Natural resources, particularly forests, are vital for ecosystem regulation but face threats like wildfires, increasingly common in Sub-Saharan Africa (SSA). This study evaluates wildfire intensity in central Cameroon (Yoko and Nanga-Eboko) from 2003 to 2023 using geospatial techniques, based on burned area (MCD64A1) and active fire (MCD14DL and Near Real-Time, NRT) images classification. Remote sensing tools like Google Earth Engine (GEE), Moran’s I, and Getis-Ord [Formula: see text] hotspot analysis identified vulnerable areas. Results reveal fire intensities of 49.5–2651.7 ha, with densities of 0–10.89 fires/km2. Temporal trends show increasing burned areas but declining active fires, confirmed by Mann-Kendall and Sen’s slope tests. Anomalies highlight increases of 3.5 ha and two additional fires during specific periods. Fire-prone months are October, November, and December, with an average monthly burned area of 1629.70 ha. Spatial analyses reveal clustering of high-intensity zones, especially outside protected areas (Moran’s I = 0.009081; p < 0.001). These findings highlight the urgent need for improved fire mitigation strategies and landscape restoration to safeguard ecosystems.
In order to achieve Germany’s climate neutrality targets by 2045, the electrification of public transport is a significant lever for reducing greenhouse gas emissions in the transport sector. Dynamic Wireless Power Transfer (DWPT) technology, which enables inductive charging of electric buses while they are in motion, is a promising solution for reducing charging times and battery sizes. Its technical potential at the national level in Germany has not yet been assessed. This study presents a novel methodology to assess the municipal-level potential of DWPT in Germany using open-source geospatial and public transit schedule data. By applying a shortest-path routing algorithm to bus stops, we estimate bus routes, and through recursive optimization techniques, the study identifies the most productive locations for Wireless Charging Lanes (WCL) in 696 German municipalities with over 20,000 inhabitants. The focus of this study is methodological scalability and data applicability rather than algorithmic optimality guarantees. The analysis estimates transferable energy potentials for WCLs of 1000 m, 2000 m, and 5000 m lengths and evaluates their impact on operating emissions and battery sizing. Results show that cities with centralized bus networks – such as Wiesbaden and Aachen – exhibit the highest DWPT potential. While the overall national operating emission reduction potential is low, certain municipalities demonstrate comparatively high local benefits. Furthermore, case studies using real vehicle schedules from Balingen suggest that DWPT can significantly reduce required battery capacities, especially in scenarios with limited downtime. This research highlights the value of combining open data and open-source tools for energy system planning and provides a replicable framework for assessing DWPT infrastructure deployment in other regions.
Accurately estimating ambient population density, which refers to the number of people present in a location at a specific time, is critical for effective urban planning, disaster response, and infrastructure management. However, most existing methods rely on proprietary data sources such as GPS traces or mobile phone records, limiting both accessibility and reproducibility. This study demonstrated that the spatial density of cellular base stations, available as an open infrastructure dataset, can serve as an effective proxy for ambient population density. Additionally, we introduce a practical method for identifying high-demand areas during daytime using only open data, which supports more accurate population estimation without the need for mobility logs or commercial datasets. Using Tokyo Metropolis as a case study, we evaluated the correlation between the cellular base station density, derived from the open dataset OpenCelliD, and ground-truth ambient population data. The results indicate that daytime correlations exceed 0.85 at the citywide level and reach up to 0.90 in estimated high-demand areas, which are higher than those obtained when using points of interest (POI) density as the comparative proxy. These findings confirm that cellular base station density is a reliable and reproducible proxy for ambient population density in urban environments. The proposed method provides a scalable and low-cost alternative for geospatial population analysis, particularly in contexts where proprietary data are unavailable.
