The spatiotemporal knowledge graph (STKG) represents the most effective framework for organizing and disseminating multimodal, heterogeneous spatiotemporal knowledge. As a cutting-edge frontier in geographic information science, STKGs fulfill critical requirements for high-quality national development and modern industrial governance. Despite significant advances in STKG construction, large-scale systematic industrial applications remain nascent. Key challenges persist, particularly in modeling complex spatiotemporal features/relationships and delivering intelligent knowledge services.This paper systematically examines the connotations, characteristics, and classifications of spatiotemporal knowledge while conducting an in-depth analysis of seven core challenges: (1) STKG representation models, (2) spatiotemporal data acquisition and quality assessment, (3) spatiotemporal information extraction and alignment fusion, (4) STKG construction and dynamic updating, (5) STKG computational reasoning and knowledge discovery, (6) efficient visualization/retrieval mechanisms, and (7) multi-scenario application services. Corresponding methodological solutions are elaborated.
Focusing on six spatiotemporal data types—professional spatial datasets, mobile trajectories, IoT sensor feeds, scientific literature, web text, and geo-semantic web resources—this research achieves automated construction and intelligent services spanning “data → information → knowledge.” Core advancements include: (1) Adaptive Representation Model: A spatiotemporal correlation-aware knowledge graph expression framework integrating embedded representations for diverse spatiotemporal features. (2) Quality Control System: Data source/content-oriented evaluation indices and quality factor extraction methods for spatiotemporal knowledge graph development. (3) Multimodal Information Mining: Hybrid shallow mapping/deep learning approaches for high-precision spatiotemporal information extraction, coupled with collaborative time-space-semantic modeling for tuple alignment/fusion. (4) Inference & Discovery: Ontology-constrained joint embedding for vector-based spatiotemporal reasoning, combined with textual inference and structural learning for new knowledge discovery. (5) Visualization Efficiency: Adaptive hierarchical nebula visualization and spatiotemporal index-integrated pruning algorithms to enhance large-scale STKG retrieval performance. (6) Engineering Applications: Three-tiered service architecture—(i) foundational spatiotemporal big data governance, (ii) business-process-oriented core knowledge services, and (iii) large-model-empowered knowledge fusion.
To address urgent demands from AI advancement—particularly high-quality large-model dataset development—future STKG research must prioritize: (1) refining modeling methodologies and core technologies, (2) establishing large-scale high-fidelity STKG infrastructure,(3) deepening STKG-large model integration, and (4) accelerating engineering/business implementation.

This study presents a systematic review of current domestic and international research on automated recognition methods across three distinct categories. For confidential scale recognition, existing approaches attempt to infer scale from map elements such as element density or detail levels, yet they frequently suffer from imprecision and operational complexity. In the domain of sensitive annotation recognition, technological evolution has progressed from conventional optical character recognition (OCR) systems to deep learning-based models. Nevertheless, significant challenges persist in handling complex Chinese typography, rotated text orientations, and annotations overlapping with other cartographic features. Regarding graphical misrepresentation recognition, while object detection frameworks derived from deep learning have shown promise, they demonstrate limited generalization capacity when confronted with diverse boundary deformation patterns and topological errors. This systematic review confirms an inevitable technological trajectory: the transition from purely manual inspection processes and data-driven algorithms toward a more robust paradigm known as hybrid intelligence computing.
Hybrid intelligence, which integrates human domain expertise with machine computational capabilities, emerges as a promising solution. Despite its potential, practical implementation in problematic map recognition remains nascent and confronts four fundamental scientific and technical challenges: (1) Knowledge extraction for map inspection continues to rely excessively on subjective, non-standardized expert experience; (2) The coupling mechanism required for effective integration of human intelligence (domain knowledge) with machine intelligence (algorithmic models) remains undefined; (3) A formal computational framework for hybrid intelligence applications in this domain is currently absent; and (4) There exists no unified hybrid intelligence computing model specifically tailored for problematic map recognition tasks.
