TerraMosaic Daily Digest: Feb 18, 2026
Daily Summary
This digest synthesizes 147 selected papers spanning landslide initiation–runout mechanics, deformation monitoring, and engineering performance under multi-hazard loading. Nearly two-thirds of the set concentrates on either observing deformation (InSAR, LiDAR, UAV/TLS, feature tracking) or resolving process physics (granular rheology, pore-pressure evolution, fracture and liquefaction), underscoring a field that is tightening the loop between measurement and mechanism.
The most decision-relevant contributions pair physical constraints with uncertainty-aware inference: rainfall-trigger studies move beyond fixed “I–D lines” to probabilistic thresholding under measurement uncertainty; monitoring workflows push toward near-real-time updates in rapidly decorrelating terrain; and coupled models of sediment-laden flows, landslide-dam breaches, and coastal hazards increasingly report quantities that matter operationally—timing, runout, and loads—rather than descriptive maps alone.
Key Trends
- Uncertainty is becoming part of the trigger definition: rainfall thresholds and temporal prediction frameworks increasingly quantify how event delineation, gauge gaps, and rainfall uncertainty propagate into warning performance.
- Deformation monitoring is moving toward “always-on” time series: sequential multi-temporal InSAR, TLS-based 4D tracking, and large-deformation offset methods aim to maintain coverage where decorrelation and nonlinearity previously broke pipelines.
- Cascading mass-flow hazards are treated as coupled sediment–fluid systems: studies model flash-flood–to–debris-flow transitions, landslide-dam breach amplification, and sediment effects in tsunami-like waves with multi-phase numerics and calibrated experiments.
- Infrastructure-facing geotechnics is becoming more parameterized: vulnerability and performance are increasingly expressed through curves, capacity envelopes, and experimentally constrained constitutive behavior for buildings, tunnels, offshore foundations, and lifelines.
- Method papers emphasize trust over raw accuracy: interpretability (e.g., Shapley-style decompositions), causal structure, and distribution-shift robustness are being designed in from the start, reflecting the safety-critical setting of geohazard decisions.
Selected Papers
This digest features 147 selected papers from 899 deduplicated papers analyzed (out of 2588 raw papers scanned; 899 new papers after deduplication). Each paper has been evaluated for its relevance to landslide and broader geohazard research and includes links to the original publications.
1. A comparative analysis of AHP, FR, AHP‐FR and LR models for landslide susceptibility mapping in Sikkim Himalaya, India
Core Problem: Landslides are recurrent and destructive geohazards in mountainous regions like the Sikkim Himalaya, but existing landslide susceptibility mapping (LSM) studies often lack methodological advancement and sufficient validation in data-scarce environments.
Key Innovation: The study systematically evaluated deterministic, statistical, and hybrid models for LSM, finding that the hybrid AHP-FR model achieved the highest predictive accuracy (AUC=0.85) and spatial reliability, demonstrating that structured hybridization effectively reduces subjectivity and improves spatial predictability in data-limited regions.
2. Tephra seismites—Understanding seismic hazard of hidden faults by analyzing liquefied tephra layers in lakes
Core Problem: Assessing seismic hazards in regions with hidden or poorly expressed faults is a major challenge in paleoseismology due to the difficulty in constraining paleoearthquake recurrence intervals and magnitudes.
Key Innovation: Develops a methodology using computed tomography imaging to quantify "tephra seismites" (liquefaction structures in tephra layers) in lakes, demonstrating their direct relation to ground shaking from near-field fault ruptures, and uses this to constrain recurrence intervals and magnitudes of paleoearthquakes from hidden fault systems.
3. Landslides, knowledge, and shared learning for safer societies
Core Problem: Technical advances in landslide science do not consistently translate into safer decisions, in part because knowledge is fragmented across disciplines and unevenly shared with practitioners and communities.
Key Innovation: A perspective article that frames landslide risk reduction as a shared-learning problem, outlining how open knowledge exchange, co-production with stakeholders, and iterative training can improve preparedness and response.
4. Numerical modeling of high-magnitude mass movement in data-scarce Himalayan terrain: a study using Lagrangian based modified DualSPHysics tool
Core Problem: Accurately simulating complex granular debris flow behavior in data-scarce, steep Himalayan terrain using numerical models, particularly for high-magnitude events.
Key Innovation: Presented a high-resolution numerical simulation of the 2017 Kotrupi debris flow using the fully Lagrangian Smoothed Particle Hydrodynamics (SPH) method in DualSPHysics, employing a modified Herschel–Bulkley–Papanastasiou (HBP) rheological model with pressure-dependent yield stress. The simulation accurately reproduced key flow characteristics and offers a reproducible, open-source framework for hazard assessment.
5. Research on landslide evolution stage identification and prediction algorithm based on real-time monitoring data
Core Problem: Traditional real-time landslide monitoring relies solely on displacement thresholds, failing to capture dynamic evolution or predict displacement trends, thus limiting risk assessment.
Key Innovation: Proposes an automated framework for landslide displacement stage recognition and short-term risk prediction based on the three-stage theory. It includes a time-series segmentation algorithm for automatic stage identification, integrates real-time data with prior information for risk assessment, and combines sequential prediction algorithms with an improved tangent angle model for short-term forecasting, achieving high accuracy and enhancing early warning systems.
6. Optimising rainfall characteristics for determining landslide thresholds
Core Problem: Empirical rainfall thresholds for landslide triggering are highly sensitive to the definition of "rainfall event" (minimum inter-event time (MIT) and triggering event (TE)), leading to variability and uncertainty in threshold estimation, especially in data-limited regions.
Key Innovation: Develops a new framework using Bayesian inference (BI) and nonlinear least-squares (NLS) to evaluate how MIT and TE definitions affect rainfall threshold estimation in Event Rainfall Duration and Intensity Duration spaces. It finds BI-derived thresholds are more stable and both methods perform best at MIT = 48h, demonstrating how robust Bayesian methods can downscale global thresholds to data-scarce regions and improve landslide prediction.
7. Improved 4D feature-based deformation tracking for high-resolution real-time landslide and slope deformation monitoring based on terrestrial laser scanning
Core Problem: Real-time monitoring of terrain deformation is crucial for predicting geohazards, but existing methods need improvement in computational efficiency and robustness for challenging natural terrains.
Key Innovation: Presents an improved 4D feature-based approach for real-time deformation analysis using terrestrial laser scanning, focusing on detecting and tracking features on hillshade models. It integrates contour line analysis and feature detection (KAZE and ORB algorithms) to enhance computational efficiency and accuracy, demonstrating sub-millimeter precision in lab settings and robustness in natural landslide environments (Hochebenkar rock glacier).
8. Influence of building height and openings on the physical vulnerability of masonry buildings to debris flow
Core Problem: Accurate vulnerability assessments and quantitative risk analyses for buildings threatened by debris flow in mountainous regions are crucial but often lack detailed consideration of building-specific factors like height and openings.
Key Innovation: Proposes an improved vulnerability curve for masonry buildings to debris flow, based on over 180 field survey data, by using the ratio of debris flow depth to building height and explicitly considering the influence of building openings. It quantifies damage into six levels and demonstrates that the improved curves, which account for building height and openings, align more closely with field survey results, enhancing the accuracy of quantitative debris flow risk analysis.
9. Exploring the feasibility and challenges of AI-based rainfall-induced landslides prediction
Core Problem: The urgent need for a landslide early warning information system (LEWS) in regions prone to rainfall-induced landslides, and the challenges in developing effective AI-based predictive models due to class imbalance, data gaps, and resolution limits.
Key Innovation: Explores the feasibility of developing an AI-based LEWS using rainfall and soil moisture data, training various machine learning models for grid-level landslide prediction. XGBoost consistently outperformed others, achieving high precision, recall, F1-score, and AUC, demonstrating its effectiveness in minimizing false negatives crucial for early warning. The study also highlights challenges and deploys an open-source geo-portal for visualizing predictions.
10. Lithologically-constrained, machine learning-based temporal landslide prediction models using rainfall time series for the Benguet First Engineering District, Philippines
Core Problem: There is a substantial gap in global research concerning the temporal prediction of landslides, especially in disaster-prone countries with limited resources, and conventional models may not fully leverage spatial components like lithology.
Key Innovation: Developed a landslide prediction system using minimal rainfall and landslide data, incorporating a spatial component by subdividing data based on lithologic domains. Machine learning algorithms, particularly Random Forest, showed high performance (AUROC 82.7%), demonstrating that lithologically-constrained datasets enhance temporal landslide prediction models.
11. Hydrogeological Risk in Florence: Memory and Perception in the Context of Climate Change
Core Problem: A persistent gap between awareness and adaptive behavior regarding hydrogeological risk in Florence, influenced by collective and individual memory, and the need for differentiated communication and governance strategies.
Key Innovation: Integrating memory as an analytical dimension within risk perception studies using a mixed-methods approach (historical-literary analysis, large-scale questionnaire, statistical modeling) to show how direct memory influences preparedness and institutional trust, and how socio-demographic factors shape perception of floods and landslides.
12. An Interpretable Physics Informed Multi-Stream Deep Learning Architecture for the Discrimination between Earthquake, Quarry Blast and Noise
Core Problem: Reliable discrimination of tectonic earthquakes from anthropogenic quarry blasts and transient noise remains a critical challenge in single-station seismic monitoring.
Key Innovation: A novel Physics Informed Convolutional Recurrent Neural Network (PI CRNN) that embeds seismological domain knowledge into feature extraction. It uses a multi-stream architecture (Time Domain SincNet, MultiResolution Spectrogram, Physics Branch) with fusion and a bidirectional LSTM. Achieves 97.56% accuracy, outperforming baselines, and demonstrates perfect precision in noise rejection. Interpretability analysis confirms learning of distinct physical signatures.
13. Quantifying the Role of 3D Fault Geometry Complexities on Slow and Fast Earthquakes
Core Problem: Traditional models of slow slip events (SSEs) oversimplify fault geometry, failing to account for the observed segmentation and complexity of real subduction faults, thus limiting understanding of how these complexities influence slip behavior and the generation of both slow and fast earthquakes.
Key Innovation: 3D quasi-dynamic earthquake sequence simulations of parallel faults demonstrate that complex fault geometry can naturally generate both slow and fast earthquakes through evolving traction heterogeneities, identifying four distinct slip regimes and introducing a geometry-dependent metric to quantify fault interaction.
