TerraMosaic Daily Digest: September 27, 2026

1. Stage-based deformation preceding rainfall-induced landslide failure in Songxi County, China

Core Problem: Rainfall totals alone do not describe the deformation state preceding failure.

Key Innovation: Two Songxi landslides are investigated with field, laboratory and monitoring evidence; one slope's 2022 deformation and 2024 failure link antecedent effective rainfall to sustained movement and final acceleration. Effective rainfall is a wetness proxy, not a pore-pressure measurement.

2. Reliability analysis method for soil slopes permanent displacement under mainshock-aftershock sequences

Core Problem: Mainshock-only analysis can omit displacement accumulated under aftershocks.

Key Innovation: A probability-density-evolution framework evaluates permanent displacement under seismic sequences; cumulative absolute velocity emerges as a controlling measure among 21 tested inputs, rather than relying on peak acceleration alone.

3. Rainfall Induced Degradation of Rockbolt Anchorage: Experimental Tests and Numerical Simulation

Core Problem: Sustained infiltration changes the bond strength of resin-anchored bolts.

Key Innovation: Interface shear and pull-out experiments inform moisture-dependent cable-element updates. Strength and load reductions are tied to tested moisture conditions and a specified rainfall simulation, not universal degradation rates.

4. State-bounded directional accumulation in post-liquefaction sand: A hypoplastic formulation for asymmetric cyclic deformation

Core Problem: Sand models can accumulate unbounded directional strain after liquefaction.

Key Innovation: A state-bounded hypoplastic mechanism uses one additional parameter to reproduce extension-biased but finite accumulation, with unchanged-parameter and cross-path tests plus long-cycle ablations.

5. Hidden rock mass state theory: A physics-informed state estimation framework for next-generation Digital Twins in open-pit rock engineering

Core Problem: Measured deformation does not uniquely reveal the physical state governing rock response.

Key Innovation: The HRMS framework separates hidden physical state, external forcing and observations across mining environments. Omission, perturbation and withheld-observation tests support state inference, but do not establish a universal failure-warning threshold.

6. Explaining Ground-Motion Residuals at Two Strong-Motion Stations in Southeastern New York: Sediment Resonance and Topographic Amplification

Core Problem: Proxy site descriptors do not explain large station-specific ground-motion residuals.

Key Innovation: Surface-wave and noise measurements support sediment resonance at Caumsett and directional topographic amplification at the Palisades following the Tewksbury earthquake. This is an arXiv preprint.

7. CA-DSUNet: A Novel Physics-Prior-Guided Dual-Stream Deep Coupling Network for Flood Inundated Area Forecasting in Emergent Support

Core Problem: Unconstrained sequence models can predict implausible overflow.

Key Innovation: Cellular-automaton expansion envelopes bound a dual-stream network combining flood history and terrain. Poyang and Dongting tests support the approach; overall IoU must be distinguished from skill at changing flood margins.

8. Efficient On-Board Processing of Oblique AAV Video for Rapid Flood Extent Mapping

Core Problem: Embedded hardware limits rapid segmentation of oblique aerial video.

Key Innovation: Temporal token reuse bypasses repeated feature computation in stable image patches, giving 1.58–1.69-fold speedups on tested videos with less than one percentage point mIoU loss; still-image benefits are not implied.

9. From Coseismic Disturbance to Short-Term Dynamic Recovery: Nighttime Light Perspective on the Disaster Response to the Myanmar Mw 7.7 Earthquake

Core Problem: Immediate damage and month-scale recovery have distinct spatial trajectories.

Key Innovation: Black Marble nighttime lights identify heterogeneous post-Myanmar-earthquake recovery patterns; light restoration is an activity proxy, not direct verification of repaired buildings or full social recovery.

10. Spatiotemporal evolution of land subsidence in coastal reclamation areas: coupling anthropogenic land-use intensity and delayed soft-soil compression

Core Problem: Changing land use and delayed soft-soil compression interact in reclaimed ground.

Key Innovation: Borehole-constrained modeling and settlement observations support conditional Ningbo screening scenarios. Equivalent parameter corrections do not simulate constitutive creep, and spatial consistency with InSAR is not independent full-domain validation.

11. From waste to a watershed-scale solution: Transforming oyster shell and expansive soil into a hydrophobic barrier for intercepting rainfall recharge and stabilizing slopes

Core Problem: Cracks provide infiltration routes that conventional fill materials may not reliably seal.

Key Innovation: Oyster-shell, expansive-soil and octadecylamine mixtures reduce cracking and moisture fluctuations in laboratory and 134-day model tests. The evidence supports a hydrophobic material, not demonstrated watershed-scale stabilization or a universal superhydrophobic threshold.

12. Debris Flow Susceptibility Mapping Using Explainable Deep Learning in Eastern Hindukush Pakistan

Core Problem: Mountain hazard inventories must be related to terrain, rainfall and human disturbance.

Key Innovation: A 30 m CNN–SHAP model uses 13 factors in eastern Hindu Kush and reports AUC 0.94 with density-based checking. The available abstract does not establish spatially independent or cross-region validation; SHAP describes model associations rather than causal controls.

13. Instability risk assessment of submarine strata based on data-driven modelling and interpretable machine learning algorithms

Core Problem: Small borehole datasets provide incomplete evidence of engineering instability.

Key Innovation: Clustering-derived grades and interpretable classifiers use 120 soil samples; the best reported accuracy is 0.639. Agreement with derived labels is not validation against an independent failure inventory.

14. Time-dependent response of a loaded pile under excavation- and dewatering-induced non-Darcy reverse consolidation

Core Problem: Excavation changes consolidation history and pile contact conditions.

Key Innovation: A sequential one-way soil-to-pile model combines non-Darcy flow, fractional soil response and unilateral contact, with analytical and numerical checks. It excludes pile-to-soil feedback.

15. Research on Mining Area Surface Subsidence Based on Time-Series InSAR Images and Deep Learning Models

Core Problem: Regional subsidence retrieval and prediction depend on stable interferometric references.

Key Innovation: PS-stable-point constraints improve SBAS processing before a spatiotemporal model predicts mining deformation. Pixel-scale test error does not establish mine-to-mine generalization.

16. The Role of Multi-Sensor Satellite Remote Sensing in Characterizing Volcanic Plumes Relevant to Climate Forcing

Core Problem: Different sensors observe different plume constituents and atmospheric effects.

Key Innovation: A review combines ultraviolet through thermal-infrared perspectives on recent eruptions, separating plume characterization from downstream climatic interpretation.

17. Radar-Video Fusion for Tiny Rockfall Detection and Tracking in Complex Field Environments

Core Problem: Tiny moving blocks are intermittently visible and difficult to associate across sensors.

Key Innovation: Radar-guided search regions and gated assignment maintain fused identities in five field sequences with 22 trajectories. Reported F1 is 0.731 and identity F1 is 0.702 at the primary 20-pixel criterion; these results characterize a single-site prototype, not broad operational validation.

18. A distributed parallel processing framework for sentinel-1 wide-area time series InSAR: Application in Jining City

Core Problem: Wide-area processing and mosaicking can be costly on modest hardware.

Key Innovation: Burst-level parallel processing, polynomial correction and weighted blending map Jining mining subsidence, with field-observed road and building damage supporting site relevance.

19. Volcanic stratigraphy offshore East Montserrat: Selective preservation and subaqueous remobilisation in marine records

Core Problem: Offshore deposits may not preserve a complete eruption history.

Key Innovation: Two Montserrat cores show selective ash preservation and remobilisation; multi-proxy analysis cautions against assigning every marine layer directly to a monitored onshore event.

20. A multi-dimensional cloud-model-based fuzzy evaluation method for grading nonlinear tunnel large deformation

Core Problem: Uncertain grade boundaries and indicator weights hinder cross-project assessment.

Key Innovation: Asymmetric cloud memberships and coupled indicator weights are evaluated on three tunnel datasets; grading accuracy alone is not calibrated event probability.

