TerraMosaic Daily Digest: August 31, 2026
Daily Summary
August 31, 2026 is defined by a shift from trigger-based hazard diagnosis toward state-dependent failure physics. A physics-based attractor coupled to manifold detection argues that Hikurangi slow-slip trajectories contain week-ahead earthquake predictability; landslide studies use InSAR-refined inventories, mechanism-aware graph learning, shaking-table monitoring of anti-dip rock slopes, and reliability analysis to resolve susceptibility and failure stage in active mountain belts; and geotechnical papers show that drainage regime, anisotropy, stress rotation, grading, and coupled stress metrics govern liquefaction or instability in fly ash, calcareous sand, Zanda silty sand, loess, frozen soils, mixed rockfill, and expansive ground. Underground studies extend the same logic to delayed rockburst, axial chain rockburst, strainburst, and mine-roof failure, identifying precursor statistics, thermal energy storage, layering effects, and damage evolution as practical controls on violent rupture.
Hydroclimatic and surface-process papers likewise emphasize coupled transport, spatial completeness, and operational timing. Flood work ranges from satellite validation of SFINCS across 499 historical river floods to sediment-conditioned risk mapping for Pakistan, probabilistic completion of FEMA flood-hazard coverage in Louisiana, national forecast-system design for Denmark, and reservoir operation under successive Meiyu storms, while radar- and radiometer-based precipitation studies push toward finer and better-calibrated forcing fields. Related process papers show that tributary mouth bars, cascading check dams, debris-affected bridge scour, trestle safety, and landslide-debris-flow-flash-flood cascades depend on feedbacks among erosion, deposition, and hydraulics rather than on fixed geometry alone.
Cryosphere, drought, wildfire, and transferable observation papers broaden the hazard frame while keeping direct findings distinct from enabling methods. Supraglacial lake fate becomes predictable well before season end; multilayer soil-moisture drought is linked to stronger ecosystem losses; snow, glacier, sea-ice, freeze-thaw, and permafrost studies quantify secondary surface change and infrastructure exposure; and aerial LiDAR resolves wildfire losses and carbon emissions while bushfire work advances building-level vulnerability assessment. In parallel, geometry-aware InSAR restoration and persistent-scatterer selection, geodetic 3D reconstruction, drone-to-satellite registration, geospatial pretraining, damage-detection networks, calibrated AI weather models, neural-operator surrogates, and related imaging or time-series tools strengthen remote-sensing and scientific-ML capacity, but most remain transferable frameworks rather than demonstrated geohazard validations.
Key Trends
Five trajectories organize the August 31, 2026 corpus: state-dependent failure physics, deformation-informed landslide analysis, operational probabilistic flood support, secondary surface-change accounting in cold-region and fire settings, and geometry- plus uncertainty-aware observation methods.
- Failure physics are being resolved as evolving state: Across earthquake forecasting, liquefaction, slope shaking, rockburst, loess, and frozen-soil studies, instability is framed through precursor evolution, coupled mechanics, and stress-path dependence rather than through single threshold metrics.
- Landslide analysis is becoming deformation-informed and mechanism-specific: The landslide cohort combines InSAR-refined inventories, spatially constrained sampling, mechanism-aware graph learning, staged slope-failure experiments, and probabilistic reliability analysis, indicating a move away from generic susceptibility surfaces toward process-structured diagnosis.
- Flood support is turning operational, probabilistic, and sediment-aware: Global hydrodynamic validation, national forecast-system design, reservoir operation under repeated storms, sediment-enhanced flood risk, conditional-diffusion completion of hazard coverage, and improved precipitation estimation all push flood products toward decision-ready forecasting with explicit uncertainty.
- Cold-region and fire studies prioritize secondary change and exposure: Supraglacial lake outcomes, freeze-thaw colluvial inventories, permafrost-corridor distress, snow and glacier retrievals, sea-ice wave attenuation, and wildfire LiDAR differencing focus on how surface change propagates into runoff, infrastructure stress, or carbon and exposure consequences.
- Transferable Earth-observation methods emphasize geometry, calibration, and uncertainty: The broader methods cohort repeatedly encodes acquisition geometry, geodetic reference, compositional structure, or calibrated uncertainty in InSAR, 3D reconstruction, weather forecasting, spectral sensing, and PDE surrogates, but usually stops short of hazard-specific validation.
Selected Papers
The selected papers combine direct studies of earthquakes, landslides, floods, cryosphere change, wildfire effects, and geotechnical instability with a secondary methods cohort in Earth observation, 3D reconstruction, and scientific machine learning. The first group reports hazard-specific findings; the second contributes transferable sensing and inference capacity that still requires domain validation for geohazard use.
1. Supraglacial Lake Fate Is Knowable Long Before the Season Ends
Core Problem: Lake fate classifiers are accurate only after the melt season, leaving no early warning on whether meltwater reaches the ice bed.
Key Innovation: Measures the earliest date each lake-outcome class becomes predictable and shows rapid and slow drainage can be flagged months before full-season closure.
2. Is Seismic Forecasting Possible with Physics-based AI?
Core Problem: Noise and standard filtering obscure deterministic precursor structure linking slow slip, subduction physics, and regional seismicity.
Key Innovation: Combines a physics-based subduction attractor with AI-assisted manifold detection to forecast a Hikurangi earthquake and estimate predictability limits.
3. Centrifuge Modeling of Variable-Rate Cone Penetration Testing in Fly Ash and Its Implications for State Interpretation
Core Problem: Determine how CPT drainage conditions affect state interpretation and undrained failure vulnerability in fly ash.
Key Innovation: Uses variable-rate centrifuge CPTs to derive a drained-versus-undrained resistance framework for identifying contractive, failure-prone layers.
4. Ice jam formation at river confluences: comprehensive field investigation and comparison to laboratory-derived predictive equations
Core Problem: Determine how breakup processes and hydro-environmental factors control ice-jam formation at confluences.
Key Innovation: Supplies rare multi-winter field observations and tests laboratory-derived predictive equations against real confluence jams.
5. Validation of the open-source hydrodynamic model SFINCS on historical river floods at the global scale
Core Problem: Quantify how well SFINCS reproduces historical river floods worldwide.
Key Innovation: Satellite-based validation of 499 flood events plus sensitivity analysis on gauges and DEM resolution.
6. Multilayer soil moisture depletion intensifies drought impacts on global ecosystems
Core Problem: Assess how vertically compound soil moisture drought changes ecosystem impacts globally.
Key Innovation: Frames multilayer soil depletion as a distinct, intensifying drought hazard class.
7. Volumetric Impact Characterization of the 2025 Palisades and Eaton Fires Using Aerial LiDAR
Core Problem: Measure vegetation and building losses from the 2025 Palisades and Eaton fires.