14. AIFL: A Global Daily Streamflow Forecasting Model Using Deterministic LSTM Pre-trained on ERA5-Land and Fine-tuned on IFS
Core Problem: Data-driven models for global streamflow forecasting often suffer from a performance gap when transitioning from historical reanalysis data to operational forecast products, hindering reliable flood preparedness and water resource management.
Key Innovation: AIFL, a deterministic LSTM-based model, addresses the reanalysis-to-forecast domain shift using a novel two-stage training strategy: pre-training on ERA5-Land and fine-tuning on IFS control forecasts. It achieves high predictive skill (median KGE' of 0.66, NSE of 0.53) on an independent test set and demonstrates exceptional reliability in extreme-event detection, providing a robust baseline for global hydrological forecasting.
15. Statistical-Geometric Degeneracy in UAV Search: A Physics-Aware Asymmetric Filtering Approach
Core Problem: Post-disaster survivor localization using UAVs faces a fundamental challenge from Non-Line-of-Sight (NLOS) propagation in collapsed structures, introducing non-negative ranging biases that cause 'Statistical-Geometric Degeneracy' in standard symmetric robust estimators.
Key Innovation: Proposes AsymmetricHuberEKF, a physically-grounded asymmetric filtering approach that explicitly incorporates the non-negative physical prior of NLOS biases. This, combined with a co-designed active sensing strategy, resolves SGD and significantly accelerates convergence in UAV search operations.
16. Prototype-scale simulation of key wave parameters and force induced by black tsunamis
Core Problem: Black tsunamis, characterized by high suspended sediment concentrations, pose serious threats to marine infrastructure, and current assessments may underestimate forces by neglecting sediment effects.
Key Innovation: Applies the FS3M multiphase flow model to simulate black tsunami impacts, demonstrating that suspended sediment can attenuate forces under non-breaking conditions but amplify them under breaking conditions, highlighting the need to include sediment effects in tsunami impact assessments.
17. A new DEM-integrated dual-branch swin transformer model for landslide detection along the China Sichuan-Tibet railway
Core Problem: Ensuring the safety of the Sichuan-Tibet Railway is challenging due to complex geological conditions and frequent geological hazards, necessitating effective landslide detection.
Key Innovation: A new DEM-integrated dual-branch Swin Transformer model was developed for landslide detection, specifically applied along the China Sichuan-Tibet railway to enhance safety.
18. Fatal debris avalanche on an anthropogenically disturbed, earthquake-perturbed slope during antecedent rainfall
Core Problem: Understanding the complex interplay of predisposing factors (anthropogenic disturbance, previous landslide, earthquake) and proximate triggers (antecedent rainfall) leading to a fatal debris avalanche on a specific, chronically susceptible hillslope.
Key Innovation: Integrated multi-source remote sensing (MAXAR, LiDAR, UAV, Sentinel-1 InSAR) and field observations to reconstruct pre-failure behavior and analyze the roles of anthropogenic modification, antecedent rainfall, and a moderate earthquake. Delineated three high-susceptibility sectors for targeted monitoring and stabilization, illustrating how antecedent rainfall governs timing on a progressively weakened slope.
19. Experimental study on spatial vibration characteristics and early damage detection of unstable rock mass
Core Problem: Unidirectional vibration-based methods for unstable rock mass damage detection are directionally dependent and insufficient for identifying multidirectional structural plane degradation.
Key Innovation: Proposed a three-dimensional approximate entropy (ApEn3D) algorithm to quantitatively characterize spatial motion states for damage identification in toppling-type unstable rock masses. Experimental results showed ApEn3D is negatively correlated with structural plane damage and is less influenced by crack evolution direction, making it more effective than dominant frequency for identifying structural plane degradation.
20. Seismic hazard assessment based on predominance of local earthquake sources: a methodological approach using PSHA applied to the main cities in Peru
Core Problem: The need for a deeper comprehension and representation of seismic hazard at a city-size level for assessing existing buildings and infrastructure, which often deviates from seismic design standards, especially in seismically active regions like Peru.
Key Innovation: Proposes a novel methodological approach for city-level seismic hazard assessment using uniform hazard spectra calculated through a comprehensive probabilistic seismic hazard assessment (PSHA) framework. It considers a vast earthquake catalog, multiple source models, ground motion prediction models, and return periods, evaluated with a logic tree-based weighting system accounting for soil classification and fault activity, revealing specific local fault segments influencing seismic activity.
21. AE Characteristics and Dynamic Damage Evolution of Fissured Sandstone Under Multi-level Cyclic Loading Using Wave-Velocity Tomography
Core Problem: Understanding the deformation response, crack evolution, and damage accumulation characteristics of pre-existing fissure sandstone under multi-level cyclic loading is crucial, along with developing real-time monitoring capabilities for rock failure.
Key Innovation: Investigated the dynamic damage evolution of fissured sandstone under cyclic loading using AE monitoring and wave velocity tomography, revealing the influence of fissure inclination on mechanical properties and establishing a constitutive equation for damage evolution, providing real-time monitoring capability for rock failure early warning systems.
22. Anchoring Mechanism of Tunnel Free Face Subjected to Dynamic–Static Combined Load
Core Problem: There is a need to understand the failure characteristics and anchorage mechanisms of tunnel face anchorage systems under combined dynamic and static loading to improve reinforcement effectiveness.
Key Innovation: Investigated the performance of different rock bolt types (conventional rebar vs. Ductile-Expansion) in anchoring tunnel free faces under dynamic-static combined loads, demonstrating that Ductile-Expansion rock bolts significantly enhance crack suppression, energy absorption, and confining pressure stability, reducing brittle fracture.
23. Stability Appraisal of Road Cut Slopes Along NH-109 Using Multiple Rock Mass Classification Systems
Core Problem: The vulnerability of road cut slopes in the Himalayan region to slope failures due to complex geology, jointed rock mass, and active tectonics, requiring a robust stability assessment.
Key Innovation: An integrative approach combining multiple empirical rock mass classification systems (RMRbasic, Q-Slope, SMR, CoSMR, ChSMR) and kinematic analysis to appraise the stability of road cut slopes, deriving correlations between methods, and constructing validated stability charts from integrated stability domains of CoSMR and Q-Slope.
24. When Moderate Storms Become Disasters: Operationalising the Co-Production of Flood Hazard in Iraq
Core Problem: Understanding why moderate storms lead to damaging floods in Iraq, beyond just hydroclimatic forcing, by considering the role of rapidly changing socio-environmental conditions, governance gaps, and reactive coping mechanisms.
Key Innovation: Operationalizing the co-production of flood hazard by integrating national-scale geospatial analysis, long-term precipitation data, flood event databases, and qualitative interviews to demonstrate how urban expansion, loss of natural drainage, infrastructure underinvestment, and governance issues transform moderate rainfall into disasters.
25. Conceptual clarification and systematic comparison of the PESERA and (R)USLE methodologies, applied in the NEMEA wine producing zone, Greece
Core Problem: Limited understanding and systematic comparison of the strengths and limitations of empirical (R-USLE) and physical (PESERA) models for estimating regional soil erosion rates, leading to discrepancies in risk assessment.
Key Innovation: A systematic comparison and conceptual clarification of the PESERA and (R)USLE methodologies applied to a wine-producing zone, revealing significant differences in estimated erosion rates (RUSLE overestimates compared to PESERA and conventional mapping) and identifying key limitations for future model improvement.
26. Effects of liquefaction on V-H-M capacity envelopes for monopile foundations of multi-megawatt offshore wind turbines for Indian coastlines
Core Problem: The impact of liquefaction on the combined vertical-horizontal-moment (V-H-M) capacity of monopile foundations for offshore wind turbines in seismically active Indian coastal regions is not adequately quantified, hindering efficient and site-specific design.
Key Innovation: Evaluates liquefaction potential for Indian coastal regions, investigates failure mechanisms under non-liquefiable and liquefied conditions, and assesses V-H-M capacity envelopes for monopile foundations, providing directly applicable design charts and reduction factors for liquefaction-prone offshore environments.
27. Effects of flow intensity and vibration–flow alignment angle on riprap protection around a vibrating monopile: An experimental study
Core Problem: The performance of riprap protection for offshore wind turbine monopile foundations against local scour under monopile vibration remains insufficiently understood.
Key Innovation: Experimentally investigates riprap stability and scour development under varying flow intensities and vibration–flow alignment angles, demonstrating the dominant role of the dynamically enlarged blockage area in controlling scour depth and proposing empirical equations for scour prediction.
28. Soil surface change data of high spatio-temporal resolution from the plot to the catchment scale
Core Problem: Limitations in process-based soil erosion models due to constraints in integrating novel data, uncertainties in parameterization, and difficulties in integrating different process scales.
Key Innovation: Presents a unique dataset of high-resolution spatio-temporal soil surface elevation changes during precipitation events, obtained through nested Structure from Motion (SfM) photogrammetry (plot to micro-catchment scale) over nearly four years, intended to enhance understanding and improve calibration/evaluation of soil erosion models.
29. Real-time coastal disaster monitoring system for typhoon-induced wave overtopping
Core Problem: The challenge of real-time observation and quantification of dynamic wave overtopping processes in coastal areas using traditional methods.
Key Innovation: Developed a novel automated binocular wave overtopping monitoring system based on Binocular Vision-Large Scale Particle Image Velocimetry (BV-LSPIV) integrated with a deep learning object detection algorithm. This system enables pixel-level identification of wave run-up/overtopping and calculates overtopping discharge in real-time, validated during a typhoon event.
30. Machine learning-driven flood hazard assessment: integrating SAR and elevation data for inundation mapping and depth estimation
Core Problem: The need for high-resolution flood hazard mapping and depth estimation, particularly in data-scarce regions, to support disaster preparedness and infrastructure planning.
Key Innovation: Presents a region-based dual machine learning framework for flood hazard mapping, combining flood-prone area classification (using bagged decision tree with 93.2% accuracy) and flood depth estimation (using bagged regression tree with RMSE 0.99m and R2 0.6). It integrates Sentinel-1 SAR imagery and DEM-derived conditioning factors, enabling high-resolution (30m) flood risk assessment and providing a valuable decision-support tool.
31. An integrated exploration framework for evaluating adverse geology and grouting quality underlying roadbeds
Core Problem: Karst-related hazards threaten transportation infrastructure stability, and single geophysical methods are often limited by interpretive non-uniqueness in evaluating adverse geology and grouting quality underlying roadbeds.
