GIS-based multi-criteria landslide susceptibility assessment in the Northwestern Himalaya, Jammu and Kashmir, India

Abstract

This study presents a landslide susceptibility assessment using multiple parameters by integrating remote sensing, GIS, and field observations in the Jammu and Kashmir region. Eight contributing variables were selected for landslide susceptibility analysis: Land Use and Land Cover (LULC), proximity to roads, streams, slope gradient, slope orientation (aspect), geology, geomorphology, and elevation. In addition, an extensive landslide inventory consisting of 669 landslide events was developed using the Field Landslide Inventory Mapping (FLIM) application, LISS-IV satellite data, and field observations, covering an area of 42,950.43 km². Landslide susceptibility mapping (LSM) was carried out using the Analytical Hierarchy Process (AHP) approach and validated with MaxEnt software and field-generated landslide data. The resulting landslide susceptibility map was classified into five categories: very high, high, medium, low, and very low susceptibility zones. Based on the AHP approach, these zones cover 3.65% (1,569.0483 km²), 24.43% (10,492.8912 km²), 51.56% (22,147.2369 km²), 18.81% (8,079.2199 km²), and 1.54% (662.0301 km²) of the study area, respectively. The weighted overlay and MaxEnt models proved effective for landslide vulnerability mapping, with MaxEnt achieving AUC values of 0.82 for training data and 0.807 for testing data, indicating good predictive performance. Field validation further showed that 87% of landslides occurred within high and very high susceptibility zones. The Jackknife test identified road proximity, slope, and stream proximity as the most significant independent variables influencing landslide occurrence. The generated landslide susceptibility map provides important insights for reducing landslide risk and serves as a valuable tool for infrastructure planning, community development, and disaster management in the region.

Introduction

Landslides are among the most destructive natural hazards in mountainous regions, causing significant damage to infrastructure, human life, and the natural environment. Globally, landslides frequently occur in tectonically active and high-relief regions where geological, geomorphological, hydrological, and anthropogenic factors interact to destabilize slopes. The Himalayan region is particularly vulnerable due to steep terrain, fragile lithology, intense rainfall, active tectonics, and rapid infrastructure development, making landslide susceptibility assessment a critical component of disaster risk reduction and sustainable land-use planning^1–3. The Union Territory of Jammu and Kashmir, located in the northwestern Himalaya, is highly prone to landslides due to its complex geological structure, rugged topography, and extreme climatic variability. The region experiences frequent slope failures triggered by monsoonal rainfall, river erosion, seismic activity, and anthropogenic interventions such as road construction and deforestation. Lithological heterogeneity, including sedimentary, metamorphic, and unconsolidated formations, combined with steep slopes and intense drainage dissection, significantly increases slope instability across the region. In addition, the presence of active faults and thrust systems associated with the Himalayan tectonic framework contributes to geomorphological instability and slope weakening, making landslide susceptibility mapping essential for hazard mitigation, infrastructure planning, and disaster risk management^4–6. Landslide occurrence in Jammu and Kashmir is primarily controlled by rainfall-induced slope instability, geomorphological processes, and anthropogenic disturbances. Rainfall infiltration increases pore water pressure and reduces shear strength of slope materials, particularly in weathered and fractured lithological units, leading to slope failure along structural discontinuities and weak zones. River erosion at slope toes and road cutting further destabilize slopes by removing lateral support, while seismic activity acts as a secondary triggering mechanism by weakening already unstable terrain. The general landslide mechanism involves rainfall infiltration, slope saturation, reduction of soil cohesion, and eventual mass movement along lithological or structural planes. A schematic representation of this landslide mechanism, including rainfall infiltration, slope instability, drainage erosion, and road cutting, is presented in Fig. 1 to provide a conceptual understanding of slope failure processes in the study area, following recent conceptual and process-based landslide mechanism studies^7,8. Over the past two decades, landslide susceptibility mapping has gained significant attention due to advancements in Geographic Information Systems (GIS), remote sensing, and spatial modelling techniques. Various approaches have been developed, including statistical models, machine learning algorithms, and multi-criteria decision-making techniques. Among these, the Analytic Hierarchy Process (AHP) has been widely used because of its ability to integrate expert knowledge and multiple conditioning factors in a structured decision-making framework^9,10. AHP is particularly useful in mountainous regions where detailed landslide inventory or large training datasets may be limited. However, AHP relies on expert judgment for weight assignment, which can introduce subjectivity and affect prediction reliability. To overcome this limitation, hybrid approaches combining multi-criteria decision-making with probabilistic or machine learning models have been increasingly adopted. The Maximum Entropy (MaxEnt) model is a presence-only probabilistic approach that estimates landslide occurrence probability based on environmental constraints and has shown strong predictive performance in susceptibility studies^11–13. MaxEnt effectively captures nonlinear relationships between conditioning factors and landslide occurrence and provides variable importance through jackknife testing and response curves. Therefore, integrating AHP and MaxEnt offers a balanced approach by combining expert-based weighting with data-driven probabilistic modelling, thereby improving the robustness and reliability of landslide susceptibility assessment^14,15.

Fig. 1.

Schematic representation of rainfall-induced landslides¹⁸.

Several studies have applied GIS-based AHP, MaxEnt, and machine learning models for landslide susceptibility mapping in different parts of the Himalaya and other mountainous regions. However, most of these studies are limited to district-level or watershed-scale analyses, and comprehensive regional-scale assessments covering the entire Jammu and Kashmir Union Territory remain limited^3,9,15. In addition, previous research often applies multi-criteria decision-making methods or machine learning models independently, with limited integration of geomorphological, hydrological, and anthropogenic factors within a unified modelling framework^13,14,16. Rainfall is widely recognized as a major triggering factor for landslides in the Himalayan region, as intense and prolonged precipitation increases pore-water pressure, reduces shear strength, and destabilizes slopes, particularly in tectonically active mountainous terrain^1,17. In the present study, rainfall has been incorporated as an important conditioning factor using spatial rainfall data derived from available meteorological and satellite-based datasets, allowing a more comprehensive representation of landslide-triggering conditions across the Jammu and Kashmir region. The inclusion of rainfall along with slope, lithology, geomorphology, drainage proximity, road proximity, and land use/land cover improves the reliability of the susceptibility model and better captures the interaction between climatic and geo-environmental factors controlling landslide occurrence¹⁴. Despite this improvement, future research can further enhance landslide prediction by integrating high-resolution time-series rainfall intensity, soil moisture, and seismic triggering parameters to develop dynamic landslide forecasting and early warning systems for the Himalayan region⁸. Recent studies have highlighted the importance of incorporating spatiotemporal rainfall variability, antecedent moisture conditions, and deep learning approaches for improving landslide prediction and early warning systems^18–21;. These advanced models integrate rainfall randomness, soil moisture dynamics, and hydrological processes to enhance prediction accuracy and reduce uncertainty. However, such data-driven approaches require continuous high-resolution rainfall and geotechnical datasets, which are often limited in large mountainous regions like Jammu and Kashmir. Figure 1 visualizes a schematic view of the rainfall-induced landslides. Moreover, many machine learning studies focus primarily on overall accuracy and AUC while paying limited attention to operational performance metrics such as false negatives, which are critical for hazard mitigation and early warning applications. Therefore, regional-scale hybrid modelling approaches that combine expert knowledge with probabilistic models remain highly relevant for large and data-scarce Himalayan regions.