High-resolution multi-temporal remote sensing imagery is essential for land-cover mapping and long-term environmental monitoring, but radiometric inconsistencies often reduce the reliability of cross-date comparison and change analysis. This problem is particularly severe in karst regions, where rugged terrain, terrain-induced shadowing, seasonal phenological variation, rock – vegetation spectral coupling, and atmospheric heterogeneity make pseudo-invariant feature (PIF) extraction difficult and weaken conventional relative radiometric normalization (RRN) methods. To address these challenges, we propose a multi-stage cascaded RRN framework, termed GCCM, which integrates Gaussian mixture model (GMM)-based probabilistic screening, dual-branch convolutional neural network (CNN)-based deep feature learning, and canonical correlation analysis (CCA) – multivariate alteration detection (MAD)-based statistical refinement. GCCM first screens candidate invariant regions, then learns robust spectral – spatial representations for reference and target images, and finally refines invariant-region selection in the learned feature space. Experiments were conducted on multi-temporal GF-1 and Sentinel-2 image pairs over a karst area in northern Guangxi, China, under large-temporal-span and pronounced seasonal-variation conditions. The results show that GCCM generally outperforms SAD-ED and IR-MAD in radiometric normalization accuracy and PIF quality, although improvements vary among datasets and sensors. In the challenging GF-1 large-temporal-span experiments, GCCM achieved a mean R2 of 0.800 and a mean RMSE of 0.007. In the seasonal-variation and Sentinel-2 experiments, most band-wise R2 values exceeded 0.900 with low RMSE values. Ablation analysis further showed that the complete GCCM achieved a mean R2 of 0.9370 and an RMSE of 0.0126 on four representative datasets. These results indicate that GCCM provides a robust and terrain-adaptive solution for RRN in complex karst environments.
Urban land use is increasingly shaped by evolving work modalities, particularly the rise of Work from Home (WFH) arrangements. This study introduces a hybrid Artificial Neural Network – Agent-Based Modeling (ANN – ABM) framework to simulate urban expansion under differentiated work behaviors in the Roanoke Area, Virginia, USA. By integrating temporally stratified household behavior data with spatial predictors, the model captures how heterogeneity between WFH agents and Work on Site (WOS) agents influences land-transition probabilities. The simulation framework comprises four ANN classifiers trained on 2011–2019 data, generating land-use suitability surfaces, behavioral preference maps, and local adaptability scores. These were fused into a dynamic land-use change model calibrated to reflect zoning constraints and infrastructure access. Land Use Changeability Scores (LCS) were computed to prioritize high-probability transition cells, guiding annual allocation from 2019 to 2032. Results show that incorporating behavioral agents increased classification accuracy from 86.7% to 89.8%, particularly improving predictions of medium- and high-intensity development. Contrary to expectations, WFH prevalence had a limited effect on urban sprawl, with densification remaining dominant – reflecting persistently low WFH ratios observed in the 2011–2019 baseline data. Projected land transitions emphasize fringe intensification and infill development, with forest and open-space classes in decline. This research contributes to GIScience by enhancing land-use simulation accuracy through the integration of behavioral agents into an intelligent, GIS-based ANN – ABM framework. By incorporating spatial predictors, zoning constraints, and agent heterogeneity, it offers an interpretable and transferable method to simulate urban expansion patterns under emerging work modalities.
Remote sensing using light detection and ranging (LiDAR) and hyperspectral imaging is widely applied in underground exploration for non-contact detection and mapping of subsurface features, particularly in mining environments. This study presents a methodology for alignment and 3D georeferencing using the tunnel coaxial hyperspectral scanning system (TCHSS), which employs a rotating sensor head integrating a hyperspectral camera, 2D LiDAR unit, and inertial measurement unit (IMU). By applying the principles of sensor-based orientation estimation in a local reference frame, exterior orientation parameters for each hyperspectral scan line were derived as a composition of IMU-based attitude data and LiDAR-based geometric data. A multi-stage protocol was implemented to correct system-induced geometric errors, including systematic lever-arm offsets and quantified boresight misalignments. The methodology was validated in an ilmenite (Fe-Ti-V) mine drift, where 20 scan sessions were registered into a continuous 3D hyperspectral point cloud, achieving a mean distance error of 86.44 mm relative to a reference 3D LiDAR model. Results demonstrate that the proposed approach generates geometrically coherent 3D models suitable for semi-quantitative applications, such as geological mapping and mine reconciliation, providing a technological foundation for fully automated underground exploration.
