TerraMosaic Daily Digest: June 4, 2026
Daily Summary
The June 4 literature is led by studies that make failure processes observable at operational scale. A real-gully debris-flow experiment quantifies how baffle arrays and a flexible barrier redistribute velocity, impact force, erosion, and deposition. LandslideSegFormer targets the persistent loss of small landslide bodies and sharp scar boundaries in transformer segmentation. Mine-slope papers move from static weak-layer description to coupled thermo-hydro-mechanical creep and freeze-thaw damage models, while two thaw-slump studies combine InSAR deformation, interpretable convolutional learning, and transferable hemispheric mapping to treat permafrost failure as both a local instability and a northern-latitude landscape process.
The broader geohazard set links physical state, exposure, and sensing constraints. Antakya is analyzed as a multi-hazard heritage landscape where landslides, earthquakes, floods, erosion, and other threats intersect with cultural assets. Seismic-risk papers refine vulnerability functions, building-function recovery, ground-motion correlation, and infrastructure response to fault dislocation. Flood, coastal, drought-flood, tree-fall, and barrier-island studies extend susceptibility and consequence modelling beyond single hazard layers. The method papers are strongest where they remove known bottlenecks in disaster observation: SAR temporal anomaly detection, SAR-optical flood translation, DEM void filling, self-supervised SAR representation learning, and foundation-model adaptation for polar and civil-infrastructure scenes.
Key Trends
Five movements define the issue: measured failure mechanics, scalable permafrost mapping, sensor-robust remote-sensing AI, geology-aware infrastructure modelling, and risk assessment tied to exposed assets and recovery.
- Failure mechanics are being measured rather than inferred: The debris-flow barrier experiment, carbonaceous-shale creep tests, fissured-conglomerate freeze-thaw study, clay-concrete seepage tests, and pressurized-granular fault simulations all tie hazard interpretation to observable deformation, force, permeability, and damage evolution.
- Cryosphere-related slope hazards are becoming scalable mapping targets: The Qinghai-Tibet susceptibility model and hemispheric thaw-slump framework combine InSAR, optical-SAR fusion, topography, and transferable deep learning to move permafrost instability from case-study inventories toward repeatable regional monitoring.
- Remote-sensing AI is shifting toward sensor failure modes: Landslide segmentation, heterogeneous disaster change detection, SAR self-supervision, SAR anomaly forecasting, flood-oriented SAR-optical translation, and DEM reconstruction address small targets, missing labels, cloud cover, voids, and modality gaps rather than only improving benchmark accuracy.
- Infrastructure geohazard models are coupling geology with system response: Tunnel fault-dislocation tests, tunnel water-bearing inversion, rock-mass discontinuity prediction, soil-rock interface identification, ballast degradation, dam seepage, mine inflow, and coal-burst warning link subsurface conditions to the behavior of engineered networks.
- Risk assessment is becoming more spatially explicit and decision-facing: The Antakya, flood-susceptibility, seismic-vulnerability, DFAA, medicanes, tree-fall, and social-vulnerability papers translate hazard fields into exposed assets, recovery functions, site-selection uncertainty, and infrastructure-scale consequences.
Selected Papers
This issue contains 50 selected papers from 1,905 papers analyzed. The selected papers are ordered by relevance score and emphasize field-validated debris-flow mitigation, small-landslide segmentation, thaw-slump mapping, mine-slope rheology, multi-hazard vulnerability, infrastructure response, and remote-sensing methods that remove bottlenecks in hazard observation.
1. Effect of a hybrid baffle-flexible barrier system for debris flow mitigation: A real-scale investigation in a natural gully
Core Problem: Debris-flow barriers are commonly designed from simplified impact assumptions, while field-scale evidence on how upstream baffles alter velocity, impact force, erosion, and deposition remains limited.
Key Innovation: Runs a real-scale natural-gully experiment using camcorders, high-speed video, UAV-LiDAR, and load cells to show that baffle arrays reduce frontal velocity and terminal-barrier load while promoting bed deposition.
