TerraMosaic Daily Digest: August 24, 2026
Daily Summary
Across the direct geohazard papers, damaging outcomes are repeatedly traced to evolving hydraulic, mechanical, and structural state rather than to forcing magnitude alone. Rainfall-driven slope studies show delayed infiltration organizing isolated-boulder instability into staged transitions, shallow loess failure responding more strongly to slope gradient than rainfall intensity and exhibiting precursor thresholds in matric suction, pore pressure, and acoustic emission, and fractured rock slopes under flood-discharge atomization collapsing through fracture-guided seepage and progressive toe-to-mid-slope weakening. Runout papers extend this process view: erosion modifies the resistance of rock-ice-avalanche debris flows through pore-pressure and dilatancy effects, and the long-runout Gansu mudflow was intensified less by transient road-dam breach alone than by added upstream water supply. Seismic contributions follow the same pattern, with elliptic ETAS kernels improving European aftershock forecasting, Qujiang simulations showing fault geometry and basin structure jointly reshape near-fault pulses, and Jishishan fatalities concentrating where vulnerable building typologies, local environmental conditions, and hanging-wall position coincided.
Hydroclimatic, wildfire, coastal, and cryospheric papers sharpen both forcing diagnosis and operational relevance. Eastern Siberia's fire season lengthens mainly through delayed autumn termination associated with accumulated August-September drought, while neighborhood-scale structure loss in the 2025 Palisades Fire is explained most strongly by building density and vegetation moisture rather than by a single topographic variable. A Canadian Fire Weather Index dataset recalculated at local solar noon improves spatial realism for daily fire-danger fields. Elsewhere, teleconnection-specific lags provide continent-scale drought lead information, extreme precipitation in the Hengduan Mountains intensifies with a stable south-to-north propagation sequence and increasingly dominant elevation control at the highest extremes, and retrogressive thaw slumps measurably reduce active-layer structural quality, water storage capacity, and thermal stability through a thermal-to-hydrological feedback chain. Coastal and polar studies add more defensible sea-level uncertainty quantification, measurable satellite-ground gains for compound humid-heat and oxidant warning, improved vegetation-based wave-attenuation parameterization, and a continent-scale Antarctic grounded-iceberg baseline.
Methodological advances proceed along two distinct tracks. Hazard models become more interpretable and more tightly coupled through Bayesian and physics-informed landslide susceptibility frameworks, dimensionless-feature pluvial flood mapping, gauge-localized streamflow assimilation, and seismic resilience or fragility analyses showing that building-water coupling, epicentral-distance grouping, and joint safety-serviceability criteria can materially shift risk estimates. In parallel, transferable studies advance multimodal disaster assessment, remote-sensing tokenization and co-registration, hyperspectral super-resolution, cavity detection, point-cloud registration, sparse-view reconstruction, imbalance-robust evaluation, calibrated rare-event uncertainty, neural operators, and differentiable data assimilation. These methods materially enlarge the technical base for geohazard analysis, but most validate sensing, inference, or computation in their stated domains rather than demonstrating geohazard performance by themselves.
Key Trends
Five scientific and methodological trajectories define the August 24 literature: state-resolved hazard mechanics, geometry-aware forecasting, operational sensor fusion, coupled probabilistic assessment, and a distinct but largely transferable methods stream in remote sensing and scientific AI.
- Failure is being resolved through hidden state evolution: The strongest hazard papers identify progressive internal change as the decisive control: delayed infiltration weakens boulder stability over time, loess slopes cross precursor thresholds in suction and pore pressure before collapse, thaw slumps reorganize active-layer structure and heat transfer, and erosion, wet-dry cycling, or transient saturation alter the mobility or strength of debris flows, rock slopes, and tailings dams.
- Geometry and structural context are now first-order hazard controls: Several studies replace isotropic or overly simplified representations with explicit geometry. Elliptic ETAS kernels improve aftershock forecasts, Qujiang rupture simulations show curved fault segments and basin structure reshape pulse behavior, Tibetan radial anisotropy maps localize weak zones beneath rifts, SWOT gravity resolves buried Gulf of Mexico structures, and Jishishan mortality patterns depend on fault-block position and building configuration as much as shaking alone.
- Operational warning is moving toward fused, deployable observations: Operational studies increasingly combine observations that can function under real monitoring constraints. Debris-flow warning integrates InSAR with recurrent sequence models, groundwater and radon records are decomposed to separate seismic from meteorological signals, coastal compound-warning skill improves when latency-aware MODIS context is added to ground data, Canadian fire-weather fields are recalculated at local solar noon, and streamflow reanalysis gains from gauge-adaptive localization.
- Hazard assessment is becoming coupled, probabilistic, and interpretable: Direct hazard models increasingly embed physical structure while retaining explainability. Landslide susceptibility is linked to slope-stability logic through PINN and SHAP or to causal factor interactions through Bayesian networks; pluvial flood mapping uses dimensionless physically based features; sea-level inference adopts infinite-dimensional Bayesian inversion; and seismic infrastructure papers show that functional coupling, restoration uncertainty, epicentral-distance stratification, and joint serviceability criteria change inferred resilience and fragility.
- Transferable AI methods are advancing faster than hazard validation: A large methods cohort improves multimodal retrieval, tokenization, co-registration, hyperspectral reconstruction, 3D scene recovery, anomaly detection, uncertainty calibration, rare-event metrics, neural PDE surrogates, and differentiable assimilation. These studies are important for Earth-observation and geoscience workflows, but the evidence presented here generally supports methodological transferability rather than established geohazard validity unless a hazard application is explicitly tested.