To address these challenges, this investigation proposes a comprehensive strategy organized around a “knowledge-model-method- application” pipeline. This strategic framework encompasses four critical actions: First, overcoming the knowledge acquisition bottleneck through development of automated extraction, refinement, and parsing methodologies for map inspection knowledge, thereby transforming implicit expert cognition into machine-processable representations. Second, constructing a unified hybrid intelligence computing model predicated on a three-stage closed-loop mechanism involving “human knowledge embedding—machine learning modeling—expert feedback optimization.” Third, developing novel families of hybrid intelligence-based recognition methods customized for each problem category. Fourth, implementing an integrated prototype system to empirically validate the efficacy and reliability of the proposed approach.
This work systematically delineates the critical pathway from labor-intensive, inefficient manual map inspection toward automated, intelligent, and reliable recognition systems. By establishing this hybrid intelligence-based processing pipeline, the research aims to enable high-precision automated recognition of problematic maps, thereby providing a robust technical foundation supporting national geographic information security and enhancing regulatory oversight capabilities in the digital era.

Spatiotemporal information is increasingly recognized as a critical component of national infrastructure. Accurate and autonomous spatiotemporal services are essential to advancing the digital economy and realizing Digital China. However, spatiotemporal computing faces significant challenges, including multi-source heterogeneous data, incompatible analytical tools, and diverse application scenarios. These limitations hinder service flexibility, efficiency, and cross-domain collaboration. Large models—with their robust capabilities in natural language understanding, multi-modal data integration, and intelligent reasoning—offer promising solutions to address these challenges, enhancing the flexibility, scalability, and autonomy of spatiotemporal computing services.
This paper systematically reviews advancements in large model technologies, tracing the evolution from general-purpose large language model (LLM) to domain-specific variants, and examines the emergence of LLM-driven agents capable of autonomous reasoning and task execution. We then focus on two complementary approaches for empowering spatiotemporal computing: domain-specific spatiotemporal large models and spatiotemporal computing agents.
Domain-specific spatiotemporal large models address inherent limitations of general models in handling spatiotemporal constraints, integrating domain knowledge, and supporting complex scenarios. Their development is analyzed through two lenses: data modalities and application contexts. For heterogeneous spatiotemporal data: Temporal knowledge graphs employ prompt engineering to fuse structured knowledge with LLMs. Spatiotemporal graphs explicitly model node adjacency and spatial topology to encode spatiotemporal dependencies. Videos leverage multimodal alignment techniques to harmonize visual-temporal representations with textual semantics. 3D data utilize three pathways: direct processing, fusion of multi-view 2D images with 3D positional encoding, and 2D image encoding. Across application domains, these models typically adopt pretrained foundations, augmented with domain-specific fine-tuning.
Spatiotemporal computing agents mitigate intelligence gaps in traditional systems. Key challenges include: automated goal comprehension, decomposition of complex tasks into executable steps, and feedback-driven self-optimization. An LLM-powered agent architecture— centered on core LLM and supported by planning, memory, and tool modules—enables autonomous environmental perception, reasoning, and action. This framework provides a novel methodological pathway for constructing spatiotemporal computing agents.
To demonstrate practical applicability, we present two case studies based on our team's prior work: 3D spatial perception and farmland protection. In the 3D scenario, Route3D integrates open-source 3D models with lightweight routing modules to meet diverse perceptual requirements. In farmland protection, a multi-agent system—developed via knowledge base construction, single-agent design, and collaborative workflows—delivers specialized spatiotemporal knowledge services. Collectively, these cases validate the feasibility of leveraging spatiotemporal large models and computing agents to advance precision and autonomy in spatiotemporal services.

Land cover is a comprehensive term that encompasses various coverings and their characteristics on the Earth's surface, reflecting the distribution and evolution of material types. It primarily includes surface vegetation, glaciers, lakes, marsh wetlands, and various buildings, serving as essential foundational data in fields such as food security, greenhouse gas emissions, ecosystem structure and function, water resource circulation and redistribution, and biogeochemical cycles. However, due to the influence of production technology, land cover data may encounter issues of misclassification and omission, which directly impact the effectiveness of the data in subsequent applications. To ensure the reliability of future applications, the accuracy assessment of land cover data is an indispensable component of the production process and has become a key research focus in the production and application of remote sensing data both domestically and internationally.
