TerraMosaic Daily Digest: November 1, 2026
Daily Summary
A warning statistic is meaningful only when its target is clear. The Fujian crackmeter study separates genuine from false threshold-triggered alerts using asynchronous rainfall and displacement observations; it does not test the detection of every slope failure. A wider review of rainfall-induced landslide warning systems shows why forecasting, operational warning and community preparedness must also be distinguished. Across both scales, the practical question is which evidence reaches a decision, for whom, and under what conditions.
Two inventory studies extend the evidence base in complementary directions. HazMiner extracts located and timed geohydrological events from multilingual news, improving documentation where existing records are sparse. A global rock-glacier preprint maps more than 300,000 polygons under common morphological criteria. The first remains conditioned by what is reported; the second provides an occurrence baseline rather than a map of present activity or failure probability. Their value lies in expanding comparable observations while preserving those distinctions.
Independent comparisons expose limits that headline accuracy can conceal. Alternative GRACE solutions and GNSS observations do not support the proposed large-scale precursors to the Tohoku and Maule earthquakes. Spatial validation along the Fujian–Guangdong border separates an inventory-based landslide model from a weakly discriminating terrain-only alternative. In flood mapping, matched experiments on Sen1Floods11 find a much larger contribution from input preparation than network architecture, followed by substantial performance loss on unseen events.
Hydrological connections shape the conditions a warning system must face. Indian reservoir modelling combines substantial peak attenuation with rapid post-peak storage saturation, raising questions about sequential inflows. Godavari floods depend on different rainfall accumulation periods across sub-basins, while European routing experiments show how river networks redistribute drought extent. In frozen fractured rock, UPMFrost adds a smaller-scale counterpart: drainage persistence depends on which connected pathways freeze, not simply on the total amount of frozen material.
Key Trends
The selection links observation coverage, clearly defined warning targets and tests of whether a result survives a change of product, location or physical pathway.
- Warning targets need explicit denominators: Alert classification in Fujian, system-level review through Early Warnings for All, and corridor rockfall screening address different decisions. Precision among triggered alerts, operational service status and qualitative risk consistency cannot substitute for one another.
- Larger inventories still need coverage tests: HazMiner expands multilingual event documentation, while the rock-glacier preprint standardizes global morphological mapping. Reporting bias, extraction errors and omitted features remain distinct from the precision of retained records; inventory extent alone does not establish complete hazard coverage.
- Replication can overturn an attractive signal: The GRACE/GNSS study challenges a specific earthquake-precursor interpretation. Spatially separated landslide validation and held-out flood events likewise test whether apparent skill persists beyond the observations or products that generated it.
- Inputs and reference support shape apparent skill: Flood mapping is sensitive to radiometric normalization, river-stage retrieval to bathymetry and environmental conditions, and drought occurrence to catchment landscape context. Their results do not reduce to one architecture ranking, one native pixel size or one universal vulnerability index.
- Connectivity governs storage and response: Reservoir capacity after a peak, rainfall accumulation across Godavari sub-basins, river-network drought routing and freezing of connected fractures all show why the location and timing of storage or blockage matter alongside bulk state variables.
Selected Papers
Landslide warning, multilingual event documentation and a global rock-glacier inventory lead the selection. Geodetic hypothesis testing, corridor surveys and spatially independent mapping comparisons examine how observations support hazard decisions. Hydrological and coupled-process studies extend the analysis to storage, routing and subsurface connectivity, with supporting methods placed according to their relevance scores.
1. A global inventory of rock glaciers
Core Problem: Regional rock-glacier inventories differ in mapping criteria and completeness, obstructing comparisons of mountain permafrost terrain and priorities for field monitoring.
Key Innovation: A common imagery–segmentation–expert workflow maps 301,524 polygons across 19 regions. Independent inspection of 3,015 polygons yields 90.9% feature-level morphological precision. This is an occurrence baseline: neither that precision nor agreement with modelled permafrost establishes inventory recall, ground-ice volume, present motion or instability probability.
