TerraMosaic Daily Digest: October 29, 2026
Daily Summary
A slope can retain damage from an earlier earthquake without failing, leaving less resistance for a later event. The loess study organizes this inherited state by geological architecture and distinguishes later reactivation from a slope becoming critical between events. A separate reservoir-fault model finds a related history effect: pressure recovery can reload a previously slipped fault even below its historical maximum pressure. Both studies make loading history part of the assessment, within different physical systems and modelling assumptions.
Monitoring and forecasting constrain different parts of that evolution. Camera attitude correction improves displacement measurement in a field campaign, while the Zaoshuwa forecasting framework tests multiple horizons with rolling preprocessing and scenario rainfall. Regional InSAR learning and microseismic graph analysis add spatial and temporal information, but measurement agreement, displacement error and a precursor association each need their own interpretation before they can support a warning decision.
Flood and drought studies examine how information is aggregated across time and space. Annual-maximum residual correction moves process-based flood quantiles toward long-record benchmarks in three Canadian watersheds. Detention-basin rules make the allocation of modelled flood losses explicit, while Amazon lake observations reveal uneven storage deficits and timing during drought. The exposure-period slope metric, flood return quantile and observed storage anomaly answer different questions and should retain their definitions.
Hazard-learning studies also expose limits hidden by a single average score. AFSeq broadens active-fire sequence evaluation toward shorter and smaller events. Building-damage experiments find that neighborhood context can override evidence from the building itself. Seismic pretraining comparisons redistribute skill between tasks, and groundwater-transfer tests find a trade-off between average error and the number of wells harmed. These results support evaluation by event, location, target and failure mode alongside headline performance.
Key Trends
Inherited state, observation support and validation design connect the strongest studies across slope, flood, fire and infrastructure hazards.
- Loading history changes the remaining resistance: Successive-earthquake loess analysis and depleted-reservoir fault modelling retain the changes left by earlier loading. The transferable point is the need to preserve state; calculated reactivation boundaries remain conditional on architecture and model assumptions.
- Monitoring accuracy and forecast skill need separate tests: Attitude-corrected images address false apparent movement. Multi-horizon displacement models address future trajectories, while magnitude fusion and microseismic indicators address other targets. None of these performance measures alone validates a complete early-warning system.
- Aggregation changes the risk question: Flood-frequency correction operates at annual maxima, detention rules optimize an assumed loss function, and satellite lake estimates resolve distributed storage changes. Keeping these supports explicit helps prevent a fitted score or model statistic from becoming an unqualified event probability.
- Transfer evaluation reveals who loses: Groundwater blending can protect wells while changing pooled error, building-damage scores can hide isolated-building failures, and a seismic objective can improve one task at another's expense. Forward, event-grouped and spatially disjoint tests reveal different weaknesses.
- Inventory definitions determine coverage: The Caucasus assessment scores a subset of its mapped mountain lakes, AFSeq represents observed active fire, and fuel-moisture models primarily test held-out dates at represented sites. Class definitions and observation conditions matter when these resources are used in a new region or hazard workflow.
Selected Papers
The selection follows how physical state is inherited, how movement and hazard conditions are observed, and how models behave beyond their fitted examples. It brings together slope and fault mechanisms, flood and fire assessment, practical monitoring, and specific supporting methods. Papers are ordered by relevance score in one continuous list.
1. Engineering-geological architecture governs inherited-state evolution and reactivation-threshold erosion of loess landslides under successive earthquakes
Core Problem: Later earthquakes act on slopes already changed by earlier shaking and the intervening evolution of resistance.
Key Innovation: State-preserving comparisons of three geological architectures distinguish persistent subcritical damage, later-event reactivation and a state becoming critical before that event. State patterns transfer more consistently than numerical transition boundaries. This supports architecture-aware interpretation of remaining resistance, while the calculated acceleration brackets remain conditional on the modelled endmembers and histories.
2. A kinematics-informed dual-stream deep learning framework for multi-timescale landslide displacement forecasting and risk-informed decision support
Core Problem: Short- and long-horizon displacement forecasts need deformation-state information without leaking future observations into preprocessing or rainfall inputs.
Key Innovation: A dual-stream temporal–kinematic model uses causal rolling decomposition, scenario rainfall and an unseen monitoring target at Zaoshuwa. Tests span one to ninety days. The kinematic features are interpretable descriptors rather than mechanical constraints; performance at one additional target does not establish regional transfer or calibrated warning thresholds.
3. A real-time visual monitoring method for landslide creep using platform attitude self-calibration
Core Problem: Camera-platform vibration and settlement can appear as movement of the monitored slope.