Key Innovation: Developed an integrated exploration framework combining electrical resistivity tomography (ERT), multi-channel analysis of surface waves (MASW), borehole drilling, and water pressure tests to evaluate adverse geology and grouting quality in karst regions. Proposed a synergistic strategy (ERT-anchored, MASW-supplemented, drilling-validated) to improve detection reliability and established a multi-parameter evaluation system for grouting effectiveness.
32. Evaluation on the internal erosion resistance of gap-graded sand reinforced by microbially induced carbonate precipitation
Core Problem: Internal erosion of gap-graded sand is a common cause of hydraulic infrastructure failures, necessitating effective methods to enhance erosion resistance.
Key Innovation: Demonstrated that Microbially Induced Carbonate Precipitation (MICP) significantly reduces hydraulic conductivity and enhances the erosion resistance of gap-graded sand, with the number of treatment cycles being the most impactful factor.
33. Present‐Day Tectonic and Non‐Tectonic Crustal Deformation of and Around the Indian Plate From GNSS and GRACE Measurements
Core Problem: The internal deformation of the Indian plate, particularly the interplay between tectonic forces, seasonal hydrological loading, and anthropogenic processes (like groundwater depletion), and its implications for earthquake activity and ground movement, are not fully characterized.
Key Innovation: By using continuous GNSS and GRACE measurements from over 1600 sites, the study quantified present-day tectonic and non-tectonic crustal deformation of the Indian plate, identifying regions of higher strain rates and earthquake productivity (e.g., Gujarat), uplift due to groundwater depletion, and the significant influence of hydrological loading on land movement.
34. The Influence of Tropopause Temperature Biases on Climate Model Simulations of Tropical Cyclones
Core Problem: Significant variability in tropical cyclone potential intensity (PI) across global climate models, even with identical sea surface temperatures, and an underappreciated role of upper atmospheric model biases in modulating TC activity.
Key Innovation: Demonstrating that differences in PI across GCMs are primarily driven by outflow temperature (a consequence of upper atmospheric temperatures) and showing that altering temperature profiles in idealized experiments significantly impacts global TC frequency, hurricane frequency, and lifetime maximum intensity.
35. Unprecedented Look at Lightning Propagation and Ground Attachment in Ultraviolet
Core Problem: Poor understanding of streamers' crucial role in lightning propagation and attachment, particularly because they primarily emit UV light which usually goes undetected, leading to mysteries and debates in the field.
Key Innovation: Presenting unique UV images of lightning, including interaction between positive and negative streamers, leading to new inferences on lightning propagation and attachment, observing an abrupt change in propagation path and revealing streamer zones are an order of magnitude longer than previously thought.
36. Characterization of Preferential Flow Occurrence During Freeze‐Thaw Cycles
Core Problem: Quantification and mechanisms of preferential flow (PF) in frozen soils remain poorly understood due to observational challenges, despite its critical influence on water and energy dynamics relevant to soil stability.
Key Innovation: Developed a novel method to identify PF in frozen soils by analyzing soil temperature response times across depths, revealing its significant role in energy transfer and identifying key spatial and temporal drivers related to soil pore structure and meteorological forcing.
37. Graph neural network for colliding particles with an application to sea ice floe modeling
Core Problem: Traditional numerical methods for sea ice modeling are computationally intensive and less scalable, hindering efficient forecasting of sea ice dynamics, particularly in marginal ice zones.
Key Innovation: The Collision-captured Network (CN), a Graph Neural Network (GNN) model that leverages the natural graph structure of sea ice and integrates data assimilation to efficiently learn and predict sea ice dynamics, accelerating simulations without compromising accuracy.
38. Multi-Class Boundary Extraction from Implicit Representations
Core Problem: Existing surface extraction methods from implicit neural representations are limited to single-class surfaces and do not guarantee topological correctness or water-tightness for multi-class scenarios, which are crucial for geological modeling.
Key Innovation: A 2D boundary extraction algorithm for multi-class implicit representations that guarantees topological consistency and water-tightness, allows for minimum detail restraint, and is validated using complex geological modeling data.
39. Generative deep learning improves reconstruction of global historical climate records
Core Problem: Historical instrumental climate data is sparse, fragmented, and uncertain, leading to conventional reconstructions that excessively smooth local features, create unphysical artifacts, and systematically underestimate intrinsic variability and extremes.
Key Innovation: Presents a unified, probabilistic generative deep learning framework that leverages a learned generative prior of Earth system dynamics to recover spatiotemporally consistent historical temperature and precipitation fields from sparse observations, preserving higher-order statistics and improving assessment of extreme weather events and climate variability.
40. Frequency-Aware Vision Transformers for High-Fidelity Super-Resolution of Earth System Models
Core Problem: Traditional deep super-resolution methods for Earth System Model outputs suffer from spectral bias, failing to adequately reconstruct valuable high-frequency details crucial for fine-scale climate science and localized hazard analysis.
Key Innovation: Introduces ViSIR (Vision Transformer-Tuned Sinusoidal Implicit Representation) and ViFOR (Vision Transformer Fourier Representation Network), two frequency-aware frameworks that mitigate spectral bias by combining vision transformers with sinusoidal activations or explicit Fourier-based filtering, achieving significant improvements in super-resolution of Earth System Model outputs.
41. Scour development around spudcan foundations of varying geometries
Core Problem: Understanding the influence of spudcan geometry and loading on local scour development around foundations is crucial for the stability and design of marine infrastructure.
Key Innovation: Experimentally investigates scour development around spudcan foundations, revealing that applied vertical load substantially accelerates scour, sharper spudcans generate deeper scour, and the bearing area ratio exhibits a two-phase trend, providing new insights for foundation design.
42. A new limit analysis framework for bearing capacity of layered coral sand foundations considering the influence of gradation and particle recombination
Core Problem: Assessing the bearing capacity and stability of layered coral sand foundations is challenging due to complex failure mechanisms influenced by gradational contrasts and particle recombination, impacting offshore and coastal infrastructure.
Key Innovation: Developed a novel upper-bound limit analysis framework that explicitly incorporates gradation and particle recombination effects through a multi-block failure mechanism and a new parameter (B_D), providing a robust tool for accurate foundation assessment and design.
43. Using seasonal forecasts to enhance our understanding of extreme wind and precipitation impacts from extratropical cyclones
Core Problem: Estimating the risk posed by extreme wind and heavy precipitation from extratropical cyclones (ETCs) for insurance and reinsurance companies, particularly for high-return-period storms, where observed data is limited.
Key Innovation: Utilizes nearly 700 years of extended wintertime seasonal forecast model output and the UNSEEN method to quantify the likelihood of unprecedented ETC impacts (wind and precipitation) across Europe, demonstrating the influence of the North Atlantic Oscillation (NAO) on wind impacts and providing data for high-return-period storm impact estimation.
44. A benchmark laboratory calibration dataset for tipping-bucket rain gauges: comparison of manual burette and automated methods
Core Problem: Reliable calibration data are essential for ensuring the accuracy and traceability of precipitation measurements from tipping-bucket rain gauges (TBRGs), and publicly available benchmark datasets comparing manual and automated calibration methods are scarce.
Key Innovation: Presents a benchmark laboratory calibration dataset for TBRGs, generated under controlled conditions using both manual burette and automated methods, including raw measurements, summary statistics, and comprehensive uncertainty evaluation.
45. Increasing synchronicity of global extreme fire weather
Core Problem: The increasing synchronicity of extreme fire weather globally creates conditions for widespread large fires, complicating fire suppression and degrading air quality, with over half the increase attributable to anthropogenic climate change.
Key Innovation: Identifies significant increases in intra- and interregional synchronous fire weather (more than twofold in most regions since 1979), attributes over half of this increase to anthropogenic climate change, and links SFW to regional fire-sourced PM2.5, highlighting growing challenges for firefighting and human health.
46. New shear constitutive model for rock discontinuities based on maximum entropy theory: performance analysis and engineering application
Core Problem: Conventional damage constitutive models for shear deformation of rock discontinuities often rely on empirically selected damage distributions, neglecting the intrinsic uncertainty associated with the distribution function, which limits their robustness and adaptability.
Key Innovation: Proposed a novel shear constitutive model for rock discontinuities based on a generalized distribution function derived from maximum entropy theory. The model accurately simulates shear deformation across all stages, showing superior accuracy compared to classical models, and offers a strengthened tool for geoengineers.
47. Development of a new method for evaluating the geological strength index (GSI) by applying image analysis to in-situ rock mass
Core Problem: Evaluating discontinuity spacing (Sd) and Rock Quality Designation (RQD) for the Geological Strength Index (GSI) using conventional methods is challenging due to variation with scanline orientation and lack of standardized criteria, complicating block size distribution assessment.
Key Innovation: Developed a new method for evaluating GSI by applying image analysis to in-situ rock mass. Analysis revealed a good correlation between GSI and the combination of the discontinuity condition rating (Rcd) and median block size (F50), leading to a simplified new relation for GSI assessment.
48. Experimental and numerical investigation of the mechanical behavior of lime-treated silty clay under various loading and temperature conditions
Core Problem: There is a need to evaluate the strength characteristics of lime-treated silty clay under various loading and temperature conditions, particularly in cold environments, for its use in slopes and foundations.
Key Innovation: Investigated the tensile, compressive, and shear strengths of lime-treated marine soils under varied loading rates and temperatures, developing a nonlinear relationship between strength parameters and temperature, and validating FEM models for predicting soil behavior and failure patterns in cold environments.
49. Key Factors Influencing the Dynamic Response of Pile Supported Railway Embankments under Train Loading
Core Problem: There is a need to understand the dynamic behavior of geogrid-reinforced pile-supported railway embankments under high-speed train loads and identify key factors influencing vibration acceleration for optimal design.
Key Innovation: Developed a validated 3D finite element model to investigate the dynamic response of pile-supported railway embankments under train loading, identifying embankment height as the greatest influence on ground vibrations and developing a regression model for forecasting peak accelerations to aid in optimal vibration control design.
50. Floating wood volume determination using non-contact measurements and RANSAC method
Core Problem: The challenge of accurately determining the volume of floating wood debris in rivers, which accumulates on hydraulic structures, reduces flow capacity, and increases the risk of flooding, erosion, and structural damage.
Key Innovation: Development of a non-contact measurement method using a 2D industrial laser scanner and camera system, combined with the Random Sample Consensus (RANSAC) algorithm, to accurately calculate the individual volumes of floating wood logs from point clouds, offering higher accuracy than previous estimation methods.
