TerraMosaic Daily Digest: July 18, 2026
Daily Summary
Three landslide studies resolve different links in the chain from initiation to consequence. Field, UAV and laboratory evidence from the Chinese Loess Plateau identifies sinkholes as preferential flow and sediment pathways that connect rainfall infiltration to landsliding and erosion-amplified mudflow. A probabilistic framework then replaces runout extent with the probability that spatially variable landslide kinetic energy exceeds structural resistance. At regional scale, comparison of eight explainable machine-learning models shows that high predictive skill does not remove the need to test whether inferred controls remain stable across model families.
Flood modelling advances on both computation and decision. Physics-regularized operator learning predicts depth and velocity fields on a 91,959-cell unstructured mesh in seconds, while a momentum-constrained multi-task model preserves depth-velocity coupling under drainage-sedimentation scenarios and accelerates simulation by 113-277 times. A probabilistic warning framework converts forecast uncertainty and stakeholder risk tolerance into expected loss, and companion studies show that storm rotation can raise probable maximum precipitation by more than 25% and that climate-demographic interaction disproportionately increases future flood exposure among older adults.
Across sensing and geomechanics, uncertainty is treated as part of the physical inference. Bayesian inversion of InSAR and GPS deformation resolves lateral aquifer permeability and finds that one InSAR map can be more informative than multiple GPS time series in the Nevada test. Satellite and historical imagery recover changing glacier and ice-shelf dynamics, whereas tunnel and rockburst studies quantify how saturated faults and mechano-electrical signals alter seismic amplification and warning sensitivity. Spatial support, repeat frequency and physical coupling therefore determine which permeability, deformation and failure mechanisms can be resolved.
Key Trends
Five linked shifts connect process chains, impact intensity, physical constraints, decision theory and observation scale.
- Hazard chains are represented as connected processes: Sinkhole formation, preferential flow, landslide initiation, sediment entrainment and mudflow amplification are analysed within one evolutionary sequence.
- Landslide consequence moves beyond runout footprint: Spatial kinetic-energy distributions are compared directly with structural resistance, turning impact intensity and exposure tolerance into a probabilistic risk measure.
- Physical constraints discipline flood surrogates: Shallow-water residuals, momentum coupling and unstructured meshes reduce non-physical leakage while retaining the speed needed for scenario screening.
- Warnings and design thresholds become decision variables: Forecast uncertainty, risk preference, storm rotation and demographic structure are propagated into warning cost, design precipitation and exposed population.
- Observation scale is tested rather than assumed: InSAR versus GPS information content, DEM resolution, altimetry geometry and probabilistic soil classes reveal how measurement support changes the inferred environmental signal.
Selected Papers
The papers below connect process evidence to quantities that can change a decision: hazard-chain stage, impact energy, evacuation completion, warning loss, flood depth, seismic demand and observation uncertainty.
1. Development characteristics and hazard mode of loess sinkhole-landslide-mudflow hazard chain: A typical case study of loess tableland
Core Problem: Loess sinkholes are commonly mapped as isolated features even though intense rainfall can connect them to landslides and mudflows at catchment scale.
Key Innovation: Field surveys, UAV mapping and laboratory tests identify sinkholes as preferential flow and sediment pathways, show how connected sinkholes trigger slope failure, and organize the resulting sinkhole-landslide-mudflow chain into six evolutionary stages.
2. A probabilistic approach for landslide risk assessment: Integrating kinetic energy and structural resistance
Core Problem: Runout-distance maps describe where a landslide may reach but not whether its impact energy exceeds the resistance of exposed structures.
Key Innovation: A random material point framework propagates spatial shear-strength variability into landslide kinetic energy and defines risk as the probability that this energy exceeds structural resistance, reducing the overestimation produced by footprint-only mapping.
3. Comparative Landslide Susceptibility Mapping in Longchuan, Guangdong Province, China, Using Explainable Machine Learning
Core Problem: High landslide-susceptibility accuracy can coexist with unstable explanations of which environmental factors control the mapped pattern.
Key Innovation: Eight tuned models are compared on 363 landslides and matched non-landslides; all exceed 0.938 test AUC, while SHAP analysis identifies water index, relief and river distance as the most stable controls and GBDT as the best-performing model at 0.952 AUC.