Human activities, such as mobility, manufacturing, and energy consumption, are major drivers of nitrogen dioxide (NO2) emissions. Understanding the relationship between human activities and NO2 emissions can help design effective air quality and sustainable planning policies. The premise is that human activities can be paused or stopped, which is not a foreseeable scenario. COVID-19 (Coronavirus disease 2019) pandemic has changed socioeconomic conditions, providing a unique opportunity to investigate the spatio-temporal response of NO2 to human activities. This study examined the spatio-temporal associations between daily human activities and NO2 concentrations across 255 counties in the U.S. using point-of-interest (POI) visit data. A Light Gradient-Boosting Machine (LightGBM) regression framework was employed to model NO2 concentrations primarily based on human activities data, with model interpretability enhanced through SHapley Additive exPlanations (SHAP) analysis. The NO2 simulation accuracy was improved by incorporating human activities data, with SHAP analysis confirming its consistently positive contributions. Sensitivity analysis across two perspectives were conducted to investigate the responses of NO2 to changes in human activities i.e. (i) in highly urbanized cities, the complete restrictions of human activities corresponded to a 28.2% decrease in NO2 concentrations, whereas an ambitious doubling of human activities was associated with a 20.3% rise, and (ii) before the pandemic, a reduction in visits corresponded to a 2.1% decrease and only a 0.6% decrease during the lockdown. These findings suggest that integrating high-resolution human activity data can enhance NO2 modeling compared to using only meteorological, topographical, road network, land-use, and building data. This study further highlights the differentiated sensitivities observed across urbanization levels and restriction periods, underscoring the potential of targeted sustainable travel management strategies in mitigating urban NO2 pollution.
China’s rapid urbanization has led to the emergence of urban villages (UVs), whose governance is crucial for urban quality and built environment management. However, conventional identification methods rely on high-resolution remote sensing images and extensive localized annotated data, inadequately supporting rapid surveys over large areas. Therefore, this study developed an efficient framework for UV identification. Initially, point of interest, real estate, and building footprint data were integrated to construct a comprehensive feature system of urban function, building properties, and 2D and 3D morphologies. Subsequently, using street blocks as spatial units, the GraphSAGE model achieved deep socioeconomic and morphological feature aggregation. Finally, transfer learning (TL) enabled efficient cross-regional knowledge transfer through domain adversarial training. Results revealed that the Southern Jiangsu region comprised 705 UVs in 2020, covering 167.49 km2, with obvious variations in number, scale, and distribution across cities. With Nanjing as the source domain, the GraphSAGE model achieved recognition accuracy of 93.0%. In target domains of Zhenjiang, Changzhou, Wuxi, and Suzhou, the fused TL model achieved accuracies of 79.3%, 87.3%, 83.3%, and 85.5%, respectively. Per-feature Wasserstein distance analysis attributed the cross-city performance disparity to morphological heterogeneity, social-sensing feature misalignment, and graph topology differences, with Zhenjiang exhibiting the largest domain shift. The model converged with only 50% labeled samples, reducing training time by approximately 40%. Ablation and comparative experiments further verified the efficacy of integrating social sensing and building footprint data. The proposed framework provides a technical reference for cross-region, small-sample, large-scale UV identification, supporting decision-making in scientific urban management and planning.
