Rapid urbanization and infrastructure expansion are fundamentally reshaping land surface thermal dynamics in Sub-Saharan African metropolises, yet the combined microclimatic impacts of simultaneous urban greening and corridor redevelopment remain understudied. This study systematically evaluated the spatiotemporal thermal and biophysical changes in Addis Ababa, Ethiopia, driven by the Corridor Development Projects between 1990 and 2025. Utilizing multi-temporal Landsat satellite imagery, we employed Random Forest machine learning classification alongside the Normalized Difference Vegetation Index (NDVI) and Normalized Difference Built-up Index (NDBI) to quantify land-use and land-cover (LULC) transformations, surface temperature variations, and ecological severity via the Urban Thermal Field Variance Index (UTFVI). Over the 35-year period, built-up areas expanded nearly fourfold from 14.8% to 53.3%, primarily encroaching upon agricultural lands, while net vegetation cover contracted from 13.8% to 8.1%. Consequently, mean Land Surface Temperature (LST) increased by 5.06 °C (from 29.43 °C to 34.49 °C), expanding severe ecological heat stress (UTFVI ≥ 0.02) to 48.0% of the municipal area, with Bole, Akaki-Kality, and Kirkos emerging as considerable vulnerability hotspots. Bivariate spatial correlations confirmed strong cooling effects driven by elevation (r = −0.60 to − 0.68) and vegetation (r = −0.57 to − 0.71), contrasting with intense surface heating from built-up density (r = 0.54 to 0.63). The surface urban heat island (SUHI) intensity for Addis Ababa increased consistently from 0.40 °C in 1990 to 1.31 °C in 2025, reflecting faster urban warming than rural areas and a strengthening SUHI effect at an estimated rate of 0.026 °C per year over the study period. These findings expose a vital socio-ecological trade-off: localized corridor greening alters microclimate structure but cannot fully counterbalance city-wide warming from impervious surface expansion. Integrating thermal comfort metrics and minimum vegetation cover ratios into spatial planning is essential for climate-resilient urban development.
Climate change poses significant challenges to Morocco, which is increasingly affected by desertification, recurrent droughts, rising temperatures, and water scarcity. These impacts threaten climate-sensitive sectors such as agriculture, water resources, energy, urban planning, and environmental management, highlighting the urgent need for innovative and data-driven resilience strategies. In this context, artificial intelligence has emerged as a promising tool for strengthening prediction, planning, risk management, adaptation, and mitigation. This review assesses how artificial intelligence, machine learning, and deep learning have been applied to climate resilience in Morocco across seven climate-sensitive sectors and identifies methodological, data, and deployment gaps. A structured search of Google Scholar, IEEE Xplore, ScienceDirect, and SpringerLink identified 80 eligible publications from 2020–2025, which were analysed by sector, model family, dataset and access type, geographic scale, evaluation metric, and stated or implicit research gap. The findings show a marked increase in research activity after 2022, concentrated particularly in agriculture, water resources, and climate-risk applications. Tree-based ensembles were prevalent for heterogeneous tabular and geospatial data, whereas neural models were used mainly for imagery and time-series forecasting. Public satellite and reanalysis products enabled broad coverage, but sparse ground observations, restricted institutional data, resolution mismatch, limited spatial transfer testing, and minimal use of explainable AI constrained reproducibility and operational uptake. Most applications supported adaptation, while mitigation remained less developed. By jointly connecting AI methods, data provenance, spatiotemporal integration, interpretability, sector-specific limitations, and policy readiness, this Morocco-specific cross-sector synthesis extends narrower domain reviews and provides an evidence-based roadmap for reproducible, locally generalisable, and trustworthy climate-AI systems.
