TerraMosaic Daily Digest: October 13, 2026
Daily Summary
A geological record and a monitoring correction lead this issue. In Grise Fiord, bathymetry and sediment cores reveal more than 100 submarine landslides, with three dated failures associated with past cold-to-warm transitions. Sediment evidence suggests that the failures are more than a millennium old, supporting climate-sensitive preconditioning rather than a record of ongoing collapse. In the Alps, correcting phase-unwrapping errors raises the maximum recovered rock-glacier velocity at Grosses Gufer from about 0.5 to 2.5 m/year. The two studies address different timescales but share a practical lesson: interpreting changing slope behaviour first requires establishing what the observations actually resolve.
The Jinping investigation connects a long deformation history with a rapid landslide–debris-flow chain, while a new detector improves sparse-scar screening with thresholds calibrated on validation data. Neither result removes the need to test transfer: Jinping's peak velocity comes from retrospective simulation, and the detector still degrades substantially outside its source region. The Valencian exposure assessment adds a complementary planning scale by locating residential clusters within susceptible terrain. Together, these studies distinguish the evidence needed to explain failure, identify scars and locate exposed people; none of those components alone supplies a complete risk estimate.
Beyond landslides, Negros–Sulu tsunami scenarios emphasize very short coastal arrival times, and coupled tunnel-collapse modelling follows the creation of a water-inrush pathway. Liquefaction models examine whether predictions obey physical monotonicity, while flood-network and wildfire preprints test efficient forecasts against simulated references. Probabilistic bias correction strengthens extended-range weather guidance; using it for local hazards still requires rainfall and slope-response validation. New multi-sensor observation resources and geometric methods widen the technical options, but their contribution to disaster decisions will depend on tests tied to the relevant terrain, event and operating conditions.
Key Trends
The selected studies link measurement quality, physical processes and consequences while exposing where validation stops.
- Correct the observation before interpreting acceleration: Rock-glacier unwrapping errors can hide much of the measured motion. The Arctic submarine record likewise requires chronology to separate ancient instability from present conditions.
- Follow the pathway through a disaster chain: Jinping's transition from slope failure to debris flow and the tunnel collapse–inrush reconstruction show why source conditions and the evolving transport pathway need to be considered together.
- Use external tests to judge screening value: Source-disjoint landslide detection improves at fixed validation-calibrated thresholds, yet cross-region degradation persists. Multi-station seismic monitoring draws on a different strength: coherent evidence across a physical sensor network.
- Keep exposure and hazard scenarios distinct: Residential-cluster mapping quantifies who occupies susceptible land; tsunami propagation estimates consequences of specified ruptures. Combining these products into risk requires additional probability and vulnerability information.
- Make physical consistency part of model evaluation: Liquefaction constraints reduce implausible responses, and injection-induced seismicity models test physically motivated features. Efficient flood and fire forecasts also need independent event observations beyond their simulation references.
Selected Papers
Submarine sediment archives, corrected deformation measurements and long-runout reconstruction provide complementary views of slope instability. The selected studies then follow scar detection, exposed settlements and linked hazards in coastal and underground settings. Meteorology-linked landslide records, automated building-damage models, cross-modal geometry and extended-range weather guidance connect this work with broader observation methods. Field measurements, retrospective simulations and benchmark tests establish different kinds of evidence, and their implications are kept distinct.
1. Evidence for Submarine Landslides at Climate Transitions in a High Arctic Fjord (Grise Fiord, Nunavut)
Core Problem: The timing and preconditioning of Arctic submarine slope failure remain poorly constrained as warming changes sea ice, meltwater and sediment delivery.
Key Innovation: Bathymetry, acoustic profiles and sediment cores identify more than 100 failures in Grise Fiord. Chronologies for three examples suggest failure during cold-to-warm transitions; sediment evidence indicates that the failures are older than 1000 years. The record supports climate-sensitive preconditioning, without establishing a modern failure threshold or an imminent event.
2. Correcting Unwrapping Errors Toward Reliable Rock Glacier Velocity From InSAR Time Series
Core Problem: Phase-unwrapping errors can systematically suppress InSAR estimates of moving rock glaciers and obscure their actual deformation rates.
