TerraMosaic Daily Digest: July 30, 2026
Daily Summary
Landslide studies converge on a common result: prediction improves when mechanics and motion are represented before classification. The scaLr model derives rupture geometry from terrain and material strength, while a Richards-equation-constrained warning model connects transient soil moisture to slope failure. Sentinel-1 analyses recover more than 130 active loess slopes and large-gradient Baige displacement; in Mandi, InSAR adds 36 previously unmapped landslides, increasing inventory completeness by 18%. Controlled inventory removal further shows that a higher AUC can accompany a less faithful map, making physical consistency and label completeness part of model evaluation.
Operational monitoring is expanding across the hazard chain. Background snow depth raises avalanche-forecast AUC from 0.72 to 0.83 for dry events and from 0.85 to 0.94 for wet events. Cryoseismic energy recorded along a two-kilometre fibre array images ice-rich permafrost; physics-informed hemispherical mapping reduces GNSS multipath residuals by 37-40% at three mining-subsidence stations; and three satellite sensors track a volcanic plume at 13-15 km altitude from Ethiopia to India. A rapid EERI assessment then organizes earthquake, tsunami, landslide and infrastructure evidence collected within 48 hours, while stating where field confirmation remains absent.
Large baselines and transfer tests are becoming part of hazard modelling itself. A Chinese archive standardizes 1,057,817 multi-hazard warnings, and U.S. Disaster Normals estimate a recent 30-year annual loss burden of $42.8 billion in 2022 dollars. Controlled inventory removal and InSAR-refined labels show that missing landslides change both maps and apparent validation skill. A global streamflow model pre-trained across 18,588 basins is then adapted to operational weather forecasts, while multimodal flood mapping is tested across seven regions and 38-event mapping evaluates near-real-time performance. Together, these studies treat domain shift, missing evidence and local failure modes as measured properties rather than caveats attached to aggregate accuracy.
Key Trends
Across slope mechanics, monitoring and warning, the strongest results connect antecedent state and evolving geometry to explicit decision thresholds.
- Measured Motion Is Reshaping Susceptibility: InSAR and pixel-offset time series are being used to repair static inventories, locate active slopes and test whether susceptibility maps contain the terrain already moving.
- Monitoring Is Moving from Snapshots to Trajectories: InSAR, GNSS, fibre-optic sensing and multisensor plume tracking resolve displacement, subsurface structure and atmospheric transport through time.
- Forecasts Are Adding Antecedent Physical State: Snow depth, soil moisture, river-ice configuration and lake-entry geometry add the pre-event conditions that weather forcing or static terrain alone cannot recover.
- Rapid Evidence Is Becoming an Operational Product: National warning archives, disaster-loss normals and 48-hour reconnaissance reports turn heterogeneous observations into explicit baselines for warning and deployment.
- Validation Is Shifting toward Transfer and Failure Modes: Spatial blocks, inventory perturbation, cross-region tests and provenance checks show where models fail, what evidence is missing and when review is still required.
Selected Papers
The July 30 literature is anchored by process-based landslide mechanics, deformation-aware inventories and physically constrained warning. Avalanche forecasting, fibre-optic permafrost imaging, GNSS subsidence correction, volcanic-plume tracking and rapid earthquake reconnaissance extend that emphasis into operational monitoring. National warning and loss baselines, inventory perturbation and cross-region models then test whether those observations remain useful under missing data and domain shift.
1. A spatiotemporal dataset of multi-hazard early warnings in China
Core Problem: Multi-hazard warnings are issued through several administrative levels, but no standardized national dataset has preserved their hazard type, location, severity and timing at operational scale.
Key Innovation: The authors release 1,057,817 warnings issued across mainland China from 2022-2025 in a harmonized spatiotemporal schema; agreement above 0.94 with official summaries supports use for warning-system analysis.
2. A physics-based morphometric model to explain the emergence of landslide rupture geometry and to identify hillslope transience
Core Problem: Landslide scaling models usually assume rupture shape, depth or angle, preventing a direct test of how hillslope geometry and material strength generate observed size distributions.
Key Innovation: scaLr analytically derives the optimal rupture depth and angle at each terrain point, clusters cells by a shared daylight point and reproduces observed area-volume, area-depth and shape scaling while also identifying transient hillslopes.
3. Space-time modeling of rainfall-induced landslides via Physics-Informed Neural Networks: Moving toward a new paradigm for regional early warnings
Core Problem: Regional rainfall thresholds omit subsurface hydrological state and static terrain controls, limiting warning reliability and increasing false alarms where observations are sparse.
Key Innovation: A Richards-equation-constrained neural network predicts time-varying volumetric water content and combines it with terrain, vegetation, soil and geology; transfer and sparse-data tests show more physically realistic warnings than rainfall-only intensity-duration thresholds.
4. U.S. Disaster Normals
Core Problem: Weather has standardized climatological normals, but disaster losses lack an equivalent baseline for judging whether an event's societal impact is ordinary or extreme.
Key Innovation: Thirty-year U.S. Disaster Normals derived from SHELDUS estimate a recent annual burden of $42.8 billion in 2022 dollars and an increase of about $20 billion per year relative to 1961-1990, while documenting historical undercounting limits.
5. Freeze–thaw‐related loess landslide accelerations detection in Ili Basin based on multi‐temporal InSAR analysis
Core Problem: Regional links between loess-slope acceleration, freeze-thaw cycling and rainfall remain poorly quantified because deformation and environmental forcing evolve at different timescales.
Key Innovation: Ascending and descending Sentinel-1 time series resolve vertical and eastward motion on more than 130 active slopes; tangential-angle and wavelet analyses link Hujierti acceleration to freeze-thaw, snowmelt and concentrated rainfall and identify angles above 85 degrees as an impending-slide signal.
6. Autofocusing-based adaptive SAR pixel offset tracking for high-accuracy landslide deformation retrieval
Core Problem: Medium-resolution SAR pixel-offset tracking is vulnerable to mismatches and noise, especially where landslides produce localized, large-gradient displacement that phase-based InSAR cannot retain.
Key Innovation: AF-TSPOT refocuses temporally stacked Sentinel-1 intensity images at pixel level and combines multiple baselines, reducing stable-area RMSE by 54.16% and recovering Baige deformation patterns comparable to higher-resolution ALOS-2 observations.
7. Use of PSInSAR for Long-Term Surface Displacement Monitoring as a Complement to the Official Landslide Susceptibility Map in the Saguenay–Lac-Saint-Jean Region, Quebec
Core Problem: A static official susceptibility map may omit slopes that are already moving, but the degree of mismatch with long-term deformation and historical failures was unknown in Quebec's Saguenay-Lac-Saint-Jean region.
Key Innovation: Six years of PSInSAR, validated against four GNSS stations, reveal anomalous scatterers beyond mapped boundaries: the official map contains about 45% of historical failures and 50% of anomalies, whereas anomalies overlap more than 80% of failures where persistent scatterers are observable.
8. Enhancing Earthquake-Induced Landslide Susceptibility Mapping Through Integration of Climatological Soil Moisture: A Hybrid CNN–Swin Transformer Approach
Core Problem: Earthquake-induced landslide susceptibility models seldom represent persistent background wetness, although soil moisture modifies the pre-event stability of mountain slopes.
Key Innovation: A CNN-Swin Transformer integrates climatological soil moisture with twelve other controls and reaches AUC 0.95, while the authors explicitly identify the need to replace the random split with spatial block validation before operational use.
9. Deep Learning Approach for Landslide Delineation in Tropical Region of Malaysia
Core Problem: Tropical landslide scars at different activity stages have weak and changing topographic expression, making automated delineation from LiDAR-derived terrain products difficult.
Key Innovation: Mask R-CNN models tested across terrain-layer combinations, resolutions and network depths perform best for active landslides using 1 m DTM, hillshade and slope inputs, but the 50.7% F-measure exposes poor transfer to dormant and relict features.
10. Is Europe Under UNSEEN Risk of Cyclones of Tropical Origin?
Core Problem: The impacts and potential multidecadal variability of cyclones of tropical origin (CTOs) remain poorly understood, particularly when these events make landfall or pass near Europe (named CTO‐Es) or cause measurable impacts in Europe (named CTO‐EIs).
Key Innovation: Following the UNprecedented Simulation of Extremes using Ensembles (UNSEEN) approach, we generate CTO event sets from twentieth‐century hindcasts. Our results indicate that CTO‐EIs pose a greater risk to the British Isles than to continental Europe.
11. Calculation of Flood Control and Drainage Benefits of Levee Projects Considering Socio‐Economic Changes
Core Problem: Traditional methods for assessing levee projects focus solely on single‐year economic data, evaluating only the reduction in flood‐related economic losses while neglecting the benefits of saving lives and the dynamic changes in the economy and population over the project's design life.
