Landslide Susceptibility Mapping Using Geospatial Modelling in the Central Himalaya

Abstract

Landslides are a persistent hazard in the tectonically active Central Himalaya, frequently affecting roads and settlements. However, quantitative assessments of their spatial drivers have remained limited. This study investigated landslide susceptibility along a 90 km section of the Uttarkashi–Gangotri highway to identify dominant triggering factors and generate a reliable risk map. We applied the AHP–GIS framework, guided by a multi-criteria decision-making approach. Nine thematic parameters, such as slope, geology, lineament density, drainage density, proximity to roads, rainfall, aspect, curvature, and land use/land cover were utilised to quantify their relative influence on slope failure. Results showed that slope (23%), geology (22%), and lineament density (21%) were the most influential factors. Secondary contributions came from drainage density (9%), proximity to roads (8%), and rainfall (>231 mm). The susceptibility map was validated using 105 landslide inventory points, with 64 events (61%) located in very high-risk zones and 31 (30%) in high-risk zones. The model achieved a predictive accuracy of 0.817 based on the Area Under the Curve (AUC) metric. High-risk zones are aligned with steep slopes (30–50°), convex curvatures, and barren land, particularly near infrastructure. These findings provide a scientific tool for hazard mitigation and support disaster risk reduction in similar mountainous regions worldwide, contributing to safer infrastructure development.

1. Introduction

Landslides are among the most damaging natural hazards, causing large-scale loss of life, property damage, and environmental degradation in mountainous regions across the world [1]. These events involve the downward movement of rock, debris, or soil under gravity and are often triggered by intense rainfall, earthquakes, snowmelt, or human activities [2,3]. Climate change has increased the frequency of extreme rainfall and disrupted hydrological patterns, resulting in more frequent slope failures, especially in tectonically active areas [4]. Countries such as Japan, Nepal, Indonesia, and the Philippines face high landslide risks due to steep terrain, heavy monsoons, and seismic activity [5].

The 2015 Nepal earthquake caused thousands of landslides, severely damaging infrastructure. Similarly, the 2016 Kaikoura earthquake in New Zealand triggered major slope failures that disrupted transport and ecosystems [6,7]. Beyond physical damage, landslides affect river flow, reduce agricultural productivity, and disturb ecosystems [8,9]. Remote sensing studies in the Andes, Alps, and Rockies show increased landslide frequency due to changing rainfall and temperature trends [10,11]. As global climate and development pressures grow, reliable hazard assessment tools are urgently needed in mountain regions.

India is highly vulnerable to landslides, with about 15% of its land area located in moderate to high-risk zones, especially across the Himalayas [12,13,14]. The collision between the Indian and Eurasian plates creates tectonic instability, while the southwest monsoon brings heavy seasonal rainfall that triggers mass movements [4]. Major incidents, such as 2003 Varunavat landslide in Uttarkashi and 2013 Kedarnath disaster highlight the destructive effects of rainfall and glacial lake outbursts. The Uttarkashi–Gangotri road, a vital route for pilgrims and locals, is frequently disrupted by landslides due to steep slopes, fragile rock formations, and intense rainfall [15]. These events cause serious delays, economic losses, and damage to transport networks, especially during the Char Dham Yatra [16]. Contributing factors include high drainage density, river proximity, deforestation, and poorly planned infrastructure [17]. Although past studies and landslide inventories exist, updated maps that combine natural and human drivers are still lacking.

Understanding landslide processes requires tools that can assess multiple physical and human factors together. Multi-criteria decision-making methods help identify key contributors to landslides [8,18,19]. Among these, the Analytic Hierarchy Process (AHP) is widely used because it allows experts to assign relative weights to different factors through pairwise comparisons [20]. It supports both qualitative knowledge and numerical data and has been used in many mountain regions to assess the role of slope angle, geology, drainage, and land use [21,22,23]. The method’s reliability is tested using the Consistency Ratio [24,25]. When combined with GIS and remote sensing, AHP allows integration of digital elevation models (DEMs), rainfall data, and land cover information to map hazard-prone areas [26,27]. Satellite platforms like Landsat, Sentinel, and Cartosat offer multi-temporal images for tracking slope changes and vegetation cover [28,29]. Platforms like Google Earth Engine (GEE) also enable efficient, large-scale data analysis [30]. These tools make the AHP–GIS approach practical and cost-effective for landslide mapping in data-poor mountain regions.

