Four types: single holistic, single embedded, multiple holistic, multiple embedded
The post-positive paradigm postulates there is one reality that can be objectively described and understood by “bracketing” oneself from the research to remove prejudice or bias. 27 Yin focuses on general explanation and prediction, emphasizing the formulation of propositions, akin to hypothesis testing. This approach is best suited for structured and objective data collection 9 , 11 and is often used for mixed-method studies.
Constructivism assumes that the phenomenon of interest is constructed and influenced by local contexts, including the interaction between researchers, individuals, and their environment. 27 It acknowledges multiple interpretations of reality 24 constructed within the context by the researcher and participants which are unlikely to be replicated, should either change. 5 , 20 Stake and Merriam’s constructivist approaches emphasize a story-like rendering of a problem and an iterative process of constructing the case study. 7 This stance values researcher reflexivity and transparency, 28 acknowledging how researchers’ experiences and disciplinary lenses influence their assumptions and beliefs about the nature of the phenomenon and development of the findings.
A key tenet of case study methodology often underemphasized in literature is the importance of defining the case and phenomenon. Researches should clearly describe the case with sufficient detail to allow readers to fully understand the setting and context and determine applicability. Trying to answer a question that is too broad often leads to an unclear definition of the case and phenomenon. 20 Cases should therefore be bound by time and place to ensure rigor and feasibility. 6
Yin 22 defines a case as “a contemporary phenomenon within its real-life context,” (p13) which may contain a single unit of analysis, including individuals, programs, corporations, or clinics 29 (holistic), or be broken into sub-units of analysis, such as projects, meetings, roles, or locations within the case (embedded). 30 Merriam 24 and Stake 5 similarly define a case as a single unit studied within a bounded system. Stake 5 , 23 suggests bounding cases by contexts and experiences where the phenomenon of interest can be a program, process, or experience. However, the line between the case and phenomenon can become muddy. For guidance, Stake 5 , 23 describes the case as the noun or entity and the phenomenon of interest as the verb, functioning, or activity of the case.
Yin’s approach to a case study is rooted in a formal proposition or theory which guides the case and is used to test the outcome. 1 Stake 5 advocates for a flexible design and explicitly states that data collection and analysis may commence at any point. Merriam’s 24 approach blends both Yin and Stake’s, allowing the necessary flexibility in data collection and analysis to meet the needs.
Yin 30 proposed three types of case study approaches—descriptive, explanatory, and exploratory. Each can be designed around single or multiple cases, creating six basic case study methodologies. Descriptive studies provide a rich description of the phenomenon within its context, which can be helpful in developing theories. To test a theory or determine cause and effect relationships, researchers can use an explanatory design. An exploratory model is typically used in the pilot-test phase to develop propositions (eg, Sibbald et al. 31 used this approach to explore interprofessional network complexity). Despite having distinct characteristics, the boundaries between case study types are flexible with significant overlap. 30 Each has five key components: (1) research question; (2) proposition; (3) unit of analysis; (4) logical linking that connects the theory with proposition; and (5) criteria for analyzing findings.
Contrary to Yin, Stake 5 believes the research process cannot be planned in its entirety because research evolves as it is performed. Consequently, researchers can adjust the design of their methods even after data collection has begun. Stake 5 classifies case studies into three categories: intrinsic, instrumental, and collective/multiple. Intrinsic case studies focus on gaining a better understanding of the case. These are often undertaken when the researcher has an interest in a specific case. Instrumental case study is used when the case itself is not of the utmost importance, and the issue or phenomenon (ie, the research question) being explored becomes the focus instead (eg, Paciocco 32 used an instrumental case study to evaluate the implementation of a chronic disease management program). 5 Collective designs are rooted in an instrumental case study and include multiple cases to gain an in-depth understanding of the complexity and particularity of a phenomenon across diverse contexts. 5 , 23 In collective designs, studying similarities and differences between the cases allows the phenomenon to be understood more intimately (for examples of this in the field, see van Zelm et al. 33 and Burrows et al. 34 In addition, Sibbald et al. 35 present an example where a cross-case analysis method is used to compare instrumental cases).
Merriam’s approach is flexible (similar to Stake) as well as stepwise and linear (similar to Yin). She advocates for conducting a literature review before designing the study to better understand the theoretical underpinnings. 24 , 25 Unlike Stake or Yin, Merriam proposes a step-by-step guide for researchers to design a case study. These steps include performing a literature review, creating a theoretical framework, identifying the problem, creating and refining the research question(s), and selecting a study sample that fits the question(s). 24 , 25 , 36
Using multiple data collection methods is a key characteristic of all case study methodology; it enhances the credibility of the findings by allowing different facets and views of the phenomenon to be explored. 23 Common methods include interviews, focus groups, observation, and document analysis. 5 , 37 By seeking patterns within and across data sources, a thick description of the case can be generated to support a greater understanding and interpretation of the whole phenomenon. 5 , 17 , 20 , 23 This technique is called triangulation and is used to explore cases with greater accuracy. 5 Although Stake 5 maintains case study is most often used in qualitative research, Yin 17 supports a mix of both quantitative and qualitative methods to triangulate data. This deliberate convergence of data sources (or mixed methods) allows researchers to find greater depth in their analysis and develop converging lines of inquiry. For example, case studies evaluating interventions commonly use qualitative interviews to describe the implementation process, barriers, and facilitators paired with a quantitative survey of comparative outcomes and effectiveness. 33 , 38 , 39
Yin 30 describes analysis as dependent on the chosen approach, whether it be (1) deductive and rely on theoretical propositions; (2) inductive and analyze data from the “ground up”; (3) organized to create a case description; or (4) used to examine plausible rival explanations. According to Yin’s 40 approach to descriptive case studies, carefully considering theory development is an important part of study design. “Theory” refers to field-relevant propositions, commonly agreed upon assumptions, or fully developed theories. 40 Stake 5 advocates for using the researcher’s intuition and impression to guide analysis through a categorical aggregation and direct interpretation. Merriam 24 uses six different methods to guide the “process of making meaning” (p178) : (1) ethnographic analysis; (2) narrative analysis; (3) phenomenological analysis; (4) constant comparative method; (5) content analysis; and (6) analytic induction.
Drawing upon a theoretical or conceptual framework to inform analysis improves the quality of case study and avoids the risk of description without meaning. 18 Using Stake’s 5 approach, researchers rely on protocols and previous knowledge to help make sense of new ideas; theory can guide the research and assist researchers in understanding how new information fits into existing knowledge.
Columbia University has recently demonstrated how case studies can help train future health leaders. 41 Case studies encompass components of systems thinking—considering connections and interactions between components of a system, alongside the implications and consequences of those relationships—to equip health leaders with tools to tackle global health issues. 41 Greenwood 42 evaluated Indigenous peoples’ relationship with the healthcare system in British Columbia and used a case study to challenge and educate health leaders across the country to enhance culturally sensitive health service environments.
An important but often omitted step in case study research is an assessment of quality and rigour. We recommend using a framework or set of criteria to assess the rigour of the qualitative research. Suitable resources include Caelli et al., 43 Houghten et al., 44 Ravenek and Rudman, 45 and Tracy. 46
Although “pragmatic” case studies (ie, utilizing practical and applicable methods) have existed within psychotherapy for some time, 47 , 48 only recently has the applicability of pragmatism as an underlying paradigmatic perspective been considered in HSR. 49 This is marked by uptake of pragmatism in Randomized Control Trials, recognizing that “gold standard” testing conditions do not reflect the reality of clinical settings 50 , 51 nor do a handful of epistemologically guided methodologies suit every research inquiry.
Pragmatism positions the research question as the basis for methodological choices, rather than a theory or epistemology, allowing researchers to pursue the most practical approach to understanding a problem or discovering an actionable solution. 52 Mixed methods are commonly used to create a deeper understanding of the case through converging qualitative and quantitative data. 52 Pragmatic case study is suited to HSR because its flexibility throughout the research process accommodates complexity, ever-changing systems, and disruptions to research plans. 49 , 50 Much like case study, pragmatism has been criticized for its flexibility and use when other approaches are seemingly ill-fit. 53 , 54 Similarly, authors argue that this results from a lack of investigation and proper application rather than a reflection of validity, legitimizing the need for more exploration and conversation among researchers and practitioners. 55
Although occasionally misunderstood as a less rigourous research methodology, 8 case study research is highly flexible and allows for contextual nuances. 5 , 6 Its use is valuable when the researcher desires a thorough understanding of a phenomenon or case bound by context. 11 If needed, multiple similar cases can be studied simultaneously, or one case within another. 16 , 17 There are currently three main approaches to case study, 5 , 17 , 24 each with their own definitions of a case, ontological and epistemological paradigms, methodologies, and data collection and analysis procedures. 37
Individuals’ experiences within health systems are influenced heavily by contextual factors, participant experience, and intricate relationships between different organizations and actors. 55 Case study research is well suited for HSR because it can track and examine these complex relationships and systems as they evolve over time. 6 , 7 It is important that researchers and health leaders using this methodology understand its key tenets and how to conduct a proper case study. Although there are many examples of case study in action, they are often under-reported and, when reported, not rigorously conducted. 9 Thus, decision-makers and health leaders should use these examples with caution. The proper reporting of case studies is necessary to bolster their credibility in HSR literature and provide readers sufficient information to critically assess the methodology. We also call on health leaders who frequently use case studies 56 – 58 to report them in the primary research literature.
The purpose of this article is to advocate for the continued and advanced use of case study in HSR and to provide literature-based guidance for decision-makers, policy-makers, and health leaders on how to engage in, read, and interpret findings from case study research. As health systems progress and evolve, the application of case study research will continue to increase as researchers and health leaders aim to capture the inherent complexities, nuances, and contextual factors. 7
An official website of the United States government
The .gov means it’s official. Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.
The site is secure. The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.
Email citation, add to collections.
Your saved search, create a file for external citation management software, your rss feed.
Affiliations.
Background: Public health surveillance involves the collection, analysis and dissemination of data to improve population health. The main sources of data for public health decision-making are surveys, typically comprised of self-report which may be subject to biases, costs and delays. To complement subjective data, objective measures from sensors could potentially be used. Specifically, advancements in personal mobile and wearable technologies enable the collection of real-time and continuous health data.
Objective: In this context, the goal of this work is to apply a mobile health platform (MHP) that extracts health data from the Apple Health repository to collect data in daily-life scenarios and use it for the prediction of stress, a major public health issue.
Methods: A pilot study was conducted with 45 participants over 2 weeks, using the MHP to collect stress-related data from Apple Health and perceived stress self-reports. Apple, Withings and Empatica devices were distributed to participants and collected a wide range of data, including heart rate, sleep, blood pressure, temperature, and weight. These were used to train random forests and support vector machines. The SMOTE technique was used to handle imbalanced datasets.
