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

TerraMosaic Daily Digest: August 26, 2026

August 26, 2026 TerraMosaic Daily Digest

Daily Summary

Direct geohazard studies converge on hidden internal states that govern whether slopes fail, how they deform, and how fast cascading impacts unfold. In landslide mechanics, water softening, weak-layer geometry, and clay degradation reorganize failure kinematics: thicker overlying soil shifts trailing-edge cracks from tension toward shear in soil-bedrock slides, while strain-softened claystone can laterally extrude under lithostatic load and rotate sandstone slabs upslope, resolving the under-dip toppling energy paradox in flysch terrains. Susceptibility studies likewise favor explicit spatial structure, either by fusing local geo-environmental variables with context or by showing that roads, rivers, and tectonic lineaments dominate Himalayan slope propensity. Event-scale and monitoring papers add operational depth: the 16 August 2024 Thame Valley cascade reconstructed a combined glacial-lake release of about 770,000 m³ that reached Thame within 22-25 min; rapid mapping in Emilia-Romagna shows that satellite and aerial automation can still support response under scarce labels; sinkhole zoning in Sardinia sharpens when ambient-noise, nanoseismic, and repeated topographic surveys are combined; and physically informed slope-deformation inversion, electrochemical weak-layer reinforcement, root-specific interface strengthening, and bench-frequency analysis all make stability assessment more process-explicit.

Earthquake, flood, and atmospheric papers extend the same shift from coarse descriptors to process-resolved hazard structure. Laboratory earthquakes validate the slow unsteady dynamics of frictional slip pulses, and analysis of 77 western Canada injection-induced sequences finds that 92% of local magnitude 3 or greater mainshocks were preceded by foreshocks consistent with three fluid-mediated nucleation modes. A PGA-stratified framework for the 2025 Dingri earthquake separates landslide controls from liquefaction controls instead of pooling them into one secondary-hazard relation, while site-response estimation in the Bornova Basin, hybrid-fiber frame fragility, storey loss and environmental-impact functions, masonry retrofits, and community-led preparedness in Nepal connect rupture, amplification, damage, and recovery. For hydroclimatic hazards, Chinese studies link flood disasters to the 3D geometry of extreme-precipitation events and show that monsoon-region precipitation change is increasingly shaped by torrential-rain frequency; typhoon gust forecasting, single-radar 4D wind retrieval, MTG-FCI temperature-humidity profiling, and continuous regional temperature fields strengthen the forcing information needed for wind, flood, and heat-risk analysis, while cryosphere and hydroclimate papers track thermokarst-lake reorganization, rock-glacier chronology, vegetation-driven precipitation recycling, and unresolved sub-shelf melt as changing background conditions.

A large secondary stream expands transferable capability without itself constituting geohazard validation. Foundation-model-conditioned precipitation downscaling, survey-scale airborne electromagnetic inversion, geometry-aware first-arrival picking, ROM-based waveform inversion, PINN-based potential-field continuation, and reduced-order geotechnical surrogates all embed physical constraints to gain speed, scale, or robustness. Remote-sensing and geospatial papers similarly emphasize multimodal fusion, geometry, and uncertainty across hyperspectral-LiDAR subpixel mapping, HSI-SAR-LiDAR fusion, UAV disaster segmentation and small-object detection, cross-view localization, 4D LiDAR annotation, underwater RGB-sonar detection and 3D reconstruction, SWOT-derived wave, water-level, and surface-velocity products, building-footprint benchmarking, roof-graph synthesis, and floorplan vectorization; planetary geospatial prediction, forecast-discussion grounding, low-cost gas-sensor analytics, structural inspection workflows, and on-chip hyperspectral imaging push that automation closer to environmental operations. Across the scientific-ML cohort, operator learning, PINN pruning and post-training correction, constrained Gaussian processes, stochastic-dynamics uncertainty, and conformal-region geometry all point toward more stable and auditable surrogates, but these advances should be read as enabling methods unless a hazard application is explicitly tested.

Key Trends

Five scientific and methodological trajectories define the August 26 literature: state-dependent slope failure, multimodal process diagnosis, rupture-to-recovery earthquake assessment, structure-aware hydrometeorological forcing, and physics- and uncertainty-aware scientific AI.

  • Slope failure is being reframed as internal-state evolution: Across toppling, soil-bedrock sliding, ecological stabilization, mine-slope reinforcement, and bench-slope vibration, the decisive variables are weak-layer softening, crack-mode partition, root growth stage, pore-scale cementation, resonance, and material weathering rather than trigger totals alone. The same emphasis on internal structure extends to hydrophobic wetting curves, freeze-thaw creep, tailings behavior, and cold-region rock damage.
  • Hazard observation is moving toward multimodal process diagnosis: The strongest monitoring papers do more than map surface change. The Thame GLOF reconstruction, Sardinian sinkhole model, sparse-data slope deformation inversion, and Emilia-Romagna rapid mapping all combine complementary evidence streams, while radar wind retrieval, MTG profile inversion, SWOT products, UAV segmentation, and standalone hyperspectral hardware show how dense, fast sensing can be operationalized even when the methods themselves are not geohazard-specific.
  • Earthquake research spans rupture physics to community recovery: The earthquake cohort links laboratory rupture mechanics, induced-seismic foreshock productivity, PGA-conditioned secondary hazards, basin-scale site response, frame fragility, loss functions, masonry retrofits, and community-led preparedness. The result is a notably continuous chain from nucleation and wave amplification to structural consequences and the social conditions of resilience.
  • Hydrometeorological forcing is becoming finer, probabilistic, and structure-aware: Flood and weather papers increasingly resolve hazard forcing as structured fields rather than single indices. Recent work couples 3D precipitation-event analysis, changing Chinese precipitation magnitudes, foundation-model-conditioned downscaling, typhoon gust prediction, physics-informed wind-field retrieval, geostationary temperature-humidity profiling, flexible regional temperature queries, and SST prediction, with cryosphere and water-balance studies supplying longer background context.
  • Transferable scientific AI is privileging physics, geometry, and uncertainty: Across neural operators, PINNs, constrained Gaussian processes, conformal regions, geophysical inversion, remote-sensing fusion, localization datasets, LiDAR annotation, and geospatial foundation-model workflows, the dominant methodological move is to encode invariants, geometry, stability, or calibrated uncertainty directly into the model or evaluation. Many of these studies are promising for hazards, but their claims remain transferable rather than hazard-validated unless field application is shown.