Key Innovation: An optical tracker and inertial sensor separate attitude-induced image motion from target displacement. Indoor tests and a 93-day Dayangshan campaign report improved measurement agreement against references. The results validate monitoring under the tested setup, rather than the ability to forecast collapse or maintain the same precision under every visibility and weather condition.
4. Hybrid deep learning approach for predicting long-term deformation of landslides
Core Problem: InSAR deformation sequences and spatial conditioning factors carry complementary information about continuing slope movement.
Key Innovation: A CNN–LSTM–attention model is evaluated with separate development, validation and independent test data in central Italy. Low held-out errors support the reported task. Radar geometry, sampled landslides and the test protocol still delimit claims about regional transfer, physical failure mechanisms and prospective intensification warnings.
5. AMFP-based probabilistic risk assessment of unsaturated slopes under stochastic rainfall
Core Problem: Rainfall occurrence, intensity–duration dependence and uncertain soil properties jointly affect exposure-period slope assessment.
Key Innovation: A copula, coupled seepage–stability model and surrogate comparisons estimate average maximum failure probability for a homogeneous Baotou slope. The reported AMFP changes with exposure duration and parameter assumptions. This model-defined statistic should be kept distinct from an independently observed lifetime failure rate or an unqualified cumulative-risk estimate.
6. Dynamic evolution of rock slope stability based on representative volume element and shear damage mechanics
Core Problem: Peak or residual strength alone cannot describe the evolution of a shear zone while a rock slope deforms.
Key Innovation: RVE-scale joint representations and numerical shear tests inform a damage-dependent strength law coupled to vector-sum stability analysis. The method links displacement and resistance evolution within the model. Its proposed monitoring integration remains a future application, rather than a field-demonstrated real-time warning system.
7. Glacial Lake Inventory and GLOF Hazard Assessment in Caucasus Region
Core Problem: Regional screening needs to distinguish glacial lake types and their susceptibility-relevant setting from the full population of mountain water bodies.
Key Innovation: A multitemporal Caucasus inventory identifies 1,384 lakes above the stated elevation and size limits, then scores 411 glacier-associated lakes using six criteria. The classification provides screening priorities. High-hazard labels are criterion-based categories, not annual outburst probabilities, and inventory thresholds and image dates bound completeness.
8. The impact of an in-situ stress gradient on the growth of a fluid-driven frictional rupture
Core Problem: Depth-varying stresses can make a fluid-driven rupture grow asymmetrically even under symmetric injection pressure.
Key Innovation: A planar constant-friction model derives coupled rupture-tip equations and compares semi-analytical results with numerical solutions. The stress gradient changes the relation between rupture and diffusion fronts. The calculation concerns an idealized aseismic-slip setting, not direct validation of earthquake magnitude, dynamic rupture or a field-safe injection limit.
9. Rapid earthquake magnitude estimation from fuzzy C-means clustering-weighted fusion of seismogeodetic PGD and PGV
Core Problem: Magnitude estimates based on one seismogeodetic peak measure can be sensitive to anomalous observations and station geometry.
Key Innovation: PGD–PGV fusion is tested across nineteen earthquakes with leave-one-event-out and two-event holdouts. Fusion provides the main improvement, with smaller event-dependent benefits from additional weighting. These magnitude errors do not alone establish real-time warning latency, complete station-outage robustness or improved tsunami predictions.
10. Magnitude scaling and ground motion factors for subduction zone earthquakes
Core Problem: Earthquake magnitude alone does not fully represent the number of damaging cycles in subduction-zone motions.
Key Innovation: Soil-dependent mixed-effects models combine magnitude, duration, intensity and spectral or rupture predictors, tested with event-grouped cross-validation. The approach extends equivalent-cycle scaling within existing assessment procedures. Japan-heavy data support the reported regional checks but do not remove all geographic, source and soil-model uncertainty.
11. Unexpected fault reactivation during underground gas storage in depleted reservoirs: defining operational pressure bandwidths
Core Problem: Previous depletion-related slip can leave a stress state that makes pressure recovery hazardous below the historical maximum reservoir pressure.
Key Innovation: A coupled flow and quasi-static fault-contact model compares geological parameters and storage scenarios. It demonstrates history-dependent reloading in the conceptual reservoir and derives conditional pressure bands. These screen frictional reactivation potential; they are neither direct seismicity predictions nor universally safe operating limits for another storage site.
12. Destructive meteotsunami with fatal consequences on 12 January 2026 at Mar del Plata (Buenos Aires Province, Argentina)
Core Problem: Moving atmospheric disturbances can produce strongly location-dependent coastal oscillations through several resonance pathways.