Selected Papers
The selected papers span direct studies of earthquakes, landslides, floods, wildfire, drought, coastal change, permafrost disturbance, and infrastructure vulnerability, alongside a broader set of remote-sensing, reconstruction, simulation, and uncertainty methods. Read the first group as domain-validated hazard evidence and the second as enabling analytical advances whose geohazard utility remains application-specific unless directly demonstrated.
1. Improving Earthquake Forecasting for Europe Using Elliptic ETAS Aftershock Distribution Kernels
Core Problem: Standard ETAS forecasts ignore fault-controlled anisotropy in aftershock spatial patterns.
Key Innovation: Replaces circular ETAS kernels with elliptic kernels estimated during inversion and shows forecast gains on European catalogs.
2. Rapid Earthquake-to-Tsunami Waveform Generation via Large-Scale Multi-GPU FFT Convolution Applied to the Cascadia Subduction Zone
Core Problem: Generate large earthquake and tsunami waveform ensembles fast enough for warning pipelines.
Key Innovation: Turns rupture-to-waveform physics into distributed FFT convolutions running in 24 ms per rupture.
3. Physically-based dimensionless features for pluvial flood mapping with machine learning
Core Problem: Fast machine-learning flood maps often fail to generalize to unseen regions, landscapes, and events.
Key Innovation: Uses Buckingham Pi-constrained dimensionless multi-scale features to improve transferability of probabilistic flood mapping against 2D hydraulic and FEMA-map baselines.
4. Stability and movement characteristics of isolated boulders under delayed cyclic rainfall: an experimental and analytical study
Core Problem: Delayed infiltration makes isolated-boulder failure under cyclic rainfall hard to predict.
Key Innovation: Combines theory, lab tests, and simulations to define staged instability evolution and a dynamic safety-factor model.
5. Physics-informed and interpretable assessment of seismic landslide susceptibility: a mechanistic-learning approach coupling PINN and SHAP
Core Problem: Seismic landslide susceptibility models struggle with pseudo-negatives, interpretability, and pre/post-event comparison.
Key Innovation: Couples Isolation Forest, PINN, and SHAP to enforce slope-stability logic and explain changing seismic-landslide controls.
6. Development of solid-liquid coupled similar materials and failure mechanism analysis for fractured rock slopes under flood discharge atomization
Core Problem: Flood-discharge atomization can destabilize fractured rock slopes, but realistic physical simulation materials are lacking.
Key Innovation: Develops solid-liquid coupled similarity materials and uses them to reproduce fracture-guided seepage and progressive slope collapse.
7. Effects of rainfall intensity and slope gradient on shallow loess landslides: flume model test and numerical simulation
Core Problem: Multi-factor controls on rainfall-induced shallow loess landslides remain poorly constrained.
Key Innovation: Combines orthogonal flume tests, numerical simulation, and synchronized hydro-mechanical-acoustic monitoring to derive precursor thresholds.
8. Effects of earth dam obstruction-collapse on dynamic motion of long-runout mudflow from 2023 Gansu Ms 6.2 earthquake in China
Core Problem: The role of temporary road-dam blockage and breach in long-runout earthquake-triggered mudflow is poorly understood.
Key Innovation: Uses field evidence, satellite imagery, and Massflow scenarios to separate dam-breach velocity effects from upstream water-supply effects.
9. Scenario-based dynamic rupture simulation for velocity pulse characteristics and basin effect on the Qujiang fault
Core Problem: Velocity-pulse characteristics and basin effects on the Qujiang fault need scenario-based rupture analysis.
Key Innovation: Applies scenario-based dynamic rupture simulation to earthquake-ground-motion behavior on a named fault system.
10. Influence of erosion process on the resistance of Debris flow induced by rock-ice avalanche (DFRIA): Insights from the flume experiments
Core Problem: Erosion effects on the resistance of debris flow induced by rock-ice avalanche are insufficiently constrained.
Key Innovation: Uses flume experiments to quantify how erosion alters the mobility resistance of rock-ice-avalanche-driven debris flow.
11. Rockfall hazard characterization and its temporal and spatial variability through change detection and detailed photographic records in post-wildfire terrain
Core Problem: Characterize where and when post-wildfire rockfall activity intensifies above a railway corridor.
Key Innovation: Combines repeat TLS M3C2 differencing, vegetation filtering, and gigapixel imagery to link rockfall clusters to slope features and seasonal triggers.
12. A hybrid FDM-MPM method for modelling earthquake-induced soil-nailed slope failure
Core Problem: Model failure of soil-nailed slopes during earthquakes.
Key Innovation: Couples FDM and MPM to capture earthquake-driven large-deformation slope failure.
13. Quasi-implicit finite volume material point method for modeling fluidized mass movements
Core Problem: Model fluidized mass movements with better stability and efficiency.
Key Innovation: Introduces a quasi-implicit finite-volume material point method tailored to fluidized mass movements.
14. Extended Fire Season Over Eastern Siberia Driven by Accumulated August-September Drought
Core Problem: Explain why eastern Siberia's fire season is lengthening and shifting later into autumn.
Key Innovation: Links delayed fire-season termination to accumulated August-September drought and large-scale circulation anomalies.
15. DamageScope: Vision-Language Retrieval at Scale for Disaster Damage Assessment from Satellite Imagery
Core Problem: Scale property-damage assessment from satellite imagery without excessive indexing cost or repeated LLM calls.
Key Innovation: Combines VLM/LLM retrieval-augmented analysis with multi-vector embedding clustering and dual-store architecture for faster, cheaper damage retrieval.
16. DySCo: Dynamically consistent data-driven downscaling of extremes in climate projections
Core Problem: Extreme-event downscaling must improve statistics without breaking dynamical consistency with coarse climate trajectories.
