Ting Xiao and colleagues analyzed a multi-temporal inventory of 1,518 landslides recorded in Wanzhou District, eastern Chongqing, from 1950 to 2020. The records originated from long-term monitoring and field surveys conducted by the Chongqing Wanzhou Institute of Geological Environment Monitoring. Landslide boundaries were redrawn from field investigations, archival photographs and Google Earth imagery.

To examine both changing event patterns and model transferability, the researchers divided the inventory into five periods, designated T1 through T5, and created modeling and prediction cases that compared individual decades with progressively longer historical windows. The study area combines a humid subtropical monsoon climate with extensive environmental transformation. Wanzhou’s central urban area expanded nearly 50-fold between 1928 and 2020, and the district underwent major changes associated with the construction and operation of the Three Gorges Project.

From 1960 to 2020, annual rainfall fluctuated without a clear trend and averaged 1,194.34 millimeters, with 70% falling from May through September. Heavy-rainfall days and annual mean temperature from 1990 to 2020 appeared to trend upward, although those shorter records did not support a robust quantitative trend estimate comparable to that for urban expansion. More than 90% of the inventoried slope failures occurred during rainfall episodes, predominantly within Jurassic sandstone-mudstone interbedded strata, and the authors identified rainfall infiltration as the primary trigger.

They initially selected 18 possible conditioning factors covering topography and geomorphology, geology, hydrology and human impact. Pearson correlation screening removed terrain wetness index and terrain roughness index because both had strong correlations with slope, leaving 16 factors for the susceptibility models. Random forest and extreme gradient boosting were used to produce maps from inventories with different temporal spans, while receiver operating characteristic curves and area under the curve, or AUC, evaluated discrimination.

The modeling procedure explicitly separated performance within a period from prediction in a subsequent period. For each modeling dataset, landslide polygons were divided at random into 70% training and 30% testing subsets, with all grid cells inside an individual polygon assigned to the same subset. The models were then evaluated against landslides from later periods.

Because each inventory subset contained a limited number of events, non-landslide samples were randomly selected in equal numbers and the modeling was repeated 200 times. Susceptibility values at test grid points stabilized after 200 repetitions for both algorithms, which the researchers selected as the final number of runs. Both algorithms showed the same broad temporal pattern.

A random forest trained only on the first inventory period obtained training and test AUC values of 0.94, but its prediction AUC declined to 0.77, 0.65, 0.64 and 0.62 for the second through fifth periods. XGBoost produced training and test values of 0.95 and 0.94, followed by prediction values of 0.74, 0.62, 0.59 and 0.61. Longer calibration windows did not restore that transferability.

Models based on the first two, three or four periods continued to have strong within-period test performance, but their cross-period predictions were generally between about 0.62 and 0.68. A benchmark model focused on the third period, 1990–1999, made the loss of temporal validity more explicit. The map achieved training and test AUC values between 0.854 and 0.886, but its performance deteriorated when tested against subsequent inventories.

Random forest AUC fell to 0.656 for landslides in the following decade and 0.627 when evaluated across the next 20 years; the value for the final decade alone was 0.608. XGBoost declined to 0.651 after 10 years and 0.613 after 20 years, reaching 0.588 for the final decade alone. Validation against earlier periods also performed poorly, indicating that the benchmark was not broadly representative of landslide conditions in those earlier periods either.

The locations classified as most susceptible changed as well. In the earliest period, high-susceptibility areas formed a narrow cluster along the upper Yangtze River, consistent with the high model importance of distance to rivers, elevation and annual precipitation. During the second period, the pattern shifted northeast along the riverbanks.

In the third, the major axis rotated perpendicular to the Yangtze, coinciding with the initial construction phase of the Three Gorges Project. Road construction, hillside excavation, reservoir works and other engineering may have disturbed slopes beyond the main river corridor, although the study did not quantify each activity’s contribution. In the fourth period, the high-susceptibility axis realigned with the river, then migrated northeast toward urban centers in the fifth.

Maps were classified into five susceptibility levels using the Jenks method, and standard deviation ellipses were used to describe the spatial clustering and direction of the highest class. The two machine-learning approaches produced similar ellipse patterns, supporting the consistency of the temporal interpretation. The analysis also examined how individual causal factors were weighted in the random forest models.

Elevation, surface texture, distance to rivers, annual precipitation, geological structure, lithology and land use remained consistently important, but their relative rankings changed among periods. Elevation’s importance rose from approximately 0.3 to 0.4, whereas lithology declined from 0.1 to 0.05 and annual precipitation from 0.25 to 0.1. These shifts in model-derived importance should not be interpreted as direct measurements of how physical controls changed.

Every model used the same static factor layers, so the differences primarily reflect relationships learned from different landslide inventories. For example, a fixed long-term mean annual precipitation layer may not represent the rainfall conditions most relevant during a particular decade. Likewise, the declining importance of lithology does not establish that human disturbance became more influential; it shows only that the fitted models assigned different relative importance to the available variables.

A frequency-ratio analysis also found temporal variation in spatial associations, although values for small spatial classes were sensitive to limited area and modest changes in event counts. Several limitations restrict how far the results can be generalized. A 1:50,000 geological-hazard investigation conducted during 2010–2020 may have improved detection of smaller landslides, contributing to that period’s larger inventory and smaller median event area.

Completeness may therefore differ among periods. High-resolution pre-2000 digital elevation models and earlier land-use layers were also unavailable, preventing fully dynamic modeling across 1950–2020. Random sampling of non-landslide locations introduced uncertainty despite 200 repeated runs, and nearby landslides assigned to training and test sets could still share spatial dependence.

The study therefore evaluates the temporal transferability of static susceptibility models; it does not directly quantify how reservoir impoundment, urbanization or other environmental changes altered landslide probability. The central result is that inventory size alone does not guarantee a more transferable susceptibility map. Extending a training window can add events while also combining periods in which the relationships among conditioning factors and landslides differ.

A narrow window may yield a stronger within-period model, but that advantage can decay quickly when the map is applied to a later period. The Wanzhou analysis supports explicit temporal boundaries for conventional susceptibility maps and periodic reassessment after relevant environmental changes, while also pointing toward future models that incorporate time-dependent information and process-based parameters.

Subject of Research: Temporal transferability of landslide susceptibility models in Wanzhou, China

Article Title: Same area, different timeframes: unveiling the evolution of landslide patterns and their impact on susceptibility maps

Article References: Xiao, T., Segoni, S., Liang, X., Liu, Q., & Jiang, J. (2026). Same area, different timeframes: unveiling the evolution of landslide patterns and their impact on susceptibility maps. Natural Hazards, 122(22), Article 678. https://doi.org/10.1007/s11069-026-08458-2

Image Credits: AI Generated

DOI: 10.1007/s11069-026-08458-2

Keywords: landslide susceptibility, temporal validation, landslide inventory, machine learning, random forest, XGBoost, Three Gorges Reservoir, environmental change, Wanzhou District, hazard mapping, Same, area

Landslides are a persistent threat in the Western Ghats, the mountain chain running along India’s western coast. Roughly fourteen percent of India’s land area is considered landslide-prone, and the country loses an estimated 500 lives and around four billion rupees each year to these events. In Kerala, the west-facing scarps of the Ghats cover about 1,400 square kilometres, and debris flows are the most frequent type of failure in the highlands. Nearly all of these mass movements occur during the monsoon, when prolonged and intense rainfall raises pore-water pressures in the soil and undermines slope stability.

The study area covers roughly 70 square kilometres around Idukki township, including Painav, Cheruthoni, Karimban, Thadiyampad, Maniyarankudy, Chelachuvadu, Keerithod and Kanjikuzhy. The terrain is rugged, with altitudes ranging from 500 to 2,600 metres, and the district receives 3,000 to 4,000 millimetres of rain annually from the southwest and northeast monsoons. The underlying geology is dominated by Precambrian crystalline basement rocks, chiefly biotite gneiss and hornblende-biotite gneiss, with no major lineaments identified in the mapped region.

What distinguishes this work from most landslide susceptibility studies is its integration of geotechnical parameters alongside the usual geomorphological factors. The researchers compiled fifteen conditioning factors in total: elevation, slope angle, slope aspect, profile and planiform curvature, topographic wetness index, geomorphology, land use, distance from roads and distance from drainage, plus five soil properties measured in the laboratory: cohesion, angle of internal friction, liquid limit, plastic limit and plasticity index. Soil samples were collected from locations with different slope-forming materials, and the point measurements were interpolated into continuous surfaces using kriging.

