TerraMosaic Daily Digest: September 19, 2026
Daily Summary
Geodetic observations sharpen earthquake-hazard models at two active fault systems. Across Hispaniola, joint GNSS-InSAR inversion with internally deformable blocks resolves localized strain on the Enriquillo-Plantain Garden fault but distributed deformation south of the mapped Septentrional trace, implicating subsidiary splays in large-earthquake hazard. Along the Haiyuan fault, a time-dependent GNSS-PS-InSAR inversion recovers pre-2022 Menyuan locking to roughly 30 km depth, slip deficits of 3.1-5.6 mm yr⁻¹ and accumulated moments consistent with the observed Mw 6.7 event. In Algiers, borehole- and downhole-validated HVSR estimates place sediment thickness within 10% of ground truth in 85% of co-located cases, extending site characterization where subsurface data are sparse.
Hydrological studies resolve environmental memory and its uncertainty rather than treating forcing products as interchangeable. In the Vakhsh basin, cross-wavelet analysis and CatBoost-SHAP attribute seasonal runoff components to snow, vegetation and climate variables, with wet-season response lags within 12 days. Across four Dongting Lake sub-basins, a daily, grid-based non-stationary drought framework estimates mean recovery times of 83.5-97.3 days and recovery probabilities below 0.61 for extreme drought, with human influence dominant in three basins. Over the Tibetan Plateau, Optuna-CatBoost fusion raises correlations for latent and sensible heat flux from 0.74 to 0.85 and 0.52 to 0.70, respectively, while a title-level global study targets propagation from precipitation extremes to root-zone soil moisture.
Remote-sensing methods increasingly combine reusable representations with geometry-aware reconstruction. A spherical dual-layer equivalent-source formulation separates magnetic wavelengths and scales to more than 1.5 million observations in an open-source implementation. Google Satellite Embeddings support annual 10 m land-cover maps for 37 Chinese metropolises, while airborne LiDAR-RGB fusion resolves farm-scale rock outcrops. Prompt-guided visible-thermal detection, frequency-aware hyperspectral classification and multi-level pansharpening address small targets, spectral-spatial continuity and fine structural detail, but their value for geohazards remains transferable rather than demonstrated.
Engineering contributions shift from isolated capacity checks toward coupled processes and resilience. Title-level records introduce energy-based slope design, coupled tunnel seepage-erosion and structural-response analysis, quantitative assessment of shield-tunneling disturbance in sensitive soils, and a nonlocal large-deformation formulation for strain-softening clay. Additional title-level studies address physics-informed river-flow inversion, reclamation-fill segregation and settlement interpretation, and coal-mine methane-plume detection with ground validation. These directions widen the monitoring and modeling toolkit, but their methods, quantitative performance and cross-site transfer remain to be established from full article evidence.
Key Trends
Across the collection, hazard inference improves when spatial coverage, environmental memory and model uncertainty are treated as first-order components rather than residual complications.
- Multi-sensor geodesy resolves concealed fault behavior: Joint GNSS-InSAR inversions recover distributed strain, fault locking and slip deficit that sparse station networks or a single mapped trace cannot represent.
- Validation is moving closer to the subsurface quantity of interest: HVSR thicknesses are tested against boreholes and downhole velocities, while field and ground observations constrain rock-outcrop maps, methane-plume estimates and regional environmental products.
- Environmental memory becomes measurable: Runoff lags, root-zone soil-moisture propagation and non-stationary drought recovery frame hydroclimatic risk as a time-dependent response rather than an instantaneous forcing relation.
- Reusable representations are entering environmental mapping: Satellite embeddings, prompt-guided multimodal detection and spectral-spatial sequence models reduce dependence on hand-crafted features, although hazard-specific generalization remains untested.
- Geotechnical design is adopting coupled and resilience-aware formulations: Energy-based slope design, seepage-structure coupling, tunneling-disturbance assessment and nonlocal strain-softening models emphasize recovery, interaction and progressive deformation.
Selected Papers
The 19 September collection is led by GNSS-InSAR studies of strain partitioning in Hispaniola and pre-earthquake locking on the Haiyuan fault, together with a daily, grid-based assessment of non-stationary hydrological-drought recovery. Borehole-validated HVSR mapping strengthens urban seismic site characterization, while companion studies address slope resilience, high-mountain runoff memory, Tibetan Plateau heat-flux uncertainty, precipitation-soil-moisture propagation, tunnel and soft-ground deformation, metropolitan land-cover reconstruction, rock-outcrop mapping and transferable multimodal remote-sensing methods.
1. Interseismic Strain Accumulation and Partitioning in Hispaniola From GNSS and InSAR
Core Problem: Sparse and uneven GNSS coverage leaves the distribution of strain and fault slip across Hispaniola incompletely resolved.
