TerraMosaic Daily Digest: July 5, 2026
Daily Summary
The July 5 literature is led by landslide data and process studies rather than by generic remote-sensing methods. The strongest contribution is the Unified Global Landslide Catalogue, which consolidates open landslide records into a standardized global framework for discovery, comparison, and large-scale risk analysis. New field-oriented studies then move from inventory to mechanism: clustered landslides triggered by Typhoon Gaemi are tied to rainfall, lithology, fault proximity, and slope geometry; Three Gorges Reservoir landslides are reframed through groundwater-level thresholds; and drainage-trench performance is shown to depend on basal shear zones and internal heterogeneity rather than homogeneous hillslope assumptions.
The methods papers are most persuasive when they solve a concrete geohazard bottleneck. MRSD-U-Net targets mining-induced landslides under sparse labels by fusing multisource remote-sensing cues with residual learning. FastKAN tests whether Kolmogorov-Arnold networks can accelerate landslide susceptibility mapping while preserving interpretability through SHAP. Debris-flow susceptibility work couples conventional classifiers with metaheuristic feature selection, while SC-Net, UAV Eco-DRR synthesis, and dock-based UAV monitoring extend the same operational thread into corridor safety, ecosystem-based disaster risk reduction, and mountainous geohazard surveillance.
Beyond landslides, the issue adds infrastructure and climate-risk context. Earthquake-impact databases, seismic building-slope interaction, and machine-learned RC hysteresis models strengthen post-event consequence accounting and structural fragility analysis. Flood studies span AI-generated seasonal weather ensembles, compound coastal inundation, extreme-rainstorm transport damage, and flood-vulnerability mapping. The AI/remote-sensing layer is selective but important: WorldTensor, GeoSAM-Lite, LEVIRDet, TESSERA v2, GlacierCastAI, and regional weather downscaling all point toward broader foundation-model infrastructure for hazard-aware Earth-system analysis.
Key Trends
Five movements define this issue: standardized landslide data, groundwater-aware slope mechanics, task-specific remote sensing, AI-supported flood risk, and engineering consequence records.
- Landslide research is shifting from isolated inventories to reusable data infrastructure: UGLC, mining-induced landslide extraction, clustered-landslide mapping, and remote-sensing corridor monitoring all treat inventories as living analytical assets for rapid updating, transfer, and risk screening.
- Hydro-mechanical explanation is gaining ground over rainfall-only warning logic: The Three Gorges groundwater-threshold study, drainage-trench analysis, and Typhoon Gaemi clustered-landslide work emphasize groundwater routing, basal shear zones, lithology, and rainfall history as coupled controls on slope response.
- Operational remote sensing is becoming more task-specific: UAS Eco-DRR, dock-based geohazard UAV systems, shaft inspection, SC-Net, LEVIRDet, GeoSAM-Lite, and TESSERA v2 show a move from generic mapping toward constrained monitoring tasks with clear deployment contexts.
- Flood-risk modeling is absorbing AI weather and compound-hazard structure: AI-based seasonal weather ensembles, compound coastal inundation modeling, high-flow volume estimation, rainstorm transport-damage analysis, and MCS projections all target physically plausible extremes rather than isolated historical events.
- Engineering resilience papers are linking damage, uncertainty, and consequence records: Earthquake-impact databases, building-slope interaction, RC hysteresis adaptation, freeze-thaw rock damage, and infrastructure-damage studies connect mechanical response to the records needed for risk and recovery decisions.
Selected Papers
The selected papers cover the Unified Global Landslide Catalogue, Typhoon Gaemi clustered landslides, mining-induced landslide extraction, groundwater-level landslide warning, drainage-trench stabilization, debris-flow susceptibility, UAS-based Eco-DRR, landslide corridor monitoring, seismic slope-building interaction, earthquake impact databases, AI-based flood-risk ensembles, compound coastal flooding, extreme-rainstorm infrastructure damage, wildfire human sensor networks, WorldTensor, GeoSAM-Lite, LEVIRDet, TESSERA v2, GlacierCastAI, weather downscaling, karst groundwater, DAS earthquake data, disaster governance, bridge damage identification, search-and-rescue UAVs, building change detection, Earth-observation foundation-model adaptation, marine DAS monitoring, porous-media reconstruction, and physics-informed geomechanics. This issue contains 57 selected papers from 3075 papers analyzed.
