TerraMosaic Daily Digest: July 9, 2026
Daily Summary
The July 9 literature shifts landslide assessment from internal model skill toward independent physical validation and decision-scale consequences. During the May 2023 Emilia-Romagna event, a susceptibility map with an inventory-split AUC near 0.945 appears random against a disturbance footprint contaminated by flooding and agriculture, but reaches a median 1.94-fold lift over chance after landslide-relevant filtering; most local blocks nevertheless remain below chance. A complementary framework carries uncertainty through initiation, runout, human vulnerability, and stochastic rainfall to construct community-scale F-N fatality curves. Together, these studies make the evaluation target explicit: not merely discrimination, but spatially credible disturbance and risk.
The mechanics studies resolve how that credibility depends on state and scale. Dry density and saturation govern whether an ancient dual-slip-zone landslide reactivates shallowly or transfers failure to deeper weak layers. Multi-site rockfall records tie seasonal activity to liquid-water input despite contrasting peak months, while debris-flow experiments quantify curvature-dependent scour, pore-pressure asymmetry, check-dam capacity loss, and the transition from impact ploughing to lubricated entrainment. Three-dimensional stress modelling further shows how landslide unloading reorganizes the surrounding mountain stress field rather than simply removing mass from the slope.
Observation and representation are advancing in parallel. Bathymetry-calibrated lake-volume models improve GLOF discharge constraints, laboratory ruptures link slip gradients directly to fault curvature, and experiment-trained CNNs estimate off-fault deformation. A seafloor observatory captures dyke propagation, deep subsidence, lava emplacement, aseismic fault slip, and triggered earthquakes within one rifting episode, while volcano-tectonic models resolve stress transfer between magma reservoirs and adjacent faults. Bias-corrected extreme precipitation, high-resolution soil moisture, multimodal registration, weak supervision, and out-of-distribution detection increasingly target transferable inputs and explicit failure recognition rather than single-benchmark accuracy.
Key Trends
Five methodological directions define the issue: independent validation, consequence-aware risk, state-dependent mobility, multi-sensor diagnosis, and geographic transfer.
- Independent observations are becoming the validation target: Event footprints, field monitoring, and cross-sensor checks expose when high inventory-split scores fail to represent local disturbance.
- Process models are being connected to societal consequence: Initiation, runout, vulnerability, rainfall uncertainty, and fatality exceedance are combined within a single community-scale risk calculation.
- Mobility is treated as a state-dependent transition: Saturation, dry density, grain size, channel curvature, substrate entrainment, and reservoir filling determine when failure accelerates, changes regime, or loses mitigation capacity.
- Multi-sensor monitoring is moving from detection to diagnosis: LiDAR, radar, optical change, GNSS, InSAR, bathymetry, and hydroacoustics constrain mechanisms, fault coupling, and warning signals across slopes, lakes, faults, and frozen ground.
- Transferability is replacing benchmark-only performance: Foundation-model adaptation, weak labels, multimodal fusion, uncertainty estimation, and out-of-distribution tests address sparse labels and geographic domain shift.
Selected Papers
The selected papers connect landslide validation, mechanics, monitoring, and community risk with GLOF volume estimation, fault coupling, volcanic processes, rainfall and soil-moisture inputs, subsidence, flooding, permafrost change, and transferable geospatial AI. This issue contains 75 selected papers from 2180 papers analyzed.
1. Independent Multi-Sensor Validation of Machine-Learning Landslide Susceptibility: Footprint Construction Decides the Verdict—May 2023 Emilia-Romagna Event
Core Problem: High inventory-split accuracy does not establish that a susceptibility map corresponds to independently observed event disturbance.
Key Innovation: Independent Sentinel-1 and Sentinel-2 validation shows that footprint construction reverses the verdict: the same map is indistinguishable from random against a contaminated footprint but reaches a median 1.94-fold lift over chance after landslide-relevant filtering, with strong spatial heterogeneity.
2. Development of F-N curves for community-level societal risk assessment subjected to rainfall-induced landslides
Core Problem: Community-scale rainfall-landslide risk lacks a coherent route from uncertain initiation and runout to annual fatality exceedance.
Key Innovation: The framework couples regional initiation-runout simulation, human fragility, stochastic rainfall, and soil uncertainty to construct community F-N curves, demonstrated across four villages in Zhejiang.
