TerraMosaic Daily Digest: August 19, 2026
Daily Summary
The August 19 papers place hydroclimatic hazards in a strongly process-based frame. Tropical-cyclone studies pair physics-constrained generative forecasting with hurricane-centric atmospheric simulation to pursue higher skill at lower computational cost, while flood papers show that risk depends on system memory and human response as much as on local forcing: water routed from Lake Victoria to the Sudd may take about 17 months to propagate, wetland storage can prolong inundation over several years, and municipality-scale damages in Japan shift when flood-induced population movement is included. Parallel hydrological and cryosphere studies resolve asynchronous peak-runoff regimes across High Mountain Asia, improve glacier wet-snow mapping under cloud and SAR ambiguity, and show that post-drought vegetation recovery in the Yangtze varies materially by indicator and cover type.
Direct slope, coastal, seismic, and volcanic studies emphasize hidden structural and hydro-mechanical controls. In the Chilean Andes, multisensor satellite mapping constrains rainfall-triggered landslides, whereas the 2025 Guowa failure is reconstructed as rainfall-driven locked-segment rupture followed by retrogressive chain sliding; related work models landslide damming with multiphase flow, shows dune protection in cyclone-exposed Mozambique falling sharply below roughly 3 m, and treats dam seepage, coral-reef slopes, mine collapse, and weathered sandstone slopes as coupled geometry-fluid-material systems. Earthquake papers similarly move toward mechanism-aware characterization through scaling-law-informed sequence forecasting, faulting-specific spectral directionality, terrain-sensitive site-response simulation, regional CPT-Vs calibration, and transferable building-vulnerability estimation, while GNSS-InSAR analysis at Krafla resolves spatially variable locking and residual deformation within an active volcanic rift.
A substantial parallel stream contributes enabling methods rather than direct geohazard validation. Landslide and flood mapping studies prioritize reliability through pixel-level uncertainty triage, boundary-aware SAR segmentation, and faster elite-pixel selection for time-series InSAR; broader Earth-observation work advances cross-sensor reflectance reconstruction, SAR-to-optical translation, hyperspectral and LiDAR fusion, open-set and open-vocabulary segmentation, and long-term disturbance mapping. Scientific-machine-learning papers extend this emphasis on structure and auditability through probabilistic precipitation downscaling, neural operators with causal memory and flux constraints, conformal anomaly monitoring, uncertainty benchmarks for segmentation foundation models, and domain-shift-robust inference frameworks that enlarge the toolkit available to hazard science without yet establishing hazard-wide generalization.
Key Trends
The August 19 papers converge on physically constrained forecasting, hydro-mechanical threshold diagnosis, reliability-focused remote sensing, more site-specific seismic characterization, and a transferable GeoAI stream that treats uncertainty and cross-sensor generalization as first-order requirements.
- Hydroclimatic Forecasting Is Becoming Physics-Constrained and State-Aware: Tropical-cyclone forecasting and hurricane-centric simulation both seek higher fidelity without prohibitive cost, while flood and cryosphere papers show that storage, travel time, and class-resolved runoff regimes strongly shape hazard expression. The same direction appears in probabilistic precipitation downscaling, where domain knowledge is built into the learning objective rather than added post hoc.
- Slope and Coastal Failure Studies Are Resolving Explicit Hydro-Mechanical Thresholds: The Guowa landslide, river-blocking simulations, dune-failure experiments, seepage analyses in dams and embankments, and wave-driven coral-reef slope studies all foreground internal structure, pore-pressure evolution, and geometry-dependent thresholds. Hazard interpretation is moving away from trigger-only narratives toward coupled material-fluid-process diagnosis.
- Hazard Mapping Is Shifting from Raw Detection to Reliability and Reviewability: The Chilean Andes landslide workflow, uncertainty-ranked landslide mapping, boundary-aware SAR flood extraction, and ConvLSTM elite-pixel selection for TS-InSAR all emphasize which pixels or boundaries can be trusted, not only whether an object is detected. This makes remote-sensing outputs more suitable for deformation monitoring, inspection triage, and operational review.
- Seismic Risk Assessment Is Moving Toward Mechanism- and Site-Specific Models: Earthquake papers incorporate recurrence structure, faulting style, terrain geometry, and local geotechnical calibration instead of relying on generic regional assumptions. Sequence forecasting, directional spectral characterization, valley-response simulation, CPT-Vs recalibration, and GIS-based building-vulnerability estimation collectively point to more differentiated seismic hazard and risk modeling.
- Transferable GeoAI Prioritizes Cross-Sensor Reconstruction, Uncertainty, and Domain Shift: Across HLS reflectance reconstruction, SAR-to-optical translation, hyperspectral-LiDAR fusion, segmentation uncertainty audits, conformal monitoring, and robust adaptation theory, the method-heavy papers aim to preserve structure and quantify failure modes across sensing conditions. These advances are enabling rather than hazard-proven: they expand what geohazard studies can use, but do not themselves establish validation across hazard domains.
Selected Papers
The selected papers combine direct analyses of tropical cyclone, flood, landslide, seismic, volcanic, coastal, drought, cryosphere, and engineered-ground hazards with enabling advances in Earth observation, geospatial AI, and scientific machine learning. The hazard papers revise mechanisms, thresholds, or risk estimates in their stated systems, whereas the method papers broaden sensing, reconstruction, and uncertainty toolkits without by themselves demonstrating geohazard validation.
1. Tianmu-TC: Physics-constraints Generative Artificial Intelligence for Global Tropical Cyclone Forecasting
Core Problem: Improve global tropical cyclone track and intensity forecasting while reducing uncertainty and compute cost.
Key Innovation: Physics-constrained generative AI produces controllable forecasts that reportedly outperform strong AI and NWP baselines across basins.
2. Plate Divergence at the Krafla Volcanic System, Iceland: Insight From GNSS Geodesy and Sentinel-1 Satellite Radar Interferometry in 2002-2024
Core Problem: Resolve plate-boundary geometry, locking depth, and local deformation at Krafla.
Key Innovation: Joint 2002-2024 GNSS and Sentinel-1 inversion maps variable locking depth and residual deflation zones.
