Initiated by Dr. Xin Wei, University of Michigan
Ongoing development by the community

TerraMosaic Daily Digest: August 20, 2026

August 20, 2026 TerraMosaic Daily Digest

Daily Summary

The August 20 selection frames landslide and slope instability as consequences of evolving hydro-mechanical state rather than rainfall magnitude alone. In the Alpes-Maritimes, extreme precipitation changes both triggering conditions and inferred shallow-landslide susceptibility, while the Chieng So study couples a dense post-event inventory, mobility analysis, and explainable learning to resolve controls on rainfall-induced landslides and debris flows. Laboratory and numerical studies extend this logic across soil-slide forecasting, liquefiable sands, internally unstable soils, anti-dip layered rock slopes, vibration-exposed loess embankments, and vegetation-modified subgrade hydrology, showing that pore-pressure buildup, seepage-pathway reorganization, fines loss, bedding geometry, and material deterioration govern whether deformation stays diffuse or localizes into failure.

Seismic, tectonic, volcanic, cryospheric, and flood papers likewise recover structure-specific controls that routine hazard metrics often miss. Slow-earthquake behavior in Cascadia is linked to both tectonic and environmental forcing over seasonal to longer-term scales; geomorphic analysis of the Kashmir seismic gap argues for unresolved dextral shear and block rotation; and co-seismic surface rupture can be timed and tracked directly from earthquake video. Additional studies constrain how fault orientation affects injection-induced slip, how reverse- and strike-slip faulting load pipelines, how the base of submarine permafrost can be inferred from seismic and hydrate-stability relations, and how porous volcanic rocks undergo time-dependent compaction. Flood and hydrologic studies similarly favor coupled representations of inundation, evacuation, seasonal water storage, snowmelt process realism, atmospheric moisture, hurricane-driven streamflow change, and ecosystem buffering over static hazard footprints.

A substantial parallel stream broadens the technical repertoire without implying geohazard validation by default. Hazard-oriented methods include graph-embedded susceptibility mapping under data scarcity, InSAR-derived subsidence susceptibility, guided-diffusion synthesis for landslide change detection, and ambient-noise or thunderquake imaging of shallow structure. More general remote-sensing and AI papers emphasize weak supervision, object-centric sampling, multimodal registration, LiDAR localization, changepoint and concept-drift detection, early classification in non-stationary streams, and physically structured surrogates or operator-learning schemes. Their shared contribution is to make sparse labels, domain shift, computational cost, and uncertainty more manageable for future hazard studies, while remaining validated primarily in their stated sensing or modeling tasks.

Key Trends

The August 20 corpus converges on five directions: state-aware failure diagnosis, more explicit hydro-mechanical thresholds, opportunistic sensing under sparse observations, infrastructure-coupled hazard modeling, and transferable GeoAI designed for sparse labels and uncertainty.

  • Hidden Internal State Is Becoming the Main Explanatory Frame: Across shallow landslides, liquefaction, internal erosion, subsidence, volcanic compaction, and slow earthquakes, the decisive controls are increasingly inferred subsurface states such as pore pressure, fines migration, thermal regime, bedding-controlled damage transition, or fault-modulated stress. The corpus repeatedly shifts interpretation away from trigger counts toward diagnosing the mechanical or hydraulic condition that makes failure possible.
  • Hydro-Mechanical Thresholds Are Being Parameterized More Explicitly: The French and Vietnamese landslide studies, cyclic liquefaction tests, rainfall-vibration slope experiments, and pipeline-fault interaction analyses all recover transition regimes or sensitivity thresholds rather than only retrospective narratives. This strengthens the path from process interpretation to forecasting, design guidance, and early-warning logic.
  • Opportunistic and Multi-Modal Sensing Is Expanding Under Sparse Observation: Earthquake videos, DAS-recorded thunderquakes, ambient-noise inversion, multi-season thermal imagery, InSAR labels, and fused open geospatial predictors are used to recover hidden deformation, thermal, or structural signals where conventional monitoring is sparse, obscured, or too costly. The common direction is practical sensing under observational constraint, not sensor novelty for its own sake.
  • Hazards Are Increasingly Modeled as Coupled Environmental-Infrastructure Systems: Flood evacuation planning, high-resolution inundation emulation, dam seismic assessment, buried and subsea pipeline deformation, tunnel inflow management, and railway-slope seepage studies all treat hazard as an interaction among forcing, engineered geometry, and operational decision-making. This produces outputs that are more directly actionable than stand-alone susceptibility surfaces.
  • Transferable GeoAI Prioritizes Sparse Labels, Generalization, and Auditability: Graph-augmented susceptibility learning, diffusion-based landslide synthesis, weakly supervised segmentation, object-centric patch sampling, robust multimodal matching, drift-aware retraining, early classification in non-stationary settings, and physics-informed surrogates all target data scarcity, domain shift, or uncertainty. These are enabling advances for geohazard research, but the selected papers largely validate them within their own remote-sensing or computational tasks rather than across geohazards broadly.

Selected Papers

The selected papers combine direct studies of landslides, seismic and fault deformation, volcanic and permafrost processes, flood systems, and engineered-ground response with adjacent hydrologic and ecosystem analyses and a substantial methodological stream. The hazard papers report mechanism, threshold, or monitoring findings in their stated domains, whereas the remote-sensing and AI papers extend segmentation, registration, uncertainty, and surrogate-model capabilities that may transfer to geohazards but are not themselves hazard-validated by default.

1. Inferring the Base of Submarine Permafrost in the Canadian Beaufort Sea From Seismic Data and the Depth of the Gas Hydrate Stability Zone

Source: Journal of Geophysical Research: Solid Earth Type: Permafrost geophysics Geohazard Type: Submarine permafrost Relevance: 8/10

Core Problem: The thermal base and present extent of submarine permafrost are hard to observe directly on the Beaufort Shelf.

Key Innovation: Infers the base of permafrost from the seismic base of gas hydrate stability and validates it against diffraction signatures of ice-bearing permafrost.

2. Cascadia Slow Earthquake Behaviors Modulated by Tectonic and Environmental Forcings

Source: Geophysical Research Letters Type: Seismology Geohazard Type: Earthquake Relevance: 8/10

Core Problem: Seasonality and long-term evolution of tremor and slow slip in Cascadia are not fully explained by hydrological forcing alone.

Key Innovation: Shows that SSE stress variations better explain tremor seasonality and identifies long-term shifts in ETS nucleation and propagation.

