TerraMosaic Daily Digest: August 23, 2026
Daily Summary
The direct geohazard papers resolve failure as a product of evolving internal state and landscape structure rather than trigger magnitude alone. A Bayesian susceptibility model for the 2015 Gorkha earthquake links coseismic landslides in Nepal to river-incision forcing projected onto coupled hillslopes; Sentinel-1, optical, and UAV evidence in Heifangtai tracks irrigation-driven loess creep shaped by groundwater, freeze-thaw, and rainfall; and a regional inventory from Southern Italy shows landslide clustering organized by lithology, geomorphology, and tectonically conditioned landscape evolution. Related slope studies extend this state-resolved view to coseismic runout prediction, explainable slope-stability surrogates, shallow interface reinforcement, tailings-dam treatment heterogeneity, and mining-induced subsidence.
Hydraulic, seismic, and deformation studies push toward coupled process representation under operational constraints. Post-wildfire channel modelling shows how far two-dimensional hydraulics can be carried with tiered remote-sensing and sparse field data, while a hydro-sediment-morphodynamic breach model captures overtopping, erosion, bed change, and side collapse in landslide dams. Flood papers add objective-specific sensor placement, teleconnection-sensitive precipitation diagnostics, and urban heat-pluvial tradeoff analysis; seismic papers show that Cascadia coupling maps shift when elastic heterogeneity and viscoelastic flow are included, near-fault video can constrain the timing of 2026 Kumamoto surface rupture, and monitoring-informed sensor placement can improve rapid estimates of post-earthquake bridge traffic capacity. Land-subsidence frameworks gain clarity when cumulative and differential deformation are separated and InSAR atmospheric residuals are more aggressively corrected.
Cryosphere and transferable-method papers broaden the observational and analytical base while remaining distinct from direct hazard validation. Daily 100 m Greenland meltwater mapping, multidecadal Tibetan Plateau glacier and albedo analysis, regional glacier velocity retrieval in Peru, active-layer moisture mapping, snowmelt-phase detection, avalanche classification, Tibetan Plateau winter-regime analysis, and a penetration-aware Arctic sea-ice microwave dataset refine climate-sensitive observations relevant to downstream risk. In parallel, multimodal remote sensing, building-footprint and damage mapping, calibration-aware forecasting, identifiability analysis for partially observed dynamics, and 3D or hyperspectral reconstruction enlarge the toolkit for hazard science, but these studies mainly validate sensing or inference performance in their stated domains rather than in geohazard deployment broadly.
Key Trends
Five trajectories organize the August 23 corpus, with direct geohazard findings on failure, deformation, and hydroclimatic forcing running alongside a strong but clearly transferable methods stream.
- Preconditioning is being resolved as structure plus state: Across earthquake-triggered landslides, irrigation-driven loess instability, Southern Italian slope inventories, land subsidence, tailings treatment, and mining deformation, the decisive variables are coupled geomorphic setting, groundwater or pore-pressure history, material heterogeneity, and cumulative deformation state rather than event magnitude in isolation.
- Flood and dam studies are converging on process-coupled operational models: Post-wildfire hydraulics, landslide-dam breaching, urban flood sensor design, precipitation-product evaluation, satellite-derived radar reflectivity retrieval, and river-flat delineation all prioritize how measurement design and coupled hydraulics, sediment transport, and topographic representation control usable hazard forecasts.
- Seismic hazard inference is becoming medium-aware and sequence-aware: Cascadia coupling inversions, Kumamoto surface-fault video analysis, monitoring-informed bridge functionality assessment, offshore wind fragility under mainshock-aftershock loading, finite-deformation pore-pressure theory, and rockburst-proneness criteria show a common shift toward representing elastic structure, observational state, sequential loading, and energetics explicitly.
- Cryosphere diagnostics are moving to finer temporal and spatial states: Greenland meltwater downscaling, Tibetan Plateau glacier and active-layer datasets, Arctic sea-ice microwave emission retrievals, Cordillera Blanca velocity estimates, snowmelt-phase detection, avalanche classification, and winter hydrology regime analysis all extract intermediate surface or subsurface states that are more actionable than coarse seasonal summaries.
- Transferable AI and remote-sensing methods are emphasizing trustworthy cross-modal inference: Multimodal object detection, SAR-optical building mapping, remote-sensing image-text retrieval, forecasting audits, clustered-threshold statistics, sequential calibration, uncertainty-aware graph models, and operational anomaly detection all focus on fidelity, calibration, and domain shift. Their value to geohazards is enabling rather than already demonstrated unless the paper validates a hazard application directly.