Selected Papers

The selected papers combine domain-tested studies of landslides, earthquakes, floods, GLOFs, sinkholes, wind hazards, and slope mitigation with a broader methods cohort in geophysical inversion, remote sensing, scientific machine learning, and geospatial reconstruction. Read the first group as validated hazard evidence and the second as transferable analytical capability whose geohazard value depends on explicit application and testing.

1. Lab earthquakes confirm the theory of frictional slip pulses

Source: ArXiv (Geo/RS/AI) Type: Preprint Geohazard Type: earthquake Relevance: 9/10

Core Problem: Theoretical predictions for frictional slip pulses needed decisive experimental validation across rupture conditions.

Key Innovation: Lab earthquakes confirm slow unsteady pulse dynamics and the transition between decaying and growing slip pulses predicted by theory.

2. Mobilized-clay-driven toppling: A new mechanism resolving the “energy paradox” of under-dip slope failure in flysch terrains

Source: Landslides Type: landslide mechanism study Geohazard Type: rock toppling landslide Relevance: 9/10

Core Problem: Resolve the mechanical 'energy paradox' of uphill-rotating toppling failures in under-dip flysch slopes.

Key Innovation: Introduces mobilized-clay-driven toppling, where strain-softened claystone laterally extrudes under lithostatic stress and forces sandstone slab rotation.

3. Synergising Local Geo-Environmental Characteristics with Spatial Context for Enhancing Landslide Susceptibility Mapping

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

Core Problem: Pixel models miss surrounding context while patch models include many weakly relevant neighboring pixels.

Key Innovation: Feature-wise fusion of local geo-environmental variables and spatial context boosts CNN-based landslide mapping.

4. Rapid landslide mapping during the 2023 Emilia-Romagna disaster: assessing automated approaches with limited training data

Source: Natural Hazards and Earth System Sciences Type: rapid landslide mapping Geohazard Type: landslide Relevance: 8/10

Core Problem: Manual mapping of tens of thousands of landslides after extreme rainfall is too slow for rapid response.

Key Innovation: An empirical assessment of automated satellite and aerial-image mapping approaches for the 2023 Emilia-Romagna disaster under scarce training data.

5. Unveiling the link between extreme precipitation events and flood disasters in China: from a 3D perspective

Source: Natural Hazards and Earth System Sciences Type: flood-disaster event linkage Geohazard Type: flood Relevance: 8/10

Core Problem: The relationship between extreme precipitation events and resulting flood disasters is complex and insufficiently characterized.

Key Innovation: A 3D connected-component analysis linking flood-causing precipitation events with historical flood disasters and their controlling factors across China.

6. The 2024 cascading glacial lake outburst flood in the Thame Valley of Everest region, Nepal: process, impacts and implications

Source: Natural Hazards and Earth System Sciences Type: GLOF process and impact study Geohazard Type: glacial lake outburst flood Relevance: 8/10

Core Problem: Small, rapidly evolving glacial lakes can trigger cascading failures that are underestimated in mountain hazard assessment.

Key Innovation: A detailed reconstruction of the August 16, 2024 Thame Valley cascading GLOF, including breach sequence, released volume, travel time, peak discharge, and damage implications.

7. Foreshock productivity and rupture nucleation in injection-induced earthquakes in western Canada

Source: Science Type: research_article Geohazard Type: induced earthquake Relevance: 8/10

Core Problem: Foreshock behavior before injection-induced mainshocks is poorly constrained for traffic-light protocols.

Key Innovation: Analyzes 77 western Canada sequences to quantify foreshock productivity and define three fluid-driven rupture nucleation models.

8. Understanding the role of lineament-controlled factors in landslide susceptibility: a machine learning approach in the Uhl Basin, Himachal Pradesh

Source: Frontiers in Earth Science Type: research_article Geohazard Type: landslide Relevance: 8/10

Core Problem: Determines how lineaments, roads, rivers, and tectonic activity control landslide susceptibility in the Uhl Basin.

Key Innovation: Combines RF and MLP susceptibility modeling with SHAP interpretation and active-tectonics metrics to show lineament dominance.

9. Effect of the thickness of overlying soil on the location of trailing edge cracks in soil-bedrock landslides induced by water softening

Source: Bulletin of Engineering Geology and the Environment Type: landslide mechanics experiment/model Geohazard Type: soil-bedrock landslide Relevance: 8/10

Core Problem: Explain how overlying-soil thickness controls trailing-edge crack location in soil-bedrock landslides.

Key Innovation: Combines 11 laboratory slope experiments with a segmented integral theoretical model that predicts trailing-edge crack position and failure-mode shifts.

10. Ground-shaking-informed regime partitioning for characterizing earthquake-induced secondary hazards

Source: Geoenvironmental Disasters Type: interpretable multi-hazard ML mapping Geohazard Type: coseismic landslide and liquefaction Relevance: 8/10

Core Problem: Earthquake-induced landslides and liquefaction are controlled by different factor combinations that standard single-regime models can blur.

Key Innovation: Uses PGA-guided regime partitioning plus regime-wise XGBoost and SHAP to reveal hazard-specific controls across shaking environments.

11. FEM-Graph Convolutional Network for Deformation Field Estimation of Slope with Limited Monitoring Data

Source: Rock Mechanics and Rock Engineering Type: physics-guided slope monitoring ML Geohazard Type: slope deformation or landslide Relevance: 8/10

Core Problem: Recover a high-precision 3D slope deformation field when monitoring points are limited and pure data-driven methods distort blind zones.

