TerraMosaic Daily Digest: Feb 12, 2026
Daily Summary
Across 230 selected papers, the field moves beyond pattern recognition toward constrained prediction. Hybrid methods now embed physical structure directly into learning workflows, including differentiable back-analysis of post-liquefaction residual strength, reciprocity-enforced neural operators for seismic waves, and Bayesian physical-mechanical forecasting of reservoir landslide displacement.
The same corpus highlights stronger attention to cascading and climate-amplified hazards: post-wildfire mudflows, rainstorm-induced landslide-debris-flow chains, hydrothermal moraine failures, and expanding compound drought-heatwave hotspots. Monitoring studies increasingly close the loop with planning through decade-scale InSAR building surveillance, 3D deformation retrieval, and long-duration camera-based water-level gauging.
Key Trends
- Physics-constrained learning is maturing from calibration aid to core inference engine: Differentiable simulators, operator constraints, and Bayesian updating are now used to recover latent mechanical parameters and improve extrapolation under sparse observations.
- Monitoring is shifting from snapshots to long-horizon decision systems: Decade-scale InSAR, 3D deformation inversion, and durable image-based gauges demonstrate sustained surveillance linked to engineering thresholds and operations.
- Cascading hazard chains are treated explicitly rather than as isolated events: New analyses connect wildfire-to-mudflow, rainstorm-to-landslide-to-debris-flow, and permafrost/hydrothermal weakening to downstream failure escalation.
- Coastal and cryosphere studies emphasize disequilibrium and feedbacks: Three-dimensional erosion amplification, sea-level-rise shoreline disequilibrium, and Antarctic meltwater-ocean feedbacks are redefining boundary conditions for risk projection.
- Adaptation frameworks are becoming measurable and distribution-aware: Equity-aware retrofit optimization and adaptive urban stormwater intelligence translate hazard science into explicit, auditable planning choices.
Selected Papers
This digest features 230 selected papers from 1129 deduplicated papers analyzed (out of 2494 raw papers scanned; 1129 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. Predicting the post-wildfire mudflow onset using machine learning models on multi-parameter experimental data
Core Problem: Understanding and predicting the timing and conditions of post-wildfire mudflow onset, driven by complex coupled effects of critical parameters (rain intensity, slope, water-entry, grain size), is crucial for effective hazard assessment and emergency response.
Key Innovation: Applies multiple machine learning algorithms (MLR, LR, SVC, K-means, PCA) to multi-parameter laboratory experimental data to effectively predict total discharge and classify failure outcomes of post-wildfire mudflows, highlighting the critical role of fine sand and initial high-intensity rainfall.
2. Differentiable Graph Neural Network Simulator for the Back-Analysis of Post-Liquefaction Residual Strength from Flow Failure Runout
Core Problem: Traditional approaches to estimate post-liquefaction residual strengths ($S_r$) from flow failure runout are computationally intensive, rely on simplified physics, manual iterations, and assumptions about runout development, limiting efficiency and accuracy.
Key Innovation: Introduces Differentiable Graph Neural Network Simulators (Diff-GNS), a physics-informed and automated framework that integrates a Graph Neural Network Simulator (GNS) for accelerated granular flow simulation with gradient-based optimization for efficient and reproducible back-analysis of $S_r$, validated on real-world liquefaction-induced flow failure case histories.
3. Edge-Aware Superpixel Dual-Graph GCN for Topographically Heterogeneous Landslide Susceptibility Assessment
Core Problem: Pixel-based landslide susceptibility assessment (LSA) in topographically heterogeneous mountains is prone to boundary blurring, salt and pepper artifacts, and unstable generalization, leading to less interpretable and accurate results.
Key Innovation: Proposes an edge-aware superpixel graph framework with a dual-graph graph convolutional network (GCN). It employs terrain-enhanced simple linear iterative clustering with watershed initialization to generate edge-aware superpixels, constructs a fused graph combining spatial adjacency and feature similarity, and stabilizes learning under class imbalance. This method significantly improves LSA accuracy and produces clearer, more contiguous susceptibility maps with reduced speckle.
4. Seismic Characterization of Lahars on Volcán de Fuego Toward the Development of a Machine Learning‐Based Detection Algorithm
Core Problem: The challenge of early detection and automated alerts for frequent lahars on Volcán de Fuego, which often rely on manual monitoring and sparse visual confirmation.
Key Innovation: Characterized varied short-term lahar behavior using seismic data, revealing increasing seismic activity and a shift to lower frequencies downstream. Developed and implemented K-nearest neighbor (KNN) based detectors using seismic signal attributes, achieving high accuracy for moderate-to-large flows and demonstrating a computationally efficient, portable solution for automated lahar detection.
5. Leveraging Laboratory Experiments of Shoreline Response to Sea‐Level Rise: A Beach Disequilibrium Perspective
Core Problem: Understanding and quantifying shoreline changes due to sea-level rise (SLR), separating passive flooding from wave-driven processes, and improving equilibrium shoreline models under SLR and wave-climate change.
Key Innovation: Analyzed and synthesized 24 laboratory experiments into a dimensionless dataset to investigate SLR-induced processes, demonstrating trends between relative wave power and wave-driven shoreline changes, and identifying wave-energy dissipation as a key variable for quantifying SLR-induced disequilibrium, offering new pathways for future model improvements.
6. Antarctic Meltwater Accelerates Southern Ocean Evolution Under Projected Atmospheric Warming
Core Problem: Understanding the feedback between increasing basal meltwater from Antarctic ice shelves and Southern Ocean properties, and how this impacts the rate of melting, especially in a high-emissions scenario with interactive melt rates.
Key Innovation: Found that interactive meltwater primarily accelerates the evolution of continental shelf warming and cooling patterns, leading to a net ~35% reduction in ice-shelf meltwater input into the Southern Ocean over the 21st century, highlighting the bias introduced by omitting this feedback in projected ocean-melt-driven ice loss.
7. Retrieving Three‐Dimensional Deformation in Groundwater Pumping Areas Based on InSAR Data
Core Problem: Conventional InSAR techniques are limited in their sensitivity to north-south displacement, making it challenging to monitor full three-dimensional (3D) surface deformation in groundwater pumping areas.
Key Innovation: Developed VG-SM3D, an inversion approach combining multi-track InSAR data with a strain model and a physical prior, successfully reconstructing 3D deformation (pronounced subsidence and horizontal motion) in the Tianjin region with good agreement to ground measurements.
8. Enforcing Reciprocity in Operator Learning for Seismic Wave Propagation
Core Problem: Data-driven methods for seismic wavefield modeling, while efficient, have yet to incorporate strict physical consistency, specifically failing to hard-code fundamental physical laws like the principle of reciprocity, which limits their accuracy and reliability.
Key Innovation: Introduces the Reciprocity-Enforced Neural Operator (RENO), a transformer-based architecture for modeling seismic wave propagation that hard-codes the reciprocity principle, guaranteeing invariance under source and receiver position swapping, leading to improved physical consistency and an order-of-magnitude inference speedup for multiple sources.
9. KAN-FIF: Spline-Parameterized Lightweight Physics-based Tropical Cyclone Estimation on Meteorological Satellite
Core Problem: Timely monitoring of tropical cyclones (TCs) is crucial but hindered by the computational inefficiency and high parameter counts of existing physics-guided models, especially on resource-constrained edge devices, and their failure to capture high-order polynomial relationships between TC attributes.
Key Innovation: Introduces KAN-FIF, a lightweight multimodal architecture integrating MLP, CNN, and spline-parameterized KAN layers. It achieves a 94.8% reduction in parameters and 68.7% faster inference for Maximum Sustained Wind prediction while maintaining superior accuracy, demonstrating promising feasibility for operational TC monitoring on edge devices.
10. Three-dimensional amplification of storm-driven beach erosion: Implications for setback lines at Narrabeen-Collaroy Beach, Australia
Core Problem: Existing approaches for estimating short-term coastal erosion (storm demand) often rely on two-dimensional (2D) beach profiles, potentially underestimating erosion due to localized three-dimensional (3D) effects, leading to inaccurate setback lines for coastal planning.
Key Innovation: An empirical analysis combining nearly five decades of 2D beach profile data with targeted 3D storm response observations to quantify storm erosion-frequency relationships, demonstrating that localized 3D processes (rip currents, wave focusing) can amplify erosion volumes by up to 46%, and developing empirical design curves for both typical (2D) and enhanced (3D) erosion scenarios to improve coastal setback line estimation.
11. Assessment of Building Subsidence Along Suzhou Metro Lines Using Decade-Scale Multitemporal InSAR
Core Problem: Metro construction in soft soil conditions can induce differential settlement in nearby buildings, necessitating effective, long-term monitoring and risk assessment methods to ensure infrastructure safety and guide urban planning.
Key Innovation: Employs decade-scale (2009–2019) multitemporal InSAR using high-resolution TerraSAR-X imagery with a customized processing chain to extract time-series deformation along Suzhou's metro network. It develops a practical risk evaluation framework by integrating InSAR-derived displacements with building safety standards, identifying specific at-risk structures, and provides the first large-scale visualization of subsidence hazards for metro planning and operational safety.
12. Rapid amplification of Compound Drought and Heatwave risk over India: a regime shift from arid northwest to humid southern and eastern hotspots
Core Problem: Quantifying the regional intensification and spatiotemporal evolution of Compound Drought and Heatwave (CDHW) events is critical under changing precipitation regimes, especially in regions like India where impacts on agriculture and urban populations are significant.
Key Innovation: This study reveals a statistically significant 86% rise in individual heatwave frequency and drought intensification in historically humid zones of India (1951–2016), leading to a 10% to 35% expansion of CDHW-affected areas and a 25-fold increase in extreme CDHWs. It identifies a regime shift of hotspots from the arid northwest to humid eastern and southern regions, highlighting cascading risk amplification and significant agricultural impacts.
13. AI image-based method for a robust automatic real-time water level monitoring: a long-term application case
Core Problem: The need for robust, automated, real-time, and low-cost water level monitoring solutions, particularly for flood detection and mitigation in ungauged or remote areas.
Key Innovation: Develops and validates an AI image-based camera gauge system that achieves high accuracy (MAE 0.96-2.66 cm) and resilience over 2.5+ years, enabling 24/7, near real-time water level monitoring for enhanced flood detection.
14. Occurrence of major earthquakes is as stochastic as smaller ones
Core Problem: Seismic hazard estimates often rely on recurrence models that assume cyclic or quasiperiodic interevent times for major earthquakes, which may be inaccurate and derived from limited datasets.
Key Innovation: Demonstrated through statistical analyses of a 6000-year lake-sediment seismic record and other data that time intervals between large earthquakes (M ≥ 6.5) robustly follow a Poisson distribution, indicating stochastic occurrence and challenging conventional recurrence models, thereby substantially increasing seismic hazard estimates.
15. Spatiotemporal response of vegetation productivity to coseismic landslides: a case study of NPP dynamics following the 2014 Ludian earthquake, China
Core Problem: Coseismic landslides substantially disturb vegetation and alter ecosystem productivity, and the spatiotemporal response of vegetation to these disturbances needs to be quantified.
Key Innovation: Used MODIS net primary productivity (NPP) data to analyze the spatiotemporal response of vegetation productivity to coseismic landslides triggered by the 2014 Ludian earthquake, revealing the dynamics of ecosystem disturbance.
16. Unveiling the drivers of rainfall-triggered landslide spatial distribution: Insights from event-based and historical landslide inventories in Lisbon region, Portugal
Core Problem: Understanding the factors influencing the spatial distribution of rainfall-triggered landslides and evaluating the impact of using different types of inventories (event-based vs. historical) in assessing landslide susceptibility.
Key Innovation: Demonstrated that predisposing factors like slope, aspect, and elevation are key, but their importance ranking differs between models using event-based and historical inventories. The study highlights the individual strengths and spatial differences in susceptibility maps, emphasizing the importance of integrating both datasets and caution in using event-based inventories for general susceptibility mapping due to localized rainfall bias.
17. A decision-support framework for fair building retrofit allocation under multi-hazard risk
Core Problem: Hazard mitigation planning often misses distributional effects, leading to unequal outcomes for vulnerable groups in building retrofit allocation under multi-hazard risk.
Key Innovation: Developed a multi-objective optimization framework that treats equity (measured by the decomposable Theil index) as a first-class objective alongside efficiency (economic loss, repair time, displacement) for building retrofit allocation under coupled earthquake-tsunami hazards, yielding transparent Pareto frontiers and spatially explicit portfolios for informed, equitable planning.
18. Orbit-specific tropospheric effects on Sentinel-1A/B interferometric synthetic aperture radar observations: insights for deformation analysis and future mission design
Core Problem: Tropospheric delays significantly complicate the accurate interpretation of surface deformation from InSAR observations, and orbit-specific tropospheric features remain underexplored.
Key Innovation: Quantified pronounced spatial heterogeneity and temporal variability of orbit-specific tropospheric errors using nine years of Sentinel-1A/B data and GACOS, providing insights into improving deformation inversion accuracy and offering recommendations for future SAR mission design.
19. Failure modes of basal-ice moraine slopes under combined hydrothermal conditions: laboratory evidence
Core Problem: Insufficient understanding of specific failure mechanisms of basal-ice moraine slopes (BIMS) under combined climate warming (degradation of basal ice) and intense rainfall, which poses a significant geohazard.
Key Innovation: Provided direct experimental evidence from instrumented laboratory flume experiments, identifying the ice-soil interface as the primary control for slope instability, and distinguishing two distinct failure pathways: progressive temperature-driven instability (misaligned sliding) and abrupt rainfall-induced instability (rapid disintegration, flow-slips, debris flows).
20. Calculation of safe roof thickness for irregular goaf underlying a subgrade based on the backward elimination method
Core Problem: Accurately determining the safe roof thickness for irregular goafs (abandoned mine workings) underlying expressway subgrades is crucial for ensuring stability, but challenging due to complex geometries and multiple influencing factors.
Key Innovation: A novel multi-factor calculation model for goaf safe roof thickness was developed using 3D laser scanning data, FLAC3D numerical simulations, and the backward elimination method, which accurately predicts roof stability based on goaf span, height, burial depth, and roof cohesion, providing a valuable reference for engineering design and construction.
21. Timescales and magma dynamics of the plumbing system feeding a Plinian eruption: the 79 CE eruption of Somma-Vesuvius, Italy
Core Problem: Incomplete knowledge of the plumbing system architecture and pre-eruptive magmatic processes of Plinian eruptions, which is crucial for forecasting eruptions and mitigating hazards.
Key Innovation: Chemical, isotopic data, clinopyroxene zoning analysis, and numerical modelling reveal a vertically extended plumbing system with deep mafic magma batches refilling a tephri-phonolitic reservoir multiple times before the 79 CE Somma-Vesuvius eruption, constraining recharge timescales from decades to less than a year, enhancing hazard assessment.
22. Adaptive infrastructure intelligence: integrated urban stormwater management framework for climate-resilient cities
Core Problem: Conventional urban stormwater management practices are insufficient for achieving climate-resilient cities, leading to issues like high peak flows, frequent floods, and pollutant discharge.
Key Innovation: Developed the Adaptive Infrastructure Intelligence (AII) framework, a socio-technical system integrating urban ecology and adaptive governance with multimodal sensing networks, demonstrating predictive accuracy of 91% and practical innovations like 67% peak flow reduction, 83% flood occurrence reduction, and 72% pollutant removal compared to conventional practices.
23. Distribution and evolution of distress in embankment-bridge transition sections of the Gonghe–Yushu Expressway in degrading permafrost regions
Core Problem: Embankment-bridge transition sections (EBTSs) of expressways in high-altitude degrading permafrost regions are highly susceptible to structural deterioration (e.g., uneven settlement) due to climate warming, with limited systematic investigation into their distress characteristics and evolutionary mechanisms.
Key Innovation: Conducted the first systematic field survey using UAV and GPR to classify and elucidate the distribution and progressive evolutionary mechanisms of five primary EBTS distress types in permafrost expressways, identifying high ground temperature and water accumulation as key factors, and proposing a novel EBTS structure to mitigate these issues.
24. Formation of a typical landslide-debris flow disaster chain induced by rainstorm
Core Problem: The increasing occurrence of complex, cascading landslide-debris flow disaster chains induced by extreme rainfall, and the limited understanding of their formation mechanisms and evolutionary dynamics in mountainous regions, leading to severe casualties and socioeconomic losses.
