TerraMosaic Daily Digest: May 29, 2026
Daily Summary
May 29 is organized around a practical problem: terrain and infrastructure fail where water, structure, and dynamic loading concentrate into local instability. The strongest slope papers move from inventories to mechanism. Loess embankment failure is traced to runoff accumulation and berm erosion; unstable rock masses are rebuilt as 3-D point-cloud objects for rockfall analysis; Pyrenean rock-slope failures are connected to acid drainage and ferricrete formation; bedding structure is shown to route hydrology through landslide-prone accretionary mountains; and reservoir-bank seepage is recast as a high-frequency transient process that requires physics-informed learning rather than smooth hydraulic surrogates.
The same process emphasis extends across earthquakes, underground works, coasts, and observation systems. InSAR resolves the segmented Dingri rupture; telecommunication fibers provide dense urban site-amplification estimates in Istanbul; decadal field measurements constrain liquefaction aging; and true-triaxial tests expose how particle loss drives water and mud inrush in fault zones. Flood and coastal studies shift from static exposure to dynamic cascades, climate-conditioned storm genesis, and surrogate wave fields. The remote-sensing and AI papers are most relevant when they solve observation bottlenecks: SAR-optical fusion under cloud and degradation, GNSS-IR snow-depth retrieval, vision-language change interpretation, foundation-model groundwater prediction, and UAV rescue detection in terrain where conventional imagery fails.
Key Trends
Five methodological movements stand out: hydrologic-structural slope mechanics, opportunistic geophysical sensing, mechanism-resolved infrastructure hazard, dynamic coastal and flood risk modelling, and AI constrained by physics or sensor limits.
- Slope failure is being treated as a coupled hydrologic-structural process: Loess embankment collapse, root-controlled infiltration, reservoir-bank seepage, bedding-guided drainage, rock-slope ferricrete systems, and freeze-thaw sandstone damage all show that slope hazard cannot be reduced to topography or rainfall intensity alone.
- Dense geophysical observations are tightening earthquake and storm hazard constraints: Dingri InSAR modelling, Istanbul fiber-optic site-response mapping, liquefied-ground aging surveys, and optical-fiber fracture monitoring show a common move toward using opportunistic or repeated geophysical data to recover hazard-relevant state variables.
- Underground and offshore infrastructure studies are moving toward explicit failure mechanisms: Water-mud inrush testing, TBM shield jamming under rock creep, stress-wave interaction in faulted tunnels, shield slurry infiltration, large-deformation anchorage, scour-modified monopile response, and hydrate-bearing anchor-pile tests all identify the physical pathway from boundary condition to failure.
- Flood and coastal risk modelling is becoming dynamic rather than map-static: Urban flood cascade networks, post-eruption tropical-cyclone simulations, fast coastal wave-climate surrogates, gravel-nourishment morphodynamics, intertidal bathymetric stability metrics, and Amazon freshwater storage retrieval all shift emphasis from hazard extent toward evolving system state.
- AI contributes most when it is constrained by physics, sensors, or deployment limits: Coordinate-injected PINNs, AlphaEarth groundwater features, GNSS-IR snow-depth retrieval, VLM/LLM change detection, degradation-robust SAR-optical fusion, cloud removal, building-height weak supervision, and UAV rescue detection are useful because they address specific observation gaps rather than generic accuracy gains.
Selected Papers
This issue contains 54 selected papers from 1,072 papers analyzed. The selected papers are led by loess embankment slope instability, point-cloud modelling of unstable rock masses, rock-slope failure geochemistry, landslide-prone catchment hydrology, reservoir-bank seepage PINNs, Dingri earthquake InSAR, fiber-optic site amplification in Istanbul, liquefied-ground aging, tunnel water and mud inrush, TBM shield jamming, and stress-wave interaction in faulted rock tunnels. The broader set connects urban flood cascade networks, tropical-cyclone genesis simulations, freeze-thaw sandstone damage, root-controlled infiltration, scour-modified offshore foundations, liquefaction response, suprapermafrost groundwater exchange, glacier peak-water hydrology, coastal morphodynamics, ultra-low-temperature rock mechanics, GNSS-IR snow-depth retrieval, and targeted AI or remote-sensing methods for cloud removal, land-cover change interpretation, groundwater prediction, rescue detection, and exposure mapping.