Key Innovation: An investigation of 17 Alpine rock glaciers combines manual correction of short-baseline interferograms with phase-closure correction of longer baselines. At Grosses Gufer, the recovered maximum velocity rises from about 0.5 to 2.5 m/year and is compared with independent GNSS observations. This corrects a measurement bias rather than predicting catastrophic failure.
3. Failure mechanism and dynamic process of the Jinping high-position and long-runout landslide in the Wumeng mountainous area, China, on 8th February 2025
Core Problem: Long deformation histories must be connected with geology, rainfall and rapid runout to explain the 2025 Jinping failure.
Key Innovation: Field, remote-sensing, InSAR, laboratory and simulation evidence links anti-dip fragmented strata and rainfall weakening to a landslide–debris-flow chain. Reconstructed motion lasts about 50 seconds and reaches a simulated peak near 48 m/s. These are retrospective model results; residual rear-scarp instability remains relevant to monitoring.
4. Target-focused supervision and fine spatial evidence improve sparse landslide detection in remote-sensing imagery
Core Problem: Small scars and background-dominated supervision reduce regional landslide screening performance at controlled false-positive rates.
Key Innovation: SLD-DETR combines focused distillation, high-resolution supervision and fine-scale decoder memory. Five-seed source-scene-disjoint tests improve mAP from 53.90% to 58.07%, with lower misses at validation-calibrated thresholds. Strong degradation on cross-region transfer remains a central limitation, and another detector leads the external comparison overall.
5. Determination of the damage range in the toe zone of gentle piedmont slopes based on hydraulic mechanisms
Core Problem: Confined water behind blocking structures can damage gentle slope toes in ways that broad susceptibility maps do not describe.
Key Innovation: Three-dimensional seepage calculations are coupled with separate beam idealizations for triangular and trapezoidal uplift-pressure zones. Geometry and rainfall scenarios identify controls on toe damage, while the idealized mechanical representations require site-specific verification.
6. GIS-based assessment of population and dwelling exposure to landslide-prone areas in the Valencian Community (Spain)
Core Problem: Regional population totals can conceal concentrated exposure in dispersed and peri-urban residential settlements.
Key Innovation: Susceptibility zones, cadastral parcels and census-grid data identify residential clusters in the Valencian Community. Approximately 0.25% of the regional population is exposed, with different inland and coastal patterns. Sensitivity checks support aggregation choices, but exposure is not a complete risk estimate without failure probability and vulnerability.
7. Quantifying submarine landslides - Spatial & morphometric insights along the Calabrian continental margins
Core Problem: Morphometric differences between confined and open-slope failures complicate regional interpretation of mobilized volumes.
Key Innovation: An inventory of 1945 Calabrian-margin failures distinguishes canyon/channel and open-slope domains. Open-slope failures average about four times the size and contribute roughly three quarters of mobilized volume. This fuller evidence extends the earlier title-level entry; morphometric associations do not directly establish individual triggers or tsunami generation.
8. A coupled DDA-MPM method for the collapse-water inrush chain disaster modeling in underwater tunnels
Core Problem: Fluid ingress and rock-block failure interact strongly during underwater tunnel accidents, challenging a single simulation method.
Key Innovation: A bidirectionally coupled DDA–MPM solver is tested on block-collapse, dam-break and wave-generation benchmarks, then used to reconstruct the Shijingshan tunnel accident. The model links excavation unloading, rock collapse and reservoir connectivity, without demonstrating a prospective accident-warning system.
9. Monotonicity-constrained machine learning for CPT-based liquefaction triggering assessment
Core Problem: High classification accuracy can coexist with physically implausible liquefaction responses to stress ratio and cone resistance.
Key Innovation: Monotonic constraints are examined in random forests, boosted trees and support-vector machines for CPT-based triggering. Tree models improve discrimination while eliminating implausible splits; SVM approaches trade some performance for fewer violations. Pointwise constraints do not guarantee global monotonicity for an RBF kernel.