Current research on accuracy assessment methods for land cover data encompasses all aspects of the evaluation process, including sampling methods, sample judgment and checking, and a variety of evaluation indicators. This paper aims to clarify the necessity, main technical components, and practical cases of land cover data accuracy assessment methods in recent years, providing a systematic reference for land cover accuracy assessment.
The main method for accuracy assessment of land cover data is the confusion matrix, from which several indicators, such as overall accuracy, Kappa coefficient, producer's accuracy, and user's accuracy, are derived. To address the challenge of unbalanced sampled probabilities in stratified sampling accuracy assessment, a weighted confusion matrix has been developed, allowing for the calculation of weighted overall accuracy, weighted producer's accuracy, and weighted user's accuracy. Additionally, metrics such as quantitative disagreement and allocation disagreement between classified maps and reference data samples, as well as the F1 score, have also been introduced.
Sampling for verification involves the selection of representative samples within the verification area based on statistical sampling principles. This serves as the first step in land cover validation and significantly influences its accuracy and objectivity. Key principles guiding this process include the probability principle, feasibility principle, emphasis on rare classes, and considerations of spatial heterogeneity and spatial correlation. The main tasks include calculating sample sizes and designing spatial layouts.
The collection of reference data for sample checking is the process of determining the true ground type at sample locations based on reference data to evaluate the correctness of classifications. This stage is also the most time-consuming and costly part of the accuracy evaluation process. Reference data selection should adhere to principles such as temporal proximity, seasonal consistency, high spatial resolution, high geometric accuracy, compatibility of classification system, and authoritative sources. Methods such as multi-scale sample checking, multi-semantic sample checking, and credibility grading can be employed.
Finally, some cases of the accuracy assessment processes of key land cover datasets at a 30 m resolution, such as GlobeLand30, GLC_FCS30, and FROM-GLC, are analyzed. Current accuracy assessment methods exhibit two notable characteristics. First, the design of sampling methods has diversified. Many validation cases for medium-resolution land cover datasets take into account spatial heterogeneity and spatial correlation, involving related index calculations, quantitative analyses, and geographic stratification designs, leading to increasingly varied sampling approaches. Second, the assessment indicators have expanded beyond traditional measures such as overall accuracy, user's accuracy, producer's accuracy, and the Kappa coefficient, now incorporating a wider range of metrics.
In conclusion, the core mission of land cover accuracy assessment focuses on achieving a deep integration of standardization, ensuring that land cover data products reliably support major decision-making in areas such as ecological security, territorial governance, and climate change response.

[Objective] Time series of three-dimensional coordinates derived from continuous global navigation satellite system (GNSS) observations serve as critical data carriers for elucidating geophysical phenomena, including crustal deformation and land subsidence. High-precision prediction of GNSS coordinate time series is essential for engineering applications such as millimeter-level reference frame maintenance and early warning systems for geological hazards. While deep learning (DL) methods have advanced GNSS coordinate time series forecasting, their “black-box” nature impedes statistical interpretability and reliability in geodetic contexts. To address this limitation, this study introduces and rigorously evaluates a least-squares-based deep learning (LSBDL) methodology.
[Method] Daily GNSS solutions from six international GNSS service (IGS) stations in and around China (2015-2019) were processed. Cartesian coordinates extracted from SINEX files were transformed into east-north-up (ENU) components, with primary emphasis on the vertical (U) dimension. Data preprocessing comprised outlier excision and gap interpolation. For each station, data spanning January 2015 to December 2018 (1461 samples) constituted the training set, while January-December 2019 (365 samples) formed the test set. The LSBDL framework integrated a nonlinear feature extractor with a closed-form least-squares (LS) estimator: A fully connected layer with ReLU activation mapped lagged inputs to a design matrix, followed by Tikhonov-regularized LS to derive linear coefficients. Training alternated between closed-form LS updates for the linear component and steepest-descent optimization for network weights. Four window lengths (7, 30, 180, 365 days; one-step-ahead prediction) were evaluated. Prediction accuracy and goodness-of-fit were quantified using root mean square Error (RMSE), mean absolute error (MAE), and Pearson's correlation coefficient (R); computational efficiency was measured via iteration time per run; stability was assessed through repeated trials with randomized initial weights. Baseline single-layer LSTM and GRU models (50 units; 200 epochs) were implemented on identical data splits for comparative analysis.