Selected Papers
The selected papers combine domain-validated studies of landslides, flood hydraulics, subsidence, seismic deformation, cryosphere change, wildfire, and tailings safety with a parallel set of remote-sensing and AI methods for sensing, mapping, forecasting, and verification. Read the first group as substantive hazard findings and the second as transferable analytical advances whose geohazard utility remains application-dependent unless explicitly tested.
1. Rapid Hydraulic Modelling in Post-Wildfire Channels Using a Tiered Framework
Core Problem: Burned watersheds need reliable 2D hydraulics even when field data are sparse or unsafe to collect.
Key Innovation: Three-tier calibration framework comparing remote sensing only, limited field data, and full bathymetry.
2. Competing Effects of Elastic Heterogeneity and Viscous Flow on Interseismic Coupling at Cascadia
Core Problem: Coupling estimates can be biased when elastic heterogeneity and viscoelastic flow are ignored.
Key Innovation: Bayesian inversion that jointly tests elastic structure and earthquake-cycle viscoelasticity in coupling maps.
3. Time-Resolved Surface-Fault Displacement During the 2026 Kumamoto Earthquake From Near-Fault Video
Core Problem: The rise time of surface fault displacement is hard to quantify because available video is noisy, secondary-copy footage with camera-motion artifacts.
Key Innovation: Calibrates optical-flow and NCC tracking against field-measured offset to estimate the timing and average rate of coseismic surface displacement.
4. Influence of river incision on landslides triggered in Nepal by the 2015 Gorkha earthquake: results from a pixel-based susceptibility model using inlabru
Core Problem: regional controls on Gorkha-earthquake landslides, especially river-incision effects on coupled hillslopes, remain uncertain
Key Innovation: Bayesian pixel-based susceptibility model that projects channel steepness onto hillslopes and jointly models landslide location and size
5. Loess landslides identification and analysis of deformation mechanism in Heifangtai, China
Core Problem: Irrigation-driven loess landslides need early identification and mechanism attribution across active clusters.
Key Innovation: Combines geomorphic interpretation with long Sentinel-1 time series and UAV evidence to map active creep and diagnose groundwater, freeze-thaw, and rainfall controls.
6. Lithological, geomorphological and tectonic control on landslide processes in Southern Italy: insights from a regional inventory and metrics of landscape evolution
Core Problem: Regional controls on landslide clustering in Southern Italy were not fully explained by lithology and topography alone.
Key Innovation: Integrates a regional landslide inventory with geothematic maps and landscape-evolution metrics to show a strong tectonic control on landslide distribution.
7. Decoupling cumulative and differential land subsidence: a multi-scenario machine learning framework using InSAR and hydro-climatic data
Core Problem: Cumulative and differential subsidence behave differently, and their predictability and transferability across basins were unclear.
Key Innovation: Builds a multi-scenario machine-learning framework that decouples cumulative and differential subsidence targets and extends them to multi-horizon forecasting with uncertainty.
8. Coupled hydro-sediment-morphodynamic model for landslide dam breaching
Core Problem: Overtopping failure of landslide dams involves coupled flow, erosion, morphodynamics, and side collapse that are difficult to simulate together.
Key Innovation: Develops a detailed coupled hydro-sediment-morphodynamic model with gravity-induced side collapse and validates it on lab data and the Baige case.
9. Keep Your Friends Close, and the Right Neighbours Closer: Disaster-Conditioned Kernel-Regularized Graph Attention for Building Damage Classification
Core Problem: Damage classification under event shift needs spatial context without oversmoothing or propagating structured errors.
Key Innovation: Disaster-conditioned graph attention with multi-scale kernel priors and residual spatial-autocorrelation regularization.
10. Matching Urban Flood Sensor Placement to Monitoring Objectives Using Bayesian Optimal Experimental Design
Core Problem: Choose flood sensors based on the prediction target they are meant to inform.
Key Innovation: Compares parameter- and goal-oriented Bayesian OED across thousands of candidate flood-sensor sites.
11. MeltwaterBench: Deep learning for spatiotemporal downscaling of surface meltwater
Core Problem: Current meltwater maps trade temporal for spatial resolution on the Greenland ice sheet.
Key Innovation: Fuses SAR, passive microwave, DEM, and climate-model data to produce daily 100 m meltwater maps and benchmark them.
