摘要 : Increasing human disturbance (HD) is accelerating the erosion of ecosystem health (EH), posing severe and far-reaching threats to human well-being and ecosystem security. Although this issue has drawn global attention, existing studies predominantly focus on linear relationships between the two, while systematic exploration of nonlinear relationships and threshold effects between HD and EH has received limited attention. This knowledge gap hinders the formulation of precise strategies. This study employs an interpretable machine learning approach that combines an XGBoost model with SHAP analysis to systematically quantify the nonlinear impacts and threshold effects of multiple HD variables on the ecosystem health index (EHI) in the Taihu Basin from 2000 to 2020. A composite human disturbance index (HDI) is further constructed, and piecewise linear regression is used to identify critical thresholds in the HDI–EHI relationship. The results indicate that: (1) Land use intensity, population density, and night light are the primary driving variables, with SHAP contribution rates of 55.90%, 21.79% and 13.72%, respectively; (2) EHI responds to land use intensity, and population density in an inverted U-shaped manner, with recommended thresholds of land use intensity <0.7 and population density < 0.8, whereas night light has a monotonically negative effect with a critical threshold at 0.6; (3) The study reveals critical inflection points of impact of HDI on EHI, categorizing them into four stages: moderate promotion, critical decline, sustained resilience, and excessive collapse, presenting typical threshold transition characteristics, and shows that EHI declines sharply once HDI exceeds approximately 0.666. Based on these findings, we propose targeted planning and management strategies at different levels, emphasizing the importance of "pre-threshold proactive intervention" in preventive ecological planning and management. This research provides a new perspective for understanding complex social-ecological system relationships and offers robust support for achieving Sustainable Development Goals.

摘要 : Reinforcement learning (RL) is increasingly applied to optimize the control of residential heating, ventilation, and air conditioning (HVAC) systems. But some important issues still limit RL's practical application in residential energy optimization control, such as the unstable early-stage exploration, the requirement of massive training data, and long training duration. To address these issues, we propose a two-stage expert-guided deep deterministic policy gradient (DDPG) control method. The proposed method uses generative adversarial imitation learning to fit the expert networks based on expert experience in the offline phase. In the online phase, we integrate expert networks with DDPG, which interacts with the environment to further optimize the policy. The dual signal design ensures the agent receives both consistent positive imitation and environment guidance for stable and advanced control. The proposed method can effectively manage the dynamic and complex nonlinearities to reduce energy consumption, temperature violations and economic cost in the HVAC control problem. Experimental results show that, compared with rule-based control, the proposed method reduces energy consumption, temperature violations, and total cost by approximately 15%, 61%, and 23%, respectively. Moreover, compared with other advanced learning-based methods, it reduces e nergy consumption, temperature violations, and total cost by about 8%, 20%, and 11%, respectively. Moreover, the proposed method achieves superior performance in two additional thermal test environments.

摘要 : Infrastructures provide essential services but demand massive material inputs and generate substantial solid waste and embodied carbon emissions. Circular economy strategies offer a promising mitigation pathway by promoting infrastructure material circularity. The promotion requires understanding infrastructure material stock and flow, which depends on material intensity coefficient (MIC)-based stock-flow quantification. Yet, reliable infrastructure MICs remain scarce in existing literature. This study develops a standardized framework for compiling the real bills of quantities (BoQs) to derive MICs. Applying the framework in 181 infrastructure projects in China derived the robust MIC distributions of roads, railways, pipelines, cable systems, and flood control levees across distinct engineering grades. Also, material composition analysis indicates a "concrete-base aggregate-steel" dominance across these infrastructure cohorts, with average shares of approximately 61.5% concrete, 22.4% base aggregates, and 6.1% steel. Furthermore, comparative analysis reveals that previous studies underestimate infrastructure material intensities by up to an order of magnitude, due to the available MICs omitting structurally intensive components (e.g., viaduct structures and tunnel structures) and certain materials (e.g., asphalt). This research contributes a more reliable and extensive MIC dataset for China's infrastructures, while demonstrating a standardized BoQ-based MIC quantification framework. The MIC dataset provides empirically grounded references for infrastructure material stock-flow and embodied environmental impact assessment, and contributes robust data to support infrastructure circularity promotion.