This paper presents the R package hgwrr, which implements Hierarchical and Geographically Weighted Regression (HGWR). HGWR draws on both Hierarchical Linear Modeling (HLM) and Geographically Weighted Regression (GWR) to enable the discovery and analysis of geographically varying relationships across multiple spatial scales – a topic of long-standing interest to geographers, environmental scientists, and others working with spatial data. Although GWR and HLM are widely used, GWR is not designed for hierarchical data of observations at one scale nested within groups at another more aggregate scale. And it is likely to fail when sample sizes vary significantly between groups, or several group-level variables are introduced. Meanwhile, HLM typically models regression relationships within a bounded, not continuous conceptualization of geographic space, which limits exploration of the geographic variation in those relationships to some “top-level” partitioning of the study area into discrete regions or similar. These restrictions are problematic when, for example, the goal is to model geographically varying relationships and to obtain geographically localized estimates from spatial hierarchical data. Such data are increasingly common and can arise when the detail of what is measured in the data is greater than the detail of where it is measured due to privacy protection, as an example. HGWR is tailored to these goals and data types. Using two-level data, where “samples” nest into “groups,” HGWR permits estimation of globally fixed and geographically varying regression coefficients, as well as inter-group random effects. It also assesses the statistical significance of the geographical variation revealed. Using the package’s simulated dataset and a Wuhan real estate dataset, we demonstrate the theory, usage, and advantages of HGWR, provide R implementation examples, and compare results with conventional GWR and HLM.
Cyclonic structure is visible on synthetic aperture radar (SAR) during a tropical cyclone (TC). Based on coupled ocean-atmosphere-wave-sediment transport (COAWST) model hindcasts, fused wind data from the Haiyang-2 (HY-2) constellation, the Advanced Scatterometer (ASCAT), the Soil Moisture Active-Passive (SMAP) radiometer, and the Advanced Microwave Scanning Radiometer-2 (AMSR2) are assimilated into the COAWST model using a three-dimensional variational (3D-Var) data assimilation scheme. Validation against along-track measurements gives root mean square errors (RMSEs) of 3.76 m/s for wind speed against the Stepped-Frequency Microwave Radiometer (SFMR), 0.12 m for sea surface height (SSH) against the HY-2 altimeter, and 0.93°C for sea surface temperature (SST) against National Data Buoy Center (NDBC) buoys, respectively. Model simulations reveal a spatially asymmetric SSH response to TCs, with a pronounced rise on the right of the storm track, weaker signals on the left and rear, and a radially outward decrease. Notably, a temporary negative SSH anomaly can occur near the storm center when the TC undergoes a recent directional change, reflecting the dominant influence of wind-driven upwelling along the prior trajectory. Furthermore, the peak sea surface height anomaly (SSHA) typically lags the maximum wind speed by 12 to 48 h. In addition, this study evaluated the SAR backscatter coefficient (σ0) response to key oceanic parameters during TCs. σ0 correlated positively with sea surface wind speed and SSH, with correlation coefficients (COR) of 0.79 and 0.73. Partial correlation yields COR of 0.51 for wind speed and 0.50 for SSH, confirming independent relationships, indicating sea surface roughness is primarily wind-driven yet positively linked to SSHA (wind-induced convergence/divergence). Furthermore, the study demonstrates a notable response of σ0 to ocean-atmosphere stratification intensity, particularly in relation to the Brunt-Väisälä frequency (N). A sharp increase in σ0 is observed under low N conditions, which may be associated with enhanced vertical mixing.
Oriented object detection in optical remote sensing faces significant challenges due to extreme scale variations, arbitrary target orientations, and highly heterogeneous backgrounds. While pursuing higher accuracy often leads to prohibitive computational costs through complex explicit-fusion architectures, generic structural re-parameterization methods fail to capture the unique spatial characteristics of geo-spatial targets. To bridge this gap, we propose an adaptive re-parameterized method for oriented object detection (AROOD). Unlike conventional approaches, AROOD implicitly enhances the network’s capacity to fit rotationally robust spatial priors and channel-wise dependencies into a unified representation through a novel structural fusion strategy. This mechanism allows the model to adaptively optimize geo-spatial prior knowledge during training and structurally merge these enhancements for zero-cost inference. Extensive experiments on the dataset for object detection in aerial images (DOTA) and FAIR1M (fine-grained object recognition in high-resolution remote sensing imagery) datasets demonstrate that AROOD effectively disentangles highly aliased spatial features, boosting the mean average precision (mAP) of widely used and competitive detectors (e.g. RetinaNet and LSKNet) by up to 3% without increasing parameter count or inference latency. Furthermore, the proposed localized training strategy exhibits superior computational efficiency compared to traditional incremental training.
