6 km)	2002–2022
LST	MODIS LST (MOD11A1)	1 km	2002–2022
NDVI	MODIS surface reflectance	500 m	2002–2022
NDWI	MODIS surface reflectance	500 m	2002–2022
LULC	MODIS MCD12Q1 global land-cover	500 m	2002–2022
DEM	SRTM DEM (USGS)	30 m	Static
Slope	Derived from SRTM DEM	30 m	Static
TWI	Derived from SRTM DEM	30 m	Static
Distance to the river	Pakistan river network	30 m	Static
Distance to road	Pakistan major-road network	250 m	Static
Soil texture	OpenLandMap global soil-texture map	250 m	Static

.

2.2.1. Climatic and Hydrometeorological Variables

To characterize the large-scale climatic controls on flooding across Pakistan, we first assembled gridded rainfall, air temperature and land surface temperature datasets. Monthly mean 2 m air temperature was obtained from the ERA5 reanalysis produced by the European Centre for Medium-Range Weather Forecasts under the Copernicus Climate Change Service [36,37,38]. This product provides a physically consistent representation of the regional thermal regime over multiple decades. Precipitation was represented using the CHIRPS dataset developed by the Climate Hazards Centre, which blends satellite-based cold cloud duration with in situ observations to provide spatially continuous rainfall estimates at approximately 5 km resolution. LST fields for the benchmark years 2002, 2012 and 2022 were extracted from the MODIS MOD11A1 product [39,40,41]. Together, these three variables describe long-term patterns and trends in atmospheric forcing and surface heating that strongly influence runoff generation and the timing and intensity of flood events in Pakistan [42,43,44].

2.2.2. Vegetation, Water, Built-Up and Land Use Characteristics

Surface conditions that regulate infiltration, evapotranspiration and overland flow were captured using a set of MODIS-based spectral indices together with a global land cover product. The NDVI and NDWI were derived from the MOD09GA surface reflectance product and used to represent vegetation greenness and surface wetness, respectively [45]. In addition to these spectral indices, we quantified LULC patterns using the MODIS MCD12Q1 global land cover product at 500 m resolution for the years 2002, 2012 and 2022. These LULC maps were used to distinguish major classes such as urban, agricultural, forest and barren land, and to track their transitions over time [46].

To better characterize the thermal imprint of urbanization, two temperature-based urban indicators were also derived from MODIS LST [47,48]. By combining NDVI, NDWI, and LULC for 2002, 2012 and 2022, the dataset captures how vegetation loss, expansion of built-up surfaces and growing urban heat stress jointly modulate flood susceptibility across Pakistan’s rapidly changing landscapes.

2.2.3. Topographic, Soil and Proximity Factors

Static physiographic and exposure-related controls on flooding were represented using elevation, slope, topographic wetness, soil texture and distance to major rivers and roads. Elevation data were taken from the 30 m Shuttle Radar Topography Mission (SRTM) digital elevation model provided by the U.S. Geological Survey, from which slope was derived to represent local terrain gradients that govern the speed and direction of surface runoff [49,50]. The TWI, computed from the same DEM, was used as an indicator of potential water accumulation and saturation zones. Soil properties were incorporated using the OpenLandMap USDA soil texture dataset, which provides spatially explicit information on texture classes relevant for infiltration capacity and storage. Finally, Euclidean distances to the nearest river and to the nearest major road were calculated from national hydrography and road network layers. Distance to rivers reflects proximity to natural flood conveyance pathways, while distance to roads was included as an indirect indicator of urban development and impervious surfaces. Road networks can modify natural drainage pathways and increase surface runoff, thereby influencing flood susceptibility patterns. Consequently, distance to roads has been widely adopted as a flood-conditioning factor in previous flood susceptibility studies. These topographic, soil, and proximity variables complement the climatic and surface indicators and allow a more comprehensive representation of both hazard and exposure conditions within the national-scale flood-susceptibility modelling framework.

2.3. Methodology

This study applied a GIS- and Remote Sensing (RS)-based multi-criteria decision framework to map national-scale flood susceptibility across Pakistan for the benchmark years 2002, 2012, and 2022 (Figure 2). These benchmark years were selected at approximately 10-year intervals to evaluate long-term changes in flood susceptibility, flood persistence, land-use exposure, and flood-risk patterns across Pakistan. The selected years also represent contrasting hydro-meteorological conditions during the study period, enabling the assessment of flood-risk evolution under different climatic settings rather than focusing on individual flood events. The methodological workflow involved: data preprocessing and indicator derivation; standardization of flood-conditioning factors; AHP-based weighting; weighted overlay modelling; district-scale correlation analysis; flood persistence and flood risk change assessment; and model validation. Image processing and index derivation were carried out primarily in Google Earth Engine (GEE; Google LLC, Mountain View, CA, USA), whereas ArcGIS Pro (version 3.4; Esri, Redlands, CA, USA) was used for terrain analysis, raster standardization, proximity mapping, and weighted overlay. Microsoft Excel was used for tabulation, graphical summaries, ROC plotting, and AUC calculation.

2.3.1. Data Preprocessing and Indicator Derivation

Spatial climate and RS products were processed in a unified geospatial workflow using GEE and ArcGIS Pro [51]. Gridded rainfall, air temperature, LST, NDVI, NDWI and LULC layers were aggregated to the benchmark years 2002, 2012 and 2022 in GEE and exported as annual or multi-year means. In ArcGIS Pro, all rasters were resampled to 1 km, aligned to a common grid, clipped to the Pakistan boundary, and masked to remove no-data artifacts. From these layers, hydro-climatic, land-surface and geophysical indicators relevant to flood generation and propagation were derived. Vegetation greenness was quantified using the NDVI [52].

Following preprocessing, a total of twelve flood-conditioning factors were selected based on their hydrological relevance and established influence on flood occurrence. These factors were categorized into hydro-climatic (rainfall, air temperature, LST), land-surface (NDVI, NDWI, LULC), topographic (elevation, slope, TWI), hydrological (drainage density, distance to rivers), and soil-related (soil texture) variables. Dynamic variables were prepared for the benchmark years 2002, 2012, and 2022 to capture temporal variability, whereas terrain and soil-related factors were treated as static inputs.

The hydro-climatic variables represent the primary drivers of flood generation. Rainfall plays a dominant role in surface runoff and flood formation, with higher precipitation intensities contributing to increased flood potential. Air temperature was incorporated to reflect climatic influences on evapotranspiration and snow/glacier melt processes, particularly in upstream regions. LST was included to represent surface thermal conditions and was primarily used to support the interpretation of land-surface processes influencing flood susceptibility.

Vegetation and surface water conditions were represented using NDVI and NDWI. NDVI was calculated using (Equation (1)).

$$\mathrm{N}\mathrm{D}\mathrm{V}\mathrm{I}={\displaystyle \frac{\mathrm{N}\mathrm{I}\mathrm{R}-\mathrm{R}\mathrm{e}\mathrm{d}}{\mathrm{N}\mathrm{I}\mathrm{R}+\mathrm{R}\mathrm{e}\mathrm{d}}}$$

(1)

where NIR and Red denote the near-infrared and red spectral bands, respectively, NDWI was computed as expressed in (Equation (2)).

$$\mathrm{N}\mathrm{D}\mathrm{W}\mathrm{I}={\displaystyle \frac{ \mathrm{Green} -\mathrm{N}\mathrm{I}\mathrm{R}}{ \mathrm{Green} +\mathrm{N}\mathrm{I}\mathrm{R}}}$$

(2)

where Green and NIR represent the green and near-infrared bands, these indices were used to characterize vegetation density and surface moisture conditions, which influence infiltration, runoff, and water retention processes.

Topographic variables were derived from the digital elevation model (DEM) to represent terrain controls on flood dynamics. Elevation and slope were used to describe surface morphology and flow behaviour, where low-lying and gently sloping areas favour water accumulation. The TWI was calculated to identify potential zones of water concentration using (Equation (3)).

$$\mathrm{T}\mathrm{W}\mathrm{I}=\ln \left({\displaystyle \frac{{\mathrm{A}}_{\mathrm{s}}}{\tan \beta }}\right)$$

(3)

where A_s is the upslope contributing area and β is the slope gradient. Higher TWI values indicate an increased likelihood of soil saturation and surface water accumulation.

Hydrological connectivity was represented using drainage density and distance to rivers. Drainage density was calculated as (Equation (4)).

$$\mathrm{D}\mathrm{D}={\displaystyle \frac{\sum \mathrm{L}}{\mathrm{A}}}$$

(4)

where ∑L is the total length of streams, and A is the area. Areas with higher drainage density reflect greater runoff concentration. Distance to rivers was used to capture flood exposure, as proximity to river channels increases the likelihood of inundation.

Soil texture was included to account for variations in infiltration capacity and runoff generation. Fine-textured soils with lower permeability tend to enhance surface runoff, whereas coarse-textured soils allow greater infiltration.