The Uttarkashi–Gangotri road in the Central Himalaya is critical for local communities and pilgrims but faces regular landslide threats. Past research has mostly focused on isolated slopes or used single-factor models. This leaves a gap in comprehensive studies that combine multiple causes with updated field data [15,16]. Also, many models lacked proper validation, reducing their usefulness for planning. This study fills the gap using the AHP model, which is well-suited for the region’s complex topography and geology. The main objectives are: (1) to identify and rank the key factors affecting landslides in the Uttarkashi–Gangotri corridor, (2) to prepare a high-resolution susceptibility map using AHP–GIS overlay, and (3) to validate the model using 105 recorded landslide locations. This method combines DEMs & satellite data to create a reliable tool for hazard planning. The results aim to support safer infrastructure, better preparedness, and improved resilience in Himalayan Mountain roads.

2. Data and Methods

2.1. Study Area

The study area is located in the Central Himalaya, focusing on a 90 km stretch of the Uttarkashi–Gangotri road (National Highway-34), which lies within the Bhagirathi Valley in Uttarkashi district, Uttarakhand, northern India (Figure 1). Geographically, the region spans between 30.43° to 30.7306° N latitude and 78.27° to 78.4437° E longitude. The district is predominantly mountainous, with a rural population that depends on agriculture, livestock, and tourism for livelihood [31]. National Highway-34 connects the town of Uttarkashi (30.73° N, 78.45° E), the district headquarters, to Gangotri (30.994° N, 78.941° E), a significant religious site on the Char Dham route. This segment of the Central Himalaya is highly prone to landslides due to active tectonics, steep topography, and fragile geological formations [32,33]. The Bhagirathi Valley exhibits rugged terrain with steep slopes, high ridges, and deeply incised valleys typical of the Lesser Himalayan zone [34]. The regional geology comprises ancient Precambrian rocks, including schists, quartzites, phyllites, and gneisses, along with younger sedimentary layers [35]. These rocks are often fractured and weathered, making slopes unstable [36].

The soil varies from sandy loam to clayey loam, with different levels of porosity and cohesion, derived from weathered bedrock [37]. Seasonal monsoon rainfall, combined with unconsolidated soil and steep terrain, significantly increases landslide risk [35]. Covering an area of approximately 1014 km², the region reflects a diverse mix of geological, climatic, and geomorphological conditions, making it a high-risk zone for landslide activity in the Central Himalaya.

2.2. Satellite Data and Thematic Layers

The study utilizes multiple datasets, including SRTM DEM (30 m resolution), Sentinel-2B imagery (10 m resolution), geological data from GSI, and rainfall data from NASA’s POWER platform (

Table 1. Data used in the study and sensor characteristics.

Data	Purpose	Source
Digital Elevation Model (DEM)	Slope, Aspect, Drainage density, Curvature	Shuttle Radar topography Mission (SRTM) Open Topography (https://opentopography.org/)
Sentinel-2B	Land use and land cover, Lineament Density	Copernicus (https://dataspace.copernicus.eu/)
Geology, Geomorphology and Fault Line	Geology, Geomorphology, Main Central Thrust (MCT)	Bhukosh Geological Survey of India (GSI) (https://bhukosh.gsi.gov.in/Bhukosh/Public)
Landslide inventories	Visual demarcation of landslide inventories (Polygon), Visual Demarcation of landslide inventories (points)	Google Earth Pro (https://www.google.com/earth/versions/)
Rainfall data	Average Monthly rainfall of June to September (1990–2020)	Power NASA (https://power.larc.nasa.gov/)
Road line	Distance to road	Open Street Map (https://www.openstreetmap.org/)