51. A phase field fracture model for rock creep: Theoretical framework and engineering applications
Core Problem: Existing models inadequately capture the coupled effects of tensile and compressive strain-induced fracture and creep in rocks, making it difficult to accurately predict rock failure patterns and creep-induced failure.
Key Innovation: Proposes a double phase field framework that separates tensile and compressive-shear cracks, accounts for mutual coupling between creep and fracture, and aligns with experimental observations and engineering practices for characterizing surrounding rock damage.
52. Seismic performance analysis of a subway station incorporating friction sliding isolation bearings with nonlinear restraining mechanisms
Core Problem: Central columns in subway stations are vulnerable to compression-shear failure during earthquakes, threatening the structure's overall seismic integrity.
Key Innovation: Proposed and validated friction sliding isolation bearings with nonlinear restraining mechanisms for subway station columns, demonstrating a two-stage stiffness response that reduces peak drift ratio by 21.1% and limits residual deformation, shifting seismic energy and deformation from columns to slabs.
53. Seismic pounding between adjacent base-isolated buildings considering soil-structure interaction and structural system
Core Problem: Adjacent base-isolated buildings face seismic pounding risk, especially on soft soil and with certain base isolator types, which can be exacerbated by soil-structure interaction and stiffness disparities between buildings.
Key Innovation: Compared seismic performance and pounding risk of different building systems (Moment Frame vs. Moment Frame + Steel Plate Shear Walls) with flexible and stiff base isolators, demonstrating that flexible base isolators are superior in controlling displacements and mitigating pounding, and that synchronized dynamic response of similar stiff structures minimizes pounding risk.
54. Distributed physics-informed neural networks via domain decomposition for fast flow reconstruction
Core Problem: Scaling Physics-Informed Neural Networks (PINNs) for flow reconstruction to large spatiotemporal domains is hindered by computational bottlenecks, optimization instabilities, and pressure indeterminacy in distributed solvers.
Key Innovation: Proposes a robust distributed PINNs framework using spatiotemporal domain decomposition, addressing pressure indeterminacy with a reference anchor normalization and decoupled asymmetric weighting, and accelerating training with CUDA graphs and JIT compilation, achieving near-linear strong scaling and high-fidelity reconstruction for complex hydrodynamics.
55. EarthSpatialBench: Benchmarking Spatial Reasoning Capabilities of Multimodal LLMs on Earth Imagery
Core Problem: Existing benchmarks for spatial reasoning on Earth imagery lack support for quantitative direction/distance reasoning, systematic topological relations, and complex object geometries beyond bounding boxes.
Key Innovation: Proposes EarthSpatialBench, a comprehensive benchmark with over 325K question-answer pairs for evaluating Multimodal Large Language Models' spatial reasoning on Earth imagery, covering qualitative/quantitative distance/direction, topological relations, and various object query types.
56. Muon with Spectral Guidance: Efficient Optimization for Scientific Machine Learning
Core Problem: Physics-informed neural networks (PINNs) and neural operators suffer from severe optimization difficulties due to ill-conditioned gradients, multi-scale spectral behavior, and stiffness.
Key Innovation: SpecMuon, a spectral-aware optimizer that integrates Muon's orthogonalized geometry with a mode-wise relaxed scalar auxiliary variable (RSAV) mechanism, adaptively regulating step sizes and controlling stiff spectral components for faster convergence and improved stability in PINNs.
57. Deep TPC: Temporal-Prior Conditioning for Time Series Forecasting
Core Problem: Existing LLM-for-time series methods treat time shallowly, limiting temporal reasoning as information degrades through layers, hindering long-term forecasting accuracy.
Key Innovation: Temporal-Prior Conditioning (TPC), which elevates time to a first-class modality by conditioning the model at multiple depths with learnable time series tokens and temporal embeddings, leading to state-of-the-art performance in long-term forecasting.
58. Rethinking Input Domains in Physics-Informed Neural Networks via Geometric Compactification Mappings
Core Problem: Physics-informed neural networks (PINNs) struggle with multi-scale PDEs due to geometric misalignment in fixed coordinate systems, leading to gradient stiffness and ill-conditioning.
Key Innovation: Geometric Compactification (GC)-PINN, a framework that reshapes input coordinates through differentiable geometric compactification mappings, introducing strategies for periodic boundaries, far-field scale expansion, and localized singular structures, resulting in improved accuracy, stability, and convergence for PINNs.
59. SEMixer: Semantics Enhanced MLP-Mixer for Multiscale Mixing and Long-term Time Series Forecasting
Core Problem: Efficiently modeling multiscale patterns for long-term time series forecasting is challenging due to redundancy, noise, and semantic gaps between non-adjacent scales.
Key Innovation: SEMixer, a lightweight multiscale model for long-term time series forecasting, featuring a Random Attention Mechanism (RAM) to enhance patch-level semantics and a Multiscale Progressive Mixing Chain (MPMC) for memory-efficient and effective temporal mixing across scales.
60. VETime: Vision Enhanced Zero-Shot Time Series Anomaly Detection
Core Problem: Existing time-series anomaly detection (TSAD) foundation models face a trade-off: 1D temporal models provide fine-grained localization but lack global context, while 2D vision-based models capture global patterns but suffer from information bottlenecks and coarse-grained detection.
Key Innovation: Proposes VETime, the first TSAD framework unifying temporal and visual modalities through fine-grained visual-temporal alignment and dynamic fusion, introducing a Reversible Image Conversion, Patch-Level Temporal Alignment, Anomaly Window Contrastive Learning, and Task-Adaptive Multi-Modal Fusion to achieve superior zero-shot anomaly localization precision with lower computational overhead.
61. Functional Decomposition and Shapley Interactions for Interpreting Survival Models
Core Problem: Standard additive explanation methods are fundamentally limited in interpreting non-additive hazard and survival functions in time-to-event prediction models.
Key Innovation: SurvFD, a principled approach for analyzing feature interactions in survival models by decomposing higher-order effects, and SurvSHAP-IQ, an estimator for higher-order, time-dependent Shapley interactions, providing interaction- and time-aware interpretability.
62. Optimal training-conditional regret for online conformal prediction
Core Problem: Online conformal prediction for non-stationary data streams subject to unknown distribution drift lacks robust performance guarantees, especially in terms of training-conditional cumulative regret.
Key Innovation: Split-conformal and full-conformal style algorithms that leverage drift detection to adaptively update calibration sets for online conformal prediction under non-stationarity, achieving minimax-optimal regret guarantees.
63. Channel Dependence, Limited Lookback Windows, and the Simplicity of Datasets: How Biased is Time Series Forecasting?
Core Problem: Evaluations in Long-term Time Series Forecasting (LTSF) are often biased due to arbitrarily set lookback windows and the simplicity/weak inter-channel correlations of standard benchmark datasets, leading to misleading performance rankings and suboptimal model choices.
Key Innovation: Empirically demonstrates the necessity of tuning lookback windows per-task, reveals that the apparent superiority of Channel-Independent (CI) models on benchmarks is an artifact of dataset simplicity, and shows that Channel-Dependent (CD) models truly excel on datasets with strong cross-channel dependencies, providing key recommendations for improving TSF research.
64. CARL: Camera-Agnostic Representation Learning for Spectral Image Analysis
Core Problem: Variability in channel dimensionality and captured wavelengths among diverse spectral cameras (RGB, multispectral, hyperspectral), which impedes the development of generalizable AI-driven methodologies for spectral image analysis and limits cross-camera applicability, especially in remote sensing.
Key Innovation: Introducing CARL, a model for Camera-Agnostic Representation Learning, featuring a novel spectral encoder with a self-attention-cross-attention mechanism and spatio-spectral pre-training, demonstrating unique robustness to spectral heterogeneity across various imaging modalities, including satellite imaging.
65. Still Competitive: Revisiting Recurrent Models for Irregular Time Series Prediction
Core Problem: Modeling irregularly sampled multivariate time series for prediction remains a persistent challenge, with unclear benefits of complex architectures over potentially simpler and more efficient RNN-based methods.
Key Innovation: GRUwE (Gated Recurrent Unit with Exponential basis functions), an RNN-based architecture that maintains a Markov state with observation-triggered and time-triggered resets, demonstrating competitive or superior performance for both regression-based and event-based predictions in continuous time for irregular time series.
66. Adaptive Sampling for Hydrodynamic Stability
Core Problem: Efficiently and accurately detecting bifurcation boundaries (stability thresholds) in high-dimensional parametrized fluid flow problems, which typically requires a large number of computationally expensive simulations.
Key Innovation: Develops an adaptive sampling approach that couples a classifier network (for bifurcation probability) with a flow-based deep generative model (KRnet) to automatically refine parameter space sampling, concentrating computational effort in regions of high uncertainty (Shannon entropy), thereby achieving accurate bifurcation boundary identification with significantly fewer Navier-Stokes simulations.
67. Numerical investigation of the interaction between waves and porous floating breakwater by the smooth particle hydrodynamics method
Core Problem: Porous floating breakwaters offer unique wave-dissipation mechanisms for coastal protection, but numerical tools, especially particle methods, for their detailed analysis and optimization are limited.
Key Innovation: Developed a new SPH-based wave–porous floating breakwater coupling model within the mixture theory framework, incorporating improved SPH discretization and particle shifting techniques, validated to accurately predict wave attenuation, motion responses, and mooring forces, and identifying optimal configurations for coastal protection.
68. A review of current best practices and future directions in assimilating GRACE/-FO terrestrial water storage data into numerical models
Core Problem: Improving water cycle reanalyses by effectively assimilating GRACE/-FO terrestrial water storage data into hydrological and land surface models remains challenging due to mismatches in spatial/temporal resolution and a lack of consensus on optimal assimilation approaches.
Key Innovation: A comprehensive review synthesizing best practices and future directions for assimilating GRACE/-FO data, highlighting effective strategies (e.g., ensemble Kalman filter, localization, error handling) and future avenues (low-latency products, enhanced observations, machine learning) to achieve more accurate and operationally useful water cycle reanalyses.
69. WTTnet: a network combining wavelet transform and transformer for denoising microseismic signal
Core Problem: Accurate denoising of microseismic signals is crucial for reliable data in microseismic monitoring, particularly for locating mining-related seismic events and analyzing rock mass stability.
Key Innovation: WTTnet, a novel network combining wavelet transform and Transformer, was proposed to effectively denoise microseismic signals by capturing cross-scale correlations, outperforming traditional methods in diverse noise conditions.