LULC maps for 2002, 2012, and 2022 were reclassified into six categories: water, vegetation, cropland, built-up land, barren land, and snow and ice cover, and later used to analyze LULC transitions and to quantify the area of each class within different flood susceptibility zones. To assess the reliability of the LULC product, classification accuracy was evaluated using a confusion matrix based on reference observations and random sample points distributed across all six classes. Standard metrics, including Producer’s Accuracy, User’s Accuracy, Overall Accuracy, and Kappa coefficient, were computed. The resulting overall accuracies were 79.8%, 81.5%, and 83.9% for 2002, 2012, and 2022, respectively, with Kappa coefficients ranging from 0.79 to 0.82, indicating substantial agreement.

Together with rainfall, air temperature, LST, NDVI, NDWI, and LULC, these variables formed the set of flood-conditioning factors used in the susceptibility model.

2.3.2. Classification of Flood Conditioning Factors

All continuous flood conditioning factors were standardized to a common ordinal scale. Each raster was reclassified into five susceptibility ranks, from 1 (very low contribution to flood risk) to 5 (very high contribution to flood risk). Classification thresholds for all continuous variables were derived using the Jenks Natural Breaks algorithm to identify natural groupings within the data and minimize within-class variance, supported by hydrological interpretation and flood-conditioning hierarchies reported in the recent literature [53]. For terrain variables, low-elevation and low-slope areas were assigned higher susceptibility ranks because gentle gradients favour water accumulation and slow drainage, whereas steep mountain slopes received lower ranks. For TWI, high values (convergent, wetter zones along valleys, piedmonts and active floodplains) were ranked as more susceptible, and low values on plateaus and dune fields as less susceptible. Distance-to-river was ranked such that pixels closer to major channels received higher susceptibility scores, with risk decreasing as distance increases. Road distance was used as a proxy for exposed infrastructure and settlement; areas nearer to major roads were given higher ranks [54].

Hydro-climatic and land-surface indicators were ranked in a similar manner. Areas with higher rainfall were assigned higher susceptibility classes, reflecting the role of intense precipitation in flood generation. Conversely, low NDVI values (sparse or degraded vegetation) were associated with higher susceptibility ranks due to reduced interception and infiltration, while denser vegetation received lower ranks [55]. Higher NDWI values, representing increased surface wetness and water presence, were assigned to higher flood-susceptibility classes. Similarly, soil types with lower infiltration capacity were considered more prone to flooding than coarse-textured soils, as they promote water retention and enhance surface runoff generation. Thermal indicator LSTs were primarily used for correlation analysis and interpretation; their influence on flood susceptibility is expressed indirectly through their coupling with LULC and climatic regimes [56].

2.3.3. AHP-Based Factor Weighting Index and Mapping

The Analytic Hierarchy Process (AHP) is a widely used multi-criteria decision-making method that assigns quantitative weights to different factors based on their relative importance in a specific problem. This method plays a crucial role in analyzing complex environmental systems, particularly in flood susceptibility mapping, as flood susceptibility mapping involves the interaction of various hydro-climatic, topographic, and surface factors. First proposed by Saaty (1980) [57], the AHP method provides a structured framework for assessing and determining the priority of influencing factors through pairwise comparisons and consistency assessments [58,59].

This study uses the AHP method to determine the relative contributions of factors influencing floods and constructs a reliable flood susceptibility model. A pairwise comparison matrix where each factor is compared to all other factors using Saaty’s 1–9 scale, where 1 indicates equal importance, and 9 indicates that one factor is extremely important relative to another. The upper triangular part of the matrix is filled based on expert knowledge, the literature support, and the physical relationship between each factor and flood occurrence, while the lower triangular matrix is filled using reciprocals to maintain consistency [60].

For this study, a 12 × 12 pairwise comparison matrix was constructed to compare the relative contributions of each flood-conditioning indicator to other indicators. The upper half of the matrix triangle was filled based on expert judgement, the literature support, and the physical relationship between each indicator and flood flow. The lower half was filled with reciprocals, where the reciprocals of 1 to 9 (1/1 to 1/9) indicate that one factor is less important than another. The final pairwise comparison matrix and the resulting AHP weights are shown in

Table 2. Pairwise comparison of 12 selected flood conditioning factors used in this study.

Factors	Dem	Slope	Drainage Density	Rainfall	Lulc	Ndvi	Ndwi	Distance from the River	Distance from Roads	Twi	Lst	Soil Type
Dem	1	3	2	2	5	6	6	2	6	5	6	3
Slope	1/3	1	1	1	4	5	5	1	7	2	4	6
Drainage density	1/2	1	1	1	4	3	3	1	7	3	5	4
Rainfall	1/2	1	1	1	2	4	4	1	6	2	5	3
Lulc	1/5	1/4	1/4	1/2	1	3	2	3	5	2	4	3
Ndwi	1/6	1/5	1/3	1/4	1/3	1	1	1	2	2	3	2
Ndvi	1/6	1/5	1/3	1/4	1/2	1	1	1/2	2	1/2	3	2
Distance from the river	1/2	1	1	1	1/3	1	2	1	2	3	4	3
Distance from roads	1/6	1/7	1/7	1/6	1/5	1/2	1/2	1/2	1	2	3	2
Twi	1/5	1/2	1/3	1/2	1/2	1/2	2	1/3	1/2	1	3	2
Lst	1/6	1/4	1/5	1/5	1/4	1/3	1/3	1/4	1/3	1/3	1	2
Soil type	1/3	1/6	1/4	1/3	1/3	1/2	1/2	1/3	1/2	1/2	1/2	1

.

After the pairwise comparison matrix, it was normalized by dividing each element by the sum of its corresponding column. Then, using the normalized matrix, the relative weights of each indicator were calculated by averaging the values in each row. The weights of each indicator are calculated as follows (Equation (5)).

$${\mathrm{W}}_{\mathrm{i}}={\displaystyle \frac{1}{\mathrm{n}}}{\displaystyle \sum _{\mathrm{j}=1}^{\mathrm{n}}} {\displaystyle \frac{{\mathrm{a}}_{\mathrm{i}\mathrm{j}}}{{\displaystyle \sum _{\mathrm{i}=1}^{\mathrm{n}}} {\mathrm{a}}_{\mathrm{i}\mathrm{j}}}}$$

(5)

where ${\mathrm{W}}_{\mathrm{i}}$ represents the AHP weight of the indicator $a_{ij}$ is the pairwise comparison value between indicator and indicator j, and n is the total number of indicators included in the matrix. In this study, n = 12. After determining the weights, the consistency of the pairwise comparison matrix is evaluated to ensure that the assigned judgments are logically sound. Since the Analytic Hierarchy Process (AHP) is based on comparative judgments, a consistency check must be performed before applying the weights to susceptibility modelling. The consistency index ($\mathrm{C}\mathrm{I}$) is calculated as follows (Equation (6)).

$$CI={\displaystyle \frac{{\lambda }_{max}-n}{n-1}}$$

(6)

where $CI$ represents the consistency index, n is the number of indicators compared in the matrix, and ${\lambda }_{max}$ represents the maximum eigenvalue of the pairwise comparison matrix. The Consistency Ratio (CR) was then calculated as (Equation (7)).

$$\mathrm{C}\mathrm{R}={\displaystyle \frac{\mathrm{C}\mathrm{I}}{\mathrm{R}\mathrm{I}}}$$

(7)

where RI represents the random index corresponding to the number of indicators contained in the matrix, and its values are obtained from a (RI) table. A (CR) value below 0.10 indicates acceptable consistency, implying that the pairwise comparisons are reliable. In this study, the ${\lambda }_{\mathrm{m}\mathrm{a}\mathrm{x}}$ was calculated as 13.237, resulting in a (CI) of 0.112. Using a corresponding (RI) value of 1.48 for a 12 × 12 matrix, the (CR) was computed as 0.076 [61]. Since the (CR) value is less than 0.10, the pairwise comparison matrix is considered consistent and reliable for further analysis [62].

After confirming consistency, the final AHP-derived weights are assigned to the standardized flood impact indicator strata. Each indicator stratum has been reclassified into five susceptibility levels, ranging from 1 to 5, where 1 represents an extremely low contribution to flood susceptibility, and 5 an extremely high contribution. The direction of the level is determined by hydrological rationality. For example, factors such as higher rainfall, lower elevation, gentler slope, higher TWI, higher drainage density, higher NDWI, proximity to rivers, sparse vegetation, and fine soil texture are assigned higher susceptibility levels, while factors such as higher elevation, steeper slope, denser vegetation, and greater distance from rivers are assigned lower susceptibility levels [63,64]. A linear weighted superposition method is used to integrate the standardized index layers and their corresponding AHP weights to generate an AHP-based flood susceptibility index (Equation (8)).

$$\mathrm{F}\mathrm{S}{\mathrm{I}}_{\mathrm{A}\mathrm{H}\mathrm{P}}={\displaystyle \sum _{\mathrm{i}=1}^{\mathrm{n}}} {\mathrm{X}}_{\mathrm{i}}\times {\mathrm{W}}_{\mathrm{i}}$$

(8)

where $\mathrm{F}\mathrm{S}{\mathrm{I}}_{\mathrm{A}\mathrm{H}\mathrm{P}}$ represents the AHP-based flood susceptibility index and ${\mathrm{W}}_{\mathrm{i}}$ denotes the derived weight of each indicator. The resulting FSI values were reclassified into five flood susceptibility zones: very low, low, moderate, high, and very high. The same AHP-derived weights were consistently applied to the 2002, 2012, and 2022 sensitivity models to maintain temporal comparability. This ensured that the differences between the resulting flood susceptibility maps reflected changes in the input data rather than changes in the weighting structure.