21. Propagation of Blast-Induced Ground Vibration and Its Effect on Building Infrastructures: A Review

Core Problem: Vibration prediction and building response are governed by site and loading conditions.

Key Innovation: Reviews wave generation, empirical and numerical models, explosive loading and structural-response limits, distinguishing blast excitation from earthquake motion.

22. A time-dependent solution for excavation-induced vertical deformation of underlying existing tunnels in two-layered saturated ground based on Biot consolidation theory

Core Problem: Underlying tunnels deform as layered saturated ground consolidates around excavations.

Key Innovation: The same-title author preprint couples Biot consolidation with beam response and compares deformation with centrifuge observations. Journal identity is verified; equivalence of precursor and final validation remains unverified.

Method evidence: Author preprint; journal-version equivalence not established.

23. Predicting ground motion intensity measures in Japan using a hybrid artificial neural network integrated with Taguchi optimization

Core Problem: Regional intensity prediction requires relationships between event, path and site information.

Key Innovation: The same-title author preprint couples neural prediction with Taguchi optimization and mixed-effects variability analysis. Journal identity is verified; precursor and final validation are not assumed equivalent.

Method evidence: Author preprint; journal-version equivalence not established.

24. KG-SubNet: A Knowledge-Guided Deep Learning Framework for 3D Mining Subsidence Retrieval from InSAR Interferograms

Core Problem: Title-level focus: Line-of-sight interferometry does not directly expose full three-dimensional displacement.

Key Innovation: The title identifies knowledge-guided neural retrieval of mining subsidence from interferograms. Methods, data and results could not be assessed from a reliable abstract.

25. BMINN: A neural network for bedrock motion inversion from ground motion

Core Problem: Title-level focus: Surface shaking must be related to the motion entering underlying bedrock.

Key Innovation: The title uses a neural network for inverse estimation of bedrock motion from ground-motion records. Methods, data and results could not be assessed from a reliable abstract.

26. GUARD-net: A geostationary satellite-based network for near-real-time active fire detection under class imbalance

Core Problem: Title-level focus: Rare fire pixels create class imbalance in continuous satellite monitoring.

Key Innovation: The title identifies a geostationary fire-detection network that explicitly addresses class imbalance. Methods, data and results could not be assessed from a reliable abstract.

27. Large-scale landslide susceptibility mapping via deep learning: A case study of Pakistan

Core Problem: Title-level focus: Large-area susceptibility modeling must generalize across heterogeneous terrain.

Key Innovation: The title applies deep learning to Pakistan-wide landslide susceptibility; inventory quality and spatial validation cannot be assessed from the available metadata. Methods, data and results could not be assessed from a reliable abstract.

28. An optimization approach for underground cavern layouts utilizing the surrounding rock disturbance contribution of primary and secondary caverns

Core Problem: Title-level focus: Primary and secondary excavations contribute differently to surrounding-rock disturbance.

Key Innovation: The title uses these contributions to optimize underground cavern layouts. Methods, data and results could not be assessed from a reliable abstract.

29. Seismic fragility and risk assessment of existing long-span precast industrial buildings retrofitted with different techniques

Core Problem: Title-level focus: Retrofit choices require comparable fragility and risk estimates.

Key Innovation: The title compares retrofit techniques for long-span precast industrial buildings. Methods, data and results could not be assessed from a reliable abstract.

30. Cyclic liquefaction behavior of weakly cemented and fines-treated sands with gradation effects

Core Problem: Title-level focus: Weak cementation, fines and gradation jointly affect cyclic soil behavior.

Key Innovation: The title evaluates liquefaction response across these material conditions. Methods, data and results could not be assessed from a reliable abstract.

31. Centrifuge modeling of tunnel-depth effects on seismic response of soil-tunnel systems

Core Problem: Title-level focus: Embedment depth changes soil–tunnel interaction under shaking.

Key Innovation: The title uses centrifuge models to isolate tunnel-depth effects on seismic response. Methods, data and results could not be assessed from a reliable abstract.

32. A Hysteresis-Based Framework for Evaluating and Selecting Streamflow Monitoring Approaches Under Unsteady Gradually Varied Flow Conditions

Core Problem: A single stage–discharge curve misses unsteady-flow hysteresis and uncertainty.

Key Innovation: A three-part metric separates flow deviation, peak timing and cumulative volume, with comparisons at three USGS stations to guide monitoring-method selection.

33. Learning coarse-step dynamics and internal mechanical response with graph networks

Core Problem: Coarse trajectory observations hide internal mechanical response.

Key Innovation: A Newmark-inspired graph update and virtual hub support coarse-step dynamics and recover aspects of beam stiffness without force supervision; geomechanical transfer requires validation.

34. AlphaEarth distinguishes cities but compresses urban variation

Core Problem: Globally shared Earth embeddings may not preserve urban detail or temporal comparability.

Key Innovation: An AlphaEarth audit across 1,000 urban areas identifies uneven within-city variation and sensor-related annual changes. Cross-region utility does not establish validity for year-to-year change detection.

35. Learning to Bias: Machine Learning-Enhanced Particle Filters

Core Problem: Poor proposals reduce particle-filter efficiency.

Key Innovation: Learned one-step conditional proposals are corrected by importance weights, preserving the filtering target under stated support and density assumptions while improving nonlinear benchmarks.

36. When 10,000 Windows Are Not 10,000 Tests: Auditing Statistical Confidence in Sliding-Window Time-Series Classification

Core Problem: Overlapping windows inflate apparent test sample size.

Key Innovation: An audit separates recording- and subject-level estimands and dependence-robust intervals, showing that more windows do not provide proportional independent evidence.

37. Mechanism-Aware Ensemble Conditioning for Data-Limited Emulation of Extreme Events

Core Problem: Short trajectories undersample transient instability and distribution tails.

Key Innovation: Coarse-ensemble covariance conditions learned dynamics through FiLM; controlled chaotic and quasi-geostrophic tests assess exceedance statistics rather than only mean error.

38. Adaptive Interaction Graphs for Particle Simulation

Core Problem: Fixed interaction graphs miss spatially concentrated simulation complexity.

Key Innovation: A variance head expands particle neighborhoods using the preceding step's uncertainty, improving WaterDrop and producing more modest Sand gains; long-horizon benefits are problem-dependent.

39. Learning Hierarchical Causal Representations of the Effects of Forcings on Temperature in Climate Models

Core Problem: Black-box climate emulators blur forced response and internal variability.

Key Innovation: Hierarchical representations separate atmospheric dynamics from greenhouse-gas and aerosol forcing and test responses on unseen climate scenarios.

40. Aurora-X: Built for Extreme Time Series Forecasting

Core Problem: Forecasting backbones must accommodate covariates, cross-variable dependence and varied resolutions.

Key Innovation: Aurora-X combines a progressive curriculum, pattern-guided sparse experts and arbitrary-quantile prediction. Benchmark breadth does not yet establish geohazard forecasting superiority.

41. Data-driven analysis of muon flux variations in muography time series at the Sos Enattos Mine

Core Problem: Predefined regions can conceal localized geophysical variability.

Key Innovation: Overlapping sightline blocks and PCA identify coherent and delayed flux changes in two mine acquisitions; flux components are not automatically resolved density-change mechanisms.

42. From Image Morphology to Multiphysics Earth Posteriors

Core Problem: Morphological priors and physical-property likelihoods provide different information.

Key Innovation: Frozen image-derived coordinates support velocity and density inversion, but smooth bases win on some targets and cross-property coupling can worsen blind predictions.

43. Wave-Propagation Geometry Emerges from Scalar Travel-Time Relations

Core Problem: Accurate travel-time fitting need not imply reliable spatial derivatives.

Key Innovation: Controlled neural and classical representations recover conditional propagation geometry; field comparisons and scalar-matched perturbations expose derivative and lateral-resolution limits.

44. Structure-dependent failure modes of neural priors in acoustic full-waveform inversion

Core Problem: Low velocity error can conceal poor prediction of unseen seismic data.

