TerraMosaic Daily Digest: July 27, 2026
Daily Summary
Unstable ground is increasingly resolved through explicit process attribution rather than pattern description alone. Seasonal subsidence in coastal Louisiana is tied to poroelastic aquifer response, earthquake triggering is reformulated through mechanism-resolved non-normal Hawkes cascades, and the Joshimath crisis is explained by linking mapped deep-seated landslide geometry, InSAR kinematics, toe erosion, and pore-pressure forcing. Rainfall-induced slope failure, seepage-driven breakdown of climbing-type landslide dams, and liquefaction susceptibility are treated in the same spirit: the main advance is not just prediction, but sharper identification of what physically controls failure.
Hazard-ready observational baselines are also becoming richer in time, scale, and field evidence. The Collazzone multi-temporal inventory, the high-resolution Kaikoura submarine landslide inventory, and the multi-lake Himalayan GLOF assessment expand the empirical footing for hazard analysis. Long geodetic records are used more diagnostically: three decades of DInSAR at Campi Flegrei are decomposed into stable and residual deformation modes, and deep learning separates candidate volcanic anomalies in Cheonji caldera lake levels from dominant hydro-meteorological variability. Better hazard inference follows from clearer separation of signal, mechanism, and uncertainty.
Method development places greater weight on operational reliability. Bayesian inversion exposes uncertainty in landslide-dam breach prediction, leakage-free validation is enforced for levee and photogrammetric workflows, and prior-matched reporting is proposed for rare-event Earth-observation screening. Transferable geospatial AI now spans tool-using reasoning over ultra-high-resolution imagery, text-conditioned forecasting, cross-view localization under adverse conditions, synthetic LiDAR priors for motion estimation, and physics-aware surrogates for flood, levee, and excavation response. The emphasis is shifting from benchmark-only gains toward explicit tests of interpretability, deployment constraints, and decision relevance.
Key Trends
Process-resolving experiments, multi-temporal inventories and uncertainty-aware sensing connect ground instability, seismicity and volcanic unrest to operational hazard assessment.
- Mechanism First, Not Just Map First: Hazard analysis is moving beyond delineation to identify the controlling physics of deformation and failure. Poroelastic aquifer forcing, non-normal earthquake triggering, seepage-structured dam failure, rainfall-driven strength loss, and deep-seated slope kinematics are resolved with models that connect observations to mechanism rather than treating instability as a black-box classification task.
- Hazard Baselines Are Becoming More Temporal and More Field-Constrained: Inventories and assessments are increasingly built from multi-date imagery, pre- and post-event bathymetry, field bathymetry, geophysics, and repeated validation rather than single static snapshots. That shift is visible in the Collazzone landslide archive, the Kaikoura submarine inventory, and the Eastern Himalaya glacial-lake assessment, and it materially improves what can be said about timing, geometry, and intervention priority.
- Uncertainty Is Treated as a Design Variable: Probabilistic dam-breach modeling, Bayesian hydraulic reconstruction, soft-label pulse classification, and reliability-aware evaluation protocols all treat uncertainty as part of the result rather than a postscript. This is especially important for geohazard workflows, where overconfident screening and under-specified parameter choices can propagate directly into emergency planning errors.
- Operational Geospatial AI Is Being Stress-Tested for Reality: Geospatial AI evaluation increasingly targets deployment friction: skewed class priors, domain shift, degraded imagery, sparse labels, limited hardware, and extremely large visual context. Progress is measured less by isolated score gains than by stability under conditions approaching real hazard-monitoring pipelines.
- Remote Sensing Is Expanding from Detection to Structured Reasoning: Remote-sensing models increasingly embed retrieval, cross-view alignment, tool use, temporal conditioning, and multimodal reasoning. Earth observation is therefore expanding from image interpretation toward integrated evidence synthesis for complex, multi-source geohazard assessment.
Selected Papers
Geohazard research is converging on mechanism, observability, and operational discipline: unstable terrain and damaged assets are not merely located, but interpreted through separated physical drivers, quantified uncertainty, and richer empirical baselines.