1. Unified Global Landslide Catalogue (UGLC): a single, standardised global-scale landslide dataset
Core Problem: Landslide risk analysis remains limited by fragmented inventories with inconsistent formats, spatial coverage, and event descriptors.
Key Innovation: Integrates multiple open landslide datasets into a standardized global catalogue designed for data discovery, comparative analysis, and large-scale risk studies.
2. Controlling factors of clustered landslides triggered by extreme rainstorms from the 2024 Typhoon Gaemi in Zixing County, China
Core Problem: Extreme typhoon rainstorms can trigger dense landslide clusters, but the relative roles of rainfall, lithology, faults, and terrain are rarely quantified after the event.
Key Innovation: Builds a field-validated landslide polygon inventory and uses XGBoost with SHAP to identify 3-day rainfall, granite strata, fault proximity, elevation, slope, and aspect as dominant controls.
3. An enhanced deep learning network for rapid extraction of mining-induced landslides in data-scarce areas
Core Problem: Rapid landslide mapping in mining regions is constrained by sparse labels and heterogeneous terrain, delaying hazard-map updating and post-event assessment.
Key Innovation: Develops MRSD-U-Net, a multisource remote-sensing and residual-learning network that improves mining-induced landslide extraction under limited labeled samples.
4. Landslide deformation mechanism and groundwater level prediction model under the competitive interaction of rainfall recharge and groundwater discharge
Core Problem: Rainfall thresholds alone cannot represent the groundwater dynamics that directly control large reservoir landslide deformation.
Key Innovation: Links rainfall recharge, groundwater discharge, geological damage, and deformation in Three Gorges Reservoir landslides and proposes groundwater-level thresholds for physical early warning.
5. Hydraulic Performance of Drainage Trenches for Landslide Stabilization
Core Problem: Drainage-trench design often assumes homogeneous landslide materials, overlooking basal shear zones and internal heterogeneity that control groundwater drawdown.
Key Innovation: Uses finite-element seepage analyses to show how shear-zone and thin-layer hydraulic properties can strongly modify trench effectiveness in slow-moving landslides.
6. A Systematic Review of UAS-Based Remote Sensing for Ecosystem-Based Disaster Risk Reduction: Applications, Policy Interfaces, and Opportunities
Core Problem: UAS remote sensing is increasingly used in Eco-DRR, but applications across hazards, ecosystems, and policy interfaces remain fragmented.
Key Innovation: Synthesizes 51 studies to show how multisensor UAS workflows support monitoring of landslides, wildfires, hydrologic hazards, coastal stability, and nature-based risk reduction.
7. Comparison of machine learning algorithms combined with metaheuristic-based feature selection methods for debris flow susceptibility assessments
Core Problem: Debris-flow susceptibility models must select informative conditioning factors from many correlated terrain, hydrologic, and environmental variables.
Key Innovation: Combines SVM and random forest with particle swarm, grey wolf, and whale optimization feature selection to improve debris-flow susceptibility mapping in Diebu County, China.
8. Very fast Kolmogorov-Arnold network for landslide susceptibility prediction
Core Problem: Landslide susceptibility mapping needs models that are accurate, computationally efficient, and interpretable enough for risk screening.
Key Innovation: Benchmarks FastKAN against boosting and neural models in Gümüşhane, Türkiye, combining susceptibility prediction with SHAP-based feature interpretation.
9. SC-Net: Structural Constrained Contrastive Learning for Landslide Extraction Toward Power Transmission Corridor Safety Monitoring
Core Problem: Power transmission corridors require rapid landslide extraction in complex terrain, where ordinary segmentation models can miss structure-relevant features.
Key Innovation: Introduces a structurally constrained contrastive-learning network for landslide extraction aimed at safety monitoring along power-transmission corridors.
10. Integrated seismic and slope stability evaluation of reinforced concrete buildings constructed on sloping ground
Core Problem: Buildings on sloping ground are governed by both structural irregularity and slope stability, but these risks are often assessed separately.
Key Innovation: Couples ETABS response-spectrum structural analysis with SLOPE/W limit-equilibrium evaluation to assess RC building configurations on seismic slopes.