Key Innovation: The study introduces a dynamic evaluation method for flood control and waterlogging drainage benefits of levee projects, incorporating economic and demographic changes. A case study demonstrates that annual benefits increase from CNY 10.6 million in the base year to CNY 34.4 million by the end of the design life, with flood control benefits accounting for 59.1% and drainage benefits for 40.9%.
12. Spatiotemporal Storm Surge Forecasting with a Deep Learning Model Using an Augmented Typhoon Track Dataset and Meteorological Updates
Core Problem: Storm-surge forecasts must follow changing cyclone tracks and intensity, whereas models driven only by fixed historical inputs cannot update the full spatial field as a storm evolves.
Key Innovation: An Adaptive Fourier Neural Operator ingests updated typhoon and meteorological data at 0.05-degree resolution; two storm tests produce 48-hour spatial RMSE of 0.27-0.28 m and tide-gauge RMSE of 0.34-0.35 m.
13. Impact of warming on rainfall changes in damaging Philippine typhoons using high-resolution convection-permitting models
Core Problem: The thermodynamic and dynamical contributions to Philippine typhoon rainfall under warming cannot be separated reliably with coarse climate simulations.
Key Innovation: Convection-permitting pseudo-global-warming experiments attribute most future rainfall growth to a 20-30% thermodynamic contribution, with case-specific dynamics partly offsetting it and local extreme rainfall increasing by more than 30-40%.
14. Assessment of 2025 cloudburst-induced flash flooding in Thunag village, Mandi, Himachal Pradesh, India
Core Problem: Cloudburst-induced flash floods are among the most destructive hydro-meteorological hazards in the Himalayas, yet their localized nature makes them difficult to assess using sparse ground observations.
Key Innovation: The study investigates the June 29-July 1, 2025 cloudburst-induced flash flood in Thunag village, Himachal Pradesh, using an integrated remote sensing and hydrodynamic modelling framework. GPM IMERG captured the spatiotemporal evolution of the localized cloudburst, while simulations reproduced rapid runoff concentration, high flow velocities, and severe channelized flooding associated with floodplain encroachment.
15. Flood risk assessment and driver analysis in mining subsidence-affected high-groundwater watersheds: a coupled HEV and machine learning approach
Core Problem: Conventional flood risk assessments often treat mining areas as ordinary land-use units and insufficiently account for the coupled effects of subsidence-induced depressions, shallow groundwater and recurrent inundation.
Key Innovation: To address this gap, this study develops an integrated flood risk assessment and attribution framework for high-groundwater mining areas. The results show that high flood risk is mainly concentrated in downstream lake districts, urbanized and riparian areas, and mining-affected regions.
16. Evaluation of Satellite-Derived Topography Differencing for Volume Estimation of Landslides Triggered by the 2016 Mw 7.8 Kaikoura Earthquake
Core Problem: Regional earthquake-landslide inventories rarely contain direct volumes because airborne LiDAR is costly, leaving most volume estimates dependent on empirical area scaling.
Key Innovation: Open-source SETSM photogrammetry differences pre- and post-earthquake satellite DSMs for 1,448 Kaikoura landslides, validates the volumes against LiDAR and quantifies errors from vegetation, registration and surface modelling.
17. Assessment of Flood Risk Using Remote Sensing and GIS Techniques Based on the Analytic Hierarchy Process (AHP) and Fuzzy AHP (FAHP) in the R’Dom Watershed (Meknes, Morocco)
Core Problem: Flooding is one of the most damaging natural hazards worldwide, particularly in data-scarce watersheds where long-term hydrometeorological records are limited.
Key Innovation: The study integrates remote sensing, GIS, AHP and fuzzy AHP to assess flood susceptibility, vulnerability and relative risk in Morocco's R’Dom watershed using seventeen conditioning factors. Validation uses 900 flood and non-flood locations, with testing AUC values from 0.767 to 0.935.
18. Tsunami-induced wave pressure and forces: a review of clear water to black tsunamis
Core Problem: Clear-water tsunami loads have mature engineering formulations, but sediment-laden black-tsunami pressure and force estimates remain too configuration-dependent for design standards.
Key Innovation: The review separates depth- and drag-coefficient approaches by structure type and shows that black-tsunami prediction is constrained by narrow experiments, limited validation and inconsistent definitions, establishing priorities for a transferable load framework.
19. Spatial relationship between earthquakes and volcanoes in Türkiye: do large earthquakes wake volcanoes?
Core Problem: Earthquakes can perturb volcanic systems, but nationwide proximity alone cannot establish whether Turkish volcanoes are likely to respond after strong shaking.
Key Innovation: A rupture-length-scaled Triggering Index classifies 383 Mw 5.5 or greater earthquakes relative to ten volcanic centres; it identifies eastern Anatolian monitoring priorities while explicitly treating low index values as screening evidence, not proof of eruption triggering.
20. Disaster patterns and quantitative risk research of dam-break debris flows in high-mountain and steep-gorge regions: a case study of the Yizhong River Basin in Deqin, Yunnan
Core Problem: In high-mountain and steep-gorge regions, traditional hydrological methods often underestimate the peak discharge of dam-break debris flows because they simplify the chain-like transformation from landslide blockage to rapid breaching.
Key Innovation: The study focuses on the Yizhong River in Deqin, Yunnan, a typical landslide-blockage-dam-break disaster chain developed in a confined V-shaped gully. The results show that topographic confinement produces a pronounced funnel effect, restricts lateral spreading, and promotes rapid conversion from stored potential energy to downstream kinetic energy during breaching.
21. An overview of the formation mechanism characteristics of glacial lake outburst floods in Tibet, China
Core Problem: However, geographical challenges (extreme cold, high altitude, steep terrain) and methodological shortcomings (singular approaches, limited evidence) hinder GLOF formation mechanism research.
Key Innovation: The study addressed these challenges through a combination of methods, including field investigation, remote sensing, and the analysis of archival data to compile 47 GLOF events from 39 Tibetan glacial lakes since the 20th century, documenting information such as locations, dates, and triggering factors.
22. Integrating multi-sensor data with LSTM for accurate landslide prediction and risk assessment
Core Problem: Threshold and single-sensor warning systems cannot represent the nonlinear interaction among rainfall, pore pressure, soil moisture and slope displacement, and minority high-risk states are easily missed.
Key Innovation: A multi-sensor LSTM fuses rain gauges, piezometers, inclinometers and tensiometers with engineered displacement features; oversampling and tuning improve minority-risk detection and reduce false alarms on Meghalaya monitoring data.
23. Groundwater and hydrogeochemical patterns of Campi Flegrei active caldera (southern Italy) for volcanic hazard assessment
Core Problem: Hydrothermal signals at Campi Flegrei are embedded in a complex volcanic aquifer, but long-term groundwater flow and chemistry had not been synthesized into a monitoring design.
Key Innovation: Decades of well, water-table and hydrochemical observations reveal an internally isolated, dome-shaped aquifer with northern bicarbonate waters and warmer southern chloride-magnesium waters, identifying wells that can track hydrothermal and bradyseismic change.
24. Shear-strength degradation and damage evolution of the potential sliding surface in a reservoir-bank deposit landslide
Core Problem: Stability estimates for heterogeneous reservoir-bank deposits depend on shear strength along concealed sliding surfaces, yet water, rate and coarse fragments influence that strength differently.
Key Innovation: Large-diameter ring-shear tests and a Weibull damage model show that water controls peak and residual strength, shear rate controls displacement to the residual state and 5-13% medium gravel has limited effect, linking strain softening to damage evolution.
25. Enhanced landslide detection via multi-module deep learning: a morphology-dependent performance analysis approach
Core Problem: Landslide segmenters are rarely ablated by architectural component or evaluated by landslide shape, obscuring why elongated scars remain harder to map than compact failures.
Key Innovation: EfficientNet-B4 with ASPP, squeeze-and-excitation and PANet reaches 90.06% IoU and 95.18% recall on Bijie; module ablations and morphology-stratified errors quantify weaker performance on long-strip landslides.
26. A case study integrating remote sensing and numerical simulation: structure characterization and mechanism investigation of a deformed rock mass
Core Problem: Different sectors of a deformed rock mass display toppling, sliding, crushing and disturbance, obscuring the structures and forcing that produced the composite deformation.
Key Innovation: UAV photogrammetry, GIS structural interpretation, kinematic analysis and UDEC simulation partition the mass into three deformation zones and show that discontinuity architecture dominates toppling-sliding while topographic amplification may strengthen seismic effects.
27. Formation mechanisms of loess-mudstone landslides under climatic control: exemplified by Doujitai landslide, Southern Chinese Loess Plateau
Core Problem: The coupled roles of rainfall and the loess-mudstone interface in Southern Chinese Loess Plateau failures are difficult to isolate from tectonic preconditioning.
