TerraMosaic Daily Digest: September 17, 2026
Daily Summary
The leading landslide study treats weathering as a transient control on both failure and runout. Across the Utuado batholith in Puerto Rico, field mapping, joint seismic-resistivity imaging, geotechnical back-analysis and inundation modelling link critical-zone architecture to slope strength and debris-flow growth. Complementary studies reconstruct three-dimensional motion and rainfall-lagged deformation of a basalt weathering-crust landslide, resolve how bidispersity and fragmentation attenuate rock-avalanche barrier loads, and use multi-station seismic alignment to detect 12 held-out debris flows 31-135 minutes before they reached a downstream reference point.
Earthquake studies recover deformation and recurrence patterns that conventional observations blur. Multi-track burst-overlap interferometry restores the north-south component of interseismic, coseismic and postseismic motion, with centimetre- and millimetre-per-year-scale agreement against GNSS. Nineteen years of Tohoku-Hokkaido waveforms reveal hierarchical repeating ruptures and systematic downward directivity, while Bayesian treatment of incomplete Türkiye-Syria building records revises fragility estimates. Sparse-observation latent dynamics, GNSS inversion at Ol Doinyo Lengai and multi-sensor reconstruction of the previously unmapped Bogo Bay Fault extend the same emphasis on constrained inference.
Hydrometeorological work connects process resolution to forecast utility. A dynamic-roughness model shows that flood-stage dune washout in the lower Po can make fixed-resistance models overestimate peak water level by more than 2 m. IAU-based 4DVar improves rapid-intensification, track-turn and heavy-rain forecasts for two typhoons; a new Southeast Asian event database links extreme precipitation to floods, flash floods, wet landslides and typhoons; and minute-scale FY-4B winds resolve inner-core, rainband and outflow structure across 15 tropical cyclones.
Disaster AI is being redesigned around sparse labels, unknown classes and constrained hardware. A 0.7-million-parameter flood model distilled from Prithvi-EO-2.0 fits a 1.5 MB INT8 engine and processes 512 × 512 scenes in 5.57 ms while retaining an explicit spectral-threshold comparison. ESIA couples category-agnostic temporal change detection with open-vocabulary recognition over six anomaly classes and applies the system to Kakhovka Dam-collapse impacts and wildfire severity. A new low-bit format and shifter-based accelerator retain vision-language accuracy while reporting 2.3-4.5-fold speedups and 1.4-2.9-fold energy reductions; separate physical-world-model experiments show that symplectic energy evolution and factorized coupling address distinct failures in long-rollout stability and changed-law transfer. Radio-frequency convolution and photonic reservoir computing point toward lower-energy, lower-latency inference, although their value for geohazard operations remains prospective.
Key Trends
The day's strongest studies replace static hazard maps with state-aware observations, uncertainty-aware inference and deployable monitoring.
- State-dependent controls are entering landslide and mass-flow models: Weathering architecture, transient hydrology, three-dimensional deformation, bidispersity and fragmentation are being treated as controls on failure, mobility and structural loading rather than as fixed background conditions.
- Earthquake monitoring is recovering poorly observed dimensions: Burst-overlap interferometry restores north-south motion, long waveform archives expose nested repeating ruptures, and physics-pretrained latent dynamics reconstruct spatial response from sparse stations.
- Process models and event databases are being tied to decisions: Dynamic river roughness, typhoon data assimilation, coupled flood forecasting and a regional precipitation-disaster database connect physical state estimation to warnings, forecast lead time and impact characterization.
- Hazard AI is becoming open-world and edge-ready: Open-vocabulary anomaly recognition, foundation-model distillation and low-bit vision-language acceleration target unknown events and constrained processors, while radio-frequency and photonic hardware explore faster, lower-power inference whose geohazard transfer remains prospective.
- Uncertainty and observation bias are becoming explicit: Bayesian correction of incomplete building inventories, uncertainty-aware prospectivity mapping and probabilistic subsurface imaging distinguish observational support from apparently precise geospatial estimates.
Selected Papers
The 17 September selection is led by a critical-zone explanation of landslide hazard in Puerto Rico, multi-track InSAR that restores north-south earthquake-cycle motion, hierarchical repeating ruptures beneath the Tohoku-Hokkaido margin and an open-world satellite system for unknown surface anomalies. Companion studies address basalt-landslide deformation, rock-avalanche barrier loading, seismic debris-flow warning, volcanic uplift, typhoon data assimilation, flood hydraulics, extreme-weather event databases, physics-structured simulation and edge-deployable multimodal inference.
1. Impacts of Transient Critical Zone Evolution on Landslide Susceptibility and Hazard Across the Utuado Batholith, Puerto Rico
Core Problem: Differential weathering changes subsurface composition, porosity, groundwater connectivity, and frictional strength, but their combined control on slope failure and debris-flow hazard is poorly resolved.
Key Innovation: Integrates landslide mapping, seismic-refraction and resistivity surveys, geotechnical observations, topographic and stream-profile analysis, three-dimensional slope-strength back-analysis, and debris-flow inundation assessment to establish a process-based hazard chain.
2. Multi-Track Time-Series Burst-Overlap Interferometry for Resolving Horizontal Deformation in Earthquake-Cycle Studies
Core Problem: Near-polar SAR viewing geometry makes conventional InSAR intrinsically insensitive to north–south crustal motion.