4. Advancing Aquifer Characterization Through the Integration of Satellite Geodesy, Geomechanics, and Bayesian Inference
Core Problem: Aquifer models remain weakly constrained where permeability varies laterally and sparse borehole observations cannot resolve that heterogeneity.
Key Innovation: Bayesian poroelastic inversion combines InSAR and GPS deformation from a Nevada pumping test to map permeability and its uncertainty, finding that one spatial InSAR map carries substantially more information than several GPS time series at this site.
5. Enhancing tsunami evacuation planning through spatial and behavioral analyses: A case study of Cilacap, Indonesia
Core Problem: Spatially feasible tsunami routes do not guarantee that residents will recognize shelters, choose them or reach them under short warning times.
Key Innovation: Network analysis and an 82-resident survey show that completion falls to 60.2% at 20 minutes and 32.5% at 10 minutes, then translate destination choice, vehicle use, perceived proximity and shelter recognition into neighborhood-level planning actions.
6. Physics-informed operator learning for rapid hydrograph-to-field prediction on an unstructured-mesh flood benchmark
Core Problem: High-resolution two-dimensional flood solvers are too slow for screening many inflow hydrographs, while unconstrained surrogates can leak water across wet-dry fronts.
Key Innovation: A shallow-water-regularized DeepONet maps hydrographs to depth and velocity on a 91,959-cell unstructured mesh, improves extreme-event routing and volume conservation, and reduces online evaluation from full simulation to seconds.
7. Physics-guided multi-task learning for urban flood prediction under drainage sedimentation scenarios
Core Problem: Urban-flood surrogates that predict depth and velocity independently can violate their hydrodynamic coupling, especially when sediment reduces drainage capacity.
Key Innovation: A multi-gate mixture-of-experts model regularized by momentum constraints jointly predicts maximum depth and velocity, generalizes to unseen rainfall-sedimentation combinations and runs 113.4-277.0 times faster than the physical model.
8. A decision-theoretic framework for probabilistic flood warnings
Core Problem: Operational flood thresholds rarely express the economic consequences of false alarms, missed events or different stakeholder tolerances for risk.
Key Innovation: A Bayesian decision framework combines probabilistic precipitation, post-processed uncertainty and multi-level loss functions in Italy's Sieve basin, increasing relative economic value while reducing false alarms and expected losses.
9. Micromechanical and electrical properties of coal during loading-induced failure and charge induction monitoring: Implications for rockburst early-warning
Core Problem: Microseismic and acoustic-emission monitoring can identify coal failure too late or incompletely for sensitive rockburst warning.
Key Innovation: Multiscale laboratory and field tests show that induced-charge signals respond about six seconds before acoustic emission, track ejection energy and stress drop, and delineate abutment-pressure and plastic zones in the mine.
10. Assessing Precipitation Effectiveness in the Context of Drought
Core Problem: Total rainfall can misrepresent drought relief because intense events may generate runoff rather than recharge root-zone soil moisture.
Key Innovation: A random-forest estimate of precipitation effectiveness reproduces event-scale soil-moisture recharge at three Illinois sites and provides information on drought improvement or deterioration that precipitation totals alone do not capture.
11. Improving FY-4B Satellite Precipitation Retrieval over Coastal Complex Terrain of Eastern China: Deep Learning Approaches with Multi-Source Underlying Surface Data
Core Problem: Satellite precipitation retrieval over coastal mountains is degraded when cloud-top observations are interpreted without the underlying terrain and land surface.
Key Innovation: Four deep networks combine FY-4B infrared data with elevation, slope, roughness and land cover; lightweight models better exploit discrete land-cover boundaries, deeper models capture continuous terrain gradients, and multi-factor fusion improves detection against gauges and IMERG.
12. Seismic response of tunnels adjacent to saturated fault zones under SV-wave: Effects of fluid-solid coupling and fault properties
Core Problem: The seismic demand on a tunnel near a saturated fault depends on poroelastic coupling and on whether the fault amplifies or shields the incident wave.
