Long-distance trails serve as essential infrastructure linking humans and nature. Currently, most trail planning relies on suitability analysis, which assesses areas rather than routes and often results in discontinuous trails. Using the Taihang Mountain National Forest Trail - Beijing Section (TMNFT-BS) as a case study, this study proposes an integrated framework for trail planning. Based on hiking trajectory data, we use landscapes and wilderness continuum map for screening conditions to locate a trail. The resulting TMNFT-BS trail is approximately 179.8 km in length and: (1) follows only existing routes, minimizing ecological impacts and costs; (2) is continuous and accessible; and (3) traverses areas with the highest average wilderness quality while linking diverse landscapes, providing a rewarding hiking experience. This study introduces a novel approach to the longest path problem in general graphs and presents a trail-planning methodology applicable in China and beyond.

Accurately predicting the grain protein content (GPC) and yield of winter wheat is of significant strategic importance amid rising food demand and intensifying global market competition. However, traditional single-model approaches struggle to achieve high simulation accuracy in complex agricultural ecosystems. This study proposes a novel multi-model ensemble (MME) framework that integrates the APSIM-NG (Agricultural Production Systems Simulator-Next Generation) process-based crop model, four machine learning algorithms (Random Forest, Extreme Gradient Boosting, Multiple Linear Regression, and Long Short-Term Memory), and two ensemble methods (AIC-weighted model averaging and simple model averaging) to enhance the predictive accuracy of GPC and yield in North China Plain. The MME framework incorporates remote sensing data, extreme weather indices, and crop growth observations from 2008 to 2020 for a comprehensive performance evaluation. Validation results for the period 2015-2020 indicate that the MME framework outperforms both the baseline APSIM-NG model and the best-performing machine-learning method, achieving a Pearson's r of 0.89 (RMSE = 0.32 %, R2 = 0.76) for GPC prediction and reducing the yield RMSE to 316.96 kg/ha (Pearson's r = 0.94, R2 = 0.91). Furthermore, importance analysis indicates that within this framework, photosynthesis-related and extreme stress factors are the most influential predictors, contributing 8-12 % to model importance, highlighting the substantial impact of including extreme weather factors on model accuracy. By effectively combining process-based modeling with data-driven methods, the MME framework significantly enhances predictive accuracy and model robustness. These findings offer a more reliable technical foundation for forecasting winter wheat yield and grain quality under variable and extreme climatic conditions.

Campus sustainability assessments increasingly rely on broad indicator systems, yet their empirical foundation remains heavily dependent on institutional self-reports, administrative records, and checklist-based scoring. These data types capture governance and educational activities effectively. However, they are less suitable for evaluating spatially heterogeneous or temporally dynamic environmental conditions, such as greenspace distribution, land use composition, and thermal environments. Such conditions are difficult to verify through self-reporting alone. Remote sensing and GIS offer potential solutions by providing externally verifiable and cross-campus comparable measurements, but their integration into formal assessment standards remains underdeveloped. The literature reveals three persistent gaps: an evidence gap (over-reliance on self-reports), a spatial gap (limited inclusion of spatially explicit indicators), and a standardization gap (case studies without transferable protocols). These gaps are relevant in China, where diverse campus forms intensify comparability challenges. To address these issues within the Chinese context, this review proposes a conceptual two-level framework. The framework comprises a comprehensive assessment structure covering governance, education, operations, environment, and participation. It is paired with a separate spatial evidence module based on remote sensing and GIS. This framework is a conceptual reference and an operational pathway, rather than an empirically validated or calibrated tool. The framework intends to guide the systematic integration of spatial evidence into future Chinese green campus standards. However, threshold values, weighting schemes, and contextual adjustment factors would require further empirical testing and policy deliberation.

