TerraMosaic Daily Digest: July 21, 2026
Daily Summary
Landslide mechanics advances from partial failure descriptions toward complete, testable motion histories. Exact analytical solutions now span initiation, acceleration, deceleration and deposition under positive or negative net driving forces. Repeated large-scale experiments define hydraulic and geometric boundary conditions that reproduce rapid basal sliding at 4.30-4.60 m/s, while red-bed experiments connect pore and mineral alteration across wet-dry cycles to strength loss and macroscopic slope failure.
Field evidence and numerical models increasingly resolve connected hazard sequences rather than isolated endpoints. The Temi palaeolandslide reconstruction follows a 35-million-cubic-metre mass from a 64 m/s runout through river damming to an outburst flood peaking near 55,300 cubic metres per second. At Kalisari, five years of UAV-LiDAR show that vegetation does not materially suppress retrogressive deformation but does retain displaced sediment, separating slope stabilization from downstream impact reduction.
Model credibility depends increasingly on representing the structure of both the process and the evaluation. Circulation-constrained simulations attribute about 55% of European summer soil-moisture decline since 1980 and 42% of vapour-pressure-deficit increase to circulation change, while showing that dynamic and thermodynamic drying reinforce one another. In hazard mapping, task-specific negatives, spatial holdouts and coordinate ablations expose gains caused by label confusion or geographic memorization; terrain-aware fusion then restores information tied to the physical setting rather than to location alone.
Key Trends
Five shifts connect complete motion models, cross-scale degradation, cascade reconstruction, evaluation geometry and driver attribution.
- Motion models span the full failure trajectory: Analytical and experimental studies now connect release, acceleration, deceleration and deposition instead of calibrating only peak velocity or final runout.
- Cross-scale degradation is made measurable: Pore evolution, mineral alteration, seepage and block interaction are linked to observable strength loss and slope-scale instability.
- Cascades are reconstructed end to end: Landslide motion, river blockage, breach hydraulics and downstream inundation are treated as one sequence with shared timing and mass-flow constraints.
- Evaluation geometry becomes explicit: Hazard-specific negatives, spatial holdouts and coordinate ablations distinguish transferable signal from ambiguous labels and geographic memorization.
- Attribution separates interacting drivers: Circulation-constrained climate experiments and process-aware hazard models quantify how dynamic forcing modifies thermodynamic or hydrological responses.
Selected Papers
The leading papers complete the landslide motion trajectory, reproduce rapid failure at experimental scale, connect red-bed deterioration to slope collapse, and reconstruct a landslide-dam-outburst cascade from field chronology to downstream hydraulics.
1. The complete landslide velocity
Core Problem: Existing analytical landslide-velocity solutions describe acceleration under positive net driving force but do not cover the full transition through deceleration and deposition.
Key Innovation: A unified set of exact solutions extends the same physical model across positive and negative net driving forces, resolving release, acceleration, weak and strong deceleration, flow-regime change and final deposition.
2. European summer drying largely driven by atmospheric circulation changes since the 1980s
Core Problem: European summer drying reflects both atmospheric-circulation shifts and thermodynamic warming, but observations alone cannot isolate their relative contributions.
Key Innovation: CESM2 simulations nudged to observed circulation attribute about 55% of the 1980-2024 soil-moisture decline and 42% of the vapour-pressure-deficit increase to circulation change, which also amplifies thermodynamic drying during warm, dry regimes.
3. Meso-structure deterioration of redbeds soft rocks and its cross-scale disaster-causing mechanism
Core Problem: Water-induced degradation of red-bed soft rock is observed from pores to whole slopes, but the scales are rarely connected within one failure model.
Key Innovation: Wet-dry experiments and a Darcy-based hydraulic-geotechnical model link mineral loss, pore growth and microcrack coalescence to macroscopic shear failure; porosity rises 9.3% after five cycles and peak strength falls 35% after three.
4. Quantitative reconstruction of the damming and outburst processes of the Temi Palaeolandslide in the Upper Jinsha River, Southeastern Tibetan Plateau, China
Core Problem: Landslide dams are commonly studied as separate slope, lake and flood problems, obscuring the timing and magnitude of the complete cascade.
Key Innovation: Field mapping, luminescence dating, Massflow and HEC-RAS 2D reconstruct the Temi sequence from a 64 m/s, 1,780 m runout to dam formation and a breach flood with a simulated peak discharge near 55,300 cubic metres per second.
5. Vegetation influence on retrogressive landslide deformation and sediment transport from multi-temporal UAV-LiDAR monitoring
Core Problem: Vegetation may reduce shallow erosion, but its effect on deep-seated retrogressive deformation and exported sediment remains poorly separated.
Key Innovation: Five annual UAV-LiDAR surveys show similar deformation in vegetated and bare sectors of the Kalisari landslide, while lower net material loss in the vegetated sector identifies debris retention rather than deformation suppression.