1. Moisture evolution and instability mechanisms in high-fill loess embankment slopes subject to water infiltration from accumulated runoff following berm erosion
Core Problem: High-fill loess embankments can fail when berm erosion concentrates runoff, yet the coupled moisture evolution and collapse sequence are rarely resolved at operational scale.
Key Innovation: Transportation Geotechnics reconstructs intermittent-rainfall infiltration, cracking, collapse, and differential settlement in a high-fill loess embankment, linking local waterlogging to a staged instability mechanism.
2. Three-dimensional modelling of unstable rock masses using discrete point clouds
Core Problem: Rockfall-prone rock masses require explicit 3-D geometry before failure modes, volumes, and numerical models can be constrained.
Key Innovation: Engineering Geology develops a discrete point-cloud modelling workflow that reconstructs unstable rock masses and extracts geometry needed for rockfall hazard analysis and mechanical simulation.
3. Relationship between rock slope failures, natural acid rock drainage and ferricretes in the Axial Pyrenees (Spain)
Core Problem: Pyrite-rich mountain slopes can link mechanical failure, acid drainage, and iron precipitate formation, but that coupling is often mapped only qualitatively.
Key Innovation: Catena combines high-resolution orthophotography, DEMs, field mapping, hydrochemistry, and mineralogy in the Axial Pyrenees to show how rock-slope failures and bedding-plane springs structure ferricrete formation.
4. Dual control of bedding structure on catchment hydrology: the interplay between direct bedding-parallel recharge and indirect landslide-driven drainage in steep accretionary complexes
Core Problem: Catchment hydrology in high-relief landslide terrain depends on bedding structure, but the balance between deep recharge and shallow landslide-driven drainage remains poorly constrained.
Key Innovation: The study uses multi-period discharge and conductivity observations in a Japanese accretionary complex to separate direct bedding-parallel recharge from indirect drainage shaped by landslide-prone hillslopes.
5. Overcoming spectral bias in transient seepage problems of reservoir bank slopes via coordinate-injected physics-informed neural networks
Core Problem: Rapid reservoir drawdown generates high-frequency seepage dynamics that standard PINNs struggle to represent, limiting slope-stability assessment.
Key Innovation: Computers and Geotechnics introduces a coordinate-injected PINN that improves representation of phreatic-surface migration and coupled physical constraints in two-dimensional reservoir-bank slope seepage.
6. Seismic Source Complexities Revealed by InSAR and Analytical Modeling: The 2025 Mw 7.1 Dingri Earthquake
Core Problem: The 2025 Mw 7.1 Dingri earthquake produced complex deformation that requires fault segmentation beyond simple single-plane source models.
Key Innovation: Remote Sensing combines Sentinel-1 and ALOS-2 InSAR with analytical inverse modelling to resolve four fault segments and evaluate coseismic stress transfer.
7. Shallow Modeling and Site‐Amplification From Telecommunications Fibers in Istanbul
Core Problem: Urban seismic hazard needs shallow velocity and amplification maps, but dense geophone deployments are difficult in cities.
Key Innovation: Geophysical Research Letters repurposes an 8 km telecommunication fiber to recover shallow shear-wave velocity and site amplification in Istanbul, with estimates checked against simulations and earthquake data.
8. Aging effects of liquefied ground for over a decade estimated from in-situ sounding tests and geophysical exploration
Core Problem: Liquefaction resistance can change after past shaking, but long-term field evidence for reclaimed ground remains sparse.
Key Innovation: Soils and Foundations combines repeated sounding tests and MASW over about a decade to quantify post-liquefaction aging effects and their limits in reclaimed ground.
9. Mechanism of water and mud inrush induced by seepage failure in the filling media of fault fracture zones: A true triaxial laboratory investigation
Core Problem: Tunnel inrush hazards are driven by coupled stress, seepage, particle migration, and strength loss in fault-zone filling media.
Key Innovation: JRMGE develops a visualized true-triaxial apparatus that tracks permeability, porosity, cohesion, shear strength, and viscosity during seepage failure and particle loss.
10. Model testing and numerical simulation study on TBM shield jamming risk induced by rock creep
Core Problem: TBM shutdowns in creeping rock can evolve into shield jamming, but allowable stoppage time is rarely tied quantitatively to depth and deformation pressure.