3. Scaling-law-informed neural point processes for earthquake sequence forecasting
Core Problem: Need nonlinear earthquake sequence models that retain ETAS-style statistical structure.
Key Innovation: Neural marked point process fused with ETAS temporal features and Gutenberg-Richter magnitude information.
4. Effect of style of faulting on the probabilistic characterization of directionality of horizontal response spectral ordinates
Core Problem: Style of faulting alters the full directional distribution of horizontal spectral ordinates, but this is under-modeled.
Key Innovation: Mechanism-specific probabilistic models of directionality from 3,862 records across 58 periods for strike-slip and reverse-slip events.
5. Rainfall-Induced Landslide Detection in the Central-Southern Andes of Chile Using Integrated SAR, Optical, and Machine-Learning Approaches in Google Earth Engine
Core Problem: Detect landslides triggered by the June 2023 extreme rainfall event in the Chilean Andes using multisource satellite data.
Key Innovation: Google Earth Engine workflow integrates Sentinel-1, Sentinel-2, PALSAR topography, and spatially validated machine-learning probability mapping.
6. Cross-Level Frequency-Domain Boundary-Aware Lightweight Network for SAR Flood Detection
Core Problem: Improve bitemporal SAR flood detection when flood boundaries are irregular and weak-scattering areas cause false alarms.
Key Innovation: Frequency-domain boundary-aware lightweight network uses cross-level semantic guidance and multiscale bottlenecks to sharpen flood extraction.
7. Integrating flood-induced population movements into future fluvial flood damage estimates in Japan
Core Problem: Estimate how flood-induced population movements alter future fluvial flood damages in Japan.
Key Innovation: Couples migration responses with future flood-damage modeling to reveal important municipality-scale effects.
8. Advancing the capabilities for efficient hurricane-centric simulations with the atmospheric model ICON
Core Problem: Run hurricane-focused atmospheric simulations at very high resolution without the cost of large conventional domains.
Key Innovation: Hurricane-centric ICON configuration follows the storm and cuts compute by up to an order of magnitude while reaching 300 m resolution.
9. Lake Victoria to the Sudd Wetland: flood wave timing, connectivity and wetland buffering across the White Nile
Core Problem: Explain prolonged 2019-2024 flooding in South Sudan by tracing flood-wave timing from Lake Victoria through the White Nile system.
Key Innovation: Shows roughly 17-month travel times and multiyear wetland buffering using satellite, rainfall, lake, and river observations.
10. Enhancing the Reliability, Performance, and Practical Value of Deep Learning Landslide Mapping via Pixel-Level Uncertainty Analysis
Core Problem: Deep learning landslide maps rarely quantify pixel-level uncertainty, reducing trust and review efficiency.
Key Innovation: Applies Monte Carlo Dropout and uncertainty-ranked false-positive and false-negative triage to improve landslide mapping accuracy and inspection.
11. Characteristics and formation mechanism of the May 22, 2025, Guowa Township landslide in Dafang County, Guizhou Province, China
Core Problem: Explain deformation buildup and catastrophic failure of the 2025 Guowa bedding rock landslide.
Key Innovation: Combines field work, rock tests, SBAS-InSAR, and UDEC to identify locked-segment failure and retrogressive chain sliding.
12. Seismic response of strongly weathered sandstone slopes: fault-tunnel effects and the reinforcement mechanism
Core Problem: Fault-tunnel interactions complicate seismic amplification and damage in weathered sandstone slopes.
Key Innovation: Joint time-frequency FLAC3D analysis tied to a combined pile-anchor-frame-beam reinforcement scheme.
13. Evaluating coastal dune vulnerability to tropical cyclone forcing using numerical modeling in a data-scarce region
Core Problem: Assess dune failure thresholds under cyclone forcing in a data-scarce coastline.
Key Innovation: Large XBeach scenario set identifying about 3 m as a critical protective dune height.
14. Landslide-induced river blocking based on a multiphase flow finite volume method
Core Problem: Simulate how landslides block rivers and form blockage structures.
Key Innovation: Multiphase-flow finite-volume method tailored to landslide-induced river blocking.
15. Research on Soil Fixation Performance of Native Crop Root-Reinforced Soil in Sandy Loess Farmland of Sandy Arid Regions: A Wind Tunnel Test Study
Core Problem: How native crop roots change sandy-loess wind erosion resistance.
Key Innovation: Wind-tunnel root-volume-density gradients linked shear strength, erodibility, and erosion reduction in a nonlinear model.
16. SeisEvo: Evolution of Seismic Data Reconstruction Algorithms by Agents
Core Problem: Search for better explicit seismic reconstruction operators than hand-designed classical methods.
Key Innovation: LLM-driven constrained program evolution discovers white-box seismic interpolation and denoising operators that outperform classic baselines.
17. COSTA: A Cluster-Centric Paradigm for Annotation-Free Open-Set Semantic Segmentation of Aerial Point Clouds with Domain Shifts
Core Problem: Adapt aerial point-cloud segmentation to unlabeled shifted domains while allowing target-specific open-set classes.
Key Innovation: Cluster-centric test-time adaptation distills features into semantic centroids and propagates open-vocabulary pseudo-labels.
18. Three-dimensional upper-bound stability assessment of multi-stage slopes in highly permeable coral-reef deposits under wave-induced seepage
Core Problem: Assess 3D upper-bound stability of multi-stage slopes in highly permeable coral-reef deposits under wave-induced seepage.
Key Innovation: Combines upper-bound slope-stability assessment with wave-seepage conditions for coral-reef deposit slopes.
19. Using machine learning for the prediction of flood-related 112 calls
Core Problem: Predict which municipalities will place flood-related emergency calls within the next hour.
Key Innovation: Machine-learning models fuse meteorology, demographics, and historical 112-call records to outperform hazard-only systems.
20. Research on fuzzy analytic hierarchy process projection model for vulnerability assessment of geological hazards and its application
Core Problem: Existing FAHP approaches lose information and weakly integrate socioeconomic and geographic drivers of geological hazard vulnerability.
Key Innovation: Combines triangular fuzzy possibility weights with a projection model for township-scale geological hazard vulnerability mapping.