10. Seismic and tsunami hazard potential of the Negros–Sulu megathrust, Philippines
Core Problem: Limited rupture history and fault geometry hinder prioritization of exposed coasts around the Negros–Sulu megathrust.
Key Innovation: Eighteen structural segmentation and dip-angle scenarios are propagated with JAGURS. Some directly facing coasts have simulated arrivals within two minutes, with strong dependence on bathymetry and directivity. Uniform average slip limits the representation of local maxima; the scenarios do not provide earthquake timing or occurrence probabilities.
11. Event-Centric Earthquake Monitoring with a Multi-Station Transformer
Core Problem: Station-by-station processing can lose coherent weak arrivals and fragment the evidence needed to associate an earthquake.
Key Innovation: EQNet2 jointly predicts phases, polarity, source properties and event association with a geometry-aware multi-station Transformer. Northern California tests and continuous Ridgecrest and Bay Area waveforms support improved catalog construction. This is earthquake monitoring, not advance earthquake prediction.
12. Magnitudes and Rates of Strength‐Drawdown Modulate Injection‐Induced Seismicity
Core Problem: Injection volume alone does not resolve how evolving overpressure changes earthquake timing, location and size.
Key Innovation: Strength-reduction magnitude and rate are embedded as physically motivated features in a two-stage learning model evaluated on Utah FORGE and EGS Collab stimulation data. Feature-inclusion comparisons separate controls on event size and localization from those on onset and rate; the evidence supports mechanistic interpretation, without proving prospective control of damaging earthquakes.
13. The 2 May 2025 Mw 7.4 Drake Passage Earthquake: Insights Into Tectonics at the Shackleton Fracture Zone–Chilean Trench Transition
Core Problem: Sparse stations complicate identification of the fault structures activated at the southern Chilean subduction and Shackleton Fracture Zone transition.
Key Innovation: Relocated aftershocks, classifier-assisted depth phases and seismic reflection evidence constrain the 2025 Mw 7.4 Drake Passage sequence. The interpretation places nucleation on the interface or within the subducting plate and links aftershocks to steeper structures, while retaining uncertainty in the mainshock fault assignment.
14. From Rainfall to Road-Network Functionality: Transferable Graph Forecasting for Flood-Disrupted Transportation Networks
Core Problem: A road-impact forecasting system must transfer across network topologies without relying on calibration in every target region.
Key Innovation: G-R2R combines terrain-conditioned graph representations with rainfall and recent road-functionality states. Tests across ten Wenzhou regions and a Fitow replay support transfer within the sampled domain. The reference functionality sequences are simulation-derived, so the results do not independently establish observed road closures or evacuation performance.
15. TC-Next: Zero-Shot Multimodal Cyclone Forecasting
Core Problem: Cyclone trackers need to transfer between weather forecasting systems without retraining for each model's fields.
Key Innovation: TC-Next combines infrared imagery with generic atmospheric forecasts for 6–24-hour track and intensity prediction. Western Pacific evaluations include transfer from GraphCast to other forecast systems and a deterministic 2025-season comparison. Zero-shot performance across forecast models does not establish global basin transfer or probabilistic warning calibration.
16. Field-validated wireless vertical graphene sensors for autonomous crack and leakage monitoring in metro tunnels
Core Problem: Operational tunnel monitoring needs non-intrusive, continuous measurements of crack opening and small leakage rates.
Key Innovation: Vertical-graphene transducers and a wireless platform are tested in the laboratory and deployed for three months in a metro tunnel. The study reports monitoring of approximately 0.53 mm crack opening and 1 mL/day leakage. The deployment demonstrates sensing feasibility, while long-term durability and multi-site performance remain unestablished.
17. Integrated orthophoto-GPR-GOCAD modeling for shallow cavity detection: a case study in Qinglong County, China
Core Problem: Sparse subsurface profiles leave uncertainty in the continuity of cavities that may threaten mine safety.
Key Innovation: Orthophotos, GPR profiles and GOCAD modelling provide a common spatial framework, and excavation supports a cave-fissure interpretation at one target. The integration improves traceability of geological interpretation; it does not demonstrate a quantified detection-accuracy gain, and inter-profile geometry remains interpolation-dependent.
