Understanding the spatiotemporal impacts of land use transition on carbon emissions is crucial for achieving regional carbon neutrality. This study presents an integrated analytical framework that combines dynamic land use modeling, the Geo-detector method (GDM), and Geographically and Temporally Weighted Regression (GTWR) to analyze land use transition and carbon emission dynamics in China's Pearl River Delta (PRD) from 2000 to 2020. Key findings include: (1) Construction land expansion was the dominant explicit transition, with land conversion sources shifting from cropland-centric patterns to diverse transfers involving woodland and water bodies. (2) The implicit land use transition index exhibited an annual growth rate of 15.6%, progressing through three phases-rapid development (2000-2010), structural adjustment (2010-2015), and high-quality transition (2015-2020). (3) Regional carbon emissions increased by 186.96%, exhibiting spatial disparities between core and peripheral regions. Construction land expansion and GDP density were primary drivers. This research advances the theoretical integration of land system science and low-carbon governance, offering actionable insights for spatially differentiated emission reduction strategies in megacity clusters.

Land use change has been demonstrated to be a key driver of soil acidification. However, there is a lack of systematic research on the interfering factors that continue to exacerbate soil acidification following land use change. Here, we compiled a soil pH dataset, including data from the Second National Soil Survey (1980s) and China Soil Series Survey (2010s), and employed a combined approach of machine learning and analysis of variance to elucidate the spatial variability of soil acidification and its influencing factors in subtropical China over the past 30 years under the context of land use change. The results indicated that over the three decades, both topsoil (0-20 cm) and subsoil (20-40 cm) have experienced severe acidification. Soil acidification following the conversion of agricultural land (paddy field and dryland) to woodland depends on the buffering capacity of the parent material. The interaction between land use change and nitrogen input (nitrogen deposition and nitrogen fertilizer application) significantly affected soil pH changes in the topsoil. Specifically, when the nitrogen input decreased by more than 50 kg N ha-1 year-1, soil pH of woodland converted to agricultural land increased, i.e., acidification was reversed. Our study results support the pursuit of a balance among agricultural land, tillage management, and climate change under the condition of stabilizing soil fertility, from which policymakers and farmers can benefit.

In the red and yellow soil regions of southern China, long-term practices in soil acidification constraint management have revealed several challenges, including fragmented knowledge systems, high technical thresholds, and pronounced regional disparities in technical capacity, which collectively hinder the systematic integration and dissemination of management experience. To address these limitations, this study proposes a lightweight autonomous construction framework for a knowledge graph of soil acidification management based on large language model (LLM) agents. By introducing an "AI associative matrix" and a semantic expansion-retrieval mechanism, the framework enables the system to automatically infer related concepts and relationships from a small set of domain-specific keywords, thereby generating an initial ontology structure and knowledge graph schema for management tasks. In addition, a network protocol-based control strategy is incorporated to semantically parse publicly available governmental reports and monitoring data, facilitating automated processes such as entity recognition and relation extraction. At the system implementation level, a cloud-edge collaborative lightweight architecture is adopted to reduce computational requirements and improve usability. Through an empirical in-situ deployment and lightweight field-level testing conducted directly at the farm sites of Chengmai County, Hainan Province, this framework demonstrated a robust capacity to adapt to users with different technical backgrounds and regions with varying levels of resource input, while providing extensibility for the continuous integration of emerging technologies and updated policy regulations. Overall, this study offers a lightweight, user-friendly, and scalable intelligent approach for cultivated land quality improvement and agricultural ecological management.