). Landslide inventories were created using Google Earth and Sentinel-2B satellite imagery, identifying 105 landslide points based on proximity to roads, rivers, and terrain features. These datasets were processed in ArcGIS (https://www.arcgis.com/index.html) for thematic mapping. Nine causative factors were considered: slope, geology, lineament density, drainage density, distance to road, rainfall, aspect, curvature, and land use/land cover (LULC). Slope was derived from the Digital Elevation Model (DEM) to evaluate its influence on erosion and mass movement, which is closely linked to hydrology, vegetation, and soil depth [38]. Geological information was compiled from regional lithological maps, emphasizing the shear strength and weathering resistance of rock types [39]. Lineament density was mapped using satellite imagery to quantify structural discontinuities such as faults and fractures, which are critical for slope instability [40]. Drainage density was calculated from hydrological networks, reflecting water infiltration, erosion, and surface runoff processes [40]. The distance-to-road factor was introduced to assess anthropogenic influences like slope cutting, excavation, and blasting linked to highway construction [41].

Rainfall data spanning 30 years were integrated to examine the role of precipitation in triggering slope failures in tectonically sensitive terrains [11]. Aspect and curvature were derived from the DEM. Aspect identifies slope orientation and solar exposure, which control moisture and weathering rates, while curvature reflects surface form and runoff concentration [41]. LULC was extracted from satellite imagery to classify vegetation, barren land, snow cover, agriculture, settlements, and river sediments, which strongly influence slope stability [22]. All thematic layers were prepared in a Geographic Information System (GIS) environment. Each parameter was classified into categories, and susceptibility mapping was performed using weighted overlays to identify high-risk areas.

2.3. Analytic Hierarchy Process (AHP)

In this study, AHP was selected for its ability to integrate multiple parameters through expert judgment, making it suitable for regions with limited data and complex terrain [42]. While machine learning models like Random Forest and XGBoost offer strong predictive capabilities, they often function as black-box models and require large, high-quality datasets [43]. In contrast, AHP provides a transparent and interpretable framework, allowing for consistent weighting of diverse contributing factors in landslide susceptibility mapping [44,45]. This methodology is particularly effective for landslide susceptibility analysis, as it allows the integration of various geological and environmental factors in a systematic framework. AHP involves three fundamental steps: decomposition of the problem into a hierarchy of criteria, pairwise comparison of the criteria, and synthesis of the results to determine weights for each parameter. These weights are then used to create a composite index for the final susceptibility analysis.

In this study, nine parameters were selected as inputs for the AHP process: slope, geology, rainfall, drainage density, lineament density, land use/land cover (LULC), curvature, aspect, and distance to roads. Pairwise comparison matrixes were constructed to evaluate the relative importance of these parameters. The comparison was based on a scale ranging from 1 to 9, where 1 indicates equal importance and 9 signifies extreme importance of one parameter over another [42]. The weights were calculated using the Eigenvector method, which ensures that the assigned weights are mathematically consistent.

The consistency of the pairwise comparison matrix was validated using the Consistency Index (CI) and the Consistency Ratio (CR). The CI was calculated using the formula Equation (1).

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

(1)

where $\lambda max$ is the largest Eigenvalue, and $n$ is the number of parameters in the matrix. The CR was then determined using the formula Equation (2).

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

(2)

where $RI$ is the Random Index, a standard value depending on the matrix size [42]. A $CR$ value below 0.1 indicates that the comparisons are consistent and the results are reliable.

Once the weights for each parameter were determined, they were applied to thematic layers in ArcGIS using a weighted overlay method. This process combined the layers based on their relative importance to produce a composite landslide susceptibility map. Model validation was performed by overlaying 105 landslide inventory points onto the susceptibility map. The Area Under the Curve (AUC) value was calculated to assess the accuracy of the model. AUC values close to 1.0 indicate high predictive capability. The methodology for identifying landslide susceptibility zones is outlined in the detailed flowchart (Figure 2), showcasing data sources, analytical processes such as the AHP, and validation steps leading to the creation of the landslide susceptibility map.