6. Design and validation of boundary conditions for rapid soil mass movement in large-scale slope experiments
Core Problem: Large-scale slope tests often fail to reproduce rapid basal sliding because seepage localization and deformation change with experimental scale.
Key Innovation: Finite-element seepage analysis identifies a 0.4 m water level and a refined berm, slope and sand-layer geometry; three repeated tests then reproduce the same basal failure mode at 4.30-4.60 m/s.
7. Characteristics and susceptibility assessment of landslides triggered by historical mega earthquakes in the northeastern margin of the Qinghai-Tibet Plateau: a case study of the 1927 Gulang M8 earthquake
Core Problem: The 1927 Gulang Mw 8 earthquake produced extensive coseismic landslides, but a spatially complete inventory and terrain-adjusted reconstruction of their shaking control had remained unavailable.
Key Innovation: A new high-resolution inventory is coupled with topographically corrected peak ground acceleration and a genetic-algorithm-optimized random forest; susceptibility reaches an AUC of 0.908, five percentage points above the uncorrected-shaking model.
8. Task-Specific Negative Sample Selection for Multi-Hazard Susceptibility Mapping of Slope-Instability Hazards
Core Problem: In multi-hazard slope mapping, a location used as a negative example for one failure type may be susceptible to another, creating task-dependent label confusion.
Key Innovation: Hazard-specific feature centroids guide a mixture of hard, easy and random negatives; across five classifiers, the strategy raises mean AUC by 14.47% and places 80% of historical events within 27.06% of the mapped slope units.
9. Landslide Susceptibility Mapping Using an Image–Tabular Joint Deep Learning Framework: A Case Study of the Tacheng Region, Xinjiang, China
Core Problem: Raster-only landslide susceptibility models underuse attribute information and spatial context, which limits transfer beyond randomly partitioned samples.
Key Innovation: A FiLM-conditioned U-Net combines image patches with soft tabular attribute priors and reaches AUC 0.953 on the independent Tacheng test set and 0.921 under spatial holdout.
10. LHEM-MSegNet: A Landslide Hazard Extraction Model Based on Multi-Source Data Fusion
Core Problem: Loess landslide hazards are difficult to separate from bare slopes, gullies and disturbed ground because their spectral contrast and boundaries are weak.
Key Innovation: LHEM-MSegNet fuses optical imagery with elevation, slope and aspect through Transformer and attention pathways; it reaches 89.29% mIoU on the new LYPLH dataset and remains competitive on Landslide4Sense.
11. How toe excavation and rainfall triggered the 2022 Aiziping colluvial landslide overlying gentle red beds in Tongjiang, SW China
Core Problem: Colluvium over gently dipping red beds can fail after road excavation, yet the interaction between the new free face, altered drainage and subsequent rainfall is poorly constrained.
Key Innovation: Field, laboratory and seepage-stability analyses show that excavation concentrated infiltration and impeded toe drainage; rainfall then raised toe pore pressure and drove soil softening, two slip surfaces and retrogressive failure.
12. Deformation, failure and run-out mechanisms of joint-controlled rock slopes: Insights from Discontinuous Deformation Analysis
Core Problem: Runout forecasts for jointed rock slopes require a model that reproduces block sliding, toppling, collision and disintegration from initiation through deposition.
Key Innovation: High-speed physical experiments validate discontinuous deformation analysis before application to the Luoma slope, where joint orientation controls energy transfer and the transition to a coupled landslide-rockfall cascade.
13. Diffusion‐Based Probabilistic Modeling for Hourly Streamflow Prediction and Assimilation
Core Problem: Daily streamflow models smooth the short, intense peaks needed for flood warning and do not naturally assimilate recent observations into probabilistic hourly forecasts.
Key Innovation: h-Diffusion generates hourly hydrographs conditioned on daily discharge and uses inpainting for assimilation; tests across more than 500 US basins outperform leading multi-timescale LSTM baselines, including for extremes.
14. Geospatial Diffusion-based Evolution Synthesis (GeoDES) for Storm-Centered Weather Augmentation
Core Problem: Weather-event datasets contain too few geographically diverse, physically coherent storm realizations for robust model training.
Key Innovation: GeoDES uses diffusion-based evolution synthesis around storm-centred geospatial states, augmenting hazardous-weather trajectories while preserving their spatial and temporal structure.
15. Juxtaposition of Shallow Reservoir-Triggered Seismicity and Deep Tectonic Locking in the Qiaojia-Dongchuan Seismic Gap
Core Problem: Reservoir-induced earthquakes can obscure the deeper tectonic state of a mature seismic gap when shallow and deep events are interpreted as one population.
Key Innovation: Depth-resolved seismicity, b-values and Coulomb stressing separate shallow reservoir-triggered activity from a deeper, low-b-value locked asperity in the Qiaojia-Dongchuan gap.
