Rainwater and Snowmelt Harvesting Suitability in Chitral, Eastern Hindu Kush, Pakistan

Abstract

Water scarcity has emerged as a major environmental challenge in rain shadow regions of mountainous areas worldwide. The Hindu Kush–Karakorum–Himalaya, although recognized as one of the world’s major “water towers”, is experiencing increasing water scarcity due to climate change and rapid population growth, threatening ecosystems, economies, and livelihoods. The Chitral River basin in northwestern Pakistan is a snow-dominated watershed where meltwater and seasonal precipitation generate substantial runoff during spring, followed by prolonged periods of acute water scarcity. Water harvesting represents a potential strategy for sustainable water resource management in this region. This study employed a hybrid approach combining Multi-Criteria Decision Analysis (MCDA), based on Analytical Hierarchy Process (AHP), with geospatial techniques to identify potentially suitable locations for rainwater and snowmelt harvesting (RSH). The results indicate that Chitral can be divided into four RSH suitability classes: very highly suitable (19.1%), highly suitable (43.8%), moderately suitable (22.1%), and less/non-suitable (15.0%). Evaluation of the suitability map using existing water-harvesting structures, satellite imagery, field observations, and expert and local resident interviews demonstrated the robustness of the methodology. The findings provide a spatial framework to support stakeholders in identifying suitable sites for RSH and contribute to the scientific basis for sustainable water management and climate change adaptation. The approach can also be adapted to other mountainous basins with similar hydroclimatic and geomorphological characteristics.

1. Introduction

Water is one of the most precious natural resources for biological survival and is vital to global socio-economic development [1,2,3]. However, water resources are depleting and becoming increasingly scarce worldwide due to climate change, shifting precipitation patterns, rapid population growth, and excessive utilization. Water crises are widespread, and water scarcity has emerged as a critical global challenge to sustainable development, particularly in developing countries [4,5,6]. This growing crisis has led the United Nations to declare an “Era of Global Water Bankruptcy” [7]. According to the UN, 1.8 billion people experienced absolute water scarcity (<500 m³/year/capita) in 2025. It is estimated that half of the world’s population will face chronic water scarcity by 2030 [8]. This situation is further exacerbated in arid and semi-arid regions, where water scarcity may displace between 24 and 700 million people by 2030 [9,10].

The Hindu Kush–Karakoram–Himalaya (HKH) region is the largest freshwater reservoir outside the polar regions and commonly known as the “water tower of Asia”. Most major South Asian rivers originate in this region, sustaining the water, food, livelihoods, and energy security of more than two billion people. However, the HKH is highly sensitive and vulnerable to climate change. Over the past three to four decades, the region has emerged as a climate change hotspot, warming at approximately twice the global average rate and significantly jeopardizing the livelihoods of more than 240 million people in the highlands and 1.65 billion in the adjacent lowlands [10]. In addition, precipitation patterns have changed, with snowfall becoming less frequent and occurring later in the season or not at all. These changes are placing increasing growing pressure on river flows, freshwater resources, and spring-fed hydrological systems, threatening water availability and security, livelihood stability, and socio-ecological resilience across the region.

Water scarcity has emerged as a one of the most pressing environmental and development challenges in Pakistan, which has been classified as a “red zone” in water stress assessments [11,12,13]. Per capita water availability in Pakistan declined from 2490 m³ in 1955 to 1000 m³ in 2007 and is projected to decrease to 500 m³ by 2035 [11,12,14]. Consequently, Pakistan was declared a water-stressed country, with deteriorating conditions leading to its classification as a water-scarce country in 1991 and 2007. In the semiarid Eastern Hindu Kush of northern Pakistan, irrigation agriculture forms the foundation of livelihoods and subsistence, with more than 70% of the population directly dependent on it [15].

To safeguard long-term water security in the face of climate change and mitigate increasing hydrological stress, it is crucial to implement integrated and climate-resilient water governance strategies throughout the HKH in alignment with the UN Sustainable Development Goals 6, 13, and 15. Rainwater harvesting (RWH) refers to the collection and storage of surplus runoff during the rainy season for water utilization during dry periods [16,17]. A RWH system consists of a catchment or runoff area, a storage facility (e.g., ponds, reservoirs), and a target (i.e., domestic use, crop production) [18]. RWH has emerged as an effective strategy for mitigating climate change-induced water scarcity by improving watershed resilience and enhancing ground water recharge [19,20,21,22].

Several review studies on RWH in arid and semi-arid regions as well as on adaptation to climate change are available [2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29]. Regional studies on RWH were carried out in many countries, for example, Australia [30,31], China [32], Egypt [33], Ethiopia [34,35,36], India [37], Iran [38], Iraq [5,39,40,41,42], Nepal [43,44]; Nigeria [45], Pakistan [46,47], Palestine [48,49], South Africa [50], and Tunisia [5]. Several studies describe RWH techniques, practices, suitability mapping and structures [51,52]. For the HKH region, authors focused on topics such as RWH in relation to climate change [53], crop yields [54,55], food security [56], gender [57], and water management [58]. Previous studies in Pakistan [46,47] and China [32] have emphasized potential harvesting zones and structural interventions, while research in Nepal and India have examined the role of micro-harvesting intervention in support of mountain agriculture and water security [43,44]. However, existing research predominantly emphasizes rainfall-driven systems solely, while ignoring the dual snow and glacier-melt contribution in high-altitude snow/ice-dominated watersheds. In rugged mountain glaciated catchments of the HKH, water is governed not only by precipitation but also by snowpack accumulation, temperature-dependent melt processes, and topographic controls. Nevertheless, the combined/hybrid assessment of rainwater and snowmelt harvesting (RSH) has not yet been explored in the HKH region.

Identifying suitable locations and appropriate technologies is essential for the successful implementation of RWH practices [59,60,61] and remains a challenging task [62,63]. Various approaches, including Boolean analysis [64,65,66], design storms, linear programming, analytical probability, random matrix-based non-parametric methods, and machine-learning algorithms such as support vector machines [67] and random forest [68], have been employed to identify suitable sites for RWH. Among these, Analytical Hierarchy Process (AHP)-based Multi-Criteria Decision Analysis (MCDA), integrated with GIS and remote sensing, has proven particularly suitable, especially because of its time efficiency [69,70,71,72,73]. This hybrid approach provides a powerful tool for evaluating multiple spatial variables and synthesizing complex biophysical and environmental conditions that are critical for identifying suitable sites for rainwater harvesting [74,75,76]. It has also been highly recommended as an effective tool for problem-solving and decision-making in rainwater harvesting planning [75].

The Chitral River basin represents one of the most hydrologically sensitive and climatically vulnerable regions within the HKH, where climate change is projected to cause major hydrological disruptions, further increasing vulnerability and exacerbating water scarcity. Addressing the growing water shortage is therefore a priority, particularly in the face of climate change, population growth, and the increasing demand for water and food. However, the potential of water harvesting in Chitral remains largely unexplored, and we found no peer-reviewed studies specifically addressing RSH suitability in the region. A study conducted in Ghazi Tehsil, in the foothills of the Pakistani Himalaya, demonstrated the importance of geospatial techniques for identifying potential rainwater harvesting sites [12]. Similarly, a different study developed a relation-based dam suitability analysis (RDSA) technique to identify suitable sites for larger dams in the Sulaiman Mountains, a southern extension of the broader Hindu Kush mountain system [77].

This research aims to identify suitable zones for water harvesting in Chitral by integrating AHP-based MCDA with geospatial techniques. Given that water availability in the semi-arid Chitral River basin depends on rainwater and snowmelt, we investigate the potential of both sources for water harvesting (hence, we use RSH rather than RWH). Our findings will provide insights for stakeholders and decision-makers in developing effective climate change adaptation and mitigation strategies for Chitral.