2.3.4. Flood Risk Change Analysis

Flood risk change analysis was conducted to examine how flood susceptibility changed between the initial and final reference years and to identify areas where flood-prone conditions increased, decreased, or remained relatively stable over time. The classified flood susceptibility maps for 2002 and 2022 were used as the main input data, as both maps were generated using the same AHP-derived weights, standardized factor classes, and vulnerability classification scheme. This ensured that the comparison reflected real temporal change and was not caused by inconsistent modelling procedures. The flood risk was calculated by subtracting the 2002 flood susceptibility layer from the 2022 flood susceptibility layer using (Equation (9)).

$$\mathrm{R}\mathrm{C}=\mathrm{F}\mathrm{S}{\mathrm{I}}_{2022}-\mathrm{F}\mathrm{S}{\mathrm{I}}_{2002}$$

(9)

where RC represents flood-risk change, FSI₂₀₂₂ represents the flood susceptibility value in 2022, and FSI₂₀₀₂ represents the flood susceptibility value in 2002. Positive values of RC indicate an increase in flood susceptibility, negative values indicate a decrease, and values close to zero represent stable or unchanged conditions. The resulting change layer was then interpreted according to the direction and magnitude of change so that areas of increasing, decreasing, and stable flood risk could be distinguished across Pakistan. To support district-level interpretation, we have summarized the spatial extent of increased flood susceptibility within each region, expressed as a percentage of the total regional area. The regional flood risk change components are calculated as follows (Equation (10)).

$$\mathrm{R}{\mathrm{C}}_{\mathrm{d}}={\displaystyle \frac{{\mathrm{A}}_{\mathrm{Increase},\mathrm{d}}}{{\mathrm{A}}_{\mathrm{Total},\mathrm{d}}}}\times 100$$

(10)

where $\mathrm{R}{\mathrm{C}}_{\mathrm{d}}$ represents the district-level flood risk change score, ${\mathrm{A}}_{\mathrm{Increase},\mathrm{d}}$ is the area within district d showing increased flood susceptibility, and ${\mathrm{A}}_{\mathrm{Total},\mathrm{d}}$ is the total area of district d. This component was later used in the FRPI to represent districts where flood susceptibility increased over the study period. Thus, the flood risk change analysis provided a temporal dimension to the assessment by showing not only where flood susceptibility was high, but also where it intensified between 2002 and 2022.

2.3.5. Flood Persistence Analysis

To identify areas consistently vulnerable to flooding over multiple time periods and to distinguish between recurring flood-prone areas and intermittently affected areas, we conducted a flood persistence analysis. This analysis provides a temporal stability perspective by highlighting areas with recurring high flood susceptibility, thus revealing the long-term vulnerability of these areas. The primary input data for this analysis consisted of flood susceptibility classification maps generated in 2002, 2012, and 2022. To ensure consistency, the same classification scheme and weight structure were used for all three years. From each flood susceptibility map, we extracted only the high susceptibility and very high susceptibility levels, as these levels represent the areas most likely to experience flooding. We then converted the extracted levels into a binary flood-susceptibility layer, where flood-prone areas were assigned a value of 1, and all other areas were assigned a value of 0. The binary representation for each year can be expressed as (Equation (11)).

$$F_t=\left\{\begin{array}{c}1, \mathrm{i}\mathrm{f} FS{I}_t\in \{ \mathrm{H}\mathrm{i}\mathrm{g}\mathrm{h}, \mathrm{V}\mathrm{e}\mathrm{r}\mathrm{y} \mathrm{H}\mathrm{i}\mathrm{g}\mathrm{h} \}\hfill \\ 0, \mathrm{i}\mathrm{f} FS{I}_t\in \{ \mathrm{V}\mathrm{e}\mathrm{r}\mathrm{y} \mathrm{L}\mathrm{o}\mathrm{w}, \mathrm{L}\mathrm{o}\mathrm{w}, \mathrm{M}\mathrm{o}\mathrm{d}\mathrm{e}\mathrm{r}\mathrm{a}\mathrm{t}\mathrm{e} \}\hfill \end{array}\right.$$

(11)

where $F_t$ represents the binary flood-prone layer for the year, and $\mathrm{F}\mathrm{S}{\mathrm{I}}_{\mathrm{t}}$ represents the classified flood susceptibility for that year. The binary flood-prone layers corresponding to 2002, 2012, and 2022 were then integrated to determine the persistence of flood-prone conditions across time. This was achieved by summing the binary layers calculated using (Equation (12)).

$$FP=F_{2002}+F_{2012}+F_{2022}$$

(12)

where FP represents the flood persistence index. The calculation results range from 0 to 3, where 0 indicates no flood persistence (no year is designated as a flood-prone area), 1 indicates low persistence (flood-prone for one year), 2 indicates moderate persistence (flood-prone for two years), and 3 indicates high persistence (flood-prone for all three years). To facilitate interpretation and integration into subsequent analyses, persistence values are further subdivided into four levels: no persistence, low persistence, moderate persistence, and high persistence. These levels clearly demonstrate the temporal stability of flood-prone areas and help identify areas with consistently high flood risk. In district-level analysis, the spatial extent of high-persistence areas is quantified by calculating the proportion of each district that remains flood-prone over multiple years. The formula for calculating the district-level flood persistence component is as follows (Equation (13)).

$$\mathrm{F}{\mathrm{P}}_{\mathrm{d}}={\displaystyle \frac{{\mathrm{A}}_{\mathrm{HighPersistence},\mathrm{d}}}{{\mathrm{A}}_{\mathrm{Total},\mathrm{d}}}}\times 100$$

(13)

where $\mathrm{F}{\mathrm{P}}_{\mathrm{d}}$ represents the flood persistence value for district d, ${\mathrm{A}}_{\mathrm{HighPersistence},\mathrm{d}}$ represents the area within district d classified as high persistence FP = 3, and ${\mathrm{A}}_{\mathrm{Total},\mathrm{d}}$ represents the total area of the district.

2.3.6. LULC-Based Exposure Analysis

Flood exposure analysis based on LULC aims to assess the spatial interaction between flood-prone areas and land use systems and to quantify the extent to which key land cover types are exposed to flood susceptibility and persistent flood conditions. This analysis provides a comprehensive understanding of human-land vulnerability by integrating flood susceptibility and flood persistence with land-use distribution. The primary input data for this analysis are LULC maps from 2002, 2012, and 2022. The flood susceptibility maps were spatially integrated with the corresponding LULC maps for each baseline year. This integration allowed us to identify the distribution of different land cover types within different flood-prone areas. The spatial extent of each LULC type within each susceptibility category was quantified and expressed as a proportion of the total area of that susceptibility category. This relationship can be represented as (Equation (14)).

$$\mathrm{L}\mathrm{U}\mathrm{L}{\mathrm{C}}_{\mathrm{i},\mathrm{j}}={\displaystyle \frac{{\mathrm{A}}_{\mathrm{i},\mathrm{j}}}{{\mathrm{A}}_{\mathrm{j}}}}\times 100$$

(14)

where $\mathrm{L}\mathrm{U}\mathrm{L}{\mathrm{C}}_{\mathrm{i},\mathrm{j}}$ represents the percentage of LULC class within the flood susceptibility class $A_{i,j}$ is the area of the LULC class within the susceptibility class, and ${\mathrm{A}}_{\mathrm{i},\mathrm{j}}$ is the total area of susceptibility class. This analysis enabled the identification of land cover types that are more associated with high and very high flood susceptibility zones.

The assessment of LULC exposure is based on persistent flood conditions. The flood persistence layer is obtained by integrating multi-temporal flood susceptibility data and spatially combining it with the LULC layer to assess the extent to which land cover types are repeatedly exposed to flood-prone conditions. Emphasis is placed on high persistence areas, representing regions that have been consistently classified as flood-prone over many years. The exposure of each LULC class to persistent flooding was calculated as (Equation (15)).

$$\mathrm{L}{\mathrm{E}}_{\mathrm{i}}={\displaystyle \frac{{\mathrm{A}}_{\mathrm{i},\mathrm{H}\mathrm{P}}}{{\mathrm{A}}_{\mathrm{i},\mathrm{total} }}}\times 100$$

(15)

where $\mathrm{L}{\mathrm{E}}_{\mathrm{i}}$ represents the exposure of LULC class to high-persistence flood conditions, ${\mathrm{A}}_{\mathrm{i},\mathrm{H}\mathrm{P}}$ is the area of the LULC class within high persistence zones, and ${\mathrm{A}}_{\mathrm{i},\mathrm{total} }$ is the total area of that LULC class. To extend the analysis to the administrative level, we calculated the exposure levels for the main land-use types within each district. We paid particular attention to cropland and built-up areas, as these are closely related to agricultural productivity and human settlements. The district-level LULC exposure levels were calculated as follows (Equation (16)).

$$\mathrm{L}{\mathrm{E}}_{\mathrm{d}}={\displaystyle \frac{{\mathrm{A}}_{\mathrm{CroplandHP},\mathrm{d}}+{\mathrm{A}}_{\mathrm{BuiltupHP},\mathrm{d}}}{{\mathrm{A}}_{\mathrm{Total},\mathrm{d}}}}\times 100$$

(16)

Here $\mathrm{L}{\mathrm{E}}_{\mathrm{d}}$ represents the LULC exposure at the district level, ${\mathrm{A}}_{\mathrm{CroplandHP},\mathrm{d}}$ and $A_{\mathrm{BuiltupHP},d}$ represent the areas of cropland and built-up land within high-persistence flood zones in the district, and ${\mathrm{A}}_{\mathrm{Total},\mathrm{d}}$ represents the total area of the district.