18. Experimental study on similar simulation of overburden fracture evolution in shallow-buried coal fire zones
Core Problem: Combustion-related damage can connect overburden fractures to the surface, feeding air leakage and ground deformation.
Key Innovation: Physical similarity tests, heated-rock testing and UDEC scenarios examine how coal thickness, damage progression and temperature affect fracture breakthrough. Reported millimetre displacements and centimetre lengths belong to the scaled experimental model and are not field subsidence predictions.
19. Impacts of Sea Surface Temperature Warming Pattern Biases on Tropical Cyclone Activity and Hazard
Core Problem: Sea-surface warming-pattern biases can distort regional tropical-cyclone hazard projections even when models share a warming scenario.
Key Innovation: Downscaled cyclone simulations compare biased and bias-corrected coupled climates. Correcting the Pacific warming pattern changes basin responses and increases projected North Atlantic activity relative to the biased model, demonstrating sensitivity to the climate-model background rather than a single robust regional forecast.
20. Divergent demographic responses to the Kikai-Akahoya volcanic catastrophe in mid-Holocene Japan
Core Problem: Large eruptions can produce spatially uneven demographic consequences that collapse-or-continuity narratives miss.
Key Innovation: Statistical demographic reconstruction of the Kikai-Akahoya eruption finds severe declines near the source alongside contrasting proximal and distal responses. The work connects archaeological evidence with longer population trajectories; it does not directly translate ancient outcomes into modern evacuation or recovery estimates.
21. Advancing subseasonal forecasting with machine learning
Core Problem: Systematic forecast biases limit useful information two to six weeks ahead.
Key Innovation: A learned correction of forecast distributions improves dynamical and AI ensembles, including real-time competition tests. This supports extended-range preparedness research; landslide-warning performance needs independent local validation.
22. MajorTOM:Elliot Pretrain, A Multi-Sensor, Multi-Temporal Pretraining Dataset for Earth Observation
Core Problem: Generalist observation models need spatially aligned sensors, terrain and multiple temporal sampling regimes in an accessible training resource.
Key Innovation: MajorTOM:Elliot Pretrain releases 279,166 tiles across single-date, monthly and burst acquisitions, with optical, radar, DEM, land-cover and contextual layers. Short OlmoEarth experiments examine which inputs help transfer. The resource has clear hazard-mapping potential, but its benchmark gains do not establish landslide-specific accuracy.
23. EarthAlign: Learning Language-Aligned Representations of Earth System Data
Core Problem: Hazard records and meteorological fields need compatible representations for interpreting event conditions across places and times.
Key Innovation: EarthAlign links ERA5 fields with event-grounded language, including COOLR landslide records among its ten event sources. MeteoLIP supports retrieval and condition interpretation. Its single-snapshot input and uneven source coverage constrain use for local triggering analysis; it does not establish landslide forecast skill.
24. TartanMatch: Towards Universal Dense Correspondence Across Modalities
Core Problem: Dense matching across heterogeneous sensors usually requires separately trained modality pairs, limiting reuse of geometrical evidence.
Key Innovation: TartanMatch handles all 25 ordered pairs among RGB, thermal, event, depth and projected LiDAR inputs in one model. Reported zero-shot matching gains make it a substantial candidate for multi-sensor terrain workflows, but landslide mapping, sensor calibration and geohazard decision performance remain untested.
25. UnSAMv2: Self-Supervised Learning Enables Segment Anything at Any Granularity
Core Problem: A single visual prompt can refer to objects at several scales, making controllable segmentation difficult without dense supervision.
Key Innovation: UnSAMv2 discovers mask-granularity pairs and adds continuous granularity control to SAM-2 using 6000 unlabeled images. Broad image and video benchmarks support a substantial new control capability with potential for terrain delineation, while landslide boundaries and geological scale consistency require direct testing.
26. The Last Remote Sensing Model Built by Humans? Casting Any Earth Observation Task as Recursive Self-Improvement
Core Problem: Disaster-response teams need adaptable damage models when local training labels and model-development expertise are scarce.
