Artificial intelligence (AI) has rapidly advanced soil organic carbon (SOC) estimation by improving spatial resolution and predictive accuracy. However, whether these advances support reliable carbon stock accounting under complex environmental conditions remains insufficiently examined. This commentary argues that mountain systems provide a natural stress test for evaluating the robustness and applicability of AI-based SOC estimation. Characterized by steep climatic gradients, strong topographic control, variable soil thickness, geomorphic instability, and sampling bias, mountain environments amplify structural weaknesses that may remain hidden in more homogeneous landscapes. We identify three failure modes revealed under mountain conditions: regime-dependent shifts in predictor meaning that undermine transferability, scale mismatch between point prediction and stock aggregation, and weak process constraints that lead to physically implausible spatial patterns. Building on this diagnosis, we outline the core capabilities required for next-generation AI frameworks, including spatially structured representation learning, context-conditioned inference across environmental regimes, process-aware constraints, and explicit applicability-domain assessment. We further propose a mountain-tested benchmark that links these failure modes to cross-regime generalization, scale-and-depth coherence, process plausibility, uncertainty attribution, and applicability-domain mapping. By positioning mountains as diagnostic environments rather than merely challenging terrain, this perspective reframes AI-based SOC estimation from tool driven prediction toward defensible, process-aware carbon accounting.
Ecology is entering a data-rich era driven by rapid advances in sensor technologies and the expansion of environmental observations across spatial and temporal scales. In parallel, machine learning (ML) has been increasingly adopted as a powerful tool for analysing complex ecological systems, where interactions among organisms and their environments are often nonlinear, heterogeneous and scale-dependent. Over the past three decades, ML applications in ecology and global biology have expanded rapidly, supporting advances in pattern detection, prediction, monitoring and process understanding, while still facing major challenges in data quality, sampling bias, interpretability, uncertainty quantification and causal inference. Emerging developments in embodied Artificial Intelligence (AI), foundation models, large language models and intelligent agent systems now create new opportunities for ecological research by enabling adaptive observation, improved integration of heterogeneous data and more efficient analytical workflows. These advances are beginning to steer the field towards an AI-assisted research paradigm, in which ecologists supervise coordinated AI systems for data collection, integration, analysis and modelling, while providing interpretation, validation and iterative feedback. Realizing the full potential of such human-AI synergies in ecological studies will require closer integration between ecological knowledge and advanced AI methods, as well as sustained collaboration among ecologists, data scientists and technology developers.
The 15-minute city concept, which promotes urban liveability by ensuring that essential services are accessible within a short walking distance, has gained global attention as a sustainable alternative to car-dependent urban sprawl. While existing accessibility studies predominantly rely on map-based methods and non-pedestrian-focused mobility assessments, significant gaps remain in understanding pedestrian accessibility from a perceptual and experiential standpoint. This study addresses these gaps through a systematic review of 53 scholarly articles sourced from Web of Science, Scopus, and Google Scholar, critically examining the methodologies, metrics, techniques, and data inputs used to assess pedestrian accessibility in 15-minute city models. The review particularly focuses on data acquisition, tool digitalization, and methodological refinement. Findings highlight global trends in the implementation of pedestrian accessibility assessment, revealing variations in demographic representation, data diversity, and scale of analysis. The results show a strong reliance on geospatial data-driven approaches but a limited incorporation of perceived urban attributes, such as visual environments, travel impedance, and street-level experiences, in pedestrian-focused accessibility models. Furthermore, the study underscores the potential of integrating perceptual indicators into accessibility assessments to bridge the gap between urban infrastructure and pedestrian experience. By synthesizing advancements in data-driven urban analytics, this review contributes to the refinement of pedestrian accessibility measurement (PAM) frameworks, advocating for more inclusive, perception-based urban planning strategies.