Key Innovation: MTSB combines optimized phase linking, a unified time-series model separating deformation from residual misregistration, and block-wise spectral analysis to recover absolute ITRF-referenced along-track deformation while reducing orbital artifacts.
3. Hierarchical repeating earthquakes with strong downward directivity
Core Problem: How repeatable are nominally repeating earthquakes, and how does hierarchical fault geometry control their rupture behavior?
Key Innovation: Uses 19 years of waveforms to define hierarchical repeaters, reveal persistent fault structure and systematic downdip directivity, and show unexpectedly limited rupture repeatability.
4. Earth Surface Immune System for Rapid Monitoring of Unknown Anomalies
Core Problem: Rare and unpredictable anomalies are poorly served by category-specific models, and localization alone does not provide enough actionable information.
Key Innovation: ESIA combines category-agnostic temporal change detection, negative-selection filtering of text prompts with multimodal recognition, and lightweight test-time embedding adaptation from one reference image pair.
5. Characterizing deformation patterns and potential hazard zones of a basalt weathering crust landslide using multi-source remote sensing observations
Core Problem: Resolve the three-dimensional motion, thickness, hydrologic forcing, and downstream threat of a slow-moving basalt weathering-crust landslide.
Key Innovation: Integrates InSAR, UAV photogrammetry, surface-parallel 3D displacement, mass-conservation thickness inversion, wavelet analysis, geophysics, and RAMMS scenarios.
6. Investigation of dynamic fragmentation in bidisperse flow impact against a rigid barrier
Core Problem: Determine how bidispersity and particle breakage during flow-to-impact evolution control barrier forces.
Key Innovation: Uses centrifuge-validated DEM to show how fine-particle cushioning and interfacial crushing alter boulder impact and reduce peak force by nearly 50%.
7. Debris-flow identification and early warning based on seismic observation records
Core Problem: Seismic debris-flow monitoring suffers from severe class imbalance, environmental noise, and high false-alarm rates.
Key Innovation: Combines dynamic noise resampling, Lag-UNet multi-station temporal alignment, and joint representation learning in the DTCD model.
8. Dune Evanescence at Flood Stage in Lowland Sand-Bed Rivers: A 2D Modeling Approach With Dynamic Resistance
Core Problem: Fixed-resistance models omit flood-stage dune washout and can therefore misrepresent stage–discharge behavior.
Key Innovation: The study dynamically updates bed resistance in a 2D depth-averaged model and tests it along 200 km of the lower Po River, where conventional fixed resistance overestimated peak water levels by more than 2 m.
9. Spatiotemporal Analysis for Frequency-Magnitude Distribution of Earthquakes Across Mainland China: Comparing Classical and b-positive b-values Across Tectonic Regimes
Core Problem: Classical b-values are sensitive to changing detection capability and short-term aftershock incompleteness, complicating precursor interpretation across tectonic regions.
Key Innovation: More than 50 years of seismicity are used to compare classical and b-positive estimators across major seismogenic zones; several later rupture areas overlap low-b patches identified in earlier long-term windows.
10. Cross-Architecture Foundation-Model Distillation for Edge Flood Segmentation
Core Problem: Large geospatial foundation models are difficult to deploy near disaster operations, while manually labeled flood scenes are scarce.
Key Innovation: A 300-million-parameter Prithvi-EO-2.0 teacher labels additional Sentinel-2 imagery and is distilled into a 0.7-million-parameter EfficientViT-B0 student; the quantized 1.5 MB model runs at 5.57 ms per 512×512 image on a Jetson Xavier NX.
11. Seismic Site Response Prediction from Sparse Observations Using Finite-Element-Pretrained Latent Dynamics
Core Problem: Finite-element site-response simulations can deviate from observations, yet field records are sparse across sensors and events.
Key Innovation: FLARE-T learns forced low-dimensional dynamics from dense finite-element simulations, maps sparse sensor responses into that latent space, and calibrates the dynamics with limited measured records.
12. Accounting for Biases in the Analysis of Building Damage Data for the 2023 M7.8 Türkiye/Syria Earthquake Sequence
Core Problem: Incomplete and preferentially sampled building-damage observations can bias estimates of building inventory and fragility.
Key Innovation: Introduces a Bayesian framework that jointly infers building-stock composition and fragility while explicitly representing missing attributes, observation bias, and uncertainty.
13. A high spatiotemporal resolution atmospheric motion vector dataset from Fengyun-4B Geostationary High-speed Imager observations for 15 tropical cyclones over the western North Pacific and South China Sea (2022-2025)
Core Problem: Minute-scale observations of tropical-cyclone inner cores, rainbands, and outflow are sparse.
Key Innovation: Releases validated FY-4B atmospheric motion vectors for 15 cyclones at 3 km visible and 24 km infrared resolution with 1-, 2-, 6-, or 7-minute sampling.
14. Inflation at Ol Doinyo Lengai driven by shallow magma intrusion during the 2022-2023 uplift episode in an early-phase continental rift
Core Problem: Determine the source of the 2022–2023 non-eruptive uplift at Ol Doinyo Lengai.
Key Innovation: Inverts millimeter-precision continuous GNSS data across four source geometries and identifies a shallow sill-like intrusion near 3.1 km depth with quantified volume change.