[Result] The LSBDL model achieved millimeter-level accuracy and robust fit for short-term predictions. Across all stations, 7-day and 30-day forecasts exhibited RMSE values of 3.1-4.7 mm and R coefficients of 0.72-0.9, with typical computation times less than 1 second(≤1 s for 7/30/180-day windows; 1-2 s for 365-day windows). At the CHAN station, the 30-day window yielded an RMSE of 3.6 mm and R = 0.818. For 7-30-day horizons, LSBDL matched or marginally surpassed LSTM/GRU performance while demonstrating 1-2 orders of magnitude faster computation. However, for extended windows (180/365 days), LSBDL showed diminished fit quality (R degradation), whereas LSTM/GRU maintained higher correlation (R ≈ 0.82) with RMSE ≈ 3.55 mm. Stability analyses revealed minimal variability in short-term forecasts (ΔR ≤ 0.01) but pronounced dispersion at 365-day horizons, indicating heightened sensitivity to initialization when long-term dependencies dominate.
[Conclusion] Experimental results demonstrate that LSBDL constitutes an efficient and accurate predictive framework suitable for short-term feature extraction in GNSS coordinate time series. Nevertheless, prediction accuracy and stability require enhancement as temporal horizons extend and the complexity of time-scale features increases. This work establishes a foundation for developing interpretable DL architectures tailored to geodetic time series analysis, with implications for real-time monitoring applications demanding both precision and computational economy.

[Objective] Urban deformation monitoring is crucial for ensuring infrastructure safety and preventing geological hazards in rapidly urbanizing regions. However, existing frameworks for assessing the reliability of urban deformation monitoring systems remain underdeveloped, and available methods for enhancing this reliability are insufficient. This study aims to establish a comprehensive reliability assessment system for urban deformation monitoring and propose effective strategies for improving the reliability of such systems.
[Method] To address the challenges in urban deformation monitoring, we propose a reliability assessment framework based on a dual-dimensional evaluation system that considers both algorithmic performance and result consistency. The framework incorporates key reliability indicators, including completeness, robustness, precision, consistency, and applicability. Additionally, we develop a multidimensional collaborative monitoring network by integrating sky-ground-well platforms—such as InSAR, GNSS, leveling, and underground sensors—to enhance data integrity and robustness. Multi-source data fusion techniques, including the integration of multi-track InSAR data, are employed to identify and remove outliers, thereby reducing deformation errors. The study also investigates deformation mechanisms, with a focus on the six-month lag between groundwater extraction and surface subsidence. An intelligent classification method is proposed to improve the fine-scale interpretation of deformation signals, combining 3D spatial correlation analysis with local elevation statistical filtering.
[Result] The proposed methods were validated through case studies conducted in the Binhai New Area of Tianjin, China. Results demonstrate that the reliability enhancement framework significantly improves the precision, consistency, and applicability of urban deformation monitoring. For instance, after applying multi-track InSAR data fusion, 11.1% of gross errors were successfully identified and removed from homonymous permanent scatterer (PS) sets, leading to a 12% reduction in overall deformation errors. The analysis of deformation features provided valuable insights into the mechanisms underlying surface subsidence. Moreover, the fine-scale interpretation method achieved a classification accuracy exceeding 95%, effectively linking deformation signals to specific ground features, such as buildings and roads. This greatly enhances the utility of the results for urban safety and infrastructure health assessment.