Key Innovation: Ring-shear tests and hydro-mechanical simulation reveal non-monotonic strength changes with water content, a hardening-to-softening transition and rapid stability loss under intense rainfall, identifying extreme rain as the external trigger on a preconditioned slope.
28. Optimization of landslide parameter sequences based on time window denoising and dynamic process noise kalman filtering technique
Core Problem: Unstable data sequences pose significant challenges to the accurate characterization of landslide trends and the early warning of such disasters.
Key Innovation: An optimized method based on the time window method, Kalman filtering technique (KFT), Pauta criterion, and Weibull distribution has been proposed to eliminate abnormal fluctuations and data loss. The effectiveness of the optimization method was validated through the processing of monitoring data from the Hainan-Shilu Iron Mine in China.
29. Learning from past earthquakes: a comparative analysis and unified framework for proactive seismic risk management of vernacular heritage in historic urban landscapes
Core Problem: Post-earthquake surveys of vernacular heritage remain reactive, institutionally fragmented and biased toward demolition, preventing observations from becoming reusable risk-reduction evidence.
Key Innovation: A four-pillar Resilient Vernacular Heritage framework links pre-event baselines, tiered reconnaissance, interoperable data and knowledge reuse, drawing comparative design rules from Mexico, Chile, Italy and Croatia.
30. An enhanced GRACE-based index for nationwide flood-prone region monitoring in China from 2003 to 2022
Core Problem: Floods are among the most destructive hydrometeorological disasters worldwide, highlighting the need for reliable large-scale monitoring of flood-prone conditions.
Key Innovation: The study constructed the Optimal Flood Potential Index (OFPI) for higher-accuracy and broader-scale monitoring of storage-based flood-prone regions, based on a monthly terrestrial water storage anomaly (TWSA) dataset with improved spatial resolution (0.5°).
31. AIFL: A Global Daily Streamflow Forecasting Model Using a Deterministic LSTM Pre-Trained on ERA5-Land and Fine-Tuned on IFS
Core Problem: Global streamflow models trained on reanalysis lose skill when driven by operational forecasts with different biases and error structure.
Key Innovation: AIFL pre-trains an LSTM on 40 years of ERA5-Land across 18,588 CARAVAN basins and fine-tunes it on IFS forecasts; on 2021-2024 data it reaches median KGE' 0.66 and NSE 0.53 and outperforms single-stage forcing strategies.
32. Effect of landslide location on glacial lake outburst flood at Badongcuo, Tibetan Himalaya
Core Problem: GLOF scenarios often simplify where a landslide enters a lake, although entry position, water depth and lakebed geometry determine impulse-wave energy and dam overtopping.
Key Innovation: A field-, UAV- and satellite-constrained multiphase model simulates the full landslide-wave-piping/overtopping-breach chain at Badongcuo; severe scenarios peak at 87,345 cubic metres per second and reach planned railway infrastructure in about 44 minutes.
33. A Simplified Mitigation Model of Chure Landslide: A Case Study of Setebhir Landslide, Makwanpur, Nepal
Core Problem: The unstable Setebhir slopes lack a tested sequence of affordable measures that can raise factors of safety across contrasting Chure-region sections.
Key Innovation: Two-dimensional finite-element tests compare drainage, vegetation and regrading; slope modification raises factors of safety from 1.00 to 1.81 and 1.73, while the full result set correlates 0.956 with prior evidence.
34. Geospatial Integration of Landslide Susceptibility and Road Network Vulnerability in Mountainous Terrain
Core Problem: Transportation networks in hilly regions are often subject to landslides, which induce large socio-economic losses and impede emergency response and regional communication.
Key Innovation: This research develops a geospatial machine learning framework for landslide susceptibility mapping and road network vulnerability assessment in Kaski, Nepal, a mountainous region characterized by complicated topography, intense monsoonal rainfall, and fast land use change. The model showed good predictive performance with 93% accuracy, recall, and F1-score, 98% AUC, and a Kappa value of 0.861.
35. Inventory Incompleteness-Induced Uncertainty in Machine Learning-Based Landslide Susceptibility Modeling: A Comparative and Interpretable Analysis
Core Problem: Landslide inventories are systematically incomplete, but standard random train-validation splits can hide how missing labels alter susceptibility maps and feature attribution.
Key Innovation: Nested 10-40% removal, accessibility-biased omission and spatially clustered omission across RF, MLP and logistic models show non-monotonic AUC and map divergence; a higher AUC does not necessarily imply closer agreement with the complete-inventory map.
36. Landslide susceptibility prediction in a mediterranean environment using multi-source geospatial data and bivariate probabilistic methods
Core Problem: Steep terrain, weak marl and clay, monsoon rainfall and road disturbance combine in northwestern Setif, but their spatial contribution to landslide susceptibility had not been compared across bivariate models.
Key Innovation: An inventory of 61 field-validated landslides and nine conditioning factors is evaluated with four probabilistic methods; Weight of Evidence performs best at AUC 0.925 and consistently locates high susceptibility near weak lithology, drainage and roads.
37. Machine-Learning-Based Landslide Susceptibility Mapping Using an InSAR- Refined Inventory in the Mandi district, Northwestern Himalaya, India, Himalaya region
Core Problem: Static Himalayan inventories omit deforming slopes and therefore age quickly, while non-spatial validation can overstate susceptibility-model transfer.
Key Innovation: Persistent-scatterer and small-baseline InSAR add 36 landslides and increase inventory completeness by 18%; nested spatial blocks then show 12-15% AUC gains for the refined inventory, with XGBoost performing best at a 3 km block size.
38. Spatial distribution of earthquake-induced landslides in Southern Tianshui Loess area, Gansu Province
Core Problem: The spatial controls on landslides from the historical 1654 Tianshui earthquake remain difficult to reconstruct from incomplete event evidence.
Key Innovation: Remote sensing and field verification map 753 failures, while principal-component analysis isolates elevation, lithology, intensity and land use; all mapped landslides lie at intensity IX or greater and 28.02% occur within 4 km of faults.
39. Combining UAV SAR Tomography and Photogrammetry to study an Active Volcanic Vent in Iceland
Core Problem: Optical surveys reveal an active volcanic vent's surface but cannot resolve the density structure of its shallow conduit and loose cinder cone.
Key Innovation: Helical P-band UAV tomography combined with RGB, thermal and photogrammetric surveys images up to 20 m below the Sundhnukur vent, separating a high-intensity central structure from lower-intensity cone slopes.
40. Effectiveness of Engineered Tsunami Mitigation Measures: A Review of Current Approaches and Research Needs, Part I: Field Surveys
Core Problem: The performance limits of seawalls, breakwaters and water-filled canals cannot be generalized from a few tsunami events without separating geometry, maintenance and integration with evacuation.
Key Innovation: A field-survey synthesis shows that engineered barriers can dissipate wave energy and delay inundation, but identifies long-term reliability, cost, configuration and adaptation across tsunami regimes as unresolved design constraints.
41. Flash Flood Forecasting Based-EF5 Model using Distinct Interpolation Methods: An Ensemble Framework
Core Problem: Flash-flood forecasts in a multi-river district depend on spatially interpolated rainfall, but the effect of interpolation choice and lead time on usable warning skill was unresolved.
Key Innovation: EF5 tests show spline interpolation exceeding 0.9 for both correlation and Nash-Sutcliffe efficiency, inverse-distance weighting near 0.8 and Kriging near 0.4-0.6, with skill falling rapidly at longer lead times.
42. Mapping Recovery Resilience Pathways After the 2018 Palu Liquefaction: A Multi-Index Google Earth Engine Framework for Post-Disaster Land Systems
Core Problem: Vegetation-only recovery metrics conflate return to prior conditions, land-use transformation, reconstruction and persistent degradation after liquefaction.
Key Innovation: A Google Earth Engine Recovery Resilience Index combines Sentinel-2 vegetation, built-up and bare-soil signals with Dynamic World labels; 962-cell and 14,157-cell analyses recover four distinct post-Palu pathways and remain stable across grid scales.
43. Modeled Subsidence of Permafrost Terrain 2025–2050: North‐Central Alaska
Core Problem: Long-term permafrost subsidence observations cover only a few field sites, leaving the 2025-2050 deformation expected along Alaska's Dalton Highway spatially unresolved.
Key Innovation: A thermal-regime model coupled to land-cover and excess-ice maps estimates 1.79-2.33 cm per year in the southern study area under 25% and 50% ice scenarios, compared with less than 0.17 cm per year on the North Slope.
44. Near Real-Time Flood Mapping from Sentinel Data Using Machine Learning Techniques
Core Problem: Public flood datasets do not jointly provide the coverage, consistency and thematic diversity required for near-real-time Sentinel flood mapping.