Key Innovation: Matched-budget acoustic inversions compare grids and neural priors; rankings change with structure, and the grid baseline predicts held-out shots better in the reported tests.

45. More Sensors Only One Field: Rethinking Continual Spatio-Temporal Forecasting

Core Problem: Sensor expansion changes observations without necessarily changing field dynamics.

Key Innovation: A shared field-evolution operator separates observation/query interfaces from latent dynamics and conditions propagation on spatial regime, tested on evolving sensor networks.

46. ContraFM-S2O: Flow Matching-Based One-step SAR-to-Optical Image Translation Model with Contrastive Learning

Core Problem: Image translation must balance detail with sampling latency.

Key Innovation: Contrastive flow matching learns average transport velocity for one-step synthesis. Generated optical appearance is not a substitute for an independent optical observation.

47. NEXT: Physics-Informed Neuro-Spectral Exponential Time Differencing Architectures

Core Problem: Spectral neural integration can diverge on stiff dynamics.

Key Innovation: NEXT integrates the linear stiff component through exponentials while learning the remaining dynamics, with forward and inverse PDE tests.

48. Training Graph Foundation Models on The Web Graph

Core Problem: New graphs change feature dimensions, labels and task structure.

Key Innovation: Acacia is trained on a web graph to support several graph tasks without added heads or LLM components; transfer to physical hazard networks is not yet demonstrated.

49. ChronoFuseGS: Multi-Temporal Gaussian Fusion with Per-Splat Persistence and Change Visualization

Core Problem: Independent scene models waste persistent structure and obscure fine changes.

Key Innovation: ChronoFuseGS fuses time-specific Gaussian models with per-primitive persistence, evaluated over eight surveys of a flood-management area across seven months.

50. AUWave: A Data-Driven Model for Reconstructing Significant Wave Heights Using Sparse Observations

Core Problem: Sparse stations cannot directly resolve regional wave-height fields.

Key Innovation: AUWave combines station encoders and a multi-scale attention U-Net, evaluates buoy removal and cross-basin transfer, and distinguishes near-station accuracy from remote reconstruction.

51. FVT: Feature-to-Video Transition Modeling for Semantic Change Detection

Core Problem: Two-date features can lose directional semantic transitions.

Key Innovation: Feature-to-video representations and bidirectional evolution modules model from–to changes on SECOND and Landsat-SCD. Synthesized transitions are representations, not observed intermediate images.

52. Revisiting the Effectiveness of Self-Supervised Learning for Change Detection in Remote Sensing Imagery

Core Problem: Self-supervised training may be credited for useful features already present without learning.

Key Innovation: Within a specified contrastive-pretraining and unsupervised-postprocessing setup, random and single-layer controls challenge attribution of performance gains to pretraining; the finding is not universal to all self-supervised methods.

53. TempLocate: Endpoint-Supervised Temporal Change Localization for Time-Series Change Detection

Core Problem: Intermediate-time labels are expensive to obtain.

Key Innovation: Endpoint supervision calibrates adjacent temporal differences and consistency constraints, evaluated on DynamicEarthNet and SpaceNet7.

54. A Review of Hyperspectral Unmixing Considering Spectral Variability

Core Problem: Spectral variability undermines fixed-endmember assumptions.

Key Innovation: Reviews model-driven, data-driven and hybrid unmixing across single- and multi-temporal settings, including datasets and evaluation comparisons.

55. Few-Shot Hyperspectral Image Classification: A Review of Deep Architectures, Paradigms, and Benchmarks

Core Problem: Reported low-label classification gains are difficult to compare across protocols.

Key Innovation: A review and common benchmark compare architectures and learning paradigms, keeping scarce-label performance distinct from cross-domain validation.

56. Daily snow depth in the Southern Andes (2010-2024): a quality-controlled dataset from Chile and Argentina

Core Problem: Sparse and inconsistent gauges impede snowpack assessment.

Key Innovation: Compiles daily depths from 81 Andean stations for 2010–2024 with physical-consistency checks; quality filtering improves reliability while reducing availability.

57. Extension of the Japan Aerospace Exploration Agency long-term Northern Hemisphere snow cover extent product beyond half a century

Core Problem: Sensor transitions complicate extension of long snow records.

Key Innovation: Extends JAXA Northern Hemisphere snow observations beyond half a century, with ground checks and separate wet-snow performance; this is an ESSD discussion preprint.

58. Effectively assimilate satellite land surface temperature into offline land surface models within ensemble-based assimilation frameworks

Core Problem: Rapidly changing skin temperature complicates assimilation into land models.

Key Innovation: Joint soil-temperature and soil-moisture updates improve some subsurface and snow variables despite marginal surface-temperature gains, separating target fit from state improvement.

59. Investigating terrestrial water storage change in a western Canadian river basin with GRACE/GRACE-FO and fully-integrated groundwater - surface water modelling

Core Problem: Satellite total storage does not isolate groundwater, soil, surface and snow components.

Key Innovation: GRACE/GRACE-FO comparisons support integrated HydroGeoSphere analysis in western Canada; correlation of total storage does not independently validate each component.

60. Feature attention model for soil-geocomposite interface strength

Core Problem: Small geotechnical datasets contain interacting controls on interface shear strength.

Key Innovation: Attention-based prediction and an extracted empirical expression use large-scale direct-shear data; test accuracy and engineering-formula validation are reported separately.

61. A Freeze-thaw damage model for rock based on equivalent pore structure

Core Problem: Porosity, strength and permeability need not deteriorate together.

Key Innovation: An equivalent-pore model captures contrasting damage modes in granite and sandstones but underestimates sandstone permeability, limiting hydraulic interpretation.

62. Thermoplastic stabilisation of floating energy pile settlement: A TSL-enhanced constitutive model and finite-element analysis

Core Problem: Thermal cycles can cause pile settlement that existing models over- or under-accumulate.

Key Innovation: A stabilization-line enhancement links element tests and centrifuge observations to pile response; fifty-cycle behavior is a model projection rather than a fifty-cycle field record.

63. Efficient reliability analysis of laterally loaded offshore monopile in spatially variable clays: A DeepONet-based approach

Core Problem: Title-level focus: Spatially variable clay makes repeated lateral pile reliability calculations costly.

Key Innovation: The title proposes a DeepONet surrogate for laterally loaded offshore monopile reliability; no hazard-specific loading outcome is inferred. Methods, data and results could not be assessed from a reliable abstract.

64. HotSAM: A Hotspot-Spatial-Adaptive Multi-Resolution framework for efficient long-term deformation monitoring in terrestrial radar interferometry

Core Problem: Title-level focus: Long interferometric records can require expensive uniform-resolution processing.

Key Innovation: The title describes hotspot-adaptive multi-resolution deformation monitoring; abstract-level performance evidence is unavailable. Methods, data and results could not be assessed from a reliable abstract.

65. A knowledge-informed cascaded model for rainfall erosivity estimation in China

Core Problem: Title-level focus: Rainfall erosion potential requires estimates beyond rainfall totals alone.

Key Innovation: The title uses a knowledge-informed cascade to estimate erosivity across China. Methods, data and results could not be assessed from a reliable abstract.

66. Stress-dependent permeability of chemo- and bio-grout rock fractures: Implications for flow control in underground storage systems

Core Problem: Title-level focus: Grouting performance can change as fracture stress and aperture evolve.

Key Innovation: The title compares permeability of chemically and biologically grouted fractures under stress, with underground-storage flow-control relevance. Methods, data and results could not be assessed from a reliable abstract.

67. A parametrically optimized multichannel impact-echo system for void detection in tunnel linings

Core Problem: Title-level focus: Hidden voids in linings require interpretable nondestructive observations.

Key Innovation: The title describes a parametrically optimized multichannel impact-echo system for void detection. Methods, data and results could not be assessed from a reliable abstract.