3. Anatomy of a seafloor spreading event captured by in situ seismogeodesy
Core Problem: Short-timescale deformation at submarine spreading centres is rarely observed directly, obscuring how magmatic intrusion, fault slip, and seismicity share plate extension.
Key Innovation: A combined hydroacoustic, acoustic-ranging, bottom-pressure, and repeated-bathymetry observatory captures an entire rifting episode: about 4 m of subsidence, metre-scale extension, 160 million m3 of lava, large aseismic normal-fault slip, and triggered transform-fault earthquakes.
4. Multi-site Characterization of Rockfall Seasonality in Western Colorado Using 3D Point Cloud Monitoring
Core Problem: Rockfall seasonality is difficult to transfer across sites because lithology, snowmelt, rainfall, and freeze-thaw forcing differ sharply.
Key Innovation: Multi-year terrestrial LiDAR and daily photogrammetry at four Colorado highway slopes show site-specific peak months but a common alignment between elevated rockfall activity and seasonal liquid-water input.
5. Experimental investigation on erosion mechanisms and morphological evolution of meandering debris flow channels: A case study from Jiwozi, China
Core Problem: Debris-flow erosion in bends depends on density and curvature, but the pore-pressure and morphology controls on concave-bank scour and convex-bank deposition remain unresolved.
Key Innovation: Flume experiments identify a nonlinear density-curvature control, quantify asymmetric pore-pressure loading, derive a lateral water-surface relation, and resolve four stages from vertical erosion to selective deposition.
6. Spatial-Temporal Evolution of Proglacial Lake Volumes and Estimation Models in the Himalaya and Nyainqentanglha Ranges
Core Problem: GLOF models require lake volume and maximum depth, yet direct bathymetry is sparse and global scaling laws transfer poorly across High Mountain Asia.
Key Innovation: Ten bathymetric surveys support region-specific models that outperform 14 published formulae and reconstruct strong 1990-2020 volume growth, including a 92.9% increase in the Nyainqentanglha Mountains.
7. Stress redistribution following landslides: Insights from 3D stress modelling of mountain topography
Core Problem: Removing a landslide mass changes near-surface stresses, but conventional models struggle to represent realistic mountain topography directly.
Key Innovation: A probabilistic rockfall model coupled to Finite Cell stress simulation removes 1.8 km3 of rock from the Hochkonig massif, showing short-term detachment cascades followed by declining shear stress, coherent elastic relaxation, and a more uniform long-term stress field.
8. Influence of non-landslide sampling extent on machine learning-based landslide susceptibility assessment, a case study of Wuyi County, China
Core Problem: Landslide susceptibility scores depend strongly on where non-landslide samples are drawn, yet sampling extent is rarely quantified.
Key Innovation: Twenty-two sampling scenarios show that restricting the non-landslide domain improves discrimination but must be balanced against spatial realism; the preferred coverage is 71.42-89.92% of the study area.
9. Hydro-mechanical reactivation of a dual-sliding-zone ancient landslide controlled by dry-density-dependent water sensitivity
Core Problem: Ancient landslides with multiple slip zones respond nonlinearly to groundwater and rainfall because residual strength depends on dry density and saturation.
Key Innovation: Ring-shear-derived strength relations embedded in limit-equilibrium analysis reveal threshold groundwater responses, migration from shallow to deep failure surfaces, and localized rainfall-driven toe instability.
10. Evolution of sediment-trapping effectiveness of check dams under multiple debris flows: an experimental study
Core Problem: Check-dam performance degrades across repeated debris flows, but maintenance decisions lack a process-based measure of remaining sediment-trapping capacity.
Key Innovation: Experiments separate slit-blocking and reservoir-filling phases and derive filling-ratio controls for slit and solid dams across sediment concentrations.
11. A generalized phenomenological method for extracting and warning of landslide deformation based on ground-based radar
Core Problem: Ground-based radar supplies dense deformation images, but robust landslide-body extraction and warning indices remain underdeveloped.
Key Innovation: Otsu segmentation with region growing isolates the moving body, while standardized regional deformation and CSRDI curves detect temporal acceleration and issue case-validated warnings.
12. Particle size effect on the impact processes and geomorphology of granular flow deposition
Core Problem: Granular-flow runout and deposit morphology change with particle size, limiting scale-independent mobility prediction.
Key Innovation: Flume, PIV, seismic, and laser observations identify a transition from continuous low-impact flow to independent high-impact grains and motivate separate physically based runout models for the two regimes.
