TerraMosaic Daily Digest: August 16, 2026
Daily Summary
Direct geohazard papers resolve instability as an evolving state rather than a static susceptibility surface. National and basin-scale landslide studies couple physically based rainfall and seismic forcing with probabilistic formulations, showing that sequential rainfall-earthquake scenarios, time-varying susceptibility, and nonvertical wave incidence can enlarge or redistribute failure likelihood beyond single-trigger assumptions. Case studies of karst collapse, leak-driven urban slopes, rotational earthflows, granular-flow bifurcation, fault-zone creep, blast damage, thermal cycling, and drying-wetting cracking extend this process view, while earthquake rupture energetics, seismic vulnerability mapping, and tailings-facility screening push the same logic toward source physics and infrastructure prioritization.
Flood and hydroclimatic papers similarly move from bulk totals to structured, connected, and socially mediated hazards. Studies of riparian vegetation disturbance, dual-drainage inundation, physics-informed flood emulation, nonproportional sample selection, transfer learning in transboundary basins, and rainfall-event typologies all emphasize flow-path representation, antecedent conditions, and data-efficient generalization. Regional assessments of tidal inundation, Shanghai road exposure, long-horizon urban flood risk under climate change, heavy-precipitation structure, hydrological drought, weather whiplash, and paleoflood detection together show that coastal and inland risk depends on compound forcing, spatial heterogeneity, and event timing, not on intensity metrics alone.
A parallel methodological stream strengthens observation and inference while remaining domain-specific in its validation. Deformation-monitoring advances include robust InSAR phase restoration, GPU-scaled denoising, D-GBSAR repositioning correction, TomoSAR densification, land-subsidence typing, soil-moisture and precipitation-product evaluation, and hyperspectral or SAR sensing of thaw slumps, volcanic sulfate, and polar lows. More general remote-sensing and scientific-ML papers converge on multimodal fusion, boundary-preserving change detection, domain adaptation, self-supervised representation learning, uncertainty-aware bathymetry, and physics-structured surrogates; these are enabling advances for hazard science, not blanket proof of performance across geohazards.
Key Trends
The August 16 selection converges on five trajectories: compound-trigger probability, state-aware failure mechanics, denser operational sensing, structure-preserving Earth observation, and more physics-constrained scientific machine learning.
- Compound-Trigger Hazard Models Are Replacing Single-Driver Views: Scenario-based rainfall-earthquake landslide assessment, PSHA-coupled susceptibility, chain-level rainstorm-flood networks, and climate-conditioned urban and coastal flood studies all treat hazard as the outcome of interacting forcings rather than isolated triggers.
- Failure Controls Are Being Located in Hidden State Variables and Geometry: Leak-driven saturation, seismic-wave incidence angle, karst post-seismic wetting, vegetation-morphodynamic feedbacks, crack-controlled overland flow, and thermally or mechanically damaged rock masses show that internal state, structure, and routing geometry often govern failure style more strongly than bulk forcing alone.
- Operational Monitoring Is Shifting Toward Dense, Multiscale Observation: UAV repeat surveys, ERT moisture imaging, improved InSAR and D-GBSAR processing, TomoSAR, land-subsidence classification, and evaluation of CYGNSS, GSMaP, and radiance-assimilation products indicate a clear move toward hazard monitoring systems that are faster, denser, and more diagnostic of evolving conditions.
- Transferable Earth-Observation Methods Prioritize Edges, Cross-View Fusion, and Modality-Aware Representation: Cross-view self-supervised change detection, edge-integrity learning, cognitively structured change detection, fine-grained segmentation, multimodal spatial-frequency routing, dehazing, and LiDAR representation learning share a design priority: preserve boundaries and structural consistency when imagery is heterogeneous, weakly labeled, or degraded. These results remain method validations in their tested sensing tasks, not direct geohazard proof.