3. Results

3.1. Landslide Inventories

Using Google Earth Pro and Sentinel-2A data, a total of 105 landslide inventory points were identified based on proximity to roads, rivers, geomorphology, geology, slope, and rainfall intensity. Major landslides were delineated through polygons, with extensive landslides located predominantly along the Bhagirathi valley near the road section (Figure 3a,b). The total calculated area of landslides was 5.96 km², with the majority occurring at altitudes between 2000–3000 m in moderately to highly dissected hills. Smaller landslides were dispersed throughout the region. For documentation and mapping, polygonal and point-based landslide inventories were created using Google Earth tools (Figure 4a,b).

3.2. Causative Factors

3.2.1. Slope, Aspect, Curvature and Geology

The study area was categorized into six slope classes: below 10°, 10–20°, 20–30°, 30–40°, 40–50°, and above 50°. Most landslides were concentrated in the 30–40° and 40–50° ranges, covering 2.06 km² and 2.24 km², respectively. Steep slopes exceeding 50° were concentrated near Dharali and Gangotri, while gentler slopes below 20° were dominant near Uttarkashi and Gangori (Figure 5a).

The aspect map categorized the slope orientation into several classes: flat terrain (–1), followed by eight directional classes, such as north (0–22.5°), east (67.5–112.5°), northeast (22.5–67.5°), south (157.5–202.5°), southeast (112.5–157.5°), west (247.5–292.5°), southwest (202.5–247.5°), and northwest (292.5–337.5°) (Figure 5b). The majority of landslides occurred on southeast-facing slopes, accounting for 1.43 km² (22.14%), followed by southwest-facing slopes, covering 1.027 km² (15.90%). South- and southeast-facing slopes were particularly prominent near Bhatwari and Dharali, reflecting higher solar radiation exposure, which affected slope stability. West-facing slopes dominated near Uttarkashi and Gangori, while northwest-facing slopes were more concentrated near Gangotri. Northeast-facing slopes were evenly distributed across the region, particularly closer to the NH34 road corridor. Flat areas were minimal, primarily found in river valleys and low-elevation zones, which were less prone to landslides.

The terrain was classified into three curvature types: concave, convex, and flat (Figure 5c). Convex slopes covered the largest area (554.24 km²), followed by concave slopes (382.82 km²), while flat terrain covered the least area (76.94 km²). Convex slopes accounted for the highest percentage of landslides (58.47%), reflecting their association with surface runoff and erosion. Concave slopes, covering 28.47% of landslides, indicated areas prone to material accumulation and water retention, increasing instability. Flat areas, despite their small surface coverage, exhibited minimal landslide occurrences due to stable terrain conditions.

The geological map revealed diverse lithological units, including basic meta-volcanics, crystalline rocks, the Garhwal group (comprising the Beringa, Chamoli, and Nagthani formations), the Martoli group, and the Rakcham Granite (Figure 5d). The study area, located in the Garhwal Lesser Himalaya, exhibited significant geological variation, with basic meta-volcanics concentrated in the southern region near Uttarkashi and Gangotri.

3.2.2. Drainage Density, Lineament Density, Distance-to-Road and Rainfall

The drainage density map categorized the region into five classes: very low (32–120 km/km²), low (130–200 km/km²), moderate (210–280 km/km²), high (290–370 km/km²), and very high (380–450 km/km²) (Figure 6a). Areas with high drainage density accounted for 33.70% of landslides, followed by very high drainage density areas with 28% of landslides. High and very high drainage density zones were predominantly concentrated near Dharali, Gangotri, and Sukhi, indicating active hydrological and erosional processes in these regions. Moderate drainage density dominated the central part of the study area along NH34, where the road interacted with hydrological networks. Very low drainage density areas, primarily near Uttarkashi and Gangotri, corresponded to only 0.97% of landslides, indicating a comparatively stable hydrological regime.

The lineament density was categorized into five classes: very low (0–0.56 km/km²), low (0.56–1.13 km/km²), moderate (1.13–1.70 km/ km²), high (1.71–2.26 km/km²), and very high (2.27–2.8 km²) (Figure 6b). The maximum area under landslide occurrences fell in the moderate lineament density category, covering 3.43 km² (61.33%). Regions of very high lineament density were concentrated near Sukhi and Dharali, indicating increased tectonic activity and structural control in these areas. Moderate density was widespread along the NH34 corridor, emphasizing its influence on road stability and alignment. Very low lineament density was observed near Uttarkashi and Gangotri, reflecting more stable geological conditions.