1. Remote sensing–based analysis of accelerated surface movements in Joshimath Town, Chamoli, India: first insights into deep-seated gravitational slope processes
Core Problem: Joshimath's accelerating ground deformation needed a process-level explanation rather than surface-displacement description alone.
Key Innovation: Combines UAV mapping, field validation, Sentinel-1 InSAR, failure-surface modeling, and stability analysis to resolve deep-seated landslide geometry and drivers.
2. Poroelastic aquifer response drives seasonal vertical land motion in southern Louisiana
Core Problem: Seasonal vertical land motion in coastal Louisiana is hazard-relevant, but its physical driver has been difficult to isolate.
Key Innovation: Links multi-year geodetic oscillations to confined-aquifer poroelastic response and fault-river-controlled pressure diffusion rather than surface loading.
3. Non-normal amplification in multitype Hawkes-ETAS models of earthquake triggering
Core Problem: Scalar ETAS models miss mechanism-dependent structure in earthquake-triggering cascades.
Key Innovation: Introduces a three-type Hawkes-ETAS framework showing how non-normal cross-mechanism coupling can strongly amplify otherwise subcritical sequences.
4. Probabilistic modeling of landslide dam breach based on Bayesian inference
Core Problem: Landslide-dam breach forecasts remain limited by poorly constrained erosion parameters and opaque predictive uncertainty.
Key Innovation: Embeds a simplified breach model in a Bayesian multilevel inversion to estimate posterior erosion parameters and partition uncertainty in peak discharge.
5. The multi-temporal landslide inventory map of the Collazzone study area, central Italy
Core Problem: Temporal landslide analysis is constrained by the scarcity of reproducible multi-date inventories.
Key Innovation: Releases a rigorously compiled multi-temporal inventory for more than 3500 landslides with age, movement type, and depth information.
6. Comparative analysis of debris flow mitigation by geotechnical and biological measures: a case study of Shaer Gully, Sichuan, China
Core Problem: The joint hazard-reduction value of engineering and biological debris-flow measures has rarely been quantified.
Key Innovation: Simulates multiple rainfall scenarios and forest-belt roughness to show that integrated geotechnical-biological mitigation outperforms engineering works alone.
7. Kinematic Decomposition of Three Decades of Multi-Mission DInSAR Time Series Reveals Persistent Ground Deformation Geometry at Campi Flegrei Caldera
Core Problem: Long caldera deformation records are difficult to interpret when multiple processes overlap across space and time.
Key Innovation: Uses three decades of multi-mission DInSAR and kinematic decomposition to separate a stable dominant deformation mode from weaker residual trends.
8. Submarine landslide inventory for coseismic landslides triggered by the 2016 Kaikōura earthquake in the upper Kaikōura Canyon, New Zealand
Core Problem: Coarse or single-epoch bathymetry obscures the true source characteristics of coseismic submarine landslides.
Key Innovation: Applies tectonically corrected pre- and post-event high-resolution bathymetry to map 853 small canyon-head failures relevant to tsunami and seabed risk.
9. Comprehensive field-based hazard assessment and mitigation strategies of glacial lakes in the Eastern Himalaya
Core Problem: Satellite-only GLOF screening lacks the field precision needed for site-specific mitigation planning.
Key Innovation: Builds a multi-lake field assessment using bathymetry, geophysics, moraine integrity, and hydro-meteorology, then formalizes a mitigation workflow.
10. Comprehensive analysis of rainfall-induced landslides using a multi-field coupling discrete element model
Core Problem: Rainfall-triggered landslides require full-process models that reproduce heterogeneous saturation damage and failure depth.
Key Innovation: Develops and validates a coupled discrete-element seepage model that links saturation-sensitive strength loss, weak layers, and landslide depth.
11. Experimental study on the seepage failure mechanism of climbing-type landslide dams with spatially nonuniform depositional structures
Core Problem: Climbing-type landslide dams fail by seepage through spatially nonuniform deposits, but the failure sequence is poorly constrained.