While associations between PM2.5 exposure and respiratory morbidity are well documented, the temporal dynamics and non-linear nature of these effects across pollution levels remain poorly understood. This knowledge gap limits the development of timely and evidence-based mitigation strategies. This study applied the distributed lag model (DLM) and distributed lag non-linear model (DLNM) within a generalised additive model (GAM) framework to assess the relationship between short-term PM2.5 exposure and respiratory morbidity at four pollution levels in Nakhon Ratchasima, Thailand, during 2019–2021. PM2.5 exposure was positively associated with respiratory morbidity, with risks increasing in magnitude and persistence at higher pollution levels. At low PM2.5 concentrations, DLM and DLNM produced comparable instantaneous and peak risk estimates. However, substantial divergence emerged at higher pollution levels. At 75th and 95th percentiles of PM2.5 exposure, DLNM estimated higher risks (RR = 1.23; 95% CI: 1.17,1.28) and (RR = 1.21; 95% CI: 1.17,1.25), respectively than DLM (RR = 1.10; 95% CI: 1.07,1.14) and (RR = 1.15; 95% CI: 1.11,1.19), respectively. Likewise, the maximum risks estimated by DLNM were greater (RR = 1.39; 95% CI: 1.33,1.46) at lag 2; (RR = 1.34; 95% CI: 1.28,1.40) at lag 3 than DLM (RR = 1.19; 95% CI: 1.16,1.22) at lag 2 and (RR = 1.20; 95% CI: 1.15,1.25) at lag1, respectively. At the 95th percentile exposure level, DLNM projected a substantially higher social burden (86.76 million THB) than DLM (47.23 million THB) arose from the differences in estimated risks. These findings highlight the importance of appropriate model specification for accurate epidemiological inference and policy development.
Rapid urban expansion and sustained growth in road traffic in Prague have led to elevated nitrogen dioxide (NO2) concentrations in complex street-canyon settings, demanding accurate microscale dispersion modelling to guide effective health and policy interventions. This study evaluates the performance of two commonly applied urban dispersion models, ADMS-Urban and GRAL, driven by the 2024 hourly meteorological data, traffic emissions, and building geometries. Model outputs were evaluated against the matched 2024 monthly passive sampling data collected across five receptor locations in the Smichov district. Measured overall mean NO2 concentrations varied substantially across sites. ADMS-Urban exhibited marked site-independent bias, overestimating concentrations at background site Graficka (e.g., 32.3 versus 22 μg/m3 observed) while underestimating levels at traffic-dominated site Plzenska (40.3 versus 54.6 μg/m3). In contrast, GRAL consistently underestimated concentrations acrosss receptors but with smaller discrepancies (typically within 2.5–11.2 μg/m3). When evaluated across all paired receptor-month observations, GRAL demonstrated stronger statistical agreement, achieving higher correlation (e.g., r = 0.59 at Graficka receptor point) and lower root mean square error (6.09 μg/m3 at Gymnasium receptor point). Taylor diagram analysis and geospatial mapping further confirmed GRAL's improved representation of concentration gradients and localized hotspots, particularly within complex street-canyon geometries. By comparison, ADMS-Urban produced larger and less consistent deviations, especially along densely built traffic corridors. The findings underline the critical role of model selection and receptor-specific evaluation when assessing urban air quality in heterogeneous urban environments.
Fine particulate matter (PM2.5) poses a significant health risk to residents of Bangkok and its surrounding provinces, yet many do not consistently adopt protective behaviors, such as wearing masks or limiting outdoor activities, even when pollution levels are high. This “adoption gap” between awareness of a risk and acting on it is not well understood. This study examined this gap by developing and testing an integrated behavioral model combining the Protective Action Decision Model (PADM), which explains how people come to recognize and appraise an environmental hazard, and the Theory of Planned Behavior (TPB), which explains how attitudes, social pressures, and perceived control shape behavior. A cross-sectional study was conducted among 1214 adults across Bangkok, Nonthaburi, and Samut Prakan provinces from April to May 2025. Structural Equation Modeling (SEM) was used to examine the pathways among knowledge, perceived threat, attitudes, subjective norms, perceived behavioral control, and preventive behaviors. The integrated model showed excellent fit to the data (CFI = 0.955, TLI = 0.947, RMSEA = 0.056) and explained 21.3% of the variance in preventive behavior. Three findings stood out. First, perceived behavioral control was the strongest correlate of preventive behavior (β = 0.386, p < .001), more important than attitudes or social norms. Second, while perceived threat was positively associated with attitudes (β = 0.857) and subjective norms (β = 0.424), perceived threat itself showed a negative direct association with behavior (β = −0.209, p = .005), suggesting that fear about a health risk, without a corresponding sense that protective action will help, can be associated with avoidance rather than action. Third, knowledge about PM2.5 showed a positive direct association with behavior (β = 0.106, p < .001) but a negative association with perceived threat (β = −0.121, p < .001), indicating knowledge operates through multiple pathways. These findings suggest that public health messaging that emphasizes PM2.5 dangers without building residents' confidence and practical capacity to protect themselves may be insufficient and, in some cases, counterproductive for closing the gap between awareness and protective action in urban Southeast Asia.