Key Innovation: A modular pipeline co-registers and fuses satellite, terrain and Copernicus Rapid Mapping labels from 38 events in 2022-2025, producing a validation IoU of 0.70 and an operationally extensible flood-mask workflow.
45. Persistent Mining-Induced Subsidence Two Decades After Underground Coal Exploitation: Evidence from Multi-Temporal GNSS Monitoring
Core Problem: Ground deformation can continue long after coal extraction ends, but residual subsidence is seldom measured over multi-decadal intervals at the same benchmarks.
Key Innovation: Seventeen GNSS benchmarks reoccupied after 19 years show subsidence at every site, with cumulative displacement from -0.082 to -3.853 m and normalized annual rates up to -0.203 m per year.
46. Probabilistic assessment of warning-stage exceedance during ice-jam flooding using physics-guided river-ice simulations and interpretable stacking ensembles
Core Problem: Ice-jam stages rise abruptly, but nonlinear ice-hydraulic interactions and sparse observations make warning-stage exceedance probabilities difficult to estimate directly.
Key Innovation: RIVICE generates 10,000 physics-consistent scenarios and an interpretable stacking surrogate reproduces stages with NSE 0.95; SHAP identifies jam-toe location, upstream discharge and downstream ice thickness as the principal controls.
47. Reliability-adaptive graph-cut phase unwrapping for InSAR geohazard deformation monitoring
Core Problem: Dense residues, decorrelation and Itoh-condition violations destabilize phase unwrapping in high-gradient InSAR scenes used for geohazard monitoring.
Key Innovation: EGCPU builds a reliability-weighted graph from coherence, wrapped gradients and circular dispersion, then solves robust integer ambiguities by graph cuts; GNSS and extensometer tests yield 1.81 cm RMSE and 1.10 mm MAE, respectively.
48. Seismicity-Based Clues of Crustal Fluids in the 2021 M6.4 Yangbi Earthquake Sequence, Yunnan, China
Core Problem: Foreshocks and a rapidly changing background rate preceded the 2021 M6.4 Yangbi earthquake, but the role and direction of crustal-fluid migration were unresolved.
Key Innovation: ETAS background-rate and Coulomb-stress analyses show an increase from 0.2 to 15 events per day, an approximately 50% post-mainshock stress-rate rise and southward migration consistent with fluid diffusion.
49. Site-Specific Probabilistic Seismic Hazard Analysis for Sakarya Province
Core Problem: Regional seismic maps and national code values do not resolve basin-scale site amplification in Sakarya near the North Anatolian Fault.
Key Innovation: OpenQuake logic-tree PSHA combines homogenized 1900-2023 seismicity with 2,044 in-situ Vs30 measurements and USGS data, yielding basin hazard estimates above both TBEC-2018 and ESHM20.
50. Spatio-Temporal Characteristics of Extreme Precipitation and Flood Risk Assessment: A Case Study of the Yangtze River Delta Region in China
Core Problem: Yangtze River Delta flood assessments rarely combine the frequency, duration and intensity of extreme rain with exposure, environmental sensitivity and prevention capacity.
Key Innovation: Four ETCCDI indices from 107 stations over 1960-2024 feed an AHP-entropy risk model; very wet-day precipitation rises 1.7 mm per decade and the spatial analysis separates southern typhoon-topography controls from broader regional risk.
51. Ground motion characteristics of the 2022 Mw 5.9 Keng Tung earthquake in the northern Sunda Block
Core Problem: Sparse seismic networks and limited formal felt-report systems leave ground-motion behaviour poorly constrained in the northern Sunda Block.
Key Innovation: Community reports and instrumental records independently recover an elongated intensity field consistent with southwestward rupture directivity and identify regional intensity and ground-motion equations suitable for future hazard analysis.
52. Multi-hazard environmental risk assessment for electricity substations: integrating climate projections with atmospheric corrosion modelling *
Core Problem: Electricity substations face accelerating environmental stress from climate change, atmospheric corrosion, seismic hazard, and ageing infrastructure, yet quantitative multi-hazard risk frameworks at substation resolution remain scarce.
Key Innovation: The authors present an open, replicable framework that integrates CMIP6 climate projections, ISO 9223:2012 atmospheric corrosion classification, and a five-state Markov degradation model to produce location-specific expected time to critical-condition (ETTC) estimates for 159 720 substations across 23 OECD countries plus Greenland.
53. Agricultural flood risk under seasonally varying rainfall extremes and crop vulnerability
Core Problem: Annual design storms ignore crop phenology, seasonal exposure and dependence among rainfall durations, biasing agricultural flood-loss estimates.
Key Innovation: A copula-based multivariate frequency model, stochastic rainfall disaggregation, hydraulics and crop depth-duration functions generate monthly risk; long events drive the largest losses and single-duration storms understate dependence uncertainty.
54. Large scale cross-regional remote sensing flood monitoring framework for operative mapping and impact analysis
Core Problem: Flood segmenters trained in one region often fail across diverse landscapes, while annotated events are too sparse for dependable large-area impact mapping.
Key Innovation: A 21-channel SAR-optical-terrain framework compares supervised U-Net++ with self-supervised AnySat across seven Russian regions and carries the best inundation map into official-formula impact estimates that closely reproduce the 2019 Tulun assessment.
55. Forecasting different types of avalanches based on snowpack and snowfall/snowmelt conditions on the Southeastern Tibetan Plateau
Core Problem: Avalanche forecasts based only on snowfall or snowmelt omit background snowpack state and may miss the different triggering behaviour of dry- and wet-snow avalanches.
Key Innovation: Separate logistic models for 37 observed events show that adding snow depth raises AUC from 0.72 to 0.83 for dry avalanches and from 0.85 to 0.94 for wet avalanches while reducing wet-event false positives.
56. Distributed Acoustic Sensing Recordings of Cryoseismicity Enable Seismic Imaging of Permafrost in Remote Arctic Regions
Core Problem: Remote Arctic permafrost is spatially heterogeneous, but conventional seismic surveys are too sparse and episodic to support repeat characterization of ice-rich layers.
Key Innovation: A two-kilometre distributed-acoustic-sensing array uses natural thermal-contraction cryoseisms for dispersion analysis and shear-wave inversion, imaging deep massive ground ice without an active source.
57. EERI Virtual Earthquake Reconnaissance Team (VERT) Phase 1 Report on the 2026 Mindanao, Philippines Earthquake, in EERI 2026 Mindanao, Philippines Earthquake Reconnaissance
Core Problem: The first hours after a major earthquake contain actionable but uneven public evidence, and rapid virtual assessment must distinguish provisional observations from field-confirmed damage.
Key Innovation: EERI's Phase 1 VERT report organizes evidence collected within 48 hours of the M7.8 Mindanao earthquake across aftershocks, response, tsunami, landslides, buildings, health and lifelines, with limited ground truth stated explicitly.
58. Long Range Transport of the Hayli Gubbi Volcanic Plume Over the Indian Subcontinent
Core Problem: Volcanic ash and sulphur dioxide can cross national boundaries at high altitude, but no single sensor resolves source, horizontal transport, concentration and plume height.
Key Innovation: MODIS, TROPOMI and IASI jointly trace the Hayli Gubbi plume from Ethiopia to India, measuring peak sulphur dioxide near 0.095 mol/m2 and placing the plume at roughly 13-15 km under a strong subtropical jet.
59. Physics‐Informed Hemispherical Mapping for Global Navigation Satellite System Multipath Mitigation in Mining Subsidence Monitoring
Core Problem: Directional multipath remains a centimetre-scale error source in GNSS precise-point-positioning for mining-subsidence monitoring, especially around reflective infrastructure.
Key Innovation: A physics-informed hemispherical map combines repeatability, smoothness and elevation attenuation with a compact neural model, reducing residual RMS by 37.1-40.1% and one-hour PPP errors from 3.56/5.66 cm to 2.31/3.44 cm horizontally/vertically.
60. Independent Component Analysis (ICA) as a Superior Atmospheric Correction Method for InSAR Time Series
Core Problem: Weather-model atmospheric corrections perform poorly where subtle volcanic deformation covaries with steep topography and seasonal delay.
Key Innovation: LiCSAtmo uses spatial- or temporal-domain independent component analysis to remove empirical atmospheric modes; at Vesuvius it corrects a false seasonal reversal and lowers GNSS-referenced RMSE from 0.017 to 0.006 m.
61. Hydrodynamic interference and load evolution of staggered cylinder pairs under dam-break waves: An active learning-based Kriging modeling framework
Core Problem: Evaluating hydrodynamic loading on twin-pier coastal bridges under extreme tsunamis is vital, yet interference mechanisms remain insufficiently understood.
