TerraMosaic Daily Digest: August 18, 2026
Daily Summary
The August 18 corpus treats geohazards as products of concealed mechanical, hydraulic, and tectonic state rather than simple trigger counts. Seismological studies map crustal attenuation and shallow velocity structure to active faults, permeability barriers, and fluid pathways, while the Xiaojiang watershed analysis links strike-slip-driven drainage reorganization to long-term concentration of landslides and debris flows. Slope and ground-failure papers extend the same logic across extreme-rainfall subgrade collapse, vegetation-mediated infiltration, static liquefaction in dams, microcrack-driven rock degradation, bedding-controlled shear fracture, cryogenic sandstone behavior, and MICP-modified loess hydraulics, showing that instability depends on evolving pore pressure, anisotropy, and material condition.
Event-scale hazard papers sharpen quantitative thresholds and operational prediction. The Purepu glacial lake outburst flood reconstruction derives discharge, erosion-deposition budgets, and hydrodynamic thresholds for bank erosion and slope failure; coastal studies show that overwash forecasts depend materially on exceedance choice and wave contribution; flood-generation attribution across U.S. catchments is sensitive to soil-layer representation; and compound heatwave-extreme precipitation sequences are becoming more frequent and intense. Parallel monitoring advances include low-cost resonance tracking of an unstable coastal rock block, ground-based passive microwave detection of eruptive activity under cloud and ash, coastal-camera wave estimation, cloud-robust flood segmentation from SAR-synthetic NDWI fusion, FY-4A-based heavy-rainfall prediction, and uncertainty-aware post-earthquake component damage assessment.
A large parallel stream expands the technical base for hazard science without collapsing disciplinary boundaries. Marine-flood metamodels, coastal ENSO and satellite-SST forecasting, runoff and sediment-regime learning, sea-clutter wave inversion, sea-level and Antarctic mass-balance analyses, microclimate and evapotranspiration datasets, and aerosol-cloud observing syntheses broaden the environmental context in which hazards are interpreted. Transferable remote-sensing and scientific-machine-learning papers emphasize open-vocabulary segmentation, structural optical-SAR change detection, geometry-preserving depth and 3D reconstruction, geospatial foundation-model augmentation, confidence-aware test-time adaptation, causal and conformal forecasting, physics-informed surrogates, geostatistics-informed diffusion, and coupled thermoporomechanical solvers; these are enabling methods in their validated sensing or modeling domains, not demonstrated geohazard solutions by default.
Key Trends
The August 18 selection converges on five trajectories: state-aware failure diagnosis, thresholded hydro-mechanical reconstruction, resilient sensing under observational limits, infrastructure-coupled hazard analysis, and transferable GeoAI that treats structure, uncertainty, and physical constraint as first-order design requirements.
- Hidden State Variables Are Central to Failure Diagnosis: Across earthquake, landslide, rock, and dam studies, the governing controls are increasingly framed as poorly observed internal state variables: fluid pathways and permeability structure in the Nankai Trough, pore-pressure and stress-path evolution in static liquefaction, hydraulic gradients and groundwater level in rainfall-driven slopes, and anisotropy or unfrozen water content in rock and loess materials. Hazard interpretation is moving away from bulk forcing alone toward state-aware diagnosis.
- Thresholded Hydro-Mechanical Event Reconstruction Is Strengthening: Direct hazard papers increasingly recover explicit process thresholds or sensitivity regimes rather than only retrospective narratives. The Purepu GLOF study estimates discharge, erosion, deposition, and bank-failure thresholds; coastal inundation work isolates the influence of exceedance-value choice and wave setup; catchment-scale flood attribution tests soil-layer-depth assumptions; and runoff-sediment studies distinguish diagnostic from predictive regime formulations.
- Operational Sensing Is Expanding Into Obscured or Sparse-Data Conditions: Several studies target settings where conventional observation fails or is too costly: passive microwave radiometry for volcanoes under cloud and ash, force-balance IoT resonance monitoring for unstable rock blocks, SAR-derived synthetic NDWI for overcast flooding, FY-4A neighborhood rainfall prediction, coastal-camera wave retrieval, pre-seismic ionospheric anomaly networking, and multimodal earthquake-damage fusion. The common direction is resilient sensing under practical observational constraints.
- Infrastructure and Engineered Terrain Are Being Modeled as Coupled Hazard Systems: Hazard is repeatedly assessed through the interaction of natural forcing with engineered geometry and exposure. Himalayan highway rockfall protection links slope characterization to trajectory demand and embankment performance; tunnel-aboveground systems in liquefiable soils are analyzed in fully coupled 3D response; hurricane debris volume is inferred for recovery operations; and structural plan conversion, bridge crack evaluation, and building-height estimation all support consequence-aware infrastructure assessment rather than detached hazard mapping.