Despite the growing number of landslide susceptibility studies in the Himalayan region, comprehensive regional-scale assessments integrating multi-criteria decision-making and probabilistic modelling approaches remain limited for the Jammu and Kashmir Union Territory. The novelty of this study lies in the integrated application of the AHP–MaxEnt hybrid modelling framework at a regional scale, combining expert-based weighting with data-driven probabilistic modelling to reduce subjectivity and improve prediction reliability. The study incorporates multiple geo-environmental and anthropogenic conditioning factors and provides a mechanism-based interpretation of landslide processes in the northwestern Himalaya. Furthermore, the study contributes to the scientific community by developing a regional-scale susceptibility framework that can support infrastructure planning, hazard mitigation, and sustainable development in mountainous terrain. Therefore, the main objectives of this study are: (1) to identify and prepare landslide conditioning factors including slope, lithology, geomorphology, drainage proximity, road proximity, and land use/land cover; (2) to develop landslide susceptibility maps using the AHP method; (3) to apply the MaxEnt model for probabilistic landslide prediction and variable importance analysis; (4) to integrate AHP and MaxEnt results to generate a reliable landslide susceptibility zonation map for the Jammu and Kashmir region. The results of this study are expected to contribute to regional-scale landslide hazard assessment and provide a methodological framework for integrating multi-criteria decision-making and probabilistic modelling approaches in mountainous regions. The generated susceptibility maps can support policymakers, planners, and disaster management authorities in identifying high-risk zones and implementing effective mitigation strategies for sustainable development in the Himalayan region.

Study area

The study area covers the Jammu and Kashmir region of the India ranging from a broad spectrum of landscapes, mountain ranges and plateaus to deep gorges and canyons (Fig. 2). Southern slopes in the region tend to be barren and sparsely vegetated, while northern slopes are more densely vegetated. These areas are prone to frequent landslides due to factors such as steep slopes, heavy rainfall in the monsoon season, seismic activity, and anthropogenic activities. The area is characterized by rugged terrain, mountainous regions, and a variety of geomorphological features covering an approximate area of 59,000 square kilometres, which make them highly susceptible to landslides. The Indus River, a lifeline of the region, flows through Ladakh, shaping the hydrology and influencing local geology. The Chenab River, renowned as one of the mightiest rivers in the world, flows through districts including Kishtwar, Doda, Ramban, Reasi, and the Akhnoor area of the Jammu district, and its tributaries contribute to the geomorphic diversity of the region and pose flood risks during monsoons. The study area focuses on analyzing the landslide susceptibility, frequency, and impact in these high-risk zones. The region faces a high risk of landslides due to natural and anthropogenic factors. Heavy rainfall during the monsoon season often triggers debris flows, rockfalls, and slope failures, as the increased pore water pressure destabilizes slopes. Earthquakes, common in this tectonically active zone, also play a major role in triggering widespread landslides. Human activities, including road construction, deforestation, and urbanization, further aggravate slope instability. Major highways, such as NH-44, NH-244, and NH-1 A are particularly vulnerable due to traffic-induced vibrations and land-use changes. These combined factors highlight the need for detailed geotechnical studies and sustainable land management practices to mitigate landslide hazards.

Fig. 2.

Location map of the study area (a) India boundary showing J&T UT (b) Detailed location map of J&K UT (c) Aster Digital Elevation Model (DEM), 30-m resolutions of the study area.

Geology and geomorphology of the study area

The geology of Jammu and Kashmir (J&K) reflects a complex history of tectonic activity, volcanism, and sedimentary deposition associated with the evolution of the Himalayan orogeny²². Located in the northwestern part of the Indian Himalayan Region, the area comprises a series of mountain ranges, including the Shiwalik Hills in the south, followed by the Pir Panjal Range, the Greater Himalaya, the Zanskar Range, and the Karakoram in the north. The region contains rock formations ranging from Archean crystalline basement rocks to recent alluvial deposits, representing a wide spectrum of lithological and geomorphological units that significantly influence landslide occurrence. Broadly, the geological formations can be categorized into consolidated rocks (limestone, granite, quartzite, slates, and Panjal traps), semi-consolidated sedimentary rocks (sandstone, siltstone, and claystone), and unconsolidated Quaternary deposits consisting of clay, sand, gravel, pebbles, and boulders. The Shiwalik Group, widely exposed in the Jammu region along the Suruin–Mastgarh anticline, consists of Lower, Middle, and Upper Shiwalik formations characterized by folded sedimentary sequences and variable dip angles, making slopes structurally unstable and prone to landslides. Paleozoic and Precambrian formations, including Dogra slates, Salkhal crystalline rocks, Muth quartzites, and Syringothris limestone, are exposed across the Pir Panjal and Greater Himalayan ranges, reflecting intense tectonic deformation and structural discontinuities such as thrusts, folds, and faults. In the Kashmir Valley, Quaternary Karewa deposits represent lacustrine and fluvial sediments derived from uplifted mountain ranges, indicating past climatic and tectonic activity that has shaped the present geomorphology. The region also exhibits glacial deposition at lower elevations and widespread river incision, which, together with steep slopes, weak lithology, and active structural features, contributes significantly to slope instability and landslide susceptibility. Therefore, the geological and geomorphological framework of J&K region plays a crucial role in controlling landslide occurrence across this large Himalayan terrain.

Material and methodology

Data preparation

The data preparation for landslide susceptibility mapping divided into two groups as: (1) landslide events data (2) causative factors of landslide. The methodology framework used in the study have been described in the Fig. 3.

Fig. 3.

Methodology flowchart of the study.

Landslide events data

A comprehensive, multi-temporal landslide events data was developed for this study by integrating satellite-based image interpretation with field-based observations. Landslide locations were delineated through detailed field surveys, particularly along roads and stream channels, and field photographs were taken using Field Landslide Inventory Mapping (FLIM) Application. This app is designed by NRSC, ISRO to collect landslide data in the field. This event data comprises landslides identified both through remote sensing analysis and in-situ verification, providing a robust spatial dataset for assessing the distribution of landslides and their relationship with contributing factors. Multi-date satellite imagery was utilized to map landslide occurrences and characterize their spatial attributes over time. Field investigations were focusing on slope failure zones along transportation and drainage networks. These field observations also served as ground truth data to validate the landslide features interpreted from satellite imagery. Landslide identification on imagery was based on expert knowledge, assessing factors such as spectral contrast, terrain configuration, geometry, tone, texture, and observable changes in slope gradient. In particular, small to medium-scale rotational landslides and shallow failures were documented along exposed or unprotected slopes, frequently associated with rockfalls and slumps. To reduce uncertainties like size or magnitude of landslides, the landslide as a polygon was used in this study.

Data set used

The data set used in the study are Resourcesat-2 LISS IV with resolution 5.8 m, ASTER DEM with resolution 30 m. Another dataset includes SISDP-LULC10K (2018-19), geomorphological map and lithological map of GSI, WRIS-India, J&K administrative map, road map & district map.