The foundation of any susceptibility model is a reliable inventory of past events. The team assembled a catalogue of 502 landslide locations by combining government reports with interpretation of pre- and post-event high-resolution satellite imagery, cross-checked against Geological Survey of India datasets. Most of the mapped landslides were triggered during the extreme rainfall of August 2018, though significant activity continued through 2019 and 2020. Slides accounted for 54 percent of the inventory and flows for 45 percent. For modelling, a balanced dataset of 503 landslide and 533 non-landslide points was split 70/30 into training and testing subsets.

The AHP method works by having experts compare factors pairwise on Saaty’s nine-point scale, producing weights that reflect each factor’s relative importance. The resulting comparison matrix passed the standard consistency check, with a consistency ratio below 0.10. The Random Forest model, by contrast, learned relationships directly from the data. Implemented in Python with the scikit-learn library, it used an ensemble of 100 decision trees, each trained on random subsets of the features, with predictions combined by majority vote. This ensemble approach reduces overfitting while capturing nonlinear interactions among factors that expert weighting cannot easily express.

The two models produced noticeably different maps. Under AHP, 28.43 percent of recorded landslides fell in high susceptibility zones and 37.78 percent in very high zones, with the highest-risk areas concentrated in the central and western parts of the study area, including Maniyarankudy, Thadiyampad and Karimban. The Random Forest model distributed landslides more evenly across classes, assigned the largest share of area to the medium susceptibility class (24.60 percent, or 17.18 square kilometres), and flagged additional high-risk zones around Kanjikuzhy in the northwest and eastern Karimban, while classifying Keerithodu and Painavu as very low susceptibility.

Validation told a clear story. The Random Forest model achieved a receiver operating characteristic area under the curve of 0.99, placing it in the excellent category, compared with 0.85 for AHP, which falls in the very good range. Accuracy was 0.947 for Random Forest against 0.774 for AHP, precision was 0.922 versus 0.777, recall was 0.947 versus 0.753, and the F1-score was 0.945 versus 0.765. The authors attribute the gap to AHP’s reliance on subjective expert judgement, which can introduce inconsistencies, while Random Forest learns directly from historical landslide data.

Despite the machine learning model’s superior numbers, the researchers are careful not to dismiss the expert-based approach. AHP remains transparent, easy to interpret, and valuable in regions where comprehensive landslide inventories do not exist, since it functions independently of historical event data. Its main drawbacks are the subjectivity of weight assignment and the assumption that each factor’s importance is uniform across the landscape. The authors argue that the two methods should be seen as complementary rather than competing, and that combining them can strengthen susceptibility assessments.

The study also acknowledges its limitations: the analysis rests on a finite historical inventory and a limited number of geotechnical sampling locations, and the resulting maps represent present-day conditions without explicitly accounting for changing rainfall patterns, groundwater conditions or land-use practices. Future work, the team suggests, could incorporate time-dependent factors such as rainfall intensity and groundwater fluctuations, along with higher-resolution topographic data and more advanced hybrid machine learning techniques. As extreme rainfall events intensify in a warming climate, tools like these are becoming central to land-use planning, early warning systems and disaster risk reduction in the world’s mountainous regions.

Subject of Research: Landslide susceptibility mapping in the Western Ghats using Random Forest and AHP with integrated geomorphological and geotechnical parameters

Article Title: Landslide susceptibility assessment in the Western Ghats adjoining Idukki township, Kerala, India, using integrated geomorphological and geotechnical parameters with AHP and machine learning

Article References: P, A. S., S, A., S, A. S., & P, J. K. (2026). Landslide susceptibility assessment in the Western Ghats adjoining Idukki township, Kerala, India, using integrated geomorphological and geotechnical parameters with AHP and machine learning. Discover Geoscience, 4(1), Article 406. https://doi.org/10.1007/s44288-026-00778-x

Image Credits: AI Generated

DOI: 10.1007/s44288-026-00778-x

Keywords: landslide susceptibility, Western Ghats, Idukki, Random Forest, Analytic Hierarchy Process, machine learning, geotechnical parameters, geomorphology, Kerala, disaster risk reduction, ROC-AUC, slope stability

The team’s central innovation is a unified spatial framework that links three separate analyses into a single decision-making pipeline. First, they mapped susceptibility to floods, landslides, and urban heat using a Multi-Criteria Evaluation (MCE) model built on the Analytic Hierarchy Process (AHP), a structured method for weighing competing factors such as slope, drainage density, land cover, population density, vegetation indices, and Land Surface Temperature (LST). Second, they overlaid a suitability analysis identifying which parts of the city are physically and socially appropriate for Nature-based Solutions (NbS) — interventions like riparian buffers, slope stabilization plantings, urban forests, and green corridors. Third, they grounded both layers in the lived experience of residents through a socioeconomic survey of 305 households, supplemented by focus group discussions and key informant interviews. The result is not just a hazard map but a blueprint for where green infrastructure can realistically be deployed.

The numbers are striking. According to the study, 34.4 percent of Dessie’s land area falls into the high-susceptibility category for landslides, a figure that reflects the city’s steep topography and rapid, often unplanned, expansion onto fragile slopes. Meanwhile, 57.6 percent of the city is moderately susceptible to flooding, driven by intense seasonal rainfall, degraded upstream catchments, and impervious surfaces that prevent water from infiltrating the soil. Urban heat, the third hazard, is widespread across the city because extensive built-up surfaces absorb and re-radiate solar energy while limited vegetation cover reduces the cooling that trees and green spaces normally provide. Together, these three hazards create a compound risk profile that traditional single-hazard planning approaches systematically underestimate.

What makes this research particularly valuable is its insistence that physical susceptibility alone is not enough to guide action. The authors argue that many urban planning paradigms in the region evaluate environmental hazards in isolation, producing fragmented responses that miss the compound nature of climate risk. By layering hazard susceptibility, NbS suitability, and community perception into one framework, the study bridges a persistent gap between what the landscape can physically support and what residents are actually willing and able to maintain. This integrated approach is especially relevant for rapidly urbanizing, data-scarce cities across sub-Saharan Africa, where comprehensive hazard modeling is often unavailable but the need for evidence-based adaptation is urgent.

The NbS suitability analysis reveals that 42.4 percent of Dessie is highly or very highly suitable for green infrastructure interventions, while 24.2 percent has low suitability due to biophysical constraints such as steep rocky terrain or competing land uses like dense informal settlement. The strategic hotspots for intervention are concentrated in upstream and mid-slope zones, where riparian buffers along drainage channels can slow runoff, slope stabilization plantings can anchor unstable soils, and urban forests and green corridors can break up heat islands while providing recreational and ecological benefits. These are not abstract recommendations: the authors argue that municipal authorities could use the maps to legally enforce blue-green zoning along highly susceptible drainage channels, prioritize structural watershed interventions in upstream catchment zones, and direct resources toward underserved, climate-exposed informal settlements that face the highest combined exposure.

Community voices form the third pillar of the study, and they reveal a nuanced picture of local readiness. The survey of 305 households, along with focus group discussions and key informant interviews, found strong awareness among residents of the environmental, socioeconomic, and health benefits that Nature-based Solutions can deliver — from reduced flood damage to cooler streets and improved wellbeing. Yet awareness does not translate automatically into implementation. Residents and local stakeholders identified several persistent barriers: limited financial resources, high maintenance costs for green infrastructure, weak institutional coordination between municipal departments, and land tenure insecurity that discourages long-term investment in trees and vegetation on land people do not formally own. These findings underscore that successful NbS deployment is as much a governance and equity challenge as it is a technical one.

The study’s methodological rigor deserves attention for anyone interested in how spatial analysis can inform climate adaptation. The AHP-based MCE approach allows researchers to assign explicit weights to different contributing factors — for example, how strongly slope gradient versus drainage density influences landslide susceptibility — and then combine them into composite indices. The inclusion of socio-environmental variables such as population density and vegetation indices alongside purely physical parameters reflects a growing recognition that hazard exposure is fundamentally a function of where people live, not just where the land is unstable. Land Surface Temperature, derived from satellite thermal imagery, provides a direct measure of urban heat rather than relying solely on air temperature stations, which are sparse in many African cities.

For Dessie specifically, the implications are concrete. The city sits in the Amhara region at the edge of the Ethiopian Highlands, where steep escarpments funnel seasonal rains into narrow valleys and where decades of deforestation and agricultural expansion have left upstream catchments degraded. Rapid population growth has pushed informal settlement onto slopes that were once considered too risky to build on, placing low-income households directly in the path of landslides and floodwaters. The study’s maps offer municipal planners a way to prioritize interventions where they will deliver the greatest risk reduction per unit of investment, a critical consideration in a resource-constrained context where every birr spent on infrastructure must be justified.