- Transferable GeoAI Prioritizes Structure, Uncertainty, and Physical Constraint: The method-heavy portion of the corpus favors models that preserve geometry, respect domain shift, or expose uncertainty: optical-SAR change detection with structural context, geometry-preserving monocular depth, hierarchical 3D reconstruction, open-vocabulary remote-sensing segmentation, confidence-aware pansharpening adaptation, conformal multivariate forecasting, causal inference for dynamical systems, physics-composed generative models, PINN and surrogate formulations, and geostatistically constrained diffusion. These advances enlarge the toolkit for geohazard science but are validated in their stated tasks, not yet across hazards at large.
Selected Papers
The selected papers combine direct geohazard studies with adjacent Earth-system observations and transferable computational methods. Earthquake, slope, flood, coastal, volcanic, and liquefaction papers contribute the clearest hazard findings, while the broader remote-sensing and scientific-ML set expands the observational and modeling repertoire available for future hazard applications without itself constituting geohazard validation.
1. Quantifying Crustal Seismic Attenuation Across Continental China: A Comprehensive Coda-Wave Tomography From Ambient Noise Correlations
Core Problem: Continental China lacks a high-resolution attenuation model for interpreting crustal heterogeneity and hazard patterns.
Key Innovation: Uses ambient-noise coda-wave tomography to build the first frequency-dependent continental Qc model linked to tectonic blocks and active faults.
2. Detailed Shallow Shear Velocity Structure in the Nankai Trough From Ambient Noise Differential Adjoint Tomography: Implications for Earthquake Generation Mechanisms
Core Problem: How shallow velocity and fluid-pathway structure above the plate boundary controls seismic and aseismic slip.
Key Innovation: Ambient-noise differential adjoint tomography links high-Vs barriers and low-Vs high-Vp/Vs zones to pore-pressure conditions and earthquake style.
3. Low-Cost Ambient-Vibration Monitoring of an Unstable Coastal Rock Block: Identification of the Fundamental Resonance of Kounopetra (Kefalonia, Greece) with a Force-Balance IoT Node
Core Problem: Whether a low-cost IoT accelerometer node can reliably identify resonance in an unstable coastal rock block.
Key Innovation: Demonstrates a Raspberry Pi-based force-balance node can isolate the block's fundamental resonance baseline for affordable dense monitoring.
4. The Suitability of a Remote Microwave Radiometer for Detecting Volcanic Activity
Core Problem: How to monitor eruptive thermal activity when infrared methods fail under clouds and dense ash.
Key Innovation: Demonstrates the first ground-based passive microwave radiometer for continuous volcano monitoring at Mount Etna.
5. Effects of storm surge exceedance value on overwash and inundation forecasts
Core Problem: How exceedance-value choices affect probabilistic forecasts of storm overwash and inundation.
Key Innovation: Validates forecast skill against post-storm observations and quantifies how exceedance thresholds alter inundation and overwash predictions.
6. Rockfall protection for himalayan highway slopes: Integrated hazard assessment and performance evaluation of a slender reinforced embankment
Core Problem: How to connect geological instability, impact demand, and protection-structure performance for Himalayan highway rockfall risk.
Key Innovation: Builds an integrated workflow from field characterization to hazard ranking to dynamic FE evaluation of a slender reinforced embankment.
7. Numerical analysis of subgrade slope stability in catchment area under extreme rainfall
Core Problem: Quantify how runoff scour, seepage, groundwater level, and slope geometry jointly control subgrade-slope failure under extreme rainfall.
Key Innovation: Combines hydrologic catchment evolution, toe-erosion mechanics, and limit-analysis stability modeling to resolve coupled runoff-seepage landslide triggering.
8. Hydrodynamic controls on geomorphic responses to the 2025 Purepu glacial lake outburst flood in a Himalayan gorge
Core Problem: Explain how hydrodynamics controlled geomorphic impacts of the 2025 Purepu GLOF in a Himalayan gorge.
Key Innovation: Title indicates an event-focused analysis linking flood hydraulics to downstream geomorphic response after a major recent GLOF.
9. Drainage reorganization driven by strike-slip faulting along the Xiaojiang Fault, Southeastern Tibet: A new explanation for geohazards in the Xiaojiang River watershed
Core Problem: Explain why the Xiaojiang watershed is exceptionally prone to landslides and debris flows under strike-slip fault control.
Key Innovation: Links fault-driven divide migration and river capture to enhanced erosion, landsliding, and debris-flow concentration in a classic hazard basin.