Key Innovation: Systematic analysis of the Gaojiawan landslide-debris flow disaster chain using an integrated methodology (UAV mapping, GIS spatial analysis, and PFC3D numerical simulation) to reconstruct the entire process, analyze energy transformation, and summarize a five-stage formation mechanism. This elucidates evolutionary dynamics and provides a methodological framework for risk mitigation of similar disaster chains.
25. Forecasting step-like reservoir landslide using a physical-mechanical-numerical framework
Core Problem: Significant challenges in accurately forecasting the complex creep behavior and dynamic evolution of step-like reservoir landslides, and the need for improved interpretability of predicted displacements and mechanical parameters in landslide geological models.
Key Innovation: A novel Physical-Mechanical-Numerical (PMN) framework for forecasting step-like reservoir landslide displacements, which integrates a Seepage-mechanical-deformation (SMD) block model with data-driven Bayesian updating and Markov Chain Monte Carlo (MCMC) methods. This framework enhances the interpretability of key geological parameters and monitoring data, improving forecasting accuracy and efficiency compared to traditional models.
26. Toward Systematic Modeling of Volcano Deformation Sources Using Automatically‐Generated InSAR Products
Core Problem: The need for a robust, systematic, and (semi-)automated approach to catalog, model, and compare volcano deformation globally using routinely acquired InSAR data, which is challenging due to the need for effective pre-processing and signal footprint delineation.
Key Innovation: Developed GBIS-BULK, a framework combining filtering, clustering, ICA for noise reduction, and Otsu thresholding for image classification to semi-automatically locate and delimit volcano deformation signal footprints from InSAR data. This enables systematic point source modeling and is designed for global application with routinely processed Sentinel-1 data.
27. How Do Projections of Meteorological Droughts Vary Across Models and Regions?
Core Problem: Quantifying how and where climate change will alter meteorological drought properties (frequency, intensity, length, duration, starting date, severity) and understanding inter-model spread in these projections.
Key Innovation: Used the standardized precipitation index with Earth system models to project future changes in moderate drought events, identifying consistent changes and regional 'dry spots' where severity increases due to prolonged duration, while also acknowledging substantial inter-model spread.
28. Pleistocene Smoothing and Resurfacing of Appalachian Ridgelines by Periglaciation
Core Problem: Isolating the geomorphic impact of Pleistocene periglacial conditions on mid-latitude landscapes from modern climate, tectonics, and rock strength, particularly in the Appalachian ridgelines.
Key Innovation: Found that hilltop curvature and hillslope length in the Appalachians vary with paleotemperature (not modern climate or uplift), suggesting frost cracking and solifluction enhanced hilltop lowering and valley infilling, leaving a long-lived geomorphic signature of permafrost-driven processes.
29. Why Firn Quakes
Core Problem: Firnquakes, audible propagating collapse events in firn, are poorly understood, lacking a stable theory for their conditioning, triggering, and propagation, despite their potential hazardous implications.
Key Innovation: Proposed a theory combining granular and continuum mechanics to explain firnquakes, suggesting that unconsolidated firn at depth, supported by solid-like structures formed by pressure sintering, undergoes dynamic amplification leading to brittle failure and a cascade of collapse propagation, with flexural wavespeed matching recorded firnquake velocities. The framework is extended to similar compacting granular systems with hazardous consequences like landslides or avalanches.
30. High‐Precision Aftershock Distribution Highlights the Complex Fault Geometry of the 2024 Mw 7.5 Noto Peninsula Earthquake
Core Problem: The complex rupture behavior and underlying fault geometry of the 2024 Noto Peninsula earthquake, particularly near its hypocenter, were unclear due to limitations in conventional aftershock detection.
Key Innovation: Deployed 30 temporary seismic stations and used machine learning to precisely locate 46,252 aftershocks, revealing several fault planes and suggesting a new scenario of successive ruptures on adjacent subparallel faults, providing a foundation for improved fault geometry models.
31. Physics‐Informed, Differentiable Hydrologic Models for Capturing Unseen Extreme Events
Core Problem: Machine learning models often perform poorly on unseen extreme events, and it is unclear how well hybrid differentiable models generalize to such events or if optimizing for extremes compromises spatial generalizability and physical significance.
Key Innovation: Evaluated differentiable models (δHBV) against LSTM for predicting unseen extreme floods, finding that δHBV outperformed LSTM for events with longer return periods, especially in peak flow prediction, while maintaining spatial generalization and physical interpretability.
32. Real-Time Proactive Anomaly Detection via Forward and Backward Forecast Modeling
Core Problem: Reactive anomaly detection methods are insufficient for applications demanding timely intervention, while existing proactive methods struggle with handling heterogeneous multivariate data and maintaining precision under noisy conditions.
Key Innovation: Introduces Forward Forecasting Model (FFM) and Backward Reconstruction Model (BRM), hybrid architectures combining TCNs, GRUs, and Transformer encoders to model directional temporal dynamics for proactive anomaly detection, supporting continuous and discrete multivariate features.
33. A Large Language Model for Disaster Structural Reconnaissance Summarization
Core Problem: Previous AI-aided vision-based Structural Health Monitoring (SHM) methods generate only discrete outputs (e.g., damage labels, coordinates), requiring engineers to manually reorganize and analyze results for evaluation and decision-making, especially for rapid post-disaster reconnaissance.
Key Innovation: Proposes LLM-DRS, a novel LLM-based Disaster Reconnaissance Summarization framework. It introduces a standard reconnaissance plan for data collection, processes and integrates text-based metadata and image-based vision data (using DCNNs to extract attributes), and feeds all data into an LLM with designed prompts to generate comprehensive summary reports for individual structures or affected regions, improving post-disaster reconnaissance efficiency and resilience.
34. U-Net with Hadamard Transform and DCT Latent Spaces for Next-day Wildfire Spread Prediction
Core Problem: The need for a lightweight and computationally efficient tool for accurate next-day wildfire spread prediction using multimodal satellite data.
Key Innovation: Develops TD-FusionUNet, a deep learning model incorporating trainable Hadamard Transform and Discrete Cosine Transform layers to capture essential 'frequency' components in orthogonalized latent spaces, along with custom preprocessing, achieving high accuracy and efficiency for wildfire prediction.
35. Experimental study of clear-water local scour around a large-diameter monopile in silty sand
Core Problem: Conventional scour predictors are inadequate for large-diameter monopiles in silty sand due to distinct scour behaviors (prolonged equilibrium times, armoring, shallower depths) not captured by small-scale sand-bed tests.
Key Innovation: Experimental study of clear-water scour around a large-diameter monopile in silty sand, revealing three scour stages, bed armoring, and the influence of approach velocity and flow shallowness. Highlights the need to incorporate partial-depth boundary-layer development, pile-to-sediment size ratio, and armoring effects for improved scour design.
36. Assessing Geo-Foundational Models for Flood Inundation Mapping: Benchmarking Models for Sentinel-1, Sentinel-2, and Planetscope
Core Problem: Despite the potential of geo-foundational models (GFMs) for fast and reliable flood inundation mapping, it remains unclear whether they consistently outperform traditional models like U-Net, and a systematic comparison across various sensors and data availability scenarios is lacking.
Key Innovation: Systematically benchmarks three GFMs (Prithvi 2.0, Clay V1.5, DOFA) and UViT against traditional models for flood inundation mapping using PlanetScope, Sentinel-1, and Sentinel-2 data. It demonstrates that GFMs offer small to moderate improvements in accuracy at lower computational cost and labeling effort, with Clay V1.5 showing competitive performance, superior ability to retain fine details, and better efficiency, especially in few-shot learning scenarios.
37. Collective risk modelling of multi-peril events: correlation of European windstorm gust and precipitation annual severity
Core Problem: Understanding and modeling the correlation of annual losses from multi-peril events (e.g., windstorms and floods) is complex, with existing frameworks needing to better capture the drivers of such correlations, including event dispersion, joint hazard distributions, and interannual modulation.
Key Innovation: This study develops and illustrates three collective risk frameworks of varying complexity, explaining how multi-peril correlation depends on event dispersion and joint hazard variables. It demonstrates that only the framework incorporating interannual modulation of hazard variables can quantitatively capture negative correlations between annual wind and precipitation severity at high thresholds over Europe.
38. ConcentrationTracker: Landlab components for tracking material concentrations in sediment
Core Problem: Existing landscape evolution models often lack the capability to track specific material concentrations within sediment across a landscape, limiting detailed studies of sediment provenance, mixing, and property evolution.
Key Innovation: Development of "ConcentrationTracker" Landlab components, which use a generic mass-balance approach to track any user-defined sediment property (mass, volume, or number concentration) in both sediment and bedrock across hillslope and fluvial environments, enabling detailed studies of sediment provenance and landscape dynamics.
39. Refining the Lagrangian approach for moisture source identification through sensitivity testing of assumptions using BTrIMS1.1
Core Problem: Lagrangian moisture tracking models rely on several assumptions that are seldom thoroughly tested, potentially affecting the accuracy of moisture source identification and limiting their broader applicability.
Key Innovation: Comprehensive sensitivity testing of key assumptions in the Lagrangian model BTrIMS (e.g., number of parcels, release height, vertical movement, identification methods), leading to an improved BTrIMS1.1 model and providing critical information for more realistic and accurate moisture source identification, particularly for understanding precipitation patterns.
40. How well do hydrological models simulate streamflow extremes and drought-to-flood transitions?
Core Problem: Hydrological models struggle to accurately simulate and detect compound extreme events like drought-to-flood transitions, and the factors influencing their performance are not fully understood.
Key Innovation: Demonstrates that general model performance (KGE) doesn't guarantee accurate extreme event detection, highlights the critical role of streamflow timing, and identifies model structure, catchment characteristics, and meteorological forcings as key factors. Reveals poor model representation of these transitions, especially in challenging environments, providing insights for model improvement.
41. Community Flood Resilience Factors; A Community’s Perspective
Core Problem: There is a lack of understanding and tools to truly comprehend and measure flood resilience in detail at a community level, particularly by incorporating the experiences and lay knowledge of at-risk community members.
Key Innovation: Designed and distributed a community flood resilience survey to assess current understanding and identify key factors from the perspective of different community groups, highlighting the importance of integrating lay knowledge and proposing a community-specific flood resilience framework with both established and novel factors.
42. Resilience assessment of interdependent urban infrastructure systems under intensifying typhoon scenarios: Comparing recovery strategies for coupled networks
Core Problem: Assessing the resilience of interdependent urban traffic and power infrastructure systems under intensifying typhoon scenarios and systematically comparing recovery strategies for these coupled networks.
Key Innovation: Developed a comprehensive simulation framework incorporating Monte Carlo analysis to evaluate system performance under intensifying typhoons, explicitly modeling functional, spatial, and restoration interdependencies, quantifying cascading failures, and comparing four distinct recovery strategies to prioritize infrastructure investments for enhanced resilience.
43. Influence of subgrade frost heave on ballasted track dynamics via a hybrid DEM-vehicle model
Core Problem: Subgrade frost heave poses a significant threat to the operational safety of ballasted railways in cold regions, but its detailed impact on the vehicle-track coupled system and the underlying microscopic mechanical evolution are not fully understood.
Key Innovation: A hybrid DEM-vehicle model was established and non-iteratively coupled to systematically analyze the influence of subgrade frost heave on ballasted track dynamics, revealing how frost heave disrupts ballast structure, shifts load-bearing paths, creates unsupported sleepers, and significantly elevates wheel-rail forces and vehicle accelerations.
44. Spatiotemporal prediction and mapping of fractures from successive tunnel faces: an integrated SMF and ST-LSTM framework
Core Problem: Accurate spatiotemporal prediction of fracture distribution ahead of the tunnel face is crucial for construction safety and mitigating geological risks, but existing methods struggle to effectively model the complex spatiotemporal evolution of fractures.
Key Innovation: Development of an enhanced convolutional recurrent neural network (PredRNN with zigzag-shaped spatiotemporal memory flow and dual-channel ST-LSTM) for spatiotemporal prediction and mapping of fracture sequences, demonstrating superior stability and accuracy in forecasting fracture distribution in unexcavated zones during tunnel excavation.
45. Experimental apparatus and mechanical properties of shield tunnel segment joint under cyclic bending moment
Core Problem: Research on the mechanical behavior of shield tunnel segment joints under cyclic loading, crucial for seismic performance design, is scarce, particularly regarding their response to positive and negative bending moments.
Key Innovation: Development of a novel cyclic loading device capable of applying both positive and negative bending moments to tunnel segment joints, enabling comprehensive investigation of their cyclic mechanical properties (failure modes, hysteretic behavior, stiffness degradation, energy dissipation) and providing a basis for seismic design.
46. Experimental and numerical research on the tensile performance of a novel three-layer ring spring flexible joint for shield tunnels
Core Problem: Shield tunnel joints are prone to tensile failure under longitudinal ground deformation in zones with abrupt geological changes, and conventional flexible joints may suffer from waterproofing failure due to excessive deformation.
Key Innovation: Proposal and validation of a novel three-layer ring spring (TRS) flexible joint for shield tunnels, which reduces joint stiffness, provides energy-dissipation and self-centering characteristics, incorporates a displacement-limiting mechanism to prevent waterproofing failure, and significantly improves the safety factor against tensile damage under ground deformation.
47. Integrating distributed hydrologic simulation with low-flow resilience: a spatiotemporal perspective
Core Problem: Limited attention has been given to understanding the spatiotemporal dynamics of hydrological resilience, particularly under prolonged low-flow conditions, especially in poorly gauged basins.
Key Innovation: Leveraged a fully distributed watershed hydrologic model to investigate low-flow resilience in a poorly gauged basin, characterizing recessions by duration and magnitude, computing area-normalized metrics, and examining temporal dynamics, spatial distribution, and progression patterns, identifying a 3-day rainfall deficit as a critical intervention window for drought preparedness.
48. Unsteady diffusion characteristics of pulsed grouting considering grout-rock hydromechanical coupling effect
Core Problem: Grouting in deep fractured rock masses is challenging due to restricted grout propagation and fracture clogging, and the underlying mechanisms by which pulsed grouting enhances grout penetration under coupled grout-rock hydromechanical conditions are insufficiently understood.
Key Innovation: Developed an unsteady grouting diffusion model for Bingham fluids incorporating spatiotemporal grout rheology and coupled grout-rock hydromechanical interactions. Demonstrated that low-frequency pulsed grouting significantly enhances fracture groutability (e.g., 314.5% increase in injected volume at 4 Hz) through a synergistic mechanism of yield-threshold rectification and cubic law amplification with elastic discharge, mitigating grout front blockage.
49. Modeling non-Newtonian fluid–solid flows containing non-spherical particles by the SPH-DEM coupling model
Core Problem: Accurately simulating complex interaction dynamics in fluid-solid systems composed of non-Newtonian fluids and non-spherical particles, which are common in geological phenomena like debris flows.
Key Innovation: Develops a resolved SPH-DEM coupling model using the power-law model for non-Newtonian fluids and super-ellipsoid/polyhedral models for non-spherical particles, with improved boundary treatment, validated for complex fluid-solid interaction scenarios relevant to geological studies.
50. Rapid desorption-induced multiscale structural alterations in gas-bearing coals from Sydney Basin: Implications for coal and gas outburst
Core Problem: The lack of understanding of multiscale structural alterations in coal induced by rapid gas desorption, which is crucial for managing severe coal and gas outbursts in deep underground mining.
Key Innovation: A novel methodology integrating high-resolution 3D micro-computed tomography (μ-CT), scanning electron microscopy (SEM), low-pressure gas adsorption (LPGA-N2 and LPGA-CO2), and helium-based void volume measurements to quantify multiscale structural alterations in coal during rapid gas desorption. This establishes a clear link between gas pressure, desorption dynamics, and coal structural alteration, providing a scientific basis for refining gas threshold limits and improving outburst-risk management.
51. Revisiting the strainburst intensity of roadways in coal mines: Theory and case study
Core Problem: The need for a more accurate and comprehensive criterion to evaluate the intensity of strainbursts, a severe dynamic hazard frequently encountered in underground coal mine roadways, to improve safety.
Key Innovation: Development of a new critical energy storage density index for strainburst initiation and a novel assessment criterion for the bursting intensity of strainburst-prone roadways, based on a damage model. The study quantifies the effects of coal brittleness, rock strength, and roadway support, and validates the theory with in situ rockburst cases.
52. Monitoring the In Situ Nonlinear Elasticity Near the Dalk Glacier Area, Antarctica, Using Dense Seismic Arrays
Core Problem: The need for effective monitoring of disturbances in the subsurface medium of Antarctic glaciers and their connection to environmental changes to address sea-level rise and understand low-amplitude precursors to systemic imbalances.