Cultural ecosystem services (CESs) represent a crucial link in enhancing public well-being through the provision of ecological products. A scientific understanding of CESs characteristics at the national scale is essential for coordinating the integrated and sustainable development of society, economy, and nature, as well as for advancing ecological civilization. However, systematic evaluation methods that couple natural geography with socio-economic dynamics remain underdeveloped. This study, from a supply–demand perspective, constructs a CESs potential evaluation framework encompassing five key dimensions: historical, aesthetic, scientific, health-related, and recreational. By employing both supply–demand matching degree and supply–demand coordination degree models, the study conducts an empirical assessment of CESs across 581 national protected areas within 31 provincial-level administrative units in China. Key findings include: 1) The CESs supply capacity in 159 protected areas and the demand capacity in 187 protected areas exhibit statistically significant spatial clustering; 2) Provincial-level CESs supply potential exhibits multi-dimensional non-equilibrium characteristics, closely linked to differences in resource endowments and development strategies; 3) Demand potential displays a clear east–west gradient, with stronger demand concentrated in eastern regions; 4) Pronounced spatial mismatches exist between supply and demand, and substantial regional disparities are observed in coordination levels, highlighting the urgent need for differentiated governance strategies. This research offers a theoretical foundation and empirical evidence to support regionally tailored CESs governance and spatial optimization in protected areas.
How to enhance human well-being while maintaining ecosystem services represents a core concern in achieving the sustainable development goals. Efficiency and fairness are key dimensions during the conversion process from ecosystem services to human well‑being, but prioritizing one often comes at the expense of the other. Therefore, this study develops a comprehensive analytical framework to explore the global synergistic relationship between ecological well‑being conversion efficiency and fairness in the years of 2005, 2010, 2015, and 2020. The key findings are as follows: (1) Global human well-being increased by approximately 12 % whereas ecosystem services declined by approximately 9 % during the study period. High‑income countries exhibited high ecological well-being conversion efficiency but deficit-type fairness, while low‑income countries were trapped in a dual dilemma of low conversion efficiency and fairness; (2) The synergistic development of ecological well-being conversion efficiency and fairness shows an improving trend, with the annual average dynamic local and tele-coupling coordination degree (LTCCD) increasing steadily from 0.48 to 0.51; (3) The spatial correlation of dynamic LTCCD remains weak at the global scale, and social factors demonstrate stronger explanatory power for dynamic LTCCD than other influencing factors. This research provides a new perspective for understanding the interactions between ecosystem services and human well-being, and offers a valuable foundation for designing regionally differentiated policies aimed at enhancing ecological well-being outcomes and promoting equitable ecological well-being distribution.
Under climate change, ecosystems are increasingly exposed to multiple disturbances and stresses. Ecological engineering is widely regarded as an effective strategy for protecting and restoring ecosystems; however, whether such interventions enhance ecosystem resilience remains unclear. This study evaluates the impact of the Grain for Green Program (GFGP) on ecosystem resilience in northern China. Ecological engineering and policy timing were identified using long-term land-cover trajectories. Ecosystem resilience was quantified using lag-one autocorrelation (AR(1)) derived from detrended satellite-based vegetation index time series and evaluated using a staggered difference-in-differences approach. The results show divergent patterns among restoration types. Grassland restoration is associated with a broader spatial extent of declining AR(1) and a sustained post-treatment reduction, indicating stable and persistent resilience gains. In contrast, forest restoration exhibits a non-linear response. Although AR(1) initially decreases after implementation, it increases during the middle and later post-treatment periods, implying a potential long-term decline in resilience. In terms of overall effects, cropland-to-grassland conversion exhibits a statistically significant overall increase in resilience, whereas the average treatment effect of cropland-to-forest conversion is not statistically significant. These findings improve understanding of how large-scale ecological engineering impacts ecosystem resilience and provide evidence to inform more effective climate change mitigation and adaptation policies.