2. LandslideSegFormer: A non-local context-aware dual-branch network for accurate small-scale landslide segmentation from remote sensing imagery
Core Problem: Small landslide bodies and scar boundaries are often missed by transformer-based segmentation models because global context is gained at the cost of local detail.
Key Innovation: Redesigns the SegFormer-style encoder-decoder as a dual-branch landslide network with non-local feature interaction, improving small-target representation and boundary preservation.
3. GIS-based spatial multi-hazard vulnerability assessment of cultural heritage sites: the case of Antakya
Core Problem: Cultural heritage risk is rarely assessed with a spatial framework that can represent simultaneous exposure to landslides, earthquakes, floods, erosion, and anthropogenic threats.
Key Innovation: Integrates remote sensing, GIS, and multi-hazard vulnerability analysis for Antakya after the 2023 Kahramanmaras earthquake, identifying heritage assets exposed to seven hazard classes.
4. Shear creep characteristics and fractional modeling of carbonaceous shale under thermo-hydro-mechanical coupling
Core Problem: Carbonaceous shale weak interlayers in high mine slopes soften under water and temperature fluctuation, making time-dependent deformation a direct landslide-prevention concern.
Key Innovation: Combines direct shear, creep testing, Caputo fractional modelling, and nonlinear damage functions to represent water-content and temperature-cycle controls on shale creep.
5. Impacts of prefabricated fissure on micro-macro damage behavior of conglomerate under freeze-thaw cycles
Core Problem: Fissured conglomerate in active-fault and plateau landslide settings degrades under freeze-thaw cycling, but the link between pore evolution and macroscopic unloading failure is not well quantified.
Key Innovation: Uses Lagangcun landslide conglomerate, triaxial unloading tests, CT-derived pore metrics, permeability, fractal measures, and a damage constitutive model to connect fissure geometry with freeze-thaw weakening.
6. Thaw slump susceptibility assessment in the central Qinghai-Tibet Plateau permafrost region based on an interpretable 2D-CNN framework with SBAS-InSAR deformation
Core Problem: Permafrost thaw-slump assessments often omit surface deformation and spatially coupled environmental interactions, limiting mechanistic interpretation under climate warming.
Key Innovation: Adds SBAS-InSAR deformation to an interpretable 2D-CNN susceptibility framework, allowing thaw-slump controls on the Qinghai-Tibet Plateau to be mapped with spatial-context learning.
7. Hemispheric-scale mapping of thaw slumps using a cloud-native and transferable deep learning framework
Core Problem: Thaw-slump detection across northern high latitudes is constrained by fragmented workflows, heterogeneous environments, and weak transferability between regions.
Key Innovation: Couples Google Earth Engine multi-source fusion with a modular PyTorch pipeline and physically interpretable channels to support transferable hemispheric thaw-slump mapping.
8. Fine-grained heterogeneous change detection in complex disaster response with wavelet-based spatial-frequency coupled learning
Core Problem: Optical-SAR change detection is essential for all-weather disaster response, but modality mismatch and lack of fine-grained training data limit reliable damage mapping.
Key Innovation: Builds sub-meter heterogeneous datasets for earthquake building damage and flood assessment and introduces a wavelet-based spatial-frequency coupled Siamese network.
9. From Satellites to Safety: An Open-Source SBAS Workflow for Ground Deformation Monitoring
Core Problem: Active mines require reproducible deformation monitoring, but operational SBAS processing can be difficult to deploy consistently across safety and environmental workflows.
Key Innovation: Implements an open-source HyP3-MintPy SBAS workflow with Sentinel-1 data for semi-automated deformation monitoring at the Bjorkdal gold mine.
10. A new global vulnerability model for regional seismic risk assessments: Part 1 - structural vulnerability
Core Problem: Large-scale earthquake risk models depend on vulnerability functions that often simplify building behavior and underrepresent structural diversity.
Key Innovation: Expands the GEM structural vulnerability database to nearly 1000 building classes and uses efficient multi-degree-of-freedom numerical modelling for regional seismic risk analysis.