Results: Accuracy and f1-macro scores were in line with state-of-the-art models for stress prediction above 60% for the majority of analyses and samples analysed. Apple Watch sleep features were particularly good predictors, with most models with these data achieving results around 70%.
Conclusions: A system such as the MHP could be used for public health data collection, complementing traditional self-reporting methods when possible. The data collected with the system was promising for monitoring and predicting stress in a population.
Keywords: Apple Health; Public health; app; heart rate; mHealth; machine learning; sleep; stress.
© The Author(s) 2024.
PubMed Disclaimer
The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Study protocol.
MHP interface.
Division into datasets per device…
Division into datasets per device and per device + sleep.
Training generalised models.
NCBI Literature Resources
MeSH PMC Bookshelf Disclaimer
The PubMed wordmark and PubMed logo are registered trademarks of the U.S. Department of Health and Human Services (HHS). Unauthorized use of these marks is strictly prohibited.
Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.
Scientific Reports volume 14 , Article number: 21783 ( 2024 ) Cite this article
Metrics details
As the demand for regional geological disaster risk assessments in large cities continues to rise, our study selected Hangzhou, one of China’s megacities, as a model to evaluate the susceptibility to two major geological hazards in the region: ground collapse and ground subsidence. Given that susceptibility assessments for such disasters mainly rely on knowledge-driven models, and data-driven models have significant potential for application, we proposed a high-accuracy Random Forest—Back Propagation Neural Network Coupling Model. By using nine evaluation factors selected based on field surveys and expert recommendations, along with disaster data, the model's predictive results indicate a 3–40% improvement in model performance metrics such as AUC, accuracy, precision, recall, and F1-score, compared to single models and traditional SVM and logistic regression models. Ultimately, using the predictive results of this model, we created susceptibility maps for individual disasters and developed a muti-hazards susceptibility map by employing the expert weight discrimination method and the overlay evaluation method. Furthermore, we discussed the feature importance in the prediction process. Our study validated the feasibility of using advanced machine learning models for urban geological disaster assessment, providing a replicable template for other cities.
With the continuous expansion of urban construction, the conflict between urban geological disasters and urban development has become increasingly prominent 1 , 2 . The prevention and management of geological disasters have thus become critical considerations for every city 3 . Hangzhou, one of China’s megacities, boasts a population exceeding 10 million and achieved a GDP of $259 billion in 2022 4 . However, due to its location in the Hangjiahu Plain and ongoing human-induced modifications to the geological environment 5 , 6 , issues like land subsidence and ground collapse in Hangzhou’s main urban areas have emerged as significant challenges during its development. Specifically, subsidence primarily results from extensive urban development and the prevalent distribution of Quaternary soft soil layers, which are susceptible to compression under structural loads. In contrast, collapses are mainly triggered by the presence of artificial fill and the failure of underground pipelines. Like many cities, Hangzhou has started to undertake regional geological disaster assessments to ensure urban geological safety. The assessment and prediction of these disasters necessitate susceptibility evaluation, as it reflects geological conditions and enables reasonable zoning of the study area 7 , 8 , 9 , 10 . However, assessments based solely on single disaster types might not accurately represent the geological safety of an area, especially as more cities begin to emphasize the overall impact of regional geological issues. Therefore, comprehensive evaluations encompassing multiple types of disasters, or multi-hazards, are commonly employed 11 , 12 , 13 . Current assessments, especially those concerning subsidence and collapse, primarily rely on knowledge-driven models such as the Analytic Hierarchy Process (AHP), which entails subjective scoring of selected indicators to determine their weights for susceptibility analysis 14 , 15 , 16 , Particularly for collapse-related disasters, which occur over small areas, machine learning-based susceptibility assessments of such disasters are, to our knowledge, currently an unexplored area. Meanwhile, in the context of comprehensive disaster evaluations, experts commonly employ AHP to score and determine weights 17 , 18 , 19 , when assessing the relative weights of different types of disasters within a specific area, the lack of learning samples means that reliance on expert judgment is the only viable option.
With advancements in technology and the growing complexity of geological disaster conditions, the limitations of the AHP are becoming increasingly apparent 20 . Its reliance on expert judgment for weighting factors is inherently subjective and can be influenced by factors such as emotions, fatigue, among others, particularly with numerous evaluation criteria. Consequently, researchers are increasingly turning to data-driven models. Unlike traditional models, data-driven models are capable of handling more complex datasets, thereby improving prediction accuracy. They have been extensively applied in evaluating susceptibility to large-scale geological disasters, such as landslides and earthquakes 21 , 22 , 23 , 24 , 25 . Recently, the integration of coupling machine learning models, such as the combination of BP neural networks with SVM models, has demonstrated potential in optimizing predictions by leveraging each model’s strengths and mitigating single-model flaws, thus achieving higher accuracy in case studies 26 , 27 . However, there is still significant potential for development in the field of ensemble models, particularly in urban geological disasters. When facing datasets that include complex geological environments and human activity factors, enhancing the predictive performance of models is crucial for the accuracy of prediction results.
Our study assesses the susceptibility of multi-hazards, including ground subsidence and ground collapse, in commonly filled and silty areas in Hangzhou by employing Random Forest - Back Propagation Neural Network coupling model. We explored the feasibility of utilizing coupled models over single models to enhance evaluation effectiveness, particularly providing a viable template for applying machine learning to small-scale, collapse-type disasters. Our research flowchart is illustrated in Fig. 1 .
Research flowchart.
Selection of the study area.
The silty clay soil in Hangzhou City is primarily distributed in the plain areas. Due to the topographic and geomorphological characteristics, the plain areas with accumulative, alluvial, and marine plains are the most typical regions. Considering the comprehensive disaster situation, this paper selects the filled soil-silty clay typical area along the south bank of the Qiantang River in Binjiang District as the study area. Covering an area of 31,970,000m 2 (31.97 km 2 ), it includes urban arterials such as subways, expressways, and elevated roads, marking it as a region in Hangzhou City with significant human activity modifications. The study area is located along the western edge of the Xiaoshao Plain, north to the Hangzhou duplex hill, and south to the Puyang River Plain. The terrain is dominated by plains with a few low hills, featuring monotonous geomorphological types with clear boundaries, flat terrain, a dense river network, and mainly consists of sandy silty soil and silt, with the sand and gravel layer buried at a depth of about 35–50 m. Human activities have profoundly affected the area, with high-rise buildings typically using sand-gravel layers or bedrock as the bearing stratum. The sediments mainly consist of gray to dark gray silty mud clay and silty fine clay, characterized by distinct horizontal stratification and high-water content. The Quaternary strata mainly belong to the lower part of the Holocene, formed during the early Fuyang marine transgression, primarily through flood alluviation, often appearing as river valley plains, river terraces, and other landforms. The strata are composed of sand and gravel, with good sorting and rounding, relatively loose structure, and a thickness of 2–12 m. The geographical location of the study area is shown in Fig. 2 .
Copyright © 2020 Esri and its licensors. All rights reserved. Esri, NASA, NGA, USGS, Sources: Esri, USGS, Esri, © OpenStreetMap contributors, TomTom, Garmin, FAO, NOAA, USGS.)
Location of the study area (This figure was created using ArcGIS Pro software. Desktop GIS Software | Mapping Analytics | ArcGIS Pro (esri.com)) (Map image is the intellectual property of Esri and is used herein under license.
Considering the complexity of the urban geological environment and the unique characteristics of the study area, it is imperative to select evaluation factors that comprehensively consider the intensity of human activities in the area and conduct in-depth analyses in conjunction with geological conditions and cultural characteristics. To determine which influencing factors are most closely associated with ground collapse and subsidence in Hangzhou, we collaborated with experts from the Hangzhou urban geological safety assessment at the Zhejiang Geological Survey.
Based on field inspections and expert analyses, ground collapses in the study area are primarily attributed to the following factors: geological conditions such as artificial fill, sandy loam, and concealed ditches; deficiencies in underground drainage pipe structures; hydrogeological influences; and disruptions caused by human engineering activities. Meanwhile, ground subsidence is mainly due to the study area being in a plain, with the distribution of soft soil layers, human activities, and hydrological factors being the primary reasons for ground subsidence occurrences. Given these factors, we identified 9 key factors as preliminary evaluation indicators and conducted a correlation analysis to prevent issues such as overfitting or frequent misjudgments in the machine learning process, as depicted in the correlation heatmap (Fig. 3 ). The evaluation factors used for the disaster, data sources, and data attributes are detailed in Table 1 and Fig. 4 .
Correlation heatmap of evaluation factors.
Assessment factors used in our research.
To further enhance model accuracy and ensure that when predicting ground collapse disasters, the small scope of disaster areas leading to fewer data points does not impact the prediction results, we propose the Random Forest - Back Propagation Neural Network coupling model. Random Forest represents a collaborative approach in machine learning, comprising numerous decision trees working in unison 28 , 29 . Each tree receives a randomized subset of data and features, enhancing the model’s overall accuracy and versatility. This approach excels at managing a diverse array of inputs and discerning obscure patterns. The operation of Random Forest can be concisely expressed using the following equations to enhance the understanding of its ensemble methodology:
where \(Var(S)\) represents the variance of the target variable in the entire dataset \(S\) , \(|{S}_{left}|\) and \(|{S}_{right}|\) denote the number of samples in the left and right subsets post-split, respectively. Equation ( 1 ) highlights how Random Forest effectively reduces overfitting by incorporating diverse data samples.
where \({p}_{i}^{2}\) indicates the proportion of the samples in set \(S\) that belong to class \(i\) , and \(n\) is the total number of classes. This measure is critical in determining the best split at each node within the trees. The Eq. ( 2 ) not only allows Random Forest to handle a variety of input types effectively but also enhances its capability to detect subtle patterns in complex datasets.
Conversely, the Backpropagation Neural Network is a pivotal component in machine learning, particularly adept at complex tasks that defy linear analysis 30 , 31 . It comprises multiple layers and enhances its predictive capability through iterative adjustments of its strategy (weights and biases). To further elucidate the learning process within a Backpropagation Neural Network, the following equations can be considered:
where \({W}_{ij}^{\left(old\right)}\) and \({W}_{ij}^{\left(new\right)}\) are the old and new values of the weight between nodes \(i\) and \(j\) , \(\eta\) is the learning rate, and \(\frac{\partial \mathcal{L}}{\partial {W}_{ij}}\) represents the gradient of the loss function \(\mathcal{L}\) with respect to the weight \({W}_{ij}\) Eq. ( 3 ) delineates the core mechanism by which the network learns by iteratively adjusting its weights.
where \(\sigma \left(x\right)\) is the sigmoid activation function, which normalizes the input \(x\) into an output range between 0 and 1, enabling the network to handle non-linear relationships within the data. Equation ( 4 ) illustrates how the sigmoid activation function enables the neural network to transform linear inputs into outputs bounded between 0 and 1, facilitating the handling of probabilistic decisions and non-linear complexities.