Key Innovation: DySCo pairs trajectories through data-driven nudging and trains a dynamically and statistically consistent two-stage downscaler.
17. The spatial anatomy of urban wildfire vulnerability: a spatially validated GeoAI framework reveals the roles of building density and vegetation moisture in structure loss during the 2025 Palisades Fire
Core Problem: Identify which built, vegetation, and terrain factors explain neighborhood-scale structure loss during an urban wildfire.
Key Innovation: Spatial-block-validated GeoAI linking inspections and remote-sensing covariates, exposing building density and moisture controls while correcting random-CV optimism.
18. SkyNative: A Native Multimodal Architecture for Remote Sensing Vision-Language Understanding
Core Problem: Modular remote-sensing VLMs separate visual encoding from language reasoning and weaken fine-grained spatial inference.
Key Innovation: A native multimodal remote-sensing architecture that interleaves visual and text tokens with modality-aware decoupling.
19. A daily gridded dataset of the Fire Weather Index across Canada, with calculations based on the sun's elevation
Core Problem: Provide physically improved daily Fire Weather Index fields across Canada.
Key Innovation: Gridded FWI dataset computed from ERA5 using local solar-noon conditions for better spatial realism.
20. Landslide susceptibility study using Bayesian networks combined with landslide physical mechanisms: a case study in Hebei Province, China
Core Problem: Black-box landslide susceptibility models often lack physical interpretability.
Key Innovation: Builds a Bayesian-network susceptibility model tied to landslide mechanisms and interpretable factor interactions.
21. Regional heterogeneity-aware domain adversarial training for trans-regional landslide susceptibility assessment
Core Problem: Regional heterogeneity and sparse inventories limit cross-region landslide susceptibility transfer.
Key Innovation: Embeds heterogeneity-aware domain labels into adversarial training for trans-regional landslide mapping.
22. Early warning method for rainfall-induced debris flow based on LSTM and InSAR technology
Core Problem: Reliable debris-flow warning needs better fusion of deformation and rainfall precursors.
Key Innovation: Integrates InSAR deformation, rainfall features, BiLSTM prediction, and probabilistic warning levels for debris-flow alerts.
23. Flood susceptibility mapping in a data-scarce arid catchment of Southwestern Saudi Arabia
Core Problem: Data-scarce arid basins lack robust flood susceptibility baselines.
Key Innovation: Applies a GIS-based fuzzy AHP framework with validation against documented flood-prone locations in Najran Basin.
24. Stability evaluation of rock slope considering the strength failure of weak structural planes induced by blasting vibration
Core Problem: Weak structural planes in bedding slopes lose shear strength under blasting vibration, complicating stability evaluation.
Key Innovation: Links wave-induced dynamic stress with post-disturbance structural-plane strength to compute slope safety factors.
25. Statistical insights from long-term time-series of hydrogeochemical signatures and radon in groundwater of the Matese area, Central-Southern Italy
Core Problem: Seismic hydrogeochemical anomalies must be separated from seasonal recharge and meteorological noise.
Key Innovation: Combines PCA and STL decomposition on long groundwater and radon records to isolate possible seismic signals.
26. Runoff simulation and flood risk assessment of heavy rainfall-flood events in the Juhe River Basin, Beijing
Core Problem: Mountain-basin heavy-rainfall floods need accurate runoff simulation and spatial inundation assessment under land-use change.
Key Innovation: Couples optimized CNN-BiLSTM-Attention runoff modeling with integrated flood simulation for scenario-based flood-risk mapping.
27. Fault-block patterns and local structural vulnerabilities associated with fatalities in the 2023 Jishishan earthquake
Core Problem: Moderate earthquakes can produce highly uneven fatality patterns that shaking intensity alone does not explain.
Key Innovation: Integrates fatality-site surveys, building vulnerability, terrain, ground motion, and fault-side position to explain within-event lethality differences.
28. Influence of Interbedded Soft and Hard Rock on Rock Burst and Microseismic Activities in Deeps: Case Studies
Core Problem: Interbedded lithology complicates prediction of rock burst occurrence and microseismic precursors in deep tunnels.
Key Innovation: Uses long-duration microseismic monitoring to link rock-burst intensity and clustering to lithology-dependent stress response.
29. Field Calibrated Stability Assessment of a Red Mud Tailings Dam Under Extreme Rainfall
Core Problem: Tailings-dam rainfall stability assessments often misrepresent permeability and storm-pattern effects.
Key Innovation: Field-calibrates coupled seepage-stress modeling and shows shallow transient saturation controls localized extreme-rainfall vulnerability.
30. Seismic serviceability and resilience assessment of water distribution systems considering functional coupling with building systems
Core Problem: Assess seismic serviceability and resilience of water distribution systems while accounting for coupling with buildings.
Key Innovation: Couples water-network and building-system functionality in a resilience-oriented seismic assessment framework.
31. Seismic fragility analysis of subway station employs ground motion ensembles with different epicentral distances
Core Problem: Estimate how subway-station fragility changes with ground-motion ensembles at different epicentral distances.
Key Innovation: Compares distance-dependent ground-motion ensembles within a seismic fragility framework for subway stations.
32. Analytical modeling of injection-induced fault reactivation potential in poroelastically anisotropic rock masses
Core Problem: Predict when fluid injection may reactivate faults in anisotropic rock masses.
Key Innovation: Derives an analytical poroelastic model for injection-induced fault reactivation potential under anisotropy.
33. Lagged global relationships between teleconnection indices and hydrological drought: a continental-scale analysis using ERA5-Land runoff data
Core Problem: Resolve lagged continental-scale links between teleconnection patterns and hydrological drought.