Key Innovation: Tunnelling and Underground Space Technology combines model tests and simulations to relate creep deformation, contact pressure, and maximum shutdown time in a Xinjiang tunnel case.
11. Quantitative visualization of stress wave propagation and dynamic wave–interface coupling in faulted rock tunnels using 3D-printed photoelasticity
Core Problem: Fault interfaces can concentrate dynamic stresses during blasting or impact, yet full-field transient wave interactions are hard to measure.
Key Innovation: International Journal of Rock Mechanics and Mining Sciences uses 3-D printed photoelastic tunnel models to quantify stress-wave reflection, transmission, scattering, and fault-geometry-dependent stress concentrations.
12. System risk assessment under urban flooding disaster based on spatialized cascade network
Core Problem: Urban floods propagate through interdependent facilities, while historical disaster information is often locked in unstructured text.
Key Innovation: Reliability Engineering & System Safety extracts flood-evolution relations with NLP and builds a spatialized urban flood cascade network to rank vulnerable facilities and district-level risk.
13. Evolution of damage and permeability in sandstone subjected to direct shear loading after freeze-thaw cycles
Core Problem: Cold-region rock slopes and excavations depend on how freeze-thaw cycling changes shear damage and flow pathways.
Key Innovation: JRMGE combines direct shear tests, permeability and acoustic-emission monitoring, and CFD-DEM modelling to resolve permeability evolution in freeze-thaw damaged sandstone.
14. Spatial heterogeneity of soil infiltration affected by tree roots: Insights from in-situ double-ring infiltrometer tests and soil physicochemical analysis
Core Problem: Rainfall-triggered slope stability assessments often treat infiltration as spatially uniform and neglect root-controlled heterogeneity.
Key Innovation: Engineering Geology couples in-situ infiltration, dye tracing, root measurements, and soil property analysis to quantify how tree roots alter saturated conductivity and slope-relevant infiltration patterns.
15. Scour effects on the dynamic pile-soil interaction in transversely isotropic layered saturated soils under coupled static-dynamic loads
Core Problem: Scour removes lateral confinement around offshore foundations and changes dynamic response in layered saturated soils.
Key Innovation: Ocean Engineering derives a coupled static-dynamic model for monopiles in transversely isotropic layered saturated soils with explicit scour and excess pore-pressure effects.
16. Three-dimensional scour evolution and equilibrium scour depth prediction model around large-diameter cylinders under combined waves and currents
Core Problem: Scour around large offshore cylinders under combined waves and currents remains difficult to predict for foundation design.
Key Innovation: Ocean Engineering quantifies three-dimensional scour evolution and develops an equilibrium scour-depth prediction model for large-diameter cylinders.
17. Energy-based liquefaction assessment of saturated sand mixed with granulated rubber
Core Problem: Rubber-included granular materials are increasingly used in seismic subbases and backfills, but their liquefaction response needs mechanistic calibration.
Key Innovation: Soil Dynamics and Earthquake Engineering evaluates saturated sand mixed with granulated rubber using an energy-based liquefaction assessment framework.
18. Numerical framework for calculating stiffness of liquefied sands during post-shaking re-solidification
Core Problem: The recovery of stiffness after liquefaction affects post-earthquake deformation but is not well represented in routine models.
Key Innovation: Computers and Geotechnics proposes a numerical framework for calculating stiffness of liquefied sands during post-shaking re-solidification.
19. Experimental investigation and failure analysis of the adobe Hittite city wall in the UNESCO archaeological site of Arslantepe (Turkey)
Core Problem: Earthen archaeological structures can be highly vulnerable to earthquake loading, yet their damage mechanisms require geometry-aware testing.
Key Innovation: Soil Dynamics and Earthquake Engineering combines digital survey, experimental investigation, and failure analysis for the Hittite adobe wall at Arslantepe.
20. Application of the FCM algorithm and improved grey target theory on the prediction of rock burst
Core Problem: Rock bursts are high-risk underground geohazards whose intensity prediction is distorted by correlated indicators and redundant information.
Key Innovation: Frontiers in Earth Science integrates fuzzy C-means clustering with improved grey target theory to quantify rock-burst intensity in an engineering case.
21. Identifying the State Dependence of Effective Material Properties in a Simplified Hydrologic Hillslope Model
Core Problem: Hydrologic closure relationships at hillslope scale change with state, affecting runoff and pore-pressure controls relevant to slope response.