摘要 : Historic urban areas (HUAs) are visually and culturally sensitive environments where blue-green infrastructure (BGI) plays an increasingly important role in shaping spatial identity and environmental quality. While BGI's ecological functions are well documented, its influence on human visual perception, particularly within HUAs, remains largely unexplored. Addressing this gap, this paper proposes an integrative framework to assess how BGI affects visual experiences in heritage contexts, bridging methodological, perceptual, and user-group dimensions. By combining UAV-based photogrammetry with a three-layered perception model, the research integrates spatial analysis and empirical methods across seeing (eye-tracking), feeling (questionnaire), and understanding (interviews) layers. Street-level BGI exposure was spatially quantified and used to inform perception experiments involving both expert and general public groups. This multi-methodological, multi-layered, cross-group approach extends existing research by providing a comprehensive examination of BGI's visual impact at different cognitive levels, particularly within historic settings. Findings reveal that BGI enhances perceptual diversity, visual preference evaluation, and cognitive engagement across both groups. Although it may slightly divert attention from dominant heritage features, BGI fosters broader visual exploration and higher environmental ratings. Experts interpret BGI through more systemic and functional perspectives, while the public emphasizes emotional, aesthetic, and recreational values. Overall, this study presents a replicable framework integrating digital spatial modeling with layered perception analysis, offering new insights for evaluating and enhancing visual environments in HUAs. It supports more inclusive visual assessments and provides a basis for informed planning and selective design interventions in heritage contexts.

摘要 : Urban vitality is widely recognized as a key criteria of urban quality, yet achieving a real-time evaluation of urban vitality and linking it with the effectiveness of urban renewal strategies presents ongoing challenges in policy and planning decision-making. To fill in this gap, this study developed an integrated evaluation framework and applied it to assess urban renewal projects in Suzhou through three quantitative dimensions: spatial behavior patterns, social perception dynamics, and policy intervention impacts. Key findings include: (1) In old communities, parking and structural upgrades significantly improve urban vitality (coefficients 0.127 and 0.099). (2) Public-facility renewal shows strong gains from traffic improvements (0.208), while excessive leisure space reduces effectiveness (− 0.299). (3) Cultural identity, lighting, and service facilities generate notable spatial spillover effects, underscoring multi-intervention synergies in mixed-use areas. The effectiveness of urban renewal initiatives is shaped by joint efforts of multiple interventions, especially in mixed-functional zones with residential, commercial, and public facilities. It is expected that this pioneering framework would support decision-making in future urban planning and design from a more scientific and data-driven perspective.

摘要 : In HVAC systems of large buildings, particularly in regions with significant cooling demands due to hot climates, chillers account for the bulk of energy consumption. Their operation is influenced by a range of factors, including both indoor conditions and outdoor environmental factors. Consequently, optimizing chiller control policies remains a central challenge in building energy conservation. Consequently, optimizing chiller control policies remains a central challenge in building energy conservation, especially in environments where there is a high demand for cooling. To date, few study has combined large language models (LLMs) with domain knowledge for chiller control. This paper proposes a Retrieval-Augmented Cooperative LLM-based Control Framework (Co-LLM). The framework constructs a chiller-specific knowledge base from industry standards, expert experiences, and operational logs. High-quality prior policies are retrieved via a Retrieval-Augmented Generation (RAG) mechanism, after which a Selective Recalibration module locally adjusts only those sub-policies that are strongly affected by external disturbances. Finally, a RAG-tuned Exploration performs multi-round optimization using a composite energy-comfort metric. A field study on the central HVAC system of a tertiary hospital in Taizhou shows that Co-LLM yields a 1.9% energy-saving gain over the traditional expert policy and achieves higher system COP than all other baselines, outperforming mainstream control methods. In particular, the retrieval-augmented mechanism effectively suppresses 92.7% comfort plunges and significantly enhances the robustness of the control system.