10. Assessing the static liquefaction potential in earth dams using the bifurcation domain concept in numerical workflows
Core Problem: Conventional strength or factor-of-safety methods do not identify the onset and spatial evolution of static liquefaction in weakly compacted dam materials.
Key Innovation: Introduces an instability-line-based liquefaction instability susceptibility index and probabilistic FLAC3D workflow to map bifurcation-domain liquefaction potential.
11. Rapid Debris-Volume Estimation from Post-Hurricane Aerial Imagery
Core Problem: Post-hurricane debris volume is still estimated with slow or badly biased methods despite rapid aerial surveys.
Key Innovation: Combines debris segmentation and monocular height estimation from single-pass aerial RGB imagery to produce spatially explicit debris-volume estimates with uncertainty.
12. To Remove or Not to Remove Clouds: A Comparative Analysis and Fusion of Raw SAR and Synthetic NDWI for Overcast Water Segmentation
Core Problem: Segmenting water under complete cloud cover when optical imagery is unavailable.
Key Innovation: Demonstrates synthetic NDWI from SAR outperforms raw SAR alone and then fuses both for best overcast water mapping.
13. MAGPIE-Net: Predicting short-duration heavy-rainfall events in station neighborhoods from multitemporal FY-4A AGRI observations
Core Problem: Issuing short-lead warnings for neighborhood-scale heavy-rainfall events directly from satellite observations.
Key Innovation: Embeds a differentiable adaptive grid-to-station mapping so event losses supervise the full satellite-to-station pipeline.
14. Evaluating and improving crop-yield forecasting methods during extreme drought
Core Problem: Crop-yield forecasters break when drought-year meteorology falls outside historical training distributions.
Key Innovation: Head-to-head ML versus deep-learning evaluation under drought shift with sample weighting and feature selection.
15. Exploring Efficient Open-Vocabulary Segmentation in the Remote Sensing
Core Problem: Open-vocabulary remote-sensing segmentation lacks a benchmark and struggles with domain gap.
Key Innovation: Builds an OVRSIS benchmark and proposes a remote-sensing-specific open-vocabulary segmentation framework.
16. A multi-view contrastive learning framework for spatial embeddings in risk modelling
Core Problem: Spatial climate, weather, and demographic data are hard to fuse into predictive risk models.
Key Innovation: Learns coordinate-driven multi-view spatial embeddings from satellite imagery and OpenStreetMap features.
17. Nearshore wave height estimation using coastal cameras: comparing physics-informed spectral and conventional image representations
Core Problem: Estimating nearshore wave height from coastal imagery requires robust physically meaningful representations.
Key Innovation: Compares physics-informed spectral and conventional image representations for camera-based wave-height estimation.
18. Investigating metamodeling capability to predict sea levels and marine flooding maps for early-warning systems: application on the Arcachon Lagoon (France)
Core Problem: Overcome the computational burden of predicting sea levels and marine flooding maps fast enough for warning systems.
Key Innovation: Compares three data-driven metamodeling approaches built on precalculated simulation databases for rapid flooding prediction.
19. Global escalation of more frequent and intense compound heatwave-extreme precipitation events
Core Problem: Determine how often and how intensely heatwaves followed by extreme precipitation are escalating globally.
Key Innovation: Frames heatwave-extreme precipitation sequences as a distinct compound hazard requiring integrated warning and adaptation systems.
20. Study on the infiltration mechanism of extreme rainfall under typical vegetation coverage in the three gorges reservoir area
Core Problem: How vegetation cover changes infiltration, pore pressure, erosion, and stress under extreme rainfall in reservoir slopes.
Key Innovation: Uses soil-column experiments and an extended Green-Ampt model to quantify vegetation interception and root-uptake effects on infiltration.
21. ROC-based sensitivity analysis of predisposing factors and landslide susceptibility zonation of Khazawl district, Mizoram
Core Problem: Which conditioning factors most strongly control landslide susceptibility in Khazawl district and how to optimize zonation models.
Key Innovation: Uses ROC-based sensitivity analysis to prune factor sets and improve FR and WOE susceptibility mapping.
22. Dynamic multimodal posterior network: Trustworthy post-earthquake damage assessment for structural components by fusing heterogeneous data
Core Problem: Infer structural-component damage from heterogeneous, partially observed post-earthquake sensing data.
Key Innovation: Fuses crack imagery and distributed strain within an uncertainty-aware multimodal posterior network that explicitly scores informativeness and modality confidence.
23. Three-dimensional seismic response and stone-column mitigation mechanisms of a shield tunnel-aboveground structure system in liquefiable soils
Core Problem: Assess how a tunnel-aboveground structure system responds seismically in liquefiable soils and how stone columns mitigate that response.