Landslide causative factors

To understand the mechanisms underlying slope failures and identify the primary causes of landslides, a comprehensive set of terrain, environmental, and anthropogenic variables was assessed. Previous research in the North-Western Himalaya indicates that seismotectonic activity plays a critical role in landslide initiation, particularly due to tectonic stress and earthquake occurrences. Additionally, intense and recurrent rainfall events significantly elevate soil moisture content and slope saturation, increasing shear stress and contributing to slope instability. Human-induced land disturbances, including road construction and land use changes, further exacerbate landslide risks. Moreover, the continuous buildup of compressive forces in the Himalayan belt is known to induce seismic events, which in turn trigger landslide activity. Field investigations and a comprehensive set of geospatial data was collected and used to generate Eight georeferenced thematic layers. These includes Land Use and Land Cover (LULC) 2024, proximity to roads, streams, slope gradient, slope orientation (aspect), geology, geomorphology and elevation. A structured methodology was applied to analyse and synthesize these parameters for susceptibility mapping. Each variable was meticulously selected, standardized, and reclassified according to its relevance and influence on slope stability. These layers were then integrated within a GIS-based multi-criteria decision framework. The assigned weights and scores for each factor, based on their contribution to landslide occurrence are explained in Table 2. This systematic approach facilitated the identification of zones with varying degrees of landslide susceptibility across the study area.

Table 2.

Spatial distribution of landslide susceptibility zones.

Risk zone	Area (Km2)	Area (%)
Very High Risk	1,569.0483	3.65
High Risk	10,492.8912	24.43
Medium Risk	22,147.2369	51.56
Low Risk	8,079.2199	18.81
Very Low Risk	662.0301	1.54

Topographic factors

The slope, elevation and aspect, all derived from the 30 m Aster DEM using ArcGIS 10.8.1.

Aspect

Aspect refers to the compass direction that a slope faces, typically measured in degrees from 0° to 360°, and is commonly derived from Digital Elevation Models (DEMs). In landslide susceptibility assessments, aspect is a critical parameter for identifying terrain orientations that may influence slope stability. Slope orientation affects microclimatic conditions, such as solar radiation, soil moisture, and vegetation cover, which in turn impact the likelihood of landslides. South-facing slopes are generally more susceptible due to greater exposure to solar radiation, which can enhance soil moisture through snowmelt and support denser vegetation cover factors that may contribute to slope instability. In contrast, north-facing slopes typically receive less solar radiation, resulting in reduced vegetation and relatively lower landslide susceptibility. The aspect map (Fig. 4(a) of the study area has been divided into five distinct categories based on direction: North, North-East, East, South-East to South-West and West to North-West²³.

Fig. 4.

The landslide conditioning parameters used in the study (a) Aspect (b) Slope (c) Elevation (d) Geology (e) Geomorphology (f) Stream proximity (g) LULC (2024) (h) Road proximity.

Slope

Slope represents the degree of inclination or steepness of the Earth’s surface at a given location, commonly expressed in degrees. It is a fundamental factor in landslide susceptibility assessment, as it directly affects terrain stability²³. Steep slopes are particularly prone to landslides due to their limited capacity to resist gravitational forces and increased vulnerability to water infiltration. The slope angle also influences critical soil properties such as cohesion and internal friction, with higher angles typically associated with reduced shear strength and overall stability. The slope map of the study area has been divided into five distinct categories: 0–5°, 5°−15°, 15°−30°, 30°−45° and > 45° (Fig. 4(b). Slopes categorized within the range of > 45° are identified as highly susceptible to landslides due to their increased steepness relative to other gradient classes.

Elevation

Elevation denotes the vertical distance of a location above a reference datum, typically obtained from Digital Elevation Models (DEMs). Elevation data is instrumental in identifying areas susceptible to landslides, as higher elevations often exhibit greater vulnerability due to a combination of geomorphological and climatic factors. The elevation map of the study area has been divided into five distinct categories: 232–1560 m, 1561–2890 m, 2891–4220 m, 4221–5550 m, 5551–6885 m (Fig. 4(c). Elevations categorized as high and very high are assigned greater weight due to their increased landslide susceptibility, while low and very low elevations are given lower weight in the susceptibility assessment.

Geological factors

The lithology and lineaments, reflecting the complex tectonic setting of the region, marked by the juxtaposition of Central and Lesser Himalayan rock units across the MCT.

Geology

Geological data offers critical insights into the lithological and structural composition of the study area, encompassing rock types, soil classifications, and other geological features. Such data are typically acquired from GSI (https://www.gsi.gov.in/) and are classified according to the distinct geological characteristics of the terrain. In the present study, geological information was obtained from the Geological Survey of India, and the study area was categorized into four types: igneous, metamorphic, sedimentary, unconsolidated sediments (Fig. 4(d).

Geomorphology

Geomorphological data pertains to the physical characteristics and surface features of the terrain, including landforms such as ridges, valleys, escarpments, and slopes. In the present study, geomorphological information was obtained from the Bhukosh portal of the Geological Survey of India (https://bhukosh.gsi.gov.in/) and the study area is categories into nine types: Highly dissected structural hills and valleys, moderately dissected structural hills and valleys, low dissected structural hills and valleys, snow cover, active flood plain, lacustrine marsh, piedmont alluvial plains, valley fill (alluvial plains), water bodies (Fig. 4(e). Geomorphological data also aid in identifying potential landslide source areas and movement pathways. Regions characterized by steep slopes or highly sinuous drainage channels are typically associated with landslide initiation, whereas areas with gentle slopes or broad valleys often serve as potential pathways for landslide movement.

Hydrological variables

Steam proximity

Stream proximity is a key factor in understanding the influence of drainage features on slope stability and landslide susceptibility. Drainage networks can act as conduits for surface runoff, leading to increased soil saturation and a reduction in soil cohesion, which in turn destabilizes slopes. Furthermore, these hydrological systems alter the local infiltration dynamics and contribute to enhanced soil erosion, both of which are critical processes in landslide initiation. Therefore, assessing the proximity to streams is essential for accurately evaluating landslide risk. This parameter informs land use planning, infrastructure development, and the implementation of effective mitigation strategies in areas prone to slope failure. To facilitate the analysis of stream proximity, a stream buffer was generated, segmenting the study area into five distinct categories based on distance from drainage features 0–200 m, 200–500 m, 500–1000 m, 1000–1200 m and > 1200 m (Fig. 4(f).

Land use/land cover

Land Use and Land Cover (LULC) play a critical role in assessing landslide susceptibility and evaluating its potential consequences. In this study, LULC mapping was conducted using the base map functionality within ArcGIS software, with high-resolution satellite imagery from IRS LISS IV data & Google Earth serving as the primary data source. This imagery forms the foundational dataset for developing detailed and accurate land cover classifications, facilitating the visualization and analysis of diverse land use types. The relevance of LULC in landslide susceptibility analysis lies in its capacity to identify and evaluate areas at risk²⁴. Specific land use changes, such as urban development, deforestation, and agricultural expansion, can significantly alter the natural terrain and increase the likelihood of landslides. The LULC (2024) map of the study area was generated using the digitization method and categorized into six types: Barren land, scrub land, agriculture, built-up, forest, snow and waterbody. Among the various LULC categories, barren land exhibit notably elevated susceptibility to landslides (Fig. 4(g).