Agricultural sustainable development for food and environmental security relies largely upon our understanding of the functions of an agricultural ecosystem and their relationships. This study aims to define functional zones by identifying the relationships among the individual functions of the Guangdong Province’s agricultural land resources for sustainable development. Therefore, a specific Bayesian belief network (BBN) was built to identify the trade-off/synergy relationships between functions, and the density peaks clustering (DPC) algorithm was used to perform function-oriented zoning. As a result, significant synergies of food production function are identified with other functions, while a trade-off exists with food cleanliness. The 1 km grid scale set as the zoning unit can be more appropriate for a land use plan to consider functional relationships and specific management options. Ultimately, the Guangdong agricultural land is zoned into five functional zones (green agriculture, tropical agriculture, productive agriculture, ecological agriculture, and urban agriculture) with specific options and stakeholders’ policies that are suggested to weaken existing and potential trade-offs accordingly. Evidently, coupling BBN with DPC demonstrates an enhanced capability to optimize an agricultural land use plan by balancing multifunctional synergies/trade-offs for advancing sustainable agricultural development.

The coupling relationship between land use transition and carbon emissions is a critical scientific issue for achieving regional carbon neutrality. This study integrates dynamic land use models, geographically and temporally weighted regression (GTWR), and a framework to assess the exposure, sensitivity, and adaptive capacity to carbon emissions in the Pearl River Delta (PRD) region from 2000 to 2020. The key findings include the following. (1) Construction land expansion dominated explicit land use transition (88.79%). The land use sources shifted from cropland to multiple land use/cover types, including cropland, woodland, and water. (2) The implicit land use morphology index exhibited an annual growth rate of 15.6%, characterized by three phases (rapid development, steady adjustment, and high-quality transition), indicating increasing resilience to advanced transformation. (3) Total carbon emissions increased by 186.96% with significant spatial heterogeneity (Guangzhou > Foshan > Shenzhen > Dongguan). The area of construction land scale and GDP per unit area were key drivers, and industrial structure optimization contributed 26.7% to emission reduction. (4) The carbon emission resilience index (CRI) rose from 0.53 to 0.58, with high values in Shenzhen and Zhangshan and low values in Jiangmen and Zhaoqing, indicating technological innovation and policy synergy were critical pathways. This research provides scientific support for low-carbon land management strategies in the Guangdong-Hong Kong-Macao Greater Bay Area.

The Qinghai–Tibet Plateau is a crucial ecological security barrier in China and Asia. Its grassland ecosystem has high ecological service value. Scientific assessments and classifications of grasslands are crucial for determining the value of grassland resources and implementing refined management. Traditional grassland classification methods have used expert knowledge and linear models, which are subjective and cannot describe complex nonlinear relationships. We conducted a case study in Hongyuan County, Sichuan Province, in the water conservation area of the Qinghai–Tibet Plateau, using multi-source data including Landsat 8 (15 m/30 m), MOD15A2 (500 m), ALOS imagery (12.5 m), and 435 field survey samples, combined with machine learning models such as convolutional neural network (CNN), extreme gradient boosting (XGBoost), light gradient boosting machine (LightGBM), histogram gradient boosting (HistGradientBoosting), and random forest (RF). The objective was to develop a novel grassland classification method that integrates multi-source remote sensing data with machine learning algorithms. Based on the evaluation metrics of SHAP values, mean annual precipitation (MAP, 0.675), >0 °C Accumulated Temperature (AT, 0.591), and aspect (ASPECT, 0.548) were the most critical factors influencing alpine grasslands, revealing a driving mechanism characterized by climate dominance, topographic regulation, soil support, and vegetation response. The XGBoost model demonstrated the best performance (with an accuracy of 0.829, Precision of 0.818, Recall of 0.829, weighted F1-score of 0.820, and an AUC value of 0.870). The pixel-by-pixel absolute difference calculation between the model-predicted and the actual classification results showed that regions with no discrepancy (absolute value = 0) accounted for 75.82%, those with a minor discrepancy (absolute value = 1) accounted for 23.63%, and regions with a major discrepancy (absolute value = 2) accounted for only 0.54%. This study has established a replicable paradigm for the precise management and conservation of alpine grassland resources. Through the synergistic application of deep learning and machine learning, it generated superior baseline data, quantitatively uncovered a grassland differentiation mechanism dominated by hydrothermal factors and fine-tuned by topography in the complex Qinghai–Tibet Plateau, and delivered high-precision spatial distribution maps of grassland classes.