Key Innovation: The study integrates high-fidelity Computational Fluid Dynamics (CFD) dam-break simulations with an adaptive Kriging surrogate model using a weighted K-means active learning strategy. Results reveal that the impulse stage is dominated by staggered angle θ, yielding a load reduction of 58.15%, while θ ≈ 40° marks the boundary between load reduction and amplification during the quasi-steady stage.
62. Seismic performance of large-scale OWT monopiles in liquefiable soils: A numerical study and analytical relationships
Core Problem: Large offshore-wind monopiles interact nonlinearly with liquefying soil and the turbine superstructure, yet design relationships cover only a narrow range of turbine, soil and earthquake conditions.
Key Innovation: More than 700 validated three-dimensional simulations span turbines up to 15 MW and produce simplified front-end design expressions; scour-protection configurations also materially improve seismic response.
63. Analytical study of transient seabed liquefaction under nonlinear wave-current effect: Mechanisms and regime identification
Core Problem: Transient seabed liquefaction, driven by effective stress reduction associated with wave-current-induced pore pressure oscillations, is a major geohazard affecting offshore infrastructure.
Key Innovation: The study develops an analytical model to investigate seabed response under wave-current loading and its implications for liquefaction potential. The model integrates a nonlinear wave-current interaction model with Biot's poroelastic theory, yielding closed-form solutions for pore pressure and stress fields that are validated against classical analytical solutions and published experimental data.
64. Storm training set selection for landfalling tropical cyclone surge surrogate modeling
Core Problem: Storm-surge surrogates can look accurate in aggregate while systematically underpredicting the rare high surges that control hazard decisions.
Key Innovation: Tests across interpolation, Kriging and neural surrogates show extreme-tail bias persists across sampling schemes; larger training sets reduce but do not remove it, and Kriging is more sensitive to storm sampling than neural networks.
65. Atmospheric and cryospheric observations in the high-altitude Zarafshon River Basin and the Hydrographic Party Glacier (GGP), Tajikistan, 2018–2025
Core Problem: Despite their importance, observations of cryospheric and atmospheric variables are scarce in this area but are essential to assess the temporal and spatial changes induced by climate change.
Key Innovation: To address this gap, we present a diverse data set of cryospheric and atmospheric variables from the Zarafshon River Basin and the Hydrographic Party Glacier (GGP) in Tajikistan, spanning 2018-2025.
66. Drought-Triggering Thresholds and Vegetation Resilience Across Aridity Gradients in Central Asian Grasslands
Core Problem: Drought is a key climatic driver of grassland degradation; however, the coupling mechanisms between drought-triggering thresholds and ecosystem resilience under different hydroclimatic conditions, as well as their spatial heterogeneity, remain insufficiently understood.
Key Innovation: Here, we investigated Central Asian grasslands by integrating Copula-based joint probability analysis, drought-triggering threshold identification, and quantitative assessment of vegetation resilience. Results showed that the probability of LAI loss (LAI ≤ 40th percentile) increased from 36.35% under mild drought to 39.99% under extreme drought, while more severe losses (LAI ≤ 10th percentile) increased from 11.16% to 12.62%.
67. An experiment and application of flood inundation simulation in urban areas
Core Problem: Due to significant differences in scale between individual streets and the whole urban area, modeling urban areas poses challenges and requires special considerations.
Key Innovation: The study presents an experiment using simplified urban areas represented by square blocks to highlight the influence of street width and building obstruction on flood progression. Additionally, the two-dimensional (2D) surface flood model (the UOL model) and the porosity hydrodynamic model (the MLIT model) are employed to simulate the flooding within building areas, with experimental results serving as validation and evaluation benchmarks for both modeling approaches.
68. A hybrid LightGBM model for urban flood susceptibility mapping based on meta-heuristic algorithms
Core Problem: Urban susceptibility models require both stable tuning and an explanation of how terrain, vegetation and rainfall shape the resulting map.
Key Innovation: WHD-LightGBM combines three metaheuristics for tuning and pairs SHAP with partial-dependence analysis; it reaches F1 0.8315 on a Sentinel-2 Nanchang inventory and identifies altitude, NDVI and rainfall as dominant controls.
69. Women facing flood risk: exploring perceptions and indexing social vulnerability in the Vega Baja del Segura, Alicante (Spain)
Core Problem: Flood-risk indices often omit gendered experience and preparedness, limiting their ability to explain why social vulnerability differs among exposed communities.
Key Innovation: A Female Social Vulnerability Index combined with questionnaires and narratives from 53 women links mapped vulnerability, direct flood experience and perceptions of municipal management to awareness and preparedness.
70. Assessment of drought hazard change across agro-climatic regions of India under 2 °C, 3 °C and 4 °C warming levels based on CMIP6 projections
Core Problem: However, comprehensive evaluations of composite drought hazards across the nation’s agro-climatic zones at specific warming levels are limited.
Key Innovation: Objective A composite drought hazard index (DHI) was developed for 15 agro-climatic zones in India. The analysis was further refined using grade-transition matrices, seasonal aggregation, independent historical validation, and the attribution of precipitation versus potential evapotranspiration.
71. Seismic response of monopile-supported offshore wind turbines considering post-scour conditions and wind-wave loads
Core Problem: Offshore-wind seismic assessments usually separate scour and wind-wave loading even though both alter monopile stiffness, deformation and serviceability during earthquakes.
Key Innovation: Centrifuge-validated simulations show that scour effects depend on motion frequency and that combined wind-wave and seismic loads amplify lateral displacement, rotation and permanent deformation as scour deepens.
72. Beyond coherence: scale-invariant phase quality estimation via gradient-field modeling
Core Problem: Conventional coherence confuses deterministic phase gradients with noise and changes with window size, biasing quality maps in dense-fringe InSAR scenes.
Key Innovation: A low-order gradient-field model estimates noise from fitting residuals and normalizes it to a scale-invariant quality metric that remains stable across fringe density, gradient and window size in simulations and two real cases.
73. Analytical solution for longitudinal response of shield tunnel under normal fault dislocation considering segment joints based on the state-space method
Core Problem: Continuous-beam tunnel models cannot reproduce segment-joint opening, offset and discontinuous longitudinal deformation across a normal fault.
Key Innovation: A nonlinear state-space solution couples joint shear, rotation and axial stiffness with yielding foundation zones; validation and sensitivity tests distinguish controls from fault offset, dip, fracture-zone width and joint stiffness.
74. Nonlinear dynamic analysis and damage evaluation of Pipe-Roof Pre-Construction Tunnels under Mainshock-Aftershock Sequences
Core Problem: Although widely used in urban underground engineering, their seismic behavior under mainshock-aftershock (MS-AS) sequences remains unclear, particularly regarding nonlinear response evolution and cumulative damage.
Key Innovation: The study investigates how mainshock-induced damage affects subsequent aftershock responses in PPTs. The results show that damage growth is limited when α ≤ 0.6, whereas for α ≥ 0.8 tensile damage propagates significantly along the arch-foot-sidewall region, indicating accelerated deterioration.
75. Climate network-based synchronized structural identification of extreme drought and pluvial events in cross-basin regions
Core Problem: Global warming has intensified the frequency and intensity of synchronous climate extremes, posing severe threats to the water‒food‒energy‒ecosystem nexus and challenging regional sustainability.
Key Innovation: Our analysis reveals a distinct wet‒dry co-variability between the Yangtze and Yellow River Basins and elucidates the physical coupling between extreme hydroclimatic events and circulation anomalies, as well as the moisture transport pathways.
76. Distributed multi–shaking-table array tests on long shield tunnels: Deformation control mechanism of a polyurethane isolation layer
Core Problem: Conventional shared soil boxes reflect waves and cannot reproduce spatially varying input along long segmented tunnels, obscuring how isolation layers control deformation.
Key Innovation: A synchronized distributed shaking-table array tests a 1:18 tunnel and shows polyurethane isolation reduces critical strain, ring shear and joint opening by about 25-56% through deformation redistribution rather than acceleration reduction.
77. An energy-constrained PD–DEM framework for seismic soil–structure interaction and residual-performance assessment of underground systems
Core Problem: Underground seismic models struggle to conserve energy while resolving contact slip, fracture, damping and residual deformation through the static-to-dynamic transition.
Key Innovation: An adaptive energy-constrained peridynamic-discrete-element framework separates three dissipation modes and adds residual-performance metrics; two shaking-table benchmarks reproduce peak-response ranking while exposing coarse-graining and boundary sensitivity.
78. The impact of construction method on the seismic response of tailings sand dams
Core Problem: Tailings-dam construction sequence sets the pre-earthquake stress state, but downstream and centreline methods are rarely compared under identical calibrated material and ground-motion conditions.