Key Innovation: Utilized ambient noise data from dense seismic arrays and coda wave interferometry to measure in situ nonlinear elasticity (dv/v) near Dalk Glacier. Demonstrated that semi-diurnal dv/v variations are controlled by tidal strain, humidity, and ice melt dynamics, with sensitivity to melt rate exceeding total melt volume, providing a cost-effective, high-resolution monitoring technique for glacier mass changes and potential geohazard forecasting.
53. High‐Resolution Channel Geometry Reveals Contrasting Styles of Gravel River Adjustment
Core Problem: Existing descriptions of downstream scaling between channel geometry and drainage area for rivers rely on limited observations, potentially obscuring high-resolution, catchment-specific variations and underlying controls on channel adjustment.
Key Innovation: Developed a novel method to automatically extract bankfull width and determine high-resolution (10-m), catchment-specific width-area scaling for six gravel rivers, revealing that average width becomes slope-independent below a threshold slope and that slope and width deviations display contrasting patterns depending on the presence of knickpoints.
54. Frequency‐Dependent Seismic Velocity Variations Reveal Layered Aquifer Behavior Under Groundwater Fluctuations
Core Problem: Disentangling the contemporaneous effects of pore saturation, pore pressure change, and mass loading on aquifer behavior under groundwater fluctuations, and how these manifest in frequency-dependent seismic velocity variations.
Key Innovation: Demonstrated that pore saturation and pore pressure change influence shallow aquifer layers, while mass loading governs deeper responses, highlighting the potential of passive seismic techniques to monitor layered hydromechanical processes in aquifer systems.
55. Tide‐Modulated Ocean‐to‐Earth Energy Conversion Quantified With Coastal Fiber Sensing
Core Problem: Quantifying how incident ocean waves transfer energy into seismic surface waves along the nearshore, and the scarcity of time-resolved in-situ estimates of this process.
Key Innovation: Used a beach-deployed distributed acoustic sensing array to directly measure the conversion efficiency from wave impacts to Rayleigh-type ground motion (on the order of 10^-6), finding this efficiency is strongly modulated by tide, and establishing a framework for coastal monitoring using existing fiber infrastructure.
56. Flow History Effects on River Bifurcation Dynamics in a Himalayan River
Core Problem: Understanding how variations in seasonal discharge patterns cause instability in channel position and flow distribution at river bifurcations, and how these changes impact the location and extent of flooding.
Key Innovation: Analyzed 30 years of satellite images and daily discharge records of the Karnali River, Nepal, demonstrating that two large peak flows in a monsoon season consistently precede major changes in flow partitioning, suggesting a history-dependent threshold related to bed armoring.
57. Quantifying and Regionalizing Land Use Impacts on Catchment Response Times With High‐Frequency Observations
Core Problem: It is difficult to generate robust and quantitative evidence of land use impacts on hydrological response times at the catchment scale, which affects the development of rainfall-runoff models for ex-ante predictions.
Key Innovation: Analyzed high-frequency observational data from 16 paired catchments in the tropical Andes, finding a statistically significant impact of intensive land use on hydrological response time (quicker response), and developed a robust methodology to quantify and regionalize these impacts.
58. Hierarchical Testing of a Hybrid Machine Learning-Physics Global Atmosphere Model
Core Problem: While ML-based atmospheric models show high skill, their fidelity across various timescales, under out-of-distribution forcings (like uniform warming), and the sources of their biases remain underexplored, hindering their reliability for Earth science applications.
Key Innovation: Conducts hierarchical testing of NeuralGCM, a hybrid ML-physics global atmosphere model, across synoptic, interannual, and out-of-distribution uniform-warming scenarios. It demonstrates comparable performance to ESMs for synoptic-scale phenomena and teleconnection patterns, and reasonable responses to warming, while identifying specific weaknesses in cyclone tracking, teleconnected wave trains, and stratospheric circulation responses.
59. Experimental study of hydrodynamics and soil responses around a sandbar and their effects on bed-level evolution
Core Problem: Understanding the complex wave-sediment interactions and their effects on sandbar evolution and erosion, especially under long-period waves, is crucial for accurate erosion prediction.
Key Innovation: Conducted a scaled experimental study to analyze hydrodynamics, pore pressure, and bed-level changes around a sandbar. Findings clarify that long-period waves enhance seepage, promote liquefaction, and increase sediment transport, with erosion correlating strongly with relative wave height and Lagrangian drift velocity.
60. Shaking table tests of four-bucket jacket foundation for offshore wind turbines in soft clay
Core Problem: Evaluating the dynamic characteristics and seismic behavior of four-bucket jacket foundations for Offshore Wind Turbines (OWTs) in soft clay is crucial for their seismic design, as existing knowledge on their failure mechanisms and soil-structure interaction under strong earthquakes is limited.
Key Innovation: Conducted shaking table experiments under various seismic conditions to summarize seismic response laws and identify failure mechanisms. Findings include significant superstructure acceleration, potential exceedance of serviceability limits for lateral displacement, and the role of dynamic soil-structure interaction in excess pore water pressure accumulation, with bucket walls suppressing EPWP inside.
61. THE INFLUENCE OF STRUCTURE PERMEABILITY ON WAVE OVERTOPPING AT ROCK ARMOURED SLOPES ALONG EUROTOP
Core Problem: The EurOtop Manual's roughness factor (γf) for rock-armoured slopes, which accounts for structure permeability, lacks consistent validation across various layer compositions, leading to potential inaccuracies in wave overtopping predictions.
Key Innovation: The study conducted 591 new small-scale physical model tests, revealing that core permeability only affects overtopping for low steepness waves on steep front slopes. It revises the roughness factor for impermeable cores to γf = 0.40 (matching permeable cores) and updates the influence of wave steepness, significantly improving prediction accuracy (reducing error by 2-3 times).
62. Characterizing Short-Duration Rainfall Events on the Tibetan Plateau Based on Ground Observation, Satellite Remote Sensing, and Reanalysis Data
Core Problem: The duration of precipitation is a critical control on hydrological and geomorphological responses (e.g., runoff, soil erosion, sediment transport) on the Tibetan Plateau, but the spatial distribution and characteristics of short-duration rainfall events, and the accuracy of satellite and reanalysis data in capturing them, are not well understood.
Key Innovation: Analyzes the spatial distribution of precipitation events with different durations across the Tibetan Plateau using ground observations, satellite remote sensing (IMERG), and reanalysis data. It finds that short-duration (1–3 h) events contribute over 50% of total rainfall in the central and western TP, and demonstrates that IMERG performs well in estimating precipitation event contributions, while reanalysis data tend to systematically underestimate short-duration events.
63. The newly developed Multi-ensemble Biomass-burning Emissions Inventory (MBEI): characterizing and unraveling spatiotemporal uncertainty in global biomass burning emissions
Core Problem: Large discrepancies among existing inventories hinder a consensus on the true magnitude and long-term trends of global biomass burning emissions, making it difficult to accurately quantify emission uncertainty and understand spatiotemporal variability.
Key Innovation: This study develops the Multi-ensemble Biomass-burning Emissions Inventory (MBEI), a framework integrating bottom-up and top-down approaches to generate eight sub-inventories (2003–2023), explicitly quantifying emission uncertainty at a 0.1° grid scale. It reveals significant spatial heterogeneity in uncertainty and identifies a decadal shift in global emissions, providing a robust central estimate and uncertainty bounds for atmospheric modeling and climate assessments.
64. Projects on silty Himalayan rivers raise alarms
Core Problem: The potential hazards and risks associated with projects on highly silty rivers in the geologically active Himalayan region.
Key Innovation: The paper likely highlights specific environmental or geological concerns, or new insights into the risks posed by such projects.
65. Meso-structural characterization and discrete element modeling of freeze-thaw damage in sandstone using polarized light images
Core Problem: The microscopic mechanisms and damage evolution of sandstone under freeze-thaw cycling, particularly how mineral anisotropy and water-ice phase transitions drive crack formation and material degradation, are not fully understood.
Key Innovation: A combined approach using polarized light microscopy and discrete element modeling was used to characterize meso-structural damage and simulate freeze-thaw evolution in sandstone, revealing a 'mineral-phase transition' coupling damage mechanism and quantifying the reduction in strength and elastic modulus.
66. Intelligent and accurate recognition method of lining cracks under complex seam interference and engineering application
Core Problem: Accurate early detection of micro-scale lining cracks in tunnels is challenging due to their elusive nature and interference from construction seams, impeding effective maintenance and structural integrity assessment.
Key Innovation: An intelligent recognition method combining an enhanced U-Net (with multi-view fusion and dynamic snake convolution) for micro-crack segmentation, a compound loss function (Dice and CLDice) for topological preservation, and a two-step skeleton processing strategy with a Graph Convolutional Network (GCN) for robust seam elimination, significantly improving crack detection precision and recall.
67. Experimental investigation of frictional healing in granite: temperature-mineral composition interplay
Core Problem: Understanding the influence of temperature and mineral composition on the frictional healing behavior of rock fractures, crucial for the long-term stability of geological repositories and general rock mass stability.
Key Innovation: Demonstrated that frictional healing rates in granite are influenced by lithology-dependent thermo-mechanical-chemical effects, with quartz/feldspar-rich granite showing greater healing enhancement at elevated temperatures, and provided insights into the potential for unstable slip based on rate-and-state friction analysis.
68. Study on critical mechanical parameters for fault slip instability
Core Problem: The key mechanical parameters governing fault instability, particularly in tectonic coal under varying stress and slip rates, are poorly understood, compromising safety and efficiency in deep coalbed methane extraction.
Key Innovation: Conducts slip experiments on tectonic coal to identify critical normal stress and slip rate, demonstrating that a drop in shear stress (driven by friction from convex surface structures) triggers instability, and quantifies the relationship between stress drop, crack width, and fault failure risk, identifying a critical threshold for slip-induced instability.
69. Experimental and numerical investigations on shear-compression behavior of thick rubber bearings
Core Problem: The stability and compressive capacity of Thick Rubber Bearings (TRBs) under large lateral displacements, especially when used for mitigating earthquake-induced vibrations, are not fully understood, potentially compromising their effectiveness.
Key Innovation: Conducted experimental and numerical analyses of TRB shear-compression behavior, validated FE models, developed an improved prediction method for normalized critical load, and performed probabilistic studies to assess the critical factor distribution, demonstrating TRBs' robust compressive capacity even under high shear strains.
70. Investigation on seismic isolation performance of surface structures based on concrete-rubber layered periodic foundations
Core Problem: Building collapse during earthquakes remains a major cause of catastrophic losses, highlighting the need for enhanced seismic resilience in structures.
Key Innovation: Investigated the seismic isolation performance of a novel Concrete-Rubber Layered Periodic Foundation (CRLPF) through shaking table tests and validated numerical simulations, demonstrating its effectiveness in attenuating both acceleration (up to 34.41%) and inter-story drift (up to 17.19%) of superstructures during seismic events.
71. Behaviour of tuned hysteretic inerter viscous damper assisted structure subjected to soil dependent stochastic seismic excitation and recorded ground motions
Core Problem: Enhancing the seismic performance of structures to reduce displacement and acceleration responses during earthquakes remains a critical challenge.
Key Innovation: Proposed and evaluated a novel Tuned Hysteretic Inerter Viscous Damper (THIVD) configuration, demonstrating its superior performance in reducing structural displacement (53%-62%) and acceleration (37%-45%) compared to uncontrolled structures and linear tuned viscous mass dampers, even under soil-dependent stochastic seismic excitations.
72. Multi-scale degradation of polypropylene geostrips retrieved from a 21-year failed geosynthetic-reinforced soil retaining wall
Core Problem: The challenge in predicting the long-term performance of geosynthetic-reinforced soil (GRS) retaining walls due to limited understanding of in-service degradation mechanisms of geosynthetic materials and a scarcity of field-obtained data, leading to potential structural failures.
Key Innovation: A multi-scale analytical investigation of polypropylene geostrips retrieved from a 21-year failed GRS wall, integrating dimensional measurements, tensile testing, SEM, EDS, FTIR, and DSC. The study reveals pronounced, depth-dependent degradation profiles (oxidative chain scission, surface oxidation, cracking, and fibrillation) and demonstrates that current design reduction factors may underestimate actual strength loss, highlighting the need for refined durability-oriented design approaches.
73. Multiphysics modeling of thermo-hydraulic fracturing during CO2 sequestration in multilayered reservoirs at Ordos, China
Core Problem: Unexplained permeability variations and reduced CO2 storage capability in multilayered geological CO2 sequestration (GCS) reservoirs indicate significant thermo-hydraulic fracturing, requiring a comprehensive multiphysics model to understand the underlying mechanisms.
Key Innovation: Developed a multiphase and thermal-hydraulic-mechanical (THM) coupled numerical model for multilayered GCS reservoirs, which successfully matched monitored pressure and revealed that low-rate CO2 injection can induce significant thermal-hydraulic fracturing and permeability increases (e.g., 4 orders of magnitude), providing a full picture of system variation and benefiting fluid injection works.
74. Quantifying Vein Network Permeability in Dehydrated Serpentinites Using Thermodynamics and Generative AI
Core Problem: Poor understanding of the mechanisms of fluid escape in subduction zones, despite fluids influencing volcanism, tectonics, and geochemical cycling.
Key Innovation: Combined X-ray tomography, drone imagery, generative machine learning, electron microscopy, and equilibrium thermodynamics to model and analyze fluid pathways in exhumed meta-serpentinites. Demonstrated that dehydration vein networks act as efficient drainage systems, enabling rapid fluid percolation even at low porosities (<1%), with maximum network permeability several orders of magnitude higher than intact serpentinite.
75. Irreversible Transitions of the Ocean Circulation in Antarctic Ice‐Shelf Cavities
Core Problem: Understanding the mechanisms driving cold-to-warm transitions in Antarctic ice-shelf cavities, which can lead to dramatic increases in sea-level rise, and the role of brine rejection and ocean circulation in these transitions.
Key Innovation: Developed a generic low-dimensional box model demonstrating that brine rejection in coastal polynyas promotes bistable dynamics in ice-shelf cavities, making most cavities susceptible to irreversible abrupt transitions for realistic sea-ice formation rates, highlighting the robustness of bistability to parameterization changes.
76. Seismic Waves Do Sense Fracture Connectivity: Experimental Validation
Core Problem: Lack of experimental validation for the theoretical suggestion that seismic waves are sensitive to fracture connectivity, a key parameter for fluid flow in the Earth's crust.
Key Innovation: Developed a novel methodology to fabricate synthetic rock samples with controlled connected/unconnected fluid-saturated fractures and experimentally demonstrated that P-wave velocities are lower in samples with connected fractures, corroborated by numerical simulations attributing the difference to wave-induced fluid pressure diffusion.
77. RESIdual STability (RESIST) Calibration for Improved Hydrological Model Time Generalizability
Core Problem: Hydrological models calibrated on specific periods often show decreased accuracy when extrapolated to periods with different climate conditions, limiting their temporal generalizability for climate-impact studies.
Key Innovation: Developed RESIST, a novel calibration objective that jointly calibrates model accuracy and time-invariance of residuals, demonstrating reduced accuracy loss during extrapolation and weakened dependence of residuals on forcing variations, thereby improving temporal generalizability.
78. Adaptive Physics Transformer with Fused Global-Local Attention for Subsurface Energy Systems
Core Problem: Full physics numerical simulations of subsurface energy systems are computationally expensive due to geological heterogeneity, high resolution requirements, and tightly coupled physical processes with distinct propagation time scales.
Key Innovation: Proposes the Adaptive Physics Transformer (APT), a geometry-, mesh-, and physics-agnostic neural operator that fuses a graph-based encoder with a global attention mechanism to efficiently model subsurface systems, outperforming state-of-the-art architectures, demonstrating super-resolution, and learning from adaptive mesh refinement simulations.
79. Advancing Digital Twin Generation Through a Novel Simulation Framework and Quantitative Benchmarking
Core Problem: The generation of digital twins from real-world objects via photogrammetry involves many design choices, and differences between approaches are largely judged qualitatively, lacking repeatable, quantifiable experiments.
Key Innovation: A novel pipeline for generating synthetic images from high-quality 3D models and programmatically generated camera poses, enabling repeatable, quantifiable experiments to benchmark digital twin generation methods against ground-truth knowledge.
80. MDE-VIO: Enhancing Visual-Inertial Odometry Using Learned Depth Priors
Core Problem: Traditional monocular Visual-Inertial Odometry (VIO) systems struggle with accurate pose estimation in low-texture environments due to insufficient sparse visual features, and existing dense Monocular Depth Estimation (MDE) models are often too computationally demanding for real-time edge deployment.