68. Investigation of Radial Pore-Water Pressure Response and Thermo-Hydraulic Coupling Mechanisms in Tunnel Surrounding Rock Influenced by Lining Stiffness: Insights from One-Dimensional Freeze? Thaw Tests under Elastic Confinement

Core Problem: Title-level focus: Lining stiffness constrains thermally induced pore-pressure evolution.

Key Innovation: The title examines radial pressure and thermal–hydraulic coupling through one-dimensional freeze–thaw tests under elastic confinement. Methods, data and results could not be assessed from a reliable abstract.

69. First Application of a Coastal-Dune Model (AeoLiS) to Simulate Fluvial-Aeolian Morphodynamics

Core Problem: Coastal aeolian models may not describe river-corridor sediment behavior.

Key Innovation: AeoLiS reproduces observed Grand Canyon sandbar changes under simplified grain size, vegetation and sediment-supply assumptions.

70. Noise Stability and Rotational Equivariance of Image-Induced Hamiltonian Spectra for Robust Unsupervised Segmentation

Core Problem: Sensor noise and rotation can destabilize spectral image segmentation.

Key Innovation: Derives spectral-gap-dependent perturbation bounds and mask-stability conditions, then tests them on noisy imagery; transfer to remote sensing remains untested.

71. When Does Advection-Aware Graph Nowcasting Help? A Controlled Study of Distributed Solar Ramp Forecasting with a Self-Supervised Cloud-Motion Estimator

Core Problem: Motion-informed sensor graphs may add complexity without improving forecasts.

Key Innovation: Controlled cloud-motion simulations separate motion-estimation quality from graph design; useful advection requires forecast displacement to fit within the sensor network. Real-network validation remains open.

72. Staged Depth Training: A Representation Curriculum for PINNs

Core Problem: Implicitly learned representations can hinder PINN optimization.

Key Innovation: Staged Depth Training freezes shallow learned prefixes before adding depth, improving many equal-budget PDE benchmark configurations without changing deployed architecture.

73. CSCWD: Cross-Scale Channel-wise Knowledge Distillation for Lightweight Tiny Object Detection on Edge Devices

Core Problem: Tiny targets lose spatial evidence in lightweight detectors.

Key Innovation: Cross-scale distillation transfers teacher high-resolution features to a compact detector without increasing its inference architecture, with aerial benchmark and edge-hardware tests.

74. Structure-Guided Masked Autoencoders for Ultra-High Resolution Scientific Image Understanding

Core Problem: Uniform tokens and random masks poorly match very large structured images.

Key Innovation: A quadtree tokenizer and structure-guided masking compress scientific imagery into bounded token sequences, tested in microscopy and CT rather than geohazard imagery.

75. SAGE: Source-Anchored Guidance via Frequency Equalization for Hierarchical RGB-T Alignment and Fusion

Core Problem: Misregistration produces ghosting and uneven cross-modal information.

Key Innovation: Frequency equalization propagates low-frequency geometric guidance into hierarchical alignment and wavelet-subband fusion, evaluated under real and synthetic misalignment.

76. NEMSim: Learning Control-Conditioned Multi-Event Physical Dynamics via Executable Event-Mechanism Priors

Core Problem: Limited trajectories make controlled multi-event systems difficult to emulate.

Key Innovation: Executable event rules constrain neural transition structure on a kinetic Monte Carlo benchmark; the evidence concerns that benchmark, not natural-hazard event chains.

77. Deep Pseudo-Proximal Map: A Self-Supervised Data-Fitting Agent for Iterative Reconstruction

Core Problem: Repeated inner solvers dominate data-fitting updates.

Key Innovation: A noise-trained pseudo-proximal network replaces iterative likelihood updates, with linear-operator equivalence and tests in deblurring, super-resolution and CT.

78. Learning Provable Neural Network Observer for Uncertain Dynamical Systems

Core Problem: Certifying stable learned observers can require expensive optimization.

Key Innovation: Point-guided pretraining is followed by LMI certification, separating rapid fitting from a global stability constraint on nonlinear control benchmarks.

79. Reliability-Regulated Trajectory Optimization for Progressive COLMAP-Free 3D Gaussian Splatting

Core Problem: Early pose errors propagate through progressive 3D reconstruction.

Key Innovation: Bidirectional consistency gates motion initialization and weights sliding-window corrections, improving COLMAP-free Gaussian reconstruction without external geometric priors.

80. EPOC: Endpoint-Preserving Online Correction With Compressed Residual State for Multi-Horizon Time Series Forecasting

Core Problem: Full residual histories consume memory in multi-horizon adaptation.

Key Innovation: Low-order cosine coefficients plus a shared endpoint provide compact online residual correction, with explicit accuracy–state-size comparisons across fixed forecasters.

81. OneWorld: Learning Consistent Physics Across Actions in World Models

Core Problem: Individually plausible action-conditioned futures can imply incompatible masses or friction.

Key Innovation: OneWorld constrains branches through a shared latent mechanism and tests cross-intervention consistency in controlled environments; this is not a validated disaster simulator.

82. DAPEVO: Deep Adaptive Patch Frame-Event Visual Odometry

Core Problem: Sparse or degraded RGB observations weaken camera tracking.

Key Innovation: Patch-level gating fuses independently estimated frame and event correlations and permits event-only updates, evaluated under frame reduction and image degradation.

83. Gradient Surgery for Physics-Informed Neural Networks

Core Problem: Physics and data objectives can conflict in direction or magnitude.

Key Innovation: Physics-aware gradient surgery adapts to observed conflict phases on four PDE benchmarks; the result is an optimization contribution, not a demonstrated hazard solver.

84. TRACKGRAPH: Online Open-Vocabulary 3D Scene Graphs via Image-Space Tracking

Core Problem: Repeated segmentation and vision-language inference burden persistent 3D mapping.

Key Innovation: TRACKGRAPH propagates masks between sparse keyframes before 3D fusion, with quadruped deployment and aerial-view tests; disaster-scene performance remains unestablished.

85. Band-Selection Stability and Semantic Segmentation Performance: A Study on Hyperspectral City

Core Problem: Stable band subsets do not necessarily preserve downstream segmentation.

Key Innovation: Repeated-sample comparisons separate band stability, subset size and segmentation performance, finding no consistent stability–accuracy association.

86. SAGE: A sampling-aware global evaluation benchmark for species distribution modeling

Core Problem: Sampling bias and prevalence distort performance comparisons of occurrence models.

Key Innovation: A global species benchmark stratifies sampling effort and prevalence and shows the importance of thinning and reweighting; transfer to landslide inventories is methodological.

87. DepthEvidence: Unifying Metric Depth Prediction and Geometric Reasoning in Multimodal Language Models

Core Problem: Language answers may lose numerical information from predicted depth.

Key Innovation: Object-anchored geometry tokens connect camera-conditioned metric depth with language reasoning, evaluated on depth and spatial-query benchmarks.

88. Bayesian Tensor Autoencoder with Physics-informed Predictive Prior for Multi-dimensional Time Series Anomaly Detection

Core Problem: Flattening multidimensional observations discards correlations.

Key Innovation: A tensor autoencoder incorporates a predictive prior through Bayesian fusion. Its stated physical prior is low-rank tensor structure, not a validated conservation law.

89. WorldTS: World Modeling for Multimodal Covariate-aware Time Series Forecasting

Core Problem: External information should influence latent dynamics rather than only final predictions.

Key Innovation: WorldTS learns covariate-conditioned state evolution before fitting a decoder, evaluated on 21 time-series datasets; physical-state identifiability is not established.

90. Self-Supervised Representation Learning: From Spectral Foundation Models to Auroral Emission Spectra

Core Problem: Sparse labels and incompatible wavelength windows constrain spectral pretraining.

Key Innovation: Auroral masked pretraining is compared with optical and infrared foundation models, finding that spectral-domain compatibility matters; geological spectroscopy transfer requires separate validation.

91. WeaveAgent: A Two-Stage Tool-Routing Agent for Ultra-High-Resolution Remote Sensing Imagery

Core Problem: Large imagery makes token allocation and reliable tool invocation difficult.