Key Innovation: Title indicates a coupled 3D seismic-liquefaction mitigation analysis rather than isolated tunnel or ground response.
24. Flood-generating mechanism attribution across U.S. catchments: the role of soil-layer depth and framework sensitivity
Core Problem: Attribute dominant flood-generating mechanisms across U.S. catchments while testing sensitivity to soil-layer depth assumptions.
Key Innovation: Title indicates a continental attribution framework linking flood generation to subsurface representation choices.
25. Forecasting Coastal ENSO Warming in the Niño 1+2 Region Using ConvLSTM: Toward Improved Early Warning in Peru and Ecuador
Core Problem: Operational guidance for Niño 1+2 coastal ENSO warming remains weaker than for broader Pacific indices.
Key Innovation: Builds a ConvLSTM ensemble with uncertainty estimates that gives region-focused SST anomaly forecasts up to six months ahead.
26. MoRAX: Mobility-based Representation Augmentation for Geospatial Foundation Models
Core Problem: Injecting human functional structure into geospatial foundation model embeddings without losing transferability.
Key Innovation: Mobility-conditioned teacher-student augmentation that adds urban functional connectivity to EO-derived embeddings.
27. GS-Voxel: Fitting-Free Structured Latents for Large-Scale 3DGS Generation
Core Problem: Pre-optimized 3D Gaussian splat scenes lack structured latents for scalable aerial generation.
Key Innovation: Fitting-free voxelized latent conversion for image-conditioned large-area aerial 3DGS synthesis.
28. Composing Flow-Matching Energies with Known Physics: Generation, OOD Detection, and Inversion on PDE Fields
Core Problem: Learned PDE-field priors rarely compose cleanly with governing-physics constraints at inference.
Key Innovation: Flow-matching energies that enable physics-composed generation, OOD detection, and inversion.
29. Initialization-Free Bundle Adjustment Revisited: A Controlled Experimental Study
Core Problem: Low object-space error does not guarantee valid metric reconstruction in initialization-free bundle adjustment.
Key Innovation: Controlled benchmark exposing the optimization-reconstruction gap in InitFree BA.
30. WONDER: A Radio World Model-based Negotiation Framework for Multi-Agent UAV Coverage Optimization
Core Problem: Local observations and hidden radio fields hinder distributed UAV coverage restoration after disasters.
Key Innovation: JEPA radio world model plus multi-round negotiation in ray-traced metropolitan scenes.
31. Spatially explicit feature importance for building height estimation using research-access high-resolution SAR and optical sensors
Core Problem: City-scale footprint-level building heights are hard to recover where LiDAR and commercial imagery are scarce.
Key Innovation: Geographically weighted random forest fusing TerraSAR-X, PlanetScope, and Sentinel-1 with local feature importance.
32. Scientific Machine Learning of Chaotic Systems Learns Reduced-Order Equations for Neural Populations
Core Problem: Recovering interpretable governing equations from noisy chaotic observations is difficult.
Key Innovation: Combines prediction-error methods with universal differential equations to recover reduced-order dynamics from noisy data.
33. Neural Operator-Based Nonlinear Nudging for Chaotic Dynamical Systems
Core Problem: Effective nudging terms are hard to design for nonlinear state-space systems.
Key Innovation: Learns nonlinear nudging operators with neural networks and tests them on chaotic benchmark systems.
34. A Network-Based Framework for Characterizing Pre-Seismic Ionospheric Disturbances Using the TEC Anomaly Significance Index
Core Problem: How to characterize pre-seismic TEC disturbances across a GNSS network more robustly than peak anomalies alone.
Key Innovation: Introduces the TASI index to summarize persistence, variability, and entropy of TEC anomalies before an earthquake.
35. A Case Study on the Triggering and Maintenance Mechanisms of Dual Squall Lines over North China Within a Cold Vortex Environment
Core Problem: Why two squall lines in the same cold-vortex event formed and evolved differently over North China.
Key Innovation: Disentangles the distinct mesoscale triggers and maintenance mechanisms controlling each squall line segment.
36. Joint learning of frequency-spatio-temporal dependencies and structural context for optical-SAR change detection
Core Problem: Detect changes more robustly across optical-SAR time series by jointly modeling frequency, spatio-temporal, and structural context.
Key Innovation: A multimodal change-detection architecture with obvious downstream use for rupture, landslide, and damage mapping.
37. A regime-based framework for linking runoff initiation and event-scale sediment export in a small temperate agricultural loess catchment
Core Problem: Understand how uncertainty in runoff initiation propagates into event-scale sediment-yield prediction in a loess catchment.
Key Innovation: Separates diagnostic and predictive value of runoff-regime formulations when coupled to sediment-export modeling.