2. Reducing false alarms in crackmeter-based early warning of small-scale slope deformation via deep learning-driven asynchronous displacement–rainfall data fusion
Core Problem: Threshold-triggered crackmeter alerts include noise and benign displacement, while rainfall and displacement arrive at different sampling rates.
Key Innovation: A dual-branch temporal Transformer fuses asynchronous observations from Fujian slopes. Sensor-separated tests of field/expert-labelled alerts report 90.32% precision and 89.38% recall. These metrics concern discrimination after a threshold alert; they do not establish detection of all slope failures or calibration of failure probability. External settings require validation.
3. Review article: rainfall-induced landslide early warning systems through the lens of the early warnings for all framework – advances, gaps, and perspectives
Core Problem: A forecasting model is only one component of an operational warning system; service coverage also depends on communication, preparedness and institutional capacity.
Key Innovation: Assessment through the Early Warnings for All pillars distinguishes 45 reviewed systems from 27 operational systems, including one forecasting-focused system. Rainfall thresholds dominate, while risk mapping, documented community engagement and forecast verification remain limited. The review argues for locally supported implementation rather than equating predictive skill with an effective warning service.
4. Unravelling information on impactful geo-hydrological hazard events with HazMiner, a multilingual text mining method developed through a global scale coverage application
Core Problem: News-derived hazard inventories are uneven across languages and regions, leaving important gaps in documented impacts.
Key Innovation: HazMiner applies paragraph-level multilingual transformers to about 6.37 million articles in 58 languages, extracting located and timed flood, landslide and flash-flood events, including 6,836 landslides. It extends documentation in data-scarce settings; media reporting and extraction coverage still differ from complete event incidence or independently measured hazard frequency.
5. Reevaluating Large‐Scale Deformation Precursors of Great Subduction Earthquakes: Insights From Satellite Gravity and GNSS Data
Core Problem: Proposed large-scale deformation precursors require replication across gravity products and consistency with independent geodetic measurements.
Key Innovation: A statistical comparison of GRACE solutions and GNSS observations finds no support for the proposed precursors before the 2011 Tohoku and 2010 Maule earthquakes. The disputed gravity signals cannot be distinguished from product-specific errors or non-tectonic variation. The result constrains this particular precursor interpretation, rather than excluding every possible earthquake precursor.
6. Comparative Evaluation of Statistical and Morphometric Models for Rainfall-Induced Shallow Landslide Susceptibility Mapping: A Case Study Along the Fujian–Guangdong Border
Core Problem: Inventory-based statistical susceptibility and terrain-only morphometric screening need comparison on the same independent geographic support.
Key Innovation: On spatially separated validation samples along the Fujian–Guangdong border, IV-LR achieves AUC 0.7537 versus 0.5225 for the morphometric alternative and captures 53.06% versus 23.89% of validation landslides in the top-ranked fifth of the area. Balanced case–control Brier scores are not automatically regional failure probabilities, and performance remains event- and inventory-dependent.
7. Integrating Multi-Source Geospatial Data and Evaluating Model Architecture for Deep Learning-Based Flood Inundation Mapping
Core Problem: Architecture comparisons can confound network design with radiometric preprocessing and event-specific evaluation.
Key Innovation: Matched models and repeated factorial experiments on Sen1Floods11 attribute about 95% of tested IoU variance to input representation and about 1% to architecture. No significant event-level difference separates the four convolutional/hybrid models; all lose 24–26 IoU points on held-out events. The variance allocation applies to these tested factors and data, not universally to flood mapping.
8. Rockfall risk management on transportation corridors: coupling risk assessment with terrestrial mobile LiDAR survey on the A22 motorway (South Tyrol, Italy)
Core Problem: Road operators need repeatable slope inventories and survey coverage across long corridors, including locations that vehicle-based scanning cannot observe.