Key Innovation: A novel framework, MDE-VIO, that integrates learned dense depth priors directly into the VINS-Mono optimization backend, enforcing affine-invariant depth consistency and pairwise ordinal constraints with variance-based gating, achieving robust metric scale recovery and significant accuracy gains (up to 28.3% ATE reduction) while adhering to edge device computational limits.
81. Toward Adaptive Non-Intrusive Reduced-Order Models: Design and Challenges
Core Problem: Projection-based Reduced Order Models (ROMs) are typically static surrogates, limiting their utility when a system's dynamics evolve beyond the initial training manifold.
Key Innovation: Formalization and study of adaptive non-intrusive ROMs, including Adaptive OpInf, Adaptive NiTROM, and a hybrid approach, which update both the latent subspace and reduced dynamics online, demonstrating robust performance and near-exact energy tracking for transiently perturbed systems, even under regime changes and minimal offline data.
82. A Dual-Branch Framework for Semantic Change Detection with Boundary and Temporal Awareness
Core Problem: Existing Semantic Change Detection (SCD) methods from bi-temporal remote sensing images suffer from blurred boundaries and inadequate temporal modeling, limiting segmentation accuracy.
Key Innovation: Proposed DBTANet, a dual-branch framework with a frozen SAM branch for global semantic context/boundary priors and a ResNet34 branch for local details. It includes a Bidirectional Temporal Awareness Module (BTAM) for multi-scale feature aggregation and temporal dependencies, and a Gaussian-smoothed Projection Module (GSPM) for boundary-aware constraints, achieving state-of-the-art performance.
83. pycopm: An open-source tool to tailor OPM Flow geological models
Core Problem: Engineers need to quickly update and adjust geological models for reservoir simulations as new information becomes available, but this process can be slow and difficult.
Key Innovation: Introduces pycopm, an open-source tool that simplifies and accelerates the process of tailoring OPM Flow geological models by allowing users to adjust grids, focus on specific reservoir parts, and change model shapes and positions, supporting efficient scenario testing and troubleshooting.
84. Efficient Segment Anything with Depth-Aware Fusion and Limited Training Data
Core Problem: Existing Segment Anything Models (SAM) require massive datasets and rely solely on RGB inputs, limiting their efficiency and applicability, especially when depth information could provide valuable geometric priors for segmentation.
Key Innovation: Proposes a lightweight RGB-D fusion framework that augments EfficientViT-SAM with monocular depth priors. By fusing depth maps mid-level with RGB features, the method achieves higher accuracy than EfficientViT-SAM with significantly less training data (less than 0.1% of SA-1B), demonstrating the strong geometric priors provided by depth cues for segmentation.
85. Robust Optimization Approach and Learning Based Hide-and-Seek Game for Resilient Network Design
Core Problem: Designing resilient and reliable communication networks with limited signal transfer distance and uncertain link lengths/node availability, ensuring full connectivity under worst-case scenarios while minimizing regenerator deployment costs.
Key Innovation: A robust optimization framework combined with scalable solution methods (column-and-constraint generation, Benders decomposition, iterative robust optimization) and a learning-based hide-and-seek game to design communication networks that guarantee full connectivity under dynamic budgeted uncertainty in link lengths and regenerator installation costs.
86. Momentum LMS Theory beyond Stationarity: Stability, Tracking, and Regret
Core Problem: In large-scale data processing, especially with sequential streams from complex systems, data often exhibit nonstationarity (drifting distributions, time-varying parameters), challenging theoretical analysis and requiring computationally efficient, real-time adaptive algorithms.
Key Innovation: Investigates the Momentum Least Mean Squares (MLMS) algorithm as an adaptive identification tool for nonstationary settings. It theoretically derives tracking performance and regret bounds for MLMS in time-varying stochastic linear systems, demonstrating its rapid adaptation and robust tracking capabilities for modern streaming and online learning applications.
87. EO-VAE: Towards A Multi-sensor Tokenizer for Earth Observation Data
Core Problem: State-of-the-art generative models lack effective tokenizers for Earth observation (EO) data, which presents unique challenges due to diverse sensor specifications and variable spectral channels, hindering efficient latent representation.
Key Innovation: EO-VAE, a multi-sensor variational autoencoder designed as a foundational tokenizer for the EO domain. It uses a single model with dynamic hypernetworks to encode and reconstruct flexible channel combinations, achieving superior reconstruction fidelity and establishing a robust baseline for latent generative modeling in remote sensing.
88. Function-Space Decoupled Diffusion for Forward and Inverse Modeling in Carbon Capture and Storage
Core Problem: Accurate characterization of subsurface flow, critical for CCS (and other geological applications), is challenged by the ill-posed nature of inverse problems with sparse observations.
Key Innovation: Fun-DDPS, a generative framework combining function-space diffusion models with differentiable neural operator surrogates for both forward and inverse modeling. It learns a prior distribution over geological parameters and uses a Local Neural Operator surrogate for physics-consistent guidance, achieving significant error reduction in forward modeling with sparse data and physically consistent inverse solutions.
89. Hybrid operator learning of wave scattering maps in high-contrast media
Core Problem: Existing neural operators struggle with accurately modeling wave propagation and scattering in high-contrast heterogeneous media (e.g., subsurface models with salt bodies) due to strong scattering and phase sensitivity.
Key Innovation: Proposes a hybrid architecture that decomposes the scattering operator into a smooth background propagation (learned by FNO) and a high-contrast scattering correction (modeled by a vision transformer using attention), achieving substantially improved phase and amplitude accuracy for high-frequency Helmholtz problems relevant to seismic imaging.
90. Latent Generative Solvers for Generalizable Long-Term Physics Simulation
Core Problem: Generalizable long-term physics simulation across diverse PDE systems is challenging due to rollout drift in autoregressive prediction and the need for scalable pretraining and efficient adaptation to out-of-distribution data.
Key Innovation: Introduces Latent Generative Solvers (LGS), a two-stage framework that maps PDE states to a shared latent space (VAE) and learns probabilistic latent dynamics (Transformer with flow matching). Key mechanisms include an uncertainty knob to correct rollout drift and flow forcing to align conditioning, enabling generalizable, uncertainty-aware long-term forecasting with significantly lower computational cost and efficient adaptation.
91. The Cost of Learning under Multiple Change Points
Core Problem: Online learning in environments with multiple change points is challenging, as classical 'high confidence' detection schemes can fail due to 'endogenous confounding,' leading to high regret.
Key Innovation: Proposes a new class of horizon-free online algorithms called Anytime Tracking CUSUM (ATC) that implement a selective detection principle, balancing ignoring small shifts with quickly reacting to significant ones, achieving nearly minimax-optimal regret performance.
92. A Survey on Dynamic Neural Networks: from Computer Vision to Multi-modal Sensor Fusion
Core Problem: Large Computer Vision models are challenging to deploy on embedded devices due to computational demands, and static optimization techniques fail to adapt computations to varying input complexities.
Key Innovation: Provides a comprehensive survey and logical taxonomy of Dynamic Neural Networks, highlighting their benefits for adaptivity, noise reduction, and information prioritization, particularly in multi-modal sensor fusion, which is crucial for efficient deployment.
93. Remote Sensing Retrieval-Augmented Generation: Bridging Remote Sensing Imagery and Comprehensive Knowledge with a Multi-Modal Dataset and Retrieval-Augmented Generation Model
Core Problem: Existing remote sensing Vision-Language Models (VLMs) typically rely on closed-set scene understanding and lack the ability to incorporate external domain-specific or world knowledge, limiting their semantic reasoning capabilities.
Key Innovation: Introduces RSWK, a multimodal dataset of high-resolution satellite imagery and detailed textual descriptions for 14,141 landmarks, and RS-RAG, a framework that integrates a Multi-Modal Knowledge Vector Database with a Knowledge Retrieval and Response Generation module to enhance VLM performance in remote sensing tasks.
94. Beyond Accuracy: A Stability-Aware Metric for Multi-Horizon Forecasting
Core Problem: Traditional time series forecasting methods optimize solely for accuracy, neglecting temporal consistency (stability) in multi-horizon predictions, which is crucial for reliable decision-making.
Key Innovation: The 'forecast accuracy and coherence score' (forecast AC score), a stability-aware metric for probabilistic multi-horizon forecasts that accounts for both accuracy and temporal consistency, leading to substantial improvements in forecast stability and medium-to-long-horizon accuracy when used as a differentiable objective.
95. Learning Physics-Grounded 4D Dynamics with Neural Gaussian Force Fields
Core Problem: Predicting physical dynamics from raw visual data to generate consistently physically plausible videos remains a major challenge, with existing methods being computationally expensive or lacking robustness in complex real-world scenarios.
Key Innovation: Neural Gaussian Force Field (NGFF), an end-to-end neural framework that integrates 3D Gaussian perception with physics-based dynamic modeling to generate interactive, physically realistic 4D videos from multi-view RGB inputs, achieving significantly faster simulation and strong generalization in physical reasoning.
96. Controlling Dynamical Systems into Unseen Target States Using Machine Learning
Core Problem: Controlling complex dynamical systems into previously unseen target states, especially those with significantly different or chaotic dynamics, is challenging with existing model-free and data-driven methodologies.
Key Innovation: A novel, model-free, data-driven methodology using parameter-aware next-generation reservoir computing (NGRC) to accurately predict and control complex dynamical systems into arbitrary, even fundamentally different, unseen target states, demonstrated on a nonlinear power system model to ensure fast, transient-free transitions and avoid system collapse.
97. Quantifying pore-scale heterogeneity of drying and wetting soil water retention behavior using X-ray computed tomography
Core Problem: Effective methods are lacking for determining local degree of saturation and matric suction, and explaining hydraulic hysteresis in individual pores of unsaturated soils with high spatial resolution.
Key Innovation: A novel workflow using X-ray CT images was proposed to quantify local saturation and matric suction at individual pores, revealing significant spatial variations and the governing roles of pore geometry, size, and connectivity in hysteretic water retention behavior.
98. A physics-informed enhanced transformer network on wave overtopping forecasting for composite breakwaters
Core Problem: Existing physics-driven or data-driven methods for wave overtopping prediction for composite breakwaters lack accuracy, efficiency, and generalization due to the nonlinear and multi-scale processes involved.
Key Innovation: Proposes a Physics-Informed enhanced Transformer Network (PITN) that combines Transformer's temporal modeling with Boussinesq equation physical constraints. It outperforms existing models in reproducing spatiotemporal distributions and predicting average overtopping discharge, showing improved accuracy and stability.
99. Oblique wave interaction with (M+1) U-shaped thick porous structures interconnected with (M) thin porous boxes in the presence of dual trenches near a partially reflected wall
Core Problem: Need for comprehensive understanding of hydrodynamic behavior of complex hybrid coastal structures under oblique wave incidence to optimize their design for various coastal applications.
Key Innovation: Presents a multi-domain boundary element framework to investigate wave scattering, reflection, and forces on a hybrid coastal structure. Reveals that increasing interconnected structures, radius, thick porosity, and porous effect parameter reduce wave run-up, forces, and reflection, with U-shaped structures playing a major role in mitigating forces.
100. Reply to Discussion by A. Khayyer and C.H. Lee on “comparative study on volume conservation among various SPH models for flows of different levels of violence, coastal engineering, volume 191, August 2024, 104521”
Core Problem: Ensuring accurate volume and mechanical energy conservation in various Smoothed Particle Hydrodynamics (SPH) models, particularly for complex and violent flow phenomena relevant to coastal engineering, and addressing specific concerns raised by other researchers regarding model application and performance.
Key Innovation: A further investigation into volume conservation of SPH models, introducing new volume error indicators, numerical cases (including wave shoaling, breaking, and violent sloshing), and analyses, providing new insights into model performance, volume and energy conservation, and the effects of boundary conditions for flows of different violence levels.
101. DSM Accuracy Enhancement via Satellite Stereo Imagery Super-Resolution: Multiscale Attention and Coordinate Mapping Optimization
Core Problem: Inherent limitations in resolution and elevation accuracy of Digital Surface Models (DSMs) extracted from contemporary stereo optical satellite imagery, and the failure of existing super-resolution methods to address geometric model degradation in stereo pairs.
Key Innovation: Introduced a novel stereo image pair super-resolution (SPSR) methodology using a dual aggregation transformer with multiscale attention (DATMAT) network, which significantly enhances DSM vertical accuracy (e.g., 2.46m RMSE on WV2) by improving geometric model recovery and rectifying geolocational fidelity loss post-SR.
102. A Model for Land Cover Ecological State Assessment of Southern Ukraine Based on Remote Data for Analyzing the Consequences of the Kakhovka Reservoir Shallowing
Core Problem: There is a lack of comprehensive quantitative assessments that combine multiparametric analysis of physical, hydrological properties of the surface soil layer and land cover parameters using high spatial resolution satellite data to evaluate the ecological state of land, especially in war-damaged territories and after events like the Kakhovka Reservoir shallowing.
Key Innovation: Creates a model for assessing land cover ecological state of southern Ukraine based on high spatial resolution (10 m) satellite data (Sentinel, Landsat, MODIS) and machine learning. It forms an integrated ecological state index (ESI) using principal component analysis on NDVI, land surface temperature, soil moisture deficit, and drought severity index, revealing a significant deterioration in the ecological state of the studied territory, particularly after the Kakhovka HPP destruction.
103. PolyU2025 SLA: A global 0.25°×0.25° monthly sea level anomaly dataset (1993–2024) determined from satellite altimetry for sea-level and climate change research
Core Problem: Long-term and spatially consistent sea-level anomaly (SLA) products from satellite altimetry are fundamental for sea-level and climate change studies, but independent, regularly updated datasets are valuable for multi-product assessments and understanding regional variability.
Key Innovation: This study develops PolyU2025 SLA, a new global 0.25°×0.25° monthly gridded sea level anomaly product (1993–2024) generated using a fully independent data-processing framework. It demonstrates high consistency with existing products and tide-gauge observations, providing a stable and complementary dataset for climate-scale applications and multi-product assessments of regional sea-level variability.
104. Preserving water under megacities is crucial — and urgent
Core Problem: The critical and urgent need to preserve water resources beneath megacities, likely due to implications for urban sustainability and ground stability.
Key Innovation: Highlighting the paramount importance and urgency of managing and preserving subterranean water resources in urban environments.
105. Revive Brazil’s soy moratorium
Core Problem: The weakening of Brazil's soy moratorium, an agreement that successfully reduced Amazon deforestation, due to being deemed 'anticompetitive' and facing regulatory sanctions, thereby threatening forest conservation efforts and indirectly increasing the risk of deforestation-related geohazards.
Key Innovation: Advocating for strengthening multilateral partnerships between businesses and governments to revive and enforce the soy moratorium for the sake of forest conservation, which serves as a mitigation strategy against deforestation-related geohazards.
106. Bridging disaster resilience gaps: exploring adaptation preferences and time commitment willingness in isolated island communities
Core Problem: Strengthening disaster resilience in geographically isolated communities requires understanding how residents prioritize and engage in adaptation activities under conditions of limited institutional support.
Key Innovation: Explores adaptation preferences and time commitment willingness in isolated island communities to bridge disaster resilience gaps, providing insights into resident engagement in adaptation activities.
107. An Integrated Assessment of Heat Hazard, Vulnerability, and Accessibility to Climate Shelter Networks for Identifying Urban Adaptation Priority Areas in Andalusia (Spain)
Core Problem: Climate change is intensifying extreme heat in urban areas, posing risks to public health, but the deployment of climate shelters often lacks systematic consideration of spatial patterns of heat hazard and social vulnerability.
Key Innovation: Developed an integrated framework combining heat hazard and vulnerability indices with network-based accessibility analysis to identify urban adaptation priority areas for climate shelter deployment, revealing significant spatial mismatches and highlighting areas of compounded vulnerability in Andalusia.
108. Optimizing post-disaster road restoration with reinforcement learning: A traveler-behavior-aware approach
Core Problem: Optimizing short-term road network restoration in disaster-affected areas is complex due to the need to consider travelers' behavior, their gradual adaptation to network changes, limited recovery resources, and uncertainties in recovery times.
Key Innovation: The Traveler-Adaptive Restoration Mechanism (TARM), which integrates Reinforcement Learning, Markov Decision Process, and optimization-based day-to-day traffic simulation to optimize post-disaster road restoration, highlighting the influence of traveler route choices and information dissemination on optimal policies.