Anthropogenic factor

Distance to roads was analyzed using road networks digitized from Google Earth and imported into ArcGIS 10.8.1 for buffer analysis.

Road proximity

Road proximity, which is critical in landslide susceptibility because roads may contribute to or exacerbate hazards, quantifies how close roads are to a region in terms of distance. In the study area, there are several roads that are under upgradation, like NH-44, NH-244 (Batote-Srinagar Highway), NH-1D (Srinagar-Leh Highway), NH-3 (Leh-Manali Highway), NH-144 A (Jammu-Poonch Highway), NH-144 (Domel to Katra), and Mughal Rd (Poonch to Shopian). The upgradation of all these roads leads to the disruption of natural slopes, and therefore landslide events have increased in the study area. Road embankments can change the water flows direction which can lead to erosion and runoff that causes landslides. A road buffer has been established, dividing the study area into five distinct categories 0–200 m, 200–500 m, 500–1000 m, 1000–3000 m and > 3000 m (Fig. 4 (h).

Methods

The methods for landslide susceptibility mapping are divided into two groups as (1) Weighted overlays analysis (2) Field work.

Weighted overlay analysis

Using the relative importance or weight of each input layer, the Weighted Overlay (WO) analysis is a geographic technique used in GIS to merge several raster layers into a single composite layer. This technique creates a final output layer by combining the input layers using a mathematical formula after giving each one a weight according to its importance or relevance to the analysis. Each thematic layer is given a numerical weighting factor that indicates its relative position in relation to other layers; this weighted approach is called weighted overlay, and it is based on analysis. After the reclassified layers are added to the weighted overlay model, they are reclassified into five classes according to their attributes, importance, and alignment with the goals: According to Table 2, there are five different risk categories: Class 5 [very high (VH)], Class 4 [high (H)], Class 3 [Medium (M)], Class 2 [low (L)], and Class 1 [very low (VH)]. The polygon to raster tool in QGIS software is used to reclassify the LULC, geomorphology, geology, the slope, aspect, road, and stream. The weighted overlay analysis incorporates and nurtures each of the theme factors. Each thematic layer has been assigned the proper weight as a weighting factor, and additional weights have been assigned to each class of all the layers as weights for class based on the impact of the thematic layer and class on the landslide susceptibility study.

Weightage assignment

The weight assigned to each conditioning factor was determined based on its relative contribution to landslide susceptibility using the Analytical Hierarchy Process (AHP) supported by expert judgment, geomorphological understanding of the Himalayan terrain, and previous landslide susceptibility studies. Slope, with a weight of 25%, was assigned the highest importance due to its critical role in controlling gravitational forces and slope instability in mountainous regions. Road proximity was also weighted at 20%, as road construction significantly disturbs slope stability through cutting, excavation, and modification of natural drainage patterns. Land use/land cover (LULC), weighted at 15%, reflects the influence of vegetation cover and human activities on soil stability and erosion processes. Aspect (15%) was assigned moderate importance because of its indirect control on solar radiation, soil moisture, and weathering intensity. Elevation (10%) was considered important for representing topographic variation and climatic influence on slope processes, while geology (20%) and geomorphology (15%) were included to capture lithological strength and landform characteristics controlling slope failure. Stream proximity (5%) was assigned a lower weight as it primarily influences localized erosion and slope undercutting near drainage channels. Therefore, the term AHP in this study refers to an expert-informed weighting framework rather than a fully implemented mathematical AHP model. The final distribution of weights for all conditioning factors is presented in Table 1, ensuring a balanced representation of both natural and anthropogenic factors in landslide susceptibility modelling.

Table 1.

Weights assigned to various parameters and scores given to their respective classes in the weighted overlay analysis for assessing landslide susceptibility.

S.No.	Factors/Theme	Weightage factor	Classes	Weight for class
1	Slope (°)	25	0°−5° 5°−15° 15°−30° 30°−45° > 45°	1 2 3 4 5	VL L M H VH
2	Road Proximity (m)	20	0–200 m 200–500 m 500–1000 m 1000–3000 m > 3000 m	5 4 3 2 1	VH H M L VL
3	Aspect	15	South-East to South-West West to North-West East North-East North	5 4 3 2 1	VH H M L VL
4	Geomorphology	15	Highly dissected structural hills and valleys Moderately dissected structural hills and valleys Low dissected structural hills and valleys Piedmont alluvial plains Valley fill (alluvial plains) Snow cover Active flood plain Lacustrine marsh Water bodies	5 4 3 2 2 1 1 1 1	VH H M L L VL VL VL VL
5	Geology	20	Sedimentary Unconsolidated sediments Metamorphic Igneous	5 4 3 2	VH H M L
6	Elevation (m)	10	232–1560 m 1561–2890 m 2891–4220 m 4221–5550 m 5551–6885 m	1 2 3 4 5	VL L M H VH
7	Stream proximity (m)	5	0–200 m 200–500 m 500–1000 m 1000–1200 m >1200 m	5 4 3 2 1	VH H M L VL
8	LULC	15	Barren land Scrub land Agriculture Built-up Forest, Snow Waterbody	5 4 3 3 2 1 1	VH H M M L VL VL

Field work

The field survey covered all districts of the Jammu and Kashmir Union Territory (UT) however, certain areas remained inaccessible due to a lack of connectivity (Fig. 13). The primary types of landslides observed during the survey include debris slides, debris flows, rockslides, and rockfalls. In Ramban district, a significant landslide occurred near Pernote village, situated above the Chenab River along the Ramban-Gool Road. This landslide disrupted road communication with Reasi district and caused substantial damage to residential areas, roads, and houses. In Doda district, the Thathri Nai Basti landslide destroyed multiple houses, infrastructure, and roads. In Udhampur district, numerous landslides of varying sizes were observed along NH-44, particularly between Udhampur and Banihal. These landslides originated from the slopes above the road, resulting in damage to buildings and the disruption of road connectivity between Jammu and Srinagar. In Kishtwar district, landslides occurred at multiple locations along NH-244 and the Kishtwar–Keylong Road, with one major landslide blocking the route connecting Kishtwar to Keylong in Himachal Pradesh. In Kathua district, several landslides were reported along the Basoli–Bhaderwah Road, with a significant landslide near Sukrala Mata blocking the river. In Reasi district, landslides occurred along the Chenab River, National Highway 144, and the Mohar–Rajouri Road.