3. Co-seismic surface fault displacement captured in videos: image tracking across three earthquakes

Source: ArXiv (Geo/RS/AI) Type: Preprint Geohazard Type: Surface faulting Relevance: 8/10

Core Problem: Rare videos of fault displacement are hard to convert into kinematic measurements because of ground motion and camera shake.

Key Innovation: Stabilizes CCTV footage and tracks correlated image regions to recover displacement duration and velocity across three earthquakes.

4. SAGE-XGBoost: Spatially Augmented Graph Embeddings--Machine Learning Framework for Natural Hazards Susceptibility Mapping under Data Scarcity

Source: ArXiv (Geo/RS/AI) Type: Preprint Geohazard Type: Landslide Relevance: 8/10

Core Problem: Natural-hazard susceptibility mapping is limited by scarce labels and weak spatial feature representations.

Key Innovation: Combines graph-embedded neighborhood statistics, controlled augmentation, and XGBoost to improve landslide and wildfire susceptibility maps.

5. Impact of extreme rainfall on triggering conditions and susceptibility for shallow landslides: a case study in the Alpes-Maritimes region (France)

Source: Natural Hazards and Earth System Sciences Type: landslide triggering and susceptibility Geohazard Type: landslide Relevance: 8/10

Core Problem: Determine how extreme-rainfall failures alter shallow-landslide predictive tools.

Key Innovation: Comparison with and without Storm Alex landslides showing distinct triggering conditions and susceptibility maps.

6. The missing large-scale dextral-shear and rotation within the “Kashmir seismic gap”: geomorphological inferences

Source: Frontiers in Earth Science Type: tectonic geomorphology hazard inference Geohazard Type: earthquake faulting Relevance: 8/10

Core Problem: Explain where the geodetically inferred dextral shear in the Kashmir seismic gap is being accommodated.

Key Innovation: Uses drainage geometry, divide offsets, and chi analysis to infer active dextral shear and block rotation consistent with geodesy.

7. Rainfall-induced landslides and debris flows in Chieng So commune, Son La Province, Vietnam

Source: Landslides Type: landslide inventory and susceptibility study Geohazard Type: landslide and debris flow Relevance: 8/10

Core Problem: Characterize a 2025 storm-triggered landslide-debris-flow outbreak and identify dominant triggering controls.

Key Innovation: Integrates a 1,885-landslide inventory, mobility analysis, spatially validated XGB modeling, and SHAP/ALE-derived antecedent rainfall thresholds.

8. Evaluation of the vulnerability of shallow foundation masonry houses in landslide tension zones via physical experiments

Source: Natural Hazards Type: landslide building vulnerability study Geohazard Type: landslide building vulnerability Relevance: 8/10

Core Problem: Improve vulnerability assessment of shallow-foundation masonry houses damaged by landslide-induced tensile deformation.

Key Innovation: Physical model tests produce empirical shallow-foundation damage formulas that are integrated into fuzzy vulnerability evaluation.

9. Excess pore water pressure buildup and shear modulus degradation of liquefiable sand in strain-controlled cyclic triaxial tests

Source: Bulletin of Engineering Geology and the Environment Type: Liquefaction laboratory mechanics study Geohazard Type: seismic liquefaction Relevance: 8/10

Core Problem: Design needs better strain-controlled relations between cyclic loading, pore pressure generation, and modulus degradation.

Key Innovation: Calibrates pore-pressure buildup and normalized shear-modulus degradation relations across density and confinement states.

10. Effect of Fault Orientation Relative to the Wellbore on Injection-Induced Fault Slip: Insights from Distinct-Element Modeling and Microseismic Analysis

Source: Rock Mechanics and Rock Engineering Type: Injection-induced seismicity modeling study Geohazard Type: induced fault slip Relevance: 8/10

Core Problem: How wellbore-fault orientation controls injection-induced fault slip and seismic risk remains unclear.

Key Innovation: Calibrated 3D distinct-element simulations reveal angle-dependent slip area, stick-slip behavior, pressure surges, and microseismic severity.

11. Change in Damage Behavior of Anti-dip Layered Rock Slopes Induced by Alterations in Bedding Dip Angle: Experimental and Numerical Insights

Source: Rock Mechanics and Rock Engineering Type: Rock slope failure experiment and simulation study Geohazard Type: rock slope instability Relevance: 8/10

Core Problem: The bedding dip angle that shifts anti-dip layered slopes from internal compression to volumetric collapse is poorly understood.

Key Innovation: Uses 3D-printed slopes, DIC, AE, simulations, and full-scale validation to identify a 50-60 degree transition to shear-dominated instability.

12. InSAR-derived subsidence susceptibility mapping using open-access geospatial data and random forest

Source: Transportation Geotechnics Type: Remote-sensing and machine-learning subsidence study Geohazard Type: land subsidence Relevance: 8/10

Core Problem: Subsidence and groundwater storage loss are difficult to characterize globally at locally meaningful resolution.

Key Innovation: Uses InSAR-derived labels plus open geospatial predictors and random forest modeling to map global subsidence at about 2 km and estimate aquifer storage loss.

13. Time-dependent (Creep) Compaction in Volcanic Rocks

Source: Geophysical Research Letters Type: Volcanic rock mechanics Geohazard Type: Volcanic hazard Relevance: 7/10

Core Problem: Time-dependent compaction creep in porous lava had not been experimentally characterized.

Key Innovation: Presents the first compaction-creep experiments on lava and shows slow porosity loss localizing into compaction bands.

14. 3-D numerical modelling of the feedback between deformation and thermal structure during subduction initiation for the French Lesser Antilles

Source: ArXiv (Geo/RS/AI) Type: Preprint Geohazard Type: subduction and volcanic hazard Relevance: 7/10

Core Problem: Early subduction-initiation thermal and deformation feedback in the Lesser Antilles is poorly constrained.

Key Innovation: 3D thermomechanical models reveal strong pre-slab deformational heating, ridge formation, and possible early arc-melting zones.

15. SAE-Xplainers: Rule-Based Feature Interpretation for Extreme Earth Events

Source: ArXiv (Geo/RS/AI) Type: Preprint Geohazard Type: wildfire, cyclone, and atmospheric river Relevance: 7/10

Core Problem: Deep models for extreme Earth events lack faithful human-understandable feature explanations for operational use.