- Scientific ML Is Becoming More Physics-Constrained and Reliability-Conscious: Physics-informed random forests, generative precipitation downscaling, coupling-robust PINNs, hierarchical mesh surrogates, sharp-interface diffusion models, latent operators, modernized inversion regularization, probabilistic shear-strength modeling, GP hyperparameter critiques, misspecification diagnostics, and ordinal calibration all point toward faster hazard-relevant simulation with stronger attention to uncertainty, transfer limits, and physical coherence.
Selected Papers
The selected papers combine direct studies of landslides, floods, deformation, coastal inundation, and seismic hazard with a second group of transferable sensing and AI methods. Read together, they distinguish papers that revise process understanding or risk estimates from papers that expand the measurement and modeling toolkit but were validated only in their stated domains.
1. Scenario-based physically informed assessment of rainfall- and earthquake-induced landslide failure probability across China
Core Problem: How to estimate scenario-based landslide failure probability across China under rainfall, earthquakes, and sequential triggers.
Key Innovation: Integrates TRIGRS, Newmark, Rosenblueth point estimates, and sequential rainfall-earthquake scenarios in a national probabilistic framework.
2. Modelling Flood-Induced Riparian Vegetation Disturbance and Hydro-Morphodynamic Feedbacks
Core Problem: Single-mechanism vegetation loss rules underrepresent event-scale flood morphodynamic feedbacks that control erosion risk.
Key Innovation: Integrates overturning, breakage, washout, burial, and persistence states inside a validated 2D hydro-morphodynamic flood model.
3. An energy-based perspective on the correlation between stress drop and rupture speed
Core Problem: Explain when earthquake stress drop and rupture speed should correlate positively or negatively.
Key Innovation: Unifies conflicting observations with an energy-balance framework linking fracture energy, stress drop, and rupture speed.
4. Regional modeling of the impacts of tidal flooding in the context of average sea level rise in low-lying areas of Brazil's semi-arid coast
Core Problem: Low-lying semi-arid coasts lack regional estimates of future tidal flooding extent and impacts.
Key Innovation: Regional open-data modeling of sea-level-rise-driven tidal flooding and damages across Brazil's semi-arid coast.
5. Multi-Temporal UAV Observations of Post-Seismic Surface Collapse Evolution Following the 2026 M5.2 Liuzhou Double Earthquake in a Karst Terrain, Guangxi Province, China
Core Problem: Post-seismic collapse growth in karst terrain is poorly resolved in the days after moderate earthquakes.
Key Innovation: Multi-temporal UAV mapping that quantifies rapid expansion, delayed emergence, and rainfall-coupled deepening of collapses.
6. Cascading effects and risk assessment of rainstorm and flood disaster chains based on complex network and Bayesian network
Core Problem: Rainstorm and flood impacts evolve through cascading multi-node disaster chains that are hard to quantify.
Key Innovation: Combines complex-network structure and Bayesian-network probabilities for chain-level risk and intervention analysis.
7. A probabilistic framework for regional landslide hazard assessment by coupling time-varying landslide susceptibility and probabilistic seismic hazard
Core Problem: Regional landslide hazard under potential earthquakes needs probabilistic coupling of susceptibility and seismic hazard through time.
Key Innovation: Couples time-varying susceptibility modeling with PSHA-derived PGA in a regional probabilistic hazard framework.
8. Subsurface leak detection and landslide risk assessment using ERT and 3D imaging in Northeastern Brazil
Core Problem: Water-distribution leaks can trigger landslides but remain hidden from conventional rainfall-based monitoring.
Key Innovation: ERT and 3D resistivity imaging to map leak pathways and critical saturation volumes on unstable slopes.
9. Effect of Incidence Angle on Earthquake-Induced Landslides Through FDM-DEM: The Hongshiyan Rockslide (China) Case Study
Core Problem: How oblique seismic-wave incidence angle alters slope failure mechanisms in earthquake-induced landslides.
Key Innovation: Couples FDM and DEM to model arbitrary incidence angles and shows vertical-wave assumptions can understate failure.
10. Partitioning of granular flows at valley bifurcation under varying control conditions
Core Problem: Understand how granular flows split at valley bifurcations under varying controls.