13. Erosion and entrainment in dry granular flows: Insights from large-scale experiments
Core Problem: Conflicting interpretations of substrate erosion hinder prediction of how entrainment changes rock-avalanche mobility.
Key Innovation: Large-scale experiments resolve spatially distinct ploughing, basal-layer entrainment, and distal shear localization, showing when fine-grained basal material lubricates rather than dissipates the flow.
14. Tsunami Traveling Ionospheric Disturbances Triggered by the 29 July 2025 Mw 8.8 Kamchatka Earthquake Observed in Japan and Taiwan
Core Problem: The spatial evolution of tsunami-driven ionospheric disturbances is difficult to observe across ocean basins using ground networks alone.
Key Innovation: More than 1,400 GNSS stations resolve seismic and two tsunami-driven ionospheric waves; the primary-wave source lies within about 90 km of the main-slip centre and a second signal records coastline-reflected tsunami energy.
15. Back-Propagating Earthquake Rupture Controlled by Fluid-Modulated Effective Normal Stress
Core Problem: Earthquake ruptures sometimes reverse direction, but the stress conditions that permit back-propagation remain uncertain.
Key Innovation: Fluid-coupled rupture simulations show that low shear-to-effective-normal-stress ratios favour pulse-like backward rupture, whereas higher ratios produce crack-like forward rupture; localized pore-pressure contrasts nucleate secondary reverse fronts.
16. Deep magma underpressure and connectivity drive large dike intrusions
Core Problem: The size of dike intrusions cannot be predicted from shallow deformation alone because deep reservoir pressure and connectivity are poorly constrained.
Key Innovation: Geodesy and seismicity reconstruct a 1.4 km3 Ethiopian-rift intrusion supplied for roughly three months by connected reservoirs 6-12 km deep; pressure modelling shows that deep underpressure is required to drain such a large volume.
17. Fault Geometrical Control on Fault Slip: Experimental Validation of Theoretical Models
Core Problem: Fault geometry is widely invoked to explain heterogeneous earthquake slip, but the predicted curvature-slip relation has lacked direct experimental validation.
Key Innovation: High-resolution laboratory earthquakes on bent and self-similar rough interfaces show that local slip gradients scale with curvature and provide the first experimental evidence that known fault geometry can predict slip variability.
18. Role of Volcano-Tectonic Interactions During Early-Phase Magma-Assisted Continental Rifting
Core Problem: Magma-reservoir pressure changes can load adjacent rift faults, yet topography and heterogeneous crustal structure are often omitted from stress-transfer models.
Key Innovation: PyLith models constrained by 2016-2023 GNSS data show that topography improves displacement fits and that reservoir deflation beneath Ol Doinyo Lengai can raise Coulomb stress by more than 0.1 MPa on the adjacent Natron Fault.
19. Fourier-feature-enhanced physics-informed neural networks for highly nonlinear seepage analysis in unsaturated soils
Core Problem: Standard PINNs exhibit spectral bias and poor convergence when unsaturated seepage develops steep suction gradients relevant to rainfall-triggered slope failure.
Key Innovation: Fourier feature embeddings stabilize Richards-equation solutions across three soils and sharply reduce high-gradient errors relative to a conventional PINN while retaining finite-element agreement.
20. Source Scaling of Shallow Volcano-Tectonic Earthquakes at Piton de la Fournaise Volcano (La Réunion, France) Estimated From Codas of Seismograms
Core Problem: Source scaling for shallow volcano-tectonic earthquakes is poorly constrained because attenuation and path effects obscure small-event source parameters.
Key Innovation: Coda-based estimates at Piton de la Fournaise recover earthquake size and rupture-time scaling while reducing sensitivity to complex volcanic propagation paths.
21. Earthquake Populations From Stochastic Stress Fields
Core Problem: Smooth-fault rupture models can reproduce large events but not populations containing many small earthquakes, foreshocks, and aftershocks.
Key Innovation: Spatially heterogeneous stress fields generate broad event-size distributions and show that strongly variable, locally negative stored energy makes rupture growth intermittent rather than smooth.
22. Interpretable Machine Learning Framework for Rapid Post-Earthquake Building Safety Assessment at a Regional Scale
Core Problem: Regional post-earthquake screening must classify building safety rapidly from sparse observations without becoming an opaque black-box decision.
Key Innovation: An interpretable classifier combines limited sensing inputs with explicit feature attribution to assign regional building safety classes at operational speed.