11. Multi-Scale High-Resolution Urban Flood Susceptibility Mapping Using MaxEnt and Multi-Source Geospatial Data
Core Problem: Urban flood susceptibility maps often ignore drainage-capacity representation, spatial unit choice, and sparse-event uncertainty despite high-resolution Earth-observation inputs.
Key Innovation: Builds catchment-, street-, and grid-scale MaxEnt models using high-resolution geospatial data and explicit drainage-density indicators to expose scale-dependent flood controls.
12. Response of Tunnel-track system to fault dislocation considering rock mass rupture patterns
Core Problem: High-speed railway tunnels crossing strike-slip faults can experience very different deformation and damage depending on whether rupture localizes or distributes through the surrounding rock.
Key Innovation: Uses large-scale physical model tests to compare concentrated and distributed rupture patterns and quantify their effects on coupled tunnel-track-rock response.
13. A probabilistic framework for assessing seismic functionality loss and recovery in buildings using a spatially explicit functionality metric
Core Problem: Post-earthquake functionality loss is commonly represented by coarse operational states that cannot capture spatially varying recovery of damaged buildings.
Key Innovation: Introduces a probabilistic, spatially explicit building-functionality metric for modelling loss and recovery trajectories after seismic damage.
14. Seismic vulnerability of historic masonry buildings in Vienna (Austria): influence of structural and material parameters
Core Problem: Vienna's historic unreinforced masonry stock has high cultural value but uncertain seismic performance because structural and material parameters vary widely.
Key Innovation: Evaluates parameter sensitivity for historic masonry vulnerability functions, improving earthquake risk assessment for heritage building stocks.
15. Seismic vulnerability of existing RC building stocks at territorial scale: a novel probability-tree approach
Core Problem: Territorial earthquake-risk studies need vulnerability estimates for heterogeneous RC building stocks without oversimplifying structural variability.
Key Innovation: Proposes a probability-tree method to propagate building-stock uncertainty into large-scale seismic vulnerability assessment.
16. Effective risks and dominant factors of drought-flood abrupt alternation events to global socio-ecological systems
Core Problem: Drought-flood abrupt alternation imposes coupled hazards on social and ecological systems, but global effective risk has not been systematically quantified.
Key Innovation: Combines hazard, exposure, and vulnerability in a global double-mean, double-threshold, double-driver framework to identify dominant DFAA risk factors.
17. Assessing Future Coastal Hazards due to Mediterranean Hurricanes Under Sea Level Rise and Storm Intensification
Core Problem: Mediterranean hurricanes can interact with sea-level rise to amplify coastal flooding and erosion, but future compound hazard scenarios remain underdeveloped.
Key Innovation: Evaluates medicane-driven coastal hazards under storm intensification and sea-level-rise scenarios to support regional risk planning.
18. Complex Shallow Unilateral Rupture of the 2025 Mw 7.0 Hubbard Glacier Earthquake Terminated by a Restraining Bend on the Connector Fault
Core Problem: The Connector fault near the Alaska-Canada border is poorly constrained, limiting understanding of rupture termination and transboundary seismic hazard.
Key Innovation: Combines relocation, moment-tensor inversion, and finite-fault modelling to resolve a complex shallow unilateral rupture terminated by a restraining bend.
19. Data-efficient flood depth prediction through domain-aware coreset selection and tabular foundation models
Core Problem: Near-real-time flood-depth surrogates are usually data hungry and watershed-specific, limiting transfer to new basins or sparse-event settings.
Key Innovation: Uses domain-aware coreset selection and tabular foundation-model conditioning to approach supervised flood-depth performance with less than 1 percent of the training pool.
20. T-SAR-JEPA: Self-Supervised Temporal Anomaly Detection in SAR Amplitude Stacks via Latent Prediction
Core Problem: Operational SAR monitoring needs temporal anomaly detectors that do not depend on dense labels or coherence during training.
Key Innovation: Adapts SAR-JEPA representations to temporal latent prediction over SAR amplitude stacks and validates anomaly detection against independent InSAR-coherence pseudo-ground truth.