2.3.7. District Scale Correlation Analysis

To explore statistical relationships between key indicators, a district-wise correlation analysis was carried out for 2022. The 2022 dataset was selected because it represents the most recent hydroclimatic conditions, including the context of extreme flooding events. For each administrative district, mean values of rainfall, air temperature, LST, NDVI, and NDWI were extracted. Pearson correlation coefficients were computed for all variable pairs; statistical significance was assessed at p < 0.05. The resulting correlation matrix was visualized as a heatmap to illustrate the strength and direction of relationships among variables. This analysis was used to interpret how hot-dry regimes, vegetation conditions and urban thermal stress co-vary across Pakistan and how they relate to the spatial pattern of flood susceptibility.

2.3.8. Flood Risk Priority Index (FRPI) Development

The FRPI aims to integrate the main spatial and temporal dimensions of flood risk into a single district-level prioritization framework. Since flood susceptibility, flood persistence, flood risk change, and LULC-based exposure have already been derived, these outputs are used as input variables for FRPI development. For each district, four standardized components are considered: flood susceptibility intensity, flood persistence, flood risk change, and LULC exposure. Each component is normalized using a min-max normalization method prior to integration, ensuring comparability among variables across different units and extents, as shown in (Equation (17)).

$${\mathrm{X}}_{\mathrm{d}}^{\prime }={\displaystyle \frac{{\mathrm{X}}_{\mathrm{d}}-{\mathrm{X}}_{\mathrm{m}\mathrm{i}\mathrm{n}}}{{\mathrm{X}}_{\mathrm{m}\mathrm{a}\mathrm{x}}-{\mathrm{X}}_{\mathrm{m}\mathrm{i}\mathrm{n}}}}$$

(17)

where ${\mathrm{X}}_{\mathrm{d}}^{\prime }$ represents the normalized value for the district $X_d$ is the original district-level value, and ${\mathrm{X}}_{\mathrm{m}\mathrm{i}\mathrm{n}}$ and ${\mathrm{X}}_{\mathrm{m}\mathrm{a}\mathrm{x}}$ are the minimum and maximum values of that component across all districts. Then, a weighted aggregation method is used to combine the normalized components to calculate the final FRPI, as shown in (Equation (18)).

$$FRP{I}_d=0.35F{S}_d^{\prime }+0.25F{P}_d^{\prime }+0.20R{C}_d^{\prime }+0.20L{E}_d^{\prime }$$

(18)

where $\mathrm{F}\mathrm{R}\mathrm{P}{\mathrm{I}}_{\mathrm{d}}$ represents the Flood Risk Priority Index for the district $\mathrm{F}{\mathrm{S}}_{\mathrm{d}}^{\prime }$ is the normalized flood susceptibility component, $\mathrm{F}{\mathrm{P}}_{\mathrm{d}}^{\prime }$ is the normalized flood persistence component, $R{C}_d^{\prime }$ is the normalized flood-risk change component, and $\mathrm{L}{\mathrm{E}}_{\mathrm{d}}^{\prime }$ is the normalized LULC exposure component. The weighting scheme was designed to reflect the relative contribution of each component to flood-risk prioritization within the proposed FRPI framework. Flood susceptibility was assigned the highest weight (0.35) because it represents the primary spatial likelihood of flooding and forms the foundation of flood-risk assessment. Flood persistence was assigned the second-highest weight (0.25) to account for the temporal recurrence of flood-prone conditions. Flood-risk change (0.20) and LULC exposure (0.20) were assigned equal weights because they represent complementary dimensions of temporal risk dynamics and exposure conditions. Therefore, the coefficients were determined according to the conceptual importance of each component within the proposed flood-risk prioritization framework, following the expert-based hierarchical weighting philosophy commonly adopted in integrated flood-risk assessment studies [65,66,67].

2.3.9. Model Validation

Model validation was performed to evaluate the predictive performance and reliability of the flood susceptibility and FRPI models [68]. The flood susceptibility was validated using observed flood inventory data for 2002, 2012, and 2022, while the FRPI model was validated only for 2022, as the priority index was developed for the most recent base year. The generated flood susceptibility maps were reclassified into binary classes: high and very high susceptibility zones were considered flood-prone areas (value = 1), while very low, low, and moderate classes were categorized as non-flood-prone areas (value = 0). Similarly, the FRPI output was evaluated by comparing higher-priority flood-risk zones with observed flood and non-flood reference locations. A confusion matrix was constructed by comparing predicted flood-prone areas with observed flood locations, allowing the identification of True Positive (TP), True Negative (TN), False Positive (FP), and False Negative (FN) values. Based on these parameters, the overall classification accuracy was calculated as (Equation (19)).

$$\mathrm{Accuracy} ={\displaystyle \frac{\mathrm{T}\mathrm{P}+\mathrm{T}\mathrm{N}}{\mathrm{T}\mathrm{P}+\mathrm{T}\mathrm{N}+\mathrm{F}\mathrm{P}+\mathrm{F}\mathrm{N}}}$$

(19)

where $\mathrm{T}\mathrm{P}$ represents correctly predicted flood occurrences, $\mathrm{T}\mathrm{N}$ represents correctly identified non-flood areas, $\mathrm{F}\mathrm{P}$ indicates overestimation (false flood prediction), and $\mathrm{F}\mathrm{N}$ represents missed flood occurrences. In addition to overall accuracy, sensitivity and specificity were also calculated to assess the model’s ability to identify flood-prone and non-flood-prone areas (Equation (20)).

$$\begin{array}{c} \mathrm{S}\mathrm{e}\mathrm{n}\mathrm{s}\mathrm{i}\mathrm{t}\mathrm{i}\mathrm{v}\mathrm{i}\mathrm{t}\mathrm{y}={\displaystyle \frac{\mathrm{T}\mathrm{P}}{\mathrm{T}\mathrm{P}+\mathrm{F}\mathrm{N}}}\\ \mathrm{S}\mathrm{p}\mathrm{e}\mathrm{c}\mathrm{i}\mathrm{f}\mathrm{i}\mathrm{c}\mathrm{i}\mathrm{t}\mathrm{y}={\displaystyle \frac{\mathrm{T}\mathrm{N}}{\mathrm{T}\mathrm{N}+\mathrm{F}\mathrm{P}}}\end{array}$$

(20)

In addition to accuracy assessment, Receiver Operating Characteristic (ROC) curve analysis was performed to evaluate the model’s discrimination ability [69]. The False Positive Rate (FPR) was derived from the confusion matrix and used together with sensitivity to construct ROC curves. The Area Under the Curve (AUC) was calculated to quantify model performance using Equation (21).

$$\mathrm{FPR}={\displaystyle \frac{\mathrm{F}\mathrm{P}}{\mathrm{F}\mathrm{P}+\mathrm{T}\mathrm{N}}}$$

(21)

The AUC, calculated from the ROC curve, is used to quantify the overall performance of the model. The AUC value ranges from 0 to 1, with values closer to 1 indicating stronger predictive ability.

3. Result

3.1. Geophysical Distributions Across Pakistan

The current research also investigates the geophysical distributions within Pakistan, such as DEM, slope, distance to rivers, drainage density (DD), topographic wetness index (TWI), distance to roads, and soil characteristics. All continuous parameters were re-categorized into seven ordered categories within the same grid to allow cross-comparison, and the soils were mapped in the form of texture groups. In particular, the results of the DEM indicate that the highest elevation is observed in the Hindu Kush Karakoram HHK arc and the mountain ranges along the West, and there is a steep drop in elevation towards the Indus Plain and the coastal lands of Sindh. Similarly, slope outcomes also show that steep slopes are located in Gilgit Baltistan, Chitral, Kohistan, and the adjacent highlands, and the gentle or near-level ones are found in Punjab and the lower Indus basin. Furthermore, distance-to-rivers findings indicate a constant low-distance zone of the Indus with its major tributaries, extending along the Indus headwaters to Punjab to the delta, but very large distances within interior Balochistan and the Thar fringe.