In Poonch district, multiple landslides along the Mughal Road caused road blockages and disrupted connectivity with Shopian district. In the Baramulla district of the Kashmir Valley, most landslides were recorded along the Uri–Kaman Post Highway, as well as along the Ahangerpora Road and NH-701. In Anantnag district, landslides were observed along National Highway 244, particularly between Daksum and Sinthan Top. In Ganderbal district, both small and large landslides occurred between Sonamarg and Zojila Pass Along National Highway 01. The majority of landslides across J&K occur along road corridors, with some located near water bodies. These landslides are primarily attributed to rapid urbanization, road construction, and tunnelling activities along oblique and dip slopes. Additionally, the region’s heavy rainfall and numerous geological fracture zones allow rainwater to infiltrate rock fissures, which acts as a lubricating and driving force for landslides. The smallest recorded landslide, covering an area of 498.467 square meters, occurred in Ramban district (Toposheet No. 43O/7) at coordinates 33.3537° N latitude and 75.2937° E longitude. This landslide was rainfall-induced and occurred in terrain predominantly composed of phyllites, schists, and slates. Conversely, the largest landslide, spanning 23,448.5 square meters, was recorded in Kathua district (Toposheet No. 43P/10) at coordinates 32.6627° N latitude and 75.5883° E longitude. This landslide, also triggered by heavy rainfall, occurred in an area characterized by alternating beds of sandstone and conglomerate within heavily dissected hills and valleys.

Maximum entropy (MaxEnt) modelling

The Maximum Entropy (MaxEnt) model is a machine learning approach that estimates the probability distribution of landslide occurrence based on the principle of maximum entropy using presence-only data and environmental conditioning factors. The model determines the most uniform probability distribution subject to environmental constraints derived from landslide inventory data and predictor variables^12,25. The MaxEnt algorithm calculates the probability distribution by maximizing entropy while satisfying the expected value constraints of environmental variables (Eq. 1).

1

where represents the probability distribution of landslide occurrence and is the entropy of the system. The model estimates the probability of landslide occurrence as (Eq. 2):

2

where are environmental variables, are feature weights, and Z is a normalization constant. In this study, landslide inventory data and eight conditioning factors (slope, aspect, elevation, geology, geomorphology, land use/land cover, road proximity, and stream proximity) were used as input variables. The dataset was divided into training and testing samples to evaluate model performance, and the Receiver Operating Characteristic (ROC) curve and Area Under Curve (AUC) were used for validation. MaxEnt was applied to assess the predictive capability of landslide susceptibility zones and validate the weighted overlay model results, ensuring reliability and robustness of the susceptibility assessment^11,26.

Results

Landslide events

To implement the Weighted overlay (WO) analysis, a total of 669 landslide polygon was delineated using high-resolution satellite imagery from BHUVAN and Google Earth spanning the period between April 2023-February 2024. The resulting multi-temporal landslide inventory map provides a spatial overview of landslide distribution across the study area shown in Fig. 5.

Fig. 5.

Geotagged points of the landslide using LISS-IV satellite imagery.

Landslide susceptibility map

The landslide susceptibility map for the study area was generated using the WO model. The final susceptibility zonation map is classified into five susceptibility categories using the natural breaks classification method in ArcGIS (Fig. 6). According to the results (Table 2), approximately 3.65%, 24.43% of the 51.56%, 18.81%, and 1.54% of the total area falls within the very high, high, medium, low, and very low susceptibility zones respectively.

Fig. 6.

Landslide susceptibility map of the study area.

Validation of landslide susceptibility map

Validation by MaxEnt software

This presents the analysis and evaluation of the graphs and tables generated by the MaxEnt software. As demonstrated in Fig. 7, the omission rate for the test samples (green line) aligns closely with the predicted omission rate (black line), indicating a reliable model performance. Similarly, the training samples (blue line) demonstrate a strong agreement with the predicted omission rate, further validating the MaxEnt model’s effectiveness for landslide susceptibility.

Fig. 7.

Omission and predicted area for landslide in the study area.

assessment. This indicates a strong correspondence between the chosen cumulative thresholds and the predicted values. In the Fig. 8 the Receiver Operating Characteristic (ROC) curve in which the area under the curve (AUC) quantifies the model’s performance. The red line represents the training data, indicating the model’s fit to the training dataset, while the blue line corresponds to the testing data, reflecting the model’s predictive capability. The landslide inventory dataset was divided into training and testing samples, where 70% of the landslide points were used for model training and 30% were used for testing and validation of the MaxEnt model. The model performance was evaluated using the ROC curve and AUC values derived from both training and testing datasets. The proximity of the blue line to the upper-left corner of the graph indicates the model’s effectiveness in predicting occurrences within the test dataset, serving as a reliable measure of its predictive accuracy.

Fig. 8.

Sensitivity Vs. 1- Specificity for landslide in the study area.

Table 3 presents the threshold-dependent performance metrics generated by the MaxEnt model, including cumulative values, minimum training presence, 10-percentile training presence, equal sensitivity–specificity, and maximum sensitivity–specificity thresholds. These parameters are used to evaluate the predictive reliability of the MaxEnt model under different threshold conditions. The cumulative threshold value (10.000) represents the fixed cumulative prediction used to assess omission and commission errors, while the minimum training presence threshold ensures that all training samples are included in the model prediction. The 10-percentile training presence threshold allows a small omission of training samples to reduce overfitting and improve generalization. The equal training and test sensitivity–specificity thresholds provide a balanced trade-off between omission and commission errors, indicating optimal model performance. Similarly, the maximum sensitivity plus specificity threshold highlights the most accurate predictive configuration of the model. The balance training omission and entropy-based threshold further evaluates the probability distribution of landslide occurrence and model uncertainty. The extremely low p-values reported in the table indicate strong statistical significance and high predictive reliability of the MaxEnt model. Overall, Table 3 demonstrates that the MaxEnt model provides robust threshold-based validation and confirms the effectiveness of landslide susceptibility prediction in the study area.

Table 3.

Showing common thresholds and corresponding omission rates.

Cumulative threshold	Cloglog threshold	Description	Fractional predicted area	Training omission rate	Test omission rate	P-value
1.000	0.045	Fixed cumulative value 1	0.694	0.000	0.005	1.155E-19
5.000	0.144	Fixed cumulative value 5	0.474	0.032	0.042	4.695E-41
10.000	0.269	Fixed cumulative value 10	0.367	0.086	0.089	8.924E-55
1.436	0.055	Minimum training presence	0.653	0.000	0.011	9.954E-23
11.132	0.293	10 percentile training presence	0.351	0.099	0.105	7.679E-56
24.704	0.650	Equal training sensitivity and specificity	0.252	0.252	0.326	3.745E-41
13.218	0.370	Maximum training sensitivity plus specificity	0.326	0.113	0.137	1.872E-56
21.940	0.608	Equal test sensitivity and specificity	0.265	0.218	0.263	1.339E-49
12.374	0.337	Maximum test sensitivity plus specificity	0.335	0.113	0.116	3.832E-58
2.996	0.091	Balance training omission, predicted area and threshold value	0.551	0.009	0.021	1.034E-32
4.977	0.144	Equate entropy of threshold and original distributions	0.474	0.032	0.042	6.148E-41

MaxEnt model results

The study region’s landslide susceptibility map was created utilizing the MaxEnt model (Fig. 9) with presence-only landslide inventory data, as well as a number of conditioning factors (such as slope, elevation, land use/land cover, lithology etc.). In Fig. 9 Warmer colors show areas with better predicted conditions. White dots show the presence locations used for training, while violet dots show test locations. The resulting probability raster shows the region’s spatial distribution of landslide susceptibility on a continuous scale between 0.52 and 0.85. According to the Fig. 9, road corridors, riverbanks, and steep slope zones especially in the central and southern regions of the region, are the main locations of the very high and high susceptibility zones. These regions exhibit strong correlations with the distribution of past landslide events (shown by purple and white square points), indicating that the model performed effectively. On the other hand, the north-eastern and north-western highland regions, which are distinguished by consistent lithology, thick forest cover, and minimal interference from humans, are home to the majority of very low susceptibility zones.