[Conclusion] This study establishes a comprehensive methodological framework for assessing and improving the reliability of urban deformation monitoring. By integrating multi-platform sensing, advanced multi-source data fusion, mechanism-based feature recognition, and fine-scale interpretation, the proposed approach effectively mitigates uncertainties arising from data sources, algorithms, and interpretation processes. The framework not only provides robust data support but also offers a theoretical foundation for urban geological hazard prevention and infrastructure safety evaluation. This research contributes valuable technical pathways for enhancing the reliability and actionability of urban deformation monitoring, with broad implications for urban planning, disaster mitigation, and infrastructure management.

[Objective] Lakes, reservoirs, and rivers are critical components of the terrestrial water cycle, playing essential roles in water resource management, flood warning, and climate change research. Accurate monitoring of water level and area variations is crucial for regional water security. Traditional satellite altimetry, constrained by its nadir observation mode, suffers from limited spatial coverage and low temporal resolution when applied to inland water monitoring. Although the surface water and ocean topography (SWOT) satellite, launched in December 2022, offers wide-swath observation capabilities, its performance in regional-scale surface water monitoring and the accuracy of water level and area measurements remain to be validated. China is currently conducting a nationwide basic survey of water resources, which urgently requires efficient surface water monitoring techniques. Therefore, this study aims to evaluate the observation capability and accuracy of the SWOT satellite for surface water monitoring in Guangdong Province, and to develop an improved method for enhancing area extraction precision.
[Method] This study selected Xinfengjiang reservoir and the Shakou reach of the Beijiang River in the Pearl River Basin as representative study areas, representing lake/reservoir-type and river-type water bodies, respectively. Water level information was extracted from SWOT L2_HR_Raster data through quality flag filtering (wse_qual≤1, wse_uncert≤1 m) and a three-sigma outlier removal criterion. An improved area extraction method was proposed based on Sobel gradient edge detection combined with a frustum model, which establishes the positive correlation between water level and area to correct misclassification errors in SWOT products. Multi-source data, including in-situ hydrological station measurements, ICESat-2 ATL13 laser altimetry, and Sentinel-1 SAR imagery with dual-polarized water index (SDWI), were integrated for accuracy validation. The observation coverage of SWOT was compared with traditional altimetry satellites, including Jason-3, Sentinel-3A/B, CryoSat-2, and ICESat-2, through spatial overlay analysis.
[Result] The results demonstrate that: (1) SWOT exhibits superior spatiotemporal coverage, capable of observing 94.3% of lakes/reservoirs and 97.7% of river reaches at least once within a 21-day repeat cycle, with approximately 65% of lakes observable twice, significantly outperforming traditional nadir altimetry satellites. (2) For water level accuracy, the mean absolute error is 10.3 cm for Xinfengjiang reservoir (RMSE=13.9 cm, R2=0.999) and 15.2 cm for the Shakou reach (R2=0.85), achieving decimeter-level monitoring capability that basically meets SWOT's design accuracy of 0.1 m for large lakes. (3) The original L2_HR_LakeSP product shows low correlation (R2=0.11) between extracted area and measured water level due to systematic misclassification of urban buildings as water bodies. After applying the proposed correction method, the maximum area error decreased from 29.12% to 14.77%, and the correlation coefficient increased to 0.97, consistent with the physical relationship between water level and area.
[Conclusion] This study validates that the SWOT satellite provides an effective means for regional surface water monitoring with high coverage and decimeter-level accuracy. The proposed Sobel-frustum correction method can effectively improve area extraction precision by establishing water level-area constraints. These findings provide technical support for basic water resource surveys in ungauged regions and demonstrate the application potential of integrating SWOT with multi-source remote sensing data for comprehensive surface water monitoring.

[Objective] Road cracks represent one of the most common types of pavement distress, compromising both driving safety and infrastructure longevity. Timely and precise crack detection is critical for optimizing intelligent road maintenance systems and enhancing transportation safety management. However, conventional deep learning-based detection models often struggle with blurred crack boundaries, false negatives, and fragmented structural representation due to the inherent fine-scale irregularities and complex morphologies of crack patterns. To address these limitations, this study introduces FreqFusion-YOLOv8, a frequency-aware feature fusion framework designed to improve edge-preserving capabilities and multi-scale contextual learning in road crack detection tasks.