The research also speaks to a broader global conversation about how cities in the Global South can adapt to climate change without importing expensive, engineering-heavy solutions that may not be maintainable locally. Nature-based Solutions — working with vegetation, soil, and natural hydrology rather than against them — offer a cost-effective alternative, but only if they are sited intelligently and supported by the communities who live alongside them. By demonstrating that hazard susceptibility, NbS suitability, and community perception can be integrated into a single GIS framework, the Dessie study provides a template that other data-scarce, rapidly urbanizing cities could adapt to their own contexts.

The authors acknowledge that their work is a snapshot, not a forecast. Climate change is expected to intensify rainfall variability and heat extremes in the Horn of Africa, potentially shifting the boundaries of susceptibility zones over time. Nevertheless, the framework they have built is designed to be updated as new data becomes available, whether from improved satellite sensors, expanded ground monitoring, or revised land-use plans. For now, the study stands as a practical demonstration that even in cities where comprehensive environmental data is limited, careful spatial analysis combined with genuine community engagement can produce actionable guidance for building a more climate-resilient urban future.

Subject of Research: GIS-based multi-hazard susceptibility and nature-based solution suitability mapping for urban climate resilience in Dessie City, Ethiopia

Article Title: GIS-based multi-hazard susceptibility and nature-based solution suitability analysis for urban climate resilience in Dessie City, Ethiopia

Article References: Kebede, H. A., Ali, A. M., Eshetu, A. A., Moges, G., Desta, M. A., & Mekonen, A. A. (2026). GIS-based multi-hazard susceptibility and nature-based solution suitability analysis for urban climate resilience in Dessie City, Ethiopia. Discover Sustainability. https://doi.org/10.1007/s43621-026-04777-3

Image Credits: AI Generated

DOI: 10.1007/s43621-026-04777-3

Keywords: GIS, multi-hazard susceptibility, nature-based solutions, urban climate resilience, Dessie City, Ethiopia, flood risk, landslide susceptibility, urban heat, Analytic Hierarchy Process, spatial planning, community perception

The core insight is deceptively simple: not all landslides fail the same way, so they should not be predicted the same way. In red-bed formations, interbedded siltstone, mudstone, marl, sandstone, and conglomerate are generally fractured and weakly cemented, and two dominant failure modes emerge. Retrogressive landslides, which account for 67 percent of the 164 documented events between 2004 and 2018 in the Sanshui inventory, begin at the slope toe when rising porewater pressure reduces shear strength and generates uplift forces that diminish sliding resistance, then retreat upslope in a progressive chain. Translational landslides, making up the remaining 33 percent, develop polygonal shear surfaces when rainwater infiltrates cracks at the trailing edge of mudstone slopes, increasing hydrostatic pressure and sliding stress while reducing effective normal stress on the potential failure plane. These are fundamentally different hydromechanical laws, and the researchers argue that folding them into a single statistical entity is precisely why traditional susceptibility maps fail in lithologically complex basins.

The framework unfolds in three stages. First, the team built a geographic information system database from multi-source data, including a 30-meter digital elevation model derived from a locally surveyed 1:10,000 contour map, Landsat 8 imagery for vegetation indices, meteorological records, land-use rasters, hydrographic and road networks, and geological maps. Eleven conditioning factors were derived: elevation, slope, aspect, plan curvature, profile curvature, NDVI, annual precipitation, land use, distance to river, distance to road, and lithology. The choice to generate terrain derivatives from the local contour map rather than public DEM products such as SRTM or ASTER GDEM matters, because low-relief red-bed basins hide subtle geomorphological variations that coarser global datasets smooth away, and the locally generated DEM preserved terrain details more effectively.

Second, the researchers quantified how strongly each factor interval relates to landslide occurrence using two strategies. The classical frequency ratio method computes the ratio of landslide density within a factor subinterval to the overall landslide density across the study area, a workhorse of bivariate susceptibility analysis. The Bayesian probability method goes further, integrating uncertainty and variability into the quantification by combining the conditional probability of a factor subclass given a landslide event with the prior probability of landsliding and the marginal probability of the subclass area. These quantified factor values then feed two machine learning classifiers, random forest and logistic regression, producing four hybrid models per landslide type: FR-RF, FR-LR, BP-RF, and BP-LR. Correlation screening confirmed that no pair of the eleven factors exceeded a coefficient of 0.5 under the Bayesian treatment, eliminating multicollinearity concerns before model training.

The results were unambiguous. Bayesian-enhanced models consistently outperformed their frequency-ratio counterparts for both landslide types, with the Bayesian random forest achieving the highest accuracy. For translational landslides, the BP-RF model yielded frequency ratios of 2.761 in high susceptibility zones and 6.635 in very high zones, beating FR-RF, FR-LR, and BP-LR. For retrogressive landslides, BP-based models increased frequency ratios in high and very high susceptibility categories by 4.12 to 60.41 percent compared with other methods, and the BP-RF model achieved the lowest very-low susceptibility frequency ratio of 0.037 alongside the highest very-high ratio of 4.920. The spatial patterns also made geological sense: high and very high susceptibility zones concentrated in the hilly southern region, where steep terrain, an extensive highway network, and converging rivers collectively raise landslide propensity, while the flatter northern and central basins mapped as low risk.

The third stage is where the study makes its most novel technical contribution. Rather than simply averaging the two type-specific susceptibility maps, the team introduced a mixture probabilistic neural network, or MPNN, that fuses the multi-modal susceptibility indices while respecting spatial heterogeneity. The architecture combines a mixture probabilistic model, grounded in Bayesian model averaging, with a spiking neural network stream built on back propagation. The mixture component computes weights for each landslide-type member through iterative updates of a Dirichlet distribution parameter, while the neural stream estimates geographically weighted parameters using a spatial distance matrix that relates the retrogressive susceptibility index at every raster cell to each translational landslide reference point. Unlike conventional spiking neural networks, the design does not model temporal spike propagation; instead, spatial dependence is represented through probabilistic weighting of neighboring landslide samples. A sigmoid output layer normalizes final susceptibility values between zero and one for each cell.

Performance evaluation against a unified model trained on a combined landslide inventory revealed the scale of the gain. The MPNN confined high and very high susceptibility zones to 16.42 percent of the total area yet captured 81.32 percent of all landslides, whereas the unified model designated 34.77 percent of the area as high to very high risk while containing only 76.09 percent of events. In very high susceptibility zones, the MPNN achieved a frequency ratio of 7.956, nearly triple the unified model’s 2.681. The overall AUC of 90.2 percent against 66.7 percent confirms that explicitly accounting for divergent failure mechanisms, rather than blending them, is what unlocks the accuracy. The team reserved 30 percent of the inventory strictly for independent validation, keeping all model construction procedures on the training subset.

To open the black box, the researchers applied SHAP, or Shapley additive explanations, a method rooted in cooperative game theory that quantifies each feature’s contribution to predictions. The analysis identified lithology, annual precipitation, and road proximity as dominant controls. For translational landslides, annual precipitation, road proximity, and lithology correlated positively with susceptibility, while elevation showed a negative correlation. For retrogressive failures, annual precipitation, lithology, and elevation correlated positively, while NDVI, a proxy for vegetation cover, mitigated risk. A Sankey diagram analysis of lithostratigraphic groups showed that Groups 1, 7, 12, and 13 exhibit the highest susceptibility, characterized by interbedded Cretaceous sandstone-mudstone sequences and vertically jointed siltstone layers. Group 12, a sand-mud interbed, destabilizes through permeability contrasts between sandy and muddy layers, while Group 13, siltstone-shale, provides weak shear surfaces under rainfall infiltration. These lithologies form anisotropic permeability systems that promote preferential groundwater flow, a preparatory factor for slope failure, further accentuated by gently inclined dip slopes with pervasive discontinuities where transient pore pressure rises during heavy rainfall can reduce effective stress to critical levels.

The implications extend well beyond one district. Red-bed basins underlie vast swaths of southern China, where rapid urbanization, climate change, and extreme weather have intensified landslide risk, and similar interbedded soft-rock terrains appear globally, from weathered crystalline slopes in southern Italy to progressive failures in sensitive clays of eastern Canada and Scandinavia. The framework’s lithostratigraphy-informed, mechanism-aware pathway offers a template for regional risk assessment that distinguishes fundamental geological predispositions from transient triggers, and it reduces the false classification of densely built-up urban zones as highly susceptible, a persistent flaw of unified models that overestimate risk in developed areas.