This article critically intervenes in dominant public space paradigms in planning and design of cities in the United Arab Emirates (UAE), challenging the persistent reliance on top-down, large-scale planning logics. We argue that such approaches reproduce spatial exclusivity and socio-cultural homogeneity, reducing public life to a managed spectacle or passive consumption. In doing so, they sideline the lived relations and practices through which diverse urban publics engage with and creatively reconfigure spaces to come together. Moving beyond a peripheral view, we reframe pocket spaces as relational micropublics: vital sites for intercultural negotiation, affective placemaking, and inclusive urban vitality. This approach draws on four case profiles of urban pockets in Dubai and Sharjah, UAE, informed by experiential encounters and systematic observations across these sites.Through a situated analysis of everyday public spaces, the paper realigns multidisciplinary trajectories spanning cosmopolitan urban theory, spatial practice, and urban design and planning. An intersectional and relational lens advances an adaptive, context-sensitive approach to everyday urbanism attuned to the temporalities, proximities, and contingencies of superdiverse life. Building on this foundation, we develop the Urban Vitality for Inclusivity (UVI) framework, conceived as an exploratory and conceptual contribution rather than a fully operational planning tool. It makes a twofold contribution: thematically, by clustering UVIs as socially grounded relational phenomena; and spatially, by linking them to situated practices and adaptive micro-urban conditions. This heuristic matrix addresses the planning challenges of superdiverse cities, where universalist design tropes fall short and policy must navigate the plural, contingent conditions of everyday spaces to advance open-city sustainability agendas.
Sustaining household hygiene behaviours (HBs) after COVID-19 remains a public health challenge in rapidly urbanising African cities. Accordingly, this study investigates mental determinants associated with sustained practice of HBs beyond the pandemic context. This cross-sectional study examined mental models influencing HBs among 437 residents of urban Dodoma, Tanzania, selected through stratified random sampling. A structured questionnaire measured eight cognitive constructs: Confidence in Actions (CIA), Habit Formation (HF), Trust in Authorities (TIA), Perception of Risk (POR), Outcome Expectancies (OE), Social Responsibility (SR), Injunctive Norms (IN), and Descriptive Norms (DN), alongside five HBs: handwashing, hand sanitizing, mask wearing, physical distancing, and avoiding handshakes. Partial Least Squares Structural Equation Modelling was used. The measurement model demonstrated strong reliability and validity (Cronbach's α > 0.849; composite reliability >0.908; AVE > 0.768). The structural model explained 29.5% of HBs variance (R2 = 0.295; Q2 = 0.265). There was evidence that CIA (β = 0.225), HF (β = 0.153), TIA (β = 0.149), POR (β = 0.138), OE (β = 0.092), and SR (β = 0.086) were positively associated with HBs. There was weak evidence for an association between injunctive norms and HBs, but no evidence for an association between descriptive norms and HBs. Findings indicate that sustained HBs are primarily driven by internalised cognitive factors rather than external enforcement or social norms. Importance–Performance Map Analysis identified CIA and HF as priority intervention targets. Overall, the study provides theory-driven insights to inform context-sensitive strategies for improving long-term HBs and strengthening urban environmental health resilience.
Urban agricultural land in mature urban regions faces development pressure, with implications for both land availability and residents' accessibility to local food systems. However, limited attention has been paid to how land use transitions restructure accessibility outcomes during late-stage urbanization. Using Tokyo as a representative highly urbanized and densely populated case, this study develops a conceptual framework linking socio-economic drivers, institutional and demographic moderators, agricultural land transitions, and accessibility outcomes across two periods (1991/92–2006/07 and 2006/07–2021/22). The analysis integrates a land use transition matrix, a gravity-based accessibility model, spatial explanatory power detection, and path-based modeling to identify key drivers and mechanisms. Results show that agricultural land accessibility declined more sharply in the 2006/07–2021/22 period despite smaller absolute losses of agricultural land, suggesting that the remaining agricultural land had fallen below a level sufficient to buffer the effects of ongoing urban expansion. Notably, areas with dense farmer density and a high concentration of lands designated as “productive green land” exhibited stronger pressures for residential conversion. Population aging was associated with reduced responsiveness of land use decisions to both land price and farmer density. By linking land use transitions with accessibility outcomes, this study provides a framework for understanding how urban agricultural systems are restructured during late-stage urbanization.