21. Investigating frictional instability due to pressurization in granular media: insights from coupled computational fluid dynamics discrete element method
Core Problem: Fluid pressure can reactivate granular layers in faults, slopes, and reservoirs, but grain-scale feedbacks between pressure diffusion and contact-network degradation remain unresolved.
Key Innovation: Uses coupled CFD-DEM simulations to examine how pressurization, drainage, and grain contacts interact during frictional instability.
22. Shear behavior degradation and micro-mechanism of clay-concrete interface under seepage erosion
Core Problem: Clay-concrete interfaces in ground-improvement and slope-stability systems can lose shear resistance under seepage erosion, but the micro-mechanism is poorly observed.
Key Innovation: Combines custom seepage-direct-shear testing, SEM, and particle-loss analysis to link dry density, seepage pressure, and normal stress to interface degradation.
23. Comparative study of the transient response and failure characteristics of lined tunnels under idealized and realistic plane P-wave seismic excitation
Core Problem: Tunnel seismic analyses can underestimate damage if realistic waveforms are simplified into idealized plane-wave inputs.
Key Innovation: Compares lined-tunnel transient response and failure characteristics under idealized and realistic P-wave seismic excitation.
24. Study on the degradation process and mechanism of lateral ballast resistance under seismic loading
Core Problem: The lateral ballast resistance that stabilizes ballasted tracks can degrade sharply during earthquakes, threatening track performance.
Key Innovation: Investigates seismic-loading degradation mechanisms of ballast resistance to support earthquake-safe railway track design.
25. Structure-preserving DEM void filling via optical imagery-guided diffusion
Core Problem: DEM voids compromise terrain metrics used in landslide, flood, and erosion analysis, while interpolation oversmooths and GAN methods can introduce artifacts.
Key Innovation: Introduces an optical-imagery-guided diffusion model that uses cross-modal terrain-structure priors to fill DEM voids while preserving geomorphic form.
26. SAFE: A SAR Feature Extractor Based on Self-Supervised Learning and Masked Siamese ViTs
Core Problem: SAR is central to all-weather disaster monitoring, but labelled SAR datasets remain scarce for many hazard-mapping tasks.
Key Innovation: Adapts masked Siamese vision transformers and contrastive self-supervision to learn transferable SAR features from unlabelled imagery.
27. Clustering Guided Domain-Specific Pretrained Foundation Model for Very High-Resolution Arctic Remote Sensing
Core Problem: Arctic hazard and environmental mapping require very-high-resolution models trained on diverse polar scenes rather than globally generic image corpora.
Key Innovation: Curates roughly 3 million chips from 267 TB of Arctic imagery with diversity-aware clustering and pretrains a domain-specific MAE-ViT foundation model.
28. D-PolTransGAN: Scattering Guided Hybrid Transformer-GAN for SAR-Optical Image Translation Using Dual Polarization SAR
Core Problem: SAR supports flood mapping under clouds and darkness, but speckle and limited visual interpretability complicate integration with optical workflows.
Key Innovation: Uses dual-polarization scattering streams, cross-attention, transformers, and GAN translation to generate optical-like products tailored for flood assessment.
29. DFF-Net: A Dual-Branch Feature Fusion Network for River Ice Mapping via the Fusion of Microwave and Optical Imagery
Core Problem: River ice in high-altitude cold mountainous regions is difficult to map because cloud and snow cover degrade optical satellite extraction.
Key Innovation: Fuses optical, SAR, NDSI, and DEM features in a dual-branch attention network for mountain river-ice mapping.
30. Morphological response of vegetated and urbanized barrier islands to Hurricane Ian
Core Problem: Barrier-island morphology during hurricanes depends on both natural vegetation and built infrastructure, but these controls are difficult to isolate.
Key Innovation: Combines high-resolution pre- and post-storm topo-bathymetric data with XBeach modelling to quantify Hurricane Ian erosion, deposition, and dune-crest change.