[Method] The proposed method leverages the lightweight YOLOv8n architecture as its foundational backbone, replacing the original top-down concatenation operation with the novel FreqFusion module. This module employs a frequency-domain adaptive fusion mechanism to dynamically integrate semantic and structural information across hierarchical layers. Specifically, high-frequency components are prioritized to retain intricate texture details and sharp edges, while low-frequency components capture broader contextual cues and global structural integrity. This dual-path strategy enables balanced optimization of fine-grained morphological perception and large-scale contextual comprehension, thereby enhancing the continuity and completeness of detected crack segments. Model training and validation were conducted on the RDD2022_China benchmark dataset, which encompasses diverse crack types including longitudinal, transverse, and alligator cracks under varying environmental conditions.
[Result] Experimental findings demonstrate that FreqFusion-YOLOv8 achieves absolute improvements of 1.4% and 1.3% in mAP50 and mAP50-95 metrics respectively over the baseline YOLOv8n, without imposing additional computational overhead or extending inference latency. The model exhibits pronounced advantages in detecting narrow, low-contrast cracks and fine-grained defects (e.g., potholes), outperforming state-of-the-art one-stage detectors (YOLOv8, YOLOv9, YOLOv10) and two-stage frameworks (Faster R-CNN, Cascade R-CNN, Sparse R-CNN). Visual diagnostics further validate the efficacy of FreqFusion in enhancing edge definition and maintaining structural coherence, particularly under challenging illumination variations and heterogeneous surface textures.
[Conclusion] FreqFusion-YOLOv8 establishes an efficient and scalable paradigm for intelligent road crack detection. By synergizing frequency-domain feature integration with a compact detector architecture, the approach achieves superior accuracy-efficiency tradeoffs. Beyond advancing YOLO-based crack detection robustness, this methodology offers a transferable framework for monitoring fine-structure defects in civil infrastructure systems and autonomous road inspection applications.

[Objective] Against the backdrop of growing demand for scale-free map products in next-generation foundational surveying and mapping, automated generalization of fundamental areal geo-entities emerges as a critical technological bottleneck. Contemporary methods for areal feature generalization exhibit persistent limitations—including inadequate local geometric fidelity and imbalanced area proportions across global land categories—that risk distorting geographic structural integrity through naive merging operations. This paper introduces an area-balanced generalization algorithm for foundational areal geo-entities, designed to maintain geometric morphology and spatial adjacency relationships while ensuring relative balance in area redistribution among different entity classes before and after generalization.
[Method] The proposed model achieves automated cartographic generalization through three integrated processes: (1) semantic-aware preprocessing, (2) topological relationship maintenance, and (3) multi-level area optimization. First, differential processing protocols address specialized geo-entity types: For building clusters, right-angle features are preserved while redundant boundaries undergo simplification; only entities exceeding minimum area thresholds post-merging are retained. Slender road networks undergo centerline extraction via constrained Delaunay triangulation (CDT), followed by subdivision cutting along constructed guide curves to prevent boundary disruption of adjacent features. Second, a dynamic topological model is established using spatial coding indices that continuously update shared boundaries and adjacency edge information during merging operations. Third, a hierarchical area-balancing strategy governs merging sequences: (i) Homogeneous entity merging takes precedence to preserve semantic coherence; (ii) Land categories are classified into Type I (low area variation) and Type II (high area variation) based on projected change metrics; (iii) An optimization control function minimizes cumulative area deviation across Type II categories through constrained linear programming.
[Result] Validation employed geospatial datasets from two Guangxi pilot regions—Sanjiang Dong Autonomous County and Nanning Metropolitan Area—to generate multi-scale map products spanning 1︰10000 to 1︰250000. Quantitative analysis confirms that all areal geo-entities satisfy prescribed area change rate specifications (<±5% deviation), with no observable precision degradation during progressive generalization. The model has been operationalized in Guangxi's production system for scale-free foundational mapping, generating certified outputs meeting national cartographic standards.