Key Innovation: High-frequency sea-level and atmospheric observations reconstruct the Mar del Plata event. The authors distinguish a southern response consistent with Proudman resonance from possible edge-wave amplification near the city. The mechanism assignment is evidence-based interpretation, and port geometry and forcing differences prevent a single coast-wide amplification rule.
13. Hayli Gubbi and Erta Ale volcanic eruptions (2025): InSAR-derived pre-, co-, and post-eruptive deformations
Core Problem: Nearby volcanic systems may deform together, but inferring a connection requires comparison with alternative source geometries.
Key Innovation: InSAR and imagery track Hayli Gubbi–Erta Ale changes, and a coupled dual-source model fits the co-eruptive field better than the tested single source. Model preference supports the proposed interpretation without uniquely proving a physical conduit. Temporal windows and displacement assumptions also constrain claims about separate pre-, co- and post-eruptive processes.
14. Spatio-temporal characterization of flash floods in small data-scarce watersheds of the tropical Kigezi highlands, southwestern Uganda
Core Problem: Sparse rainfall gauges can miss the localized forcing that determines whether a small tropical watershed floods.
Key Innovation: Community observers and nearby gauges document thirty confirmed flash floods and seventeen intense-rainfall events without observed flooding across seven Ugandan watersheds. Contrasts emphasize local conditions and observation scale. Land-use associations are not randomized intervention effects, and weak terrain associations in this sample do not establish that geomorphology is generally unimportant.
15. Residual error modelling for process-based flood frequency analysis
Core Problem: A hydrological model calibrated to continuous flow may still systematically miss annual flood maxima.
Key Innovation: UQ-flood combines stochastic weather, process modelling and annual-maximum residual correction in three Canadian watersheds. Corrected central estimates move toward long-record benchmarks. Agreement within benchmark intervals is not the same as universal tail calibration; record representativeness, residual assumptions and watershed differences remain consequential.
16. Economic Principles and Optimal Rules for Flood Detention Basin Operation
Core Problem: Diverting floodwater can protect downstream assets while imposing losses inside a detention basin.
Key Innovation: An analytical economic framework derives diversion rules and a risk-adjusted control threshold, with a Chinese case and sensitivity tests. The preferred allocation depends on the specified losses, forecast uncertainty and physical constraints. It is a conditional optimization result, rather than evidence that the same economic threshold is socially or operationally appropriate everywhere.
17. Uncovering Responses of the Amazon Fluvial Lake System to the 2024 Drought by Fusing SWOT and Sentinel‐1 Observations
Core Problem: Large fluvial lakes can experience uneven storage changes and different timing within the same basin-wide drought.
Key Innovation: SWOT elevations and Sentinel-1 areas characterize 838 Amazon lakes alongside basin storage anomalies. The study reports concentrated deficits and longitudinal timing patterns. Its storage estimates and lag correlations depend on baselines, geometry and observation support; they do not uniquely establish a causal propagation pathway through every connected lake.
18. From Next Day Wildfire to Next Days Wildfire: A New Multi-Continent Dataset of Active-Fire Spread for Sequence-to-Sequence Forecasting
Core Problem: Existing time-series fire benchmarks underrepresent short and small fires and provide limited environmental diversity.
Key Innovation: AFSeq adds 1,194 events across eight countries and three broad regions, with an autoregressive temporal U-Net baseline. The resource broadens sequence-forecast evaluation. Satellite active-fire masks are not complete burned perimeters, and favorable benchmark scores do not establish calibrated multi-day operational fire probabilities or uniform global coverage.
19. Neighbourhood shortcuts in building damage assessment from very-high-resolution satellite imagery: diagnosis and a building-evidence plug-in
Core Problem: A damage classifier can exploit neighborhood damage patterns and overlook a building that differs from its surroundings.
Key Innovation: Building-stratified tests and neighborhood swaps diagnose shortcuts across nine models; a building-evidence plug-in reduces several errors. Synthetic interventions reveal sensitivity while introducing their own distribution changes. Better aggregate and subgroup scores reduce the tested shortcut without proving its elimination or validating every deployment event.
20. Causal-prior-constrained deep learning for rapid-intensification warning of western North Pacific tropical cyclones
Core Problem: An observational causal prior may help warning models, but its predictive contribution needs a forward comparison and separate causal scrutiny.
Key Innovation: CausalVisionRI combines imagery and environmental fields with a graph learned from earlier cyclones, then tests a 2020–2024 holdout. Reported Brier skill improves over the chosen references, with a smaller graph-ablation gain. Predictive skill does not verify every inferred causal edge, and operational generalization requires its own testing.
