TerraMosaic Daily Digest: October 27, 2026
Daily Summary
Mass-flow mobility depends on what a moving mass encounters along its path. Dry granular experiments show that a low-density erodible bed can extend runout, but the contrast also involves friction, particle shape and deformation. A Vajont modelling study addresses another interaction: a landslide entering water within irregular reservoir geometry. Together, the studies connect propagation to the flow–bed or solid–water interface while keeping laboratory interpretation separate from a numerical event reconstruction.
Timing and source assumptions shape the interpretation of mountain hazards. The Hong Kong knowledge-graph study identifies a delayed landslide response extending through the week after heavy rainfall, while its much weaker mechanism-classification performance limits event-specific causal attribution. Around Badrinath, hanging-glacier scenarios translate release and terrain assumptions into downstream exposure. The affected-building counts describe conditional footprints rather than failure probabilities or expected losses.
The computational studies are most useful when their physical and numerical boundaries remain explicit. An ordered multi-task slope surrogate approximates instability and post-failure fields from stochastic material-point simulations, reducing the cost of a Monte Carlo workflow. Tailings-dam reinforcement modelling shows that deformation restraint can coincide with stronger local acceleration. A monitored excavation case provides a complementary engineering example in which seepage analysis, grout properties and construction control are evaluated together.
Observation studies distinguish the target being measured from the hazard inference built on it. Time-series Sentinel-1 anomalies have different flood-detection performance in natural and built-up areas. A new Arctic water-mask preprint documents overlap-sensitive holdout results and the limits of agreement with its released targets. Radar nowcasting, typhoon assimilation and lake-altimetry comparisons add further tests of spatial structure, case dependence and reference uncertainty.
Key Trends
The issue connects interface-controlled motion, delayed hazard response and computational efficiency with explicit tests of what observations and models resolve.
- Flow interfaces govern propagation: Dry-flow experiments expose coupled bed-property effects, while the Vajont method treats solid–water motion and irregular boundaries. Neither a material contrast nor a successful numerical reconstruction alone establishes a transferable runout rule.
- Timing carries information beyond the daily event count: The Hong Kong analysis preserves post-rainfall lag and the evidence behind mechanism assignments. The distinction between useful storm-stage interpretation and weak event-level attribution is central to its practical use.
- Scenario exposure and safety metrics answer different questions: Badrinath footprints identify assets within simulated avalanche paths. Tailings-dam displacement, local acceleration and post-earthquake stability respond differently, and reinforcement efficiency depends on the tested spacings and weights.
- Fast surrogates retain their reference-model limits: The stochastic slope workflow reports errors in predicted strain and displacement fields against its material-point reference, alongside total computation time. Its ordered decoders encode a mechanical sequence, while field calibration and empirical causal identification remain separate questions.
- Reference quality changes the meaning of accuracy: Flood scores differ between open and urban land cover. Arctic segmentation tests distinguish released-target agreement from independent reference accuracy, and SWOT–ICESat-2 residuals combine the uncertainties and sampling conditions of both sensors.
Selected Papers
The selected work links landslide timing and mass-flow propagation with source assumptions, material variability and observation conditions. It spans mountain hazards, flood and storm assessment, engineering mitigation, and specific methods that support those analyses. Papers appear in one continuous list ordered by relevance score.
1. Knowledge-graph reasoning for temporal lag effects and mechanism inversion of heavy-rainfall-induced landslide events: case study on Hong Kong
Core Problem: Landslide inventories grouped by calendar day can conceal delayed responses to a storm and disconnect inferred mechanisms from investigation evidence.
Key Innovation: A Hong Kong knowledge graph links rainfall, slope attributes and reports across 2,961 records, identifying an immediate high-hazard stage and an extended response through day seven. Mechanism attribution is much weaker: leakage-controlled Top-1 accuracy is 0.163, with only 43 report-anchored events. The workflow supports storm-cluster interpretation and investigation priorities rather than reliable event-specific causal diagnosis.
2. Mobility of dry granular flows over contrasting erodible bed materials in laboratory experiments
Core Problem: Entrainment can change mass-flow mobility, but bed-material contrasts simultaneously change more than particle density.