Key Innovation: WeaveAgent separates tool routing from conditional image/tool execution and reports diagnostic ablations; its small trained system does not establish universal advantage over larger zero-shot baselines.

92. Gauss What You Need: Compact Gaussian Splatting Across Scene Scales

Core Problem: A fixed Gaussian budget does not suit both small and extensive image captures.

Key Innovation: Capture-derived allowances and training feedback adjust model size across scene scales, trading reconstruction quality against storage and rendering cost.

93. LUCID: Learning Under Confounding for Inference and Discovery in Time Series

Core Problem: Latent common causes can create spurious lagged causal edges.

Key Innovation: A spectral router selects a deconfounding strategy before graph discovery, with synthetic tests spanning intermittent and heavy-tailed confounding; real hazard causality remains unverified.

94. Progressive Memory Transformer: Memory-Aware Attention for Time-Series

Core Problem: Global losses can overlook intermediate temporal motifs.

Key Innovation: Window-aligned memory supports separate local, motif and sequence objectives, evaluated through classification, forecasting and cue-retention probes.

95. Brenier Meets Adversarial Training: Optimal Transport Geometry for Robust Learning

Core Problem: Local adversarial perturbations can waste transport cost and poorly represent distribution shift.

Key Innovation: Transport-map constraints and input-convex parameterization enforce geometric consistency in robust learning, tested on regression, images and control rather than geohazards.

96. Implicit Neural Representation for Hyperspectral Video Compression

Core Problem: Framewise coding misses spectral–temporal redundancy.

Key Innovation: An implicit video representation is evaluated using both rate–distortion metrics and downstream tracking, providing a task-aware compression reference.

97. SatNav: A Scalable Benchmark for Long-Horizon UAV Vision-Language Navigation from Satellite Imagery

Core Problem: Reconstructed 3D assets restrict navigation-benchmark scale.

Key Innovation: SatNav constructs long-horizon tasks from satellite scenes and tests satellite-to-UAV transfer; satellite crops are approximations to nadir flight observations.

98. ClearGS: Reliability-Aware Gaussian Splatting from Handheld Videos

Core Problem: Uneven viewpoints and blurred frames degrade Gaussian scene reconstruction.

Key Innovation: Graded view reliability and render-guided restoration preserve useful trajectory coverage; perceptual repair is not proof of recovered survey geometry.

99. Online Learning via Learned Latent Bayesian Tracking

Core Problem: High-dimensional parameter filtering is too costly for streaming updates.

Key Innovation: A learned low-dimensional parameter state permits extended-Kalman updates and lifting to full models, tested on drifting receivers and image classification.

100. Adaptive multi-resolution Gaussian processes: Scalable exact inference with naturally data-sparse covariance matrices

Core Problem: Large Gaussian-process systems force cost–fidelity trade-offs.

Key Innovation: Adaptive local basis functions create sparse covariance structure for exact inference within the proposed model; this is not exact inference for an arbitrary dense kernel.

101. Learning to Replace MCMC in Split-Gibbs Diffusion Posterior Sampling via Deep Unfolding

Core Problem: MCMC likelihood updates can dominate inverse-problem sampling.

Key Innovation: A learned unfolded denoiser replaces the split-Gibbs likelihood step, with nonlinear phase-retrieval tests and reuse of a pretrained prior.

102. Deep-Learning Solvers and Surrogates for Infinity and p-Laplace Problems

Core Problem: Large p-Laplace parameters create difficult three-dimensional solves.

Key Innovation: PINNs and DeepONets are compared with conventional solutions across domains and p values, with conditional convergence and approximation results.

103. Conformal Prediction under Exponential-Tilt Joint Shift

Core Problem: Reweighting calibration and shifting predictions need not preserve coverage equally.

Key Innovation: Controlled comparisons find that additional predictive tilting can shorten intervals or severely damage coverage depending on identifiability; unlabeled target inputs may be insufficient.

104. Equation discovery with Bayesian tree-adjoining grammars

Core Problem: Point-estimate symbolic models conceal structural uncertainty.

Key Innovation: Bayesian tree grammars infer model structures and parameters, including a physics-seeded wave-loading application, rather than returning only one fitted equation.

105. Beyond Empirical Support: Structured Outlier Generation via Sinkhorn Optimal Transport

Core Problem: Finite datasets omit rare cases needed for robust evaluation.

Key Innovation: Sinkhorn-guided latent outliers combine support-boundary geometry with semantic constraints, tested on time series and imagery; generated extremes are not physical event probabilities.

106. Exploring the Benefits of Vision Foundation Models for Unsupervised Domain Adaptation

Core Problem: Pretrained visual generalization may interact with adaptation machinery.

Key Innovation: A segmentation study combines foundation encoders with unsupervised adaptation and tests speed and out-of-domain accuracy, without assuming all adaptation components remain useful.

107. VFM-UDA++: Improving Network Architectures and Data Strategies for Unsupervised Domain Adaptive Semantic Segmentation

Core Problem: Losses designed for older encoders can harm foundation-model adaptation.

Key Innovation: VFM-UDA++ revises feature-distance regularization and multi-scale structure and evaluates scaling with additional source and target data.

108. What is the Added Value of UDA in the VFM Era?

Core Problem: Strong source-only baselines can reduce the apparent value of UDA.

Key Innovation: A companion evaluation finds that gains shrink with stronger synthetic sources and disappear in some diverse real-source settings; target-label and source-data conditions matter.

109. Regularizing modality contribution drift in multimodal continual learning

Core Problem: New tasks can change modality reliance and erase older predictions.

Key Innovation: Interventions estimate modality-contribution drift, which is regularized with or without replay; this provides a reference for evolving sensor-fusion systems.

110. Water Vapor Pressure Profile Retrieval From GNSS Radio Occultation Without NWP Background Fields During Inference Using a Decoupled Cross-Attention Network

Core Problem: NWP background profiles may be unavailable at inference time.

Key Innovation: A decoupled attention network retrieves water-vapor pressure from refractivity and context, with COSMIC-2, ERA5 and radiosonde comparisons; training dependence differs from inference requirements.

111. Detection of Oil Spills in Simulated Coastal Marine Environments via Hyperspectral Imaging and 3-D CNN

Core Problem: Oil thickness and water backgrounds produce ambiguous optical signatures.

Key Innovation: Wavelet preprocessing and 3D CNNs classify controlled simulated spills; controlled field performance is not validation on diverse operational spills.

112. Improving Forest Canopy Cover Mapping Through ICESat-2-Derived Understory DEM and Multisource Remote Sensing Data

Core Problem: Canopy structure and terrain errors interact in regional cover estimates.

Key Innovation: ICESat-2-derived terrain and optical/SAR features are checked against independent airborne LiDAR in three regions, distinguishing fitted and external validation performance.

113. Dynamic Priority Replay and Boundary-Aware Distillation for Remote Sensing Scene Domain Incremental Classification

Core Problem: New imagery domains can erase previously learned classification boundaries.

Key Innovation: Priority replay and boundary-aware distillation preserve informative historical samples and decision boundaries across scene datasets.

114. Large-Scale Quantum Kernels for Hyperspectral Data Classification

Core Problem: Quantum-inspired kernels face scaling and bandwidth concentration problems.

Key Innovation: Tensor-network and GPU simulation tests fidelity kernels without asserting a demonstrated hardware quantum advantage.

115. MDCF-Net: Multiscale Dual-Domain Collaborative-Aware Feature Fusion Network for Hyperspectral and Multispectral Image Fusion

Core Problem: Spectral and spatial detail must be retained across resolution scales.

Key Innovation: Spatial and frequency feature branches are fused at multiple scales, evaluated on simulated and real imagery.

116. Benchmarking Hyperspectral Transfer and Graph Learning for Coastal Water-Quality and Anomaly Mapping With PACE OCI

Core Problem: Short satellite records and sparse labels constrain regional water-quality maps.