Key Innovation: Combines GNSS and InSAR in a block model that allows internal deformation and explicitly evaluates spatially correlated InSAR noise, resolving localized Enriquillo strain and a broader Septentrional deformation zone.
2. Kinematics of the Haiyuan Fault (Northwest China) Prior to the 2022 Menyuan Mw 6.7 Earthquake
Core Problem: Pre-event fault coupling and slip deficit are difficult to resolve where deformation observations are spatially incomplete.
Key Innovation: Uses a time-dependent DEFNODE inversion of GNSS and Sentinel-1 PS-InSAR to estimate locking depth, slip deficit and accumulated moment before the 2022 Menyuan earthquake.
3. Understanding hydrological drought recovery patterns in a changing environment: Insights from a non-stationary multivariate analysis
Core Problem: Quantify hydrological-drought recovery time, required water volume and recovery probability without assuming stationary runoff statistics.
Key Innovation: Integrates a GAMLSS non-stationary runoff index with run theory, copula-Bayesian recovery analysis and random-forest attribution across four Dongting Lake sub-basins.
4. Reliability of HVSR-Based Sediment Thickness Estimation Using Borehole and Downhole Data
Core Problem: Evaluate whether low-cost ambient-vibration HVSR can reliably estimate sediment thickness where borehole coverage is sparse.
Key Innovation: Validates HVSR thickness estimates against co-located boreholes and downhole velocity data, then maps sediment thickness with 66 additional measurements while flagging possible two-dimensional effects.
5. Resilience informed design of slopes: An energy perspective
Core Problem: Title-level focus: design slopes around resilience rather than static capacity alone.
Key Innovation: Title-signalled approach or contribution: The title identifies an energy-based perspective for resilience-informed slope design. Methods, data and results could not be assessed because no reliable abstract was available.
6. Magnetic Dual-Layer Equivalent Sources on the Sphere
Core Problem: Cartesian equivalent-source formulations become geometrically inconsistent when regional or global magnetic surveys are modeled over the curved Earth.
Key Innovation: Reformulates magnetic equivalent sources on the sphere, separates long- and short-wavelength content with dual source layers, and uses gradient boosting to scale inversion beyond 1.5 million observations; the implementation is open source.
7. Generating Annual 10 m Land Cover Maps for 37 Chinese Metropolises Using Google Satellite Embeddings
Core Problem: Complex urban landscapes impede consistent annual high-resolution land-cover mapping across many cities.
Key Innovation: Combines Google Satellite Embeddings, ensemble learning and spatiotemporal post-processing to produce 2017–2024 annual 10 m maps for 37 Chinese metropolises.
8. Temporal Lag and Response Characteristics of Runoff and Its Components in a Glacial Basin
Core Problem: Data scarcity limits attribution of runoff, surface runoff and baseflow variability in glaciated basins.
Key Innovation: Combines nine-method baseflow evaluation, remote sensing, reanalysis, field observations, cross-wavelet analysis and CatBoost-SHAP to quantify seasonal drivers and lags.
9. Investigating tunnel seepage erosion and induced structural mechanical response using CFD-DEM-FEM method
Core Problem: Title-level focus: represent tunnel seepage erosion together with the induced mechanical response of the structure.
Key Innovation: Title-signalled approach or contribution: The title identifies a coupled CFD-DEM-FEM framework. Methods, data and results could not be assessed because no reliable abstract was available.
10. Propagation from precipitation extremes to root-zone soil moisture and attribute changes in dry and wet events: A global paired event analysis
Core Problem: Title-level focus: characterize how precipitation extremes propagate into root-zone soil moisture and alter dry and wet events.
Key Innovation: Title-signalled approach or contribution: The title identifies a global paired-event analysis. Methods, data and results could not be assessed because no reliable abstract was available.
11. Quantitative evaluation of shield tunneling-induced soil disturbance in sensitive soft soil
Core Problem: Title-level focus: quantify soil disturbance caused by shield tunneling in sensitive soft ground.
Key Innovation: Title-signalled approach or contribution: The title identifies a quantitative evaluation framework, but the specific method is unavailable. Methods, data and results could not be assessed because no reliable abstract was available.
12. Improving the Quality of the Surface Heat Flux Data over the Tibetan Plateau by Using an Optuna-CatBoost-Shapley Additive exPlanation Method
Core Problem: Heat-flux products over the data-sparse Tibetan Plateau have inconsistent accuracy and poorly quantified spatial uncertainty.
Key Innovation: Combines Optuna-tuned CatBoost fusion of five reanalyses with SHAP interpretation and generalized three-cornered-hat uncertainty estimation.
13. Rock Detection and Arability Mapping Using Airborne LiDAR-RGB Fusion and Machine Learning
Core Problem: Farm-scale rock-outcrop and arability mapping remains manual despite a need for sub-metre spatial detail.