Key Innovation: The A22 study combines qualitative risk ratings, probabilistic susceptibility and runout-informed vulnerability with a mobile LiDAR baseline. More than 70% of potential detachment zones are detected, with occlusion, range and slope geometry limiting coverage. A later rockfall supplies qualitative consistency evidence; one event does not calibrate absolute risk or validate all screening classes.
9. A Unified Pipe Network Model for coupled thermo-hydraulic transport in frozen fractured rock: Formulation, validation and application
Core Problem: Freezing can close selected fracture pathways and reroute drainage, which equivalent-continuum descriptions may obscure.
Key Innovation: UPMFrost couples porous-medium and discrete-fracture flow with advection, phase change and freezing-dependent aperture. Analytical tests, InterFrost comparisons and a published single-cleft experiment support the formulation. A field-derived network illustrates winter cutoff and partial summer recovery, but its simulated transition is not a universal site-warning threshold or a mechanically coupled failure prediction.
10. Impact of Reservoirs on Streamflow Extremes in India
Core Problem: Reservoirs may reduce individual flood peaks while retaining little spare capacity for the next inflow pulse.
Key Innovation: Integrating 223 reservoirs in an Indian modelling system yields 46.1% median monsoon peak attenuation, while 74% reach near-full capacity the next day after a high-flow peak. The combined result motivates attention to sequential inflows and operations. These modelled aggregate responses do not establish identical protection, capacity or forecast-informed operating benefits at every dam.
11. Urbanization Enhances Inner‐Core Rainfall During Tropical Cyclone Landfall
Core Problem: Urban land cover and anthropogenic heat can interact during cyclone landfall, making their rainfall effects difficult to separate.
Key Innovation: High-resolution simulations of Bebinca's 2024 Shanghai landfall distinguish roughness-related inflow/convergence from heat-related instability. Reported percentage increases use different counterfactuals: urban land-cover change versus cropland, and heat on versus off within each surface scenario. They must not be added as independent contributions or generalized as a universal urban rainfall multiplier.
12. Role of hydrometeorological variables and catchment area to flood generation over the monsoonal climate region of India
Core Problem: Basin-mean rainfall conceals differences in accumulation duration and the subregions contributing flood-producing runoff.
Key Innovation: Observed and VIC-simulated Godavari records associate semi-humid floods with roughly 10–11 days of accumulated rainfall, versus short accumulation and a strong single-day event in the semi-arid sub-basin. EOF analysis identifies disproportionate central/downstream contributions. These basin-specific associations guide duration-aware investigation; they are not portable warning thresholds without local testing.
13. Satellite-Derived Landscape Characteristics Distinguish Streamflow Drought Occurrence
Core Problem: Catchments under similar meteorological deficits can differ in whether a streamflow drought develops, and occurrence need not track severity.
Key Innovation: Across 361 relatively undisturbed US catchments, AlphaEarth landscape components distinguish estimated drought occurrence under comparable meteorological and initial streamflow conditions. Upper-versus-lower PC1 quartiles differ by 8.1–20.9 percentage points; evidence for intensity and duration is less conclusive. The observational associations do not identify a causal land-management intervention or one complete vulnerability ranking.
14. Generated storm lifecycles fill the rare-event tail of short rainfall records
Core Problem: Short rainfall archives underrepresent complete extreme-storm lifecycles needed for catchment stress tests.
Key Innovation: StormGen conditions hourly rainfall sequences on the observed opening of storms and reports a withheld-Harvey reconstruction and coherent 24-hour scenarios over southeast Texas. This supports conditional scenario generation, not an independently calibrated return period or a forecast before the conditioning rainfall is observed. An ambiguously formatted percentage in the source abstract is not repeated.
15. High Resolution Focal Mechanisms and Stress Field Orientation in Southern Italy From Short‐Term Dense Arrays
Core Problem: Short deployments must resolve small-event focal mechanisms well enough to improve regional fault and stress interpretation.