Key Innovation: Adapts sparse autoencoders with geographic modulation and rule-based explainers to unpack climatic predictors into interpretable rules.

16. Imaging Earth’s subsurface with thunderstorm-generated seismic waves

Source: Science Advances Type: passive seismic imaging study Geohazard Type: karst and near-surface weak-zone mapping Relevance: 7/10

Core Problem: Turn thunderstorm-generated seismic signals into a practical source for shallow subsurface imaging.

Key Innovation: Uses DAS-recorded thunderquakes, interferometry, and tomography to image about 100 m deep weak zones that align with deformation evidence.

17. Systematic evaluation of soil relative density effects on buried pipeline response to reverse-faulting: Evidence from physical model tests

Source: Bulletin of Engineering Geology and the Environment Type: Physical model study of fault-crossing pipeline response Geohazard Type: surface fault rupture Relevance: 7/10

Core Problem: Backfill density effects on pipeline strain under reverse-fault permanent ground deformation are under-studied.

Key Innovation: Uses split-box experiments to show dense backfill and greater burial depth amplify bending strain, informing mitigation.

18. X-SEL: a stacked ensemble learning architecture for forecasting soil slide displacement in flume simulations

Source: Geoenvironmental Disasters Type: Machine-learning landslide displacement forecasting study Geohazard Type: landslide Relevance: 7/10

Core Problem: Physics-based displacement integration is brittle to noisy IMU data and nonlinear slide behavior.

Key Innovation: Introduces an XGBoost-based stacked ensemble with SHAP interpretation for IMU-driven slide-displacement prediction.

19. Ambient-noise imaging of internal structure in a highway slope using transdimensional Bayesian inversion

Source: Engineering Geology Type: Slope imaging method study Geohazard Type: slope stability Relevance: 7/10

Core Problem: Internal heterogeneity inside a highway slope is difficult to resolve noninvasively for stability assessment.

Key Innovation: Applies ambient-noise imaging with transdimensional Bayesian inversion to slope internal structure characterization.

20. Resilient transit-based evacuation planning under flood disruption: An integrated optimisation and agent-based simulation framework with an Australian case study

Source: International Journal of Disaster Risk Reduction Type: Flood evacuation optimization and simulation study Geohazard Type: flood risk management Relevance: 7/10

Core Problem: Transit-based evacuation plans often ignore simultaneous flood disruption and traveler-agent behavior.

Key Innovation: Integrates optimization with agent-based simulation to build resilient flood evacuation plans.

21. A synergistic thermal framework for monitoring, quantifying, and classifying volcanic activity from space

Source: Remote Sensing of Environment Type: Thermal remote-sensing review/framework Geohazard Type: volcanic thermal monitoring Relevance: 7/10

Core Problem: Reliable satellite thermal retrieval remains challenging for consistent space-based thermal monitoring.

Key Innovation: Synthesizes thermal infrared land-surface-temperature retrieval and validation approaches as a basis for spaceborne activity monitoring.

22. Guided diffusion-based image editing for landslide image synthesis in remote sensing change detection

Source: ISPRS Journal of Photogrammetry and Remote Sensing Type: Generative remote-sensing method for landslide detection Geohazard Type: landslide Relevance: 7/10

Core Problem: Landslide change detection suffers from limited labeled imagery and hard-to-obtain event pairs.

Key Innovation: Uses guided diffusion image editing to synthesize landslide imagery tailored for remote-sensing change detection.

23. A hybrid machine learning and physics-based approach to model high-resolution inundation

Source: Journal of Hydrology Type: Hybrid inundation modeling study Geohazard Type: flood inundation Relevance: 7/10

Core Problem: High-resolution inundation modeling is computationally expensive and hard to scale with pure physics models.

Key Innovation: Combines machine learning with physics-based modeling to emulate high-resolution inundation.

24. Seismic analysis and performance evaluation of hardfill dams

Source: Soil Dynamics and Earthquake Engineering Type: Dam earthquake performance study Geohazard Type: seismic dam safety Relevance: 7/10

Core Problem: Seismic behavior of hardfill dams is under-documented across dam heights, canyon stiffnesses, and shaking intensities.

Key Innovation: Runs nonlinear 2D and 3D analyses showing where tensile damage localizes and how height and foundation stiffness govern performance.

25. Slope gradient effects on train vibration induced seepage-pathway transition and shallow failure in sandy loess railway slopes

Source: Transportation Geotechnics Type: Railway slope hydromechanical instability study Geohazard Type: slope failure Relevance: 7/10

Core Problem: Slope gradient likely controls how train vibration alters seepage paths and triggers shallow failure in sandy loess railway slopes.

Key Innovation: Investigates seepage-pathway transition and shallow failure as a function of slope gradient under train vibration.

26. CVSD-Reg: Cross-Modal Visual Semantic Prior Distillation for Robust LiDAR Registration

Source: ArXiv (Geo/RS/AI) Type: Preprint Geohazard Type: Topographic change detection Relevance: 6/10

Core Problem: Geometric LiDAR registration degrades under changing density, scan pattern, viewpoint, and sensor type.

Key Innovation: Distills vision-foundation semantic priors into LiDAR descriptors for robust zero-shot cross-sensor registration.

27. Far from the Crowd: Scalable Self-Supervised Learning via Geographic Isolation

Source: ArXiv (Geo/RS/AI) Type: Preprint Geohazard Type: general EO hazard mapping Relevance: 6/10

Core Problem: Earth observation self-supervision treats all geolocated images equally despite uneven information density.

Key Innovation: Ranks samples by geographic isolation to build a label-free curriculum that improves EO pretraining faster and cheaper.

28. Causal Inference under Interference with Learned Exposure Mappings

Source: ArXiv (Geo/RS/AI) Type: Preprint Geohazard Type: exposure propagation and intervention effects Relevance: 6/10

Core Problem: Estimate causal spillover effects when exposure mappings are not known and must be learned from environmental transport data.

Key Innovation: Compares mechanistic and operator-learning transport models by how well they recover induced exposure mappings and resulting intervention effects.

29. LF-GICP: Parameter-Free Degeneracy-Aware LiDAR Odometry via a Voxel-Normal Localizability Field

Source: ArXiv (Geo/RS/AI) Type: Preprint Geohazard Type: terrain and tunnel LiDAR mapping Relevance: 6/10

Core Problem: LiDAR odometry drifts badly along unobservable directions in tunnels and corridors, and existing fixes need retuning by environment.