Key Innovation: Water Resources Research identifies state dependence in effective material properties for a simplified hillslope model, improving process representation of subsurface storage-discharge behavior.
22. Anomalous Sample Detection Tool and Its Application in Identifying Interbasin Groundwater Flow in the Yarlung Zangbo River Basin, Tibetan Plateau
Core Problem: Interbasin groundwater flow through major rifts can alter mountain water budgets but is hard to diagnose from regional observations.
Key Innovation: Water Resources Research applies an anomalous-sample detection tool to identify interbasin groundwater pathways in the Yarlung Zangbo River Basin.
23. Elevation-conditioned isotope diagnostics of seasonal reversals in surface water–suprapermafrost groundwater exchange in Tibetan Plateau headwaters
Core Problem: Permafrost headwaters exchange water seasonally between streams and suprapermafrost aquifers, but the direction and elevation dependence are difficult to observe.
Key Innovation: Journal of Hydrology develops an elevation-conditioned isotope framework that diagnoses seasonal reversals of surface water and suprapermafrost groundwater exchange in the Three-River Headwaters Region.
24. Identifying Energy Balance Drivers of Greenland Ice Sheet Surface Melt Using Causal Discovery
Core Problem: Surface melt on the Greenland Ice Sheet reflects interacting energy-balance controls that correlations alone cannot separate.
Key Innovation: Geophysical Research Letters applies causal discovery to CESM2 ensemble output to identify causal dependencies controlling Greenland Ice Sheet surface melt.
25. Tropical Peatland Restoration Reduces Fire Occurrence
Core Problem: Tropical peat fires are climate and haze hazards, but the measurable effect of restoration on fire occurrence remains contested.
Key Innovation: Geophysical Research Letters shows that tropical peatland restoration reduces fire occurrence, linking land restoration to hazard mitigation.
26. Post‐Eruption Tropical Cyclone Genesis Potential Change Over the Western North Pacific Modulated by Initial Oceanic Conditions
Core Problem: Tropical cyclone genesis after major volcanic eruptions remains uncertain because internal climate variability can obscure forcing effects.
Key Innovation: Geophysical Research Letters uses last-millennium CESM ensembles to show how eruption strength and initial oceanic conditions modulate western North Pacific cyclone genesis potential.
27. Can streamflow observations constrain snow mass reconstructions? Lessons from two synthetic numerical experiments
Core Problem: Snow water storage is central to cold-region flood and drought risk, but its reconstruction is underconstrained where direct snow observations are sparse.
Key Innovation: HESS tests whether streamflow observations can constrain snow-mass reconstructions through synthetic numerical experiments.
28. Hydrological response to glacier peak water in the largest inland river basin of China
Core Problem: Glacierized basins may shift from increasing to declining meltwater supply after peak water, with implications for downstream hazards and water security.
Key Innovation: Journal of Hydrology projects hydrological response to glacier peak water in Chinas largest inland river basin.
29. Improving spatiotemporal assessment of watershed soil redistribution by integrating source-sink theory into a multi-model framework
Core Problem: Watershed erosion studies often emphasize source detachment while underrepresenting deposition and redistribution pathways.
Key Innovation: Journal of Hydrology integrates source-sink theory into a multi-model framework for spatiotemporal assessment of watershed soil redistribution.
30. Fast coastal wave climate prediction using autoencoders-based models
Core Problem: Coastal risk applications require wave-field predictions faster than full numerical models can provide in complex nearshore settings.
Key Innovation: Ocean Engineering combines sampling, clustering, autoencoders, and MLPs to emulate SWAN wave fields for rapid coastal wave-climate prediction.
31. Morphodynamics of gravel nourishment in front of a vertical seawall under irregular waves
Core Problem: Gravel nourishment can dissipate wave energy at seawalls, but its storm-driven morphological change remains difficult to predict.
Key Innovation: Ocean Engineering integrates laboratory experiments and XBeach-G simulations to assess nourishment morphodynamics under irregular waves.
32. Assessing the temporal stability of satellite-derived bathymetry in intertidal flats using Sentinel-2A/B/C
Core Problem: Low-relief intertidal flats complicate bathymetry because apparent depth change can reflect optical variability as much as morphology.