38. Detecting and Discriminating Operator Misspecification in Hybrid PDE-Parameter Learning: a Reference-Free Instrument, with Discrimination Bounded In Sample
Core Problem: How to tell operator misspecification apart from parameter non-identifiability in hybrid PDE-parameter learning without an oracle.
Key Innovation: Introduces a reference-free instrument that detects misspecified operators from a single fit while staying quiet on merely non-identifiable designs.
39. PXDepth: Pixel-Space Modeling for Structure Preserving Monocular Depth Estimation
Core Problem: Why zero-shot monocular depth models often lose fine structures and object boundaries.
Key Innovation: Separates global context from pixel-space depth prediction to preserve high-resolution geometry and sharp boundaries.
40. Structural Plan-to-Model Conversion with Deterministic Geometry and Guarded Agentic Vision-Language Refinement
Core Problem: Turning framing-plan PDFs into editable structural models is slow and error-prone.
Key Innovation: Combines deterministic geometry extraction with guarded agentic vision-language correction for plan-to-model conversion without detector fine-tuning.
41. GeoWeaver: Accurate Long-Sequence 3D Reconstruction via Hierarchical Geometric Assembly
Core Problem: Maintaining globally consistent geometry and camera motion over long RGB video sequences.
Key Innovation: Chunk-wise geometric priors with test-time Sim(3) alignment and coarse-to-fine global refinement.
42. Revisiting Sea-level and global water budgets for the period 1993-2022 within ESA Climate change initiative
Core Problem: Reconcile global sea-level and water budgets from 1993-2022 and explain the post-2015 gap.
Key Innovation: Revisits the ESA climate-change-initiative budget closure and isolates an unresolved recent mismatch needing better deep-ocean and gravity constraints.
43. Multiyear tropical warm pool warming drives slowdown in Antarctic mass loss
Core Problem: Explain why Antarctic mass loss temporarily slowed during recent years.
Key Innovation: Attributes the slowdown to a recurring tropical warm-pool teleconnection rather than a durable global-warming-driven shift.
44. Imputation for geochemistry: log-transform analysis of highly incomplete compositional data
Core Problem: How to cluster and impute highly incomplete compositional geochemical data robustly.
Key Innovation: Defines a Gaussian-mixture model in log-ratio space with MCMC-within-EM to recover missing values and below-detection-limit distributions.
45. Geo-Consistent Centralized Multi-UAV Gaussian SLAM for Incremental Orthophoto Generation
Core Problem: How to generate online geo-consistent orthophotos from multiple UAVs efficiently and at scale.
Key Innovation: Combines GNSS-assisted multi-UAV Gaussian SLAM with collaborative densification and adaptive pruning for incremental orthophoto updates.
46. MS-SSTNet: A Scale-Aware Spatiotemporal Learning Framework for Satellite SST Forecasting via Iterative Multiscale Decomposition and Dual-Window Modelling
Core Problem: How to improve multi-day sea-surface-temperature forecasts across interacting spatial and temporal scales.
Key Innovation: Uses iterative multiscale decomposition and dual-window temporal modeling to separate persistent and transient SST dynamics.
47. Wave Parameter Inversion Based on Sea Clutter Characteristics
Core Problem: How to infer wave direction, period, and height from non-scanning radar sea-clutter data.
Key Innovation: Combines ResNet and Transformer components to map sea clutter, radar, and wind inputs to multiple wave parameters.
48. Recent Progress in Remote Sensing of Clouds and Precipitation Physics: Platforms, Applications, and Emerging Frontiers
Core Problem: How remote sensing of cloud and precipitation physics is progressing across platforms and retrieval methods.
Key Innovation: Synthesizes multi-platform observation systems, active-passive retrieval advances, and remaining uncertainty bottlenecks.
49. A Physics-Informed Neural Network for Pressurized Cavities of Arbitrary Shape in Heterogeneous Rock
Core Problem: Predict displacement and stress around arbitrarily shaped pressurized cavities in heterogeneous rock with discontinuities.
Key Innovation: Extends PINNs with signed-distance embeddings and domain-conditioned activations to handle piecewise-smooth fields across rock interfaces.
50. Microcrack-Driven Degradation in Rock Compaction Deformation and Damage Under Different Environmental Effects
Core Problem: Unify how freeze-thaw, wet-dry, chemical, and thermal effects degrade rock compaction behavior through microcrack evolution.
Key Innovation: Combines DEM microcrack tracking with a crack-volume-strain damage model to compare multiple environmental degradation pathways in one framework.