21. Integrated UAV, TEM, and microseismic monitoring reveals the hydro-structural coupling conceptual model of goaf collapse in an abandoned gypsum mine
Core Problem: Explain and monitor collapse triggering in an abandoned gypsum goaf that threatens surface infrastructure.
Key Innovation: Integrates UAV photogrammetry, TEM, and microseismic data into a hydro-structural coupling model for collapse initiation.
22. Resolving confusion during earthquakes: the role of early warning messages in supporting and expediting comprehension
Core Problem: People receiving a first EEW alert mainly face confusion rather than clean rational decision-making.
Key Innovation: Derives a four-phase Earthquake Comprehension Cycle from nearly 1,000 firsthand accounts to improve warning-message design.
23. Long-Horizon Mining Subsidence Forecasting and Ecological Time-Lag Assessment Using Multi-Source Remote Sensing
Core Problem: Long-horizon mining subsidence prediction is unstable, and ecological indicators lag real degradation.
Key Innovation: Pairs SBAS-InSAR forecasting with stratified sampling, multi-model comparison, and cross-lag ecological analysis.
24. Integrated machine-learning and clustering framework for selecting temporary shelters in flood-prone cities
Core Problem: Temporary shelters must be placed where both buildings and access routes avoid flood exposure.
Key Innovation: CatBoost flood susceptibility mapping coupled with K-means shelter allocation and route-exposure screening.
25. The flash floods in Lambaré, Paraguay, on November 2, 2023: a simulation study
Core Problem: Reconstruct a fatal urban flash flood and identify safer vehicle routes through flooded streets.
Key Innovation: Rapid urban DEM modification plus HEC-RAS hazard mapping and flood-aware routing.
26. New valley site seismic response simulation method and application validation for seismic zoning
Core Problem: One-dimensional site-response methods miss amplification patterns in valleys and steep slopes.
Key Innovation: Optimized multidimensional valley-site simulation for terrain-sensitive amplification analysis.
27. WebGIS-based decision support system for landslide risk management in Bhutan
Core Problem: Fragmented hazard, exposure, and vulnerability data limit local landslide planning in Bhutan.
Key Innovation: Operational WebGIS DSS integrating susceptibility, exposure, and vulnerability layers at Gewog scale.
28. Regional Calibration of CPT Correlations for Shear Wave Velocity in Budapest Soils
Core Problem: Global CPT-Vs correlations underperform for Budapest soils used in seismic design.
Key Innovation: Large paired database recalibrating regional CPT-Vs relations for local seismic characterization.
29. Numerical analysis of seepage through earth dam considering deterministic modeling of heterogeneous cutoff wall parameter
Core Problem: Homogeneous cutoff-wall assumptions miss localized seepage danger in earth dams.
Key Innovation: Fuses resistivity and permeability data into a heterogeneous 3D hydraulic conductivity field for seepage analysis.
30. Comparison of tsunami evacuation map designs, warnings, and risk perceptions in Aotearoa New Zealand
Core Problem: Map design and warning style can alter evacuation understanding and perceived tsunami risk.
Key Innovation: Comparative evaluation of tsunami evacuation map designs and warning formats.
31. From local data to global risk maps: A geospatial generative pre-trained transformer framework for earthquake building vulnerability estimation
Core Problem: Scale building vulnerability estimation from sparse local data to broader earthquake risk maps.
Key Innovation: Geospatial generative pre-trained transformer for transferable building vulnerability estimation.
32. Boundary-Aware Physical Constraints Refine Satellite Precipitation Occurrence Over the Qinghai-Tibet Plateau
Core Problem: Correct mistimed weak wet-hour artifacts in hourly satellite precipitation occurrence over mountains.
Key Innovation: Learns interpretable onset and cessation rules and boundary probabilities to conservatively correct GSMaP and IMERG.
33. Scalable Geospatial Machine Learning for Power-Line Asset Risk: Integrating Remote Sensing for Lightning and Vegetation Risk Modelling
Core Problem: Utilities need scalable asset-level failure probabilities under lightning and vegetation exposure.
Key Innovation: Modular explainable geospatial ML pipeline integrating remote sensing predictors and operational records.
34. DynCur-Geo: Dynamic Curiosity Reward Shaping for Multimodal Active Geo-Localization
Core Problem: UAV search must balance exploration and target convergence with sparse multimodal cues.
Key Innovation: Distance-aware dynamic curiosity weighting plus potential-based reward shaping for active geo-localization.
35. SED-FOD: Scattering-Aware Expert Decomposition for Few-Shot Cross-Sensor SAR Object Detection
Core Problem: Few-shot SAR detectors lose accuracy across sensors because useful scattering differences are suppressed.
Key Innovation: Shared-specific expert decomposition that preserves sensor-dependent scattering while aligning transferable structure.
36. Flux-form spatiotemporal neural operators for coarse-grained dynamics of multiscale PDEs
Core Problem: Coarse-grained multiscale PDE surrogates need memory, conservation, and stable long-horizon rollout.
Key Innovation: Flux-form spatiotemporal neural operator with causal temporal kernels and data-driven memory-length selection.
37. Online Conformal Anomaly Detection with Prediction-Powered Data Acquisition
Core Problem: Online anomaly detection with guarantees usually requires a steady stream of real calibration data.
Key Innovation: Context-aware prediction-powered conformal framework that mixes synthetic and real calibration data while controlling false discoveries.
38. ProtoMix: Unified Prototype Learning for Semantic Change Detection in Remote Sensing
Core Problem: Detect semantic land-cover changes despite high intraclass variation and sparse labels on changed pixels.
Key Innovation: Prototype clusters, pseudo-labeled unchanged regions, and contrastive learning improve semantic change detection.
39. Leveraging Power of Deep Learning for Fast and Efficient Elite Pixel Selection in Time Series SAR Interferometry
Core Problem: Select stable elite pixels in long InSAR time series quickly and accurately across varied terrains.
Key Innovation: ConvLSTM models learn WabInSAR-style elite-pixel selection, cutting processing to minutes while supporting faster deformation mapping.
40. Calibrated, robust, and physically consistent uncertainty quantification in deep-learning surface-wave shallow shear-velocity inversion
Core Problem: Deterministic surface-wave inversion lacks calibrated uncertainty and degrades under distribution shift.