The spatio-temporal distribution of wildfire trends reveals spatially uneven changes in fire activity across the globe. Existing studies primarily rely on temporal models and climatic variables but ignore the spatial disparities of drivers. This study develops a spatio-temporal trend-cluster attribution (STCA) model to explore joint climate–fuel–human controls on global fire-trend patterns. STCA combines spatial trend clustering, optimal parameters-based geographical detector and grid-based temporal analysis to identify spatial hotspot and coldspot regions in burned-area trends and to attribute them to individual and interactive climate, fuel and human effects. The model is implemented on global burned area derived from the Moderate Resolution Imaging Spectroradiometer (MODIS, MCD64A1) for 2001–2025, with drivers of climate and fire weather, fuel and vegetation, and anthropogenic pressure. Globally, burned area declined by 1.28 % ± 0.35 % of its 2001–2025 mean per year, but the model identified strong regional disparities, including nine trend hotspot regions and six trend coldspot regions of burned-area trends. The spatio-temporal analysis demonstrates that fire weather dominates globally, whereas savanna fuel and anthropogenic pressure become comparably important within trend clusters. In addition, warming and drying are associated with increases in hotspots, while cooling and wetting are associated with declines in coldspots. Finally, climate drivers weaken in rising hotspots and fuel control strengthens in declining coldspots. STCA provides a scalable and reproducible approach for identifying climate–fuel–human drivers on spatially structured wildfire trends.
Resilience is a key property reflecting how ecosystems cope with stress, disturbances, and environmental change. Current estimates of remotely sensed resilience based on critical slowing down are subject to inconsistency, potentially leading to false warnings of ecosystem collapse. When it comes to remote sensing-based resilience indicators, the choice of vegetation indices remains a fundamental but unresolved issue. Here, we used time series data of remotely sensed vegetation structure (characterized by leaf area index, LAI) and function (characterized by solar-induced chlorophyll fluorescence, SIF) to compare the critical-slowing-down based resilience indicators with the indicators representing recovery rate upon disturbance events. We aim to quantify which vegetation metric (structural versus functional) provides more reliable resilience estimates, as reflected by higher consistency between these two classes of resilience indicators. Our results show that the reliability of SIF is, on average, more than twice that of LAI, and the area suitable for estimating resilience with SIF is 4.83 times higher. SIF-based estimates indicated that the number of pixels with declining resilience was 1.07 times higher than those with increasing resilience. In contrast, LAI-based estimates overestimate the proportion of pixels with declining resilience by 35.5 %. Our study reveals a mismatch between the recovery rates of vegetation structure and function, and supports the use of function-based metrics for resilience estimation. The proposed reliability assessment framework addresses inconsistencies in resilience estimates from the perspective of data source selection. We recommend prioritizing vegetation functional indicators in most scenarios. Employing appropriate vegetation indicators for resilience estimation will benefit global and regional sustainable ecosystem management and early warning systems.
Artificial Intelligence (AI) is increasingly recognized as both an enabler of and a risk to sustainable development. Yet research remains dominated by expert-driven assessments, with little attention to how the public perceives and discusses AI’s sustainability implications. This study examines concern-oriented public discourse about AI in relation to the United Nations Sustainable Development Goals (SDGs) by analyzing social media posts and developing expert-derived solutions. We collected and analyzed over 700,000 posts from X (formerly Twitter) spanning three years (2022–2025), applying natural language processing techniques including Fine-tuned BERT, Zero-Shot Classification, and Pre-trained XLNet. The analysis reveals pronounced imbalances in public attention across the 17 SDGs, with 52.8 % of AI-related discussions concentrated on SDG 9 (Industry, Innovation, and Infrastructure) and 11.8 % on SDG 16 (Peace, Justice, and Strong Institutions), while critical goals such as SDG 1 (No Poverty) and SDG 2 (Zero Hunger) receive minimal attention (1.0 % and 0.3 %, respectively). To address these gaps, we convened a focus group of nine experts from academia, industry, and policy sectors to develop actionable solutions for each SDG, including AI-driven financial inclusion tools, precision agriculture models, and energy-efficient “Green AI” technologies. Our findings suggest a notable misalignment between AI’s potential to address pressing humanitarian challenges and the patterns of concern-oriented public discourse captured in this study. These findings point to the need for broader public awareness, robust ethical governance, and interdisciplinary collaboration if AI is to advance sustainability goals equitably and reach underserved communities.