In constant proportion, the drainage density outcomes indicate that DD is maximum along the upper Indus system and the related piedmont fans, which comprise the Swat, Hazara, and Potwar margins, which are indicative of close-spaced channels, and DD is minimum in interior deserts and the final alluvial plains of southern Sindh. Moreover, the TWI outcomes depict convergent and comparatively more humid areas along mountain valleys, piedmont aprons, and active floodplains, and lower TWI scores are on interfluves, plateaus, and dune fields Figure 3. Moreover, distance-to-roads outcomes indicate that there are relatively small distances throughout the Punjab Sindh urban industrial corridor and significantly larger distances in the sparsely populated western basins. Equally, as shown by soil-texture outcomes, finer textures are found to prevail on the Indus alluvium, whereas sandier soils are found to dominate on the west and south-western basins; localized pockets of silt loam are largely oriented along major floodplains. Together, these DEM, slope, distance-to-river, drainage-density, TWI, distance-to-road, and soil-texture outputs give the physical context governing runoff generation, flow concentration, and possible pathways of inundation, and form the basis of the susceptibility analysis that proceeds.

3.2. Distribution and Trends of Climate Parameters (Rainfall, Air Temperature, LST) (2002–2022)

The study reveals significant spatial and temporal variability in rainfall, air temperature, and LST across Pakistan during the period 2002–2022 Figure 4a–c.

Year wise spatial rainfall distribution indicates that northern mountainous regions consistently receive the highest precipitation, whereas southern districts remain comparatively dry. This pattern reflects the dominant influence of orographic effects and monsoonal circulation in the northern highlands.

At the national district scale, average rainfall shows notable internal variability across the study period, increasing from 405.52 mm in 2002 to 524.48 mm in 2012, followed by a sharp decline to 307.18 mm in 2022. Monthly rainfall trends further reveal pronounced seasonal variability controlled by the monsoon system Figure 4d. The highest rainfall contributions are observed during July and August, with a peak recorded in July 2022 (160.03 mm), followed by August (99.66 mm), whereas comparatively lower rainfall is observed during the winter season. The highest monthly rainfall was recorded in July 2022 (160.03 mm). To provide additional insight into regional precipitation contrasts, representative districts from different rainfall regimes are compared in Supplementary Figure S1. The figure highlights the variability between humid northern regions and comparatively drier southern districts across Pakistan. These findings confirm that rainfall is strongly concentrated in the monsoon season, while winter months remain comparatively dry.

Likewise spatial distribution of air temperature has a distinct north south gradient across Pakistan Figure 5a–c. Southern and southwestern regions are characterized by persistently warmer conditions, whereas northern mountainous areas experience comparatively cooler temperatures due to higher elevation and dense vegetation cover.

At the national district scale, air temperature shows a gradual decline across the study period, decreasing from 22.84 °C in 2002 to 21.85 °C in 2012, and further to 20.81 °C in 2022. Despite this overall decrease, the spatial thermal gradient remains consistent, with southern districts experiencing persistent thermal stress and northern regions maintaining comparatively cooler conditions. Figure S2 presents a comparison of representative districts selected from contrasting temperature environments, providing additional evidence of the pronounced thermal differences between the cooler northern highlands and the warmer southern plains.

The spatial pattern of LST closely follows the distribution of air temperature but exhibits stronger surface contrasts due to land surface characteristics Figure 6a–c.

At the national district scale, LST also shows temporal variability, with values of 32.55 °C in 2002, decreasing to 31.25 °C in 2012, and increasing again to 32.34 °C in 2022. This trend suggests a mid-period decline followed by renewed surface warming in recent years. (Figure S3) provides a comparison of representative districts from contrasting thermal environments, further illustrating the spatial variability of land surface temperature across Pakistan.

3.3. Spatial Distribution of NDVI and NDWI (2002–2022)

The spatial distribution of vegetation and surface moisture conditions across Pakistan was evaluated using NDVI and NDWI for the years 2002, 2012, and 2022. These indices reveal substantial spatial variability driven by climatic conditions, land-use patterns, and water availability across the study area. The year-wise spatial distribution of NDVI indicates that vegetation density is predominantly concentrated in the northern and northeastern regions, whereas southern and southwestern areas exhibit sparse vegetation cover Figure 7a–c. Moreover, the NDVI results indicate improved vegetation conditions during the study period. Mean NDVI increased from 0.14 in 2002 to 0.15 in 2012 and further to 0.18 in 2022, suggesting a gradual improvement in vegetation cover across Pakistan. Spatially, relatively higher vegetation density is observed in the northern and northeastern regions, whereas lower vegetation cover is concentrated in the arid and semi-arid zones of southern and southwestern Pakistan. These patterns highlight the strong influence of rainfall availability and regional climatic conditions on vegetation dynamics. To further illustrate regional differences in vegetation conditions, representative districts from contrasting environmental settings are compared in Supplementary Figure S4.

Similarly, the spatial distribution of NDWI provides important insights into surface moisture and water availability across the study area Figure 8a–c. Mean NDWI increased slightly from −0.03 in 2002 to −0.02 in 2012 and remained at approximately −0.02 in 2022, indicating relatively stable moisture conditions over the study period. Spatially, higher moisture availability is observed in northern regions, whereas southern and arid zones exhibit comparatively lower NDWI values. These spatial and temporal patterns reflect variations in hydrological conditions and precipitation regimes across the country, indicating limited surface water availability and predominance of non-water land surfaces. A comparison of representative districts across contrasting moisture environments is presented in Supplementary Figure S5, further highlighting regional differences in surface moisture conditions.

The combined analysis of NDVI and NDWI highlights a consistent environmental gradient across Pakistan. Regions with higher vegetation density generally correspond to higher moisture availability, whereas arid regions are characterized by low vegetation cover and reduced surface moisture. Consequently, these findings establish a strong linkage between climatic conditions, vegetation dynamics, and hydrological processes across the study area.

3.4. Assessment of LULC Changes and Classification Accuracy (2002–2022)

In this study, the accuracy of the LULC classification was validated based on the standard metrics, including the Kappa coefficient and overall accuracy. The Kappa coefficient value, as in Figure S6, has improved from 0.79 in 2002 to 0.82 in 2022, and the overall accuracy has improved from 79.8 to 83.9, which shows better classification reliability over the years. Also, the accuracy of the producer’s and user’s accuracy of individual land-use classes is relatively more accurate in vegetation, cropland and water classes compared to the built-up and barren land classes, which are relatively moderate, possibly because of spectral mixing and any transitional land-use conditions. The accuracy tests validate the strength of the classification findings and justify their consistency in further analysis. Spatiotemporal dynamics of LULC in 2002, 2012 and 2022 were analyzed, and the areas of the classes are given in

Table 3. Normalized pairwise comparison matrix and derived AHP weights.

Factors	Dem	Slope	Drainage Density	Rain Fall	Lulc	Ndvi	Ndwi	Distance from River	Distance from Roads	Twi	Lst	Soil Type	Weight (wi)
Dem	0.24	0.34	0.25	0.24	0.27	0.23	0.22	0.17	0.15	0.21	0.14	0.09	0.214
Slope	0.08	0.11	0.13	0.12	0.22	0.19	0.18	0.08	0.18	0.09	0.1	0.18	0.139
Drainage density	0.12	0.11	0.13	0.12	0.22	0.12	0.11	0.08	0.18	0.13	0.12	0.12	0.13
Rainfall	0.12	0.11	0.13	0.12	0.11	0.15	0.15	0.08	0.15	0.09	0.12	0.09	0.119
Lulc	0.05	0.03	0.03	0.06	0.05	0.12	0.07	0.25	0.13	0.09	0.1	0.09	0.089
Ndwi	0.04	0.02	0.04	0.03	0.02	0.04	0.04	0.08	0.05	0.09	0.07	0.06	0.049
Ndvi	0.04	0.02	0.04	0.03	0.03	0.04	0.04	0.04	0.05	0.02	0.07	0.06	0.04
Distance from the river	0.12	0.11	0.13	0.12	0.02	0.04	0.07	0.08	0.05	0.13	0.1	0.09	0.089
Distance from roads	0.04	0.02	0.02	0.02	0.01	0.02	0.02	0.04	0.03	0.09	0.07	0.06	0.036
Twi	0.05	0.06	0.04	0.06	0.03	0.02	0.07	0.03	0.01	0.04	0.07	0.06	0.045
Lst	0.04	0.03	0.03	0.02	0.01	0.01	0.01	0.02	0.01	0.01	0.02	0.06	0.024
Soil type	0.08	0.02	0.03	0.04	0.02	0.02	0.02	0.03	0.01	0.02	0.01	0.03	0.028

and the spatial patterns depicted in Figure 9a–d. The findings show that the landscape was largely characterized by bare land, which reduced between 507,431 km² (65.37) in 2002 and 473,858 km² (61.04) in 2022, representing a slow change towards more productive land uses.

In contrast, cropland increased from 154,672 km² (19.93%) to 167,819 km² (21.62%), particularly across the Indus basin, indicating agricultural expansion and intensified land utilization can be seen in

Table 4. LULC class areas (km²) for the study domain in 2002, 2012, and 2022.