21. Analytical Optimization of Polarimetric Change Detectors for Mapping Open and Grassland Flooding Using Sentinel-1. Case of Rupununi, Guyana
Core Problem: Open water and grassland inundation produce different radar changes, making one unqualified change rule unreliable.
Key Innovation: A Guyana case derives additive and multiplicative polarimetric detectors and tests several combinations, with a data-driven combination performing best overall. The available evidence supports that comparison, without establishing a universal threshold, vegetation-independent accuracy or equivalent performance across other flood regimes.
22. Landslide identification and susceptibility assessment in high-consequence areas of the oilfield based on multi-band InSAR and machine learning
Core Problem: Oilfield facilities need screening for both observed slope movement and the conditioning terrain around potential landslides.
Key Innovation: Lutan-1 and Sentinel-1 deformation, optical checks and clustering identify candidate landslides, followed by random-forest susceptibility and facility overlap. The mapped candidates and exposure support prioritization. They are not all independently confirmed failures, and facility overlap with a high class is not a calibrated failure probability.
23. A review of partial saturation in liquefaction assessment: theory, laboratory studies, and field case histories
Core Problem: Entrapped air below the groundwater table can alter cyclic resistance, challenging a blanket assumption of full saturation.
Key Innovation: The review connects theory, laboratory results, wave-velocity interpretation and field cases within a partial-saturation assessment framework. The empirical relations remain soil- and condition-dependent. A nominal P-wave velocity contrast is not a universal field cutoff, and practical corrections need the applicable saturation, stress and measurement context.
24. Identifying rockburst-prone zones and their temporal evolution using energy-weighted MS visibility graphs
Core Problem: Microseismic energy sequences can organize into spatial and temporal patterns before high-energy events.
Key Innovation: Energy-weighted visibility graphs and entropy are tested through mine back-analysis and additional mining settings. Spatial consistency and preceding entropy changes support the studied indicators. Retrospective associations and high-energy event labels should remain distinct from prospective rockburst alarms with measured false-warning and missed-event rates.
25. Are Seismic Foundation Models More Than Glorified Feature Extractors?
Core Problem: A pretraining objective can improve one seismic task while reducing another, so one strong result does not establish a general-purpose representation.
Key Innovation: Architecture-matched experiments vary objectives and evaluate frozen models on three tasks and different label budgets. Skill is redistributed across picking, classification and ground-motion-related tasks. The authors' foundation-model critique applies to this experimental family and does not prove that all seismic pretraining is ineffective or predict earthquake occurrence.
26. Skillful and realistic precipitation forecasting with reinforced diffusion
Core Problem: Generative precipitation forecasts need to preserve storm structure while retaining useful pointwise and extreme-event skill.
Key Innovation: FitzRoy refines conditional diffusion through reinforcement learning and reports comparisons across US lead times, including selected flood cases. The stated balance of skill and realism is a benchmark claim. Case examples and fast inference do not by themselves establish physical conservation, calibrated warning probabilities or reliability for future record events.
27. Study on pre-seismic anomaly extraction of borehole strain data based on Pyraformer model: a case study of the 2021 Yangbi Ms6.4 earthquake in China
Core Problem: Signals identified before a known earthquake can also reflect seasonal, instrumental or non-seismic variation.
Key Innovation: A VMD–Pyraformer analysis describes stages of borehole-strain anomalies around the Yangbi event and compares timing with other observations. A retrospective single-event correspondence is not a calibrated earthquake forecast. Prospective definitions, non-event controls, false alarms and preprocessing leakage require separate assessment.
28. A physical coastal vulnerability index for site-specific assessment
Core Problem: A generic coastal index can miss the physical differences along a locally diverse shoreline.
Key Innovation: A place-based index combines conventional and local parameters over Lancashire and adjoining Sefton, producing relative vulnerability classes. The high-class share describes this index and study coast, rather than an observed event probability, expected loss or independently demonstrated benefit of the proposed adaptation measures.
29. A GeoAI framework for landslide susceptibility mapping using Google Earth Engine, ensemble machine learning, and SHAP explainable AI
Core Problem: Data-scarce regional mapping can combine open imagery and terrain information, but reported performance depends on sampling and validation design.
Key Innovation: A central Nepal GEE workflow compares ensembles on balanced landslide and non-landslide observations and interprets predictions with SHAP. Stratified splitting and cross-validation support the reported sampled task. They do not substitute for geographic or future-event transfer, and predictor importance is not a causal contribution to failure.
30. A Unified Multisensor Machine-Learning Framework for Live Fuel Moisture Content Retrieval
Core Problem: Multisensor archives can improve fuel-moisture estimation while retaining site-specific dependencies.