Key Innovation: Fuses FEM-derived physical priors, variogram-informed residual graphs, and graph convolution to estimate global deformation fields with improved accuracy and interpretability.

12. Engineering geological model of the Cixerri Plain (Sardinia, Italy) based on integrated geophysical investigations for sinkhole hazard assessment

Source: Engineering Geology Type: sinkhole hazard assessment Geohazard Type: sinkhole Relevance: 8/10

Core Problem: Detect subsiding sectors and subsurface cavity evolution in a sinkhole-prone plain with complex cover-collapse and piping behavior.

Key Innovation: Integrates ambient-noise resonance, nanoseismicity, and multitemporal LiDAR/GNSS/UAV topography into a depth-dependent engineering geological hazard model.

13. Frequency-aware forecasting for short-term typhoon gust prediction

Source: ArXiv (Geo/RS/AI) Type: Preprint Geohazard Type: typhoon and windstorm Relevance: 7/10

Core Problem: Short-term typhoon gusts are nonstationary and hard to predict during extremes.

Key Innovation: Wavelet-decomposed dual-branch forecasting separates trends from fast fluctuations.

14. Precipitation Downscaling Using Foundation Model-Conditioned Diffusion

Source: ArXiv (Geo/RS/AI) Type: Preprint Geohazard Type: flood and landslide rainfall forcing Relevance: 7/10

Core Problem: Downscale coarse daily precipitation fields to high resolution while retaining extremes.

Key Innovation: Compares diffusion conditioning strategies and shows foundation-model cross-attention helps realism and data-limited training.

15. Planetary Prediction Engine: Autonomous Geospatial Prediction via Intelligent Data Selection and Foundation Model Embeddings

Source: ArXiv (Geo/RS/AI) Type: Preprint Geohazard Type: disaster risk and planetary prediction Relevance: 7/10

Core Problem: Automate geospatial data discovery, fusion, and model selection directly from natural-language tasks.

Key Innovation: End-to-end planetary prediction agent combining open geodata retrieval, EO foundation-model embeddings, and automated model search with overfitting guards.

16. Hypervision: An on-chip hyperspectral microsystem for online video-rate computational imaging

Source: Science Type: research_article Geohazard Type: hazard monitoring instrumentation Relevance: 7/10

Core Problem: Hyperspectral imaging is computationally too heavy for standalone real-time operation.

Key Innovation: Integrates a hyperspectral sensor, on-chip neural processor, and pruned reconstruction network for video-rate standalone imaging.

17. Field application and microscopic reinforcement mechanism of electrochemical technology for weak interlayer soft rock slopes in open-pit coal mines

Source: Frontiers in Earth Science Type: research_article Geohazard Type: slope instability Relevance: 7/10

Core Problem: Weak mudstone interlayers in open-pit soft rock slopes trigger subsidence-sliding failures and are hard to reinforce.

Key Innovation: Applies optimized electrochemical reinforcement with field monitoring and microstructural validation to raise slope safety factors.

18. DC4Former: Orientation-Stable UAV Disaster Image Segmentation via Diagonal-Complemented C4 Consistency

Source: Remote Sensing (MDPI) Type: research_article Geohazard Type: disaster image segmentation Relevance: 7/10

Core Problem: UAV disaster image segmentation is unreliable when scene orientation changes with flight heading and camera yaw.

Key Innovation: Introduces C4-consistent operators plus diagonal training exposure to stabilize segmentation across principal image orientations.

19. Community-led earthquake preparedness in Nepal: a decolonial conversational study of trauma, governance, and resilience

Source: Natural Hazards Type: earthquake resilience governance study Geohazard Type: earthquake Relevance: 7/10

Core Problem: Understand how marginalized communities experience earthquake impacts and where institutional preparedness and recovery fail them.

Key Innovation: Applies a decolonial conversational approach to foreground community-led, especially women-led, preparedness and resilience strategies.

20. Time-dependent mechanical effects of tall fescue and lucerne roots on interface layers of ecological slopes in the loess region

Source: Bulletin of Engineering Geology and the Environment Type: bio-mediated slope stabilization study Geohazard Type: shallow landslide or slope instability Relevance: 7/10

Core Problem: Quantify how root type, growth stage, and water content change reinforcement of the weak substrate-soil interface on loess slopes.

Key Innovation: Distinguishes short-term fibrous-root reinforcement from longer-term taproot reinforcement and fits predictive strength models using root area ratio, age, and water content.

21. Seismic fragility assessment of multi-scale hybrid fiber-reinforced concrete frame structures

Source: Bulletin of Earthquake Engineering Type: seismic fragility and materials study Geohazard Type: earthquake Relevance: 7/10

Core Problem: Assess whether a multi-scale hybrid fiber-reinforced concrete system improves frame seismic resilience and fragility.

Key Innovation: Pairs a new HFRC constitutive model with nonlinear time-history and IDA analyses to show reduced severe-damage and collapse exceedance probabilities.

22. GeoFormer: Geometry-Aware Transformer and its application to 5D First-Arrival Picking

Source: ArXiv (Geo/RS/AI) Type: Preprint Geohazard Type: subsurface seismic characterization Relevance: 6/10

Core Problem: Improve noisy prestack seismic first-arrival picking using acquisition geometry.

Key Innovation: Transformer with token-, normalization-, and attention-level geometry injection.

23. Tropospheric temperature and humidity profile retrieval from Meteosat Flexible Combined Imager based on deep learning

Source: ArXiv (Geo/RS/AI) Type: Preprint Geohazard Type: weather and rainfall-triggered hazards Relevance: 6/10

Core Problem: Retrieve all-sky tropospheric temperature and humidity profiles from Meteosat FCI without NWP backgrounds.

Key Innovation: Spatially aware Residual U-Net using all 16 channels to derive profile fields independently.