To address the issues regarding the unclear impacts and driving mechanisms of large-scale photovoltaic (PV) development on the land surface thermal environment in Northwest China, this study, based on MODIS data, defined the vegetation growing season as May to September and monitored and attributed the daytime temperature changes of 3,589 PV plant sites in the region before and after construction. The study found that the overall land surface temperature (LST) change in PV plants presented a significant "seasonal reversal" characteristic of "cooling in the growing season and warming in the non-growing season," and this characteristic showed a distinct dependence on land cover type. Specifically, for sites that were bareland before construction, the growing season showed an offsetting of cooling and warming effects, with cooling sites accounting for about 50% (temperature difference -1~1°C); whereas the non-growing season showed significant warming, with the proportion of cooling sites dropping to 30%~35% (temperature difference 0~3°C). In contrast, for sites that were cropland before construction, the proportion of cooling sites remained around 60% in both the vegetation growing season and non-growing season, with the majority of sites showing cooling. For sites that were grassland before construction, about 60% of the sites showed warming during the vegetation growing season, with the distribution mainly concentrated in the 0-2°C temperature change interval; while in the vegetation non-growing season, the proportions of cooling and warming sites were close, and the distribution range extended to -1~2°C. Attribution analysis revealed a seasonal switch of dominant factors: the biophysical mechanism dominates in the growing season, where vegetation evapotranspiration produces a cooling effect, and grassland Enhanced Vegetation Index (EVI) showed a strong negative correlation with temperature difference (r=-0.84, p<0.001); the radiative forcing mechanism dominates in the non-growing season, where the low albedo of PV modules increases net radiation absorption, and White Sky Albedo (WSA) on bareland showed a significant negative correlation with tem-perature difference (r=-0.58, p<0.001). This study quantitatively confirmed the mechanism of "vegetation transpiration dominating cooling in the growing season, and albedo radiative forcing dominating warming in the non-growing season," providing a scientific basis for assessing the ecological and environmental effects of the PV industry in Northwest China.

Reliable and intelligent retrieval of leaf traits from hyperspectral reflectance is crucial for assessing ecosystem functions, yet conventional approaches struggle with spectral complexity and nonlinearities. To address these challenges, we developed the Leaf Trait Retrieval Network (LTRN), a novel deep learning framework that integrates Kolmogorov-Arnold Network (KAN), Transformer, and Temporal Convolutional Networks (TCN) for end-to-end trait estimation. Model validation was carried out using a large spectral-trait database covering hundreds of plant species and four functional traits. Experimental results demonstrated that LTRN model outperforms state-of-the-art deep learning models, achieving R2 values greater than 0.78 for estimating chlorophyll content (Chla+b), equivalent water thickness (EWT), carotenoid content (Ccar), and leaf mass per area (LMA). Further analyses indicated that the LTRN model delivers stable estimation performance across spectral resolutions of 10-25 nm. Moreover, the model demonstrates strong stability across varying proportions of training samples. These findings underscore the robustness and stability of LTRN for large-scale vegetation trait retrieval, offering a valuable framework for advancing the intelligent estimation of other ecological parameters.

Precise estimation of rice aboveground biomass (AGB) is essential for assessing rice straw resources and their potential utilization as animal feed and lignocellulosic materials. Vegetation indices (VIs) often perform poorly at extreme AGB levels due to spectral saturation and canopy structure complexity. Texture characterizes the spatial distribution and structural traits of vegetation and could compensate for the limitations of spectral information in monitoring AGB, particularly under complex canopies and soil backgrounds. In this study, hyperspectral images covering the entire rice growth cycle were used to derive VIs and texture indices (TIs). Based on the mechanistic linkage between spectral and textural information, two spectrum-texture composite indices (STCIs) were developed: spectrum and corresponding-band texture composite index (SCBTCI) and spectrum and full-band texture composite index (SFBTCI). The optimal VIs, SCBTCIs, and SFBTCIs with the highest correlations with AGB were selected through correlation analysis and used to build models using linear regression. The results showed that STCI-based models consistently outperformed VI-based models. The three optimal SCBTCI models achieved an overall validation rRMSE of 0.271, while the three optimal SFBTCI models performed better with an overall validation rRMSE of 0.240. Compared with the VI1[872,784], SCBTCIs improved the estimation at low AGB level (overall=15.656%) and pre-heading stages (overall=14.706%), whereas SFBTCIs showed greater improvements at high AGB level (overall=27.082%) and post-heading stages (overall=23.424%). STCIs showed a stronger linear relationship with AGB and achieved accuracy comparable to or higher than machine learning methods. The great performance of STCIs under extreme AGB conditions demonstrated their capability to mitigate spectral saturation and reduce background interferences. STCIs provide an efficient framework for AGB monitoring, supporting the assessment of rice straw resources and their sustainable utilization as lignocellulosic biomass.