To optimize our Random Forest Classifier, we conducted a grid search exploring various hyperparameters 32 , including the number of trees (ranging from 50 to 200 in increments of 50) and max features (options including 'auto', 'sqrt', and 'log2'). The optimal configuration was determined to be 100 trees, ensuring repeatability with a fixed random seed of 42. Similarly, for the Neural Network Classifier, our grid search covered different network architectures, focusing on varying the number of neurons in the hidden layers (options including 50, 100, and 150 for the first layer and 25, 50, and 75 for the second layer) and the maximum number of iterations (500, 1000, and 1500 iterations were tested). The most effective configuration found consists of two hidden layers with 100 and 50 neurons respectively, with a maximum of 1000 iterations, again using the random seed of 42 to maintain consistency. Each set of parameters was evaluated using 5-fold cross-validation 33 , ensuring that our selection process was robust and the results reliable. This comprehensive approach to hyperparameter tuning enhances the individual models' predictive accuracy before integrating them into the Stacking Classifier. The Stacking Classifier utilizes these two models as base models and integrates their predictions using another Random Forest configured with 100 trees as the final meta-model to make the ultimate decision. Subsequently, the Backpropagation Neural Network refines the analysis with its specialized capabilities. Together, they constitute a robust model adept at assimilating extensive information, distilling it, and producing precise predictions. The principle of the ensemble model is depicted in Fig. 5 .
Schematic diagram of RF-BP neural network coupling model structure.
Model performance analysis.
In this study, we selected 27898 data points using the "Raster to Point" function in ArcGIS, the sample size was determined by dividing the study area and evaluation factors into 32m x 32m pixels. Under these sample conditions, the dataset maintained relative balance, the computation time was reasonable, and subsequent model evaluation metrics also demonstrated the high accuracy of the model predictions. For ground subsidence susceptibility evaluation, susceptible areas were delineated using regional cumulative subsidence data. To ensure model training accuracy, during the ground subsidence susceptibility evaluation, 70% of the total data were randomly extracted for the training set using Python software. For ground collapse susceptibility evaluation, a 1:1 ratio of ground collapse disaster points to non-disaster points was employed to segment the training set (with disaster points labeled as 1 and non-disaster points as 0). Due to the limited scope of collapse areas, we obtained a total of 300 collapse data points, from which we randomly selected 210 collapse points and 210 non-collapse points to serve as the training set. This balanced approach helps to ensure that the model is not biased towards the more prevalent non-collapse conditions. The trained model was subsequently used to predict the remaining points, providing the probability of ground collapse for each pixel (ranging from 0 to 1) and the predicted cumulative subsidence value for each pixel.
To further demonstrate that our Random Forest-Back Propagation Neural Network Coupling Model outperforms individual models and ensures optimal results for susceptibility mapping, we evaluated the model's performance using multiple metrics. Since ground collapse prediction is a binary classification problem in our multi-hazard assessment, we primarily used metrics including ROC curves, AUC values, Accuracy, Precision, Recall, and F1-score.
The AUC (Area Under the ROC Curve) represents the model's overall performance across all classification thresholds, with a higher AUC indicating superior classification capability 34 . Accuracy measures the proportion of correctly predicted instances (true positives and true negatives) out of the total instances, reflecting the model's general effectiveness. Precision evaluates the proportion of true positive predictions among all predicted positives, indicating the model's reliability in predicting positive instances. Recall assesses the proportion of actual positive instances correctly identified by the model, demonstrating its ability to capture positive cases. The F1-score, which is the harmonic mean of precision and recall, provides a balanced measure of both metrics, especially useful for imbalanced datasets. Higher values for these metrics signify better model performance on the given dataset 35 .
Results showed that the stacked model outperformed others in all metrics in ground collapse, as illustrated in Fig. 6 a and b. Following the validation of the stacked model's performance, we employed it to predict susceptibility to non-binary tasks of ground subsidence. The results yielded high AUC values: 0.99 for less than 24mm (low susceptibility), 0.99 for 24mm-64mm (moderate susceptibility), 0.97 for 64mm-96mm (high susceptibility), and 0.99 for 96mm-128mm (very high susceptibility), indicating the model’s high predictive accuracy. The average AUC value of over 0.98 confirms the model's accuracy in predicting ground subsidence in our study and underscores its potential application for non-binary classification tasks. The specific ROC curves and AUC values can be seen in Fig. 6 c.
( a ) The ROC curves and AUC values in the ground collapse prediction task; ( b ) The Accuracy, Precision, Recall, and F1-score in the ground collapse prediction task; ( c ) The ROC curves and AUC values of the stacking model in the ground subsidence prediction task.
After the model’s prediction was completed, the RF-BP Neural Network Model’s results were used as a benchmark to import all grid points into ArcGIS 10.8 for evaluating the susceptibility zones of ground collapses and ground subsidence. First, regarding single-hazard susceptibility maps, for ground collapse, the probability of occurrence in the study area (ranging from 0 to 1) was reclassified using the equal interval method. This led to its division into four categories: low susceptibility (0–0.25), medium susceptibility (0.25–0.5), high susceptibility (0.5–0.75), and very high susceptibility (0.75–1). Specifically, Fig. 7 a shows the susceptibility map of ground collapses in the study area under this classification, and Table 2 details the specific susceptibility area, its percentage of the total area, and the distribution of disaster cases in the study area. For ground subsidence, the model’s predicted subsidence map was compared with the actual monitored cumulative subsidence map. Susceptibility areas based on cumulative subsidence were categorized as low (0–24 mm), medium (24 mm–64 mm), high (64 mm–96 mm), and very high (96 mm–128 mm) susceptibility areas. Subsequently, the overlap degree was calculated using the grid pixel data. Statistically, the overlap between the actual and the model-simulated cumulative subsidence was 90.31%, with specific overlap degrees for each interval detailed in Table 3 . This also reflects the accuracy of the stacking model. Figure 7 b shows the susceptibility map of ground subsidence in the study area.
( a ) Susceptibility map of ground collapses; ( b ) Susceptibility map of ground subsidence.
For the comprehensive susceptibility map of hazards, we employed two approaches. First, three experts from the Zhejiang Geological Survey were invited to reference the basic data of the two primary hazards affecting the study area, ground collapse and ground subsidence. They combined this information with their understanding of the study area, using the Analytic Hierarchy Process (AHP) to determine the impact magnitude (weights) of these two hazards on the study area. This process entails decomposing decision-making elements into hierarchies, including objectives, criteria, and alternatives, followed by qualitative and quantitative analysis to determine the weights of these disasters in the study area. The objective layer in the judgment process focused on the weight values of ground collapse and subsidence, while the criteria layer comprised ten indicators, including the impacts on the economy and human safety, relationship with geological conditions, current monitoring and management status, susceptibility assessment results, and causes and probabilities of occurrence. Of these, the first five indicators were weight determinants for ground collapse, and the latter five pertained to ground subsidence. Finally, by summing the weights of the disaster indicators for ground collapse and subsidence, the overall weight of the disaster was ascertained, leading to an aggregate judgment on the weights of multi-hazard indicators. The specific AHP judgment table and the results are shown in Table 4 and Fig. 8 .
( a ) The result of AHP assessment; ( b ) Comprehensive susceptibility map based on the AHP assessment. (This figure was created using ArcGIS Pro software. Desktop GIS Software|Mapping Analytics|ArcGIS Pro (esri.com)).
Second, we adopted a common overlay evaluation method used in multi-hazard susceptibility assessments 36 , 37 . This method involves overlaying susceptibility maps of individual hazards to determine the risk of different areas experiencing various hazards. For example, it helps to identify regions at high risk for a single hazard, areas at high risk for multiple hazards simultaneously, or areas where multiple hazards are unlikely to occur. Although this overlay method is relatively simple, it greatly reduces subjective judgments and, assuming the individual hazard susceptibility maps are accurate, provides highly valuable reference data. We selected areas with high and very high susceptibility levels from the four susceptibility categories. In Fig. 9 , the three colors represent areas in the study region with high risks of ground subsidence, ground collapse, and both disasters simultaneously. The remaining areas represent regions with low risk for both types of disasters.
Comprehensive susceptibility map using overlay evaluation method. (This figure was created using ArcGIS Pro software. Desktop GIS Software | Mapping Analytics | ArcGIS Pro (esri.com)) (Map image is the intellectual property of Esri and is used herein under license.
In the process of assessing regional disaster susceptibility using machine learning models, optimization is essential to achieve more accurate predictions. In our study, we focused on two main aspects of optimization: first, we employed grid search to obtain the optimal hyperparameters for individual models, ensuring the best performance of the stacked model; second, we optimized the dataset by using multi-source data and selecting an appropriate pixel size (32 m*32 m) to ensure that the dataset, particularly for the binary classification task of ground subsidence prediction, was not extremely imbalanced between disaster and non-disaster points.
Although we adopted reasonable methods to achieve good predictive accuracy, we acknowledge there is still room for improvement. For instance, we could explore different data partition ratios. In our case, we used a 70%:30% ratio for the training and test sets. While this ratio is widely used in many established cases of geological hazard susceptibility mapping, other ratios, such as 80%:20% 38 , might be more optimal for different scenarios. Also, Sameen 39 proposed a systematic subdivision method that captures training samples well-representative of the entire dataset. By subdividing the training/testing set based on Hellinger distance and applying a minimization method to reduce feature interdependence, they successfully enhanced the predictive capability of landslide susceptibility models.
Additionally, considering more multi-source data and evaluation metrics can improve the assessment results and increase the credibility of model predictions. In our predictions of ground subsidence and ground settlement susceptibility, we selected 7 and 6 evaluation metrics, respectively, based on field investigations and expert discussions. While these were reasonable choices, future work could include more metrics to enhance model robustness. Finally, for model hyperparameter optimization, we chose the commonly used grid search method due to the relatively simple nature of our dataset, leading to satisfactory optimization results. However, when considering more evaluation metrics and additional sample points, evaluating and quantifying model uncertainty 40 or opting for more efficient search methods, such as ant colony clustering algorithm in random search 41 or Bayesian optimization 42 , can yield better optimization results.
The "black box" effect of machine learning models often leads to their predictive results not being directly trusted by urban decision-makers. Understanding the influence of data on model decisions and the model judgment process has become increasingly important among scholars. In our model, we used a Random Forest (RF) model, which acts as the final meta-model, to assess feature importance, specifically, feature importance is derived by calculating each feature's cumulative contribution to impurity reduction across multiple decision trees, then averaging and standardizing these contributions. As shown in Fig. 10 , in predicting ground subsidence, the thickness of the surface fill layer and rainfall (weekly) (2022) contribute significantly. In predicting ground settlement, aside from the density of the drainage pipe network and burial depth of the underground confined water level, other factors also have notable contributions.
