TerraMosaic Daily Digest: August 17, 2026
Daily Summary
The August 17 corpus treats geohazards as coupled, state-dependent systems rather than as isolated trigger-response pairs. Direct studies indicate increasing rainfall-induced landslide susceptibility across inhabited parts of China under projected climate forcing, identify hidden subglacial water storage as a plausible amplifier of the 2024 La Berarde flood, and show that vertically heterogeneous landslide dams breach through seepage-contact interactions that can accelerate flood peaks. Related slope and ground-failure papers locate hazard control in groundwater redistribution, moisture cycling, lithologic layering, and chemically assisted degradation, from irrigation-driven cascades on the Heifangtai loess terrace to expansive soils, dispersive soils, karst depressions, frost-heave loading, and reservoir-bank deterioration.
Observation-focused papers move hazard assessment from static mapping toward deformation-aware surveillance. Sentinel-1 backscatter anomalies in Nepal precede documented slope failures by weeks; polygon-based inventories plus InSAR deformation sharpen susceptibility estimates in the Upper Jinsha; combined ascending-descending SBAS and PS-InSAR resolve active landslides, rainfall sensitivity, and engineering influence in the Batang reach; and UAV photogrammetry, GNSS-R soil-moisture retrieval, differentiable resistivity inversion, and bridge-deformation sensing broaden the measurable precursors of instability and damage. Seismic and cryospheric studies extend the same logic: the Palghar swarm is best explained by coupled fluid-assisted deformation and transient stress transfer, Bayesian regionalization improves ground-motion estimates in Turkey, and label-efficient crevasse mapping shows how self-supervised features can support hazard delineation where expert annotation is scarce.
A large parallel stream advances transferable Earth-observation and scientific-machine-learning methods. New data products, including global bias-corrected wave and storm-surge fields, CAMELS-KR catchment records, peatland DTMs, all-weather land-surface temperatures, and multi-product groundwater validation in Chile, strengthen the observational substrate for regional hazard analysis. At the same time, high-resolution weather and ocean forecasting, diffusion-refined data assimilation, neural operators, symbolic and interpretable surrogate methods, multimodal remote-sensing foundation models, boundary-aware segmentation, cross-view localization, calibration audits, and anomaly-monitoring schemes all aim to preserve structure, quantify uncertainty, or reduce data and compute demands. These papers expand the technical repertoire available to hazard science, but they are validated in their stated sensing, forecasting, or reconstruction domains rather than as already generalized geohazard solutions.
Key Trends
The August 17 selection converges on hazards governed by hidden hydrological or mechanical state, denser deformation sensing, infrastructure-conditioned exposure, and method papers that prioritize structural fidelity and uncertainty over domain-free claims.
- Hidden State Variables Are Reframing Hazard Mechanics: Subglacial water storage, seepage-contact erosion in layered landslide dams, irrigation-driven groundwater rise, moisture-deformation coupling in expansive soils, and temperature- or wetting-controlled degradation in limestone, loess, and dispersive soils show that poorly observed internal state variables can dominate hazard magnitude and timing.
- Deformation-Constrained Landslide Monitoring Is Maturing: Sentinel-1 anomaly baselines, polygon-based inventories with InSAR deformation, combined SBAS/PS-InSAR, and UAV or GNSS-R observations shift slope assessment from static susceptibility surfaces toward active surveillance of evolving deformation and hydrological precursors.
- Engineered Terrain Is Being Treated as a Coupled Hazard System: Road corridors, hydropower reaches, landslide dams, underground chambers, breakwaters, bridges, tunnels, reservoir banks, and post-earthquake building stocks recur as settings where natural forcing and engineered geometry interact, making hazard assessment inseparable from infrastructure configuration.
- Transferable EO Methods Emphasize Structural Fidelity and Reliability: Across optical-SAR fusion, boundary-aware segmentation, source-only change detection, calibration audits, source-aware damage review, and flood-response reasoning benchmarks, the methodological emphasis is on boundary preservation, domain shift, label efficiency, and auditable evidence rather than raw accuracy alone.
- Scientific ML for Earth Systems Is Scaling with Physics and Uncertainty: High-resolution Earth modeling through data scaling, unstructured-mesh ocean forecasting, diffusion-based super-resolved data assimilation, neural operators, transport under covariate shift, conformal monitoring, and symbolic or interpretable surrogate methods point toward faster Earth-system prediction that remains constrained by structure and uncertainty.
Selected Papers
The selected papers range from direct analyses of landslide, flood, seismic, cryospheric, erosion, and subsurface hazards to transferable advances in Earth observation and scientific machine learning. Read together, they separate papers that revise hazard mechanisms or risk estimates from papers that contribute enabling datasets, sensors, and models validated only in their stated domains.
1. CrevasseSeg: A Label-Efficient UAV Crevasse Segmentation Framework
Core Problem: Crevasse mapping from UAV imagery is data-scarce and pixel annotation is expensive for glacier experts.
Key Innovation: Provides a label-efficient UAV crevasse segmentation benchmark and shows strong performance from self-supervised plus non-linear readout pipelines.
2. Coupled Transient Processes Govern Intraplate Earthquake Swarm Evolution: Insights from the 2019-2020 Palghar Sequence
Core Problem: Fluid-only and slow-slip-only explanations do not fully explain the evolution of intraplate earthquake swarms.
Key Innovation: Uses a large ML-enhanced catalog and high-resolution relocations to show coupled fluid-assisted deformation and transient stress transfer across interacting faults.
3. Projecting changes in rainfall-induced landslide susceptibility across inhabited areas of China under climate change
Core Problem: Quantify how rainfall-induced landslide susceptibility in inhabited China changes under future climate conditions.
Key Innovation: Combines national landslide records with high-resolution precipitation projections to map future susceptibility hotspots.
4. Potential glacier contributions to the 2024 La Bérarde flood
Core Problem: Explain the unusually large June 2024 La Berarde flood beyond rain, snowmelt, and surface-lake drainage.
Key Innovation: Estimates hidden subglacial water storage as an overlooked flood-amplifying source.
5. Failure evolution and mechanisms of vertically heterogeneous landslide dams under coupled seepage and overtopping
Core Problem: Vertical heterogeneity complicates seepage-overtopping failure evolution in landslide dams.
Key Innovation: Physical model tests show how layered structures alter seepage erosion, breach timing, and peak flood discharge.
6. A Sentinel-1 SAR time-series framework for seasonal road surface vulnerability assessment and landslide precursor detection in monsoon-affected Nepal
Core Problem: Data-sparse mountain regions lack tools to flag vulnerable road segments and pre-failure signals.
Key Innovation: Uses Sentinel-1 seasonal backscatter contrasts and anomaly baselines to detect hydrological precursors 4-6 weeks before failure.
7. Landslide Susceptibility Modeling Constrained by Multi-Scale Polygon Sampling and InSAR Deformation for High-Relief Mountainous Areas: A Case Study in the Upper Jinsha River, Southwest China
Core Problem: Point-sampled static models underperform for large landslides in high-relief mountainous terrain.
Key Innovation: Combines polygon-based multi-scale sampling, a large landslide inventory, and InSAR deformation constraints in an RF-OFR framework.