70. Effect of Strain Rate and Porosity on the Mechanical Response of Rocks
Core Problem: A comprehensive understanding of the combined effects of porosity and strain rate on the mechanical behavior of rocks under compressive and tensile loading is lacking.
Key Innovation: Quantified the combined effects of porosity and strain rate on rock strength and dynamic Young's modulus, showing that higher porosity reduces strength and strain rate sensitivity varies with loading mode, validating the Kimberley model for dynamic strength behavior.
71. A global framework for subsurface soil moisture estimation: Coupling fractal Richards equation with Bayesian optimization
Core Problem: The critical gap in global subsurface soil moisture (SM) monitoring, as current satellite missions only reliably measure near-surface SM, limiting comprehensive understanding of a key landslide trigger.
Key Innovation: A global, satellite-based framework (ExpF–FRE) that extends surface SM to 20 cm and 50 cm at 400 m daily resolution, by integrating the Exponential Filter with a fractal-diffusion representation and replacing the empirical transfer time parameter with a physically derived, pixel- and depth-specific timescale computed from globally available soil hydraulics and satellite inputs, validated against extensive in situ networks.
72. Deciphering the role of anisotropy in elastic incremental behavior and path-dependent shear stiffness of granular materials
Core Problem: The directional dependence of elastic incremental behavior and path-dependent shear stiffness in granular materials under anisotropic conditions is insufficiently quantified, hindering advanced constitutive model development.
Key Innovation: Uses DEM to systematically investigate elastic incremental responses and shear stiffness under various anisotropic stress states, establishing a relation between strain response envelopes, stiffness anisotropy, and fabric anisotropy, and proposing a unified formulation to predict normalized stiffness.
73. The influence of fabric and mineralogy on the dynamics of carbonate silty clays
Core Problem: The role of carbonate silt inclusions in clay matrices on the dynamic properties (shear modulus, damping) of soils, particularly for offshore wind farm foundations in regions like the Gulfs of Mexico and Texas, is not well defined.
Key Innovation: Utilizes artificial soils with controlled fabric and mineralogy (aragonite, calcite, silicate silt) to systematically investigate dynamic soil properties through triaxial and resonant column tests, elucidating how plasticity index and grading heterogeneity influence mechanical responses and providing insights into the dynamic behavior of carbonate silty clays.
74. A Comprehensive Survey on Deep Learning-Based LiDAR Super-Resolution for Autonomous Driving
Core Problem: Low-resolution LiDAR sensors produce sparse point clouds that miss critical details, posing challenges for autonomous driving and cross-sensor compatibility.
Key Innovation: Presents the first comprehensive survey of deep learning-based LiDAR super-resolution methods for autonomous driving, categorizing approaches, establishing fundamental concepts, and identifying future research directions for enhancing sparse point clouds.
75. Feature-based morphological analysis of shape graph data
Core Problem: The need for a computational pipeline to statistically analyze shape graph datasets, distinguishing variations in both connectivity structure and geometric differences of network branches, beyond traditional abstract graph analysis.
Key Innovation: Introduces a computational pipeline for feature-based morphological analysis of shape graph data, extracting a curated set of topological, geometric, and directional features with key invariance properties. This representation enables tasks like group comparison, clustering, and classification on cohorts of shape graphs, demonstrated on urban road networks, neuronal traces, and astrocyte imaging.
76. Geometric Neural Operators via Lie Group-Constrained Latent Dynamics
Core Problem: Neural operators suffer from instability and reduced accuracy in multi-layer iteration and long-horizon rollouts due to unconstrained Euclidean latent space updates that violate geometric and conservation laws of physical systems.
Key Innovation: MCL (Manifold Constraining based on Lie group), a plug-and-play module that enforces geometric inductive bias by constraining latent manifolds with low-rank Lie algebra parameterization, significantly improving long-term prediction fidelity for partial differential equations.
77. EasyControlEdge: A Foundation-Model Fine-Tuning for Edge Detection
Core Problem: Producing crisp raw edge maps with limited training samples remains challenging in real-world edge detection tasks, and the pretrained priors of image-generation foundation models are underexploited for this purpose.
Key Innovation: Proposes EasyControlEdge, an edge-specialized adaptation of image-generation foundation models that incorporates an edge-oriented objective with an efficient pixel-space loss and guidance based on unconditional dynamics, enabling crisp and data-efficient edge detection with controllable edge density.
78. SCAR: Satellite Imagery-Based Calibration for Aerial Recordings
Core Problem: Existing aerial visual-inertial systems suffer from calibration degradation over long-term field deployment, requiring dedicated maneuvers or manually surveyed ground control points, which limits their accuracy and robustness.
Key Innovation: SCAR, a method for long-term auto-calibration refinement of aerial visual-inertial systems, exploits georeferenced satellite imagery, orthophotos, and elevation models as a persistent global reference, significantly reducing reprojection and localization errors without manual intervention.
79. HPMixer: Hierarchical Patching for Multivariate Time Series Forecasting
Core Problem: Effectively capturing both periodic patterns and residual dynamics is essential for accurate long-term multivariate time series forecasting within standard deep learning benchmark settings.
Key Innovation: Proposes HPMixer, a framework that models periodicity and residuals in a decoupled yet complementary manner, utilizing a learnable cycle module for periodicity and a Learnable Stationary Wavelet Transform with a two-level hierarchical patching mechanism for residuals, achieving competitive or state-of-the-art performance.
80. Benchmarking Adversarial Robustness and Adversarial Training Strategies for Object Detection
Core Problem: Progress in defending object detection models against adversarial attacks lags behind classification due to a lack of standardized evaluation, making it difficult to compare attack or defense methods consistently across different datasets and metrics.
Key Innovation: Proposes a unified benchmark framework for digital, non-patch-based adversarial attacks on object detection, introducing specific metrics to disentangle localization and classification errors, and demonstrates that a mixed-objective, high-perturbation adversarial training dataset is the most robust defense strategy, while revealing a lack of transferability of modern attacks to transformer-based architectures.
81. Interpretability-by-Design with Accurate Locally Additive Models and Conditional Feature Effects
Core Problem: Generalized additive models (GAMs) offer interpretability but underfit when interactions are present, while GA^2Ms improve accuracy but sacrifice interpretability.
Key Innovation: Proposes Conditionally Additive Local Models (CALMs), a new model class that balances interpretability and accuracy by allowing multiple univariate shape functions per feature, active in different input regions defined by simple logical conditions, capturing interactions locally.
82. FEKAN: Feature-Enriched Kolmogorov-Arnold Networks
Core Problem: Existing Kolmogorov-Arnold Networks (KANs) offer enhanced interpretability but suffer from high computational cost and slow convergence, limiting scalability and practical applicability.
Key Innovation: Introduces Feature-Enriched Kolmogorov-Arnold Networks (FEKAN), an extension that preserves KAN advantages while improving computational efficiency and predictive accuracy through feature enrichment, accelerating convergence and increasing representation capacity without increasing trainable parameters.
83. A Contrastive Learning Framework Empowered by Attention-based Feature Adaptation for Street-View Image Classification
Core Problem: Street-view image attribute classification is computationally demanding, and existing adaptation methods for pre-trained vision-language models like CLIP often rely on global image embeddings, limiting their ability to capture fine-grained, localized attributes essential for complex street scenes.
Key Innovation: CLIP-MHAdapter, a lightweight contrastive learning framework, enhances CLIP adaptation by appending a bottleneck MLP with multi-head self-attention operating on patch tokens to model inter-patch dependencies. This achieves superior or competitive accuracy across eight attribute classification tasks on the Global StreetScapes dataset with low computational cost, enabling better capture of fine-grained localized attributes.
84. Unpaired Image-to-Image Translation via a Self-Supervised Semantic Bridge
Core Problem: Existing unpaired image-to-image translation methods (adversarial diffusion, diffusion-inversion) suffer from limitations such as requiring target-domain adversarial loss (limiting generalization) or producing low-fidelity translations due to imperfect inversion.
Key Innovation: Proposes the Self-Supervised Semantic Bridge (SSB) framework, which integrates external self-supervised semantic priors into diffusion bridge models to enable spatially faithful translation without cross-domain supervision, by leveraging visual encoders to learn appearance-invariant but structure-capturing representations.
85. Causally-Guided Automated Feature Engineering with Multi-Agent Reinforcement Learning
Core Problem: Existing automated feature engineering (AFE) methods rely on statistical heuristics, yielding brittle features that fail under distribution shift.
Key Innovation: CAFE, a causally-guided AFE framework that combines causal discovery with a cascading multi-agent deep Q-learning architecture for robust feature construction, improving performance and stability under covariate shifts.
86. Towards a Science of AI Agent Reliability
Core Problem: The discrepancy between high benchmark accuracy and practical failures of AI agents, due to current evaluations obscuring critical operational flaws like inconsistency, lack of robustness, unpredictability, or unbounded error severity.
Key Innovation: Proposing a holistic performance profile for AI agent reliability by introducing twelve concrete metrics across four key dimensions (consistency, robustness, predictability, and safety), which complement traditional evaluations and offer tools for reasoning about how agents perform, degrade, and fail.
87. FOCUS on Contamination: Hydrology-Informed Noise-Aware Learning for Geospatial PFAS Mapping
Core Problem: Large-scale monitoring and understanding of PFAS environmental contaminants are severely limited by high sampling costs and logistical challenges, leading to sparse observations and difficulty in simulating their spread with physical models.
Key Innovation: Introduces FOCUS, a geospatial deep learning framework that integrates sparse PFAS observations with rich environmental context (hydrology, land cover, source proximity, sampling distance) and a principled, noise-aware loss, consistently outperforming baselines for large-scale PFAS contamination mapping and risk assessment.
88. KnowIt: Deep Time Series Modeling and Interpretation
Core Problem: Users need a flexible and comprehensive environment to build powerful deep learning models for complex time series data and effectively interpret their behavior for knowledge discovery, without being constrained by task-specific assumptions.
Key Innovation: KnowIt, a Python toolkit that provides a flexible framework for deep time series modeling and interpretation, decoupling dataset, neural network architecture, and interpretability techniques through well-defined interfaces, enabling on-the-fly modeling and explanation of diverse time series data.
89. Robust Causal Discovery in Real-World Time Series with Power-Laws
Core Problem: Existing Causal Discovery (CD) algorithms are highly sensitive to noise, leading to spurious causal inferences in real-world time series, especially those exhibiting power-law distributions.