16. Tensor Network Machine Learning for Wildfire Susceptibility Mapping: from Grokking Dynamics to Quantum Mixedness of Class Representations
Core Problem: Wildfire susceptibility classes are difficult to separate when neighbouring hazard levels occupy overlapping environmental feature space.
Key Innovation: Matrix-product-state classifiers operate on AlphaEarth embeddings and add reduced-density-matrix diagnostics, turning class confusion into a level-resolved measure of representation mixedness.
17. Attenuation of tsunami-like wave runup by roughness elements
Core Problem: Tsunami runup over rough coasts depends on how obstacle geometry redistributes momentum, but the attenuation is difficult to isolate from other boundary effects.
Key Innovation: Controlled tsunami-like waves quantify how roughness elements reduce runup, resolving the relationship among element configuration, energy dissipation and shoreline response.
18. Copula-based multi-hazard risk assessment of coastal marine disasters: A case study of storm surges and extreme waves in Zhoushan, China
Core Problem: Coastal assessments often treat storm surge and extreme waves separately even though their dependence governs compound marine hazard.
Key Innovation: A copula-based framework models joint surge-wave behaviour and translates it into multi-hazard risk for Zhoushan, preserving dependence that single-variable return periods omit.
19. A fault-based application to model seismicity rates for seismic hazard assessment in the southern Apennines (Italy)
Core Problem: Fault-based seismic hazard models must reconcile mapped fault geometry with incomplete earthquake records and spatially variable recurrence.
Key Innovation: The southern Apennines application derives fault-specific seismicity rates for probabilistic hazard assessment, connecting geological sources to the event-rate model instead of relying only on area-wide catalog statistics.
20. Europe's transport infrastructure is not ready to face climate change
Core Problem: European transport networks are assessed with climate data and design standards that do not fully represent changing heat, precipitation, flood and slope-related extremes.
Key Innovation: A cross-European infrastructure analysis identifies where present standards and adaptation planning diverge from projected hazards, exposing systemic readiness gaps across transport modes.
21. Deep Learning Emulation of Multivariate Climate Indices: A Case Study of the Fire Weather Index in the Iberian Peninsula
Core Problem: Daily fire-danger assessment depends on a multivariate index whose physical calculation is costly to reproduce at scale from climate-model output.
Key Innovation: Deep networks emulate the reference Fire Weather Index from the same daily meteorological inputs used by conventional proxies, with evaluation focused on spatial fidelity and fire-weather extremes across Iberia.
22. WSDS-CAN: Wildfire Spread Prediction Dataset for Canadian Boreal Forests
Core Problem: Wildfire-spread learning in Canadian boreal forests lacks event-aligned, machine-ready data that pair observed perimeters with weather, terrain and fuels at the scale of individual fires.
Key Innovation: WSDS-CAN aligns 2,565 fires from 2001-2020 with fire-adapted environmental covariates and event-scale rasters, creating an open basis for segmentation and fire-front prediction across both small ignitions and complex perimeters.
23. Ensemble forecasts of isolated and compound wind and precipitation extremes in Europe using HC-SWG (v3.1) and MA-SWG (v1.1) Stochastic Weather Generators
Core Problem: Useful early warning for European extremes must represent isolated and sequential wind-rainfall events beyond the skill horizon of direct numerical reforecasts.
Key Innovation: Two circulation-analogue stochastic weather generators improve station-scale extreme forecasts to lead times of ten days and reproduce observed frequencies of simultaneous and sequential wind-precipitation events.
24. The 2022–2023 snow drought in the Italian Alps doubled glacier contribution to summer streamflow
Core Problem: The extent to which glacier melt buffers downstream flow during severe snow drought has rarely been quantified at catchment scale.
Key Innovation: Distributed cryospheric modelling and stream gauges show that glacier contributions doubled to tripled during the 2022-2023 Italian Alpine snow droughts through earlier melt onset and stronger summer ablation.
25. Spatial Leakage in Classifying NASA FIRMS Thermal Anomalies as Wildfire Incidents: A Leakage-Controlled Evaluation of Radiometric, Temporal, and Spatiotemporal Features
Core Problem: Random train-test splits can make wildfire classifiers memorize where fires occurred, while non-causal clustering can leak future detections into current predictions.
Key Innovation: Event-aware, spatial-block and temporal tests reveal the inflation: LightGBM falls from F1 0.985 under a random split to 0.627 under spatial holdout, whereas coordinate-free causal clustering raises spatial-transfer F1 to 0.868.
26. GTSNet: A Global Topography-Aware Segmentation Network for Remote Sensing Identification of Unstable Rock Masses
Core Problem: Unstable rock masses are small, irregular targets whose appearance changes with relief and viewing geometry, weakening ordinary image segmentation.
Key Innovation: GTSNet injects global topographic context into the segmentation pathway to retain terrain-scale controls while delineating local unstable-rock boundaries.