Key Innovation: Targets bifurcation partitioning behavior rather than single-channel runout.
11. A multiple rotational slide-earthflow in Southern Apennines: The Pietrapertosa landslide
Core Problem: Characterize the Pietrapertosa multiple rotational slide-earthflow in the Southern Apennines.
Key Innovation: Focuses on a complex multi-lobe rotational slide-earthflow case.
12. MEFMA: A Multiscale Empirical Fringe-Manifold Adaptive Filtering Framework for Robust InSAR Phase Restoration
Core Problem: Denoise severely corrupted low-coherence InSAR phase while preserving dense deformation fringes needed for unwrapping and inversion.
Key Innovation: Builds a multiscale fringe-manifold filtering framework that alternates sparse local fringe extraction with gradient-preserving phase reconstruction smoothing.
13. Symmetry-Enhanced AI and GIS Framework Using Open Data for Simplified Physical Stability Assessment of Tailings Storage Facilities in Chile
Core Problem: Scale physical-stability assessment of tailings facilities using open data instead of costly site-specific inspections alone.
Key Innovation: Integrates satellite imagery, GIS layers, and deep-learning detection and classification into a hierarchical instability-and-impact prioritization framework.
14. An improved method for detecting event layers in lake sediment archives
Core Problem: Traditional sediment proxies miss some flood layers, especially organic-rich deposits, in lake archives.
Key Innovation: Non-destructive CT-thresholding workflow that isolates flood event layers at sub-centimeter resolution.
15. Classification of Urban Land Subsidence Types in Fuzhou from Time-Series InSAR Using FFT-Based Filtering and Ensemble Learning
Core Problem: Urban subsidence signals mix scales and drivers, complicating type classification.
Key Innovation: FFT-based separation of regional and local deformation plus ensemble-learning classification of five subsidence types.
16. A new framework for landslide negative sample selection based on farthest point sampling constrained by random forest importance weights
Core Problem: Random negative-sample selection causes background overfitting in landslide susceptibility assessment.
Key Innovation: Random-forest-weighted farthest-point sampling to mine hard negatives in geo-environmental feature space.
17. Integration of climate change impacts and agent-based model to flood risk evaluation in an urbanized area
Core Problem: Long-term urban flood risk under climate change depends on both physical flooding and adaptive behavior.
Key Innovation: Couples climate projections, hydrologic-hydraulic simulations, surveys, and ABM for stochastic long-horizon flood risk evaluation.
18. Mapping flood susceptibility in the transboundary Tumen River Basin at the China-North Korea border using transfer learning
Core Problem: Transboundary flood susceptibility mapping is constrained by sparse data on one side of the border.
Key Innovation: Validates simplified LightGBM and XGBoost models and transfers them across the China-North Korea boundary.
19. Spatiotemporal classification of flood-triggering rainfall in urban watersheds of Mamminasata, Indonesia
Core Problem: Daily rainfall extremes alone do not explain urban flood occurrence across heterogeneous watersheds.
Key Innovation: Event-based spatiotemporal typology linking antecedent and trigger rainfall structure to hydrologic response.
20. Road asset exposure and risk assessment of extreme storm flooding under a changing climate: a case of Shanghai
Core Problem: Future climate-driven storm flooding impacts on urban road networks are insufficiently quantified.
Key Innovation: Probabilistic exposure-damage analysis of Shanghai road assets across scenarios and return periods.
21. A Hybrid GeoAI Framework Integrating Smile Random Forest for Comprehensive Seismic Vulnerability Assessment: A Case Study of Mersin, Türkiye
Core Problem: Urban earthquake vulnerability assessment must combine physical, demographic, access, and facility factors coherently.
Key Innovation: Hybrid GeoAI framework combining ANP-style weighting, fuzzy distance functions, and Smile Random Forest mapping.
22. Urban inundation prediction using seasonal autoregressive integrated moving average model under extreme rainfall events
Core Problem: Fast prediction of inundation depth, flow, and velocity during extreme rainstorms.