Key Innovation: Adds conformal and augmentation-calibrated uncertainty quantification plus physics-consistency checks to deep shear-velocity inversion.
41. Rocking analysis of the moai-pukao monuments of Rapa Nui: directional topknot release and a sacrificial-fuse effect
Core Problem: Determine how moai-pukao assemblies fail under earthquake rocking.
Key Innovation: Two-stage rigid-block model revealing directional topknot release and sacrificial-fuse behavior.
42. DAMSE, an integrated system to support emergency management and monitoring of dams in earthquake prone areas
Core Problem: Responders need integrated real-time tools for dams during earthquakes.
Key Innovation: Links WebGIS, digital inspection forms, and mobile field collection into one emergency workflow.
43. HLS-GPT: A generative pretrained transformer (GPT) for continental-scale NASA harmonized Landsat and Sentinel-2 (HLS) reflectance reconstruction across all bands on arbitrary dates
Core Problem: Reconstruct gap-free multispectral HLS reflectance on arbitrary dates at continental scale.
Key Innovation: GPT-style model for all-band arbitrary-date Landsat-Sentinel reflectance reconstruction.
44. Spatio-Temporal Evolution Analysis of Flood-Chain Disasters for Water Diversion Project Safety
Core Problem: Trace how flood-chain disasters evolve around water diversion project safety.
Key Innovation: Spatio-temporal framework for analyzing cascading flood impacts on diversion projects.
45. Probabilistic Parameter Inversion for Water-Soil Scour Using an SPH-Based Bayesian Updating Framework
Core Problem: Infer scour parameters under uncertainty in coupled water-soil erosion problems.
Key Innovation: SPH-based Bayesian updating for probabilistic water-soil scour inversion.
46. Influence of stagnant water effect on microstructural loess-mudstone interface under seepage conditions
Core Problem: Explain how stagnant water alters loess-mudstone interface microstructure under seepage.
Key Innovation: Microstructural analysis of seepage-driven weakening at a loess-mudstone interface.
47. Seismic Exploration for Magmatic Brine Lenses: An Example From Soufrière Hills Volcano, Montserrat
Core Problem: How to image subvolcanic brine lenses beneath Soufriere Hills.
Key Innovation: Joint P-wave and S-wave travel-time inversion uses velocity structure and hypocenters to infer a fluid-saturated lens at 2-3 km depth.
48. Pore Occupancy of Ice
Core Problem: Which pores ice, water, and gas occupy in frozen soils and sediments at equilibrium.
Key Innovation: Free-energy minimization shows ice prefers intermediate pores and yields tractable pressure relationships in three-phase frozen media.
49. CoMVS-GS: Collaborative Multi-View Stereo and 3D Gaussian Splatting for Surface Reconstruction
Core Problem: Stabilize mesh reconstruction from Gaussian splatting in weakly observed and occluded regions.
Key Innovation: Couples dense multi-view stereo initialization and mutual supervision with Delaunay graph-cut meshing for better geometry.
50. A Critical Synthesis of Uncertainty Quantification and Foundation Models for Semantic Segmentation
Core Problem: Foundation-model segmentation lacks trustworthy uncertainty under out-of-domain conditions.
Key Innovation: Systematic UQ benchmark on a SAM2-based segmentation backbone across in- and out-of-domain settings.
51. GrabVG: Graph-Attentive Binding for Visual Grounding in UAV Imagery
Core Problem: UAV grounding suffers from dense small objects and ambiguous repeated spatial layouts.
Key Innovation: Two-stage grounding that filters object hypotheses then reasons over a sparse graph with language-guided attention.
52. Regularised Iterative Generalised Least Squares with Optimal Selection of the Hyper-Parameter for Identifying Nonlinear Phenomenological Models
Core Problem: Confounded parameters and correlated noise make quasi-physical phenomenological models unstable to fit.
Key Innovation: Regularized iterative GLS with automated ridge hyperparameter selection from information-theoretic criteria.
53. Dual Modality Prompted Diffusion Priors for Zero Shot Hyperspectral Pansharpening
Core Problem: Reconstruct high-resolution hyperspectral imagery from PAN and low-resolution hyperspectral inputs without losing spectral fidelity.
Key Innovation: Injects spectral and spatial prompt tokens into a frozen diffusion prior and adds PAN-guided structural regularization.
54. Synergistic Use of Time-Series Multispectral and Synthetic Aperture Radar Data for Adaptive Glacier Surface-State Mapping
Core Problem: Map glacier wet-snow extent and surface-state evolution despite cloud contamination and SAR ambiguity.
Key Innovation: Uses optical wet-snow probabilities to guide adaptive SAR thresholding, then stabilizes results with temporal and spatial refinement.
55. STG-Net: Shared Tensor Guidance Hyperspectral and Multispectral Remote Sensing Image Fusion Network Under Locally Overlapping Condition
Core Problem: Fuse hyperspectral and multispectral imagery when the two modalities overlap only locally.
Key Innovation: Learns a shared deep Tucker core in the overlap region and propagates it scene-wide with multiscale attentive synthesis.
56. UniRS-Instruct: A Principle-Guided Unified Instruction-Following Dataset for Remote Sensing Understanding
Core Problem: Overcome the lack of unified instruction-following supervision for multimodal remote-sensing foundation models.
Key Innovation: Creates a unified dataset spanning captioning, VQA, grounding, and region descriptions through hierarchical prompting.
57. CoastMamba: A Boundary-Enhanced Mamba Framework for Sea-Land Segmentation in Optical Remote Sensing Imagery
Core Problem: Segment sea-land boundaries in complex coastal imagery with blurred and highly variable coastlines.
Key Innovation: Mamba-based feature gating plus frequency-domain boundary enhancement and a new high-resolution coastline dataset.
58. HARC-Net: Hierarchical Multiaxis Representation and Adaptive Residual Calibration for End-to-End SAR-to-Optical Image Translation
Core Problem: Translate noisy SAR imagery into structurally faithful optical-like imagery without deformation or artifact amplification.
Key Innovation: Hierarchical sparse transformer encoder and statistics-guided skip calibration improve noise-resilient SAR-to-optical translation.