This study proposes a Digital-Twin-enabled Spatial Decision Support System (DT-SDSS) for explainable and sustainability-aware urban traffic routing under dynamic and uncertain conditions. The framework integrates spatiotemporal data harmonization, machine-learning-based travel-time prediction, ontology-driven semantic reasoning, and controlled candidate-route evaluation within a closed analytical workflow. It is evaluated through a corridor-scale case study in western Tehran, characterized by heterogeneous traffic conditions and recurrent congestion. Historical observations collected from 3 April to 5 July 2024 produced 8439 causal ten-minute-ahead forecasting samples, which were partitioned chronologically into 6596 model-development samples, a 97-sample temporal gap, and an untouched 1746-sample final holdout. Seven forecasting approaches were evaluated under common temporal boundaries: persistence, historical time-slot mean, Random Forest Regression, a deterministic random-feature approximation to radial-basis-function ε-Support Vector Regression, a deterministic second-order XGBoost-style gradient-boosted regression-tree approximation, Long Short-Term Memory, and Bidirectional Long Short-Term Memory. Random Forest Regression achieved the lowest final-holdout error, with an MAE of 0.606 min, an RMSE of 1.493 min, a MAPE of 2.410%, and an R2 of 0.959. The approximate XGBoost-style model achieved an MAE of 0.957 min, while the LSTM and BiLSTM models achieved MAEs of 1.546 and 1.565 min, respectively. The SDSS interprets predicted traffic states together with weather, incidents, temporal context, road characteristics, and policy constraints to produce traceable route evaluations. Commercial navigation services were excluded from quantitative benchmarking; instead, the routing component was evaluated against Dijkstra using the same scenario-adjusted candidate-union graph. Fuel and CO₂ values are reported as controlled scenario estimates derived from an explicitly documented speed-dependent fuel function rather than as locally calibrated fleet measurements. The results demonstrate the feasibility of integrating reproducible predictive analytics, semantic reasoning, and sustainability-aware route evaluation within a local corridor-scale research prototype.
Parks play a crucial role in promoting human health. However, most studies on park visits rely on subjective constructs rather than objective measurements. We used the global positioning system to objectively determine the number of visits by older adults to 25 parks within a 1600 m buffer zone from the homes of 40 older adults residing in suburban areas of Japan. Multilevel negative binomial regression analysis showed a positive association between the number of park visits and park size, but no significant association between the number of park visits and home-to-park distance. These findings suggest that park size may be associated with older adults' park visits. Location data revealed that park visitors were concentrated in large grounds and on walking trails. These exploratory findings suggest that park size, large grounds, and walking paths may be relevant considerations when planning neighborhood environments that support older adults' health and well-being.
Urban green infrastructure (UGI) and their ecosystem services (ES) strengthen urban biodiversity and resilience—but are threatened by increasing drought and heat risks. This study applies the Drought and Heat Risk (DHR) Assessment Framework in an urban park in Plauen, Germany, using an indicator-based approach to assess multi-risks for selected ES. The risk system is first delineated by considering hazards, exposure, and vulnerability to define key endpoints and derive descriptors. Local stakeholders/experts in urban planning, UGI management, biodiversity conservation, water management, climate, meteorology, soil, and environmental science then appraise the system, select appropriate indicators, and define evaluation criteria with weights and thresholds. Risk indicators are calculated using station measurements, remote sensing, microclimate modeling, and GIS analysis. Multi-risks are evaluated with the TOPSIS method, aggregating risks for the provisioning, regulating, and cultural dimensions of ES. Empirical testing confirms that the framework is able to capture system complexity through interconnected multi-risk indicators with attributes from diverse tiers, while underscoring the need for flexible, multi-method approaches in the face of data limitations. The results, presented as sub-city spatial maps, offer decision makers valuable insights into the spatiotemporal risks affecting UGI's ES and support efforts to safeguard their benefits for both society and the environment.