Key Innovation: Replaces Hessian-based degeneracy cues with a voxel-normal localizability field and parameter-free Fisher-information weighting that generalizes across datasets and sensors.

30. SoilWaterNow: Soil water nowcasting for mapping plant available water (PAW) across paddocks for improved on-farm decision-making

Source: ArXiv (Geo/RS/AI) Type: Preprint Geohazard Type: soil moisture and hydrologic triggering Relevance: 6/10

Core Problem: Produce daily paddock-scale plant-available water and root-zone soil-moisture estimates from satellite, climate, and soil data.

Key Innovation: Builds a physically consistent 30 m daily water-balance model that nowcasts root-zone moisture nationally from fused EO and climate inputs.

31. Object-aware graph matching network for cross-domain remote sensing image localization

Source: ArXiv (Geo/RS/AI) Type: Preprint Geohazard Type: remote-sensing transfer Relevance: 6/10

Core Problem: Geo-localize heterogeneous remote-sensing images across severe cross-modal appearance gaps.

Key Innovation: Object-aware dual-graph matching with a new infrared-visible localization dataset.

32. Influence of loading mode and water content on the mechanical properties and deterioration mechanisms of various lithologies: insights from acoustic emission characterization

Source: Canadian Geotechnical Journal Type: rock deterioration mechanics Geohazard Type: rock instability Relevance: 6/10

Core Problem: Explain lithology-dependent mechanical degradation and failure under different water contents.

Key Innovation: Comparative granite-slate acoustic-emission and microstructural analysis revealing distinct softening pathways.

33. Fate of heat approaching the Filchner-Ronne Ice Shelf mediated by continental shelf eddies

Source: Science Advances Type: cryosphere ocean-process simulation Geohazard Type: ice-shelf instability and sea-level hazard Relevance: 6/10

Core Problem: Explain why warm shelf heat near the Filchner-Ronne Ice Shelf does not fully translate into strong basal melt.

Key Innovation: High-resolution modeling shows continental shelf eddies divert incoming heat upward and away from the ice shelf.

34. Mitigating Class Imbalance and False-Negative Supervision in Remote Sensing Semantic Segmentation Using Object-Centric Patch Sampling

Source: Remote Sensing (MDPI) Type: remote-sensing segmentation sampling method Geohazard Type: hazard extent mapping transfer Relevance: 6/10

Core Problem: Reduce background dominance and false-negative supervision caused by conventional patch sampling.

Key Innovation: Object-centric patch sampling anchors training chips on labeled targets and materially improves segmentation across sensors and models.

35. Weakly Supervised Remote Sensing Segmentation via Decoupled Cross-Modal Distillation and Semantic-Guided Refinement

Source: Remote Sensing (MDPI) Type: weakly supervised remote-sensing segmentation Geohazard Type: hazard extent mapping transfer Relevance: 6/10

Core Problem: Generate complete and spatially coherent segmentation labels from image-level supervision only.

Key Innovation: Decoupled CLIP distillation plus DINOv2 clustering and SAM-guided refinement produce stronger pseudo-labels without foundation models at inference.

36. Hyper-VMIL: Topology-Aware Variational Hypergraph Multiple-Instance Learning for Weakly Supervised Hyperspectral Target Detection

Source: Remote Sensing (MDPI) Type: weakly supervised hyperspectral detection Geohazard Type: hyperspectral material mapping transfer Relevance: 6/10

Core Problem: Handle bag ambiguity, boundary over-smoothing, and latency in weakly supervised hyperspectral target detection.

Key Innovation: Dual spatial-spectral hypergraphs with variational inference, posterior refinement, and teacher-student distillation enable accurate fast pixel inference.

37. Internal erosion and post-erosion stress-strain behavior of internally unstable soil under fluctuating hydraulic loading

Source: Bulletin of Engineering Geology and the Environment Type: internal-erosion geomechanics study Geohazard Type: piping and embankment instability Relevance: 6/10

Core Problem: Determine how fluctuating hydraulic gradients intensify internal erosion and degrade post-erosion soil strength.

Key Innovation: Coupled erosion and triaxial tests tie fines loss to conductivity, volume change, dilatancy, and peak-friction-angle reduction across loading patterns.

38. Surface ages of alluvial fans in the southern Central Andes recorded by ground- and space-based hyperspectral reflectance

Source: Earth Surface Processes and Landforms Type: Geomorphology hyperspectral remote sensing Geohazard Type: Sedimentary landforms Relevance: 5/10

Core Problem: Absolute dating and mapping of alluvial-fan surfaces are expensive and difficult to scale across large basins.

Key Innovation: Calibrates ground and satellite hyperspectral weathering signatures against cosmogenic ages to map fan-surface age at high spatial resolution.

39. RIPE++: Reinforced Keypoint Learning from Positive Pairs Only

Source: ArXiv (Geo/RS/AI) Type: Preprint Geohazard Type: None Relevance: 5/10

Core Problem: Robust keypoint and matching models usually require poses, depth, or carefully curated negative pairs.

Key Innovation: Derives a positive-pairs-only reinforcement learning reward and extends it to end-to-end sparse matching.

40. Unified and Efficient Point-Line Local Features

Source: ArXiv (Geo/RS/AI) Type: Preprint Geohazard Type: terrain and scene mapping Relevance: 5/10

Core Problem: Separate point and line pipelines are too heavy for real-time robust multi-view geometry.

Key Innovation: Jointly extracts points, lines, and descriptors in one lightweight network with accelerated line recovery.

41. Gravity-aware partially calibrated absolute pose estimation from affine- or rotation-covariant features

Source: ArXiv (Geo/RS/AI) Type: Preprint Geohazard Type: photogrammetry and mobile mapping Relevance: 5/10

Core Problem: Semi-calibrated absolute pose estimation under IMU gravity cues underuses local geometry encoded in features.

Key Innovation: Derives minimal gravity-aware solvers that estimate pose and focal length from one affine or two orientation correspondences.

42. Evaluating Neural Cartographic Relief Shading for Urban Environments: A Downtown Calgary Study Using High-Resolution DEM and DSM Data

Source: ArXiv (Geo/RS/AI) Type: Preprint Geohazard Type: terrain representation Relevance: 5/10

Core Problem: Neural relief shading trained on alpine terrain may fail on dense urban topography.

Key Innovation: Systematically compares analytical and neural hillshading under tuned parameters on high-resolution urban DEM and DSM data.