Key Innovation: Models combine optical products, environmental information and vegetation-specific training, with paired comparisons on shared test observations. Most primary tests hold out dates at represented sites, while unseen-site accuracy is lower. Improved LFMC retrieval is useful fuel information, not a direct demonstration of ignition or spread forecast skill.
31. FAME-Net: A Lightweight Feature Alignment and Multi-Scale Enhancement Network for UAV Forest Fire Detection
Core Problem: Small flames and smoke need to remain detectable when UAV inference is constrained by memory and computation.
Key Innovation: FAME-Net combines lightweight encoding, feature alignment and shared multiscale detection, with two-dataset and Jetson measurements. The work supports the tested accuracy–throughput trade-off. Image detection scores and device frame rates do not determine missed-fire risk under new visibility conditions or the complete warning-system latency.
32. GCDI: A Composite Drought Index on a 1-km Grid for China via Multi-Source Geo-Aware Feature Learning
Core Problem: Spatially continuous drought maps can depend strongly on nearby station information despite a fine output grid.
Key Innovation: GCDI-1 combines a station-derived SPEI auxiliary field with environmental predictors and temporal tests. Its improvement over leave-one-station-out Kriging is modest, and the station-derived feature supplies much of the skill. One-kilometre output spacing is not evidence of independently resolved drought variability at that scale.
33. A novel interpretation framework to derive mathematical model for flood susceptibility mapping integrating combined impact of flood conditioning factors from machine learning insights
Core Problem: Individual predictor summaries may not expose the combined response represented by a fitted flood classifier.
Key Innovation: A symbolic-regression interpretation derives compact nonlinear expressions and compares them with the study's SHAP summaries. These are descriptive approximations to model behavior. They do not identify causal flood mechanisms, prove that SHAP cannot analyze interactions, or validate physical compound flooding merely by including interacting terrain predictors.
34. Integrating damage load into strategic identification of vital streets for reducing emergency relief time in earthquake disasters
Core Problem: Street importance can depend on both travel time and the concentration of vulnerable parcels served.
Key Innovation: A pre-disruption allocation and routing framework accumulates damage-related load along medical-access paths and ranks intervention priorities. The concentration of assigned load supports the chosen planning criterion. It does not directly measure reduced post-earthquake relief time or represent every possible blockage and response pattern.
35. An energy-based collapse criterion for reinforced-concrete and masonry structures
Core Problem: An element-scale energy marker needs to be distinguished from independent validation of structural collapse.
Key Innovation: The proposed criterion is compared with complete and descriptor-truncated quasi-static histories, showing perfect conditional agreement in the reported comparisons. The labels depend on conventional response descriptors. That result is not perfect prediction of independently observed dynamic collapse, and extrapolation to whole structures and earthquakes remains a separate step.
36. Advance identification of where the 2026 Kumamoto earthquake would begin
Core Problem: A spatial correspondence between an earlier published candidate area and one later earthquake needs a broader test before it can establish forecasting skill.
Key Innovation: The authors report that a low-b-value area identified earlier contained the start of the 2026 event. This digest treats the advance-identification statement as the paper's case claim. One correspondence does not quantify timing, magnitude, spatial selectivity or false-alarm performance, and no further location prediction is endorsed here.
37. OutageDiT: A Foundation Model for Power Outage Forecasting and Scenario Generation
Core Problem: Power-system planning requires joint scenarios of outage magnitude, timing and restoration rather than isolated point predictions.
Key Innovation: OutageDiT generates seven-day trajectories from outage, weather and admissible future covariates, with held-out-region tests. Reported transfer and scenario scores support the studied planning task. Extreme-event calibration, the availability of covariates and operational decision benefits require evaluation beyond the foundation-model label.
38. Fleet-scale scenario simulation of offshore wind turbines under sequential earthquake–tsunami hazards: An application in China
Core Problem: An offshore fleet can experience correlated sequential hazards that a single-turbine analysis cannot fully describe.
Key Innovation: The verified study examines fleet-scale earthquake–tsunami scenarios in China. With no matching abstract, the entry does not infer fleet loss, failure correlations, turbine ranking or a demonstrated improvement in resilience.
39. A probabilistic framework for flood damage assessment under levee-breach scenarios for insurance applications
Core Problem: Insurance-oriented damage estimates require a consistent link between breach scenarios, inundation and exposed assets.
Key Innovation: The verified article proposes a probabilistic framework for this chain. Title-level evidence supports its scope but not a specific breach model, calibrated loss distribution, premium implication or demonstrated accuracy.