109. Multi-objective optimisation of arctic shipping routes considering navigational risk, voyage time, and fuel consumption
Core Problem: Optimizing Arctic shipping routes requires simultaneously balancing conflicting objectives of navigational risk, voyage time, and fuel consumption, while also adapting to the temporal variability of the Arctic environment.
Key Innovation: Develops a multi-objective dynamic route-planning framework for the Arctic Northeast Passage, employing an improved A* algorithm for initial route determination and the LPA* algorithm for real-time adaptation to daily updated environmental datasets, enabling adaptive optimization based on vessel type, departure date, and operational requirements.
110. Bioturbation and bio-geomorphic control of pedogenesis along a catena in the Atlantic Forest of southeastern Brazil
Core Problem: The largely unaddressed role of biota in soil formation and its influence on pedogenesis along catenas in the Atlantic Forest, especially concerning processes that affect erosion and soil stability.
Key Innovation: Demonstrated that organisms (plants, earthworms) play key roles in catena dynamics by enhancing infiltration, reducing erosion, and preserving organic matter, thereby significantly affecting pedogenetic processes and influencing soil evolution and conservation strategies in environments under high denudation pressure.
111. Simulation of brittle particle breakage using a spherical harmonic-based discrete element method
Core Problem: Conventional discrete element methods are limited in capturing irregular particle morphology and complex brittle breakage mechanisms, which significantly influence material strength and deformation in geotechnical engineering.
Key Innovation: Develops a 3D breakage model based on SH-DEM with a dual-center definition and a novel breakage identification method for irregular particles, validated against FEA and experimental data, demonstrating the strong influence of particle morphology on brittle breakage mechanisms.
112. Inner-radius-controlled tuning of surface-wave bandgaps in hexagonal locally resonant periodic pile barriers: experiments and complex band-structure modelling
Core Problem: Effectively isolating surface waves generated by rail and road traffic to control environmental vibrations, particularly through the design of periodic pile barriers, requires a better understanding of how structural parameters influence attenuation zones.
Key Innovation: Investigated Hexagonally Latticed Locally Resonant Periodic Pile Barriers (HLRPPBs) through experiments and complex band-structure modeling, demonstrating that decreasing the inner radius of pipe piles significantly improves surface wave attenuation (105–478 Hz range) and providing mechanistic insights into bandgap widening.
113. Strength prediction using binders’ water-absorption capacity and a saturation-based compaction quality control framework for treated clays
Core Problem: Conventional compaction control methods for treated clays in road embankments do not adequately capture water-binder interactions, especially for high-absorption binders, leading to inconsistencies in field compaction performance and strength prediction.
Key Innovation: Proposed a new strength prediction model integrating binder water absorption capacity and a saturation-based compaction quality control framework (using degree of saturation and air void volume) for treated clays, offering improved consistency and predictability for embankment construction.
114. Effects of structural plane on shearing behavior of anchored rocks under true triaxial disturbance: Morphology, roughness, and lithology
Core Problem: An incomplete understanding of the mechanical behaviors and failure mechanisms of anchored rock mass structural planes under true three-dimensional stress conditions, particularly concerning the effects of foliation morphology, roughness, and lithology, which are critical for deep underground engineering stability.
Key Innovation: Development and application of a new true triaxial dynamic–static combined shear testing system to systematically examine the effects of foliation morphology, initial roughness, and wall plate lithology on the disturbance shear strength, deformation, and fracture evolution of anchored structural planes in various rock types.
115. Determination of well stability and sand risk minimization parameters for gas condensate field conditions using geomechanical and CT-based approaches
Core Problem: Ensuring well stability and minimizing sand production risks in gas condensate fields, especially in weakly cemented formations, requires a comprehensive understanding of rock deformation, fracture, and filtration processes under operational stresses.
Key Innovation: Integrated geomechanical and CT-based approaches (mechanical properties, true triaxial modeling, sand production studies, digital CT-analysis) to determine relationships between permeability, deformation, and stresses, identify strength anisotropy, and provide optimal parameters for wellbore stability, sand control, and fracture localization, thereby reducing operational risks.
116. Quantifying Changes in Water Loading in the U.S. Southwest via Comparison of GNSS, GRACE, and SWE Data Sets
Core Problem: Unraveling hydrological partitioning and surface mass changes in the complex Colorado River Basin is difficult due to the limited number of GNSS stations and the resolution of GRACE data.
Key Innovation: Compared GNSS vertical displacement, GRACE surface mass change, and SWE data using elastic surface displacement modeling and signal localization, revealing region-dependent seasonal partitioning and timing differences in sensing terrestrial water storage changes.
117. GAC-KAN: An Ultra-Lightweight GNSS Interference Classifier for GenAI-Powered Consumer Edge Devices
Core Problem: GenAI applications on edge hardware impose immense computational burdens, leaving limited resources for fundamental security tasks like robust GNSS interference classification, compounded by data scarcity for training.
Key Innovation: Proposing GAC-KAN, an ultra-lightweight framework that uses physics-guided simulation for data synthesis and a Multi-Scale Ghost-ACB-Coordinate (MS-GAC) backbone with a Kolmogorov-Arnold Network (KAN) decision head to achieve high accuracy (98.0%) with significantly fewer parameters, ensuring GNSS reliability on GenAI-native edge chips.
118. Credal Concept Bottleneck Models: Structural Separation of Epistemic and Aleatoric Uncertainty
Core Problem: Most methods for decomposing predictive uncertainty into epistemic (model ignorance) and aleatoric (data ambiguity) components estimate both from the same predictive distribution, leading to strong correlations that blur their semantics and hinder reliable decision-making.
Key Innovation: A credal-set formulation instantiated in a Variational Credal Concept Bottleneck Model that structurally separates epistemic and aleatoric uncertainty through distinct geometric properties and disjoint training objectives, reducing their correlation by over an order of magnitude and improving alignment with prediction error and ground-truth ambiguity.
119. TimeSynth: A Framework for Uncovering Systematic Biases in Time Series Forecasting
Core Problem: Debates persist regarding whether complex nonlinear architectures truly outperform simple linear models in time series forecasting, often due to benchmarks lacking diverse temporal dynamics and employing biased evaluation protocols.
Key Innovation: Introduces TimeSynth, a structured framework that emulates key properties of real-world time series to uncover systematic biases, demonstrating that linear models collapse to simple oscillation regardless of signal complexity, while nonlinear models (especially Transformers and CNNs) gain clear advantages as signal complexity increases.
120. AltTS: A Dual-Path Framework with Alternating Optimization for Multivariate Time Series Forecasting
Core Problem: Multivariate time series forecasting faces an optimization conflict when a single model attempts to capture both stable within-series autoregressive dynamics and intermittent cross-dimension interactions, leading to brittle training and degraded long-horizon accuracy.
Key Innovation: Proposes ALTTS, a dual-path framework that explicitly decouples autoregression and cross-relation modeling, using a linear predictor for AR and a Transformer with Cross-Relation Self-Attention for CR, coordinated via alternating optimization to isolate gradient noise.
121. ArGEnT: Arbitrary Geometry-encoded Transformer for Operator Learning
Core Problem: Learning solution operators for physical systems with complex, varying geometries and parametric settings is challenging, requiring surrogate models that generalize across geometries and allow flexible evaluation at arbitrary spatial locations, especially in many-query regimes.
Key Innovation: Introduces Arbitrary Geometry-encoded Transformer (ArGEnT), a geometry-aware attention-based architecture that encodes geometric information directly from point-cloud representations, enabling improved prediction accuracy and generalization for operator learning on arbitrary domains in scientific machine learning applications like fluid dynamics and solid mechanics.
122. Potential-energy gating for robust state estimation in bistable stochastic systems
Core Problem: Robust state estimation in bistable stochastic systems is challenging, especially with observation noise and outliers, and existing filters treat all regions of state space identically.
Key Innovation: Introducing potential-energy gating, which modulates observation noise covariance in Bayesian filters based on a potential energy function, trusting observations near potential minima and discounting them near barriers, leading to significantly improved RMSE in state estimation.
123. GSO-SLAM: Bidirectionally Coupled Gaussian Splatting and Direct Visual Odometry
Core Problem: Existing SLAM methods either incur high computational costs by tightly coupling tracking and mapping or introduce redundancies with loose integration, hindering real-time dense reconstruction.
Key Innovation: Proposing GSO-SLAM, a real-time monocular dense SLAM system that bidirectionally couples Visual Odometry and Gaussian Splatting within an EM framework for joint optimization, achieving simultaneous refinement of depth estimates and scene representation, along with a novel Gaussian Splat Initialization.
124. Code2Worlds: Empowering Coding LLMs for 4D World Generation
Core Problem: Extending coding LLMs to 4D dynamic world generation faces challenges in balancing local object structures with global environmental layouts (multi-scale context entanglement) and ensuring physical fidelity (semantic-physical execution gap).
Key Innovation: Introduces Code2Worlds, a framework for language-to-simulation code generation for 4D worlds, featuring a dual-stream architecture for disentangled object/environment generation and a physics-aware closed-loop mechanism with a PostProcess Agent and VLM-Motion Critic for dynamic fidelity.
125. CAAL: Confidence-Aware Active Learning for Heteroscedastic Atmospheric Regression
Core Problem: In heteroscedastic regression settings with limited and costly labeling budgets (e.g., estimating atmospheric particle properties from routine observations), standard active learning strategies waste budget by conflating reducible epistemic uncertainty with irreducible aleatoric uncertainty.
Key Innovation: Proposes the Confidence-Aware Active Learning (CAAL) framework, which includes a decoupled uncertainty-aware training objective to stabilize uncertainty estimation and a confidence-aware acquisition function that dynamically weights epistemic uncertainty using predicted aleatoric uncertainty as a reliability signal. This enables efficient and robust sample selection in heteroscedastic settings.
126. WorldTree: Towards 4D Dynamic Worlds from Monocular Video using Tree-Chains
Core Problem: Existing dynamic reconstruction methods from monocular video lack a unified spatiotemporal decomposition framework, leading to inefficiencies in holistic temporal optimization or coupled hierarchical spatial composition, limiting their practical application for dynamic scene understanding.
Key Innovation: WorldTree, a unified framework comprising a Temporal Partition Tree (TPT) for coarse-to-fine hierarchical temporal decomposition and Spatial Ancestral Chains (SAC) for recursively querying ancestral hierarchical structure, providing complementary spatial dynamics and improving 4D dynamic world reconstruction from monocular video.
127. Temporally Unified Adversarial Perturbations for Time Series Forecasting
Core Problem: Existing adversarial attack methods for time series forecasting ignore temporal consistency, leading to impractical divergent perturbations for the same timestamp across overlapping samples.
Key Innovation: Temporally Unified Adversarial Perturbations (TUAPs) which enforce temporal unification, and a novel Timestamp-wise Gradient Accumulation Method (TGAM) to efficiently generate TUAPs, demonstrating superior attack performance and transferability for time series forecasting models.
128. UPDA: Unsupervised Progressive Domain Adaptation for No-Reference Point Cloud Quality Assessment
Core Problem: No-Reference Point Cloud Quality Assessment (NR-PCQA) models suffer significant performance degradation when a distribution gap exists between training (source) and testing (target) data, and limited attention has been paid to cross-domain transfer.
Key Innovation: Proposes UPDA, the first unsupervised progressive domain adaptation framework for NR-PCQA. It uses a two-stage coarse-to-fine alignment paradigm, including a discrepancy-aware coarse-grained alignment with a novel quality-discrepancy-aware hybrid loss, and a perception fusion fine-grained alignment with symmetric feature fusion and a conditional discriminator to identify and transfer domain-invariant, quality-relevant features.
129. A DMD-Based Adaptive Modulation Method for High Dynamic Range Imaging in High-Glare Environments
Core Problem: Achieving accurate photomechanics measurements and high-fidelity imaging in high-glare, extreme illumination environments where conventional CCD/CMOS sensors suffer from saturation and loss of detail, limiting applications like welding arc monitoring and polished metallic surface analysis.
Key Innovation: Presenting an HDR imaging system leveraging a digital micromirror device (DMD) for spatial modulation, enabling autonomous regional segmentation and adaptive exposure control to achieve a 127 dB dynamic range, effectively eliminating saturation and reducing strain error in DIC.
130. Oscillators Are All You Need: Irregular Time Series Modelling via Damped Harmonic Oscillators with Closed-Form Solutions
Core Problem: Existing Transformer models struggle with irregular time series due to assumptions of uniform intervals, while Neural Ordinary Differential Equations (NODEs), though effective, suffer from computational bottlenecks due to heavy numerical solvers.
Key Innovation: A novel approach that replaces NODEs with a linear damped harmonic oscillator analogy, providing a closed-form solution for irregular time series modeling. This eliminates computational overhead, preserves expressivity, and maintains universal approximation properties, achieving state-of-the-art performance and speed.
131. It's TIME: Towards the Next Generation of Time Series Forecasting Benchmarks
Core Problem: Existing time series forecasting benchmarks suffer from constrained data composition, compromised data integrity, misaligned task formulations, and rigid analysis perspectives, hindering the development and evaluation of generalizable Time Series Foundation Models (TSFMs).
Key Innovation: TIME, a next-generation task-centric benchmark comprising 50 fresh datasets and 98 forecasting tasks, designed for strict zero-shot TSFM evaluation. It features a human-in-the-loop construction pipeline for data integrity, real-world aligned task formulations, and a novel pattern-level evaluation perspective for generalizable insights.
132. Amortised and provably-robust simulation-based inference
Core Problem: Existing simulation-based inference methods struggle to account for outliers and extreme values in data, leading to unreliable results, and often incur high computational costs.
Key Innovation: Introduces a novel approach to simulation-based inference grounded in generalised Bayesian inference and a neural approximation of a weighted score-matching loss, resulting in a method that is amortised, provably robust to outliers, and computationally efficient without the need for Markov chain Monte Carlo sampling.
133. Sample-Free Safety Assessment of Neural Network Controllers via Taylor Methods
Core Problem: Neural network controllers, while effective in complex scenarios, lack the verification guarantees and trustworthiness required for adoption in safety-critical applications.
Key Innovation: Develops a sample-free method to assess the safety of trained neural network feedback controllers by embedding the network into system dynamics, approximating flow with high-order Taylor polynomials, and using automatic domain splitting and polynomial bounding to rigorously constrain and analyze closed-loop outcomes.
134. HyperDet: 3D Object Detection with Hyper 4D Radar Point Clouds
Core Problem: Radar-only 3D object detection trails LiDAR-based systems because radar point clouds are sparse, irregular, and often corrupted by multipath noise, yielding weak and unstable geometry.
Key Innovation: Presents HyperDet, a detector-agnostic framework that constructs a task-aware hyper 4D radar point cloud by aggregating returns, applying geometry-aware cross-sensor consensus validation, and integrating a foreground-focused diffusion module to densify object structures and lift radar attributes, narrowing the radar-LiDAR gap.
135. Safety Beyond the Training Data: Robust Out-of-Distribution MPC via Conformalized System Level Synthesis
Core Problem: Ensuring safety and robustness in planning and control when using learned dynamics models, especially when operating beyond the training data distribution (out-of-distribution scenarios).
Key Innovation: A novel framework combining weighted conformal prediction for high-confidence model error bounds with an SLS-based robust nonlinear MPC formulation, providing theoretical guarantees on coverage and robustness under distributional drift, and empirically demonstrating improved safety and robustness for nonlinear systems.
136. Learning to Control: The iUzawa-Net for Nonsmooth Optimal Control of Linear PDEs
Core Problem: There is a lack of real-time solvers for a class of nonsmooth optimal control problems of linear partial differential equations (PDEs), which are fundamental in many scientific and engineering domains.
Key Innovation: Proposes iUzawa-Net, an optimization-informed deep neural network that unrolls an inexact Uzawa method, replacing classical preconditioners and PDE solvers with learnable neural networks, proving universal approximation and asymptotic optimality, and demonstrating promising numerical efficiency for nonsmooth optimal control problems.
137. Learning a Neural Solver for Parametric PDE to Enhance Physics-Informed Methods
Core Problem: Physics-informed deep learning faces optimization challenges when solving PDEs due to large solution spaces, numerous iterations, unstable training, and ill-conditioning. Traditional methods also solve only single PDE instances.
Key Innovation: Proposes learning a neural solver that conditions a gradient descent algorithm to adapt to each PDE instance, significantly accelerating and stabilizing optimization. It also extends to parametric PDEs, allowing solving over a distribution of parameters, enhancing physics-informed methods.
138. Test-Time Efficient Pretrained Model Portfolios for Time Series Forecasting
Core Problem: Training a single, large monolithic model for time series forecasting may not always be the most compute-efficient or effective approach, especially when considering diverse forecasting tasks.