31. Predicting the risk of individual tree fall along powerlines in Norway with a mechanistic wind risk model and machine learning
Core Problem: Tree fall along powerlines creates localized hazard for linear infrastructure, yet risk is often managed without individual-tree vulnerability modelling.
Key Innovation: Combines drone-based inventory, the ForestGALES mechanistic wind-risk model, and machine learning to estimate individual tree-fall probability.
32. Spatial correlation of multiple ground-motion intensity measures based on the updated New Zealand ground-motion database
Core Problem: Portfolio and distributed-infrastructure seismic risk estimates require regional spatial-correlation models for ground-motion intensity measures.
Key Innovation: Uses 1839 recordings from 143 New Zealand earthquakes to develop region-specific spatial-correlation models for PGA and spectral acceleration periods.
33. Design and performance analysis of a new type of guide-rail isolation device applicable to the seismic protection of cultural relics
Core Problem: Cultural relics require seismic isolation systems that move easily, dissipate energy, and remain reliable under repeated museum-storage or exhibition conditions.
Key Innovation: Designs and tests a guide-rail isolation device with orthogonal motion, variable stiffness, and rotary damping for relic-scale seismic protection.
34. Influence of grain group deletion on the properties and pore-cementation synergistic mechanism of EICP-consolidated sandy clayey purple soil
Core Problem: Sandy clayey soils are structurally vulnerable to erosion, and EICP treatment performance depends on grain-size composition and pore-cementation patterns.
Key Innovation: Tests grain-group deletion effects to clarify how enzymatic calcium carbonate precipitation improves the pore-cementation structure of purple soil.
35. A dual-stream LSTM-XGBoost framework for mine water inflow forecasting: a case study of the Longfeng coal mine
Core Problem: Mine inflow forecasts must combine temporal water-inflow history with spatial hydrogeological and operational controls, which many models treat separately.
Key Innovation: Uses parallel LSTM and XGBoost streams with two-stage training to fuse temporal and spatial drivers for daily coal-mine inflow prediction.
36. Investigation of underground seepage problem of a clay core rockfill dam and its fast and cost-effective treatment method
Core Problem: Persistent seepage beneath clay-core rockfill dams can threaten stability and water-retention performance when karstic limestone and low-permeability bedrock coexist.
Key Innovation: Diagnoses leakage mechanisms at an operating dam and evaluates a rapid treatment strategy after prior blanket and shallow-grouting measures proved insufficient.
37. Coal Burst Failure and Probabilistic Early Warning for Combined Coal-Rock Based on Electric Potential Under True Triaxial Stress
Core Problem: Coal burst in deep coal-rock composites requires warning indicators that detect precursor anomalies under realistic three-dimensional stress states.
Key Innovation: Uses electric-potential monitoring in true triaxial tests to construct a probabilistic early-warning procedure for coal-rock burst failure.
38. Identification of Soil-Rock Interfaces Using Multiscale Wavelet-Based Sparse Signal Peak Detection of Vibration During Shield Tunnelling
Core Problem: Unexpected soil-rock interfaces during shield tunnelling can change cutting behavior and hazard exposure, but continuous interface positions are difficult to detect in advance.
Key Innovation: Uses multiscale wavelet-based sparse vibration peaks and two-level clustering to locate soil-rock interfaces from shield-tunnelling vibration signals.
39. Prediction of Applied Forces on the TBM Disk Cutter by Numerical and Artificial Intelligence Methods
Core Problem: Real-time TBM cutter-force prediction is difficult because numerical models are slow while AI models require reliable training data.
Key Innovation: Generates a FEM-based dataset and integrates numerical simulation with AI models to predict disc-cutter forces under variable rock and stress conditions.
40. MESA-Net: A Multi-Directional Edge-Aware Network with Scale Adaptation for Water Body Segmentation in Karst Landscapes
Core Problem: Karst water bodies are elongated, fragmented, shadow-affected, and often obscured by vegetation, limiting accurate boundary delineation from satellite imagery.
Key Innovation: Introduces a CNN-Mamba segmentation network with edge awareness and scale adaptation for small and irregular water-body extraction in karst terrain.
