TerraMosaic Daily Digest: Mar 24, 2026
Daily Summary
This March 24, 2026 digest distills 30 selected papers from 873 analyzed records. Today's literature converges on a common scientific question: how hidden system state becomes legible before failure. Landslide and rock-avalanche papers show that instability is organized by inherited structure, freeze-thaw forcing, and fragmentation-controlled mobility rather than by static terrain descriptors alone. In parallel, embankment and tunnel studies recast underground infrastructure as an evolving geosystem whose safety depends on heterogeneous stress transfer, seepage organization, and recovery after damage.
A second shift is operational. Flood, drought, tsunami, wildfire, and earthquake-warning studies seek not merely to detect hazards but to estimate consequences in forms that can guide action: impact forecasts, displacement hotspots, flood depth, building damage, and route-level evacuation performance. Across the set, the strongest methods remain physically interpretable. Thresholds, mechanisms, and uncertainty structures are preserved, allowing monitoring and machine learning systems to function as scientific instruments rather than opaque predictors.
Key Trends
The strongest March 24 studies move geohazard science from descriptive mapping toward state diagnosis, uncertainty-aware prognosis, and decision-ready inference.
- Slope instability is being treated as path-dependent: DSGSD activity, freeze-thaw forcing, susceptibility scale effects, and fragmentation-flow coupling are all quantified as history-dependent controls on failure.
- Underground geotechnics is becoming probabilistic: Embankments and tunnels are assessed through spatial variability, seepage uncertainty, resilience, and progressive failure rather than single deterministic safety margins.
- Monitoring now targets latent failure state: AE-DIC coupling, 3D tunnel scanning, scour sensing, flood imagery, and wildfire localization are used to infer instability variables that are not directly observable.
- Early warning is shifting toward impact anticipation: The best flood, drought, tsunami, and earthquake-warning papers focus on consequences, communication, and actionable lead time rather than on hazard indication alone.
- AI methods matter most when physically legible: The most convincing data-driven studies retain thresholds, uncertainty, or mechanistic attribution instead of offering generic black-box prediction.
Selected Papers
This digest features 30 selected papers from 873 papers analyzed.
1. Impact of freeze-thaw on landslide activity under diverse topographies using geospatial data mining: Insights from Qilian Permafrost Region, Tibetan Plateau
Core Problem: The quantitative role of freeze-thaw forcing, and its interaction with topography, remains insufficiently resolved in permafrost-region landslide activity.
Key Innovation: Geospatial data mining isolates freeze-thaw as a measurable secondary trigger, identifies a critical annual cycle threshold, and shows how its effect amplifies with slope gradient and elevation.
2. Multidisciplinary approach to revisit the state of activity of a deep‐seated gravitational slope deformation in the frame of the Quaternary geomorphological evolution of the Central Apennines (Italy)
Core Problem: Determining whether a deep-seated gravitational slope deformation remains active is difficult when present-day deformation is entangled with long Quaternary valley-slope evolution.
Key Innovation: The study reconstructs the geomorphological framework of a Central Apennine DSGSD with a multidisciplinary dataset to reassess its present state of activity in a morphoevolutionary context.
3. Numerical simulation of rock avalanche dynamics using a coupled fragmentation-flow model: insights from the Liangshui Village landslide
Core Problem: Existing numerical approaches struggle to represent the full transition from initial rock failure to fragmentation-controlled granular flow in long-runout landslides.
Key Innovation: A coupled fragmentation-flow framework integrates damage mechanics with field-derived structural planes to simulate progressive failure, dynamic fragmentation, and runout in the 2024 Liangshui Village landslide.
4. Global Stability and Failure Performance of Geosynthetic-Reinforced Column-Supported Embankments
Core Problem: The progressive global failure of geosynthetic-reinforced column-supported embankments is hard to resolve because columns, reinforcement, and soft foundation interact nonlinearly.
Key Innovation: Centrifuge tests and numerical simulations reveal how soil arching degrades, columns bend progressively, and added basal reinforcement improves load transfer and system stability.
5. Probabilistic assessment on seismic resilience of cross-fault tunnels via Bayesian updating
Core Problem: Fragility analyses alone do not describe how cross-fault tunnels recover from coupled fault-rupture and shaking damage.
Key Innovation: The paper builds a Bayesian-updated framework that connects fragility, loss, recovery time, and resilience grades for probabilistic seismic performance assessment.
6. Quantifying the impact of stratigraphic uncertainty on tunnel seepage field based on a 3D geostatistical interpolation model
Core Problem: Deterministic tunnel seepage analyses overlook how stratigraphic uncertainty can localize drainage amplification and produce exceedance risk in mountainous settings.
Key Innovation: A transition-probability geostatistical model, validated with multi-source field data, quantifies uncertainty-driven seepage amplification and seasonal drainage reversals along tunnel sections.