Key Innovation: Benchmarks heuristic, Random Forest, DeepLabV3+, Mask R-CNN and U-Net approaches on 10 cm LiDAR-RGB composites and integrates rock, tree-cover and slope layers.
14. ProG-Net: Prompted Guidance Network for Visible-Thermal Tiny-Object Detection
Core Problem: Tiny targets lose discriminative information during downsampling while visible and thermal streams have unequal feature distributions.
Key Innovation: Uses a frozen vision foundation model, an auxiliary point-prompt branch and prompt-guided cross-modal fusion to preserve target-focused spatial priors.
15. MSFFusion: Multi-Level Shallow Feature Fusion Network for Pansharpening
Core Problem: Deep pansharpening models often underuse shallow features, weakening spatial-detail reconstruction and spectral fidelity.
Key Innovation: Introduces memory feature supplementation, adaptive gating, cross-scale aggregation and progressive multi-level fusion in MSFFusion.
16. FCD-Mamba: A Frequency-Enhanced and Center-Pixel-Guided Dual-Branch Mamba Network for Hyperspectral Image Classification
Core Problem: Mamba hyperspectral classifiers insufficiently preserve shallow frequency cues, spatial-neighborhood continuity and center-pixel semantics.
Key Innovation: Combines Fourier and DCT frequency enhancement, Hilbert-curve spatial serialization, center-preserving spectral scans and guided dual-branch fusion.
17. Adaptive background suppression-based plume identification, emission estimation, and ground validation of coal mine methane in Changzhi, Shanxi, China
Core Problem: Title-level focus: identify coal-mine methane plumes against variable backgrounds and estimate their emissions.
Key Innovation: Title-signalled approach or contribution: The title indicates adaptive background suppression coupled with plume identification, emission estimation and ground validation. Methods, data and results could not be assessed because no reliable abstract was available.
18. Physics-informed neural networks for hydrodynamic inversion and discharge estimation in river flows over complex bed topography
Core Problem: Title-level focus: infer hydrodynamic states and estimate river discharge over complex bed topography.
Key Innovation: Title-signalled approach or contribution: The title identifies a physics-informed neural-network formulation. Methods, data and results could not be assessed because no reliable abstract was available.
19. Segregation and self-weight consolidation of dredged fills: A site-specific laboratory-field interpretation framework for estimating segregation distance in the Saemangeum reclamation area
Core Problem: Title-level focus: estimate segregation distance and consolidation behavior of dredged fills at Saemangeum.
Key Innovation: Title-signalled approach or contribution: The title identifies a site-specific laboratory-field interpretation framework. Methods, data and results could not be assessed because no reliable abstract was available.
20. An incrementally updated nonlocal RITSS framework for strain-softening clay with application to cyclic caisson penetration
Core Problem: Title-level focus: model strain-softening clay during cyclic caisson penetration without pathological localization.
Key Innovation: Title-signalled approach or contribution: The title identifies an incrementally updated nonlocal RITSS framework. Methods, data and results could not be assessed because no reliable abstract was available.
21. Cross-Regional Classification of Rice Cropping Systems Based on Within-Year Seasonal Composition Using Multimodal Remote-Sensing Time Series
Core Problem: Cropping-intensity products do not distinguish the within-year seasonal combinations that define regional rice production systems.
Key Innovation: Integrates SAR-assisted optical reconstruction, phenology-constrained time-series splicing and an Ordered-Cycle Query Transformer for cross-regional classification.
22. Motion-Guided Multi-Offset Detector-Native ReID Readout for Efficient UAV Multi-Object Tracking
Core Problem: Detector-center identity readout becomes unstable in UAV video under small targets, localization jitter and camera motion.
Key Innovation: Samples detector-native identity features at multiple offsets tied to compensated track predictions and conditions the readout on a motion prior.
23. The Five-Decade Landsat Legacy in Inland Water Quality: A Systematic Review of Technological and Methodological Evolution
Core Problem: The evolution, transferability limits and operational maturity of Landsat-based inland-water-quality retrieval have not been synthesized consistently.
Key Innovation: Synthesizes 229 studies from 1983–2026 and proposes a Landsat-to-Global Freshwater Intelligence Framework.
24. Mitigating cascading failures through intentional line removal: A conditional reinforcement learning approach
Core Problem: Title-level focus: mitigate cascading failures in a network through controlled line removal.
Key Innovation: Title-signalled approach or contribution: The title identifies conditional reinforcement learning for intentional line-removal decisions. Methods, data and results could not be assessed because no reliable abstract was available.
25. A physics-informed, soft-causal ensemble deep neural network for robust probabilistic water quality assessment in arid regions
Core Problem: Title-level focus: produce robust probabilistic water-quality assessments in arid regions.
Key Innovation: Title-signalled approach or contribution: The title identifies a physics-informed, soft-causal ensemble deep neural network. Methods, data and results could not be assessed because no reliable abstract was available.