24. Learning Continuous Regional Temperature Fields with Lead-Time and Resolution Queries

Source: ArXiv (Geo/RS/AI) Type: Preprint Geohazard Type: temperature-related hazard forecasting Relevance: 6/10

Core Problem: Forecast 2 m temperature continuously across lead times and spatial resolutions.

Key Innovation: Query-conditioned neural field with consistency losses across time, space, and scale.

25. Continually learning neural-operator surrogate for three-dimensional airborne electromagnetic Bayesian inversion

Source: ArXiv (Geo/RS/AI) Type: Preprint Geohazard Type: subsurface conductivity imaging Relevance: 6/10

Core Problem: Make Bayesian inversion of 3D airborne electromagnetic data computationally feasible at survey scale.

Key Innovation: Continually learning neural-operator surrogate with validity checking inside MCMC inversion.

26. ROMNet: a hybrid reduced order modeling and machine learning approach to waveform inversion

Source: ArXiv (Geo/RS/AI) Type: Preprint Geohazard Type: subsurface and seismic imaging Relevance: 6/10

Core Problem: Waveform inversion is computationally expensive and vulnerable to cycle skipping when recovering wave speed in heterogeneous media.

Key Innovation: Uses a neural network to map reduced-order-model matrices to simpler nearby operators that enable cheaper wave-speed recovery.

27. SwinMSNet: A Swin Transformer-Based Multiscale Network for Spatio-Environmental Changes

Source: IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing Type: landslide change detection Geohazard Type: landslide Relevance: 6/10

Core Problem: Remote-sensing change detection misses small or incomplete environmental changes that matter in landslide and disaster contexts.

Key Innovation: A Swin Transformer multiscale change detector with cross-scale fusion, contextual enhancement, and false-positive-guided label refinement.

28. Two-Stage Physics-Informed Neural Network for 4-D Wind Field Reconstruction From Single-Doppler Phased-Array Radar Observations

Source: IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing Type: radar wind-field reconstruction Geohazard Type: storm wind / severe weather Relevance: 6/10

Core Problem: Single-Doppler radar wind retrieval with PINNs is hampered by nonconvex optimization and underestimation of peak velocities.

Key Innovation: A two-stage PINN that pretrains on a proxy background field and then enforces radar observations plus anelastic Navier-Stokes physics for 4D retrieval.

29. A review of hydrological monitoring and early warning technologies and equipment for geological hazards

Source: Environmental Earth Sciences Type: geohazard monitoring review Geohazard Type: multi-hazard geological hazards Relevance: 6/10

Core Problem: Monitoring and warning technologies for hydrologically triggered geological hazards have evolved unevenly and remain fragmented.

Key Innovation: Reframes the review around an equipment-development timeline and proposes a next-generation roadmap for full-area perception, intelligent fusion, and resilient early warning.

30. Stability Analysis of Multi-Stage Bench Slopes based on Frequency Characteristics

Source: Rock Mechanics and Rock Engineering Type: bench slope dynamic stability analysis Geohazard Type: blast-induced slope instability Relevance: 6/10

Core Problem: Assess how blasting-induced vibrations affect the stability of multi-stage bench slopes with different bench-specific frequency responses.

Key Innovation: Uses natural-frequency characterization of individual benches to define a critical frequency and run bench-aware dynamic stability analysis.

31. Storey loss and environmental impact functions for existing RC buildings

Source: Bulletin of Earthquake Engineering Type: seismic loss modeling study Geohazard Type: earthquake Relevance: 6/10

Core Problem: Need transferable storey-level economic-loss and environmental-impact functions for poorly detailed existing RC buildings.

Key Innovation: Generalizes storey loss and environmental impact functions from fragility and consequence models and demonstrates country-specific adaptation.

32. Experimental study of confined masonry-wall retrofitted using various ferrocement configuration

Source: Bulletin of Earthquake Engineering Type: seismic retrofit experiment Geohazard Type: earthquake Relevance: 6/10

Core Problem: Evaluate whether ferrocement retrofit materially improves confined masonry wall performance under lateral cyclic loading.

Key Innovation: Full-scale tests compare retrofit configurations and derive a simple in-plane strength method spanning shear- and flexure-dominated failure.

33. Estimation of S-wave velocity structure and site response characteristics by microtremor array explorations in the Bornova Basin (İzmir, Türkiye)

Source: Engineering Geology Type: seismic site-response characterization Geohazard Type: earthquake Relevance: 6/10

Core Problem: Estimate S-wave velocity structure and site-response characteristics in the Bornova Basin.

Key Innovation: Microtremor array exploration appears to be the central method; the supplied abstract does not state additional methodological novelty.

34. FORMSpoT: Revealing fine-scale forest disturbances from nation-wide 1.5 m forest canopy height time series

Source: Remote Sensing of Environment Type: forest disturbance remote sensing Geohazard Type: indirect land-surface change relevance Relevance: 6/10

Core Problem: Detect fine-scale forest disturbances from nationwide canopy-height time series.

Key Innovation: The title indicates a 1.5 m canopy-height time-series framework for disturbance revelation, without an explicit geohazard application.

35. The Frame Kernel Method for Multiscale Operator Learning

Source: ArXiv (Geo/RS/AI) Type: Preprint Geohazard Type: cross-cutting Relevance: 5/10

Core Problem: Learn accurate operators for inherently multiscale PDE systems on grids and point clouds.

Key Innovation: A multiscale kernel-frame approximation learns frame coefficients and preserves decomposition at inference.

36. What Should a Large Language Model See? Physical Invariants as a Data Representation for PDE Discovery

Source: ArXiv (Geo/RS/AI) Type: Preprint Geohazard Type: cross-cutting Relevance: 5/10

Core Problem: Raw spatiotemporal fields are hard to feed into LLMs for automated PDE discovery.

Key Innovation: Adds a data-interpretation stage that converts fields into physical invariants before prompting the model.