Feature importance in predicting ground subsidence and collapse.
While our method provides the contribution of each evaluation factor to the model output, this explanation might not be comprehensive. Recently, SHAP (SHapley Additive exPlanations), which uses Shapley values from game theory to analyze and illustrate the impact of each input feature on model decisions, has become a popular choice among scholars 43 , 44 . However, its application in ensemble models, especially those combining different types of models, has not yet been reliably verified. This is an area that we may need to consider in future research. Additionally, integrating physical models with machine learning models to evaluate disaster mechanisms, rather than solely focusing on susceptibility assessment, often yields more convincing results 45 , 46 .
This study focused on a typical plain area in Hangzhou affected by urban geological hazards, creating susceptibility maps for the primary geological hazards in the study area: ground subsidence and ground collapse. The specific conclusions are as follows:
To obtain reliable and reasonable susceptibility assessment results, we conducted field investigations and discussions with the Zhejiang Geological Survey, and selected nine evaluation factors with low correlation coefficients, including seven for ground collapse and six for ground subsidence. Then we used a pixel size of 32 m × 32 m, with 27,898 data samples for model prediction. After optimizing the model using the grid search method, we obtained an optimized Random Forest—Back Propagation Neural Network Coupling Model. By introducing single models and comparing with SVM and logistic regression models, the results showed that the ensemble model improved the AUC, Accuracy, Precision, Recall, and F1-score by 3% to 40% in the binary classification task of ground collapse prediction. For the non-binary task of ground subsidence prediction, the average AUC value reached 0.98.
To achieve the highest accuracy in susceptibility mapping results for the study area, we ultimately selected the best-performing stacked model's prediction results to create susceptibility maps for ground collapse and ground subsidence. To further evaluate the overall hazard situation in the study area, we sought assistance from three experts from the Zhejiang Geological Survey. Using the AHP method, the study assessed the impact weights of ground collapse and ground subsidence on the study area and created a comprehensive hazard susceptibility map. Additionally, we used the traditional overlay method to objectively determine the hazard risk in different areas. After completing the mapping, we investigated the feature importance during the prediction process. The results indicated that for ground collapse prediction, the thickness of the surface fill layer and weekly rainfall (2022) were the most significant contributors. For ground subsidence prediction, besides the density of the drainage pipe network and burial depth of the underground confined water level, other factors showed significant contributions.
All data generated or analyzed during this study are included in this published article.
Name of the code/library: RF-BP Neural Network Coupling Model Contact:[email protected]. Hardware requirements: Processor: A multi-core processor is recommended for efficient running of machine learning models. Memory: At least 8 GB RAM is advised for effective processing of the datasets and machine learning models. Storage Space: Sufficient storage space to accommodate the datasets, model files, and output files. Program language: Python. Software required: Python Interpreter, preferably version 3.8 or higher. Python Libraries: pandas, scikit-learn, joblib, matplotlib. Excel or a compatible software for reading and writing Excel files. Program size: code:4 KB data:818 KB model:5463 KB. The source codes are available for downloading at the link: https://github.com/xygbb/RF-BP-Neural-Network-Coupling - Model2.git.
Zhou, N.-Q. & Zhao, S. Urbanization process and induced environmental geological hazards in china. Nat. Hazards 67 , 797–810 (2013).
Google Scholar
Mitchell, J. K. Megacities and natural disasters: A comparative analysis. GeoJournal 49 , 137–142 (1999).
Chai, J. & Wu, H.-Z. Prevention/mitigation of natural disasters in urban areas. Smart Constr. Sustain. Cities 1 , 4 (2023).
Qian, Z. Hangzhou. Cities 48 , 42–54 (2015).
Zhou, L. et al. Insar time series analysis of natural and anthropogenic coastal plain subsidence: A case of hangjiahu plain. Geod. Geodyn. 7 , 16004–23 (2024).
Yue, W., Liu, Y. & Fan, P. Measuring urban sprawl and its drivers in large chinese cities: The case of Hangzhou. Land Use Policy 31 , 358–370 (2013).
Pourghasemi, H. R., Teimoori Yansari, Z., Panagos, P. & Pradhan, B. Analysis and evaluation of landslide susceptibility: A review on articles published during 2005–2016 (periods of 2005–2012 and 2013–2016). Arab. J. Geosci. 11 , 1–12 (2018).
Arabameri, A. et al. Performance evaluation of gis-based novel ensemble approaches for land subsidence susceptibility mapping. Front. Earth Sci. 9 , 663678 (2021).
Oh, H.-J. & Lee, S. Assessment of ground subsidence using gis and the weights-of-evidence model. Eng. Geol. 115 , 36–48 (2010).
Ansari, A., Seshagiri Rao, K. & Jain, A. K. Seismic microzonation of the himalayan region considering site characterization: Application toward seismic risk assessment for sustainable tunneling projects. Nat. Hazards Rev. 25 , 04023052 (2024).
Zhou, Y., Liu, Y., Wu, W. & Li, N. Integrated risk assessment of multi-hazards in china. Nat. Hazards 78 , 257–280 (2015).
Pourghasemi, H. R., Gayen, A., Edalat, M., Zarafshar, M. & Tiefenbacher, J. P. Is multi-hazard mapping effective in assessing natural hazards and integrated watershed management?. Geosci. Front. 11 , 1203–1217 (2020).
Eshrati, L., Mahmoudzadeh, A. & Taghvaei, M. Multi hazards risk assessment, a new methodology. Int. J. Health Syst. Disaster Manag. 3 , 79 (2015).
Wu, Y., Jiang, X., Guan, Z., Luo, W. & Wang, Y. Ahp-based evaluation of the karst collapse susceptibility in Tailai basin, Shandong province, China. Environ. Earth Sci. 77 , 1–14 (2018).
ADS Google Scholar
Xie, Y.-H. et al. Evaluation of the karst collapse susceptibility of subgrade based on the ahp method of arcgis and prevention measures: A case study of the quannan expressway, section k1379+ 300–k1471+ 920. Water 14 , 1432 (2022).
CAS Google Scholar
Wei, A., Li, D., Zhou, Y., Deng, Q. & Yan, L. A novel combination approach for karst collapse susceptibility assessment using the analytic hierarchy process, catastrophe, and entropy model. Nat. Hazards 105 , 405–430 (2021).
Chen, L. et al. Integrating expert opinion with modelling for quantitative multi-hazard risk assessment in the eastern Italian alps. Geomorphology 273 , 150–167 (2016).
Pourghasemi, H. R., Gayen, A., Panahi, M., Rezaie, F. & Blaschke, T. Multi-hazard probability assessment and mapping in Iran. Sci. Total Environ. 692 , 556–571 (2019).
ADS CAS PubMed Google Scholar
Skilodimou, H. D., Bathrellos, G. D., Chousianitis, K., Youssef, A. M. & Pradhan, B. Multi-hazard assessment modeling via multi-criteria analysis and Gis: A case study. Environ. Earth Sci. 78 , 1–21 (2019).
Munier, N., Hontoria, E., Munier, N. & Hontoria, E. Shortcomings of the ahp method. In Uses Limitations AHP Method: A Non-Mathematical Ration. Analysis (eds Munier, N. & Hontoria, E.) 41–90 (Springer International Publishing, 2021).
Ansari, A., Rao, K., Jain, A. & Ansari, A. Formulation of multi-hazard damage prediction (mhdp) model for tunnelling projects in earthquake and landslide-prone regions: A novel approach with artificial neural networking (ann). J. Earth Syst. Sci. 132 , 164 (2023).
Ansari, A., Rao, K., Jain, A. & Ansari, A. Deep learning model for predicting tunnel damages and track serviceability under seismic environment. Model. Earth Syst. Environ. 9 , 1349–1368 (2023).
PubMed Google Scholar
Dou, J. et al. An integrated artificial neural network model for the landslide susceptibility assessment of Osado island, Japan. Nat. Hazards 78 , 1749–1776 (2015).
Huang, Y. & Zhao, L. Review on landslide susceptibility mapping using support vector machines. Catena 165 , 520–529 (2018).
Tang, R.-X., Kulatilake, P. H., Yan, E.-C. & Cai, J.-S. Evaluating landslide susceptibility based on cluster analysis, probabilistic methods, and artificial neural networks. Bull. Eng. Geol. Environ. 79 , 2235–2254 (2020).
He, Q., Jiang, Z., Wang, M. & Liu, K. Landslide and wildfire susceptibility assessment in southeast Asia using ensemble machine learning methods. Remote Sens. 13 , 1572 (2021).
Li, Y. et al. Evaluation of landslide susceptibility based on cf-svm in Nujiang prefecture. Int. J. Environ. Res. Public Health 19 , 14248 (2022).
PubMed PubMed Central Google Scholar
Biau, G. & Scornet, E. A random forest guided tour. Test 25 , 197–227 (2016).
MathSciNet Google Scholar
Speiser, J. L., Miller, M. E., Tooze, J. & Ip, E. A comparison of random forest variable selection methods for classification prediction modeling. Expert. Syst. Appl. 134 , 93–101 (2019).
Wythoff, B. J. Backpropagation neural networks: A tutorial. Chemom. Intell. Lab. Syst. 18 , 115–155 (1993).
Henseler, J. Back propagation. Artif. Neural Networks: An Introd. to ANN Theory Pract. 37–66 (2005).
Amiri, E., Keshavarz, H., Ohshima, N. & Komaki, S. Resource allocation in grid: A review. Procedia-Soc. Behav. Sci. 129 , 436–440 (2014).
Ghojogh, B. & Crowley, M. The theory behind overfitting, cross validation, regularization, bagging, and boosting: tutorial. Preprint at https://arXiv.org/quant-ph/1905.12787 (2019).
Huang, J. & Ling, C. X. Using auc and accuracy in evaluating learning algorithms. IEEE Trans. Knowl. Data Eng. 17 , 299–310 (2005).
Naidu, G., Zuva, T. & Sibanda, E. M. A review of evaluation metrics in machine learning algorithms. In Computer Science On-line Conference (ed. Naidu, G.) 15–25 (Springer, 2023).
Pourghasemi, H. R. et al. Assessing and mapping multi-hazard risk susceptibility using a machine learning technique. Sci. Rep. 10 , 3203 (2020).
ADS CAS PubMed PubMed Central Google Scholar
Youssef, A. M., Mahdi, A. M., Al-Katheri, M. M., Pouyan, S. & Pourghasemi, H. R. Multi-hazards (landslides, floods, and gully erosion) modeling and mapping using machine learning algorithms. J. Afr. Earth Sci. 197 , 104788 (2023).