11. Earthquake impact on built environment and on the population: a database for Greece
Core Problem: Earthquake consequence records are often incomplete or inconsistent, limiting national-scale risk and impact analysis.
Key Innovation: Constructs the Greek Earthquake Impact Database with focal parameters, building damage, fatalities, injuries, intensity metadata, and references for 253 earthquakes.
12. Evaluation of AI-based seasonal weather ensembles as input for fluvial flood risk estimation: a case study over the Elbe basin
Core Problem: Flood-risk analysis needs physically plausible synthetic weather ensembles that extend beyond the historical record while preserving spatial coherence.
Key Innovation: Uses an SFNO-based AI weather model with a precipitation diagnostic model to generate large seasonal ensembles for Elbe River flood-risk estimation.
13. Pronounced Spatiotemporal Differences of Compound Flooding Inundation in China's Coastal Cities Under Climate Change
Core Problem: Coastal cities face compound inundation from rainfall and sea level, but driver dominance and future change vary strongly across space.
Key Innovation: Couples statistical design events with SFINCS simulations to quantify compound flooding across 16 Chinese coastal cities under historical and future climate scenarios.
14. A harmonised dataset for Earth system foundation models
Core Problem: Earth-system foundation models need training data that combine climate, land, ocean, cryosphere, infrastructure, hazard, and socioeconomic variables on common grids.
Key Innovation: Introduces WorldTensor, a harmonized global 0.25-degree annual dataset aligning environmental and socioeconomic layers for multimodal Earth-system learning.
15. LEVIRDet: A Million-Scale 159-Category Dataset and Foundation Model for Universal Remote Sensing Object Detection
Core Problem: Remote-sensing object detection remains fragmented by category sets, sensor types, spatial scales, and annotation granularity.
Key Innovation: Introduces a million-scale 159-category detection dataset and a scale-hierarchy-aware foundation detector for universal remote-sensing object detection.
16. GeoSAM-Lite: A Lightweight Foundation Model for Onboard Remote Sensing Segmentation
Core Problem: Large segmentation foundation models are difficult to deploy on resource-constrained Earth-observation platforms and suffer from geospatial domain shift.
Key Innovation: Distills geospatial priors into a lightweight prompt-free segmentation model with feature-fusion layers for efficient onboard remote-sensing use.
17. Evaluation and Future Changes of Mesoscale Convective Systems Over the Conterminous United States in High-Resolution Global and Regional Simulations
Core Problem: Mesoscale convective systems drive major U.S. extreme precipitation, but their future intensity, frequency, and spatial structure remain hard to constrain.
Key Innovation: Compares observations, ERA5, CESM-HR, and CONUS404 to evaluate MCS simulation skill and project intensified MCS-related precipitation under warming.
18. Flood vulnerability assessment in Mopani District Municipality, South Africa. Part 1: Physical factors and adaptation framework
Core Problem: Rural flood vulnerability depends on terrain, rainfall, drainage, land cover, soils, and infrastructure access, but local adaptation planning needs spatially explicit assessment.
Key Innovation: Uses AHP and weighted-overlay mapping to identify physical flood-vulnerability patterns and adaptation priorities in Mopani District, South Africa.
19. "23.7" extreme rainstorm: investigation and analysis of transportation infrastructure damage in Mentougou District
Core Problem: Extreme rainstorms can isolate residents and disrupt emergency response when transportation infrastructure fails.
Key Innovation: Integrates documents, literature, interviews, and field visits to analyze infrastructure damage, cascading access failures, and recovery lessons from the 2023 Mentougou rainstorm.
20. Vision-based human sensor networks for wildfires
Core Problem: Wildfire monitoring needs real-time local evidence, but formal sensor systems can be overwhelmed during fast-moving disasters.
Key Innovation: Frames social-media images and videos as a human sensor network for wildfire detection, severity mapping, and emergency situational awareness.
21. On the Seismic Response of Irregular Concrete Building with Shark Dampers Considering SSI Effects
Core Problem: Plan-irregular RC buildings near faults are vulnerable to torsion and soil-structure interaction under pulse-like ground motions.
Key Innovation: Uses nonlinear time-history analysis to show how foundation flexibility and short-stroke metallic yielding dampers jointly control drift and torsional response.