Key Innovation: Dry quartz releases over contrasting beds show longer runout over the low-density material, while equal- and high-density beds respond similarly. Friction, shape, deformation and entrainment jointly explain the asymmetry. Surface measurements do not isolate density as the cause or resolve basal momentum exchange, and the results should not be transferred directly to saturated debris flows.
3. A surface-mesh-assisted semi-implicit two-phase double-layer MPM for 3D simulation of the Vajont disaster on GPUs
Core Problem: Irregular reservoir boundaries complicate pressure projection and phase-compatible boundary conditions in two-phase material point simulations.
Key Innovation: A surface-mesh-assisted semi-implicit formulation combines boundary treatments, particle shifting, projection-error compensation and anti-penetration enforcement for GPU computation. Analytical and experimental benchmarks precede a three-dimensional Vajont application. This supports a numerical treatment of complex boundaries, not a demonstrated forecast of another reservoir failure or an independently established universal run-up accuracy.
4. A stochastic BSMPM-derived causal multi-task U-Net surrogate model for quantitative risk assessment of spatially variable slopes
Core Problem: Monte Carlo slope assessment is expensive when each realization must resolve instability and large post-failure displacement.
Key Innovation: A BSMPM-trained U-Net predicts strength reduction, plastic strain and displacement, with an ordered, gradient-isolated decoder coupling. Runout and volume extracted from predicted fields are checked against the simulator, while a 300-realization workflow saves 87.2% total time. Here “causal” describes mechanical ordering in the architecture; simulator agreement and stage-specific acceleration do not establish field-calibrated risk or equally large end-to-end speedups.
5. Vulnerability assessment of hanging-glacier avalanches in Badrinath Township, Central Himalayas, India
Core Problem: Steep Himalayan terrain exposes settlements to hanging-ice releases whose downstream effects depend on source volume and the intervening valley.
Key Innovation: Remote-sensing release estimates and RAMMS mobility scenarios compare three sites around Badrinath and identify the most exposed corridor. The reported affected buildings and infrastructure are conditional exposure counts, rather than demonstrated vulnerability, expected losses or avalanche probabilities. Limited variation within the tested friction ranges does not encompass every source and model uncertainty.
6. Three‐Dimensional Resistivity Imaging Constrains Crustal Heterogeneity Beneath the 2010 Yushu Ms 7.1 Earthquake Rupture
Core Problem: The Yushu rupture gap and evolving aftershock pattern require constraints on crustal structure that distinguish resolved anomalies from poorly sampled regions.
Key Innovation: A 70-site magnetotelluric study combines alternative inversions and targeted forward tests to identify robust resistivity contrasts near the mainshock and early aftershocks. These features support a possible structural influence on rupture segmentation. Sparse coverage leaves the late conjugate swarm unresolved, preventing a single electrical explanation for all stages of the sequence.
7. Dynamic response and spacing optimization of a tailings dam reinforced with grid-pattern waste rock inclusions under representative ground motions
Core Problem: Permeable waste-rock inclusions can reduce deformation while increasing local acceleration, making reinforcement spacing a multi-objective choice.
Key Innovation: FLAC3D comparisons of three spacings under four motions balance drainage, stiffness, displacement and material volume. Dynamic calculations use the established fixed pore-pressure field and continuous shared-gridpoint connections. The preferred 40 m efficiency result depends on the tested alternatives and weighting, and does not validate cyclic excess-pore-pressure generation, interface slip or a universal optimum.
8. Mechanism of pile-bottom leakage and grouting control in deep foundation pits of water-rich sand strata: a case study of the Xiong’an–Xinzhou high-speed railway project
Core Problem: Leakage below retaining piles in water-rich sand can enlarge through hydraulic forcing and grain migration, threatening deformation control in deep excavations.
Key Innovation: A railway-pit case combines seepage–stress analysis, monitoring and grout/process optimization, reporting lower leakage and restrained pile displacement after treatment. The material, pressure–diffusion relation and cost benefits are project-specific. They provide an engineering treatment example rather than a transferable recipe for all permeable ground.