15. Source fault, impacts, and seismotectonic implications of the 2025 magnitude (MW) 6.9 Northern Cebu earthquake, Philippines
Core Problem: Identify the causative fault and explain the coseismic deformation and spatial pattern of impacts from the 2025 Northern Cebu earthquake.
Key Innovation: Integrates seismicity, DInSAR, pixel offsets, geomorphic interpretation, field observations, and stress modeling to identify the Bogo Bay Fault.
16. Intra-Seasonal Variability of Serial Clustering of Severe European Winter Windstorms
Core Problem: The within-season timing of severe windstorm clustering and its relationship to large-scale circulation modes remain insufficiently characterized.
Key Innovation: Uses ERA5 and multiple fixed and developing windows to quantify serial clustering, identify two enhanced periods late in the winter half-year, and investigate associations with major circulation patterns.
17. Capturing Typhoon Rapid Intensification and Abrupt Recurvature Using an IAU-Based 4DVar Scheme
Core Problem: Dynamical imbalance in model initialization limits predictability of rapid tropical-cyclone intensification and sudden track changes.
Key Innovation: Evaluates an incremental-analysis-update-based 4DVar scheme on Typhoons Doksuri and Khanun, linking balanced initialization to improvements in track and heavy-rainfall prediction.
18. Linking Extreme Precipitation to Pluvial Hydrometeorological Disasters in Southeast Asia Using a Long-Term Event Database
Core Problem: The absence of a unified subnational event database limits regional risk analysis and early-warning research for Southeast Asian hydrometeorological disasters.
Key Innovation: Integrates multiple disaster-reporting and precipitation sources to characterize hazard types, spatiotemporal patterns, impacts, and empirical rainfall references while explicitly examining reporting-completeness effects.
19. Discrete Boltzmann Modeling of Sediment Transport in Dam-Break Flows Over Erodible Beds
Core Problem: Strongly nonlinear coupling among flow, suspended and bed load, and mobile-bed morphology is difficult to simulate consistently and stably during dam-break events.
Key Innovation: Combines a discrete Boltzmann hydrodynamic model with lattice-Boltzmann sediment equations, an advection-diffusion treatment of suspended sediment, and an Exner equation for bed load, with one- and two-dimensional mobile-bed tests.
20. The Brittle-Plastic Transition in Quartz-Albite Mixtures: New Insights From Shear Deformation Experiments at Mid-to-Lower Crustal Depth Conditions
Core Problem: The microscale mechanisms controlling bulk transitional strength, weakening, and deformation stability in the upper crust remain uncertain.
Key Innovation: Quartz–albite shear experiments spanning simulated depths of 7–30 km link depth-dependent mechanical behavior to cataclastic comminution, grain-boundary sliding, nanograin formation, and dynamic recrystallization.
21. PhyRestore: Physics-Structured Latent-Factor Restoration
Core Problem: Rare, high-magnitude soil-loss changes are difficult to recover when rainfall erosivity and cover-management factors are noisy or corrupted.
Key Innovation: Rather than directly predicting soil loss, PhyRestore reconstructs corrupted RUSLE factors and then computes change through the known physical relationship.
22. DEM study on the thaw settlement mechanism of permafrost embankments based on phase-transition thermo-mechanical coupling
Core Problem: Explain how ice-water phase change, embankment geometry, ice content, and asymmetric heating govern permafrost embankment settlement.
Key Innovation: Develops a phase-transition thermo-mechanical DEM that resolves ice melting and quantifies seasonal and sunny-shady slope controls on settlement and thaw-bowl evolution.
23. Stratospheric wave predictability enhances surface forecasts over North America
Core Problem: The contribution of stratospheric variability to week-two prediction of North American cold extremes is not well constrained.
Key Innovation: Identifies a lower-stratospheric wave-1 mode that improves second-week temperature forecasts by 15% and conditioned cold-extreme forecast skill by 28%.
24. The effect of hybrid models integration sequence: taking Fushun area as an example
Core Problem: The effect of coupling order in hybrid WoE, multilayer-perceptron, and random-forest susceptibility models is poorly understood.
Key Innovation: Systematically compares integration sequences and identifies WoE-MLP-RF as the best tested configuration in Fushun, with an AUC of 0.841.
25. An Ensemble-Based Transfer Learning Framework Using EfficientNetV2 and MobileNetV2 for Satellite Image Classification of Wildfires
Core Problem: Improve the robustness of binary wildfire classification across transfer-learning architectures.
Key Innovation: Ensembles EfficientNetV2 and MobileNetV2 predictions, reporting 0.967 accuracy and 0.995 ROC AUC on the evaluated dataset.
26. Assessment of Sea Surface Wind Measurements from Wave Gliders in Tropical Cyclones
Core Problem: Reliable in situ sea-surface wind observations are scarce under tropical-cyclone conditions, and Wave Glider measurement reliability is insufficiently assessed.
Key Innovation: Combines internal quality checks, inter-platform comparison, and validation against multiple satellite products and a numerical weather model.
27. Dampened Diurnal Cycles of Hot-Wet Extremes in Urban Areas Contrast With Thermally Driven Daytime Stress in Rural Landscapes of Beijing-Tianjin-Hebei Region
Core Problem: Fine-scale diurnal differences in urban and rural hot-wet extremes, including the relative roles of temperature and humidity, remain unclear.