59. RDSRNet: A Recursive Decoupling Super-Resolution Network for Meteorological Remote Sensing Data
Core Problem: Super-resolve meteorological radar and satellite data without losing small intense convective features.
Key Innovation: Recursive structural-detail decoupling network explicitly separates broad background structure from high-value storm detail.
60. Remote Sensing Lithology Interpretation via a Response-Optimized Structure-Aware Multimodal Fusion Network
Core Problem: Interpret rock and soil types from multimodal remote-sensing data despite inconsistent modality responses.
Key Innovation: Responsiveness-aware channel substitution and dynamic structure-aware fusion with SAR support improve lithology interpretation.
61. A multi-decadal global Landsat-derived dataset of forest fire patches from 1984-2022
Core Problem: Provide global long-term information on the size, shape, and spatial organization of forest fire patches.
Key Innovation: Builds a 30 m global Landsat-derived fire-patch dataset from 1984 to 2022 containing nearly 12 million patches.
62. Spatial-Spectral Decoupling-Enhanced Lightweight Network for Few-Shot Hyperspectral Anomaly Detection in Remote Sensing Imagery
Core Problem: Few-shot hyperspectral anomaly detection overfits because anomalies are sparse and labels are scarce.
Key Innovation: Uses lightweight spatial-spectral decoupling and projection-gating fusion to reduce capacity and overfitting.
63. GCF-Net: Stage-Aligned Optical-Elevation Fusion for Aerial Remote Sensing Semantic Segmentation
Core Problem: Cross-modal inconsistency and decoder attenuation limit optical-elevation semantic segmentation.
Key Innovation: Stage-aligns correction, fusion, and geometry-aware refinement across the encoder-decoder pipeline.
64. Multiple Vegetation Indicators Reveal Contrasting Post-Drought Recovery Time in the Yangtze River Basin Following the 2022 Extreme Drought
Core Problem: Single vegetation indices inadequately capture post-drought recovery timing across ecosystems.
Key Innovation: Compares NDVI, EVI, LAI, GPP, and SIF to map contrasting recovery times after the 2022 Yangtze drought.
65. Asynchronous Responses of Ecosystem Carbon Gain and Groundwater Storage Under Ecological Restoration in the Loess Plateau
Core Problem: Restoration gains in carbon uptake may mask groundwater depletion and delayed hydrologic costs.
Key Innovation: Jointly analyzes GPP, ET, soil moisture, and GRACE groundwater trends to show asynchronous restoration outcomes.
66. Historical legacies and contemporary mismanagement aggravated the 2015-2020 drought impacts in Qwaqwa, South Africa
Core Problem: Explain why drought impacts worsened in Qwaqwa beyond climatic forcing alone.
Key Innovation: Links historical legacies and contemporary mismanagement to observed drought impacts.
67. Provenance fingerprinting in open systems: compositional tracer selection and the limits of geochemical invariance revealed by mineralogical benchmarking
Core Problem: Geochemical provenance fingerprinting fails when open-system alteration breaks tracer invariance.
Key Innovation: Compositional-data tracer selection benchmarked against heavy-mineral endmember modeling.
68. Domain-knowledge-guided deep learning for probabilistic precipitation downscaling over southeastern Tibetan Plateau
Core Problem: Produce probabilistic high-resolution precipitation estimates over complex Tibetan terrain.
Key Innovation: Domain-knowledge-guided deep learning for probabilistic precipitation downscaling.
69. A study on sediment forecasting using machine learning with structure-aware mechanisms
Core Problem: Improve sediment forecasting by respecting structural relationships in the data.
Key Innovation: Machine-learning framework with structure-aware mechanisms for sediment prediction.
70. Ecohydrological thresholds of alpine meadow slopes: Quantifying resilience limits under a warming-wetting climate
Core Problem: Quantify resilience limits of alpine meadow slopes under warming-wetting climate change.
Key Innovation: Ecohydrological threshold framework for slope resilience under climate forcing.
71. Projected depletion of high mountain Asia glaciers by 2100: linking mass loss to asynchronous peak runoff
Core Problem: Link glacier mass loss trajectories to asynchronous peak runoff across High Mountain Asia.
Key Innovation: Regional projection connecting glacier depletion timing to runoff transitions.
72. Numerical analysis of the effects of radial centrifugal acceleration on centrifuge modelling of steady-state seepage in embankments
Core Problem: Obtain reliable unsaturated hydraulic characteristics and understand centrifuge seepage-modeling effects.
Key Innovation: Centrifuge permeameter methodology linked to steady-state seepage interpretation for embankment problems.
73. A mesh resolution criterion for the apparent heat capacity method in multidimensional freezing soil modelling
Core Problem: Apparent heat-capacity freezing models can silently skip latent heat when meshes are too coarse.
Key Innovation: A priori mesh and time-step criteria for multidimensional freezing-soil simulations.
74. Impacts of Particle Density and Relative Submergence on the Sediment Entrainment Process: An Experimental Study Using a Smart Sediment Particle
Core Problem: How particle density and relative submergence affect threshold entrainment.
Key Innovation: Uses an instrumented smart sediment particle to capture onset accelerations at incipient motion.
75. Safe Domain Adaptation for Physics: Overcoming Nuisances, Label Shifts, and Simulation Priors
Core Problem: How to adapt simulation-trained models when nuisance, physics, and target-label distributions all shift.
Key Innovation: Adaptive reweighting isolates physical mismatch and avoids bias from simulation-prior spectrum alignment.
76. Bound-Aware Per-Organ Recall Risk Control for Multi-Organ CT Segmentation under Clinical Domain Shift
Core Problem: How to maintain per-class recall guarantees when segmentation models transfer across clinical domains.
Key Innovation: Compares RCPS and CRC under domain shift and uses WSR betting bounds to reduce local recertification cost.
77. When Does Dynamic Ensembling Pay Off? Diagnosing Regionwise Gains in Regression under Distribution Shift
Core Problem: Predict when regionwise dynamic ensembling beats the best static blend under target shift.
Key Innovation: Small labeled probes estimate realizable regionwise gain and gate deployment of dynamic ensembles.
78. From Inference to Adaptation: A Unified Optimal Transport View of Vision Language Model
Core Problem: How to make VLM test-time adaptation robust when pseudo-labels are noisy under shift.