Key Innovation: PACE spectral analogues, conventional predictors and graphs are compared; station prediction, dense inference and historical MODIS extrapolation remain distinct evidence levels.

117. PGTFusion: Progressive Group-Feature Transmission Fusion Network for Pansharpening

Core Problem: Bandwise details can be lost in undifferentiated feature fusion.

Key Innovation: Adjacent-band groups progressively exchange features before spatial fusion, evaluated on simulated and real datasets.

118. Mountain glacier evolution since the last interglacial

Core Problem: Sparse geomorphic evidence leaves historical glacier extent uncertain.

Key Innovation: Ensembles simulate nine mountain regions over 130,000 years and compare past and modern constraints; the product is a discussion preprint and not a present-day hazard forecast.

119. ChinaTidalVSC: a 10 m wall-to-wall dataset of vegetation structural complexity in China’s tidal wetlands derived from LiDAR-based relative entropy and AlphaEarth Foundation data

Core Problem: Coastal vegetation structure is incompletely captured by common indices.

Key Innovation: LiDAR-derived relative entropy and AlphaEarth support 10 m structural-complexity mapping in Chinese tidal wetlands; the discussion preprint does not demonstrate flood-protection performance.

120. The NASA-GISS ModelE2.1-CC2 ESM: development and evaluation

Core Problem: Coupled climate–ecosystem feedbacks require integrated model evaluation.

Key Innovation: NASA-GISS ModelE2.1-CC2 is assessed against diverse observations; global model agreement does not establish local hazard skill.

121. A hierarchical well-log-based interpretable machine learning workflow for identifying reservoir, seal, transition, and basal barrier domains in geological CO₂ storage

Core Problem: Functional subsurface labels can mask imbalance and uncertain boundaries.

Key Innovation: Blocked-depth evaluation and class-balanced metrics test geological storage domains; moderate discrimination and absent cross-well validation limit extrapolation.

122. Forest Canopy Height Retrieval from WorldView-3 Stereo Imagery Using Learned Feature Matching and Local Geometric Rectification

Core Problem: Repetitive crowns and stereo geometry degrade feature matching.

Key Innovation: Learned correspondences, local rectification and refinement improve tested stereo pairs against LiDAR at one pine-forest site, with convergence-angle dependence.

123. Extension of the WPS Model with Hierarchical Scene Representation and Analytic Ray-Primitive Intersection for Native Support of RAMI Actual Scene Descriptions in Radiative Transfer Simulation

Core Problem: Detailed canopy scenes impose geometry and memory costs.

Key Innovation: Hierarchical instancing and analytic ray–primitive intersections reduce tessellation uncertainty and memory in RAMI comparisons; these are simulation, not field retrieval, gains.

124. FAHRNet: Frequency-Aware Hybrid Refocusing Network for Remote Sensing Image Super-Resolution

Core Problem: Detail recovery must be compared under matched losses and compute.

Key Innovation: FAHRNet combines local, directional and frequency processing; matched-objective baselines reveal dataset- and metric-dependent advantages rather than universal superiority.

125. Development of drought index and quantitative assessment of spatiotemporal distribution characteristics of the Loess Plateau using multi-source remote sensing

Core Problem: Individual drought indices represent different physical components.

Key Innovation: A composite index is compared with soil moisture and regional trends. Hurst persistence is not an independently validated future drought projection.

126. A theoretical model for slurry infiltration in horizontally layered strata considering interlayer slurry transport

Core Problem: Interlayer slurry exchange complicates infiltration through stratified ground.

Key Innovation: The same-title author preprint combines laboratory tests with a Darcy- and mass-conservation model allowing interlayer transfer. Early/intermediate test agreement is not proof of field-scale seepage control.

Method evidence: Author preprint; journal-version equivalence not established.

127. Reflected Intensity Correction Based on Global 3D Plant Structure Acquired by Hyperspectral LiDAR

Core Problem: Title-level focus: Three-dimensional structure can bias reflected intensity.

Key Innovation: The title uses global plant geometry to correct hyperspectral LiDAR intensity. Methods, data and results could not be assessed from a reliable abstract.

128. Gaussian Mixture Model based Pseudo-Label Learning for Semi-Supervised Hyperspectral Image classification

Core Problem: Title-level focus: Hyperspectral classification is limited by scarce labels.

Key Innovation: The title uses Gaussian-mixture pseudo-labels in semi-supervised learning. Methods, data and results could not be assessed from a reliable abstract.

129. A Spatial-Spectral Collaborative Method for Radiometric Cross-Calibration of Ocean Color Satellite Sensors

Core Problem: Title-level focus: Different sensor responses impede comparison of spectral measurements.

Key Innovation: The title couples spatial and spectral information for ocean-color cross-calibration. Methods, data and results could not be assessed from a reliable abstract.

130. DDFGFormer: Dual-Decoupled Frequency-Guided Transformer for Hyperspectral Image Classification

Core Problem: Title-level focus: Spatial and spectral cues may interfere in hyperspectral representations.

Key Innovation: The title introduces decoupled frequency guidance in a classification Transformer. Methods, data and results could not be assessed from a reliable abstract.

131. Compact Multi-level-prior Tensor Representation for Hyperspectral Image Super-resolution

Core Problem: Title-level focus: Joint spatial–spectral recovery needs compact structural priors.

Key Innovation: The title combines multilevel priors in a tensor representation for hyperspectral super-resolution. Methods, data and results could not be assessed from a reliable abstract.

132. UHR-DETR: Efficient End-to-End Small Object Detection for Ultra-High-Resolution Remote Sensing Imagery

Core Problem: Title-level focus: Very large images and tiny objects create computational and localization challenges.

Key Innovation: The title describes an efficient end-to-end detector designed for ultra-high-resolution remote-sensing scenes. Methods, data and results could not be assessed from a reliable abstract.

133. A Unified Novel Architectural Paradigm for Remote Sensing Change Detection

Core Problem: Title-level focus: Change detection requires cross-time representations and consistent feature comparison.

Key Innovation: The title proposes a unified architecture; no specific benchmark advantage is inferred from its wording. Methods, data and results could not be assessed from a reliable abstract.

134. A new approach for pressure-fluid substitution

Core Problem: Title-level focus: Pressure and pore-fluid effects complicate rock-property interpretation.

Key Innovation: The title proposes a pressure–fluid substitution approach; governing equations and validation remain unavailable. Methods, data and results could not be assessed from a reliable abstract.

135. Decoupling Spectral Confusion in Cold-Temperate Mountain Forests: A Topography-Phenology-Spectral Decoupling Classification Framework

Core Problem: Title-level focus: Topography and phenology can confound mountain-forest spectra.

Key Innovation: The title separates terrain, seasonal and spectral contributions in cold-temperate forest classification. Methods, data and results could not be assessed from a reliable abstract.

136. Interpreting the Altimetric Radar Cross Section of Young Sea Ice

Core Problem: Title-level focus: Radar backscatter from young ice is difficult to interpret physically.

Key Innovation: The title examines altimetric radar cross sections of young sea ice. Methods, data and results could not be assessed from a reliable abstract.

137. Self-Supervised Learning for Very High-Resolution Forest Mapping with TanDEM-X InSAR Data

Core Problem: Title-level focus: Very-high-resolution forest mapping requires efficient use of interferometric data.

Key Innovation: The title applies self-supervised learning to TanDEM-X InSAR forest mapping. Methods, data and results could not be assessed from a reliable abstract.

138. RSLayout-DiT: Parameter-Efficient Layout-Controlled Remote Sensing Image Generation with Diffusion Transformers

Core Problem: Title-level focus: Synthetic scenes require spatial control and efficient adaptation.

Key Innovation: The title combines layout control and parameter-efficient diffusion-Transformer generation; generated images are not environmental observations. Methods, data and results could not be assessed from a reliable abstract.

139. FGPMamba: Frequency-geometry prior-guided Mamba for building change detection

Core Problem: Title-level focus: Frequency and geometric cues must be combined to distinguish building changes.