Key Innovation: Combines 2016–2024 observations from 1,962 automatic weather stations with wet-bulb-temperature decomposition to separate thermal and humidity contributions across the Beijing-Tianjin-Hebei region.
28. Co-Occurring Weather Systems Vary by Atmospheric River Scale Across the United States
Core Problem: How weather-system co-occurrence changes across atmospheric-river intensity classes, seasons, and U.S. regions is not well quantified.
Key Innovation: Combines ERA5-derived feature masks with satellite information to quantify spatial and temporal overlaps and identify increasingly complex system combinations for stronger atmospheric rivers.
29. Anticipating household rescue demand in hurricanes using socio-demographic data and machine learning
Core Problem: Title-level focus: anticipating household-level rescue demand so limited response capacity can be allocated more effectively.
Key Innovation: Title-signalled approach or contribution: The title indicates the use of socio-demographic data and machine learning; no abstract is available, so model design, sample size, spatial transfer, and performance are unknown. Methods, data and results could not be assessed because no reliable abstract was available.
30. Coupled hydromechanical modeling of mine water rebound-induced seismicity at the Gardanne mine, France
Core Problem: Title-level focus: understanding how post-mining water-level rebound changes pore pressure and stress to trigger seismicity.
Key Innovation: Title-signalled approach or contribution: The title indicates a coupled hydromechanical model applied at the Gardanne mine; no abstract is available, so constraints, event matching, and predictive capability are unknown. Methods, data and results could not be assessed because no reliable abstract was available.
31. A coupled flood forecasting framework of HYDRUS-1D and Xinanjiang models based on the transfer of physically-based soil moisture fields
Core Problem: Title-level focus: connecting physically simulated soil-moisture states to a conceptual rainfall-runoff model to improve flood forecasting.
Key Innovation: Title-signalled approach or contribution: The title couples HYDRUS-1D and the Xinanjiang model through transfer of physically based soil-moisture fields; no abstract is available, so basin, lead time, and forecast gain are unknown. Methods, data and results could not be assessed because no reliable abstract was available.
32. Efficient and high-precision urban flood forecasting model combining numerical weather prediction with real-time correction
Core Problem: Title-level focus: combining forecast meteorology with real-time information to improve urban-flood prediction efficiency and precision.
Key Innovation: Title-signalled approach or contribution: The title combines numerical weather prediction with real-time correction; no abstract is available, so the correction method, lead time, case coverage, and achieved accuracy are unknown. Methods, data and results could not be assessed because no reliable abstract was available.
33. Near-fault seafloor ground motions: A pure time-domain analytical solution to the 2D marine Lamb problem
Core Problem: Title-level focus: obtaining a time-domain analytical solution for near-fault ground motion in the two-dimensional marine Lamb problem.
Key Innovation: Title-signalled approach or contribution: The title claims a pure time-domain analytical solution; no abstract is available, so assumptions, boundary conditions, applicable regimes, and validation are unknown. Methods, data and results could not be assessed because no reliable abstract was available.
34. Radio-Frequency Convolutional Neural Networks
Core Problem: Modern neural networks are difficult to run on edge devices because added accelerators increase size, weight, power consumption, and cost.
Key Innovation: RF-CNN maps multi-channel convolutions onto frequency tones so a radio frequency mixer performs them in one pass; experiments cover networks up to 26.4 million parameters and report energy as low as 0.72 femtojoules per multiply-accumulate.
35. High-speed reservoir computing using photonic integrated circuit optical parametric oscillators
Core Problem: Integrated photonic neural systems have struggled to provide linear and nonlinear computation together at very high speed without repeated optical-electronic conversion.
Key Innovation: Implements an optical-parametric-oscillator reservoir computer on thin-film lithium niobate, exceeding 93% accuracy on several benchmarks at about 10 GHz and supporting a path to subnanosecond latency.
36. MiX: Micro-Inverted-Scaling for End-to-End Low-Bit Vision-Language Model Acceleration
Core Problem: Shared-exponent low-bit formats can underflow ordinary multimodal tokens when a large outlier dominates the scale, degrading attention and preventing efficient end-to-end VLM deployment.
Key Innovation: MiX inverts microscaling by grouping per-element exponents under a shared mantissa and maps the format to a shifter-based accelerator through an adaptive MiX-MX framework.
37. Continental-scale probabilistic resistivity imaging of Australia using deep learning: Implications for geology, groundwater, and critical minerals
Core Problem: Probabilistic inversion of tens of millions of airborne electromagnetic soundings is computationally expensive and difficult to harmonize across surveys.
Key Innovation: Eleven invertible neural networks consistently invert more than 26.6 million soundings over 352,600 line-km in 4.37 GPU hours, producing an uncertainty-aware Australian resistivity model to roughly 650 m depth.
38. Water saturation dependent acoustic emission event structures and damage accumulation paths in limestone with a central circular through hole
Core Problem: Determine how water saturation changes strength degradation, acoustic-emission event structure, and pre-peak damage accumulation in holed limestone.
Key Innovation: Combines UMAP-HDBSCAN event grouping, physics-based event merging, energy analysis, and an event-weighted acoustic-emission damage index.
39. Acquisition-Level Optimization of Split-Spectrum TEC Observability in L-Band InSAR
Core Problem: Split-spectrum ionospheric correction accuracy is constrained by acquisition bandwidth, subband signal-to-noise ratio, and frequency separation.