27. Rapid Strong Earthquake Magnitude Estimation Based on Near-Field High-Rate GNSS Data Using Deep Learning
Core Problem: Rapid earthquake magnitude estimates saturate or destabilize when they rely on short near-field waveform windows from a changing high-rate GNSS network.
Key Innovation: A deep model learns the early displacement evolution recorded by high-rate GNSS stations, targeting stable strong-earthquake magnitude estimates before complete rupture observations are available.
28. Near-Real-Time Flood Disaster Monitoring Based on Moonlight and Multi-Source Remote Sensing: A Case Study of Zhengzhou Rainstorm
Core Problem: Near-real-time flood assessment is delayed when daytime optical coverage and radar revisits leave a night-time observation gap.
Key Innovation: Moonlit optical observations are combined with multi-source imagery for the Zhengzhou rainstorm, reducing the effective acquisition interval to about one day and mapping 722.93 square kilometres of inundation.
29. CLIFF: A Multi-Modal Remote Sensing Model for Geological Hazard Monitoring Based on Bitemporal UAV Images
Core Problem: Bitemporal UAV surveys can document hazardous terrain change, but optical differences alone confound true deformation with illumination and viewpoint shifts.
Key Innovation: CLIFF combines complementary modalities in a bitemporal remote-sensing model designed to separate geological-hazard change from non-structural image variation.
30. Multiparametric investigation of Etna volcano paroxysms in February 2021 from a multilayer geophysical perspective
Core Problem: Volcanic paroxysms perturb several coupled systems, yet seismic, atmospheric, ionospheric and geomagnetic observations are often interpreted independently.
Key Innovation: The February 2021 Etna sequence is examined across those layers; localized total-electron-content anomalies are separated from regional space-weather variability and evaluated as part of lithosphere-atmosphere-ionosphere coupling.
31. Unravelling the gap between the probabilistic volcanic hazard and risk perception: A comparative study of tephra fallout at Etna and Vesuvius volcanoes
Core Problem: Probabilistic tephra hazard does not translate directly into perceived personal risk, weakening evacuation and preparedness decisions around active volcanoes.
Key Innovation: A comparative Etna-Vesuvius study places mapped fallout hazard beside survey-based risk perception to identify where physical exposure and public interpretation diverge.
32. Machine learning-based wildfire susceptibility assessment in California's WUI: Multi-source feature selection and analysis
Core Problem: Wildfire occurrence in California's wildland-urban interface reflects nonlinear interactions among fuels, terrain, weather and human exposure that single-source models cannot resolve.
Key Innovation: Multi-source predictors, feature selection and machine-learning susceptibility models are combined to map WUI fire probability and identify the dominant nonlinear controls.
33. A multiplex network-based framework to assess interdependent infrastructure asset criticality and priority under natural hazard scenarios
Core Problem: Natural hazards propagate through electricity, telecommunications, health and other infrastructure, so asset priority cannot be inferred from single-network centrality.
Key Innovation: A node-weighted multiplex model combines interdependency type, multidimensional consequence, exposure and vulnerability; a Torbay flood scenario identifies electricity and telecommunications assets with the largest cascading influence.
34. Multi-temporal mapping and analyses of post-wildfire surface soil moisture modeled from Landsat Thermal-IR and Sentinel-1 SAR
Core Problem: Wildfire changes soil-water balance through vegetation loss, altered throughfall and hydrophobicity, but field networks alone cannot resolve the spatial response.
Key Innovation: A random forest combines Sentinel-1 SAR, Landsat thermal imagery and in-situ measurements to produce 30 m soil-moisture time series, showing consistently drier post-fire surfaces than unburned controls.
35. Cascade structure and event-level attribution of three-dimensional compound drought–heat events in China: contrasting meteorological and agricultural systems
Core Problem: Compound drought-heat events evolve through different meteorological and agricultural pathways that event counts alone cannot distinguish.
Key Innovation: A three-dimensional event framework resolves cascade structure and attributes individual Chinese events, separating atmospheric drought-heat coupling from the delayed response of agricultural systems.
36. Dynamic response and failure mechanisms of pipeline-tunnel portal section in strong earthquake area: Insights from large-scale shaking table test
Core Problem: Pipeline-tunnel portals in strong-motion regions concentrate demand at the interface between an underground structure and a deformable entrance slope.
Key Innovation: A large-scale shaking-table experiment resolves nonlinear amplification and divides portal-slope deformation into initiation, progressive growth and failure stages.
37. Failure mechanisms and seismic stability of purely cohesive slopes with internal karst cavities under surcharge and seismic loading
Core Problem: Internal karst cavities redirect slip surfaces and amplify seismic weakening in cohesive slopes subjected to crest surcharge.
Key Innovation: Finite-element limit analysis maps the coupled effects of cavity geometry, soil strength, surcharge and shaking, then condenses them into a cavity-seismic stability ratio and design charts for rapid factor-of-safety estimation.