Key Innovation: Applies SARIMA to multi-point inundation forecasting under 100-year rainfall scenarios.
23. Flood susceptibility mapping using analytic hierarchy process hybrid with machine learning model
Core Problem: Map flood susceptibility by combining expert weighting and machine learning.
Key Innovation: Hybridizes AHP with machine learning for flood-susceptibility assessment.
24. Enhanced simulations of overland-gully-sewer flow interactions in coupled three-layer dual drainage modeling
Core Problem: Better simulate interactions among overland flow, gullies, and sewer networks.
Key Innovation: Develops a coupled three-layer dual-drainage model for integrated urban flood dynamics.
25. Slope-weighted cost distance for modeling flood dynamics: a physics-informed random forest approach
Core Problem: How to forecast flood inundation rapidly without the computational cost of physical models or the weak generalization and interpretability of conventional data-driven models.
Key Innovation: Introduces slope-weighted cost distance as a physics-informed random-forest feature and a hydrograph-space sampling strategy that sustains high F1 performance with four representative training cases.
26. Beyond proportional training set selection in data-driven models: A novel two side distributional decomposition method for urban flood prediction
Core Problem: How to select informative, non-proportional training samples for urban flood models instead of relying on arbitrary fixed train-test fractions.
Key Innovation: Proposes two-side distributional decomposition, which improves RF, LSTM, CNN, and Transformer predictions across six decomposition methods and high-frequency drainage observations.
27. Climate-informed streamflow forecasting based on a Bayesian autoregressive exogenous stochastic volatility model and its use for hydrological drought forecasting
Core Problem: Improve streamflow forecasting and extend it to hydrological drought prediction using climate information.
Key Innovation: Uses a Bayesian autoregressive exogenous stochastic-volatility model for climate-informed drought forecasting.
28. The Overprint of Transient Rheology on Laterally Heterogeneous Viscosity: Influences on Sea-Level Change Driven by Antarctic Ice Sheet Loss
Core Problem: How transient rheology interacts with lateral viscosity structure in Antarctic ice-loss deformation predictions.
Key Innovation: First joint 3D transient and laterally heterogeneous viscoelastic modeling, showing much higher uplift than Maxwell-only cases.
29. GPU-Accelerated InSAR-BM3-D and OC-InSAR-BM3-D for Scalable Interferometric Phase Filtering
Core Problem: Make state-of-the-art nonlocal InSAR phase filtering practical for operational-scale interferograms.
Key Innovation: Provides GPU-accelerated InSAR-BM3-D and OC-InSAR-BM3-D with memory-optimized tiling, float32 acceleration, and multi-GPU scaling.
30. TLSPGM: A Phase-Gradient-Based Approach for Efficient 3-D Repositioning Error Compensation in D-GBSAR Slow Deformation Monitoring
Core Problem: D-GBSAR repositioning correction is slowed and destabilized by phase unwrapping and nonconvex optimization.
Key Innovation: Phase-gradient least-squares repositioning compensation that avoids 2D phase unwrapping.
31. Comprehensive inter-comparison of generative AI models for super-resolution precipitation downscaling across hydroclimatic regimes
Core Problem: Coarse climate products miss storm-scale extremes needed for hazard modeling, and model tradeoffs are unclear.
Key Innovation: Systematic comparison of U-NET, WGAN, and diffusion models for 8x-16x precipitation super-resolution across regimes.
32. Stability Charts for Blast Disturbed Zone of Rock Slope Using Hoek-Brown Failure Criterion
Core Problem: How to represent depth-varying blast damage when estimating rock-slope safety.
Key Innovation: Introduces a two-zone disturbance model and stability charts that outperform constant-disturbance assumptions.
33. Rockburst risk assessment in deep mining using ensemble models with Interpretable features
Core Problem: How to improve rockburst risk classification under class imbalance while retaining interpretable links between mechanical indicators and predicted hazard levels.