Classes	2002	2012	2022
water	12,804.5	12,755.3	13,503.8
vegetation	78,356.2	68,328.3	85,207.7
crops	154,672.3	162,195.2	167,818.8
built	20,187.9	23,682.0	31,127.1
bare	507,431.4	506,380.9	473,857.8
snow	2792.2	2902.8	4729.0

. Similarly, built-up areas exhibited the most pronounced growth, expanding from 20,188 km² (2.60%) to 31,127 km² (4.01%), highlighting rapid urbanization and increasing anthropogenic pressure on land resources. The vegetation cover showed a moderate increase, rising from 78,356 km² (10.09%) to 85,208 km² (10.98%), suggesting localized ecological recovery. Meanwhile, water bodies remained relatively stable, with a slight increase from 12,804 km² (1.65%) to 13,504 km² (1.74%), indicating consistency in surface water distribution. Notably, snow and ice cover increased from 2792 km² (0.36%) to 4729 km² (0.61%), reflecting changes in cryospheric conditions. The observed increase may reflect interannual variability in snow accumulation and seasonal snow persistence in the high-altitude northern regions of Pakistan. This is particularly significant in the context of flood susceptibility, as glacier melt and snow dynamics contribute directly to runoff generation and flood hazards in northern Pakistan. The overall transition pattern highlights that the reduction in barren land is primarily compensated by increases in cropland and built-up areas, indicating a shift toward more intensive land-use practices. These transformations demonstrate the growing interaction between land-use change and hydrological processes.

The findings indicate that LULC changes in Pakistan are driven by urban expansion, agricultural intensification, and environmental variability, which collectively influence flood dynamics and spatial susceptibility patterns across the region.

3.5. Flood Susceptibility and LULC Association (2002–2022)

Flood susceptibility analysis revealed clear spatial variations across Pakistan, with high-susceptibility zones concentrated in riverine, floodplain, and coastal-deltaic environments. Across all study years, high and very high flood susceptibility classes were primarily concentrated along the Indus River corridor, its major tributaries, lower floodplains, and coastal regions of Sindh Figure 10a–d.

District-wise results further confirm this spatial concentration of flood susceptibility Figure 11. In 2022, the largest proportions of high and very high flood-susceptible areas are observed in Rann of Kutch, accounting for approximately 66.6%, followed by Jakobabad (65.0%), Jafarabad (61.1%), Badin (56.5%) and Kashmore (54.1%). These districts represent the most critical flood-susceptibility zones because a substantial part of their area falls within the high and very high classes. Moreover, their spatial location near low-lying floodplains, deltaic environments and river-connected agricultural zones explains why these districts remain highly exposed to flood-prone conditions.

Similarly, several other districts show notable levels of high and very high susceptibility, although at comparatively lower proportions than the most exposed districts. Larkana records approximately 44.6%, followed by Thatta (40.6%), Shikarpur (32.3%), Gujranwala 2 (30.3%), Muzaffargarh (28.5%) and Mirphurkhas (28.5%). Likewise, Rajan Pur (23.1%), Sialkot (21.6%), Matiari (21.1%), Sanghar (20.4%) and Tando Muhammad Khan (20.3%) indicate moderate to high district-level exposure. These values show that flood susceptibility is not confined to one administrative region; rather, it extends across multiple floodplains and irrigated agricultural landscapes where low elevation, drainage concentration and proximity to river systems increase flood sensitivity.

Flood susceptibility patterns remained largely stable between 2002 and 2022, with Rann of Kutch, Jakobabad, Jafarabad, Badin, Kashmore, Larkana, Thatta, and Shikarpur consistently appearing among the most vulnerable districts.

Conversely, a number of western, dry and high-elevation districts tend to exhibit smaller numbers of high and very high susceptibility. At the district level, high-susceptibility classes are less dominant in such districts as Chagai, Kharan, Quetta, Kalat, Pishin, Gwadar, Chitral, Gilgit and Swat. However, it does not mean that there is no risk of floods in such places, since local flash floods, hill torrents and valley-based floods can take place. Instead, the percentages of lower district-levels reveal that high-susceptibility areas are more spatially constrained than the lower Indus floodplain and coastal-deltaic districts.

Further information about the land-cover features of the individual flood-susceptibility classes is obtained with the help of the LULC association analysis Figure 12a–o. In the very low susceptibility class, barren land remains the dominant LULC category across all three years, contributing approximately 81% in 2002, 77% in 2012 and 72% in 2022. Meanwhile, vegetation increases from 15% to 18% and then to 21%, whereas water, cropland, built-up land and snow cover remain minor components.

This composition means that the very low susceptibility areas are primarily linked to dry and poorly developed landscapes where repeated exposure to floods is minimal. Similarly, barren land also predominates in the low susceptibility category, but this category decreases to around 70% in 2002 to 65% in 2012 and 60% in 2022. Simultaneously, the proportion of cropland rises to 23–24% and 26% with the vegetation proportion increasing to 4–6% and 8%. There is also a slight increase in built-up land, which rises to 3% or 5%.

Mixed land cover composition is exhibited by the moderate susceptibility class. This class is dominated by water and barren land in 2002, with an approximation of 45% and 47% respectively. Water contributed approximately 43% in 2012, with barren land contributing 45%. By 2022, water reduces to 40% and, barren land is crucial at 41% and cropland for approximately 10%. In addition, snow cover is seen more in 2022, adding around 5%. These results indicate a mixed LULC composition in moderate susceptibility zones. In the high susceptibility category, the importance of cropland is heightened. Its share increases to about 37% in 2002, to 40% in 2012, and 43% in 2022, and is the prevailing LULC category in recent years. Conversely, barren land decreases from 36% to 32% and then to 27%, while water remains relatively stable, increasing slightly from 12% to 14%. Vegetation remains close to 11%, and built-up land increases from 4% to 5%. The very high susceptibility class shows the most important land-cover transition. In 2002, barren land was the largest component, accounting for approximately 40%, followed by cropland (23%), water (22%) and vegetation (12%). However, by 2022, cropland becomes the dominant class, reaching approximately 48%, while barren land declines to 22%. Water remains important at about 19%, whereas vegetation decreases to 8%, and built-up land remains around 3%. Cropland becomes the dominant LULC class in very high susceptibility zones by 2022.

Overall, the district-wise and LULC analyses indicate a transition from barren land-dominated low susceptibility zones to cropland-dominated high susceptibility zones. High and very high susceptibility classes show increasing cropland exposure, particularly along the Indus floodplain and coastal deltaic belt.

3.6. Flood Persistence and Risk Change Analysis

The flood-risk change analysis reveals that most of Pakistan remained within the no-change category, while increased and decreased risk zones occurred as localized transition areas. The strongest increases were observed in western upland districts, particularly Kalat (16.8%) and Qilla Abdullah (11.3%), indicating spatially concentrated flood-risk expansion rather than widespread national change. These patterns suggest that local topography, drainage conditions and terrain-controlled runoff pathways play an important role in shaping flood-risk transitions across the country. The results show that most of the country remained within the no-change category, whereas increased and decreased risk zones appeared as spatially limited but important transition areas. These changes were not randomly distributed; rather, they were concentrated in selected western, north-western and river-connected districts where local topography, drainage behaviour and land-surface conditions influenced the direction of flood-risk transition. At the district scale, the strongest increase in flood risk is observed in Kalat, where the increased-risk area reaches approximately 16.8%. This is followed by Qilla Abdullah (11.3%), Ziarat (9.2%), Quetta (8.7%), Pishin (8.4%) and Mastung (6.6%) Figure 13a,b.

Alongside these increased-risk areas, several districts show a clear reduction in flood risk extent. The highest declines are observed in Khuzdar (13.2%), Loralai (13.0%) and Qilla Saifullah (12.8%). Additional declines are seen in Kalat (8.6%), South Waziristan (7.3%), Zhob (6.6%) and Musakhel (5.9%). Building on the risk change outcomes, the flood persistence analysis reports areas of repeated high flood susceptibility over the study period. The persistence map demonstrates that the high flood persistence is concentrated primarily along Indus floodplain, the agricultural region of Sindh and the coastal-deltaic region. This spatial pattern follows major drainage and floodplain corridors, indicating that the most exposed areas are not only susceptible in one period but remain repeatedly flood-prone over time. Thus, persistence emphasizes the chronic aspect of flood exposure and the risk change analysis demonstrates the places of transitions Figure 13c.

This persistent flood exposure is especially clear in the district-wise results. Rann of Kutch records the highest high-persistence proportion, with approximately 66.6% of its area falling within the high flood-persistence class. It is followed by Jakobabad (65.0%), Jafarabad (61.1%), Badin (56.5%) and Kashmore (54.1%), Since more than half of the area in these districts remains repeatedly associated with high flood-prone conditions, they represent the strongest persistence hotspots. This persistent flood exposure is especially clear in the district-wise results. Rann of Kutch records the highest high-persistence proportion (66.6%), followed by Jakobabad (65.0%), Jafarabad (61.1%), Badin (56.5%) and Kashmore (54.1%). Larkana (44.6%) and Thatta (40.6%) also show considerable flood persistence within the lower Indus and deltaic environments Figure 13d. These findings indicate that recurrent flood susceptibility is primarily concentrated within river-connected agricultural and floodplain landscapes.