9. Large-scale flood monitoring based on time-series Sentinel-1 images and Z -score
Core Problem: Vegetation, geometry and double-bounce scattering make a single backscatter-change rule unreliable across open and built-up flood areas.
Key Innovation: A Sentinel-1 time-series classification tree combines polarization-specific standardized anomalies with auxiliary information for the Weihui flood. Separate natural and building-area metrics expose differing performance. Z-scores depend on the reference distribution, and observed backscatter change should not be equated with water depth or uninterrupted real-time coverage.
10. UAV monitoring of geomorphic trajectories in foredune resilience
Core Problem: Foredune recovery involves changes across the whole profile and alongshore variation, which infrequent surveys can miss.
Key Innovation: Eight UAV surveys spanning several storms on Jekyll Island classify trajectories from scarping toward distributed infilling and rebuilding. Initial geometry and directional transport history help interpret persistent impact and recovery zones. Local boardwalk associations do not establish causation, and the observed trajectories are not a forecast of resilience under every future storm.
11. Probabilistic modeling of wind-borne debris interactions for pre-disaster hurricane damage prediction in coastal communities
Core Problem: Building-by-building wind assessments omit debris that transfers damage between neighboring structures.
Key Innovation: A probabilistic generation, trajectory and damage chain incorporates community layout and is tested against Hurricane Laura damage in Lake Charles. Agreement improves relative to the specified HAZUS-based baseline. The reported gain is in damage-class accuracy for this testbed, not an observed reduction in disaster losses or proof of transfer to different building stocks.
12. Spatial distribution patterns of sediment transport after debris flow into the river: Insights from physical experiments in the Lower Yarlung Zangbo River
Core Problem: A debris flow entering a river changes sediment routing and deposition across the channel rather than ending at its outlet.
Key Innovation: The verified title identifies physical experiments on sediment-transport patterns in the Lower Yarlung Zangbo setting. Matching results are unavailable, so this scope summary does not assign deposition fractions, blockage thresholds or downstream transport distances.
13. Enhancing debris flow susceptibility studies through machine learning: Introducing a new data screening method based on geomorphological principles
Core Problem: Training-data selection can determine whether a susceptibility model reflects plausible geomorphic processes or artifacts in the inventory.
Key Innovation: The verified study introduces geomorphology-based screening for machine-learning debris-flow susceptibility analysis. With no reliable matching abstract, its specific screening rules, validation design and accuracy gains are not inferred.
14. A novel model for effective erosion rate (EER) in debris flow susceptibility assessment: integrating InSAR and hydrological connectivity
Core Problem: Susceptibility analysis needs a representation of sediment supply that connects surface change to hydrological pathways.
Key Innovation: The verified effective-erosion-rate study integrates InSAR and hydrological connectivity. Title-level evidence establishes that modelling scope, without supporting a specific erosion formula, field validation result or improvement over alternative conditioning factors.
15. Reconstruction of the May 2025 long-runout rock-ice avalanche in the Swiss Alps
Core Problem: Reconstruction of the May 2025 Swiss Alps event can help constrain how a large mixed rock-and-ice mass propagated through steep terrain.
Key Innovation: The article identity and authors are verified, but a matching abstract was not retrieved. The entry therefore identifies the reconstruction without importing volumes, velocity estimates or mechanisms from other papers about the same event.
16. Liquefaction susceptibility assessment using joint inversion of Rayleigh and love waves: A case study from the Granada Basin (Spain)
Core Problem: Subsurface shear-wave structure used in liquefaction screening is sensitive to the wave types and inversion constraints available.
Key Innovation: The verified Granada Basin case jointly inverts Rayleigh and Love waves for liquefaction-susceptibility assessment. A matching abstract is unavailable; neither improved layer resolution nor a change in mapped susceptibility is asserted from the title alone.
17. FloodDuo: a dual foundation-encoder architecture for generalizable high-resolution flood mapping
Core Problem: High-resolution flood models need to retain useful representations when geographic conditions and image sources change.
Key Innovation: The verified FloodDuo record proposes a dual foundation-encoder architecture for generalizable flood mapping. It is retained as a directly relevant method, with title-level evidence that does not establish cross-event superiority, training efficiency or operational warning capability.