Key Innovation: Combines ADASYN reconstruction, five ensemble strategies, Bayesian hyperparameter optimization, and SHAP/ICE/PDP interpretation to expose nonlinear thresholds in rockburst risk.
34. Consensus-gated Multi-Agent Neural Architecture Search for Seismic Fault Segmentation
Core Problem: Design data-efficient seismic fault segmentation models under tight labeled-data budgets.
Key Innovation: Uses consensus-gated multi-LLM architecture search to discover a compact best-performing seismic fault network.
35. Beyond Control Points: Arcsecond Relative-Motion Estimation of Vision Measurement Platforms With Incomplete or Absent Control Fields
Core Problem: Estimate platform motion accurately when vision monitoring lacks complete control fields.
Key Innovation: Derives a control-adaptive differential estimator that recovers arcsecond rotation and millimeter translation without nonlinear pose optimization.
36. Rethinking Auxiliary Modalities in Multimodal Zero-shot Anomaly Detection: From Semantic Fusion to Conditional Modulation
Core Problem: Use auxiliary modalities to improve zero-shot anomaly detection without corrupting RGB foundation-model semantics.
Key Innovation: Refines RGB anomaly features through global-to-local conditional modulation driven by auxiliary cues and uncertainty.
37. FIRM: Fine-Grained Intra-Token Representation of Masks for Remote Sensing Reasoning Segmentation
Core Problem: Recover small targets and fine boundaries lost when one visual token covers many remote-sensing patches.
Key Innovation: Predicts intra-token mask codes plus a lightweight continuous renderer for fine-grained mask recovery.
38. Extending Occam's inversion with lasso fusion, overcomplete dictionaries, and isotropic total variation regularisation
Core Problem: Modernize Occam-style geophysical inversion with sharper l1 regularization options and clearer solver choices.
Key Innovation: Unifies fused lasso, isotropic total variation, overcomplete dictionaries, and three solver families inside the Occam inversion framework.
39. Accelerating Large-scale Bundle Adjustment for LiDAR Mapping via Parallel Computing
Core Problem: Speed up large-scale LiDAR bundle adjustment without sacrificing mapping accuracy.
Key Innovation: Fully parallel framework combining asynchronous loading, bottom-up planar voxelization, and parallel residual, Hessian, and increment computation.
40. Coupling-Robust Accuracy in Multiphysics Physics Informed Neural Networks via Kronecker-Preconditioned Optimization
Core Problem: Prevent PINN accuracy collapse as coupling strength increases in strongly coupled multiphysics systems.
Key Innovation: Shows coupling failure is a preconditioner-structure problem and uses Kronecker-preconditioned SOAP plus GradNorm to make performance coupling-robust.
41. Cross-View Change Detection via Self-Supervised Learning
Core Problem: Cross-view change detection is hard under viewpoint, scale, and modality gaps without labels.
Key Innovation: Self-supervised contrastive and predictive learning with cross-attention for unlabeled satellite-UAV change detection.
42. Revealing the structure of heavy precipitation events: a spatio-temporal wavelet approach
Core Problem: Risk depends on rainfall structure, but event timing, length, and motion are poorly quantified.
Key Innovation: Spatio-temporal wavelet framework that extracts key structural characteristics from radar-observed heavy-rainfall events.
43. Anisotropic behavior of high liquid limit clay subjected to two maximum drying stress paths
Core Problem: How drying stress paths change anisotropy and cracking in high liquid limit clay relevant to slope collapse.
Key Innovation: Combines triaxial tests and microstructural analysis to contrast MDS and MDS-SS induced anisotropy.
44. Progressive Damage of Sandstone and Shale Under Heating and Cooling Cycles: Implications for Stratified Rock Slope Stability
Core Problem: Assess how repeated heating-cooling cycles progressively damage sandstone-shale slopes.
Key Innovation: Links cyclic thermal loading, microcrack evolution, and property degradation to slope-stability implications.
45. Simulated AVIRIS-3 detects species-specific patterns of vegetation succession in retrogressive thaw slumps
Core Problem: Detect species-specific vegetation succession patterns associated with retrogressive thaw slumps.