43. Ultra-High-Definition Restoration Transformers with Correlation Matching Transformation

Source: ArXiv (Geo/RS/AI) Type: Preprint Geohazard Type: weather-degraded imagery Relevance: 5/10

Core Problem: Ultra-high-definition restoration needs strong performance across multiple degradations without huge models.

Key Innovation: Coordinates high- and low-resolution learning spaces with correlation matching and adaptive channel modulation for compact restoration.

44. skchange: Fast and Flexible Algorithms for Changepoint Detection

Source: ArXiv (Geo/RS/AI) Type: Preprint Geohazard Type: monitoring and anomaly detection Relevance: 5/10

Core Problem: Provide fast, modular, and high-dimensional changepoint and anomaly-segment detection in a usable open-source package.

Key Innovation: Packages modern search methods, penalties, and high-dimensional change detectors into a composable scikit-learn-style library.

45. Learning piecewise-smooth dynamical systems

Source: ArXiv (Geo/RS/AI) Type: Preprint Geohazard Type: threshold and regime-switching processes Relevance: 5/10

Core Problem: Identify piecewise-smooth dynamical systems directly from trajectories, including switching boundaries and sliding motion.

Key Innovation: Separates hyperplane discovery from region-wise dynamics learning and introduces a discontinuity-aware neural architecture with theory.

46. Tracking Total Precipitable Water Vapor: A Multi-Instrument Comparative Analysis

Source: Remote Sensing (MDPI) Type: atmospheric moisture validation study Geohazard Type: extreme rainfall and atmospheric moisture Relevance: 5/10

Core Problem: Benchmark six remote-sensing instruments for PWV retrieval against reference radiosondes.

Key Innovation: Updated 2024-2025 multi-instrument comparison identifies GPS as the most robust high-accuracy PWV benchmark and highlights calibration needs.

47. Numerical Investigation on the Mechanical Properties and Damage Mechanisms of Xiyu Conglomerate Under Freeze-Thaw Cycles

Source: Rock Mechanics and Rock Engineering Type: Freeze-thaw rock damage numerical study Geohazard Type: cold-region rock deterioration Relevance: 5/10

Core Problem: Freeze-thaw degradation mechanisms in Xiyu conglomerate are hard to observe and quantify experimentally.

Key Innovation: Builds a CT-based discrete-element model coupling thermal conduction and ice-expansion damage to track crack evolution.

48. A Novel Dual-Configuration Peridynamics Method for Simulating Compression-Shear Fracture in Brittle Rocks

Source: Rock Mechanics and Rock Engineering Type: Rock fracture simulation method paper Geohazard Type: rock fracture mechanics Relevance: 5/10

Core Problem: Bond-based peridynamics struggles to represent independent shear failure in brittle rocks under compression-shear loading.

Key Innovation: Introduces a dual-configuration peridynamics formulation with Hoek-Brown shear failure and crack-contact friction.

49. Mechanisms of vegetation effects on hydrological processes in subgrade slopes: novel models for SWRC and moisture dynamics response analysis

Source: Journal of Hydrology Type: Slope hydrology modeling study Geohazard Type: slope moisture processes Relevance: 5/10

Core Problem: Vegetation effects on soil-water retention and moisture dynamics in engineered slopes are not well mechanistically represented.

Key Innovation: Develops new SWRC and moisture-response models to explain vegetation controls in subgrade slopes.

50. An adaptive PD-SBFEM coupling method incorporating quadtree refinement with application to efficient dynamic fracture analysis in rock-like materials

Source: Computers and Geotechnics Type: Dynamic fracture simulation method paper Geohazard Type: rock dynamic fracturing Relevance: 5/10

Core Problem: Efficiently simulating dynamic fracture in rock-like materials with local refinement remains challenging.

Key Innovation: Couples peridynamics and SBFEM with quadtree refinement for adaptive dynamic fracture analysis.

51. Real-time tracking of rock-joint morphology and contact evolution during direct shear using DEM and digital image processing

Source: Computers and Geotechnics Type: Fractured rock review and uncertainty study Geohazard Type: fractured rock flow and stability Relevance: 5/10

Core Problem: Predicting flow, transport, deformation, and uncertainty in fractured rock remains difficult across scales.

Key Innovation: Synthesizes recent field, laboratory, numerical, and machine-learning advances for fractured-rock prediction and uncertainty quantification.

52. A topology-preserving dimensional transformation framework for arbitrary non-planar three-dimensional crack propagation

Source: Computers and Geotechnics Type: 3D rock crack-propagation simulation method Geohazard Type: fractured rock mass failure Relevance: 5/10

Core Problem: Non-planar intersecting 3D crack growth in rock masses is hard to represent without topological instability.

Key Innovation: Uses an auxiliary thick-layer transformation to preserve topology while explicitly meshing and propagating complex crack networks.

53. SceneGTMM: A Conformal Mapping-based Scene-Aware Transferable GNN-Transformer Dual-Graph Interaction Framework for Map Matching

Source: ArXiv (Geo/RS/AI) Type: Preprint Geohazard Type: None Relevance: 4/10

Core Problem: Map matching needs stronger noise robustness, transfer across regions, and better interpretability.

Key Innovation: Combines conformal local coordinates, dual-graph GNN-Transformer fusion, and CRF decoding for transferable map matching.

54. Does Marginal Coverage Guarantee Class-Conditional Safety for Zero-Shot VLMs Under Shift?

Source: ArXiv (Geo/RS/AI) Type: Preprint Geohazard Type: None Relevance: 4/10

Core Problem: Marginal conformal coverage can mask severe worst-class failures under deployment shift.

Key Innovation: Provides a systematic audit showing tail-coverage collapse and compares source-side and target-side calibration remedies.

55. HiRA-CAM: Preserving Fine-Grained Spatial Relevance in Gradient-Based Visual Explanations

Source: ArXiv (Geo/RS/AI) Type: Preprint Geohazard Type: None Relevance: 4/10

Core Problem: Gradient-based CAM explanations blur or lose fine-grained spatial relevance in CNNs.

Key Innovation: Aggregates activation maps across CNN layers to preserve higher-resolution relevance in saliency maps.

56. When to Retrain: An Empirical Study of Retraining Policies for Streaming ML Under Concept Drift, Budget, and Latency Constraints

Source: ArXiv (Geo/RS/AI) Type: Preprint Geohazard Type: None Relevance: 4/10

Core Problem: Practitioners lack grounded guidance on when retraining helps under drift, budget, and deployment latency.