Human-engineered “gray” infrastructure (e.g., artificial reservoirs) and ecosystem-based “green” infrastructure (e.g., forests) both provide essential water storage capacities to buffer hydrological variability under rising climate uncertainty. However, a quantitative assessment of gray (Sgray) and green (Sgreen) water storage capacities across global major river basins remains lacking. In this study, we estimate ecosystems’ root zone storage capacity as a proxy for Sgreen using the mass curve technique. Combined with artificial reservoir storage capacities, our results showed that Sgray and Sgreen exhibit highly spatial heterogeneity, with the largest volumetric Sgray in Yenisei (459 km3) and Sgreen in Amazon (1,427 km3). Over the period 1959–2020, Sgray experienced a consistent increase, while Sgreen exhibited fluctuations. Notably, in 12 % of the global major river basin area (1.3 × 107 km2), including the Colorado, Yenisei, and Yangtze, indicating that human-engineered regulation of hydrology has become more influential than terrestrial ecosystem in these regions. These gray dominated basins are also projected to experience a decline in Sgreen, suggesting the potential need to protect and enhance green infrastructure. This study provides a scientific basis for the integrated management of gray and green infrastructures, offering insights to enhance the resilience and sustainability of water resources at the basin scale.
Human-wildlife conflict in drylands is worsening due to climate change and increasing human pressures, yet its spatiotemporal dynamics remain poorly understood. Here, we present an integrated modeling framework combining system dynamics, the Patch-generating Land Use Simulation (PLUS) model, and the BIOMOD2 species distribution platform to project future potential conflict trends under three socioeconomic pathways (i.e., SSP1-2.6, SSP2-4.5, and SSP5-8.5). Our results show that 23,900 km2 of drylands are currently potentially affected by spatial overlap with conflict-prone areas, with projections under SSP5-8.5 suggesting that potential conflict areas could nearly double by 2100. Spatial simulations highlight oasis-desert transition zones, particularly near protected areas, as key hotspots where increasing aridity and habitat fragmentation drive heightened resource competition. Our analysis suggests that management aligned with Nature-based Solutions principles—including land-use restrictions, ecological restoration and protected area management—could reduce spatial overlap between human land use and wildlife habitat, supporting biodiversity and sustainable development. These findings underscore the need to integrate such strategies into global sustainability frameworks, like the Kunming-Montreal Global Biodiversity Framework, to enhance the resilience of dryland ecosystems amid rapid environmental change.
Managing interactions among food production, water resources, and ecological systems is a major sustainability challenge in semi-arid and rapidly urbanizing socio-ecological regions. Using Big Earth Data and a two-scale analytical framework, this study quantifies spatiotemporal synergies and trade-offs within the food-water-ecology (FWE) nexus in the Beijing-Tianjin-Hebei region from 2005 to 2022. A mechanical equilibrium model reveals persistent north-south differentiation, with strong trade-offs in mountainous ecological zones and weak synergies across the agricultural plains. XGBoost-SHAP analysis identifies nonlinear ecological and anthropogenic thresholds, including forest cover above 60 %, elevations exceeding 780 m, and construction-land development near 0.98, that influence transitions between synergy and trade-off states. Cross-scale comparison shows that natural constraints dominate grid-level patterns, while human activities increasingly control township-level outcomes. These findings highlight the necessity of threshold-informed and spatially differentiated governance, including groundwater recovery and agricultural intensity management in the plains, long-term ecological stabilization in uplands, and integrated water-ecology regulation in metropolitan areas. The framework provides a tool for advancing coordinated progress toward Sustainable Development Goals 2, 6, 11, and 15.