Key Innovation: Replicates depositional sorting in a large physical flume to trace seepage deformation, particle redistribution, and spatial failure progression during impoundment.
12. AI-Enhanced High-Resolution Liquefaction Hazard Mapping Using Ensemble Machine Learning and Borehole Big Data
Core Problem: Liquefaction maps are often too coarse and too weakly tied to subsurface evidence for site-level hazard use.
Key Innovation: Combines stacked ensemble learning with large borehole datasets to produce high-resolution, AI-assisted liquefaction hazard mapping.
13. Deep learning-based decoupling of hydro-meteorological signals from Cheonji caldera lake water levels for volcanic precursor detection
Core Problem: Meteorological forcing masks volcanic lake-level anomalies that could act as unrest precursors.
Key Innovation: Uses deep learning on long Landsat and Sentinel-2 records to decouple hydro-meteorological variability from volcanic water-level signals.
14. LLM as Forecasting Planner: Training-Free Text Conditioning for Time-Series Foundation Models
Core Problem: Numerical time-series forecasters do not naturally absorb contextual information expressed in language.
Key Innovation: Uses an LLM as a training-free planner that converts text into forecast conditioning for time-series foundation models.
15. Automatic Knowledge Graph Construction and Query for Earthquake Catalogs
Core Problem: Growing earthquake catalogs are hard to query interpretively without rigid temporal windows or manual expert reading.
Key Innovation: Applies graph-based retrieval-augmented generation directly to raw catalog tables for open-ended earthquake-sequence analysis.
16. Leak-Free Cross-Validated Stacking with Per-Architecture Calibration for Sand-Boil Segmentation in Earthen Levees
Core Problem: Sand-boil segmentation results can be inflated by leakage across folds and model architectures.
Key Innovation: Builds a leakage-free stacking and calibration protocol for levee seepage segmentation under stricter validation.
17. Few-Shot Open-Vocabulary Remote Sensing Segmentation via Textual Inversion
Core Problem: Open-vocabulary segmentation transfers poorly to remote sensing partly because text labels are poorly aligned with the imagery domain.
Key Innovation: Improves few-shot remote-sensing segmentation by adapting the textual prompt space through textual inversion.
18. Beyond Zooming: Learning Multi-Tool Visual Reasoning for Ultra-High-Resolution Remote Sensing
Core Problem: Ultra-high-resolution remote-sensing scenes overwhelm multimodal models because evidence is sparse, local, and widely distributed.
Key Innovation: Trains multi-tool visual reasoning for when, where, and how to inspect ultra-large scenes instead of relying on a single fixed view.
19. SwiftGS: Episodic Priors for Immediate Satellite Surface Recovery
Core Problem: Fast multi-date satellite surface reconstruction is limited by illumination change, sensor heterogeneity, and costly per-scene optimization.
Key Innovation: Uses episodic priors to predict geometry-radiation-decoupled Gaussian primitives and a lightweight SDF in a single forward pass.
20. Operational evaluation of data-driven forest fire forecasting models
Core Problem: Operational wildfire forecasting cannot be judged by standard classifier metrics alone.
Key Innovation: Evaluates data-driven forest-fire forecasting models against operational fire-danger use rather than benchmark skill in isolation.
21. Fully Automatic Trace Gas Plume Detection
Core Problem: Future imaging spectrometers will generate plume-search workloads too large for manual detection and labeling.
Key Innovation: Delivers a fully automated trace-gas plume detection and labeling pipeline with operational-scale performance.
22. Extreme Event Aware ($\eta$-) Learning
Core Problem: Rare extreme events are difficult to learn because the available data may contain few or no extremes.
Key Innovation: Introduces eta-learning, which extrapolates toward extremes without requiring extreme samples in the training data.
23. Field experimental study on the temperature and humidity field and deformation evolution characteristics of a cut-fill subgrade slope in a high-altitude loess area
Core Problem: High-altitude loess cut-fill slopes undergo coupled thermal, moisture, and deformation changes that are rarely observed in the field.