Key Innovation: Maps delayed global drought relationships using teleconnection indices and ERA5-Land runoff data.
34. Reshaping of the active layer by retrogressive thaw slump in permafrost regions: coupled soil structural, hydrological, and thermal responses based on high-resolution in situ observations
Core Problem: Explain how retrogressive thaw slumps reorganize the active layer's structure, hydrology, and thermal state.
Key Innovation: Uses high-resolution in situ observations to capture coupled soil, hydrological, and thermal responses to thaw-slump disturbance.
35. Spatiotemporal variation of extreme precipitation and its terrain modulation in the Hengduan Mountains: machine learning and interpretable analysis
Core Problem: Map how extreme precipitation varies through time and is modulated by topography in the Hengduan Mountains.
Key Innovation: Combines machine learning with interpretable analysis to resolve terrain controls on precipitation extremes.
36. Seismic fragility assessment of high-speed railway subgrade-bridge transition: a coupled three-dimensional safety-serviceability framework
Core Problem: Assess seismic safety and serviceability of high-speed railway subgrade-bridge transition zones.
Key Innovation: Builds a coupled 3D fragility framework combining safety and serviceability criteria.
37. Fidelity-Diversity-Consistency (FDC): Data Pruning for Remote Sensing Change Detection
Core Problem: Existing data-pruning methods do not reliably improve remote-sensing change detection training subsets.
Key Innovation: Identifies change-distribution fidelity as the key pruning signal and turns it into a two-stage FDC pruning method.
38. DECO: Depth-Guided Co-Visibility Reasoning for Low-Altitude UAV Visual Localization
Core Problem: Low-altitude UAV images contain many features not co-visible in orthographic reference maps.
Key Innovation: Depth-guided co-visibility scoring that filters keypoints using local geometry plus detector saliency.
39. Self-Calibrating Dense Displacement Fields for Reliable Co-Registration of Large Optical Satellite Imagery
Core Problem: Operational optical satellite pairs still have offsets large enough to corrupt multi-temporal analysis.
Key Innovation: Training-free self-calibrating dense displacement fields with pairwise threshold calibration and sub-pixel matching.
40. A scalable Bayesian framework for modern sea-level inference
Core Problem: Standard satellite-based sea-level inference underestimates uncertainty and omits important physics.
Key Innovation: Infinite-dimensional Bayesian inversion with adjoint sea-level physics and scalable matrix-free computation.
41. Tracing the Unlabeled Storm: Cross-Variable Transfer in a Lagrangian Atmospheric JEPA Framework
Core Problem: Direct rainfall pretraining learns weak latent representations for monsoon convection and heavy-rain prediction.
Key Innovation: Cross-variable JEPA pretraining on continuous atmospheric proxies over Lagrangian patches, then transfer to precipitation forecasting.
42. HeatTok: Enhancing Remote Sensing Image Understanding via Thermodiffusion-based Tokenization
Core Problem: Patch tokenization fragments irregular geo-objects and mixes semantics in remote-sensing MLLMs.
Key Innovation: Thermodiffusion-based irregular tokenization with Gaussian positional encoding for object-aligned remote-sensing tokens.
43. Interpretable Landsat-to-Hyperspectral Dual Super-Resolution Without Large Matrix Inversion
Core Problem: Generate AVIRIS-like hyperspectral detail from 7-band Landsat while avoiding prohibitive large matrix inversions.
Key Innovation: Interpretable PGD-ADMM network with spectral continuity prior and physically grounded pan sharpening that removes large-matrix inversion bottlenecks.
44. Lightweight Multi-scale Hierarchical Anomaly Detection and Localization for Geospatial Big Data Applications at the Edge
Core Problem: Real-time geospatial data streams are too large for exhaustive centralized anomaly detection and localization.
Key Innovation: Uses H3-based multiscale drill-down anomaly detection to localize persistent signals with major computational savings.
45. ADDA: a Modular Framework for Representing, Simulating and Assimilating Dynamics with End-to-end Differentiability
Core Problem: Data-assimilation comparisons and hybrid methods are hard because simulation and assimilation codes are fragmented, inflexible, and often not differentiable.
Key Innovation: Introduces an end-to-end differentiable framework with flexible state, mesh, observation, and parallel data-assimilation abstractions for geoscience systems.
46. A Modular Multitask Reasoning Framework Integrating Spatio-temporal Models and LLMs
Core Problem: Enable multi-task spatio-temporal inference with grounded long-form reasoning instead of narrow single-task prediction.
Key Innovation: Modular program-based integration of spatio-temporal models and LLM reasoning with benchmarked grounded explanations.
47. Shallow cavity detection using MASW, Rayleigh-wave scattering, and machine learning at a controlled test site
Core Problem: Localize small buried cavities that conventional surface-wave methods laterally smear.
Key Innovation: Combined MASW, Rayleigh-wave back-scattering, and Random-Forest probability mapping for sharper cavity detection.
48. Multiscale characterization of curved bank erosion under wave-current interaction: a combined turbulence-image analysis approach
Core Problem: Resolve multiscale deformation and turbulence controls on curved bank erosion under wave-current interaction.
Key Innovation: Combined turbulence and image-analysis framework for bank-erosion characterization.
49. Upslope expansion of active fire detections weakened following intensified fire-prevention policies in the southeastern margin of the Tibetan Plateau
Core Problem: It is unclear whether intensified fire-prevention policy altered elevational wildfire expansion in the Hengduan Mountains.
Key Innovation: Combines long MODIS fire records with interrupted time-series analysis to isolate policy-linked shifts in fire occurrence.