18. Reconciling flood control and hydropower objectives through optimized spatially-explicit capacity equivalency of cascade reservoirs
Core Problem: Cascade-reservoir operation must reconcile spatially distributed flood-storage functions with hydropower objectives.
Key Innovation: The verified study addresses optimized, spatially explicit capacity equivalency across reservoirs. Without a matching abstract, no flood-peak reduction, energy gain or operating rule is assigned to the proposed approach.
19. The coupling effect of varying water-sediment conditions and levee spacing on the alluvial river-A case study of the Lower Yellow River
Core Problem: Water and sediment conditions can interact with levee spacing to alter an alluvial river's response.
Key Innovation: The verified Lower Yellow River case considers these coupled controls. Title-level evidence does not establish a preferred spacing, conveyance benefit, sedimentation rate or resulting flood-risk reduction.
20. A spatiotemporal graph-based robust optimization framework for data-driven disaster recovery scheduling
Core Problem: Recovery schedules must accommodate spatially uneven demand, limited resources and new disruptions rather than a fixed deterministic sequence.
Key Innovation: A spatiotemporal graph, demand model and robust rolling-horizon optimization are evaluated on a Türkiye–Syria earthquake case. Reported scheduling gains concern computational experiments, including synthetic disruptions and centralized coordination. Field performance with disrupted communications and humanitarian operations remains to be demonstrated.
21. The 2009–2010 Ouazzane seismic sequence: evidence for a possible active horsetail-like termination of the Jebha Fault, Rif belt, Northern Morocco
Core Problem: Diffuse seismicity near the end of a mapped fault can signal distributed deformation without proving a continuous active connection.
Key Innovation: Relocation, declustering, focal mechanisms and regional geology support a possible horsetail-like Jebha termination near Ouazzane. The inferred branches remain conceptual concealed structures rather than directly mapped active fault surfaces, constraining how the interpretation should enter seismic-hazard models.
22. Terrain-Modulated Multi-Polarimetric Fusion for Efficient Volumetric Radar Nowcasting
Core Problem: Volumetric radar prediction must retain vertical storm structure without the cost of dense three-dimensional operations.
Key Innovation: TerraCast3D combines grouped-altitude encoding, polarimetric inputs and terrain conditioning, improving intense-echo categorical scores in GridRad-Severe comparisons. The reported 40 dBZ CSI is 0.3044, leaving substantial event error. Reflectivity skill and lower arithmetic cost do not themselves establish rainfall accuracy, warning reliability or deployed throughput.
23. Sensitivity Experiments on Assimilating GIIRS Temperature and Humidity Retrieved Profiles for Three Typhoons Affecting Zhejiang
Core Problem: Satellite profile assimilation can modify both the cyclone vortex and its surrounding steering environment, with different effects on track and rainfall.
Key Innovation: Three Zhejiang-affecting typhoon cases compare clear-, cloudy- and all-sky GIIRS assimilation, including vortex relocation tests. All-sky experiments perform best for the reported cases, while narrower sampling can be adverse. The small case set supports conditional sensitivity findings rather than a universal operational advantage.
24. A Comparative Study of Convolutional and Recurrent Neural Networks for Storm Surge Prediction in Tampa Bay
Core Problem: Model ranking on ordinary water-level variation may differ from ranking during the rarer storms most relevant to coastal hazards.
Key Innovation: A Tampa Bay comparison of LSTM, CNN and hybrid models finds different advantages for overall trends and a hurricane subset. Reported error values depend on the data and windows, and lower average error alone does not establish reliability during extreme surges. This is a newly screened preprint record whose original identifier dates to 2024.
25. Important but Overlooked Reactive Organic Carbon (ROC) Air Pollution From Permafrost Wildfires in Canada
Core Problem: Deep organic-soil burning can emit reactive compounds whose effects are large relative to their share of emitted carbon.
Key Innovation: Three fire inventories, GEOS-Chem simulations and observations are used to assess Canadian permafrost wildfire emissions in 2023. Modelled volatile organics and secondary aerosols contribute to persistent pollution. The findings concern emissions and air-quality impacts, rather than predicting ignition, spread or population-specific health outcomes.