Key Innovation: Provides field-scale measurements linking temperature-humidity evolution to deformation in a high-altitude loess subgrade slope.
24. Relationships between Yield Acceleration and Static Factor of Safety
Core Problem: Yield acceleration and static factor of safety are widely used in seismic slope assessment but are not always interpreted jointly.
Key Innovation: Clarifies the relationship between yield acceleration and static factor of safety in pseudostatic embankment, dam, and slope analysis.
25. A ground motion prediction model for the Italian region based on a mixture of experts framework
Core Problem: Regional ground-motion prediction degrades when a single model must span heterogeneous tectonic settings.
Key Innovation: Uses a mixture-of-experts XGBoost framework to regionalize ground-motion prediction for Italy more flexibly.
26. Assessment of seismicity and risk from gas injection in the Groningen gas field
Core Problem: Induced seismic risk in Groningen persists even after gas production stops because the subsurface system keeps re-equilibrating.
Key Innovation: Assesses ongoing post-production seismicity and risk with explicit attention to pressure gradients and delayed gas redistribution.
27. Spatiotemporal assessment and structural attribution of urban flood resilience using interpretable machine learning: evidence from the Beijing–Tianjin–Hebei Urban Agglomeration
Core Problem: Urban flood resilience is difficult to resolve structurally across space and time.
Key Innovation: Maps flood resilience in Beijing-Tianjin-Hebei on a grid using an interpretable resistance-recoverability-adaptability framework.
28. Geohazard susceptibility assessment in gaochang district: a comparative study based on AHP-INF and multiple models
Core Problem: Fine-scale geohazard susceptibility mapping is needed in Gaochang's fragile arid terrain.
Key Innovation: Compares AHP-INF and multiple predictive models for district-scale geohazard susceptibility assessment.
29. Real-time flood severity classification using MobileViT and visual data in resource-constrained environment of Bangladesh
Core Problem: Resource-constrained regions need flood monitoring systems that can classify severity in real time on lightweight hardware.
Key Innovation: Builds a MobileViT-based visual classifier optimized for real-time flood-severity recognition.
30. A machine learning method for predicting groundwater conditions in tunnels using geological information and TBM operational data
Core Problem: Groundwater conditions ahead of TBM excavation must be inferred despite sparse direct labels and water-inrush risk.
Key Innovation: Fuses geological information with abundant TBM operational data in a machine-learning framework for groundwater prediction.
31. Multi-level earthquake damage assessment using InSAR-derived displacement
Core Problem: Earthquake damage mapping from SAR is limited when deformation is not represented directly.
Key Innovation: Introduces an InSAR-displacement-based machine-learning framework for multi-level earthquake damage assessment.
32. Estimation of earthquake indirect economic losses in mainland China with Bayesian optimization-based elastic net regression algorithm
Core Problem: Indirect earthquake losses are prolonged, system-wide, and difficult to estimate from direct damage alone.
Key Innovation: Uses Bayesian-optimized elastic-net regression to estimate earthquake-induced indirect economic losses in mainland China.
33. Laboratory-AE-driven wavelet-enhanced CNN for in-situ microseismic recognition
Core Problem: Roof-fracture microseismic events in mines are easily confused with operational disturbances.
Key Innovation: Combines wavelet enhancement and CNN classification to recognize in-situ fracture-related microseismic signals.
34. Three‐Fourths of Carbon Emissions From 2023 Record‐Breaking Wildfires in Canada Traced to Soil and Peat Combustion
Core Problem: The split between biomass and soil carbon emissions in Canada's record 2023 wildfires was unresolved.
Key Innovation: Combines ground, satellite, and fire-weather evidence to show that most emissions came from soil and peat combustion.
35. DisasterTD: Disaster Toponym Disambiguation Using Multimodal LLMs and Cross-View Geolocalization
Core Problem: Disaster social-media imagery is useful only if ambiguous place names in the accompanying text can be localized correctly.
