TerraMosaic Daily Digest: September 14, 2026
Daily Summary
Landslide research expands from transport corridors to planetary terrain. A national railway study compares five machine-learning models, adds a stacking ensemble and projects susceptibility under three emissions pathways; highly susceptible corridor area increases from 23.5% historically to 27.1-28.8%, with the largest transition in the Changbai Mountains. On Mars, the first broad comparison of modern segmentation architectures on seven-band multimodal data shows that joint local geomorphic features and transformer-scale context improve geographically separated landslide delineation. In Scandinavia, a transformer trained on SNOWPACK simulations recovers broad variations in satellite-observed avalanche activity, although smoothing, uncertain detection timing and the absence of an untouched test winter preclude an operational forecasting claim. New studies also target DEM-scale effects in debris-flow susceptibility and long-term creep of a water-level-driven riverbank landslide.
Fault mechanics papers challenge two simplifying assumptions: that faults are infinitely thin and that failure always relaxes stored energy. Mollified elastic-dislocation kernels remove on-fault singularities while representing finite-width zones independently of mesh scale, enabling stable boundary-element models with heterogeneous materials and topography. Three-dimensional simulations show that geometric interaction alone can switch a fault system among periodic earthquakes, slow-slip events and irregular sequences. Injection experiments further link permeability change to earthquake frequency-magnitude statistics, while granular-fault measurements reveal avalanches that dilate yet increase elastic energy by reorganizing force chains outside the shear band.
Hydroclimatic hazards are increasingly represented as evolving spatial objects rather than station statistics. A true-state constraint applied to GraphCast traces upstream forecast errors into the large-scale dynamics of two extreme events. Sixty-one years of three-dimensional cold-wave tracking over China reveal shorter and thermodynamically weaker events whose spatial reach remains extensive; global atmospheric-river analysis reconciles multiple classification schemes around a wind-driven, cyclone-associated archetype; and an extended tropical-cyclone record shows that the QBO-track relationship changed sign around the 2000s. Storm Gloria shows that multivariate, spatial and sequential compound hazards co-occurred across the western Mediterranean, altered joint return-period estimates and coincided with the largest losses where hazard levels were highest.
Observation and uncertainty are being connected more directly to decisions. Decision-oriented calibration for Earth-system foundation models reduces regret and missed events by optimizing action-conditional risk rather than forecast coverage alone. Multi-temporal InSAR and explainable learning identify stage-dependent groundwater thresholds for Beijing subsidence, controlled comparisons separate algorithmic from sensor-driven uncertainty in loess-plateau deformation, and two-component InSAR resolves opposing horizontal motion and widespread subsidence across the Erta Ale volcanic segment. Groundwater-drought analysis resolves differences in aquifer memory, Cantareira geodesy links storage deficits to surface deformation, and OceanTACO harmonizes multisensor sea-surface observations for reproducible analysis.
Key Trends
The day's studies converge on finite-width fault mechanics, event-based hazard analysis and decision-calibrated uncertainty.
- Fault geometry and internal state are becoming explicit: Finite-width dislocation kernels, interacting three-dimensional faults, permeability-b-value coupling and force-chain reorganization replace planar, singular or purely relaxing failure assumptions.
- Event-based analyses are replacing station summaries: Cold-wave tracking preserves lifecycle and footprint, avalanche models retain spatial and temporal activity, and compound-flood studies resolve simultaneous and sequential hazards; atmospheric-river work instead reconciles competing global classifications.
- Risk assessment is moving beyond point accuracy: Utility-aware calibration quantifies missed events and action-conditional risk, while groundwater-drought and railway-landslide studies expose recovery and scenario-dependent planning constraints.
- Multi-sensor geodesy is separating drivers from measurements: InSAR comparisons isolate algorithm and sensor effects, while groundwater thresholds, water-storage loading and horizontal-vertical volcanic deformation connect observed motion to distinct physical controls.
- Remote sensing is testing transfer, evidence grounding and resource limits: Geographically separated landslide evaluation, deterministic tools for change VQA and onboard satellite-image restoration test domain shift, evidence use and constrained computation.
Selected Papers
The 14 September papers connect landslide mapping, fault mechanics and hydroclimatic extremes across scales. Direct studies address railway susceptibility, Martian landslide segmentation, induced seismicity, avalanche activity, compound flooding, drought, subsidence and volcanic deformation; methodological work focuses on decision-calibrated uncertainty, explainable InSAR and physics-constrained modeling.
1. Landslide susceptibility assessment along China’s railways assisted by machine learning
Core Problem: Build a unified national framework and project railway landslide susceptibility under multiple SSPs through 2100.
Key Innovation: Five machine-learning models are evaluated before a stacking ensemble raises AUC to 0.9135; SSP precipitation projections then quantify expansion of highly susceptible corridor area.
2. Three dimensional non-singular mollified elastic dislocation theory for extended width fault zones and inhomogeneous boundary element models
Core Problem: Remove fictitious on-fault singular stresses and the infinite-thinness assumption of classical elastic dislocation theory.
Key Innovation: Analytically integrated mollified displacement-discontinuity kernels for triangular elements, with width scale decoupled from mesh size and support for material heterogeneity and topography.
3. Global-Local Contextual Progressive Expansion Network for Martian Landslide Segmentation in Multimodal Remote Sensing Imagery
Core Problem: Segmenting sparse, irregular, morphologically variable Martian landslides under heterogeneous multimodal observations.
Key Innovation: TransCPLES couples progressively expanded local geomorphic features with Transformer global context and is assessed on geographically distinct MMLSv2 samples.
4. Correlation of Permeability Evolution With Frequency‐Magnitude Distributions for Injection‐Induced Seismicity Under Varying Initial Stress States
Core Problem: How initial fault stress controls seismic failure mode and permeability evolution during fluid injection.
Key Innovation: Couples controlled smooth/rough-fault experiments and acoustic-emission mapping to a Gutenberg–Richter-based permeability inversion.
5. Quantifying the Role of 3D Fault Geometry Complexities on Slow and Fast Earthquakes
Core Problem: Determining whether fault interaction geometry alone can generate slow slip and diverse earthquake sequences.