Mangroves are crucial to coastal ecosystem stability and blue-carbon sequestration. Timely identification of current growth hotspots is essential for effective management and protection, which typically requires a large amount of multi-temporal high-resolution image series. In this study, we are aiming to explore the potential of identifying growth hotspots of mangrove using single-temporal high-resolution remote sensing imagery. A straightforward yet effective machine learning method is employed, with the core idea of modeling the indirect relationship between mangrove growth rates and single-temporal image-derived features. We first derive a reference indicator for hotspot intensity by quantifying mangrove changes over multiple years, namely the Mangrove Growth Hotspots Index (MGHI). Second, we compute features from a single high-resolution satellite image, including band-wise spectral features, spectral indices, and texture descriptors, and further incorporate geographical constraints to account for the characteristic expansion of mangroves from edge areas toward tidal flats. Three machine learning algorithms were employed to quantify the relations between image-derived features and reference MGHI, including random forest, eXtreme Gradient Boosting (XGBoost), and Categorical Boosting (CatBoost). The best-performing XGBoost regressor achieves R2 = 0.780, indicating that single image-derived features explain a substantial proportion of the variance in MGHI. The corresponding Root Mean Square Error (RMSE) =1.513, Mean Absolute Error (MAE) = 0.605, and Mean Bias Error (MBE)=−0.007 quantify the typical magnitude of prediction errors, supporting accurate and reliable hotspot estimation. The visual comparison and analysis also indicate that mangrove edges with blurred boundaries and rough textures may have relatively high growth rates, and a greater probability of being growth hotspots. Thus, we advocate for prioritizing attention to these areas to underpin the restoration and sustainable conservation of mangroves.
Ecological water replenishment (EWR) can alleviate groundwater depletion and related land subsidence caused by overexploitation. In the Beijing Plain, China, the Miyun-Huairou-Shunyi (MHS) region experiences long-term groundwater overexploitation and recent EWR. This study investigates the spatiotemporal patterns associated with the effects of EWR on surface water, groundwater, and ground deformation. We built an integrated observation and analysis framework using multi-source datasets, including SAR-derived water coverage, groundwater levels, and InSAR-derived deformation. Independent component analysis (ICA) is employed to separate and extract differentiated deformation responses of individual strata from ground surface deformation, while k-means clustering is applied to identify the dominant spatiotemporal response patterns of surface water and groundwater. Results show that EWR expands river coverage, drives aquifer recharge through infiltration and lateral flow, and promotes groundwater recovery, producing stratified deformation: persistent shallow-layer uplift, gradual middle-layer rebound, and continued but decelerating deep-layer subsidence. Compared with the pronounced effects of EWR, extreme rainfall events induce only short-lived and minor responses. Meanwhile, continuous shallow groundwater rise substantially increases hydrostatic uplift pressures and spatial heterogeneity of deformation, threatening the anti-floating stability of foundation slabs and generating localized differential deformation hazards. These findings highlight EWR’s dual role in supporting hydrological restoration while introducing engineering challenges, offering scientific insights for groundwater management.