Key Innovation: Jointly optimizes acquisition and processing parameters through a TEC observability index and validates correction performance with a measured UAVSAR pair.
40. Direct quantification of solar-induced chlorophyll fluorescence using compact solar-blind optical radiometers
Core Problem: Conventional solar-induced fluorescence retrieval depends on atmospheric and illumination conditions, bulky hyperspectral instruments and complex spectral inversion.
Key Innovation: Measures within a saturated atmospheric oxygen line to optically isolate fluorescence from reflected sunlight, enabling compact radiometry without spectral retrieval or a reference measurement.
41. Assimilation of SWOT Water Surface Elevation Enhances Simulations of River Dynamics in the Ohio River Basin
Core Problem: The benefit and spatial propagation of assimilating irregular SWOT observations into regulated, large-basin river simulations are uncertain.
Key Innovation: Assimilates node-level SWOT water-surface elevation with an ensemble Kalman filter and river-network-aware localization in the Ohio River Basin, including reaches without direct satellite overpasses.
42. Observation-Masked Localized Assimilation of Satellite-Derived Snow Cover Data for Enhancing Runoff Prediction in the Rhône River Basin, France
Core Problem: Sparse, irregular satellite snow observations, complex terrain, and high-dimensional assimilation costs constrain operational snow-data assimilation.
Key Innovation: Introduces an observation-masked localized ensemble Kalman filter that updates observed areas while preserving neighborhood covariance, applied to daily Copernicus snow-cover data in the Rhône basin.
43. Lowering Barriers to Disaster Housing Recovery: A Collaborative and Action-Based Planning Framework
Core Problem: Title-level focus: reducing planning, coordination, and implementation barriers that hinder housing recovery after disasters.
Key Innovation: Title-signalled approach or contribution: The title presents a collaborative, action-based planning framework; no abstract is available, so the hazard context, case coverage, and demonstrated effectiveness are unknown. Methods, data and results could not be assessed because no reliable abstract was available.
44. A coupled evacuation model considering physiological tolerance and dynamic speed adaptation for fire smoke
Core Problem: Title-level focus: representing how physiological tolerance and changing movement speed affect evacuation during smoke exposure.
Key Innovation: Title-signalled approach or contribution: The title indicates a coupled evacuation model incorporating physiological tolerance and dynamic speed adaptation; no abstract is available, so its formulation, validation setting, and performance are unknown. Methods, data and results could not be assessed because no reliable abstract was available.
45. CEDA-Net: A cascade empirical-deep network for mitigating InSAR tropospheric delay in complex atmospheric and topographic conditions
Core Problem: Title-level focus: mitigating tropospheric-delay contamination of small deformation signals under complex atmospheric and topographic conditions.
Key Innovation: Title-signalled approach or contribution: The title proposes CEDA-Net, a cascade of empirical modeling and a deep network; no abstract is available, so inputs, geographic generalization, baselines, and correction accuracy are unknown. Methods, data and results could not be assessed because no reliable abstract was available.
46. HTFU-Net: A Cross-Modal Deep Learning Framework for Antarctic Surface Melt Mapping Using Multi-Source Microwave Remote Sensing Data
Core Problem: Title-level focus: fusing multiple microwave remote-sensing sources to identify and map Antarctic surface melt.
Key Innovation: Title-signalled approach or contribution: The title proposes HTFU-Net as a cross-modal deep-learning framework; no abstract is available, so sensor combinations, fusion mechanics, mapped extent, and accuracy are unknown. Methods, data and results could not be assessed because no reliable abstract was available.
47. Conservation Buys Stability and Factoring Buys Counterfactuals in Physical World Models
Core Problem: Learned physical simulators can drift during long rollouts and can remain locked to the training law when a governing coupling changes.
Key Innovation: Separates the two failure modes structurally: symplectic evolution of learned energy preserves bounded dynamics, while explicit factorization of physical coupling enables counterfactual transfer to an unseen coupling sign.
48. Quantifying Differences Between Floodplain Waterbody Connectivity and River-Floodplain Connectivity in a Coastal Meandering River
Core Problem: The relationship between floodplain-waterbody connectivity, river–floodplain exchange, and changing discharge is poorly quantified.
Key Innovation: The analysis distinguishes active from effective connections and identifies a transition from topographic control at lower discharge to hydraulic control at high discharge, with the most uneven exchange occurring during full bank inundation.
49. PerSeM: Persistent Semantic Memory for Long-Horizon Open-Vocabulary UAV Mapping
Core Problem: Frame-wise open-vocabulary segmentation is temporally inconsistent across repeated observations and changing UAV viewpoints.
Key Innovation: A training-free framework associates semantic observations with persistent world-space voxels and refines memory through history-preserving spatial refinement, trust-aware replay, and contextual verification.
50. 3DSO: Quantifying forest three-dimensional structural organization beyond the canopy surface using LiDAR data
Core Problem: Conventional LiDAR metrics summarize vegetation amount, vertical profiles or canopy surfaces without representing local three-dimensional configurations and their spatial deployment through a stand.
Key Innovation: 3DSO combines symmetry-canonicalized occupied-empty voxel patterns with their spatial deployment in an area-normalized information-theoretic metric.