Key Innovation: Fuses multimodal LLM reasoning with cross-view geolocalization for disaster toponym disambiguation.
36. Crustal and upper mantle model of the Middle East based on full-waveform inversion
Core Problem: Seismic structure across the Middle East remains insufficiently resolved for regional tectonic interpretation.
Key Innovation: Builds a new crust-upper mantle tomographic model for the Middle East from full-waveform inversion.
37. Freq-RemoteVAR: Next-Frequency Autoregressive Modeling for Remote Sensing Change Detection
Core Problem: Bi-temporal remote-sensing change detection misses structure when treated as a standard classification problem.
Key Innovation: Recasts change detection as next-frequency autoregressive generation in the frequency domain.
38. RRS-10K: A Multitask Vision-Language Model Benchmark for Rare Remote Sensing Image Interpretation
Core Problem: Rare remote-sensing scenarios are underrepresented in vision-language evaluation.
Key Innovation: Introduces a 10K-scale benchmark for rare remote-sensing image interpretation.
39. Forecasting Ionospheric Irregularities on GNSS Lines of Sight Using Dynamic Graphs with Ephemeris Conditioning
Core Problem: Gridded ionospheric models discard the moving observation geometry of GNSS line-of-sight data.
Key Innovation: Models ionospheric irregularities as a dynamic graph over time-varying pierce points with ephemeris conditioning.
40. Polar sea vector wind ensemble forecasting with TD3 guided multi-objective optimization
Core Problem: Polar vector-wind forecasting is difficult under nonstationary dynamics and multiple performance objectives.
Key Innovation: Uses TD3-guided dynamic weighting to optimize an ensemble for polar sea vector-wind prediction.
41. A physics-aware image-to-image surrogate model for levee reliability analysis under spatially heterogeneous soil conditions
Core Problem: Probabilistic levee reliability analysis under spatially heterogeneous soils is too expensive with full finite-element sampling.
Key Innovation: Builds a physics-aware image-to-image surrogate for low-cost levee reliability analysis.
42. Evaluation of seismic demand and structural damage caused by the Mw 6.2 Silivri (İstanbul) earthquake (April 23, 2025)
Core Problem: Moderate earthquakes can produce ambiguous demand-damage patterns across large, low-quality building stocks.
Key Innovation: Pairs instrumental demand estimates with inspections from 38,049 buildings after the 2025 Silivri earthquake.
43. Comparative evaluation of spectral indices and random forest for burned area mapping: A case study of the 2023 Çanakkale Wildfire (Türkiye)
Core Problem: Post-fire burned-area maps can vary strongly with index choice even on the same Sentinel-2 scene.
Key Innovation: Systematically compares spectral indices and random forest for mapping the 2023 Canakkale wildfire.
44. A data-driven framework for reconstructing hydraulic responses of soil slopes from sparse measurements by combining Bayesian inference and deep operator network
Core Problem: Sparse measurements make inverse reconstruction of slope hydraulic response ill-posed.
Key Innovation: Couples DeepONet with Bayesian inference to reconstruct full hydraulic fields and quantify parameter uncertainty.
45. Near-fault ground motion identification using soft-label fusion and multi-scale convolutional neural networks
Core Problem: Near-fault pulse-like motions are hard to identify reliably with single-detector, hard-label schemes.
Key Innovation: Uses soft-label fusion and multiscale CNNs for uncertainty-aware near-fault ground-motion identification.
46. Track-Leakage-Free Hold-Out Self-Validation for Photogrammetric Reconstruction: Protocol, Sensitivity, and Limits
Core Problem: Photogrammetric inspection reconstructions rarely quantify their own reliability without external ground truth.
Key Innovation: Tests self-validation protocols and sensitivity limits for leakage-free hold-out reliability estimation.
47. Physics-Informed CNN-LSTM for Street-Scale Urban Flood Prediction: Reconciling Aggregate Accuracy and Street-Level Plausibility
Core Problem: Urban-flood surrogates can be statistically accurate while violating street-scale flow physics.