Key Innovation: Large controlled comparison shows incremental learning dominates retraining policy choice and exposes reactive-policy failure modes.

57. Where Grounding Accuracy Lives on the IoU Curve: Label-Free Inference-Time Boundary Refinement

Source: ArXiv (Geo/RS/AI) Type: Preprint Geohazard Type: None Relevance: 4/10

Core Problem: Grounding models often identify the right referent but return imprecise bounding boxes.

Key Innovation: Adds label-free localized re-observation with geometric guards and midpoint fusion to refine grounded boxes.

58. Complementary, Not Cumulative: Interaction Effects in Physics-Informed Neural Networks for Navier-Stokes Vortex Shedding

Source: ArXiv (Geo/RS/AI) Type: Preprint Geohazard Type: None Relevance: 4/10

Core Problem: It was unclear whether popular PINN training tricks actually compose on unsteady flow problems.

Key Innovation: Systematically tests interactions and finds that SIREN activations plus causal weighting are the load-bearing combination.

59. Rationally Enriched Chebyshev Trunk Bases for DeepONet Surrogates of High P\'eclet Entrance Transport

Source: ArXiv (Geo/RS/AI) Type: Preprint Geohazard Type: None Relevance: 4/10

Core Problem: DeepONet surrogates struggle with thin localized boundary layers in high-Peclet transport regimes.

Key Innovation: Adds rationally enriched Chebyshev trunk bases with AAA-derived rational dictionary elements.

60. Coupled Optimal Transport with Landmark Constraints

Source: ArXiv (Geo/RS/AI) Type: Preprint Geohazard Type: deformation mapping Relevance: 4/10

Core Problem: Pure cost-minimizing optimal transport can miss the physically meaningful deformation between shapes.

Key Innovation: Couples landmark-guided deformation fields with transport plans under a mutual-consistency constraint.

61. Simulation-to-Real First-Break Segmentation for Efficient Inversion in Musculoskeletal Ultrasound Tomography

Source: ArXiv (Geo/RS/AI) Type: Preprint Geohazard Type: subsurface imaging Relevance: 4/10

Core Problem: Weak noisy first arrivals undermine initialization and stability in ultrasound waveform inversion.

Key Innovation: Uses simulation-to-real first-break segmentation plus hybrid traveltime and waveform inversion to stabilize reconstructions.

62. Multi-Source Wasserstein Distributionally Robust Graph Learning

Source: ArXiv (Geo/RS/AI) Type: Preprint Geohazard Type: sensor-network analysis Relevance: 4/10

Core Problem: Naive pooling of heterogeneous source graph signals distorts topology inference when target data are scarce.

Key Innovation: Fuses sources with a Wasserstein barycenter and robust ambiguity set to estimate graphs under heterogeneity.

63. Point-Based 3D Reconstruction from Sparse Views under Known Illumination

Source: ArXiv (Geo/RS/AI) Type: Preprint Geohazard Type: terrain and outcrop reconstruction Relevance: 4/10

Core Problem: Sparse-view surface recovery often needs huge primitive counts or weak physical constraints.

Key Innovation: Optimizes compact opacity-bearing surfels with adjoint light-transport gradients for geometry recovery.

64. End-to-end Early Classification of Time Series in Non-Stationary Environments

Source: ArXiv (Geo/RS/AI) Type: Preprint Geohazard Type: streaming early warning Relevance: 4/10

Core Problem: Separately trained early classifiers and trigger policies adapt poorly under concept drift.

Key Innovation: Jointly learns representation, classification, and trigger timing with an end-to-end RL architecture.

65. Physical-Support Confidence Sets for Highly Coherent Dictionaries

Source: ArXiv (Geo/RS/AI) Type: Preprint Geohazard Type: generalizable spatial inference Relevance: 4/10

Core Problem: Sparse dictionary supports can look physically precise even when calibration uncertainty makes that interpretation unjustified.

Key Innovation: Builds cross-dictionary physical-support confidence sets and an active endpoint bracketing procedure that abstains from unsupported over-localization.

66. Active Spiking Perception: The Membrane Potential as a Belief State for Anytime 3D Point Cloud Recognition

Source: ArXiv (Geo/RS/AI) Type: Preprint Geohazard Type: LiDAR and 3D scene interpretation Relevance: 4/10

Core Problem: Use spiking point-cloud networks more efficiently by deciding what spatial chunks to inspect next and when to stop.

Key Innovation: Treats membrane potential as a Bayesian belief state to drive adaptive chunk selection and confidence-based early exit.

67. Heteroscedastic Neural Surrogate Modeling for Robust and Rapid Bayesian Inference in Fusion Plasma Diagnostics

Source: ArXiv (Geo/RS/AI) Type: Preprint Geohazard Type: general inverse modeling under noise Relevance: 4/10

Core Problem: Make MCMC-based inference practical for noisy physical diagnostics without losing uncertainty awareness.

Key Innovation: Uses a dual-head neural surrogate with heteroscedastic likelihood training to emulate forward models and intrinsic noise simultaneously.

68. Learning Deterministic and Stochastic Forced Hamiltonian Systems

Source: ArXiv (Geo/RS/AI) Type: Preprint Geohazard Type: thresholded physical dynamics Relevance: 4/10

Core Problem: Learn deterministic and stochastic forced Hamiltonian systems without losing geometric structure or long-horizon stability.

Key Innovation: Defines Lagrange-d'Alembert neural maps and structure-preserving forced Hamiltonian networks with convergence and approximation guarantees.

69. The impact of feature engineering and an optimisation framework for ocean colour machine learning

Source: ArXiv (Geo/RS/AI) Type: Preprint Geohazard Type: earth observation feature engineering Relevance: 4/10

Core Problem: Quantify how much feature engineering, not just model tuning, controls ocean-colour ML performance.

Key Innovation: Defines a seven-stage feature-engineering optimization pipeline and shows tuned FE can dominate model choice.

70. Gravitational-wave parameter estimation with machine-learning generated surrogate waveforms

Source: ArXiv (Geo/RS/AI) Type: Preprint Geohazard Type: method transfer Relevance: 4/10

Core Problem: Waveform generation is too expensive for large-scale gravitational-wave parameter estimation.