8. Integrating Ascending-Descending SBAS and PS-InSAR to Monitor Landslide Deformation in the Jinsha River Batang Reach, China
Core Problem: How to monitor active canyon landslides comprehensively in steep vegetated terrain.
Key Innovation: Fuses ascending-descending SBAS and PS-InSAR to map 38 active landslides and quantify rainfall and infrastructure links.
9. Regionalization of global ground-motion models with partial non-ergodicity based on hierarchical Bayesian approach: a study in Turkey
Core Problem: Global ground-motion models miss region-specific and site-specific behavior in Turkey.
Key Innovation: Uses hierarchical Bayesian regionalization with partial non-ergodicity to improve seismic intensity prediction.
10. Irrigation partitioning triggers multi-hazard cascades mediated by rising groundwater in Heifangtai loess terrace, NW China
Core Problem: Irrigation and rising groundwater trigger interacting hazard cascades on the Heifangtai loess terrace.
Key Innovation: Frames irrigation partitioning as the driver of coupled groundwater, slope-failure, and related hazard cascades.
11. Snow avalanche susceptibility assessment using machine-learning and hybrid ensemble models in the NW Himalaya, India
Core Problem: Avalanche-prone Himalayan terrain needs robust, explainable susceptibility mapping under complex controls.
Key Innovation: Builds and validates a high-performing hybrid CNN-RF-CatBoost avalanche susceptibility framework with SHAP interpretation.
12. Frictional contact based on signed distance functions for simulating landslide-induced tsunamis by two-phase double-point material point method
Core Problem: Accurate frictional contact remains difficult in two-phase simulation of landslide-generated tsunamis.
Key Innovation: Introduces signed-distance-function contact within a two-phase double-point material point method for landslide-tsunami simulation.
13. Pushing the Limits of High-Resolution Weather Forecasting through Data Scaling
Core Problem: Overcome the lack of long high-resolution training data for global 0.1-degree machine-learning weather forecasts.
Key Innovation: Shifts resolution transfer from model fine-tuning to variable-wise super-resolution data synthesis and demonstrates power-law gains from scaling the synthetic-plus-real dataset.
14. Looks Can be Deceiving: Annotator and Reviewer Performance Across Imagery Sources in Crowd-Sourced Aerial Damage Assessment
Core Problem: Quantify annotator and reviewer disagreement across drone, crewed, and satellite building-damage imagery.
Key Innovation: Shows disagreement varies strongly by imagery source and recommends source-aware review allocation.
15. FloodReasonBench: Benchmarking VLM Reasoning Segmentation for Embodied Flood Response at the Edge
Core Problem: Benchmark reasoning segmentation for embodied flood response on resource-constrained edge platforms.
Key Innovation: A flood-specific reasoning-segmentation dataset and benchmark that jointly measures accuracy, latency, energy, and communication costs.
16. RingMo-Agent: A Unified Remote Sensing Foundation Model for Multi-Platform and Multi-Modal Reasoning
Core Problem: Existing remote-sensing vision-language models do not unify heterogeneous platforms, modalities, and reasoning tasks.
Key Innovation: Introduces a 3M-pair multi-platform RS dataset and an instruction-following model spanning optical, SAR, and infrared imagery.
17. A global dataset of bias-corrected waves and storm surge (1950-2023) with improved extremes for hazard mapping
Core Problem: Existing global wave and surge products do not represent extremes well enough for hazard mapping.
Key Innovation: Creates a bias-corrected 1950-2023 global waves and storm-surge dataset tailored to hazard applications.
18. A Geo-AI approach for rapid post-earthquake damage assessment based on UAV photogrammetry and deep learning
Core Problem: Manual post-earthquake infrastructure surveys are slow, risky, and difficult to scale.
Key Innovation: Integrates UAV photogrammetry, U-Net, and YOLOv11 to quantify crack geometry and map condition over a 55 ha site in one day.
19. Quantifying Lateral Fluvial Dynamics Using Sentinel-2: Monitoring Medium-Large Rivers in Italy
Core Problem: Operational agencies lack scalable tools to monitor channel migration and bank retreat.
Key Innovation: Uses Sentinel-2 and random forests to quantify lateral mobility, link it to drivers, and simulate future channel-infrastructure interactions.
20. Spaceborne GNSS-R Soil Moisture Retrieval over Expansive Soils Using an Attention-Enhanced Spatio-Temporal Graph Convolution Network
Core Problem: Retrieve soil moisture accurately enough to track deformation-prone expansive soils.
Key Innovation: Uses an attention-enhanced spatiotemporal graph network on GNSS-R data and links moisture cycles to measured ground deformation.
21. Damage Evolution of Limestone Under Chemically Assisted Wet-Dry Cycles: An Integrated Analysis Using Recurrence Quantification, Machine Learning and Chemical Modeling
Core Problem: Chemically assisted wet-dry cycles degrade limestone strength in reservoir fluctuation zones.
Key Innovation: Combines acoustic emission, recurrence analysis, machine learning, and geochemical modeling to separate multiscale damage modes.
22. Desertification threshold signals and their driving factors in China
Core Problem: Where desertification threshold behavior emerges in China and what drives it.
Key Innovation: Targets threshold signals and their drivers rather than only mapping desertification extent.
23. An automatic-differentiation framework for time-lapse electrical resistivity tomography inversion of hydrologic dynamics
Core Problem: Make long time-lapse ERT inversion flexible and fast enough for hydrologic monitoring workflows.
Key Innovation: Uses automatic differentiation to unify inversion components in one differentiable chain, yielding major speedups and direct water-content inversion.
24. Iterative Refinement Diffusion for Super-Resolved Data Assimilation of Multiscale Physical Systems
Core Problem: Infer high-resolution physical states from sparse low-resolution observations in multiscale systems.
Key Innovation: Builds hierarchical forecast-analysis diffusion refinement across resolutions instead of one-shot super-resolution.
25. Registration-Free Hyperspectral Reconstruction from RGB via a Permutation-Invariant Gram-Matrix Principle
Core Problem: Recover high-resolution hyperspectral images from RGB and low-resolution HSI without registration or known response functions.
Key Innovation: Uses permutation-invariant abundance Gram matching to learn RGB-to-HSI reconstruction without paired alignment.
26. Geometry-Calibrated Closed-Form Shrinkage for SAR Despeckling
Core Problem: Remove multiplicative speckle without erasing scattering structure across SAR sensors.
Key Innovation: Derives a closed-form geometry-calibrated shrinkage estimator that is training-free and sensor-agnostic.
27. TERRA: A Hierarchical Parallel Training and Memory Orchestration Framework for High-Resolution AI-based Earth Modeling
Core Problem: Memory and parallelization limits block training and long-rollout finetuning of massive high-resolution Earth models.
Key Innovation: Combines sampling-aware hierarchical parallelism with rollout-aware memory orchestration to scale Earth models to 11.4B parameters.
28. Boundary-Aligned Contribution Routing for Robust Optical--SAR Object Detection
Core Problem: Naive optical-SAR fusion can induce negative transfer when spatial or semantic correspondence is imperfect.
Key Innovation: Learns task-conditioned modality contribution routing before fusion with feature-level and semantic-level routers.