Land cover conversions (LCC) have substantially reshaped terrestrial carbon dynamics, yet their net impact on carbon sequestration remains uncertain. Here, we use the remote sensing-driven BEPS model and high-resolution HILDA+ data to quantify LCC-induced changes in net ecosystem productivity (NEP) from 1981 to 2019. Despite global forest loss and cropland/urban expansion, LCC led to a net carbon gain of 229 Tg C. Afforestation and reforestation increased NEP by 1559 Tg C, largely offsetting deforestation-driven losses (-1544 Tg C), with newly established forests in the Northern Hemisphere driving gains that counterbalanced emissions from tropical deforestation. Regional carbon gains were concentrated in East Asia, North America, and Europe, while losses occurred mainly in the Amazon and Southeast Asia. Although smaller in area, newly established forests exhibited higher sequestration efficiency than degraded older forests, emphasizing the role of forest age in shaping global carbon sink dynamics. These findings highlight the critical importance of afforestation, forest management, and spatially informed land-use strategies in strengthening carbon sinks and supporting global carbon neutrality goals.

Arable land is a crucial natural resource for human survival and development, which supports food production, ecological services, and material–energy cycling. It is not only an important production resource for agriculture but also a key guarantee for ensuring food security and sustainable agricultural development. Understanding the current utilization of arable land, exploring the spatial–temporal evolution characteristics, and analyzing the driving mechanisms behind its pattern changes are essential for the rational allocation and sustainable utilization of arable land resources. This study focuses on the utilization of arable land in Guangzhou from 2005 to 2018, employing methods such as statistical analysis and spatial econometrics to provide an in-depth analysis of the spatial–temporal distribution characteristics and driving mechanisms of arable land changes. The results show that from 2005 to 2018, the issue of the conversion of arable land to non-agricultural uses was quite severe in Guangzhou, with the primary form being the conversion of arable land into urban residential construction land. Kernel density analysis revealed that non-agriculturization in Guangzhou exhibited spatial clustering, mainly concentrated in areas with lower elevation. Using standard deviation ellipses and centroid migration analysis, it was found that the center of gravity of non-agriculturization in Guangzhou was generally distributed in a southwest–northeast direction, with a more distinct dispersion compared to the northwest–southeast direction. From 2005 to 2010, the rapid increase in the non-agriculturization rate of arable land in Guangzhou was mainly driven by population density and per capita income, both having a positive impact. From 2010 to 2015, the main driving factor shifted to regional GDP. From 2015 to 2018, regional GDP and the value of the tertiary industry became the main driving factors, but unlike the impact of GDP, the tertiary industry exerted a negative influence on non-agriculturization.

Due to the rapid expansion of urban areas, the aging of agricultural labor, and the loss of rural workforce, some regions in China have experienced farmland abandonment. The use of remote sensing technology allows for the rapid and accurate extraction of abandoned farmland, which is of great significance for research on land-using change, food security protection, and ecological and environmental conservation. This research focuses on Qiaotou Town in Chengmai County, Hainan Province, as the study area. Using four high-resolution satellite imagery scenes, digital elevation models, and other relevant data, the random forest classification method was applied to extract abandoned farmland and analyze its spatial distribution characteristics. The accuracy of the results was verified. Based on these findings, the study examines the influence of four factors—irrigation conditions, slope, accessibility, and proximity to residential areas—on farmland abandonment and proposes corresponding governance policies. The results indicate that the accuracy of abandoned farmland extraction using high-resolution satellite imagery is 93.29%. The phenomenon of seasonal farmland abandonment is more prevalent than perennial farmland abandonment in the study area. Among the influencing factors, the abandonment rate decreases with increasing distance from road buffer zones, increases with greater distance from water systems, and decreases with increasing distance from residential areas. Most of the abandoned farmland is located in areas with gentler slopes, which have a relatively smaller impact on farmland abandonment. This study provides valuable references for the extraction of abandoned farmland and for analyzing the abandonment mechanisms in the study area, which have a profound impact on agricultural economic development and help to support the implementation of rural revitalization strategies.