51. Advancing hydrological and cryospheric simulations in Indus River Basin: First comparative evaluation of GESSI-3/ESSI-3 and a CNN-LSTM deep learning framework
Core Problem: Title-level focus: compare process-based hydrological and cryospheric simulation systems with a CNN-LSTM framework in a data- and hazard-sensitive basin.
Key Innovation: Title-signalled approach or contribution: Presents the first stated comparative evaluation of GESSI-3/ESSI-3 against a CNN-LSTM approach for the Indus River Basin. Methods, data and results could not be assessed because no reliable abstract was available.
52. Enhancing river flow prediction accuracy in data scarce catchments using a time varying ensemble deep learning approach
Core Problem: Title-level focus: improve river-flow predictions where observations are sparse and model performance varies over time.
Key Innovation: Title-signalled approach or contribution: Uses a time-varying ensemble of deep-learning predictors rather than a fixed ensemble. Methods, data and results could not be assessed because no reliable abstract was available.
53. Integrating SWOT data and machine learning for high-resolution river surface velocity retrieval in data-scarce headwater regions
Core Problem: Title-level focus: retrieve high-resolution river surface velocity in data-scarce headwater regions.
Key Innovation: Title-signalled approach or contribution: Integrates SWOT observations with machine learning for surface-velocity retrieval. Methods, data and results could not be assessed because no reliable abstract was available.
54. A method for analysis of soil segregation frost heave
Core Problem: Title-level focus: provide a method for analyzing soil segregation frost heave.
Key Innovation: Title-signalled approach or contribution: Proposes a dedicated analytical treatment of segregation-driven frost heave. Methods, data and results could not be assessed because no reliable abstract was available.
55. Drone-based camera discharge derived from the entropy-based probability concept and image velocimetry algorithms
Core Problem: Title-level focus: deriving river discharge reliably from drone-camera imagery.
Key Innovation: Title-signalled approach or contribution: The title combines an entropy-based probability concept with image-velocimetry algorithms; no abstract is available, so test reaches, flow regimes, accuracy, and operational maturity are unknown. Methods, data and results could not be assessed because no reliable abstract was available.
56. Physics-informed hydrological graph neural network for distributed daily streamflow prediction in directed river networks
Core Problem: Title-level focus: predicting spatially distributed daily streamflow while respecting directionality and hydrological structure in river networks.
Key Innovation: Title-signalled approach or contribution: The title proposes a physics-informed hydrological graph neural network for directed river networks; no abstract is available, so the physical constraints, basins, baselines, and accuracy are unknown. Methods, data and results could not be assessed because no reliable abstract was available.
57. Process-oriented evaluation of machine learning and physics-based models for wave parameter prediction under extreme conditions in a fetch-limited sea
Core Problem: Title-level focus: comparing machine-learning and physics-based wave models in a process-aware manner under extreme fetch-limited conditions.
Key Innovation: Title-signalled approach or contribution: The title specifies a process-oriented comparison of machine-learning and physics-based models; no abstract is available, so datasets, models, metrics, and outcomes are unknown. Methods, data and results could not be assessed because no reliable abstract was available.
58. Vertical Attenuation of Thermoremanent Magnetic Anomalies: Implications for Drone-Borne Archaeological and Shallow Geophysical Surveys
Core Problem: Small, shallow magnetic anomalies weaken and broaden rapidly as UAV sensor altitude increases.
Key Innovation: Synthetic modeling and upward continuation of field fluxgate data produce quantitative attenuation curves, including retention of only about 11% of near-surface peak amplitude at 2 m and about 3% at 4 m for the synthetic target.
59. Instance Segmentation and Fine-grained Classification for Urban Buildings with Adaptive Region Dividing and Spatially-Supervised Contrastive Learning
Core Problem: Fixed point-cloud blocks fragment buildings and require artificial partitions, while fine-grained building classification remains underdeveloped.
Key Innovation: The method derives structure-aligned training regions from BEV segmentation and combines building geometry, color, and local context with a spatially supervised contrastive loss.
60. Towards Active Cross-View Object Geo-Localization
Core Problem: Conventional cross-view object geolocation assumes a fixed query image and cannot acquire more informative observations or decide when to stop.
Key Innovation: ActiveMoPT combines multi-view prompt-preserving adaptation, trajectory-guided policy initialization, and cost-aware reinforcement learning; it is evaluated on MoP-UAV and the 858-scene ActiveGeo-858 benchmark.
61. Socialized UAV Cross-Task Learning: Towards Cross-Granularity Collaboration through Hierarchical Interaction
Core Problem: Mismatched detection and segmentation granularities make simple cross-task sharing prone to interference, teacher bias, and one-way collapse.
Key Innovation: The CrossUAV benchmark and Cross-Granularity Socialized Collaboration framework progressively activate and adaptively regulate hierarchical information exchange between tasks.
62. CitySTAR: Structured and Topology-Aware Reasoning for Open-Vocabulary Urban 3D Grounding
Core Problem: Similarity-based grounding does not adequately connect natural-language intent to semantic, geometric, and topological structure in billion-scale urban point clouds.
Key Innovation: CitySTAR builds an open-vocabulary 3D scene graph, represents target–context topology with paired hypergraphs, applies bidirectional topology verification, and incorporates candidate-centered 2D visual evidence.