29. On the Adversarial Robustness of Remote Sensing Semantic Change Detection
Core Problem: Conventional adversarial-evaluation protocols miss coupled vulnerabilities in bitemporal semantic change detection.
Key Innovation: Defines a task-specific robustness framework separating output objectives and temporal perturbation access across the full bitemporal pathway.
30. AlignJEPA: Predictive Vision-Language Alignment for Remote Sensing Foundation Models
Core Problem: Remote-sensing foundation models remain weakly aligned with language for search and retrieval tasks.
Key Innovation: Uses a JEPA-style mask-aware predictive aligner that predicts text embeddings from multiscale visible EO tokens.
31. Hierarchical Adaptive Feature Refinement Network for VHR Remote Sensing Image Segmentation
Core Problem: Pretrained hierarchical encoders for VHR imagery are hard to adapt without losing local detail or structural guidance.
Key Innovation: Combines stage-adaptive fusion, frequency-residual refinement, and tri-prior decoding for stronger remote-sensing segmentation.
32. BASeg: Boundary-Aware Remote Sensing Segmentation with Structural Penalties
Core Problem: Remote-sensing segmentation often misses fine object boundaries and struggles across diverse urban morphologies.
Key Innovation: Introduces a structural boundary loss, a boundary-aware segmentation architecture, and a 10-city benchmark dataset.
33. ChainSpace: A Chained-Reasoning Paradigm for Spatial Intelligence
Core Problem: Spatial reasoning benchmarks treat questions independently, allowing shortcut answers without maintaining spatial state.
Key Innovation: Introduces chained multi-round spatial reasoning benchmarks and simulator-generated supervision for state-preserving spatial intelligence.
34. MITE-Net: SWaP-Optimized 4K Video Tiny Target Perception for Embodied Edge SAR
Core Problem: Edge UAVs cannot reliably detect tiny targets in 4K SAR mission imagery without sacrificing recall or power efficiency.
Key Innovation: Pairs a bio-inspired tiny-target proposal stage with an ultralight detector and standardized 4K tiny-target benchmarks.
35. OceanLight: Efficient Global Ocean Forecasting via Geometry-Adaptive Unstructured Mesh Representation
Core Problem: Structured-grid ocean AI wastes computation and struggles to represent heterogeneous ocean dynamics efficiently.
Key Innovation: Geometry-adaptive unstructured-mesh tokenization with a GNN backbone for accurate, efficient global ocean forecasting.
36. PCT-Prompt: A Prompt-Guided Transformer Framework for Dense Prediction Tasks in Point Clouds
Core Problem: Standard transformers lack suitable priors for dense point-cloud prediction in complex scenes.
Key Innovation: Prompt-guided local geometric feature branch fused with pretrained transformer features for dense prediction.
37. Convolution-Free Holistic Multivariance Decomposition Layer for Efficient Hyperspectral Image Classification Tensor Networks
Core Problem: Hyperspectral classifiers either use rigid tensor decompositions or heavy convolutions that miss complex spatio-spectral interactions.
Key Innovation: End-to-end holistic multivariance decomposition layer capturing independent and cooperative spatio-spectral variation without standard convolution.
38. Remote-Sensing City Layout Extraction with MLLM
Core Problem: Remote-sensing masks and boxes do not by themselves yield executable layouts with object identities, relations, and topology.
Key Innovation: Reframes top-down urban extraction as constrained code generation that outputs editable city graphs and regenerable 3D layouts.
39. Beyond Accuracy: Assessing Calibration of Geospatial Foundation Models and Their Sensitivity to Distribution Shifts
Core Problem: GeoFMs are usually ranked by accuracy even though operational EO deployment also depends on calibration and shift robustness.
Key Innovation: Systematically measures calibration and corruption sensitivity across 16 encoders and shows EO-pretrained models remain overconfident under shift.
40. How Sampling Strategy Affects Imbalance Mitigation in LiDAR Segmentation: A Study of Structured vs. Random Point-Based Architectures
Core Problem: It is unclear which imbalance mitigation strategies actually help point-based LiDAR segmentation under different sampling schemes.
Key Innovation: Benchmarks reweighting and loss choices across datasets and shows effectiveness depends strongly on structured versus random sampling.
41. Bridging the Gap between Labeled and Unlabeled Data via Unified Flow with Feature Memory Bank
Core Problem: Separate optimization of labeled and unlabeled remote-sensing data yields biased pseudo-labels and persistent feature gaps.
Key Innovation: Unifies teacher-guided pseudo-labeling with labeled training and aligns class features through a dynamic memory bank.
42. SplatGuide: Geometric Priors from 3D Gaussians for Pose-Free Novel View Synthesis
Core Problem: Prior unposed-view pipelines underuse the geometry, visibility, and learned features available from a single 3D Gaussian reconstruction.
Key Innovation: Reuses one 3DGS scene for geometric conditioning, occlusion-aware reference voting, and token-level feature guidance.
43. Koopman early warning signals for bifurcation and rate-induced tipping
Core Problem: Classical critical-slowing indicators miss rate-induced tipping in stochastic nonautonomous systems.
Key Innovation: Extends residual Koopman mode decomposition with control augmentation and learned embeddings to detect tipping before transitions.
44. A Deep Learning Model for Spatially Clustered Data via Differentiable Cluster Assignment
Core Problem: Model covariate-response relationships that change across unknown spatial subregions.
Key Innovation: Differentiable location-based cluster assignment with graph and occupancy penalties to learn spatial partitions and cluster-specific regressors together.
45. Convolution Smoothed Quantile Regression for XGBoost
Core Problem: Equip XGBoost with calibrated conditional quantiles and extreme-outcome probability estimates.
Key Innovation: A convolution-smoothed quantile loss that restores usable gradients and Hessians for dense CDF and exceedance estimation.
46. Eigenanalysis framework for autoregressive neural emulators of multi-scale chaotic dynamics
Core Problem: Autoregressive neural emulators of chaotic systems diverge and lack a priori stability diagnostics.
Key Innovation: Links forecast instability to Jacobian spectral radius and adds stability-promoting regularization based on that theory.
47. Snapshot Compressive Imaging under Saturation: Theory, Mask Design, and Reconstruction
Core Problem: Sensor saturation breaks the linear reconstruction assumptions of snapshot compressive imaging.
Key Innovation: Derives saturation-aware recovery theory, optimal mask-density guidance, and a plug-and-play reconstruction network.
48. PhyxMamba: Chaotic System Reconstruction from Short Context Observations with Generative State-Space Models
Core Problem: Short and discontinuous observations make faithful reconstruction of chaotic systems difficult.
Key Innovation: Combines time-delay embeddings, Mamba state-space models, and geometry-aware regularization to preserve local trajectories and global invariants.
49. Recurrent Cross-View Object Geo-Localization
Core Problem: One-shot cross-view object geo-localization is vulnerable to noise and lacks iterative correction.
Key Innovation: Recasts localization as recurrent refinement with task tokens, SAM distillation, and hierarchical reference-feature enhancement.
50. CSMoE: An Efficient Remote Sensing Foundation Model with Soft Mixture-of-Experts
Core Problem: Remote-sensing foundation models remain computationally heavy and waste training cost on redundant imagery.