Key Innovation: A Biot-poroelastic analytical solution, validated against ABAQUS and recorded motions, predicts up to 65.8% amplification when tunnel and wave share a fault side and up to 24.7% reduction when the fault lies between them.
13. Re-evaluating probable maximum precipitation estimates: sensitivity to transposition domains and storm rotation using modern datasets
Core Problem: Probable maximum precipitation is treated as a design bound even though storm selection, transposition and rotation remain weakly standardized.
Key Innovation: A modern NOAA gridded-data analysis for Iowa shows that PMP is strongly assumption-dependent, can exceed the legacy spillway-design value, and rises by more than 25% from storm rotation alone.
14. Hindcasting compound coastal-inland flood events caused by post-tropical storms
Core Problem: Compound coastal and inland flood reconstructions lose hydraulic connectivity when observed high-water marks and small drainage structures are omitted from terrain models.
Key Innovation: A unified HEC-RAS 2D hindcast of post-tropical storms Fiona and Dorian integrates rainfall, coastal marks and a breached high-resolution DEM, identifies 2,895 hydraulic structures and resolves contrasting island-wide inundation patterns.
15. Projecting flood exposure in China under future climate and population change: an age-structured analysis based on the SSP-RCP scenarios
Core Problem: Future flood-exposure estimates based on total population conceal age-specific vulnerability and interactions between demographic and climate change.
Key Innovation: Coupled climate, hydrodynamic and age-structured projections indicate 2.9-fold overall exposure growth in China by 2100 and up to 6.9-fold growth for older adults, with climate-population interaction explaining as much as 53% of the elderly increase.
16. Seismic isolation of shallow footings using tire derived aggregate
Core Problem: Geotechnical seismic isolation with tire-derived aggregate lacks broad dynamic evidence on rocking, recentering and residual deformation beneath shallow footings.
Key Innovation: Shake-table tests show strong hysteretic damping, negligible residual shear, small settlement and a 2.15-fold period lengthening, while also identifying structural mass and resonance as more consequential than aggregate-layer thickness alone.
17. Three Decades of Glacial Changes on the Western Antarctic Peninsula Revealed by Historical Aerial and High-Resolution Satellite Imagery
Core Problem: Mass-balance records before 2000 are sparse for climate-sensitive glaciers on the western Antarctic Peninsula.
Key Innovation: Historical aerial photographs and high-resolution satellite imagery reconstruct 203 glaciers from 1989 to 2020, revealing retreat-driven mass loss and a post-2016 shift toward faster frontal thinning and broader surface lowering.
18. Small Tropical Islands Also Exposed to Extreme Humid Heat by the End of the Century
Core Problem: Coarse global climate models do not resolve small tropical islands, leaving their future humid-heat exposure poorly quantified.
Key Innovation: Bias-corrected model output at island weather stations projects dangerous heat-index levels and longer humid heatwaves by century end, with the fastest intensification toward the equator.
19. SWOT High-Rate Raster Data Reveals Antarctic Ice Shelf Motion and Change in Response to Ocean and Ice Dynamics
Core Problem: Ice-shelf thickness, flexure and crevasse growth change at scales poorly sampled by conventional repeat altimetry.
Key Innovation: Repeated SWOT high-rate swaths, cross-checked against ICESat-2, resolve tidal and atmospheric height response, grounding-zone flexure and crevasse deepening across dynamic Antarctic ice shelves.
20. A Multi-Method Approach to the Analysis of Trends and Cyclical Variability in Sea Level Along the Southern Baltic Coast
Core Problem: Sea-level trends derived from tide gauges and satellite altimetry diverge with record length, coastal distance and nonstationary cycles.
Key Innovation: Harmonic and wavelet analysis raises cross-platform trend correlation from 0.72-0.80 to 0.92-0.95 and separates a roughly 2 mm/yr coastal-gauge trend from a 4.3 mm/yr altimetric estimate offshore.
21. From abandonment to gratitude: Unmet needs and perceived support after the 2024 Valencia DANA flash flood
Core Problem: Post-flood assessments often separate utility disruption from psychosocial recovery even when loss of electricity and communication shapes perceived abandonment.