Key Innovation: Defines a geometry-dependent Coulomb-interaction metric and identifies four slip regimes in 3D quasi-dynamic simulations.
6. Avalanches can increase stored energy in a granular fault
Core Problem: Determining whether abrupt granular-fault failures always relax and compact the system.
Key Innovation: Simultaneous torque, layer-thickness, acoustic-emission, and photoelastic force-network measurements reveal avalanches that dilate while increasing transmitted elastic energy.
7. Data-driven Prediction of Satellite-observed Avalanche Activity from Snowpack Simulations
Core Problem: Predict next-day SAR-derived avalanche activity from five-day SNOWPACK histories where field observations are sparse.
Key Innovation: A transformer maps five-day, multi-elevation and multi-slope SNOWPACK histories to a SAR-derived activity index over five winters in Norway and Sweden.
8. Attribution analysis and threshold identification of land subsidence in the Beijing Plain based on MT-InSAR technology and explainable artificial intelligence
Core Problem: Separate changing groundwater drivers and stage-dependent land-subsidence responses.
Key Innovation: Multi-sensor MT-InSAR combined with Random Forest and SHAP threshold attribution across hydrological periods.
9. Geodetic Assessment of Drought Intensity and Hydrological Dynamics in the Cantareira System, Southeastern Brazil
Core Problem: Monitor drought propagation and storage dynamics in a complex managed basin.
Key Innovation: Integrated GPS, Sentinel-1 InSAR, GRACE/GRACE-FO, groundwater, reservoir, and meteorological records with correlation and EOF analysis.
10. Horizontal and Vertical Deformations of the Erta Ale Volcanic Segment in the Northern Rift Region of Ethiopia Using Sentinel-1 InSAR Time-Series Analysis with LiCSBAS
Core Problem: Constrain tectono-magmatic motion beyond single-volcano line-of-sight studies.
Key Innovation: Four-year ascending/descending Sentinel-1 LiCSBAS decomposition across six volcanoes, identifying approximately 110 mm separation and major subsidence.
11. Storm Gloria (2020): coexisting types of compound flooding in the West Mediterranean Region
Core Problem: Characterize how multivariate, spatial and temporal compound flooding coexisted during Storm Gloria and related to observed losses.
Key Innovation: Joint rainfall, wind, wave, river/coastal flooding, exposure, vulnerability, and insurance analysis.
12. Integrating propagation and recovery dynamics into groundwater drought vulnerability assessment through exposure, pressure, and aquifer system response
Core Problem: Integrate rainfall deficits, pumping pressure, storage response, and recovery into vulnerability assessment.
Key Innovation: An exposure-pressure-aquifer-response framework applied to three aquifer groups; apparent non-recovery in the deep confined system remains exploratory because that group contains only two monitoring wells.
13. The Asynchronous Evolution of Cold Waves Over China Revealed by 3D Spatiotemporal Tracking
Core Problem: Station statistics do not resolve the joint spatial and temporal evolution of contiguous cold-wave events.
Key Innovation: Applies 3D connected-component labeling to 1960–2020 observations and reveals a shift toward shorter, thermodynamically weaker but spatially extensive cold waves.
14. Linkage in the Diversity of Atmospheric Rivers: A Global Perspective on Multi‐Framework Classification
Core Problem: Multiple atmospheric-river classification systems remain fragmented and their statistical relationships are unknown.
Key Innovation: Provides the first global linkage analysis across frequency, moisture-wind and cyclone-association frameworks, resolving a convergent wind-driven, cyclone-associated archetype.
15. QBO–TC Relationship in the Western North Pacific Is Nonstationary
Core Problem: The proposed QBO control on tropical-cyclone tracks has uncertain robustness through time.
Key Innovation: Extends the record through 2024 and identifies a sign reversal around the 2000s; the change is likely influenced by internal climate variability and weakens the case for QBO as a stable track predictor.
16. Decision-Oriented Uncertainty Quantification for Risk Control in Earth System Spatiotemporal Foundation Models
Core Problem: Convert forecast uncertainty into reliable action-conditional risk rather than merely calibrated prediction intervals.
Key Innovation: A decision-risk adapter and utility-aware calibration layer for spatiotemporal Earth foundation models, with reported reductions in regret, missed events, and decision calibration error.
17. Diagnosing Forecast Error Propagation and Large‐Scale Dynamics of Weather Extremes With an AI Weather Model
Core Problem: Tracing how upstream state errors and circulation anomalies propagate into failed regional-extreme forecasts.
Key Innovation: True-state and climatology constraints use GraphCast as a perturbation-based diagnostic of error propagation; the resulting responses are consistent with dynamical nudging experiments.
18. The overlooked scale effect on seepage failure modes in cylindrical cofferdams: An axisymmetric correction factor and behavioral regimes for design
Core Problem: Account for radius-to-depth scale effects missing from major geotechnical standards.
Key Innovation: A 3D axisymmetric correction factor, four behavioral regimes, and parameter hierarchy supported by numerical and field evidence.
19. Sustainable and Seismically Resilient Confined Masonry: Global Progress in Research and Design Practice
Core Problem: Expand safe, sustainable confined-masonry design, schools, and retrofits across diverse codes and regions.
Key Innovation: Expert synthesis from more than 14 countries covering codes, research, schools, retrofitting, durability, and climate adaptation.
20. Km-scale regional coupled system in the Northwest European shelf for weather and climate applications: RCS-UKC4
Core Problem: Represent atmosphere-ocean-wave-land-river-biogeochemistry coupling at kilometer scale.
Key Innovation: RCS-UKC4 couples atmosphere, ocean, waves, land, rivers and biogeochemistry at kilometer scale, with demonstrations of near-real-time ensembles, storm-wave skill and 10-minute coupling.
21. Comparative Multi-Data and Multi-Method InSAR for Deformation Monitoring and Visualization: A Case Study of Baode, China
Core Problem: Separate algorithm-induced from sensor-induced discrepancies and quantify uncertainty in surface-deformation products.
Key Innovation: Controlled Radarsat-2/Sentinel-1 comparison across PS-, SBAS-, and IPTA-InSAR plus turning-point analysis and precipitation linkage.