Key Innovation: A one-year southern Italian array yields 289 mechanisms and a predominantly normal-faulting stress regime, with stable principal orientations and possible seasonal stress-ratio variation. A catalogue comparable in count to several conventional-monitoring years does not imply equivalent coverage or prediction skill. Hydrological modulation remains an interpretation, not an isolated causal result.
16. A probabilistic spatiotemporal framework for landslide hazard assessment
Core Problem: The verified title targets a joint representation of where and when landslide hazards occur.
Key Innovation: A probabilistic spatiotemporal framework is proposed for landslide hazard assessment. A matching abstract was unavailable, so the probability definition, inventory construction, calibration, temporal validation and improvement over existing methods cannot be assessed from the available record.
17. From deglaciation to collapse: A thermo-mechanical reconstruction of permafrost control on the Köfels rock slide (Austria, 9.5 ka BP)
Core Problem: The Köfels rock slide offers a historical case for examining deglaciation, frozen-rock conditions and slope collapse.
Key Innovation: The verified study reconstructs permafrost control through a thermo-mechanical approach. Title-level evidence establishes the topic, but does not resolve the inferred thermal history, relative causal contributions, model fit or implications for modern slope-warning thresholds.
18. Next 24 h rockburst probabilistic warning using microseismic monitoring and machine learning
Core Problem: The verified paper targets a next-24-hour rockburst warning horizon using microseismic records and machine learning.
Key Innovation: A probabilistic warning formulation is indicated by the title. No matching abstract was available to assess event definitions, prospective testing, class imbalance, probability calibration or warning performance; a 24-hour horizon alone does not establish an operational lead time.
19. Effects of Mesoscale Fracture Networks on Injection-Induced Slip and Seismicity in Fault Systems During CO2 Geological Storage
Core Problem: Geological CO2 storage can interact with pre-existing fault and fracture networks.
Key Innovation: The verified title examines how mesoscale fracture networks affect injection-induced slip and seismicity. The available metadata do not establish whether a particular network increases or decreases risk, which hydraulic boundary conditions apply, or whether simulated event rates are validated against observations.
20. Reliability assessment of river-crossing pipelines under flood action considering dependence between yield and resonance failures
Core Problem: Yield and resonance failure modes may be dependent under flood loading.
Key Innovation: The verified study incorporates dependence in pipeline reliability assessment. The supplied text is an unrelated marine-structure thesis abstract and is excluded from interpretation. Model assumptions, dependence estimates, failure probabilities and mitigation benefits cannot be recovered from the verified title alone.
21. Selection of Ground Motions for Hazard‐Consistent Incremental Dynamic Analysis Using Genetic Algorithms
Core Problem: Hazard-consistent collapse analysis depends on how records represent site-specific spectral shape and duration, beyond the number of ground motions.
Key Innovation: A genetic algorithm selects 100 records for Wellington from 2,322 candidates and tests two steel-frame models. The gain concerns sampling precision and conditional-hazard alignment within the existing HC-IDA method. Building-height-dependent median changes and comparison-set differences preclude treating lower dispersion as a universal reduction in physical collapse risk.
22. Objective Ground Motion Model Selection and Ranking Using Likelihood and Bayesian Criteria: A Case Study for Türkiye
Core Problem: Logic-tree model selection depends on residual assumptions, data subsets and treatment of epistemic uncertainty.
Key Innovation: Likelihood and Bayesian criteria rank candidate ground-motion models using 12,160 Turkish recordings. DIC estimates residual dispersion while holding median functions and regression coefficients fixed. Magnitude, event and prior sensitivity are examined; the exercise is not a full Bayesian refit or a universal ranking valid for every Turkish site.
23. Sentinel-1 SAR Monitoring of River Water Surface Elevation: Effects of Wind, Waves, River Morphology and Vegetation
Core Problem: An inundation-area–stage relation needs submerged topography, while vegetation and water roughness alter SAR boundaries.