63. Task-Oriented Semantic Feature Transmission for Multi-Task Satellite Remote Sensing over Low-SNR Channels
Core Problem: Reconstruct-then-infer transmission optimizes pixel fidelity rather than downstream task accuracy, especially under low signal-to-noise ratios.
Key Innovation: A lightweight channel adapter compresses features from a frozen multitask backbone, while a feature restorer and task heads are jointly trained under random-SNR corruption.
64. A long-term multiscale Critical Zone dataset integrating environmental monitoring and an open-air laboratory: The Alento River Catchment Observatory
Core Problem: Critical-zone and hydrological models lack long-duration observations that connect water storage and movement across soil, vegetation, groundwater, hillslope and catchment scales.
Key Innovation: Integrates continuous environmental monitoring with targeted hydrogeophysical, isotope, soil, spectral and high-resolution UAV campaigns, with documented processing, calibration, quality control and missing-data records.
65. Uncertainty quantification of deep learning algorithms for mineral prospectivity mapping
Core Problem: Deep-learning mineral prospectivity maps rarely separate data, model, and prediction uncertainty.
Key Innovation: Combines and visualizes multiple uncertainty sources to convert deterministic prospectivity outputs into probabilistic decision support.
66. A new Earth Observation-based WRF configuration for urban regional climate simulations over Paris
Core Problem: Urban regional climate simulations inadequately represent city-specific morphology and energy processes.
Key Innovation: Incorporates city-specific Earth-observation inputs into the BEP-BEM urban canopy scheme within WRF.
67. Divergent drainage-scale responses to lithological heterogeneity
Core Problem: A transferable diagnostic is lacking for where and at what drainage scale lithologic contrasts reorganize river networks.
Key Innovation: Introduces the rock proportion-contributing area order relationship and finds a robust V-shaped scale response using simulations and natural networks.
68. LitModDyn: quantifying dynamic topography induced by slab-mantle interaction beneath the Adria microplate
Core Problem: Estimate dynamic topography consistently from thermochemical lithospheric structure and mantle flow.
Key Innovation: Provides a Python post-processing tool that couples thermodynamic structure, rheological layering, density, and flow to calculate dynamic topography.
69. Cross-Year Crop and Land-Cover Classification with Limited Labels
Core Problem: Produce annual crop and land-cover maps when dense semantic labels are available for only a few source years.
Key Innovation: Tests fixed source-year segmentation models with spectral and vegetation-index inputs across multiple unlabeled target years.
70. A DEM investigation into the nature of shear strength in bio-cemented grains
Core Problem: Clarify whether bio-cemented granular strength arises primarily from cohesion or from friction, geometric interlocking, and mineral breakage.
Key Innovation: Uses a stochastic crystal-growth DEM to represent calcium-carbonate geometry and isolate confinement, friction, bond-strength, and fracture effects.
71. Learning instance-level semantic Gaussian representations for occlusion-aware urban scene reconstruction from UAV imagery
Core Problem: Title-level focus: recover instance-aware semantic urban geometry from UAV imagery under occlusion.
Key Innovation: Title-signalled approach or contribution: Learns instance-level semantic Gaussian representations designed for occlusion-aware scene reconstruction. Methods, data and results could not be assessed because no reliable abstract was available.
72. Vegetation canopy height retrieval in complex mountainous regions based on data calibration and CNN model
Core Problem: Title-level focus: retrieve vegetation canopy height reliably in complex mountainous terrain.
Key Innovation: Title-signalled approach or contribution: Combines data calibration with a CNN-based canopy-height retrieval approach. Methods, data and results could not be assessed because no reliable abstract was available.
73. Salinity effects on freezing behavior and moisture redistribution under freeze-thaw-evaporation coupling
Core Problem: Title-level focus: determine how salinity affects freezing behavior and moisture redistribution under coupled freeze-thaw and evaporation conditions.
Key Innovation: Title-signalled approach or contribution: Examines water-salt interactions within a coupled freeze-thaw-evaporation framework. Methods, data and results could not be assessed because no reliable abstract was available.
74. Long-term monitoring of lake dynamics across the Yangtze River Basin using multi-source altimetry fusion
Core Problem: Title-level focus: construct continuous long-term records of lake dynamics across the Yangtze River Basin.
Key Innovation: Title-signalled approach or contribution: Fuses multiple satellite-altimetry sources for basin-wide lake monitoring. Methods, data and results could not be assessed because no reliable abstract was available.
75. Hydrologic Implications of GNSS Vertical Displacements in Cold Regions: Insights From Northeast China
Core Problem: It is uncertain whether sparse GNSS vertical-displacement observations reliably represent water-storage variability when hydrologic loading is weak and freeze-thaw deformation is present.
Key Innovation: Compares 2012–2023 GNSS, GRACE, and Noah estimates at regional and station scales and separates long-term, interannual, seasonal, and wintertime consistency.
76. Geostatistical Discharge Estimation in Hydroelectric Canals: Integration Kriging, Conditional Simulation, and Sampling Design
Core Problem: Conventional current-meter procedures do not rigorously propagate uncertainty in cross-section-integrated discharge or optimize measurement density.
Key Innovation: Combines integration kriging and conditional simulation to produce discharge estimates, confidence intervals, and sampling-density versus uncertainty trade-offs across 36 surveys.