Accurate and efficient monitoring of surface coal mines’ distribution and change dynamics is essential for sustainable land resource management. However, existing studies have mainly focused on disturbances to surrounding vegetation, with relatively little attention to internal components and their dynamic evolution. To provide a more comprehensive assessment of mining-induced land impacts, this study develops a novel approach for mapping the extent, components, and dynamics of surface coal mines using multi-decadal Landsat time-series imagery. A Global and Local Composite Threshold Segmentation (GLCTS) is proposed to delineate mining extent, followed by a Multiple Hierarchical Index-based Classification (MHIC) to extract internal components. The relationship between extent changes and component succession is further revealed across the mining life cycle. Applied to three major energy-producing provinces in northern China, the method achieves an overall accuracy exceeding 90%. The results show that the accumulated destruction area reached 3811.09 km2, while reclamation covered 2628.55 km2 over the past four decades, with waste rock identified as the dominant component (58.47%). The findings reveal that internal components follow a regular pattern of succession during the mining process, uncovering the heterogeneous and dynamic characteristics of mining interiors and offering new insights into their spatiotemporal processes and life-cycle evolution.
Global context encoding is crucial for robust semantic segmentation, especially in complex multimodal scenarios. Previous methods struggle with the complexity of modeling inter- and intra-modality long-range dependencies. Moreover, focusing only on spatial-domain context aggregation often overlooks crucial local textural information. To overcome these limitations, we propose an efficient framework designed for comprehensive spatial-frequency domain feature synergy, termed DSMNet. This study establishes an efficient and globally aware foundation by leveraging the VMamba architecture as the dual-stream backbone for feature extraction. Subsequently, the global correlation module and the multi-direction correlation module are introduced to systematically model all necessary long-range dependencies. To enforce the learning of contextually global features, a global context loss is employed to explicitly constrain the established dependencies during training. Furthermore, to compensate for the overlooked local details, the frequency-compensated refinement module and frequency-synergy fusion module are integrated. This mechanism mines and integrates local information from the frequency domain, achieving an effective synergy between spatial global semantics and frequency local details. Extensive experiments on the MFNet and PST900 benchmarks demonstrate that DSMNet consistently achieves state-of-the-art performance in both segmentation accuracy and robustness. The source code is available at https://github.com/qiwenjjin/IR_SEG.
Continuous extraction-induced deformation from oil and gas fields can trigger destructive geological hazards, significantly threatening underground infrastructure safety. Therefore, reliable surface deformation prediction is crucial for early warning of potential risks in oil and gas fields. However, no existing studies focus on spatiotemporal prediction in oil and gas fields. Moreover, previous spatiotemporal prediction studies do not consider the complementary information among multi-directional deformations and their joint prediction. They also ignore the constraint of empirical physical model during data normalization. To address the above issues, this study presents a novel approach for spatiotemporal prediction of two-dimensional (EW and vertical) deformation in oil and gas fields by incorporating the Multi-dimensional Small Baseline Subset InSAR (MSBAS-InSAR) method and deep learning. Specifically, we propose a dual-branch coupled encoder-decoder spatiotemporal prediction model enhanced with a Multi-Scale Channel-Spatial Attention (MSCA) mechanism, which jointly learns from both deformation components. Additionally, we introduce a function-fitting normalization strategy based on a time decay model to better conform with geomechanical laws and improve prediction stability. In this paper, we tested the proposed model separately on three gas reservoirs within the Sebei gas field. Experimental results demonstrate that the proposed model achieves smaller errors and better predictive performance across the three gas reservoirs compared to several baseline models, indicating its generalization capability in different reservoir deformation scenarios. Ablation studies further confirm the soundness of the model design. This study offers a generalizable framework for predicting diverse reservoir deformation scenarios, helping to ensure safe production and early geohazard prevention.