23. The impact of land subsiding on pluvial flooding in a mega-city: a century of evidence from Shanghai
Core Problem: The cumulative influence of land subsidence on urban pluvial flooding is rarely quantified over century-long timescales.
Key Innovation: A century-scale Shanghai reconstruction demonstrates that subsidence increases both flood depth and inundated area, with amplification strongest under extreme rainfall.
24. Counterfactual modeling isolates sand mining impacts, revealing it as a key driver of Mekong Delta destabilization
Core Problem: Sand-mining effects on delta stability are difficult to separate from upstream dams and climate-driven hydrological change.
Key Innovation: Counterfactual simulations attribute 25-30% of total Mekong channel incision to extraction, with mining volumes six to fifteen times natural sediment supply and erosion affecting about 65% of channels.
25. SpectralX: Parameter-efficient domain generalization for spectral Remote Sensing Foundation Models
Core Problem: Remote-sensing foundation models must ingest different spectral configurations without full retraining or loss of cross-domain generalization.
Key Innovation: SpectralX adds parameter-efficient spectral adaptation and a two-stage training strategy to reuse a common foundation backbone across heterogeneous spectral inputs.
26. Remote sensing change captioning meets large language and vision models
Core Problem: Change captioning must distinguish meaningful surface change from appearance differences and evaluate whether generated descriptions are factually complete.
Key Innovation: Large vision-language models are integrated with remote-sensing change captioning, alongside FMScore, an evaluation metric based on explicit reference facts and LLM-assisted fact matching rather than surface text similarity alone.
27. Stereo Matching in Satellite Imagery: A Depth Estimation Foundation Model-Assisted Iterative Approach
Core Problem: Satellite stereo reconstruction remains unreliable in textureless areas, repetitive patterns, occlusions, and large temporal or viewing-angle differences.
Key Innovation: IFMA-Stereo injects monocular depth-foundation priors into multi-scale matching and iterative disparity refinement, improving US3D and GaoFen-7 accuracy and transfer to unseen urban scenes.
28. Estimating Off-Fault Deformation Using Convolutional Neural Networks Trained on Experimental Strike-Slip Faults
Core Problem: Off-fault deformation is a major but sparsely measured component of strike-slip strain, limiting fault-slip and seismic-hazard estimates.
Key Innovation: CNNs trained on scaled physical fault experiments exceed 90% accuracy on unseen single-material experiments and transfer best to crustal fault maps when training materials reproduce comparable loading conditions.
29. High-Resolution Subsurface Geophysical Characterisation of Icelandic Volcanic Layering
Core Problem: Single geophysical methods yield non-unique velocity structures in fractured volcanic terrain, obscuring contrasts relevant to site response.
Key Innovation: Joint onshore measurements and Scholte-wave modelling distinguish low-Vs30 young fractured lava from consolidated Miocene basalt and show that legacy marine seismic data can extend regional velocity characterization.
30. Dataset of daily vertical displacements observed by GPS between 1994 and 2023 for hydrogeodetic studies over Europe
Core Problem: Hydrogeodetic studies lack a harmonized long-term record that separates daily vertical crustal motion from hydrological loading across Europe.
Key Innovation: The dataset provides daily GPS displacement series from 1994-2023 and demonstrates regional consistency with precipitation, hydrological models, InSAR, GRACE, and ERA5-Land water-storage variations.
31. Static and Dynamic Stability Analysis of the Erkenek Tunnels Damaged During the February 6, 2023, Türkiye Earthquake
Core Problem: The severe damage to the Erkenek tunnels during the 2023 Turkiye earthquake challenges the assumption that underground structures are uniformly protected from strong shaking.
Key Innovation: Integrated field evidence and static-dynamic stability analysis identify the geological and structural conditions that concentrated damage in the twin tunnels.
32. Three-year post-fire soil erodibility and erosion dynamics in a Mediterranean microbasin
Core Problem: Post-fire erosion recovery is commonly inferred from the first year, although soil hydraulic and strength changes can persist longer than vegetation loss.
Key Innovation: Three years of field measurements and InSAR show deformation-related erosion peaking in year two before falling below pre-fire levels in year three, and establish burn-severity and time-dependent K-factor multipliers.
33. A framework for earthquake seismic signal classification with the aid of hybrid feature selection schemes and robust machine learning models
Core Problem: Automated earthquake-signal classification must remain robust when informative features are redundant, noisy, or strongly dataset dependent.