Key Innovation: A robust CD method that leverages the power-law spectral features of real-world time series to amplify genuine causal signals, consistently outperforming state-of-the-art alternatives on synthetic and real datasets.
90. Uncertainty Matters in Dynamic Gaussian Splatting for Monocular 4D Reconstruction
Core Problem: Reconstructing dynamic 3D scenes from monocular input using dynamic Gaussian Splatting is fundamentally under-constrained, leading to motion drifts under occlusion and degraded synthesis due to uniform optimization of Gaussians regardless of observation reliability.
Key Innovation: USplat4D, a novel Uncertainty-aware dynamic Gaussian Splatting framework that estimates time-varying per-Gaussian uncertainty and leverages it to construct a spatio-temporal graph for optimized 4D reconstruction, yielding more stable geometry and high-quality synthesis.
91. INQUIRE-Search: Interactive Discovery in Large-Scale Biodiversity Databases
Core Problem: Complex ecological phenomena (e.g., species interactions, responses to disturbance) are difficult to observe and sparsely documented, and current methods for discovering evidence in large biodiversity image databases rely on inefficient manual inspection.
Key Innovation: INQUIRE-Search, an open-source system that uses natural language to enable scientists to rapidly search, verify, and export observations of specific phenomena from large ecological image databases, demonstrating significant efficiency gains for ecological inference, including analyzing forest regrowth after wildfires.
92. GEPC: Group-Equivariant Posterior Consistency for Out-of-Distribution Detection in Diffusion Models
Core Problem: Existing diffusion-based Out-of-Distribution (OOD) detectors largely ignore the approximate equivariances inherited by learned score fields, potentially missing crucial OOD signals related to equivariance breaking.
Key Innovation: Group-Equivariant Posterior Consistency (GEPC), a training-free probe that measures how consistently the learned score transforms under a finite group, detecting equivariance breaking for OOD detection, demonstrating competitive performance on image benchmarks and strong target-background separation in SAR imagery.
93. Large Language Models for Water Distribution Systems Modeling and Decision-Making
Core Problem: Computational tools like EPANET for water distribution system (WDS) management are underutilized due to technical or expertise barriers, limiting data-driven decision-making.
Key Innovation: LLM-EPANET, an agent-based framework combining retrieval-augmented generation and multi-agent orchestration to enable natural language interaction with EPANET, translating user queries into executable code, running simulations, and returning structured results, thereby democratizing WDS modeling.
94. FindAnything: Open-Vocabulary and Object-Centric Mapping for Robot Exploration in Any Environment
Core Problem: Achieving real-time, open-vocabulary semantic understanding and geometrically accurate mapping of large-scale unknown environments for robot deployment and task planning, especially given computational constraints.
Key Innovation: FindAnything, an open-world mapping framework that integrates vision-language information into dense volumetric submaps, efficiently storing open-vocabulary information through object-level feature aggregation, enabling scalable deployment on resource-constrained devices like MAVs for tasks such as search and rescue.
95. A CNN-LSTM-Attention model for risk prediction of offshore wind turbines
Core Problem: Offshore wind turbines are exposed to complex environmental loads, leading to progressive structural degradation and increased failure risk over time, necessitating accurate risk prediction for intelligent structural health management.
Key Innovation: Proposes a hybrid deep learning model (CNN-LSTM-Attention) that integrates CNN for local spatial feature extraction, LSTM for temporal dynamics, and an Attention mechanism for learning critical patterns, demonstrating superior performance in predicting structural failure risk across multiple limit states.
96. Propagation of the Madden-Julian oscillation as a deterministic chaotic phenomenon
Core Problem: The mechanism and predictability of Madden-Julian oscillation (MJO) propagation, a planetary-scale tropical weather disturbance causing severe weather and climate events, remain elusive due to poorly understood multiscale processes.
Key Innovation: Revealed chaotic MJO propagation arising from cross-scale nonlinear interactions based on 4000-member ensemble simulations, demonstrating that multiple propagation regimes emerge depending on equatorial sea surface temperature asymmetry and tropical-extratropical interplay, contributing to a more complete MJO conceptual model.
97. Optimal choice of proxy for cloud condensation nuclei reduces uncertainty in aerosol-cloud-climate forcing
Core Problem: Aerosol-cloud interactions (ACI) are the largest uncertainty in anthropogenic climate forcings, and observation-based estimates of instantaneous radiative forcing from ACI (RFaci) are highly dependent on the choice of aerosol quantities as proxies for cloud condensation nuclei (CCN).
Key Innovation: Evaluated diverse aerosol proxies for CCN using observations and models, identifying surface CCN as the optimal proxy with the smallest bias (+5%) in predicting RFaci, which reduces the uncertainty in RFaci estimates from 66% to 43% and yields a less negative RFaci (-1.0 W m-2).
98. Centrifuge modeling to investigate the axial behavior of multi-helix piles with varied helix spacing in sand
Core Problem: There is insufficient understanding of internal load transfers and axial resistance distribution in multi-helix piles in sand, leading to conservative designs.
Key Innovation: Used centrifuge modeling to investigate load distribution between shaft sections and helix zones in multi-helix piles, showing that shaft resistance near the tip and within helix spacing is higher, and providing improved helix bearing/breakout factors for pile design.
99. Empirical Modelling of Tunnelling Induced Ground Deformations Based on Comprehensive Case Assessment
Core Problem: The need for accurate prediction of ground deformations (surface and subsurface settlements) during tunnelling operations to ensure the safety and stability of surrounding infrastructure.
Key Innovation: Development of new, more accurate empirical predictive equations for surface and subsurface settlement and trough width, based on a comprehensive assessment of case studies and incorporating critical variables like soil type, volume loss, tunnel diameter, and cover depth.
100. Rheological experiment and transport simulation of conditioned soil in EPB shield tunneling through fully weathered granite strata
Core Problem: Optimizing the efficiency and safety of Earth Pressure Balance (EPB) shield tunneling in challenging fully weathered granite strata requires a better understanding of the rheological behavior and transport characteristics of foam-conditioned soil.
Key Innovation: Investigated the rheological behavior of foam-conditioned soil using a self-developed device and established a coupled CFD model to simulate soil transport in EPB shields, providing insights into pressure distribution, transfer efficiency, and optimal cutterhead opening ratios for managing ground conditions.
101. Element differential solvers for nonlinear Biot’s poroelasticity equations in porous media
Core Problem: Conventional numerical methods like FEM have limitations in flexibility and direct discretization for nonlinear poroelastic problems in porous media.
Key Innovation: Proposes an improved element differential method using Lagrange elements and Chebyshev polynomials for accurate and flexible numerical analysis of nonlinear 2D/3D poroelastic problems, discretizing coupled governing equations directly without numerical integration.
102. Deep Neural Network Based Scale‐Adaptive Cloud Vertical Overlap Parameterization
Core Problem: Difficulty in accurately simulating cloud vertical overlap in climate models due to its spatiotemporal complexity, leading to errors in cloud radiative effects.
Key Innovation: Introducing a scale-adaptive decorrelation length (L) parameterization scheme developed by a deep neural network (DNN) that considers atmospheric statistics and dynamic factors, significantly surpassing traditional overlap schemes in simulation accuracy and having potential to improve cloud radiation effects in climate models.
103. Sediment Legacy Organic Matter Amplifies Nutrient‐Driven Coastal Hypoxia: A Coupled Benthic‐Pelagic Modeling Study
Core Problem: Quantitative understanding of how sediment legacy organic matter perpetuates coastal hypoxia via benthic-pelagic coupling and amplifies hypoxia sensitivity to nutrient loading remains limited.
Key Innovation: Developed a two-layer sediment model integrated with a physical-biogeochemical framework for the Pearl River Estuary, demonstrating that sediment oxygen consumption (44% of total oxygen depletion) is significantly supported by legacy organic matter, which amplifies hypoxia sensitivity and can lead to hysteresis and regime shifts.
104. Early to Mid‐Holocene Cool Temperature‐Induced Drought in Asian Interior
Core Problem: Hydroclimatic variations in mid-latitude Asia during the early to mid-Holocene and their associated mechanisms remain disputed, hindering understanding of atmospheric circulation controls on regional climate changes.
Key Innovation: Reporting Holocene alkenone records from Siberian lakes and synthesizing regional data to show that early to mid-Holocene drought in the Asian interior was associated with an enhanced anticyclonic system induced by prevailing cold air masses, explaining spatial heterogeneity of hydrological changes.
105. County‐Scale Climate Projections Over Minnesota and the Effects of Lakes
Core Problem: Global climate models (GCMs) lack sufficient resolution and lake effects for accurate county-scale climate projections over Minnesota, hindering informed decision-making for various sectors.
Key Innovation: Dynamically downscaled CMIP6 GCMs with a lake model at 4-km resolution to provide more detailed climate projections for Minnesota, showing stronger increases in spring/early summer precipitation and specific impacts on snow depth and lake ice.
106. Genetic Generalized Additive Models
Core Problem: Manually configuring the structure of Generalized Additive Models (GAMs) to balance predictive accuracy and interpretability is challenging.
Key Innovation: Proposes using the multi-objective genetic algorithm NSGA-II to automatically optimize GAMs by jointly minimizing prediction error and a complexity penalty, resulting in simpler, smoother, and more interpretable models that outperform or match baselines with lower complexity.
107. IT-OSE: Exploring Optimal Sample Size for Industrial Data Augmentation
Core Problem: There is no theoretical research or established method for estimating the optimal sample size (OSS) in data augmentation, particularly in industrial scenarios, leading to suboptimal model performance and high computational costs.
Key Innovation: Proposes an information-theoretic optimal sample size estimation (IT-OSE) framework and an interval coverage and deviation (ICD) score to reliably estimate OSS for industrial data augmentation, improving model accuracy, reducing computational costs, and enhancing interpretability.
108. A Study on Real-time Object Detection using Deep Learning
Core Problem: The need for dynamic analysis of visual information and immediate decision-making across various domains, which can be addressed by real-time object detection, leveraging advanced deep learning algorithms.
Key Innovation: Provides a detailed study on how deep learning algorithms enhance real-time object recognition, covering different models, benchmark datasets, applications, comparative studies, and future research directions.
109. Geometry-Aware Uncertainty Quantification via Conformal Prediction on Manifolds
Core Problem: Existing conformal prediction methods assume Euclidean output spaces, leading to poorly calibrated prediction regions when responses lie on Riemannian manifolds.