Soil acidification has become a major constraint on agricultural productivity and sustainable land use. Traditional knowledge graph construction methods rely on manual annotation and rule-based extraction, limiting efficiency and scalability. This study proposes a few-shot learning framework based on large language models (LLMs) for automated information extraction and knowledge graph construction in acidified farmland management. Four representative LLMs-ChatGLM-4, LLaMA-3, Mistral-7B, and ERNIE-Bot 4.0-were evaluated. By integrating iterative prompt engineering with ontology-guided graph building, the framework organizes domain knowledge covering acidification causes, remediation measures, and ecological impacts. Experimental results show that the method substantially improves extraction speed, consistency, and scalability, while reducing manual effort and enhancing structural completeness. Nevertheless, the accuracy of extracted knowledge is sensitive to prompt design, suggesting the importance of domain-specific optimization. This study offers a scalable approach to agricultural knowledge graph construction and provides methodological support for intelligent knowledge services in farmland management.

(1) Terrestrial ecosystems are critical carbon sinks, and the accurate assessment of their carbon storage is vital for understanding global carbon cycles and formulating climate change mitigation strategies. (2) This study integrated vegetation indices, meteorological factors, land use data, soil/vegetation types, field sampling, and a convolutional neural network (CNN) model to estimate the carbon storage of terrestrial ecosystems in Guangdong Province. (3) Total carbon storage increased by 0.11 Pg from 2000 to 2021, with vegetation carbon gains (+0.19 Pg) offsetting soil carbon losses (−0.08 Pg), with the latter primarily being driven by reduced soil carbon in forest ecosystems. (4) Northern and eastern Guangdong exhibit high potential for enhancing carbon storage capacity, which is crucial for achieving regional carbon peaking and neutrality targets.

Solely focusing on the agricultural production function of cultivated land resources is not conducive to the various demands for meeting the UN's Sustainable Development Goals. Recognizing the multifunctionality of cultivated land and understanding the interrelationships between individual functions are crucial for the rational planning and utilization of resources. This paper introduces an "element coupling-function synergy" analytical framework for the sustainable utilization of cultivated land resources. The proposed framework is based on the causal relationships between elements and functions within the cultivated land system. Subsequently, a causal Bayesian belief network was constructed to identify trade-offs and synergies among multiple functions of cultivated land resources in Guangdong, China. The findings reveal trade-offs between food cleanliness and food production/social security, and synergistic relationships among food production, social security, and ecological regulation, as well as among ecological regulation, habitat maintenance and landscape culture. The study area was divided into eight functional zones: Green Agricultural Zone, Agro-inputs Control Zone, Urban Agricultural Zone, Major Grain-producing Zone, Modern Agricultural Zone, Agro-ecological Preservation Zone, Agro-ecological Tourism Zone, Quality Improvement Zone. Multi-objective management plans were formulated for optimizing multifunctional relationships within each zone. The analysis result reveals the importance of nutrient conditions and ecological environments for the sustainable management of cultivated land. Consequently, specific policy recommendations are proposed accordingly. This paper may not only advance understanding of the multifunctionality of cultivated land but can also provide valuable insights for land-use planning to ensure the judicious and sustainable management of cultivated land resources.