Key Innovation: Hybrid feature-selection schemes are paired with multiple machine-learning classifiers to identify a compact, stable representation for seismic event discrimination.
34. A coupled geotechnical and sediment transport model to predict dune erosion during storms
Core Problem: Storm-erosion models often move sediment without representing the geotechnical failure of an over-steepened dune face.
Key Innovation: The process-based framework couples sediment transport with geotechnical instability so that hydraulic scour and episodic dune-face collapse evolve within one model.
35. Surface Heat Source Variations and Driving Factors in Typical Permafrost Areas of the Tibetan Plateau
Core Problem: The surface-energy controls that accelerate continuous-permafrost warming can differ from those at isolated permafrost sites.
Key Innovation: Multi-year observations detect a 2.2 W m-2 yr-1 increase at the continuous-permafrost site, and random-forest attribution assigns more than 88% of variability to soil temperature, shortwave radiation, and albedo.
36. Vegetation and permafrost thermal responses to highway engineering under a warming climate on NE Qinghai-Tibet Plateau
Core Problem: Highway corridors in warm discontinuous permafrost experience interacting vegetation disturbance and ground-temperature change that are rarely assessed together.
Key Innovation: Remote sensing, borehole observations, and GeoDetector analysis separate precipitation-elevation controls on vegetation from elevation-soil controls on deep ground temperature.
37. Time-series framework integrating rainfall and satellite soil moisture for multi-scale urban flood risk assessment
Core Problem: Urban flood assessments often omit antecedent soil moisture or treat rainfall at a single temporal scale.
Key Innovation: The framework fuses hourly rainfall with three-hourly SMAP soil moisture across eight Korean flood events to resolve flood risk across temporal and spatial scales.
38. Time-dependent structural reliability under seismic actions using adaptive Polynomial-Chaos-Kriging model
Core Problem: Seismic limit-state functions can change abruptly over a structure's service life, defeating smooth time-dependent reliability approximations.
Key Innovation: An adaptive Polynomial-Chaos-Kriging surrogate targets non-differentiable response transitions while reducing the number of expensive dynamic analyses.
39. Theoretical study of active earth pressure under seismic-anisotropic seepage coupling field
Core Problem: Retaining-earth-pressure solutions rarely combine earthquake loading with direction-dependent seepage, despite their coupled influence on failure demand.
Key Innovation: A coupled analytical solution and parameter study show an approximately linear seismic-seepage control on active earth pressure across anisotropy conditions.
40. Seismic response of rigid drainage pile groups in liquefied lateral spreading sites
Core Problem: Lateral spreading in liquefied marine sediments combines pore-pressure accumulation with large kinematic demand on pile groups.
Key Innovation: Shaking experiments show that rigid drainage piles reduce excess pore-pressure ratios by as much as 40% relative to conventional piles while modifying group interaction under spreading.
41. CIPCast 5.0: A multi-hazard Decision Support System for risk and resilience assessment of buildings and infrastructure
Core Problem: Emergency and adaptation planning require one platform to translate multiple hazards into comparable damage, risk, and infrastructure-service consequences.
Key Innovation: CIPCast 5.0 integrates earthquakes, floods, extreme precipitation, and heatwaves with building and network models to produce damage maps, risk indicators, and resilience metrics for response and long-term planning.
42. Quantifying potential ground-ice meltwater release and seasonal active-layer water-storage change using decade-long InSAR-derived ground deformation in the Qilian Mountains
Core Problem: Regional observations rarely constrain how thawing ground ice changes both meltwater release and seasonal active-layer storage.
Key Innovation: A decade of InSAR deformation in the Qilian Mountains converts seasonal and long-term vertical motion into spatial estimates of active-layer water storage and potential ground-ice meltwater release.
43. Iron as a driver of organic carbon fate in permafrost regions
Core Problem: Permafrost carbon projections incompletely represent how iron stabilizes, transports, or releases organic carbon during thaw.
Key Innovation: A process framework links redox-dependent iron-carbon interactions to changing hydrology and thaw state, clarifying when iron protects carbon and when it promotes mobilization.
44. WaterPLNet: A noise-robust weakly supervised network for water extraction from SAR imagery using point labels
Core Problem: Operational SAR water mapping is limited by the cost of dense labels and by noise in pseudo-masks derived from sparse supervision.
Key Innovation: WaterPLNet learns water boundaries from point labels with explicit noise-robust refinement, reducing annotation demand for large-area flood and surface-water mapping.