Key Innovation: Adds soft mixture-of-experts and diversity-aware training-set reduction to cross-sensor masked autoencoding.
51. JAPE: Joint Anomaly Prediction and Intrinsic Explanation in Multivariate Time Series
Core Problem: Forecast-based anomaly detectors miss dependency-structure changes and usually lack native variable-level explanations.
Key Innovation: Predicts evolving dependency graphs and reuses them for joint alerting and intrinsic variable-level explanation.
52. From Canopy Phenology to Lithological Signals: Evaluating Biophysical Traits with Machine Learning in the Hațeg Basin
Core Problem: Vegetation indices blur lithologic signals in vegetated terrain.
Key Innovation: Replaces empirical indices with PROSAIL-derived traits plus machine learning for lithology discrimination.
53. An Identification Framework for Mesoscopic Shear Deformation Stages of Rock Joints Based on Integration of Multi-source Monitoring Techniques
Core Problem: Single-sensor monitoring misses staged mesoscopic-to-macroscopic shear failure in joints.
Key Innovation: Integrates DIC, infrared, and acoustic emission into a staged joint-failure identification framework.
54. Coupled density-hazard navigation for evacuation reliability in dense urban informal settlements
Core Problem: Static or density-only routing performs poorly in dense informal settlements under evolving hazards.
Key Innovation: Couples crowd density and hazard fields in a tail-sensitive dynamic navigation framework.
55. Orbital imaging spectroscopy of Earth’s mineral surface: Results and stability from EMIT’s first two years
Core Problem: Need stable, large-scale orbital mineral-surface mapping from EMIT observations.
Key Innovation: Evaluates first-two-year EMIT mineral spectroscopy performance and stability at Earth scale.
56. Experimental investigation on hydraulic erosion of soil-rock mixtures using transparent soil technology
Core Problem: Hydraulic erosion mechanisms in soil-rock mixtures are hard to observe inside the material.
Key Innovation: Uses transparent soil and 3D reconstruction to visualize internal soil-rock-mixture erosion initiation and evolution.
57. Normal-stress unclamping promotes cascading instability in deep mining-induced seismicity
Core Problem: Deep mining seismicity can cascade when normal-stress unclamping destabilizes the rock mass.
Key Innovation: Highlights unclamping as a trigger for cascading mining-induced seismic instability.
58. Multi-scale periodic characteristics of seismic activity along the Sunda subduction system inferred from singularity analysis
Core Problem: Periodic structure in Sunda subduction seismicity remains poorly characterized across adjacent systems.
Key Innovation: Combines singularity and wavelet analysis to resolve multi-scale periodicities in earthquake frequency and energy.
59. Determining the trends and drivers of extreme flow in permafrost rivers
Core Problem: Drivers and trends of extreme river flow in permafrost basins are changing under climate forcing.
Key Innovation: Targets attribution of extreme-flow trends specifically in permafrost river systems.
60. Rapid hydrologic flushing and biogeochemical recovery after Hurricane Helene in a heterogeneous coastal karst system
Core Problem: How a major hurricane flushes and resets hydrologic and biogeochemical behavior in coastal karst.
Key Innovation: Examines rapid post-Hurricane Helene recovery dynamics in a heterogeneous coastal karst system.
61. Effects of microscopic parameters on granular flow down rough inclines
Core Problem: Microscopic friction parameters controlling rough-incline granular flow behavior remain unclear.
Key Innovation: Systematically ties sliding and rolling friction to stopping height, velocity scaling, and dissipation pathways.
62. Seismic response simulation of cross-fault tunnels using synthetic across-fault ground motions based on multi-parameter permanent displacement model
Core Problem: Across-fault ground motions are too scarce for realistic tunnel seismic simulations.
Key Innovation: Synthesizes fling-step across-fault motions with a multi-parameter permanent-displacement model and validates tunnel damage patterns.
63. Impact of Heterogeneous Mantle Viscosity on the Glacial Isostatic Adjustment Correction for GRACE(-FO): Implications for Surface Mass Trends
Core Problem: Quantify how 3D mantle viscosity changes GRACE-derived surface-mass trends through glacial isostatic adjustment corrections.
Key Innovation: Builds GRACE-compatible GIA models with laterally heterogeneous mantle viscosity and shows large regional impacts on inferred mass trends.
64. Satellite Altimetry Reveals an Eddy Train Along the Ross Sea Shelf Break
Core Problem: Resolve poorly observed Ross Sea shelf-break circulation and its variability.
Key Innovation: Combines SWOT and AVISO to reveal an alternating eddy train and a phase reversal missed by reanalyses.
65. PIKFNO: An Interpretable Neural Operator Based on Physics Informed Kernel Function
Core Problem: Build an interpretable neural operator that embeds governing-equation kernels into operator learning.
Key Innovation: Constrains the trunk with physics-informed kernel functions so the operator mirrors meshless kernel expansions while preserving accuracy and generalization.
66. Frequency and Edge-Guided Segment Anything Model for Remote Sensing Image Semantic Segmentation
Core Problem: Adapt SAM to heterogeneous land-cover classes and boundary ambiguity in remote-sensing segmentation.
Key Innovation: Combines frequency-modulated adapters with multi-scale edge refinement for sharper remote-sensing segmentation.
67. SA-GEM: Scale-Adaptive and Geospatial Evidence-Modulated Token Pruning for Efficient Remote Sensing Large Vision-Language Models
Core Problem: Reduce RS-LVLM token cost while preserving task-critical geospatial evidence.
Key Innovation: Adds query-adaptive resolution routing and geospatial evidence modulation for token pruning.
68. MiNO: Cotangent-bundle propagator learning for PDEs
Core Problem: Field- or operator-learning methods struggle with discontinuities and caustics in PDE solutions.
Key Innovation: Learns the propagator itself in phase space via a microlocal neural operator and reconstructs solutions with oscillatory integrals.
69. Robust structure from motion for aerial-ground images via detector-free feature matching and multi-view track refinement
Core Problem: Large viewpoint, scale, and rotation gaps make aerial-ground feature matching brittle in incremental SfM.
Key Innovation: Combines rotation-robust detector-free matching, quadtree attention, symmetric coarse-to-fine correspondence, and multi-view track refinement.
70. UAV Video Deblurring via Motion-Aware Diffusion: A Path to Robust Target Detection
Core Problem: Motion blur from UAV maneuvers and vibration degrades downstream target detection.
Key Innovation: Introduces motion-aware diffusion with adaptive latent scaling and gated multi-frame alignment for efficient video deblurring.
71. Earth Observation Foundation Models for Terrestrial Ecohydrology: From Representation Learning to Process Inference
Core Problem: EO foundation models are not yet aligned with ecohydrological inference depth, uncertainty, and process-timescale needs.
Key Innovation: Provides a process-aware observation-to-inference framework plus meta-analysis and benchmark audit for ecohydrology-focused EOFMs.
72. FAST-DeepONet: Factor-Augmented Branch Representations for High-Dimensional PDE Inputs in the Small-Sample Regime
Core Problem: DeepONets become statistically unstable when PDE inputs are observed at many correlated sensors but training samples are few.
Key Innovation: Uses a factor-augmented branch with a fixed spectral path and regularized residual projection to stabilize operator learning.