Key Innovation: Adaptive geodesic conformal prediction, a framework that uses geodesic nonconformity scores and a cross-validated difficulty estimator to produce position-independent, adaptively sized prediction regions (geodesic caps) on spheres, improving conditional coverage uniformity for manifold-valued responses.
110. Axle Sensor Fusion for Online Continual Wheel Fault Detection in Wayside Railway Monitoring
Core Problem: Traditional railway fault detection methods require manual feature engineering or suffer from performance degradation in online settings due to evolving operational patterns, hindering reliable and cost-effective predictive maintenance.
Key Innovation: A semantic-aware, label-efficient continual learning framework for railway fault diagnostics that fuses VAE-encoded accelerometer signals with semantic metadata from fiber Bragg grating sensors (axle counts, wheel indexes, strain-based deformations). It uses a lightweight classifier and replay-based continual learning to adapt to evolving conditions and detect minor wheel imperfections.
111. Investigating GNN Convergence on Large Randomly Generated Graphs with Realistic Node Feature Correlations
Core Problem: Existing studies on GNN convergence on large random graphs often fail to model correlations between node features, which are common in real-life networks, thus not accurately reflecting GNN expressive power on realistic graphs.
Key Innovation: Introduces a novel method to generate random graphs with correlated node features, ensuring correlation between neighboring nodes. Theoretical analysis and empirical validation on these graphs demonstrate that GNN convergence can be avoided in some cases, suggesting GNNs may be more expressive on realistic graphs than previously thought.
112. Differentially Private Non-convex Distributionally Robust Optimization
Core Problem: Traditional Empirical Risk Minimization (ERM) degrades under distribution shifts and data sensitivity, while Differentially Private Distributionally Robust Optimization (DP-DRO) is challenging due to its minimax structure.
Key Innovation: Developing novel $(\varepsilon, \delta)$-DP optimization methods, DP Double-Spider and DP Recursive-Spider, for DP-(finite-sum)-DRO with $\psi$-divergence and non-convex loss, achieving improved utility bounds and outperforming existing DP minimax optimization approaches.
113. Uncertainty-Guided Inference-Time Depth Adaptation for Transformer-Based Visual Tracking
Core Problem: Transformer-based single-object trackers incur unnecessary computational cost due to fixed-depth inference, regardless of visual complexity in video sequences.
Key Innovation: UncL-STARK, an architecture-preserving approach enabling dynamic, uncertainty-aware depth adaptation in transformer-based trackers, achieving significant computational and energy savings while maintaining tracking accuracy.
114. Amortized Predictability-aware Training Framework for Time Series Forecasting and Classification
Core Problem: Time series data are prone to noise and contain low-predictability patterns, leading to training instability and suboptimal performance in time series forecasting and classification tasks.
Key Innovation: Proposes the Amortized Predictability-aware Training Framework (APTF) which introduces a Hierarchical Predictability-aware Loss (HPL) to dynamically identify and penalize low-predictability samples, and an amortization model to mitigate predictability estimation errors, thereby improving model performance.
115. A Self-Supervised Approach for Enhanced Feature Representations in Object Detection Tasks
Core Problem: The increasing complexity of deep learning models, especially for object detection, is bottlenecked by the significant time and resources required for labeled data, leading to high investment costs.
Key Innovation: A self-supervised learning strategy that trains a model on unlabeled data to enhance feature extractors, outperforming state-of-the-art ImageNet pre-trained models for object detection and improving reliability by focusing on relevant object aspects.
116. Explainability for Fault Detection System in Chemical Processes
Core Problem: Understanding and explaining the fault diagnosis decisions made by highly accurate deep learning models (like LSTMs) in complex industrial processes is crucial for identifying root causes and improving system reliability.
Key Innovation: Application and comparison of state-of-the-art eXplainability Artificial Intelligence (XAI) methods (Integrated Gradients and SHAP) to explain an LSTM classifier's fault diagnosis decisions in the Tennessee Eastman Process, demonstrating their utility in identifying fault subsystems and potential root causes.
117. ReMoRa: Multimodal Large Language Model based on Refined Motion Representation for Long-Video Understanding
Core Problem: Long-form video understanding remains a significant challenge for Multimodal Large Language Models (MLLMs) due to the computational intractability and high redundancy of processing full RGB frame streams, which have quadratic complexity with sequence length.
Key Innovation: ReMoRa, a video MLLM, processes videos by operating directly on compressed representations (sparse RGB keyframes for appearance and a refined motion representation for temporal dynamics), scaling linearly with sequence length and outperforming baselines on long-video understanding benchmarks.
118. Transfer Learning of Linear Regression with Multiple Pretrained Models: Benefiting from More Pretrained Models via Overparameterization Debiasing
Core Problem: Understanding when and how using multiple overparameterized pretrained models can improve transfer learning for linear regression, and addressing the overparameterization bias that can compromise learning.
Key Innovation: Analytically formulates the test error for transfer learning with multiple pretrained models, elucidates conditions for beneficial transfer, and proposes a simple debiasing via multiplicative correction factor to reduce overparameterization bias and leverage more pretrained models.
119. Let's Split Up: Zero-Shot Classifier Edits for Fine-Grained Video Understanding
Core Problem: Video recognition models are typically trained on fixed, often coarse taxonomies, making it costly to accommodate emerging fine-grained distinctions without collecting new annotations and retraining.
Key Innovation: Introduces 'category splitting' and proposes a zero-shot editing method that leverages the latent compositional structure of video classifiers to expose fine-grained distinctions without additional data, improving accuracy on newly split categories without sacrificing performance elsewhere.
120. TeCoNeRV: Leveraging Temporal Coherence for Compressible Neural Representations for Videos
Core Problem: Existing Implicit Neural Representations (INRs) for video compression struggle to scale to high-resolution videos while maintaining encoding efficiency, and hypernetwork-based approaches have low quality, large compressed size, and high memory needs.
Key Innovation: Introduces TeCoNeRV, which decomposes weight prediction spatially and temporally, uses a residual-based storage scheme for differences between segment representations, and applies temporal coherence regularization, achieving substantial PSNR improvements, lower bitrates, and faster encoding speeds for high-resolution videos.
121. ReasonNavi: Human-Inspired Global Map Reasoning for Zero-Shot Embodied Navigation
Core Problem: Embodied agents often struggle with efficient navigation due to reliance on partial egocentric observations, lacking global foresight and leading to inefficient exploration, unlike human map-based planning.
Key Innovation: Introduces ReasonNavi, a human-inspired framework that couples Multimodal Large Language Models (MLLMs) with deterministic planners for zero-shot embodied navigation, converting top-down maps into a discrete reasoning space for MLLM querying and grounding selected waypoints into executable trajectories, outperforming prior methods without fine-tuning.
122. MARVL: Multi-Stage Guidance for Robotic Manipulation via Vision-Language Models
Core Problem: Designing dense reward functions for robotic Reinforcement Learning (RL) is challenging due to manual engineering, and naive VLM rewards often misalign with task progress, struggle with spatial grounding, and lack task semantic understanding.
Key Innovation: Proposes MARVL, a multi-stage guidance framework for robotic manipulation that fine-tunes a VLM for spatial and semantic consistency, decomposes tasks into subtasks with task direction projection, significantly outperforming existing VLM-reward methods in sample efficiency and robustness on sparse-reward manipulation tasks.
123. VGGT-based online 3D semantic SLAM for indoor scene understanding and navigation
Core Problem: Achieving robust, memory- and speed-efficient 3D semantic scene understanding and mapping for autonomous and assistive navigation, especially with long video streams.
Key Innovation: Presents SceneVGGT, an online spatio-temporal 3D semantic SLAM framework that uses a sliding-window pipeline for memory-efficient mapping, lifts 2D instance masks to 3D objects with temporal coherence, and supports interactive assistive navigation, all while maintaining low GPU memory usage.
124. Adaptive Illumination Control for Robot Perception
Core Problem: Robot perception in low light or high dynamic range conditions is limited by captured image quality, and predicting the impact of programmable onboard light is complex due to nonlinear interactions with depth, surface reflectance, and scene geometry.
Key Innovation: Introduces Lightning, a closed-loop illumination-control framework for visual SLAM. It combines a Co-Located Illumination Decomposition (CLID) relighting model, an offline Optimal Intensity Schedule (OIS) optimization, and an Illumination Control Policy (ILC) distilled through behavior cloning, substantially improving SLAM trajectory robustness and reducing power consumption.
125. Machine Learning in Epidemiology
Core Problem: Epidemiologists face increasing amounts of complex, high-dimensional data, requiring powerful tools like machine learning for effective analysis.
Key Innovation: Lays methodological foundations for applying machine learning in epidemiology, covering principles of supervised and unsupervised learning, important ML methods, strategies for model evaluation and hyperparameter optimization, and interpretable machine learning, accompanied by practical R code examples.
126. Explainable AI: Context-Aware Layer-Wise Integrated Gradients for Explaining Transformer Models
Core Problem: Existing explainability methods for Transformer models rely on final-layer attributions, lack context-awareness, and fail to capture how relevance evolves across layers and how structural components shape decision-making.
Key Innovation: Proposing the Context-Aware Layer-wise Integrated Gradients (CA-LIG) Framework, a unified hierarchical attribution method that computes layer-wise Integrated Gradients and fuses them with class-specific attention gradients to provide more faithful, context-sensitive, and semantically coherent explanations of Transformer decision-making.
127. Model-Agnostic Dynamic Feature Selection with Uncertainty Quantification
Core Problem: Existing Dynamic Feature Selection (DFS) methods require models specifically designed for sequential acquisition, limiting compatibility with pre-trained models, and provide insufficient uncertainty quantification for high-stakes decisions.
Key Innovation: A model-agnostic DFS framework compatible with pre-trained classifiers, which formalizes new uncertainty sources in DFS and proposes efficient subset reparametrization strategies, achieving competitive accuracy and highlighting the need for uncertainty-aware DFS.
128. Pinet: Optimizing hard-constrained neural networks with orthogonal projection layers
Core Problem: Neural networks often struggle to satisfy hard convex constraints in their outputs, which is crucial for parametric constrained optimization problems, leading to slower and less robust solutions compared to traditional solvers.
Key Innovation: Introduces $\Pi$net, an output layer for neural networks that ensures satisfaction of convex constraints using operator splitting for rapid projections and the implicit function theorem for backpropagation, achieving faster and more robust solutions than traditional solvers and state-of-the-art learning approaches.