Key Innovation: Explores building portfolios of smaller, pretrained forecasting models, demonstrating that collections of specialist models (especially those created by post-training a base model) outperform generalists, and that ensembling/model selection are more compute-efficient than test-time fine-tuning.
139. Prominence-Aware Artifact Detection and Dataset for Image Super-Resolution
Core Problem: State-of-the-art generative single-image super-resolution (SISR) models produce visual artifacts that vary widely in perceptual impact, but existing detection methods treat them uniformly, failing to account for their prominence to human observers.
Key Innovation: Proposes characterizing artifacts by their prominence, introduces a novel dataset (1302 examples) with crowdsourced prominence scores for SISR artifacts, and trains a lightweight regressor to produce spatial prominence heatmaps, demonstrating superior performance over existing detectors and effectiveness in guiding SR model fine-tuning.
140. Feature-Based Interpretable Surrogates for Optimization
Core Problem: Existing inherently interpretable optimization models often rely on decision trees, limiting interpretability and decision-maker freedom, and potentially sacrificing solution quality.
Key Innovation: A methodology using mixed-integer programming and heuristics to find more general feature-based optimization rules that map to sets of solutions, increasing interpretability and decision-maker freedom while demonstrating improved solution quality compared to existing interpretable surrogates.
141. Storage effects on tube specimens of a highly plastic marine clay
Core Problem: The influence of storage time on the compressibility and undrained shear strength of highly plastic marine clay tube specimens, particularly concerning different sampling methods, was not well understood.
Key Innovation: The study found that for highly plastic soils, storage time has a minor influence on mechanical properties when using low-disturbance fixed-piston samplers, with mechanical disturbance from sampling being the primary factor affecting properties in Shelby samples. This allows for confident use of fixed-piston samples even after long storage.
142. Numerical analysis of post-cyclic ultimate bearing capacity of helical anchors in clay
Core Problem: Accurately assessing the post-cyclic loading ultimate bearing capacity of helical anchors in clay, which serve as foundations for floating offshore structures, remains challenging despite significant impacts from cyclic loading.
Key Innovation: Develops a 3D finite element model incorporating the E-R model to investigate the effects of cyclic loading patterns and soil parameters on ultimate bearing capacity, exploring mechanisms of soil strength degradation and failure, and developing a predictive expression for post-cyclic capacity.
143. Numerical investigation of the influence of vertical loading on the lateral response of disconnected piled raft foundation
Core Problem: The influence of vertical loading on the lateral response of Disconnected Piled Raft Foundations (DPRFs) in sand, particularly under combined vertical-horizontal (V-H) loading, is not fully understood.
Key Innovation: A three-dimensional numerical investigation using FLAC 3D analyzes the coupled V-H loading effects on DPRFs, revealing how vertical load affects lateral capacity and pile load sharing, and proposing new accurate predictive equations for DPRF capacity.
144. Comparative Analysis of Structure-From-Motion Underwater Point Cloud Processing Techniques for a Steep Mountain Creek
Core Problem: Accurate mapping of underwater topography in steep mountain creeks using autonomous aerial vehicles and Structure-From-Motion (SfM) photogrammetry is challenging due to steep slopes, rough substrates, and complex refraction effects.
Key Innovation: Investigated and compared Simple and Multicam bathymetric correction methods for SfM underwater point clouds, demonstrating that limiting the refraction angle in the Multicam method significantly improved accuracy (RMSE 0.15m for nadir configuration) for mapping steep mountain creek topography.
145. An Advanced Algorithm for Constructing Elevation Change Time Series With Satellite Altimetry Observations
Core Problem: Establishing accurate, long-term time series of elevation changes using satellite altimetry data, particularly for polar regions, is crucial for monitoring mass balance but is challenged by data utilization and uncertainty reduction.
Key Innovation: Introduces the indirect adjustment method (IAM), a novel algorithm for constructing elevation change time series. IAM maximizes data utilization by integrating a larger number of altimetric cycles, reduces uncertainty in elevation values (average 30% reduction in annual trend standard error), and yields significantly more effective cycles in the time series, improving spatiotemporal resolution and continuity even with partial missing data.
146. A benchmark dataset for half-hourly evapotranspiration estimation in China from 2000 to 2024
Core Problem: Existing ChinaFlux observations for evapotranspiration are limited by short observation periods and extensive data gaps, hindering long-term and multi-scale hydrological studies.
Key Innovation: Development of a continuous, half-hourly, ground-based benchmark dataset for latent heat flux across China (2000-2024) using an AutoML-H2O framework, integrating ERA5-Land and MODIS data for accurate gap-filling and temporal prolongation.
147. The countrywide historical gravity dataset of Lithuanian territory
Core Problem: Lack of a standardized, countrywide historical gravity dataset for Lithuania in modern reference systems, limiting its use for geodetic and geological applications.
Key Innovation: Compilation and standardization of a countrywide historical gravity dataset for Lithuania (1951-1962 measurements), converting raw data into modern gravity and coordinate reference systems, suitable for quasi-geoid modelling, Earth geopotential models, and geological interpretation.
148. HUST-Grace2026s: A high-resolution static gravity field product from GRACE and GRACE-FO observations (2002–2025)
Core Problem: Existing GRACE-only static gravity field models have limitations in spatial resolution and accuracy, and merely adding GRACE-FO observations offers limited improvement without refined processing strategies.
Key Innovation: Development of HUST-Grace2026s, a high-resolution static gravity field product (up to degree/order 180) from GRACE and GRACE-FO data, achieving over 50% accuracy improvement through a stochastic model based on postfit residuals and optimized rate term estimation strategies.
149. AGILE v0.1: The Open Global Glacier Data Assimilation Framework
Core Problem: Existing glacier models struggle to dynamically and consistently integrate heterogeneous observations, leading to uncertainties in glacier volume and area estimates, and making it difficult to reconstruct past glacier states.
Key Innovation: AGILE, an open global glacier data assimilation framework built on PyTorch, uses a time-dependent variational method and automatic differentiation to efficiently optimize multiple control variables, substantially improving glacier bed topography and distributed ice volume recovery.
150. More than mitigation: The role of forests in climate adaptation
Core Problem: Understanding and leveraging the full potential of forests beyond carbon mitigation for climate adaptation, specifically their effects on local temperature and hydrology.
Key Innovation: Synthesis of data and findings on forests' effects on temperature and hydrology, discussing how these effects vary depending on the environmental context, with implications for forest management and climate adaptation planning.
151. The effect of digitalization on disaster response in local government: Quasi-experimental evidence from smart city infrastructure in China
Core Problem: Local governments face a critical challenge in enhancing their disaster response capacity amidst increasing extreme natural disasters.
Key Innovation: Quasi-experimental evidence demonstrating that smart city infrastructure significantly shortens disaster response times by improving information transmission efficiency, optimizing resource allocation, and enhancing interdepartmental collaboration.
152. Don’t quit your day job. Part-time firefighters in rural Norway.
Core Problem: Understanding how part-time firefighters in rural areas, despite limited resources and training, effectively respond to diverse incidents, and the potential risks of losing these qualities during organizational restructuring.
Key Innovation: Analysis revealing how social networks, reciprocal relations, and the embeddedness of fire and rescue services within rural communities are key factors in mobilizing resources for emergency preparedness and contributing to community resilience.
153. Frequency-domain approach to automated and efficient multivariate kernel density estimation for probabilistic modeling
Core Problem: Accurate and efficient multivariate Kernel Density Estimation (KDE) for probabilistic modeling of large-scale, multidimensional datasets is challenged by high computational cost, suboptimal bandwidth selection, and density leakage.
Key Innovation: Proposes a frequency-domain approach that reformulates bandwidth selection as a gradient-based optimization task, using discrete cosine transform to decouple complexity from dataset size, and constructing a differentiable objective function with fidelity loss and regularization, leading to improved efficiency and accuracy over classical methods for one- to multi-dimensional data-driven density estimation.
154. Learning with less: label-efficient land cover classification at very high spatial resolution using self-supervised deep learning
Core Problem: The need for large volumes of manually annotated training data is a significant barrier to widespread adoption of deep learning for very high-resolution land cover mapping over large areas.
Key Innovation: A novel label-efficient approach for statewide 1-m land cover classification using only 1,000 annotated reference image patches by pre-training with self-supervised deep learning (BYOL) on a large amount of unlabeled data, achieving high accuracy and addressing the data annotation limitation.
155. Time-dependent mechanical response of buried water pipelines to cold waves under coupled multiphysics loading
Core Problem: The mechanical response and failure risk of buried water pipelines under time-varying multiphysics coupling, especially during extreme cold waves, are poorly understood.
Key Innovation: Development of a validated 3D FE model to simulate transient pipeline response under coupled loads (thermal, earth pressure, traffic, groundwater, soil-pipe interaction), revealing non-linear field interactions and identifying critical factors (soil elastic modulus, permeability) for enhancing pipeline resilience.
156. Surface explosion induced damage effects and optimization of blast resistance performance of buried pipelines: a case study of concrete pipelines with bell-and-spigot joints
Core Problem: The dynamic response, crack propagation, and blast resistance of shallow-buried reinforced concrete pipelines with bell-and-spigot joints against accidental surface explosions are not adequately understood, leading to vulnerabilities in urban infrastructure.
Key Innovation: An integrated approach combining field explosion tests and fully coupled numerical simulations (LS-DYNA) to analyze blast-induced damage in RC pipelines, identifying bell-and-spigot joints as vulnerable points, and providing validated reinforcement design strategies (e.g., increased rebar diameter and density) to enhance blast resilience.
157. Influence of free surface on microwave-induced rock fracturing: An experimental and numerical study
Core Problem: Insufficient understanding of how free surfaces influence crack initiation and propagation in rocks under microwave irradiation, relevant for enhancing rock breakage efficiency in mining and tunneling.
Key Innovation: Identified an optimal range of free surface distance (5–10 cm) for maximum crack propagation in microwave-induced rock fracturing, and elucidated the role of stress concentration and material properties in directional crack propagation.
158. A high-accuracy spatiotemporal numerical manifold method with thermal DOF inheritance and its application in rock thermal fracturing analysis
Core Problem: Accurately and efficiently modeling rock thermal fracturing problems involving large spatial scales, long time spans, and highly time-dependent loading conditions.
Key Innovation: Proposed a high-accuracy spatiotemporal numerical manifold method with a thermal DOF inheritance strategy to precisely compute transient temperature fields and model crack propagation under transient heat transfer conditions, balancing accuracy and efficiency.
159. Hydro-mechanical responses of irregular twin tunnels with unequal burial depths in anisotropic soil layer
Core Problem: Existing studies on underground structures often neglect the complex hydro-mechanical behavior of irregular multi-tunnel systems with unequal burial depths and noncircular sections in anisotropic soils, hindering accurate safety and durability assessments.
Key Innovation: Developed a novel hybrid Physics-Informed Neural Network (PINN)-Finite Element Method (FEM) framework to efficiently and reliably analyze steady-state seepage and stress fields for complex irregular twin tunnels in anisotropic soil, revealing that tunnel number and burial-depth asymmetry significantly dominate hydro-mechanical responses.
160. Experimental evaluation on bearing performance of tire cell-geogrid reinforced subgrades
Core Problem: There is a need for experimental quantification of the bearing performance and deformation behavior of subgrades reinforced with sustainable materials like tire cells and geogrids to optimize their use in transportation engineering.
Key Innovation: Experimentally evaluated the synergistic interaction of tire cells and geogrids in reinforced subgrades, demonstrating significant reduction in surface settlement, improved stress distribution, and higher load-bearing stiffness, highlighting its potential for sustainable subgrade stabilization.
161. Horizontal vibration of offshore wind turbines supported by monopile-friction wheel composite foundation in multilayered saturated soil: Theoretical approach
Core Problem: Accurately predicting and mitigating horizontal vibration in offshore wind turbine foundations supported by monopile-friction wheel composite foundations in multilayered saturated soil is crucial for ensuring dynamic stability and optimal design.
Key Innovation: Developed a comprehensive theoretical model combining 3D continuum mechanics, Novak's plane strain model, Biot's porous media theory, and radiation wave theory to accurately investigate horizontal vibration, providing insights into soil property influences and guidelines for improving dynamic stability in offshore wind turbine foundation design.
162. Airblast waves and noises induced by tunnel delay blasting: Field measurement and tempo-spatial analysis
Core Problem: Residents near drill-and-blast tunnels experience disturbances from blasting-induced airblast waves and noise, necessitating a comprehensive understanding of their tempo-spatial characteristics and a reliable predictive model for mitigation.
Key Innovation: Conducted field measurements and tempo-spatial analysis of airblast waves and noise from tunnel blasting, revealing complex propagation patterns (e.g., wave superposition at far-field distances) and proposing a validated predictive model for external noise levels that accounts for anisotropic propagation, aiding in disturbance management.
163. A Physically Consistent Particle Size Distribution Modeling of the Microphysics of Precipitation for Weather and Climate Models
Core Problem: Current models for the probability density function of precipitation drops make problematic assumptions (e.g., allowing negative values for mean distribution) that hinder physically consistent modeling for weather and climate models.
Key Innovation: Developed a new particle size distribution model that satisfies mathematical and physical consistency, seamlessly integrates into microphysics parameterizations, and outperforms existing models, offering substantial practical and theoretical advantages for weather and climate modeling.
164. Automated Optimization Modeling via a Localizable Error-Driven Perspective
Core Problem: Existing automated optimization modeling approaches using Large Language Models (LLMs) suffer from suboptimal performance due to the scarcity of error-specific problems and sparse rewards associated with difficult problems in post-training.
Key Innovation: Proposing MIND (automated optimization modeling via a localizable error-driven perspective), an error-driven learning framework that customizes model training from data synthesis to post-training, leveraging localizable error patterns and Dynamic Supervised Fine-Tuning Policy Optimization (DFPO) for localized refinement.
165. KBVQ-MoE: KLT-guided SVD with Bias-Corrected Vector Quantization for MoE Large Language Models
Core Problem: Directly applying Vector Quantization (VQ) to Mixture of Experts (MoE) Large Language Models (LLMs) leads to substantial performance degradation due to redundant representations among experts and amplified cumulative output bias, hindering deployment in resource-constrained environments.
Key Innovation: Proposing KBVQ-MoE, a novel VQ framework that integrates input-driven redundancy elimination (KLT-guided SVD) and bias-corrected output stabilization (channel-wise affine compensation) to enhance ultra-low-bit quantization for MoE-based LLMs, preserving accuracy substantially better than existing methods.
166. Spectra: Rethinking Optimizers for LLMs Under Spectral Anisotropy
Core Problem: Gradient signals in Large Language Model (LLM) training are highly anisotropic, with dominant spectral directions suppressing learning in the long tail, leading to slower training and suboptimal performance.
Key Innovation: Proposing Spectra, a spike-aware optimizer that suppresses the dominant low-rank spike subspace without amplifying noise-sensitive spectral tail, using cached power iteration and low-rank spectral shaping, resulting in faster training, reduced memory, and improved accuracy for LLMs.
167. Time-TK: A Multi-Offset Temporal Interaction Framework Combining Transformer and Kolmogorov-Arnold Networks for Time Series Forecasting
Core Problem: Existing time series forecasting methods using independent token embedding destroy crucial multi-offset temporal correlations in long sequences, leading to an information bottleneck and performance degradation.
Key Innovation: Introduces a Multi-Offset Time Embedding (MOTE) method to mitigate performance degradation from standard token embedding, and proposes Time-TK, a novel forecasting architecture combining Multi-Offset Interactive KAN and an efficient Multi-Offset Temporal Interaction mechanism for state-of-the-art accuracy.
168. DD-MDN: Human Trajectory Forecasting with Diffusion-Based Dual Mixture Density Networks and Uncertainty Self-Calibration
Core Problem: Human Trajectory Forecasting (HTF) models often lack robust uncertainty modeling, calibration, and reliable forecasts from short observation periods, which are crucial for downstream tasks like path planning and collision avoidance.
Key Innovation: Proposes DD-MDN, an end-to-end probabilistic HTF model that combines a few-shot denoising diffusion backbone and a dual mixture density network to achieve high positional accuracy, calibrated uncertainty, and robustness to short observations by learning self-calibrated residence areas and probability-ranked anchor paths.
169. Structured Hybrid Mechanistic Models for Robust Estimation of Time-Dependent Intervention Outcomes
Core Problem: Purely data-driven models for estimating intervention effects in dynamical systems often fail in out-of-distribution (OOD) regimes, while purely mechanistic models can be oversimplified.