40. STRATA: A Structure-Guided Transfer Architecture for Multi-Event Building Damage Assessment from Post-Disaster VHR Optical Imagery under Limited Target Supervision
Core Problem: Transfer between disaster events becomes difficult when target labels are scarce and building patterns differ.
Key Innovation: The verified STRATA record proposes structure-guided transfer for post-disaster VHR imagery under limited target supervision. Matching results are unavailable, so cross-event accuracy, label savings and robustness to neighborhood shortcuts remain unassigned.
41. Hillslope erosion after tethered logging on steep slopes in beetle-killed subalpine forests
Core Problem: Logging on steep beetle-affected slopes can change the surface and flow pathways relevant to erosion.
Key Innovation: The verified study examines hillslope erosion after tethered logging. Without a matching abstract, this entry does not attribute a measured erosion increase or reduction to the practice, compare harvest systems or quantify downstream hazard.
42. Diffusion-assisted vision-based detection of Tunnel-lining icicles for maintaining high-speed railways in cold regions
Core Problem: Ice formations on tunnel linings can threaten railway operation and require dependable inspection.
Key Innovation: The verified record applies diffusion-assisted visual detection to lining icicles. Title-level evidence does not establish accuracy, synthetic-image validity, field false alarms or a demonstrated maintenance-safety benefit.
43. Interpretable soil moisture drought forecasting using graph attention networks
Core Problem: Forecasting a drought index requires spatiotemporal information and an interpretation that distinguishes associations from drivers.
Key Innovation: The verified study uses graph attention for interpretable soil-moisture drought prediction. A matching abstract is unavailable; lead times, geographic transfer, uncertainty calibration and comparative skill are not inferred.
44. Morphology-intensity relationships and long-term evolution of extreme precipitation over China based on an adaptive sector scanning algorithm
Core Problem: Storm morphology and intensity can change together, complicating trends based on one scalar rainfall measure.
Key Innovation: The verified China study uses adaptive sector scanning to examine these relationships over time. Title-level evidence does not support a particular trend, storm-type mechanism or causal climate attribution.
45. A probabilistic framework for modeling compound flooding hazard in coastal-inland transition zones
Core Problem: Hazard in a coastal transition zone depends on the joint behavior of upstream and marine drivers.
Key Innovation: The verified article develops a probabilistic compound-flooding framework. Without a matching abstract, dependence structure, nonlinearity, event probabilities and validation results remain unspecified.
46. Quantifying multidimensional drought risks under climate change: A deep learning-based downscaling and DDF curves approach
Core Problem: Drought intensity, duration and frequency can change in combinations that a single index does not summarize.
Key Innovation: The verified study combines downscaling with drought duration–frequency analysis. Available title-level evidence does not establish local projection accuracy, scenario changes or a calibrated multi-dimensional risk estimate.
47. Sparse-monitoring-enhanced physics-informed reconstruction of full-field excavation-induced soil deformation
Core Problem: Sparse monitoring leaves most of an excavation's deformation field unobserved.
Key Innovation: The verified study uses physics-informed reconstruction enhanced by limited measurements. A matching abstract is unavailable, so field accuracy, identifiability and transfer to different excavation geometries or soils are not asserted.
48. Numerical investigation of progressive submarine landslides and the resulting wave response using a two-phase two-layer SPFEM
Core Problem: A progressive submarine failure and the wave it generates require treatment of both moving sediment and water.
Key Innovation: The verified record proposes a two-phase two-layer SPFEM investigation of that coupled process. With title-level evidence, no runout, wave-height accuracy, validation case or general numerical advantage is inferred.
49. Multi-data source based quantifying urban flood severity in major Chinese cities (2000–2024) using a hybrid machine-learning weighting framework
Core Problem: Long-term city comparisons depend on how fragmented impact records are assembled and weighted.
Key Innovation: Official, media and social records support an event database and severity index for twenty Chinese cities. The study reports regional differences and substantial interannual variability. Severity scores are not direct physical flood measurements, and changing reporting coverage and weighting affect interpretation of trends.
50. Effect of earthquake duration on seismic assessment of elevated water tanks using damage and energy-based indicators
Core Problem: Duration can affect accumulated tank damage even when peak intensity is similar.
Key Innovation: A nonlinear finite-element case compares damage and energy indicators under different shaking durations. Water mass is represented with simplified hydrostatic loading. The model therefore cannot substantiate full fluid–structure or sloshing effects, and its duration-dependent indices do not alone establish general failure probabilities.
51. Impact of Spatial Variability in Root‐Zone Storage Deficit on Runoff Generation
Core Problem: Spatially varying root-zone deficits can alter when different parts of a basin contribute drainage.