2. Study Area

The Chitral River Basin is situated in the extreme northwestern part (35–37° N; 71–74° E) of Pakistan (Figure 1). Administratively, it is composed of two districts, Chitral Lower and Chitral Upper, situated in the Khyber Pakhtunkhwa province. It covers an area of 14,850 km² with a density of 35 persons/km². In 2023, Chitral was home to a population of 515,935 [78].

Figure 1. (A) Chitral in northwestern Pakistan; (B) mean summer land surface temperature in Chitral in the north of the Khyber Pakhtunkhwa province; (C) topography of Chitral.

Chitral’s climate is the subtropical continental highland type, characterized by long cold winters and short warm summers. The semi-arid high mountain climate shows seasonal extremes in both temperature and precipitation, with a mean annual temperature of 16 °C and mean annual precipitation of 460 mm. Temperatures remain high during the entire summer season. Precipitation is inconsistent and unevenly distributed during the year, with the highest mean precipitation of 345.2 mm falling mainly as snow from Western Disturbances during the winter season from December to April, accounting for around 74% of the total annual precipitation [10]. Precipitation increases from December on and reaches a maximum in March.

The hydrological regime is predominantly determined by three main water sources: glacier and snowmelt, rainfall, and springs. Glacier and snowmelt are the principal contributors to streamflow, particularly during the spring and early summer months, followed by glacier melt during peak summer. Both play an important role in regulating seasonal water availability. The mean annual discharge of the Chitral River is about 292 m³/s, with peak flows reaching up to 856 m³/s in the summer season, while minimum flows of about 68 m³/s occur in winter. Precipitation contributes much less to streamflow because of the semi-arid conditions and high interannual variability. The mean annual precipitation is relatively low and unevenly distributed, with a significant portion occurring as winter snowfall. Mountain springs always have been the primary source of freshwater, with more than 70% of people relying on them for drinking water, agriculture, and domestic use [8,79,80]. Currently, springs are drying up and disappearing rapidly due to climate change as well as land use and land cover change (LULCC).

3. Materials and Methods

3.1. Data Sources

This study used satellite imagery analyses to describe topography, lineament density, drainage density, gridded precipitation, Land Use Land Cover (LULC), and runoff (using temperature). Secondary data were retrieved for geology, soils, and precipitation (Table 1).

Table 1. Data sources used in this study.

Category	Dataset	Source
Geology	Lithology map	Geological Survey of Pakistan
Lineaments, LULC	Landsat 8 (30 m)	US Geological Survey—Earth Explorer (https://earthexplorer.usgs.gov/, accessed on 15 March 2026)
	Sentinel 2B (10 m)	Copernicus Data Space Ecosystem (https://dataspace.copernicus.eu/, accessed on 20 March 2026)
Slope, Drainage	ALOS PALSAR DEM (12.5 m)	Alaska Satellite Facility (ASF) (https://asf.alaska.edu/)
Soils	Soil map	Soil and Water Conservation Department, Khyber Pakhtunkhwa province
Precipitation	Precipitation (1980–2025)	Regional Meteorological Center, Peshawar
	CHIRPS	UCSB CHC Data Portal/GEE (https://www.chc.ucsb.edu/data/chirps, accessed on 15 April 2026)
Temperature (Snowmelt)	ERA5-Land Daily Air Temperature	European Centre for Medium-Range Weather Forecasts (ECMWF)—Copernicus Climate Change Service (C3S)
Surface Runoff	Arc SWAT (SCS-Curve Number method)	Daily precipitation (CHIRPS), LULC, hydrologic soil group, and DEM, ERA5-Land temperature data

3.2. Preparation of Thematic Layers

The selection of suitable sites for RSH is influenced by geomorphological, hydrological, and socio-ecological factors such as LULC [4,76,81]. In this study, eight parameters were selected that represent important characteristics of the region and are widely used in RSH studies: geology, lineament density, slope, drainage density, soil type, precipitation, LULC, and runoff (Figure 2).

Figure 2. Methodological framework of this study.

3.2.1. Geology

Geology significantly determines soil development in a watershed. Geological structures regulate groundwater flow, which involves the movement from surface streams to subsurface aquifers, and vice versa [12]. Data about the general geology was acquired from the Geological Survey of Pakistan and divided into nine classes: Carboniferous rocks; Cretaceous rocks; Devonian and Silurian rocks; igneous, metamorphic and mafic intrusive rocks; Late Paleozoic rocks; Mesozoic rocks; snow caps and glaciers; and unmapped areas.

3.2.2. Lineament Density

Lineaments are linear features in the landscape that represent zones of structural weakness, such as faults, joints, and fractures [75] that have a great influence on water flow, water storage, groundwater recharge, and subsurface flow of water [82]. While areas with high lineament density are not suitable for RSH, areas with low lineament density are [83]. Lineament density mapping was carried out using Landsat-8 satellite images from 2025. Principal Component Analysis (PCA), a widely used method for enhancing geological features, was employed for extraction of lineaments using the Multivariate tool, and PC2 was selected based on its improved contrast. Lineaments were delineated, converted to vector format, and used to generate/compute lineament density using the Line Density tool. The final map displays five classes of lineament densities: very low—0.5 km/km²; low—0.5–0.9 km/km²; moderate—1.0–1.5 km/km²; high—1.6–2 km/km²; very high—>2 km/km².

3.2.3. Slope

Slope is important for the determination of runoff patterns and velocities as well as groundwater recharge [12,84]. The slope map was extracted from an ALOS PALSAR digital elevation model (DEM). Slope was then divided into five classes: gentle—<15°; moderate—15–30°; steep 31–45°; very steep—45–60°; extremely steep/cliff—>60° [85].

3.2.4. Drainage Density

Runoff volume and infiltration are directly related to the drainage density; lower drainage density generally relates to lower RSH potential [84]. Drainage density is defined as the total stream length per unit catchment area. Drainage density is expressed as:

$$\mathit{D}\mathit{D}={\displaystyle \frac{{\sum }_{\mathit{i}=\mathbf{1}}^{\mathit{n}}\mathit{L}}{\mathit{A}}}$$

(1)

where DD is the drainage density, n is the number of streams, L is the stream length (km), and A is the watershed area (km²).

Drainage density was calculated through DEM-based hydrological modeling (Figure 3). Initially, the DEM was preprocessed using the Fill function to remove sinks. Then, flow direction was computed using the D8 (Eight direction) algorithm, followed by the Flow Accumulation tool to delineate primary drainage channels. Stream initiation was defined using a flow-accumulation threshold of 5000 cells, corresponding to a critical contributing area of approximately 0.78 km². This threshold was selected through sensitivity testing across contributing areas ranging from 0.25 to 2.0 km² and spatial validation against the global HydroRIVERS hydrographic database. The selected threshold provided a reasonable representation of the drainage network while minimizing ephemeral hillslope artifacts and numerical DEM noise characteristic of high-mountain landscapes. Streams were extracted and classified into eight orders using the Horton–Strahler method. The stream network was converted into vector format using the Stream to Feature tool. Finally, the drainage density was calculated using the Line Density tool. Drainage density was divided into five classes: very low—<2 km/km²; low—2–2.6 km/km²; moderate—2.7–3.3 km/km²; high—3.4–4.0 km/km²; very high—>4.0 km/km² [41].

Figure 3. Workflow for the DEM-based calculation of drainage density.