12. Dynamic response and seismic fragility analysis of offshore wind turbines under mainshock-aftershock
Core Problem: mainshock-aftershock sequences can degrade offshore wind-turbine response and fragility estimates
Key Innovation: explicit seismic fragility framing for offshore turbines under sequential earthquake loading
13. Two-stage classification of avalanche problems and danger levels across elevation bands in Colorado
Core Problem: many avalanche regions lack the dense snowpack instrumentation assumed by feature-rich forecasting systems
Key Innovation: two-stage elevation-band ML framework that infers avalanche problems and danger levels from widely available meteorology
14. Coseismic landslide runout distance prediction based on hybrid stacking ensemble model
Core Problem: Single models predict coseismic landslide runout poorly under complex terrain and varied failure mechanisms.
Key Innovation: Uses a Bayesian-optimized stacking ensemble with SHAP interpretation to improve and explain runout-distance prediction.
15. Modeling the cooling influence of the Agulhas Current System on the heritage day floods at Theewaterskloof Dam, Cape Town, South Africa
Core Problem: The role of the Agulhas Current System in shaping a severe South African flood event was uncertain.
Key Innovation: Couples HEC-RAS flood modeling, satellite validation, and MPAS sensitivity experiments to quantify the cooling effect of the current system on flood severity.
16. Leveraging explainable artificial intelligence and ensemble learning as surrogate models for slope stability evaluation
Core Problem: Slope safety-factor evaluation for monitoring and digital twins remains computationally heavy and often opaque.
Key Innovation: Combines ensemble surrogate models with explainable AI on a 497-case database to rank the field variables most important for slope stability.
17. Numerical Assessment of Bio-grouting Strategies for Tailings Dams Considering Treatment-induced Heterogeneity
Core Problem: The stability gains from bio-grouting layouts and treatment heterogeneity in tailings dams were poorly understood.
Key Innovation: Uses hydro-mechanical modeling and stochastic porosity fields to compare continuous and intermittent bio-grouting strategies under seepage.
18. Monitoring data-driven framework for assessing the post-earthquake traffic capacity of urban bridges
Core Problem: Regional bridge inventories lack enough structural detail for timely and reliable post-earthquake traffic-capacity assessment.
Key Innovation: Combines cost-aware heterogeneous sensor placement with interpretable ensemble learning to infer bridge traffic capacity from seismic-response monitoring data under incomplete information.
19. TopoSurfel: Closing the Loop between Gaussian Surfels and Meshes for Surface Reconstruction
Core Problem: Accurate surface extraction from unstructured Gaussian splats is difficult in ambiguous or textureless regions.
Key Innovation: Differentiable loop between Gaussian surfels and proxy meshes with mesh-guided surfel evolution and robust re-initialization.
20. M2Depth: Unifying Monocular Depth Foundation Priors with Multi-View Stereo
Core Problem: MVS generalizes poorly in occluded or sparse-overlap scenes, and monocular depth priors are often fused too crudely.
Key Innovation: Bidirectionally couples monocular depth foundation priors with cascade MVS so each refines the other and sharpens depth structure.
21. Triangulation-Free Bundle Adjustment with Graduated Non-Convexity for Camera Pose Refinement from Coarse Priors
Core Problem: Prior-seeded bundle adjustment can degrade already good camera priors because triangulated structure bakes in prior error before optimization.
Key Innovation: Eliminates committed triangulation by optimizing ray-depth variables with symmetric cross-projection residuals and graduated non-convexity.
22. CoST: Semantic-Aware Urban Understanding via Spatial-Temporal Alignment
Core Problem: Satellite representation learning often lacks semantic interpretability and generalizes poorly across regions.
Key Innovation: Aligns spatial context with multi-year urban semantic change to learn more transferable geospatial representations.
23. Stream3Dv2: Geometric-Semantic Fusion Enhanced Streaming Zero-Shot 3D Scene Understanding
Core Problem: Streaming zero-shot 3D scene understanding is too slow and brittle to noisy masks for real-world deployment.
Key Innovation: Training-free nested local-to-historical fusion with manifold-distance point-cloud refinement for robust streaming segmentation and detection.
24. Revisiting the hydromechanical formulation of a micromechanics-based phase-field model for poro-elastoplastic media
Core Problem: Common phase-field formulations for poro-elastoplastic fracture create discontinuous strength surfaces and incorrect fracture driving forces.
Key Innovation: Revised hydromechanical phase-field formulation combining associative Drucker-Prager plasticity with the missing coupling term for continuous strength behavior.
25. When Generated Images Look Right and Retrieve Wrong: Coverage-Guided Cross-Scale Re-Indexing for Knowledge-Faithful Generative Perception
Core Problem: Generated images can look plausible yet lose retrievable multiscale semantic content.