Liu, S., Wang, L., Zhang, W., He, Y. & Pijush, S. A comprehensive review of machine learning-based methods in landslide susceptibility mapping. Geol. J. 58 , 2283–2301 (2023).
Sameen, M. I., Pradhan, B., Bui, D. T. & Alamri, A. M. Systematic sample subdividing strategy for training landslide susceptibility models. Catena 187 , 104358 (2020).
Abbaszadeh Shahri, A., Shan, C. & Larsson, S. A novel approach to uncertainty quantification in groundwater table modeling by automated predictive deep learning. Nat. Resour. Res. 31 , 1351–1373 (2022).
Liu, X. & Fu, H. An effective clustering algorithm with ant colony. J. Comput. 5 , 598–605 (2010).
Frazier, P. I. Bayesian optimization. In Recent Advances in Optimization and Modeling of Contemporary Problems (ed. Frazier, P. I.) 255–278 (Informs, 2018).
Li, Z. Extracting spatial effects from machine learning model using local interpretation method: An example of shap and xgboost. Comput. Environ. Urban Syst. 96 , 101845 (2022).
Zhang, J. et al. Insights into geospatial heterogeneity of landslide susceptibility based on the shap-xgboost model. J. Environ. Manag. 332 , 117357 (2023).
Willard, J., Jia, X., Xu, S., Steinbach, M. & Kumar, V. Integrating physics-based modeling with machine learning: A survey. arXiv 1 , 1–34 (2020).
Han, J., Zhang, L. et al. Integrating machine learning with physics-based modeling. Preprint at https://arXiv.org/quant-ph/2006.02619 (2020).
Download references
This research is supported by the China Geological Survey, Nanjing Center, Zhejiang Geological Survey and China University of Geosciences, Wuhan. The work described in this paper was funded by Laboratory of Geological Safety of Underground Space in Coastal Cities, Ministry of Natural Resources (Project No. BHKF2022Z02), and the China Geological Survey, Nanjing Center (Project No.DD20190281).
Authors and affiliations.
China Geological Survey Nanjing Center, Nanjing, 210016, People’s Republic of China
Bofan Yu & Huaixue Xing
China University of Geosciences (Wuhan), The Institute of Geological Survey of China University of Geosciences (Wuhan), Wuhan, 430074, People’s Republic of China
Bofan Yu & Jiaxing Yan
You can also search for this author in PubMed Google Scholar
Bofan Yu: Conceptualization, Investigation, Methodology, Software, Writing—original draft, Writing—review & editing Jiaxing Yan: software, validation Huaixue Xing: Resources, Project administration, Funding acquisition.
Correspondence to Huaixue Xing .
Competing interests.
The authors declare no competing interests.
Publisher's note.
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/ .
Reprints and permissions
Cite this article.
Yu, B., Xing, H. & Yan, J. Susceptibility assessment of multi-hazards using random forest—back propagation neural network coupling model: a Hangzhou city case study. Sci Rep 14 , 21783 (2024). https://doi.org/10.1038/s41598-024-71053-7
Download citation
Received : 25 June 2024
Accepted : 23 August 2024
Published : 18 September 2024
DOI : https://doi.org/10.1038/s41598-024-71053-7
Anyone you share the following link with will be able to read this content:
Sorry, a shareable link is not currently available for this article.
Provided by the Springer Nature SharedIt content-sharing initiative
By submitting a comment you agree to abide by our Terms and Community Guidelines . If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.
Sign up for the Nature Briefing: Anthropocene newsletter — what matters in anthropocene research, free to your inbox weekly.
You have full access to this open access article
Linear agricultural nutrient regimes are the principal cause for perturbation of the geochemical cycles for nitrogen (N) and phosphorus (P) and other planetary boundaries. Nutrient cycles are characterized by high spatial disparity and China is a hotspot due to high fertilizer application rates. Using substance flow analysis, this study identified and quantified nutrient flows from agricultural production to residue management of Huangyan tangerines ( Citrus reticulata ) and water bamboo ( Zizania latifolia ) in a case study of Huangyan district (Taizhou City, Zhejiang province). About 754 Mg/a of N and 105 Mg/a of P can theoretically be recovered in the tangerines and water bamboo systems from currently untapped material flows. This could replace 59% of the N and 15% of the P currently applied as chemical fertilizer, reducing environmental impacts. Combining the nutrient recovery of both systems and upscaling the results to Taizhou City, the goal from the 14th Five-Year Plan for Agricultural and Rural Modernization to save 1182 Mg of nutrients per year could be exceeded by almost 12 times. This study’s data have varying degrees of uncertainty. The analysis of data representativeness shows potential for improvements, especially in the agricultural production of water bamboo and the nutrient contents of material flows.
A multi-regional soil phosphorus balance for exploring secondary fertilizer potential: the case of norway.
Explore related subjects.
Avoid common mistakes on your manuscript.
The biogeochemical cycles have exceeded the proposed safe operating space for humans [ 1 ]. According to Campbell et al., agriculture is the main driver for the transgression of this and other planetary boundaries [ 2 ]. Specifically, the production of nitrogen (N) fertilizer is linked to a high energy demand (the Haber–Bosch process for ammonia synthesis is responsible for 1–2% of the world’s total energy consumption) [ 3 ], it consumes 2% of the world’s natural gas as feedstock [ 4 ], and contributes to 1.4% of all anthropogenic CO 2 emissions [ 5 ]. Phosphorus (P) is a finite resource mainly mined in China, Morocco and the United States, making phosphate rock as well as phosphorus critical raw materials [ 6 ]. Apart from climate change and resource depletion, the high N and P application in agricultural regions leads to air pollution, soil acidification, and eutrophication of water bodies—which ultimately results in water dead zones, fish kills, algal blooms, and water contamination [ 7 ]. This unsustainable nutrient management attempts against the achievement the United Nations Sustainable Development Goals (e.g. 6—Clean water and sanitation, 12—Responsible consumption and production, 14—Life below water, 15—Life on land).
Industrial agricultural systems are commonly characterized by linear regimes. Since the nineteenth century, natural nutrient sources and recycled materials were no longer enough to cover the needs of the increasing population [ 7 ]. Therefore, the use of Haber–Bosch N and mined P started to increase rapidly, becoming the main nutrient input for crops—since the 1960s, the use of synthetic N fertilizers increased ninefold and the P fertilizer use tripled [ 7 ]. Although fertilizer production is concentrated in a few countries (mainly China with 25%, Russia with 10%, and United States and India both with 9%), its use is widely distributed globally, meaning that many countries rely on imports to cover their fertilizer demand [ 8 , 9 ]. After application, an important part of these nutrients (over 80% of the N and 25–75% of the P) gets lost to the environment, causing pollution [ 7 ]. The inefficient fertilizer use has been recognized as a global issue and the UN Environment Programme highlighted the need for coordinated policies to manage N and P pollution effectively at global, national, and regional levels [ 10 ]. Moreover, fertilizers are one of the carbon-intensive products covered by the EU’s Carbon Border Adjustment Mechanism (CBAM) to encourage global reduction of carbon emissions instead of moving their production to countries with less stringent climate policies [ 11 ].
China (in particular eastern China) is one of the few agricultural regions with very high P and N application rates that are the main contributors to the transgression of the biogeochemical flows boundary [ 1 ]. It is the world’s largest consumer of fertilizers (42.31 Mio Mg/a of nutrients) [ 12 ], and the country with the second highest greenhouse gas (GHG) emissions from agriculture after India (663 Mio. Mg of CO2 eq/a) [ 13 ]. The reason is that China’s agriculture has been transitioning from a self-subsistence extensive agriculture to an intensive industrial cash crop production system [ 14 ]. After the establishment of the People’s Republic of China in 1949, hundreds of millions of farmers were organized in collective agriculture communes, where all agriculture production decisions were made by local leaders following a higher level production plan [ 15 ]. This changed after the economic reform in 1978: collectively owned land was leased to individual farmers, who had to deliver a fixed quota of “strategic crops” to the state for a fixed price, but were also free to produce more and sell their surpluses in rural markets [ 15 ]. To increase the production on their small plots, farmers started to apply excessive amounts of chemical fertilizer—China’s yearly application per hectare of cropland is one of the highest of the world with 319 kg (for comparison, the world’s average is 119 kg and Europe’s average, 75 kg) [ 16 ]. This resulted in a series of environmental impacts, including groundwater contamination with nitrates, eutrophication of surface waters, soil acidification, greenhouse gas emissions, loss of biodiversity, and micronutrients deficiencies in feed and food [ 17 ].
The Chinese government has set goals and introduced policies to improve sustainability in agriculture. In 2015, the Ministry of Agriculture launched the “zero-growth” action plan, regulating fertilizer and pesticide use by 2020 [ 18 ]. Later, in 2021, the 14th Five-Year Plan for Agricultural Green Development was released, emphasizing the reduction of chemical fertilizer and pesticide use and the improvement of the utilization level of agricultural resources [ 19 ]. China wants to become carbon neutral by 2060, which requires a substantial transformation of the agricultural sector as well [ 20 ].
In addition, local governments are increasingly taking measures to promote agricultural practices that are more circular and sustainable. In Taizhou, a city in the Zhejiang province, the Bureau of Agriculture and Rural Affairs launched its own 14th Five-Year Plan for Agricultural and Rural Modernization to improve productivity and living standards in rural areas (its focus is slightly different of that of the national Plan for Agricultural Green Development mentioned above, which mainly aims at reducing environmental impacts). This action plan aims to increase the use of commercial organic fertilizer, digestate from anaerobic digestion, and green manure to improve soil structure and organic matter content. The utilization rate of manure (from livestock and poultry) and rice straw should also be improved to 92% and 95%, respectively, by 2025. Besides, chemical fertilizer use in the crop sector should be reduced with the help of soil testing and formula fertilization. In this way, Taizhou expects to save 0.6 kg of “pure nutrients” (N, P 2 O 5, and K 2 O) per Mu—or more than 1182 Mg per year—while reducing air pollution (mainly caused by straw burning), eutrophication of water bodies, soil erosion, and salinization. [ 21 ]
Implementing circular systems to achieve these goals requires the detailed study of locally available, potentially valuable organic material flows, which can be done by using Material Flow Analysis (MFA) [ 22 ]. This is a case study for Huangyan (黄岩), a district in Taizhou city (Fig. 1 ) [ 23 ]. Huangyan is a typical example of a region with a linear, cash-crop-oriented, agricultural system and a high demand for chemical fertilizers. While the majority of Taizhou’s economic activities take place in Luqiao district, the agricultural production is based in Huangyan. The agricultural sector, with a yearly contribution of over 10 billion yuan, plays an important role in the region’s economy [ 24 ]. In 2021, 1,918,000 Mu of agricultural land were sown in Taizhou, with a total output of 560,000 Mg of grain and 1,456,900 Mg of fruit [ 25 ]. By quantifying the nutrients that could be recovered in this region, the possible contribution to Taizhou’s goal can be assessed and additional measures can be taken (e.g. to effectively recover those nutrients, to formulate stricter goals if necessary, or to introduce new regulations).