Key Innovation: At Mississippi's Thebes gauge, 265 Sentinel-1 images give best overall results for log10(VV), with 1.61 m RMSE, yet it is the most accurate metric in only 44.5% of acquisitions. Topobathymetry, flow regime and wind matter jointly. Finer resampling does not create new native-resolution information or consistently improve stage accuracy.
24. SNAPxtra: An Open-Source Python Workflow for Automated Multi-Temporal InSAR Deformation Monitoring with SNAP
Core Problem: Producing reproducible PS and SBAS stacks requires many preprocessing and export steps.
Key Innovation: SNAPxtra integrates master selection, multi-frame/swath handling, masks, trial processing and network generation, with exports for StaMPS and LiCSBAS and three case demonstrations. It automates a workflow; it does not remove atmospheric artefacts, guarantee stable targets or independently deliver validated warning thresholds.
25. Applications of Deep Learning for Correcting Tropospheric Phase Delays in InSAR: A Review
Core Problem: Atmospheric phase and real displacement can share spatial or temporal patterns, so smoother interferograms do not necessarily mean more accurate deformation.
Key Innovation: The review organizes deep-learning correction by spatial, temporal, generative and regression strategies. It emphasizes label reliability, deformation preservation, physical consistency and independent geodetic checks. It supplies evaluation priorities rather than a new universal correction algorithm or pooled estimate of performance.
26. Joint probability and interpretability analysis of urban infrastructure flood resilience: a case study of the Beijing-Tianjin-Hebei region
Core Problem: The verified title examines joint probability and interpretability in Beijing–Tianjin–Hebei infrastructure flood resilience.
Key Innovation: The bibliographic record establishes the regional resilience topic. Without a matching abstract, the dependence model, resilience indicators, explanatory method and validation cannot be judged; an interpretable index must not be assumed to be a calibrated flood-loss probability.
27. Analysis of drought and desertification dynamics in Diwaniya Governorate, Iraq using remote sensing and unsupervised machine learning
Core Problem: The verified study addresses drought and desertification dynamics in Iraq's Diwaniya Governorate.
Key Innovation: Remote sensing and unsupervised machine learning provide the stated analysis approach. No matching abstract was available to establish the time window, trend magnitude, drought definitions, reference data or attribution to climate and land use.
28. Global-to-local seismic performance analysis for three-layer ring spring flexible shield tunnel structures
Core Problem: Seismic performance can differ between overall tunnel motion and local demand in flexible multi-layer structures.
Key Innovation: A global-to-local analysis of three-layer ring-spring shield tunnels is identified in the verified title. The supplied record does not establish joint behaviour, damage thresholds, validation quality or an improvement over conventional segment designs.
29. 3D seismic wavefield evolution and adverse geology imaging in shield tunnels: Large-scale physical modeling and numerical insights
Core Problem: Three-dimensional wave propagation affects how adverse ground ahead of a shield tunnel can be imaged.
Key Innovation: The verified study combines large-scale physical modelling and numerical analysis of seismic wavefields. Without a matching abstract, detectable feature sizes, resolution, uncertainty and field transfer remain unknown; imaging adverse geology is distinct from predicting a collapse event.
30. Multi-objective stochastic optimisation of lead-rubber-isolated elevated liquid storage tanks with supplemental clutching inertial system considering soil-structure interaction
Core Problem: Isolation, sloshing and force transmission compete, with soil–structure interaction altering the design trade-offs.
Key Innovation: A stochastic multi-objective design combines lead-rubber bearings and a clutching inertial system, followed by checks with 11 near-fault motions. Displacement-focused and base-shear-focused solutions trade different demands rather than improving every response simultaneously. Benefits remain conditional on soil, tank geometry and the adopted linearization and design strategy.