37. OpenCVL: An Open, Diverse, and Large-Scale Dataset for Fine-Grained Cross-View Localization

Source: ArXiv (Geo/RS/AI) Type: Preprint Geohazard Type: general geospatial mapping Relevance: 5/10

Core Problem: Cross-view localization lacks open diverse data with reliable wild geotags.

Key Innovation: Builds a 617k-pair open dataset with pose curation and snow and cross-area tests.

38. PIVOT: A Multi-Trajectory Dataset and Testbed for Pose, Intrinsics, and Novel Viewpoint Evaluation in Real-World 3D Reconstruction

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

Core Problem: 3D reconstruction benchmarks hide sensitivity to real poses, calibration, and unseen trajectories.

Key Innovation: A multi-trajectory drone dataset isolates pose, intrinsics, and viewpoint-coverage effects.

39. Bootstrapping a 4D LiDAR Annotation Tool from Video Foundation Models

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

Core Problem: Dense 4D LiDAR labels are too expensive to scale manually.

Key Innovation: Bootstraps LiDAR supervision from SAM2 video masks via projection and temporal aggregation.

40. Gaussian Splatting Underwater: A Controlled Cross-Regime Study

Source: ArXiv (Geo/RS/AI) Type: Preprint Geohazard Type: coastal and subaqueous mapping transfer Relevance: 5/10

Core Problem: Underwater Gaussian splatting is evaluated under inconsistent conditions that hide failure modes.

Key Innovation: A controlled cross-regime benchmark separates effects of turbidity, lighting, and SfM failure.

41. Physics-Informed Foresight Pruning for Sparse PINN Solvers of Nonlinear PDEs

Source: ArXiv (Geo/RS/AI) Type: Preprint Geohazard Type: process modeling transfer Relevance: 5/10

Core Problem: PINN pruning criteria ignore parameters important to PDE residual dynamics.

Key Innovation: Residual-sensitive saliency preserves physics-side training behavior under sparsity.

42. A Constitutive Markov Physics-Informed Neural Operator (MPNO) for Autoregressive Stability in Transient Dynamics

Source: ArXiv (Geo/RS/AI) Type: Preprint Geohazard Type: physics-based hazard simulation Relevance: 5/10

Core Problem: Prevent autoregressive instability in neural-operator simulation of transient dynamics.

Key Innovation: Constructs a Markov row-stochastic propagator with spectral-radius control from constitutive physics.

43. Forecasting Multiple Observables with SCROLL: Score-Trained Uncertainty for Stochastic Dynamics

Source: ArXiv (Geo/RS/AI) Type: Preprint Geohazard Type: probabilistic hazard forecasting Relevance: 5/10

Core Problem: Forecast multiple future observables from stochastic systems without manual task-loss balancing.

Key Innovation: Free-routed last-layer likelihood composition that learns uncertainty jointly across tasks.

44. Uncertainty-Guided Latent Diffusion Models for Faithful Super Resolution

Source: ArXiv (Geo/RS/AI) Type: Preprint Geohazard Type: remote-sensing image enhancement Relevance: 5/10

Core Problem: Diffusion super-resolution methods struggle to improve fidelity without sacrificing perceptual quality.

Key Innovation: Guides latent diffusion with estimated reconstruction uncertainty so high-frequency restoration is concentrated in uncertain regions.

45. Multi-output Gaussian process prediction of physical fields under linear equality constraints

Source: ArXiv (Geo/RS/AI) Type: Preprint Geohazard Type: physical field prediction Relevance: 5/10

Core Problem: Predict multiple high-dimensional physical fields while enforcing linear physical constraints.

Key Innovation: Row-wise PCA plus constrained multi-output Gaussian processes that preserve constraints in latent space.

46. Multisource Feature Embedding Network for Joint Subpixel Mapping of Hyperspectral and LiDAR Data

Source: IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing Type: hyperspectral-LiDAR subpixel mapping Geohazard Type: terrain-mapping transfer Relevance: 5/10

Core Problem: HSI-only subpixel mapping struggles with mixed pixels and spectral variation, while LiDAR spatial detail is underused.

Key Innovation: A multisource embedding network that combines LiDAR multigranularity spatial cues with enhanced spectral-context modeling for subpixel maps.

47. SKANet: A Single-Stream Spectral-KAN Framework for Multisource Remote Sensing Fusion and Interpretation

Source: IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing Type: multisource remote-sensing fusion Geohazard Type: multi-hazard EO transfer Relevance: 5/10

Core Problem: Multisensor remote-sensing fusion struggles to bridge modality gaps while preserving local detail and global structure efficiently.

Key Innovation: A single-stream fusion network that couples FFT-based global perception with KAN-style nonlinear modeling and adaptive multimodal integration.

48. DeepMelt-GL v1: a neural network emulator of sub-shelf melt rates for the unrepresented regions of ice-shelf cavities in ocean models

Source: Geoscientific Model Development Type: research_article Geohazard Type: cryosphere instability Relevance: 5/10

Core Problem: Coarse ocean models cannot resolve sub-shelf melt processes beneath Antarctic ice shelves.

Key Innovation: Introduces a neural-network emulator for unresolved sub-shelf melt rates so long multi-century simulations can include ice-ocean interactions.

49. Spatiotemporal variations in precipitation of different magnitudes in China during 1961-2020 and their contribution analysis

Source: Frontiers in Earth Science Type: research_article Geohazard Type: flood climatology Relevance: 5/10

Core Problem: Separates how changes in precipitation days and intensity at different rain magnitudes reshape total precipitation across China.

Key Innovation: Introduces a partial-derivative contribution decomposition showing torrential-rain frequency dominates recent monsoon-region increases.

50. Decoupled Aquatic Greening and Water-Area Dynamics in Northeast Siberian Arctic Thermokarst Lakes from 2000 to 2025

Source: Remote Sensing (MDPI) Type: research_article Geohazard Type: permafrost and thermokarst change Relevance: 5/10

Core Problem: Quantifies how lake area and aquatic vegetation have changed across thousands of Northeast Siberian thermokarst lakes.