Key Innovation: Coastal Engineering introduces an Intertidal Bathymetric Stability Index using multi-temporal Sentinel-2A/B/C composites and in-situ validation.
33. Towards comprehensive multi-task land cover change detection leveraging vision-language model and LLM-driven agents
Core Problem: Change detection for hazard applications needs methods that can connect visual evidence to task-level reasoning across land-cover changes.
Key Innovation: ISPRS Journal of Photogrammetry develops a comprehensive multi-task land-cover change-detection framework using vision-language models and LLM-driven agents.
34. SAMDFuse: Semantic-Aware Multi-scale Degradation-robust Fusion network for SAR and optical images
Core Problem: Disaster mapping often requires fusing SAR and optical imagery under clouds, noise, and modality-specific degradation.
Key Innovation: ISPRS Journal of Photogrammetry proposes SAMDFuse, a semantic-aware multi-scale network for degradation-robust SAR and optical image fusion.
35. A new AI paradigm for groundwater level prediction: integrating the AlphaEarth foundations model in the Leizhou Peninsula
Core Problem: Groundwater prediction in coastal settings is limited by sparse wells and nonlinear environmental controls.
Key Innovation: Frontiers in Earth Science integrates AlphaEarth foundation-model embeddings with TabPFN to predict groundwater levels in the Leizhou Peninsula.
36. A Lightweight Multi-Level Feature Fusion Detector for UAV-Based Tiny Personnel Detection in Hilly Road Safety Monitoring and Rescue
Core Problem: Aerial rescue in hilly terrain is limited by small target size, terrain occlusion, and cluttered backgrounds.
Key Innovation: Remote Sensing proposes a lightweight frequency-aware YOLO detector for UAV-based tiny personnel detection in hilly road safety monitoring and rescue.
37. A review of multi-category forest fuel load estimation methods from traditional measurements to remote sensing and future perspectives: towards fire prevention
Core Problem: Wildfire risk modelling depends on fuel-load estimates that remain inconsistent across measurement and remote-sensing approaches.
Key Innovation: Remote Sensing of Environment reviews forest fuel-load estimation from field measurements to remote sensing and identifies pathways toward operational fire prevention.
38. Experimental study on dynamic response of a calcareous seabed around a trenched pipeline under non-linear wave–current loading
Core Problem: Subsea pipelines in calcareous seabeds face coupled wave-current loading that can modify seabed response and pipeline stability.
Key Innovation: Ocean Engineering investigates dynamic response of a calcareous seabed around a trenched pipeline under nonlinear wave-current loading.
39. Centrifuge model test of bearing capacity of anchor piles in hydrate-bearing silty sand
Core Problem: Hydrate-bearing seabed sediments can change foundation bearing capacity for offshore and deep-sea infrastructure.
Key Innovation: Ocean Engineering uses centrifuge model tests to evaluate anchor-pile bearing capacity in hydrate-bearing silty sand.
40. Micromechanical analysis of suction pile-soil interactions using a three-dimensional DEM-FEM approach: Combined effects of pull angle and friction
Core Problem: Suction pile design needs particle-scale understanding of pile-soil interaction under different pull angles and interface conditions.
Key Innovation: Ocean Engineering applies a coupled DEM-FEM model to resolve micromechanical suction pile-soil interactions.
41. Undrained uniaxial and combined VHM capacities of horizontal plate anchors in anisotropic and non-homogeneous clays
Core Problem: Anchor capacity estimates are uncertain where seabed clays are anisotropic and non-homogeneous.
Key Innovation: Ocean Engineering evaluates undrained uniaxial and combined VHM capacities of horizontal plate anchors in anisotropic, non-homogeneous clays.
42. Infiltration behavior of shield slurry in highly permeable sandy strata: non-newtonian rheological effects and pressure transmission mechanisms
Core Problem: Slurry shield face stability depends on how non-Newtonian slurry infiltrates permeable sandy strata and transmits support pressure.
Key Innovation: Transportation Geotechnics analyzes infiltration behavior and pressure transmission mechanisms of shield slurry in highly permeable sands.
43. Experimental modulus degradation and damping curves for sands under pure compressional loading
Core Problem: Earthquake site-response analyses require soil nonlinearity curves that are appropriate for compressional loading, not only shear loading.
Key Innovation: Soil Dynamics and Earthquake Engineering reports experimental modulus-degradation and damping curves for sands under pure compressional loading.