50. Process-Informed Satellite-Ground Fusion for Coastal Compound Humid-Heat and Photochemical Oxidant Early Warning
Core Problem: It is unclear whether satellite context improves coastal warning decisions under explicit false-alarm limits.
Key Innovation: Builds a validation-locked satellite-ground fusion benchmark showing measurable warning gains from latency-aware MODIS context.
51. Integrated experimental and numerical investigation of the seismic behavior of historic timber-supported masonry mosques
Core Problem: Historic timber-supported masonry mosques need validated seismic-performance assessment beyond wall-strength checks alone.
Key Innovation: Calibrates 3D finite-element mosque models with modal testing and identifies repeat local damage mechanisms before collapse.
52. Gully morphological characteristics and evolution in the compound water-wind erosion region of the Loess Plateau
Core Problem: Quantify long-term gully growth and controls in the water-wind erosion zone of the Loess Plateau.
Key Innovation: Fuses UAV mapping and historical topography to derive volume scaling, retreat rates, land-use effects, and initiation thresholds.
53. Revealing buried tectonic structures in the Gulf of Mexico using SWOT-derived marine gravity
Core Problem: Detect buried tectonic structures in the Gulf of Mexico from satellite-derived gravity.
Key Innovation: Applies SWOT-derived marine gravity to reveal subsurface tectonic patterns not obvious from surface data.
54. Locally relevant streamflow reanalysis using ensemble data assimilation and a gauge-based adaptive localization scheme
Core Problem: Improve streamflow reanalysis with gauge-informed adaptive localization.
Key Innovation: Combines ensemble data assimilation, locally adaptive localization, and large-scale hydrological reanalysis output.
55. A thermodynamically consistent variational neural constitutive update framework for geomaterials with non-associated flow
Core Problem: Learn constitutive updates for geomaterials with non-associated flow without violating thermodynamics.
Key Innovation: Builds a variational neural constitutive update framework that enforces thermodynamic consistency.
56. Normalized Regime Persistence: A Simple Metric to Diagnose Land Feedback States
Core Problem: Global land-feedback states are hard to diagnose consistently from multivariate flux estimates.
Key Innovation: Introduces normalized regime persistence using satellite soil-moisture variability alone to classify coupling regimes.
57. Seismic Radial Anisotropy of the Southern Tibetan Rifts From Ambient Noise Adjoint Tomography: Basal Shear and Localized Crustal Weak Zones Promote Rifting
Core Problem: Determine how deep basal shear localizes into southern Tibetan rifts.
Key Innovation: Ambient-noise adjoint tomography maps radial anisotropy and partial melt to trace upward focusing of ductile shear into weak zones.
58. GeoQ: Geometry-Aware Conditional Quantile Error Estimation for Scientific Surrogate Models
Core Problem: Scientific surrogate models need query-specific error estimates in extrapolative settings.
Key Innovation: Geometry-aware conditional quantile calibration that combines anchor-relative errors with local support diagnostics.
59. Read, Write, Relax: Why Neural PDE Surrogates Need Both Global and Local Processing
Core Problem: Global-token and local-message-passing surrogates each fail on large, industrial-scale meshes.
Key Innovation: Interleaves latent attention with message-passing relaxation to correct both low- and high-frequency simulation errors.
60. FlashReg: GPU-Accelerated 3-Clique Point Cloud Registration for Real-Time Correspondence-to-Pose Estimation
Core Problem: Graph-based correspondence-to-pose estimation is too slow and memory-heavy for real-time onboard use.
Key Innovation: GPU-optimized sparse second-order graph construction and 3-clique enumeration for fast correspondence-to-pose registration.
61. GaussVid: Sparse-View Gaussian Splatting with 3D-Aware Video Diffusion Priors
Core Problem: Sparse-view Gaussian splatting reconstructions contain artifacts because video priors lack camera-geometry awareness.
Key Innovation: Camera-conditioned, boundary-anchored 3D-aware video diffusion prior for restoring sparse-view Gaussian splats.
62. ORBIT++: Benchmarking SfM in the Wild with 360° Video
Core Problem: SfM lacks hard real-world benchmarks with reliable ground truth under dynamic, messy video conditions.
Key Innovation: A 360-video-derived benchmark that preserves robust ground truth while creating challenging perspective clips.
63. Robust Global Structure-from-Motion via View Graph Pruning
Core Problem: Erroneous view-graph edges destabilize global SfM in ambiguous image collections.
Key Innovation: Subgraph-guided pruning that uses locally consistent reconstructions to remove unreliable cross-subgraph edges.
64. Symbolic Neural ODEs: Learning interpretable models from time-series data
Core Problem: Learning sparse dynamical equations from time series often overfits one-step dynamics and becomes unstable.
Key Innovation: Multi-step neural ODE training with sparsity regularization to recover stable symbolic dynamics.
65. DAW: Dynamics-Aware Weighting for Deep Learning Forecasts of Chaotic Systems
Core Problem: Uniform training underweights rare high-complexity states that dominate long-rollout errors in chaotic systems.
Key Innovation: Local-dimension-based loss reweighting that shifts capacity toward dynamically rare high-error regimes.
66. Seeing the Unseen: Semantic-in-Gaussian for Sparse-View 3D Generalization
Core Problem: Sparse-view Gaussian splatting collapses in partially observed or occluded regions.
Key Innovation: Semantic-conditioned Gaussian-space refinement using cross-view entropy-aware embeddings and a conditional Gaussian transformer.
67. LagrangeGS: Non-Conservative Lagrangian System on Dynamic 3D Gaussian Splatting
Core Problem: Dynamic Gaussian splatting extrapolates with physically inconsistent trajectories, poor reversibility, and geometric collapse.