31. Stochastic site response analysis of Rayleigh wave in soil stratum with spatially varying shear wave velocity
Core Problem: Spatially variable shear-wave velocity can change surface response and its spatial dependence.
Key Innovation: Two-dimensional random fields and finite elements test Rayleigh-wave propagation after a deterministic Thin Layer Method comparison. Velocity variability dominates the tested response dispersion, while correlation-length effects are weak and non-monotonic in this setup. That result is not a general reason to discard spatial correlation in other wavefields or geologies.
32. Seismic resilience assessment of reinforced concrete moment-resisting frames considering embodied carbon and recovery performance
Core Problem: The verified title combines recovery performance and embodied carbon in frame resilience assessment.
Key Innovation: The work broadens the stated assessment objectives beyond structural response. No matching abstract was available to determine damage, repair, carbon or recovery models, their validation, or whether any design reduces both environmental cost and seismic risk.
33. Path duration of ground motion in shallow crustal earthquakes based on the generalized reflection/transmission coefficient matrix method
Core Problem: Path-related duration is part of how seismic demand accumulates beyond a single amplitude measure.
Key Innovation: The verified title uses a generalized reflection/transmission coefficient matrix method. Available metadata do not establish duration predictors, regional calibration, fit statistics, frequency dependence or implications for site-specific design motions.
34. Strength–stiffness divergence and fracture-inclination effects in frozen green sandstone: implications for surrounding rock stability in cold-region tunnels
Core Problem: Strength and stiffness may evolve differently as fractures and freezing affect surrounding rock.
Key Innovation: The verified title examines divergence between strength and stiffness and fracture-inclination effects in frozen green sandstone. Without matching abstract evidence, the trend directions, temperature range, tested geometry and transfer to in-situ tunnel stability are not inferred.
35. SATFRONT: Detecting Atmospheric Fronts with Satellite Data using Machine Learning
Core Problem: Manual front analyses vary across services, complicating automated detection from rapid satellite observations.
Key Innovation: SATFRONT detects European cold, warm and occluded fronts and reports CSI 0.428 from satellite data at 150 km tolerance, versus 0.591 when ERA5 is added. This explicitly tolerance-dependent comparison against service analyses does not establish equally precise rainband location or flood forecast skill.
36. Chronos-2: A foundation model for multivariate time series forecasting
Core Problem: Forecasting multiple interacting variables and heterogeneous covariates usually requires domain-specific model fitting.
Key Innovation: Chronos-2 extends zero-shot probabilistic forecasting to cross-series interactions and covariates, with broad benchmarks and a runoff application. Reported average gains and GPU timings motivate hazard-domain evaluation, but neither establish extreme-flow calibration nor transfer under every spatial, temporal or covariate shift.
37. How the River Network Shapes Spatiotemporal Drought Behavior Across Europe
Core Problem: Drought extent in routed rivers differs from the footprint of runoff deficits that generated it.
Key Innovation: European simulations compare unrouted runoff with routed discharge to isolate a modelled network effect. Routing usually enlarges drought extent, while lakes and reservoirs can decouple runoff and discharge through storage. The controlled model comparison clarifies propagation, not a universal observed effect size for every basin or drought definition.
38. Sparse‐Well Diffusion Reconstructions Enable Groundwater Transport Equation Recovery and Plume Refinement
Core Problem: Sparse monitoring-well reconstructions may look plausible while losing the transport structure needed for plume interpretation.
Key Innovation: SW-DSRR links diffusion reconstruction to sparse equation recovery and consistency-guided refinement under tested well layouts, heterogeneity and sources. Checks include mass, migration and non-negativity. It offers a method for subsurface diagnosis; no landslide warning or universal recovery of the true field equation is demonstrated.
39. Experimental Evidence of Reaction‐Driven Fracturing and Ultramafic Rock Weakening Induced by CO 2 Mineralization
Core Problem: Mineral formation can alter fracture pathways and strength during subsurface storage.
