Landslide Research: State of the Art and Innovations

1. Introduction

Landslides, defined as downslope movements of rock, soil, or debris under the action of gravity [1], are among the most pervasive geohazards on Earth. They occur across all climatic zones and on every continent wherever slopes exist, affecting both continental and submarine environments, and are frequently triggered or exacerbated by other geophysical processes—including earthquakes, intense or prolonged rainfall, rapid snowmelt, and volcanic activity [2,3]. Unlike many other natural hazards, landslides often act as secondary events superimposed on primary triggers, causing them to be systematically under-reported in global disaster databases [4,5].

In terms of global societal impact, landslides rank among the most lethal categories of natural hazards. According to the Emergency Events Database (EM-DAT) maintained by the Centre for Research on the Epidemiology of Disasters (CRED/UCLouvain), mass movements (wet) recorded 19 events and 958 deaths in 2025 alone, marginally above the 2005–2024 annual averages of 18 events and 897 deaths per year [6]. These records capture only disasters meeting EM-DAT entry criteria and substantially underestimate actual landslide losses: the Global Fatal Landslide Database suggests that EM-DAT undercounts fatal events by up to 1400–2000% and associated deaths by 330–430% relative to ground-level documentation [4,7]. The most comprehensive long-term analysis, by Froude and Petley [7], examined a 13-year (2004–2016) global database of fatal non-seismic landslides and documented 4862 distinct lethal events claiming a total of 55,997 lives, with a mean annual toll exceeding 4000 victims. Asia concentrates approximately 67.7% of fatal events and 74% of associated fatalities, driven by the convergence of high relief, intense monsoon precipitation, dense population, and rapid urbanization in mountain belts of South, Southeast, and East Asia [7,8]; in 2025, Asia as a whole accounted for 72.8% of all natural-disaster deaths recorded by EM-DAT [6]. The Americas account for approximately 15% of landslide events and 11% of fatalities [7], and Europe, despite lower absolute exposure, experiences significant losses, particularly in the Alpine, Apennine, and Iberian mountain systems [7,9] (Figure 1). Among the most notable landslide events of 2025, a catastrophic mass movement triggered by intense rainfall in late August in the Marrah Mountains of Central Darfur, Sudan, engulfed the village of Tarasin, resulting in approximately 400 casualties and ranking as the seventh-deadliest natural disaster of that year globally [6].

Annual global economic losses attributable to landslides have been conservatively estimated at USD 20 billion or more [5,10], with the United States alone sustaining USD 1–3.6 billion per year [11]. EM-DAT does not disaggregate economic losses for mass movements separately from other hazard types in 2025, but the database records zero direct economic losses attributed to wet mass movements in that year, reflecting severe under-reporting of landslide economic impact in low- and middle-income countries rather than an absence of losses [6]. The increasing trend in fatal landslide occurrence documented by [7] is driven by a combination of rising anthropogenic pressures—including informal settlement on unstable slopes, illegal mining, and uncontrolled road construction—and the growing frequency of extreme precipitation events attributable to anthropogenic climate change [12,13]. In 2025, the Sudan Tarasin landslide exemplified both drivers: deforestation of the Marrah highland massif and an anomalously intense late-summer rainfall event combined to produce one of the deadliest single landslide events recorded in Africa in recent decades [6].

Against this global backdrop, the scientific community has invested heavily in advancing landslide research across all thematic domains, from fundamental geomorphological characterization to operational early-warning systems.

2. Landslide Research: State of the Art and Innovations

The Special Issue, Landslide Research: State of the Art and Innovations, was conceived as a vehicle for collecting and disseminating innovative peer-reviewed contributions across the full spectrum of landslide science. This Editorial provides the global and methodological context for this collection, describes all 11 published articles, and identifies priority directions for future research.

Asgary et al. [14] reported a virtual reality (VR) educational framework based on a 3D simulation of the Chiradzulu landslide disaster triggered by Cyclone Freddy in southern Malawi (March 2023). Three distinct terrain configurations were reproduced to illustrate how topographic factors modulate landslide run-out and impact zones. The VR environment enables interactive, risk-free scenario exploration without physical exposure to hazardous terrain, constituting a particularly valuable approach in low- and middle-income country contexts where field-based professional training is severely resource-constrained. The study demonstrates that photorealistic VR reconstructions of documented events can be produced at relatively low cost from freely available topographic and remotely sensed data, offering a scalable model for landslide disaster education.

Vishnu et al. [15] addressed the effect of intra-seasonal rainfall variability on landslide early-warning threshold performance in the Western Ghats monsoon belt of India. Using gauge-corrected precipitation products and a multi-source landslide catalogue, the authors developed separate empirical intensity–duration thresholds for the early and late monsoon phases, incorporating antecedent rainfall windows of 1–40 days. The approach substantially reduced false alarm rates relative to conventional single-season thresholds, demonstrating that antecedent soil moisture—approximated through lagged cumulative rainfall—is a primary modulator of slope hydrological response. The methodology constitutes a transferable reference for threshold regionalization in seasonally contrasted climatic contexts, with direct applicability to the monsoon belts of South and Southeast Asia.

