TerraMosaic Daily Digest: April 14, 2026
Daily Summary
This April 14, 2026 digest distills 40 selected papers from 1,494 analyzed records. The April 14 digest is strongest once the landslide papers are put back at the center. The key slope studies now fall into three concrete groups. First are event-scale susceptibility papers, where the Nepal Himalaya, Noto, Western Quebec, and the Yangtze infrastructure corridor show that InSAR deformation, channel steepness, explainable predictors, and time-varying climate windows add real value beyond static factor stacks. Second are deformation and morphology papers, where LiDAR in soft-rock New Zealand, PS- and SBAS-InSAR at Lumei, phase-gradient strain fields at Taoping, and UAV LiDAR at Surami are used to separate moving sectors, instability regimes, or boundary geometry directly. Third are landslide mapping and segmentation papers such as SGH-Net, FANet, DeepLrn-Source, and the text-semantic positive-unlabeled framework, which matter less as generic AI exercises and more because they make source zones, susceptibility structure, and model logic easier to interpret.
The non-landslide papers are useful when they stay equally concrete. The Yangtze paper measures changing flood-regulation capacity at basin scale rather than just discussing exposure. The groundwater-reservoir digital twin turns drought mitigation into an optimization problem. The Sunkoshi landslide-dam paper follows breach routing downstream instead of stopping at the dam crest. Flood-loss sensitivity, Central America translation-to-action work, Tibetan ice-rock avalanche tracking, glacier retreat in Peru, and continental baseflow analysis are all strongest when they estimate the variable a practitioner would actually need. Overall, this is not a day of abstract 'AI for hazards' claims. It is a day of studies that either add physically meaningful variables to hazard models or make model outputs more defensible and operational.
Key Trends
The best papers today do not just map hazards; they add physically meaningful variables and make the result easier to defend or act on.
- Event-scale landslide models improved by adding better predictors: The Nepal, Noto, Western Quebec, and Yangtze papers all move beyond static factor stacks by adding InSAR deformation, channel steepness, explainable variables, or explicit climate windows.
- Deformation products were used mechanistically, not just illustratively: LiDAR morphometry in soft-rock terrain, PS/SBAS-InSAR at Lumei, strain fields at Taoping, and UAV LiDAR at Surami all help separate instability regimes, moving sectors, or landslide boundaries directly.
- The strongest operational papers estimated the variable people actually need: Yangtze retained floodwater, groundwater-reservoir control actions, Sunkoshi breach routing, and flood-loss sensitivity are more useful than a generic hazard map because they point at a decision variable.
- Several AI papers stood out because they made hazards more interpretable: Text-semantic positive-unlabeled learning, knowledge-guided landslide reasoning, explainable coseismic modelling, and source-zone identification all improve how the model can be defended in practice.
Selected Papers
This digest features 40 selected papers from 1,494 papers analyzed, led by event-scale landslide susceptibility, deformation diagnostics, and source-zone mapping, then widening into flood operations, drought management, cryosphere change, and mechanism-aware geohazard monitoring.
1. Integrating InSAR and Channel Steepness for AI-Based Coseismic Landslide Modeling in the Nepal Himalaya
Core Problem: Event-based coseismic landslide mapping still misses physically meaningful predictors when InSAR deformation and channel steepness are left outside the susceptibility model.
Key Innovation: A Nepal Himalaya case study shows that combining InSAR-derived deformation with channel steepness materially improves rapid post-earthquake landslide prediction under strong class imbalance.
2. Where the Hills Slide Slowly: A LiDAR-Based Morphometric Framework for Landslide Instability Regimes in Soft-Rock Terrains
Core Problem: Soft-rock hill country is hard to classify regionally because slow-moving landslide systems mix lithologic, fluvial, and hillslope signals in the same terrain.
Key Innovation: This LiDAR-based framework separates instability regimes in New Zealand soft-rock terrain and shows how lithology, incision, and valley confinement organize slow landslide behavior.
3. Landslide susceptibility assessment based on textual semantic-numerical embedding and positive-unlabeled learning
Core Problem: Conventional susceptibility models underuse geological text knowledge and rely on questionable negative samples, which limits both interpretability and reliability.
Key Innovation: The paper combines geological text semantics, numerical factors, spatial similarity, and positive-unlabeled learning to build a more interpretable landslide susceptibility workflow in the Three Gorges area.
4. Geomorphometry-Informed Ground-Motion Modeling for Earthquake-Induced Landslides
Core Problem: Earthquake-induced landslide models are weakened when their shaking inputs ignore terrain-controlled site effects in complex mountains.
Key Innovation: A geomorphometry-informed ground-motion model uses DEM-derived site proxies and finite-fault metrics to produce better triggering inputs for earthquake-induced landslide applications.