41. Regulatory role of permanent gullies in dissolved nitrogen and phosphorus transport under different rainfall types
Core Problem: Permanent gullies can regulate runoff and nutrient transport under rainfall, but their event-scale role in hillslope-to-channel transfer is poorly resolved.
Key Innovation: Monitors paired gully heads and outlets across rainfall classes to quantify dissolved nitrogen and phosphorus transport under different storm types.
42. Where social vulnerability indices diverge: Navigating the decision cascade for local site selection
Core Problem: Local hazard planning often relies on social-vulnerability indices whose construction choices are undocumented and can change site-selection outcomes.
Key Innovation: Analyzes how indicator, normalization, weighting, and aggregation decisions propagate through SVI outputs used for local resource allocation.
43. Advanced spatial-temporal drought mapping using machine learning and object-based methods
Core Problem: Conventional drought indices can miss spatial-temporal dynamics in semi-arid watersheds with complex vegetation-climate interactions.
Key Innovation: Combines machine-learning algorithms, object-based methods, and multiple optical-SAR drought indicators for spatial-temporal drought mapping in the Taleqan watershed.
44. Disaster risk in Brazil: trends, challenges and policy insights
Core Problem: Brazil has expanded disaster-risk policies, but the relationship between institutional change and long-term disaster outcomes remains unclear.
Key Innovation: Links Sendai-aligned disaster indicators from 1991-2024 with national DRR legislation to evaluate trends, policy gaps, and persistent regional disparities.
45. Deciphering extreme wave events in the Guadiana River-Estuary (SW Spain): a multidisciplinary approach
Core Problem: Extreme wave events in estuarine records are difficult to distinguish without integrating stratigraphy, geomorphology, and hazard interpretation.
Key Innovation: Uses multidisciplinary trench and palaeoenvironmental evidence to reconstruct extreme-wave signatures in the Guadiana River estuary.
46. Multi-Task Crack Foundation Model for Engineering-Reliable Crack Representation and Topology Preservation in Civil Infrastructure
Core Problem: Crack segmentation models can achieve high overlap scores while fragmenting crack topology and providing no calibrated uncertainty under domain shift.
Key Innovation: Combines a frozen foundation backbone with crack-specific mask, skeleton, and uncertainty heads to preserve connected crack geometry for infrastructure assessment.
47. RadiusFPS: Efficient Farthest Point Sampling on CPUs and GPUs via Spherical Voxel Pruning
Core Problem: Point-cloud downsampling is a latency bottleneck for LiDAR perception, SLAM, and mobile mapping pipelines used in terrain and infrastructure surveys.
Key Innovation: Accelerates farthest-point sampling with spherical voxel pruning while preserving the standard sampling update rule for large point clouds.
48. UltraVR: A Diagnostic Ultra-Resolution Image-VQA Benchmark for Evidence-Grounded Reasoning
Core Problem: Vision-language models may fail when evidence in ultra-resolution remote-sensing images is tiny, spatially distant, or distributed across a scene.
Key Innovation: Introduces a diagnostic benchmark for evidence-grounded visual reasoning across ultra-resolution domains, including remote sensing and industrial anomaly detection.
49. Q-SOCO-Field: A Quantum-Enhanced Observation Capability Representation Framework for Earth Observation Sensor Planning
Core Problem: Fine-grained Earth-observation sensor planning becomes computationally difficult as dynamic observation capability fields increase in spatial and temporal resolution.
Key Innovation: Proposes a quantum-enhanced representation framework for computing time-varying EO sensor capabilities, with potential relevance to rapid hazard-tasking systems.
50. ESFCNet: Efficient Spatial-Frequency Complementary Network for Real-Time Semantic Segmentation of Uncrewed Aerial Vehicle Images
Core Problem: Fine-grained UAV mapping needs real-time segmentation that preserves spatial detail without expensive global-context computation.
Key Innovation: Uses an efficient spatial-frequency complementary architecture to support low-latency UAV semantic segmentation, a transferable capability for rapid hazard mapping.