Tiranti [16] developed a GIS-based classification scheme for evaluating sediment gravity flow (SGF) hazard in Alpine catchments on the basis of dominant lithology and the percentage of outcropping bedrock. The method enables rapid, reproducible screening of catchment-scale SGF hazard applicable at regional scales with limited site-specific geotechnical data, directly integrable into land-use planning and civil protection frameworks. The contribution advances the operationalization of SGF hazard management in Alpine environments where dense geotechnical investigation networks are economically unfeasible.

Forno et al. [17] investigated the Pointe Leysser deep-seated gravitational slope deformation (DSGSD) in the Aosta Valley (northwestern Italy) using deep electrical resistivity tomography (ERT) integrated with geological and geomorphological field mapping. The Aosta Valley hosts numerous DSGSDs that produce complex buried morpho-structures partially invisible to surface survey methods owing to subglacial abrasion and sedimentary burial. Multiple ERT profiles revealed three wide buried glacial valleys infilled with glacial sediments and mapped the spatial distribution of gravitational morpho-structures at depth, establishing a clear relationship between pre- and post-glacial slope movements and the overdeepened glacial trough. The study illustrates the indispensable role of deep geophysical imaging for characterizing slope instabilities that cannot be adequately constrained by surface data alone.

Psarropoulos et al. [18] presented a comprehensive analysis of the mechanisms through which climate change (CC) is expected to alter the stability of soil slopes, addressing both hydrological (rainfall intensity and duration, groundwater table fluctuations, soil erosion) and geotechnical (pore pressure, effective shear strength) dimensions. The authors demonstrated that CC-induced modifications of precipitation characteristics may either increase or decrease slope stability depending on site-specific soil conditions, drainage configurations, and the direction of change in seasonal moisture regimes—contrary to a simplistic narrative of universal destabilization. The analysis emphasizes the necessity of site-by-site assessment incorporating long time series of climatic and geotechnical parameters and provides a theoretical and practical framework directly relevant to climate change adaptation in landslide risk governance.

Ben-Yehoshua et al. [19] quantified the destabilizing effect of ongoing glacial thinning on the Svarthamrar slope, a large volcanic slope instability adjacent to the Svínafellsjökull outlet glacier in Southeast Iceland. Since the turn of the 20th century, progressive deglaciation has exposed mountain flanks around Iceland’s outlet glaciers; the Svarthamrar instability is characterized by a fracture system exceeding 2 km in length. Integrating updated glacier bed topography with stratigraphical and structural assessment, the authors showed that pre-existing structural discontinuities and a strongly overdeepened glacial trough predisposed the slope to instability, while glacial unloading from 1890 onward controlled progressive destabilization, amplified by transiently elevated hydraulic gradients during rapid glacier thinning in the late 1990s–2000s. Future glacial retreat is predicted to further reduce slope stability, and the study offers a globally relevant perspective on ice-loss-driven mass movement hazards under accelerating cryosphere change.

Cantonati et al. [20] extended the Alpine catchment classification framework of Tiranti [16] to the practical problem of monitoring system design, demonstrating how evidence-based characterization of lithology, sediment source areas, and historical debris flow occurrence directly informs the optimized design of sensor networks for debris flow detection on alluvial fans. Sensor type selection (seismic geophones, tipping-bucket rain gauges, ultrasonic flow-depth sensors) and spatial placement decisions were guided by catchment typology. Near-real-time data recorded during a 2024 debris flow event validated both the theoretical predictions of the classification model and the functionality of the installed monitoring array, confirming that typology-driven design yields operationally effective sensor configurations at reduced deployment costs.

Wu et al. [21] proposed a systematic framework for constructing multi-temporal knowledge graphs of landslide hazard chains, integrating heterogeneous datasets through the Resource Description Framework (RDF) and a quadruple (subject–predicate–object–timestamp) representation scheme. The temporal ontology model, aligned with a three-phase disaster chain structure (pre-event–event–post-event), enables multidimensional semantic querying along the temporal axis, overcoming the principal limitation of conventional static knowledge graphs. A web-based interface was developed and validated as a prototype for an operational landslide information platform capable of continuous knowledge updating as new monitoring data, event reports, and field assessments become available. The framework is directly applicable to national and regional landslide information systems and to heterogeneous data stream integration from sensor networks and earth observation archives.