45. Physics-Informed Semantic Prompt Learning for Few-Shot Low-Altitude Radar Target Recognition in Remote Sensing
Core Problem: Few-shot radar recognition is difficult when weak targets share similar trajectories and labelled tracks are scarce.
Key Innovation: Physical radar descriptors are encoded as semantic prompts for a partly tuned GPT-2 and relation network, reaching a 90.63% F1 score across four classes in the 20-shot setting.
46. A Modular and Transferable Framework for Enhancing Satellite-Derived Daily Precipitation: Adjusting Values, Aligning Distributions, and Preserving Extremes
Core Problem: Daily satellite precipitation retains systematic bias in gauge-sparse regions, especially in the distribution tail that controls flood and rainfall-triggered hazard analysis.
Key Innovation: A linear-scaling, extreme-tail quantile-mapping, and confidence-weighted CNN chain corrects IMERG Late Run over Indonesia, reducing relative bias from -11.4% to -0.6% and bringing the 99th-percentile ratio from 0.71 to 1.01 at independent stations.
47. FFCDNet: Remote sensing image change detection method based on fourier frequency domain feature enhancement
Core Problem: Change-detection networks struggle to preserve local boundaries while modelling long-range variation without excessive computation or directional bias.
Key Innovation: FFCDNet enhances Fourier-domain features to couple local structure with global frequency context for bitemporal remote-sensing change maps.
48. T2Net: Taxonomy-Guided Tree-Structured Prompt Learning for Few-Shot Remote Sensing Scene Classification
Core Problem: Few-shot remote-sensing scene classification is confounded by high within-class diversity and visual similarity between related classes.
Key Innovation: T2Net organizes semantic prompts along a class taxonomy so a vision-language model can share information hierarchically while retaining class-specific distinctions.
49. Few-Shot Remote Sensing Scene Classification via Fusion of Zigzag Scanning Feature Sequence and Riemannian Geometric Barycenter Network
Core Problem: Euclidean few-shot classifiers underuse spatial structure and poorly represent the non-Euclidean distribution of remote-sensing scene features.
Key Innovation: ZSFS-RGBN preserves spatial order with zigzag feature sequences, maps autoregressive descriptors to an SPD manifold, and classifies scenes using Riemannian barycentres, outperforming representative baselines in 1-shot and 5-shot tests.
50. Structure-Confidence Guided Phase Congruency and Cascade Matching for Registration of Optical and SAR Images
Core Problem: Optical-SAR registration loses correct correspondences when speckle and repeated textures create unstable structural keypoints.
Key Innovation: Structure-confidence-weighted phase congruency suppresses pseudo-structure and cascade matching restores correspondences; across 60 pairs it increases correct matches by at least 2.43-fold and accuracy by more than 19.85%.
51. Attention-Guided Generative Adversarial Network for False Alarm-Resistant Change Detection in Remote Sensing Orthophotos
Core Problem: Residual misregistration, shadow displacement, and illumination differences produce false change alarms in multitemporal orthophotos.
Key Innovation: Attention-GAN trains on controlled geometric and radiometric negative perturbations, reducing false changes to 4.9% while achieving F1 scores of 91.2-93.18% on three benchmarks.
52. Relative Energy Learning for LiDAR out-of-distribution detection
Core Problem: LiDAR models can assign confident labels to unfamiliar geometry, undermining reliability when scenes depart from training data.
Key Innovation: Relative-energy learning separates in-distribution structure from unknown point-cloud inputs and supplies an explicit out-of-distribution signal for downstream screening.
53. Navigating Hierarchies in Hyperbolic Space: A Knowledge-Driven Graph Network for Semantic Segmentation of Remote Sensing Imagery
Core Problem: Remote-sensing segmentation must represent nested geographic classes and multi-scale spatial relations without flattening their hierarchy.
Key Innovation: A knowledge-driven graph network performs hierarchy-aware representation in hyperbolic space, integrating geographic relations directly into semantic segmentation.
54. Geospatial foundation models enable data-efficient tree species mapping in temperate mountain forests
Core Problem: Mountain-forest species mapping is label limited and vulnerable to environmental gradients and mixed stands.
Key Innovation: Geospatial foundation-model embeddings improve data efficiency and reach a weighted F1 of 0.83, outperforming conventional representations under limited supervision.