Despite numerous global initiatives and policy framework to mitigate ongoing biodiversity decline, progress remains limited due to lack of biodiversity indicators that are timely, scientifically rigorous, and representative. Furthermore, databases underlying previous indicators are spatially, temporally, geographically and taxonomically biased, making it difficult to track biodiversity change dynamics and set proper biodiversity targets. Here, we constructed a new version of the global human footprint, and used it to infer temporally explicit annual shifting patterns of biodiversity across all scales by incorporating remote sensing and mapping out human pressures. This indicator (the Ecological Integrity Index- EII) successfully differentiates high- and low- biodiversity biomes, especially for deserts and tundra. Moreover, shifting annual patterns can identify global hotspots (e.g., major rainforests and regional hotspots), and shows biodiversity change dynamics at regional level based on estimating biodiversity change over time. Changes of evolving human footprint were further analyzed with relationships to biodiversity patterns. At more a local level, the patterns perfectly reflect biodiversity and intactness. Compared to other indicators (e.g., BII, BHI) in the Kunming-Montreal Global Biodiversity Framework (GBF) and biodiversity models (e.g., GLOBIO), the EII can better reflect biodiversity. EII shows a good performance, with the potential to inform biodiversity conservation efforts, and support the implementation of the post-2020 global biodiversity framework.

Abstract Across the globe, spatial mismatches between work and residence reshape human mobility and influence regional air quality, yet their effects across urban–rural systems remain poorly quantified. China's ongoing urbanization and rural revitalization underscore the urgent need to understand these dynamics. Here, using multi‐source geospatial data, we conduct a nationwide analysis across China to assess how work–residence mismatches within and between urban and rural areas affect air pollution, accounting for seasonal variation. We find that overall, urban, and rural mismatch degrees follow an inverted‐U trend with city tier, with disparities between urban and rural degrees of mismatch intensifying within lower tiers. We show that increased mismatch aggravates pollution, particularly in rural regions, higher‐tier cities, and during winter. Furthermore, we identify six distinct urban–rural synergy patterns of spatial mismatch, each associated with varying pollution levels that eventually decline after peaking. Seasonal effects, especially in winter, exacerbate these disparities. Our results emphasize the need to incorporate rural planning into air quality strategies and to adopt seasonally and contextually tailored policies.

Unlocking the potential of abandoned cropland is widely recognized as a promising strategy for enhancing landuse efficiency and achieving land sustainability. However, the specific pathways and achievable yield potential remain poorly quantified. Here, we integrated geospatial data, multi-modeling, and scenario analysis methods to evaluate the potential for food production, carbon sequestration, and photovoltaic (PV) power generation via recultivation, afforestation, and PV deployment incentives on China's abandoned cropland. Our results show that China's abandoned land has the potential to yield 17.63 - 42.53 Pcal yr-1 of food, sequester 4.58 - 6.21 Tg C yr-1 of carbon, and generate 2.18 - 5.50 PWh yr-1 of clean energy, depending on the scenario of land reuse. We found that prioritizing regionally tailored strategies could further amplify land-use efficiency. For example, afforesting 0.93 Mha of abandoned land in Southwest China (24.6 % of the total abandoned area) has the potential to sequester 2.13 Tg C yr-1, accounting for 34.3 % of the maximum carbon sequestration potential. Further analysis reveals that transforming from rain-fed to irrigated practices could enhance food production potential from abandoned land by 30.4 - 55.3 %, and future climate changes may lead to potential spillover effects (increasing 1.67 +/- 2.36 %, standard deviations across different Shared Socioeconomic Pathway scenarios) of PV power generation. Collectively, our findings underscore that multiple pathways for repurposing abandoned land are essential for harnessing its full suite of land resources.