Key Innovation: Coupled hydro-mechanical finite-element simulations reconstruct construction and operation before the 1985 and 2015 Chile earthquakes, revealing construction-dependent deformation mechanisms that initiate at the dam-slimes interface.
79. Three-directional shaking table test of seismic response on twin stacked tunnels consider subway train load in clayey soil
Core Problem: Twin stacked tunnels are highly susceptible to engineering disasters under the combined action of subway train load and seismic excitation.
Key Innovation: The study conducted a series of 3D shaking table tests to investigate the natural frequency and damping ratio, and seismic responses of soil and tunnel considering subway train load in clayey soil. The results indicate that with the PGA increase, the natural frequency initially decreases from 6.84 Hz to 5.86 Hz and then rises to 6.40 Hz.
80. Deformation Pattern Modifications Induced by 2021 Brentonico Earthquake: Insights from EGMS Ortho Products
Core Problem: It remains uncertain whether an ML 3.5 earthquake can produce detectable, spatially coherent changes in satellite-derived deformation trends.
Key Innovation: European Ground Motion Service time series around the 2022 Brentonico event reveal localized acceleration, deceleration and direction reversal, defining an observational test for subtle low-magnitude coseismic effects.
81. Monitoring and Mapping of fast and slow subsidence in hard rock metal mining using SAR Interferometry Techniques on high resolution TSX/TDX Satellite Data
Core Problem: Hard-rock metal mines contain both fast and slow deformation, but no single InSAR chain reliably preserves both modes over complex workings.
Key Innovation: High-resolution TerraSAR-X/TanDEM-X data processed with coherent small-baseline time series, stacked DInSAR and single-reference PSI locate persistent fast and slow motion around one trough while showing most of Mine-B remained stable in 2023-2024.
82. Seismic Fragility of Gravity and Semi‐Gravity Retaining Walls and Its Impact on the Functionality Loss of Road Infrastructures
Core Problem: Retaining walls are key geotechnical components of road infrastructure, whose seismic performance directly affects the resilience of transportation networks.
Key Innovation: The study presents a new methodology for developing fragility and functionality loss functions of gravity and semi‐gravity cantilever retaining walls, suitable for large‐scale applications. Such influence was considered by calibrating a predictive equation of the displacement as a function of the peak ground velocity, joint to the peak ground acceleration normalized by the critical acceleration of the wall.
83. Spatiotemporal Dynamics of Meteorological and Agricultural Drought on the Northern Slope of the Tianshan Mountains: Trends, Propagation Processes, and Triggering Thresholds
Core Problem: Meteorological drought does not translate immediately or uniformly into agricultural drought, leaving seasonal lag and triggering thresholds uncertain across the Tianshan agro-pastoral zone.
Key Innovation: Event tracking, wavelet coherence and copulas identify 92 meteorological and 65 agricultural droughts, dominant 2-8 month propagation and a typical 1-3 month agricultural lag, with strong seasonal variation in transfer speed.
84. Stochastic Emulation of Seismic Structural Responses Under Recorded Ground Motions and the Significance of Intensity Measure Selection
Core Problem: Surrogate models offer a practical route to reduce the computational cost of nonlinear response history analysis by approximating expensive finite‑element simulations from a limited set of training runs.
Key Innovation: The authors construct a large input–response database by pairing 3,400 PEER NGA West‑2 recorded ground motions with 2,000 structural parameter realizations across steel and reinforced‑concrete archetypes (2, 5, and 10 stories). Emulator performance is evaluated both globally and locally by comparing predicted EDP distributions to ensembles of nearby IM‑space realizations, enabling assessment of local variability reproduction and predictive bias.
85. Drought trends in the Arabian Peninsula using SPEI and SPEDI indices and their implications for climate adaptation
Core Problem: Recent decades have witnessed intensifying drought across the Arabian Peninsula, yet it remains unclear whether precipitation deficits or increased potential evapotranspiration (PET) are the dominant driver of this intensification.
Key Innovation: The study employs the Standardized Precipitation Evapotranspiration Index (SPEI) and the recently introduced Standardized Precipitation Evaporation Differential Index (SPEDI), which integrates SPI and EDDI to enhance sensitivity to rapid-onset, heat-driven drought. Both indices were calculated at 3-, 6-, and 12-month timescales for the Arabian Peninsula from 1975 to 2024 using ERA5-Land reanalysis data validated against observed meteorological stations.
86. A foundation model of numerical intelligence with cross-disciplinary generalization
Core Problem: Scientific foundation models typically specialize by equation family or discipline and cannot infer a new numerical system from a small set of examples without retraining.
Key Innovation: UNICON encodes graph-based input-output examples as context and infers a transferable operator; one model approaches specialist performance in unseen scientific and social systems, with further gains when paired with language-model agents.
87. A Systems Engineering Framework for Vision-Language-Enabled UAV Triage and Disaster Response
Core Problem: Vision-language disaster tools usually stop at advice, leaving operators to translate outputs into coordinated UAV tasks under time pressure.
Key Innovation: A model-based systems architecture inserts a VLM coordinator between natural-language operators, mission control and task allocation; software-in-the-loop tests and a seven-person study report lower workload and high communication clarity.
88. Beacon: Knowing When and How to Perform Agentic Visual Reasoning
Core Problem: Visual agents often invoke tools when they are unnecessary and lose on easy cases as much as they gain on hard ones, so tool availability does not guarantee net benefit.
Key Innovation: Beacon trains with necessity-aware rewards and hint-guided capability expansion to improve both adaptive tool invocation and genuine tool effect, outperforming prior agentic visual-reasoning systems across diverse benchmarks.
89. LEDGERMIND: Provenance-Constrained Multimodal Agentic Reasoning with a Structured Evidence Ledger
Core Problem: Final-answer accuracy cannot distinguish grounded multimodal reasoning from unsupported intermediate claims, language priors or errors that cancel by chance.
Key Innovation: LedgerMind constrains every downstream claim to active entries in a structured evidence ledger and implements repair as provenance-preserving state transitions, improving both answer accuracy and trajectory-level faithfulness across models and benchmarks.
90. PhiZero: A World Model Built Around Physical Language
Core Problem: Pixel-space world models leave physical state transitions implicit, making their predicted dynamics hard to inspect or reuse across actions.
Key Innovation: PhiZero learns a compact discrete physical language from video, reasons over future transitions before rendering them and demonstrates coherent generation, action-conditioned simulation and zero-shot motion transfer.
91. SpatialCLI: Learning to Reason With Spatial Tools, Then Without Them
Core Problem: However, a fundamental capability mismatch remains: general VLMs can reason about the overall task but often miss the visual details that determine success, while specialist vision models can capture those details but cannot translate them into task-level decisions.
Key Innovation: The authors further introduce SpatialCLI-Bench, a 516-example benchmark for compositional perception across localization, segmentation, depth, and pose. SpatialCLI proceeds in three stages: (1) Call exposes specialist vision models as spatial tools to augment the VLM's perception; (2) Learn uses Cold-Start SFT and agentic RL to improve tool use; and (3) Internalize verbalizes successful tool-use trajectories to internalize specialist perceptual capabilities.
92. Theia: Large-Scale Multimodal Captioning and Automated Validation of the Incidents1M Dataset for Data-Free Distillation
Core Problem: Disaster image datasets either lack descriptive text or contain image-text mismatch, weakening cross-modal data-free knowledge distillation.
Key Innovation: Theia recovers 100,000 Incidents1M images, captions them with two Qwen3.5 models and validates 173,179 label pairs with an image-blind judge; agreement reaches 78.65/100 with 77.6% precision and 46.0% recall.
93. Weather Emulators at the Frontier of Heat Extremes Predictability
Core Problem: Yet in a warming world, improving early warnings of extreme heat is an increasingly critical challenge.
Key Innovation: Here we evaluate six state-of-the-art deep learning weather emulators - Pangu-Weather, FuXi, ArchesWeather, AIFS, GraphCast and Aurora - alongside leading dynamical systems and statistical baselines in forecasting global near-surface temperature and extreme heat at lead times of 10-15 days. While all models show some degree of predictive skill for extreme heat, most emulators under-represent peak intensities, and IFS recall is greater than that of any of the emulators.
94. What Makes Graph Unified? Principles and Generative Sliding-Window Transformer for Graph Foundation Models
Core Problem: Matching node-feature dimensions across graph domains does not ensure that semantics, information and transferable relational structure survive unification.
Key Innovation: SliGFM orders features by topological smoothness, encodes them with shared sliding windows and reconstructs the originals, satisfying explicit requirements for formal uniformity, transferability, information preservation and backbone compatibility.