Global change and rural abandonment are restructuring Mediterranean fire regimes, as landscapes once characterized by heterogeneous agricultural mosaics are progressively replaced by homogenized, drought-limited fuel complexes. This transformation increases the likelihood of high-intensity wildfire events and their associated impacts. This study applied a stochastic fire-spread modeling framework based on thousands of simulated wildfire events. We quantified the effects of land abandonment in the Matarranya region in eastern Spain, comparing current landscape conditions with a counterfactual “no-abandonment” scenario considering agricultural lands non-burnable in both cases. Results indicate that active agriculture serves as a critical passive infrastructure protection system by maintaining fuel discontinuity, yielding a 68% mean reduction in building exposure (β = 0.32). Conversely, land abandonment functions as an exposure multiplier, triggering a 7-fold increase in the number of structures reaching high and very high burn probability threshold. Abandonment also raises burning intensity (flame length), which limits suppression capability. Overall, the cessation of management roughly tripled potential fire size and quadrupled exposure of human infrastructure, highlighting the importance of identifying “high-leverage plots” where targeted agricultural restoration maximizes safety. By strategically prioritizing these parcels, planners can transform abandoned land into revenue-generating firebreaks and functional strategic fire zones (SFZs), establishing an evidence-based framework for rural development and wildfire risk mitigation.
The escalating pressure of global warming necessitates practical and effective climate governance policies to meet the urgent goals of the Paris Agreement. However, for conventional policy simulation models like integrated assessment models (IAMs), their reliance on predetermined scenarios makes it challenging to adequately address the substantial uncertainties inherent in climate-social system evolution. This study proposes an adaptive decision-making framework that integrates deep reinforcement learning (DRL) with a climate–social system model to identify governance strategies that can reduce the risk of transgressing planetary boundaries. In this framework, we operationalize Social Tipping Elements (STEs) as targeted actions. A reward function provides feedback to optimize these interventions, ensuring the system remains within critical planetary boundaries. Our results demonstrate that, compared to conventional static models, the framework discovers adaptive policies through dynamic learning, enabling real-time responses to evolving climate-social conditions. These adaptive policies exhibit an “early-stage intensive intervention, mid-term moderation, and late-stage reinforcement” pattern that reduces planetary boundary overshoot time by 55 years while stabilizing global warming below 1.5 °C by 2100. By varying governance objectives, we further reveal the trade-off mechanisms between competing climate and socioeconomic goals. Notably, the designed multi-objective reward function enables a synergistic balance across competing objectives, resulting in the shortest planetary boundary overshoot duration (15 years) among all evaluated scenarios. This framework overcomes the limitations of static simulations, offering a robust and interpretable tool to design adaptive strategies crucial for navigating the competing objectives of time-sensitive climate governance.
The year 2030 marks a pivotal milestone for China, algining with both the completion of the United Nations 2030 Agenda for Sustainable Development and the country’s commitment to peak carbon emissions under the Paris Agreement. Given limited time and resources, harmonizing decarbonization targets with Sustainable Development Goals (SDGs) is a core governance challenge. This study proposes a framework integrating Multi-spatial Convergence Cross-Mapping (MCCM), network analysis, and Graph Convolutional Network (GCN)-based scenario forecasting. By incorporating causal effects, structural importance, and evolutionary potential, the framework quantitatively derives SDG priority orders across Shared Socioeconomic Pathways (SSPs) under China’s carbon peaking. The results show that SDG indicators 7.2.L, 3.d.1, 1.1.L, 3.8.1, and 9.5.1 are core cross-pathway levers, generating spillover effects that are 60.7 %–91.5 % greater than the system average. Targeted resource allocation to these indicators can catalyze SDG advancement. At the goal level, prioritizing SDG 7 as a key driver (with spillover effects 57.3 % above mean) while addressing the persistent underperformance of SDG 15 provides the backbone for achieving cross-pathway synergies. Implementation sequences are path-dependent. In SSP1, SDG 4 and SDG 17 are the foundational pillars for harmonizing climate commitments with SDGs. Under SSP3 and SSP5, structural imbalances elevate the urgency of SDG 10 by 38.7 % and 37.3 % relative to SSP1, increasing its strategic priority. In SSP2 and SSP4, SDG 9 becomes pivotal, driven by lagging progress (10.7 % and 17.3 % behind SSP1) and amplified spillover effects (15.2 % above average). This study provides evidence-based guidance for China to optimize policies and maximize dual-goal synergies across pathways, while offering transferable insights for aligning sustainability and climate action globally.