Map of Huangyan, a district in Taizhou city, located in the Jiaojiang River Delta, southeast of Zhejiang Province. Huangyan has an area of 988 km 2 and a registered population of around 610,000
The study focuses on the two crops with the highest production amounts in the region, namely Huangyan tangerines ( Citrus reticulata, annual production in Taizhou reaches 546,000 Mg/a) and water bamboo ( Zizania latifolia, also called Manchurian wild rice, 77,000 Mg/a) [ 25 ]. Both species are native to China, which leads the global market—China produces 61% of the world’s tangerines, mandarins, and clementines, and 27% of the rice [ 26 , 27 ]. The goal of the study is to identify and quantify all the by-product flows in these agri-food systems, to later derive the amounts of N and P that could theoretically be recovered and recycled. These theoretical recycling potentials will then be related to the amount of nutrients that Taizhou expects to save according to its last Five-Year Plan for Agricultural and Rural Modernization. In this way, the potential contribution of nutrient recovery in these two agri-food systems can be measured.
The methodology of material flow analysis, deeply described in Brunner & Rechberger’s Handbook of Material Flow Analysis , is defined as “a systematic assessment of the flows and stocks of materials within a system defined in space and time” [ 22 ]. MFA is an iterative method consisting of several steps, including goal and system definition (which lead to a qualitative model), determination of mass flows and stocks, and mass balance (resulting in a quantitative model). The quantified material flows can also be linked with the substances’ mass fractions to identify the substance-specific sources and sinks, allowing the development of specific recommendations to improve substance flows.
The goal of this substance flow analysis (SFA) is to quantify the amount of N and P that could be recovered within the agri-food systems of the two main agricultural products (Huangyan tangerines and water bamboo) in Huangyan district, Taizhou City, Zhejiang province, China, in 2019.
The system model was defined according to the material life cycle stages identified during the data acquisition, namely: agricultural production, manufacturing, distribution and sales, consumption, and waste treatment and disposal (Figs. 2 , 3 ). The processes shown in the SFAs represent the average and not the best available technology. All solid organic material flows were considered. Gaseous, water, and energy flows were not regarded due to a lack of data.
System definition for Huangyan tangerines
System definition for water bamboo
The required data were acquired by both top-down (secondary data comprising statistical and aggregated data) and bottom-up (primary data from interviews) data collection.
For the bottom-up data collection, 25 relevant stakeholders from governmental institutions, educational organizations, agricultural and manufacturing sectors, markets, and waste management companies were interviewed between October and November 2019 (supplementary material, Figure a). The interview participants were selected through snowball sampling. Interview guides were prepared in advance to provide a framework of topics to be covered (see questions in supplementary material). However, the interviews were semi-structured, including open questions and allowing new ideas to be raised during the discussion. A translator was always present to translate from Chinese to English and vice versa.
Top-down data were collected through literature research, including statistical yearbooks (from Huangyan and Taizhou), scientific publications, and official reports. This secondary data were combined with the primary data from the interviews to have a solid foundation for the MFA modelling.
For the SFA, the material composition data was taken from FoodData Central, a data system launched by the U.S. Department of Agriculture (USDA), whenever available [ 28 ]. If the needed data were not available in that system, different sources were revised, and the average of various values was calculated.
To account for uncertainties and varying reliability, as well as the temporal and spatial representativeness of the collected data, four data representativeness levels [ 29 ] were defined to classify the quality of the gathered data (Table 1 ). Level 1 was assigned to educational organizations (interviewees were experts in their fields); level 2 to governmental institutions and waste treatment companies (statistical data derived from reports, measurements, and census in the study region); level 3 to farmers, food manufacturing companies and markets (specific information, not valid for the entire region); and level 4 to literature data (related to other geographic regions and/or periods).
Mass flows of fresh matter (FM) were determined using an Excel model. All the inputs and outputs of each process were related to a reference flow of 1 Mg of freshly harvested product (Huangyan tangerines or water bamboo). When a range of values was available, an average was used. The calculations to determine each of the FM flows and the assumptions made can be found in the supplementary material (Tables a, b).
The flows for the N and P SFAs were calculated by multiplying the FM flows by the N and P mass fractions of each flow. These mass fractions were obtained from secondary sources, except for a few cases, where the interviews revealed relevant data (supplementary material, Tables c, d).
The models for Huangyan tangerines and water bamboo were introduced in the MFA software e!Sankey to obtain a graphical visualization of the nutrient flows. Finally, the theoretical recycling potential of P and N was calculated by adding up all the by-product flows and compared to the amount of N and P already being recycled.
These results were compared with the goal set by Taizhou’s Five-Year Plan for Agricultural and Rural Modernization (saving 1,182 Mg/a of “pure nutrients”). To do that, the values for the P recycling potential were multiplied by 2.29 to account for P 2 O 5 and, hence, allow the comparison with the above-mentioned goal.
The results for Huangyan tangerines and water bamboo are described in this section (the graphic descriptions of the entire life cycles of the tangerines and the water bamboo can be found in Figs. 2 , 3 ). The SFAs for N are illustrated in Fig. 4 (tangerines) and Fig. 5 (water bamboo), and those for P are shown on Fig. 6 (tangerines) and Fig. 7 (water bamboo).
Nitrogen SFA for Huangyan tangerines. Values represent annual flows
Nitrogen SFA for water bamboo. Values represent annual flows
Phosphorus SFA for Huangyan tangerines. Values represent annual flows
Phosphorus SFA for water bamboo. Values represent annual flows
Agricultural production.
Around 4,200 ha in Huangyan are covered by tangerine orchards. This area is shared between almost 19,500 farms, i.e. the average area of a tangerine farm in Huangyan is only 0.2 ha [ 24 ]. Most of this area belongs to families who own a small piece of land for self-supply. Recently, the government is investing to improve the quality of Huangyan tangerines by funding large, modern tangerine gardens managed by just a few partners [ 30 ]. An average tree produces around 24 kg of tangerines per year [ 31 ]. Trees in their most productive phase, however, can produce up to 10 times more [ 32 ]. The young tangerine gardens in the district are the main reason for the low average [ 30 ].
Tangerine farmers apply 20–30% (relative to N) organic fertilizer and 70–80% chemical fertilizer (15% N, 15% P 2 O 5 , 15% K 2 O) [ 31 ]. In 2017, Taizhou launched a policy to promote the use of organic fertilizer: farmers who use at least 50% organic fertilizer can get access to subsidies [ 31 ]. Currently, there are less than 50 of these “green agriculture” farms in Huangyan—including but not only tangerine farms [ 32 ]. Since there is no fertilizer production in Huangyan, this product has to be imported from other regions [ 31 ]. Chemical fertilizers add 975 Mg of N and 426 Mg of P to the agricultural soils every year, constituting the largest nutrient input of this system. Common organic fertilizers include rapeseed compost, a mixture of sheep manure and mushrooms, compost made out of cassava and residues from alcohol production, and compost made out of water bamboo and sugar cane waste [ 31 , 32 ]. These organic fertilizers introduce 325 Mg of N and 195 Mg of P to the tangerine system.
Most of the tangerines produced in Huangyan (95%) are exported to other regions [ 24 ]. 173 Mg of N and 11 Mg of P leave the system in this way. From the rest, 95% is sold as fresh fruit directly on the farm or in markets and supermarkets (or grocery stores), and 5% serves for the further manufacturing of different products like canned tangerines, Chinese medicine, oil, or candies [ 33 ]. The largest by-product flow in the entire tangerine system comes from the agricultural stage. Every year, 16,140 Mg of by-products are produced in Huangyan’s tangerine farms. This agricultural waste, consisting mostly of pruning residues, contains 311 Mg of N and 20 Mg of P. It is partly used to make fire for heating and cooking [ 33 ], while the rest is either left on the ground, burned in open fires, or landfilled [ 30 , 31 ]. Farmers mentioned a lack of space (it is not permitted to construct buildings in agricultural areas) and a lack of animal manure as the main reasons why they do not compost or digest their waste [ 32 ].
The biggest company for canned fruit in Taizhou City, “Yiguan Food”, contributes to 10% of China’s canned tangerines production [ 31 , 32 , 33 , 34 ]. The majority of their products are exported and only small amounts (5%) stay in Taizhou for supermarkets, cake and jelly production [ 35 ]. The tangerines for the manufacturing process are purchased from different cooperatives in Taizhou, Hubei, Hunan, and, mostly, Linhai [ 36 ]. The production process includes size separation (small tangerines are picked out), heating in water to facilitate the peel separation, and an acid–base treatment to separate the pith. Afterward, tangerines are separated into slices by hand, which in some cases damages them. Peels, damaged or rotten tangerines, pith, and wastewater are the major by-product flows in this production chain. The peels are sold to a company that produces spices. Wastewater is sent to an industrial treatment plant that processes wastewater from this and other companies, before going to the municipal wastewater treatment plant.
Huangyan’s tangerines for fresh consumption are either sold directly from the farm (30%) or in markets and supermarkets (70%) [ 37 ]. Only 9% of the products in the market come from Huangyan, while the rest is imported from other regions, for example from the wholesale market in Luqiao [ 38 ]. Imported tangerines introduce 60 Mg of N and 4 Mg of P into the system. Around 5% of the products sold in the market get wasted [ 39 ]. A small part of these by-products is collected by people who use it to feed their animals, but most of it goes to Huangyan’s landfill [ 40 ]. In contrast, the wholesale market in Luqiao treats part of its by-products: the “clean” fraction (without plastic, paper, or other contaminants) is treated at high temperatures to produce fertilizer. The “contaminated” fraction is compressed. The wastewater then goes to the wastewater treatment plant and the compressed waste is incinerated at Luqiao waste to energy plant. [ 39 ]
Around 6% of the produced tangerines, containing 50 Mg of N and 2 Mg of P, get wasted in households [ 41 ]. The consumed nutrients are excreted and leave the households with the wastewater (16 Mg of N and 2 Mg of P).