7. Scale discrepancies in landslide susceptibility assessment – a comprehensive analysis in Southwestern China
Core Problem: Large-scale landslide susceptibility maps can change substantially with analysis scale, weakening their interpretability for planning and risk reduction.
Key Innovation: The paper systematically diagnoses how scale reshapes susceptibility outcomes in Southwestern China, clarifying where and why model behaviour diverges across mapping resolutions.
8. Shield tunnel reliability assessment considering spatial variability: a heterogeneous ensemble surrogate modeling framework with physics-informed interpretability
Core Problem: Geological spatial variability introduces low-probability but high-consequence tunnel response modes that simplified design checks do not capture.
Key Innovation: An ensemble surrogate model with physics-informed interpretation and large-scale Monte Carlo simulation resolves tail-sensitive failure behaviour under heterogeneous ground conditions.
9. Air escape and disintegration behaviors of granite residual soil
Core Problem: Granite residual soil disintegration is closely tied to rainfall-triggered geohazards, but the role of pore-air escape in accelerating breakdown has remained unclear.
Key Innovation: An improved disintegration apparatus quantifies air escape and particle loss simultaneously, revealing distinct disintegration modes and the microscale mechanisms by which escaping air accelerates failure.
10. Comparing flood forecasting and early warning systems in northwestern Europe
Core Problem: After the 2021 flood disaster, the practical gap between flood forecasts, warnings, and early action across northwestern Europe's transboundary basins remained unresolved.
Key Innovation: A comparative operational assessment identifies post-2021 improvements and pinpoints persistent bottlenecks, especially the slow adoption of impact-based forecasting and harmonized warning practice.
11. Inversion analysis of service performance of shield tunnel: Driven by 3D laser scanning data and analytical models
Core Problem: Tunnel inspection now produces dense geometric data, but these measurements are difficult to translate into structural performance diagnoses.
Key Innovation: The study fuses 3D laser-scanning deformation fields with analytical tunnel models and geological constraints to infer service-performance state more reliably.
12. Longitudinal soil arch evolution in cobble tunnels using shallow tunnelling method: Experimental and DEM insights
Core Problem: Stress transfer and arching evolution in heterogeneous cobble ground remain insufficiently understood during shallow tunnelling, despite their control on deformation risk.
Key Innovation: Experiments and DEM reveal a shift from load receiving to load transferring during excavation and show how cover depth governs longitudinal soil-arch continuity and stability.
13. Toward early warning of drought impacts: a framework for predicting drought impacts in the UK
Core Problem: Operational drought warning still struggles to predict impacts across regions, sectors, and lead times with a generalized framework.
Key Innovation: Random-forest impact forecasting identifies long-accumulation drought indices and deep soil moisture as key predictors and delivers spatially explicit drought-impact forecasts.
14. Adjacent ultra-long deep excavation induced deformation of metro station–tunnel system: mechanism and intelligent prediction
Core Problem: System-level deformation of interconnected station-tunnel infrastructure under adjacent ultra-long excavation is not well captured by section-based analyses.
Key Innovation: The study resolves deformation mechanisms across the coupled metro system and develops an intelligent prediction framework for differentiated control of excavation effects.
15. Advancements in predicting bearing capacity of foundations near slopes: a comparative review of LEM/LAM/FEM to AI-driven models
Core Problem: Foundation capacity near slopes depends on coupled footing-slope mechanisms that are treated inconsistently across analytical, numerical, and AI-based methods.
Key Innovation: This review synthesizes conventional and emerging approaches and argues for hybrid, explainable physics-AI workflows for safer foundation design near unstable slopes.
16. Correlations between strain localisation evolution and acoustic emission characteristics of granite subjected to uniaxial compression
Core Problem: Monitoring indicators that reliably mark the transition from localized damage to catastrophic rock failure remain poorly constrained.
Key Innovation: Joint acoustic-emission and digital image correlation measurements tie AE quiet periods, shear-fracture surges, and energy transfer to accelerating strain localization before failure.
17. Efficacy of Prompt Evacuation and Route Selection for Tsunamis Considering Building Collapse: A Case Study of the 2024 Noto Peninsula Earthquake
Core Problem: Evacuation performance under compound tsunami and building-collapse conditions remains difficult to evaluate realistically.
Key Innovation: Coupled inundation and agent-based simulations quantify how prompt departure and route choice alter survival under blockage-prone tsunami scenarios.
18. Hydrodynamic responses and cumulative displacement damage mechanism of beam bridges subjected to continuous flood actions
Core Problem: Continuous flood actions can accumulate damage in bridge systems in ways not captured by static or event-isolated assessments.