38. Geological hazard susceptibility and segment-based risk assessment of mountainous expressways: A case study of Lianying Expressway in Guangdong, China
Core Problem: Mountain-road risk assessments often stop at regional susceptibility and do not translate distributed hazard into priorities for individual route segments.
Key Innovation: Information-value conditioning factors and five classifiers are coupled with distance-weighted susceptibility and traffic load; the random-forest model reaches AUC 0.874 and identifies distinct risk concentrations for cuts, bridges and tunnels.
39. Global Performance of Remote Sensing‐Based and Reanalysis‐Driven Models to Estimate Open Water Evaporation
Core Problem: Open-water evaporation products are widely applied without a common global test against lake observations, leaving their climatic and geographic transfer uncertain.
Key Innovation: Remote-sensing and reanalysis-driven models are evaluated at 27 lakes; an ensemble reduces global error to an RMSE of 1.3 mm per day with a 0.3 mm per day bias and reveals where individual formulations fail.
40. Physics-Aware Complex-Valued State Space Model with Scattering-Prior Feature Modulation for PolSAR Image Classification
Core Problem: PolSAR classifiers must retain amplitude-phase coupling while learning long-range spatial structure and respecting the scattering information carried by polarimetric channels.
Key Innovation: CV-SSMNet operates in the complex domain and modulates state-space features with scattering priors, combining long-range dependency modelling with physically interpretable polarimetric cues.
41. Memory-Augmented Multimodal Large Language Models for Small Object Understanding in Streaming Aerial Videos
Core Problem: Streaming aerial video presents tiny targets over long sequences, exceeding the spatial resolution and temporal memory of general multimodal language models.
Key Innovation: DroneEyes contributes 2,140 high-definition videos and 176,623 referring-segmentation pairs, while a memory-augmented model preserves object evidence across frames for open-vocabulary queries.
42. OffNadirLoc: Benchmark and Framework for Challenging UAV-to-Satellite Geo-Localization under Large Off-Nadir Views
Core Problem: Large off-nadir angles create perspective distortion, occlusion and appearance shifts that defeat pairwise UAV-to-satellite geolocation.
Key Innovation: OffNadirLoc introduces a dedicated benchmark and set-level supervision that learns consistency across multiple views rather than treating each aerial-satellite pair independently.
43. Post-Training in Time Series Foundation Models: A Unifying Framework
Core Problem: Pretraining alone does not determine how a time-series foundation model will adapt, calibrate uncertainty or meet deployment constraints on a new forecasting task.
Key Innovation: The review organizes post-training into parameter adaptation, context augmentation, model composition, output and uncertainty control, and compression or specialization, clarifying where each intervention acts.
44. RIM: A Retrieval-In-Matching Framework for Cross-Domain Global Visual Localization of UAVs
Core Problem: UAV localization against remote-sensing maps degrades when platform, season and viewing angle produce large cross-domain appearance and geometry shifts.
Key Innovation: RIM adapts retrieval with pose-near positives and geographic hard negatives, reuses one DINOv2 token field for retrieval and matching, and improves zero-shot Recall@1 while completing retrieval and geometric verification in 67.9 ms.
45. RS-RIE-Bench: Benchmarking Reasoning-Guided Remote Sensing Image Editing
Core Problem: Remote-sensing image editors are rarely tested on instructions that require both scene reasoning and faithful, localized visual change.
Key Innovation: RS-RIE-Bench evaluates eight models with strict joint criteria and expert-checked multimodal judging; the strongest system reaches only 24.28% overall accuracy, exposing a large reasoning-editing gap.
46. PIER: Physics-Informed Environmental Retrieval for Time-Series Modeling
Core Problem: Environmental time-series retrieval by embedding similarity can select historical analogues that look alike statistically but violate the target system's flux-response dynamics.
Key Innovation: PIER adds a physics-aware retrieval stream with local verifiers built from flux features, ranking candidate contexts by both representation similarity and physical consistency.
47. Multimodal Large Language Models for Remote Sensing Image Understanding: Domain-Specific or General-Purpose?
Core Problem: It remains unclear when remote-sensing-specific multimodal language models outperform general vision-language systems and where either class fails across tasks.
Key Innovation: A common diagnostic evaluation maps task-level capability boundaries and finds that general-purpose models can match or exceed specialized systems on several remote-sensing understanding tasks without domain fine-tuning.
48. Interval and fuzzy physics-augmented neural networks (iPANN and fPANN) for uncertainty quantification and propagation in constitutive modeling
Core Problem: Constitutive models trained on sparse or noisy stress-deformation data need uncertainty bounds that remain consistent with material physics.
Key Innovation: Interval and fuzzy physics-augmented neural networks propagate input uncertainty through hyperelastic response and learn prediction envelopes that contain noisy observations while generalizing to held-out loading states.