Key Innovation: A distributed water-balance analysis in Sierra Nevada basins shows differences that persist after spatial aggregation. The result supports retaining heterogeneity even in simple models. Inferred drainage is not a direct observation of every runoff pathway or an independently validated flood forecast.
52. Selective Transfer Learning for Groundwater Level Anomaly Forecasting: Accuracy-Risk Trade-offs
Core Problem: Transfer learning can improve a mean score while harming many individual monitoring wells.
Key Innovation: A Netherlands study compares fixed transfer, contextual blending and a simple global blend under an exact monthly forecast task. Contextual average gains are small, and the global blend protects more wells in several seeds. Partial rescue can follow mechanically from blending; it is not proof of contextual superiority or rolling-deployment validity.
53. RainCheck: Harnessing Historical Weather Data with Machine Learning for Rainfall Forecasting in Bangladesh
Core Problem: Random station-day splits can reward shared calendar and same-day spatial information rather than future rainfall skill.
Key Innovation: RainCheck compares temporal and spatial holdouts with climatology, persistence and learned models in Bangladesh. Genuine short-lead gains are modest, and several high headline scores disappear under stronger splits. The result concerns the tested historical data and targets, not an argument that all rainfall machine learning lacks value.
54. Operator-Aware Posterior Sampling for Multivariate Weather Downscaling
Core Problem: A diffusion prior can be reused across coarse weather sources only if the observation relationship and variable scales are handled appropriately.
Key Innovation: An operator-aware guidance loop learns the coarse-to-fine observation mapping and normalizes channel gradients while holding the prior fixed. This addresses the stated inverse problem. The source and paired-data requirements still matter, and the method alone does not validate extreme-event probabilities or downstream hazard impacts.
55. What Survives Label Downsampling? Exact Region Survival for Coarse Segmentation Targets
Core Problem: Coarse training labels can erase small regions before a segmentation model has any opportunity to learn them.
Key Innovation: MatchDown uses region-to-cell matching to maximize label survival under a one-label-per-cell representation. Benchmarks separate survival from shape recovery. Its guarantee concerns coarse targets under defined constraints, not accurate reconstruction of every small landslide, building or fire region in Earth-observation imagery.
56. NeurEO: dissecting Earth observation embeddings with computational neuroscience
Core Problem: Embedding features may encode useful land-surface structure without having an obvious semantic interpretation.
Key Innovation: NeurEO adapts correlation and tuning analyses to AlphaEarth, relating feature responses to land cover and elevation. The results reveal structured associations and finer residual representations. They do not identify causal physical variables or demonstrate that a feature is reliable for a new hazard task.
57. Effect of tailings composition on enzyme-induced carbonate precipitation stabilisation: a baseline study
Core Problem: Tailings chemistry can inhibit the enzyme activity needed for carbonate-based biocementation.
Key Innovation: Laboratory tests compare metals and flotation collectors, then examine cysteine-assisted treatment and mechanical response. Improved precipitation and strength support the tested baseline. They do not establish field-scale tailings stability, durable containment, metal immobilization or the treatment's complete environmental performance.
58. Experimental and theoretical study on passive earth pressure: effect of backfill width
Core Problem: Backfill width can change passive thrust and soil–wall interaction even where the displacement to the limit state changes little.
Key Innovation: Model tests, image-based strain measurements and a theoretical solution examine arching and shear localization in narrow backfills. The comparison supports the tested nonlimit response. Higher passive thrust is a demand or resistance quantity within a particular boundary setup, not by itself a general improvement in wall safety.
59. Direct measurement of shear mobilization in 3D-printed root analogues using distributed fibre-optic sensing
Core Problem: The reinforcement contribution of a root depends on its embedded length and mobilized deformation, not only its material stiffness.
Key Innovation: Scanned, printed root analogues and fibre-optic strain measurements distinguish local short-root response from tensile mobilization along longer branches. This provides controlled mechanism measurements. Resin analogues and one geometry do not reproduce all biological root networks or establish field slope-stability benefits.
60. Systematic Validation of GNSS-R Wind Retrievals From the 22-Satellite Tianmu-1 Constellation Over Chinese Coastal Waters: Multi-GNSS, Intersatellite, and Rainfall Impacts
Core Problem: GNSS-reflectometry wind errors vary by signal, satellite, wind regime and rainfall, which average validation can obscure.
Key Innovation: A Tianmu-1 comparison uses buoys and cyclone-related matchups to identify heterogeneity and rain-associated degradation. The result constrains product use in the sampled settings. Low-wind overestimation and stronger-wind underestimation matter for hazard inputs, and controlled associations do not uniquely isolate every physical error mechanism.