22. Macroscopic mechanical response and microscopic action mechanism of hydrogel–modified Yili loess in seasonal frozen region
Core Problem: Prevent pore coarsening and strength loss in seasonally frozen loess.
Key Innovation: Hydrogel treatment evaluated with triaxial tests, SEM, NMR, FTIR, and 20 freeze-thaw cycles, producing an 82.5 percent strength increase.
23. Seismic energy dissipation and vibration mitigation through a PVDF-based hybrid base isolation system with electromagnetic control
Core Problem: Combine base isolation, energy harvesting, and active electromagnetic control without external power.
Key Innovation: PVDF-electromagnetic hybrid isolation validated in ANSYS and shaking-table tests on eight scaled frames under 11 earthquake records.
24. Seismic resilience of hollow core slab – beam connections with novel ductile detailing: experimental and numerical simulations
Core Problem: Develop ductile detailing that retains support and moment capacity during earthquakes.
Key Innovation: Two new connection configurations tested cyclically to plus or minus 5.36 percent drift and reproduced with nonlinear 3D finite elements.
25. Multitemporal InSAR analysis of long-term creep and active thickness changes in a paleo riverbank landslide driven by anomalous water levels
Core Problem: Resolving persistent deformation and changing active-layer thickness in a riverbank landslide.
Key Innovation: Title indicates multitemporal InSAR coupling of kinematics, thickness evolution, and hydrologic forcing.
26. Scale-aware debris-flow susceptibility and runout modelling: implications of DEM configuration and sampling design for model discrimination and spatial coherence
Core Problem: Understanding how scale, DEM configuration, and sampling alter model discrimination and coherent hazard patterns.
Key Innovation: Title explicitly integrates scale-aware susceptibility and runout evaluation with data-design effects.
27. Mitigating Small-Target Omission in Post-Earthquake Collapsed-Building Segmentation: A Dual-Branch Feature-Decoupling Network
Core Problem: Reducing omission of small collapsed structures in post-event imagery.
Key Innovation: Dual-branch feature decoupling tailored to small-target segmentation.
28. R2F-YOLO: Multi-Scale Cross-Modal Residual Fusion for RGB-Thermal Wildfire Flame Detection
Core Problem: Detecting flames across scales and sensing modalities.
Key Innovation: Multi-scale cross-modal residual fusion within a YOLO detector.
29. Quantitative evidence for the advantage of fourier-spectrum based evaluation of site amplifications: Towards utilization of small-magnitude records
Core Problem: Quantifying site amplification when strong-motion observations are limited.
Key Innovation: Fourier-spectrum evaluation aimed at exploiting small-magnitude records.
30. Grain size and thickness of event layers in relation to the flood magnitude: examples from floodplains in the middle and lower Yangtze River, China
Core Problem: Linking event-layer grain size and thickness to flood magnitude.
Key Innovation: Title indicates empirical examples across middle and lower Yangtze floodplains.
31. Tracking diurnal aerosol dynamics and smoke transport over East Asia using multi-geostationary satellite fusion (GK-2B, Himawari-9, FY-4B)
Core Problem: Resolving rapidly changing aerosol and smoke motion across satellite domains.
Key Innovation: Title indicates fusion of three geostationary satellite systems.
32. OceanTACO: a multi-sensor global ocean sea surface state dataset
Core Problem: Remove archive incompatibilities across satellite, float, and reanalysis sea-surface observations.
Key Innovation: Consistent access to sea-surface height, temperature, salinity, and winds from multiple observing systems.
33. Bay Assessment Model: A Python-based Salinity Projection Tool for Florida Bay, USA
Core Problem: Projecting water level and salinity across Florida Bay under hydrologic boundary changes.
Key Innovation: Open-source mass-conservative 54-basin Python model validated over 1999-2026 with explicit regime-dependent bias analysis.
34. Experimental Study on Shear Fracture Characteristics of Bedded Shale Based on Fractal Quantification
Core Problem: Replace subjective joint roughness inputs with objective fractal measures under bedding and stress variation.
Key Innovation: Direct shear tests, eight fractal estimators, and a fractal-strength constitutive relation with reconstructed curves.
35. Experimental study and equivalent parameter analysis of a semi-active negative stiffness friction damper
Core Problem: Derive usable equivalent parameters and control friction through loading-state transitions.
Key Innovation: Negative-stiffness friction device, analytical equivalent linearization, microcontroller control, and cyclic tests.
36. PhysBrain 1.5: From Vision-Language Models to Physical Foundation Models
Core Problem: Unify physical-scene understanding, action generation, and future-state prediction in one model.
Key Innovation: Joint autoregressive tokenization of language, end-effector motion, RGB, depth, and robot masks, pretrained from human interaction video and adapted with robot and simulation data.
37. Effects of Side Bileaders on Parent Positive Leader
Core Problem: The influence of side bileaders on propagating parent positive leaders is poorly constrained.
Key Innovation: Combines high-speed video with LF/VLF channel mapping and identifies connection-driven luminosity and branch formation in eight of nine observed side-bileader cases.
38. Translation-Invariant Tile-Based Phase Unwrapping with Residual-Weighted Multipath Averaging
Core Problem: Separating true phase discontinuities from wrapping, noise, undersampling, and decorrelation without tile seams.
Key Innovation: A translation-invariant DCT/least-squares tiling ensemble weighted by Poisson residuals, including a ground-truth-free cyclic metric.
39. A Machine Learning API for Earth Observation Data Cubes Based on openEO
Core Problem: Standardizing ML workflows over heterogeneous spatiotemporal EO data-cube backends.
Key Innovation: A process-level openEO specification covering model initialization, actions, management, classical ML, deep learning, and foundation models across independent implementations.
40. Attention Is All You Need (to Avoid Spurious Oscillations)
Core Problem: Moving discontinuities over large time steps without spurious oscillations or loss of conservation.
Key Innovation: A CFL-conditioned attention flux learns a state-dependent upstream stencil and transfers from Burgers transport to shallow-water equations.
41. Multimodal Foundation Models Adaptation based on Domain-Aware Relaxed Orthogonal Subspace for Remote Sensing
Core Problem: Fixed low-rank PEFT subspaces poorly match downstream remote-sensing activation geometry.