95. Evaluation of soil liquefaction and earthquake-induced deformations based on ground motion parameters
Core Problem: Peak ground acceleration alone cannot explain why sites with similar shaking develop different liquefaction-triggering and permanent-deformation outcomes.
Key Innovation: The review separates triggering from post-liquefaction settlement and lateral spreading, associating the first mainly with PGA and the deformation stage more strongly with PGV, PGD, duration, frequency content and energy.
96. SVM Machine Learning Implementation on Geophone Earthquake Detection System
Core Problem: Low-cost geophones require a reproducible way to separate earthquake-like vibration from background noise before they can support local warning.
Key Innovation: A Raspberry Pi and geophone prototype compares six SVM kernels under five laboratory vibration amplitudes; quadratic and fine-Gaussian kernels reach 96% classification accuracy, with field validation identified as the next requirement.
97. Dynamic Assessment and Optimization of Bridge Resilience under Multi-Hazard Coupling
Core Problem: Bridge-resilience assessments built for isolated hazards cannot represent amplification when an earthquake and flood occur in sequence or interact across recovery phases.
Key Innovation: A system-dynamics model links absorption, adaptation and recovery with a hazard-coupling coefficient and finds coordinated interventions across all disaster phases outperform single-phase strategies in a coastal bridge case.
98. Hotspot magmatism and volcanic hazard at Cape Verde through the lens of shipborne surveying and dredging
Core Problem: Cape Verde's subaerial volcanism is well studied, but sparse observations of submarine vents limit interpretation of plume magmatism and the archipelago's distributed volcanic hazards.
Key Innovation: High-resolution multibeam bathymetry, sub-bottom profiling and targeted dredging sample young cones on debris fans and find no substantial edifices inside the island ring, constraining where melt generation is expressed.
99. Seismic Response of Existing Masonry‐Infilled RC Frame Structures Retrofitted with Steel Exoskeletons
Core Problem: Steel exoskeleton retrofits are often designed from bare reinforced-concrete frames even though masonry infills can change stiffness, resistance and inelastic interaction with the new bracing.
Key Innovation: Static and dynamic nonlinear analyses compare serviceability- and collapse-prevention retrofit designs and quantify how infills alter the performance indicators and effectiveness of exterior steel bracing.
100. Study on stability assessment of buckling failure and rainfall-induced instability mechanism for steeply inclined rock slope
Core Problem: Buckling-prone steep rock slopes require mechanical thresholds and rainfall-sensitive risk zones that can be reconciled with field displacement and strain observations.
Key Innovation: Theory, orthogonal sensitivity analysis and FLAC3D identify layer thickness as the strongest positive control and heavy rainfall as a failure threshold, increasing maximum displacement by 175% and localizing the core risk and shear-outlet zones.
101. Integrated ICESat-2 and Sentinel-2 bathymetry and its application to high-resolution wave modeling
Core Problem: The scarcity of high-resolution bathymetric data is a major limitation for accurate wave modeling in coral reef environments.
Key Innovation: The study focuses on Yongle Atoll in the South China Sea and integrates ICESat-2 laser altimetry with Sentinel-2 optical imagery to produce a 10 m-resolution satellite-derived bathymetry (SDB) dataset. Numerical simulations of typhoon-induced waves indicate that the fused bathymetry reduces the root mean square error of significant wave height within the lagoon by approximately 40%, demonstrating a substantial improvement in model performance.
102. Efficient reconstruction of nearshore sea state and bathymetry from marine radar images
Core Problem: Point instruments cannot resolve the spatial evolution of nearshore waves and bathymetry, while existing radar processing is computationally heavy and vulnerable to cross-sea interference.
Key Innovation: Lightweight Image Assimilation combines Radon transforms, signal processing and outlier histograms across X-band radar and video, recovering wave height at R-squared 0.79 and bathymetry at 0.70 with lower computational demand.
103. A Novel Soil Moisture Retrieval Model Using Dynamic Spatiotemporal Error Propagation for CYGNSS
Core Problem: CYGNSS soil-moisture retrieval from reflectivity alone retains time-varying, spatially structured bias, while adding many external predictors reduces independence.
Key Innovation: A dual-branch wavelet network learns gridwise error propagation and feeds it back into reflectivity retrieval, lowering SMAP-referenced RMSE by 20.97% and raising correlation from 0.876 to 0.923, with larger gains over farmland.
104. Geometric Accuracy Assessment of Large-Scale ZY-3 Imagery Based on Inter-Image Consistency
Core Problem: Large-area optical-image accuracy is difficult to measure without ground control because relative consistency and absolute position error are usually assessed separately.
Key Innovation: A unified error model validates Google Earth and SRTM against WorldView before assessing 1,368 ZY-3 scenes over 1.9 million square kilometres, obtaining 3.33 m planimetric and 4.27 m vertical RMSE.
105. Validation Design Governs Reported Accuracy in Small UAV Hyperspectral Datasets: An Interpretable Feature-Optimization Case Study of Maize Canopy Nitrogen Concentration
Core Problem: Small UAV hyperspectral studies can leak feature selection across folds and report within-site accuracy that collapses under geographic domain shift.
Key Innovation: Nested feature selection lowers R-squared from 0.712 to 0.612-0.627, and every leave-one-field-out test is negative; the result attributes most apparent gain to feature optimization rather than model complexity or LiDAR.
106. PCINet: A Prior-Guided Correlation Interaction Network for High-Resolution Remote Sensing Image Change Detection
Core Problem: However, illumination differences, seasonal variation, and complex background variation can generate pseudo-change responses, making it difficult to preserve detection accuracy under lightweight computational constraints.
Key Innovation: The study proposes PCINet, a prior-guided correlation interaction network designed to balance reliable change discrimination and computational efficiency. On the LEVIR-CD, SYSU-CD, and GZ-CD datasets, PCINet achieved F1 score values of 91.30%, 82.61%, and 88.62% and IoU values of 83.99%, 70.37%, and 79.57%, respectively.
107. Sentinel-1 SAR and Temporal Lag Soil Moisture Estimation at Instrumented Field Sites: A Stacked Ensemble Approach
Core Problem: Field-scale soil moisture (SM) estimation from Sentinel-1 C-band SAR alone is challenged by vegetation, roughness, and spatial heterogeneity.
Key Innovation: The study proposes a Stacked Additive Boosting-based Model (SABM) that combines Sentinel-1 SAR, Sentinel-2 optical, and ancillary geophysical features with temporal lag SM features (SMlag1, SMlag2) derived from a station’s own antecedent in situ record, exploiting SM persistence at 12-day Sentinel-1 repeat intervals; the framework is accordingly intended for instrumented sites with historical SM observations rather than as a satellite-only retrieval method for ungauged locations.
108. Relative Radiometric Normalization of Multisource Optical Satellite Imagery via Automatic Construction and Intelligent Refinement of Radiometric Reference Sample Set
Core Problem: However, existing RRN methods often suffer from low automation in radiometric reference sample selection and are highly sensitive to anomalous samples, resulting in limited normalization accuracy and robustness.
Key Innovation: To address these issues, this study proposes an automated cross-sensor RRN framework based on automatic construction and intelligent refinement of radiometric reference sample set (RRSS). Subsequently, a dual-stage sample refinement strategy integrating statistical outlier detection and topographic prior knowledge is developed to progressively remove anomalous samples and improve sample reliability.
109. DAChanger: A remote sensing change detector inspired by differential amplifier circuit for signal energy Decoupling and geometric consistency
Core Problem: Although deep learning has improved detection performance, pseudo-change interference caused by non-semantic factors and boundary geometric degradation remain major challenges in complex real-world scenes.
Key Innovation: To address these issues, this paper proposes DAChanger, a remote sensing CD framework inspired by the principle of differential amplification. Instead of directly discriminating changes from mixed bi-temporal features, DAChanger explicitly organizes differential-mode and common-mode states in the feature space, and progressively improves the separation between change-related responses and shared interference responses through progressive purification and edge geometry enhancement.
110. PI-MLP: A physics-informed multilayer perceptron surrogate for rapid tunnel support assessment under heterogeneous geological conditions
Core Problem: Tunnel support assessment during construction is increasingly challenged by heterogeneous and time-varying geological conditions.
Key Innovation: To address this issue, this study proposes PI-MLP, a physics-informed multilayer perceptron surrogate for rapid tunnel support assessment under heterogeneous geological conditions. To improve physical plausibility, the composite loss combines sparse FLAC3D supervision with reduced equilibrium-related differential regularization, constitutive-consistency regularization, and soft boundary constraints.
111. Deep learning for automated rock core image analysis: Weathering classification, fracture detection, and 3D geological modelling
Core Problem: Rock core images are a primary source of subsurface information, but conventional manual logging is slow, labour-intensive and prone to subjective error.