The distance-to-road map categorized the area into five buffer zones: 0–200, 200–400, 400–600, 600–800, and 800–1000 m (Figure 6c). Approximately 33.76% of landslides occurred within 200–400 m of the road, highlighting the influence of proximity to NH34 on slope stability. The highest density of landslides was observed near townships such as Bhatwari, Sukhi, and Dharali, where road construction activities were most prominent. Landslides were much less frequent in areas beyond 600 m from the road, with the lowest occurrences recorded within the 600–800 m buffer zone.

The rainfall map was categorized into five classes of average annual rainfall over the past 30 years: 155–170 mm, 171–190 mm, 191–210 mm, 211–230 mm, and 231 mm and above (Figure 6d). Most landslides occurred in areas receiving 231 mm and above rainfall, with significant occurrences also observed in zones with 191–210 mm rainfall. The highest rainfall areas, concentrated near Uttarkashi and Gangori, corresponded to regions of increased landslide susceptibility due to intense precipitation and runoff. Moderate rainfall (191–210 mm) dominated the central part of the study area near Bhatwari, which also experienced a notable number of landslides. Lower rainfall areas, such as Dharali and Gangotri (155–170 mm), exhibited fewer landslides, reflecting reduced precipitation impact on slope stability.

3.2.3. LULC

The LULC map categorized the region into rivers, vegetation, snow cover, barren land, river sediments, settlements, and agriculture (Figure 7). Vegetation dominated the landscape (386.17 km²) (

Table 2. Area covered by each LULC class along the Uttarkashi-Gangotri road section.

Land Use/Land Cover	Area (km2)
River	22.65
Vegetation	386.17
Snow cover	280.83
Barren Land	168.76
River Sediment	29.34
Settlement	84.79
Agriculture	46.67

), especially near Uttarkashi and Bhatwari, providing stability to slopes. Snow cover (280.83 km²) was prominent at higher altitudes near Gangotri and Dharali, reflecting minimal anthropogenic activity in these areas. The maximum area of landslides (54.49%) was observed in barren land, indicating a strong correlation between vegetation loss and slope instability. Settlements and agricultural land were concentrated in lower elevations near Uttarkashi and Gangori, reflecting human activities in more stable regions. River sediments along the Bhagirathi River indicated active fluvial processes that could destabilize adjacent slopes.

3.3. Pair-Wise Comparison Matrix and Weights of Thematic Layers for Landslide Susceptibility Analysis

The proportional significance of different criteria in influencing landslide susceptibility was shown by the pairwise comparison matrix of thematic layers and their related weights (

Table 3. Pair-wise comparison matrix of thematic layers and weights.

Parameters	Slope	Geology	Lineament density	Drainage density	Distance to road	Rainfall	Aspect	Curvature	LULC	Weight	Weight Percentage
Slope	1	1	1	2	3	4	5	9	9	0.23	23
Geology	1	1	1	2	3	4	5	6	8	0.22	22
Lineament density	1	1	1	2	3	4	5	6	9	0.21	21
Drainage density	0.5	0.5	0.5	1	1	1	2	3	4	0.09	9
Distance to road	0.333	0.333	0.333	1	1	1	2	3	4	0.08	8
Rainfall	0.25	0.25	0.25	1	1	1	1	2	3	0.06	6
Aspect	0.2	0.2	0.2	0.5	0.5	1	1	2	2	0.05	5
Curvature	0.111	0.167	0.167	0.333	0.333	0.5	0.5	1	2	0.03	3
LULC	0.111	0.125	0.111	0.25	0.25	0.333	0.5	0.5	1	0.02	2
SUM	4.506	4.575	4.561	10.083	13.083	16.833	22	32.5	42	1.00	100
CR value										0.0127