5. A data-knowledge-model synergistic reasoning framework for landslide identification
Core Problem: Landslide identification in complex terrain remains brittle when remote-sensing evidence is not connected to explicit geoscience knowledge and reasoning.
Key Innovation: The study combines improved InSAR processing, a geoscience knowledge graph, and graph-based reasoning to make landslide identification more interpretable and transferable.
6. Hydrodynamic Analysis of Landslide Dam Breach Formation and Outburst Flood Propagation in the Sunkoshi River Basin, Nepal
Core Problem: Landslide-dam emergencies are often assessed with simplified routing assumptions that do not resolve how terrain reorganizes breach discharge and downstream force.
Key Innovation: A 2-D hydraulic simulation of the Jure landslide dam reconstructs breach discharge, velocity, depth, and downstream flood evolution in terrain-constrained Himalayan valleys.
7. Space-time variability modelling of landslide susceptibility for strategic infrastructure under changing climate scenarios: The case study of the mega clean energy transmission network (Yangtze River Basin, China)
Core Problem: Strategic infrastructure still tends to rely on static susceptibility maps even though landslide clustering and climate forcing vary through time.
Key Innovation: This study links historical clustering windows to future climate scenarios and shows how landslide susceptibility along the Yangtze clean-energy network shifts across decades.
8. Observation‐Based Spatiotemporal Analysis of the Evolving Flood Regulation Capacity for the Yangtze River Basin
Core Problem: Flood-regulation capacity is usually summarized reservoir by reservoir, leaving basin-scale changes in retained versus unregulated floodwater poorly quantified.
Key Innovation: Observation-based basin analysis directly measures how the Yangtze’s flood-regulation capacity has evolved through time and space.
9. Simulation-driven digital twin framework for drought-risk mitigation in groundwater reservoirs via hierarchical optimization
Core Problem: Groundwater-reservoir drought management remains fragmented when simulation, optimization, and reservoir-state estimation are not solved in one operational loop.
Key Innovation: A simulation-driven digital twin couples reservoir dynamics with hierarchical optimization to turn drought mitigation into an explicit control problem.
10. Application of PSInSAR Monitoring for Large-Scale Landslide with Persistent Scatterers from Deep Learning Classification
Core Problem: Large-area landslide monitoring remains difficult when persistent-scatterer selection is not adapted to heterogeneous surface conditions.
Key Innovation: A deep-learning-assisted PSInSAR workflow improves persistent-scatterer classification and supports wide-area landslide deformation monitoring.
11. SGH-Net: An Efficient Hierarchical Fusion Network with Spectrally Guided Attention for Multi-Modal Landslide Segmentation
Core Problem: Multi-modal landslide segmentation still struggles to fuse RGB, multispectral, and terrain cues without adding excessive noise or model cost.
Key Innovation: SGH-Net uses guided attention and hierarchical fusion to improve landslide boundary delineation while keeping the architecture efficient.
12. Fanet: Landslide recognition in remote sensing images based on multi-source data
Core Problem: Landslide recognition remains unstable across settings when multi-source inputs are fused without a feature-selection mechanism tuned to terrain-relevant signals.
Key Innovation: FANet combines frequency attention with remote-sensing imagery and geological factors, showing that carefully chosen multi-source inputs improve landslide recognition accuracy and generalization.
13. Deformation Characteristics of Lumei Landslide in the Tibetan Plateau Combined with PS-InSAR and SBAS-InSAR
Core Problem: Long-term slope monitoring on the Tibetan Plateau remains incomplete when a single InSAR method captures either detail or continuity, but not both.
Key Innovation: Joint PS-InSAR and SBAS-InSAR analysis resolves the Lumei landslide deformation field and highlights strong internal heterogeneity relevant to monitoring and mitigation.
14. Strain Modeling and Revealed Slope Motion Mechanisms of the Taoping Paleo-Landslide from InSAR Observations
Core Problem: Conventional deformation maps often miss the localized strain structure that actually reveals how large paleo-landslides are reorganizing internally.
Key Innovation: An InSAR phase-gradient method reconstructs strain fields across the Taoping paleo-landslide and ties compressional and extensional sectors to slope-motion mechanisms.
15. Mapping and Spatiotemporal Analysis of Landslides Along the Costa Viola Transportation Network (Italy)
Core Problem: Road-corridor landslide hazard is hard to manage without linking long-term failure mapping to actual damage records and recurrence patterns.
Key Innovation: The Costa Viola study combines geomorphological mapping with a 75-year road-damage archive to show where recurrent landslides keep disrupting transport infrastructure.
16. Machine Learning Analysis of Landslide Susceptibility in the Western Québec Seismic Zone of Canada
Core Problem: The Western Québec Seismic Zone has lacked event-linked landslide inventories robust enough to support regional susceptibility modelling.