Key Innovation: Tomography and displacement correlation follow serpentinized dunite exposed to reactive brine, with cracking progressing along existing veins and flaws. Mineral observations and calculated crystallization pressure support a reaction-driven interpretation. Specimen-scale conditions and modelled pressures are not a field-scale leakage or induced-seismicity forecast.
40. Drone-PUR: From Visual Clue Discovery to Understanding and Reasoning in UAV Scenes
Core Problem: Aerial reasoning can fail because small objects, view changes and cross-modal observations are perceived incorrectly.
Key Innovation: Drone-PUR separates perception, understanding and reasoning across 11 tasks with human-reviewed tests. Fine-tuned variants improve benchmark multiple-choice accuracy. The resource may inform inspection of disaster imagery, but its scores do not validate disaster detection, geographic transfer or operational decision quality.
41. Frozen in Time: How a Physics Foundation Model Forgets Under Sim-to-Real Adaptation
Core Problem: Adapting a continuum-dynamics foundation model to a new simulation and then real measurements can erase earlier skills.
Key Innovation: A Walrus-1.3B cylinder-flow study tracks forgetting and finds that replaying 1% simulation data during real-data adaptation recovers much of the lost simulation performance. This concrete retention test is relevant to future hazard emulators, but is not itself a landslide application or a guarantee across physical systems.
42. When Does Jacobian Fine-Tuning Stabilize Neural PDE Emulators?
Core Problem: Autoregressive error includes both physical instability and excess amplification caused by the emulator.
Key Innovation: A solver-Jacobian counterfactual separates these contributions in small neural models across four PDEs. Fine-tuning gains concentrate where excess amplification is high; directional Jacobian matching and input-noise injection are compared. The tested amplification diagnostic is not a universal stability certificate for large hazard models.
43. Know When the Equation Is Wrong: Structural Uncertainty for LLM-Guided Symbolic Regression under Extrapolation
Core Problem: Several equations can fit observations similarly while diverging outside the fitted range.
Key Innovation: LLM-SR-UQ retains competing functional structures and uses disagreement to widen uncertainty under extrapolation. Benchmark improvements do not restore distribution-free conformal guarantees under arbitrary shift, prove causal identification or ensure the retained equation set contains the true mechanism.
44. PDE-Residual-Informed Kriging
Core Problem: Pointwise equation constraints may miss residual behaviour over the rest of the domain.
Key Innovation: PDE-residual-informed Kriging builds integrated residual information into its covariance and derives posterior-mean and variance results under stated regularity and stability assumptions. It is a mathematical and numerical contribution; residual energy and contraction should not be treated as field-calibrated hazard uncertainty without additional validation.
45. Beyond IID: How General Are Tabular Foundation Models, Really?
Core Problem: IID tabular benchmarks can hide failures on temporal, grouped and larger datasets relevant to hazard mapping.
Key Innovation: BeyondArena and DataFoundry compare 11 models across 142 datasets, finding that tabular foundation models' advantages depend on task and scale, with conventional approaches remaining strong on grouped or large data. This motivates appropriate holdouts; it does not establish a best model for a particular landslide inventory.
46. Network-based subbasin-scale mapping of streamflow alteration in Ontario, Canada
Core Problem: Incomplete operating records make regulated-versus-natural labels difficult to interpret along connected rivers.
Key Innovation: Ontario's Streamflow Alteration Index propagates drainage-area-weighted exposure to mapped alteration sources. It is a proxy for potentially affected flow, not the magnitude or direction of hydrological change. Incomplete correspondence with regulated-gauge labels reinforces that screening and actual flow-response estimation are different tasks.
47. NH-SDN: A Deep Learning-Derived Daily 0.25° Snow Density Dataset for the Northern Hemisphere from 1982–2025
Core Problem: Long, spatially continuous snow-density records are needed for snow-water and thermal studies.