Key Innovation: Unifies inference and adaptation through sample-level Wasserstein optimal transport pseudo-labels and OT-equivalent InfoNCE training.
79. Depth Anything V4: Dynamic 4D Scene Reconstruction via Riemannian Flow Matching on 4D Gaussian Splatting
Core Problem: Improve monocular dynamic 4D scene reconstruction while keeping Gaussian parameters physically valid.
Key Innovation: Applies Riemannian flow matching directly on 4D Gaussian-splatting manifolds and isolates its gain from optimization and pretraining.
80. MorphoGP: A Nonparametric Framework for Predicting Equilibrium Beach Profiles Under Tidal Influence
Core Problem: Beach-profile prediction under tidal influence is nonlinear and site dependent.
Key Innovation: Category-specific Gaussian-process experts gated by learned beach morphology classes.
81. SPARC: Subspace Position-Aware Robust Few-Shot Calibration for Distribution-Shifted Industrial Anomaly Detection
Core Problem: Anomaly detectors fail after illumination or sensor shifts when only a few normal target images are available.
Key Innovation: Closed-form per-cell subspace projection using a handful of normal images and no retraining.
82. CDGP: Contrastive Dual Gaussian Processes for Weakly Supervised Anomaly Segmentation
Core Problem: Normality-only anomaly models confuse true defects with unusual but valid regions.
Key Innovation: Dual Gaussian processes with posterior-dominance statistics plus reconstruction residuals.
83. Evaluation of Image Matching Methods for Visual Odometry on UAVs
Core Problem: Need robust image matching for downward-looking UAV visual odometry when GNSS fails.
Key Innovation: Benchmark of modern image matchers inside a UAV VO setup showing both RoMa gains and strong SIFT baselines.
84. SAM2Dual: Training-Free, Dual Memory for Long-Term Video Object Segmentation
Core Problem: SAM2 memory drifts during long occlusions and scene changes.
Key Innovation: Dual short- and long-term memory with text-aware reweighting for training-free long-video segmentation.
85. The Impact of CutMix on Reliability and Robustness in Semantic Segmentation
Core Problem: The reliability effects of CutMix on dense prediction are poorly understood.
Key Innovation: Systematic in- and out-of-domain study showing CutMix chiefly improves calibration and uncertainty quality.
86. CL4D: Contrastive Language-4D Pretraining for Vision-Language Reasoning in Dynamic Scenes
Core Problem: Existing encoders do not jointly model geometry and motion in dynamic point clouds.
Key Innovation: Contrastive language-4D pretraining and a 4D VLM operating directly on dynamic point clouds.
87. Multi-stage neural operator learning with application for convolutions
Core Problem: Standard one-shot neural operators underuse staged basis enrichment for accurate convolution operators.
Key Innovation: Multi-stage supervised and weak-form neural-operator schemes that iteratively learn residual bases to near machine precision.
88. Simple, Safe, and Overlooked: Reclaiming Sustainable Domain Generalization with Statistical Color Matching
Core Problem: Existing domain-generalization augmentations either lack diversity or hallucinate structure.
Key Innovation: Training-free RGB mean-std color matching that preserves structure while broadening domain coverage.
89. Score the Algebra, Not the Span: Dimension Reduction for Transfer Operator Models of Dynamical Systems
Core Problem: Rank-based spectral reductions can entirely mask weakly interacting dynamical components.
Key Innovation: Scores generated sigma-algebras instead of spans, preserving full operator spectrum with few coordinates.
90. ReWEIGH the Evidence: Calibrating Token-Level Ordinal Visual Evidence to Mitigate Hallucinations in Large Vision-Language Models
Core Problem: LVLMs hallucinate unsupported objects because visual-token evidence is poorly calibrated across positions.
Key Innovation: Training-free decoding penalty based on token-specific ordinal visual evidence referenced to unlabeled-image baselines.
91. SPK: Eliciting Structured Prior Knowledge for Interpretable Out-of-Distribution Detection in Real-Time Object Detection
Core Problem: Detectors hallucinate confident predictions on out-of-distribution objects and hide the priors behind those errors.
Key Innovation: Elicits part-level semantic, geometric, and contextual priors into a compact interpretable five-dimensional OoD representation.
92. Detecting Backdoors in Object Detection via Pre-NMS Prediction Distribution Shift
Core Problem: Scene-level backdoors in object detectors evade prior architecture-specific or trigger-inversion defenses.
Key Innovation: Detects backdoored models from clean-data shifts in pre-NMS class distribution relative to training frequencies.
93. Pretraining Reusable Inference Across Views with Synthetic Task Priors
Core Problem: Downstream tasks repeatedly relearn how to combine heterogeneous views and cope with missingness.
Key Innovation: Synthetic task prior plus a hierarchical in-context learner that pretrains reusable within-view and cross-view reasoning.
94. Discretizing Continuous Time Series for Imputation with Masked Diffusion Training
Core Problem: Continuous-value diffusion imputers mix missing and observed tokens and learn noise instead of the signal.
Key Innovation: Masked diffusion imputation with a structurally separate MASK token and stochastic discretization of continuous values.
95. Leaf Values as Coordinates: Exact Contrastive Explanation for Gradient-Boosted Ensembles
Core Problem: Contrastive explanations for tree ensembles are usually approximate and hard to audit.
Key Innovation: Reinterprets leaf values as coordinates, yielding exact contrastive explanations and recourse in model space.
96. L\'evy Attention: Single-Pass Predictive Uncertainty for Continuous-Time Attention
Core Problem: Continuous-time attention models make predictions at arbitrary times but do not quantify trust.
Key Innovation: Poisson-measure attention layer that emits closed-form evidence and predictive disagreement in one deterministic pass.
97. Image-Guided Pavement Defect Recognition in GPR Data with novel 3D Deep Learning Architecture
Core Problem: Automated 3D GPR pavement inspection lacks scalable labels and architectures tailored to volumetric radar data.
Key Innovation: Uses aligned RGB orthomosaics to transfer labels into 3D GPR volumes and trains a specialized attention-enhanced 3D CNN.