Key Innovation: A new Val-des-Bois earthquake landslide inventory supports ML susceptibility models that clearly outperform the currently used Hazus-style baseline.
17. Prediction of Coseismic Landslides by Explainable Machine Learning Methods
Core Problem: Coseismic landslide models are often accurate but opaque, which makes them harder to trust in rapid-response settings.
Key Innovation: An explainable ML framework for the Noto Peninsula identifies slope, rupture distance, and PGA as the dominant controls behind coseismic landslide patterns.
18. DeepLrn-Source: Deep learning powered landslide source identification
Core Problem: Rapid post-event assessment is slowed when analysts cannot reliably separate landslide source areas from runout and deposition using early imagery.
Key Innovation: DeepLrn-Source uses pre- and post-event imagery to isolate source zones directly, improving rapid source-area mapping after storms or earthquakes.
19. Regional landslide susceptibility mapping of earthflows using deep neural networks: A case study in the southern Apennines, Italy
Core Problem: Regional earthflow mapping remains difficult in data-limited terrain where inventories exist but not at the density needed for more manual workflows.
Key Innovation: A deep neural network trained on lithology, slope, land use, aspect, and curvature provides regional earthflow susceptibility screening for the southern Apennines.
20. Identification of Landslide Boundaries and Key Morphological Features Using UAV LiDAR Data: A Case Study in Surami, Georgia
Core Problem: Landslide boundary mapping remains unreliable in disturbed or vegetated slopes when morphology is inferred from imagery with limited relief precision.
Key Innovation: High-resolution UAV LiDAR and terrain derivatives sharpen landslide boundary and feature detection in an anthropogenically modified slope at Surami, Georgia.
21. Land Cover and Land Use Controls on Landslide Morphometry and Occurrence in a Heterogeneous Mountain Watershed
Core Problem: Watershed-scale landslide inventories often describe terrain classes but do not isolate how vegetation structure and land use alter landslide morphometry.
Key Innovation: A long inventory from the Upper Ciliwung Watershed shows that tree-dominated and herbaceous or urban terrain produce distinct landslide sizes, shapes, and mobility patterns.
22. Application of machine learning and numerical simulation for monitoring and early warning systems of landslides and rockfalls in geohazard-prone regions
Core Problem: Warning systems remain shallow when machine-learning prediction is not tied back to the physical slope response it is supposed to anticipate.
Key Innovation: This paper combines RF-SVM-PCA forecasting with numerical simulation to build a more integrated warning workflow for landslides and rockfalls.
23. Towards global sensitivity analysis of large-scale flood loss models
Core Problem: Large-scale flood-loss models are widely used, but uncertainty attribution is often too local and too limited to support confident scenario interpretation.
Key Innovation: A global sensitivity framework identifies which assumptions dominate flood-loss uncertainty and makes large-scale risk modelling easier to interrogate.
24. Earth observations to mitigate flood impacts in Central America: translating research to decision-making and actions
Core Problem: Flood-monitoring products often stall at the research stage instead of becoming decision-ready tools for mitigation agencies.
Key Innovation: The paper focuses on how earth-observation flood products can be translated into operational decisions and concrete actions in Central America.
25. Evident Dependence of Dynamics of Baseflow on Groundwater Across the Contiguous United States
Core Problem: Baseflow is assumed to be groundwater-fed, but direct multi-scale evidence for that control has remained surprisingly thin at continental scale.
Key Innovation: A synthesis of stream gauges and groundwater records shows that groundwater dominates daily to seasonal baseflow variability across the contiguous United States.
26. Climate-Driven Glacier Retreat and Biocrust Colonization Dynamics in the Quelccaya Ice Cap, Peru
Core Problem: Glacier retreat is often mapped as ice loss alone, without tracking how the newly exposed terrain is reorganized after the ice pulls back.
Key Innovation: Remote-sensing analysis of the Quelccaya Ice Cap links glacier retreat to rapid biocrust colonization and shows how cryosphere change reorganizes fresh surfaces.
27. A Context-Aware Flood Warning Framework Integrating Ensemble Learning and LLMs
Core Problem: Flood-warning systems often stop at hazard prediction and do not adapt outputs to context, uncertainty, and communication needs.
Key Innovation: A context-aware warning framework combines ensemble learning with LLM support to extend flood prediction into adaptive communication and decision assistance.
28. The Evolution and Impact of Glacier and Ice-Rock Avalanches in the Tibetan Plateau with Sentinel-2 Time-Series Images
Core Problem: Large cryospheric mass movements on the Tibetan Plateau remain under-documented despite their downstream hazard potential.