Key Innovation: Introduces closed-loop cross-scale re-indexing with coverage verification and SOTA pansharpening gains.
26. Interannual Glacier Variability and Accelerated Albedo Decline in Northeastern Tibetan Plateau: Multidecadal Remote Sensing Insights (1986-2024)
Core Problem: long-term glacier retreat and albedo decline in the northeastern Tibetan Plateau remain poorly quantified
Key Innovation: machine-learning fusion of Landsat boundaries and MODIS albedo to track annual glacier and debris-cover change
27. Global Spatial-Spectral and Frequency-Domain Mamba for Hyperspectral Change Detection
Core Problem: hyperspectral change detection misses global context and confuses fine-grained relevant versus irrelevant changes
Key Innovation: global spatial-spectral attention with frequency-domain refinement and Mamba-based local enhancement
28. Elevation-Dependent Glacier Velocity and X-Band SAR Backscatter Indicate Seasonal and Topographic Controls on Tropical Glacier Dynamics in the Central and Southern Cordillera Blanca, Peru
Core Problem: seasonal and topographic controls on tropical glacier motion and radar backscatter are poorly resolved regionally
Key Innovation: multi-year TerraSAR-X velocity and X-band backscatter analysis across 195 glaciers with elevation-stratified interpretation
29. Analysis and Correction of Atmospheric Residual Errors in Time-Series InSAR for the Guangzhou-Zhanjiang High-Speed Railway
Core Problem: GACOS-corrected time-series InSAR still retains large atmospheric residuals during extreme water-vapor fluctuations
Key Innovation: GNSS-ERA5 precipitable-water-vapor fusion model that removes residual delay and sharpens subsidence detection
30. Robust Optical-to-SAR Image Registration via Dense Tukey-Weighted Gradient Histogram and Structural Saliency Weight
Core Problem: cross-modal registration fails in textureless or speckled regions because uniform weighting amplifies mismatches
Key Innovation: training-free dense gradient histogram descriptor with structural saliency weighting for robust optical-SAR alignment
31. Enhancing InSAR Tropospheric Delay Correction by Combining Phase-Elevation Correlation and Common-Scene Stacking: Validation and Application to Southern California
Core Problem: single-component tropospheric corrections leave major InSAR deformation errors in mixed topographic settings
Key Innovation: adaptive PEC-plus-common-scene-stacking correction that markedly improves GNSS-validated SBAS deformation time series
32. A first 90 m resolution active layer moisture dataset across the Qinghai-Tibet Plateau permafrost region
Core Problem: plateau-wide active-layer moisture conditions have lacked high-resolution, uncertainty-aware mapping
Key Innovation: 90 m machine-learning fusion dataset with applicability mask and per-pixel uncertainty across Tibetan permafrost
33. Pore-pressure evolution in low-permeability weak layers: a comparison between small-deformation and finite-deformation poro-viscoelastic theories
Core Problem: small-deformation poro-viscoelastic theory can misestimate pore-pressure evolution in weak low-permeability faulted media
Key Innovation: finite-deformation poro-viscoelastic algorithm showing sizable pore-pressure differences under drained and sealed loading
34. Climate Teleconnection Indices and Their Influence on Wildfire Activity in Serbia
Core Problem: Many teleconnection indices are redundant, making it hard to identify the large-scale climate signals most associated with wildfire activity.
Key Innovation: Combines PCA, partial correlation, and LASSO to isolate the most informative teleconnection predictors for burned area in Serbia.
35. Cross-Scale Performance Evaluation of GPM IMERG V07 Precipitation Products in a Typical Mountainous Monsoon Region
Core Problem: The newest IMERG version may propagate scale-dependent biases over complex mountainous wet and dry surfaces.
Key Innovation: Performs cross-scale V07 versus V06 evaluation stratified by elevation, intensity, and season to expose systematic bias amplification.
36. Retrieval of Warm-Season Radar Composite Reflectivity in Sichuan by Integrating FY-4A Multi-Channel Satellite Data and DEM Topographic Information
Core Problem: Mountainous radar blockage and sparse coverage create large gaps in precipitation monitoring.
Key Innovation: Fuses FY-4A multispectral data with DEM constraints in an EMA-U-Net to retrieve composite reflectivity over complex terrain.
37. Teleconnection controls on precipitation variability and extremes across Iran’s climate zones
Core Problem: Nationwide precipitation hazard behavior and its nonlinear teleconnection controls were poorly quantified across Iran's climate zones.