Key Innovation: The title uses frequency–geometry priors in a Mamba-based change detector. Methods, data and results could not be assessed from a reliable abstract.

140. A multi-scale retrieval of three-dimensional ocean temperature by merging satellite and in situ observations

Core Problem: Title-level focus: Satellite surface measurements incompletely constrain subsurface temperatures.

Key Innovation: The title merges satellite and in-situ observations in a multiscale three-dimensional retrieval. Methods, data and results could not be assessed from a reliable abstract.

141. Plants alter vertical hydrological connectivity to influence moisture recycling through varying water use strategies in China's Loess Region

Core Problem: Title-level focus: Plant water strategies alter connections between subsurface stores and atmospheric recycling.

Key Innovation: The title examines vertical hydrological connectivity and moisture recycling in the Loess Region. Methods, data and results could not be assessed from a reliable abstract.

142. Multi-stage diagnostic and characterization framework for bored pile defects driven by distributed thermo-mechanical sensing fingerprints

Core Problem: Title-level focus: Different pile defects can produce overlapping sensor responses.

Key Innovation: The title combines distributed thermal and mechanical sensing in a multi-stage diagnostic framework. Methods, data and results could not be assessed from a reliable abstract.

143. Hierarchical source-zone inversion and century-scale transport interpretation of landfill-derived groundwater contamination

Core Problem: Title-level focus: Long-term contaminant signals can be difficult to attribute to source zones.

Key Innovation: The title combines hierarchical source-zone inference with century-scale transport interpretation; geohazard use is methodological. Methods, data and results could not be assessed from a reliable abstract.

144. Phase-field model and experimental study of acid fractures coupled with hydro-mechano-reactive flow and strain-softening

Core Problem: Title-level focus: Reactive transport, deformation and softening interact during fracture evolution.

Key Innovation: The title combines phase-field modeling and experiments for acid fractures; application to natural weathering or slope failure is not established. Methods, data and results could not be assessed from a reliable abstract.

145. Stress and stiffness anisotropies of sands during high-stress oedometer test

Core Problem: Title-level focus: Loading history changes stress and stiffness anisotropy in sand.

Key Innovation: The title examines these anisotropies during high-stress oedometer testing. Methods, data and results could not be assessed from a reliable abstract.

146. GAUDI: Geometry-Aware Diffusion for Calibrated Air-Quality Time-Series Imputation

Core Problem: Contiguous sensor outages differ from isolated missing values.

Key Innovation: Conditional diffusion tests variable and mask conditioning for block-missing air-quality data; the demonstrated gain is restricted to the reported dataset and window protocol.

147. LiTe-GS: Oracle-Efficient Next Best View Selection for 3D Gaussian Splatting

Core Problem: Information-guided reconstruction can spend too much computation ranking candidate views.

Key Innovation: LiTe-GS samples candidate subsets to reduce information-oracle calls while maintaining comparable reconstruction on two benchmarks; field survey efficiency remains to be tested.

148. ProCAP: Probabilistic Cross-Attentive Prompt Learning for Vision-Language Models

Core Problem: Prompt adaptation can overfit small labelled datasets and domain shifts.

Key Innovation: Probabilistic visual and textual prompts interact through cross-attention while the CLIP backbone remains frozen; hazard-class transfer is prospective.

149. The Shape of Events: Edge-Based Inductive Biases via Cross-Domain Distillation

Core Problem: Texture dependence can weaken visual robustness.

Key Innovation: Event-to-RGB distillation promotes shape and colour invariance but exposes complementary frequency and geometry vulnerabilities; it does not deliver universal robustness.

150. Image Reconstruction from Phase with Untrained Neural Priors

Core Problem: Missing spectral magnitudes leave intensity and spatial structure underconstrained.

Key Innovation: Untrained neural guidance is followed by phase/support refinement; microscopy tests show that lower phase residual need not mean a more accurate image.

151. Learning Polarization Image Restoration with General Restoration Priors

Core Problem: Coupled acquisition degradations impair polarization cues.

Key Innovation: A normalized-Stokes representation separates intensity and polarization branches, transferring general restoration priors and evaluating composite degradation.

152. LLPR: Location-aware learning and physics-based reconstruction for raindrop removal from a single image

Core Problem: Lens raindrops obscure monitoring imagery.

Key Innovation: Training-only location guidance and a transparency-based reconstruction model improve single-image restoration without retaining the auxiliary location branch at inference.

153. Query-Conditioned Prototype Adaptation for Cross-Domain Few-Shot Learning: Single-Query Inference, Controlled Comparisons, and Failure Modes

Core Problem: Sparse target labels may not support robust prototype adaptation.

Key Innovation: A frozen-encoder controlled study finds query-conditioned benefits on some low-shot targets, including EuroSAT, but not universal improvement across domains or shot counts.

154. Missingness-Aware Conformal Prediction Under Cross-Hospital Distribution Shift

Core Problem: Pooled coverage can hide site-specific errors associated with missing observations.

Key Innovation: Clinical experiments and decompositions show that missingness-group calibration improvements can reverse within sites; the transferable result is the evaluation warning, not clinical-to-hazard accuracy.

155. Skip the Talk, Re-Focus on Vision: Latent Reasoning for Reasoning Segmentation in Multimodal Large Language Models

Core Problem: Verbose language reasoning can interfere with fine visual localization.

Key Innovation: Compact spatially aligned latent tokens replace explicit verbal reasoning before segmentation, evaluated on reasoning-segmentation benchmarks with prospective remote-imagery use.

156. Counterfactual Online Conformal Prediction Under Adaptive Logging

Core Problem: Actions affect which outcomes are available for subsequent calibration.

Key Innovation: Inverse-propensity and doubly robust updates target counterfactual coverage under positivity; application to adaptive environmental monitoring is prospective.

157. Robust to Which Model Change? A Unified Evaluation of Robust Counterfactual Explanations

Core Problem: Counterfactual explanations can fail differently under retraining and parameter perturbation.

Key Innovation: A fixed-instance protocol compares eight model-change families and distinguishes robustness from coverage and intervention cost; guarantees for one change do not establish robustness to another.

158. IDM-Net: A Lightweight Illumination-Decoupled Modulation Network for Low-Light Image Enhancement

Core Problem: Small image-enhancement models struggle to recover illumination without distorting colour.

Key Innovation: A separate luminance encoder guides spatial modulation and refinement, evaluated on low-light benchmarks; improved appearance is not proof of accurate event detection.

159. LipSSM: Structurally Lipschitz-Bounded Cascaded State-Space Model via Metric Transfer between Consecutive SSM Layers

Core Problem: Layerwise robustness constraints can be overly restrictive in long-memory models.

Key Innovation: LipSSM transfers metric information between cascaded state-space layers to obtain structural Lipschitz bounds, with prospective use in robust sensor-sequence modeling.

160. Self-Supervised Perceptually Interpretable Monocular Depth Estimation

Core Problem: RGB depth predictions obscure the contribution of individual visual cues.

Key Innovation: Separate perceptual feature branches produce interpretable depth estimates and explicit fusion, tested on KITTI without ground-truth depth supervision.

161. PhoenixSR: Generative Heterogeneous Distillation Unleashes Efficient Models for Real-World Super-Resolution

Core Problem: Generative priors improve perceptual detail but increase inference cost.

Key Innovation: Distribution-matching distillation transfers diffusion guidance to feed-forward restorers; reconstruction and perceptual fidelity remain distinct evaluation targets.

162. Robust Graph Clustering Network for Multiple Missing Data

Core Problem: Jointly missing attributes and links propagate imputation errors.

Key Innovation: Separate imputation branches, mixture priors and boundary-aware contrastive training address simultaneous feature and topology incompleteness on graph benchmarks.

163. Bayesian Optimization with Fisher Information Geometry: Gradient Bounds and Trust-Region Methods

Core Problem: High-dimensional Bayesian optimization can have weak acquisition gradients.