Within the context of the UN 2030 Agenda for Sustainable Development and Education for Sustainable Development (ESD), this study draws on a Web of Science dataset (n = 815, 1991-2025) and employs a mixed approach combining scientometric mapping with framework analysis and tool comparison. It systematically reviews the knowledge structure, methodological evolution, and tool genealogy of Campus Sustainability Assessment (CSA). The results reveal a paradigmatic shift from an operations-oriented focus to a whole-of-institution and impact-oriented perspective. Representative tools can be grouped into five categories by purpose-improvement-oriented, ranking and benchmarking, education and curriculum, standards and certification, and policy advocacy and recognition-and can be mapped onto the four domains of governance, academics, operations, and engagement in alignment with the Sustainable Development Goals (SDGs). Synthesizing quantitative and qualitative evidence, three systemic shortcomings are identified: excessive reliance on self-reporting with limited verification, insufficient evidence of learning outcomes and key competencies, and weak interoperability of indicators across educational stages and frameworks. Looking ahead, four actionable research pathways are proposed: (1) assessment of key competencies centered on learning outcomes with stronger curriculum-practice alignment; (2) policy-indicator interoperability and vertical integration grounded in SDGs and national or sectoral standards; (3) stakeholder co-design enabling an assessment-improvement loop; and (4) remote-sensing-based multi-scale monitoring and data governance. The contribution of this study lies in advancing a unified four-domain framework under a process-outcome-impact evidence chain, while suggesting cross-stage and cross-tool alignment and complementarity. This provides methodological support and an implementation roadmap for shifting CSA from measuring performance to empowering improvement.

Climate change challenges rice production to sustain yield while improving nitrogen-use efficiency and reducing environmental impacts. Yet the role of stage-specific nitrogen management across diverse rice systems remains unclear. Here we evaluated nitrogen management scenarios under historical and future climates. RiceGrow, DSSAT-CERES-Rice and ORYZA v3 were calibrated on agro-meteorological observations, and literature-derived field trials benchmarked their nitrogen responses. RiceGrow was then coupled to machine-learning emulation and multi-objective optimization in a transferable framework, with yield constrained to at least 90% of the simulated maximum. Under historical climate, optimized practice reduced subregion-mean nitrogen inputs by 15% to 37% and raised agronomic efficiency of nitrogen (AEN) and gross margin by 10 to 21 kg·kg-1 and 0.3 to 2.5 ×103 USD·ha-1, relative to farmers' practice. These gains required one to two additional splits, with greenhouse gas emissions (GHGs) varying from a 6% decrease to a 26% increase, a range driven by total nitrogen and yield rather than by stage reallocation. The framework consistently favored mid-season topdressing over large basal applications, although the specific split percentages are conditional on the backbone model. Future climate altered nitrogen demand unevenly, decreasing optimal nitrogen rates by 2% for middle-lower Yangtze single rice and 12% for late rice, but increasing them by 12% in other single-rice subregions and 19% for early rice. Within this modeling framework, soil properties accounted for the largest unique share of variation in optimized practice. Shifting nitrogen from basal dressing to mid-season topdressing offers a practical pathway to climate-resilient rice production, though GHG mitigation in some subregions depends on complementary water-regime and residue management.

Abstract. Climate and habitat change are reshaping global biodiversity, yet long-term, spatially explicit datasets that capture these dynamics across ecologically meaningful regions remain limited. Here, we present a global spatiotemporal dataset of ecoregion-level biodiversity indicators for terrestrial vertebrates derived from species-specific Area of Habitat (AOH). By integrating species occurrence records, expert-derived range maps, climate data, and temporally explicit habitat maps, we reconstructed distributions for 19146 species of amphibians, birds, mammals, and reptiles. We then derived four complementary biodiversity indicators—species richness, threatened species richness, species endemism, and AOH density—for global terrestrial ecoregions at five time points (1990, 2020, 2030, 2050, and 2100), with future projections under SSP245 and SSP585. Validation at both species and dataset levels demonstrated good performance. Predicted AOH consistently assigned higher suitability to independent occurrence records than to the broader range background, and estimated richness closely reproduced the major spatial patterns of occurrence-derived richness across ecoregions. In 2020, the four indicators showed distinct but broadly concordant spatial patterns, with the highest overall values concentrated in tropical ecoregions. Future projections revealed marked spatial heterogeneity, with declines in multiple indicators concentrated in many tropical ecoregions by 2100 and generally becoming stronger under SSP585. This dataset provides a globally consistent and ecologically meaningful resource for biodiversity monitoring, macroecological analysis, and conservation assessment under ongoing climate and habitat change. The dataset supporting this study is publicly available at https://doi.org/10.5281/zenodo.20119261.