51. The sediment sources availability index: Application of a new semi-quantitative index to assess the potential connectivity of sediment sources at the catchment scale
Core Problem: Assess catchment-scale availability and connectivity of sediment sources using a new semi-quantitative index.
Key Innovation: Introduces a screening index for sediment-source connectivity that could support erosion and debris-flow source appraisal.
52. Solid waste geomorphology: A new classification of anthropogenic gravitational processes
Core Problem: Classify anthropogenic gravitational processes associated with solid-waste landforms.
Key Innovation: Title suggests a new geomorphologic classification for gravity-driven waste-landform instability.
53. Measuring the spatial explicitness of GeoAI and geospatial foundation models: A systematic review and future roadmap
Core Problem: Review how spatially explicit current GeoAI and geospatial foundation models actually are.
Key Innovation: Offers a roadmap for spatially aware GeoAI, which is directly relevant to improving hazard-mapping model design.
54. El Niño-Southern oscillation and changes in agricultural practices as drivers of sediment flux in the Pirapó River, Southern Brazil
Core Problem: Attribute multi-decadal sediment-flux changes to ENSO variability versus conservation agriculture in a river basin.
Key Innovation: Combines in-situ and Landsat-based sediment records to isolate climate and land-management controls on sediment export.
55. SuPerPoV: Score and Evolution of the Stratospheric Polar Vortex via Persistent Homology
Core Problem: Existing polar-vortex event definitions rely on unstable empirical thresholds and poorly capture event evolution.
Key Innovation: Introduces persistent-homology scores that classify and track vortex displacement and splitting without fixed thresholds.
56. A Hierarchical Deep Learning Framework for Runoff Prediction Using Raster-Based Spatial Representations
Core Problem: How much explicit raster-based spatial encoding improves runoff prediction and transfer to ungauged basins.
Key Innovation: Systematically compares averaged, fixed-grid, and adaptive multi-patch spatial encodings in a hierarchical deep-learning framework.
57. RoBell-RVFL: A Robust Generalized Bell Random Vector Functional Link Network
Core Problem: How to make randomized neural networks robust to imbalance, outliers, and noisy labels.
Key Innovation: Adds quality-aware sample weighting with generalized-bell membership functions that preserve minority signals and suppress noisy majority samples.
58. Certified but Private: Scalable Zero-Knowledge Proofs for Neural Network Guarantees
Core Problem: How to prove robustness and fairness guarantees of neural networks without revealing proprietary weights.
Key Innovation: Builds scalable zero-knowledge proofs for CROWN-style neural-network certifications on million-parameter models.
59. Dynamic Regime-Aware Conformal Calibration for Reliable Economic Forecast Intervals under Multiple Distribution Shifts
Core Problem: How to maintain calibrated prediction intervals under covariate shift, concept drift, local heterogeneity, and latent regimes.
Key Innovation: Combines weighted conformal calibration with regime-aware weighting and an online significance controller.
60. From Abductive Explanations to Global Logical Rules for Node Classification in SGCs
Core Problem: How to derive compact global logical rules from node-classification GNNs without redundant subgraph detail.
Key Innovation: Uses minimal abductive explanations as an intermediate representation for rule extraction from SGC networks.
61. OV3D-Bench: A Diagnostic Benchmark for Open-Vocabulary Monocular 3D Detection
Core Problem: Open-vocabulary monocular 3D detectors are hard to compare and often rely on unrealistic deployment oracles.
Key Innovation: Creates a deployment-realistic benchmark that decouples geometry, semantics, and cross-domain transfer across seven datasets.
62. Causal Discovery in Equal Variance Linear Gaussian DAGs via SURE-Tuned Ridge Regression
Core Problem: How to recover equal-variance Gaussian DAGs in sample-limited and compute-limited regimes.
Key Innovation: Uses parallel SURE-tuned ridge regressions plus adaptive thresholding for fast non-iterative DAG recovery.
63. SCENARIODIFF: A Scenario-level Guidance Framework for Multimodal Time Series Forecasting--Extended Version
Core Problem: How to use noisy textual context in time-series forecasting without asking LLMs to directly predict numbers.
Key Innovation: Builds a three-level scenario-and-anchor guidance pipeline conditioning a multimodal diffusion forecaster.
64. UniQuery4R: Unified 4D Scene Reconstruction from a Single Query
Core Problem: How to jointly reconstruct correspondence, depth, camera motion, and scene flow without repeated pairwise encoding.
Key Innovation: Encodes a multi-frame clip once and answers arbitrary source-target geometry and flow queries through query-conditioned decoding.
65. Rethinking Irregular Time Series Forecasting from the Perspective of Basis Functions
Core Problem: Predefined basis-function models for irregular series suffer from sampling-density bias and limited adaptability.