Aiming at the challenges of low detection accuracy, susceptibility to complex background interference, difficulty in detecting small objects, and multi-scale object issues in aerial images, our proposed an improved YOLOv8-based object detection algorithm, named YOLO-RLDW. Leveraging the advantages of Receptive Field Attention Convolution (RFAConv), we designed a feature extraction module named C2f-RFA to enhance the feature extraction capability for small objects in aerial images. Inspired by the concept of Large Separable Kernel Attention (LSKA), we developed the SPPF-LSKA module, which effectively reduces the interference of aerial backgrounds in object detection. We replaced the YOLOv8 detection head with a Dynamic Head (DyHead), further enhancing the model’s generalization and adaptability. Finally, we employed as boundary box regression loss based on a dynamic focusing mechanism, WIoU, as the loss function, which accelerates model convergence while improving the localization capability for multi-scale objects. Experimental results demonstrate that on the VisDrone2021 dataset, the proposed algorithm achieves improvements of 5.5%, 3.9%, 5.4%, and 3.7% in precision (P), recall (R), mean average precision (mAP50), and mAP95, respectively, compared to the original algorithm. On our self-built remote sensing image dataset RSI, the accuracy, recall, and mean average precision reach 94.2%, 91.0%, and 95.4%, respectively, demonstrating good performance in detecting objects in aerial images. Comparison with other mainstream object detection algorithms validates the effectiveness and superiority of the proposed method.

Optimal phosphorus (P) levels in lateritic soils are key for sustainable crop production. However, the effect of various fertilizers on soil phosphorus pools, crop phosphorus uptake and crop yields remains unclear. This study investigated the effect of different fertilizer application strategies on plant growth and soil P fractions and determined the contribution of biotic and abiotic factors to insoluble P release. We found that resin-P, NaHCO3-P and NaOH-P represented the primary active P pools in the lateritic soil, contributing 59.8% of the total P in conventional fertilizer application (CF), with Fe/Al-bound P (NaOH-P) being 42.1% of the active P pool. Combining Nangbowang (NBW), a microbial organic fertilizer inoculated with Bacillus subtilis (>= 2 x 10(7) million CFU g(-1)), with reduced chemical fertilizer (NBW + CR) increased soil P availability and promoted the release of Fe/Al-bound P and residual-P, decreasing the Fe/Al-bound P to 21.7%. Soil biological factors mainly influenced the P transformation process. With the consumption of soil active P, NBW bio-organic fertilizer enhanced the niche filtration of P-solubilizing bacterial communities (Gemmatimonadetes, Sphingomonas and Halomonas), altered the soil functional microbial community structure and promoted P form conversion. The NBW + CR treatment also enhanced nitrogen and P nutrient uptake by pepper plants (Capsicum annuum), with increased total P concentrations in pepper fruit and stem, and improved crop yield. NBW increased active P concentrations in the soil and promoted Fe/Al-P release and transformation by impacting autochthonous microbes involved in the conversion of P chemical species. These results can guide the improvement of P availability and release in lateritic red soils using bio-organic fertilizer.

Cropland is a comprehensive system influenced by the natural environment and human activities. This article collects the data of cropland use, soil, and other geographic, social and economic factors in the study area and then uses the methods of system analysis, induction and deduction to propose a new research perspective for establishing a cognitive framework and analyzing cropland resources and their functions. The framework is used to assess the rapidly urbanizing region of Guangzhou and investigate the production, ecological, and living functions provided by cropland resources. Synergistic relationships between functions are analyzed using the hot and cold spot methods. The results indicate that the production function of cropland resources in Guangzhou is good, the ecological function is favorable, and the living function is relatively low. A synergistic relationship between the three functions is observed in 91% of areas of Guangzhou, whereas a balanced relationship occurs in some areas of the southern part of Zengcheng, the northwestern and northeastern parts of Conghua, and the western part of Nansha. This research provides guidance for managing cropland resources and ensuring their sustainable utilization.