Key Innovation: DROS learns activation-conditioned relaxed-orthogonal subspaces and shares transformations across modalities without inference overhead.
42. Selective Tool Use for Agentic Change Visual Question Answering in Remote Sensing
Core Problem: Prevent VLM errors on explicit measurements and spatial reasoning in bi-temporal remote-sensing VQA.
Key Innovation: A learned gate selectively invokes deterministic change-analysis tools and conditions answers on structured evidence.
43. Unsupervised Point Cloud Registration via Training-Time Semantic Guidance
Core Problem: Overcome geometric ambiguity and noisy pseudo-labels in sparse outdoor point-cloud registration.
Key Innovation: Train-only semantic guidance, dual-cue re-matching, semantic-geometric pseudo-label mining, and predictive distillation with no inference overhead.
44. MambaMPD: A Mamba-Driven Segmentation Framework for Marine Pollution Detection from Remote Sensing Imagery
Core Problem: Segment subtle marine pollution with ambiguous boundaries and sea-like appearance.
Key Innovation: Wavelet frequency-aware augmentation and multi-scale Laplacian edge-guided attention in an efficient Mamba U-Net.
45. Transfer Learning for Socioeconomic Estimation in Forced-Displacement Settings
Core Problem: Adapting general-population EO wealth models to camps and host communities.
Key Innovation: Transfers a multimodal spatiotemporal transformer pretrained on 1.2 million households and validates it across three African displacement settings.
46. Aries: A Proprietary Medium-Range Weather Prediction Model for the Energy Industry
Core Problem: Building competitive industry-scale global weather forecasts from ERA5.
Key Innovation: A SwinTransformer predicts 85 atmospheric variables and is evaluated strictly out of sample against ECMWF HRES and AIFS.
47. SAR-FAH: A Frequency-Adaptive Hybrid Network based on Neural ODEs for Structural-Preserving SAR Despeckling
Core Problem: Suppress multiplicative SAR speckle without blurring heterogeneous-region edges and textures.
Key Innovation: Wavelet separation of homogeneous and heterogeneous content, Neural-ODE low-frequency smoothing, and deformable U-Net high-frequency restoration.
48. CLIMATEAGENT: Multi-Agent Orchestration for Complex Climate Data Science Workflows
Core Problem: Execute reliable end-to-end analyses across large heterogeneous climate datasets and APIs.
Key Innovation: Climate-specific multi-agent orchestration with dynamic API introspection, code generation, self-correction, and an 85-task benchmark.
49. Online Reinforcement Learning in the Met Office Unified Model through Distributed Model-Agent Coupling
Core Problem: Train stable online corrections inside the Met Office Unified Model.
Key Innovation: Distributed rank-local reinforcement-learning agents applying bounded column-aware corrections with low runtime overhead.
50. A Global Database of Soil Methane Uptake (SMUD)
Core Problem: Compile previously fragmented field measurements of a major methane sink.
Key Innovation: Open multi-timescale observations from the 1980s with moisture and temperature covariates across biomes.
51. Spatial pattern regression for meteorological fields interpolation
Core Problem: Reconstruct daily meteorological fields where observations are sparse.
Key Innovation: Spatial Pattern Regression combines past high-resolution simulations with current observations and outperforms common methods.
52. Seafloor Topography Prediction from Altimetry-Derived Gravity Data Using a Wavelet-Assisted and High-Frequency Enhancement Neural Network
Core Problem: Infer detailed seafloor topography from altimetry-derived gravity beyond linear approximations.
Key Innovation: Wavelet-assisted low-frequency retention and high-frequency enhancement, validated against shipborne soundings.
53. Occlusion-Aware Topology Refinement for Robust Road Graph Extraction from Satellite Imagery
Core Problem: Recover connected road topology from obscured satellite imagery and improve geographic generalization.
Key Innovation: Synthetic occlusion training, adaptive line extension, topology hard mining, and node-guided resampling.
54. Exploiting Projection Trajectories Discrepancy for Multipath Suppression and Detailed Feature Extraction of Buildings in SAR Adjacent Sub-Aperture Images
Core Problem: Distinguish true elevated structures, walls, and multipath artifacts from few adjacent sub-apertures.
Key Innovation: Projection-offset theory plus segmentation, wall tracking, and pixel-level offset estimation on UAV circular SAR.
55. Foundation-Model Embeddings for Land-Cover Mapping and Annual Change Detection in a Hyper-Arid Region: A Case Study of Saudi Arabia (2017–2024)
Core Problem: Evaluate foundation-model embeddings without conflating agreement to noisy source labels with independent accuracy.
Key Innovation: AlphaEarth multisensor embeddings, spatially blocked validation, design weighting, independent interpretation, and an explicitly limited change test.
56. Mountain unit extraction based on surface runoff simulation
Core Problem: Automatically delineate mountain-scale terrain objects from DEM topology.
Key Innovation: Peaks, ridges, saddles, D8 runoff paths, and tunable scale parameters yield stable hierarchical mountain partitions.
57. Three decades of Antarctic Ice Sheet surface melt dynamics with disentangled radiative and turbulent controls
Core Problem: Resolve long-term melt dynamics and physical drivers.
Key Innovation: No substantive abstract is available.
58. Climate change both increases and decreases winter snowmelt across North America
Core Problem: Determine where warming increases or decreases North American winter melt.
Key Innovation: No substantive abstract is available.
59. Reach‐Scale Analysis of Bedforms at Low Flows on the Lower Missouri River: Depth‐Scaling Behavior and Extreme Value Prediction
Core Problem: Existing depth-scaling relations do not adequately predict both reach-average bedform dimensions and their extreme variability.
Key Innovation: Uses 428 Lower Missouri River bathymetric surveys and a scale-aware lognormal model that reduces errors in variability and extreme-bedform estimates by 5–50%.
60. From Core to Field: Pattern Selection in Dissolving Fractures
Core Problem: No predictive theory has linked core-scale fracture-dissolution patterns to field-scale behavior.
Key Innovation: Combines experiments, simulations and stability analysis in a universal phase diagram spanning 0.1–30 m and derives linear scaling of the optimal injection rate with fracture length.
61. From Fires to Floods: A Participatory Case Study on Stakeholder Requirements for AI-Supported Satellite Services in Disaster Management
Core Problem: Aligning satellite-AI services with stakeholder needs in disaster management.