129. Vision and Language: Novel Representations and Artificial intelligence for Driving Scene Safety Assessment and Autonomous Vehicle Planning
Core Problem: Effectively integrating vision-language representations into autonomous driving perception, prediction, and planning pipelines to improve driving scene safety assessment and decision-making, particularly for detecting diverse road hazards and incorporating semantic constraints.
Key Innovation: Demonstration of three system-level use cases for VLMs in autonomous driving: a lightweight, category-agnostic CLIP-based hazard screening; investigating VLM embeddings in transformer-based trajectory planning; and using natural language as explicit behavioral constraints to improve safety-aligned behavior.
130. COOPERTRIM: Adaptive Data Selection for Uncertainty-Aware Cooperative Perception
Core Problem: Cooperative perception for autonomous agents is hindered by the tension between limited communication bandwidth and the rich sensor information, making it challenging to efficiently share relevant data while maintaining performance.
Key Innovation: Introduction of COOPERTRIM, an adaptive data selection framework that proactively exploits temporal continuity and a novel conformal temporal uncertainty metric to gauge feature relevance, dynamically determining the sharing quantity. This achieves significant bandwidth reduction while maintaining comparable accuracy.
131. ToaSt: Token Channel Selection and Structured Pruning for Efficient ViT
Core Problem: Vision Transformers (ViTs) have prohibitive computational costs hindering deployment, and existing efficiency solutions like structured weight pruning and token compression suffer from prolonged retraining or global propagation issues.
Key Innovation: ToaSt, a decoupled framework applying specialized strategies (coupled head-wise structured pruning and Token Channel Selection) to distinct ViT components, achieving superior accuracy-efficiency trade-offs and outperforming existing baselines.
132. Statistical Inference Leveraging Synthetic Data with Distribution-Free Guarantees
Core Problem: Safely and efficiently enhancing statistical inference by combining high-quality synthetic data with real data, especially in scenarios with limited labeled data, while ensuring robust, distribution-free guarantees on error bounds.
Key Innovation: The GEneral Synthetic-Powered Inference (GESPI) framework, which adaptively combines synthetic and real data to boost statistical power, yet defaults to standard inference with real data if synthetic data quality is low, providing distribution-free error bounds and seamless integration with various inference procedures without modification.
133. Imaging with super-resolution in changing random media
Core Problem: Achieving super-resolution imaging in complex, changing random media where strong scattering typically degrades image quality.
Key Innovation: An imaging algorithm that exploits strong scattering, combining sparse dictionary learning, clustering, and multidimensional scaling to reliably extract unknown medium properties and achieve super-resolution beyond homogeneous medium limits, especially with abundant data.
134. GFC2020: a global map of forest land use for year 2020 to support the EU Deforestation Regulation
Core Problem: Comprehensive global information on forest cover, capturing both physical characteristics and land use components as defined by FAO, remains limited.
Key Innovation: Presents GFC2020, a harmonized and globally consistent 10m resolution map of forest presence or absence for 2020, combining multiple Earth observation datasets within an open science framework with 91% overall accuracy.
135. Inertia-driven amphibious robot with asymmetric microundulatory fin arrays
Core Problem: Existing centimeter-scale amphibious robots often rely on complex, active, and multiple mechanisms for environmental interaction, leading to sealing challenges and reliability issues at small scales.
Key Innovation: Developed an inertia-driven actuation strategy for a fully sealed amphibious robot using a variable-output voice coil motor and passive tilted fins, enabling versatile locomotion (jumping, rapid terrestrial motion, steerable aquatic propulsion) through a compact and reliable design, demonstrated by a 24-gram prototype.
136. Study on the mechanism of water injection pressure and fractures on wetting and diffusion of hard rock during water injection
Core Problem: Insufficient understanding of the combined effects of water injection pressure and fracture characteristics on moisture diffusion in high-density, low-porosity hard rock, which is crucial for optimizing water injection parameters in applications like mining dust suppression.
Key Innovation: Systematically investigated water migration using coupled low-temperature high-pressure nuclear magnetic resonance tests and COMSOL simulations. Found that high pressure effectively enhances seepage and that wetting radius exhibits an exponential relationship with pressure (stabilizing after 11 MPa) and a linear relationship with fracture length, providing theoretical guidance for optimizing water injection.
137. Advances in the monitoring and forecasting of urban extreme meteorological events: a bibliometric review
Core Problem: Understanding the dynamic interactions between rapid urbanization and environmental changes, particularly the intensification of urban heat islands, extreme rainfall, pollution, and temperature anomalies, and the challenges in integrating localized urban features into predictive models.
Key Innovation: A bibliometric review analyzing 1,000 articles on urban meteorology, identifying global research patterns, thematic insights (LULC, aerosols, socio-economic factors), and the pivotal role of advanced technologies (satellite remote sensing, GIS, models) in monitoring and predicting urban atmospheric phenomena. It highlights the need for interdisciplinary, data-driven approaches and leveraging satellite data for better risk management.
138. Hydrochemical characteristics and influencing factors of the surface water and the groundwater in the Mingyong River Basin of the Meili Snow Mountains
Core Problem: Understanding the complex hydrochemical characteristics and the natural and anthropogenic factors influencing surface and groundwater in mountainous river basins is essential for water resource management.
Key Innovation: Systematically analyzed hydrochemical characteristics of surface and groundwater in the Mingyong River Basin using a combination of hydrogeochemical and statistical methods, identifying dominant ions, primary solute sources (rock weathering), and quantifying contributions from various influencing factors including human activities.
139. Mapping vegetation phenology and its response to climate change in Southwest China using solar-induced chlorophyll fluorescence
Core Problem: Accurately extracting vegetation phenology in complex terrains like Southwest China, especially for evergreen forests, is challenging with traditional remote sensing, leading to uncertainties in ecosystem dynamics and carbon sink assessments.
Key Innovation: Demonstrated that solar-induced chlorophyll fluorescence (SIF) outperforms traditional remote sensing in extracting vegetation phenology, and used SIF to analyze spatial-temporal changes in phenology (SOS, EOS, LOS) and their responses to preseason climatic factors in Southwest China, improving understanding of regional ecosystem dynamics.
140. Compacted red soil-bentonite liners: effect of compactive effort, wet-dry cycles and chemical contamination
Core Problem: There is a need for sustainable liner materials with low hydraulic conductivity for waste containment that can withstand environmental stresses like wet-dry cycling and chemical exposure.
Key Innovation: Demonstrated that red soil-bentonite mixtures compacted with modified Proctor effort maintain hydraulic conductivities within design limits even after wet-dry cycles and chemical exposure, offering durable and efficient performance as liner materials.
141. Deformation Behavior and Shakedown Response of Subgrade Soils Under Cyclic Loading
Core Problem: Long-term deformation and stability of heavy-haul railway subgrades under repeated traffic loading, which can lead to track performance issues and maintenance demands.
Key Innovation: Systematic analysis of irreversible deformation and shakedown characteristics of compacted silty clay subgrade soils under varying conditions, leading to an empirical model for permanent deformation and a framework for determining shakedown thresholds.
142. A proactive decision support system for managing transient fire risks in construction environments
Core Problem: Systematic safety management for transient fire risks in construction environments is hindered by the structural opacity of accident data and reliance on subjective judgment, lacking an objective, data-driven assessment framework independent of inaccessible internal site data.
Key Innovation: A novel hybrid AI framework utilizing XGBoost with SHAP analysis and K-means clustering on national fire and meteorological data to objectively derive risk weights and thresholds for weather variables, providing a quantifiable decision support tool for proactive fire safety management in construction.
143. Environmental gradients mediate divergent patterns of microbial nutrient use efficiency and metabolic limitation in arid desert ecosystems
Core Problem: How microbial metabolic plasticity and resource use trade-offs respond to steep environmental gradients in riparian-to-desert transition zones, and their implications for desertification, is poorly understood.
Key Innovation: Revealed divergent microbial nutrient limitations along an environmental gradient, with nitrogen limitation dominating but carbon limitation more pronounced in mesic areas. Abiotic constraints indirectly modulated carbon and nitrogen use efficiency by driving shifts in microbial biomass and community diversity, providing a theoretical basis for understanding desert ecosystem vulnerability and resilience under desertification.
144. Aridity-dependent biodiversity and multi-trophic associations drive soil multifunctionality in dryland ecosystems
Core Problem: The relative contributions of different taxonomic groups' diversity and distinct types of multi-trophic associations to soil multifunctionality under aridity gradients in dryland ecosystems are poorly understood.
Key Innovation: Demonstrated that while soil biodiversity strongly predicts soil multifunctionality across aridity gradients, the contributions vary by taxa (e.g., bacterial and saprotrophic fungal diversity were key). Aridity weakened biotic associations, with microbe-microbe and microbe-microfauna associations having the strongest positive relationships with soil multifunctionality, highlighting the need to conserve both diversity and functional ecological associations for dryland ecosystem health.
145. The impact of data quality and outlier detection in high-frequency water quality data on water management and process understanding
Core Problem: Outliers in high-frequency environmental data (water quality) are a common challenge, and their quantitative impact on water quality metrics and the effectiveness of various detection methods are not well-evaluated.
Key Innovation: Quantitatively evaluated the impact of outlier detection methods on water quality metrics using a four-year dataset, demonstrating that tailored quality control is crucial and that aggregating results across multiple transformations shows promising potential, while no single method is universally superior.
146. Water quality spatial-temporal imputation using diffusion graph convolutional networks: A case study in Georgia, USA
Core Problem: Water quality monitoring often suffers from sparse sensor deployments and missing spatio-temporal data, and conventional imputation methods struggle with complex spatio-temporal, multi-parameter dynamics and noisy, evolving sensor networks.
Key Innovation: Proposed a novel Water Quality Auxiliary-Enhanced Spatio-Temporal Imputation Framework based on Diffusion Graph Convolutional Networks, which adaptively integrates auxiliary variables and uses subgraph-sampling to accurately impute missing water quality data, outperforming state-of-the-art baselines.
147. Arching effect in a full-scale reinforced piled embankment on very soft soils, considering long-term performance and plate load tests
Core Problem: Understanding the long-term arching effect in reinforced piled embankments on very soft soils to ensure infrastructure stability and optimize design, as existing analytical and numerical solutions show discrepancies with measured data.
Key Innovation: Comprehensive long-term monitoring (almost 5 years) of a full-scale piled embankment, revealing the magnitude of the arching effect, its reduction due to cyclic annual rainy seasons and seismic events, and verifying its presence under simulated aircraft loads, providing insights into soil-structure interaction.