73. Dual-Branch State-Displacement Network for Sea Surface Temperature Super-Resolution
Core Problem: Coarse satellite SST misses fine thermal structures such as ocean fronts.
Key Innovation: Combines wavelet-frequency processing, structural state-space modeling, and displacement-gated geometry-aware refinement for SST super-resolution.
74. TinyCast: Probabilistic Zero-Shot Forecasting with Computed Periodicity
Core Problem: Zero-shot probabilistic forecasters are usually too large or too learned to suit embedded deployments.
Key Innovation: Computes dominant periodicities explicitly and combines them with a tiny attention-free probabilistic forecaster.
75. Beyond Similarity Matching: Structured Reasoning for Open-Vocabulary Referring Segmentation in 3DGS
Core Problem: Similarity-only 3DGS referring segmentation fails on relations, attributes, parts, and granularity shifts.
Key Innovation: Query-conditioned Gaussian slots, relation-aware slot graphs, and granularity-adaptive mask routing.
76. LaGSplat: Inferring Physics-Governed Interactive Simulation from Monocular Video Using Latent Lagrangian Gaussian Splatting
Core Problem: Monocular video reconstructions rarely yield interactive dynamics that remain physically plausible under unseen external forces.
Key Innovation: Couples a learned dissipative Lagrangian latent state with Gaussian splatting so image-space forces map into generalized dynamics.
77. Towards Real-Time and Adaptable LiDAR Scene Completion
Core Problem: Current LiDAR completion methods are slow and sensor-specific because initialization is hand-tuned.
Key Innovation: Learns adaptive point expansion for coarse completion and efficient voxel-BEV refinement for real-time scene completion.
78. Supervising the Path to Fine Scales: GalerkinFlow for Scientific-Field and Image Super-Resolution
Core Problem: Final-output-only supervision gives little control over intermediate coarse-to-fine reconstruction dynamics in super-resolution.
Key Innovation: Supervises the full reconstruction path with intermediate residual-velocity targets, coarse-endpoint supervision, and scale-conditioned operator mixing.
79. Ultra: Unsupervised Cross-Task Optimization for Reliable Restoration Segmentation Collaboration under Adverse Weather
Core Problem: Unsupervised restoration-segmentation collaboration can propagate hallucinations under severe weather without target labels.
Key Innovation: Filters cross-task updates through candidate direction selection and causal effect estimation to stabilize collaboration.
80. PixRestore: Unified Image Restoration via Pixel Diffusion Transformer
Core Problem: Latent diffusion restoration discards fine details and can hallucinate content inconsistent with the low-quality input.
Key Innovation: Trains a VAE-free pixel-space diffusion transformer with reliability-aware feature conditioning and efficient one-step generation.
81. Characterization of Thermal Systems from Noisy and Low-resolution Measurements Using Dynamic Mode Decomposition
Core Problem: Standard dynamic mode decomposition is unstable when observations are noisy, sparse, or low resolution.
Key Innovation: Studies preprocessing and mode-truncation choices that stabilize DMD on sparse thermocouple and degraded thermal-image data.
82. A Low-Cost IoT Device for Environmental Monitoring and Embedded Solar Forecasting with On-Device Incremental Learning
Core Problem: Dense environmental sensing and local forecasting are too expensive or cloud-dependent for many distributed deployments.
Key Innovation: Builds a low-cost ESP32 monitoring node with local sensors, offline-trained neural forecasting, and on-device incremental updates.
83. On Stopping Rules and Spatial Adaptation for CART
Core Problem: Clarify how CART stopping rules affect statistical rates and spatial adaptation.
Key Innovation: Proof that minimum-impurity-decrease stopping can achieve near-minimax spatial adaptation, unlike minimum leaf size.
84. Coded Hankel Polynomial Chaos: Spectral Identification of Dominant Polynomial-Chaos Modes
Core Problem: Identifying dominant polynomial-chaos modes is difficult with sampled multivariate dictionaries.
Key Innovation: Recasts dominant-mode discovery as a spectral Hankel problem that recovers sparse polynomial-chaos structure without full design matrices.
85. When Single-Dataset Conclusions Fail: A 45-Task Study of Threshold Tuning and Resampling for Imbalanced Classification
Core Problem: Single-dataset conclusions about threshold tuning and resampling are unreliable for imbalanced classification.
Key Innovation: Runs a leakage-free 45-task benchmark showing when threshold tuning and SMOTE help or fail across imbalance regimes.
86. LiD-GLM: Lipschitz-constrained Deep Generalized Linear Models
Core Problem: Hybrid GLM-neural models usually sacrifice either interpretability or flexibility.
Key Innovation: Uses Lipschitz-constrained invertible residual networks that preserve stochastic monotonicity while quantifying deviation from classical GLMs.
87. OccamView: Object-Conditioned View Selection for Frame-Budgeted Active 3D Gaussian Reconstruction
Core Problem: Geometry-only next-best-view planning misses partially occluded objects when frame budgets are tight.
Key Innovation: Adds object-conditioned hidden-region proxies and Geo-Floor reranking to active 3D Gaussian reconstruction.
88. STAG-VIO: Stabilized Prompt-to-Geometry Interface for Robust Dynamic Visual--Inertial Odometry
Core Problem: Semantic-assisted VIO degrades when prompts and segmentation masks flicker under occlusion in dynamic scenes.
Key Innovation: Stabilizes prompt generation with uncertainty-adaptive tracking and geometry-oriented mask refinement for dynamic VIO.
89. WATCH: Adaptive Monitoring for AI Deployments via Weighted-Conformal Martingales
Core Problem: Post-deployment AI monitoring lacks adaptive anytime-valid methods that detect and diagnose harmful distribution shifts.
Key Innovation: Generalizes conformal test martingales into weighted adaptive martingales that detect changepoints and distinguish concept shifts from extreme covariate shifts.
90. VisDom: Sparse Novel View Synthesis with Visible Domain Constraint
Core Problem: Sparse-view NeRF and Gaussian-splatting methods overfit and create floating artifacts because geometry is underconstrained.
Key Innovation: Adds a visible-domain geometric constraint that requires multi-view visibility beyond silhouette consistency, with no learned parameters.
91. Multilook Coherent Imaging: Theoretical Guarantees and Algorithms
Core Problem: Speckle noise and weak theory limit likelihood-based multilook coherent imaging.
Key Innovation: Derives a first MSE bound for deep-image-prior MLE and improves PGD with Newton-Schulz inverses and bagging.
92. Development of a Labeled Dataset for Convection Initiation Events over China’s Central and Eastern Mainland During the Warm Season
Core Problem: AI nowcasting lacks long-term labeled datasets for convection initiation over China.
Key Innovation: Builds a quality-controlled CI dataset with a forward-search and backward-verification labeling strategy and multiple lead times.
93. Source-Only Cross-Dataset Building Change Detection with Frozen DINOv3 and Hierarchical Evidence Fusion
Core Problem: Source-only cross-dataset building change detection breaks under strong domain shifts.
Key Innovation: Uses frozen DINOv3 features, lightweight adapters, and hierarchical evidence fusion for target-free transfer.