Timely and reliable prediction of fish biodiversity across extensive river systems is pivotal for biodiversity monitoring and sustaining ecosystem stability. However, large-scale field surveys remain difficult, and broadband multispectral imagery captures habitat attributes only coarsely, limiting predictive precision. To address these challenges, Sentinel-2 multispectral observations, Gaofen-5A hyperspectral imagery, and the Google Satellite Embedding dataset were integrated with environmental DNA metabarcoding, and 7 machine learning algorithms were implemented to predict fish taxonomic diversity. A total of 149 environmental DNA samples were compiled from field campaigns and literature sources across 15 river basins in China. Habitat-informative vegetation indices were derived from pixel-level spectra, and 4 feature categories were examined, namely, multispectral variables, hyperspectral indices, deep embeddings, and fusion features combining hyperspectral information with satellite embeddings. Hyperspectral indices markedly improved prediction accuracy relative to that of Sentinel-2 multispectral predictors, achieving an R 2 of 0.80 ± 0.03 compared with 0.65 ± 0.07 for the multispectral baseline. Model performance remained stable across spectral resolutions using Gaofen-5A data. Notably, deep embedding features enabled multiple models to achieve an R 2 above 0.90, reflecting their strong capacity to integrate multisource information and characterize fish habitat conditions. Fusion of hyperspectral and embedding features produced the most accurate estimates of fish taxonomic richness with R 2 up to 0.94 ± 0.01. Overall, these outcomes demonstrate the feasibility and robustness of combining hyperspectral data with deep learned representations to predict riverine fish taxonomic richness, providing a multisource remote sensing pathway for biodiversity assessments across large spatial extents.

Protected areas (PAs) are vital for biodiversity conservation and ecosystem services, but their benefits are unevenly distributed. However, the extent of this inequality remains unclear. A quantitative accessibility-availability framework is developed to evaluate the spatial equality of PAs in China. Using data on 3,710 PAs, a county-level spatial distribution index is constructed by combining road-network accessibility (distance to the nearest PA) and availability (PA quantity and coverage) across 2,859 county-level divisions. Distributional equality is summarized with the Gini coefficient, supplemented by spatial clustering and Geographically Weighted Regression (GWR). Results reveal four main patterns: (i) approximately one-third of county-level divisions lack any PA sites, and PA quantity is more unevenly distributed than coverage rates; (ii) accessibility analysis shows an average road distance of 56.57 km to the nearest PA (maximum 503.45 km), with half of all county-level divisions located within 40 km; (iii) pronounced regional asymmetries emerge-eastern, northeastern, and central China host 62.9% of PA sites but only 8.3% of total PA area, whereas western China contains 37.1% of sites and 91.7% of area-indicating fragmented, small PAs in the east versus large, sparse PAs in the west; and (iv) GWR identifies population, road length, GDP per capita, and biodiversity value as the dominant correlates of spatial equality, reflecting the combined demographic, infrastructural, economic, and ecological determinants shaping inequality. Policy implications for PA system planning: (i) integrating spatial equality into PA planning objectives; (ii) prioritizing new PAs near urban and densely populated regions; (iii) enhancing connectivity and gateway infrastructure in remote regions; (iv) supporting small and community-based PAs; and (v) integrating public health into PA planning. Embedding the Gini-based accessibility-availability metric in national and provincial planning can advance an inclusive, just, and ecologically effective PA system.