Key Innovation: Two-stage conditional autoencoder surrogate with residual calibration and posterior bias correction.

71. Transfer Learning in Nonparametric Regression with Deep ReLU Networks

Source: ArXiv (Geo/RS/AI) Type: Preprint Geohazard Type: method transfer Relevance: 4/10

Core Problem: Learn group-specific nonparametric regressions while exploiting shared structure across groups.

Key Innovation: Two-stage pooled mean plus group-offset framework with error bounds for deep ReLU models.

72. Multi-Modal Graph Interaction for Multi-Graph Convolution Network in Urban Spatiotemporal Forecasting

Source: ArXiv (Geo/RS/AI) Type: Preprint Geohazard Type: method transfer Relevance: 4/10

Core Problem: Fuse several spatial relationship graphs in region-level forecasting.

Key Innovation: Grouped GCN plus higher-layer multi-linear relationship GCN for multimodal interaction.

73. ReAugment: Model Zoo-Guided RL for Few-Shot Time Series Augmentation and Forecasting

Source: ArXiv (Geo/RS/AI) Type: Preprint Geohazard Type: method transfer Relevance: 4/10

Core Problem: Improve forecasting when training data are scarce and overfitting is likely.

Key Innovation: RL-guided augmentation that selects overfit-prone anchors using model-zoo diversity.

74. Table2Image: Lightweight Tabular Learning with Generated Proxy Representations and Reliability Diagnostics

Source: ArXiv (Geo/RS/AI) Type: Preprint Geohazard Type: method transfer Relevance: 4/10

Core Problem: Balance predictive performance, compactness, and robustness in tabular learning.

Key Innovation: Generated proxy representations plus severity-controlled reliability diagnostics.

75. Deformation and global buckling analysis of long-distance shallowly embedded subsea pipeline under strike-slip fault displacement

Source: Ocean Engineering Type: fault-displacement engineering Geohazard Type: fault deformation Relevance: 4/10

Core Problem: Assess subsea pipeline deformation and buckling under strike-slip fault movement.

Key Innovation: Global buckling analysis for long shallowly embedded pipelines crossing active faults.

76. Multi-model hydrological seasonal forecasts for Europe in the C3S climate data store

Source: ESSD Type: seasonal hydrological forecasting Geohazard Type: hydrologic forecasting Relevance: 4/10

Core Problem: Provide and evaluate Europe-wide seasonal forecasts of flow, runoff, snow, and soil moisture.

Key Innovation: Operational multi-model service with monthly updates and six-month outlooks in C3S.

77. Grounding-line dynamics in a Stokes ice-flow model (Elmer/Ice v9.0): improved numerical stability allows larger time steps

Source: Geoscientific Model Development Type: cryosphere model numerics Geohazard Type: cryosphere hazard transfer Relevance: 4/10

Core Problem: Remove restrictive time-step limits in Stokes ice-sheet grounding-line simulations.

Key Innovation: Ice-ocean interface stabilization that permits larger stable time steps.

78. Optimizing Gaussian process emulation and generalized additive model fitting for rapid, reproducible earth system model analysis

Source: Geoscientific Model Development Type: earth system model emulation Geohazard Type: climate model method transfer Relevance: 4/10

Core Problem: Speed up reproducible analysis of large earth-system model ensembles.

Key Innovation: Gaussian-process emulation and GAM workflow up to 25 times faster without accuracy loss.

79. Process diagnostics of snowmelt runoff in global hydrological and land surface models - Part 1: A systematic evaluation across basins of increasing complexity

Source: Hydrology and Earth System Sciences Type: snowmelt hydrology evaluation Geohazard Type: hydrologic process Relevance: 4/10

Core Problem: Diagnose how large-scale models simulate snowmelt runoff volume, peak, and timing.

Key Innovation: Systematic process evaluation of 15 models across 1,455 snow-dominated basins.

80. Slow bleaching of water-insoluble brown carbon from biomass burning: Implication for direct radiative effect

Source: Science Advances Type: atmospheric brown-carbon process model Geohazard Type: wildfire smoke and climate forcing Relevance: 4/10

Core Problem: Represent bleaching of water-insoluble brown carbon across the full biomass-burning aerosol matrix.

Key Innovation: Develops humidity- and temperature-sensitive bleaching parameterization that materially increases estimated BrC radiative effects.

81. Evaluating ENSO-vegetation interactions and variability using remote sensing-derived vegetation indices and principal component analysis

Source: Frontiers in Earth Science Type: climate-vegetation remote-sensing analysis Geohazard Type: drought and ecosystem stress Relevance: 4/10

Core Problem: Assess how ENSO phases modulate vegetation response across monsoon-dependent agro-climatic zones.

Key Innovation: Builds a PCA-based vegetation activity component and lagged correlation maps to identify phase-specific stress and recovery hotspots.

82. Dual-Level Spatial-Frequency Collaborative Detector for Oriented Object Detection in Remote Sensing Images

Source: Remote Sensing (MDPI) Type: remote-sensing object-detection method Geohazard Type: hazard inventory mapping transfer Relevance: 4/10

Core Problem: Improve oriented object detection when targets are small, rotated, and weakly represented in frequency space.

Key Innovation: Unified image-level and instance-level wavelet spatial-frequency fusion preserves both global scene semantics and local target detail.

83. HRRP Reconstruction Method for Coded Interrupted Sampling Radar Echoes Based on Multi-Frame Sequential Priors

Source: Remote Sensing (MDPI) Type: radar signal reconstruction method Geohazard Type: radar monitoring transfer Relevance: 4/10

Core Problem: Recover accurate HRRPs from coded interrupted-sampling radar echoes under low duty ratio and low SNR.

Key Innovation: Candidate-interval-assisted OMP injects multi-frame sequential priors to improve support recovery while shrinking atom search space.

84. DA-GDNet: A Data-Augmented Gather-and-Distribute Network for Robust SAR Target Detection

Source: Remote Sensing (MDPI) Type: SAR target-detection method Geohazard Type: all-weather hazard target mapping transfer Relevance: 4/10

Core Problem: Improve SAR target detection when labeled samples poorly cover target types, angles, and backgrounds.

Key Innovation: Pairs 3D-model-based data augmentation with gather-distribute multi-scale fusion and spatial enhancement modules.