In contrast to these persistent floodplain districts, several western and northern districts show relatively low high-persistence proportions. Gwadar and Chagai record only 0.8% and 0.6%, respectively, while Quetta and Kharan show almost no high flood persistence in the selected district-wise comparison. Similarly, Swat (5.2%) and Gilgit (6.4%) show limited district scale high persistence, although localized flood hazards may still occur along narrow valleys and mountain drainage channels. This contrast is important because it separates broad district level persistence from localized flood events, particularly in mountainous and arid environments where flood prone zones are spatially narrow. The LULC composition further explains the distribution of flood persistence classes Figure 14. The no-persistence class is dominated by barren land (49.3%), followed by vegetation (25.1%) and cropland (23.1%), indicating that non-persistent zones are mainly associated with less flood-connected landscapes. In the low persistence class, vegetation (35.1%), barren land (33.1%) and cropland (28.1%) represent the dominant land-cover categories. Moderate persistence zones are characterized by vegetation (42.4%), barren land (41.0%) and cropland (16.4%), representing transitional environments exposed to periodic flooding.

The strongest exposure pattern occurs in the high flood-persistence class, where cropland becomes the dominant LULC category (45.7%), followed by vegetation (36.3%). Barren land decreases to 12.6%, indicating a transition from barren-dominated non-persistent zones to cropland-dominated persistent zones. These findings demonstrate that recurrent flood-prone areas are strongly associated with agricultural and vegetated floodplain landscapes.

3.7. LULC Exposure to Flood Susceptibility

The LULC exposure values were classified into low, moderate, high, and very high categories based on the district-wise distribution of cropland and built-up areas within high and very high flood-susceptibility zones. The spatial distribution of exposure is presented in Figure 15a, while the district-wise comparison of cropland and built-up exposure is shown in Figure 15b. The results show a clear spatial variation in LULC exposure particularly within high and very high flood-susceptibility zones. Cropland exposure exhibits a wider range compared to built-up areas indicating a noticeable difference in the intensity of land-use exposure across districts. The highest exposure is observed in Badin (17%), followed by Larkana (10.1%), Attock (8.1%), Mirpurkhas (7.1%), and Chakwal (6.1%). These districts fall under the very high exposure category and are primarily located along the Indus floodplain, where agricultural activities are highly concentrated in flood-prone areas.

While built-up area exposure was relatively low, it remains significant as it represents at-risk settlements and infrastructure. The highest built-up area exposure was observed in Atok (2.10%), followed by Larkana (1.90%), Badin (1.80%), Mirpurkhas (1.70%), and Jacobabad (1.60%). This shows that although built-up exposure is smaller than cropland exposure, some districts still contain notable settlement exposure within flood-prone areas.

The results suggest that cultivated land is the dominant LULC type, while built-up area exposure is more localized. This pattern aligns with the results of the flood susceptibility, persistence, and FRPI, with high-risk areas primarily located in the Indus floodplain and other flood-prone riparian zones. Therefore, the LULC exposure analysis confirms that flood risk in these areas is influenced not only by physical flood susceptibility but also by the exposure of agricultural land and built-up areas. Moreover, the identified high-exposure districts were also inundated by floods during the 2010 and 2022 floods, further adding to the exposure. These findings are consistent with the observed LULC expansion trends between 2002 and 2022, particularly the increase in cropland and built-up areas within flood-prone regions.

3.8. Integrated Flood Risk Priority Index Analysis

The FRPI values represent the combined effect of flood susceptible, flood persistence, flood risk change and LULC exposure at the district level. The results show clear differences in flood-risk priority across districts, indicating that the intensity of flood risk varies spatially rather than remaining uniform. The highest FRPI values are observed in Jakobabad (0.742), Rann of Kutch (0.738), Badin (0.711), Kashmore (0.637), and Jafarabad (0.600). These districts constitute the highest priority category, where flood-prone conditions, recurrent flooding, increasing risk tendency and vulnerable land-use areas occur simultaneously with more intensity, leading to a higher priority than the other districts Figure 16a,b.

A moderate priority level is observed in Larkana (0.556), where flood-related conditions remain important but are comparatively less intense than those in the highest-ranked districts. Thatta (0.398), Mirpurkhas (0.385), Shikarpur (0.373), and Muzaffargarh (0.308) have lower FRPI values, which indicate lower combined effects of the flood-related factors. Districts in the very high and high FRPI categories reveal a higher intensity of flood-related conditions, suggesting a greater priority for flood management, drainage improvement, agricultural protection, and land-use planning. By contrast, the districts in moderate and low categories show a relatively less intense influence of these conditions. The FRPI pattern aligns with known flood-prone regions along the Indus floodplain, where higher-priority districts are concentrated. This pattern demonstrates the impact of recurring flood events on the geographical distribution of FRPI values.

3.9. Correlation Between Hydro-Climatic and Land Surface Indicators

To better understand how hydro-climatic forcing interacts with land-surface conditions, a Pearson correlation analysis was carried out for 2022 using LST, air temperature, precipitation, NDVI, and NDWI, as shown in Figure 17. LST is positively correlated with air temperature (r = 0.83), which means that districts with higher near-surface air temperatures tend to have higher skin temperatures. Conversely, LST has a strong negative correlation with precipitation (r = −0.66), and air temperature shows a strong negative correlation with precipitation (r = −0.65). These trends indicate a consistently hot, dry regime across large parts of southern and south-western Pakistan. For example, districts such as Chagai, Kech, and Kharan have very high mean LST (41–42 °C), low annual precipitation (less than 150 mm), reflecting thermally stressed and moisture deficient environments.

Precipitation has a moderate positive relationship with NDVI (r = 0.35), implying that greener districts are usually wetter and have denser vegetation cover, whereas arid areas are less green. The correlation between vegetation and thermal indicators is weaker: NDVI shows only a weak negative correlation with LST (r = −0.25), suggesting that vegetation cover is influenced not only by temperature but also by localized environmental conditions and land-use heterogeneity. Cooler and wetter northern regions, such as Gilgit, Swat, and Kohistan, exhibit relatively lower LST and higher NDVI values.

NDWI is also an important moisture-based complement of NDVI. It is moderately positively correlated with precipitation (r = 0.50), which suggests that those districts that are receiving more rains have more surface moisture. Similarly, NDWI correlates with NDVI in a moderate positive manner (r = 0.40), which implies that wetter areas are also prone to have more vegetation cover. Contrarily, NDWI is moderately to strongly negatively correlated with LST (r = −0.49), air temperature (r = −0.60). These trends indicate that hotter and more thermally stressed districts are generally drier and more moisture-deficient. For example, districts such as Gilgit, Chitral, Kohistan, Swat, and Battagram show relatively higher NDWI values, whereas districts such as Disputed Area 1, Jamshoro, Kholu, Mastung, and Barkhan have lower NDWI values, reflecting comparatively drier surface conditions.

3.10. Performance Evaluation of the Model

The AUC values of the flood susceptibility model were 0.86 in 2002, 0.85 in 2012, and 0.87 in 2022, all within the “very good” performance range (0.80–0.90). This indicates that the model can reliably distinguish between flood-inundated and non-flood-inundated areas within Pakistan. Furthermore, the AUC values remained highly similar over time, indicating that the model’s performance remained stable and reliable despite interannual variations in hydroclimatic forcing and surface conditions Figure 18.

The confusion matrix statistics also support the robustness of the flood susceptibility model. The overall accuracy for 2002, 2012, and 2022 was 84%, 83%, and 85%, respectively, indicating that most pixels were correctly classified. The sensitivity values were 86%, 85%, and 87%, respectively, indicating that most observed flood-inundated areas were located within the predicted high-risk and very high-risk areas. The specificity values were 82%, 81%, and 83%, respectively, indicating that non-flood-inundated areas were also reliably identified.

The FRPI model’s AUC of 0.85 is also within the “very good” performance range. Its confusion matrix results further confirm its reliable performance, with an overall accuracy of 82%, sensitivity of 83%, and specificity of 85%. This figure demonstrates that the FRPI model, after integrating susceptibility, persistence, flood risk change, and LULC based exposure information, can effectively identify flood-vulnerable and non-flood-vulnerable areas.

The high consistency between the 2022 flood susceptibility validation results and the FRPI validation results indicates that this integrated risk prioritization framework maintains strong predictive power even after incorporating exposure-based risk information. The validation results confirm that the outputs of both the susceptibility and priority indices are robust and suitable for supporting district-level flood risk assessment and planning.