). With a weight of 0.23 (23%), slope stood out as the most significant factor. With a weight of 0.22 (22%), geology ranked second, emphasizing the role of structural features and lithological traits in regulating mass movements. With a weight of 0.21 (21%), lineament density ranked third, highlighting the importance of fault zones and fracture systems in promoting slope failures. Road distance (0.08, 8%) and drainage density (0.09, 9%) were significantly weighted, indicating their considerable, but indirect impact on slope modification and erosion. Rainfall (0.06, 6%) and aspect (0.05, 5%) were assigned relatively lower weights, suggesting a lesser degree of influence. The two factors with the lowest weights, such as curvature (0.03, 3%) and land use/land cover (0.02, 2%) showed the least direct influence on the occurrence of landslides. The analysis based on the AHP confirmed the internal consistency and reliability of the assigned weights, with a Consistency Ratio (CR) of 0.0127, which was well below the acceptable level of 0.1 [42]. According to Saaty’s AHP theory, a consistency ratio (CR) below 0.1 indicates an acceptable level of consistency in pairwise comparisons, ensuring that the assigned weights are logically sound and reliable.

The normalized weight and rank of thematic parameters highlighted the relative significance of sub-classes in landslide susceptibility analysis (

Table 4. Assigned weights of thematic parameters, normalized weights, and ranks for landslide susceptibility analysis.

Sl. No.	Thematic Parameters	Sub-Classes	Rank	Normalized Weight
1	Slope	Below 10	5	0.096
		10–20	5	0.086
		20–30	4	0.127
		30–40	2	0.275
		40–50	1	0.278
		Above 50	3	0.138
2	Geology	Meta-Vocanics	5	0.052
		Central Crystalline	1	0.311
		The Garhwal group	4	0.085
		The Martoli group	2	0.290
		The Rakcham granite or Mandi	3	0.263
3	Lineament Density	Very Low	5	0.072
		Low	3	0.228
		Moderate	1	0.243
		High	2	0.228
		Very High	4	0.228
4	Drainage Density	Very Low	5	0.093
		Low	4	0.217
		Moderate	3	0.217
		High	1	0.236
		Very High	2	0.236
5	Distance to Road	0–200	2	0.247
		200–400	1	0.269
		400–600	3	0.247
		600–800	4	0.123
		800–1000	5	0.114
6	Rainfall	157–170	5	0.103
		171–190	4	0.111
		191–210	1	0.284
		211–230	2	0.262
		231 & above	3	0.241
7	Aspect	Flat	5	0.024
		North	3	0.113
		North East	4	0.088
		East	4	0.068
		South East	1	0.191
		South	2	0.140
		South West	1	0.213
		West	5	0.047
		North West	5	0.039
		North	2	0.113
8	Curvature	Concave	2	0.416
		Flat	3	0.126
		Convex	1	0.458
9	LULC	River	4	0.105
		Vegetation	1	0.222
		Snow cover	3	0.129
		Barren Land	1	0.222
		River Sediment	5	0.077
		Settlement	5	0.060
		Agriculture	2	0.184

). Within the slope parameter, the 40–50° category received the highest normalized weight (0.278), followed closely by the 30–40° category (0.275), indicating that steep slopes were the most critical in triggering landslides. Among geological sub-classes, the central crystalline formation was assigned the highest weight (0.311), reflecting its high susceptibility to slope failures, followed by the Martoli group (0.290). In lineament density, the moderate category (0.243) received the highest weight, emphasizing that intermediate density areas were more vulnerable to landslides, likely due to their structural weaknesses. The high and very high drainage density categories received equal weights (0.236), suggesting that these areas were hotspots for material saturation and slope failures. For distance to road, the 200–400 m buffer zone was weighted the highest (0.269), showing the significant role of proximity to roads in destabilizing slopes. Rainfall sub-classes between 191–210 mm had the highest normalized weight (0.284), emphasizing the critical impact of moderate to high precipitation levels on landslide activity. Lastly, curvature and LULC showed the highest weights for convex curvature (0.458) and vegetation/barren land (0.222 each), underlining their direct influence on slope stability and erosion processes.