Key Innovation: Non-conservative Lagrangian dynamics for dynamic 3DGS with tractable Hessian approximation and local rigid alignment.
68. SiZeUp: Fast 3D Proxy from Aerial Images via Depth Ordinal Loss
Core Problem: Recover scalable building-height proxies from oblique aerial imagery without dense reconstruction or reliable metric depth.
Key Innovation: Differentiable footprint extrusion optimized with ordinal depth consistency and dynamic view selection for large-scale proxy modeling.
69. A 3D VTI factored eikonal solver using six-tetrahedron pyramidal stencil
Core Problem: Accurate factored eikonal updates in 3D VTI media require solving difficult quartic systems on oblique stencil faces.
Key Innovation: Six-tetrahedron pyramidal stencil with Ferrari-based quartic updates and specialized constrained-face solvers.
70. Substrate constraint effects on internal deformation and desiccation cracking in drying fine sediments: an X-ray CT study
Core Problem: Explain how substrate constraint drives internal deformation and crack initiation in drying fine sediments.
Key Innovation: X-ray CT with embedded markers to reconstruct 3D strain fields and a mechanism for substrate-controlled cracking.
71. Modeling wave attenuation by rigid vegetation under combined wave-current conditions: Improved prediction of effective bulk drag coefficient
Core Problem: Predict how vegetation attenuates waves under combined wave-current forcing relevant to coastal erosion.
Key Innovation: Laboratory-numerical framework linking attenuation to wavelength scaling and wave nonlinearity.
72. Controlled weakening of model ice: Replicating the mechanical response of summer-warmed sea ice in laboratories and test basins
Core Problem: Replicate the mechanical response of summer-warmed sea ice in controlled laboratory settings.
Key Innovation: Controlled weakening protocol for model ice that mimics warmer-season sea-ice mechanics.
73. Geospatial micro-estimates of slum populations in 129 Global South countries using machine learning and public data
Core Problem: Estimate slum populations spatially across data-sparse countries at continent-scale coverage.
Key Innovation: Generalized bottom-up ML framework producing gridded vulnerability estimates for 129 countries.
74. Grounded icebergs around Antarctica: a high-resolution dataset derived from deep learning and Sentinel-1 synthetic aperture radar
Core Problem: Map grounded Antarctic icebergs continent-wide despite their small size and sparse prior records.
Key Innovation: Deep-learning extraction of nearly 40,000 grounded icebergs from Sentinel-1 SAR imagery.
75. A JRC3D shear strength model considering failure surface spatial anisotropy of soft rock
Core Problem: 2D roughness indices miss 3D fracture anisotropy that governs joint shear strength.
Key Innovation: Defines 3D roughness parameters and a JRC3D-JCS model that improves soft-rock shear-strength prediction.
76. Criterion and evolution characteristics of the cyclic secant shear modulus of compacted loess under cyclic shearing
Core Problem: Dynamic stress-state transitions in compacted loess before large deformation and flowslide are not well characterized.
Key Innovation: Defines stiffness-index criteria and a stiffening-decay line to diagnose stable, metastable, and unstable cyclic states.
77. Multi-scale deterioration mechanism and failure precursor identification in limestone considering fissure angle and prestress under wet-dry cycles
Core Problem: Identify how fissure orientation and prestress control limestone deterioration and failure warning signals under wet-dry cycling.
Key Innovation: Tracks multiscale degradation and precursor behavior in weathered fractured limestone.
78. Simulation of rainfall-runoff flow using three-dimensional smoothed particle hydrodynamics
Core Problem: Examine whether rainfall and surface runoff can explain valley-network erosion patterns.
Key Innovation: Frames the problem around rainfall-runoff simulation and morphological evidence rather than groundwater-only explanations.
79. Emplacement Conditions and Alteration of Volcanic Deposits Inferred From Magnetic Properties of the ICDP Eger Drill Site S4 (Bažina Maar), Czechia
Core Problem: Infer emplacement and post-eruptive alteration histories in maar deposits from ambiguous magnetic signatures.
Key Innovation: Combines temperature-dependent magnetic susceptibility, anisotropy, and mineralogy to distinguish primary magmatic signals from hydrothermal overprints.
80. CLSC DETR: Reliable Candidate Ranking via Cross Layer Geometric Support for UAV Small Object Detection
Core Problem: Single DETR queries rank UAV small objects unreliably because geometric evidence is weak.
Key Innovation: Aggregates cross-layer local geometric support and calibrates classification-localization consistency for better ranking.
81. Reaching the Tail: Calibration Diversity Drives Conformal Coverage under Data Scarcity
Core Problem: Explain and improve conformal coverage for scarce, autocorrelated rare events over multiple forecast horizons.
Key Innovation: Shows calibration diversity drives coverage and proposes a diversity-maximizing selector with exact quantile-reach interpretation.
82. StereoDiffuer: Diffusion-based Progressive Geometry Modeling with Saliency Attention Perception for Stereo Matching
Core Problem: Stereo methods blur edges and suppress fine geometric detail in disparity maps.
Key Innovation: Diffusion-based iterative disparity refinement conditioned on saliency-aware geometric cues.
83. HP-UniIF: Hierarchical Prompt Learning for Unified Image Fusion
Core Problem: One model must handle heterogeneous fusion, restoration, and downstream perception without entangling objectives.
Key Innovation: Hierarchical prompt modulation across network depth to decouple task, degradation, and application controls.
84. Simple data fusion from several ocean and atmosphere hindcast models improves surface drifter trajectory prediction
Core Problem: Surface-drifter prediction is limited by uncertain ocean-current forcing in hindcast products.