Key Innovation: Sub-meter LiDAR differencing for volumetric loss and bottom-up carbon estimates cross-validated against atmospheric inversion.
8. Multidimensional Quantification of Engineering Distresses and Secondary Periglacial Hazards Along Linear Infrastructure in the Permafrost Region of Northeast China Using UAV-LiDAR and Synchronous Visible-Light Imagery
Core Problem: Quantify infrastructure distress and secondary periglacial hazards along cold-region corridors.
Key Innovation: Synergistic UAV optical-LiDAR framework for multidimensional quantification of several permafrost-related failure modes.
9. Improvement of Flood Risk Model Performance by Incorporating Sediment Factors
Core Problem: Explain and map the 2022 Pakistan flood while accounting for sediment amplification.
Key Innovation: Adds near-real-time satellite sediment information to flood risk modeling and quantifies underestimation in traditional models.
10. Multi-level InSAR coupling and spatially-probabilistic constrained imbalanced sampling for refined landslide susceptibility assessment in active tectonic orogens
Core Problem: Reduce sampling bias and label distortion in landslide susceptibility assessment for the Eastern Himalayan Syntaxis.
Key Innovation: Couples refined InSAR deformation information with spatially constrained negative sampling and semi-supervised imbalanced learning.
11. A regional landslide susceptibility mapping framework considering mechanism uncertainty and cross-class adversarial constraints
Core Problem: Handle mechanism uncertainty and ambiguous typology boundaries in regional landslide susceptibility mapping.
Key Innovation: Encodes cluster-derived landslide types and cross-class contrastive relations in a heterogeneous graph attention framework.
12. Quantitative assessment of the liquefaction resistance of Zanda silty sand using SHAP analysis
Core Problem: Quantify how stress state, density, and loading jointly control liquefaction resistance of Zanda silty sand.
Key Innovation: Uses cyclic simple shear plus SHAP interaction analysis to expose factor coupling and a critical brittle-failure loading threshold.
13. Seismic Response and Progressive Failure of Steep Anti-dip Bedding Rock Slopes with Various Slope Angles via Multi-monitoring Data: Insights from Shaking Table Tests
Core Problem: Determine how slope angle governs seismic response and progressive failure of steep anti-dip bedding rock slopes.
Key Innovation: Combines shaking-table multi-monitoring data to define four failure stages, damage thresholds, and angle-dependent failure modes.
14. Engineering-geological inventory and hazard-oriented classification of freeze-thaw colluvial deposits from multi-epoch Landsat (1990-2024) along the Yunnan-Tibet transportation corridor
Core Problem: Inventory and classify freeze-thaw colluvial deposits along the Yunnan-Tibet transport corridor.
Key Innovation: Pairs multi-epoch Landsat mapping with a hazard-oriented engineering-geological classification for corridor-scale screening.
15. Mechanism of Time-Delayed Rockburst-Inducing disasters in Deep-Buried Tunnels: A microseismic Data-Driven Methodology
Core Problem: Explain and forecast time-delayed rockburst-inducing disasters in deep tunnels.
Key Innovation: Uses microseismic data in a data-driven methodology to identify mechanisms and timing of delayed rockburst hazards.
16. Characteristics and formation mechanism of axial chain rockbursts in deep-buried high-geothermal tunnels: Insight from field monitoring and thermo-mechanical experiments
Core Problem: Explain formation mechanisms of axial chain rockbursts in high-geothermal deep tunnels.
Key Innovation: Combines field monitoring with thermo-mechanical experiments to characterize a distinctive rockburst pattern.
17. Integration of vertical and lateral erosion in modelling landslide mass flows, debris flows and flash floods
Core Problem: Model landslide mass flows, debris flows, and flash floods while accounting for both vertical and lateral erosion.
Key Innovation: Integrates two erosion dimensions within one mass-flow modelling framework for several related hazards.
18. Micromechanical insights into grading effects on cyclic liquefaction resistance of sands
Core Problem: Explain how grading affects cyclic liquefaction resistance of sands at the micromechanical level.
Key Innovation: Links pre-liquefaction contact degradation rates to liquefaction resistance across graded DEM assemblies.
19. Spatial Shift of Dominant Controls and Development Patterns of Tributary Mouth Bars in the Xiaolangdi Reservoir
Core Problem: Managers lack a spatial explanation for tributary mouth bar growth along the reservoir gradient.
Key Innovation: Long-term four-tributary comparison isolates how dominant controls on mouth-bar development shift with distance to dam.
20. FiLM-GPNet: Geometry-Aware Pseudo-Supervised Phase Restoration with Zero-Shot Generalization for Large Temporal InSAR Stacks
Core Problem: Fixed classical interferogram filters break under heterogeneous SAR acquisition geometry, hurting temporal consistency and unwrapping.
Key Innovation: Conditions wrapped-phase restoration on acquisition geometry with FiLM, pseudo-supervision, closure-physics regularization, and per-pixel uncertainty estimation.
21. GeoRay: Gauge-Aware Feed-Forward Satellite 3D Reconstruction in the Geodetic Frame
Core Problem: Feed-forward 3D models do not directly solve absolute geodetic satellite reconstruction with RPC cameras and height-datum ambiguity.
Key Innovation: Introduces gauge-aware absolute-height reconstruction with RPC-ray adapters and calibrated fusion of monocular and multi-view relief cues.
22. Sensitivity-Constrained Neural Operators for Data-Efficient Forward and Inverse Modeling of Partial Differential Equation Systems
Core Problem: Standard neural operators lack sensitivity supervision, limiting forward accuracy and inverse stability for PDE-based hazard modeling.
Key Innovation: Adds sampled Jacobian supervision to neural operators and demonstrates better tsunami source inversion from sparse early gauges.
23. Diffusion-Based Refinement for Kilometer-Scale Probabilistic Precipitation Nowcasting
Core Problem: Existing nowcasts miss fine-scale extreme rainfall while providing weak probabilistic detail for high-impact events.
Key Innovation: Uses residual conditional diffusion to turn a deterministic radar nowcaster into a fast 1 km ensemble that both corrects bias and resolves fine-scale uncertainty.
24. Quantum-Grassmann-Plucker Token Mixing for Deep Learning-Based Post-Disaster Damage Assessment
Core Problem: Post-disaster building damage mapping suffers from class imbalance, ambiguous intermediate states, and weak transfer across events.
Key Innovation: Introduces Grassmann-Plucker token mixing and quantum-inspired variants for paired pre/post disaster image classification with strong unseen-event results.
25. Uncertainty-Aware End-to-End AI Weather Forecasting: Disentangling Observation and Model Contributions
Core Problem: End-to-end AI weather models are deterministic and do not separate observation-driven from model-driven uncertainty.
Key Innovation: Adds aleatoric encoder noise and epistemic MC dropout to produce calibrated, source-attributed probabilistic forecasts.