Key Innovation: A seven-month mixed-method survey of 300 Valencia flood survivors traces needs across nine time points and shows that basic relief and utility failures dominate abandonment in the first 72 hours, followed by legal and administrative deficits.
22. Learning from a 43-year experience of post-disaster relocation in Anaa Atoll, French Polynesia
Core Problem: Planned relocation is frequently evaluated as a one-time move rather than as a process that can recreate exposure and land insecurity over decades.
Key Innovation: A 43-year study of Anaa Atoll shows that initial flood-risk reduction eroded as unsafe areas were resettled, protective housing deteriorated and land conflict produced new vulnerability and maladaptation.
23. Assessing the availability of emergency shelters under Natech scenarios: A coupled accessibility-functionality framework
Core Problem: Emergency-shelter plans under earthquake-triggered technological accidents can appear adequate when reachability is assessed without network failure, exclusion zones or demand overload.
Key Innovation: The N-SAAF framework couples probabilistic hazard simulation with risk-aware two-step catchment analysis, exposing nonlinear shelter unavailability caused by severed links and capacity saturation under delayed Natech escalation.
24. Spatiotemporal changes of soil profile salinization in Xinjiang and their response to glacial runoff variations
Core Problem: The response of root-zone salinity to glacier retreat is poorly resolved because runoff change and soil controls vary among basins and with depth.
Key Innovation: Random-forest soil profiles, OGGM runoff reconstructions and Geodetector interactions reveal increasing salinization, basin-specific meltwater histories and a strengthening runoff influence in the 60-100 cm layer across Xinjiang.
25. Nonlinear response of soil erosion to rainfall erosivity: effects of vegetation cover and rainfall intensity
Core Problem: Widely used erosion models assume soil loss rises linearly with rainfall erosivity despite vegetation-dependent thresholds during extreme rain.
Key Innovation: Field monitoring and rainfall experiments reveal nonlinear inflection points under crops, forests and orchards; 40% vegetation cover cuts erosion by 78.5% at 50 mm/hr, but protection weakens as rainfall intensity increases.
26. Data-driven simultaneous accounting of diverse agronomic and eco-hydro-climatological stressors for guiding targeted watershed restoration strategies
Core Problem: Watershed-restoration priorities are distorted when hydroclimatic, ecological, agronomic and terrain stressors are ranked separately or protective indicators cancel hazardous ones.
Key Innovation: Satellite observations and process simulations feed a sign-consistent, entropy-weighted two-stage vulnerability index across 98 sub-watersheds, preserving domain interpretation while exposing persistent and time-sensitive degradation regimes.
27. Hydrological effectiveness of multiple natural flood management interventions in rapid responses catchments
Core Problem: Evidence for natural flood management in small, fast-response catchments is limited by short records and weak counterfactual designs.
Key Innovation: High-frequency monitoring across five English catchments and 662 events finds 20-80% peak-flow reduction and 7-50% longer lag times, with performance governed by hydraulic connectivity, antecedent moisture and intervention type.
28. Residual Magnetization Accumulation Enables Rapid Pre-Polarized Surface NMR Groundwater Detection
Core Problem: Surface NMR detects groundwater directly, but conventional pre-polarization improves weak signals at the cost of long acquisition cycles.
Key Innovation: Repeated short pulses exploit residual magnetization to retain nearly conventional signal amplitude while reducing the excitation cycle from about 10 seconds to under one second in simulation and field tests.
29. Least-Resistance Path as a Proxy for Efficient Simulation of Solute Transport in Heterogeneous Porous Media
Core Problem: First-arrival prediction in heterogeneous aquifers normally requires computationally expensive particle tracking through uncertain conductivity fields.
Key Innovation: A least-resistance path derived from the flow field approximates the fastest transport path and its arrival time while reducing computational cost by more than an order of magnitude.
30. Revealing scale-dependent controls on possible sunshine duration in mountains: A machine learning simulation of topographic nonlinearities
Core Problem: DEM resolution changes mountain solar-exposure estimates, but linear scale analyses cannot explain the resulting nonlinear topographic controls.
Key Innovation: Bayesian-tuned XGBoost with SHAP shows that dominant controls shift from aspect and local relief at 30 m toward terrain undulation and curvature on coarser grids, with strong seasonal differences.