Key Innovation: Uses simulated AVIRIS-3 hyperspectral sensing to resolve thaw-slump vegetation signatures.
46. Regional atmospheric pattern shifts associated with weather whiplash events in the upper Yangtze River Basin
Core Problem: Identify atmospheric pattern shifts associated with weather whiplash in the upper Yangtze.
Key Innovation: Links regional atmospheric-regime changes to rapid hydroclimatic swings.
47. Bathymetry Prediction With SWOT Gravity Anomaly Using Machine Learning Methods: Paper 2-Model Evaluation and Uncertainty Analysis
Core Problem: Global gravity-based bathymetry remains uncertain in unsounded regions and needs better confidence estimation.
Key Innovation: Combines SWOT gravity with five ML models to improve bathymetry and produce confidence and target-mapping layers.
48. Bathymetry Prediction With SWOT Gravity Anomaly Using Machine Learning Methods: Paper 1-Model Development
Core Problem: Classical gravity-to-bathymetry inversion is too coarse for unresolved seamounts and trenches.
Key Innovation: Shows physics-aware ML with SWOT gravity substantially improves global bathymetric accuracy and resolution.
49. Lidar Ratios and Fluorescence Properties of Stratospheric Volcanic Sulfate
Core Problem: How to isolate the optical properties of pure stratospheric volcanic sulfate and distinguish it from wildfire smoke.
Key Innovation: Combines Raman and fluorescence lidar to derive pure sulfate lidar ratios and a clear low-fluorescence signature.
50. Overland Flow Dynamics on a Cracked Soil Slope Under Drying-Wetting Cycles: An Integrated Hydrological and Infrared Thermographic Study
Core Problem: How drying-wetting cracking alters runoff generation, flow organization, and sediment yield on a soil slope.
Key Innovation: Integrates rainfall simulation, crack image analysis, and infrared thermography to track evolving flow connectivity.
51. Multiphase-Diff: Diffusion-Based Generative Modeling for High-Contrast Multiphase Physical Systems with Sharp Interfaces
Core Problem: Diffusion models struggle to represent high-contrast multiphase fields with discontinuous coefficients and sharp interfaces.
Key Innovation: Combines conservative flux residuals, positivity-preserving latent transforms, and preconditioned likelihood balancing.
52. HI-MeshGraphNets: Efficient and Accurate Mesh-based Physics Learning with Hierarchical Multi-scale Graph Neural Networks
Core Problem: Flat message passing makes mesh surrogates inefficient and memory-heavy on large physical domains.
Key Innovation: Hierarchical graph coarsening and learned interpolation enable longer-range physics communication at lower cost.
53. Probabilistic indirect models for undrained shear strength: addressing significant data missing and variability with advanced imputation and machine learning techniques
Core Problem: Predict undrained shear strength from sparse, incomplete CPTU and index-property data.
Key Innovation: Benchmarks missing-data imputation and couples the best strategy with an attention-based probabilistic neural network.
54. From Fixed Grids to Moving Particles:A Transferable Latent Operator for Fluid Dynamics
Core Problem: Neural operators trained on Eulerian fields rarely generalize to Lagrangian particle trajectories without dedicated supervision.
Key Innovation: Introduces a transferable latent operator that shares latent flow evolution across field prediction and particle rollout, enabling zero-shot Lagrangian forecasting.
55. PISA: A Pseudo-Individual Source-Domain Feature Adaptation Framework for Test-Time Open-Vocabulary Object Detection
Core Problem: Open-vocabulary detectors degrade under test-time domain shift, and pseudo-label-based adaptation collapses when initial predictions are poor.
Key Innovation: Uses corruption-invariant features and pseudo source-domain feature alignment instead of pseudo labels for source-free test-time adaptation.
56. Connected Subspace Clustering: Hardness, a Scalable Heuristic, and an Application to Sea Level Geodesy
Core Problem: Spatially coherent regionalization of high-dimensional sea-level measurements is hard under connectivity constraints.