55. Toward a Unified Dual-Path Framework for Multimodal Remote Sensing Data Fusion: Integrating Temporal, Resolution, and Spectral Dimensions
Core Problem: Spatiotemporal fusion, pan-sharpening, and change detection are usually optimized as separate tasks despite sharing cross-sensor alignment problems.
Key Innovation: DPLNet uses a common dual-path architecture to fuse temporal, spatial-resolution, and spectral information across all three task families.
56. Weakly supervised learning for snow cover segmentation in mountainous areas from Sentinel-1 SAR images using interpolated NDSI time series
Core Problem: Mountain snow mapping needs all-weather SAR observations, but dense SAR labels are difficult to produce and optical labels are cloud limited.
Key Innovation: Interpolated optical NDSI supplies weak supervision for a Sentinel-1 segmentation model, combining optical snow specificity with SAR cloud penetration.
57. Baseline-Dependent Retrieval of Underlying Terrain and Forest Height From Hongtu-1 Bistatic X-Band InSAR
Core Problem: X-band InSAR estimates of ground and canopy height vary with baseline and canopy penetration, complicating terrain recovery beneath vegetation.
Key Innovation: Hongtu-1's six bistatic baselines are used to quantify baseline-dependent penetration and jointly retrieve underlying terrain and forest height.
58. A spaceborne lidar guided TanDEM-X InSAR phase height histogram approach for robust ground elevation and canopy height mapping considering canopy penetration capability
Core Problem: TanDEM-X phase-height distributions mix canopy and ground returns, biasing elevation in forests with variable penetration.
Key Innovation: Spaceborne LiDAR anchors a phase-height histogram model that explicitly accounts for canopy penetration to recover ground elevation and canopy height.
59. Generating a defect-enriched geometric digital twin for tunnel operation and maintenance using a bottom-up unsupervised approach
Core Problem: Tunnel digital twins remain expensive to scale because many workflows depend on labelled defects, component libraries, or rigid templates.
Key Innovation: A bottom-up unsupervised pipeline reconstructs long curved tunnels and enriches the geometry with detected defects without pre-labelled training data.
60. Morphological dynamics and controlling factors of gully erosion in two contrasting loess landforms of Western Shanxi, China
Core Problem: Gully growth controls sediment production on the Loess Plateau, but landform-specific trajectories and drivers remain poorly separated.
Key Innovation: Comparative morphology across two loess landforms identifies distinct erosion-development patterns and controlling factors for targeted soil-water conservation.
61. 100 years of soil erosion monitoring: advances, limitations and pathways for improved validation of soil erosion models
Core Problem: Soil-erosion models are widely applied despite fragmented validation records and incompatible monitoring methods accumulated over a century.
Key Innovation: The review synthesizes monitoring advances and limitations and defines validation pathways linking field measurements, remote sensing, and model evaluation.
62. A Fractional-Calculus Unified Framework for Manhole Shaft Uplift Analysis: Predicting Velocity and Displacement via Thermodynamic Phase-Field Theory of Liquefaction (FC-ULSM)
Core Problem: Construction-induced liquefaction can uplift buried shafts, but existing models do not jointly predict uplift velocity and displacement.
Key Innovation: FC-ULSM combines fractional calculus with a thermodynamic phase-field description to model the time-dependent onset and magnitude of shaft uplift in saturated ground.
63. Estimation of High-Resolution Multi-Layer Soil Moisture Using Land Data Assimilation and the Three-Cornered Hat Method
Core Problem: Coarse soil-moisture products cannot resolve local wetness or root-zone state needed for hydrological and slope-response modelling.
Key Innovation: Sixteen-metre multi-sensor soil moisture and three-cornered-hat uncertainties are assimilated into Noah-MP with an Ensemble Kalman Filter, improving surface and root-zone estimates during a Yunnan spring drought.
64. ECDF-Stabilized Sentinel-1 InSAR Coherence for Verification-Oriented Heritage Monitoring: A TabICLv2-SHAP and Three-Horizon DEMATEL Architecture
Core Problem: Multi-epoch Sentinel-1 coherence values are difficult to compare directly across acquisition intervals and sectors, weakening verification-oriented monitoring.
Key Innovation: Fixed-support sampling, empirical-CDF rank alignment, ordinal recoding, TabICLv2-SHAP interpretation, and multi-horizon DEMATEL convert 12 interferometric intervals and 1,193 points into a cross-epoch evidence register.
65. A neighborhood change detection method for soil moisture retrieval using single-temporal Sentinel-1 data over vegetated fields
Core Problem: Short-term SAR soil-moisture retrieval assumes temporal invariance and dense time series, making it sensitive to changing vegetation and surface roughness.