49. PG-KINN: A Physics-Informed Petrov-Galerkin Kolmogorov-Arnold Network for Solving Forward and Inverse PDEs
Core Problem: Strong-form physics-informed losses require high derivatives, whereas common variational forms can fail for non-self-adjoint or inverse problems.
Key Innovation: PG-KINN places a Kolmogorov-Arnold network in the trial space and independent compact polynomial functions in a Petrov-Galerkin test space, avoiding reliance on a known fundamental solution.
50. Reliability-Aware Hard--Soft Physics-Informed Neural Networks for Robust Learning of Challenging Partial Differential Equations
Core Problem: Physics-informed networks become unstable when exact constraints, localized solutions and uneven data reliability compete within one loss.
Key Innovation: RA-HSPINN retains embedded hard constraints but learns a bounded spatial reliability field that modulates the interior representation, improving difficult localized and multi-mode PDE cases.
51. Trend strength predicts when generative foundation models win: a power-controlled benchmark, a mechanism, and an actionable selection rule
Core Problem: Zero-shot generative forecasters do not outperform classical time-series models uniformly, so aggregate leaderboards give little guidance for model choice.
Key Innovation: A power-controlled benchmark links relative performance to trend strength: the generative model wins 78% of low-trend series but 44% of high-trend series, yielding a simple pre-forecast selection signal.
52. Theory-to-Practice Gap for Neural Networks and Neural Operators
Core Problem: Approximation capacity does not by itself determine how many samples neural networks or neural operators require to learn a target mapping.
Key Innovation: A unified Lp analysis derives best-possible sample-convergence bounds for relevant approximation spaces, sharpening the theoretical gap between representational and statistical complexity.
53. GLUSE: Enhanced Channel-Wise Adaptive Gated Linear Units SE for Onboard Satellite Earth Observation Image Classification
Core Problem: Onboard Earth-observation classification must preserve channel selectivity under severe limits on memory, power and inference time.
Key Innovation: GLUSE adds adaptive gated channel recalibration to a lightweight ResNet, targeting real-time satellite inference without the parameter cost of larger attention modules.
54. Prompt-Calibrated SAM 3 for Open-Vocabulary Remote Sensing Semantic Segmentation
Core Problem: A single class-name prompt underspecifies remote-sensing categories, but naive prompt expansion adds redundant text encoding and propagates noisy masks.
Key Innovation: ProC-SAM3 calibrates prompt construction, caches text features and filters multi-prompt responses; across eight benchmarks it reaches mean mIoU 56.1%, 3.9 points above the previous training-free method.
55. UAV-OVVIS: Unmanned Aerial Vehicles Also Need Open-Vocabulary Video Instance Segmentation
Core Problem: UAV video systems largely track fixed categories with boxes, limiting open-vocabulary queries and pixel-level analysis of dense moving targets.
Key Innovation: AeroVIS supplies 8,279 trajectories across nine categories, and the AeroTrack variants combine open-vocabulary transfer with temporal instance segmentation for long aerial sequences.
56. Hourly-scale modeling of storm transitions in Southern Brazil with Markov Chains
Core Problem: Flood risk in southern Brazil depends on sub-daily storm persistence and transitions that daily rainfall statistics do not represent.
Key Innovation: Season-specific Markov chains fitted to hourly observations from 15 stations over 2007-2024 quantify transitions among storm-intensity states and their persistence.
57. The UWO dataset – long-term observations from a full-scale field laboratory to better understand urban hydrology at small spatio-temporal scales
Core Problem: Urban drainage models are constrained by scarce, internally consistent observations of rainfall, sewer flow, wastewater conditions and sensor performance.
Key Innovation: The UWO release provides three years of 1-5 minute observations from 124 sensors, infrastructure geodata and a validated hydrodynamic model from a full-scale Swiss field laboratory.
58. Benchmarking reservoir operation schemes for large-scale hydrological models
Core Problem: Large-scale hydrological models represent reservoir operations with competing routines that are rarely compared against common observations and calibration targets.
Key Innovation: Five operation schemes are benchmarked across 164 US reservoirs; the mHM routine performs best overall, while its dependence on local demand data limits global deployment.
59. Characterization of mining-induced overburden deformation beneath extremely thick conglomerate using optical fiber frequency shift variation
Core Problem: Extremely thick conglomerate makes it difficult to observe how mining fractures overburden and transfers load to key strata through time.
Key Innovation: BOTDA fibre-frequency shifts are linked to a key-stratum fracture model and validated with physical and numerical experiments, providing continuous signatures of overburden deformation and face weighting.
60. Flood responses of sub-regional river networks to basin-scale flood regulation: a case study of the Hangjiahu Plain, China
Core Problem: Basin-scale regulation can change downstream boundary conditions without revealing how flood levels and drainage safety respond within connected sub-regional channels.
Key Innovation: A Hangjiahu Plain network model shows that the 2023 rules raise peaks by as much as 0.12 m and extend time above safe levels by about 25%, especially during Meiyu rainfall.