Key Innovation: Large-scale Landsat event analysis shows strong long-term decoupling between aquatic greening and water-area dynamics.

51. A reduced order model framework suitable for geotechnical problems

Source: Acta Geotechnica Type: geotechnical reduced-order modeling Geohazard Type: general geotechnical modeling transfer Relevance: 5/10

Core Problem: High-fidelity geotechnical simulations are expensive, and reduced-order models depend strongly on architecture choice.

Key Innovation: Combines latent-space reduction via autoencoders or PCA with DeepONet to predict spatiotemporal geotechnical responses efficiently.

52. When Does Frequency Decomposition Benefit Physics-Informed Neural Networks? A Preliminary Ablation Study

Source: ArXiv (Geo/RS/AI) Type: Preprint Geohazard Type: cross-cutting Relevance: 4/10

Core Problem: Identify when frequency decomposition actually helps PINNs approximate high-frequency or multi-scale PDE solutions.

Key Innovation: Dual-branch spectrally gated PINN isolates low- and high-frequency components for controlled ablation.

53. AFDBench: A Reasoning-First AI Scientist for NationalWeather Service Forecast Discussions

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

Core Problem: LLMs hallucinate numbers and style when drafting professional forecast discussions from weather model inputs.

Key Innovation: Introduces AFDBench and domain-reward RL to improve numerically grounded NWS-style forecast discussions.

54. Physics-Informed Error Field Learning: A Post-Training Optimization Framework for Physics-Informed Neural Networks

Source: ArXiv (Geo/RS/AI) Type: Preprint Geohazard Type: cross-cutting Relevance: 4/10

Core Problem: Late-stage PINN optimization spends compute for diminishing accuracy gains.

Key Innovation: Adds a physics-constrained auxiliary error network that learns the residual field after primary training.

55. Learning spatially varying regularisation parameters of low regularity for image reconstruction

Source: ArXiv (Geo/RS/AI) Type: Preprint Geohazard Type: cross-cutting Relevance: 4/10

Core Problem: Understand what regularity assumptions are appropriate for learned spatially varying regularization weights.

Key Innovation: Synthesizes theory and examples showing low-regularity adaptive weights can improve reconstruction detail preservation.

56. Lowering the Barrier to AI-Driven Inspection: A No-Code Workflow for Automated Structural Defect Detection

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

Core Problem: Adoption of AI-based structural defect detection is limited by labeling, training, and deployment complexity.

Key Innovation: YOLOEZ packages labeling, training, and inference in a no-code GUI for reproducible structural defect workflows.

57. WAVE: Reversing the Guidance Hierarchy for Coarse-to-Fine Guided Depth Super-Resolution

Source: ArXiv (Geo/RS/AI) Type: Preprint Geohazard Type: terrain and depth mapping transfer Relevance: 4/10

Core Problem: Fine-to-coarse guidance pipelines blur depth boundaries and inject texture artifacts.

Key Innovation: Reverses guidance order with wavelet sub-bands and semantic gating.

58. Neither Precision Nor Architecture Alone: Controlled Tests of Failure Remedies for Physics-Informed Neural Networks

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

Core Problem: PINN failures on hard PDEs are blamed on single remedies without controlled comparison.

Key Innovation: Seed-paired studies separate precision, stopping, backbone, and alignment effects.

59. RSFusionDet: Underwater RGB-Sonar Multimodal Object Detection

Source: ArXiv (Geo/RS/AI) Type: Preprint Geohazard Type: coastal and subsea sensing transfer Relevance: 4/10

Core Problem: RGB and sonar detections remain misaligned and separately incomplete underwater.

Key Innovation: Cross-attention fusion and an object-matching head align complementary RGB-sonar detections.

60. Saliency-Depth Conditioning for Zero-Shot Segmentation of Communication-Tower Components in Cluttered UAV Imagery

Source: ArXiv (Geo/RS/AI) Type: Preprint Geohazard Type: infrastructure and remote-sensing transfer Relevance: 4/10

Core Problem: Zero-shot tower segmentation fails in cluttered UAV scenes with confusing background structures.

Key Innovation: Combines saliency and monocular depth into a coarse tower prior for Grounded-SAM and SAM 3.

61. When Should a Network Emit Geometry, and When Should It Detect It? Readout, Reconciliation, and Representation in Floorplan Vectorization

Source: ArXiv (Geo/RS/AI) Type: Preprint Geohazard Type: built-environment exposure mapping transfer Relevance: 4/10

Core Problem: Floorplan vectorization performance depends on output representation and domain shift.

Key Innovation: Directly compares sequence and detection readouts, then shows output-level fusion helps most.

62. LDAC-Net: A Learnable Multi-Lag Differencing Attention-Convolution Network for Drift-Robust Recognition with Low-Cost MOX Gas Sensors

Source: ArXiv (Geo/RS/AI) Type: Preprint Geohazard Type: volcanic or air-quality gas monitoring Relevance: 4/10

Core Problem: Robustly recognize gases from drifting low-cost MOX sensor signals.

Key Innovation: Learnable multi-lag differencing with normalization and attention-convolution modeling.

63. Diffusion Transformers for Roof Graph Synthesis and Reconstruction

Source: ArXiv (Geo/RS/AI) Type: Preprint Geohazard Type: built-environment exposure mapping Relevance: 4/10

Core Problem: Generate and reconstruct structured roof graphs from footprints and images.

Key Innovation: Diffusion transformer for vertices plus geometry-aware edge prediction.

64. Modeling spatio-temporal locality in multi-step forecasting of geo-referenced time series

Source: ArXiv (Geo/RS/AI) Type: Preprint Geohazard Type: distributed environmental sensor networks Relevance: 4/10

Core Problem: Forecast multi-step measurements from geographically distributed sensors with local dependencies.

Key Innovation: Linear model trees that inject spatial features only within trend-similar local groups.