[Conclusion] This framework transcends the unilateral optimization constraints of conventional approaches—which prioritize either local morphology preservation or global area balance—by establishing a dual-constraint mechanism integrating “topological fidelity maintenance” and “area equilibrium control.” It successfully achieves synergistic regulation of: (1) local geometric integrity, (2) global land category proportionality, and (3) continuous spatial adjacency relationships. These capabilities fully satisfy the stringent technical requirements of scale-free mapping in modern foundational surveying, providing robust data infrastructure support for land use statistics, ecological reserve delineation, and quota allocation for urban-rural construction projects.

[Objective] The rapid integration of personalized maps into geographic information systems and location-based services necessitates optimized design strategies to mitigate cognitive load and enhance visual search efficiency. Contemporary approaches lack a systematic, cognition-informed framework for understanding how key cartographic attributes (e.g., dimensionality, feature salience) interact with user-specific variables such as prior spatial knowledge. Addressing this gap, the present study investigates the combined effects of map familiarity, dimensionality (2D vs. 3D isometric), and building height salience on visual search performance. Grounded in cognitive load theory and predictive coding models of visual attention, this research disentangles top-down versus bottom-up processing dynamics during map interpretation, establishing an empirical foundation for user-adaptive cartographic design.
[Method] A controlled within-subjects experiment enrolled 21 university students (normal or corrected-to-normal vision). Employing a 2 (Familiarity: familiar campus vs. unfamiliar campus) × 2 (Dimensionality: 2D planar vs. 3D isometric) × 2 (Target salience: high vs. low) full-factorial design, participants completed two computerized tasks across separate sessions one week apart: (1) Visual search task: rapid localization/clicking of target buildings; (2) Dot-probe task: quantification of early attentional orienting speed post-map onset.Dependent measures included search time (ms), first-click accuracy (%), and attentional response time (ms). Stimulus presentation order and trial sequences were fully counterbalanced and randomized.
[Results] Three-way repeated-measures ANOVA revealed: Main effect of dimensionality: 2D maps significantly outperformed 3D maps in search speed [F(1,20)=11.48, p=0.003, ηp2=0.365], accuracy [F(1,20)=6.06, p=0.023, ηp2=0.232], and attentional engagement speed [F(1,20)=12.70, p=0.002, ηp2=0.388]. Counterintuitive familiarity effect: Unfamiliar maps elicited faster searches than familiar maps [F(1,20)=6.51, p=.019, ηp2=0.245]. Interaction effects: Salience did not independently influence search time but moderated relationships between familiarity and dimensionality through significant two-way and three-way interactions. Simple effects analysis demonstrated: In familiar contexts, low-salience targets in 2D maps yielded fastest detection; In unfamiliar contexts, high-salience targets accelerated searches. Salience primarily enhanced late-stage target identification rather than initial attentional capture, with pronounced effects on accuracy contingent upon scene context.
[Conclusion] Optimal map design exhibits strong context-dependency. While 2D representations universally reduce cognitive load, visual search efficiency emerges from dynamic interplay between user expertise and visual properties. Familiarity induces a dual-stage “guide-and-verify” strategy that may prolong searches, whereas salience functions as a conditional facilitator. These findings establish a cognitive psychology basis for adaptive personalized mapping systems, advocating dynamic adjustments aligning user states and task requirements to maximize information processing efficiency.

[Objective] Amidst the escalating urgency of global climate change, the precise assessment and analysis of carbon emissions have ascended to a paramount research priority within climate science and environmental management. For policymakers and governmental agencies, a comprehensive understanding of the spatiotemporal dynamics of carbon emissions, alongside the socioeconomic and environmental determinants that govern them, is indispensable for formulating targeted mitigation strategies and enacting effective carbon reduction policies. Nevertheless, the accurate quantification of carbon emissions presents significant challenges; conventional methodologies, predominantly reliant on energy consumption statistics, are constrained by their limited scope, coarse spatial detail, and inadequacy in capturing localized emission patterns.