The authors are candid about limitations. Extrapolation to regions with distinct geological or climatic settings remains uncertain because the model relies on localized lithostratigraphic features. The framework uses static environmental factors rather than dynamic triggers such as rainfall intensity or piezometric monitoring, so it does not yet support real-time early warning, and although Bayesian elements are incorporated, predictive uncertainty from parameter variability and model structure is not fully quantified. Validation relied on a random training-testing split, which may still allow spatial dependence among neighboring landslides to produce optimistic estimates; the team proposes spatial block cross-validation and temporally independent validation as remedies. Future work includes integrating real-time monitoring data, automating landslide type classification through deep learning-based semantic segmentation of satellite imagery, and building hybrid data assimilation frameworks using UAV-LiDAR derived scarps. Even with these caveats, the message is clear: when the ground fails in more than one way, the map should know it.

Subject of Research: Earth Science

Article Title: Lithostratigraphically-constrained landslide susceptibility modeling in red-bed basins using a bayesian probability-based mixture probabilistic neural network

Article References: Liu, Z., Yi, A., Huang, G., Li, Q., & Zheng, T. (2026). Lithostratigraphically-constrained landslide susceptibility modeling in red-bed basins using a bayesian probability-based mixture probabilistic neural network. Frontiers in Earth Science, 14, Article 1910278. https://doi.org/10.3389/feart.2026.1910278

Image Credits: AI Generated

DOI: 10.3389/feart.2026.1910278

Keywords: landslide susceptibility, red-bed basin, Bayesian probability, mixture probabilistic neural network, machine learning, Sanshui District, SHAP analysis, lithostratigraphy, Lithostratigraphically-constrained, landslide, susceptibility, modeling

The study’s lead finding is stark: roughly 19 percent of Nepal’s lowland areas fall into medium to very high flood susceptibility zones, threatening approximately 900,000 people and more than 3.4 million buildings. In the hillier terrain, the picture is equally sobering, with about 40 percent of the country susceptible to slope failure, endangering around 200,000 people and some 600,000 buildings. These are not abstract statistics in a nation where, over the past half century, disasters have killed an average of two people every day and affected more than 300 families annually. Landslides alone claim an average of 113 lives each year, while floods account for more than nine percent of disaster deaths and touch over sixty percent of affected families.

What sets this research apart from previous hazard assessments is its methodological machinery. Traditional susceptibility mapping in Nepal has relied heavily on multi-criteria analysis, a GIS technique in which experts assign subjective weights to variables such as slope, elevation and land cover. That approach struggles to capture the tangled, non-linear relationships that govern where water flows and where slopes fail. The ICIMOD team instead deployed a Random Forest model, an ensemble machine learning algorithm that builds hundreds of decision trees, each trained on random subsets of the data, and averages their predictions into a continuous probability score between zero and one. Hyperparameter tuning across 400 to 600 trees showed that 600 delivered the best performance, and the resulting models achieved strong discrimination, with an area under the receiver operating characteristic curve of 0.84 for floods and 0.85 for landslides.

The variable lists themselves tell a geomorphological story. For floods, accessibility and flow accumulation ranked as the most important predictors, followed by landform, land cover and slope, confirming that monsoon runoff concentrates in river corridors where human development presses closest to the water. For landslides, slope steepness and aspect dominated, with vegetation greenness, captured through the Normalized Difference Vegetation Index, coming in third. Validation against independent datasets added credibility: the flood model agreed 74 percent with an established hydrological inundation product, while the landslide model matched a manually digitized inventory of landslides triggered by the devastating 2015 Gorkha earthquake with 81 percent agreement.

The truly novel step came after the single-hazard maps were complete. The researchers fed the flood and landslide probability surfaces into an unsupervised k-means clustering algorithm, which partitioned every spatial unit of the country into three natural groupings without any expert-defined thresholds: low-hazard zones, flood zones and landslide zones. A subsequent spatial adjacency analysis, a GIS operation that identifies polygons sharing boundaries, then carved out a fourth category: interaction zones, where landslide-susceptible slopes physically touch flood-susceptible river corridors. Nationally, this data-driven classification assigned 81 percent of the territory to the low-hazard category, nine percent to flood zones, five percent to landslide-only zones and five percent to interaction zones.

Those interaction zones matter because of what they represent physically. In mountainous terrain, the boundary between a steep, fragile hillslope and an active river channel is not an arbitrary line on a map; it is the slope-toe interface where debris from slope failures enters the drainage network and can amplify downstream inundation. By applying the adjacency operator a second time, linking interaction zones to downstream flood-prone polygons, the team delineated what they call critical multi-hazard flow zones: roughly 7,588 square kilometres of connected corridors that funnel hazard material from unstable headwaters toward populated valleys. These corridors overlap 88 square kilometres of built-up land, about twelve percent of Nepal’s total urban footprint, and 1,722 square kilometres of cropland.

The authors are careful to note that these flow zones represent susceptibility-based spatial connectivity, a screening-level proxy, rather than physically simulated cascading processes. Full hydrodynamic modelling of landslide-dam formation, breach dynamics and downstream flood propagation was beyond the scope of the study. Yet the screening framework has already proven its relevance against real events. The catastrophic 2021 Melamchi disaster, in which intense upstream rainfall, landslides and debris flows cascaded into a flood that devastated downstream settlements and the Melamchi drinking water supply project serving Kathmandu, followed precisely the source-to-sink pathway the mapping is designed to flag. The July 2025 transboundary outburst flood from Nepal’s border with China, which swept away a critical trade bridge at Rasuwaghadi, underscores the stakes for infrastructure planning in similar corridors.

The study also exposes an uncomfortable institutional blind spot. In Nepal, flood hazards are managed by the Department of Hydrology and Meteorology while landslides fall under the Department of Mines and Geology, a siloed arrangement that treats physically coupled processes as administratively separate phenomena. Meanwhile, current national building regulations rely on uniform river-buffer setbacks, thirty metres from rivers, fifty metres from lakes, ten metres from canals, that bear no relationship to actual topographic susceptibility. Satellite imagery documenting the transformation of the Sardu Khola floodplain in Sunsari District, from sparse settlement and intact farmland in 2004 to dense housing and road networks by 2025, illustrates how rapidly exposure is accumulating in exactly the wrong places.

The researchers argue their maps can shift Nepal from reactive post-disaster relocation toward proactive risk screening. Low-hazard zones, often found on stable interfluves and elevated terraces even within high-risk basins like Melamchi, should be prioritized for settlement expansion and critical infrastructure. Interaction zones, by contrast, warrant strict regulatory controls and could even be designated no-development areas. The framework extends beyond government planning: banks and infrastructure investors, including those financing hydropower projects and strategic road corridors across the Himalaya, could use the flow zones as preliminary geospatial due diligence, embedding geological risk assessment into lending decisions before capital is committed to hazardous ground.

Limitations remain, and the authors flag them candidly. The models were trained on historical inventories that treat hazard patterns as stationary, even as climate change intensifies rainfall extremes and vegetation recovery alters slope stability, a tension visible in the divergent bias patterns between the flood and landslide validations. Detailed lithology, groundwater dynamics and real-time hydrometeorological data were unavailable, and spatial block cross-validation was not implemented. Exposure figures should therefore be read as indicative screening estimates rather than precise measurements. Still, as a replicable, computationally efficient template for data-scarce mountain nations, the framework aligns squarely with the Sendai Framework for Disaster Risk Reduction and the Sustainable Development Goals, offering a pathway for risk-informed land-use planning that could save lives before the next monsoon arrives.