Rapid urbanization increases outdoor thermal discomfort in residential areas, especially in areas with limited vegetation and dense urban forms, thereby intensifying the heat island effect. This study investigates how tree typologies, shape, density, height, trunk size, and spatial arrangement can enhance outdoor thermal comfort across different residential morphologies in Hebron, Palestine. Two small-scale residential settings were selected, and ENVI-met simulations were used to assess thermal performance in the case studies in Hebron's Mediterranean climate, after validation with field-collected data. The results for Case 1 (dense heart-shaped crowns) show a 2 °C reduction in operative temperature, a 5% increase in relative humidity (%RH), and a reduction in wind speed. PMV improved from 4 to 2.5. For Case 2 (semi-open outdoor area), results show greater improvements, with a 4 °C reduction in temperature, a 10% increase in %RH, and a 1 m/s decrease in wind speed. PMV decreased from 4.5 to 2.8. Finally, tree configurations significantly improved the microclimate, but they are not sufficient on their own. The study recommends combining vegetation with passive strategies, such as vertical gardens and shading devices, to improve thermal performance.
In the context of global climate warming, optimizing land spatial patterns to enhance ecosystem carbon sequestration offers a cost-effective and immediate strategy for carbon absorption. This approach not only mitigates the pressure of emission reductions but also secures a critical window for industrial decarbonization. While many studies optimize land spatial patterns by simulating land use scenarios and forecasting land demand, few integrate ecological red lines, control zones, and restricted areas into their models. Additionally, existing studies often fail to balance food security, ecological protection, and economic development. To address these gaps, it is essential to explore a broader range of potential land use scenarios to define the upper and lower limits of land use changes, facilitating multi-objective optimization analysis. This study takes Shandong Province as a case study, employing the PLUS and InVEST models to simulate five scenarios: Current Continuation (CC), Food Security (FS), Ecological Protection (EP), Urban Expansion (UE), and Comprehensive Optimization (CO). The objective is to analyze spatial patterns and spatiotemporal distribution characteristics of carbon sequestration under different scenarios by 2030. Key findings are as follows: (1) Ecologically sensitive and vulnerable areas in Shandong are primarily concentrated in the central, northeastern, and eastern mountainous and hilly regions. (2) Compared to 2020, land use structures exhibit significant changes across all five scenarios in 2030, with a notable expansion of construction land, particularly in the UE scenario. Carbon stocks increase under all scenarios, but the CO scenario maximizes carbon sequestration potential, achieving the optimal balance between food security, ecological protection, and economic development. (3) Future land use planning in Shandong should be guided by the spatial patterns derived from the CO scenario, with a strong emphasis on preserving ecologically sensitive and vulnerable regions in the central and northeastern mountainous areas. These findings provide theoretical support for land use management in Shandong Province and offer valuable insights for achieving national carbon peak and carbon neutrality goals.
Digital infrastructure usage in historic urban cores is shaped by a fine-grained interaction among commercial service quality, spatial permeability, heritage agglomeration, and environmental comfort. Using anonymized operator-side cellular records from Beijing's Capital Core Functional Area, this study constructs Digital Infrastructure Usage Intensity (DIUI) as a grid-level indicator that combines passive ambient population presence with active foreground cellular traffic. An interpretable modelling framework integrating XGBoost, SHAP, and geographically weighted regression is then used to identify dominant correlates, nonlinear association ranges, and spatially varying local relationships. The results indicate that commercial service quality and block openness are the most influential predictors, but their associations with DIUI are conditional rather than monotonic: commercial upgrading is associated with higher DIUI only when the surrounding morphology provides sufficient permeability and visibility, whereas overly enclosed or over-saturated settings weaken this relationship. Heritage concentration shows an inverted-U association with DIUI, suggesting that moderate agglomeration may support digital activity more effectively than excessive clustering. Spatially, transitional historic neighborhoods show stronger associations with commercial improvement than already saturated core areas, while street-level greenery appears to amplify digital activity in permeable settings. Overall, the results support a context-specific interpretation of spatial-digital coupling and caution against translating model-derived ranges into universal planning prescriptions without local validation.