Key Innovation: The neighborhood change-detection formulation transfers the invariance assumption from time to space and solves target-neighbour observation equations to retrieve soil moisture from a single Sentinel-1 acquisition.
66. Rectify-Match-Reconstruct (RMR): Framework for dense and accurate metric 3D reconstruction via neural homography alignment
Core Problem: Image-based infrastructure reconstruction loses metric accuracy under wide baselines, repeated texture, scale ambiguity, and geometric misalignment.
Key Innovation: RMR rectifies views before correspondence and reconstruction, combining neural homography alignment with metric constraints to produce dense field-scale models for structural condition assessment.
67. Spatiotemporal fusion of radar backscatter for enhanced monitoring of surface melt dynamics over Antarctic ice shelves
Core Problem: Antarctic surface-melt monitoring faces a resolution trade-off between daily coarse scatterometers and intermittent high-resolution SAR.
Key Innovation: SRBDF fuses ASCAT and Sentinel-1 C-band backscatter into daily high-resolution observations, extending optical spatiotemporal fusion principles to radar-based ice-shelf melt monitoring.
68. Soil erosion risk and prediction under land use changes based on RUSLE and PLUS models in the black soil region of Northeast China
Core Problem: Land-use transitions alter erosion exposure, but long-term observations and future planning scenarios are rarely evaluated within one spatial framework.
Key Innovation: Coupled RUSLE and PLUS modelling reconstructs 1980-2020 erosion change and shows that the 2030 farmland-protection scenario produces substantially higher erosion than natural-development and ecological-conservation alternatives.
69. Hydraulic mechanisms underlying vegetation mitigation of ephemeral gully erosion in Mollisols from Northeast China - insights from a flume experiment
Core Problem: Vegetation reduces ephemeral-gully erosion, but its hydraulic controls across slope and inflow conditions remain poorly quantified in humid Mollisols.
Key Innovation: Flume experiments show that vegetation deepens and slows runoff, raises hydraulic roughness, and reduces soil detachment by 96.3%, with stronger mitigation at steeper slopes and larger inflows.
70. Beyond the meso level: A multilevel analysis of hospital resilience to floods
Core Problem: Hospital flood resilience is commonly assessed at whole-facility level, overlooking dependencies among staff, departments, infrastructure, and external services.
Key Innovation: A resilience-engineering framework traces cross-level interactions and informal practices that support or impede continued care during floods.
71. A Multi-Stage Outlier Removal Method for Point Clouds with High Outlier Ratio
Core Problem: High outlier ratios can destroy valid point-cloud geometry and cause conventional filters to remove structural detail.
Key Innovation: MORPH combines density-peak coarse filtering, structure-adaptive refinement, and scale-consistent local-plane completion to remove heavy contamination while preserving geometry.
72. Robust small object detection via multimodal pre-trained representations and lightweight task-specific adaptation
Core Problem: Small-object detection remains unstable when targets occupy few pixels, backgrounds are complex, and task-specific labels are scarce.
Key Innovation: A foundation-model-driven framework adapts multimodal pretrained representations with a lightweight task head to improve feature discrimination under limited labels.
73. DSDGNet: Dual-Scale Dynamic Graph Network for Hyperspectral Image Change Detection
Core Problem: Hyperspectral change detection must distinguish real land-cover transitions from high-dimensional spectral variability across dates.
Key Innovation: DSDGNet couples local and broader graph interactions at two scales to propagate spectral-spatial evidence across bitemporal hyperspectral imagery.
74. PRGD: Proximal Residual-Guided Diffusion with pattern-aware experts for remote sensing image super-resolution
Core Problem: Diffusion-based remote-sensing super-resolution can accumulate reconstruction error and drift from the measured low-resolution signal.
Key Innovation: PRGD combines residual-shifted generation, measurement-consistent proximal sampling, and pattern-aware guidance to preserve data fidelity throughout iterative image enhancement.
75. CAMELS-PE: Hydrometeorological time series and catchment attributes for 136 catchments in Peru
Core Problem: Tropical Andean hydrological modelling lacks standardized large-sample observations spanning strong climatic and physiographic gradients.
Key Innovation: CAMELS-PE provides harmonized daily forcing, observed and simulated streamflow, geospatial layers, and catchment attributes for 136 Peruvian basins under a reproducible screening workflow.