Key Innovation: A structured hybrid mechanistic-data-driven approach that decomposes the dynamical system's transition operator into parametric and nonparametric components, leveraging mechanistic anchors while learning residual patterns from data, demonstrating superior robustness, especially OOD, compared to purely data-driven or mechanistic models.
170. Arbitrary Ratio Feature Compression via Next Token Prediction
Core Problem: Existing feature compression methods lack flexibility, requiring dedicated models for specific compression ratios and retraining for new ratios, limiting generalization.
Key Innovation: Proposed Arbitrary Ratio Feature Compression (ARFC) framework with Arbitrary Ratio Compressor (ARC), an auto-regressive model using next-token prediction for flexible compression. It includes a Mixture of Solutions (MoS) module for refining compressed tokens and an Entity Relation Graph Constraint (ERGC) for preserving semantic/structural relationships, outperforming existing methods and even uncompressed features in some cases.
171. TS-Memory: Plug-and-Play Memory for Time Series Foundation Models
Core Problem: Adapting Time Series Foundation Models (TSFMs) to downstream domains under distribution shift remains challenging, with existing solutions facing a trade-off between catastrophic forgetting/costly maintenance (Parametric Adaptation) and high inference latency (Non-Parametric Retrieval).
Key Innovation: Proposes TS-Memory, a lightweight memory adapter that augments frozen TSFMs using Parametric Memory Distillation. It trains an offline kNN teacher to synthesize confidence-aware quantile targets from retrieved futures, then distills this into the adapter via confidence-gated supervision, enabling efficient, retrieval-free deployment with constant-time overhead.
172. Perception-based Image Denoising via Generative Compression
Core Problem: Distortion-driven image denoising methods often produce over-smoothed reconstructions, especially under strong noise and distribution shift, failing to preserve structural details and perceptual realism.
Key Innovation: Introduces a generative compression framework for perception-based denoising. Restoration is achieved by reconstructing from entropy-coded latent representations that enforce low-complexity structure, while generative decoders recover realistic textures via perceptual measures (LPIPS, Wasserstein distance). Two instantiations are proposed: a conditional WGAN-based denoiser and a conditional diffusion-based reconstruction strategy.
173. Move What Matters: Parameter-Efficient Domain Adaptation via Optimal Transport Flow for Collaborative Perception
Core Problem: Fast domain adaptation for deploying multi-agent systems across diverse environments in Vehicle-to-Everything (V2X) collaborative perception is challenging, as direct application of PEFT leads to significant performance degradation and training instability due to inter-frame redundancy and erosion of fine-grained semantics.
Key Innovation: Proposes FlowAdapt, a parameter-efficient framework grounded in optimal transport theory. It minimizes information transport costs across data distributions and network hierarchies using a Wasserstein Greedy Sampling strategy to filter redundant samples and a Progressive Knowledge Transfer module to inject compressed early-stage representations into later stages, alleviating semantic degradation.
174. GP2F: Cross-Domain Graph Prompting with Adaptive Fusion of Pre-trained Graph Neural Networks
Core Problem: The effectiveness of Graph Prompt Learning (GPL) in cross-domain scenarios, where pre-training source and downstream target data distributions differ substantially, is not well understood, and existing GPL methods often do not significantly outperform simpler baselines.
Key Innovation: Provides a theoretical analysis demonstrating the benefits of integrating pre-trained knowledge and task-specific adaptation in cross-domain GPL, and proposes GP2F, a dual-branch GPL method that explicitly fuses a frozen branch (pre-trained knowledge) and an adapted branch (task-specific adaptation) with adaptive fusion under topology constraints, outperforming existing methods.
175. U-DAVI: Uncertainty-Aware Diffusion-Prior-Based Amortized Variational Inference for Image Reconstruction
Core Problem: Existing diffusion-based image reconstruction methods are computationally intensive and struggle with fine details and complex textures.
Key Innovation: Extending amortized variational inference by injecting spatially adaptive perturbations guided by uncertainty estimates during training, leading to more realistic and computationally efficient reconstructions.
176. STVG-R1: Incentivizing Instance-Level Reasoning and Grounding in Videos via Reinforcement Learning
Core Problem: Misalignment between textual descriptions and visual coordinates in Vision-Language Models (VLMs) leads to hallucinations, especially in dense prediction tasks like spatial-temporal video grounding (STVG), with existing solutions incurring high costs.
Key Innovation: Proposing a novel visual prompting paradigm that reformulates per-frame coordinate prediction as instance-level identification using temporally consistent IDs embedded as visual prompts, and introducing STVG-R1, the first reinforcement learning framework for STVG, to jointly optimize temporal accuracy, spatial consistency, and structural format regularization.
177. U-Former ODE: Fast Probabilistic Forecasting of Irregular Time Series
Core Problem: Existing Neural Controlled Differential Equation (Neural CDE) approaches for probabilistic forecasting of irregularly sampled time series are slow, inherently sequential, and restrict access to global context.
Key Innovation: Introduces UFO (U-Former ODE), a novel architecture integrating U-Nets, Transformers, and Neural CDEs to achieve faster (up to 15x) and more accurate probabilistic forecasting of irregular time series, providing a global receptive field with strong sensitivity to local temporal dynamics.
178. Latent-Variable Learning of SPDEs via Wiener Chaos
Core Problem: Existing deep learning approaches for learning stochastic partial differential equations (SPDEs) from spatiotemporal observations either require explicit access to driving noise/initial conditions or rely on deterministic models that fail to capture intrinsic stochasticity.
Key Innovation: Proposes a structured latent-variable formulation combining spectral Galerkin projection with a truncated Wiener chaos expansion to separate deterministic evolution and stochastic forcing. This allows joint inference of latent dynamics and stochastic forcing via variational learning, recovering stochastic structure without explicit noise observation during training.
179. DiffPlace: Street View Generation via Place-Controllable Diffusion Model Enhancing Place Recognition
Core Problem: Existing multi-view diffusion models struggle to generate place-aware and background-consistent urban street scenes from text, BEV maps, and object bounding boxes, limiting their effectiveness in producing realistic samples for place recognition tasks.
Key Innovation: DiffPlace, a novel framework that introduces a place-ID controller into a multi-view diffusion model, enabling place-controllable image generation that synthesizes images with consistent background buildings while flexibly modifying foreground objects and weather conditions, thereby enhancing place recognition.
180. Universal Diffusion-Based Probabilistic Downscaling
Core Problem: Enhancing spatial resolution and uncertainty representation in operational weather forecasting from deterministic low-resolution inputs without model-specific fine-tuning.
Key Innovation: A universal diffusion-based downscaling framework that lifts deterministic low-resolution weather forecasts into probabilistic high-resolution predictions, demonstrating improved ensemble mean and probabilistic skill across diverse weather models.
181. Empirical Gaussian Processes
Core Problem: Traditional Gaussian Processes are limited by handcrafted kernel functions, requiring expert knowledge and lacking adaptivity to diverse data, imposing strong assumptions on the hypothesis space.
Key Innovation: Introduces Empirical GPs, a principled framework that empirically estimates mean and covariance functions from historical observations, enabling flexible, data-driven priors that reflect rich, non-trivial covariance structures. It converges to the KL-divergence closest GP and uses an EM algorithm for learning from independent datasets.
182. Geometry of Uncertainty: Learning Metric Spaces for Multimodal State Estimation in RL
Core Problem: Estimating the state of an environment from high-dimensional, multimodal, and noisy observations is a fundamental challenge in reinforcement learning (RL), with traditional probabilistic models often requiring explicit noise assumptions that limit generalization.
Key Innovation: Contributes a novel method to learn a structured latent representation where distances between states directly correlate with the minimum number of actions required to transition between them, providing a geometric interpretation of uncertainty without explicit probabilistic modeling. It uses a multimodal latent transition model and inverse distance weighting for adaptive sensor fusion.
183. WaveFormer: Wavelet Embedding Transformer for Biomedical Signals
Core Problem: Standard transformer architectures struggle to capture long sequences, complex temporal dynamics, and multi-scale frequency patterns inherent in biomedical signals.
Key Innovation: WaveFormer integrates wavelet decomposition into transformer architecture at two stages: multi-channel Discrete Wavelet Transform (DWT) for embedding construction to extract frequency features, and Dynamic Wavelet Positional Encoding (DyWPE) for adapting position embeddings to signal-specific temporal structure, achieving competitive frequency-aware processing.
184. Explaining AI Without Code: A User Study on Explainable AI
Core Problem: Most Explainable AI (XAI) methods require technical expertise, limiting their value for novices and creating a gap in no-code ML platforms that aim to democratize AI.
Key Innovation: A human-centered XAI module integrated into DashAI, an open-source no-code ML platform, which combines Partial Dependence Plots (PDP), Permutation Feature Importance (PFI), and KernelSHAP. A user study demonstrated high task success and improved perceived predictability and confidence for novices.
185. Mitigating Error Accumulation in Continuous Navigation via Memory-Augmented Kalman Filtering
Core Problem: Existing Vision-Language Navigation (VLN) models for UAVs suffer from accumulated position errors ('state drift') over time due to their dead-reckoning approach, compromising full trajectory prediction.
Key Innovation: Proposes NeuroKalman, a novel framework that uses a memory-augmented Kalman filter approach, decoupling navigation into Prior Prediction and Likelihood Correction. It associates Kernel Density Estimation with attention-based retrieval to rectify latent representations using retrieved historical anchors, significantly reducing drift accumulation with minimal fine-tuning.
186. Adaptive Power Iteration Method for Differentially Private PCA
Core Problem: Developing differentially private algorithms for computing the top singular vector (PCA) that offer improved utility beyond worst-case guarantees, especially for matrices with low coherence.
Key Innovation: Designs an adaptive power iteration method for differentially private PCA based on a filtering technique that adjusts to the input matrix's coherence parameter, providing better utility for low-coherence matrices compared to existing methods.
187. Multimodal Fact-Level Attribution for Verifiable Reasoning
Core Problem: Existing multimodal grounding benchmarks fail to assess fact-level attribution in complex multimodal reasoning scenarios, where Multimodal Large Language Models (MLLMs) need to ground outputs in heterogeneous sources and provide verifiable citations.
Key Innovation: Introduces MuRGAt, a benchmark for evaluating fact-level multimodal attribution in complex reasoning settings, requiring MLLMs to generate answers with explicit reasoning and precise citations (modality and temporal segments) from diverse inputs, and reveals a trade-off between reasoning depth/structured grounding and accuracy.
188. Estimation of instrument and noise parameters for inverse problem based on prior diffusion model
Core Problem: Estimating observation parameters (response and error) in inverse problems, especially when regularization is introduced in a Bayesian framework with a diffusion prior, where posterior sampling is challenging.
Key Innovation: A strategy that enables optimal estimation for both observation parameters and the image of interest, provides uncertainty quantification, and uses MCMC algorithms for efficient computation, building on a recent effective solution for posterior sampling.
189. Aggregate Models, Not Explanations: Improving Feature Importance Estimation
Core Problem: Expressive machine learning models suffer from unstable and inaccurate variable importance estimates due to data sampling and algorithmic stochasticity, and the optimal strategy for ensembling (model vs. explanation aggregation) is unclear.
Key Innovation: Theoretical analysis and empirical validation demonstrating that ensembling at the model level provides more accurate variable-importance estimates, particularly for expressive models, by reducing the leading error term (excess risk), contrary to prior literature.
190. A Comparative Study of MAP and LMMSE Estimators for Blind Inverse Problems
Core Problem: Maximum-a-posteriori (MAP) approaches for blind inverse problems are unstable and require extensive parameter tuning due to non-convexity and non-identifiability, while (Linear) minimum mean square error (MMSE) estimators offer a compelling alternative.
Key Innovation: A comparative study demonstrating that LMMSE estimators provide a robust and reliable baseline for blind inverse problems, even in highly controlled settings, and can serve as an effective initialization for MAP approaches to improve their performance and reduce sensitivity.
191. Data-Driven Trajectory Imputation for Vessel Mobility Analysis
Core Problem: Large gaps in vessel trajectories from AIS data significantly degrade data quality, leading to inaccurate or incomplete analysis, and existing imputation methods for vehicles do not adequately account for unique vessel motion patterns.
Key Innovation: HABIT, a lightweight, configurable H3 Aggregation-Based Imputation framework, imputes missing vessel trajectory segments by extracting, analyzing, and indexing motion patterns from historical AIS data, performing comparably to baselines in accuracy while offering better latency and accounting for vessel characteristics.
192. Think like a Scientist: Physics-guided LLM Agent for Equation Discovery
Core Problem: Existing LLM-based systems for symbolic equation discovery often guess equations directly from data, lacking the multi-step scientific reasoning process of inferring physical properties (like symmetries) as priors to restrict the search space.
Key Innovation: Introduces KeplerAgent, an agentic framework that explicitly follows a scientific reasoning process by coordinating physics-based tools to extract intermediate structure and using these results to configure symbolic regression engines, achieving substantially higher symbolic accuracy and robustness to noisy data.
193. Learning A Physical-aware Diffusion Model Based on Transformer for Underwater Image Enhancement
Core Problem: Underwater images suffer from complex degradations, limiting the efficiency of underwater vision tasks, and existing diffusion models for UIE fail to incorporate physical properties.
Key Innovation: Introduces PA-Diff, a physics-aware diffusion model that uses a Physics Prior Generation branch, Implicit Neural Reconstruction branch, and Physics-aware Diffusion Transformer branch to guide the diffusion process with physical knowledge, achieving SOTA UIE performance.
194. Scale Contrastive Learning with Selective Attentions for Blind Image Quality Assessment
Core Problem: Existing multi-scale blind image quality assessment (BIQA) algorithms fail to effectively replicate human visual perception across scales, leading to misleading feature fusion and diluted quality-critical features.
Key Innovation: Proposes CSFIQA, a framework that uses a selective focus attention mechanism to filter redundant cross-scale information and a scale contrastive learning strategy to distinguish quality variations across and within scales, significantly improving BIQA performance.
195. Efficient and Sharp Off-Policy Learning under Unobserved Confounding
Core Problem: Standard off-policy learning assumes unconfoundedness, leading to biased estimates and potentially harmful policies when unobserved factors influence both treatment assignment and outcomes.
Key Innovation: Develops a semi-parametrically efficient estimator for a sharp bound on the value function under unobserved confounding, which avoids unstable minimax optimization and leads to optimal confounding-robust policies.
196. A Multi-Fidelity Control Variate Approach for Policy Gradient Estimation
Core Problem: Many reinforcement learning (RL) algorithms are data-intensive, making them impractical for training in operational systems or computationally expensive high-fidelity simulations.
Key Innovation: Proposes Multi-Fidelity Policy Gradients (MFPGs), a sample-efficient RL framework that mixes scarce target-environment data with a control variate from abundant low-fidelity simulation data to construct an unbiased, variance-reduced estimator for on-policy policy gradients.
197. GraphPFN: A Prior-Data Fitted Graph Foundation Model
Core Problem: Graph foundation models face challenges in transferability and data scarcity, and existing PFN-based approaches for graphs are limited to hand-crafted features or lack graph-specific pretraining.
Key Innovation: Proposes GraphPFN, a PFN-based model designed and pretrained for graph node-level tasks using a novel prior distribution of synthetic attributed graphs, achieving strong in-context learning and state-of-the-art results.
198. DRIFT-Net: A Spectral--Coupled Neural Operator for PDEs Learning
Core Problem: Existing PDE foundation models using multi-scale windowed self-attention suffer from weak global coupling and error accumulation/drift during long rollouts due to their locality.
Key Innovation: Proposes DRIFT-Net, a dual-branch neural operator with spectral and image branches that captures global low-frequency information and local details, respectively, and fuses them via bandwise weighting, achieving lower error, higher throughput, and fewer parameters for PDE learning.
199. OpenTSLM: Time-Series Language Models for Reasoning over Multivariate Medical Text- and Time-Series Data
Core Problem: Large Language Models (LLMs) are limited in their ability to natively handle time series data, which is crucial for synthesizing information in domains like medicine.
Key Innovation: Introduces OpenTSLM, a family of Time Series Language Models that integrate time series as a native modality into pretrained LLMs (via SoftPrompt or Flamingo architectures), enabling reasoning over multivariate medical text- and time-series data and outperforming baselines.
200. Hilbert-Guided Sparse Local Attention
Core Problem: The quadratic compute and memory costs of global self-attention severely limit its use in high-resolution images, and conventional local attention patterns often fail to deliver significant speedups with block-sparse kernels due to non-contiguous token arrangements.