Key Innovation: Title indicates a participatory cross-hazard requirements case study.
62. Unraveling the hydrological dimension of ecosystem resilience: Drought-induced responses of water retention and nonlinear drivers
Core Problem: Explaining ecosystem water-retention behavior under drought.
Key Innovation: Title emphasizes nonlinear hydrological drivers of resilience.
63. Dual urban runoff response to urbanization and contradictory precipitation trends driven by climate change
Core Problem: Explaining dual runoff responses to human and climatic drivers.
Key Innovation: Title highlights contradictory precipitation trends within a coupled urbanization analysis.
64. A dual-channel framework for video-based rainfall intensity estimation using weakly supervised rain streak extraction
Core Problem: Estimating rainfall intensity from video with limited labels.
Key Innovation: Dual-channel modeling with weakly supervised rain-streak extraction.
65. Assessing future soil erosion responses to freeze–thaw cycle representation in SWAT in the seasonally frozen Songhua River watershed
Core Problem: Representing freeze-thaw effects on modeled soil erosion.
Key Innovation: Title indicates a SWAT-based future-response assessment in a seasonally frozen basin.
66. A multi-output constitutive physics-informed neural network (MOC-PINN) for seepage problems with free surface
Core Problem: Solving constitutive seepage problems with multiple outputs and a moving free surface.
Key Innovation: Multi-output constitutive PINN formulation.
67. Uncertainty-aware pseudo-static seismic pullout capacity of inclined plate anchors in clay-over-sand seabeds using finite element limit analysis and Bayesian surrogates
Core Problem: Estimating pseudo-static seismic pullout capacity in layered seabeds.
Key Innovation: Finite-element limit analysis coupled with Bayesian surrogates.
68. Land Art as a Big-Data Climate Sensor
Core Problem: Testing whether image-complexity signals around Spiral Jetty reliably track climate or Great Salt Lake elevation.
Key Innovation: Combines 1,744 co-registered image chips with interpretable texture features, deep embeddings, lag analysis, and sensor/season controls.
69. GradRepair-ODE: Certified Gradient Repair for Neural ODE Training
Core Problem: Numerical solvers can return finite but directionally wrong gradients under stiffness, chaos, or discontinuities.
Key Innovation: Checks multiple gradient candidates against finite differences and solver diagnostics, then repairs or rejects unsafe optimizer steps.
70. SomBench: Benchmark Dataset for Advancing Machine Learning in Lunar Science
Core Problem: Harmonizing heterogeneous lunar orbital observations for ML.
Key Innovation: Thirty-plus aligned layers, leakage-safe tiles, and multiple planetary-process benchmarks.
71. Multimodal-Multiresolution Foundation Model for Lunar Remote Sensing
Core Problem: Pretraining one model across lunar modalities and spatial scales.
Key Innovation: Acquisition-geometry conditioning, joint scales, and flexible modality adaptation.
72. HGSQ: Heatmap-Guided Sparse Query Detector for Real-Time Aerial Small Object Detection
Core Problem: Preserving small-object localization while avoiding computation over large aerial backgrounds.
Key Innovation: A learned heatmap jointly controls query selection, local refinement, and adaptive decoder budget at deployment.
73. ADEPTS: An auto-differentiable framework for time-dependent nonlinear thermo-chemical mantle convection inversion
Core Problem: Recovering high-dimensional initial conditions and rheological parameters through nonlinear time-dependent mantle convection.
Key Innovation: Compares unrolled and implicit gradients in a differentiable Stokes/advection framework and jointly recovers fields and physical parameters.
74. YOLO12-MambaScan: An Efficient Object Detector with High-Frequency Enhancement and State-Space Modeling
Core Problem: Preserving high-frequency small-object cues while modeling global aerial context efficiently.
Key Innovation: Combines triple-path frequency enhancement, receptive-field coordinate attention, and a Mamba global-context module.
75. A Variational Optimal Transport Operator on Incompressible Flow
Core Problem: Replacing slow per-instance optimization for incompressible density transport.
Key Innovation: Enforces incompressibility by construction and amortizes full 2D/3D trajectories with a Fourier neural operator across grids.
76. Benchmarking Optimizers to Solve Inverse Problems with Differentiable Physics Simulators
Core Problem: Selecting optimizers for difficult differentiable-simulator inverse problems.
Key Innovation: Twelve simulators spanning discrete/continuous mechanics and other physics, evaluated across first-order, approximate second-order, and global optimizers.
77. SkyAnchor: Updating Metric-scale Aerial 3D Gaussian Scenes from Unposed Ground-View Sequences
Core Problem: Update an existing aerial 3D scene using an unposed ground-view sequence without scale or pose.
Key Innovation: Geometrically verified aerial anchors, anchor-constrained submaps, and filtered ground-Gaussian insertion.
78. Should All Noises Be Treated Equally: Impact of Input Noise Variability on Neural Network Robustness
Core Problem: Understand how noise type, scale, and complexity govern neural-network generalization on field seismic data.
Key Innovation: Controlled stochastic noise generators, compound-noise curricula, and a cross-condition robustness matrix.
79. LIMODENet: Attention-Free Compact Encoders for Information-Preserving Onboard Satellite Image Restoration
Core Problem: Restore channel-degraded satellite images within neuromorphic hardware constraints.
Key Innovation: A 0.69M attention-free ODE-style encoder verified for spiking deployment.
80. Tackling Failure Modes of PINNs and PIKANs Using Conflict-Free Gradients
Core Problem: Conflicting residual, boundary, and interface gradients destabilize domain-decomposed physics-informed solvers.
Key Innovation: Norm-PCGrad plus selective separable subdomain architectures, demonstrated across PINN and PIKAN solvers.
81. LiftGCN: Efficient Energy-Preserving Graph Learning via Joukowski Spectral Lifting for Finite Element Stress Prediction
Core Problem: Preserve sharp stress concentrations and high-frequency components on irregular finite-element meshes.
Key Innovation: A Joukowski spectral-lifting recurrence with energy-preserving propagation and one sparse aggregation per layer.