26. Scour depth prediction at debris-affected bridge piers using interpretable hybrid machine learning models
Core Problem: Synthesize the physics, modeling, monitoring, and design state of knowledge for scour at bridge foundations.
Key Innovation: Provides an up-to-date holistic review spanning sediment motion, predictive models, monitoring practice, design assessment, and research gaps.
27. Impact of Radar-Constrained Effective Drop-Shape Relations on Polarimetric Radar Quantitative Precipitation Estimation in Typhoons
Core Problem: Reduce precipitation bias caused by mismatched drop-shape relations in typhoon radar QPE.
Key Innovation: Derives radar-constrained effective drop-shape relations that transfer better than surface-only typhoon relations.
28. Probabilistic Spatial Completion of FEMA Special Flood Hazard Area Coverage in Louisiana Using Conditional Diffusion and Distributionally Trustworthy Explanation
Core Problem: Estimate missing or unresolved SFHA shares across Louisiana block groups.
Key Innovation: Calibrated hurdle conditional diffusion with spatial blocking and distribution-level explanations for flood-hazard completion.
29. Data-driven reliability evaluation of soil slopes using FOSM-integrated hybrid extreme gradient boosting models
Core Problem: Estimate slope failure probability more accurately under uncertain soil parameters.
Key Innovation: Integrates FOSM with XGB and metaheuristic tuning to predict reliability indices and failure probabilities.
30. Prediction of rockburst tendency of horizontal layered hard rock tunnel in high ground stress environment: A case study
Core Problem: Correct rockburst prediction criteria for layered dolomite tunnels under ultra-high stress.
Key Innovation: Bedding-aware modified energy criterion supported by unloading tests, simulation, and field records.
31. Microscopic Failure Mechanisms of Granite Strainburst: Insights from PFC3D-GBM Simulation
Core Problem: Explain grain-scale failure and energy partitioning during granite strainburst.
Key Innovation: Builds a 3D grain-based DEM with staged damping, crack tracking, and energy accounting to link grain size and depth to strainburst intensity.
32. Only a matter of time? Evaluating modeling strategies while building a national operational flood forecast system for Denmark
Core Problem: Compare modeling strategies while building a national operational flood forecast system for Denmark.
Key Innovation: Evaluates alternative national-scale operational forecasting strategies rather than a purely academic hydrology exercise.
33. DPNet: A disaster perception network for building damage detection using single-temporal remote sensing imagery
Core Problem: Detect building damage from single-temporal remote-sensing imagery.
Key Innovation: Introduces DPNet to infer disaster damage without requiring pre-event imagery.
34. Optimal reservoir flood-control operation under successive rainfall events during the Meiyu season
Core Problem: Optimize reservoir flood-control operation during successive Meiyu rainfall events.
Key Innovation: Targets multi-event operational decision-making for flood-control releases.
35. A Composition-Aware Pretraining Framework for Geospatial Foundation Models
Core Problem: Single-concept geospatial pretraining fails to encode the compositional mixtures that dominate real satellite scenes.
Key Innovation: Predicts fractional land-cover histograms with Earth Mover's Distance to learn composition-aware geospatial embeddings.
36. Ice-thickness based scaling of wave attenuation in sea ice: Application and assessment of wave spectra
Core Problem: Operational sea-ice wave attenuation schemes need better representation of ice-thickness effects and field validation.
Key Innovation: Validates an ice-thickness-aware attenuation scheme against satellite and buoy data and exposes marginal-ice-zone ice-thickness bias.
37. Layout Optimization of Urban Emergency Shelter Sites Under Compound Disaster Scenarios Based on MOGWO
Core Problem: Optimize urban emergency shelter locations under coupled geological and flood disaster scenarios.
Key Innovation: Combines RF hazard mapping, disaster-chain coupling adjustment, G2SFCA suitability, and MOGWO optimization.
38. A granular thermodynamic model for unsaturated frozen soil considering water-ice phase transition
Core Problem: Represent coupled thermal-hydraulic-mechanical behavior of unsaturated frozen soil during phase change.
Key Innovation: Granular thermodynamic constitutive model embedding water-ice transition and frost-heave effects.
39. Moisture migration patterns and deformation characteristics of coarse-grained fill under freeze-thaw cycles
Core Problem: Identify moisture-migration controls and fine-content thresholds governing frost heave in coarse fills.
Key Innovation: Open-system freeze-thaw experiments revealing a sharp frost-heave threshold near 8% fine content.
40. Development and evaluation of an AI-based framework for bushfire vulnerability assessment of buildings
Core Problem: Assess building vulnerability to bushfire with an AI-based framework.
Key Innovation: Applies AI to building-level bushfire vulnerability assessment for disaster-risk reduction.
41. PSSformer: A persistent scatters selection method for SAR interferometry based on temporal-spatial vision transformers
Core Problem: Improve persistent-scatterer selection for SAR interferometry.
Key Innovation: Uses temporal-spatial vision transformers to strengthen persistent-scatterer selection for InSAR time-series analysis.
42. Snow What? Strengths and Limitations of Different Snow Products in Western U.S. Mountains
Core Problem: Western U.S. snow products disagree drastically on peak snow water equivalent.
Key Innovation: LiDAR-based intercomparison identifies which SWE products best capture magnitude and spatial distribution.
43. Pixel-wise Geo-registration of Drone and Satellite Images
Core Problem: Existing cross-view localization benchmarks only provide one GPS label per image rather than dense pixel-wise alignment.
Key Innovation: Introduces SkyReg and trains geometry-aware dense drone-to-satellite registration for per-pixel geolocation.
44. mmIR: Frequency-Space Inverse Rendering for 3D Millimeter-Wave Radar ADC Synthesis
Core Problem: High-resolution 3D mmWave radar training data are scarce and real sensors have limited angular resolution.
Key Innovation: Fits a differentiable FMCW radar renderer to real captures and re-renders dense virtual apertures using LiDAR-guided geometry.
45. Ground-to-Satellite Localization in Unconstrained Image Collections for 3D Scene Reconstruction
Core Problem: Ground-to-satellite localization from unconstrained image collections is unreliable without panoramas or initial GPS.
Key Innovation: Aggregates coarse-to-fine cross-view pose hypotheses over SfM models with KDE consensus to geo-localize and metrically scale 3D reconstructions.
46. Compact Snapshot Spectral Imaging with Calibration-Free Aperture Diffraction
Core Problem: Snapshot spectral imaging is bulky and calibration-heavy, limiting deployment.
Key Innovation: Uses a diffractive lens plus Bayer sensor and theory-derived PSFs with an unfolding transformer to recover full-resolution spectra without calibration.
47. A Controlled Evaluation of Model Rankings and Input Reliance in Surface Water Segmentation
Core Problem: Aggregate IoU rankings obscure stability, input reliance, and scope limits in surface-water segmentation.