3.2.5. Soil Type

In soil, its texture determines the rate of infiltration, volume of runoff, and water holding capability [85]. A texture that permits a high infiltration rate is unsuitable for RSH because water easily percolates to deeper layers. Fine-grained soil with smaller pores, on the other hand, provides effective water storage [81]. Soil data obtained from the Soil and Water Conservation Department were divided into four classes: glaciers and snow caps; shallow loamy and calcareous soil; shallow, mainly loamy soil; and very shallow loamy soil and rock outcrops.

3.2.6. Precipitation

Monthly gridded precipitation data from CHIRPS were used due to the lack of ground-based observations in higher elevations. Only two met stations are available in Chitral, both situated in valley bottom locations at around 1500 m asl., but they are not adequately representing the precipitation patterns in Chitral, particularly in higher elevations.

CHIRPS gridded precipitation data were converted into points, and an interpolation was performed using Inverse Distance Weighting (IDW). CHIRPS precipitation data of 0.05° resolution were preprocessed, harmonized, and downscaled to a common spatial resolution of 30 m to ensure consistency and comparability in the GIS-based AHP analysis. CHIRPS estimates for Chitral and Drosh were validated against ground-based meteorological observations from the corresponding stations (Figure 4). Following validation, monthly CHIRPS precipitation data were aggregated to derive total annual precipitation. Based on regional climate studies [12,86], and to account for the pronounced orographic moisture gradient across the eastern Hindu Kush mountain range in Chitral, total annual precipitation was divided into five classes: very low—<250 mm; low—251–350 mm; moderate—351–450 mm; high—451–550 mm; very high—>550 mm. This classification distinguishes hyper-arid valley floors (with annual precipitation below 250 mm) from high-altitude moisture accumulation zones (receiving more than 550 mm annually), primarily sustained by Western Disturbances. Furthermore, these precipitation thresholds are consistent with regional agroclimatic aridity benchmarks established by the Pakistan Meteorological Department (PMD).

Figure 4. Validation of CHIRPS precipitation data against ground-based stations of Chitral and Drosh, Pakistan.

3.2.7. Land Use and Land Cover

LULC determines surface water runoff and infiltration [66,81]. Sentinel-2B multispectral satellite images from 2025 were classified using a supervised classification procedure and maximum likelihood algorithm. The seven LULC classes are barren land, snow/ice, water, forest, cropland, rangeland, and built-up area. The validation of the classes included field surveys and visual interpretation of high-resolution Google Earth imagery. Kappa statistics indicate an overall accuracy of >88%.

3.2.8. Runoff

The ArcSWAT model has been widely used for runoff estimation. In this study, both surface runoff and snowmelt runoff were simulated using ArcSWAT version 2012.10_8.26, with the ArcGIS 10.8 interface. Surface runoff was estimated using the Soil Conservation Service (SCS) Curve Number method, whereas snowmelt runoff was simulated using a temperature index (Degree-Day) approach. The input datasets included precipitation, temperature, topography, soil, land use/land cover (LULC), and river discharge data from the Chitral gauging station. All spatial datasets were projected to a common coordinate system using UTM Zone 43° N. A 12.5-m-reolution ALOS PALSAR DEM was used to delineate the entire watershed, which was subsequently divided into 23 sub-basins. Soil, LULC, and slope datasets were then incorporated into the SWAT model. Following the overlay of these spatial datasets, the model generated 156 hydrological response units (HRUs), with 100% spatial coverage within the delineated watershed boundaries.

The soil conservation service curve number (SCS-CN) method was developed by the United States Natural Resource Service (NRCS) in 1950 to estimate surface runoff from precipitation. It is widely used because of its simplicity, consistency, and broad applicability. The method incorporates land use/land cover (LULC), hydrologic soil group (HSG), and precipitation data. The LULC and HSG characteristics are combined to derive a Curve Number (CN), which generally ranges from 30 to 100; higher CN values indicate greater runoff potential and, consequently, a larger portion of precipitation becoming surface runoff. In this study, the SCS-CN was calculated as:

$${\mathit{Q}}_{\mathit{S}\mathit{u}\mathit{r}}={\displaystyle \frac{({\mathit{P}-{\mathit{I}}_{\mathit{a}})}^{\mathbf{2}}}{\left(\mathit{P}-{\mathit{I}}_{\mathit{a}}\right)+\mathit{S}}}$$

(2)

where Q_sur is the depth of surface runoff (mm), P is the depth of daily precipitation (mm), I_a is the initial storage (mm), and S is the surface storage. The surface storage was calculated as follows:

$$\mathit{S}=\mathbf{25.4}\left({\displaystyle \frac{\mathbf{1000}}{\mathit{C}\mathit{N}}}-\mathbf{10}\right)$$

(3)

where CN is the catchment curve number ranging from 0 to 100.

Snowmelt runoff can be estimated using various hydrological models, including SRM [87,88], UBC [89,90], CRHM [91,92], ARHYTHM [93], and SWAT [94,95]. Among these, the SWAT model has been widely applied to simulate and predict snowmelt runoff. Therefore, this study employed ArcSWAT with the temperature index (Degree-Day) method.

In the ArcSWAT model, daily mean air temperature (TA) is used to partition precipitation into snowfall and rain. When TA is below the snowfall threshold temperature (T_snow), precipitation is classified as snowfall. When TA exceeds 0 °C, precipitation is classified as rainfall. Snowfall accumulates on the ground as snowpack, and the amount of water contained within the snowpack is expressed as snow water equivalent (SWE). The snowpack increases with additional snowfall and decreases through processes such as snowmelt and sublimation. The snowpack mass balance was computed using:

$$\mathit{S}\mathit{W}{\mathit{E}}_{\mathit{i}}=\mathit{S}\mathit{W}{\mathit{E}}_{\mathit{i}-\mathbf{1}}+{\mathit{R}}_{\mathit{d}\mathit{a}\mathit{y}\mathit{i}}-{\mathit{E}}_{\mathit{s}\mathit{u}\mathit{b}\mathit{i}}-\mathit{S}\mathit{N}{\mathit{O}}_{\mathit{m}\mathit{e}\mathit{l}\mathit{t}}$$

(4)

where SWE_i is the snow water equivalent (mm H₂O), R_dayi is the precipitation as snowfall (mm H₂O), E_sub is the snow sublimation (mm H₂O), SNO_melt is the amount of snowmelt on a given day (mm H₂O), and SWE_i−1 is the snow water equivalent from the previous day (mm H₂O).

Snow is not uniformly distributed throughout the Chitral River Basin because of snow drifting, terrain shading, and complex topography. In this study, an aerial depletion curve was used to represent the seasonal expansion and reduction of the snow-covered area as a function of the amount of snow stored in the basin. The snow-covered area in the ArcSWAT model was estimated as:

$$\mathit{S}\mathit{N}{\mathit{O}}_{\mathit{c}\mathit{o}\mathit{v}\mathit{e}\mathit{r}}={\displaystyle \frac{\mathit{S}\mathit{N}\mathit{O}}{\mathit{S}\mathit{N}\mathit{O}\mathit{C}\mathit{o}\mathit{v}\mathit{M}\mathit{x}}}{\left[{\displaystyle \frac{\mathit{S}\mathit{N}{\mathit{O}}_{\mathit{i}}}{\mathit{S}\mathit{N}\mathit{O}\mathit{C}\mathit{o}\mathit{v}\mathit{M}\mathit{x}}}+\mathit{e}\mathit{x}\mathit{p}\left(\mathit{C}\mathit{o}{\mathit{v}}_{\mathbf{1}}-\mathit{C}\mathit{o}{\mathit{v}}_{\mathbf{2}}\times {\displaystyle \frac{\mathit{S}\mathit{N}{\mathit{O}}_{\mathit{i}}}{\mathit{S}\mathit{N}\mathit{O}\mathit{C}\mathit{o}\mathit{v}\mathit{M}\mathit{x}}}\right)\right]}^{-\mathbf{1}}$$

(5)

where SNOcovi is the percentage of snow-covered area, SNOCovMx is the minimum snow content corresponding to 100% snow cover, and Cov₁ and Cov₂ are coefficients that define the shape of the snow cover depletion curve.