26. A Sentinel-2 semantic segmentation dataset for thermokarst-associated lake and pond water surfaces in Arctic permafrost lowlands
Core Problem: Small Arctic water bodies require georeferenced training pairs with clear treatment of ambiguous margins and geographic separation.
Key Innovation: The release provides 5,513 Sentinel-2 image–mask pairs across 23 selected lowland areas, with QA and split information. A geometry audit and revised holdout tests expose overlap sensitivity. Reported IoU measures agreement with released targets; independent human-reference accuracy remains unestablished, and visible summer water is not direct evidence of active thermokarst or future ground failure.
27. OceanTracker 0.5: fast adaptable Lagrangian particle tracking in structured and unstructured grids
Core Problem: Large particle ensembles can make transport statistics and sensitivity studies expensive within coastal hydrodynamic models.
Key Innovation: OceanTracker combines efficient tracking, in-run statistics and a modular pipeline across supported structured and unstructured grids. Comparisons report substantial speed gains for the tested workloads and computers. Computational efficiency expands feasible ensembles but does not validate the underlying currents, particle behavior or a particular pollution-hazard scenario.
28. Influence of matrix strength and block content on the mechanical characteristics of block-in-matrix rocks (bimrock)
Core Problem: The effect of block content depends on matrix strength, so bimrock properties cannot be derived by simply adding separate material effects.
Key Innovation: Orthogonal experiments and response-surface analysis distinguish coupled strength regimes and changes in crack paths. The reported 45–55% transition band applies to the tested weak-matrix materials. It informs characterization and model selection without establishing a field-scale stability threshold for all block-in-matrix rock slopes.
29. Progressive fracture and energy response of granite subjected to rapid localized flame heating
Core Problem: Rapid localized heating can produce uneven fracture and energy changes within granite, relevant to interpreting heat-damaged rock.
Key Innovation: The verified study focuses on progressive fracture under flame heating. Matching results are unavailable; this title-level entry does not infer thermal thresholds, post-heating strength loss or wildfire-driven rockfall rates.
30. Nested multi-source data assimilation coupled with a hydrodynamic model for enhancing water-quantity monitoring accuracy
Core Problem: Sparse station observations limit reach-wide estimates of water level, discharge and velocity in regulated rivers.
Key Innovation: A lower Lancang framework combines manual measurements, radar estimates and nested hydrodynamic assimilation, with a reported validation correlation of 0.935. The case supports monitoring improvements at the tested station and reach. Correlation does not establish calibrated uncertainty, extreme-flood accuracy or independence from the hydropower setting.
31. A Dual-Stage Deep Learning Framework for Fine-Crack Segmentation in High-Resolution UAV Images for Bridge Inspection
Core Problem: Fine cracks in very large UAV images are easily missed, while full-resolution segmentation is computationally expensive.
Key Innovation: A screening stage retains candidate neighborhoods for graph-enhanced segmentation across a bridge-image collection. The approach improves the measured quality–cost balance, but an end-to-end crack IoU of 0.4110 still leaves substantial segmentation error. Crack masks are inspection evidence rather than a validated measure of structural safety or disaster damage.
32. Cross-Sensor Comparison of SWOT Level-2 Pixel Cloud and ICESat-2 ATL13 Water-Surface Elevations over Tibetan Plateau Lakes
Core Problem: Cross-sensor water-elevation differences combine measurement, sampling and temporal effects rather than isolating one satellite's error.
Key Innovation: Quality-controlled SWOT–ICESat-2 matches across 50 Tibetan Plateau lakes show decimeter-scale agreement with scene-dependent residuals. The reported mean RMSE of 0.231 m is a cross-sensor consistency measure, not absolute SWOT accuracy against ground truth or a lake-outburst warning threshold.
33. Quantifying aleatoric and epistemic uncertainty in remote sensing spectral super-resolution via Bayesian deep learning
Core Problem: Spectral super-resolution needs to distinguish variability in the observations from uncertainty associated with the learned model.