82. G-ray: Ray-Level Relative Geometric Position Encoding in Multi-View Vision Transformers under Camera Heterogeneity
Core Problem: Make multi-view geometric encoding invariant to heterogeneous camera projections.
Key Innovation: Camera-local ray-angle rotary phases that improve heterogeneous-view reconstruction without learned parameters.
83. Beyond Numerical Time Series: A Unified Benchmark for Multimodal Forecasting with Heterogeneous Context
Core Problem: Evaluate whether forecasting models use heterogeneous temporal context reliably.
Key Innovation: MUSE-Bench unifies fourteen datasets, six context types, probabilistic metrics, and aligned forecast windows across model families.
84. Closed-form Bayesian homography estimation from noisy point correspondences
Core Problem: Estimate projective transforms while propagating correspondence noise.
Key Innovation: Closed-form Bayesian posterior mean with an iterative treatment of nonlinearities and posterior uncertainty.
85. Single-condition neural solvers encode transferable response spaces for parametric differential equations
Core Problem: Adapt neural PDE solvers to new parameters without global cross-condition training.
Key Innovation: Output-Jacobian response spaces, residual-minimizing linearized transfer, and active acquisition of additional local spaces.
86. Where to Compute and How to Interact: Operator-Readable Adaptation with Gauge-Aware Transport
Core Problem: Prevent mesh relocation from making node representations geometrically incompatible during aggregation.
Key Innovation: Physics-informed adaptive allocation plus geometry-conditioned gauge transport, with inspectable interventions and theoretical consistency conditions.
87. Backward SDEs-based Diffusion for Physics-Constrained Generation
Core Problem: Make pretrained diffusion priors satisfy terminal measurement or physics constraints with feasibility guarantees.
Key Innovation: Terminal-conditioned backward SDE inversion with existence/uniqueness results and a neural BSDE solver that leaves the score prior frozen.
88. SURE-Map: Self-Correcting Streaming Geometric Foundation Model
Core Problem: Accumulated geometric distortion and scale drift under limited streaming context.
Key Innovation: Cross-view geometric uncertainty plus fast/slow multi-timescale recalibration and optional loop closure.
89. From objective discovery to prediction of global ocean eco-provinces: A pathway for trustworthy learning
Core Problem: Defining meaningful eco-provinces and predicting them from satellite-observable ocean color.
Key Innovation: Explainable dense ensembles link simulation-derived provinces to modeled ocean-color fields while explicitly assessing uncertainty and input sufficiency.
90. Calibrating subgrid parametrizations of single-column ocean models via simulation-based inference
Core Problem: Inferring poorly observed turbulent-mixing coefficients with quantified uncertainty and limited simulator calls.
Key Innovation: Blockwise PCA preserves physical output structure and enables low-budget simulation-based posterior inference.
91. From Words to Wavelengths: VLMs for Few-Shot Multispectral Object Detection
Core Problem: Adapt vision-language detectors to multispectral imagery under label scarcity.
Key Innovation: Text, visual, and thermal fusion for Grounding DINO and YOLO-World with strong few-shot transfer.
92. Architecture--Optimization Co-Design for Physics-Informed Neural Networks via Layer-wise Coordinate Adaptation and Gradient Conflict Resolution
Core Problem: Reduce representation limits and gradient conflicts in PINNs.
Key Innovation: Layer-wise adaptive coordinate fusion combined with conditional gradient projection across PDE, boundary, and initial losses.
93. Noise-Adaptive Conformal Classification with Marginal Coverage
Core Problem: Maintain conformal marginal coverage when labels violate exchangeability through random noise.
Key Innovation: Noise-adaptive conformal classification with theoretical coverage and validation including BigEarthNet.
94. FAO estimates of net forest emissions and removals, 1990–2025
Core Problem: Update net forest emissions and removals through 2025.
Key Innovation: FAO country submissions cross-checked against Paris Agreement reporting and current evidence.
95. Boundary-Guided Dual-Perspective Cross-Modal Fusion Network for RGB-IR Object Detection
Core Problem: Reconcile scene-level modality importance with locally varying reliability while preserving boundaries.
Key Innovation: Sobel-guided shallow structure and dual-perspective adaptive fusion validated on three RGB-IR benchmarks.
96. Adaptive Plug-and-Play Image Restoration for Diffractive Remote Sensing with a Latent Diffusion Prior
Core Problem: Handle spatially varying blur, order aliasing, and unknown noise without large paired datasets.
Key Innovation: MAP plug-and-play restoration with a latent relationship encoder and adaptive latent diffusion prior.
97. Annual Gridded Anthropogenic CH4 Emissions Estimation in China (2019–2025) Integrating Multisource Data: SHAP-Based Driver Attribution and Spatio-Temporal Patterns
Core Problem: Improve the resolution, timeliness, and nonlinear attribution of anthropogenic methane inventories.
Key Innovation: Multisource geographic and remote-sensing data, LightGBM, SHAP, and spatial autocorrelation at 0.1 degrees.
98. Cross-Validation of Collocated ICESat-2 and CALIPSO Cloud-Aerosol Discrimination
Core Problem: Cross-validate ICESat-2 operational and CNN cloud-aerosol discrimination against CALIPSO.
Key Innovation: 9,289 global collocations with feature-scale, latitude, resolution, and sampling-strategy sensitivity analysis.
99. Weighted Strong Product Graph Laplacian Regularization for Hyperspectral Image Mixed-Noise Removal with Superpixel Segmentation
Core Problem: Preserve pure spatial and spectral edges omitted by standard product graphs.
Key Innovation: Weighted strong-product graph Laplacian regularization in a low-rank plus sparse model, tested on simulated and real HSIs.
100. Benchmarking a bounded-coordinate DeepONet for unsaturated flow and solute transport under time-varying infiltration
Core Problem: Benchmark a bounded-coordinate DeepONet for transient subsurface transport.
Key Innovation: Only the title and bibliographic information are available.
101. Understanding data sufficiency and temporal informativeness for hydrological model calibration in data scarce regions
Core Problem: Determine how much and which timing of observations are needed for hydrological calibration.
Key Innovation: No substantive abstract is available.
102. Atlantic moisture transported by the westerlies drives the expansion of the Yili River watershed in the Tian Shan, central Asia
Core Problem: Explaining the climatic driver of Yili watershed expansion.