Rapid urbanization degrades near-surface wind environments and microclimate regulation in southern China, whereas traditional Jiangnan gardens offer climate-responsive spatial models through integrated corridor–water–building systems. Yet how these configurations regulate wind–thermal conditions beyond summer heat remains unclear. This study examined the Humble Administrator's Garden in Suzhou during the cool-to-mild Qingming–Guyu period to clarify how spatial sequences organize airflow, radiant exposure, and thermal conditions. A 1:1 geometric model coupled steady RNG k–ε Reynolds-averaged Navier–Stokes simulation with Radiance-based radiation analysis and Universal Thermal Climate Index (UTCI) post-processing. Numerical checks and Pingjiang Road measurements assessed airflow performance (RMSE = 0.23 m s−1; NMSE = 0.09). Of 20,330 points, 13,937 cells supported strong-hierarchy response-surface modelling and leakage-controlled validation. The quadratic model reproduced the calculated UTCI field (R2 = 0.9961; RMSE = 0.215 °C) and remained accurate in 20 m block cross-validation (R2 = 0.996; RMSE = 0.219 °C), although buffered east–west transfer weakened (R2 = 0.975; RMSE = 0.542 °C). The robust statistical feature was a nonlinear conditional response rather than a stable directional wind × MRT interaction. Open water edges, lawns, and inflow-aligned gaps formed low-obstruction ventilation paths, whereas corridors, buildings, rockeries, and large-canopy-tree geometry redirected flow and generated localized wakes, shade, and shelter. These findings identify seasonal coordination between ventilation exposure and protected refuge as a key adaptive mechanism, providing a transferable, mechanism-based framework for climate-sensitive heritage management and contemporary open-space renewal.
In the context of rapid urbanization, heat island effects and microclimate degradation caused by poor ventilation have become important challenges for sustainable urban development. Taking the comprehensive functional area along the Qingnian Street metro corridor in Shenyang, China, as the study area, this study combines multi-season CFD simulations, urban block morphological indicators, the XGBoost machine learning model, and the SHAP interpretability method to systematically evaluate the influence mechanisms of block-scale morphological factors on the near-ground wind environment. The results show that the wind environment in the study area exhibits significant seasonal variation and spatial heterogeneity. The area of stagnant-wind zones follows the order of summer > winter > autumn > spring. Ventilation conditions are relatively weak in the high-density blocks in the northern part of the study area, while comfortable wind zones are mainly distributed along major roads and open spaces in the southern part. Further analysis indicates that enclosure degree, sky view factor, frontal area index, building density, building height, building volume density, green space ratio, and impervious surface ratio are the main morphological factors affecting the wind environment, and their effects show obvious seasonal sensitivity. Highly enclosed urban forms significantly suppress ventilation in summer, whereas the winter wind environment is more dependent on the guidance of airflow by road networks and spatial openness. The SHAP marginal effect results further reveal the nonlinear responses and threshold characteristics of key morphological factors. For example, when the enclosure degree is approximately 2, building density is approximately 0.3, sky view factor is above 0.85, and green space ratio is above 0.3, the effects of some indicators on the wind environment change significantly. This study verifies the interpretability and applicability of the XGBoost-SHAP method in urban wind environment modeling, reveals the nonlinear influence mechanism of urban block morphology on the wind environment, and provides quantitative evidence and methodological references for high-density urban block planning oriented toward ventilation optimization and microclimate improvement.
This study examines the geography of urban data centres (DCs) and develops a spatial strategy framework that explains the concentration of DCs and the infrastructural conditions that support these clusters. The paper addresses the lack of integrated, spatially specific site models by considering DC location as a decision-making problem influenced by power and network infrastructure, environmental aspects, and urban factors. Four metropolitan case studies: Paris (Île-de-France), Ashburn (Northern Virginia), Beijing, and Riyadh are examined as hubs in the centre network. The methodology combines DBSCAN clustering and kernel density estimation with a GIS-based multi-criteria suitability analysis. Six factors (power-grid substations, fibre hubs, telecom hubs, rivers/water sources, major roads and central business districts) are transformed and scaled into 1–5 suitability scores and integrated through a weighted overlay. Results show that DCs are strongly clustered rather than randomly distributed in all four cities, with distinct morphologies in each case. Clustered facilities consistently exhibit higher suitability scores than non-clustered sites (0.7–1.7 on a 1–5 scale). Regression models demonstrate strong correlations between cluster membership and suitability (R = 0.71–0.90) and contribute 50% to over 80% of the variance in suitability (R2 = 0.50–0.81, all p < 0.001), confirming the significant influence of proximity to high-capacity power and fibre infrastructure, enhanced by road access and urban demand. The framework provides planners and policymakers with a useful tool to identify current and prospective DC corridors, align grid and fibre investments with land-use planning, and predict the challenges associated with the rapid growth of digital infrastructure.