65. StablePDENet: Enhancing Neural Operator Stability through Physics-Informed Residual-Sensitivity Regularization

Source: ArXiv (Geo/RS/AI) Type: Preprint Geohazard Type: hazard-process surrogate transfer Relevance: 4/10

Core Problem: Physics-informed neural operators can be accurate yet unstable under input perturbations.

Key Innovation: Residual-sensitivity regularization within a physics-informed adversarial min-max training scheme for more stable operator learning.

66. Loss Landscape Geometry of Partial Differential Equation Emulators: Or, Symmetry Learning via Gradient Alignment

Source: ArXiv (Geo/RS/AI) Type: Preprint Geohazard Type: hazard-process surrogate transfer Relevance: 4/10

Core Problem: It is hard to tell whether neural PDE emulators have truly internalized physical symmetries.

Key Innovation: A gradient-alignment diagnostic on group orbits that probes symmetry learning through loss-landscape geometry rather than only forward predictions.

67. CPTu-Based Interpretation of Partially Drained Responses in Thickened and Conventional Mine Tailings

Source: Journal of Geotechnical and Geoenvironmental Engineering Type: tailings site characterization Geohazard Type: tailings geotechnics transfer Relevance: 4/10

Core Problem: Partially drained CPTu responses in mine tailings are hard to interpret using standard drained or undrained frameworks.

Key Innovation: A field-to-lab interpretation procedure combining variable-rate CPTu, supporting tests, and simulations to constrain tailings state and consolidation behavior.

68. Downward Continuation Method for Potential Field Data Based on Physics-Informed Neural Network

Source: IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing Type: physics-informed potential-field processing Geohazard Type: geophysical mapping transfer Relevance: 4/10

Core Problem: Downward continuation of potential-field data is severely ill-posed and amplifies noise.

Key Innovation: A PINN formulation with Laplace-equation and boundary-condition regularization to stabilize downward continuation at depth.

69. Hyper Look-Ahead Network for Remote Sensing Object Detection

Source: IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing Type: remote sensing object detection Geohazard Type: multi-hazard EO transfer Relevance: 4/10

Core Problem: Remote-sensing detectors struggle to localize objects in low-resolution wide-area imagery because feature-pyramid information is underused.

Key Innovation: A detector with look-ahead neck design, conspicuous-feature attention, and multiscale feature processing for faster small-object localization.

70. LLM-Enhanced Architecture Design for Lightweight Small-Object Detection in Low-Altitude Remote Sensing

Source: IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing Type: LLM-guided low-altitude detector design Geohazard Type: multi-hazard EO transfer Relevance: 4/10

Core Problem: Designing lightweight small-object detectors for diverse low-altitude remote-sensing scenes is hard to do manually.

Key Innovation: An LLM-driven architecture-planning loop that maps scene characteristics into adaptive NAS policies and replans from search feedback.

71. SHCFN: Superpixel-Based Hybrid Cooperative Fusion Network for Hyperspectral Image Classification

Source: IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing Type: hyperspectral classification Geohazard Type: terrain-mapping transfer Relevance: 4/10

Core Problem: Graph-based HSI classifiers capture local relations well but are computationally heavy and weak on global dependencies.

Key Innovation: A superpixel-based hybrid network combining multihop graph attention, agent attention, channel attention, and gated fusion.

72. Acoustic-Hyperspectral Multimodal Generation and Fusion for Cotton Drought Stress Detection Based on UAV Remote Sensing

Source: IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing Type: agricultural drought-stress detection Geohazard Type: drought-related remote-sensing transfer Relevance: 4/10

Core Problem: Single-modal sensing is too noisy and incomplete for early UAV-based cotton drought-stress detection.

Key Innovation: A bimodal acoustic-hyperspectral fusion network with multiscale time-frequency and regional-global spectral branches plus adaptive fusion.

73. Research on Sea Surface Temperature Prediction Method Based on Multiscale Temporal Dynamics and Frequency-Spatial Attention

Source: IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing Type: sea-surface-temperature forecasting Geohazard Type: meteorological hazard transfer Relevance: 4/10

Core Problem: Existing SST models inadequately capture multiscale temporal evolution and frequency-domain structure.

Key Innovation: A ConvLSTM architecture that combines circular dilated-time modeling with frequency-spatial attention in the DCT domain.

74. CMMS-DETR: Cross-Modal Multiscale Feature Fusion Network for UAV Remote Sensing Detection

Source: IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing Type: visible-infrared UAV detection Geohazard Type: multi-hazard EO transfer Relevance: 4/10

Core Problem: Visible-infrared UAV detection suffers from scale variation, weak detail expression, and ineffective multimodal fusion.

Key Innovation: An RT-DETR-based detector with multiscale residual extraction, mixed channel attention, and channel-spatial cross-modal fusion.

75. A Frequency-Aware and Multiscale Aligned Framework for Fine-Grained Object Detection in Remote Sensing Imagery

Source: IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing Type: remote sensing object detection Geohazard Type: multi-hazard EO transfer Relevance: 4/10

Core Problem: Remote-sensing object detection is degraded by small targets, subtle class differences, and multiscale feature mismatch.

Key Innovation: A frequency-aware preprocessing and multiscale alignment framework that strengthens structural detail and scale-consistent YOLO features.

76. A New Approach to Estimate Ocean Surface Velocity from High-Resolution Surface Water and Ocean Topography (SWOT) Data

Source: IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing Type: ocean remote-sensing retrieval Geohazard Type: coastal-ocean hazard transfer Relevance: 4/10

Core Problem: Geostrophic assumptions break down at submesoscale, biasing surface-current estimates from high-resolution altimetry.

Key Innovation: A curvature-based cyclostrophic correction for SWOT-derived velocities validated against models and coherent chlorophyll structures.