Key Innovation: Title attributes watershed expansion to Atlantic moisture carried by westerlies.
103. Construction and assessment of evaluation indicators for the preparedness of special-needs evacuation shelters in urban areas in Taiwan
Core Problem: Building and assessing indicators for special-needs shelter readiness.
Key Innovation: Title indicates both indicator construction and empirical assessment.
104. A DEMATEL-Based Sustainability Evaluation of Post-Disaster Temporary Housing Grounded in the Geneva UN Charter
Core Problem: Structuring sustainability criteria for post-disaster accommodation.
Key Innovation: Title combines DEMATEL causal evaluation with Geneva UN Charter principles.
105. A novel method for estimating urban building heights using bi-temporal high resolution satellite imagery and reprojected nDSM
Core Problem: Estimating urban building heights from bi-temporal satellite imagery.
Key Innovation: Title combines high-resolution image change with reprojected nDSM information.
106. How does rapid stochastic injection–withdrawal govern wellbore integrity in underground energy storage: A phase-field modeling study
Core Problem: Quantifying stochastic cycling effects on wellbore integrity.
Key Innovation: Phase-field modeling of rapid injection-withdrawal sequences.
107. Supercritical CO2-induced mechanical damage and strain localization evolution in coal based on distributed optical fiber sensing
Core Problem: Tracking damage localization caused by supercritical CO2.
Key Innovation: Title indicates distributed optical-fiber sensing of evolving strain localization.
108. Role of adsorbed water in the thermo-hydro-mechanical modeling of unsaturated expansive soils
Core Problem: Representing adsorbed water in coupled expansive-soil behavior.
Key Innovation: Title identifies adsorbed-water effects in a THM formulation.
109. A derivative-fidelity failure mode in physics-informed neural networks: strengthened benchmark evidence from function-value training
Core Problem: Function-value accuracy can conceal large automatic-differentiation errors in learned derivatives.
Key Innovation: A compact benchmark protocol separately tests second-derivative fidelity across sampling densities, activations, and boundary regions.
110. Pixel-wise Planarity for High-Precision Monocular Plane Segmentation
Core Problem: Reducing geometrically inconsistent and false plane segments from a single RGB image.
Key Innovation: Combines pretrained depth/normals with a per-pixel planarity head and geometry-aware region growing.
111. Forward-Facing Near-Infrared Adds Little to Colour for Farm-Machinery Traversability: A Site-Disjoint Evaluation of Sensor-Dependent Spatial Leakage
Core Problem: Testing whether apparent NIR gains persist when agricultural traversability data are split by location.
Key Innovation: Rotated site-held-out evaluation shows spatial leakage reverses sensor conclusions.
112. RIGOR: Rig-Informed Geometry for Omnidirectional Reconstruction
Core Problem: Correcting long-trajectory inconsistency in feed-forward 3D reconstruction under weak texture, repetition, and dynamic objects.
Key Innovation: Treats each panorama as a four-view virtual rig for inconsistency repair, consensus loop closure, and Sim(3) pose-graph optimization.
113. MomentBA: Second-order Spatial Moments for Anisotropic Correspondence Uncertainty in Differentiable Bundle Adjustment
Core Problem: Bundle adjustment commonly ignores anisotropic localization ambiguity in visual matches.
Key Innovation: Derives interpretable covariance from local similarity-response moments and embeds it as information matrices in differentiable BA.
114. FFVO: A Feedforward Pose Decoder for Long-Horizon Visual Odometry
Core Problem: Efficient, temporally stable camera-pose estimation over long sequences.
Key Innovation: Compact camera tokens, hierarchical local-to-global temporal decoding, and intermediate trajectory supervision.
115. An Uncertainty-Aware Hybrid Mathematical-Machine-Learning Model for Smart Irrigation Decision Support
Core Problem: Make irrigation decisions from uncertain soil-moisture forecasts.
Key Innovation: A calibrated water-balance model corrected by a residual random forest, conformal intervals, and an interval-aware trigger rule.
116. Physically Typed and Geometry-Aware Representations for Earth Foundation Models
Core Problem: Determine whether Earth foundation models benefit from explicit physical transformation laws.
Key Innovation: A staged falsification program comparing conventional, augmentation-matched, equivariant, and Hodge/Helmholtz representations.
117. Tabby: An Open Pretraining Recipe for Time Series Foundation Models
Core Problem: Build a reusable long-context probabilistic model across time-series tasks.
Key Innovation: Open 145M-parameter recipe combining real and causal-synthetic data, deep quantile supervision, and prompt tuning.
118. Robust low-rank tensor completion via factorized weighted tensor schatten-p norm minimization
Core Problem: Recover low-rank multidimensional data under missing and corrupted observations.
Key Innovation: Weighted Schatten-p tensor factorization with SVD-free updates and rank pruning.
119. Small Object Detection in Drone Aerial Imagery with LAF-YOLOv10
Core Problem: Test whether individually promising detector modules compose effectively on UAV imagery.
Key Innovation: Multi-seed, cross-dataset, held-out, TIDE, and ablation analysis identifying a harmful P2/P5-head interaction.
120. Diagnosing Temporal Misalignment in Multichannel Time-Series Classification with Minimum Description Length
Core Problem: Identify and correct relative delays among multichannel sensor streams without labels or a trusted reference.
Key Innovation: Minimum-description-length alignment diagnostics applicable to both training and deployment data.
121. SCOUT-SLAM: Structurally-Coupled Dual Uncertainty-Aware 3DGS SLAM in the Wild
Core Problem: Break the circular dependency between tracking uncertainty and unstable 3D reconstruction.
Key Innovation: Shared-network dual uncertainty with LoRA tracking adaptation and a spatially adaptive reconstruction prior.
122. Evaluating Contextual Bias in CNN Image Classification: Evidence from Agricultural Benchmark Datasets
Core Problem: Determine whether CNN accuracy is driven by intended objects or incidental background context.
Key Innovation: Cross-dataset background-patch tests with chance, class-balance, and macro-metric controls.
123. PC^2-AD: Point Cloud Upsampling to Safeguard 3D Anomaly Detection with Resolution-constrained Edge Devices
Core Problem: Address train-test point-density mismatch in 3D anomaly detection.