Efficient monitoring of urban water consumption is essential for sustainable resource management and early detection of distribution network irregularities. Single-method anomaly detection approaches often fail to capture the diversity of patterns present in real consumption data, such as leaks, meter malfunctions, abrupt demand shifts, and data inconsistencies. This study proposes an ensemble anomaly detection framework that combines complementary time series techniques, including decomposition, forecasting, density-based, and smoothing-based methods. The framework was applied to neighborhood-level daily water consumption data from Valencia, Spain, where six individual neighborhoods were selected to reflect distinct consumption patterns. Each time series was analyzed independently to preserve local consumption behavior. The ensemble approach demonstrated improved robustness in identifying diverse irregularities compared to individual methods, without requiring labeled training data. Results were explored through a geospatial visualization interface that enabled spatial interpretation of detected anomalies across neighborhoods. The proposed framework is scalable, interpretable, and adaptable to evolving consumption patterns, supporting data-driven decision-making for urban water management in smart city contexts.
With the acceleration of urbanization, the problem of air pollution has become increasingly prominent. The prediction of PM2.5 concentration has become an important issue in air quality management and public health research. However, the complex meteorological conditions and multi-site spatiotemporal coupling make it challenging for traditional data-driven models to fully capture both predictive performance and physical constraints. Therefore, we propose a multi-site hybrid framework that integrates BiLSTM and PINN, embedding the wind-driven advection transport, diffusion, and pollutant attenuation processes into the network training, and constructing PDE residual constraints in the form of finite differences. This model integrates multi-scale features and employs a fixed physical loss weight (λPDE = 0.01). A physically motivated loss formulation is used to improve the model's robustness under high-concentration pollution conditions. On this basis, a mixed source term structure is constructed, which includes linear contribution terms of CO, NO2, SO2, O3, and PM10, as well as nonlinear interaction terms such as CO–NO2 and PM10–O3. Moreover, the diffusion coefficient and pollution source parameters are jointly optimized as learnable physical parameters. The experimental results based on 46,754 observation samples from 5 air quality monitoring stations in Kunming City show that the proposed model achieved good prediction performance on the test set. The RMSE was 5.95μg/m3, the MAE was 4.36 μg/m3, and the R2 reached 0.8684, indicating that the model can effectively learn the spatiotemporal diffusion and pollution source contribution characteristics of PM2.5. This model provides a physics-constrained spatio-temporal framework for PM2.5 prediction with improved interpretability through physically meaningful parameters. Experimental results show that, in the 1-h-ahead forecasting task, the proposed model does not achieve the highest numerical accuracy compared with the simple persistence model and several data-driven baselines. However, it offers improved physical consistency and interpretability, providing a trade-off between predictive accuracy and physical interpretability.
Accurate quantification of urban carbon dioxide (CO2) emissions is essential for evaluating local mitigation strategies. Yet discrepancies can occur when comparing city emission inventories with flux estimates from measurement-based methods such as eddy covariance (EC). This study investigated whether urban vegetation can explain such inventory-flux discrepancies in Vienna, Austria. Residuals between downscaled, temporally resolved inventories and tall tower CO2 flux measurements generally exhibited a temporal pattern following seasonal and diel dynamics in vegetation CO2 uptake. To evaluate this potential influence, correlations of inventory-flux residuals with satellite-derived Leaf Area Index (LAI) and Fraction of Absorbed Photosynthetic Active Radiation (FAPAR) within flux footprints, as well as meteorological variables, were tested. Results revealed positive correlations with daytime data (ρ = 0.74 for LAI vs. relative residual), and negative correlations with nighttime data (ρ = −0.65 for LAI vs. absolute residual), suggesting a role of biogenic processes in urban CO2 fluxes, driven by daytime plant net assimilation and nighttime plant and soil respiration. Subsequently, a first estimate of net ecosystem exchange (NEE) was quantified using a semi-empirical urban vegetation flux model and directly compared with the observed residuals. Monte Carlo simulations (n = 3000) indicate that NEE remains comparatively small, resulting in only a marginally improved agreement between emission inventory and EC observations (R2 = 0.361 without NEE; R2 = 0.361–0.390). Although diurnal and seasonal variations in biogenic fluxes are noticeable, they are too small to reconcile the systematic divergence between EC observations and the downscaled inventory, pointing to additional sources of uncertainty.