Fig. 9.

The outcome of MaxEnt model showing the landslide probability in the study area.

Analysis of variable contributions

Table 4 gives estimates of relative contributions of the environmental variables to the MaxEnt model. To determine the first estimate, in each iteration of the training algorithm, the increase in regularized gain is added to the contribution of the corresponding variable or subtracted from it, if the change to the absolute value of lambda is negative. For the second estimate, for each environmental variable in turn, the values of that variable on training presence and background data are randomly permuted. The model is reevaluated on the permuted data, and the resulting drop in training AUC is shown in the Table 4, normalized to percentages. The parameters such road proximity and slope contribute more in the MaxEnt model prediction. As with the variable jackknife, variable contributions should be interpreted with caution when the predictor variables are correlated.

Table 4.

Analysis of variable contributions to the MaxEnt model.

Variable	Percent contribution	Permutation importance
Road proximity	71.3	65.8
Slope	16.3	18.1
Stream proximity	8.6	7.6
Geomorphology	2.5	1.4
Elevation	1	6.3
Aspect	0.2	0.7
LULC	0.1	0
Geology	0	0

MaxEnt offers the jackknife test as a robust method for evaluating the importance of environmental variables in the model. The training data represents the portion of the dataset used to develop and calibrate the model (Fig. 10). The test reveals that road proximity contributes the most to the model when used independently, followed by Stream, elevation and Slope, as highlighted by the dark blue bars. These variables provide significant standalone predictive power. Conversely, variables such as geomorphology, geology, aspect, and LULC offer minimal contributions individually. The light blue bars indicate the decrease in model gain when a variable is omitted, showcasing the unique information that each variable contributes. Road proximity, elevation, slope and Stream exhibit the most significant reduction in gain when excluded, emphasizing their critical role in predicting landslide susceptibility. This highlights that while other variables provide complementary information, road proximity, elevation, slope and stream are indispensable for achieving a reliable model.

Fig. 10.

Training data set used in Jackknife test for MaxEnt model validation.

A jackknife test was performed in the MaxEnt model to evaluate the relative importance of conditioning factors using the test dataset (30% of the total samples). The landslide inventory dataset (n = 669) was divided into training (70%, n = 468) and testing (30%, n = 201) samples for model calibration and validation, and this dataset splitting procedure was defined earlier in the methodology section to ensure transparency and reproducibility of ROC–AUC–based model performance evaluation. The jackknife test results for the test dataset are shown in Fig. 11, indicating that road proximity, slope, and stream proximity are the most influential factors controlling landslide occurrence in the study area. The geomorphological setting, characterized by severe river erosion and steep slopes, supports these findings as key landslide triggers. In contrast, LULC and aspect show relatively lower standalone predictive power due to widespread vegetation cover and uniform terrain conditions. The integration of multiple conditioning factors within the MaxEnt model ensures that both natural and anthropogenic influences are effectively represented, enhancing the reliability of the landslide susceptibility assessment.

Fig. 11.

Testing data set used in Jackknife test for MaxEnt model validation.

The Fig. 12 represents the jackknife test for Area Under Curve (AUC). Since the test data consists of different information, including distinct landslide locations, the resulting AUC differ significantly from those of the training data. Stream proximity produced the highest AUC when used alone and its exclusion causes a notable drop in model performance. This confirms it as the most important factor influencing landslide susceptibility, likely due to fluvial erosion, toe-cutting, and valley-side instability. Slope shows a strong AUC when used in isolation, indicating its significant role in determining slope failures. Its influence aligns with known mechanics of landslides on steep gradients. Road proximity provides moderate predictive performance on its own but is less influential than slope or stream. Nonetheless, it reflects the impact of anthropogenic activities, such as road cuts destabilizing slopes. The parameters such as elevation, geomorphology, geology & LULC contributes reasonably to the model’s discriminatory power but does not outperform the top predictors. Their individual AUCs are lower, but their absence slightly reduces the model’s performance, suggesting complementary roles. Aspect has the lowest AUC when used alone, indicating limited stand-alone predictive capability. Its exclusion barely affects the overall model AUC, suggesting it is not a critical variable in this regional context.

Fig. 12.

Jackknife test of AUC for landslide validation.

To ensure the reliability and accuracy of the landslide susceptibility map, extensive field validation was conducted across all districts of Jammu and Kashmir. A total of 669 landslide sites previously delineated from satellite data and inventory maps were cross-verified during field surveys. Field validation involved ground truthing, visual assessment of slope failures, and verification of geomorphological and lithological conditions contributing to landslides (Fig. 12). During validation, 87% of the mapped landslide locations showed strong spatial correspondence with the field-identified landslides, confirming the reliability of the remote sensing-based inventory. Most of the landslides observed in the field were located in areas classified as high to very high susceptibility zones, supporting the accuracy of the Weighted Overlay (WO)-based modelling. Smaller and shallow landslides, which were difficult to detect in medium-resolution imagery, were effectively captured and verified through field inspections, enhancing the completeness of the landslide inventory. Moreover, several old and reactivated landslide scars were identified in the field that corresponded well with the geomorphological features evident in the DEM and slope curvature maps. These field observations also validated the spatial relationship between landslides and contributing factors such as slope gradient, lithology, proximity to roads and streams, and land use. Overall, the field validation confirms that the susceptibility model performs well in identifying areas at risk of landslides, particularly along critical infrastructure like roads and stream banks. The integration of field-verified landslide points has significantly strengthened the robustness of the final susceptibility map.

Discussion

The results of the present study demonstrate that landslide susceptibility in the Jammu and Kashmir region is strongly influenced by a combination of geomorphological, geological, hydrological, and anthropogenic factors, with high-risk zones predominantly concentrated along major transportation corridors and structurally weak mountainous terrain. The spatial distribution of high and very high susceptibility areas indicates significant clustering in districts such as Ramban, Doda, Kishtwar, Poonch, and Rajouri, particularly along major highways including NH-44, NH-244, NH-144 A, Mughal Road, and Kishtwar–Kelong Road (Fig. 13). These transportation corridors traverse fragile geological formations and steep slopes, making them highly prone to slope instability and frequent landslide occurrences. The close association between landslide-prone areas and road networks highlights the critical role of anthropogenic activities in destabilizing slopes through excavation, slope cutting, and removal of natural vegetation. Similar findings have been reported in recent Himalayan studies, where road construction and infrastructure development were identified as major contributors to landslide occurrence due to slope modification and increased erosion in structurally weak zones^4,5,27. This indicates that infrastructure expansion in mountainous terrain significantly increases landslide vulnerability and requires careful planning and mitigation strategies.