4. Discussion

This study used a GIS framework based on the Analytic Hierarchy Process to integrate hydro-climate, topographic, surface, and LULC indicators to assess flood susceptibility and risk prioritization patterns in Pakistan [70,71]. The results show that flood susceptibility is not determined by a single factor, but is influenced by a combination of factors including rainfall variability, thermal conditions, topography, river proximity, surface humidity, vegetation status, and land use exposure [14]. In 2002, 2012, and 2022, high-susceptibility and very high-susceptibility areas were mainly concentrated along the Indus River, in low-lying floodplains, and in coastal delta regions. This pattern indicates that flood risk in Pakistan is spatially concentrated rather than spatially extensive; therefore, even a relatively small high-risk area can have major management importance when it overlaps with agricultural floodplains and settlements [21].

Our findings at the district level also confirm this trend. These districts are mainly located within or near the lower Indus floodplain and coastal Sindh, where low elevation, gentle slopes, river proximity, alluvial plains, and dense drainage networks favour flood accumulation. In contrast, western arid and high-elevation districts generally show lower susceptibility, although localized flash-flood risk may still occur in narrow valleys and piedmont areas. These findings are consistent with previous studies conducted in the Indus floodplain and lower Sindh regions, which identified low elevation, river proximity, and alluvial floodplain environments as major controls of flood susceptibility.

The hydro-climatic and land-surface results suggest that rainfall alone does not fully explain the flood-susceptibility pattern while air temperature and LST remained higher in lowland and southern regions, `hotter regions are generally more moisture-deficient. These relationships indicate that flooding behaviour is affected by the interaction of precipitation, temperature, surface moisture, and soil moisture, not just by rainfall magnitude. Similar interactions among precipitation, temperature, vegetation condition, and surface moisture have also been reported in previous flood susceptibility studies in South Asia and Pakistan.

The LULC results show that flood risk is increasingly becoming an exposure-related issue. This transition indicates a gradual shift from open surfaces toward agricultural and settlement landscapes. This means that even if the overall high-risk area is not expanding dramatically, exposure of productive agricultural land within flood-prone zones is increasing. Flood persistence analysis adds a temporal dimension to this interpretation. This shift indicates that repeated flood-prone conditions are closely linked with agricultural floodplain environments. Previous studies have also highlighted that land-use change, particularly agricultural expansion and urban growth, increases flood exposure and vulnerability in floodplain regions. However, most earlier studies considered LULC mainly as a flood-conditioning or vulnerability factor, whereas the present study explicitly quantifies LULC-based exposure within flood-prone and persistent flood-risk zones. This provides a clearer understanding of how agricultural and built-up land systems intersect with recurrent flood-prone environments. This finding agrees with earlier studies reporting that agricultural expansion and settlement growth within floodplains substantially increase flood exposure and vulnerability.

Although the LULC classification achieved acceptable overall accuracies (79.8–83.9%) and Kappa coefficients (0.79–0.82), some uncertainty may remain in the discrimination of built-up, barren land, and transitional land-cover classes. Consequently, minor uncertainty may propagate into district-level LULC exposure estimates. However, the achieved accuracy levels are consistent with previous national-scale studies and are considered suitable for regional-scale flood-risk assessment and prioritization.

Flood-risk change analysis shows that Pakistan’s flood-risk pattern has both stable and changing components. Much of the country remains under no-change conditions, confirming that the national flood-susceptibility structure is relatively stable. However, the presence of localized increases and decreases in flood risk suggests that flood-prone environments are not static and may evolve over time in response to changing hydro-climatic and land-surface conditions. This suggests that flood risk includes both persistent hotspots along established flood corridors and emerging transition zones where susceptibility is changing over time. The FRPI integrates these separate results into a district-level priority framework by combining susceptibility, persistence, flood-risk change, and LULC exposure. This integration enables the identification of areas where flood-prone conditions, repeated flooding, changing risk, and exposed land-use systems overlap most strongly. Unlike conventional flood-susceptibility studies that primarily produce static, single-period hazard maps, the FRPI moves the analysis beyond susceptibility mapping by integrating susceptibility, persistence, flood-risk change, and LULC-based exposure into a unified, temporally explicit, district-level priority framework. This approach enables the identification not only of where flood-prone areas exist, but also where flood conditions persist, intensify, and increasingly intersect with human land-use systems.

The Google Earth-based regional interpretation also supports the spatial meaning of the modelled results. Representative locations across floodplains, agricultural regions, and coastal environments illustrate diverse flood-risk settings, including low-lying terrain, river proximity, agricultural expansion, settlement development, drainage connectivity, and coastal exposure. These observations confirm that the mapped high-risk zones correspond well with real-world land-surface conditions and flood-prone environments. The validation results further support the reliability of the proposed framework. The obtained validation performance indicates a high level of predictive accuracy and demonstrates the robustness of the integrated approach in identifying flood-prone areas and supporting district-level flood-risk prioritization. These findings suggest that the framework can provide a reliable basis for flood-risk assessment and management in data-scarce environments.

Overall, the findings shows that flood risk in Pakistan is not only a question of where flooding is physically likely to occur. It is also shaped by where flood-prone conditions repeat, where risk is changing, and where cropland and settlements are increasingly exposed. The main contribution of this study is therefore the integration of susceptibility, persistence, change, LULC exposure, FRPI, and regional visual interpretation into a single flood-risk understanding.

Planning and Flood Risk Management Implications

Regional flood risk interpretation combines spatial flood risk patterns with high-resolution Google Earth imagery to more clearly demonstrate the correspondence between model results and actual conditions Figure 19. Flood risk maps show that high-risk areas are mainly concentrated along the Indus River, particularly in the middle and lower floodplains and the coastal areas of Sindh. These areas are typically characterized by low-lying terrain, flat topography, and dense drainage networks.

The spatial distribution of these high-risk areas closely matches areas with intensive agricultural activity and dense settlements, indicating that flood-prone areas are influenced not only by topographical conditions but also by land use patterns. Representative locations selected, including Tata, Dadu, Naushera, Rajanpur, Chiniot, and Gwadar, demonstrate how different environmental and land use patterns interact within identified flood risk areas.

Google Earth-based illustrations visually confirm these spatial patterns by showcasing river proximity, irrigated farmland, settlement expansion, and low-lying floodplain environments within high-risk and very high-risk areas. The high consistency between model outputs and ground observations demonstrates the reliability and robustness of the applied analytical framework. Furthermore, the observed spatial patterns indicate that the northern and high-elevation regions are generally less prone to flooding due to steep terrain and limited floodplain development; while the lowlands and delta regions are more susceptible to flooding due to gentler slopes, poor infiltration capacity, and concentrated human activities.

These outcomes could assist national and provincial disaster risks management authorities through the provision of spatially explicit data on the areas where interventions are the most required. Structural measures like drainage rehabilitation and embankment maintenance and floodwater diversion planning can be concentrated in high-persistence districts and monitoring, early warning and preparedness may be focused in increased-risk districts. Further, the cropland-controlling high-persistence areas can be identified to inform land-use zoning and adaptation approaches in agriculture, especially in the lower Indus floodplain and coastal-deltaic areas. This study offers a more specific framework to minimizing flood losses by incorporating flood susceptibility, risk change, persistence and LULC exposure, which facilitates better resource allocation and climate-resilient land-use planning in Pakistan.

5. Conclusions

This study developed an integrated GIS-AHP framework to assess flood susceptibility, temporal flood dynamics, LULC-based exposure, and district-level flood-risk priority across Pakistan for 2002, 2012, and 2022. By combining hydro-climatic, topographic, hydrological, land-surface, and LULC indicators, the study provides a comprehensive national-scale understanding of how physical flood susceptibility interacts with changing land use and exposure.

The results show that high and very high flood susceptibility is consistently concentrated along the Indus River corridor, lower floodplains, and coastal Sindh, where low elevation, gentle slopes, drainage concentration, and river proximity create favourable flood-prone conditions. Beyond identifying flood-prone areas, the study demonstrates how flood persistence, flood-risk change, and LULC exposure interact spatially and temporally, providing a more comprehensive understanding of flood-risk dynamics than conventional single-period susceptibility assessments. Although the highest susceptibility zones occupy a relatively limited proportion of the national area, they overlap strongly with agriculturally and socio-economically important landscapes. A key finding of this study is that flood risk in Pakistan is increasingly shaped by exposure. Cropland and built-up areas have expanded within flood-prone zones, indicating that productive land, settlements, and infrastructure are becoming more vulnerable to flooding. Flood persistence analysis further identified recurrent hotspots in the lower Indus and coastal districts, while flood-risk change analysis highlighted both stable flood corridors and localized emerging risk zones. The FRPI strengthened the assessment by integrating susceptibility, persistence, risk change, and LULC exposure into a single district-level priority index. High FRPI values in Jacobabad, Rann of Kutch, Badin, Kashmore, and Jafarabad indicate districts where multiple flood-risk dimensions overlap. Validation results confirmed the reliability of the framework, with strong AUC values for both flood susceptibility and FRPI.

Overall, this study demonstrates that flood risk in Pakistan is not merely a natural disaster issue but also an exposure-driven risk problem. This integrated framework provides a solid foundation for identifying priority flood risk zones and can support future flood risk assessments, land use planning, and climate adaptation decisions. The proposed FRPI framework provides a practical district-level prioritization tool that can support targeted flood-risk management, agricultural protection, land-use planning, and climate adaptation strategies in data-limited environments.