Key Innovation: Uses a geographical detector framework on multi-decadal station data to isolate seasonal and interaction effects of major teleconnections on precipitation extremes.
38. Unraveling the spatial paradox of multi-hazard mitigation: divergent impacts of urban form and function on heat and pluvial flood hazards
Core Problem: Urban form and function can reduce one hazard while worsening another, but the spatial tradeoffs were poorly diagnosed.
Key Innovation: Uses a PCA-MGWR framework to separate synergistic and conflicting urban drivers of heat and pluvial flood hazards.
39. A novel criterion for estimating rockburst proneness based on energy release rate
Core Problem: Rockburst proneness needs a more physically grounded criterion than standard empirical indicators.
Key Innovation: The title indicates an energy-release-rate-based criterion for estimating rockburst proneness.
40. Shared Physics Responses Recover Hidden Rankings in Neural Operator Libraries
Core Problem: Selecting the best neural-operator prediction at deployment is hard without ground truth.
Key Innovation: Shared anchor-based physical-response diagnostic that recovers model rankings across operator libraries.
41. Sparse Light Field Sampling Improves Casual 3D and 4D Reconstruction
Core Problem: Novel-view pipelines underuse synchronized multi-view observations when exposures or sensors are limited.
Key Innovation: Demonstrates that sparse light-field angular sampling materially boosts single-shot and dynamic 3D or 4D reconstruction.
42. Fuzzy-MoE: Interpretable Regime-Conditioned Expert Routing for Non-Stationary Multivariate Time Series Forecasting
Core Problem: Standard deep forecasters blur heterogeneous temporal regimes into one mapping and give poor interpretability.
Key Innovation: Uses a fuzzy-rule Mixture-of-Experts router to infer latent regimes and route variables to specialized forecasting experts.
43. Semantically Compatible Knowledge Distillation for Cross-Domain Object Detection with Vision Foundation Models
Core Problem: VFM teachers and student detectors are semantically misaligned across scale and domain, weakening adaptation and pseudo-labeling.
Key Innovation: Adds a semantic-localization adapter to produce dense teacher features that are spatially compatible with the student detector.
44. SuppreSensing: Expert-Guided Feature Recalibration and Discrepancy Augmentation for Multimodal Object Detection
Core Problem: Multimodal remote-sensing detection suffers from semantic heterogeneity, symmetry traps, and noisy modality-specific features.
Key Innovation: Uses expert-driven recalibration, discrepancy augmentation, and iterative feature purification for selective multimodal collaboration.
45. A Critical Audit of Spatiotemporal Forecasting Benchmark Datasets and Baselines
Core Problem: Common spatiotemporal forecasting benchmarks may overstate progress and understate strong simple baselines because of structural bias.
Key Innovation: Uses classical time-series analysis to expose dataset bias and shows how the audit can guide stronger hybrid forecasting models.
46. TRACE-C: Rank-Calibrated Relational Anomaly Detection for Multi-Stream Operational Telemetry
Core Problem: Important events can be jointly anomalous across telemetry streams even when each individual stream looks normal.
Key Innovation: Strictly prior rank-calibrated fusion of local, dependence, and temporal anomaly channels with transparent ablation reporting.
47. Uncertainty propagation in auto-regressive random neural network models
Core Problem: Neural dynamical systems lack tractable methods to propagate joint input and parameter uncertainty over time.
Key Innovation: Activation-pattern-aware local linearization with recursive state-parameter covariance propagation for autoregressive random networks.
48. The Exceedance Design Effect: Effective Sample Size for Thresholds under Clustering
Core Problem: Estimate effective sample size for quantile thresholds under clustered dependence.
Key Innovation: Derives a threshold-specific design effect rather than reusing mean-based corrections.
49. Truthful Calibration Measures for Sequential Prediction
Core Problem: Define calibration measures that remain truthful for sequential probabilistic prediction.
Key Innovation: Shows exact truthfulness is impossible, then gives approximately truthful sound and complete reductions.
50. Structure is information: structural identifiability mappings for machine learning with partially observed dynamical systems
Core Problem: Machine learning on mechanistic dynamical systems breaks when models are structurally unidentifiable.
Key Innovation: Uses structural identifiability mappings to constrain learning and improve generalization.
51. Doctor Rashomon and the UNIVERSE of Madness: Variable Importance with Unobserved Confounding and the Rashomon Effect
Core Problem: Variable-importance estimates shift across equally good models and omitted variables.
Key Innovation: Bounds true variable importance using Rashomon sets under unobserved confounding.