Fig. 13.

Field-photos of the landslides in different areas of the study region in which (a, b & c) shows the rockfall, debris slide, rockslide at the Samroli, Ramnagar & Basantgarh respectively; (d & e) debris slide at Reasi near Salal & Pernote, Ramban; (f) debris slide at Chanderkote, Ramban; (g) rock slide at Thatri, Doda; (h & i) debris slide at Mugal Maidan, Kishtwar & Peer ki gali, Poonch respectively; (j) debris slide at Srinagar near Pantha chownk; (k & l) debris slide at Anantnag, Dakshum & Ganderbal, Sonamarg; (m & n) debris flow and debris slide at Baramulla, near City town & Baramulla, Uri; (o) debris slide at Kathua, Bani.

Landslides are inherently complex phenomena influenced by multiple interacting environmental and human-induced factors, and the present study successfully integrates geospatial techniques with landslide conditioning parameters to identify spatial patterns of slope instability across Jammu and Kashmir. The developed landslide susceptibility map provides a comprehensive understanding of landslide-prone zones by analyzing the interaction between slope, lithology, geomorphology, drainage proximity, road proximity, and land use/land cover. Previous global and Himalayan studies have emphasized that integrating multiple conditioning factors within GIS-based frameworks significantly improves landslide hazard assessment and enhances prediction accuracy^3,15,16. The present study supports this approach by demonstrating that landslide susceptibility is not controlled by a single factor but rather by the cumulative effect of multiple geo-environmental variables. Slope angle emerged as one of the most dominant factors controlling landslide occurrence in the study area, with higher susceptibility observed in slope categories ranging from approximately 28° to more than 37°. Steep slopes increase gravitational stress and reduce slope stability, particularly in regions characterized by weak lithological formations and structural discontinuities. The presence of fractured rock units such as schists, slates, and weathered sedimentary formations further contributes to slope instability by reducing shear strength and increasing the likelihood of failure during rainfall events or anthropogenic disturbances. These findings are consistent with recent Himalayan studies, which reported that slope gradient is a primary controlling factor for landslide occurrence due to its direct influence on gravitational forces and material strength^1,6,8. Similar slope thresholds have been observed in the Uttarakhand and Sikkim Himalaya, where slopes exceeding 30° were found to be highly susceptible to landslides due to increased mass movement potential and reduced cohesion in weathered materials. The study also highlights that landslides can occur on moderately steep slopes when additional triggering factors such as road cutting, drainage erosion, and land-use changes are present. Field observations confirmed that slope modification due to human activities significantly increases landslide occurrence even in areas with moderate slope gradients. This observation aligns with the findings of Stanley and Kirschbaum¹⁷, who reported that anthropogenic disturbances alter natural slope equilibrium and increase landslide susceptibility in mountainous environments. Moreover, prolonged rainfall during the southwest monsoon season plays a crucial role in triggering slope failures by increasing pore water pressure and reducing soil cohesion. Rainfall infiltration leads to slope saturation and weakening of lithological units, ultimately resulting in mass movement along structural planes. Recent Himalayan studies have similarly emphasized the role of rainfall-induced slope instability and hydrological processes in triggering landslides, particularly in regions experiencing intense monsoonal precipitation^1,7,8. The rainfall analysis of Jammu and Kashmir from 2014 to 2024 shows a fluctuating but overall declining trend in both monsoon (July–September) and yearly average rainfall (Fig. 14). The rainfall data collected from CRU TS (0.5° x 0.5°) grid. The monsoon rainfall decreased gradually over the study period, while annual rainfall also showed significant variability with lower values in recent years. Since the Jammu and Kashmir region receives a major portion of rainfall during the southwest monsoon, any decline or irregularity directly affects hydrological conditions, soil moisture, and slope stability. Variations in rainfall patterns are mainly controlled by the interaction of southwest monsoon and western disturbances, along with regional climatic variability. These changes in rainfall patterns may contribute to increased environmental instability, including landslides and hydrological hazards in the region.

Fig. 14.

Rainfall variability in Jammu and Kashmir from 2014 to 2024: (a) average monsoon rainfall (July–September) and (b) yearly average rainfall showing interannual variability and gradual reduction over time.

Drainage proximity was identified as another critical factor influencing landslide occurrence, with higher susceptibility observed near rivers and streams. River erosion at slope toes reduces lateral support and increases slope instability, leading to mass movement and slope collapse. The strong relationship between drainage and landslide occurrence observed in this study is consistent with global and Himalayan research, where river incision and toe erosion were identified as key drivers of slope failure^2,3,5. Field verification further confirmed that many landslides occurred near drainage channels and riverbanks, supporting the reliability of the susceptibility mapping results.

Lithology and tectonic structures also play a significant role in controlling landslide susceptibility in Jammu and Kashmir. Weak and weathered rock formations were found to be more prone to slope failure compared to compact and stable lithological units, particularly in areas located near major tectonic structures such as the Main Boundary Thrust and Main Central Thrust. These structural features create zones of weakness that facilitate slope instability under rainfall and seismic stress. Similar findings have been reported in northwestern Himalayan studies, where tectonic structures and lithological heterogeneity were identified as major factors influencing landslide occurrence^4,6. In contrast, forested areas exhibited relatively low landslide susceptibility due to the stabilizing effect of vegetation roots and evapotranspiration, which reduces soil moisture and enhances slope stability. This observation is consistent with previous studies that highlighted the importance of vegetation cover in reducing landslide risk by increasing soil cohesion and reducing surface runoff⁹. The performance of the AHP and MaxEnt models demonstrates the effectiveness of the integrated susceptibility mapping approach used in this study. The MaxEnt model achieved AUC values of 0.82 for training data and 0.807 for testing data, indicating good predictive performance and reliable landslide prediction capability. These values fall within the acceptable range for landslide susceptibility modelling and are comparable to recent studies conducted in the Himalayan region and other mountainous areas, where AUC values between 0.75 and 0.90 were reported for MaxEnt and hybrid machine learning models^13–15. Field validation showed that approximately 87% of landslides occurred in high and very high susceptibility zones, further confirming the accuracy and reliability of the model outputs. Similar validation results have been reported in recent Himalayan susceptibility studies, demonstrating the effectiveness of integrated GIS and machine learning approaches in hazard assessment^5,15.

The LULC analysis indicates that barren and scrub land fall within the very high and high susceptibility class in the AHP evaluation. This imply that barren land causes landslides; rather, forests in Jammu and Kashmir are predominantly located on steep and structurally fragile Himalayan terrain where slope, lithology, and road proximity strongly control landslide occurrence. Snow and waterbody areas show very low susceptibility, while agriculture, built-up and forest regions exhibit low to moderate susceptibility due to limited soil movement. Therefore, LULC should be interpreted as a terrain-associated conditioning factor, while slope, road proximity, and drainage proximity remain the primary controlling variables.