Key Innovation: Adds physics-informed constraints to CNN-LSTM flood prediction to preserve realistic urban flow behavior.
48. ObliCity: A Benchmark and Baseline for Roof-to-Ground Projection Displacement Correction
Core Problem: Oblique urban imagery suffers roof-to-ground displacement that distorts building geometry.
Key Innovation: Creates the first large benchmark for roof-to-ground projection correction across UAV and satellite views.
49. ReLATE: Reliability-Guided Evidence Fusion for Robust UAV--Satellite cross-view Geo-Localization
Core Problem: UAV-satellite geo-localization benchmarks understate failure under degraded imaging conditions.
Key Innovation: Introduces a degraded-condition robustness benchmark and reliability-guided evidence fusion for cross-view matching.
50. Long-Term PM2.5 Forecasting Using a DTW-Enhanced CNN-GRU Model
Core Problem: Long-horizon PM2.5 forecasts become unstable in cities with sparse monitoring networks.
Key Innovation: Combines DTW-based station selection with a CNN-GRU framework for extended-horizon air-quality forecasting.
51. A spatio-temporal route planning algorithm with stopovers for ship navigation in Arctic ice-covered regions
Core Problem: Arctic route planning must jointly handle ice constraints, time variation, and stopovers.
Key Innovation: Formulates a spatiotemporal routing algorithm with stopovers for navigation in ice-covered seas.
52. Fourth-order flexural–gravity waves on a uniform current in finite depth: Nonlinear buckling, blocking, and energy analysis
Core Problem: Finite-amplitude flexural-gravity waves on currents are poorly described beyond low-order theory.
Key Innovation: Derives a fourth-order solution that captures nonlinear buckling, blocking, and energy behavior.
53. Infragravity oscillations in a coupled-basin marina under typhoon conditions: physical model experiments and numerical modeling
Core Problem: Marina design under typhoon forcing still underestimates long-period infragravity oscillations.
Key Innovation: Combines physical experiments and XBeach modeling to resolve oscillations in a coupled-basin marina.
54. Toward brash ice recognition for convoyed ship following: a segmentation framework combining segment-anything and morphological Chan-Vese models
Core Problem: Convoy ice resistance is hard to assess when brash-ice geometry is not segmented automatically.
Key Innovation: Combines Segment Anything with morphological Chan-Vese modeling for brash-ice recognition.
55. Deep learning-based shipborne sea-ice segmentation for ice-channel perception during Arctic convoy operations
Core Problem: Safe icebreaker-assisted convoying needs reliable onboard perception of sea-ice channels.
Key Innovation: Develops a shipborne deep-learning framework for sea-ice segmentation and channel perception.
56. A phase-state-dependent bounding surface model for predicting the thermo-hydro-mechanical response of gas hydrate-bearing marine sediments
Core Problem: Gas-hydrate-bearing sediments weaken through coupled thermal, hydraulic, and mechanical processes during dissociation.
Key Innovation: Proposes a phase-state-dependent bounding-surface constitutive model for thermo-hydro-mechanical response.
57. Towards a robust approach for satellite-derived waterline definition in complex macrotidal beaches
Core Problem: Satellite-derived waterlines lose robustness on morphologically complex macrotidal beaches.
Key Innovation: Works toward a more reliable framework for waterline definition in complex tidal settings.
58. Finite physics-informed deep learning for predicting deformations and mitigating risks induced by deep excavation
Core Problem: Deep excavations induce risky deformations that are difficult to predict efficiently.
Key Innovation: Introduces a finite physics-informed deep-learning framework for excavation-induced deformation prediction.
59. Towards a sustainable groundwater extraction: analysis of pumping-induced deformations in viscoelastic soil containing gas bubbles
Core Problem: Groundwater pumping can induce damaging deformation, complicating sustainable extraction decisions.
Key Innovation: Analyzes pumping-induced deformation in gas-bubble-bearing viscoelastic soils to assess subsidence risk.