Subject of Research: National-scale flood and landslide susceptibility and multi-hazard interaction mapping in Nepal using machine learning and GIS

Article Title: Flood and landslide susceptibility assessment and multi hazard interaction mapping using machine learning and GIS for sustainable settlement planning in Nepal

Article References: Thapa, N., Uddin, K., Thapa, R. B., & Udas, E. (2026). Flood and landslide susceptibility assessment and multi hazard interaction mapping using machine learning and GIS for sustainable settlement planning in Nepal. Discover Geoscience, 4(1), Article 287. https://doi.org/10.1007/s44288-026-00670-8

Image Credits: AI Generated

DOI: 10.1007/s44288-026-00670-8

Keywords: Nepal, flood susceptibility, landslide susceptibility, machine learning, Random Forest, k-means clustering, GIS, multi-hazard, spatial adjacency analysis, Google Earth Engine, land-use planning, disaster risk reduction

The research, led by Jianqi Zhuang of Sichuan University together with colleagues from institutions across China, set out to answer a question that has long frustrated hazard scientists: how do the rainfall changes projected for a warming world translate into changes in landslide susceptibility on the ground? Susceptibility mapping is a well-established technique, but most existing maps are static snapshots, built from historical records of where landslides have occurred and the terrain conditions that favored them. They say little about how those danger zones might expand, contract, or shift as precipitation patterns evolve. The team’s approach was to fuse two strands of data that rarely meet so thoroughly: a massive inventory of past landslides and a continuous rainfall record stretching from 1980 all the way to 2060, derived from the latest generation of global climate models.

The foundation of the study is its dataset, and it is a formidable one. The researchers assembled 8,373 documented historical landslide locations across the Qinling Mountains. To train machine learning models fairly, they paired each landslide point with an equal number of non-landslide samples drawn from stable terrain, producing a balanced dataset of nearly 17,000 labeled locations. This balance matters more than it might appear. When models are trained on datasets where landslides are vastly outnumbered by safe ground, they tend to underestimate danger, and a growing body of literature has highlighted how sampling strategy can make or break susceptibility predictions. By keeping the classes equal, the team gave their algorithms a level playing field on which to learn what distinguishes failing slopes from stable ones.

Onto each of those locations, the researchers overlaid fifteen conditioning factors: the environmental variables that govern whether a slope can fail. These included topographic attributes such as elevation and slope angle, distances to faults, rivers, and roads, vegetation cover, land use type, landform classification, and, critically, rainfall itself. The analysis of where historical landslides cluster is strikingly consistent. Slopes fail most often at elevations between 500 and 1,000 meters, on gradients of 15 to 30 degrees, within 250 meters of geological faults, within 2,500 meters of rivers, and within 250 meters of roads. In other words, landslides in the Qinling concentrate where tectonic fracturing has weakened the rock, where rivers undercut the toes of slopes, and where road construction has steepened and loaded hillsides, all in places where annual rainfall exceeds 650 millimeters and vegetation cover is sparse or the land has been converted to dryland farming or built development.

To quantify which of these factors matter most, the team first applied frequency ratio analysis, a statistical technique that compares the density of landslides within each class of a factor to the overall density across the landscape. A frequency ratio above one indicates that a class is over-represented among landslide sites and therefore more hazardous. The results pointed to three dominant controls: slope, rainfall, and landform. That triad is physically intuitive. Slope determines the gravitational shear stress acting on hillside material, rainfall controls the pore water pressure that reduces friction along potential failure surfaces, and landform encapsulates the broader geomorphological setting that concentrates or disperses both water and sediment. The fact that rainfall ranks alongside terrain as a first-order control is precisely what makes the study’s forward-looking component so consequential.

That component rests on the Coupled Model Intercomparison Project Phase 6, or CMIP6, the international ensemble of global climate models that underpins much of modern climate projection. The researchers constructed a continuous rainfall series for the Qinling region spanning 1980 to 2060 under the SSP2-4.5 scenario, one of the intermediate emissions pathways that assumes roughly current trajectories of greenhouse gas output continue through mid-century. Using projected future rainfall as an input, they could then re-evaluate susceptibility across the mountains for coming decades rather than for the present alone. The projections show annual rainfall rising over the region, and with it, an expansion of the zones classified as having high to very high landslide susceptibility. Areas that are marginal today, where slopes sit just below the rainfall and saturation thresholds needed for failure, are the most likely to cross into the danger zone as precipitation increases.

The modeling itself was a three-way contest between statistical and machine learning approaches, and the comparison is one of the study’s most instructive contributions. The team used logistic regression, the classical statistical baseline that models landslide probability as a weighted sum of factors; random forest, an ensemble method that aggregates hundreds of decision trees, each trained on random subsets of the data and factors; and a convolutional neural network, a deep learning architecture designed to recognize spatial patterns. Convolutional networks, famously successful in image recognition, excel at capturing local spatial context, effectively learning the texture of hazardous terrain rather than treating each pixel in isolation. That ability to encode the nonlinear, spatially correlated interactions among factors is exactly what landslide susceptibility demands, since a steep slope near a fault and a river is not merely the sum of three independent risks.

The performance metrics tell a clear story. Measured by the area under the receiver operating characteristic curve, or ROC-AUC, the convolutional neural network scored 0.913, compared with 0.865 for random forest and 0.826 for logistic regression. The same ranking held for the precision-recall curve, with AUC values of 0.906, 0.858, and 0.818 respectively, and for accuracy, recall, and F1-score, where the neural network achieved 0.881, 0.874, and 0.880. All three models performed well by the standards of the field, but the deep learning model was consistently superior, confirming that in a geologically and climatically transitional landscape like the Qinling, the added complexity of a convolutional architecture pays for itself. For hazard practitioners, the message is that the choice of model is not a technical footnote; it directly changes which slopes appear on the warning map.

Why does this matter beyond the mountains themselves? Landslides are among the deadliest of climate-driven hazards, responsible for thousands of fatalities worldwide each year, and they disproportionately strike rural communities and infrastructure corridors where warning systems are thin. The Qinling region is crossed by transport links, settlements, and agricultural land that press directly against its unstable mid-elevation slopes. The study’s identification of the specific combinations of terrain, proximity, land use, and rainfall that concentrate historical landslides gives planners a concrete template: which distances to faults and roads to treat as buffer zones, which elevation and slope bands to prioritize for monitoring, and which land use categories to scrutinize when approving new development. The finding that landslides cluster within 250 meters of roads is particularly pointed, since it implicates human engineering as a co-conspirator with climate in shaping future risk.

The broader significance of the work lies in its demonstration of a replicable workflow. By coupling a large, balanced landslide inventory with CMIP6 rainfall projections and benchmarking multiple learning algorithms, the researchers have produced a template that can be applied to other mountain belts facing similar pressures, from the Himalaya to the Alps, where recent studies have also documented intensifying landslide activity under changing climate. The authors note that their findings can inform land-use planning and climate adaptation in the Qinling, aiding efforts to reduce future landslide risks. As global temperatures climb and the hydrological cycle accelerates, the ground beneath some of the world’s most densely populated mountain regions is becoming less forgiving. Maps like these, built on deep learning and forward-looking climate data, may become the difference between communities that see the danger coming and those that do not.

Subject of Research: Machine learning-based landslide susceptibility mapping under future rainfall changes in the Qinling Mountains

Article Title: Landslide susceptibility assessment in the Qinling mountain driven by rainfall changes under future climate scenarios

Article References: Zhuang, J., Ma, Z., Di, B., Heng, S., Chang, L., Dou, J., Guo, X., & Peng, J. (2026). Landslide susceptibility assessment in the Qinling mountain driven by rainfall changes under future climate scenarios. Natural Hazards, 122(21), Article 670. https://doi.org/10.1007/s11069-026-08437-7

Image Credits: AI Generated

DOI: 10.1007/s11069-026-08437-7

Keywords: landslide susceptibility, Qinling Mountains, climate change, CMIP6, rainfall, convolutional neural network, random forest, logistic regression, machine learning, natural hazards, SSP2-4.5, disaster risk

The study, published in the journal Discover Geoscience, was conducted by Fumani Nkanyane and Fhatuwani Sengani, who set out to answer a deceptively simple question: which modelling technique best captures landslide risk in terrain where geology, rather than climate or topography alone, dominates slope behaviour? To find out, they pitted three fundamentally different approaches against one another using the same data, the same landslide inventory and the same validation framework. The contenders were the Analytical Hierarchy Process, or AHP, a structured expert-judgment method; Fuzzy Logic, an uncertainty-based technique that embraces gradual transitions rather than sharp thresholds; and XGBoost, an ensemble machine-learning algorithm that has swept through geospatial prediction problems in recent years.

The stakes are far from academic. Landslides rank among the most destructive geological hazards in mountainous regions worldwide, and the Fetakgomo-Tubatse Municipality presents an especially challenging case. The area sits within the Eastern Limb of the Bushveld Igneous Complex, a vast layered intrusion of mafic and ultramafic rocks interbedded with sandstones and quartzites of the Transvaal Supergroup. Major structures such as the Steelpoort Fault and the Thabazimbi-Murchison Lineament cut through the landscape, introducing discontinuities that act as ready-made failure planes. Elevations range from roughly 490 to 2,331 metres above sea level, and the semi-arid climate concentrates its modest 400 to 600 millimetres of annual rainfall into intense summer storms that can rapidly raise pore-water pressures in weakened slopes.