94. Investigation of mechanical properties of saturated loess soils near an artificial lake in Xi'an, China and its evaluating method
Core Problem: Saturated loess near artificial lakes shows highly variable in-situ strength and deformation behavior.
Key Innovation: Introduces a composite physical index linking void ratio and liquid index to field mechanical parameters.
95. Terracing and ridge planting enhances soil organic carbon stability on water-eroded farmland slopes in Northeast China
Core Problem: Water-eroded farmland slopes lose soil carbon stability under conventional cultivation.
Key Innovation: Shows terracing and ridge planting improve soil organic carbon stability on erosion-prone slopes.
96. A Mixture-of-Experts framework for AMSR2 passive microwave snow depth retrieval over the Northern Hemisphere
Core Problem: Passive microwave snow-depth retrieval remains inaccurate across Northern Hemisphere snow regimes.
Key Innovation: Applies a Mixture-of-Experts framework to improve AMSR2 snow-depth retrieval.
97. Numerical modeling of particle-laden flows interacting with flexible structures: A coupled SPH-DEM-FEM approach
Core Problem: Coupled behavior of particle-laden flows and flexible structures is hard to simulate consistently.
Key Innovation: Couples SPH, DEM, and FEM to model flow-structure interaction in particle-laden flows.
98. A Roadmap for Identifying and Interpreting Physical Processes and National Water Model Prediction Bias Associated With Baseflow Index Regimes Across the Contiguous United States
Core Problem: Relate baseflow-index regimes to National Water Model skill and watershed controls across the contiguous United States.
Key Innovation: Clusters monthly baseflow signatures and uses explainable random forests to diagnose regime-specific model biases.
99. Learning Discrete Riemannian Metrics for Physical Fields with Cochain-Frame Equivarianc
Core Problem: Learn geometry-dependent propagation on meshes without violating topological conservation structure.
Key Innovation: Introduces Riemannian Hodge Message Passing with learned symmetric positive-definite cochain metrics and exact complex identities.
100. Efficient Neural-Network-Based High-Resolution Radiative Transfer for CO___ Retrieval, and Application to Interferometric Sensing
Core Problem: Speed up high-resolution radiative-transfer calculations for spaceborne CO2 and CH4 retrieval.
Key Innovation: Trains a neural surrogate that preserves both radiances and Jacobians, then couples it to an interferometric sensing model.
101. A Novel Fourier Feature Network for Solving Partial Differential Equations
Core Problem: Improve neural approximation of PDE solutions over standard single-layer ELM baselines.
Key Innovation: Replaces standard hidden activations with Fourier feature bases and tunes a global scaling factor for higher-accuracy solutions.
102. Generative Learning of Separatrices
Core Problem: Reconstruct under-sampled separatrix manifolds in high-dimensional multistable dynamical systems.
Key Innovation: Pairs basin-classification uncertainty with score-based generative sampling to trace separatrices.
103. ER-KANs: Efficient and Robust Kolmogorov-Arnold Networks for Data-Scarce Scientific Machine Learning
Core Problem: Make efficient KAN architectures robust under noisy, data-scarce scientific learning.
Key Innovation: Combines shared Gaussian RBF bases, curriculum noise injection, and entropy-weighted regularization in ER-KAN.
104. PathFinder: Joint Decompositions of Linked Multimodal Datasets
Core Problem: Jointly factor linked multimodal datasets when not all matrices share the same dimensions.
Key Innovation: Introduces path-based global joint decomposition across partially shared matrix relationships.
105. ProjFormer: Point Cloud Completion via Geometric-Projective Transformer and Cross-Modal Semantic Constraints
Core Problem: Complete sparse partial point clouds with consistent 2D-3D fusion.
Key Innovation: Uses deterministic projection-guided multi-view attention and geometry-aware point-wise routing.
106. VGGT-Align: Bridging Local Reconstruction and Global Consistency for Long-Sequence 3D Reconstruction
Core Problem: Chunk-based 3D reconstruction accumulates unconstrained scale drift over long sequences.
Key Innovation: Adds scene geometric invariant anchoring and lightweight test-time adaptation to suppress inter-chunk scale drift.
107. Beyond Field Accuracy: Two-Axis Diagnosis of Inverse-PINN Parameter Error
Core Problem: Inverse PINNs can fit observed fields well while returning the wrong physical parameter.
Key Innovation: Separates observation-resolution limits from residual-landscape parameter preference through a two-axis post-training diagnosis.
108. HistReNeRF: Historic Image Relocalisation within Contemporary Neural Radiance Field Reconstructions
Core Problem: Historic photographs are hard to localize in modern scene models because appearance and layout shift across time.
Key Innovation: Matches adapted DINO patch features to rays sampled from a contemporary NeRF for historic-photo 6-DoF relocalization.
109. QSMP: finding representative time series subsequences through Quick Shift+Matrix Profile
Core Problem: Long time series need scalable discovery of representative waveforms for summarization and downstream analysis.
Key Innovation: Combines Quick Shift with Matrix Profile to cluster subsequences by density at lower space cost.
110. CrossView: Can Vision-Language Models Reason Across Cameras?
Core Problem: Existing multimodal models are not tested on cross-view reasoning across simultaneous cameras.
Key Innovation: Introduces CrossView, a benchmark spanning multiple domains to measure evidence integration across viewpoints.
111. Catching Hallucinated Citations in Video-LLM Question Answering: A Self-Verification Pipeline and Verifier Ablation Study
Core Problem: Video-LLMs emit confident timestamped claims that may not be supported by the cited frame.
Key Innovation: Replaces sycophantic self-checking with frame recaptioning plus an NLI verifier to catch fabricated grounded claims.
112. Generalised Transportability via Causal Abstractions
Core Problem: Classical transportability handles one causal query at a time and gives little guidance when exact transfer fails.
Key Innovation: Recasts transport as model-level causal abstraction and derives certified approximate transport intervals.
113. Counterfactual Sensitivity Is Not Repairability: Auditing Replay Probes for Video Evidence
Core Problem: Retrieved video evidence in tool-using agents may not actually influence the final answer.
Key Innovation: CARVE measures answer sensitivity to matched destroy versus sham replays to audit grounding dependence.
114. What You Ask is What You Ground: Bridging Question Intent to Temporal Evidence for Grounded VideoQA
Core Problem: Grounded VideoQA models often localize nearly the same temporal segment regardless of the question intent.
Key Innovation: Conditions video features on question intent before localization and couples answer selection with grounded temporal evidence.
115. YOLO26-RD: An End-to-End Road Damage Detection Network With Learnable Contrast Enhancement and Edge-Guided Downsampling
Core Problem: Road-damage detectors overemphasize small-object design despite survey annotations being mostly large instances.
Key Innovation: Reallocates anchor budget and adds learnable contrast enhancement plus edge-guided downsampling for road-distress detection.
116. Learning Auditable Classifier Models: Source-Disjoint Tree Ensembles
Core Problem: Accurate tree ensembles are hard to audit because overlapping rules and sequential boosting obscure per-prediction logic.
Key Innovation: Learns source-disjoint shallow trees and fits sparse global coefficients so each prediction decomposes into non-overlapping rule contributions.