Snowmelt begins when the snowpack and temperature conditions exceed the threshold required for snowmelt runoff generation. The ArcSWAT snowmelt module uses a temperature index (Degree-Day) approach to estimate snowmelt. The Degree-Day factor represents the rate of snowmelt per degree of temperature above the melting threshold and may vary with time. Snowmelt was calculated as:

$$\mathit{S}\mathit{N}{\mathit{O}}_{\mathit{m}\mathit{e}\mathit{l}\mathit{t}}={\mathit{b}}_{\mathit{m}\mathit{e}\mathit{l}\mathit{t}}\times \mathit{S}\mathit{N}{\mathit{O}}_{\mathit{c}\mathit{o}\mathit{v}\mathit{i}}\left[{\displaystyle \frac{{\mathit{T}}_{\mathit{s}\mathit{p}\mathit{i}}+{\mathit{T}}_{\mathit{m}\mathit{a}\mathit{x}\mathit{i}}}{\mathbf{2}}}-\mathit{S}\mathit{M}\mathit{T}\mathit{M}\mathit{P}\right]$$

(6)

where SNO_melt is the amount of snowmelt on the current day, T_maxi is the maximum temperature on the current day, T_spi is the snowpack temperature on the current day, and SMTMP is the snowmelt onset temperature (i.e., the temperature threshold for snowmelt), and b_mlti is the snowmelt factor for the current day, which was calculated using:

$${\mathit{b}}_{\mathit{m}\mathit{l}\mathit{t}\mathit{i}}={\displaystyle \frac{\mathit{S}\mathit{M}\mathit{F}\mathit{M}\mathit{x}+\mathit{S}\mathit{m}\mathit{F}\mathit{M}\mathit{n}}{\mathbf{2}}}+{\displaystyle \frac{\mathit{S}\mathit{M}\mathit{F}\mathit{M}\mathit{x}-\mathit{S}\mathit{m}\mathit{F}\mathit{M}\mathit{n}}{\mathbf{2}}}\times \mathit{s}\mathit{i}\mathit{n}\left[{\displaystyle \frac{\mathbf{2}\mathit{\pi }}{\mathbf{365}}}\left(\mathit{t}\mathbf{-}\mathbf{81}\right)\right]$$

(7)

where SMFMN and SMFMX are the minimum and maximum snowmelt factors, corresponding to 21 June and 21 December, respectively, and i denotes the day of the year.

In SWAT, the snowpack temperature is estimated as a function of the mean air temperature of the previous day and a temperature damping factor. The snowpack temperature of the previous day influences the current snowpack temperature through the delay factor (TIMP) that accounts for factors affecting the response of snowpack temperature to changes in air temperature, including snow depth, snowpack density, and degree of exposure. The snowpack temperature was calculated as:

$${\mathit{T}}_{\mathit{s}\mathit{p}\mathit{i}}={\mathit{T}}_{\mathit{s}\mathit{p}\mathit{i}-\mathbf{1}}\left(\mathbf{1}-\mathit{T}\mathit{I}\mathit{M}\mathit{P}\right)+{\mathit{T}}_{\mathit{a}\mathit{i}}\times \mathit{T}\mathit{I}\mathit{M}\mathit{P}$$

(8)

where T_spi is the snowpack temperature on the current day, T_spi−1 is the snowpack temperature on the previous day, Tai is the air temperature of the current day, and TIMP is the snowpack temperature delay factor.

3.2.9. Evaluation of Runoff Simulation

The SWAT Calibration and Uncertainty Program (SWAT-CUP) was used for sensitivity analysis, model calibration, and validation. SWAT-CUP incorporates several algorithms, including Parameter Solution (ParaSol), Generalized Likelihood Uncertainty Estimation (GLUE), Markov Chain Monte Carlo (MCMC), Particle Swarm Optimization (PSO), and Sequential Uncertainty Fitting 2 (SUFI-2). In this study, we used the latter. Sensitivity analysis was conducted to assess parameter uncertainty and identify the key parameters governing runoff generation and, consequently, RSH potential. Fifteen parameters were evaluated using approximately 200 simulation iterations (Table 2). Parameter sensitivity was quantified using the $\mathit{t}$-statistic and $\mathit{p}$-value. The SCS runoff curve number (CN₂) was identified as one of the most sensitive parameters in the study area. Parameters showing high statistical significance ($\mathit{p}\mathbf{<}\mathbf{0.05}$) were prioritized for model calibration and validation.

Table 2. SWAT parameters evaluated for sensitivity analysis in Chitral.

Parameter	Description	Calibrated Range	Fitted Value	t-Statistic	p-Value	Sensitivity Level
CN2	SC runoff curve number (±%)	−0.2 to +0.2	−0.07	15.9	0.000	very high
PLAPS	Precipitation Lapse Rate (mm/km)	100 to 700	350	13.8	0.001	very high
SFTMP	Snowfall temp (°C)	−2 to +3	1.25	12.3	0.003	High
ALPHA_BF	Baseflow alpha factor	0.01 to 0.1	0.45	9.4	0.016	High
GW_DELAY	Groundwater delay	10 to 150	69.0	10.6	0.008	High
SMTMP	Snowmelt base temp (°C)	−2 to +3	1.8	8.6	0.013	Moderate
GWQMN	Groundwater return flow	0 to 2000	125	8.3	0.021	Moderate
TIMP	Snowpack temperature lag factor	0.01 to 1.0	0.36	7.7	0.025	moderate
SMFMX	Maximum snowmelt factor (mm H2O)	0.25 to 8.0	5.0	6.3	0.028	Moderate
GW_REVAP	Groundwater movement toward root zone	0.02 to 2.0	0.52	6.7	0.035	low
RCHRG_DP	Deep aquifer recharge	0.00 to 0.2	0.05	3.5	0.052	low
SOL_AWC	Soil available water capacity	−0.25 to +0.25	0.07	3.8	0.123	low
CH_N2	Main-channel roughness	0.03 to 0.15	0.06	4.8	0.089	low
SLSUBBSN	Average slope length (m)	−0.2 to +2.0	−0.10	3.8	0.065	low

The ArcSWAT model was simulated for the period 2000–2025 using discharge data from the Chitral gauging station for model calibration and validation (Figure 5). The first two years, 2000–2001, were used as a warm-up period to minimize the effects of initial model conditions. The period 2002–2012 was used for model calibration, while 2013–2025 was used for model validation.

Figure 5. Simulated and observed river discharge during (A) calibration and (B) validation.

The performance of the ArcSWAT model was evaluated in SWAT-CUP using several statistical metrics, including the Nash–Sutcliff Efficiency (NSE), Percentage Bias (PBIAS), coefficient of determination (R²), p-factor and r-factor based on comparisons between observed and simulated discharge.