To build their models, the researchers first compiled an inventory of 100 documented landslide events, identified through field surveys, photographic documentation and interpretation of high-resolution satellite imagery from Google Earth and Landsat 8. The inventory captured mostly shallow translational failures and rockfalls affecting engineered road cuts and natural slopes. Each landslide was georeferenced as a point at the crown of the failure, and the dataset was randomly split into 70 percent for model training and 30 percent for independent validation. The team then assembled ten conditioning factors known to influence slope stability: slope angle, aspect, elevation, lithology, soil type, rainfall, land use and land cover, and proximity to roads, rivers and geological lineaments. All layers were standardised to a common 30-metre grid resolution derived from the Shuttle Radar Topography Mission digital elevation model.

Each modelling philosophy handled these inputs in its own distinctive way. The AHP method, introduced by Thomas Saaty in 1980, relies on expert-derived pairwise comparisons to weight each factor’s relative importance, and the team’s comparison matrix achieved a consistency ratio of 0.074, comfortably within the accepted threshold of 0.10. Fuzzy Logic took a different route, assigning each pixel a membership value between 0 and 1 that expresses its degree of belonging to the fuzzy set of landslide-susceptible ground. Rather than forcing terrain into rigid classes, membership functions allowed susceptibility to fade gradually across the landscape, and the layers were combined using a fuzzy gamma operator set at 0.90, a compromise between the strict intersection and liberal union of fuzzy sets. XGBoost, by contrast, learned directly from the data, building a sequence of decision trees in which each new tree corrected the errors of its predecessors, with regularisation terms built in to prevent overfitting.

When the maps were validated against the held-out landslide data, the results were striking. The Fuzzy Logic model achieved the highest predictive performance, with a Receiver Operating Characteristic Area Under the Curve, or ROC-AUC, of 0.769. The AHP model followed closely at 0.757, while XGBoost trailed at 0.739. A confusion matrix analysis reinforced the ranking: Fuzzy Logic posted the highest overall accuracy at 0.7708, followed by AHP at 0.7292 and XGBoost at 0.6857. Notably, both AHP and Fuzzy Logic significantly outperformed the no-information rate, meaning their classifications were far better than random guessing, whereas XGBoost’s accuracy merely matched the majority-class baseline. McNemar’s tests confirmed that the differences between predicted and observed classifications were statistically significant for all three models.

Why did the old-school uncertainty approach beat the machine? The authors attribute Fuzzy Logic’s edge to its ability to represent the continuous, progressive nature of slope instability. In mountainous terrain, landslides rarely obey discrete thresholds; instead, they emerge from gradual interactions between slope geometry, lithological weathering, structural discontinuities, rainfall infiltration and human excavation. Fuzzy membership functions capture those soft transitions, whereas conventional weighted overlays impose hard class boundaries. XGBoost’s comparatively modest showing likely reflects the reality of data-scarce environments: machine-learning algorithms typically need large, representative landslide inventories to generalise well, and a 100-event inventory in a geologically heterogeneous setting offers limited training material. The finding challenges the widespread assumption that artificial intelligence automatically trumps expert-informed methods in every hazard-mapping context.

The maps themselves tell a coherent geological story. Under the AHP model, 40.48 percent of the municipality fell into the low susceptibility class and 56.76 percent into the moderate class, with just under 3 percent classified as high or very high. The Fuzzy Logic model redistributed those figures dramatically, placing 82 percent of the area in the very low class while elevating 12.4 percent to high susceptibility, concentrated in the north-eastern region along the R37 and its tributaries. XGBoost identified roughly 13 percent of the district as high or very high susceptibility. Across all three models, the danger zones clustered in the same places: steep road-cut sections where excavations have disturbed natural slope equilibrium, weathered quartzite units with reduced shear strength, and slopes intersected by joints, faults and bedding planes that act as potential failure surfaces. Field observations confirmed rockfalls and combined sliding-toppling failures in exactly these locations.

The practical implications extend well beyond Limpopo Province. The high-susceptibility zones identified along the R37 and R555 give provincial engineers a spatially explicit priority list for slope monitoring, drainage management and road maintenance, and the study’s framework is explicitly designed to be transferable to other mountainous regions where layered intrusions, fault systems and infrastructure development intersect. The authors acknowledge limitations that future work should address: the 30-metre DEM may miss small-scale terrain features, event-based rainfall intensity was not modelled, expert weighting in AHP inevitably introduces some subjectivity, and a larger landslide inventory would strengthen any machine-learning application. They suggest that integrating geotechnical parameters such as shear strength and cohesion, along with higher-resolution topographic data and climate-change projections, could further sharpen predictive reliability.

Perhaps the study’s most valuable contribution is its methodological honesty. By testing three paradigms head-to-head in the same terrain, Nkanyane and Sengani have demonstrated that model choice should be guided by geological context and data availability rather than by fashion. In structurally complex, data-limited environments, uncertainty-based modelling and well-calibrated expert judgment remain formidable tools, and the best hazard assessments may come from combining them with machine learning rather than replacing them. As climate variability intensifies rainfall extremes and infrastructure expansion pushes roads ever deeper into unstable mountainsides, that lesson resonates far beyond the Bushveld.

Subject of Research: GIS-based comparative landslide susceptibility mapping using AHP, Fuzzy Logic and XGBoost in the Fetakgomo-Tubatse Municipality, South Africa

Article Title: Application of GIS techniques for landslide susceptibility mapping in the mountainous regions of Fetakgomo-Tubatse Municipality

Article References: Nkanyane, F., & Sengani, F. (2026). Application of GIS techniques for landslide susceptibility mapping in the mountainous regions of Fetakgomo-Tubatse Municipality. Discover Geoscience, 4(1), Article 297. https://doi.org/10.1007/s44288-026-00649-5

Image Credits: AI Generated

DOI: 10.1007/s44288-026-00649-5

Keywords: landslide susceptibility, GIS, Fuzzy Logic, AHP, XGBoost, machine learning, Bushveld Igneous Complex, South Africa, slope stability, road-cut slopes, ROC-AUC, disaster risk reduction

Shallow landslides, including debris slides, debris avalanches and debris flows, involve the failure of soft surficial material at rooting depth. They are overwhelmingly triggered by rainfall, which raises porewater pressure in the soil until the shear strength of the slope is overcome. Under climate change scenarios, heavy precipitation over mid-latitude and tropical mountainous regions is expected to intensify, making reliable predictions of where these failures will occur ever more urgent. Yet the field of landslide susceptibility mapping has a dirty secret: many recent studies report near-perfect prediction accuracies that the Norwegian team argues are almost certainly artefacts of overfitting rather than genuine predictive skill.

The root of the problem lies in space itself. Landslides tend to occur in clusters, and the environmental variables that influence them, from elevation to precipitation, vary smoothly across the landscape. When researchers randomly split their data into training and test sets, as standard machine learning practice dictates, points in the test set sit tantalisingly close to points in the training set, sharing nearly identical conditions. The model appears to predict brilliantly while merely memorising local patterns. The team quantified this effect with variogram analysis, showing that under random cross-validation, the residual spatial correlation of key predictors such as relative precipitation approached unity at the distances separating training and test observations, meaning a held-out point was essentially indistinguishable from its nearest training neighbour.

To confront this challenge head-on, the researchers built an exceptionally clean dataset. Rather than relying on Norway’s crowd-sourced landslide database, which misses many events and clusters along roads, they systematically mapped landslide scars using satellite-derived vegetation change indices, validated with high-resolution imagery, orthophotos and fieldwork. This effort detected 648 landslides, a 246 percent increase over the crowd-sourced record, and reduced spatial bias towards roads by 35 percent. After excluding debris floods, which follow a different release mechanism, the final inventory contained 571 landslide starting points, each characterised by twenty explanatory variables spanning topography, geology, hydrology and forest structure.

The team then ran thirty-two gradient boosted decision tree models, systematically varying three dimensions: whether hyperparameters were tuned with nested or simple cross-validation, whether the model included an explicit spatial random effect through the Gaussian Process Boosting framework, and which of four cross-validation schemes was used, ranging from fully random splits to spatially clustered folds and large and small spatial blocks. The results were sobering. Models with random cross-validation posted the highest performance metrics, with test AUC values reaching 0.90, but left significant spatial autocorrelation in their residuals, a statistical fingerprint of over-optimism. Only spatial models combined with block cross-validation fully eliminated residual autocorrelation, and they paid for this honesty with lower predictive power, achieving a test AUC of 0.77 at best.