Key Innovation: Proposes a novel method for constructing attention windows and neighborhoods based on reordering image tokens along a Hilbert curve, which significantly increases block sparsity and improves the efficiency of 2D local attention (e.g., 4x for window attention, 18x for slide attention) with minimal accuracy loss.
201. Self-Adaptive Graph Mixture of Models
Core Problem: Graph Neural Network (GNN) performance gains are plateauing, and selecting the most suitable GNN model for a given graph task or dataset remains a significant challenge, as existing mixture-of-experts approaches often rely on variations of a single base model.
Key Innovation: Proposes Self-Adaptive Graph Mixture of Models (SAGMM), a modular framework that learns to automatically select and combine diverse GNN architectures using a topology-aware attention gating mechanism, outperforming or matching leading GNN baselines across 16 benchmark datasets for various graph tasks.
202. Minimum distance classification for nonlinear dynamical systems
Core Problem: Classifying trajectory data generated by distinct nonlinear dynamical systems, which is challenging due to the complexity of nonlinear dynamics.
Key Innovation: Dynafit, a kernel-based method that learns a distance metric between training trajectories and the underlying dynamics by approximating the Koopman operator, enabling classification of new observations based on the similarity of their dynamics in a feature space.
203. Landscaper: Understanding Loss Landscapes Through Multi-Dimensional Topological Analysis
Core Problem: Traditional low-dimensional analyses of loss landscapes often miss complex topological features crucial for understanding neural network optimization and generalization.
Key Innovation: Presents Landscaper, an open-source Python package for arbitrary-dimensional loss landscape analysis, combining Hessian-based subspace construction with topological data analysis and introducing Saddle-Minimum Average Distance (SMAD) to quantify landscape smoothness, revealing geometric structures and capturing training transitions missed by conventional metrics, with potential for out-of-distribution generalization in scientific ML.
204. Reducing Estimation Uncertainty Using Normalizing Flows and Stratification
Core Problem: Current methodologies for estimating the expectation of a function from sample data often assume (semi-)parametric distributions, leading to significant estimation uncertainty if these assumptions do not hold for unknown data distributions.
Key Innovation: A flow-based model integrated with stratified sampling leverages a parametrized neural network to offer greater flexibility in modeling unknown data distributions, significantly reducing estimation uncertainty across multiple datasets compared to crude Monte Carlo and Gaussian mixture models.
205. Optimal Cross-Validation for Sparse Linear Regression
Core Problem: K-fold cross-validation for hyperparameter selection in sparse linear regression is computationally expensive, requiring solving many mixed-integer optimization problems (MIOs).
Key Innovation: Derivation of computationally tractable relaxations of the k-fold cross-validation loss, which facilitates hyperparameter selection while solving 50-80% fewer MIOs, making exact MIO-based cross-validation competitive with mature software packages.
206. Designing lensless imaging systems to maximize information capture
Core Problem: Optimizing mask-based lensless imaging systems to maximize information capture, considering the object-dependent nature of imaging and the interdependence between object sparsity, encoder multiplexing, and noise.
Key Innovation: A mutual information-based approach to design information-optimal lensless encoders that tailor multiplexing to object sparsity, demonstrating improved downstream reconstruction performance without requiring joint optimization with a specific reconstruction algorithm.
207. Prediction of impact damage to Fe-based amorphous coatings via interpretable machine learning
Core Problem: Predicting impact damage to Fe-based amorphous coatings is challenging due to the complex, nonlinear effects of their properties.
Key Innovation: This study develops an interpretable machine learning model to predict the impact damage of Fe-based amorphous coatings.
208. Dynamic behavior of deep-sea floor drill rigs under the rheological contact force
Core Problem: Traditional Kelvin models fail to accurately describe the rheological behaviors of sediments during the landing impact process of deep-sea floor drill rigs, posing a challenge for dynamic calculation.
Key Innovation: Proposes an improved contact force model that innovatively incorporates the Zener model to characterize rheological phenomena of sediments, establishes a dynamic model of the drill rig, and determines the safe landing parameter range, providing a theoretical reference for structural design and safety validation.
209. Meta-Learning With Unlabeled Query Updating and Consistency Learning for Few-Shot OCT Image Classification
Core Problem: Achieving automatic diagnosis for rare diseases using deep neural networks when there is insufficient training data (few-shot learning problem) in fields like optical coherence tomography (OCT).
Key Innovation: A novel meta-learning algorithm for few-shot OCT image classification that integrates unsupervised learning with unlabeled query data updating, cross-set consistency learning to reduce the gap between meta-knowledge, and data mixup, significantly improving generalization to unseen tasks and rare disease diagnosis.
210. PASG-Net:Spatial-Guided Frequency Compensation Polarized Attention Fusion Network for Hyperspectral and Multispectral Image Fusion
Core Problem: Existing hyperspectral and multispectral image fusion methods often struggle to effectively extract spatial-frequency features and achieve complete and efficient integration of complementary information, leading to fused images that fail to maintain spatial-spectral consistency.
Key Innovation: Proposes PASG-Net, a spatial-guided frequency compensation polarized attention fusion network, which integrates a grouped spatial feature extraction module, a spatial-guided frequency compensation module (based on 'phase similarity and magnitude complementarity'), and a symmetric polarized cross-attention module. This design effectively integrates complementary information from both domains while maintaining low computational complexity, outperforming current state-of-the-art methods in HR-HSI reconstruction.
211. MD2F-Mamba: Multidirectional Depthwise Convolution and Dual-Branch Mamba Feature Fusion Networks for Hyperspectral Image Classification
Core Problem: Existing hyperspectral image (HSI) classification methods struggle to effectively model both intricate local variations and long-range spectral-spatial dependencies, often neglecting directional information or employing simplistic fusion, leading to inadequate feature representations and reduced discriminative ability while maintaining computational efficiency.
Key Innovation: MD2F-Mamba, a novel dual-branch architecture, integrates a multidirectional depthwise convolution module to capture spatial features from multiple orientations and a hierarchical state-space Mamba for global feature extraction. A cosine similarity feature fusion module adaptively merges local and global features, achieving competitive HSI classification results with a minimal parameter count.
212. Reconstructing two-decadal global daily high-resolution XCO2 records using a hybrid Transformer–BiLSTM model
Core Problem: Satellite-based XCO2 observations are often spatially incomplete and temporally discontinuous, and existing products suffer from coarse spatial resolutions, hindering accurate and continuous global monitoring of atmospheric carbon dioxide and the detection of fine-scale emission changes.
Key Innovation: This study develops GlobalHighXCO2, a novel spatiotemporal Transformer-BiLSTM deep-learning network, to reconstruct global, daily, and seamless XCO2 at 0.1° resolution from 2003 to 2022. It achieves excellent agreement with TCCON measurements and reliably resolves XCO2 variability across scales, capturing climate-driven signals and short-lived enhancements from major wildfire events.
213. Implementation of a multi-layer snow scheme in the GloSea6 seasonal forecast system: impacts on land–atmosphere interactions and climatological biases
Core Problem: Traditional single-layer snow schemes in land surface models inadequately represent the insulating effect of snowpack, leading to cold and warm biases and poor simulation of land-atmosphere interactions in seasonal forecast systems.
Key Innovation: Implementing a multi-layer snow scheme in the GloSea6 seasonal forecast system improves the simulation of Northern Hemisphere snow seasonality, delays snowmelt, enhances soil moisture memory, and mitigates near-surface warming biases, leading to improved temperature and precipitation forecasts.
214. River intermittency: mapping and upscaling of water occurrence using unmanned aerial vehicle, Random Forest and remote sensing landscape attributes
Core Problem: Measurements and modeling of non-perennial rivers are scarce, making it challenging to predict and understand water scarcity scenarios and the spatio-temporal dynamics of river intermittency.
Key Innovation: Develops a scalable modeling framework integrating UAV surveys, Random Forest, and remote sensing landscape attributes (e.g., Sentinel MNDWI) to successfully map and upscale the spatio-temporal dynamics of river intermittency with high accuracy, providing a broader understanding of this complex hydrological process.
215. Self-powered vibration sensor for wearable health care and voice detection
Core Problem: Developing compact, high-fidelity, and self-powered vibration sensors for wearable applications.
Key Innovation: A high-fidelity, self-powered vibration sensor constructed from densely packed arrays of capacitors, suitable for wearable health care and voice detection.
216. Diffusiophoretic transport of colloids in porous media
Core Problem: Diffusiophoretic migration of colloids driven by chemical gradients in porous media flows has largely been ignored, despite chemical gradients being ubiquitous and potentially influencing colloid transport beyond stagnant pockets.
Key Innovation: Uncovered that even moderate solute gradients can markedly alter colloid transport in porous media by inducing cross-streamline phoretic migration within preferential flow pathways, which changes macroscopic dispersion by orders of magnitude and suppresses the impact of geometric disorder, challenging classical colloid transport models.
217. Gas vortex discovery in butterfly microcavities for constructing ultrasensitive gas sensors
Core Problem: Conventional gas sensors face a sensitivity-stability trade-off in trace-gas detection, often relying on reactive surface modifications that risk stability or inefficient gas-solid interaction time.
Key Innovation: Discovered gas vortex effects in butterfly wings that prolong molecular residence time and applied this bioinspired mechanism to gas sensor design. Established a universal design rule for periodic microcavities to generate centralized vortices, enabling ultralow detection limits (0.8 to 30 ppb) with long-period stability for various metal oxide sensors.
218. Navigating high-dimensional processing parameters in organic photovoltaics via a multitier machine learning framework
Core Problem: The challenge of optimizing organic photovoltaic (OPV) performance due to the high-dimensional, interdependent processing parameters governing bulk heterojunction morphology.
Key Innovation: Developing a three-tiered machine learning framework (using gradient boosting regression trees) based on a standardized database of OPV experimental results, which achieves high accuracy (>80%) in identifying optimal multiparameter configurations and robust generalization for accelerating OPV photoactive layer optimization.
219. Nonlinear optical extreme learner via data reverberation with incoherent light
Core Problem: The growing energy and computational demands of artificial neural networks and the limitations of optical neural networks due to weak optical nonlinearities.
Key Innovation: Demonstrating a low-power, incoherent-light-compatible optical extreme learner that uses 'data nonlinearity' from optical pattern reverberations in a tailored optical cavity, achieving nonlinear transformations at extremely low optical power and outperforming linear digital networks in image classification tasks.
220. Data and knowledge driven Markov random field inference for geological cross-section reconstruction (Shangyuanmen floodplain, China)
Core Problem: The lack of geological knowledge often limits the interpretability and geological realism of Markov Random Field (MRF) models used for geological cross-section reconstruction.
Key Innovation: Proposed a novel MRF inference framework that integrates borehole records and geological knowledge, including a lithology classification strategy and Monte Carlo Boolean decision algorithm for fault modeling, significantly improving stratigraphic continuity and reducing uncertainty in subsurface geological interpretation.
221. Multi-satellite data fusion for improved field-scale evapotranspiration mapping on Google Earth Engine
Core Problem: Insufficient temporal sampling of Landsat data for field-scale evapotranspiration (ET) mapping, hindering accurate characterization of surface energy and water balance dynamics.
Key Innovation: A GEE-based framework integrating TIR observations from ECOSTRESS and VIIRS with Harmonized Landsat-Sentinel (HLS) data, using a sharpened LST and DisALEXI model, to produce daily 30-m ET estimates with improved accuracy.
222. Integrating very-high-resolution imagery, Sentinel-2 time-series data, and machine learning to map shrub fractional abundance across arid and semi-arid ecosystems in China
Core Problem: Challenges in large-scale shrub fractional abundance (SFA) mapping in arid/semi-arid regions due to small crown size, sparse distribution, and spectral overlap, hindering accurate detection with coarser satellite data or traditional surveys.
Key Innovation: A two-step approach integrating VHR imagery (for benchmark shrub crown maps via deep learning) and Sentinel-2 time-series data (for SFA prediction via XGBoost, leveraging phenological information) to achieve accurate, spatially continuous SFA maps.
223. Knowledge-data-model-driven multimodal few-shot learning for hyperspectral fine classification: Generalization across sensor, category and scene
Core Problem: Challenges in fine-grained land-cover mapping, especially for unseen or rare species/classes, due to limited samples and significant cross-sensor, cross-category, and cross-scene variations in hyperspectral images.
Key Innovation: Knowing-Net, a knowledge-data-model-driven multimodal few-shot learning network, which leverages prior sensor knowledge for cross-sensor image reconstruction, embeds multimodal data (textual descriptions, natural images) for unseen classes, and uses a cross-alignment mechanism and a sliding discriminant window for spatial context, achieving generalization across sensor, category, and scene.
224. Vegetation productivity and soil CO₂ correlation were decoupled during post-wildfire recovery in karst landscapes
Core Problem: Understanding the coupling between aboveground vegetation recovery and subsurface carbon processes after wildfire, especially in complex karst areas, and the potential long-term impacts on karst weathering.
Key Innovation: Demonstrated a pronounced decoupling between rapid surface vegetation recovery (driven by shallow-rooted plants) and lagged deep carbon dynamics (depressed deep-soil pCO₂) post-wildfire in karst landscapes, suggesting an overestimation of carbon sequestration if only surface indicators are used and potential long-term weakening of karst weathering sinks.
225. Climatic seasonality as a driver of soil carbon stability and soil quality in tropical semiarid regions
Core Problem: Understanding how seasonal climatic oscillations and different land-use systems affect soil quality and carbon stability in tropical semiarid regions.
Key Innovation: Demonstrated that soil quality indices (SQI) and carbon stabilization (TOC, Prot-C stocks) are significantly enhanced during the dry season in tropical semiarid regions, driven by organic carbon, microbial biomass, and clay-mediated protection, highlighting the importance of maintaining organic carbon pools and structural stability, especially during the rainy season when degradation risk is highest.
226. A climate-adapted GIS-based simulation-optimization method for optimal basin-scale wetland placement to mitigate nonpoint source pollution under shared socioeconomic pathways
Core Problem: The long-term efficacy of constructed wetlands (CWs) for mitigating non-point source (NPS) pollution under changing climate conditions is uncertain, and optimal placement strategies need to account for future climate scenarios.
Key Innovation: Developed a climate-adapted GIS-based simulation–optimization method (CA-GSOM) integrating SWAT, delta change, CTI analysis, and a genetic algorithm to optimize wetland configurations for NPS pollution mitigation under future climate scenarios (CMIP6 GCMs across SSPs), demonstrating shifts in optimal placement areas and reduced costs, providing a framework for climate-adaptive watershed planning.
227. Hydro-mechanical–chemical modelling of solute–consolidation coupling in near-saturated soils with entrapped bubbles
Core Problem: Understanding and predicting long-term solute transport and its coupled hydro-mechanical-chemical effects in deformable, low-permeability porous media, especially under unsaturated conditions.
Key Innovation: Presents a 3D HMC model for near-saturated soils with entrapped bubbles, incorporating osmotic pressure and chemically induced strain, demonstrating their relative importance and identifying conditions where chemical effects become significant.
228. Genetic programming-based closed-form solutions for predicting the compressive strength of cement-treated soils
Core Problem: Existing closed-form models for predicting the unconfined compressive strength (UCS) of cement-treated soils are often oversimplified, rely on a narrow set of input variables, or are tailored to specific soil types, limiting their general applicability.
Key Innovation: Developed accurate and physically meaningful closed-form predictive models for UCS of cement-treated soils using Gene Expression Programming (GEP) and Multi Expression Programming (MEP), incorporating a wide range of geotechnical and treatment parameters and demonstrating high predictive accuracy.
229. An LLM-driven agent framework for resilient modulus prediction: integrating modified consistency index model and residual compensation
Core Problem: Existing methods for predicting the resilient modulus (Mr) of subgrade soil exhibit significant shortcomings in testing efficiency, generalization capability, and interpretability, especially for multiple soil types and small-sample scenarios.
Key Innovation: Proposed a hybrid prediction model (MCI-XGBoost-RC) integrating a modified consistency index model with Bayesian-optimized XGBoost residual compensation, demonstrating superior accuracy, generalization, and interpretability for Mr prediction, and introduced an LLM-driven agent framework for intelligent engineering decision support.
230. A dual attention-based deep learning model for lithology identification while drilling
Core Problem: Traditional lithology identification models while drilling struggle with accurate feature extraction and adaptability in complex geological conditions, limiting their reliability.
Key Innovation: Proposed a dual attention-based deep learning model integrated with an LSTM network and optimized by the crayfish optimization algorithm (COA), significantly enhancing accuracy (97.15% lab, 91.96% field) and robustness for real-time lithology identification while drilling.