Current shale gas well production capacity predictions primarily rely on analytical and numerical simulation methods, which necessitate extensive calculations and manual parameter tuning and produce lowly accurate predictions. Although employing neural networks yields highly accurate predictions, they can easily fall into local optima. This paper suggests a new way to use Cuckoo Search (CS)-optimized neural networks to make shale gas well production capacity predictions more accurate and to solve the problem of local optima. It aims to assist engineers in devising more effective development plans and production strategies, optimizing resource allocation, and reducing risk. The method first analyzes the factors influencing the production capacity of shale gas wells in a block located in western China through correlation coefficients. It identifies the main factors affecting the gas test absolute open flow as organic carbon content, small-layer passage rate, fracture pressure, acid volume, pump-in fluid volume, brittle mineral content in the rock, and rock density. Subsequently, we used the CS algorithm to conduct the global training of the neural network, avoiding the problem of local optima, and established a neural network model for predicting shale gas well production capacity optimized by the CS algorithm. A comparative analysis with other relevant methods demonstrates that the CS-optimized neural network model can accurately predict production capacity, enabling a more rational and effective exploitation of shale gas resources, which lower development costs and increase the economic returns of oil and gas fields. Compared to numerical simulation, SVM, and BP neural network algorithms, the CS-optimized BP neural network (CS-BP) exhibits significantly lower prediction error. Its correlation coefficient between predicted and actual values reaches as high as 0.9924. Verification experiments conducted on another shale gas well also demonstrate that, in comparison to the BP neural network algorithm, CS-BP offers superior prediction performance, with model validation showing a prediction error of only 0.05. This study can facilitate more rational and efficient exploitation of shale gas resources, reduce development costs, and enhance the economic benefits of oil and gas fields.

High nature value farmland (HNVf) plays an important role in improving biodiversity and landscape heterogeneity, and it is effective in curbing soil non-point source pollution and carbon loss in sustainable eco-agricultural systems. To this end, we developed high-resolution (2 m × 2 m) indicators for the identification of potential HNVf based on GF1B remote sensing imaging, including the land cover (LC), normalized difference vegetation index (NDVI), Shannon diversity (SH), and Simpsons index (SI). The statistical results for LC with high resolution (2 m × 2 m) showed that there was 41.05% of intensive farmland in the study area, and the pixel proportion of the HNVf map (above G3) was 44.30%. These HNVf patches were concentrated in the transition zone around the edge of the intensive farmland and around rivers, with characteristics of HNVf type 2 being significantly reflected. Among the real-life areas from Map World, elements (i.e., linear forests, rivers, and semi-natural vegetation etc.) of HNVf accounted for more than 70% of these regions, while a field survey based on potential HNVf patches also exhibited significant HNVf characteristics in comparison with intensive farmlands. In addition, from 2002 to 2020, the total migration distance of the gravity center of intensive farmland in the study area was 7.65 km. Moreover, four landscape indices (patch COH index, landscape division index, SH, and SI) slowly increased, indicating that the species richness and biodiversity were improved. It was also found that a series of ecological protection policies provide effective guarantees for an improvement in species diversity and the development of HNVf in the study area. In particular, the average contents of As, Cr, Cu, Ni, and Zn in the HNVf were 20.99 mg kg−1, 121.11 mg kg−1, 21.97 mg kg−1, 29.34 mg kg−1, and 41.68 mg kg−1, respectively, which were lower in comparison with the intensive farmland soil. This is the first HNVf exploration for landscape and soil pollution assessment in a coastal delta in China, and could provide powerful guidance for the ecological protection of farmland soil and the high-quality development of sustainable agriculture.

The ecological restoration and protection of territorial space is a systematic project for the protection and restoration of ecosystems damaged or degraded by human disturbance. Effectiveness evaluation is of great significance to the optimization, adjustment, and sustainability of ecological restoration and protection. Current research and practices tend to focus on a single element and the site scale. Based on the study on the connotation of ecological restoration and protection of territorial space and the multi-scale characteristics of ecosystems, we constructed a multi-scale effectiveness evaluation system for ecological restoration and protection of territorial space and a full-cycle monitoring system for effectiveness evaluation. The multi-scale effectiveness evaluation system consisted of the regional/watershed scale, the protection and restoration unit scale, and the sub-project scale. The full-cycle monitoring system contained the basic information investigation system before construction, the construction monitoring system during construction, and the multi-scale effectiveness evaluation system after construction. At the regional/watershed scale, structure, quality, and services of ecosystem were concerned and remote sensing was used as the main method to capture data. At the protection and restoration unit scale, ecological stress factors, ecosystem quality and services were concerned, and the main methods were remote sensing combined with field survey. At the sub-project scale, engineering technology measures were concerned and the field survey was used as the main method. In the implementation of the multi-scale effectiveness evaluation, it would be necessary to focus on and solve the key issues including the spillover effect, transmission mechanism, and potential impact of ecological restoration.