22. TestMate: Test-Time Domain Adaptation Aided by Lightweight Vision Foundation Model
Core Problem: Dense prediction models used in remote sensing and hazard mapping often degrade under streaming distribution shifts without labels.
Key Innovation: Uses a lightweight vision foundation model and memory-based test-time adaptation to improve segmentation robustness without backpropagation-heavy updates.
23. iVISION-2DCD: A Long-Term Change Detection Dataset for Large-Scale Outdoor Construction Monitoring
Core Problem: Construction and infrastructure monitoring need change detection that works across long time spans and arbitrary camera viewpoints.
Key Innovation: Introduces a long-term 2D change-detection dataset for outdoor construction scenes, supporting automated progress and site-change monitoring.
24. Phase-Preserving Trimodal Transformer for Tropical Forest Biomass Estimation Using Optical and PolInSAR Data
Core Problem: Tropical biomass estimation is limited by optical cloud cover and SAR saturation in dense forests.
Key Innovation: Fuses optical, P-band, and L-band PolInSAR data with complex-valued encoders to preserve phase information for above-ground biomass estimation.
25. Exploring SAM Supervision for Fine-Grained UAV Target Segmentation under Data Scarcity
Core Problem: UAV segmentation often lacks dense annotations for small objects in cluttered scenes.
Key Innovation: Uses SAM-generated pseudo-labels and localized refinement to train lightweight UAV segmentation models under data scarcity.
26. TESSERA v2: Scaling Pixel-wise Earth Foundation Models
Core Problem: Pixel-level Earth-observation foundation models must scale while retaining dense spatial information useful for mapping and monitoring.
Key Innovation: Extends the TESSERA line toward scalable pixel-wise Earth foundation modeling for downstream geospatial prediction tasks.
27. GlacierCastAI: Predicting Glacier Retreat from Multi-Modal Satellite Imagery and Climate Signals
Core Problem: Glacier retreat prediction requires integrating satellite imagery with climate drivers across spatial and temporal scales.
Key Innovation: Builds a multimodal AI framework that combines satellite imagery and climate signals to forecast glacier retreat.
28. From Global to Local: Efficient Regional Weather Downscaling with Global Weather Foundation Model
Core Problem: Hazard forecasting needs local-scale weather information, but regional downscaling can be computationally expensive.
Key Innovation: Adapts global weather foundation-model information for efficient regional downscaling relevant to local hazard and hydrologic modeling.
29. U3DWind: A Low Altitude Wind Field Dataset and Benchmark for Urban Air Mobility
Core Problem: Low-altitude wind hazards affect urban air mobility and emergency UAV operations, but city-scale benchmark data remain limited.
Key Innovation: Introduces a building-resolved wind-field benchmark generated with GPU-accelerated LBM-LES simulations across five megacities.
30. TRISTAR: Triple-Signal Stair Recognition and Vision-Only Indoor Navigation for Search-and-Rescue Micro-UAVs
Core Problem: Indoor search-and-rescue operations often lack GNSS, safe human access, and low-cost aerial navigation options.
Key Innovation: Combines monocular depth estimation, classical vision, lightweight learning, victim detection, and stair-ascent recognition for low-cost micro-UAV search-and-rescue navigation.
31. FM-ChangeNet: Learning Change through Pathwise Feature Transport
Core Problem: Bi-temporal change detection can confuse true structural change with illumination shifts, misregistration, and endpoint-only supervision artifacts.
Key Innovation: Reframes change detection as continuous feature-space transport, using pathwise supervision and velocity magnitude as an interpretable spatial change cue.
32. SIMPLER: Efficient Foundation Model Adaptation via Similarity-Guided Layer Pruning for Earth Observation
Core Problem: Fine-tuning Earth-observation foundation models is often too expensive for rapid downstream mapping and deployment.
Key Innovation: Selects redundant transformer layers before adaptation using representation similarity, reducing training and inference cost across Prithvi-EO-2 and TerraMind-style models.
33. Statistical Characterization of High Flow Volumes Across the Conterminous United States Supporting Managed Aquifer Recharge
Core Problem: Managed aquifer recharge using floodwater needs national-scale estimates of high-flow volume frequency, especially in ungaged basins.