Key Innovation: Controlled stress tests separate ranking stability, ancillary-input dependence, and geographic weighting effects on flood datasets.
48. A review of the causes and consequences of the 1930s Dust Bowl Drought, Southern Great Plains, U.S.A
Core Problem: Explain climatic, land-use, and health drivers of the Dust Bowl.
Key Innovation: Integrates circulation, land-surface, agricultural, and remote-sensing evidence into a modern Dust Bowl synthesis.
49. Deep Learning-Based Quantitative Precipitation Estimation Using Ground-Based Microwave Radiometer and Micro-Rain Radar Observations
Core Problem: Improve quantitative precipitation estimation from ground-based microwave radiometer and radar data.
Key Innovation: Fuses dual-polarization radiometer signals and micro-rain radar profiles in deep retrieval models.
50. Machine-Learning-Based Radar Quantitative Precipitation Estimation Using Intra-Hour Temporal Features from Dual-Polarization Observations
Core Problem: Improve hourly radar precipitation estimates using intra-hour dual-polarization temporal features.
Key Innovation: Encodes ten 6-minute radar scans into physically meaningful temporal features for ML QPE.
51. Degradation trigger mechanism in compacted loess under seepage-stress coupling: correlated with microstructural evolution
Core Problem: Explain how seepage-stress coupling drives structural degradation in compacted loess.
Key Innovation: Combines seepage-shear testing with NMR and SEM to separate stress-dominant and seepage-dominant degradation pathways.
52. Mechanisms of moisture-induced deformation in deep loess under constant head conditions
Core Problem: Explain moisture-induced deformation in deep loess under constant head conditions.
Key Innovation: Focuses on wetting-driven deformation mechanisms in deep loess, a key precursor for loess instability.
53. Applied Numerical Modelling of Landscape Evolution in Anthropogenic Landscapes: A Systematic Review
Core Problem: Practical use of landscape evolution models in human-modified landscapes is poorly synthesized.
Key Innovation: PRISMA review of 55 studies clarifying applications, uncertainties, and validation gaps for risk management.
54. Lithospheric Cliff-Induced Asthenospheric Upwelling: Insights From High-Resolution Broadband Lg Attenuation Structure Beneath the Carpathian-Pannonian Region
Core Problem: The mechanism driving intraplate volcanism in the Carpathian-Pannonian region remains uncertain.
Key Innovation: High-resolution broadband Lg attenuation links asthenospheric upwelling patterns to late Cenozoic magmatism.
55. Flat-Slab Subduction and Overriding Plate Architecture Beneath the Northwestern Andes Revealed by Receiver Functions
Core Problem: Northwestern Andes slab geometry and overriding-plate structure were poorly constrained.
Key Innovation: New receiver functions image flat-slab geometry, dual Mohos, and hydrated underplated sediments.
56. Testing the Seismic Detectability of Magmatic Underplating Beneath the Hawaiian Ridge: A Synthetic Wavefield Approach
Core Problem: It is unclear whether modern seismic data should detect Hawaiian magmatic underplating.
Key Innovation: Wavefield simulations show thick laterally extensive underplates would leave signatures absent from observations.
57. Understanding Temporal Semantic Stability in Open-Vocabulary UAV Perception through Metric 3D Fusion
Core Problem: Open-vocabulary UAV segmentation flickers over time despite high aggregate agreement.
Key Innovation: Metric 3D voxel fusion quantifies semantic belief drift, persistence, and uncertainty in world space.
58. Multi-Sensor Mapping of Vulnerable Urban Settlements Using SAR, Multispectral, and Hyperspectral Imagery: A Case Study in Córdoba, Argentina
Core Problem: Informal settlements are hard to map consistently from heterogeneous EO data and incomplete inventories.
Key Innovation: Compares fusion strategies and shows late fusion plus hyperspectral support best balances detection and selectivity.
59. Distributed Semantic Segmentation With Improved Rate-Distortion Trade-Off
Core Problem: Existing split-segmentation codecs are suboptimal in the extreme low-bitrate regime.
Key Innovation: Introduces new source codecs that improve rate-distortion trade-offs for distributed semantic segmentation.
60. Variable-Granularity Tokenization for High-Resolution Object Detection
Core Problem: Uniform ViT token grids waste compute and miss tiny objects in high-resolution aerial imagery.
Key Innovation: Training-free variable-granularity tokenizer preserves salient small targets while cutting encoder cost.
61. ReconSplat: Generalizable 3D Scene Reconstruction Beyond Observed Views
Core Problem: Feed-forward 3D reconstruction methods struggle to balance plausible extrapolation with geometric consistency.
Key Innovation: Combines 3D Gaussian splatting with diffusion-guided appearance and geometry refinement for consistent novel views and depth.
62. RoSe-SLAM: Robust Semantic-Aware Gaussian Splatting SLAM from Dynamic Monocular Videos
Core Problem: Dynamic objects break standard monocular SLAM assumptions and degrade tracking and mapping.
Key Innovation: Combines foundation-model semantics, motion masking, and Gaussian splatting for dynamic-aware camera tracking and static-scene reconstruction.
63. Di²CycleSB: Towards High-Quality Unsupervised Nighttime Visibility Enhancement via Schrödinger Bridge Transformer
Core Problem: Nighttime enhancement methods struggle with non-uniform glow and light-effect contamination under unsupervised settings.
Key Innovation: Formulates light suppression as a cycle Schrodinger bridge with dynamic integral-image priors and transformer guidance.
64. AdapToPASS: Ambiguity-aware Adaptive Spherical Transformer for Panoramic Semantic Segmentation
Core Problem: Panoramic segmentation models are brittle to unseen spherical transformations and viewpoint ambiguity.
Key Innovation: Adds ambiguity-aware spherical attention, bifocal representation, and boundary supervision for robust PASS.
65. NFAD: Nuisance-Filtered Anomaly Detection Under Distribution Shift
Core Problem: Anomaly detectors trained on controlled imagery break when illumination, background, or viewpoint shift normal samples.
Key Innovation: Learns a nuisance subspace from content-preserving perturbations and removes it differently for image-level and pixel-level anomaly detection.
66. GramLoop: Training-Free Gram-Gated Replay for Robust Dense Prediction
Core Problem: Frozen vision backbones degrade under distribution shift and standard test-time adaptation often changes weights or heads.
Key Innovation: Replays selected transformer blocks and accepts updates through Gram-consistency gating to improve shifted dense prediction without retraining.
67. Dynamic-Robust Photometric-Semantic Reconstruction for Open-Vocabulary 3D Scene Understanding
Core Problem: Static-scene assumptions break joint novel-view synthesis and open-vocabulary segmentation in dynamic environments.