Key Innovation: RSI4EO includes optical–SAR building-damage assessment on BRIGHT, reaching 71.6% hidden-test mIoU with full training labels and 49.4% with 14 training pairs. Separate labeled verifier data remain necessary, and searches use up to four GB200 GPUs for 12 hours. The experiment supports automated model development rather than label-free operational deployment.
27. A Generalized Analytical Model for Marine Gas Hydrate Stability Zones With Thermodynamically Constrained Pressure–Temperature Phase Boundary
Core Problem: Idealized hydrate phase boundaries can misplace natural marine stability zones used in assessments of dissociation and overpressure.
Key Innovation: A field-calibrated, thermodynamically constrained phase boundary yields a closed-form stability-zone depth solution and is checked against global observations. It improves characterization of hydrate stability; slope failure and tsunami consequences are not directly simulated.
28. Probabilistic Forecasting of Wildfire Size Using Conditional Flow Matching
Core Problem: Fast estimates of fire-area growth need uncertainty information without the cost of repeated coupled atmosphere-fire simulations.
Key Innovation: Conditional flow matching predicts three-hour acreage increments from fire state, weather, terrain and fuels. Training uses 140 WRF-SFIRE simulations and multi-step testing includes 12 held-out cases. The output describes total area rather than a spatial fire perimeter, and validation remains against simulated fires.
29. A Decade of PhotoMappers: A Longitudinal Study of Crowdsourced Disaster Photos Geolocalization with Vision-Language Models and Geospatial Reasoning
Core Problem: Ground photographs can supply timely disaster evidence, but uncertain location and uneven collection reduce their operational value.
Key Innovation: A decade-scale PhotoMappers analysis combines volunteered imagery, street views and remote sensing. Retrieval reaches 40.90% top-1 accuracy overall, with substantial variation by year. The results support candidate-location assistance and continued human interpretation, rather than automatic geolocation of every observation.
30. When Satellite and Street Views Disagree: Cross-View Complementarity in Disaster Damage Assessment
Core Problem: Roof and facade observations can disagree, and average fusion accuracy hides whether complementary damage evidence is used.
Key Innovation: The study isolates satellite/street-view conflicts in Eaton wildfire and hurricane datasets. Visibility-aware gating improves the wildfire conflict subset, whereas gains do not consistently extend to panoramic hurricane imagery. The findings identify capture-dependent limits of evidence fusion rather than a universally superior combination rule.
31. Earthquake Dynamics as a Benchmark for Scale-Invariant Behavior
Core Problem: A scale-free exponent alone provides an incomplete test of whether a synthetic event process resembles earthquake energy release.
Key Innovation: Southern California and Japan catalog statistics are compared with synthetic distributions under controlled activity. The analysis narrows earthquake-like exponent ranges under an additional energy-budget criterion. It supplies a statistical benchmark, without estimating the time or location of future earthquakes.
32. Seabirds and native vegetation drive ecosystem processes that support atoll island resilience
Core Problem: Claims that island ecosystem restoration improves physical resilience need mechanisms linking ecology to sediment supply and stabilization.
Key Innovation: Observations and geo-ecological analysis connect seabirds and native vegetation to productivity, reef accretion, sediment production and root development. Restoration could support island maintenance under the evaluated warming scenario, while realized island survival remains conditional on multiple physical and ecological processes.
33. Hydrodynamic simulation and risk evolution of urban waterlogging based on integrated SWMM–LISFLOOD modeling: a case study of Dongxihu District, Wuhan
Core Problem: Drainage-network overload and surface-water routing must be evaluated together to locate urban inundation hotspots.
Key Innovation: A linked SWMM–LISFLOOD framework tests a 20-year design storm in Wuhan's Dongxihu District and identifies overflow nodes and affected road sections. Without continuous depth and hydraulic observations, the contribution is scenario-based hotspot assessment rather than fully field-validated forecasting.