60. A probabilistic deformation-based framework for assessing the post-earthquake operability of high-speed railway bridges in near-fault regions
Core Problem: Bridge operability after near-fault earthquakes must be judged quickly from deformation-sensitive response measures.
Key Innovation: Builds a probabilistic deformation-based framework for post-earthquake operability of high-speed rail bridges.
61. Human-in-the-Loop Signature Bootstrapping for UAV Hyperspectral PFM-1 Mine Detection
Core Problem: Operational UAV hyperspectral mine detection is constrained by false-alarm burden as much as by spectral separability.
Key Innovation: Uses human-in-the-loop signature bootstrapping to evaluate and improve PFM-1 mine detection under UAV hyperspectral imaging.
62. A Unified Benchmark and Modality-Adaptive Network for Day-and-Night Drone-View Geo-Localization
Core Problem: Day-night drone-view geo-localization lacks a unified benchmark with co-registered visible, infrared, and satellite imagery.
Key Innovation: Builds the IRCHN benchmark and a modality-adaptive network for cross-illumination localization.
63. GeoMFD: Continual Drone-View Geo-Localization with Geometry-Aware Adapter and Margin-Field Distillation
Core Problem: Drone-view geo-localization models forget previous environments when adapted continuously to new ones.
Key Innovation: Introduces geometry-aware adapters and margin-field distillation for continual cross-view localization.
64. SynFlow: Scaling Up LiDAR Scene Flow Estimation with Synthetic Data
Core Problem: LiDAR scene-flow learning is bottlenecked by the scarcity of dense motion annotations.
Key Innovation: Scales synthetic LiDAR scene-flow generation to learn motion priors that transfer to real benchmarks.
65. T2LDM++: A Self-Conditioned Representation Guided Diffusion Model for Realistic Text-to-LiDAR Scene Generation
Core Problem: Text-to-LiDAR generation is limited by scarce text-LiDAR pairs and weak geometric detail.
Key Innovation: Adds self-conditioned representation guidance and new large benchmarks for more controllable LiDAR scene synthesis.
66. Prior-matched evaluation of operational Earth-observation classifiers: a three-number reporting method demonstrated on Sentinel-1 internal-wave detection
Core Problem: Balanced-test metrics can badly overstate operational precision in rare-event Earth-observation screening.
Key Innovation: Defines a three-number prior-matched reporting protocol and demonstrates it on Sentinel-1 internal-wave detection.
67. Effective Receptive Field Ordering Matters for Infrared Small Target Detection
Core Problem: Infrared small-target detection has underexplored sensitivity to the order of receptive-field refinement.
Key Innovation: Introduces RFONet, which schedules effective receptive fields through a multigrid-inspired V-cycle.
68. A review on mining damage characteristics and protective technology of buildings
Core Problem: Mining subsidence damages buildings, but mitigation knowledge is scattered across mechanisms, monitoring, and reinforcement methods.
Key Innovation: Synthesizes damage characteristics and protective technologies for buildings in mining-subsidence areas.
69. MUSTI: Multi-scale spatiotemporal tensor imputation for global leaf area index time series reconstruction
Core Problem: Atmospheric contamination and fixed-rank retrieval artifacts degrade global LAI time series.
Key Innovation: Introduces multi-scale spatiotemporal tensor imputation with adaptive rank, wavelet attention, and biome-specific regularization.
70. Semi-supervised time series classification for real-time geological condition prediction in shield tunneling
Core Problem: Real-time geological condition prediction in shield tunneling is constrained by scarce labeled data.
Key Innovation: Uses semi-supervised time-series classification with augmentation and confidence filtering to exploit unlabeled excavation data.
71. A multiphysics thermo-hydro-mechanical framework for thermo-poro-elasto-viscoplastic modeling of frozen soils, Part I: Theoretical formulation and validation
Core Problem: Frozen soils require a consistent multiphysics description of phase change, deformation, and energy exchange.
Key Innovation: Presents a thermodynamically consistent thermo-hydro-mechanical framework with thermo-elasto-viscoplastic behavior.