In 2022, approximately 31,800 new HIV cases were reported in the United States, and an estimated 67% occurred among gay, bisexual, and other men who have sex with men (GBMSM). The literature indicates that gay bars and other GBMSM social venues often function as community gathering spaces and may reflect the spatial concentration of GBMSM populations and associated social and sexual activity patterns. These populations continue to experience disproportionate HIV burdens due to persistent structural and systemic factors. Although previous studies have examined venue-related contexts and HIV outcomes, less focus has been given to examining how the association between gay bar visitation and HIV prevalence varies across space using large-scale mobility data. This study investigated spatial patterns of visitation to gay bars across six major U.S. cities at the ZIP code level in 2022. To the best of our knowledge, this is among the first efforts to examine these patterns at such a fine geographic scale. Across all cities, visitation to gay bars was positively and statistically significantly associated with HIV prevalence (p < 0.001). Including gay bar visitation in global spatial models increased the coefficient of determination by approximately 4% to 13%, indicating that this measure provides additional explanatory value beyond socioeconomic and demographic characteristics.
Urban vitality (UV) is a crucial indicator of the prosperity, livability, and sustainability of cities, yet its complex and non-linear relationship with the urban environment (UE) remains underexplored. Therefore, taking Nanjing as the study area, this study integrated multi-source data of mobile phone signaling, POI, street view images (SVI), and building footprints, etc. First, SVI features were extracted by the DeepLab V3+ model to reflect human perception visually, and Technique for Order Preference by Similarity to Ideal Solution (TOPSIS) was employed to quantitatively assess UV from three dimensions of population activity, space quality, and zonal connection. Then, the Geographically Weighted Random Forest (GWRF) and Gaussian Constraint Line (GCL) methods were combined to explore the non-linear influences and threshold effects of 16 built-up and ecological environmental indices on UV. The results demonstrated that: (1) UV in Nanjing exhibited a clear gradient decline from the center to the periphery, with the highest UV in central regions such as Xuanwu and Qinhuai districts, accounting for over 20%, and the lowest UV in fringe regions such as Gaochun and Luhe districts. (2) All UE indices exhibited pronounced spatial heterogeneity of non-linear influences on UV by pseudo t-values from GWRF, where four indices (e.g. building height and transportation pollution) displayed predominantly positive effects and three indices (e.g. building quality and carbon emission) showed mainly negative effects. (3) Except for building age, the other 15 UE indices exhibited significant inverted-U threshold effects, with high reliability and robustness by acceptable R2 and χ2 values, while thresholds of building height, land development intensity, and land surface temperature were 14.54 m, 0.63, and 22.19°C, respectively. Further experiments showed high segmentation accuracy of SVI features with an F1-score larger than 0.68 to significantly improve UV assessment, and synergistic effects among UE indices. This proposed framework exploring the UE-UV non-linear relationship could help to realize scientific spatial planning and sustainable urban management.
Rapid urban expansion reshapes ecosystem services (ES) and their interactions, yet the structural responses of these interactions across urban expansion gradients remain insufficiently understood. Here, we investigate these dynamics in the Middle Reaches of the Yangtze River Urban Agglomeration (MRYRUA) in China through a combined framework of the urban expansion differentiation index (UEDI), integrated valuation of ecosystem services and trade-offs (InVEST), and network analysis. This framework enables a spatially explicit assessment of ES interaction changes at different urbanization stages. Our results indicate that ES interaction networks exhibit marked structural sensitivity to urban scale and expansion speed, as evidenced by 40% and 25% declines in network density and connectance. With rapid urbanization, ES networks shift toward a core–periphery structure, where carbon sequestration and water yield function as primary hubs. Redundancy analysis identifies UEDI as the dominant driver of ES network structural shift, accounting for 52.40% of the cumulative variance (p < 0.01). At the agglomeration level, the Poyang Lake Urban Agglomeration, characterized by slower expansion (UEDI: 0.90), maintains the most intricate ES network (density: 0.80; connectance: 0.78). Conversely, the Wuhan Urban Agglomeration, at a more advanced urban development stage (UEDI: 1.01), exhibits the lowest network connectance (0.55) and ES synergies. These findings highlight the destabilizing effects of advanced urbanization on ES interactions and provide insights for stage-specific urban planning and ecosystem management.