85. Development and Verification of an Automatic Tower-Based SIF Observation System Based on Narrow Field-of-View Scanning and DOAS Atmospheric Correction

Source: Remote Sensing Type: tower-based vegetation fluorescence sensing Geohazard Type: drought and vegetation-stress monitoring Relevance: 4/10

Core Problem: Stabilize tower-based SIF retrieval under variable atmospheric water-vapor conditions.

Key Innovation: DOAS-based atmospheric correction with adaptive reference spectra enables automated high-fidelity narrow-FOV SIF monitoring.

86. SFSMamba-DETR: Selective Feature Scanning with State Space Models and Dual-Scale Window Attention for Remote Sensing Object Detection

Source: Remote Sensing (MDPI) Type: remote-sensing object-detection architecture Geohazard Type: hazard inventory mapping transfer Relevance: 4/10

Core Problem: Improve remote-sensing object detection under scale variation, dense small targets, and high computational cost.

Key Innovation: Combines Mamba selective scanning, dual-scale window attention, and cross-scale aggregation for efficient detection.

87. Integrating Kernel-Based Vegetation Indices and Ensemble Learning for Mangrove Canopy Height Mapping Using GEDI and Sentinel Data

Source: Remote Sensing (MDPI) Type: mangrove structure mapping Geohazard Type: coastal resilience and ecosystem buffering Relevance: 4/10

Core Problem: Retrieve mangrove canopy height where optical saturation and tidal-zone noise degrade estimates.

Key Innovation: Kernel-based vegetation indices act as nonlinear stabilizers when fused with GEDI and Sentinel data for more robust canopy-height mapping.

88. Altered streamflow dynamics following Hurricane Michael and their linkage to forest damage in South Carolina, USA

Source: Natural Hazards Type: post-hurricane watershed hydrology Geohazard Type: storm runoff response Relevance: 4/10

Core Problem: Separate hurricane impacts on quickflow and baseflow recovery and relate them to forest damage.

Key Innovation: Component-wise recovery analysis across 21 watersheds shows much slower baseflow recovery and links stronger changes to bottomland forest damage.

89. Real-time prediction of the natural decline rate of groundwater inflow from a fault zone or fractures during tunnel excavation and application to deciding on whether to implement grouting

Source: Bulletin of Engineering Geology and the Environment Type: tunnel groundwater-inflow prediction Geohazard Type: fractured-rock groundwater inflow Relevance: 4/10

Core Problem: Predict whether groundwater inflow from faulted rock will decline naturally enough to avoid tunnel pre-grouting.

Key Innovation: Uses power-law log-log inflow decay from pilot boreholes to make rapid grouting decisions during excavation.

90. Structural and thermal analysis of geothermal prospectivity in the Midyan Basin, NEOM, Saudi Arabia

Source: Environmental Earth Sciences Type: Integrated remote-sensing structural prospectivity study Geohazard Type: fault and fracture mapping Relevance: 4/10

Core Problem: Geothermal fluid pathways in the Midyan Basin are poorly constrained despite tectonic activity.

Key Innovation: Fuses multi-season Landsat thermal anomalies, DEM lineaments, magnetic lineaments, and fracture-density mapping to rank target corridors.

91. Exploration of Point Load Testing under In Situ Pore Pressure: Mechanical Behaviour and Systematic Characterization

Source: Rock Mechanics and Rock Engineering Type: In-situ rock core testing method Geohazard Type: deep rock characterization Relevance: 4/10

Core Problem: Deep cores lose in-situ stress and pore conditions before mechanical testing, biasing parameter estimates.

Key Innovation: Embeds point-load testing inside a condition-preserved coring tool to measure rock mechanics immediately after retrieval.

92. Mechanisms and Applications of Gas Flow in Coal Seams Considering the Effects of Cross-Section Shape and Surface Roughness

Source: Rock Mechanics and Rock Engineering Type: Coal seam gas-flow modeling study Geohazard Type: coal-seam dynamic disaster mitigation Relevance: 4/10

Core Problem: Simplified pore geometry assumptions distort permeability predictions in rough coal-seam pore networks.

Key Innovation: Builds a geometry-aware permeability model coupling roughness, adsorption, stress, and pore shape, then links it to fracturing plus CO2 injection.

93. RIPC: A novel FFT-Based multimodal image matching framework with radiometric, scale and rotation invariance

Source: ISPRS Journal of Photogrammetry and Remote Sensing Type: Remote-sensing image matching method Geohazard Type: geospatial image registration Relevance: 4/10

Core Problem: Multimodal image matching remains sensitive to radiometric, scale, and rotation differences.

Key Innovation: Introduces an FFT-based framework designed for invariant multimodal image matching.

94. Vis-NIR, MIR, and XRF spectral data fusion for predicting soil weathering indices - A case study using 150 Alfisols

Source: CATENA Type: Proximal sensing soil-weathering study Geohazard Type: weathering and regolith characterization Relevance: 4/10

Core Problem: Rapid estimation of soil weathering indices across contrasting horizons and parent materials is difficult.

Key Innovation: Shows MIR-centered spectral fusion with Vis-NIR and XRF can predict multiple weathering indices accurately.

95. Performance analysis of the OS LISFLOOD hydrological model using nearly three decades of satellite laser ranging data

Source: Journal of Hydrology Type: Hydrological model evaluation study Geohazard Type: flood hydrology Relevance: 4/10

Core Problem: Long-term large-scale hydrological model behavior needs independent evaluation with geodetic observations.

Key Innovation: Assesses OS LISFLOOD using nearly three decades of satellite laser ranging data.

96. Dealing with nonlattice data in three-dimensional probabilistic site characterizations using 3D-MSBL

Source: Computers and Geotechnics Type: Probabilistic site characterization method paper Geohazard Type: 3D subsurface uncertainty modeling Relevance: 4/10

Core Problem: Three-dimensional probabilistic site characterization methods often assume lattice-structured data that field campaigns do not provide.

Key Innovation: Extends 3D-MSBL to handle nonlattice subsurface data for probabilistic ground characterization.

97. An Integrated Method for Identification and Predictive Modeling of Rock Discontinuities in Tunnels: Methodology and Experimental Validation

Source: Journal of Rock Mechanics and Geotechnical Engineering Type: Rock discontinuity identification and modeling method Geohazard Type: rock mass structural characterization Relevance: 4/10

Core Problem: Reliable identification and prediction of rock discontinuities in tunnels is methodologically challenging.

Key Innovation: Presents and experimentally validates an integrated workflow for detecting and predictive modeling of tunnel rock discontinuities.