52. Spatially Aware Dictionary-Free Koopman Eigenfunction Identification for Modeling and Control
Core Problem: Identify Koopman eigenfunctions without fixed dictionaries or neural parameterizations.
Key Innovation: Combines dictionary-free least-squares discovery with spatial KPDE regularization.
53. Modality-Aware Fusion for Optical and SAR Images via Complementary Structure-Texture Decomposition and Enhancement
Core Problem: generic fusion methods mishandle modality-specific structure and texture in optical-SAR imagery
Key Innovation: modality-aware structure-texture decomposition with hierarchical cross-fusion and detail enhancement
54. Collaborative Optimization of Receptive Field and Texture Preservation for Remote Sensing Small Object Detection
Core Problem: small-object detectors lose texture detail when expanding receptive fields in cluttered remote-sensing scenes
Key Innovation: collaborative receptive-field and texture-preservation network with global-local extraction and multiscale attention fusion
55. GNSRNet: Geometric Guided Noise Reduction Super-Resolution Network for Remote Sensing Tiny Object Detection
Core Problem: tiny objects are buried in remote-sensing backgrounds and detection is sensitive to localization errors
Key Innovation: noise-reduced super-resolution pyramid plus geometry-aware regression for tiny-object detection
56. Snow melting phases detection using apparent thermal inertia, radar data and numerical modelling
Core Problem: Operational snowmelt timing is hard to estimate consistently across sensors and models in alpine terrain.
Key Innovation: Cross-compares apparent thermal inertia, Sentinel-1 backscatter, and SNOWPACK simulations to extract snow warming, ripening, and output phases with quantified timing uncertainty.
57. ACBDT: SAR-Optical Cross-Modal Distillation for Sentinel-1/2 Building-Footprint Mapping in Heterogeneous Yangtze River Delta Cities
Core Problem: Medium-resolution optical and SAR imagery struggle to separate buildings from mixed backgrounds and speckle.
Key Innovation: Introduces a SAR-optical cross-modal distillation bridge with fused teacher and refinement decoder for Sentinel-1 and Sentinel-2 building mapping.
58. An Enhanced Nonlinear Grid Transformation Method for Weather Radar Echo Extrapolation
Core Problem: Radar echo extrapolation often fails to jointly predict motion, deformation, and intensity.
Key Innovation: Extends nonlinear grid transformation into a reflectivity-aware 3D formulation for echo position, shape, and intensity extrapolation.
59. Overburden Fracture Evolution and Surface Subsidence in Peak-Cluster Karst Longwall Mining A Physics-Constrained Multimodal Prediction Framework
Core Problem: Karst longwall mining creates complex overburden fracturing and surface subsidence that standard empirical methods handle poorly.
Key Innovation: Builds a physics-constrained multimodal predictor that fuses simulation images, stress-displacement sequences, geology, and field data.
60. Hydroxypropyl methylcellulose-enhanced interfacial bonding in eco-engineered slopes: Applications to shallow interface stabilization
Core Problem: Shallow eco-engineered slope interfaces need stronger bonding and stabilization.
Key Innovation: The title suggests hydroxypropyl methylcellulose is used to enhance interfacial bonding in eco-engineered slopes.
61. A Subtle Atmospheric Shift Is Redefining Winter Hydrology on the Tibetan Plateau
Core Problem: Winter warming may be weakening the Tibetan Plateau snow-storage buffering function.
Key Innovation: Compound temperature-precipitation regime framework linking circulation shifts to snow, soil moisture, and runoff changes.
62. Wrong-Physics Backdoors in Neural PDE Operators
Core Problem: Neural PDE operators can be poisoned to output physically plausible but parameter-wrong solutions.
Key Innovation: Wrong-physics backdoor attack via cross-parameter relinking across multiple neural-operator families.
63. Frozen CLIP Priors for Robust Self-Supervised Poisson Inverse Problems
Core Problem: Poisson inverse problems need self-supervised priors that remain robust under acquisition shifts.
Key Innovation: ADMM-style plug-and-play solver using frozen CLIP multi-scale features with re-corruption and equivariant regularization.
64. Lift, Associate, and Fuse: A Decision-Centric Framework for 2D-to-3D Foundation Model Transfer
Core Problem: Current taxonomies hide where 2D foundation-model evidence becomes irrecoverable in 3D transfer systems.
Key Innovation: Decision-centric five-operator framework and structured audit protocol spanning 161 systems.
65. Generating Multi-view Adversarial Examples for Visual Geometry Grounded Transformer
Core Problem: Feed-forward multi-view 3D foundation models may be vulnerable to consistent cross-view adversarial perturbations.