Key Innovation: Sentinel-2 time-series imagery is used to trace the evolution and consequences of glacier and ice-rock avalanches across the Tibetan Plateau.
29. Landslide Susceptibility Mapping Using Geospatial Modelling in the Central Himalaya
Core Problem: Road-corridor landslide risk in the Central Himalaya is still often assessed qualitatively rather than with explicitly weighted conditioning factors.
Key Innovation: An AHP-GIS workflow along the Uttarkashi-Gangotri highway identifies slope, geology, and lineament density as the dominant susceptibility controls.
30. The Complex Application of Geophysical and Engineering Geological Methods in a Landslide Body for Analysis of Structural Characteristics and Reduction of Landslide Risk (Tumanyan Landslide, Armenia)
Core Problem: Complex landslide bodies are still difficult to characterize structurally without combining multiple survey methods.
Key Innovation: An integrated geophysical and engineering-geological survey resolves the Tumanyan landslide structure for better hazard interpretation and local risk reduction.
31. Geological and Social Factors Related to Disasters Caused by Complex Mass Movements: The Quilloturo Landslide in Ecuador (2024)
Core Problem: Complex mass-movement disasters are often described geologically or socially, but not with both strands kept together.
Key Innovation: The Quilloturo case links geological controls to disaster consequence, making the event more useful as a multi-dimensional landslide case study.
32. Experimental Study and THM Coupling Analysis of Slope Instability in Seasonally Frozen Ground
Core Problem: Slope instability in seasonally frozen ground remains hard to interpret because thermal, hydraulic, and mechanical controls evolve together.
Key Innovation: Experiments and THM coupling analysis clarify how freezing and thawing reorganize instability in seasonally frozen slopes.
33. Assessing Post-Fire Rockfall Hazards: A Case Study of Hazard System Adaptation and Application in Evros, Greece
Core Problem: Post-fire slope studies still focus more often on burn severity than on how rockfall systems adapt after fire.
Key Innovation: A case study from Evros treats post-fire rockfall as a system-level hazard adaptation problem rather than a narrow aftereffect.
34. A Multi-Temporal Knowledge Graph Framework for Landslide Monitoring and Hazard Assessment
Core Problem: Landslide monitoring data remain fragmented when observations, terrain factors, and hazard relations are not stored in a time-aware structure.
Key Innovation: A multi-temporal knowledge graph organizes evolving landslide evidence in a way that makes monitoring and hazard assessment easier to query and connect.
35. Quantitative Assessment of Drought Risk in Major Rice-Growing Areas in China Driven by Process-Based Crop Growth Model
Core Problem: Agricultural drought risk is often inferred from weather indices alone, without explicit crop-process response.
Key Innovation: A process-based crop-growth model quantifies drought risk across China’s major rice-growing regions by linking meteorology to crop development.
36. Quantifying Causal Impact of Drought on Vegetation Degradation in the Chad Basin (2000–2023) with Machine Learning-Enhanced Transfer Entropy
Core Problem: Vegetation degradation in drought-prone regions is frequently described correlatively rather than causally.
Key Innovation: Machine-learning-enhanced transfer entropy is used to isolate the causal influence of drought on vegetation degradation across the Chad Basin.
37. Divergent Sensitivity of Gross Primary Productivity to Compound Drought and Heatwaves Across China’s Three Major Urban Agglomerations
Core Problem: Compound drought-heat stress is spatially heterogeneous, but that heterogeneity is often flattened in broader productivity analyses.
Key Innovation: Satellite-based analysis shows that gross primary productivity responds differently to drought-heat compounding across China’s three major urban agglomerations.
38. Monitoring and Prediction of Subsidence in Mining Areas of Liaoyuan Northern New District Based on InSAR Technology
Core Problem: Mining subsidence papers often stop at retrospective mapping instead of adding a forward-looking prediction component.
Key Innovation: An InSAR-based workflow for Liaoyuan combines deformation monitoring with predictive assessment of mining subsidence evolution.
39. Flood Susceptibility Mapping of the Kosi Megafan Using Ensemble Machine Learning and SAR Data
Core Problem: Flood susceptibility over megafan terrain remains difficult to resolve when avulsive morphology and SAR-observable water signals are not modeled together.
Key Innovation: An ensemble machine-learning framework uses SAR and geomorphic information to improve flood susceptibility mapping across the Kosi Megafan.
40. Hydrothermally Altered Rocks and Their Implications for Debris Flow Generation in the Monarch Butterfly Biosphere Reserve, Mexico
Core Problem: Hydrothermal alteration is still underused as a first-order conditioning factor in debris-flow studies.
Key Innovation: The paper shows how hydrothermally altered rocks create source-material conditions favorable to debris-flow generation in the Monarch Butterfly Biosphere Reserve.