[Method] To surmount these limitations, this study employs a multi-source data integration framework designed to enhance the precision and robustness of carbon emission estimations. Specifically, we synergistically combine nighttime light remote sensing data, population distribution datasets, land use maps, socioeconomic indicators, and traditional energy consumption statistics within a unified predictive model. This integrated approach, drawing upon both physical and social dimensions, yields annual, high-resolution (1 km×1 km) gridded maps of carbon emissions across the Yangtze River economic belt from 2014 to 2020. Subsequently, these detailed gridded datasets undergo a suite of analyses, including time-series analysis to identify temporal trends, spatial autocorrelation to evaluate clustering phenomena, and geographic detectors to discern the primary drivers underpinning the observed spatial heterogeneity in emissions.
[Result] The analytical outcomes reveal profound spatial heterogeneity in carbon emissions throughout the Yangtze River economic belt, manifesting as a distinct east-high, west-low gradient accompanied by significant spatial clustering. Downstream provinces emerge as the principal contributors, collectively accounting for over 50% of the region's total emissions. Furthermore, substantial disparities exist among land use types; impervious surfaces exhibit the highest per-unit-area emissions, peaking at approximately 25500 t/km², whereas emissions from croplands and forested areas are markedly lower. An investigation into the driving forces confirms that regional gross domestic product (GDP) exerts the most potent and consistent influence on the spatial distribution of carbon emissions, thereby underscoring the strong coupling between economic activity and emission intensity.
[Conclusion] By leveraging multi-source data and applying a comprehensive suite of spatial and statistical analyses, this study successfully mitigates the inherent limitations associated with single-source, coarse-resolution carbon emission estimation. The findings not only provide a refined characterization of the spatiotemporal dynamics of carbon emissions in the Yangtze River economic belt but also establish a novel methodological framework with broad applicability for carbon emission research in other large-scale regions.

[Objective] Collapsible loess regions, characterized by loose soil structures and substantial deformation upon wetting, exhibit heightened susceptibility to geological hazards such as landslides and collapses. These disasters critically threaten the safety and stability of oil/gas pipelines traversing such terrain. To improve disaster prevention and risk management in pipeline engineering, this study investigates the spatiotemporal evolution patterns and dominant driving mechanisms of geological hazards along pipelines in typical collapsible loess zones of the Loess Plateau.
[Method] A representative pipeline segment in a collapsible loess area was selected for investigation. Using high-resolution GF-1 and GF-2 remote sensing imagery from 2015, 2020, and 2024, multi-temporal geological hazard data were extracted via visual interpretation validated by field surveys. Temporal trends and spatial distribution shifts in hazard occurrences were analyzed through time-series comparison and kernel density estimation. Additionally, a geographical detector model quantified the explanatory power (q-value) of key environmental and climatic factors—including precipitation, NDVI, slope, curvature, and aspect—on hazard spatial differentiation.
[Result] Landslides and collapses dominate the study area's hazard landscape. Hazard counts increased from 2015 to 2020, followed by marginal decline by 2024, revealing a distinct “rise-peak-decline” temporal trajectory. Spatial migration occurred progressively south-to-north, demonstrating stage-dependent evolution. Geographical detector analysis identified quarterly precipitation (q = 0.134) as the primary climatic driver. NDVI (q = 0.11) and slope (q = 0.081) emerged as significant environmental regulators, while curvature (q = 0.045) and aspect (q = 0.007) showed negligible influence. Results underscore the synergistic impact of climate variability and topographic conditions on hazard development in collapsible loess regions.
[Conclusion] This study establishes an integrated framework combining multi-temporal remote sensing, field validation, and quantitative spatial analysis. The approach effectively characterizes temporal dynamics and spatial heterogeneity of pipeline-related geological hazards in collapsible loess zones. Findings provide a scientific basis for dynamic monitoring, risk assessment, and engineering planning of oil/gas pipelines in analogous loess terrains. Future research should integrate multi-source environmental datasets and high-frequency monitoring to deepen understanding of disaster dynamics under evolving climatic conditions.