NSE is widely used in hydrological modeling and measures the relative magnitude of the residual variance compared with variance of observed data. The NSE metric ranges from −∞ and 1.0, with the value of 1.0 indicating perfect agreement between observed and simulated values. In general, NSE values between 0.0 and 1.0 indicate acceptable model performance, whereas values ≤ 0.0 indicate poor model performance. NSE was computed as:

$$\mathit{N}\mathit{S}\mathit{E}=\mathbf{1}-\left[{\displaystyle \frac{{\sum }_{\mathit{i}=\mathbf{1}}^{\mathit{n}}({{\mathit{Q}}_{\mathit{i}}^{\mathit{o}\mathit{b}}-{\mathit{Q}}_{\mathit{i}}^{\mathit{s}\mathit{i}\mathit{m}})}^{\mathbf{2}}}{{\sum }_{\mathit{i}=\mathbf{1}}^{\mathit{n}}({{\mathit{Q}}_{\mathit{i}}^{\mathit{o}\mathit{b}}-{\overline{\mathit{Q}}}^{\mathit{obs}})}^{\mathbf{2}}}}\right]$$

(9)

where ${\mathit{Q}}_{\mathit{i}}^{\mathit{o}\mathit{b}}$ and ${\mathit{Q}}_{\mathit{i}}^{\mathit{s}\mathit{i}\mathit{m}}$ are the ith observed and simulated values, respectively, and ${\overline{\mathit{Q}}}^{\mathit{obs}}$ is the mean of the observed values.

PBIAS measures the average tendency of the simulated values to overestimate or underestimate the corresponding observed values. The optimal PBIAS value is 0.0, indicating perfect agreement between the observed and simulated values. Positive values indicate model underestimation, whereas negative values indicate model overestimation. PBIAS was calculated as:

$$\mathit{P}\mathit{B}\mathit{I}\mathit{A}\mathit{s}=\left({\displaystyle \frac{{\sum }_{\mathit{i}=\mathbf{1}}^{\mathit{n}}\left({\mathit{Q}}_{\mathit{i}}^{\mathit{o}\mathit{b}}-{\mathit{Q}}_{\mathit{i}}^{\mathit{s}\mathit{i}\mathit{m}}\right)\times \mathbf{100}}{{\sum }_{\mathit{i}=\mathbf{1}}^{\mathit{n}}{\mathit{Q}}_{\mathit{i}}^{\mathit{o}\mathit{b}}}}\right)$$

(10)

where ${\mathit{Q}}_{\mathit{i}}^{\mathit{o}\mathit{b}}$, and ${\mathit{Q}}_{\mathit{i}}^{\mathit{s}\mathit{i}\mathit{m}}$ are the ith observed and simulated values, respectively.

R² describes the agreement between the simulated and observed values of each simulation. A well-performing model generally has an R² value close to 1, whereas values greater than 0.5 are generally considered acceptable. R² was calculated as:

$${\mathit{R}}^{\mathbf{2}}={\left({\displaystyle \frac{{\sum }_{\mathit{i}=\mathbf{1}}^{\mathit{N}}\left({\mathit{Q}}_{\mathit{i}}^{\mathit{o}\mathit{b}}-{\mathit{q}}^{\mathit{o}\mathit{b}}\right)\times ({\mathit{Q}}_{\mathit{i}}^{\mathit{s}\mathit{i}\mathit{m}}-{\mathit{q}}^{\mathit{s}\mathit{i}\mathit{m}})}{\sqrt{{\sum }_{\mathit{i}=\mathbf{1}}^{\mathit{N}}{\left({\mathit{Q}}_{\mathit{i}}^{\mathit{o}\mathit{b}}-{\mathit{q}}^{\mathit{o}\mathit{b}}\right)}^{\mathbf{2}}\sqrt{{\sum }_{\mathit{i}=\mathbf{1}}^{\mathit{N}}{({\mathit{Q}}_{\mathit{i}}^{\mathit{s}\mathit{i}\mathit{m}}-{\mathit{q}}^{\mathit{s}\mathit{i}\mathit{m}})}^{\mathbf{2}}}}}}\right)}^{\mathbf{2}},\mathbf{0}\le ={\mathit{R}}^{\mathbf{2}}\le \mathbf{1}$$

(11)

The model achieved an NSE of 0.76 during calibration and 0.89 during validation (Table 3). The corresponding PBIAS values were −3.2% and 2.1%, respectively, while the R² values were 0.87 during calibration and 0.79 during validation. Further, the model performance was also evaluated using p-factor and r-factor during calibration and validation. These statistics confirm the low model uncertainty and accurate representation of runoff generation (Table 3).

Table 3. Model performance metrics.

Period	NSE	PBIAS (%)	R2	p-Factor	r-Factor
Calibration (2002–2012)	0.76	−3.2	0.87	0.84	0.81
Validation (2013–2025)	0.89	2.1	0.79	0.86	0.78

3.3. Weighting and Normalization

Assigning weights to each input layer is a crucial step as results significantly rely on their accuracy. Various methods have been used in water harvesting studies, for instance, Weighted Linear Combination (WLC), Entropy Weight Method, Multi-Influencing Factor (MIF), frequency ratio, certainty factor, weights-of-evidence, fuzzy logic, and multi-influencing factor. However, the Analytic Hierarchy Process (AHP) is the most popular and widely used heuristic method. This study integrated AHP with Multi-Criteria Decision Analysis (MCDA) and GIS. All spatial datasets including the 12.5 m DEM, 30 m Landsat imagery, 10 m Sentinel-2 data, and precipitation CHIRPS (0.05°) were preprocessed, harmonized, and standardized to a common spatial resolution using appropriate resampling, upscaling and downscaling methods to ensure consistency and comparability in the GIS-based analysis. This hybrid approach employed multiple steps.

For AHP-Based RSH suitability analysis, an 8 × 8 pairwise comparison matrix (P) was constructed to evaluate the relative importance of selected thematic layers and assign quantitative weights. The criteria comprised C₁: geology, C₂: lineament density, C₃: slope, C₄: drainage density, C₅: soil type, C₆: precipitation, C₇: LULC, and C₈: runoff. The matrix is expressed as:

$$\mathit{P}=\left[\begin{array}{cccc}{\mathit{P}}_{\mathbf{11}}& {\mathit{P}}_{\mathbf{12}}& \dots & {\mathit{P}}_{\mathbf{1}\mathit{n}}\\ {\mathit{P}}_{\mathbf{21}}& {\mathit{P}}_{\mathbf{22}}& \dots & {\mathit{P}}_{\mathbf{2}\mathit{n}}\\ .& .& .& .\\ .& .& .& .\\ {\mathit{P}}_{\mathit{n}\mathbf{1}}& {\mathit{P}}_{\mathit{n}\mathbf{2}}& \dots & {\mathit{P}}_{\mathit{n}\mathit{n}}\end{array}\right]$$

(12)

The Saaty’s scale assigns the rank to two selected criteria based on their significance [96]. The ranks range from 1 to 9, where 1 indicates equal importance, while 9 indicates an extremely higher importance of one criterion over the other (Table 4). For example, in this study, an area with the lowest rainfall was assigned Rank 1, and an area with the highest rainfall was assigned Rank 9, i.e., conditions favorable for RSH were assigned the highest rank. An evaluation of the pairwise ranking was carried out using past studies, observed field conditions, and local experts.

Table 4. Scale of relative importance of criteria after Saaty [97].