Perhaps the most unsettling finding concerned variable importance. The top predictors identified by spatial models differed markedly from those of non-spatial models, and rankings shifted dramatically across cross-validation schemes. The researchers attribute this instability largely to spatial confounding, a phenomenon in which predictors correlate with unmodelled spatial structure and absorb its signal. In non-spatial models, variables with strong spatial footprints, such as total precipitation and bedrock weatherability, dominated, likely acting as proxies for spatial pattern rather than true causal factors. When an explicit spatial term was added, those variables lost much of their apparent importance. No single model configuration emerged as optimal, a candid admission that underscores how much work remains in this field.

Yet across the ensemble of models, a coherent physical story did emerge. Landslides were most likely on south-facing slopes, at elevations below roughly 480 metres, and where water inputs were greatest, whether from intense three-hour rainfall bursts, high flow accumulation, or elevated pre-storm groundwater. The team interprets elevation and aspect as indicators of soil conditions that existing datasets fail to capture directly: lower elevations and south-facing slopes in Norway’s climate experience more weathering, producing thicker and more stratified soils in which impermeable layers can channel water and build the critical pore pressures that trigger failure. Notably, bedrock weatherability, a novel reclassification of Norway’s national bedrock map into five classes, proved influential, with easily weathered shales and phyllites increasing landslide probability while resistant gneisses and granites reduced it, explaining why some drenched regions escaped unscathed.

The forest analysis delivered the study’s most surprising result. In models restricted to forested terrain, tree type consistently ranked among the top predictors, with deciduous stands showing markedly higher landslide susceptibility than spruce or pine forests. This runs counter to simple mechanical expectations, since spruce roots provide substantial reinforcement. The authors suggest the pattern reflects where deciduous forest, often birch-dominated, typically grows in Norway: steep, thin-soiled terrain near the tree line that is unsuitable for production forestry. The findings also revealed that forests suppress landslides on gentler slopes, with landslide probability in forest rising only above slope angles of about 30 degrees, compared with a much wider range in open land. Canopy cover above 75 percent further reduced susceptibility, supporting the role of forests as nature-based solutions for landslide mitigation.

The implications reach well beyond Norway. The team recommends that future machine learning studies of landslide susceptibility test for spatial autocorrelation in both response and predictor data, explore explicitly modelled spatial structure, employ multiple spatial cross-validation techniques, and report methods in full detail. Performance metrics and predictor rankings from studies that ignore spatial structure, they caution, should be interpreted with considerable scepticism, particularly when models are meant to generalise to new locations. As extreme rainfall events grow more frequent in a warming climate, the difference between an honest model and an overfitting one may determine whether hazard maps guide evacuation in time. The Hans storm, which forced the evacuation of around 4,600 people with no loss of life, showed what good early warning can achieve; this study shows how the science of predicting where the ground will fail can be made considerably more trustworthy.

Subject of Research: Spatial machine learning modelling of explanatory factors for rainfall-triggered shallow landslide release in southern Norway

Article Title: Spatial machine learning modelling reveals that soil indicators and tree type best explain shallow landslide release

Article References: Rüther, D. C., Haualand, K. F., Peeters, I. L. J., & Gillespie, M. A. K. (2026). Spatial machine learning modelling reveals that soil indicators and tree type best explain shallow landslide release. Natural Hazards and Earth System Sciences, 26(10), 4675-4721. https://doi.org/10.5194/nhess-26-4675-2026

Image Credits: AI Generated

DOI: 10.5194/nhess-26-4675-2026

Keywords: shallow landslides, machine learning, spatial cross-validation, landslide susceptibility, storm Hans, Norway, soil thickness, bedrock weatherability, forest cover, tree type, gradient boosting, spatial autocorrelation

The stakes could hardly be higher. Landslides rank among the world’s most destructive natural hazards, and a global analysis of fatal non-seismic landslide events recorded 4,862 such disasters between January 2004 and December 2016, killing 55,997 people, with Asia the hardest-hit continent. India carries a disproportionate share of the burden: roughly 0.42 million square kilometres, or 12.6 percent of the country’s land surface excluding snow-covered terrain, is susceptible to landslide hazard. In the Himalayan region alone, landslides claimed more than 5,000 lives between 2007 and 2015, and the June 2013 extreme rainfall disaster in Uttarakhand destroyed over 250 villages and killed approximately 6,000 people. The 2021 Chamoli disaster further devastated hydropower infrastructure, bridges, and settlements in the Rishiganga and Dhauliganga valleys.

The research team, led by Sidra of Jamia Millia Islamia in New Delhi, built their analysis on an unusually detailed foundation: a landslide inventory of 852 manually digitised landslide locations, compiled from Google Earth satellite imagery spanning 2012 to 2022. Each landslide was mapped as a polygon and then converted to presence points, which were split into a training set of 627 points and an independent testing set of 225. To give the machine learning classifiers the negative examples they need, the team generated 852 non-landslide points in stable areas, carefully avoiding a 150-metre buffer around known landslides to reduce spatial autocorrelation and enforcing a minimum spacing of 50 metres to prevent clustering.

Fourteen conditioning factors fed the models, each chosen for its documented influence on slope instability and its relevance to the geology, hydrology, and human geography of the corridor. Topographic variables—elevation, slope, aspect, and curvature—were derived from the ALOS PALSAR digital elevation model at 30-metre resolution. Hydrological factors included drainage density and distance to streams, while geological inputs comprised lithology, geology, distance to faults, and distance to lineaments drawn from Geological Survey of India data at 1:50,000 scale. Environmental and anthropogenic layers included land use and land cover, the Normalized Difference Vegetation Index derived from Sentinel-2 imagery, distance to roads from OpenStreetMap, and rainfall data from the India Meteorological Department.

The Frequency Ratio model, a classical bivariate statistical technique, quantified the association between each factor class and landslide occurrence by dividing the percentage of landslide pixels in a class by the percentage of the study area that class occupies. Values above one signal elevated susceptibility. The results were telling: the steepest slopes, between 45 and 78.18 degrees, showed the highest frequency ratio of 1.36, while barren land registered a striking 3.76 and built-up areas 2.64. Bare soil and rock with sparse vegetation scored 3.62, and areas within 486 metres of streams scored 1.56. Intriguingly, susceptibility was highest at the lowest elevations, between 815 and 1,130 metres, likely reflecting the concentration of road construction, settlements, and land-use modification at these altitudes.

Where the statistical model treats each factor in isolation, the machine learning approaches capture the nonlinear interplay that governs real slope failures. Random Forest, an ensemble method that grows hundreds of decision trees on bootstrap samples and aggregates their votes, was trained with 500 trees using the scikit-learn library. XGBoost, a gradient boosting algorithm that sequentially adds trees to correct the residuals of previous ones while penalising model complexity through a regularised objective function, was trained with 400 trees at a maximum depth of five and a learning rate of 0.05. Both models received the same balanced presence-absence dataset.

Performance was evaluated using the area under the receiver operating characteristic curve, a standard measure of a model’s ability to discriminate between landslide and non-landslide locations, alongside success rate curves and a rigorous spatial cross-validation scheme. The GroupKFold algorithm partitioned the data into spatial blocks so that all samples within a block went entirely to either training or testing, minimising spatial leakage. Random Forest achieved the highest ROC-AUC of 0.880, with a mean spatial cross-validation score of 0.888 plus or minus 0.021. XGBoost followed closely at 0.867, with a spatial mean of 0.876 plus or minus 0.026. The Frequency Ratio model trailed at 0.761, with a spatial mean of 0.805 plus or minus 0.025.

The spatial concentration of predicted risk was equally revealing. In the Random Forest map, the very high susceptibility class covered just 10.4 percent of the corridor yet contained 66.8 percent of all mapped landslides, while the very low class spanned 26.6 percent of the area with a negligible 0.3 percent of events. XGBoost performed similarly, concentrating 65.3 percent of landslides within 13.4 percent of the area. The success rate curves for both machine learning models rose steeply at the outset, indicating sharp discrimination between stable and unstable ground, whereas the Frequency Ratio curve climbed gradually, reflecting its weaker ability to delineate high-risk zones.