117. FirstDiff: One-Step Diffusion-Based Anomaly Detection for Multivariate Time Series via Initial Noise Prediction
Core Problem: Diffusion-based anomaly detection is expensive and underuses informative intermediate denoising signals.
Key Innovation: Shows the first reverse-step noise prediction is sufficient for strong anomaly detection with only one denoising evaluation.
118. RoofGS: Roofline-Guided End-to-End Acceleration of 3D Gaussian Splatting
Core Problem: Different stages of 3D Gaussian rendering are bottlenecked by different GPU limits, so generic optimization leaves performance untapped.
Key Innovation: Applies roofline-guided stage-specific optimizations including compressed depth keys and fast bounded exponential approximation.
119. PWLR: Pairwise Witness Local Rejection for Boundary-Aware Out-of-Distribution Detection
Core Problem: Near-OOD samples close to in-distribution class boundaries evade many image OOD detectors.
Key Innovation: Uses pairwise local visual witnesses between confusing class pairs and combines them with global scores for rejection.
120. Feasible and Novel Synthetic Population Generation with Tabular and Sequential Travel Attributes
Core Problem: Survey-based synthetic populations miss valid combinations and generate infeasible travel behaviors.
Key Innovation: Uses a regularized two-stage generator for tabular demographics plus sequential trip chains with novelty-aware evaluation.
121. Beyond Visual CoT: Internalized Visual Thinking for Proactive Video Reasoning
Core Problem: Visual chain-of-thought improves proactive video reasoning but is too costly because it generates intermediate images at inference.
Key Innovation: Internalizes visual prediction during training so future-frame reasoning improves without explicit image generation at inference.
122. Deploying Frontier Agentic Technology in MOOSEnger, a Multiphysics-Capable AI Assistant
Core Problem: Specialized multiphysics frameworks are hard for non-experts to use effectively in realistic simulation workflows.
Key Innovation: Builds a tool-using agent that retrieves framework knowledge, validates executable inputs, and stores lessons in persistent memory.
123. Depth-guided Multi-view Exposure Bracketing for HDR Robot Vision
Core Problem: Single-shot HDR perception remains unreliable in multi-sensor systems under extreme lighting.
Key Innovation: Depth-guided multi-view exposure bracketing with confidence-aware fusion plus a dedicated HDR perception dataset.
124. AsyTO: Asymmetric Temporal Operator for Parameter-Efficient Multivariate Time Series Forecasting
Core Problem: Multivariate forecasting trades off shared efficiency against variable-specific temporal dynamics.
Key Innovation: Asymmetric temporal operator factorizing history-reading and future-writing modes with periodic prototypes.
125. OceanDepths: A Global Dataset of Paired Subsurface and Surface Ocean Observations
Core Problem: Surface and subsurface ocean observations are not jointly available in a high-resolution AI-ready global dataset.
Key Innovation: Pairs satellite sea-surface fields with co-located subsurface profiles and reanalysis at weekly 0.1 degree global scale.
126. Synthetic Data Augmentation for Satellite-Based Analysis of Battle-Damaged Agricultural Fields in Ukraine
Core Problem: Labeled satellite imagery of battle-damaged agricultural fields is scarce and class-imbalanced.
Key Innovation: Shows that balanced diffusion-based synthetic augmentation materially improves real-image classification performance.
127. Reference-free logged energy-oracle recovery for neural approximations of symmetric coercive variational problems: conforming Riesz reconstruction and archive-level selection
Core Problem: Neural PDE checkpoint archives lack a reference-free way to identify the model with the lowest true energy error.
Key Innovation: Derives a computable conforming Riesz monitor with oracle-recovery guarantees for archive-level model selection.
128. Variational Outlier-Robust Gaussian Process Regression with Generative Modeling
Core Problem: Standard Gaussian-process regression is brittle to contaminated observations and outliers.
Key Innovation: Introduces observation-specific contamination in a generative GPR and fits it with variational generalized EM.
129. X²Localizer: Cross-grained Alignment for Progressive Cross-view Video Geo-localization
Core Problem: Conventional cross-view video geo-localization assumes fixed-length inputs and handles partial or interrupted observations poorly.
Key Innovation: Jointly aligns temporal prefixes and frame-to-tile evidence and adds sliding-window re-localization for progressive deployment.
130. Turning spectra into images improves plant trait retrieval with 2D-CNNs
Core Problem: One-dimensional spectral models underuse long-range inter-band dependencies in reflectance data.
Key Innovation: Reshapes spectra into 2D images and shows simple 2D CNNs outperform 1D baselines, with interpretable band attribution.
131. Binarized High-Efficiency RAW Video Restoration and Beyond
Core Problem: Binary neural networks struggle to preserve temporal coherence and activation fidelity in RAW video restoration.
Key Innovation: Introduces a binarized information interaction module and distribution-aware binarized convolutions to make lightweight RAW video restoration practical.
132. SpotlessGS: Relightable 3D Gaussian Splatting under Dynamic Illumination for Robotic Perception
Core Problem: Uneven onboard lighting breaks 2D enhancement approaches and harms geometrically consistent perception in dark scenes.
Key Innovation: Jointly optimizes lighting, spherical-harmonic illumination, and BRDF within Gaussian Splatting for relightable reconstruction.
133. MotionGS-SLAM: Event-Modulated Gaussian Splatting for Motion-Blur Robust SLAM
Core Problem: Prevent visual SLAM failure under severe motion blur.
Key Innovation: Event-modulated Gaussian splatting that models blur formation directly with spatial and temporal motion-aware modulation.
134. Scale-Consistent Posterior Dynamics for Diffusion Inverse Problems
Core Problem: Make posterior sampling with pretrained diffusion priors tractable for inverse problems.
Key Innovation: A scale-consistent surrogate posterior SDE with Langevin correction and IMEX discretization that preserves posterior structure.
135. Identifying parameter couplings and uncertainties of mixed-noise stochastic systems via full-covariance Gaussian mixture network
Core Problem: Infer coupled parameters and uncertainties in partially observed stochastic systems with mixed noise.
Key Innovation: A full-covariance Gaussian-mixture network that learns likelihood structure, parameter couplings, and non-identifiability.
136. Inferential Evaluation of Surrogate-Derived Models under Covariate Shift
Core Problem: Estimate target-population model performance when gold labels are scarce and only surrogate labels are available.
Key Innovation: Cross-fitted transport estimators with density-ratio and outcome-regression correction for ROC and AUC inference under shift.
137. SEER: Long-Context Reasoning via Selective Visual-Text Compression
Core Problem: Long-context reasoning is expensive, while uniform visual compression loses query-relevant detail.
Key Innovation: Learns query-relevant image selection and localized text retrieval through supervised tool-interaction trajectories.
138. Structured Prediction for Scalable Spreadsheet Table Understanding: From Cell Types to Table Ranges (Extended Version)
Core Problem: Heterogeneous spreadsheet layouts make cell typing and table detection unreliable at scale.
Key Innovation: Pairs CRF-LightGBM cell classification with a deterministic five-stage table-range extraction pipeline.
139. Asymptotics-guided learning and symbolic regression for dispersive resonances
Core Problem: Asymptotic resonance approximations remain inaccurate away from leading-order behavior.
Key Innovation: Learns theory-guided residual corrections and compresses them into compact symbolic formulas.