77. Improving simulation of Earth system variability through weakly coupled ocean data assimilation in E3SM

Source: Geoscientific Model Development Type: ocean data assimilation Geohazard Type: climate-hazard transfer Relevance: 4/10

Core Problem: Climate prediction skill is limited by inaccurate ocean initial conditions and potential initialization shocks.

Key Innovation: A weakly coupled ocean data-assimilation system in E3SM that improves variability simulation and teleconnected hindcasts without strong shock.

78. Hydrological implications of vegetation-associated precipitation recycling during peak growing season over the Loess Plateau

Source: Hydrology and Earth System Sciences Type: research_article Geohazard Type: hydroclimate stress Relevance: 4/10

Core Problem: Tests whether vegetation-driven precipitation recycling can offset revegetation water demand on the Loess Plateau.

Key Innovation: Combines rainfall tracking and water-balance analysis to show added rainfall usually does not compensate for extra plant water use.

79. FBR-DETR: An Efficient End-to-End Network for Real-Time Small-Object Detection in UAV Imagery

Source: Remote Sensing (MDPI) Type: research_article Geohazard Type: UAV remote-sensing analytics Relevance: 4/10

Core Problem: Improves small-object detection accuracy and speed in UAV imagery without increasing model size.

Key Innovation: Combines frequency-domain feature extraction, binary attention, and re-parameterized cross-scale fusion in an efficient detector.

80. SWH Retrieval from SWOT KaRIn Data by Combining Backscattering and Interference Characteristics

Source: Remote Sensing (MDPI) Type: research_article Geohazard Type: coastal wave monitoring Relevance: 4/10

Core Problem: Improves significant wave height retrieval from SWOT KaRIn observations.

Key Innovation: Uses an MLP to fuse backscatter and interferometric features, substantially reducing retrieval error versus the reference product.

81. Benchmarking Open-Access Building Footprints: A Multi-Dimensional Assessment with High-Fidelity References

Source: Remote Sensing (MDPI) Type: remote-sensing benchmark dataset Geohazard Type: exposure mapping and built-environment remote sensing Relevance: 4/10

Core Problem: Open-access building-footprint products lack a consistent large-scale benchmark for reliability assessment across cities.

Key Innovation: Builds a 200,000-plus manually annotated reference set and a three-tier framework for statistical, spatial, and morphological evaluation of 11 footprint datasets.

82. The influence of soil hydrophobicity and its persistence on soil water retention curve under wetting path: an experimental investigation and a fractal-based model

Source: Acta Geotechnica Type: soil hydraulic modeling study Geohazard Type: indirect rainfall-induced slope failure relevance Relevance: 4/10

Core Problem: The wetting branch of the soil-water retention curve is poorly understood for hydrophobic sands with different persistence behaviors.

Key Innovation: Develops a fractal wetting-path SWRC model that incorporates contact-angle dynamics and hydrophobicity persistence.

83. Undrained creep behaviors of a compacted clay under freeze-thaw-drying-wetting effects: insights from experimental tests and MPM-based numerical simulations

Source: Acta Geotechnica Type: clay creep experiment and simulation Geohazard Type: indirect slope or earthwork deformation relevance Relevance: 4/10

Core Problem: Clarify how freeze-thaw-drying-wetting cycles alter undrained creep of compacted clay under low confining pressure.

Key Innovation: Couples triaxial creep testing with an MPM-based numerical method to interpret both macroscopic strain evolution and particle-scale behavior.

84. Topo-climatic controls on rock glacier chronology since Lateglacial period in the northwestern part of Mattertal, Swiss Alps, reconstructed by TCN dating

Source: Geomorphology Type: periglacial geomorphology chronology Geohazard Type: cryosphere and periglacial slope-process context Relevance: 4/10

Core Problem: Reconstruct when and why rock glaciers formed, advanced, stabilized, or disappeared across Alpine catchments since the Lateglacial.

Key Innovation: Uses 10Be dating across multiple catchments to tie rock-glacier chronology to elevation, aspect, deglaciation history, and debris supply.

85. Identifying regional drivers of variability in SWOT-derived water surface elevation accuracy

Source: Remote Sensing of Environment Type: remote-sensing accuracy analysis Geohazard Type: indirect flood-monitoring relevance Relevance: 4/10

Core Problem: Identify why SWOT-derived water-surface-elevation accuracy varies across regions.

Key Innovation: The title indicates a regional driver analysis for satellite altimetry accuracy; no hazard-specific innovation is described in the supplied abstract.

86. Strengthening and drainage enhancement of copper tailings using an all-solid-waste geopolymer: Multiscale evidence

Source: Journal of Rock Mechanics and Geotechnical Engineering Type: tailings stabilization material study Geohazard Type: indirect tailings-instability relevance Relevance: 4/10

Core Problem: Improve copper-tailings strength and drainage using an all-solid-waste geopolymer.

Key Innovation: The title indicates a waste-based geopolymer validated by multiscale evidence, with no explicit hazard workflow described in the supplied abstract.

87. Damage and fracture behaviors of weathered granite in cold regions: Insights from CT imaging

Source: Journal of Rock Mechanics and Geotechnical Engineering Type: cold-region rock damage imaging Geohazard Type: indirect rockfall or rock-slope degradation relevance Relevance: 4/10

Core Problem: Characterize damage and fracture behavior of weathered granite in cold regions.

Key Innovation: CT imaging appears to be used to reveal fracture evolution, but the supplied abstract gives no explicit geohazard-facing advance.

88. Disintegration Properties and Micro-Mechanisms of Red Sandstone Subjected to Coupled Effects of Acid-Alkali Corrosion and Freeze-Thaw Cycle

Source: Journal of Rock Mechanics and Geotechnical Engineering Type: rock weathering mechanics study Geohazard Type: indirect rock-slope degradation relevance Relevance: 4/10

Core Problem: Understand how combined acid-alkali corrosion and freeze-thaw cycling drive red-sandstone disintegration.

Key Innovation: The title indicates a coupled environmental-weathering micro-mechanism study, without a stated hazard-assessment application.