Key Innovation: Separates transient dynamics before latent aggregation and jointly learns motion-aware photometric and semantic reconstruction.
68. SGPDFuse: Semantically-Guided Physics-Disentanglement General Multi-Modal Image Fusion
Core Problem: Blind multimodal fusion mixes useful scene content with modality-specific physical degradations.
Key Innovation: Builds a semantic-physical bridge that disentangles invariant structure from transient degradation and regularizes fusion with semantic alignment and texture fidelity.
69. DARD: Zero-Shot Degradation-Aware Retinex-Guided Diffusion for Low-Light Image Enhancement
Core Problem: Zero-shot low-light enhancement often drifts structurally and semantically without reliable scene priors.
Key Innovation: Extracts degradation-aware Retinex priors at test time and injects them into diffusion sampling with adaptive frequency fusion and guided refinement.
70. Polis: 3D Self-Supervision at City Scale
Core Problem: Most 3D self-supervised models are pretrained on indoor or autonomous-driving data and transfer poorly to city-scale aerial point clouds.
Key Innovation: Designs a native outdoor point-cloud pretraining objective and view-sampling strategy tuned to city-scale aerial geometry.
71. SS-ESOAP: Self-Scaled Adaptive Preconditioning for Physics-Informed Learning
Core Problem: PINNs often stall on ill-conditioned objectives and existing scalable preconditioners are insufficiently accurate on stiff problems.
Key Innovation: Adds a secant-energy scalar correction and adaptive basis-update handling to SOAP-style Kronecker preconditioning.
72. GeoAgent: Evaluating VLM Geolocalization Through Embodied Navigation
Core Problem: Static image geolocalization benchmarks do not reflect embodied observation gathering needed for real geolocation tasks.
Key Innovation: Introduces an embodied Street View benchmark where VLM agents navigate, gather evidence, and refine geolocation predictions.
73. ARMOR: Manifold-Oriented Training for Adversarially Robust Aerial Object Detection under Data Scarcity
Core Problem: Aerial object detectors remain vulnerable to physical adversarial patches when site-specific training data are scarce.
Key Innovation: Uses manifold-oriented training with background masking and randomized object patching to improve robustness efficiently.
74. BEACON: Behavioral and Semantic Enrichment of AlphaEarth Embeddings through Tri-Modal Contrastive Learning
Core Problem: Image-only geospatial foundation embeddings weakly encode human activity and urban function.
Key Innovation: Tri-modal contrastive learning aligns earth observation embeddings with POI semantics and visitation behavior while keeping image-only deployment.
75. SPLG-Mamba: Structure-Preserving Local-Global Mamba Network for Salient Object Detection in Optical Remote Sensing Images
Core Problem: Optical remote-sensing salient object detection often produces fragmented and structurally incomplete targets.
Key Innovation: Combines detail recalibration, local-global Mamba modeling, and gated cross-scale fusion to preserve structure.
76. LLMODE: Aligning ODEs with LLMs via Gated Token Injection for Irregular Spatio-Temporal Forecasting
Core Problem: LLM-based spatiotemporal forecasters struggle with irregular sampling, asynchronous timing, and limited context windows.
Key Innovation: Uses graph-aware ODE encoding, fixed-budget token resampling, and gated memory injection into a frozen LLM.
77. A Calibration Audit of Confidence in Feed-Forward 3D Reconstruction
Core Problem: Feed-forward 3D reconstruction confidence is often overconfident when interpreted as calibrated uncertainty off training conditions.
Key Innovation: Provides a broad calibration audit and simple rescaling laws that partially correct uncertainty magnitude.
78. XDG: Accelerated Visual Disambiguation
Core Problem: Structure-from-motion fails on visually similar but physically distinct surfaces, and current disambiguation is costly.
Key Innovation: Lightweight LoRA adaptation of a 3D foundation model uses camera tokens for fast doppelganger classification.
79. InspectorGPT: A Comparative Reasoning Enhanced VLM for Comprehensive Industrial Anomaly Detection
Core Problem: Industrial anomaly detectors generalize poorly and often lack pixel-level localization and interpretable reasoning.
Key Innovation: Uses reference-based comparative VLM reasoning plus a separate segmentation branch fused through task vectors.
80. GridFlow: Structured Latent Flow for Seamless City-Scale 3D Point Cloud Generation
Core Problem: Existing point-cloud generators cannot create seamless city-scale 3D scenes from partial remote-sensing observations.
Key Innovation: Uses a grid-aligned latent space and flow-based geometry synthesis to maintain cross-tile consistency at scale.
81. PhasorNet: Learning Structure from Frequency for Real-Time Stereo Matching
Core Problem: Stereo matching remains unreliable in fine-structure, reflective, and transparent regions.
Key Innovation: Injects Fourier phase cues into attention and focuses refinement on edge and high-error regions.
82. Joint Spatiotemporal Spectral Neural Operators for Learning PDEs on Irregular Domains
Core Problem: Neural operators struggle to learn PDE solution maps on irregular geometry-dependent domains without costly warping or embeddings.
Key Innovation: Combines graph Laplacian spatial spectra with temporal Fourier kernels to learn geometry-aware space-time operators directly on irregular meshes.
83. Continual Test-Time Adaptation via Entropy Sensitivity-Guidance in Strict Online Setting
Core Problem: Batch-size-one online adaptation drifts or collapses when no source data are available.
Key Innovation: Uses entropy changes under structured erasures as a per-sample signal to gate continual adaptation and recovery.
84. SVI2LoD3: Agent-Driven Reconstruction of LoD3 Facade Openings in Semantic 3D City Models from Volunteered Street View Imagery using Large Language and Visual Models
Core Problem: LoD3 facade reconstruction usually needs costly labeled data and often fails to preserve valid semantic hierarchies.
Key Innovation: Builds an agent-driven zero-shot pipeline that reconstructs facade openings into CityGML-conform models and proposes a feature-based evaluation metric.
85. Selection, Representation, and Execution in Sparse Fourier Neural Operators
Core Problem: Sparse Fourier neural operators may shrink parameter counts without actually reducing deployed runtime.
Key Innovation: Separates representational sparsity, stored sparsity, operation count, and real latency to show when sparse FNOs are truly useful.
86. A Hybrid State-Space Approach for Census-Tract Population Estimation
Core Problem: Population estimation from imagery usually discards the census unit structure and introduces spatial bias through disaggregation.
Key Innovation: Represents each census tract as a polygon-masked image and uses a hybrid state-space attention model to predict tract population directly.
87. Efficient and High-Quality Depth Estimation via Pixel-Space Diffusion with Linear Attention
Core Problem: Generative monocular depth models are accurate but too expensive at high resolution.
Key Innovation: Combines one-step pixel-space diffusion, linear attention, and coarse-to-fine structural refinement to deliver sharper depth at much lower latency.