61. TCM-CR: Multi-Temporal SAR–Optical Cloud Removal with a Reference Image and Gated Bounded Residual
Core Problem: Cloud-removal models can appear effective when evaluated against weak temporal composites or scene-averaged metrics that hide errors under heavy cloud.
Key Innovation: TCM-CR uses a reference date, SAR-optical temporal cues and a gated bounded residual; on heavily obscured samples it improves peak signal-to-noise ratio by 7.93 dB.
62. An Integrated and Hierarchical Geophysical Workflow for Subsurface Cavity Assessment in Legacy Mining Districts
Core Problem: Abandoned mining galleries create subsidence and collapse hazards whose depth, geometry and continuity are ambiguous to any single geophysical survey.
Key Innovation: A hierarchical workflow combines complementary near-surface geophysical measurements to locate and characterize legacy cavities around Linares, Spain, and rank targets for follow-up investigation.
63. Groundwater contribution to tropical streamflow revealed by radon and stable isotopes
Core Problem: Groundwater support of tropical streamflow is poorly quantified where conventional discharge separation is confounded by rapid rainfall and heterogeneous subsurface flow.
Key Innovation: Radon-222 and stable isotopes trace groundwater inputs to a pristine Brunei catchment, separating subsurface contributions from surface-water mixing.
64. Stochastic Polynomial Surrogate Models for Uncertainty Quantification of Offshore Plate Anchor Capacity
Core Problem: Offshore plate-anchor capacity depends on uncertain soil, load and model parameters that are expensive to propagate through repeated numerical simulations.
Key Innovation: Bayesian inference is coupled with polynomial-chaos surrogates to quantify capacity uncertainty under combined vertical, horizontal and moment loading at reduced computational cost.
65. Communication-related vulnerability at the last mile: a mechanism-based framework
Core Problem: Warnings fail at the last mile when connectivity, trust, language and comprehension barriers interact rather than acting as independent vulnerability indicators.
Key Innovation: A mechanism-based framework traces how communication barriers alter access, interpretation and response, organizing intervention around the causal pathways that produce unequal warning outcomes.
66. Fiber Bragg grating–based monitoring of internal deformation at gully heads in cold-region black soil
Core Problem: Surface surveys cannot show the internal deformation that precedes collapse at erosion-gully heads in seasonally frozen black soil.
Key Innovation: Temperature-compensated fibre Bragg grating sensors are validated against inclinometers and one year of field observations, resolving oscillatory movement with a total displacement amplitude of 30.871 mm.
67. High-resolution floodplain topography mapping from ICESat-2 and Sentinel-2 using slope-corrected inundation frequency
Core Problem: Flood models require continuous high-resolution floodplain elevations, but satellite altimetry samples cross sections while optical imagery provides only inundation frequency.
Key Innovation: A slope-corrected fusion of ICESat-2 terrain profiles and Sentinel-2 inundation frequency reconstructs floodplain topography with validation RMSEs of 1.039-1.375 m across three river tests.
68. A Physics-Informed Extreme Learning Machine framework for steady-state unconfined seepage problems
Core Problem: Unconfined seepage is a free-boundary problem in which hydraulic head and the phreatic surface must be solved together, including across sharp material interfaces.
Key Innovation: A physics-informed extreme learning machine iteratively updates the free surface and enforces head and normal-flux continuity between heterogeneous subdomains in two- and three-dimensional benchmarks.
69. Mechanical behavior and fracturing of sandstone under combined water diffusion-confining pressure: Experimental and numerical studies
Core Problem: Rock strength and fracture development under confinement depend on nonuniform water diffusion, yet most tests assume either dry or uniformly saturated specimens.
Key Innovation: Triaxial experiments across soaking times and confining pressures are paired with meso-damage simulations to resolve how evolving moisture gradients shift sandstone deformation and failure.
70. Numerical Investigation of Storm Surge Effects on Denitrification in Nearshore Aquifers
Core Problem: Storm surges alter coastal-aquifer salinity, redox state and residence time, but their effect on nitrate removal can persist after water levels recede.
Key Innovation: Reactive-transport simulations couple surge and tide forcing and show a rapid rise in denitrification followed by sustained enhancement from lingering redox and transport changes.
71. Point-Selection Fine-Tuning Framework for Robust Point Cloud Classification
Core Problem: Fine-tuning a pretrained point-cloud model can amplify outliers and overwrite robustness learned during pretraining, while lightweight adaptation still processes corrupted points.
Key Innovation: PSFT selects reliable points before parameter-efficient adaptation, limiting the influence of corrupted tokens while retaining the pretrained representation.
72. A round robin study for benchmarking the measurement of soil shrinkage curve (SSC)
Core Problem: Soil shrinkage curves are used in expansive-soil assessment, yet differences in specimen preparation and measurement can obscure reproducibility across laboratories.