Grasslands, as the largest terrestrial ecosystem, hold significant ecological and economic value. Due to widespread grassland degradation driven by environmental changes and human activity, restoration efforts have increasingly prioritized improving the health of grassland ecosystems. Ecosystem services serve as a critical metric for evaluating the success of these ecological restoration efforts. However, quantitative evidence remains limited regarding how different restoration measures affect multiple ecosystem service categories and how these effects are regulated by environmental and baseline conditions. In this study, a meta-analysis was conducted to assess the impact of grassland restoration on ecosystem services. The results revealed that combination measures produced the most substantial improvements in ecosystem services, with increases of 94.2 % in total services, 120.1 % in provisioning services, and 102.4 % in regulating services. Natural recovery and reseeding showed stable benefits, with total, provisioning, and regulating services increasing by 50.7 % and 77.7 %, 96.6 % and 152.4 %, and 64.6 % and 101.1 %, respectively. Fertilization demonstrated the least improvement, with total and provisioning services increasing by 35.5 % and 57.4 %, respectively. Supporting services, however, proved difficult to restore, with only natural recovery showing a significant positive effect. Additionally, the effectiveness of ecological restoration measures is influenced by various factors, including time accumulation, precipitation, altitude gradients, and the baseline condition of the ecosystem. The restoration of grassland ecosystem services is a complex, nonlinear process shaped by the interaction of these factors. Therefore, it is crucial to understand the specific environmental background of a given location before selecting the most suitable restoration measures based on the restoration goals. This approach will enhance the effectiveness and sustainability of future restoration efforts.
Food security is increasingly threatened by the growing mismatch among population, farmland, and grain production. However, existing studies remain limited in spatial coverage and rarely examine the dynamic interactions among these three components at the global scale. This study analyzes the spatiotemporal patterns of per capita farmland area (PCFA) and per capita grain production (PCGP) from 2000 to 2020 and develops a population-farmland-grain production coefficient (PFGC) to quantify their matching relationships. Results show that global PCFA declined by 21.18 % during the study period, while PCGP decreased by a more moderate 14.32 %. Notably, only South America experienced a net increase in PCGP (7.76 %) among all continents. The global distribution of matching types is dominated by a declining pattern, accounting for more than 70 % of the study area, indicating increasing pressure on grain production systems. These findings reveal a widespread spatiotemporal mismatch among population growth, farmland availability, and grain production. The study provides new insights into the global dynamics of agricultural resources and offers policy implications for improving food security and sustainable farmland management.
The Belt and Road Initiative (BRI) has significantly reshaped global trade patterns while intensifying ecological pressures. To capture these ecological impacts, Human Appropriation of Net Primary Production (HANPP) provides a comprehensive metric to assess the impact of trade on biomass flows and carbon sink loss, yet previous research has lacked detailed country- and sector-specific estimates. We quantify the HANPP embodied in trade across the BRI using an environmentally extended multi-regional input-output model and project its future trajectory under socioeconomic scenario pathways from 2020 to 2050. Our findings reveal a sharp increase in HANPP embodied in trade within the BRI, with consumption-driven economies relying on biomass imports, shifting ecological burdens to resource-exporting regions. During 2004–2017, and particularly after the launch of the BRI, HANPP embodied in imports rose from 23.9 % to 29.3 % within the Belt and Road countries, while net HANPP exports concentrated in key suppliers such as Russia, Indonesia, and Brazil. China, Egypt, and Turkey emerged as the largest net importers, intensifying land-use pressure in exporting regions. Projections indicate that by 2050, HANPP could increase by up to 3.8 % annually under a fossil-fueled development pathway (SSP5), exceeding the growth under sustainability-oriented scenario (SSP1, 3.0 %) and middle-of-the-road scenario (SSP2, 2.1 %), and further widening the gap between production- and consumption-based HANPP. Agricultural products, processed food, and manufacturing sectors will dominate future HANPP consumption. This research highlights the necessity for sustainable trade policies, emphasizing green product adjustments, burden-sharing among key trading nations, and enhanced environmental cooperation within the BRI framework.