Huangyan Waste Sorting and Treatment Centre is responsible for 20% (or 27 Mg per day) of municipal perishable waste. Well-sorted perishable waste from school and office canteens, as well as the “clean” fraction of the perishable household waste is composted. This accounts for 20% of the perishable waste treated by the company. The composting process is still being tested, and the compost quality is still not good enough for agricultural applications. The rest 80% of the company’s perishable waste, corresponding to the “polluted” fraction of the household waste, is treated anaerobically. This waste is first filtered, and then the oil is separated and sold to soap or biodiesel companies. The solid fraction is treated for three to seven days under anaerobic conditions and high temperatures (around 140 °C), while the liquid fraction goes to the wastewater treatment plant. Also this treatment is still being tested to improve the quality of the resulting fertilizer. [ 42 ]
Most of Huangyan’s municipal perishable waste (80%) ends up in the landfill [ 42 ]. 40 Mg of N and 2 Mg of P get lost in this way. At the time of the field trip, the landfill was almost full, and a new incineration plant was being constructed to replace it. During the interview at the landfill, it was stated that China would ban landfills in 2020. There was a project to use kitchen waste to produce biogas in the future, but there was still no estimated starting date. [ 43 ]
Excreta is the main waste flow resulting from the tangerine consumption stage, and the second-largest waste stream in the entire system (12,009 Mg/a). Yuanqiao wastewater treatment plant in Huangyan District can treat up to 60,000 m. 3 per day. The site covers an area of more than 11 ha and applies Anaerobic/Anoxic/Oxic technology (AAO). The clean effluent is discharged into the river. [ 44 ]
Manchurian wild rice ( Zizania latifolia ), or water bamboo, is a perennial plant native to China [ 45 ]. Its seeds have been consumed as a cereal for more than 3,000 years. When infected with the black smut fungus ( Ustilago esculenta ), the stem becomes bigger and tender, making it the second most-cultivated aquatic vegetable in China. Huangyan is one of the main producing areas in the country [ 46 ]. Wetland farms growing Z. latifolia in Huangyan cover a much smaller area than tangerine farms—around 800 ha [ 37 ]. However, this area is growing because producing this crop in paddy fields is more profitable than producing rice [ 30 ].
There are two harvest seasons for water bamboo: one from May to June and another one from October to December [ 47 ]. In the summer, the production (around 100 kg/ha) is twice as much as in autumn [ 48 ]. To produce 1 Mg of water bamboo, farmers use on average 50 kg of chemical fertilizer (15–15-15) [ 24 ]. In this way, 296 Mg of N and 129 Mg of P are added to the cropland every year. Apart from that, in the summer, farmers put compost from their farms back in the field (503 Mg of N and 19 Mg of P per year) [ 48 ].
During the harvesting, waste leaves (which represent 30% of the plant’s weight) are cut off and, in most cases, left on the field [ 47 ]. This is the heaviest by-product flow in the entire system, accounting for 25,200 Mg per year (437 Mg of N and 147 Mg of P). There are some farms using these leaves to cover the soil (e.g. in tangerine fields) to improve its quality, although the transport and labour is expensive [ 49 ]. Stems are collected for further processing [ 47 ]. In the next step, the stems are peeled by hand. Thus, the water bamboo shoots (representing 50% of the plant’s weight) are separated from the peels (accounting for the rest 20%) [ 47 ].
Around 20 kg/ha of water bamboo shoots are wasted every year because of diseases, bad quality, or bad appearance. This agricultural waste stream, containing 64 Mg of N and 14 Mg of P, gets landfilled [ 24 ]. After the peeling process, the majority of the water bamboo shoots are exported, along with 648 Mg of N and 137 Mg of P. This is the largest P export in the system.
The peels are cut into little pieces on site and mixed with urea—the largest nitrogen flow in the system, adding 3,091 Mg of N—and decomposing agent. After 20 days, the compost is ready to be put back on the field. This is done every year at the end of July. 70% of the compost (1173 Mg of N and 44 Mg of P) is sent to other fruit and vegetable farms, constituting the largest N export in the system. The rest 30%, containing 503 Mg of N and 19 Mg of P, is used in the farm itself. [ 48 ]
Most of the produced water bamboo shoots (95%) are exported to other regions. The part that stays in Huangyan gets sold in markets, supermarkets, and grocery stores [ 24 ]. As mentioned for the tangerine system, only 9% of the products in the markets come from Huangyan, and about 5% of the products sold in the market get wasted.
The life stages consumption, waste treatment, and wastewater treatment are identical to the ones described for the tangerine system. Also in, this case, excreta constitutes the major waste flow coming from the consumption stage (16,032 Mg/a). 286 Mg of N and 61 Mg of P are lost through this route.
In the tangerine system, the by-product streams that could theoretically be recycled include agricultural waste; peels, damaged tangerines, and pith from the manufacturing process; and excreta, peels, and rotten tangerines from the consumption process. These flows amount to 38,870 Mg/a. The nitrogen content in those flows adds up to 382 Mg/a and the phosphorus content to 25 Mg/a.
Only the tangerine peels from the manufacturing process are currently being recycled to produce spices and a part of the rotten tangerines and peels from consumption are treated anaerobically or through composting. These account for 2,115 Mg/a or 5% of the total theoretical recycling potential. The nitrogen contained in the recycled flows accounts for only 11 Mg/a or 3%, and the phosphorus for 0 Mg/a or 2% (Fig. 8 ).
Tapped and untapped recycling potential for nitrogen and phosphorus in Huangyan’s tangerine and water bamboo systems. Solid color areas represent the amount of N (green) and P (orange) that is already being recycled. Dotted areas represent the amount of N (green) and P (orange) that is not yet recycled. t tangerine system, wb water bamboo system
Regarding the water bamboo system, the theoretically recyclable by-product streams include leaves cut and put back into the field while harvesting; peels and rotten water bamboo from the peeling process; rotten water bamboo from the markets; and excreta and rotten water bamboo from consumption. Together, these flows amount to 63,685 Mg/a. The nitrogen content in those flows adds up to 1,122 Mg/a and the phosphorus content to 292 Mg/a.
The materials that are already recycled include water bamboo leaves and peels in the field and rotten water bamboo shoots that are sent to composting or anaerobic treatment after consumption. These flows add up to 42,166 Mg/a, or 66%, of the theoretical recycling potential for fresh matter. The nitrogen contained in the recycled flows accounts for 738 Mg/a or 66%, and the phosphorus for 211 Mg/a or 72% (Fig. 8 ).
Adding up the recycling potentials of the untapped N and P 2 O 5 flows results in 426 Mg/a for Huangyan tangerines and 570 Mg/a for water bamboo (Fig. 9 ). The goal of Taizhou’s Bureau of Agriculture and Rural Affairs is to save 1,182 Mg of “pure nutrients” (N, P 2 O 5 , K 2 O) per year. Hence, recovering all the N and P from the analysed systems would contribute 36% to the achievement of the goal in the case of the tangerines and 48% in the case of the water bamboo. Combining both systems would make a very high contribution to the goal (84%).
Contribution of the potentially recyclable N and P 2 O 5 in both Huangyan systems (Huangyan tangerines and water bamboo) to the goal set by Taizhou’s Five-Year Plan for Agricultural and Rural Modernization on the mass of “pure nutrients” (N, P 2 O 5 , K 2 O) to be saved
The basis for the N and P SFAs for tangerines and water bamboo presented in Sect. “Results” are MFAs conducted in the scope of this study. Data uncertainties for the corresponding material flows are discussed hereunder. The below-mentioned reference flows correspond to 1 Mg of fresh matter (tangerines or water bamboo).
The data representativeness of all flows entering or leaving the agricultural production process in the tangerine’s SFA corresponds to the first or the second level. The “organic fertilizer” input and the “pruning residues and leaves” output account for 21% of the reference flow. The size of these flows combined with their very high data reliability form a good basis for the SFA. The flow “tangerines for export” is the largest of the system, accounting for 90% of the reference flow, and it is defined with high reliability. Due to the flows’ significance, it is recommended to further improve data representativeness and confirm flow quantities by published scientific data or an independent source such as producers or distributors. Most input and output data of the canned tangerines production process was provided by one factory owner ( Yiguan Food ) and therefore was classified as medium reliability data. It could be argued that in reality data representativeness is higher, since it is the largest company producing canned tangerines in Taizhou City [ 31 , 32 , 33 , 34 ]. Further research could be conducted by examining other similar factories or conducting own measurements. The largest flow related to this process, however, only accounts for 7% of the reference flow. Hence, efforts to improve the data quality of this process should be kept within reasonable limits. Three of the four flows of the distribution process correspond to the third data representativeness level. Due to the significance of the “imported tangerines” (31% of the reference flow) and the “tangerines for the markets” (33%) flows, it is suggested to verify the gathered information. More interviews with owners of various markets in Huangyan should be conducted. It was assumed that the consumers are “steady state” humans, i.e. the nutrients consumed in the food will end up in the excreta, based on the fact that the percentage of food retained by the human body for the synthesis of tissues can be considered negligible on average [ 50 ]. Since this data comes from literature, those flows were classified as level four. The “Excreta to wastewater” flow is the largest output (1% of the reference flow) of the consumption process. Its data with defined medium data reliability was obtained from scientific publications of other geographic regions and periods. Since this flow holds a significant nutrient recovery potential, it is recommended to reduce data uncertainty and contact Taizhou’s Environmental Protection Bureau. Unfortunately, during this study, the Bureau repeatedly refused to be interviewed. Requesting a reference letter from the Chinese government could help to obtain data from governmental institutions. All the flows entering and leaving the composting, filtration, and anaerobic treatment processes belong to the data representativeness level 2, which indicates that these flows can picture the real situation in Huangyan relatively well.
The outputs of the agricultural production, “stems” and “leaves” account for 140% and 60% of the reference flow, respectively. Both flows are based on data from one farmer and are, therefore, defined as the third data representativeness level. Roughly 29% of the stems become peels. Together with the amount of rotten water bamboo shoots, it makes ca. 53%, which corresponds with the information obtained from Huangyan Agricultural Bureau—according to which 50% of the vegetable is wasted [ 30 ]. The “leaves” and the “peels” flows are the largest by-product flows in the system and represent the highest nutrient recovery potential. Based on the statements of two farmers (third representativeness level) the leaves are left in the fields during the harvest and the peels are composted. It is recommended to approach water bamboo experts (equivalent to the scientists from the Institute of Citriculture for the tangerines), identify relevant scientific studies, or conduct surveys covering a representative number of farms, to reduce data uncertainty for these relevant material flows. The data employed to model the rest of the processes (distribution, consumption, composting, filtration, and anaerobic treatment) is the same as the one used for the tangerines. Hence, the data representativeness of those processes in both systems is similar and the same recommendations can be drawn (see “ Data uncertainties in the tangerine system ”).
The fresh matter flows served as a basis for the calculation of the N and P flows. During the SFA modelling, the uncertain material flow values were multiplied by estimates for N and P contents obtained from secondary sources. Firstly, the level of uncertainty rises with these mathematical operations. Secondly, the values from the literature refer to studies with different spatial scopes decreasing the accuracy of the calculated P and N flows in this study. To improve accuracy and reduce uncertainty, it is recommended to conduct a sampling campaign to measure the N and P contents of each flow as a basis for the SFAs, as was done in a Danish study [ 51 ]. Although this approach would have exceeded the scope of this study, it could have avoided the need for several assumptions (supplementary material, Tables c, d).