Key Innovation: A common procedure, moulds and two clay samples are circulated among 14 Australian and Indian geotechnical groups to quantify inter-laboratory variability and define a repeatable protocol.
73. Investigations on the Advanced SAR Processing Algorithms for Security Applications With Distributed Sensing
Core Problem: Distributed SAR constellations promise shorter revisits and wider access but require processing that remains coherent across separated sensors and acquisition geometries.
Key Innovation: The study evaluates five advanced algorithms across four security-oriented distributed-sensing applications, documenting their performance limits and the processing changes needed for operational use.
74. A data-driven method for identifying climate drivers of agricultural yield failure from daily weather data
Core Problem: The timing and aggregation window of weather drivers create an enormous candidate space for explaining annual crop-yield failure.
Key Innovation: A data-driven search reduces daily meteorology to ten interpretable drivers; logistic models retain correlations of 0.75-0.91 between predicted and observed failure prevalence under a shifted-climate test period.
75. Uplift and seismicity of intraplate regions of India: role of rheology, paleo-and active tectonics
Core Problem: Uplift and earthquakes within stable continental interiors require an explanation that connects inherited structures, present stress and spatially variable lithospheric strength.
Key Innovation: Indian intraplate regions are compared through rheology, palaeotectonic architecture, active faults and seismicity to identify conditions that reactivate crustal blocks and basin structures.
76. Coupled seismic hysteretic degradation in reinforced concrete components based on the Park-Ang damage index
Core Problem: Reinforced-concrete hysteresis models often calibrate stiffness loss, strength deterioration and pinching with separate damage indices that can become physically inconsistent.
Key Innovation: A single Park-Ang damage variable drives all three degradation mechanisms within a bilinear unloading formulation, reducing calibration freedom while retaining two-stage stiffness loss.
77. Monitoring disaster preparedness behaviors in Japan: social, residential, and neighborhood-related gradients
Core Problem: Preparedness is usually reduced to one composite score, concealing how specific behaviours vary with household, housing and neighbourhood conditions.
Key Innovation: A weighted Japanese survey estimates item-level prevalence and uses adjusted quasi-Poisson models to relate four community-perception measures to concrete preparedness actions.
78. Barriers and facilitators for recovery: A qualitative analysis of older Australian's disaster experiences
Core Problem: Older adults encounter distinct barriers during evacuation, assistance and long-term recovery, but these experiences are poorly represented in standardized recovery plans.
Key Innovation: Reflexive thematic analysis of Australian disaster interviews identifies unmet needs, community support, preparation and individual circumstance as interacting controls on recovery.
79. Urban sidewalk detection via a cross-view joint segmentation method using satellite and street-view imagery
Core Problem: Tree and building occlusion breaks sidewalk traces in satellite imagery, while street-view images lack a complete plan-view network.
Key Innovation: S2DB-Net aligns satellite and street-level geometry in a dual branch; it improves intersection-over-union by 6.47-9.18 points over single-view baselines across three datasets.
80. Prescribed fire effects on soil organic matter in a eucalypt forest previously affected by severe wildfire
Core Problem: Prescribed burning reduces fuel after severe wildfire, but its longer-term effect on soil carbon quantity and stability remains uncertain.
Key Innovation: A Greater Sydney field study finds that low-to-moderate prescribed fire reduces soil organic-matter stocks while increasing the thermal recalcitrance of the material that remains.
81. Excitation-time-independent seismic tomography for blast-induced rock damage characterization and its application
Core Problem: Active seismic tomography of blast-damaged rock depends on synchronized source timing and suffers from an ill-posed, computationally expensive inverse problem.
Key Innovation: An excitation-time-independent formulation removes the synchronization requirement and reconstructs velocity changes for laboratory and applied rock-damage characterization.
82. Seismic SSI analyses considering fault site effects using a high-accuracy direct method implemented in implicit ABAQUS
Core Problem: Soil-structure interaction near deep faults requires one nonlinear model to represent layered ground, fault-zone wave effects and a large structure without incompatible subdomain approximations.
Key Innovation: A three-dimensional high-accuracy direct method implemented in implicit ABAQUS performs unified static-dynamic SSI analysis for faulted layered sites.
83. Calculation method for foundation settlement of widened embankments considering lateral deformation
Core Problem: Settlement estimates for widened embankments underpredict deformation when asymmetric confinement drives lateral movement in the new soft foundation.
Key Innovation: The calculation incorporates lateral strain and a time-dependent soil modulus, then checks predicted settlement against monitoring from the widened G15 expressway.
84. An SSC-based deposit-specific framework for preloading embankments on Port Said clay with indicative carbon implications
Core Problem: Staged preloading on Port Said marine clay is difficult to control because low strength, high compressibility and creep evolve throughout construction.
Key Innovation: A deposit-specific PLAXIS Soft Soil Creep model is embedded in an observational workflow to derive deformation-based staging charts, with wider field validation explicitly left outstanding.