34. Seismic – Gravimetric analysis of the Algeciras fault system (Colombia): spatio-temporal correlations from GRACE and GNSS-InSAR
Core Problem: Gravity changes, deformation and seismicity are difficult to interpret together when their spatial scales and time responses differ.
Key Innovation: Hydrologically corrected GRACE observations are compared with GNSS, InSAR and seismic records along the Algeciras Fault System. Variable correlations and lags motivate further integrated investigation. The associations do not establish earthquake precursors, causation or forecast skill.
35. Explainable XGBoost–SHAP Modeling Coupled with GeoDetector for Forest-Fire Occurrence Susceptibility and Future Climate Sensitivity in the Dabie Mountains
Core Problem: Fire-occurrence models need interpretable regional screening while distinguishing relative susceptibility from event probability.
Key Innovation: XGBoost, SHAP and GeoDetector are applied to Dabie Mountains fire records and environmental factors, with separate climate-sensitivity scenarios. Reported AUC is 0.850 for the main model. Predictor attribution is model-dependent, and future score redistribution is not a deterministic fire forecast.
36. Initial Experimental Demonstration of Microwave Imaging for Mapping Leakage-Induced Subsurface Anomalies
Core Problem: Non-contact mapping of leak-related wet zones and cavities needs validation under controlled target and soil conditions.
Key Innovation: A broadband microwave system and back-projection imaging distinguish wet-sand, pothole and cavity surrogates in laboratory measurements. The work establishes initial imaging feasibility, without demonstrating actual buried-leak detection or performance across representative field covers and soils.
37. Physics-Constrained Transfer Learning Uncovers Mineralogical Controls on Shale Caprock Integrity During CO2 Storage
Core Problem: Limited shale samples constrain early assessment of caprock mechanical changes during carbon storage.
Key Innovation: A network pretrained on synthetic mineralogical data is adjusted with experimental measurements to estimate mechanical properties and fracture pressure. An Alberta application suggests limited matrix-scale change after short brine exposure; faulted or fractured field-scale containment and longer-term effects remain unresolved.
38. Integrating rainfall event and antecedent depths into a single probabilistic framework for deriving event depth-duration-antecedent depth-frequency curves
Core Problem: Conventional design storms can omit antecedent rainfall that changes runoff response at a given return period.
Key Innovation: A copula-based framework derives joint event-depth, duration and antecedent-depth frequency curves using three European rainfall records. Equiprobable combinations can be tested in hydrological design models; event-return-period characterization does not alone validate flood peaks or landslide thresholds.
Further Reading
- A national atlas of design-wave and water-level conditions for Turkish coastal and offshore structuresMarine Georesources & Geotechnology
- Non-typhoon extreme waves in the Northeast Asian Marginal Seas: Observation-based identification, synoptic conditions, and classificationOcean Engineering
- Evaluating multi-satellite VIIRS active fire observations as a burned area proxy across North AmericaRemote Sensing of Environment
- Study on thermal stability of a novel double-layer ventilated wide expressway embankment in permafrost regionsCold Regions Science and Technology
- Failure mechanism and stability control of roadways in weakly cemented soft rock strata supported by a closed structure consisting of mesh-reinforced shotcrete and inverted archTunnelling and Underground Space Technology
- Material-dependent interlocking mechanisms and AE precursor evolution of locked-patches instability in filled double-flawed sandstoneInternational Journal of Rock Mechanics and Mining Sciences
- Assessing multi-flood hazard zones for stormwater infrastructure in highly urbanized coastal citiesJournal of Hydrology
- Bivariate regional frequency analysis of extreme rainfall depth and intensity: comparing annual maxima and r-Largest approaches with simulated annealing regionalization across South KoreaJournal of Hydrology
- Nonlinear pipe–soil–liner interaction of corroded steel pipelines rehabilitated with FRP liners under differential ground settlementSoil Dynamics and Earthquake Engineering
- Period-dependent behavior of conditional dispersion in probabilistic seismic demand modelsSoil Dynamics and Earthquake Engineering
- Monotonicity-constrained machine learning for deformation prediction and stability-informed design of underground powerhouse cavernsTransportation Geotechnics