Key Innovation: The verified Bayesian deep-learning study addresses aleatoric and epistemic uncertainty in spectral reconstruction. With only title-level evidence, the entry does not claim calibrated intervals, reliable out-of-distribution detection or improved hazard classification.
34. Deep Bayesian Approaches for Interpretable Nonlinear Structural System Identification
Core Problem: Reconstructing a measured structural response is a different task from forecasting unobserved dynamics or identifying unique physical parameters.
Key Innovation: A nonlinear shear-frame benchmark compares deep Bayesian state-space approaches with different physical structure and inference schemes. The tutorial separates conditioned reconstruction, open-loop prediction and parameter robustness. Physical interpretability and posterior uncertainty do not guarantee predictive accuracy or calibrated field deployment.
35. Generative Refinement Learning for Temporal Interpolation in Scientific Machine Learning
Core Problem: Coarse temporal sampling leaves intermediate fields unresolved, and moving patterns do not fully determine changes in their physical values.
Key Innovation: iFLEX learns a diffusion residual around weighted temporal interpolation, tested on turbulence, radar precipitation and sea-surface temperature. Improvements concern reconstruction between two available observations. They should not be described as causal nowcasting, future-storm prediction or recovery of independently observed fine-scale detail.
36. RangeRectQuant: Compressing Scientific Fields for Every Future Rectangular Sum
Core Problem: A raster can be compressed before future regions of interest are known, making query-specific error control impractical.
Key Innovation: RangeRectQuant coordinates integer pixel changes so every axis-aligned rectangular sum meets a prescribed budget while preserving range and designated samples. Geographically held-out built-surface and water maps test storage savings. The guarantee concerns rectangular totals, not arbitrary polygons, pixel morphology or hazard-classification accuracy.
37. Bayesian Inference of Expensive Simulators: Survey and Comparison of Design Choices
Core Problem: Expensive forward models restrict the number of runs available for Bayesian parameter inference, and a poor surrogate target can waste that budget.
Key Innovation: A unified review and comparison examines design choices for Bayesian optimization-based inference, emphasizing information-preserving proxy responses over a scalar log-likelihood in the studied problems. Useful simulator-specific guidance does not make a single proxy universally optimal or demonstrate field-calibrated hazard uncertainty.
38. Scale-aware and spatially-localised optimisation of kilometre-scale MLWP models
Core Problem: Kilometre-scale prediction losses can preserve spectral structure while sacrificing the location of features important for local weather assessment.
Key Innovation: A spatially localized complex-wavelet objective improves multiscale structure and neighborhood scores in an experimental global–regional FastNet setup. Sharper precipitation and more realistic intensity distributions are benchmark results, rather than evidence of calibrated extreme-rainfall probabilities or improved landslide warnings.
39. Assessing BIOMASS Radiometric Stability using Permanent Scatterers
Core Problem: Apparent temporal backscatter changes can combine scene evolution with instrument or processing effects.
Key Innovation: BIOMASS permanent-scatterer calibration tests limited acquisitions, then examines Nairobi and Cordoba stacks alongside dedicated calibration measurements. A processing-related anomaly illustrates the diagnostic value. Results depend on scatterer selection and signal conditions; relative radiometric stability is not interferometric displacement accuracy or a direct hazard-detection result.
40. LESSViT: Robust Hyperspectral Representation Learning under Spectral Configuration Shift
Core Problem: Changes in wavelength coverage and band sampling can invalidate representations learned under one spectral configuration.
Key Innovation: LESSViT combines low-rank spatial–spectral attention, wavelength-aware input handling and masked pretraining, evaluated under simulated cross-spectral shifts. The experiments support robustness in that protocol. Simulated configuration changes are not a complete test of real sensor transfer, atmospheric differences or downstream disaster mapping.
41. Expert lithological labels do not improve learning when used to predict marine sediment properties
Core Problem: Expert categories may add little supervision when they are themselves derived from the same measurements used as model inputs.
Key Innovation: A masked embedding model uses measurements from 296 ocean-drilling boreholes and tests withheld properties at unseen holes. Adding lithological labels does not improve the studied tasks and can degrade reconstruction. This result concerns a specific label–measurement relationship and does not show that independent geological knowledge or labels are generally unhelpful.