Key Innovation: Multi-view perturbation generator with cross-view adversarial alignment that attacks VGGT without per-scene optimization.
66. Hidden Axis of Uncertainty: Latent-Posterior Alignment in Graph Neural Networks with Bayesian Output Layers
Core Problem: Predictive uncertainty behavior in GNNs with Bayesian output layers is poorly understood and not well controlled.
Key Innovation: Identifies latent-posterior alignment as the driver of uncertainty reduction and proposes alignment-guided learning to improve calibration.
67. A2DINOv3: Rethinking Multi-Modal Object Detection via Socialized Collaboration
Core Problem: Dense cross-modal fusion can corrupt pretrained priors and weaken multimodal object detection in challenging conditions.
Key Innovation: Treats RGB and infrared branches as selective collaborating experts with constrained information exchange and gradual cross-modal activation.
68. ConceptTS: LLM-Guided Concept Bottlenecks for Interpretable Multivariate Time-Series Forecasting
Core Problem: Black-box multivariate forecasters give little insight into why they predict a future trajectory.
Key Innovation: LLM-generated human-readable concepts organized into historical, interval, and horizon bottlenecks for forecasting.
69. Robust Discovery of Coarse-Grained Continuum Equations from Microscopic Dynamics
Core Problem: PDE discovery from data becomes unstable with limited samples, noise, and oversized libraries.
Key Innovation: Systematic analysis of PDE-SINDy selection probabilities showing how data volume and thresholding recover robust coarse-grained equations.
70. StateSight: Benchmarking Latent Spatial-State Reconstruction in Vision-Language Models
Core Problem: Existing multimodal benchmarks confound spatial reconstruction with OCR, world knowledge, and language priors.
Key Innovation: Procedural benchmark isolating latent spatial-state recovery plus deterministic intermediate-state supervision.
71. The Coastline as a Structural Constraint: Harnessing Scene Geometry for Autonomous Surface Vessel Localization
Core Problem: Localize autonomous surface vessels in GPS-denied coastal settings.
Key Innovation: Registers shoreline observations from LiDAR or monocular imagery against satellite-derived coastline maps.
72. Vision Foundation Model Driven Foreground-Aware Pseudo-LiDAR Generation for Monocular 3D Object Detection
Core Problem: Generate better pseudo-LiDAR for monocular 3D object detection.
Key Innovation: Fuses depth and segmentation foundation-model priors into foreground-aware pseudo-LiDAR generation.
73. Perseus: Interactive Time Series Segmentation with Sparse Supervision via Stateful Memory
Core Problem: Sparse prompts in long multivariate time series do not persist across unprompted intervals.
Key Innovation: Stores expert corrections in persistent stateful memory that is queried during inference.
74. On the Fr\'echet interaction density of certain wave equations
Core Problem: Generalize adjoint-based Fréchet sensitivity analysis to real and complex PDEs.
Key Innovation: Defines Fréchet interaction density as a unified kernel-generating object across wave equations.
75. Actively Learning Joint Contours of Multiple Computer Experiments
Core Problem: finding inputs that satisfy multiple simulator response targets simultaneously is sample-inefficient
Key Innovation: dual-acquisition joint-contour strategy with built-in stopping rule
76. Key Technologies for GNSS Water Vapor Retrieval and Their Applications in Marine Environmental Monitoring
Core Problem: marine water-vapor monitoring methods and error sources are fragmented across platforms and workflows
Key Innovation: integrated review separating column retrieval, tomography, multisource fusion, and operational constraints
77. A Systematic Quantitative Analysis of Machine and Deep Learning Architectures for Hyperspectral Mineral Mapping
Core Problem: performance tradeoffs across ML and deep-learning mineral-mapping architectures lack systematic comparison
Key Innovation: meta-analytic synthesis across four generations of hyperspectral mineral-mapping models
78. MRC-Former: A Multiscale Rotation-Invariant and Center-Preserving Transformer for Hyperspectral Image Classification
Core Problem: HSI classifiers struggle with rotation variance and spatial-spectral ambiguity across scales
Key Innovation: rotation-invariant transformer with center-preserving pooling for multiscale spectral-spatial encoding
79. Remote Detection of Soil Carbon, Nitrogen, and Soil Minerals Through Decomposition of Hyperspectral Reflectance and Multispectral Satellite Imagery
Core Problem: mixed soil spectra and heterogeneous covers hinder remote estimation of shallow soil carbon, nitrogen, and minerals