Feature importance analysis from both machine learning models converged on a consistent story. Land use and land cover emerged as the single most influential factor in Random Forest, contributing 0.1608 of the total importance, followed by elevation, aspect, NDVI, and slope—together accounting for roughly 52 percent of model importance. XGBoost likewise ranked land use and land cover first, with elevation and distance to streams next. The prominence of road proximity, which ranked sixth in Random Forest and carried weight comparable to slope itself, underscores how slope cutting, excavation, and drainage alteration during highway construction destabilise the very corridors that connect Himalayan communities.

The authors acknowledge limitations, including a temporal mismatch between the landslide inventory compiled from 2022 imagery and the 2024 land cover and vegetation datasets, and the absence of field verification of the susceptibility zones. Even so, the maps produced offer highway planners, engineers, and disaster managers a practical tool for prioritising slope stabilisation, guiding development, and targeting monitoring along one of India’s most critical and most fragile mountain routes. As climate extremes intensify and infrastructure expands across the Himalaya, the study makes a compelling case that data-driven machine learning, validated in space as well as in random splits, should become the default instrument for anticipating where the mountains will give way.

Subject of Research: Comparative landslide susceptibility mapping using Frequency Ratio, Random Forest, and XGBoost models along the NH-7 corridor in Chamoli district, Uttarakhand, India

Article Title: Landslide susceptibility mapping using Frequency Ratio, Random Forest, and XGBoost Models along NH-7 between Nandprayag and Vishnuprayag

Article References: Sidra, Dwivedi, D., Faruque, U., Raina, S. I., Sundaram, Siddiqui, L., Prakash, S., & Siddiqui, M. A. (2026). Landslide susceptibility mapping using Frequency Ratio, Random Forest, and XGBoost Models along NH-7 between Nandprayag and Vishnuprayag. Discover Geoscience, 4(1), Article 312. https://doi.org/10.1007/s44288-026-00683-3

Image Credits: AI Generated

DOI: 10.1007/s44288-026-00683-3

Keywords: landslide susceptibility, Random Forest, XGBoost, Frequency Ratio, machine learning, Himalaya, Uttarakhand, NH-7, Chamoli, GIS, remote sensing, disaster risk management

The research team, led by Kiker Singh of Kurukshetra University with colleagues from Banaras Hindu University, Netaji Subhash University of Technology and NIIT University, focused on a section of the Lesser Himalaya that is tectonically restless and structurally fragmented. Three major regional thrust faults, the Giri, Chail and Juthogh thrusts, cut across the lower, middle and upper parts of the study area, fracturing the rock and priming slopes for failure. Elevations along the corridor range from roughly 1,425 to 1,700 metres above sea level, and the entire region sits within India’s seismic zone IV, meaning strong earthquake shaking is a realistic possibility. Add intense monsoon rainfall and aggressive road widening, and the ingredients for disaster are all present.

The stakes are not abstract. India accounts for nearly fifteen percent of all reported landslide occurrences worldwide and about thirty-one percent of associated fatalities, and almost 0.31 million square kilometres of landslide-prone territory lies within the Himalayan belt. Globally, landslides cause billions of dollars in economic losses and thousands of deaths each year, and roughly 50,000 people are estimated to have died from rock slope failures during the twentieth century alone. Along the Solan-Shimla stretch, road widening has repeatedly disturbed the natural equilibrium of slopes by removing the toe of hillsides, and since that work began, two lanes of the highway have been completely or partially blocked by landsliding between Solan and Kaithlighat.

To build their map, the researchers started with a landslide inventory: forty-eight landslide locations recorded along the highway using handheld GPS, cross-checked against high-resolution imagery in Google Earth and verified during extensive field campaigns. Each landslide was delineated as a polygon and then converted to point data within a geographic information system. This inventory served as the ground truth against which the model’s predictions could be tested, a step that many older susceptibility studies skipped and one that lends the new work considerable credibility.

The heart of the method is the Analytic Hierarchy Process, a multi-criteria decision-making framework introduced by mathematician Thomas Saaty. AHP asks experts to compare every causative factor against every other in pairs, assigning relative importance on a scale from one to nine. The resulting comparison matrix yields weights through eigenvector analysis, and a consistency check ensures the judgements are internally coherent. In this study the consistency ratio came out at 0.079, comfortably below the 0.10 threshold that signals unacceptable inconsistency. The team then combined the weighted factor layers in ArcGIS to produce a continuous susceptibility index for every patch of terrain along the corridor.

What distinguishes this study from earlier work in the same region is the breadth of factors considered. Previous research had examined slope, curvature, relative relief, aspect, geology, lithology, fault density, distance to road and drainage density. The new analysis adds five previously unexplored variables: the Topographic Wetness Index, which quantifies where water accumulates in soil; the Stream Power Index, which captures the erosive force of flowing water; land use and land cover; geomorphology; and lineament density, a measure of weakness zones in the rock. All terrain-derived layers were generated from a 30-metre Shuttle Radar Topography Mission digital elevation model, while geological and geomorphological data came from the Geological Survey of India’s Bhukosh portal.

The results are striking in their clarity. Slope emerged as the single most influential factor, receiving a weight of about 30.56 percent, with the densest landslide activity concentrated on slopes between 30 and 60 degrees. Distance to road ranked second at roughly 18.68 percent, confirming what engineers have long suspected: cutting into hillsides to widen the highway is a primary trigger of instability. Geology followed at about 9.5 percent, with the Mahi-Krol A and Chhaosa formations, composed of weathered limestone, shale and siltstone, accounting for a third and a quarter of all mapped landslides respectively. Concave slopes, which trap water and raise pore pressure, hosted 62.5 percent of the events, and highly dissected structural hills experienced nearly 73 percent of them. Aspect mattered too: northeast-facing slopes alone recorded about 29 percent of the landslides, likely because uneven solar exposure alters soil moisture dynamics.

When the weighted layers were merged and classified into five categories using natural breaks, the map revealed that 12.83 percent of the study area falls into the very high susceptibility class and a further 21.98 percent into the high class. Moderate susceptibility covers 26.58 percent, while low and very low zones account for 24.02 and 14.57 percent respectively. The validation statistics are what make the map genuinely useful rather than merely illustrative. The model achieved a ROC-AUC of 0.835, placing it in the ‘very good’ range, and a Precision-Recall AUC of 0.943, which is particularly meaningful because precision-recall analysis is less prone to overestimating performance when landslide events are rare compared with stable ground. An F1-score of 0.923 at an optimal threshold of 2.664 confirmed strong classification reliability.

Spatial validation sealed the case. Of the forty-eight mapped landslides, 72.92 percent occurred in the very high susceptibility zone and 25 percent in the high zone, with just one event, or 2.08 percent, in the moderate class and none at all in the low or very low categories. Normalized landslide density climbed steadily with susceptibility class, peaking at 0.821 in the very high zone. In practical terms, the model rarely cries wolf and almost never misses a real hazard, which is precisely the behaviour a planning authority needs when deciding where to install slope protection, drainage works or early warning systems.

The authors are candid about the limitations. An inventory of forty-eight events is modest, expert-driven weighting in AHP inevitably carries subjectivity, and the linear corridor format restricted any detailed rainfall threshold analysis. Future work, they note, will integrate rainfall data to build event-triggered models and apply machine learning approaches with expanded inventories to sharpen predictions further. Even so, the map already offers immediate value: engineers can prioritise slope stabilisation and cut-slope reinforcement along the highest-risk sections, adopt bio-engineering measures such as deep-rooted vegetation and grass turfing in moderate zones, and manage surface and subsurface drainage to keep water out of slope material. For a highway that thousands of people depend on every day, knowing precisely where the mountainside is most likely to move may prove as valuable as any retaining wall.

Subject of Research: Landslide susceptibility mapping along the Solan-Shimla National Highway-5 corridor in Himachal Pradesh, India, using GIS and the analytic hierarchy process

Article Title: Landslide susceptibility mapping along the Solan-Shimla highway using the analytic hierarchy process, Himachal Pradesh, India

Article References: Singh, K., Kumar, S., Khan, I., Kumar, H., & Rawat, A. (2026). Landslide susceptibility mapping along the Solan-Shimla highway using the analytic hierarchy process, Himachal Pradesh, India. Discover Geoscience, 4(1), Article 327. https://doi.org/10.1007/s44288-026-00694-0

Image Credits: AI Generated

DOI: 10.1007/s44288-026-00694-0

Keywords: landslide susceptibility, analytic hierarchy process, Himachal Pradesh, National Highway-5, Himalaya, GIS, remote sensing, Solan, Shimla, slope stability, ROC-AUC, geohazard mapping