Key Innovation: The study links flood-driven hydrodynamic loading to cumulative displacement damage in beam bridges, providing a process-based picture of progressive degradation.
19. Interpretable machine learning approach with WGAN-GP augmented data for predicting seismic stability of horseshoe tunnels in cohesive-frictional soils
Core Problem: Fast prediction of pseudo-static seismic stability for unsupported horseshoe tunnels remains limited by data scarcity and poor interpretability.
Key Innovation: The paper combines generative data augmentation with interpretable machine learning to predict tunnel stability while exposing the main controlling variables.
20. Quantifying Global Internal Displacement Risk at the Hazard-Vulnerability-Conflict Nexus
Core Problem: Displacement risk is rarely quantified jointly across hazard exposure, social vulnerability, and conflict, masking compound hotspots.
Key Innovation: The paper introduces a hazard-specific displacement-risk index that exposes flood-drought-conflict co-occurrence and identifies global compound-displacement hotspots.
21. A review of sonar based bridge underwater scour inspection and identification
Core Problem: Bridge-scour inspection lacks a consolidated evaluation of sonar-based methods for reliable underwater diagnosis.
Key Innovation: This review benchmarks sonar modalities and inspection workflows, clarifying their roles in practical scour detection and identification.
22. Cross-modal distillation for real-time wildfire detection and localization in edge-deployed aerial vehicles
Core Problem: Real-time airborne wildfire sensing is limited by the compute and sensor burdens of deploying thermal-capable systems at the edge.
Key Innovation: Thermal-to-optical cross-modal distillation yields a lightweight student model that localizes wildfire patches quickly using only optical imagery at deployment.
23. Decoding GPR facies with internal sedimentary structures of 2019 tropical storm Pabuk: utility to infer history of storm hazards in coastal environments
Core Problem: Coastal storm deposits are difficult to interpret in ways that reliably reconstruct past hazard intensity and extent.
Key Innovation: Ground-penetrating radar facies are tied to internal storm-deposit structures to improve retrospective diagnosis of coastal hazard history.
24. Dynamic response laws of tunnel lining in mudstone under the moving load of high-speed railway trains: A case study
Core Problem: Repeated moving loads from high-speed rail disturb tunnel lining and surrounding mudstone, but their dynamic interaction remains poorly quantified.
Key Innovation: A project-based dynamic analysis clarifies how lining and surrounding rock co-evolve under train loading, improving stability evaluation for mudstone tunnels.
25. Effect of layout on local scour around pile groups under combined wave-current conditions
Core Problem: Layout effects on local scour around multi-pile systems under combined wave-current forcing are still difficult to predict.
Key Innovation: Large-eddy simulations resolve how pile-group arrangement reorganizes scour development and therefore changes foundation vulnerability under marine forcing.
26. High Resolution Flood Extent Detection Using Deep Learning with Random Forest Derived Training Labels
Core Problem: Extreme-event flood mapping is constrained by scarce labels and intermittent observations at the spatial detail needed for validation and response.
Key Innovation: A label-efficient framework uses Random Forest-derived masks and PlanetScope imagery to train U-Net flood mappers for high-resolution inundation detection.
27. Leveraging LLMs and Social Media to Understand User Perception of Smartphone-Based Earthquake Early Warnings
Core Problem: Technical performance of smartphone-based earthquake early warning systems is not matched by equally strong evidence on public perception and response.
Key Innovation: The study uses LLM-assisted analysis of social-media reactions to a real 2025 earthquake alert deployment to diagnose user perception of smartphone EEW.
28. LLM-Powered Flood Depth Estimation from Social Media Imagery: A Vision-Language Model Framework with Mechanistic Interpretability for Transportation Resilience
Core Problem: Real-time, street-level flood-depth estimation at decision-relevant resolution is still unavailable for transportation operations.
Key Innovation: A fine-tuned vision-language model converts single crowd-sourced images into continuous flood-depth estimates and links prediction skill to interpretable internal depth-encoding layers.
29. Permanent deformation characteristics of coarse-grained materials in high-speed railway subgrades under the combined effects of hydro-environmental variations and high-frequency dynamic loading
Core Problem: Subgrade serviceability deteriorates under the coupled action of moisture cycling and high-frequency dynamic loading, but the deformation envelope is not well constrained.
Key Innovation: Experiments map permanent deformation behaviour across hydro-environmental and loading regimes, providing a basis for more robust high-speed railway subgrade design.
30. The Potential of Copernicus Satellites for Disaster Response: Retrieving Building Damage from Sentinel-1 and Sentinel-2
Core Problem: Rapid building-damage mapping depends heavily on very-high-resolution imagery that is often unavailable over broad disaster footprints.
Key Innovation: The study introduces the xBD-S12 benchmark and shows that Sentinel-1/2 data can support operationally useful, wide-area building-damage assessment.