Scale	Definition	Explanation
1	Equal importance	Two objectives contribute equally to the objective
2	Weak	Compromise between 1 and 3
3	Moderate importance	Experience and judgment slightly factor one activity over another
4	Moderate plus	Compromise between 3 and 5
5	Strong importance	Experience and judgment strongly favor one activity over another
6	Strong plus	Compromise between 5 and 7
7	Very strong	An activity is favored very strongly over another; its dominance demonstrated in practice
8	Very, very strong	Compromise between 7 and 9
9	Extreme importance	The evidence favoring one activity over another is of the highest possible order of affirmation
Reciprocals	If activity i has one of the above non-zero numbers assigned to it when compared with activity j, then j has the reciprocal value when compared with j	A reasonable assumption
Rational	Ratios arising from the scale	If consistency were to be forced by obtaining n numerical values to span the matrix

During the 2024 and 2025 field campaigns, 25 experienced local experts, all older than 45 years, were selected using a stratified purposive sampling technique. The expert panel comprised five academics/researchers, five farmers, five GIS analysists, five hydrologists, and five water resource engineers. Each expert independently completed a pairwise comparison matrix using Saaty’s fundamental scale. The individual judgements were aggregated into a single group pairwise comparison matrix using the Geometric Mean Method to minimize the influence of individual judgements and derive a representative group decision. Consistency ratios were calculated for both individual and aggregated group matrices. Matrices with CR > 0.1 were reviewed and either revised or excluded from the analysis.

The normalized weight (${\mathit{W}}_{\mathit{n}}$) for each criterion was calculated as:

$${\mathit{W}}_{\mathit{n}}={\displaystyle \frac{\mathit{G}{\mathit{M}}_{\mathit{n}}}{{\sum }_{\mathit{n}=\mathbf{1}}^{\mathit{n}}\mathit{G}{\mathit{M}}_{\mathit{n}}}}$$

(13)

where ${\mathit{W}}_{\mathit{n}}$ is the normalized weight of the nth criterion, $\mathit{G}{\mathit{M}}_{\mathit{n}}$ is the geometric mean of a nth row of the pairwise comparison matrix.

The geometric mean ($\mathit{G}{\mathit{M}}_{\mathit{n}}$) can be computed as:

$$\mathit{G}{\mathit{M}}_{\mathit{n}}=\sqrt[\mathit{n}]{{\mathit{P}}_{\mathbf{1}\mathit{n}}\times {\mathit{P}}_{\mathbf{2}\mathit{n}}\times \dots \dots .\times {\mathit{P}}_{\mathit{m}\mathit{n}}}$$

(14)

where P is the pairwise comparison matrix.

The consistency analysis of the pairwise comparison was carried out using:

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

(15)

where $\mathit{C}\mathit{R}$ is the consistency ratio, $\mathit{C}\mathit{I}$ is the consistency index, and $\mathit{R}\mathit{I}$ is the random index obtained from [50].

The consistency index was computed as:

$$\mathit{C}\mathit{I}={\displaystyle \frac{{\mathit{\lambda }}_{\mathit{m}\mathit{a}\mathit{x}}-{\mathit{n}}_{\mathit{i}}}{{\mathit{n}}_{\mathit{i}}-\mathbf{1}}}$$

(16)

where ${\mathit{\lambda }}_{\mathbf{m}\mathbf{a}\mathbf{x}}$ is the maximum eigenvalue of the pairwise comparison matrix that can be computed as:

$${\mathit{\lambda }}_{\mathbf{m}\mathbf{a}\mathbf{x}}={\sum }_{\mathit{i}=\mathbf{1}}^{\mathit{n}}{\displaystyle \frac{{\left(\mathit{P}\mathit{W}\right)}_{\mathit{n}}}{{{\mathit{n}}_{\mathit{i}}\mathit{W}}_{\mathit{e}\mathit{n}}}}$$

(17)

where $(\mathit{P}\mathit{W}{)}_{\mathit{n}}$ is the weighted sum value of the ${\mathit{n}}^{\mathit{t}\mathit{h}}$ row, and ${\mathit{W}}_{\mathit{e}\mathit{n}}$ is the normalized eigenvector weight of the ${\mathit{n}}^{\mathit{t}\mathit{h}}$ criterion. A threshold of CR < 0.1 indicates acceptable consistency.

3.4. GIS-Based MCDA and Weighted Overlay Analysis

Following consistency validation, all thematic layers were reclassified into suitability ranks and integrated within a GIS environment using a weighted linear combination (WLC) approach. The Rainwater and Snowmelt Harvesting Potential Index (RSHPI) was computed as:

$$\mathit{R}\mathit{S}\mathit{H}\mathit{P}\mathit{I}={\sum }_{\mathit{i}=\mathbf{1}}^{\mathit{n}}{\mathit{W}}_{\mathit{i}}\hspace{0.17em}{\mathit{X}}_{\mathit{i}}$$

(18)

where ${\mathit{W}}_{\mathit{i}}$ is the weight derived from AHP, and ${\mathit{X}}_{\mathit{i}}$ is the rank assigned to each spatial class.

3.5. Ground Truthing of Results

The RSH results were evaluated using the Receiver Operating Characteristic (ROC) curve and the Area Under the Curve (AUC). Suitability scores of 363 existing water-harvesting structures were compared with those at an equal number of pseudo-absence points generated from locations without existing structures. The structures included check dams, farm ponds, gully plugs, terraces, and water ponds. The pseudo-absence points were spatially constrained to locations more than 500 m from existing structures and outside active stream channels to minimize potential spatial sampling bias.

4. Results

4.1. Water Harvesting Suitability Based on Thematic Layers

4.1.1. Geology

The RSH suitability in Chitral based on geology is as follows: 42% is very highly or highly suitable; 18% is moderately suitable; and 40% is less or not suitable (Figure 6, Table 5).

Figure 6. (A) Geology of Chitral; (B) RSH suitability based on geology.

Table 5. RSH suitability in Chitral based on geology.

Geology	Saaty’s Scale	Suitability	Area (km2)	Area (%)
Paleozoic Rocks	1	Unsuitable	4009	27
Mesozoic/Cretaceous Rocks	2	Less suitable	1931	13
Sedimentary Rock	3	Moderately suitable	2673	18
Snow and Glaciers	4	Highly suitable	3267	22
Igneous/Metamorphic Rock	5	Very highly suitable	2970	20

4.1.2. Lineament Density

The RSH suitability in Chitral based on lineament density is as follows: 68% is very highly or highly suitable, particularly in Lower Chitral; 15% is moderately suitable; and 17% is less or not suitable (Figure 7, Table 6).

Figure 7. (A) Lineament density in Chitral; (B) RSH suitability based on lineament density.

Table 6. RSH suitability in Chitral based on lineament density.

Lineament Density (km/km2)	Saaty’s Scale	Suitability	Area (km2)	Area (%)
Very high (>2.0)	1	Unsuitable	891	6
High (1.6–2.0)	2	Less suitable	1634	11
Moderate (1.0–1.5)	3	Moderately suitable	2227	15
Low (0.5–0.9)	4	Highly suitable	4900	33
Very low (<0.5)	5	Very highly suitable	5198	35

4.1.3. Slope

The RSH suitability in Chitral based on slope is as follows: 16% is very highly or highly suitable; 28% is moderately suitable; 56% is less or not suitable (Figure 8, Table 7).

Figure 8. (A) Slope across Chitral; (B) RSH suitability based on slope.

Table 7. RSH suitability in Chitral based on slope.

Slope (Degrees)	Saaty’s Scale	Suitability	Area (km2)	Area (%)
>60	1	Unsuitable	2376	16
46–60	2	Less suitable	5940	40
31–45	3	Moderately suitable	4158	28
15–30	4	Highly suitable	1485	10
<15	5	Very highly suitable	891	6

4.1.4. Drainage Density

The RSH suitability in Chitral based on drainage density is as follows: 49% is very highly to highly suitable; 25% is moderately suitable; 23% is less or not suitable (Figure 9, Table 8).

Figure 9. (A) Drainage density in Chitral; (B) RSH suitability based on drainage density.

Table 8. RSH suitability in Chitral based on drainage density.