Key Innovation: Defines connected subspace clustering, proves hardness, and proposes a scalable merging heuristic that returns exactly connected regions.
57. Learning to Forecast Crop Growth from Earth Observation Data
Core Problem: Sparse cloud-free LAI supervision makes sequence models fit observations while generating physically implausible crop trajectories between them.
Key Innovation: Combines Sentinel-2 and meteorology in Seq2Seq forecasting and adds a lightweight shape regularizer to enforce plausible future trajectories.
58. Weakly Supervised Polar Low Segmentation in Sentinel-1 SAR Imagery
Core Problem: Polar low segmentation lacks pixel-level masks and even experts disagree on diffuse storm boundaries.
Key Innovation: Introduces CREST, which adds connected region expansion and reliability-weighted bootstrapping to weakly supervised SAR segmentation.
59. Catching the Imposter: Self-Supervised Learning of Physical Coherence with Cross-Entity Feature Permutations
Core Problem: Standard self-supervised objectives for scientific data ignore cross-feature physical coherence governed by environmental laws.
Key Innovation: Creates an imposter pretext task that swaps subsets of physically linked environmental features to force learning of coherent dependencies.
60. GhostPoint: Self-Supervised Representation Learning by Hallucinating Occluded LiDAR Structure
Core Problem: Self-supervised LiDAR objectives learn mostly from visible returns, leaving missing or occluded structure weakly modeled for downstream 3D detection.
Key Innovation: Hallucinates neighborhood features for observed and unobserved voxels using instance voxel dilation plus predictor-level supervision.
61. Learning Unsteady Aneurysm Hemodynamics with Physics-Informed DeepONets
Core Problem: Patient-specific unsteady 3D flow prediction remains too costly with CFD and difficult for existing PI-DeepONet architectures.
Key Innovation: Uses a multi-input multi-output PI-DeepONet with aggregated latent injection and Navier-Stokes constraints to predict full flow fields from sparse labels.
62. Unknown Unknowns: Model Misspecification in Machine Learning for Physics
Core Problem: Models trained on simulations can fail on real physical data in unanticipated ways that standard validation may miss.
Key Innovation: Frames misspecification handling as an iterative loop of complementary diagnostics plus mitigation strategies rather than a single test.
63. On the Brittleness of Maximum Likelihood Estimation for Gaussian Process Hyperparameter Optimization
Core Problem: Maximum-likelihood hyperparameter fitting for Gaussian processes can be brittle and produce poor generalization in engineering workflows.
Key Innovation: Diagnoses MLE brittleness, compares stronger alternatives, and provides practical fixes that improve downstream Bayesian optimization and uncertainty quantification.
64. Meteorology-driven Causal Nowcasting of Fugitive Landfill Emissions Enables Proactive Public Health Response
Core Problem: Forecast harmful fugitive landfill-gas episodes early enough for proactive public-health intervention.
Key Innovation: Causal, meteorology-driven nowcasting framework with fast and slow memory components that reproduces sensor-network alerts and complaint records.
65. Ordinal-Aware Calibration for Ordinal Classification
Core Problem: Calibrate ordinal classifiers while preserving unimodal, order-consistent predictions.
Key Innovation: ORCU loss unifies distance-aware soft targets with an ordinal-aware log-barrier to improve calibration without architecture changes.
66. An Integrated Image Matching and Block Adjustment Method for Regional-Scale Spaceborne SAR Image Mapping
Core Problem: Improve relative geometric positioning for regional-scale spaceborne SAR image mapping.
Key Innovation: Couples enhanced SAR-SIFT tie-point extraction with DEM-aided planar block adjustment for accurate large-area orthorectification.
67. Stage-Adaptive Spatial-Frequency Decomposition and Enhancement With Multimodal Conditional Routing for Remote Sensing Image Segmentation
Core Problem: Fuse multimodal remote-sensing data more adaptively across encoding and decoding stages for segmentation.