Rapid and accurate agricultural land evaluation provides essential guidance for the supervision and allocation of agricultural land resources; it also helps to ensure food security. Previous work has mainly evaluated the land quality at the county level by using field sampling data and based on a factor approach. However, it is difficult to achieve uniform, large-scale agricultural land evaluation via conventional approaches because of its spatial heterogeneity, as well as the large temporal and economic costs associated with data acquisition. In this study, we integrated publicly available multimodal data (i.e., satellite remote sensing, environmental, and socioeconomic data) into the Google Earth Engine (GEE) platform, selected the best indicators from each modality using the geodetector, on the basis of which different combinations of input models were designed. And then we developed machine learning (random forest, RF) and deep learning (deep neural network, DNN) models to evaluate the land quality in paddy field and dry land systems in 2013 throughout Guangdong Province, China. The results showed that the performance of our combination of variables decreased in the following order: multimodal > bimodal > unimodal. With the best input combination, the RF model (R-2 = 0.91, RMSE = 97.56, and CCC = 0.95) outperformed the DNN model (R-2 = 0.89, RMSE = 108.72, and CCC = 0.94) in terms of predicting the quality of paddy field. The RF model (R-2 = 0.90, RMSE = 104.27, and CCC = 0.95) also out-performed the DNN model (R-2 = 0.86, RMSE = 124.38, and CCC = 0.93) in terms of predicting the quality of dry land. The agricultural land quality estimates obtained using the RF and DNN models were more accurate for paddy field than for dry land systems because of greater land quality homogeneity in paddy fields. This research proposed a simple, low-cost for rapid and accurate agricultural land evaluation at the provincial scale using publicly available multimodal data, which can help to achieve control of the agricultural land grade at multiple spatial and temporal scales.

The rapid development of industrialization and urbanization has posed serious challenges for coastal farmland ecosystems. Source apportionment of soil heavy metals is an effective way for the detection of non-point source pollution in farmland to help support the high-quality development of coastal agriculture. To this end, 113 surface soil samples were collected in the coastal delta of China, and the contents of As, Cd, Cr, Cu, Ni, Pb, and Zn were determined. A variety of models were integrated to apportion the source of soil heavy metals, including positive matrix factorization (PMF), geographical detector (GD), eXtreme gradient boosting (XGBoost), and structural equation modeling (SEM). The result of PMF models revealed that there was collinearity between various heavy metals, and the same heavy metal may have a mixed source. The XGBoost model analysis indicated that there were significant non-linear relationships between soil heavy metals and source factors. A synergy between air quality and human activity factors was the key source of heavy metal that entered the study area, based on the results of the GD. Furthermore, the input path effect of heavy metals in the soil of the study area was quantified by SEM. The balance of evidence from the above models showed that air quality (SO2 and NO2) and factories in the study area had the greatest impacts on Cd, Cr, and Zn. Natural sources were dominant for Pb, while As, Cu, and Ni were contributed by soil parent material and factories. The above results led to the conclusion that there was a cycle path in the study area that continuously promoted the migration and accumulation of heavy metals in farmland soil; that is, the heavy metals discharged during oil exploitation and smelting entered the atmosphere and then accumulated in the farmland soil through precipitation, atmospheric deposition, and other paths. In this study, it is shown that a variety of models can be used to more comprehensively assess the sources of soil heavy metals. This approach can provide effective support for the rapid prevention and decision-making management of soil heavy metal pollution in coastal areas.