Key Innovation: NH-SDN provides daily 0.25-degree Northern Hemisphere estimates for 1982–2025, with reported improvements against reanalysis and in-situ references. Its coarse grid and bulk density cannot directly resolve slope-scale snow stratigraphy or avalanche stability; validation and mountain representativeness need to accompany any downstream use.
48. Technical note: Benefits of Bayesian estimation of model parameters in a large hydrological model ensemble
Core Problem: Changing parameter calibration can change apparent model-structure rankings within large ensembles.
Key Innovation: Bayesian parameter estimation benefits the interpretation of complex individual models, yet does not substantially improve the tested multi-model predictions because structural diversity mitigates misspecification. The result distinguishes model-comparison inference from predictive benefit, without implying parameter uncertainty is unimportant in every catchment.
49. Technical note: Evaluation of snow water equivalent from large-scale land-surface products over Italy
Core Problem: Long reanalysis records trade broad coverage against forcing and topographic biases in mountain SWE.
Key Innovation: Comparison with IT-SNOW finds no product best on every metric: ERA5-Land tracks timing well but overestimates Alpine SWE, while finer-grid VHR-REA_IT has broad underestimation and weaker correspondence. Nominal grid spacing is therefore an insufficient basis for product selection; the reference dataset also defines the comparison's support.
50. A Second-Order Spatial Guided Sampling Method for Outlier Removal in UAV Remote Sensing Image Registration
Core Problem: Repetitive textures and viewpoint changes can create locally plausible but structurally unsupported image matches.
Key Innovation: SPGSC++ uses second-order common-neighbour support for sampling and residual-guided refinement, improving aggregate matching metrics across five datasets. Homography-based correspondence accuracy is useful preprocessing, but does not establish reliable 3D displacement on deforming slopes or eliminate relief and occlusion effects.
51. Seismic wave propagation in a viscoelastic layer with exponentially varying bounded shear modulus
Core Problem: The verified study represents a viscoelastic layer whose bounded shear modulus varies exponentially with depth.
Key Innovation: The title establishes the material profile and wave-propagation problem. No matching abstract was available to assess boundary conditions, damping, solution verification, amplification trends or the geological range in which this profile is appropriate.
52. A Comparison of Tensile Testing Techniques and Predictive Correlations
Core Problem: Indirect tensile strengths depend on loading arrangements and may not equal direct tensile strength.
Key Innovation: Digital-image-correlation tests compare platen shapes and cushions and show that curved-platen Brazilian tests approximate direct tensile tests best in this programme. Loading arc angle must be retained in comparisons; crack-initiation strength and Hoek–Brown parameter correlations are less reliable. These are material-test findings, not a slope safety assessment.
53. Streamflow prediction in natural ungauged catchments using Temporal Fusion Transformers
Core Problem: Streamflow models must transfer across natural catchments without local calibration data.
Key Innovation: Across screened Portuguese catchments, TFT improves ungauged median NSE, KGE and CRPS over calibrated HBV in the reported comparisons; specialization and limited local retraining add gains in separate tasks. Those tasks must not be pooled, and daily-flow skill does not by itself establish extreme-peak accuracy or real-time flood-warning performance.
54. Diffusion-Inversion-Net (DIN): an end-to-end direct probabilistic framework for characterizing hydraulic conductivities and quantifying uncertainty
Core Problem: The verified title targets probabilistic characterization of hydraulic conductivity and uncertainty.
Key Innovation: Diffusion-Inversion-Net is described as an end-to-end direct framework. No matching abstract was available to assess conditioning observations, prior assumptions, posterior calibration, non-uniqueness or performance on independent hydrogeological fields.
55. Investigating the spatial patterns in the dependence of daily rainfall depth on wet spell duration over Europe
Core Problem: The relation between daily rainfall depth and wet-spell duration can vary geographically.
Key Innovation: The verified paper investigates that dependence over Europe. Metadata alone do not establish its sign, functional form, seasonal differences or implications for antecedent-rainfall thresholds and design storms.