88. Multivariate Scientific Data Compression with Learned Cross-Variable Latent Decorrelation and Autoregressive Entropy Modeling
Core Problem: Improve error-controlled compression of coupled scientific fields with heterogeneous dependencies.
Key Innovation: Joint latent decorrelation and autoregressive entropy modeling inside a multivariate learned compressor.
89. Learning PDE Time-Stepping with Neural Cellular Automata
Core Problem: Learn stable long-horizon PDE time-stepping across varying initial conditions.
Key Innovation: Neural cellular automata learn a local homogeneous update rule that extrapolates in time better than several baselines.
90. HorizonNet for visual terrain navigation
Core Problem: Localize unmanned surface vessels in coastal terrain using panoramic imagery and terrain profiles.
Key Innovation: Two-stage horizon extraction with camera-leveling and Fourier-domain DEM correlation for GPS-level positioning.
91. When the Martingale Never Stops Firing: Anytime-Valid Gating on Real Forecast Streams
Core Problem: Conformal test-martingale gates can generate persistent false alarms when adaptive forecast streams violate exchangeability.
Key Innovation: Uses null-calibration controls and mechanism traces to expose systematic false firing, then demonstrates robust Huber-style update gating.
92. Minerals in the Wild: A Hyperspectral-XRF Dataset for Elemental Composition Estimation
Core Problem: Address the lack of ground-truthed datasets for hyperspectral mineral and elemental characterization.
Key Innovation: Public hyperspectral-XRF rock dataset with a USGS-dictionary pruning baseline for elemental estimation.
93. DiffSAC: Diffusion-guided Sampling for Consensus-based Robust Estimation
Core Problem: Sample-consensus estimators waste computation on many bad minimal sets before finding a valid hypothesis.
Key Innovation: Uses a diffusion model conditioned on geometric features to sample high-quality minimum sets with far fewer hypotheses.
94. Season-Aware Hybrid Convolutional-Transformer for Antarctic Sea Ice Concentration Forecasting
Core Problem: Sea-ice forecasting must jointly capture local spatial structure, long temporal dependence, and strong seasonality.
Key Innovation: Adds month-aware positional encoding and seasonal temporal bias to a hybrid convolutional-transformer forecasting model.
95. Liquid Gated Attention
Core Problem: Existing time-series models either mishandle irregular intervals or rely on sequential solvers that scale poorly.
Key Innovation: Introduces interval-aware liquid gated attention with linear-time parallel computation for continuous-time sequence modeling.
96. PixelIR: Fidelity-Perception Decoupling via Pixel-Space Image-Residual Flow Matching for Efficient One-Step Real-World Super-Resolution
Core Problem: Real-world super-resolution entangles fidelity and perceptual detail synthesis, hurting control and efficiency.
Key Innovation: Decouples faithful reconstruction and residual detail generation in pixel space, then distills the system to one-step inference.
97. VCAR: Training-Free 3DGS Segmentation via View Completeness and Axis-Aware Boundary Refinement
Core Problem: 3D Gaussian-splatting segmentation usually needs per-scene distillation and still produces blurry boundaries.
Key Innovation: Adds completeness-driven view augmentation and axis-aware anisotropic boundary refinement for training-free 3DGS segmentation.
98. Analytic Dynamics: Learning Physics-Grounded Representation for Fast Intrinsic Dynamics Inference from Monocular Videos
Core Problem: Monocular video models struggle to infer intrinsic dynamics when they lack intermediate physical-state representations.
Key Innovation: Aligns video features with privileged physics-state representations to infer dynamics and material parameters without deployment-time simulation.
99. BRF-GS: Hyperspectral Bidirectional Reflectance Factor Modeling and Image Generation Based on 3D Gaussian Splatting
Core Problem: Existing 3D radiative transfer models are too cumbersome for efficient multi-angle hyperspectral BRF generation.
Key Innovation: Combines BRDF-driven Gaussian splatting, reliable-band initialization, and staged training for hyperspectral BRF modeling.
100. Machine Learning-Enhanced Tabu Search for Tactical Wireless Network Design
Core Problem: Tabu search for realistic wireless-network design is slowed by expensive move evaluations.
Key Innovation: Trains a graph neural network on search trajectories to rank promising candidate moves.
101. From Location Phrases to Geographic Entities: Task-Adapted Retrieval for People Search
Core Problem: Free-form location phrases are hard to normalize, disambiguate, and map to structured geographic entities.
Key Innovation: Task-adapted bi-encoder with alias calibration, ambiguity-aware negatives, and editable entity documents.
102. Spectral-Embedded Operator Learning for Three-Phase Interfacial Flow: A Ternary Cahn-Hilliard-Navier-Stokes Benchmark
Core Problem: Operator-learning choices for constrained multiphase flows are underbenchmarked.
Key Innovation: Introduces a ternary CHNS benchmark and shows Chebyshev spectral trunks outperform standard coordinate encodings.
103. APPSolver: Adaptive Patch Partitioning for Point-Wise Ship Flow Prediction on Unstructured Meshes
Core Problem: Attention-based flow surrogates become computationally expensive on large, spatially nonuniform CFD point sets.
Key Innovation: Adaptive quadtree patching compresses irregular mesh slices while preserving local flow-field structure and an explicit accuracy-efficiency tradeoff.
104. Neural ODE enhanced linear mixed effect models for estimating complex association patterns of time-varying covariates with the marker trajectory
Core Problem: Linear mixed models require hand-specified exposure-history effects for time-varying covariates.
Key Innovation: Embeds a Neural ODE inside an LMM and estimates counterfactual trajectory contrasts.
105. TSExplorer: An interactive data annotation and exploration tool for time-series data
Core Problem: Researchers need interactive tooling to inspect and refine labels in high-dimensional time series.
Key Innovation: Cross-platform annotation and exploration with multiple linked 2D feature-space views.
106. GAFT: Geo-Anchored Fine-Tuning for Hazard Identification from Rare Failures
Core Problem: Rare failure labels and weak localization let models learn spurious cues instead of true hazard structures.
Key Innovation: Geo-Anchored Fine-Tuning aligns LoRA attention with geometry-derived priors to improve rare hazard identification.
107. Implementing neural network mixed-effects models in Template Model Builder (TMB)
Core Problem: Neural mixed-effects models are hard to implement because objective functions and gradients are usually derived manually.
Key Innovation: A general Template Model Builder framework using autodiff and Laplace approximation for flexible neural mixed-effects estimation.
108. Alert: Learning Trigger Functions for Early Classification of Time Series using Deep-RL
Core Problem: Early classification systems rely too heavily on handcrafted trigger rules for deciding when to predict.
Key Innovation: A Deep-RL trigger-learning framework and improved Alert+ design for better accuracy-delay tradeoffs.