The Jackknife test results indicated that road proximity, slope, and stream proximity were the most influential predictors in the MaxEnt model, highlighting the dominant role of transportation infrastructure and geomorphological factors in controlling landslide occurrence in Jammu and Kashmir. This finding is consistent with recent Himalayan research, where road proximity and slope gradient were identified as major contributors to landslide susceptibility due to slope destabilization and increased erosion along transportation corridors^15,27. The results therefore emphasize the need for sustainable infrastructure planning and slope stabilization measures in landslide-prone areas.

Recent studies conducted across the Himalayan region between 2022 and 2025 further support the findings of this research and highlight the growing importance of hybrid geospatial and machine learning approaches for landslide susceptibility mapping. Studies in the Central Himalaya, Uttarakhand, and Sikkim regions have reported that slope, drainage density, lithology, and road proximity are dominant factors controlling landslide occurrence, with high-risk zones concentrated in steep and structurally weak terrain^5,8,15. Similarly, research conducted in the northwestern Himalaya of India has demonstrated that hybrid models combining GIS, machine learning, and statistical approaches significantly improve landslide prediction accuracy and reduce uncertainty in susceptibility mapping. These studies collectively confirm that landslide occurrence in the Himalayan region is primarily governed by geomorphological and anthropogenic factors, consistent with the results obtained in the present study^28–39.

Despite the strong predictive performance of the integrated AHP–MaxEnt framework, some limitations should be acknowledged. The exclusion of high-resolution rainfall and seismic data due to limited spatial datasets may affect the dynamic prediction of landslide occurrence. Additionally, the landslide inventory used in the study may not capture all small-scale slope failures due to data availability constraints. Future research should focus on integrating high-resolution rainfall data, soil moisture information, remote sensing time-series data, and advanced machine learning techniques such as deep learning and ensemble modelling to develop dynamic landslide prediction systems. The incorporation of real-time monitoring and early warning systems can further enhance the applicability of landslide susceptibility mapping in mountainous regions.

Limitations and future work

This study provides a reliable landslide susceptibility assessment using the MaxEnt model; however, several limitations should be acknowledged. The modelling framework primarily relies on spatial conditioning factors, and although rainfall has been incorporated into the analysis, other dynamic triggering factors such as seismic activity and temporal environmental variations were not included due to limited availability of high-resolution and spatially consistent datasets for the study area. The landslide inventory used in this study is based on available historical records and satellite interpretation, which may not capture all small or recent landslide events, potentially affecting model generalization. In addition, MaxEnt is a presence-only model and does not explicitly utilize absence data, which may limit its ability to fully represent complex non-landslide conditions. Machine learning models were not considered in the present study because they generally require large and balanced datasets with both presence and absence samples, extensive parameter tuning, and high computational resources, which were beyond the scope of this work. The primary objective of this research was to develop a robust and interpretable landslide susceptibility model using MaxEnt, which performs well with limited and presence-only data and provides clear insights into the contribution of environmental factors.

Despite the strong predictive performance of the integrated AHP–MaxEnt framework, several limitations should be acknowledged. The landslide inventory used in this study may contain spatial bias, as landslides are more frequently mapped and reported along road corridors and accessible areas, which can influence model training and increase the apparent importance of road proximity. The spatial resolution of the DEM may also limit the representation of micro-topographic variations and small-scale slope failures in complex Himalayan terrain. In addition, the weighted overlay (AHP) method involves expert-based pairwise comparisons and factor weighting, which may introduce subjectivity and uncertainty in susceptibility ranking. Although rainfall was incorporated as a conditioning factor, the use of coarse-resolution spatial datasets may not fully capture localized rainfall intensity and short-duration extreme events.

Future research should focus on integrating high-resolution seismic datasets and multi-temporal environmental variables to develop dynamic landslide susceptibility models and improve hazard prediction accuracy. Future research also focuses on developing more comprehensive landslide inventories, integrating high-resolution DEM and time-series rainfall and soil moisture data to improve model robustness and support dynamic landslide prediction and early warning systems in the Himalayan region. The incorporation of advanced machine learning and hybrid modelling approaches, including Random Forest, Support Vector Machine, XGBoost, and deep learning techniques, could enhance predictive performance through comparative and ensemble modelling frameworks. Additionally, expanding the landslide inventory through detailed field surveys, UAV-based mapping, and multi-temporal remote sensing analysis would improve model reliability. Future studies should also explore multi-model comparisons and spatio-temporal landslide prediction frameworks to develop more comprehensive and operational landslide hazard assessment systems for the region.

Conclusions

This study applied geospatial techniques and an integrated AHP–MaxEnt modelling framework to assess landslide susceptibility in the Jammu and Kashmir region and to identify the spatial distribution of landslide-prone zones. The analysis classified the study area into five susceptibility classes (very low, low, moderate, high, and very high), providing a structured basis for understanding landslide risk and supporting targeted mitigation and land-use planning strategies. The model performance indicated reliable predictive capability, with AUC values demonstrating good agreement between training and testing datasets and field validation confirming that most landslides occurred within high and very high susceptibility zones.

The results clearly show that slope, road proximity, and stream proximity are the most influential variables controlling landslide occurrence in the study area, as confirmed by the MaxEnt jackknife test and spatial analysis. Steep slopes increase gravitational instability, while proximity to roads and drainage networks significantly enhances slope failure due to excavation, toe erosion, and slope modification. Geological and tectonic features, including thrust faults and weak lithological formations, act as secondary conditioning factors that further increase susceptibility in structurally unstable zones, particularly along major transportation corridors such as NH-44, NH-244, NH-144 A, Mughal Road, and Kishtwar–Kelong Road. The clustering of high susceptibility zones in districts such as Ramban, Doda, Kishtwar, Poonch, and Rajouri highlights the strong relationship between infrastructure development and slope instability in the Himalayan terrain. The landslide susceptibility map developed in this study provides a practical tool for identifying high-risk areas and prioritizing slope stabilization and infrastructure management measures. High and very high susceptibility zones require immediate attention through slope protection, controlled road construction, and continuous monitoring, while moderate susceptibility areas require careful planning and environmental assessment before development. Low susceptibility zones can support controlled and sustainable land-use activities with minimal risk.

Overall, the integration of AHP and MaxEnt models demonstrates a robust and reliable approach for regional-scale landslide susceptibility assessment in mountainous terrain. The results contribute to improved understanding of landslide controlling factors in the Jammu and Kashmir Himalaya and provide a scientific basis for hazard-aware infrastructure planning and land-use management. Future studies can further improve the model by incorporating high-resolution rainfall, soil moisture, and time-series remote sensing data to enhance predictive accuracy and support dynamic landslide monitoring in the region.

Author contributions

Avtar Singh Jasrotia contributed to writing (review and editing), conceptualization, validation, visualization, and supervision. Amit Sharma contributed to writing the original draft, methodology, and formal analysis. Iswar Chandra Das contributed to visualization and supervision. Tapas Ranjan Martha was involved in conceptualization and validation. Rajesh Kumar contributed to validation. Vikshay Kumar was responsible for software development, methodology, and data curation while Akarshan Rasyal contributed to visualization.

Funding

The corresponding author’s affiliated institution has an established financial agreement with the journal.

Data availability

The datasets used and/or analysed during the current study available from the corresponding author on reasonable request.

Declarations

Competing interests

The authors declare no competing interests.