Key Innovation: Estimates high-flow volumes for 2,115 CONUS subbasins using prediction-in-ungaged-basin methods and extreme-value modeling.
34. Toward Joint Information Selection and Policy Learning in Water Resources Management
Core Problem: Water operating policies are often limited by narrow information sets despite richer monitoring and forecasting data.
Key Innovation: Frames information selection as a core design problem and reviews methods for learning policy inputs jointly with management decisions.
35. Ecogeomorphologic Response of a Tidally Restricted Salt Marsh to Tidal Restoration and Sea-Level Rise
Core Problem: Tidally restricted marshes can have low accretion, land subsidence, and underdeveloped channels, reducing resilience to sea-level rise.
Key Innovation: Couples hydrodynamic and marsh-accretion modeling to assess restoration potential under tidal reconnection and future sea-level scenarios.
36. UAV-Based Photogrammetric Inspection in Deep Vertical Shafts: A Case Study from the KGHM GG-1 Mineshaft in Poland
Core Problem: Deep mine shafts are difficult and hazardous to inspect with conventional human-access methods.
Key Innovation: Develops and tests a UAV photogrammetric inspection system for deep vertical mine shafts under high airflow, confined geometry, and GNSS-denied conditions.
37. Machine learning-based parameter adaptation approach for modeling hysteretic axial-shear-flexure interaction in RC columns
Core Problem: RC columns under earthquake loading exhibit coupled axial, shear, and flexural degradation that is expensive to model in system-level assessments.
Key Innovation: Embeds machine-learning surrogate models in adaptive hysteretic formulations to update parameters during cyclic analysis.
38. Uncertainty-aware damage identification in short-span bridges via physics-informed variational autoencoder
Core Problem: Post-event bridge assessment needs models that identify structural damage while retaining uncertainty estimates under sparse monitoring data.
Key Innovation: Combines physics-informed constraints with a variational autoencoder to estimate short-span bridge damage and associated uncertainty.
39. A Constitutive Model of Rock Freeze-Thaw Damage Based on Irreversible Thermodynamics
Core Problem: Freeze-thaw cycling degrades rock microstructure and strength, threatening rock-engineering stability in cold regions.
Key Innovation: Links NMR-derived pore evolution to an irreversible-thermodynamics damage variable for constitutive modeling of freeze-thaw sandstone degradation.
40. Automated dock-based UAV systems for geohazard monitoring in mountainous terrain
Core Problem: Mountain geohazard monitoring is constrained by access, weather, and the need for repeatable high-resolution observations.
Key Innovation: Evaluates automated dock-based UAV workflows for repeat geohazard monitoring in mountainous terrain.
41. MultiFireSeg: A Multitask Network With Interactive Prompts and Boundary Supervision for Wildfire Segmentation
Core Problem: Operational wildfire mapping requires segmentation models that preserve boundaries and respond to heterogeneous fire scenes.
Key Innovation: Uses multitask learning, interactive prompts, and boundary supervision to improve wildfire segmentation.
42. Long-term marine acoustic and seismic monitoring using distributed acoustic sensing and deep learning
Core Problem: Submarine geophysical monitoring is limited by sparse instrumentation and the difficulty of extracting signals from massive DAS recordings.
Key Innovation: Uses deep learning on nearly four years of submarine fiber-optic DAS data to detect local earthquakes, distant T-waves, volcanic-eruption signals, whale calls, and vessel traffic.
43. SiamixFormer: a fully-transformer Siamese network with temporal Fusion for accurate building detection and change detection in bi-temporal remote sensing images
Core Problem: Post-disaster mapping and urban exposure analysis require reliable building detection and change detection from multi-temporal remote-sensing imagery.
Key Innovation: Develops a fully transformer-based Siamese architecture with temporal fusion for building extraction and bi-temporal change detection.
44. Earthquake catalog and continuous waveforms from a two-week distributed acoustic sensing experiment on Kefalonia Island, Greece
Core Problem: Dense seismic monitoring can improve earthquake detection and source characterization, but open DAS datasets remain limited.
Key Innovation: Publishes an earthquake catalog and continuous DAS waveforms from a two-week experiment on Kefalonia Island.
45. Risk assessment of karst groundwater contamination from closed goafs: mechanisms, indicators, and model application
Core Problem: Closed goafs can alter groundwater pathways and contamination risks in karst settings.