The landslide susceptibility map of the Uttarkashi–Gangotri road section illustrated five susceptibility zones: very low, low, moderate, high, and very high (Figure 8). High and very high susceptibility zones dominated the study area, especially along NH34 and near townships like Dharali and Sukhi. Moderate susceptibility zones were distributed across central parts of the study area, making up a significant portion of the landscape. Low and very low susceptibility zones were limited and mostly observed in stable areas near Uttarkashi and Gangotri. The clustering of landslide inventories within high and very high susceptibility zones validated the map’s accuracy in identifying vulnerable areas. These zones were spatially aligned with geological formations, slope categories, and anthropogenic factors such as road construction.

The area distribution of landslide susceptibility zones revealed that high susceptibility zones covered the largest area (374.6 km², 37.1%), followed by moderate zones (316.6 km², 31.4%) (Figure 9). The very high susceptibility zone occupied a substantial area of 252.2 km² (25.01%), emphasizing the significant vulnerability in these regions. Low susceptibility zones accounted for only 61.37 km² (6.1%), reflecting limited stable areas. Very low susceptibility zones, the most stable regions, covered just 3.7 km² (0.37%) of the area.

3.4. Validation

3.4.1. Field Validation

Field surveys were conducted along Uttarkashi–Gangotri road section to validate satellite-based landslide inventories and understand the geomorphological processes controlling slope failures. An active roadside landslide was observed with debris accumulation obstructing the road corridor, where movement is primarily along a surface marked by steep slopes and loose debris (Figure 10a). This type of failure is common in cut slopes with exposed fractured rock masses and intense weathering, especially during monsoonal periods. In contrast, the Varunavat mountain in Uttarkashi township exhibits an old, large-scale landslide scar, along with signs of minor, newly initiated slope instabilities (Figure 10b). The head scarp and runout zone suggest a translational slide mechanism influenced by topographic steepness, anthropogenic activities, and intense rainfall.

3.4.2. Validation of the Model

The AUC curve demonstrated the predictive performance of the landslide susceptibility map derived from the AHP (Figure 11). The curve represented the true positive rate versus the false positive rate, with an AUC value of 0.817, signifying strong predictive capability. The dashed line indicated a random guess as a baseline, while the shaded region reflected confidence intervals, showcasing model uncertainty at various thresholds. Validation was conducted using 105 known landslide locations overlaid on the final map, classified using the natural break (Jenks) method. The relationship between the cumulative percentage of landslides and the cumulative area percentage was represented by the AUC curve, validating the reliability of the model. AUC values close to 1 indicated a highly accurate model, and the AUC of 0.817 confirmed the consistency and robustness of the landslide susceptibility map.

The landslide inventory points showed that the very high susceptibility zone contained the majority of landslides (64 inventories), followed by the high zone with 31, and the moderate zone with 9 landslide inventories, while the low zones had 1 inventory, respectively (

Table 5. Distribution of landslide inventory points across different landslide susceptibility zones.

Landslide Susceptible Zone	No. of Landslide Inventories
Very Low	0
Low	1
Moderate	9
High	31
Very High	64

). These results validated the effective correlation between the susceptibility zones and real-world landslide locations, demonstrating the model’s accuracy.

4. Discussion

The integration of landslide inventory mapping with GIS and AHP methods in the Uttarkashi–Gangotri region revealed significant clustering of landslides along the NH34 corridor, especially near Sukhi, Dharali, and Gangori. This pattern underscored the vulnerability of infrastructure in steep terrains and tectonically active zones, consistent with previous findings [46,47]. The dominance of landslides in areas with steep slopes (40–50°), particularly those associated with Central Crystalline and Martoli Group lithologies, reaffirmed the role of geology and gradient in slope failures. Additionally, regions marked by high lineament density, especially near thrusts such as the Main Central Thrust (MCT), coincided with major landslide-prone zones, emphasizing the structural influence on landslide initiation [40]. These lineaments are primarily of tectonic origin, associated with regional fault systems and major thrust zones that characterize the highly deformed geology of the central Himalaya, as supported by earlier geological studies. Hydrological factors such as drainage density further influenced slope instability; areas with high and very high drainage networks exhibited greater erosion potential and saturation, aligning with global studies [12,48]. Aspect and curvature also modulated landslide susceptibility: southeast-facing and convex slopes were most vulnerable due to intense solar exposure and surface runoff. These geomorphic indicators, combined with rainfall exceeding 231 mm, indicated high potential for shallow landslides during monsoons, mirroring findings from previous studies [48]. Finally, land use played a critical role barren lands and regions undergoing construction were highly prone to slope failures, whereas vegetated areas offered greater stability.