Drainage Density (km/km2)	Saaty’s Scale	Suitability	Area (km2)	Area (%)
Very low (<2)	1	Unsuitable	1188	8
Low (2.0–2.6)	2	Less suitable	2227	15
Moderate (2.7–3.3)	3	Moderately suitable	4158	25
High (3.4–4.0)	4	Highly suitable	4604	31
Very high (>4.0)	5	Very highly suitable	2673	18

4.1.5. Soil Type

The RSH suitability in Chitral based on soil type is as follows: 7% is very highly or highly suitable; 73% is moderately suitable; and 20% is less or not suitable (Figure 10, Table 9).

Figure 10. (A) Soil types in Chitral; (B) RSH suitability based on soil type.

Table 9. RSH suitability in Chitral based on soil type.

Soil Type	Saaty’s Scale	Suitability	Area (km2)	Area (%)
Glaciers, snowcaps	1 and 2	Unsuitable or less suitable	2970	20
Loamy, shallow, non-calcareous	3	Moderately suitable	10,840	73
Mainly loamy, shallow	4	Highly suitable	594	4
Rock outcrops, loamy, very shallow	5	Very highly suitable	446	3

4.1.6. Precipitation

The RSH suitability in Chitral based on precipitation is as follows: 35% is very highly or highly suitable; 31% is moderately suitable; and 24% is less or not suitable (Figure 11, Table 10).

Figure 11. (A) Mean annual precipitation in Chitral; (B) RSH suitability based on precipitation.

Table 10. RSH suitability in Chitral based on precipitation.

Precipitation (mm)	Saaty’s Scale	Suitability	Area (km2)	Area (%)
<250	1	Unsuitable	2673	18
251–350	2	Less suitable	2376	16
351–450	3	Moderately suitable	4604	31
451–550	4	Highly suitable	2970	20
>550	5	Very highly suitable	2227	15

4.1.7. LULC

The RSH suitability in Chitral based on LULC is as follows: 75% is very highly or highly suitable; 8% is moderately suitable; and 17% is less or not suitable (Figure 12, Table 11).

Figure 12. (A) LULC in Chitral; (B) RSH suitability based on LULC.

Table 11. RSH suitability in Chitral based on LULC.

LULC	Saaty’s Scale	Suitability	Area (km2)	Area (%)
Settlement, built-up	1	Unsuitable	2079	14
Water bodies, snow, glaciers	2	Less suitable	446	3
Agriculture	3	Moderately suitable	1188	8
Forest, vegetation	4	Highly suitable	6089	41
Barren land, rangeland	5	Very highly suitable	5049	34

4.1.8. Runoff

The RSH suitability in Chitral based on runoff is as follows: 14% is very highly or highly suitable; 50% is moderately suitable; and 36% is less or not suitable (Figure 13, Table 12).

Figure 13. (A) Runoff in Chitral; (B) RSH suitability based on runoff.

Table 12. RSH suitability in Chitral based on runoff.

Runoff (mm)	Saaty’s Scale	Suitability	Area (km2)	Area (%)
<100	1	Unsuitable	1636	11
100–250	2	Less suitable	3697	25
251–500	3	Moderately suitable	7423	50
501–750	4	Highly suitable	1441	10
>750	5	Very highly suitable	653	4

4.2. Thematic Layers’ Weights and Multi-Criteria Decision Analysis

To quantify the relative importance of the individual environmental characteristics in water harvesting, we carried out an AHP-based pairwise comparison (Table 13). Precipitation (W = 0.25) is the most controlling factor, followed by runoff (W = 0.22) and slope (W = 0.14). LULC (W = 0.10) and drainage density (W = 0.09) show a moderate influence, while lineament density (W = 0.07), soil type (W = 0.06), and geology (W = 0.06) have an only low influence. Reliability testing of the results included the calculation of Eigenvalue (λ_max), Consistency Index (CI), Random Index (RI), and Consistency Ratio (CR). The CR (0.02) indicates that the pairwise comparison is logically consistent, and the derived weights are reliable.

Table 13. Statistical results for selected environmental factors in estimations of RSH suitability in Chitral. PR: precipitation, LL: Land Use Land Cover, SL: slope, DD: drainage density, LD: lineament density, SO: soil type, GE: geology, RU: runoff; (a) AHP pairwise comparison matrix and normalized pairwise comparison matrix; (b) $W_{en}$: normalized eigenvector weight, $(PW{)}_n$: weighted sum vector, CI: consistency index, RI: random index, CR: consistency ratio.

(a)
Layer	AHP Pairwise Comparison Matrix									Normalized Pairwise Comparison Matrix
	PR	LL	SL		DD	LD	SO	GE	RU	PR	LL	SL	DD	LD	SO	GE	RU
PR	1	1	3		5	5	5	7	9	0.29	0.26	0.42	0.31	0.23	0.18	0.17	0.16
SL	1/3	1/3	1		1	5	5	7	9	0.10	0.09	0.14	0.06	0.23	0.18	0.17	0.16
LL	1/3	1/5	1		1	1/3	5	7	7	0.10	0.05	0.14	0.06	0.02	0.18	0.17	0.12
DD	1/5	1/3	1/3		1	3	3	5	5	0.06	0.09	0.05	0.06	0.14	0.11	0.12	0.09
LD	1/5	1/3	1/3		1	1	3	3	7	0.06	0.09	0.05	0.06	0.05	0.11	0.07	0.12
SO	1/5	1/3	1/3		1	1	1	3	5	0.06	0.09	0.05	0.06	0.05	0.04	0.07	0.09
GE	1/7	1/3	1/5		1	1	1	3	7	0.04	0.09	0.03	0.06	0.05	0.04	0.07	0.12
RU	1	1	1		5	5	5	7	9	0.29	0.26	0.14	0.31	0.23	0.18	0.17	0.16
SUM	3.41	3.87	7.20		16.0	21.33	28.0	42.0	58.0	1.00	1.00	1.00	1.00	1.00	1.00	1.00	1.00
(b)
Layer	Normalized Eigenvector						$(\mathit{P}\mathit{W}{)}_{\mathit{n}}$		${\displaystyle \frac{{\left(\mathit{P}\mathit{W}\right)}_{\mathit{n}}}{{\mathit{W}}_{\mathit{e}\mathit{n}}}}$		${\mathit{\lambda }}_{\mathbf{m}\mathbf{a}\mathbf{x}}$	$\mathbf{C}\mathbf{I}$		RI		CR
	Weight (${\mathit{W}}_{\mathit{e}\mathit{n}}$)			%
PR	0.25			25			2.25		8.91		8.21	0.03		1.41		0.02
SL	0.14			14			1.24		8.83
LL	0.10			10			0.85		8.14
DD	0.09			9			0.78		8.79
LD	0.07			7			0.60		7.98
SO	0.06			6			0.45		7.23
GE	0.06			6			0.42		6.74
RU	0.22			22			1.97		9.05
SUM	1.00			100

4.3. Water Harvesting Suitability Map

The final RSH harvesting suitability map shows that Chitral can be divided into four distinct zones (Figure 14); the zones “less suitable” and “unsuitable” were combined because the former represents only 0.2% of Chitral’s area, preventing its display in the final map: 19.1% (2836 km²) is very highly suitable; 43.8% (6504 km²) is highly suitability; 22.1% (3282 km²) is moderately suitable; and 15.0% (2228 km²) is less/unsuitable. By combining the first three zones (85%), Chitral displays an overall strong potential for RSH. Only a smaller part of Chitral is less/unsuitable for RSH because of unfavorable factors such as high elevations, steep slopes, and built-up areas.

Figure 14. (A) RSH harvesting suitability in Chitral; (B) validation of the modeled RSH suitability zones based on traditional water harvesting structures.