Liu et al. [22] proposed a weighted-voting-based multi-classifier ensemble method (WPU) assigning category-specific weights based on producer’s accuracy (PA) and user’s accuracy (UA) for rapid co-seismic landslide detection from remote sensing imagery. The method was validated on 193 co-seismic landslides mapped in Jiuzhaigou County, Sichuan Province, China, following the M_s 7.0 earthquake of 8 August 2017. Six commonly used remote-sensing-based classification methods were fused through WPU, achieving an overall accuracy of 0.9755 and a Kappa coefficient of 0.7848, a substantial and consistent improvement over any individual classifier. The approach maintains computational efficiency and timeliness, making it well-suited for post-earthquake emergency response, where rapid inventory generation is critical for prioritizing rescue operations.

Bhattarai et al. [23] presented a data-driven susceptibility framework for the 2323 co-seismic landslides triggered by the M_s 7.5 Noto Peninsula earthquake (Japan, 1 January 2024). Random forest (RF) and logistic regression (LR) models trained on 11 conditioning factors achieved AUC = 0.914 and an overall detection accuracy of approximately 85% for the RF model, substantially outperforming a benchmark Hazus-based method (67%). Spatial analysis revealed that 75% of failures were smaller than 3220 m², that southwest-facing aspects were overrepresented, and that mean slope angle at failure sites averaged 31.8°. SHAP-based explainability analysis identified epicentral proximity, slope angle, and lithology as dominant susceptibility drivers. The study critically discusses limitations of applying global susceptibility models to region-specific co-seismic events and advocates for regionally specific training inventories as an operational priority.

Gevorgyan et al. [24] documented the interdisciplinary emergency investigation of the Tumanyan landslide (Lori Marz, northern Armenia), reactivated in January 2018 within the Arabia–Eurasia collision zone, a region with more than 2504 catalogued active landslides. The landslide threatened critical infrastructure (the Yerevan–Tbilisi railway and the M6 interstate highway), and preliminary hazard analysis indicated risk of Debed River impoundment with catastrophic downstream consequences. The investigation combined UAV-based digital terrain mapping with four complementary geophysical methods: Multichannel Analysis of Surface Waves (MASWs), microtremor recordings, Ground Penetrating Radar (GPR), and Vertical Electrical Sounding (VES). The combined dataset characterized the three-dimensional geotechnical structure of the landslide body and identified the primary sliding plane. Numerical slope stability modelling confirmed critical instability and susceptibility to seismic or groundwater reactivation. Emergency horizontal drainage drilling, designed based on these findings, successfully arrested movement and mitigated immediate infrastructure risk, exemplifying the value of rapid multi-method geophysical investigation for evidence-based emergency landslide management.

3. Future Directions

The trajectory of landslide research will be increasingly shaped by the compound pressures of anthropogenic climate change and global demographic growth. The IPCC Sixth Assessment Report confirms with high confidence that extreme precipitation events will intensify across most of the globe under all warming scenarios above 1.5 °C [25], directly translating into increased frequency and severity of rainfall-induced landslides in mountainous and hilly terrain. Simultaneously, ongoing glacial retreat and permafrost degradation are destabilizing slopes at altitudes and in geographic regions where landslide hazard has historically been lower [26,27]. Cascading hazard chains, including landslide-dam outburst floods and debris-flow–river-blockage sequences [28], are expected to increase in frequency in glaciated mountain belts with potentially transboundary impacts.

Against this backdrop, the integration of AI tools into landslide science represents arguably the most transformative frontier for improving societal resilience. Deep learning architectures, including transformer-based models and graph neural networks, demonstrate the capacity to process multi-source, multi-temporal geospatial data at scales and speeds incompatible with classical approaches, enabling quasi-real-time global susceptibility updating [29]. Physics-informed machine learning (in which deep learning architectures are constrained by geotechnical and hydrological governing equations) promises to overcome the extrapolation limitations of purely data-driven models in regions with sparse training inventories [30].

Landslide early warning systems of the next generation will need to operate in fully probabilistic, multi-hazard, multi-scale frameworks, combining numerical weather prediction outputs, real-time sensor networks (MEMS accelerometers, fibre-optic distributed sensing, GNSS, and TDR), satellite-derived rainfall products, and AI-based decision support to produce actionable alerts with quantified uncertainty [31]. Community-based warning components (leveraging mobile phone networks, social media monitoring, and citizen science) will be essential for extending coverage to rural and peri-urban populations in low- and middle-income countries that sustain the greatest landslide mortality [7].

Risk governance frameworks must systematically incorporate climate change projections into hazard zonation, land-use planning, and infrastructure design standards [12,32,33]. Risk-informed nature-based solutions—reforestation, catchment restoration, and green infrastructure for slope stabilization—will complement engineered countermeasures in a holistic approach to landscape resilience that addresses both immediate hazard reduction and long-term environmental sustainability under a changing climate [34].