77. Uncertainty analysis and calibration for regional-scale resilience assessment of building portfolios using simulation-based and empirical-based fragility approaches
Core Problem: Title-level focus: quantifying and calibrating uncertainty from simulation-based and empirical fragility approaches in regional resilience assessment.
Key Innovation: Title-signalled approach or contribution: The title indicates comparative uncertainty analysis and calibration across two fragility paradigms; no abstract is available, so hazard, data, metrics, and findings are unknown. Methods, data and results could not be assessed because no reliable abstract was available.
78. Community Resilience: A Hybrid Framework of Multi-Agent Simulation and Spatiotemporal Graph Learning
Core Problem: Title-level focus: jointly representing agent behavior and spatiotemporal dependencies in community-resilience analysis.
Key Innovation: Title-signalled approach or contribution: The title proposes a hybrid of multi-agent simulation and spatiotemporal graph learning; no abstract is available, so real-event data, validation, and performance are unknown. Methods, data and results could not be assessed because no reliable abstract was available.
79. Characterizing the microstructure of Yellow River ice using X-ray computed tomography
Core Problem: Title-level focus: three-dimensionally characterizing the internal microstructure of Yellow River ice.
Key Innovation: Title-signalled approach or contribution: The title specifies X-ray computed tomography as the characterization method; no abstract is available, so sample representativeness, derived metrics, and findings are unknown. Methods, data and results could not be assessed because no reliable abstract was available.
80. S-GRHyMoLAP: A stochastic extension of the GRHyMoLAP model with different diffusion mechanisms for probabilistic rainfall-runoff modeling
Core Problem: Title-level focus: representing stochasticity and alternative diffusion mechanisms in rainfall-runoff simulation.
Key Innovation: Title-signalled approach or contribution: The title describes a stochastic extension of GRHyMoLAP with different diffusion mechanisms; no abstract is available, so its formulation, data, and performance are unknown. Methods, data and results could not be assessed because no reliable abstract was available.
81. Quantitative monitoring of DNAPL source zone evolution in heterogeneous aquifers via deep-learning fusion of time-lapse surface and cross-borehole ERT with borehole constraints
Core Problem: Title-level focus: quantitatively tracking DNAPL source-zone evolution in heterogeneous aquifers from complementary but incomplete electrical-resistivity observations.
Key Innovation: Title-signalled approach or contribution: The title fuses time-lapse surface and cross-borehole ERT with borehole constraints through deep learning; no abstract is available, so architecture, field setting, resolution, and accuracy are unknown. Methods, data and results could not be assessed because no reliable abstract was available.
82. Flume experiments on wave attenuation through combined flexible and rigid vegetation
Core Problem: Title-level focus: understanding wave attenuation through configurations that combine flexible and rigid vegetation.
Key Innovation: Title-signalled approach or contribution: The title indicates flume experiments on combined vegetation types; no abstract is available, so layout, wave conditions, measurements, and attenuation results are unknown. Methods, data and results could not be assessed because no reliable abstract was available.
83. Enhanced tidal level forecasting: A hybrid CEEMDAN-LSTM framework with Bayesian optimization
Core Problem: Title-level focus: forecasting tidal levels from complex, nonstationary time series while tuning the predictive model effectively.
Key Innovation: Title-signalled approach or contribution: The title combines CEEMDAN decomposition, LSTM forecasting, and Bayesian optimization; no abstract is available, so forecast horizon, sites, baselines, and gains are unknown. Methods, data and results could not be assessed because no reliable abstract was available.
84. ELS²C-Net: A Lightweight Network for Hyperspectral Image Classification via Kernel Decomposition and Multi-Scale Convolution
Core Problem: Title-level focus: reducing computational burden in hyperspectral classification while representing multiscale spatial features.
Key Innovation: Title-signalled approach or contribution: The title combines kernel decomposition and multiscale convolution in ELS²C-Net; no abstract is available, so architecture details, datasets, and performance are unknown. Methods, data and results could not be assessed because no reliable abstract was available.
85. DTDC-DeepLabv3+: A Dual-Stream Terrain-Guided Network with DenseASPP and CBAM for Water Body Extraction in Ganzhou
Core Problem: Title-level focus: extracting water bodies reliably in terrain-complex remote-sensing scenes.
Key Innovation: Title-signalled approach or contribution: The title describes a dual-stream terrain-guided DeepLabv3+ network incorporating DenseASPP and CBAM; no abstract is available, so imagery, labels, comparisons, and gains are unknown. Methods, data and results could not be assessed because no reliable abstract was available.
86. A Lightweight CNN-SEEDS Fusion Model with Factor Analysis for Hyperspectral Image Classification
Core Problem: Title-level focus: performing efficient hyperspectral classification by combining learned and superpixel-level information.
Key Innovation: Title-signalled approach or contribution: The title presents a lightweight CNN-SEEDS fusion model with factor analysis; no abstract is available, so the fusion mechanism, datasets, and results are unknown. Methods, data and results could not be assessed because no reliable abstract was available.
87. A simple yet effective method for correcting geolocation errors caused by satellite jitter
Core Problem: Title-level focus: correcting image geolocation errors introduced by satellite-platform jitter.
Key Innovation: Title-signalled approach or contribution: The title claims a simple correction method; no abstract is available, so algorithmic details, sensor type, error magnitude, and validation results are unknown. Methods, data and results could not be assessed because no reliable abstract was available.