31. CPTU-Informed Unified Earth Pressure Model for Confined Excavations
Core Problem: Earth pressures in confined excavations cannot be estimated reliably without the interacting effects of wall movement, soil arching, narrow spacing and excavation-induced soil changes.
Key Innovation: A CPTU-informed unified model incorporates all four effects, reproduces model tests and monitored field pressures, and identifies increased overconsolidation and earth-pressure coefficients below the excavation base.
32. Resilience-oriented preventive dispatch for urban backbone power grid with high renewable penetration under extreme heat
Core Problem: Extreme heat simultaneously raises electricity demand and reduces generation and network capacity, but preventive dispatch rarely represents meteorological uncertainty across transmission and distribution systems together.
Key Innovation: A two-stage distributionally robust model coordinates the urban backbone grid, searches worst-case contingencies and reduces overload, renewable curtailment and load shedding under uncertain high-temperature conditions.
33. Bridging soil mapping units and digital soil mapping through probability-driven machine learning: an uncertainty-aware framework for soil class prediction in Antarctica
Core Problem: Deterministic Antarctic soil maps are difficult to reproduce and provide no direct representation of class uncertainty in sparse observations.
Key Innovation: Probability-based random forests combine legacy profiles, terrain and multispectral data into ranked class probabilities, then reconcile those probabilities with mapping units to preserve plausible class coexistence in remote ice-free terrain.
34. Machine-learning quantification of provenance of aeolian sediments using geochemical fingerprinting: A case study from the Gurbantunggut Desert, northwestern China
Core Problem: Aeolian-source attribution from high-dimensional geochemistry is difficult when several source regions contribute overlapping compositional signatures.
Key Innovation: Compositional preprocessing, recursive feature selection and three tree regressors trained on synthetic mixtures quantify source fractions that agree with independent Bayesian and frequentist unmixing for the Gurbantunggut Desert.
35. Mechanisms of fracture swarm formation in shale cores from hydraulic fracturing test sites: an experimental investigation
Core Problem: Closely spaced hydraulic-fracture swarms occur far more often than perforation spacing predicts, but the role of bedding and lithologic interfaces is unresolved.
Key Innovation: True-triaxial tests with CT, pressure and acoustic-emission monitoring show how interface contrast controls crossing, capture, deflection and secondary re-initiation, defining activation conditions for swarm development.
36. TCE redistribution and inferred entrapment in loess under groundwater table fluctuation patterns
Core Problem: Groundwater-table fluctuations redistribute volatile contaminants in collapsible loess, yet conventional transport assessments neglect how moisture-induced structural collapse changes physical entrapment.
Key Innovation: Controlled soil-column experiments separate amplitude and velocity effects on moisture and TCE transfer, then link metastable-pore collapse to vapor entrapment, surface release and secondary contamination risk.
37. A holistic index for hydrological characterization of catchments
Core Problem: Conventional hydrological indicators isolate runoff, evaporation or storage and therefore miss long-term changes in whole-catchment runoff restriction.
Key Innovation: A normalized runoff-restricting potential index integrates rainfall, storage, evapotranspiration and runoff, classifies 406 Indian catchments and identifies significant 1980-2020 trends in 93 of them.
38. A procedure for geostatistical inversion incorporating a high-resolution initial hydraulic-conductivity field from travel time inversion
Core Problem: Geostatistical hydraulic tomography can over-smooth conductivity where observations are sparse and geological zonation is unavailable.
Key Innovation: Travel-time inversion supplies a high-resolution initial conductivity field for geostatistical inversion, matching the performance of an accurately zoned model in the Herten aquifer analogue without requiring prior geological boundaries.
39. Ecohydrological trend detection in the Mobile Bay Watershed-AL using a rolling window Mann-Kendall test
Core Problem: Whole-record monotonic tests can hide intervals when low flows, high flows and ecological responses change direction or strength.
Key Innovation: A rolling 30-year Mann-Kendall analysis resolves episodic streamflow shifts across the Mobile Bay watershed and packages the reproducible workflow in the open-source rwMK R library.