44. Lateral kinematic response of single-pile bridge pier foundations under seismic waves: Linear versus non-linear soil behaviour
Core Problem: Bridge foundation response to vertically propagating seismic waves depends on whether soil behaviour is treated as linear or nonlinear.
Key Innovation: Soil Dynamics and Earthquake Engineering compares lateral kinematic response of single-pile bridge piers under linear and nonlinear soil models.
45. Analysis method for longitudinal structural response of shield tunnels incorporating iterative updates of circumferential joint bending and shear stiffness
Core Problem: Longitudinal shield-tunnel response depends on evolving joint bending and shear stiffness that simplified models can miss.
Key Innovation: Tunnelling and Underground Space Technology develops an analysis method that iteratively updates circumferential-joint stiffness in shield-tunnel structural response.
46. Brittle-to-ductile transformation of rock induced by constant-resistance large-deformation anchorage: Implications from experimental tests on rock-like material
Core Problem: Deep underground support requires reinforcement systems that transform brittle rock response into ductile deformation capacity.
Key Innovation: Tunnelling and Underground Space Technology tests constant-resistance large-deformation anchorage and its influence on brittle-to-ductile transformation in rock-like material.
47. Assessment of hydraulic fracture propagation in layered rock using optical frequency domain reflectometry technology
Core Problem: Layered rock fracture propagation is difficult to monitor with high spatial resolution during hydraulic loading.
Key Innovation: JRMGE applies optical frequency domain reflectometry to assess hydraulic-fracture propagation in layered rock.
48. Artificial intelligence-based back analysis of geomechanical parameters in tunnelling: A state-of-the-art review
Core Problem: Tunnel model reliability depends on geomechanical parameters that must often be inferred from monitoring data.
Key Innovation: JRMGE reviews artificial-intelligence-based back analysis of geomechanical parameters in tunnelling, clarifying data-driven parameter updating for underground construction.
49. Surface freshwater storage and recent trends in the Amazon Basin from spaceborne GNSS-R and multi-satellite observations (2019–2024)
Core Problem: Basin-scale flood and drought monitoring depends on spatially continuous estimates of surface freshwater storage.
Key Innovation: Science of Remote Sensing combines spaceborne GNSS-R with multi-satellite observations to map Amazon Basin freshwater storage trends from 2019 to 2024.
50. Nonlinear mechanical behavior of rocks in ultra-low-temperature conditions: classifications and micro mechanisms
Core Problem: Polar and ultra-low-temperature engineering requires rock-mechanics models that separate frost-heave effects from thermal control on strength and deformation.
Key Innovation: Earth-Science Reviews synthesizes classifications and micro-mechanisms of nonlinear rock mechanical behaviour below -40 degrees C.
51. An improved RF-XGBoost integrated approach for GNSS-IR snow depth estimation
Core Problem: GNSS-IR snow-depth retrieval is degraded by antenna-height uncertainty and terrain heterogeneity, limiting robust snow monitoring.
Key Innovation: Cold Regions Science and Technology integrates optimized signal decomposition with RF-XGBoost ensemble learning to improve GNSS-IR snow-depth estimation.
52. DCAN-CR: A Dual Cross-Modal Attention Network for Cloud Removal Based on Optical and SAR Images
Core Problem: Cloud contamination limits optical imagery exactly when disaster and land-surface monitoring need timely observations.
Key Innovation: IEEE JSTARS proposes DCAN-CR, a dual cross-modal attention network that reconstructs cloud-contaminated optical images using SAR reference information.
53. Enhancing Monocular Building Height Estimation via Weak Supervision From Imperfect Labels
Core Problem: Exposure and damage models need building-height information, but high-quality height labels are scarce outside well-mapped regions.
Key Innovation: IEEE JSTARS incorporates imperfect out-of-domain labels into monocular building-height estimation to improve scalable 3-D urban mapping.
54. Pixel’s Neighbors Are Noteworthy: Localized Vision–Language Attention for Remote Sensing Semantic Segmentation
Core Problem: Vision-language segmentation in remote sensing can miss local spatial continuity that matters for high-resolution hazard interpretation.
Key Innovation: Remote Sensing introduces LoVLANet, combining language-driven semantics with localized neighborhood attention for remote-sensing semantic segmentation.