Key Innovation: Uses stage-specific spatial-frequency fusion and conditional expert-routed frequency decomposition to adapt masks to scene content.
68. Deep-Sea Bathymetry From Geosynchronous SAR via DTCWT-Enhanced Spatial-Correlation Wave Imaging and Multiparameter Time-Series Inversion
Core Problem: Recover deep-sea bathymetry from geosynchronous SAR despite wave defocusing and ill-posed depth inversion.
Key Innovation: Pairs DTCWT-enhanced motion-compensated spatial-correlation SAR imaging with a hierarchical multiparameter time-series inversion network.
69. FSDNet: Joint Frequency-Spatial Modulation Network for Remote Sensing Image Dehazing
Core Problem: Remove haze from remote-sensing imagery without oversmoothing details or using heavy transformer or diffusion models.
Key Innovation: Jointly optimizes spatial and Fourier-domain branches and adds an FFT reconstruction loss to preserve structural frequency content.
70. Human-Cognitive Inspired Remote Sensing Change Detection Network
Core Problem: Improve remote-sensing change detection by strengthening global context, temporal encoding, and decision fusion.
Key Innovation: Organizes the model as a perception-memory-observation-decision pipeline combining foundation-model features, BiLSTM memory, Dense Mamba, and attention fusion.
71. 3-D Joint Inversion of Urban Persistent and Distributed Scatterers via SAR Tomography
Core Problem: PS-only TomoSAR underuses distributed scatterers and leaves sparse urban 3D inversion under limited acquisitions.
Key Innovation: Joint PS-DS TomoSAR with a directed-minimum-spanning-tree reference network and flexible RN construction.
72. Comprehensive Evaluation of the CYGNSS V3.2 Soil Moisture Product
Core Problem: CYGNSS soil-moisture accuracy, stability, and failure regimes need long-record evaluation.
Key Innovation: Seven-year validation tying calibration upgrades and surface conditions to retrieval performance.
73. Edge Integrity Learning for Remote Sensing Image Change Detection
Core Problem: Change edges blur into background under seasonal and illumination variation, causing false detections.
Key Innovation: Edge-region decoupled siamese network with explicit and implicit edge constraints.
74. Hydrologic model parameter estimation in snow-dominated headwater catchments using multiple observation datasets
Core Problem: Multi-dataset constraints can improve or degrade parameter estimation in snow-dominated catchment models.
Key Innovation: Assesses how multiple observation datasets reshape parameter identifiability and streamflow skill.
75. Effects of Deep Learning Observation Operators in Direct Radiance Assimilation of Microwave Radiation Imager in Land Surface Models
Core Problem: Physical observation operators limit direct microwave radiance assimilation over land.
Key Innovation: MLP-based surrogate radiative transfer operator within SEKF for land-surface soil-moisture assimilation.
76. A Consistency-Guided Collaborative Filtering Framework for Suppressing Structured Coherent Artifacts
Core Problem: Structured coherent artifacts mimic real signals and are hard to remove without losing information.
Key Innovation: Consistency-guided collaborative filtering using redundant paired observations and coherent-noise PSD modeling.
77. Shear Creep Behavior of an Undisturbed Fault Fracture Zone and Bayesian Parameter Inversion of a Fractional-Order Model: A Case Study of the F115 Fault at the LCGX Hydropower Station
Core Problem: Characterize shear creep and long-term deformation behavior of a fault fracture zone affecting dam and cavern stability.
Key Innovation: Builds a fractional-order creep model with Bayesian MCMC inversion and parameter uncertainty quantification.
78. Advancing subsurface characterization through borehole optical imaging: A comprehensive review
Core Problem: Summarize how borehole optical imaging advances subsurface characterization.
Key Innovation: Synthesizes borehole optical imaging capabilities and applications across underground settings.
79. Evaluating the effectiveness of GSMaP precipitation product in capturing precipitation events
Core Problem: Evaluate how well GSMaP captures precipitation events.
Key Innovation: Assesses event-level performance of a widely used satellite precipitation product.