Key Innovation: Geometry-aware, normality-preserving candidate generation and filtering before a frozen detector.
124. A 25-\mus/inf Event-driven Graph Neural Network Processor with Spatiotemporal Caching and Spline Convolution for Ultra-low-latency AI at the Edge
Core Problem: Run sparse event-driven graph inference at high resolution under edge memory and energy constraints.
Key Innovation: A measured EV-GNN accelerator with neighbor-parallel spline convolution and 2D/3D spatiotemporal caching.
125. When Correlations Mislead: Confounder-Aware Multi-View Urban Region Representation Learning
Core Problem: Prevent spurious shared factors from destabilizing multi-view regional embeddings.
Key Innovation: Shared-latent influence removal followed by hierarchical graph-aware residual-view fusion.
126. Robust Multi-Model Fitting through Learning Neighbor Regions
Core Problem: Robustly fit overlapping geometric models amid outliers and inefficient hypothesis sampling.
Key Innovation: Coarse learned minimum-set selection followed by neighbor-region refinement and scoring.
127. GSLAD: Prototype-Regularized Graph Structure Learning for Multivariate Time Series Anomaly Detection
Core Problem: Detect anomalies expressed first as changes in inter-variable structure across operating regimes.
Key Innovation: Condition-aware learned graphs, regime prototypes, uncertainty-normalized structural deviation, and predictive deviation.
128. What a gated sensing pipeline never looks at: bandwidth reduction and the misses behind it
Core Problem: Audit detection misses hidden by impressive sensor-bandwidth reduction ratios.
Key Innovation: Tick-level stage tracing, corrected tracker analysis, and explicit reporting of one detection versus six misses across staged flights.
129. Beyond Point Forecasts: A Survey on Probabilistic Forecasting for Time Series and Spatiotemporal Data
Core Problem: Fragmentation across uncertainty-forecasting paradigms and metrics.
Key Innovation: A pipeline-based taxonomy paired with cross-paradigm temporal and spatiotemporal experiments and domain-selection guidance.
130. A pullback-corrected scalar auxiliary variable optimizer with momentum and adaptive mobility
Core Problem: Providing curvature-aware stable optimization for composite scientific-learning objectives.
Key Innovation: Momentum and adaptive mobility are integrated with a low-rank pullback correction and an exact modified energy law.
131. When Do Learned Priors Help Visual Inertial Estimation? A Controlled Study of Prior Integration, Calibration, Initialization, and Backend Consistency
Core Problem: Separating true learned-prior value from changes in estimator setup.
Key Innovation: Matched-backend evidence layers covering local motion, trajectory, physical state, and numerical consistency.
132. Shapley Value Estimation for Multi-Site Data with Blockwise-Missing Features
Core Problem: Estimate population Shapley values when sites observe different feature blocks.
Key Innovation: Imputation-free fusion of partial sites with permutation screening for incompatible distributions.
133. Linearized PINN with pretrained nonlinear layers
Core Problem: Reduce repeated online cost and error of physics-informed neural solutions.
Key Innovation: Offline operator-compatible nonlinear bases followed by physics-constrained linear-layer fitting per instance.
134. SAMReg: SAM-enabled Image Registration with ROI-based Correspondence
Core Problem: Register images using paired semantic regions rather than dense displacement alone.
Key Innovation: SAM-derived ROI correspondence requiring no training, fine-tuning, or prompt engineering.
135. WaveHiTS: Wavelet-Enhanced Hierarchical Time Series Modeling for Wind Direction Nowcasting in Eastern Inner Mongolia
Core Problem: Forecast circular wind direction without multistep error accumulation.
Key Innovation: U-V decomposition, wavelet multiscale analysis, and hierarchical interpolation forecasting.
136. Preserving Guidance in Cost-Volume Retrieval under Extreme LiDAR Sparsity in Iterative Stereo
Core Problem: Prevent sparse LiDAR guidance from disappearing during iterative stereo cost-volume retrieval.
Key Innovation: Distinct disparity and feature prefilling strategies combined in Guided RAFT-Stereo.
137. ROVER: Robust Loop Closure Verification with Trajectory Prior in Repetitive Environments
Core Problem: Reject catastrophic false loop closures in repetitive environments.
Key Innovation: Trajectory-prior scoring of loop candidates integrated into established SLAM systems.
138. Intermittent CO2 injection enhances microstructural modification and sequestration efficiency in anthracite coal via elastic memory and threshold-limited recovery
Core Problem: Understand how intermittent injection alters anthracite and storage efficiency.
Key Innovation: Elastic memory and threshold-limited recovery are indicated by the title; no abstract is available.
139. Discussion of the paper entitled "Are there Class II rocks?"
Core Problem: Challenge potentially biased claims about whether Class II rocks exist.
Key Innovation: A reasoned comparison of axial versus lateral strain control and servo response effects.
140. Spatial distribution, connectivity, and hydrogeological significance of brittle structures in deep crystalline rock: A KURT case study for geological disposal
Core Problem: Mapping fracture distribution and hydrogeological connectivity in crystalline rock.
Key Innovation: Title indicates integrated spatial, connectivity, and hydrogeological analysis at KURT.
141. Cultural values in disaster discourse: Behavioral propositions from the Philippines
Core Problem: Formulating how cultural values shape disaster-related behavior.
Key Innovation: Title indicates behavioral propositions grounded in the Philippines.
142. Conditional diffusion model with residual decomposition generation strategy for SAR to optical image translation
Core Problem: Translating SAR observations into optical-like imagery.
Key Innovation: Conditional diffusion with residual-decomposition generation.
143. An extended subloading surface critical state constitutive model for CO2 hydrate bearing sediments
Core Problem: Modeling critical-state behavior of CO2 hydrate sediments.
Key Innovation: Extended subloading-surface formulation.
144. Real-Time Onboard Geopositioning of LEO Optical Imagery via Lightweight Algorithm and FPGA-based Heterogeneous Architecture Co-Design
Core Problem: Geolocating LEO optical imagery under onboard compute constraints.
Key Innovation: Lightweight algorithm and FPGA heterogeneous co-design.