摘要 : Underground public space is a reserve of national spatial resources with development potential under the concept of stock development, and it is also an important component of the urban public space system. However, compared with above ground public space, underground public space involves more development entities, making it difficult to effectively coordinate and form a high-quality public space system. The flexible reward system, multi-party negotiation and discussion platform, and other multi-party collaborative mechanisms can mobilize the government, development entities, users, and third parties to participate in the creation of public spaces, and can solve the problems faced by underground public space development. This study utilizes case study method to thoroughly explore the construction and operation process of a public space system centered on underground pedestrian spaces in Sapporo. The study analyzes the collaborative mechanisms and models of multiple entities throughout the entire cycle of underground public space planning, construction and operation, which focus on the collaboration between government, local developers, the third organizations and citizens. The results show that: The Sapporo case has the advantages of integrating multi-agent coordination goal, quantifying elastic incentive measures of underground public space and constructing multi-agent coordination network with organization as media, which can provide reference for the development of underground public space in China.

摘要 : Urban redevelopment has been widely implemented to address social, economic and environmental challenges. However, the geography of urban redevelopment, its underlying mechanisms and driving forces remain insufficiently understood, both theoretically and across different metropolitan scales. Here, to fill this gap, we analyze state-owned land transactions between 2012 and 2022 across 326 Chinese cities. We find that China's city hierarchy exerts a regulatory effect on urban redevelopment. Higher-ranked cities tend to prioritize government regulation over market forces, whereas lower-ranked cities rely more on market-oriented approaches, and this divergence explains varying redevelopment patterns across city hierarchies. Our findings also reflect a shift toward top–down urban governance aligned with central-government policy, with higher-tier cities leading in implementing diverse redevelopment practices to achieve economic and non-economic goals. This study advances our understanding of administrative power in urban governance and offers insights for developing tailored strategies across different city hierarchies.

摘要 : In the digital media era, geo-tagged social media (SM) has emerged as a powerful tool for understanding tourist behavior and managing destination image. This study examines how spatial quality influences buzz behavior—rapid, user-driven dissemination of destination information—through the lens of the Feedback, Sympathy, Identification, Participation, and Sharing (FSIPS) framework. Focusing on Suzhou's historic city core, we integrate ridge regression, deep learning-based text mining, and eXtreme Gradient Boosting (XGBoost) to model the relationship between built environment features and SM engagement metrics, including emotional expression and likes. Findings reveal that (1) buzz behavior, driven by emotionally resonant user-generated content, plays a critical role in shaping destination popularity' (2) spatial features such as park and catering density significantly enhance emotional responses and content diffusion' and (3) entertainment density shows a negative association with engagement, suggesting diminishing returns in over-commercialized zones. (4) Furthermore, the analysis uncovers non-linear interaction effects—e.g., the co-presence of green infrastructure and public transport density synergistically boosts perceptual response. This research contributes a theoretically grounded and data-driven framework for decoding the spatial triggers of word-of-mouth dynamics in tourism, offering actionable insights for planners aiming to enhance tourist experiences and manage spatial quality in heritage-rich urban settings.

摘要 : Green space equity is increasingly recognized as a critical environmental condition for healthy aging, yet existing research often overlooks how different green space attributes—accessibility and diversity—are associated with distinct dimensions of older adults' health. Limited attention has been paid to their nonlinear threshold effects or to the social pathways through which green spaces influence health outcomes. Using the United States county-level panel data from 2020 to 2023, this study integrates fixed-effects models, Extreme Gradient Boosting (XGBoost), and mediation analysis to examine the associations between green accessibility measured by the Two-Step Floating Catchment Area (2SFCA) method, and green diversity measured by the Shannon Index, on the general, physical, and mental health of older adults. Findings indicate that (1) higher green accessibility is associated with better general health, whereas green diversity shows a stronger association with physical health, reflecting its link to more heterogeneous ecosystem service environments. (2) Green accessibility demonstrates the threshold effect, in which the strength of association with health becomes steeper once accessibility approaches higher levels. (3) Green space equity is linked to health partly through social structures. Education clustering and marital stability mediate the associations with general health, while mental health appears to depend more on the social interaction opportunities embedded within green environments than on their physical attributes alone. The study proposes an integrated "physical environment–social structure–health outcome" framework and a threshold-oriented spatial intervention strategy, highlighting the need to prioritize improvements in green accessibility in underserved areas and prioritizing green diversity and age-friendly social functions where accessibility is already high. These findings offer evidence for designing inclusive, health-oriented urban environments for aging populations.