98. You Are What You Prompt: Prompt Quality, Domain Shift, and Uncertainty in Agrifood Vision-Language Models
Core Problem: Zero-shot prompt ensembling behaves unpredictably under distribution shift in specialized vision-language domains.
Key Innovation: Shows domain-specific prompt pools help under shift and uses prompt disagreement as epistemic uncertainty.
99. Seasonal false alarms in customer churn and decline early-warning systems: adjacent-window labels confound seasonality with decline, and a year-over-year correction
Core Problem: Adjacent-window labels turn seasonal cycles into false decline alarms.
Key Innovation: Year-over-year aligned labels that remove anchor-month confounding and cut false alerts.
100. Evaluating Structured Information Extraction with Open Models in a High Risk Public Sector Application
Core Problem: Open OCR/LLM/VLM pipelines lack realistic end-to-end evaluation for high-risk document extraction.
Key Innovation: Benchmark showing VLM advantages and strong dependence on OCR structure preservation.
101. Debiased Inference for AI-Generated Data without Gold-Standard Labels: Identification via Multiple Imperfect Measurements
Core Problem: AI-measured variables bias downstream inference when gold labels are unavailable.
Key Innovation: Multiple-imperfect-measurement framework that debiases inference without gold-standard labels.
102. Low-Power, Neuromorphic, Acoustic Anomaly Detection for Persistent Machine Monitoring
Core Problem: Persistent anomaly monitoring is limited by power and deployment constraints.
Key Innovation: Neuromorphic on-chip acoustic autoencoder delivering real-time detection at very low energy.
103. Inference and Uncertainty Quantification for Streaming r-PCA
Core Problem: Streaming PCA lacks sharp general-rank convergence and principled uncertainty quantification.
Key Innovation: Sharp Oja convergence plus Gaussian approximation and online bootstrap for subspace error.
104. Change Point--Aware Evaluation and Re-Calibration of PPG-Based Blood Pressure Estimation
Core Problem: Aggregate evaluation hides model failure during rapid state changes.
Key Innovation: Change-point-triggered evaluation and recalibration rather than static interval-wide metrics.
105. Sharper Regret Bounds for Time-Varying Gaussian Process Bandits with Constant Exploration
Core Problem: Time-varying GP bandits usually require overly conservative exploration growth.
Key Innovation: Constant-exploration GP-UCB analysis with sharper regret and information-gain bounds under drift.
106. Robust Risk Under Evolving Uncertainty: A Wasserstein Counterpart of the Entropic Value-at-Risk
Core Problem: Entropy-based robust risk misses catastrophes that nominal models assign zero probability.
Key Innovation: Wasserstein entropic value-at-risk and a belief-entropy-driven robust dynamic-programming operator.
107. Contrasting Cost-Agnostic and Cost-Sensitive Losses under Limited Model Capacity via H-consistency
Core Problem: It is unclear when cost-sensitive training truly beats cost-agnostic surrogates under limited model capacity.
Key Innovation: H-consistency analysis proving a strict gap favoring cost-sensitive objectives in finite-capacity settings.
108. DeGLIF for Label Noise Robust Node Classification using GNNs
Core Problem: Noisy node labels degrade GNN classification and noise level is often unknown.
Key Innovation: Leave-one-out influence-based node denoising and relabeling without assuming a noise model.
109. An Edge-Aware and Noise Resilient CNN-ViT Hybrid Network With Cross-Attention Feature Fusion for PolSAR Land Cover Classification
Core Problem: Classify PolSAR land cover while suppressing speckle and preserving boundaries.
Key Innovation: CNN-ViT hybrid couples edge enhancement, residual shrinkage denoising, and cross-attention local-global fusion.
110. GLUSE: Enhanced Channelwise Adaptive Gated Linear Units SE for Onboard Satellite Earth Observation Image Classification
Core Problem: Deliver accurate satellite image classification under tight onboard power, memory, and latency constraints.
Key Innovation: GLUSE dynamic channel attention yields high accuracy with much lower compute and power than heavier baselines.
111. C3Former: A Spectral-Spatial Fusion Transformer for Hyperspectral Landcover Classification
Core Problem: Model spectral-spatial dependencies for hyperspectral land-cover classification under limited labels.
Key Innovation: Fusion transformer combines adaptive multiscale spatial pooling with cross-band spectral attention and cross-layer interaction.
112. Machine Learning Models for Field-Scale Soil Moisture Mapping via UAS-Based GNSS-R
Core Problem: Estimate high-resolution soil moisture from multimodal UAS observations across crops, seasons, and validation regimes.
Key Innovation: Fuses GNSS-R, multispectral, LiDAR, and temporal features with ML under strict spatial, temporal, and cross-crop tests.
113. Remote Sensing Image Destriping via Adaptive Huber-Enhanced Anisotropic TV With Statistical Feature Constraints
Core Problem: Remove spatially varying stripe noise without destroying real remote-sensing structure.
Key Innovation: Adaptive Huber-enhanced anisotropic TV with statistical guidance and reweighted block sparsity for real-image destriping.
114. Jointly Enhancing Representation and Supervision for Semisupervised Rotated Object Detection of Remote Sensing Images
Core Problem: Improve rotated object detection when labeled remote-sensing data are scarce.
Key Innovation: Dynamic feature modulation and dense feature complement add rotation-sensitive representation and supervision.
115. Cross-Modal Consistency-Guided Information Fusion for Hyperspectral and LiDAR Classification
Core Problem: Classify HSI-LiDAR land cover under few labels and strong cross-modal heterogeneity.
Key Innovation: Cross-modal consistency guidance, pseudo-sample screening, and dual-gradient MRF refinement improve fusion under limited labels.
116. IrrEM (Irrigation in the eastern Mediterranean): Two and a half decades of seasonal high-resolution irrigated land maps of the eastern Mediterranean
Core Problem: Fill the lack of long-term high-resolution irrigation maps in the drought-prone eastern Mediterranean.
Key Innovation: Generates 26 years of seasonal irrigated-land maps from satellite observations tracking soil-moisture and irrigation patterns.
117. Multiscale Estimation of Mangrove Biomass in Fujian Province Using UAV as a Bridging Scale
Core Problem: Field plots and satellite pixels are mismatched for regional mangrove biomass estimation.