Key Innovation: Introduces debiased neural basis functions with importance sampling, multi-scale decomposition, and mass-aware fusion.
66. Beyond MSE: Rethinking the Evaluation Metric and Benchmarking for Irregular Time Series Forecasting
Core Problem: MSE gives biased assessments of irregular time-series forecasts because sampling distributions differ across samples.
Key Innovation: Defines continuous-time squared error with importance weighting and validates it on a broad irregular-forecasting benchmark.
67. Causal Local States: Scalable Simultaneous Causal Network Inference and Forecasting for Dynamical Systems
Core Problem: Inferring interaction networks while still producing accurate forecasts in heterogeneous dynamical systems.
Key Innovation: Per-node local neighborhood selection that jointly supports approximate causal discovery and prediction.
68. Picard Proximal Monte Carlo for Parallel Bayesian Imaging with Score-Based Generative Priors
Core Problem: Scaling posterior sampling for high-dimensional imaging inverse problems with learned generative priors.
Key Innovation: Picard-iterated, proximal Langevin framework that exposes parallelism across time steps and GPUs.
69. Monitoring Pasture Restoration from Satellite Image Time Series: Caveats and Opportunities
Core Problem: Detecting restoration outcomes from noisy Sentinel-2 time series without confounding year effects.
Key Innovation: Shows intra-year modeling and per-site normalization help, then audits temporal bias and label imbalance explicitly.
70. Evaluation of AI-based Visual Crack Detection in Steel Bridges Using Probability of Detection
Core Problem: Evaluating crack detection for steel bridge inspection in a way engineers can use for reliability analysis.
Key Innovation: Probability-of-detection evaluation framework that quantifies uncertainty and image-resolution effects for AI crack inspection.
71. GenRec: Knowing Where to Reconstruct and Where to Generate
Core Problem: Separating reconstructable from generative pixels in sparse-view scene synthesis.
Key Innovation: Observation-mask-gated architecture and supervision that explicitly split reconstruction from hallucination.
72. Understanding the Surprising Generalization Properties of Tabular Foundation Models
Core Problem: Why and how tabular foundation models generalize across downstream tasks is still unclear.
Key Innovation: Single-table self-supervised pretraining plus a task-centric retrieval view of in-context tabular learning.
73. SFMformer: A Spatial-Frequency Modulation Transformer for Lightweight Image Super-Resolution
Core Problem: Sparse-attention super-resolution loses quality at both token selection and feature aggregation stages.
Key Innovation: Spatial pre-enhancement plus wavelet-domain modulation placed around sparse attention.
74. TabNSM: Neural Sparse Mixer for Tabular Regression
Core Problem: Deep tabular regression struggles with noisy high-dimensional feature interactions.
Key Innovation: Sparse interaction mixer with GridLoss supervision and difficulty-aware sampling.
75. Information fusion and machine learning for sensitivity analysis using physics knowledge and experimental data
Core Problem: Sensitivity estimates degrade when both physics-based and data-driven models carry uncertainty.
Key Innovation: Physics-informed DNN and GP fusion with uncertainty-aware Sobol analysis.
76. Adaptive surrogate modeling for high-dimensional spatio-temporal output
Core Problem: Very high-dimensional spatiotemporal outputs are hard to surrogate accurately and adaptively.
Key Innovation: Latent-space surrogate with adaptive sampling driven by original-space prediction error.
77. SPACE: Sample-cloud Predictive Adaptive Conformal Ellipsoids for Multivariate Time-Series Forecasting
Core Problem: Sample-based multivariate forecasts lack calibrated joint coverage under distribution shift.
Key Innovation: Forecast-sample covariance geometry combined with adaptive conformal window selection.
78. Exact Reformulation and Optimization for Direct Metric Optimization in Binary Imbalanced Classification
Core Problem: Precision, recall, and F1 are hard to optimize exactly in binary imbalanced classification.
Key Innovation: Provides exact reformulations and penalty-based optimization for direct metric optimization.
79. TabCausal: Pretraining Across Causal Environments for Tabular Causal Discovery
Core Problem: Causal discovery foundation models do not generalize well across diverse causal environments.
Key Innovation: Pretrains a causal-discovery model across varied graph, mechanism, noise, and intervention regimes.
80. Wave power enhancement of an oscillating water column device near seawalls through gap-induced resonance
Core Problem: OWC performance near partially reflecting seawalls depends on coupled hydrodynamic parameters.
Key Innovation: Analyzes how seawall reflection, slope, and gap-induced resonance affect energy extraction and wave run-up.
81. CzechGrids Micro: microclimate temperature offset grids at 10 m resolution for the Czech Republic
Core Problem: Represent local microclimate temperature offsets across the Czech Republic at 10 m resolution.