Key Innovation: Develops mechanisms, indicators, and a model application for assessing karst groundwater contamination risk from closed mining goafs.
46. Quantitative evaluation of disaster governance policies in the Guangdong-Hong Kong-Macao greater Bay Area and policy optimization recommendations
Core Problem: Regional disaster risk reduction depends on policy coordination, but governance effectiveness is difficult to evaluate quantitatively across jurisdictions.
Key Innovation: Assesses disaster-governance policies in the Guangdong-Hong Kong-Macao Greater Bay Area and identifies optimization directions for regional risk governance.
47. CT-TreeFlow: Probabilistic Groundwater-Potential Mapping Using Remote Sensing-Derived Environmental Predictors in Karst Aquifers
Core Problem: Karst aquifer groundwater potential is spatially heterogeneous and difficult to map using sparse field observations alone.
Key Innovation: Uses remote-sensing environmental predictors and probabilistic modeling to map groundwater potential in karst aquifers.
48. Running speed limits of railway vehicles on simply-supported bridge subjected to strike-slip faulting
Core Problem: Strike-slip fault displacement can compromise bridge-track-vehicle safety, requiring operational speed limits under faulting scenarios.
Key Innovation: Analyzes railway vehicle running speed limits on simply supported bridges subjected to strike-slip faulting.
49. A Time-Domain State-Space Iterative (TDSSI) method for seismic wave propagation in layered media under oblique incidence
Core Problem: Layered media and oblique incidence complicate time-domain seismic-wave modeling needed for site-response analysis.
Key Innovation: Develops a time-domain state-space iterative method for seismic wave propagation in layered media.
50. Adaptive loss-weighted PINNs for modeling elasto-dynamic behavior of rocks
Core Problem: Physics-informed neural networks can model rock dynamics, but balancing loss terms is difficult for elastic-wave problems.
Key Innovation: Introduces adaptive loss weighting for PINNs to improve elasto-dynamic modeling of rocks.
51. Data-driven and physics-guided constitutive modeling of hard rock using hybrid PANN networks
Core Problem: Hard-rock constitutive behavior is difficult to capture with purely empirical or purely mechanistic models.
Key Innovation: Combines data-driven and physics-guided neural architecture to model hard-rock mechanical response.
52. Property-Constrained 3D Porous Media Reconstruction from 2D Images via Conditional Generative Adversarial Networks
Core Problem: Hydro-mechanical and subsurface-flow modeling often lacks realistic 3D porous structures when only 2D imagery is available.
Key Innovation: Uses conditional generative adversarial networks with property constraints to reconstruct 3D porous media from 2D images.
53. Kolmogorov-Arnold networks for consolidation modeling of dual-porosity media
Core Problem: Dual-porosity consolidation involves coupled flow and deformation processes that challenge efficient surrogate modeling.
Key Innovation: Applies Kolmogorov-Arnold networks to consolidation modeling in dual-porosity media.
54. Generative AI-enabled granular force measurement from photoelastic images and validation of stress-force-fabric relationship
Core Problem: Granular force networks control soil and rock assembly behavior but are difficult to measure from photoelastic images.
Key Innovation: Uses generative AI to infer granular forces and validate stress-force-fabric relationships.
55. A stabilized two-phase two-layer SPFEM for large deformation water-soil coupling problems
Core Problem: Large deformation water-soil coupling is central to many slope, flow-slide, and ground-failure problems but remains numerically challenging.
Key Innovation: Develops a stabilized two-phase two-layer SPFEM formulation for large-deformation hydro-mechanical simulations.
56. GeoSelect: Spatial-Program Execution for Training-Free Referring Remote Sensing Image Segmentation
Core Problem: Remote-sensing referring segmentation requires spatial reasoning over objects, sizes, ordinal relations, and proximity without task-specific training.
Key Innovation: Converts text expressions into explicit spatial programs whose intermediate fields and rankings are inspectable.
57. LBTCap: A Lightweight Bilateral Transformer for Real-Time Remote Sensing Image Change Captioning
Core Problem: Disaster and infrastructure monitoring benefit from models that can describe scene changes, not only detect them.
Key Innovation: Introduces a lightweight bilateral transformer for real-time captioning of changes in remote-sensing images.