109. Keyframe-Centric State-Space Modeling for Burst Image Super-Resolution
Core Problem: Burst image super-resolution wastes computation on non-key frames and scales poorly with burst length.
Key Innovation: A keyframe-centric BurstMamba design with gather-aggregate-scatter transfer and wavelet-conditioned state updates.
110. Large-Scale Bayesian Tensor Reconstruction via Approximate Message Passing
Core Problem: Bayesian CP tensor decomposition scales poorly because variational updates require repeated matrix inversions.
Key Innovation: CP-GAMP avoids high-dimensional inversions while estimating rank and noise and matching state-evolution benchmarks.
111. JVLGS: Joint Vision-Language Gas Leak Segmentation
Core Problem: Infrared gas plumes are blurry, non-rigid, and frequently confused with noise or background artifacts.
Key Innovation: Fuses visual and textual cues with adaptive postprocessing to segment leak events robustly in supervised and few-shot settings.
112. Emulating the Forced Response of Climate Models with Generative Machine Learning
Core Problem: Earth system models are too computationally expensive for broad scenario ensembles across forcing pathways.
Key Innovation: Generative emulator conditions on external forcings and extrapolates to unseen SSP scenarios while remaining physically consistent.
113. Neural 3D Object Reconstruction with Small-Scale Unmanned Aerial Vehicles
Core Problem: Enable autonomous high-fidelity 3D scanning with sub-100 g UAVs under severe platform constraints.
Key Innovation: Combines real-time SfM feedback, adaptive viewpoint planning, and final NeRF reconstruction in a dual-pipeline system.
114. SPADE: A Large Language Model Framework for Soil Moisture Pattern Recognition and Anomaly Detection in Precision Agriculture
Core Problem: Detect wetting events and sensor anomalies from soil-moisture time series without task-specific training.
Key Innovation: Turns time series into text and uses prompted GPT-4.1 for zero-shot joint event detection, anomaly classification, and reporting.
115. Evaluation of soil disturbance due to sonic drilling using instrumentation and cone penetration test measurements
Core Problem: The magnitude, spatial extent, and testing consequences of soil disturbance caused by sonic drilling are poorly quantified.
Key Innovation: Combines in situ motion instrumentation with before-after CPT measurements to delineate drilling disturbance and its influence on sampling.
116. Nonlinear response and damage evolution mechanisms of subsea tunnels in fault zones under static-dynamic loading
Core Problem: Explain nonlinear damage evolution of subsea tunnels in fault fracture zones under static load and blasting impact.
Key Innovation: Identifies staged failure evolution, a critical instability threshold, and fault-zone reinforcement priorities using scaled tests plus simulation.
117. Endurance time analysis-based multi-hazard seismic fragility assessment of a 10 MW monopile offshore wind turbine with nonlinear soil-structure interaction
Core Problem: Assess seismic fragility of a monopile offshore wind turbine under multi-hazard loading with nonlinear soil-structure interaction.
Key Innovation: Applies endurance time analysis to multi-hazard fragility assessment with nonlinear soil interaction.
118. Seismic uplift capacity of strip anchor plate embedded in cohesionless soil beyond slope using modified-pseudo dynamic method
Core Problem: Estimate seismic uplift capacity of a strip anchor embedded in cohesionless soil beyond a slope.
Key Innovation: Combines modified pseudo-dynamic LB-FELA with ANN and GPR surrogates to map seismic uplift capacity across slope and soil parameters.
119. Global surface mining and land reclamation of time series from 1985-2022
Core Problem: Map the global history of surface mining disturbance and land reclamation from 1985 to 2022.
Key Innovation: Builds the first global mine-disturbance and reclamation time series across 74,000 mines.
120. A global 30 m disturbance-recovery age dataset for young natural and planted forests (1985-2024)
Core Problem: Estimate disturbance-recovery age for young natural and planted forests globally at 30 m resolution.
Key Innovation: Integrates long-term satellite observations to produce a global forest-age dataset with pixel-level uncertainty.
121. Berkeley Earth Surface Temperature - High-Resolution (BEST-HR): a 0.25° Global Gridded Temperature Data Set for Climate Monitoring
Core Problem: Create a high-resolution global gridded surface-temperature reconstruction from station and ocean observations.
Key Innovation: Delivers a 0.25 degree thermometer-based global temperature dataset with unprecedented local detail.
122. Implementation of predicted rime mass in the bin microphysics scheme DESCAM 3D: evaluation for an idealized squall line system and a heavy snowfall event during ICE-POP 2018
Core Problem: Improve a bin microphysics scheme by predicting rime mass and testing it in squall-line and heavy-snowfall cases.
Key Innovation: Implements a smooth unrimed-to-graupel transition that improves agreement with snowfall observations.
123. Integrating Physical-Based Xinanjiang Model and Deep Learning for Interpretable Streamflow Simulation: A Multi-Source Data Fusion Approach across Diverse Chinese Basins
Core Problem: Improve interpretable streamflow simulation across diverse basins.
Key Innovation: Fuses Xinanjiang physics, TCN-GRU, and RF-based feature integration with interpretability.
124. Impacts of cascading check dams on sediment yield in the Middle Yellow River Basin: insights from 50 years of grid-cell-level simulation
Core Problem: Estimate long-term basin-scale sediment yield under cascading check dams.
Key Innovation: Integrates check-dam trapping into RUSLE-connectivity-sediment delivery simulation.
125. Physics-based active-learning surrogate model for rapid safety assessment of temporary steel trestle bridges under construction vehicle loads
Core Problem: Rapidly assess trestle bridge safety under scour, degradation, and vehicle loads.
Key Innovation: Uses active-learning GPR around the safety limit state to build real-time safety maps.
126. A Temporal Analysis of Wildfires in Spain Through the Use of Multi-Database Research
Core Problem: Characterize Spanish wildfire trends and database inconsistencies.
Key Innovation: Cross-database temporal analysis of wildfire counts, large fires, and documentation limits.
127. GeoAI-Driven Wetland Change Analysis in the Sangamon River Watershed (2000-2025): A Comparative Assessment of Machine Learning and Deep Learning Approaches
Core Problem: Long-term wetland loss and classifier performance need consistent evaluation across Landsat, Sentinel, SAR, texture, and terrain inputs.
Key Innovation: Benchmarks pixel- and object-based ML and deep-learning models in a cloud workflow and quantifies 2000-2025 wetland change.
128. A Linear Attention Framework with Dual-Axis Multi-Scale Fusion for Fine-Grained Eucalyptus Change Detection
Core Problem: Detect weak plantation appearance and disappearance in high-resolution imagery.
Key Innovation: Linear-attention encoder with dual-axis change extraction and multi-scale fusion.