The comparison of the simulated RSH results with the locations of the 363 existing water harvesting structures revealed that 292 structures (81%) were located in areas of very high or high suitability, 45 structures (12%) were located in areas of moderate suitability, and 26 structures (7%) were located in areas of low or no suitability (Figure 14 and Figure 15). The ROC curve lies above the diagonal reference line, with an AUC of 0.82, indicating very good discriminatory performance (Figure 16). This AUC value indicates an approximately 82% probability that a randomly selected existing water-harvesting structure location has a higher RSHPI than a randomly selected location without structures. Overall, these results indicate that the proposed approach provides a reliable framework for identifying suitable zones for RSH.

Figure 15. Traditional ponds near Qaqlast in the highly suitable RSH zone of Upper Chitral. (Photos: Z. Ahmad, April 2026).

Figure 16. Validation using ROC curve revealed an AUC of 0.82, indicating good discriminatory performance.

5. Discussion

In our study on water harvesting in Chitral, we included runoff from rain and snow and, therefore, introduced the term “rainwater and snowmelt harvesting” (RSH). Surprisingly, we could not find any other study that estimated runoff from ice and/or snow.

Our results support previous studies that advocate for a combined assessment of climatic, hydrologic, topographic, and geologic criteria when identifying suitable zones for water harvesting [12,98,99,100]. Our findings demonstrate that, in Chitral, geology, runoff, lineament density, drainage density, and soil type all influence RSH suitability, while precipitation, runoff, and slope emerge as the key controlling parameters, with precipitation exerting the strongest influence.

The emergence of precipitation as the primary governing factor can be attributed to the semi-arid, high-mountain hydroclimatic conditions of the Chitral River basin, which functions as a source-limited moisture system. Unlike transport-limited humid regions, where hydrological processes are generally less constrained by atmospheric moisture availability, such processes in Chitral are strongly governed by atmospheric moisture supply rather than landscape transport capacity. Precipitation is sparse, highly variable, and often erratic, while the basin’s hydrological yield depends substantially on high-altitude winter snowfall [10]. Thus, precipitation represents the primary physical gateway determining the macro-scale feasibility of RSH in the CRB. The southern parts of Chitral receive greater amounts of summer rainfall associated with monsoonal depressions, which helps explain the concentration of very high or high RSH suitability zones in this region. In contrast, middle and upper parts of Chitral have a more arid climate and receive a greater portion of their precipitation during the winter from Western Disturbances [10], resulting in comparatively lower RSH suitability.

Furthermore, slope is particularly important in Chitral because of its complex topography that ranges from relatively flat valley bottoms to extremely steep alpine terrain. Slope influences runoff velocity, infiltration opportunity, water accumulation, erosion, and the physical feasibility and spatial distribution of water-harvesting structures. Our findings that slope is a highly influential factor in determining water harvesting suitability are consistent with previous assessments [21,101,102,103].

An interesting contribution to the selection of suitable RWH sites comes from a study in Iraq, which advocates incorporating ecological criteria relevant to ecosystem health, including temperature, light, wind speed and direction, humidity, and evaporation [42,104]. Among these criteria, temperature and light were identified as the most influential factors. The inclusion of these ecological criteria altered the ranking of sites compared with rankings based solely on conventional suitability criteria.

The integrated temperature index (Degree-Day) and AHP-based MCDA framework provides an effective tool for RSH suitability mapping. However, while it is computationally effective and suitable for data-scarce regions like Chitral, it simplifies complex snowpack ablation dynamics by relying primarily on temperature and precipitation while omitting energy balance variables and other parameters such as solar radiation, ground heat flux, snow depth, freeze–thaw cycles, wind speed etc. Additionally, AHP-based MCDA relies on expert judgment for criteria weighting, and the spatial thematic layers (e.g., geology, lineament density, drainage density, and slope) are subject to the temporal and spatial resolution of available satellite datasets, which may smooth out micro-topographic or localized geological variations.

Our findings show that 63% of Chitral is classified as very highly or highly suitable for RSH, while an additional 22% is still moderately suitable. In contrast, a study conducted in Gilgit-Baltistan, a snow-dominated region of the HKH of Pakistan that borders Chitral to the east, found that only 4% of the region is highly suitable for RWH, and 40% is moderately suitable [99]. One possible explanation for this discrepancy in water harvesting suitability is the difference in methodological approaches. The Gilgit-Baltistan study estimated snowmelt from rainfall depths and direct runoff using the SCS-CN method, whereas our approach explicitly incorporates snowmelt runoff as a separate hydrological component. Furthermore, the study does not clearly explain the criteria or thresholds used to define the high, medium, low, and unsuitable suitability classes. A different study conducted in the Ghazi Tehsil, Khyber Pakhtunkhwa, classified 28% of the region as suitable, 52% less suitable, and 20% not suitable using the SCS-CN method [12]. Again, the contribution of snowmelt to runoff was not considered in the analysis.

An interesting water harvesting method and adaptation to seasonal water scarcity was developed by Chewang Norphelin in Ladakh of the Indian Himalaya in the mid-1990s in the form of man-made “artificial glaciers” that slowly release meltwater during the spring and summer [105,106,107,108].

6. Conclusions

Water harvesting provides a viable, practical, and climate-resilient strategy for sustainable water resource management in high-mountain regions vulnerable to climate change and population growth. Surprisingly, water harvesting suitability studies in snow-dominated mountain regions have largely focused on rainwater, despite the substantial contribution of snowmelt and, in some basins, glacier melt to surface runoff. Incorporating snowmelt based on temperature data therefore represents an important step toward improving water harvesting suitability assessments in mountain environments.

Several environmental parameters—geology, lineament density, slope, drainage density, soil type, precipitation, LULC and runoff—play critical roles in controlling surface runoff generation, infiltration potential, and water storage capacity. The integration of a temperature index (Degree-Day) and AHP-based MCDA approach with RS/GIS techniques enables the identification and spatial delineation of RSH suitability zones. In semi-arid Chitral, where snow covers the ground for many months of the year, rainwater and snowmelt harvesting (RSH) represent a promising strategy for addressing water scarcity. More than 60% of the area is either very highly or highly suitable for RSH, while an additional 22% is moderately suitable. The resulting suitability map provides a valuable decision-support tool for stakeholders by identifying priority areas for harvesting interventions aimed at enhancing water availability for agriculture and domestic use. These maps can thus support the development of strategies to address water scarcity and strengthen climate resilience in the Eastern Hindu Kush of northern Pakistan.

However, several methodological limitations remain. Future research should address these limitations in four key directions:

• First, studies should incorporate high-resolution satellite and UAV imagery, together with hydrogeological datasets such as snow water equivalent, topographic wetness index, soil infiltration capacity, and snow cover area and depth, to refine input parameterization.
• Second, advanced modeling approaches should be explored, including the integration of physically based energy-balance snowmelt models (e.g., DHSVM and iSNOBAL) with machine learning and hybrid decision-making methods (e.g., Random Forest and Fuzzy-AHP) to improve runoff estimation and reduce criteria subjectivity in criteria weighting.
• Third, research should bridge the gap between macro-scale suitability modeling and site-specific engineering design by identifying precise locations for structural water-harvesting interventions, such as check dams, snow-storage basins, and aquifer recharge shafts, supported by ground-based geophysical surveys and field validation.
• Fourth, suitability assessments should be downscaled from regional scale to the tehsil and village-council levels to provide actionable decision-support tools for municipal water authorities and local planners.
• Fifth, water-harvesting research in high-mountain environments should also account for runoff contributions from glaciers and snow. We therefore propose an integrated approach to rainwater, ice- and snowmelt harvesting (RISH) for high-mountain regions.