Key Innovation: VPCA plus stepwise regression workflow spanning field hyperspectral, lab spectra, and Sentinel-2 imagery
80. GeoCaps: Geometric and Lightweight Capsule Network for Hyperspectral Image Classification
Core Problem: scalar-neuron HSI models miss spatial hierarchy and pose information while capsule nets remain heavy
Key Innovation: lightweight capsule framework with ghost-graph interaction and noniterative geometric consensus routing
81. BiFGA: A Bidirectional Image-Text-Guided Fine-Grained Alignment Method for Remote Sensing Image-Text Retrieval
Core Problem: fine-grained cross-modal alignment is hard in remote-sensing scenes with dense objects and complex layouts
Key Innovation: bidirectionally guided image-text alignment using semantic anchors for coarse-to-fine retrieval
82. Local-to-Pixel Hyperspectral Image Restoration Network With Wave Mamba
Core Problem: HSI restoration under complex degradations fails to preserve fine spatial-spectral structure
Key Innovation: local-to-pixel restoration with distance-aware Wave Mamba local context modeling
83. A Real-Time Back-Projection Algorithm With 2-D Spectrum Compression for Ultra-High-Speed Maneuvering Platform SAR
Core Problem: back-projection imaging for ultra-high-speed maneuvering SAR is too computationally expensive for real time
Key Innovation: 2-D spectrum-compressed back-projection with azimuth and range decimation for real-time SAR imaging
84. SenSNN: Energy-Efficient Tiny-Object Pseudo-Mask Segmentation via a Dual-Stream Membrane Shortcut
Core Problem: continuous onboard tiny-object screening on nanosatellites is too power-hungry for standard CNNs
Key Innovation: spiking dual-stream membrane shortcut architecture with temporal consensus for low-energy pseudo-mask segmentation
85. Thirteen Winters (2012-2025) of Penetration-Aware Arctic Sea-Ice Emissivity and Emission Temperature from AMSR2
Core Problem: winter microwave emission properties of snow-covered Arctic sea ice are not consistently characterized for long records
Key Innovation: penetration-aware multiwinter emissivity and emission-temperature dataset from AMSR2
86. FFR-YOLO: A Frequency-Guided Fusion Reconstruction Network for Small-Object Detection in Remote Sensing Images
Core Problem: Small objects in remote-sensing imagery lose detail and fuse poorly across scales.
Key Innovation: Adds wavelet and frequency-guided backbone, neck, and head modules to improve YOLOv8 small-object detection.
87. High-Resolution Mapping of Forest Vegetation Types Using Multiplatform Imagery and Advanced Classification Techniques
Core Problem: Detailed forest type mapping is difficult in topographically complex and heterogeneous island terrain.
Key Innovation: Tests multiplatform satellite, UAV, and LiDAR inputs across ten classifiers for high-resolution forest vegetation mapping.
88. HNGT-Net: Hard-Negative Guided Topology Transfer for Lightweight Hyperspectral Small-Target Detection
Core Problem: Unknown small targets in hyperspectral imagery are difficult to detect without target priors.
Key Innovation: Transfers background topology from a teacher to a lightweight student while synthesizing hard negatives for robust anomaly detection.
89. Aerosol Optical Depth Retrieval from MODIS Using a Physically Informed Machine Learning Framework
Core Problem: Land AOD retrieval must separate weak aerosol signals from strong and variable surface reflectance.
Key Innovation: Builds a physically informed random-forest AOD retrieval using aerosol-background priors and time-series clear-sky reflectance features.
90. HSAR-DETR: Hierarchical Spatial-Frequency Attention Network for UAV Small Object Detection
Core Problem: UAV small-object detection suffers from detail loss, weak cross-scale fusion, and unstable localization.
Key Innovation: Combines hierarchical enhancement, spatial-frequency refinement, coordinate-guided adaptive convolution, and a high-resolution detection branch.
91. Effects of Cementitious Amendments on the Mechanical Strength and Hydraulic Conductivity of Mine Tailings for Reclamation Cover Systems
Core Problem: Weathered mine tailings need enough strength and low hydraulic conductivity to function as cover-system barriers.
Key Innovation: Compares limestone and slag-based binders against tailings geochemistry to optimize solidification and barrier performance.
92. Delineation of river flats with local hypsometry
Core Problem: Rainfall and snowmelt contributions to Himalayan river discharge were poorly constrained because gauge networks are sparse.
Key Innovation: Uses remotely sensed climate fields and a snowmelt model to resolve along-strike Himalayan precipitation and runoff contributions.