The anthropogenic factors analyzed showed that distance to roads was a dominant trigger of landslides, particularly within 200–400 m buffers, corroborating earlier findings on the effects of slope cutting and blasting near transport routes [40]. The AHP-derived weights ranked slope (23%), geology (22%), and lineament density (21%) as the top contributors to landslide susceptibility, highlighting their compounded effect in geomorphologically sensitive zones. Drainage density (9%) and proximity to roads (8%) had moderate importance, illustrating how water saturation and human activities interacted with natural terrain to exacerbate instability. Rainfall (6%) and aspect (5%) emerged as secondary yet significant indicators, especially in areas receiving moderate to high rainfall where slopes were already weakened. The AHP matrix’s consistency ratio (0.0127) confirmed methodological robustness, while validation using ROC–AUC (0.817) demonstrated high predictive accuracy. Compared with studies across Nepal and northern Pakistan, these performance metrics suggested that the methodology was well-suited to Himalayan topography. The spatial susceptibility zoning showed that 93.55% of the area fell under moderate to very high susceptibility, calling for urgent hazard mitigation strategies. High and very high zones together comprised over 62% of the terrain, particularly around Dharali, Bhatwari, and Gangotri, highlighting areas in immediate need of stabilization and monitoring.

Model validation further confirmed that the majority of landslide occurrences (95%) fell within high and very high susceptibility zones, demonstrating the effectiveness of the weighted overlay technique and factor selection. Only 6.46% of the area was classified under low and very low zones, indicating minimal regions with natural resistance to landslide formation. These outcomes aligned with hazard zoning models in similar terrains such as Sikkim and Himachal Pradesh, where landslide occurrence was strongly correlated with slope angle, rock type, and infrastructure proximity [49]. The study also reinforced the importance of LULC classification—barren and built-up lands coincided with high-risk areas, supporting global research on land degradation as a precursor to mass wasting events [50]. The use of both polygonal and point-based inventories ensured a comprehensive hazard representation, allowing accurate spatial correlation with influencing parameters.

Unlike past models limited to elevation or rainfall, this multi-criteria framework integrated geomorphological, hydrological, geological, and anthropogenic factors, offering a holistic perspective. This integrative approach contributed to disaster preparedness, providing policymakers with a refined hazard assessment tool for planning interventions. Future studies could incorporate real-time rainfall and seismic data to further improve landslide forecasting. While average rainfall values were used in this study, incorporating short-term extreme rainfall events and seismic activity could significantly shift hazard classifications and should be considered in future assessments. Finally, integrating climate projections could help assess how changing precipitation patterns might modify susceptibility zones, particularly in glacier-fed regions such as Gangotri. Although AHP remains useful in data-scarce regions due to its simplicity and interpretability, modern machine learning models such as Random Forest and XGBoost have emerged as more robust approaches for landslide susceptibility and predictability, providing higher accuracy, less subjectivity, and better generalization when sufficient quality data are available.

5. Conclusions

This study provided an adaptable framework for landslide susceptibility mapping with a prediction accuracy of 0.817. The AHP based geospatial analysis identified slope (23%), geology (22%), and lineament density (21%) as the main factors causing instability. Other important factors included drainage, roads, rainfall, aspect, curvature, and land cover. The landslide susceptibility map showed that over 93% of the region lay in moderate to very high-risk zones. Areas with high and very high risk accounted for more than 62% of the terrain and included 95% of known landslides. Field validation confirmed strong agreement between predicted and observed events. This framework helped to identify dangerous slopes, such as southeast-facing, convex terrains, and barren lands. It also showed that forested and flat areas were more stable. The model supported better planning for roads, infrastructure, and slope safety. It was especially useful in areas with limited data but high hazard exposure. As climate change increased rainfall and construction pressures, this model linked scientific analysis with practical planning to reduce landslide risks and promote safer development in vulnerable regions.