Key Innovation: Simple linear fusion of multiple ocean and atmosphere hindcasts that materially improves short-term drift skill over baseline simulation.
85. Learning Reduced-Order Dynamics with Singularity via Latent-Augmented Neural Ordinary Differential Equations
Core Problem: Standard neural ODEs cannot represent reduced-order dynamics with self-intersecting phase trajectories.
Key Innovation: Latent augmentation with theory for minimum augmentation dimension to handle conflicting vector fields.
86. FreKoo++: Learning Continuous Spectral Dynamics for Temporal Domain Generalization
Core Problem: Temporal domain generalization struggles with multi-scale drift and irregular sampling.
Key Innovation: Continuous Koopman-spectral dynamics with adaptive soft spectral weighting for future-domain extrapolation.
87. Gaussian process learning with flow map refinement for parameter estimation in dynamical systems
Core Problem: Local Gaussian-process derivative matching can yield globally inconsistent parameter estimates from noisy trajectories.
Key Innovation: Two-stage Gaussian-process learning followed by flow-map refinement enforcing global dynamical constraints.
88. Contextrast++: Robust Multi-Scale Contextual Contrastive Learning for Semantic Segmentation
Core Problem: Semantic segmentation still struggles to combine local and global context while handling long-tailed class imbalance.
Key Innovation: Contextual contrastive learning with adaptive multi-scale fusion plus boundary-aware hard-negative sampling.
89. Maximum-distance nonnegative matrix factorization for unmixing highly mixed grain-size distribution data: A generalization of AnalySize
Core Problem: Highly mixed grain-size distributions are hard to decompose when no samples lie near true end members.
Key Innovation: Maximum-distance NMF that pushes estimated end members apart and solves the model with hierarchical alternating least squares.
90. Bridge Damage Detection from Low-Light UAV Imagery via Degradation-Aware Mixture-of-Experts Enhancement
Core Problem: Low-light UAV imagery obscures small bridge defects and hurts automated inspection.
Key Innovation: A degradation-aware MoE restoration front end trained with ISP-aware low-light synthesis to improve defect detection.
91. Geometry-Driven Opti-Acoustic Co-Registration and View-Invariant Reflectivity Mapping for Side-Scan Sonar
Core Problem: Align optical and side-scan sonar imagery despite viewpoint and acoustic distortions.
Key Innovation: Anchors co-registration with SfM geometry, first-bottom-return correction, and view-invariant reflectivity mapping.
92. FixAnything: 3D-Consistent Rendering Refinement via Video Generative Priors
Core Problem: Fix sparse-view rendering artifacts across many 3D scene representations.
Key Innovation: Repurposes a pretrained video model with clean-pixel masks and SfM-based DPO rewards.
93. What Does CLIP Learn for Regional Geolocalization? Probing Visual Cues and Scene Configuration After Adaptation
Core Problem: Understand what visual cues adapted CLIP uses for fine-grained regional geolocalization.
Key Innovation: Compares adaptation strategies and probes cue reliance through scramble, blur, and semantic ablations.
94. One-Step Evolution for Long-Time Extrapolation: An Error-Bound-Informed and Prior-Guided Neural Residual Framework for Autonomous PDEs
Core Problem: Long-time PDE extrapolation without trajectory supervision is unstable and inaccurate.
Key Innovation: Uses a numerical prior plus weak-form residual constraints to learn a one-step evolution operator for recursive rollout.
95. Efficient Regression Models for Scan Statistics
Core Problem: Scan statistics for interval anomalies become computationally expensive on non-stationary real-valued signals.
Key Innovation: Defines regression models for scan statistics and reduces naive quartic cost to linear time under practical assumptions.
96. Scale-invariant Optimal Sampling for Rare-events Data with Sparse Models
Core Problem: Optimal rare-event subsampling can become inefficient under arbitrary feature scaling, especially with inactive variables.
Key Innovation: Derives a scale-invariant optimal subsampling rule for sparse rare-event models and improves efficiency with MSCL estimation.
97. When a neural surrogate cannot accelerate a solver: runtime share, closed-loop drift, and the economics of uncertainty gating in a stiff coupled simulation
Core Problem: A learned surrogate for an expensive stiff inner solver may still fail to accelerate or remain stable in the full simulation loop.
Key Innovation: Provides an end-to-end negative-result analysis with Amdahl-law runtime caps, a break-even deferral formula, and long-run drift quantification.
98. Which Histories Matter for Time Series Forecasting? Learning Predictive Relevance with Future Supervision
Core Problem: Historical similarity is not always the best criterion for selecting useful analogs for forecasting.
Key Innovation: Learns predictive relevance from realized futures during training while keeping inference-time scoring strictly past-only.
99. Inertial Manifold Neural Operator for Dissipative Time-Dependent Partial Differential Equations
Core Problem: Standard neural operators do not explicitly exploit the effective low-dimensional long-time structure of dissipative PDEs.
Key Innovation: Builds inertial-manifold neural operators, including a shift-equivariant variant, for more stable and interpretable autoregressive PDE prediction.
100. Residual-based attention in physics-informed neural networks
Core Problem: Vanilla physics-informed neural networks converge slowly and underfit difficult regions of the domain.
Key Innovation: Introduces a gradient-free residual attention weighting that focuses optimization on persistently misfit regions.
101. Learning in PINNs: Phase transition, diffusion equilibrium, and generalization
Core Problem: PINN training dynamics and generalization remain poorly understood under nonconvex optimization.
Key Innovation: Identifies a diffusion-equilibrium phase and proposes sample-wise reweighting to improve residual homogeneity and generalization.
102. StereoDiff: Stereo-Diffusion Synergy for Video Depth Estimation
Core Problem: Video depth models handle static and dynamic regions poorly when treated like ordinary image-depth estimation.