Key Innovation: Builds generalized additive models linking station data, topography, and forest cover to monthly microclimate offsets.
82. The CMIP6-downscaled CORDEX-Southeast Asia (SEA) ensemble: evaluation and benchmarking for megacities of SEA
Core Problem: Evaluate and benchmark CMIP6-downscaled CORDEX Southeast Asia simulations for megacities.
Key Innovation: Applies a standardized benchmarking framework to identify high-performing regional climate models for finer-scale downscaling.
83. Perspiration vapor lightens near-skin air, but hinders human evaporative cooling in arid heat
Core Problem: Why common thermoregulation models underpredict body temperature in hot, dry, stagnant air.
Key Innovation: Reveals a dueling-buoyancy mechanism from perspiration vapor and provides compact physics-informed heat-transfer models.
84. Large-scale outdoor 3D trajectory measurements enabled by drone-assisted multicamera tracking
Core Problem: How to obtain accurate large-volume outdoor 3D trajectories without cumbersome field calibration.
Key Innovation: Introduces drone-assisted multicamera calibration tied to RTK-GPS for portable georeferenced 3D tracking.
85. Atmospheric Remote Sensing Based on Satellite Oxygen-Band Observations: A Review
Core Problem: How satellite oxygen-band observations constrain atmospheric path length, clouds, aerosols, and trace-gas retrievals.
Key Innovation: Synthesizes spectroscopy, radiative-transfer theory, algorithms, and applications across O2 and O4 satellite observations.
86. Confidence-Aware Selective Test-Time Adaptation for Remote-Sensing Pansharpening
Core Problem: How to adapt pansharpening models to unseen sensors without target labels or full retraining.
Key Innovation: Uses confidence-aware selective test-time adaptation that updates only sensor-sensitive parameters under uncertainty guidance.
87. Spatial-Temporal Distribution and Microphysical Characteristics of Aerosols and Clouds over China: A Combined Satellite and Aircraft Observation Study
Core Problem: How aerosol and cloud properties vary across China and what aircraft observations add to satellite patterns.
Key Innovation: Integrates MODIS and aircraft data to characterize regional aerosol-cloud microphysics and support model parameterization.
88. True Mode III Fracture Toughness and Bedding-Induced Anisotropy of Layered Sandstone Using ANBD Brazilian Disks
Core Problem: Measure true Mode III fracture toughness and bedding-controlled anisotropy in layered sandstone.
Key Innovation: Uses ANBD tests plus DIC, AE, and surface scanning to isolate non-opening Mode III fracture and quantify strong bedding effects.
89. Thermo-mechanical behavior of medium sandstone under cryogenic conditions: Role of unfrozen water content
Core Problem: Understand how unfrozen water content controls sandstone thermo-mechanical behavior under cryogenic conditions.
Key Innovation: Title suggests a mechanistic cold-region rock study relevant to frost-weathering and frozen-ground instability.
90. Phenology and snow penman evapotranspiration model: development, validation, and products
Core Problem: Improve ET estimation by explicitly representing phenology, snow cover, and CO2-driven stomatal inhibition.
Key Innovation: Adds snow and phenology processes to a remote-sensing Penman framework, producing a long-term China ET dataset.
91. Geostatistics-informed latent diffusion for conditional hydrogeological facies simulation under sparse hard-data constraints
Core Problem: Simulate hydrogeological facies conditionally under sparse hard-data constraints.
Key Innovation: Uses latent diffusion informed by geostatistics to improve facies simulation where subsurface observations are scarce.
92. Unveiling anisotropic elasticity of granular materials under true triaxial conditions
Core Problem: Directional elastic constants of granular materials under true triaxial stress states are hard to characterize because inherent and stress-induced anisotropy interact.
Key Innovation: Combines DEM probe tests with an orthotropic hyperelastic formulation and new micromechanical tensors linking elastic anisotropy to granular fabric.
93. FFT-based solver for upscaling the thermoporomechanical behavior of infinitely contrasted porous materials
Core Problem: Periodic homogenization of porous materials with infinitely contrasted properties is difficult across coupled thermal, hydraulic, and mechanical physics.
Key Innovation: Develops an FFT-based Adaptive Eyre-Milton solver that robustly computes fully coupled thermoporomechanical operators in one framework.
94. Hydraulic property improvement of MICP-treated loess: Macroscopic performance and microscopic mechanism
Core Problem: Loess hydraulic properties need improvement, and the micro-mechanisms behind MICP treatment effects are not fully resolved.
Key Innovation: Links macroscopic hydraulic-property improvement in MICP-treated loess to underlying microscopic mechanisms.