61. Time-Domain P-Wave Impedance Inversion Using a Learning-Based Approach: A Sleipner CO 2 Monitoring Study
Core Problem: Time-lapse impedance changes are inferred through a nonunique inverse problem, making training physics and field comparison important.
Key Innovation: A synthetic-trained time-domain method is applied to Sleipner and produces plume patterns consistent with established interpretations. Agreement with prior plume geometry and processing sensitivity checks supports interpretation. It is not direct validation of every thin layer or independent proof of containment without leakage.
62. CCWEM and CCWIF: 10 m Datasets of Coastal Wetland Extent and Inundation-Frequency Gradients across China
Core Problem: Wetland extent and internal tidal inundation gradients describe different parts of a changing coastal surface.
Key Innovation: CCWEM and CCWIF combine seasonal optical mapping and tide-informed estimation across Chinese coastal subregions. Reported validation supports extent and relative inundation ordering. Ten-metre grids and class agreement do not establish absolute event-flood depth, storm-surge probability or uniform accuracy under every tidal and vegetation condition.
63. Influence of matrix cementation and soft rock blocks on the mechanical behavior of soil-rock mixture: insights from triaxial experiments and interpretable machine learning
Core Problem: Matrix cementation changes how weak rock blocks contribute to the composite's skeleton and frictional response.
Key Innovation: Triaxial tests and model comparisons identify confinement- and cementation-dependent strength relations. Interpretable curve prediction supplements the experiments. SHAP associations do not independently discover universal physical laws, and sample behavior still needs a site-specific link to slope geometry and loading before stability can be assessed.
64. Effects of grain size and normal stress on the frictional stability and healing characteristics of simulated quartz gouge
Core Problem: Grain size and normal stress may affect whether a gouge layer slides stably and how resistance recovers.
Key Innovation: The verified quartz-gouge study examines frictional stability and healing. A matching abstract is unavailable, so the direction or magnitude of these effects and their transfer to natural fault slip are not inferred.
65. Improving hydrologic prediction and interpretability through physics-informed dual-output LSTM with water and energy balance constraints
Core Problem: Multi-output hydrological learning needs to balance streamflow skill, evapotranspiration and consistency across stores and fluxes.
Key Innovation: PIDuaL tests water- and energy-constrained variants across 149 Indian watersheds, with gains especially in low-flow and difficult settings and some ET trade-offs. Independent-state correspondence supports interpretation. It does not uniquely identify physical storage or establish equivalent gains for rare flood peaks and untested climates.
66. Climate-driven shift from glacial to pluvial groundwater contributions in a central Nepal Himalayan catchment
Core Problem: Similar total discharge can conceal a substantial change in the climatic origins of groundwater and streamflow.
Key Innovation: A source-explicit Himalayan model carries rain, snow and glacier identity through subsurface storage and explores three climate scenarios. Modelled rainfall recharge increasingly replaces glacier-derived contributions. The source partition depends on model and forcing assumptions, and broadly stable groundwater discharge does not imply unchanged seasonal water-security risk.
67. HydroFlow: A robust and lightweight network for river surface velocimetry on edge devices via semi-supervised learning
Core Problem: Edge deployment of image velocimetry requires a balance between robustness, labeling effort and computation.
Key Innovation: The verified HydroFlow study proposes a lightweight semi-supervised river-surface velocimetry network. Title-level evidence does not establish velocity error, unseen-site robustness, device throughput or discharge accuracy.
68. Infiltration dynamics of vegetated soils controlled by drought-induced root-soil interface evolution
Core Problem: Drying can modify root–soil contacts and the routes by which later rainfall infiltrates.
Key Innovation: The verified study addresses interface evolution and infiltration dynamics. Without a matching abstract, this entry does not assign a universal direction to root effects or claim an observed change in landslide probability.
69. Soil moisture sensor networks for advancing understanding of hillslope and catchment hydrology: a review
Core Problem: Sensor placement and measurement support shape what soil-moisture networks reveal about hydrological processes.
Key Innovation: The verified review examines network contributions to hillslope and catchment understanding. Available title-level evidence does not establish an optimal network design, representative accuracy or a demonstrated warning benefit.
70. A novel 3D engineering geological zonation method for major transportation corridors in mountainous regions
Core Problem: Corridor planning needs geological information in three dimensions rather than a single surface zoning layer.
Key Innovation: The verified article proposes a 3D engineering-geological zonation method for mountain transport routes. A matching abstract is unavailable; no quantified mapping accuracy, preferred route or reduction in landslide risk is inferred.