Key Innovation: Uses UAV-derived bridging labels to link plot measurements and satellite biomass models.
118. Evaluating Seasonal Fidelity and Cross-Site Structural Discrimination of Sentinel-2 LAI Products in Karst Forests
Core Problem: Satellite LAI products can track seasonality yet fail to preserve true structural differences across heterogeneous karst stands.
Key Innovation: Separates within-site seasonal fidelity from cross-site structural discrimination when evaluating Sentinel-2 LAI products.
119. Midday Depression and Legacy Effect Disrupt SIF-GPP Coupling in Northern Peatlands During Combined Heat and Drought Stress
Core Problem: SIF and GPP can decouple under combined heat and drought, undermining remote stress monitoring.
Key Innovation: Shows midday and legacy-stress breakdowns in SIF-GPP coupling using high-frequency measurements.
120. AERO: Arbitrary-Scale Equivariant Resolution Operator for Remote Sensing Image Super-Resolution
Core Problem: Fixed-scale super-resolution cannot flexibly reconstruct arbitrary-scale remote-sensing imagery.
Key Innovation: Combines orientation-aware features, wavelet high-frequency modeling, and implicit terrain coordinates for arbitrary-scale super-resolution.
121. A Balanced Spectral-Spatial Cross-Fusion Network for Hyperspectral Anomaly Detection
Core Problem: Anomaly detectors struggle to balance spectral and spatial cues under complex backgrounds.
Key Innovation: Introduces bidirectional spectral-spatial cross-attention with multi-scale gated refinement in an autoencoder.
122. Unsupervised Gaussian-Noise-Robust Remote Sensing Change Detection via FRFCM-IRM Change Intensity Modeling and SEEDSAM-Constrained HCRF
Core Problem: Gaussian noise destabilizes unsupervised change detection in bi-temporal remote-sensing images.
Key Innovation: Combines noise-robust fuzzy clustering, region matching, and SEEDSAM-constrained HCRF labeling.
123. Dual-Domain Illumination Prior for Low-Light Remote Sensing Image Enhancement
Core Problem: Low-light remote-sensing scenes lose terrain detail and prior methods use only one domain of illumination information.
Key Innovation: Fuses frequency-domain and spatial-domain illumination priors in a trainable enhancement module.
124. Improving Land Surface Emissivity for Better Simulation of Microwave Radiances over Northern Latitudes
Core Problem: Land-surface emissivity assumptions degrade simulation of surface-sensitive microwave radiances at high latitudes.
Key Innovation: Compares Lambertian and specular reflection in a regional model to improve emissivity retrieval and radiance simulation.
125. Construction and Validation of a High-Fidelity Virtual Scene for Low-Stature and High-Biodiversity Ecosystems-Simulating Multi-Modal Sensing Approaches
Core Problem: Sensor limits hinder resolving small structurally complex vegetation targets.
Key Innovation: Physics-based 3D virtual scene linking TLS, SfM, radiative transfer, and simulated multisensor outputs.
126. UNet-DFH: A Semantic Segmentation Network Combining Multi-Scale Edge Fusion and Attention-Deformable Modules for Sugarcane Mapping in Heterogeneous Karst Regions
Core Problem: Fragmented fields, terrain complexity, and cloud cover weaken mapping in karst landscapes.
Key Innovation: UNet-DFH combines optical-SAR fusion with edge-preserving and deformable attention modules.
127. Dynamics of Forest Disturbance in the Canopy of Permanent Production Forests: A Multitemporal Analysis (2004-2025) Using Spectral Unmixing in the Southeastern Peruvian Amazon
Core Problem: Medium-resolution optical mapping misses understory degradation and understates disturbance.
Key Innovation: NDFI plus stratified area estimation for long-term area-corrected disturbance trajectories.
128. Predicting LiDAR-Derived Canopy Leaf Area Index in Loblolly Pine Plantations with Sentinel-2 Imagery Using a Convolutional Neural Network Approach
Core Problem: Low-cost frequent CLAI estimation is needed without repeated LiDAR acquisition.
Key Innovation: CNN regression from Sentinel-2 to LiDAR-derived CLAI with explicit domain-generalization testing.
129. Integrating Landsat-based irrigation mapping and actual evapotranspiration estimates with hydrological modelling to assess regional irrigation dynamics in a large basin experiencing extreme climate variability: Towards a national operational irrigation water accounting system
Core Problem: Track basin-scale irrigation dynamics under extreme climate variability.
Key Innovation: Integrates Landsat irrigation maps, evapotranspiration, and hydrological modeling for operational accounting.
130. Satellite embeddings for crop type classification: a comparative examination
Core Problem: Compare embedding strategies for crop type classification from satellite data.
Key Innovation: Systematic comparative examination of satellite representation learning for classification.
131. Integrating scale-matched spatial scales into water scarcity risk screening in China
Core Problem: Match analysis scale to drivers of water scarcity risk across China.
Key Innovation: Scale-matched framework for spatial water-scarcity risk screening.
132. Elevation-dependent hydrological vulnerability of lakes and reservoirs to evaporation loss in the Yarlung Zangbo River Basin, Tibetan Plateau
Core Problem: Assess how elevation controls evaporation losses from lakes and reservoirs.
Key Innovation: Elevation-dependent vulnerability analysis for Tibetan water bodies.
133. Multi-criteria seismic design for Fe-SMA slip-friction hybrid self-centering braced RC frames considering multi-stage yielding mechanism
Core Problem: Design self-centering braced RC frames under multiple seismic performance criteria.
Key Innovation: Fe-SMA slip-friction hybrid design accounting for multi-stage yielding.
134. A singular value-based secular function for artifact-free multimodal surface wave inversion
Core Problem: Recover shallow shear-wave velocity without dispersion artifacts in challenging multimodal data.
Key Innovation: Electroseismic Scholte-wave analysis and improved dispersion inversion for shallow sediments.
135. Hydro-mechanical behavior of coarse-grained red-stratum mudstone fillers considering effects of particle breakage
Core Problem: Assess particle-breakage effects on hydro-mechanical behavior of coarse-grained red-stratum mudstone fillers.
Key Innovation: Coupled hydro-mechanical characterization of breakage-sensitive mudstone filler material.