129. NWCSAF High Resolution Winds (NWCSAF GEO-I HRW) Stereo AMVs over the Atlantic Ocean
Core Problem: Evaluate stereo versus non-stereo height assignment for atmospheric motion vectors.
Key Innovation: Operational stereo AMV height assignment comparison against ERA5 and EarthCARE profiles.
130. Performance Analysis of BDS-3 PPP-B2b During the Satellite In-Orbit Upgrade Period
Core Problem: Satellite upgrades can disrupt BDS-3 PPP-B2b coverage, correction recovery, convergence, and positioning accuracy.
Key Innovation: Decodes receiver-level correction messages and quantifies service recovery, geometry, convergence, and positioning performance through the upgrade period.
131. Generalized Matérn Process for GNSS Coordinate Series Noise Modeling
Core Problem: Unify GGM and Matern noise representations for GNSS coordinate series.
Key Innovation: Introduces a generalized Matern process that continuously links two common geodetic noise families.
132. An Earth-Limb-Constrained Framework for On-Orbit Geometric Calibration of GEO Wide-Field Area-Array Cameras
Core Problem: Calibrate GEO wide-field cameras without relying on GCPs or stellar observations.
Key Innovation: Uses Earth-limb geometry plus terrain-elevation constraints for on-orbit calibration.
133. Satellite Remote Sensing of a Melting Glacier Albedo: Examples from EnMAP and an Intercomparison with Other Satellite and Ground Measurements
Core Problem: Retrieve and benchmark melting glacier broadband albedo from hyperspectral satellite data.
Key Innovation: Topography- and atmosphere-corrected EnMAP albedo retrieval with multi-source comparison and uncertainty discussion.
134. A Practical Framework for Surface Water Extraction from GF1/GF6 Wide-Field-View Imagery
Core Problem: Extract fragmented surface water reliably from GF1/GF6 wide-field imagery.
Key Innovation: Seven-channel UNet++ with residual logit refinement, edge supervision, and leakage-aware scene partitioning.
135. A nonlinear constitutive model for fine sandstone considering freeze-thaw induced compaction and damage evolution
Core Problem: Existing damage models poorly capture the prolonged compaction-hardening response of freeze-thaw fissured sandstone.
Key Innovation: Introduces strain-driven compaction compliance and an explicit statistical damage model validated across several lithologies.
136. Experimental study on the dynamic characteristics of sandstone-mudstone mixed rockfill based on large-scale cyclic triaxial tests
Core Problem: The effect of mudstone content on cyclic response and permanent deformation of sandstone-mudstone rockfill is insufficiently constrained.
Key Innovation: Large-scale cyclic triaxial tests and calibrated residual-deformation parameters isolate mudstone-content effects on seismic material response.
137. Effect of inherent anisotropy and principal stress rotation on the cyclic response of calcareous sand
Core Problem: The coupled effects of inherent anisotropy and principal stress rotation on cyclic pore-pressure, deformation, and liquefaction resistance are poorly constrained.
Key Innovation: Biaxial principal-stress-rotation tests identify a non-monotonic bedding-angle control on liquefaction resistance and provide fitted pore-pressure, strain, stiffness, and damping models.
138. Complex spatial-temporal distribution of glaciovolcanic interactions at Copahue volcano, Southern Volcanic Zone of the Andes
Core Problem: Reconstruct the spatial-temporal pattern of glaciovolcanic interactions at Copahue volcano.
Key Innovation: Maps interaction patterns that may inform later volcanic or cryospheric hazard interpretation.
139. Late Quaternary fire-induced erosion, impacts and innovations in the southern Levant
Core Problem: Synthesize evidence for wildfire-linked hillslope erosion and valley redeposition during two Late Quaternary intervals.
Key Innovation: Integrates multiple paleoenvironmental proxies to frame fire-erosion cascades as drivers of geomorphic instability.
140. Disentangling diverse findings on how urbanization, vegetation, and background climate shape the surface urban heat island at large scales
Core Problem: Reported controls on surface urban heat islands vary across regions because urbanization, vegetation, and climate effects are difficult to separate.
Key Innovation: Uses large-scale remote sensing to isolate the relative contributions and interactions of urban form, vegetation, and background climate.
141. Segmentation-guided and attention-enhanced GAN for SAR image restoration under multi-type interference
Core Problem: Restore SAR imagery contaminated by multiple interference types.
Key Innovation: Combines segmentation guidance and attention-enhanced GAN restoration for cleaner SAR inputs.
142. Physics-informed regression-residual reconstruction and group fusion dual-attention transformer for methane plume detection
Core Problem: Detect methane plumes more robustly in remote-sensing imagery.
Key Innovation: Fuses physics-informed residual reconstruction with dual-attention transformers for plume detection.
143. A multi-scale spatial-conditioned spectral routing network for hyperspectral and multispectral image fusion
Core Problem: Fuse hyperspectral and multispectral imagery across scales.
Key Innovation: Uses spatial-conditioned spectral routing to improve multi-scale image fusion for downstream mapping tasks.
144. Frost cracking improves prediction of millennial-scale erosion flux in weathering-limited catchments on the Tibetan Plateau
Core Problem: Test whether frost cracking improves prediction of millennial-scale erosion flux on the Tibetan Plateau.
Key Innovation: Shows frost-cracking intensity adds regime-dependent predictive power beyond local relief in cold weathering-limited basins.
145. Low-energy microseismic event identification during TBM tunnelling at Beishan URL with a hybrid signal-processing and deep-learning framework
Core Problem: Identify low-energy microseismic events during TBM tunnelling.
Key Innovation: Combines signal processing and deep learning for weak-event recognition in noisy tunnelling data.
146. Mechanism-informed data-driven forecasting of roof cavities in longwall mining
Core Problem: Roof cavities in longwall mines need earlier, physically interpretable forecasting from monitoring data.
Key Innovation: Combines mechanism constraints with data-driven forecasting to target cavity formation and operational warning.
147. Coastal aquifer vulnerability to salinization under sea-level rise and groundwater abstraction pressure in a Mediterranean karst system
Core Problem: Assess coastal-aquifer salinization vulnerability under sea-level rise and pumping pressure.
Key Innovation: Frames salinization as a compounded vulnerability problem in a Mediterranean karst aquifer.
148. Global assessment of river flow regimes simulated by reach-level streamflow datasets
Core Problem: The ability of reach-level streamflow datasets to reproduce global river flow regimes requires systematic evaluation.
Key Innovation: Provides a global assessment of simulated river-regime behavior to identify strengths and limitations relevant to hydrologic hazard analysis.
149. Shrinkage cracks in expansive soils: Field observations, numerical modelling, and design implications for residential structures
Core Problem: Characterize shrinkage cracks in expansive soils and their implications for residential structures.
Key Innovation: Combines field observation and numerical modelling to connect expansive-soil cracking with design implications.