Key Innovation: The study presents CoreVision, an end-to-end data-driven workflow for automated interpretation and digitisation of drilling-core imagery in linear underground infrastructure projects. At the macroscopic scale, Core_YOLO modifies the YOLOv5 FPN + PAN architecture by adding a coarser feature map tailored to elongated core segments, improving the detection of lithological and structural variability in rock-core imagery.
112. Beyond Visual Ambiguity: Guiding Robust Monocular Depth Estimation in Challenging Scenarios via Detailed Long Captions
Core Problem: Monocular depth estimation (MDE) faces challenges with non-Lambertian surfaces and adverse weather conditions due to the visual ambiguities inherent in single-image limited information.
Key Innovation: To address these limitations, we propose CapDepth, a novel framework for robust MDE that leverages guidance from detailed long captions to alleviate visual ambiguities in both challenging scenarios. Extensive experiments validate the efficacy of CapDepth, which outperforms state-of-the-art methods, achieving depth error reductions of 25.0% on non-Lambertian surfaces and 22.0% under adverse weather conditions.
113. CoRE-UIR: Prior-guided common and residual experts for efficient all-in-one remote sensing image restoration
Core Problem: Remote sensing images acquired by unmanned aerial vehicles (UAVs) and satellites are often degraded by adverse weather, illumination variation, and imaging artifacts, which may co-occur and jointly induce global distribution shifts and local structural corruption.
Key Innovation: The authors propose CoRE-UIR (Common and Residual Experts for Universal Image Restoration), a prior-guided global-local framework centered on the Common-and-Residual Expert Block (CoRE). Extensive experiments on multiple datasets show that CoRE-UIR improves the overall average PSNR by 1.05 dB while running 11.83 times faster and reducing peak memory by 85.3% relative to the strongest baseline, BaryIR, thereby maintaining a favorable quality-efficiency trade-off.
114. Finding Change in Satellite Archives from Text: How to Combine Before-and-After Images Efficiently
Core Problem: Text search over before-and-after satellite archives requires a fusion module at query time, so small design choices can determine both retrieval cost and whether subtle changes survive compression.
Key Innovation: A controlled eight-model study finds that cheap difference retrieval followed by attention re-ranking cuts query cost 10-15 times, while temporal bottleneck fusion reduces parameters 2.3 times and latency 1.6 times with only a small captioning penalty.
115. FootprintNet: State-Transition-Guided Dynamic Footprint Learning for Multi-temporal Remote Sensing Change Detection
Core Problem: Despite substantial progress in remote sensing multi-temporal change detection (MTCD), most existing MTCD methods still represent the dynamic process at each spatial location over the entire observation period using a single change category associated with the final observation.
Key Innovation: To address this limitation, we introduce Urban Building Dynamics Detection (UBDD), which identifies building-change dynamic footprints, i.e., the temporal intervals in which changes occur, from multi-temporal imagery and produces pixel-wise classification masks. It further exploits temporal change-boundary cues to enhance feature contrast across boundary sides, thereby improving the discrimination among different dynamic footprints and enabling accurate detection of dynamic footprints.
116. Uncertainty quantification for trustworthy deep learning: Methods and measures
Core Problem: The deployment of deep neural networks in safety-critical domains demands reliable estimates of predictive confidence, yet conventional architectures lack principled uncertainty quantification.
Key Innovation: The authors then review ensemble diversity theory and uncertainty measures and their decompositions, contrasting the entropy decomposition with pairwise divergence measures, and consolidate evaluation methodology so that our qualitative comparisons share a common basis.
117. VAD: Attributing Visual Evidence for Target Reconstruction in Multimodal On-Policy Distillation
Core Problem: The key challenge is to estimate which corrections are supported by visual evidence, not merely where or how strongly to distill.
Key Innovation: The authors introduce Visual Attribution Distillation (VAD), a counterfactual target-reconstruction algorithm that estimates the visually attributable part of a teacher correction. Across six fine-grained visual benchmarks at 4B and 9B scales, VAD outperforms direct privileged-view distillation and visual-advantage weighting.
118. Witness Evidence Portfolios: Single-Prefill Risk Detection for Closed Multimodal Answers
Core Problem: A confident closed-form visual answer can still rest on sparse, contradictory or poorly localized evidence, which candidate-margin scores do not reveal.
Key Innovation: Witness Evidence Portfolios extracts layerwise supporting and contradicting visual contributions from a single white-box prefill and fuses provenance or concentration routes with confidence, improving mean error average precision by 0.134 across 12 model-dataset pairs.
119. Pulse RFI Mitigation for SAR Data Based on Reduced Rank Approximate
Core Problem: Pulse radio-frequency interference obscures SAR targets, while notch filtering removes increasing amounts of useful signal as interference occupancy grows.
Key Innovation: A semi-parametric Hankel low-rank decomposition uses truncated nuclear-norm regularization to separate contaminated echoes without over-penalizing dominant singular values, retaining more Sentinel-1A signal in simulated and measured tests.
120. WetVeg-2mm: An Ultra-High-Resolution UAV Dataset for Riparian Vegetation Semantic Segmentation
Core Problem: However, riparian plant communities often exhibit fragmented patches, broad transition zones and high visual similarity among classes, making stable species-level segmentation difficult from conventional satellite imagery or lower-resolution UAV imagery.
Key Innovation: To address this gap, we present WetVeg-2mm, an ultra-high-resolution UAV dataset for fine-grained riparian vegetation semantic segmentation. Across all evaluated baseline settings, SegFormer with ImageNet pretraining achieved the best overall performance, with 76.23% mIoU, 86.01% mDice, 85.17% PA, 87.30% Recall and 85.42% Precision on the test set.
121. A Hybrid Prior-Based Framework for Infrared Image Enhancement Towards Reliable Scene Interpretation
Core Problem: Infrared scenes combine low contrast, structural blur and frame-to-frame grayscale drift, yet real-time enhancement cannot rely on expensive retraining for each sensor band.
Key Innovation: A training-free decomposition with prior-preserving bi-gamma correction runs above 25 frames per second on CPU across SWIR, MWIR and LWIR and raises average contrast-to-noise and signal-to-clutter ratios by 174.7% and 298.5%.
122. GSSA: Gaussian Surfels with Spatial Awareness for Surface Reconstruction
Core Problem: Surface extraction from rendered Gaussian-splat depth loses detail under occlusion and limited viewpoints because image-space depth does not preserve each surfel's full geometry.
Key Innovation: GSSA redistributes trained Gaussian surfels and constructs a signed-distance field directly from their primitives, improving the accuracy-time balance on terrestrial and airborne photogrammetry and transferring to other surfel pipelines.
123. Self-Supervised Hyperspectral Image Clustering via Spatial–Frequency Interaction and Amplitude–Phase Decoupling
Core Problem: However, the quadratic computational complexity of self-attention restricts practical applications in large HSI scenes.
Key Innovation: To address the above limitations, we propose a self-supervised Spatial–Frequency Interaction and Amplitude–Phase Decoupling framework, termed SFI-APD, which integrates a High-Order Spatial–Frequency Interaction Module (HSFIM), a Frequency Feature Attention Block (FFAB), and a Frequency-Domain Vision Transformer (FreqViT) into a unified architecture.
124. Aerial–ground LiDAR place recognition with patch-level self-supervised learning and expanded reciprocal re-ranking
Core Problem: The most studied ground-level LiDAR place recognition suffers from pre-visit requirements, incomplete coverage, and limited perspectives.
Key Innovation: To overcome these limitations, we present a novel retrieval and re-ranking framework for aerial-ground LiDAR place recognition. Our retrieval network integrates these patch-level self-supervised learning modules with scene-level learning to improve the discriminativeness of global features across aerial and ground point clouds.
125. AdaAnchor4D: Anchor-Conditioned Spatiotemporal Feature Aggregation for Monocular UAV 4D Reconstruction
Core Problem: However, such scenes exhibit pronounced spatiotemporal heterogeneity: different regions follow distinct temporal activity patterns, while the motion states of some dynamic regions may further evolve over time.
Key Innovation: To address this challenge, we propose AdaAnchor4D, an adaptive anchor deformation framework for monocular UAV dynamic scene reconstruction. Experiments on UAV-Arc4D, VisDrone, and UAVDT show that AdaAnchor4D achieves higher rendering quality than representative dynamic Gaussian methods while maintaining real-time rendering performance.
126. Convolutional Neural Shading for High-Quality 3D Reconstruction from Multi-View Images
Core Problem: Neural multi-view reconstruction based on independent surface points loses local geometric detail, particularly in dark, textureless regions and near image boundaries.
Key Innovation: Convolutional Neural Shading aggregates neighbouring rendering coordinates and adds a fine-detail displacement network, improving surface geometry and boundary regularity over pointwise neural-shading baselines.