Key Innovation: Pullback Fisher geometry supplies sensitivity weights for trust regions; benefits depend on surrogate and task, with prospective use in simulator calibration.

164. Teacher-Anchored Selection of Post-Training Quantized Models under Domain Shift

Core Problem: Model confidence and teacher similarity can misrank deployment choices after domain shift.

Key Innovation: A fixed-family evaluation separates label-free failures from small-label anchored selection; lower quantization distortion alone does not identify the best target-domain model.

165. TaskIR: Task-Driven Image Restoration via Degradation Adaptation and Task Feedback

Core Problem: Cleaner-looking images need not improve downstream detection.

Key Innovation: Degradation-conditioned restoration is refined using selective task feedback, coupling visual reconstruction with downstream feature requirements.

166. ALF: An Active Learning Framework for Scientific Discovery

Core Problem: Expensive simulations and labels require repeatable acquisition loops.

Key Innovation: ALF supplies a modular open-source loop for both offline benchmarks and online oracle queries; value to hazard sampling depends on the chosen acquisition policy and oracle.

167. Geometric Inconsistency Localization in Multi-View Image Sets

Core Problem: Plausible synthesized views can disagree geometrically.

Key Innovation: A pixel-labelled inconsistency benchmark and cross-view classifier distinguish inconsistent regions from consistent deformation, supporting future 3D reconstruction quality checks.

168. Different Corruptions, Different Signals: Uncertainty and Loss in Federated Data Quality

Core Problem: Uncertainty and label loss detect different errors.

Key Innovation: Federated image tests show label loss detects persistent wrong labels better than predictive uncertainty; environmental inventory audits should not equate confidence with label correctness.

169. Scaffold: Support Graph Theory Based Sparsification for Graph Neural Networks

Core Problem: Removing sensor-network edges can distort communication structure.

Key Innovation: Support-graph sparsification controls rerouting dilation and congestion, retaining predictive performance with reduced graph memory on benchmark tasks.

170. GraphWrit3R: End-to-End 3D Scene Graph Writing

Core Problem: Multistage scene-graph pipelines can propagate object-association errors.

Key Innovation: GraphWrit3R aligns point-cloud and Gaussian representations before decoding structured scene relations, with prospective use in mapped exposure inventories.

171. Persistent Homology of Time Series through Complex Networks

Core Problem: Feature construction choices complicate comparisons of topology-based classifiers.

Key Innovation: A common persistent-homology pipeline isolates graph and distance choices; no network construction dominates all signal types.

172. MARCEDES: Score-based causal discovery under non-Gaussianity with continuous optimization

Core Problem: Gaussian assumptions can misrepresent causal structure.

Key Innovation: A residual-based score combines sparsity penalties and a soft acyclicity constraint, evaluated in simulations; observational geological causation is not established.

173. Efficient Constrained Graph Search for Post-hoc Error Correction in Binary Classifiers

Core Problem: Reducing one error class can introduce damaging errors elsewhere.

Key Innovation: Rule-path search adds interpretable post-hoc corrections under explicit new-error constraints, exposing rather than eliminating the false-positive/false-negative trade-off.

174. Spatial Information Bottleneck for Interpretable Visual Recognition

Core Problem: Background correlations can contaminate visual explanations.

Key Innovation: A spatial information bottleneck regulates foreground and background gradient information, evaluated across visual explanation methods; clearer maps do not by themselves establish causal interpretation.

175. Learning from Next-Frame Prediction: Autoregressive Video Modeling Encodes Effective Representations

Core Problem: Image-only masking underuses temporal structure.

Key Innovation: Masked next-frame prediction separates semantic representation from flow-matching decoding, evaluated by downstream probing rather than hazard-video performance.

176. Learning with Volterra Neural Networks: A System Theoretic Perspective

Core Problem: Explicit higher-order interactions become computationally expensive.

Key Innovation: Kernelized Volterra layers separate interaction orders without full tensor parameterization, evaluated on vision benchmarks with prospective sensor-signal use.

177. Deep Learning and Reflectance Spectroscopy for Mapping Rare Earths in Brazil

Core Problem: Mineral mapping needs alternatives to dense chemical sampling.

Key Innovation: Laboratory spectra, hyperspectral imagery and neural models map potential rare-earth occurrences; relevance is spectral methodology rather than hazard detection.

178. DAC-Net: Divide-and-Conquer Network for Infrared Small Target Detection

Core Problem: Small targets compete with structured background clutter.

Key Innovation: Hybrid local/global feature processing and edge supervision are evaluated on infrared benchmarks; geohazard performance is not established.

179. Beyond Linear Convolutional Representations: Manifold-Aware Neighborhood Feature Learning for PolSAR Ship Detection

Core Problem: Flattening covariance matrices loses geometric and inter-pixel structure.

Key Innovation: Manifold-aware neighborhood embeddings approximate missing coherence information in multilook ship detection; transfer to terrain classification is prospective.

180. Assessing the Effects of Spartina alterniflora Management on Mangrove Carbon Storage Dynamics in the Zhangjiang Estuary Using Multi-Source Remote Sensing from 2016-2025

Core Problem: Vegetation management and canopy changes complicate carbon-stock interpretation.

Key Innovation: Optical/SAR and LiDAR inputs map mangrove change; temporal alignment with management does not isolate a causal management effect or quantify flood protection.

181. Frantic Spawners in a Lazy River: When Sea Lampreys Match Hydrology as a Factor of Sediment Mobility and Morphological Changes

Core Problem: Biological activity and moderate flows can both mobilize riverbed material.

Key Innovation: Tagged particles and centimetric topographic surveys compare lamprey spawning with winter-flow sediment mobility at two sites; transfer concerns sediment budgets, not flood forecasting.

182. A Low-Complexity Separated Space-Time IAA-LS Method With Local Joint Refinement for Forward-Looking Super-Resolution Imaging

Core Problem: Title-level focus: High-resolution forward-looking radar inversion can be computationally expensive.

Key Innovation: The title combines separated space–time estimation with local joint refinement. Methods, data and results could not be assessed from a reliable abstract.

183. Spatial-Temporal Infrared Small Target Detection via Motion-Aware Nonlocal Low-Rank Implicit Representation

Core Problem: Title-level focus: Small targets must be separated from spatially and temporally structured backgrounds.

Key Innovation: The title combines motion-aware nonlocal low-rank information with implicit representations. Methods, data and results could not be assessed from a reliable abstract.

184. Ship Detection for SAR Images Based on 0-Dimensional Topological Persistent Homology and Local Density-Suppression Diffusion Model

Core Problem: Title-level focus: Clutter and local density complicate radar object detection.

Key Innovation: The title uses zero-dimensional persistence and density suppression for ship detection; hazard transfer is prospective. Methods, data and results could not be assessed from a reliable abstract.

185. Range-Dependent Folded Clutter Suppression for IPPAW-Based Side-Looking Airborne Radar in Non-Homogeneous Environments

Core Problem: Title-level focus: Range-varying clutter affects nonhomogeneous airborne scenes.

Key Innovation: The title targets folded-clutter suppression in side-looking radar; natural-hazard detection is not established. Methods, data and results could not be assessed from a reliable abstract.

186. Multi-Sensor Fusion and Decoupled Spatial Context-Aware Feature Adapter for Crop Mapping in Sub-Hectare Fragmented Farmlands

Core Problem: Title-level focus: Small fragmented parcels complicate spatial context and sensor fusion.

Key Innovation: The title decouples spatial-context adaptation for crop mapping; transfer to hazard mapping requires new validation. Methods, data and results could not be assessed from a reliable abstract.

187. A spatially-adaptive automated sample generation framework via multi-source phenological fusion for large-scale crop mapping: demonstrated on peanuts in Shandong, China

Core Problem: Title-level focus: Large-area classification needs labels that reflect local phenology.

Key Innovation: The title combines multi-source phenological information for spatially adaptive crop-sample generation. Methods, data and results could not be assessed from a reliable abstract.