140. Improved Regret Analysis for Parallel Gaussian Process Bandit Optimization
Core Problem: Parallel Gaussian-process bandits either suffer batch-size regret penalties or require wasteful initial uncertainty sampling.
Key Innovation: Proves strong regret bounds for batched Thompson sampling without the usual upfront uncertainty-sampling phase.
141. Density-Reweighted Entropic Optimal Transport: Decoupling Geometry from Sampling Density
Core Problem: Entropic optimal transport confounds geometry with sampling-density differences.
Key Innovation: Introduces density-reweighted entropic OT that explicitly discounts sampling-density effects while preserving geometric correspondences.
142. Hide&Seek: Learning to Explain in an End-to-End Differentiable Network
Core Problem: Selector-predictor explanation models suffer information leakage and limited differentiability.
Key Innovation: Makes feature removal differentiable through feature replacement and trains selection and prediction jointly with parsimony annealing.
143. AutoSR: Automatic Symbolic Regression by Searching Research States
Core Problem: Equation search from noisy finite data loses the investigative record needed to judge scientific credibility.
Key Innovation: Searches persistent research states with proposer-reviewer agents and progressive-widening Monte Carlo tree search.
144. Prompt Engineering in Segment Anything Model: Methodologies, Applications, and Emerging Challenges
Core Problem: Prompt engineering for Segment Anything and its variants has grown rapidly without a systematic synthesis.
Key Innovation: Builds a taxonomy of geometric, textual, and multimodal prompting methods across applications including remote sensing.
145. HTTM: Head-wise Temporal Token Merging for Faster VGGT
Core Problem: Global attention makes joint multi-view geometry reconstruction slow on long scene sequences.
Key Innovation: Merges 3D tokens head-wise using spatial locality and temporal correspondence, preserving head diversity while cutting inference cost.
146. ODTlearn: A Package for Learning Optimal Decision Trees for Prediction and Prescription
Core Problem: Optimal decision-tree methods for prediction and prescription remain difficult to access and extend in practice.
Key Innovation: Packages mixed-integer optimization methods for optimal, fair, prescriptive, and distributionally robust trees in Python.
147. Projection-based multifidelity linear regression for data-scarce applications
Core Problem: High-dimensional outputs are hard to model when only a few high-fidelity samples exist.
Key Innovation: Combines projection-based output reduction with weighted low/high-fidelity data augmentation and automatic proximity weighting.
148. Hybrid numerical-AI modeling of directional wave spectra on an idealized one-dimensional nearshore slope
Core Problem: Directional wave spectra on nearshore slopes are expensive to resolve accurately.
Key Innovation: Combines numerical simulation and AI to model directional wave spectra over an idealized slope.
149. CAMELS-KR: Catchment attributes, meteorology, and reconstructed streamflow for large-sample hydrology in South Korea
Core Problem: South Korea lacks a national CAMELS-style dataset for comparative catchment hydrology.
Key Innovation: Builds the first nationwide collection of catchment attributes, meteorology, and reconstructed streamflow for Korea.
150. CDF-DETR: Cross-Stage Attention and Dual-Scale Feature Calibration for Small-Object Detection in UAV Remote Sensing Imagery
Core Problem: Small objects in UAV imagery are hard to detect accurately under dense clutter and resource limits.
Key Innovation: Adds a partial-attention backbone, dual-scale calibration, and improved localization loss to RT-DETR.
151. Experimental investigation on the influence factors of permeability reduction effect in granite fracture using EICP technology
Core Problem: Which EICP parameters best reduce flow through fractured granite.
Key Innovation: Systematically links aperture, reactant concentration, and flow rate to CaCO3 distribution and sealing efficiency.
152. Effects of Joint Roughness and Orientation on Hydrodynamic Pressure Characteristics in Plunge Pool Bedrock Joints
Core Problem: Joint roughness and orientation complicate pressure transmission driving plunge-pool bedrock scour.
Key Innovation: Quantifies how joint geometry alters dynamic pressure and resonance within bedrock joints.
153. A novel method for generating optimal DTM in tropical peatlands using GEDI and ICESat-2 observations
Core Problem: Generating accurate DTMs in tropical peatlands is difficult from conventional remote observations.
Key Innovation: Combines GEDI and ICESat-2 to derive an improved peatland bare-earth surface.
154. Real-time deformation monitoring of long-span bridges by coupling computer vision with inertial measurement
Core Problem: Real-time bridge deformation monitoring needs robust support beyond vision alone.
Key Innovation: Couples computer vision with inertial sensing for real-time structural deformation tracking.
155. Estimation of hourly All-Weather land surface temperature in an area with frequent clouds based on satellite passive microwave remote sensing and data assimilation
Core Problem: Frequent clouds prevent continuous hourly land-surface temperature retrieval.
Key Innovation: Fuses passive microwave remote sensing and data assimilation for hourly all-weather LST.
156. Stratigraphic evidence of polygenetic doline evolution and epikarst processes in the Central Yucatán Peninsula, Mexico
Core Problem: How stratigraphy records polygenetic doline development and epikarst evolution in central Yucatan.
Key Innovation: Uses stratigraphic evidence to reconstruct multi-stage doline and epikarst development.
157. Mobilization and distribution of adfreeze stresses on steel piles under frost heave conditions: a laboratory study
Core Problem: Adfreeze stress mobilization on steel piles under frost heave is poorly constrained.
Key Innovation: Laboratory study resolves how adfreeze stresses develop and distribute under frost-heave loading.
158. Study on tunnel response under the coupled action of 160 km/h high-speed metro vibration and seismic loading in the Guangdong-Hong Kong-Macao Greater Bay Area
Core Problem: High-speed metro vibration and earthquakes jointly affect tunnel response in the Greater Bay Area.
Key Innovation: Evaluates coupled train-vibration and seismic loading effects on tunnel behavior.
159. Water inrush mechanisms and critical prevention layer thickness for underground chambers in coral reef limestone stratum
Core Problem: Underground chambers in coral reef limestone face poorly constrained water-inrush risk.
Key Innovation: Identifies inrush mechanisms and estimates critical prevention-layer thickness.
160. Validation of GRACE, ERA5-Land and MODIS remote sensing products for groundwater monitoring in semiarid regions: A multi-basin study in Chile’s Coquimbo region
Core Problem: Remote-sensing groundwater indicators need basin-scale validation in semiarid Chile.
Key Innovation: Cross-validates GRACE, ERA5-Land, and MODIS signals with trend and uncertainty analysis for groundwater monitoring.
161. Interaction between dispersive soil and liquid water: From temperature-ponding driven water-salt dynamics to water content-mediated microstructural evolution
Core Problem: Temperature and ponding reshape water-salt transport and microstructure in dispersive soils.
Key Innovation: Links macro water-salt dynamics to microstructural evolution in dispersive soil-water interaction.
162. Seismic vulnerability and mitigation for mound-type breakwaters on liquefiable soils: Insights from 1g shaking table tests, FE modeling, and fragility analysis
Core Problem: Mound-type breakwaters on liquefiable soils need quantified seismic vulnerability and mitigation options.
Key Innovation: Combines shaking-table tests, FE modeling, and fragility analysis for liquefaction-prone breakwaters.