Key Innovation: Combines stereo matching for static regions with diffusion-based video depth for dynamic regions.
103. RobustGS: Unified Boosting of Feedforward 3D Gaussian Splatting under Low-Quality Conditions
Core Problem: Feedforward 3D Gaussian splatting breaks down when multi-view inputs are degraded by real-world capture conditions.
Key Innovation: Plug-in degradation-aware feature enhancement with semantic cross-view aggregation for robust 3D reconstruction.
104. HOT-POT: Optimal Transport for Sparse Stereo Matching
Core Problem: Sparse stereo matching is ill-posed and sensitive to geometry and parameter choices.
Key Innovation: Frames epipolar and ray-consistent sparse matching as efficiently solvable optimal-transport problems.
105. Robust performance metrics for imbalanced classification problems
Core Problem: Common binary-classification metrics systematically underweight the minority class as imbalance becomes extreme.
Key Innovation: Robustified MCC, kappa, and F-score variants with guarantees that minority-class sensitivity remains bounded away from zero.
106. Unified frameworks for modeling small-strain stiffness and critical state behavior in gap-graded soils
Core Problem: Conventional state variables fail to capture coarse-fine stress transmission in gap-graded soils.
Key Innovation: Micromechanically refined state variable unifying small-strain stiffness and critical-state behavior.
107. SO_SLICE: a 19-year gridded sea level anomaly dataset for the open and ice-covered Southern Ocean (2003-2021)
Core Problem: Measure Southern Ocean sea-level anomaly despite sea-ice obstruction.
Key Innovation: Merged multi-mission record using open-water and lead echoes to extend coverage into ice-covered regions.
108. Evaluation of the ALARO1-SFX (CY43T2) regional climate model over Belgium across different resolutions
Core Problem: Regional climate simulations need better representation of precipitation extremes at local scales.
Key Innovation: Shows that SURFEX coupling and finer resolution improve temperature, precipitation, and precipitation-extreme skill over Belgium.
109. Learning evaporative fraction with memory
Core Problem: Evaporative fraction models often miss ecosystem memory effects under drought stress.
Key Innovation: Uses explainable machine learning with memory to quantify lagged controls on plant water stress across biomes.
110. Towards a semi-asynchronous method for hydrological modeling in climate change studies
Core Problem: Conventional climate-change hydrological workflows struggle to preserve extreme-event behavior.
Key Innovation: Tests a semi-asynchronous hydrological modeling approach that better captures extremes than standard workflows.
111. Seasonal variation monitoring of fluvial scene using PlanetScope data and machine learning method
Core Problem: Seasonal riverscape mapping is inconsistent at high spatial resolution.
Key Innovation: Creates a transferable PlanetScope ensemble workflow with confidence voting for seasonal water-sediment-vegetation mapping.
112. Altitude and Geographic Sensitivity Characteristics of the AIRS Satellite Spectrometer and Drift Correction Using Methane (CH4) Data
Core Problem: AIRS methane products show altitude-dependent sensitivity and long-term drift relative to ground truth.
Key Innovation: Derives pressure-level correction factors that improve long-term AIRS-ground agreement across sites.
113. Does the degree of pore filling by ice particles affect rock deformation during freeze-thaw cycles? A brief discussion and key evidence based on DEM
Core Problem: It is unclear how ice-particle pore filling influences rock deformation through freeze-thaw cycles.
Key Innovation: Uses DEM-based evidence to isolate a micro-mechanical control on freeze-thaw deformation.
114. Dynamic reliability assessment and settlement-limit calibration of train-track-bridge system under data-informed non-Gaussian single- and multi-pier settlement
Core Problem: Calibrate reliable settlement limits for train-track-bridge systems under uncertain non-Gaussian pier settlement.
Key Innovation: Integrates data-informed non-Gaussian settlement models with probability-density evolution for condition-specific reliability domains.
115. Pixels, points and polygons: A dataset and benchmark for multimodal building vectorization
Core Problem: Benchmark automated building vectorization from mixed raster and point-cloud inputs.
Key Innovation: Provides a multimodal dataset and evaluation benchmark spanning pixels, points, and polygon outputs.
116. Viewing-geometry effects on UAV-based tree height retrieval using BRDF-derived vegetation indices
Core Problem: Quantify how viewing geometry biases UAV tree-height retrieval.
Key Innovation: Uses BRDF-derived vegetation indices to analyze and correct geometry effects in UAV height estimation.
117. Quantification of wear evolution and roughness degradation associated with shear deformation of rock joints using transparent rock-like material and visualization method
Core Problem: Measure how rock-joint roughness degrades during shear deformation.
Key Innovation: Uses transparent rock-like material and visualization to quantify wear evolution along sheared joints.
118. Simulation of quasi-static fracture in fractured rock materials using a two-stage joint algorithm within peridynamics
Core Problem: Simulate quasi-static fracture growth in fractured rock materials more robustly.
Key Innovation: Introduces a two-stage joint algorithm within peridynamics for fractured rock failure.
119. An improved two-phase adaptive mesh refinement lattice Boltzmann method for cross-scale flow simulation in porous media
Core Problem: Simulate multiscale two-phase flow in porous media efficiently across scales.
Key Innovation: Adds adaptive mesh refinement to a two-phase lattice Boltzmann framework for cross-scale flow.
120. Optimal frictional design of LIR-DCFP seismic isolators to mitigate internal lateral impacts
Core Problem: Tune frictional properties of seismic isolators to reduce internal lateral impacts.
Key Innovation: Optimizes LIR-DCFP isolator friction design for improved seismic protection.