The inputs and outputs of the substance flow processes are not always balanced (see Δ MB flows in the SFAs), since data on energy, soil, water, and gas flows could not be gathered in the scope of this study. The following two paragraphs discuss nutrient imbalances. The mentioned values represent the real annual flows in Huangyan.
The biggest N losses occur as agricultural waste during the tangerine production (311 ± 31 Mg/a) and as peels during the consumption phase (50 ± 10 Mg/a). The N mass fraction of the peels is higher than that of the rest of the fruit. Since most of the peels produced during the consumption phase (80%) are sent to the landfill, this organic material is lost from the system. 807 ± 161 Mg N/a are missing when comparing N from fertilizer application (inputs) and N contents in agricultural products and by-products (outputs). From the applied 1,300 ± 130 Mg N/a, only 13% (173 ± 35 Mg/a) end up in the tangerines. The findings align with previous research, which shows that agriculture is responsible for almost three quarters of all N losses (43% of those losses occur through nitrate leaching to ground and surface waters, 30% through denitrification, and 23% through ammonia emissions) [ 52 ]. Similarly, major P losses in the tangerine system occur during agricultural production. While 621 ± 62 Mg P/a are applied to the tangerine fields, 20 ± 2 Mg/a end up in agricultural waste and 12 ± 2 Mg/a (2%) in produced tangerines. The rest 589 ± 118 Mg P/a (95%) is lost. Prior research has also showed that more than half of the total P losses happen in agriculture, 88% of those losses through P accumulation in soils [ 52 ]. According to George [ 53 ], agricultural production has low P-efficiencies due to solubility and mobility interactions in soils and plants. Also, Golomb and Goldschmidt [ 54 ] stated that less than half (43.7%) of the P-uptake by the mandarin trees ends up in the fruits. The pruning residues and leaves to product ratio calculated in a Croatian study is much lower than the one resulting from this study (0.07 kg/kg versus 0.21 kg/kg) [ 55 ]. This can be explained by the new tangerine farms in Huangyan, since young trees have lower production rates leading to relatively higher amounts of residues [ 30 ]. The pruning residues and leaves to product ratios are expected to decrease in the future, with more mature trees. These are also possible reasons for the low N-uptake of tangerines in Huangyan. Furthermore, the export of tangerines implies N and P losses for Huangyan’s tangerine system, with 176 ± 36 Mg N/a (14%) and 12 ± 2 Mg P/a (2%). No information was found on the local wastewater treatment processes. The literature research indicated that wastewater treatment plants in China are still concerned about removal—and not recovery—of nutrients, and that landfilling is the main sludge treatment method [ 56 ]. Hence, the nutrients contained in excreta (16 ± 6 Mg N/a, 2.4 ± 1.0 Mg P/a) are not recovered. A study conducted in Vietnam quantified regional N and P flows from rice, fruit, and vegetable production. Major P losses modelled in that study also occur during the same life cycle stages and include leachate from agricultural production, solid waste, and excreta from households [ 57 ]. Analyses of soil, surface water, groundwater, and wastewater samples, as well as liquid and gaseous emissions, could provide details on the fate of N and P in the agricultural production and consumption phase.
The production of water bamboo is characterized by higher N and P recovery rates than that of tangerines. Assuming that all nutrients from the leaves, which are cut off during the harvest process, are absorbed by the crops in the wetland, a total of 29% N (550 ± 165 Mg/a) and 56% P (166 ± 50 Mg/a) of the applied nutrients would be recovered. The differences between added nutrients from fertilizer and leaves (inputs) compared to harvested stems (outputs) are 425 ± 127 Mg N/a (22% of applied N) and 73 ± 22 Mg P/a (20%). Other studies investigating flooded rice production systems, showed that the largest losses for P and N were caused by run-off, leaching and accumulation in the soil, as well as emissions to air [ 58 , 59 ]. These potential nutrient losses should be quantified in future studies for the water bamboo production system with leaves decomposing in anaerobic conditions in flooded paddies and peels further processed to compost (19 Mg P/a).
The nutrient losses during the peeling, distribution and consumption phase are 384 Mg N/a and 81 Mg P/a from rotten water bamboo and excreta, which end up in the landfill or the wastewater treatment plant. 50% of the N (1,821 ± 481 Mg/a) and 95% of the P (181 ± 41 Mg/a) entering the system as fertilizers are exported as water bamboo and composted leaves.
In summary, for the agricultural processes, nutrient fixation in soil and nutrient leaching in surface and ground water, as well as gaseous emissions, are assumed to be relevant contributions to the substance imbalances. For the remaining processes, data uncertainties are assumed to be the main cause for imbalances. The N and P SFAs presented in this study, as well as the derived recycling potentials, should be considered as estimates. It is strongly recommended to conduct a sampling campaign and use own measurements to correct these results.
The results suggest that, combining the already recycled streams with the currently unutilized rotten tangerines and water bamboo as well as the resulting excreta from consumption, over 1,503 Mg N/a and 316 Mg P/a (725 Mg P 2 O 5 /a) could theoretically be recovered in Huangyan. Recovering these nutrients would allow a growth of agricultural production of 15% (based on N) and 13% (based on P), while fulfilling the zero-growth action plan of the Chinese government. Upscaling these results from Huangyan district to the region of Taizhou, 11,065 Mg N/a and 1,329 Mg P/a (3,043 Mg P 2 O 5 ) could theoretically be recovered (based on the production numbers for tangerines and water bamboo published in Taizhou’s statistical yearbook) [ 60 ]. Adding up the nutrient mass for N and P 2 O 5 then results in 14,108 Mg nutrients/a. This would exceed the goal of chemical fertilizer reduction (1,182 Mg of nutrients) by almost 12 times. It is worth mentioning that this result would be even better if potassium flows would have also been accounted for. Considering the negative environmental impacts caused by the utilization of chemical fertilizers [ 61 , 62 ], recycling activities can help to enhance environmental sustainability and support China in becoming carbon neutral in 2060.
While this study calculates the theoretical potential for nutrient recovery, several practical challenges must be addressed to enhance feasibility. Primarily, technical barriers exist, such as the efficiency of recovery technologies. For example, net losses of 18% of N were reported during anaerobic digestion [ 63 ], and for composting they can reach up to 50% [ 64 ]. Besides, the agricultural by-products analyzed in this study are from plant origin, and an addition of animal manure (which could imply a collaboration with other stakeholders) might be required to improve the efficiency of the recovery process. These technologies depend on (in some cases advanced) infrastructure and expertise, requiring investment in training and capacity-building for local technicians and farmers. Economic constraints also play a significant role: the initial investment costs for installing nutrient recovery systems can be a big barrier. As a reference, a household-scale biogas digester in China costs between 368 and 792 USD [ 65 ]. To encourage adoption, financial incentives such as subsidies or low-interest loans could be provided.
An important condition for stakeholders to adopt technologies to treat organic wastes for bio-based fertilizers’ production is that there needs to be a market for those fertilizers. Users of organic fertilizers have high quality expectations (nutrient content, nutrient release rates, risk) [ 66 ]. Moreover, a Chinese study on barriers to replace mineral fertilizers with manure showed that farmers have an overall negative attitude, lack of knowledge, and limited experience [ 67 ]. Targeted marketing strategies that provide information on the benefits and costs of implementation are needed, as well as regulatory tools that ensure price stability (to maximise the market share of these fertilizers, they should be 30–46% cheaper than equivalent mineral fertilizers) [ 68 ]. Sutton et al. [ 7 ] stress the importance of a holistic approach to nutrient management, advocating for improved nutrient use efficiency throughout the entire food chain to enhance food and energy production while minimizing losses that cause environmental impacts.
This study quantifies untapped theoretical recycling potentials for N and P throughout the tangerine and water bamboo life cycle in Huangyan, China in 2019 (370 ± 47 Mg N and 24 ± 3 Mg P for tangerines, and 384 ± 136 Mg N and 81 ± 29 Mg P for water bamboo). Hotspots for nutrient recovery include inefficient agricultural residue utilization, low recovery of nutrients from organic residues in municipal solid wastes (most N and P is landfilled), and missing recovery of N and P during wastewater treatment. N and P imports of 427 ± 128 Mg/a for tangerines and water bamboo production are contrasted by N and P exports of 2,189 ± 560 Mg/a. Such imbalances are symptomatic for cash crop producing regions and result in the need of chemical fertilizer imports. Implementing circular economy management approaches such as precision farming, source waste separation, composting, or anaerobic digestion, combined with advanced wastewater treatment concepts could improve N and P recovery.
The recovery of these nutrients could substitute 59% of N and 15% of P supplied by chemical fertilizers for tangerines and water bamboo production in Huangyan, contributing to the circular economy and the achievement of political goals in Taizhou and China. Using untapped nutrients enables further production growth while reducing chemical fertilizer application, contributing to the zero-growth action plan of the Chinese Ministry of Agriculture. The use of organic fertilizers from agricultural residues and municipal solid waste and wastewater allows low-carbon agriculture and improvement of the utilization level of agricultural resources in line with the 14th 5-year plan of the Chinese government, paving the path towards carbon neutrality in 2060. On a regional level, implementing these circular concepts would allow Taizhou to reach the goal of increasing the use of organic fertilizer, digestate from anaerobic digestion, and green manure; improving soil structure and organic matter content. Taizhou could exceed the expected nutrients saving of 0.6 kg (sum of N, P 2 O 5 and K 2 O) per Mu—or more than 1,000 Mg per year (worth over 5 million yuan)—reducing air pollution (mainly caused by straw burning), eutrophication of water bodies, soil erosion and salinization.
Reliable data are crucial for potential assessments, yet currently available Chinese data for regional material and substance flows stems from statistical yearbooks, which lack scientific basis. To address this uncertain data foundation, this study defines data representativeness levels. This enables a first quantification and visualization and supports the identification of data weak points for future studies to address. The study identified theoretical recycling potentials, but several challenges related to technology efficiencies, costs, and users’ acceptance need to be addressed to ensure feasibility.
Steffen W, Richardson K, Rockstrom J et al (2015) Planetary boundaries: guiding human development on a changing planet. Science 347:1259855–1259855. https://doi.org/10.1126/science.1259855
Article Google Scholar
Campbell B, Beare D, Bennett E et al (2017) Agriculture production as a major driver of the earth system exceeding planetary boundaries. Ecol Soc. https://doi.org/10.5751/ES-09595-220408
Article MATH Google Scholar
Tanabe Y, Nishibayashi Y (2013) Developing more sustainable processes for ammonia synthesis. Coord Chem Rev 257:2551–2564. https://doi.org/10.1016/j.ccr.2013.02.010
Cherkasov N, Ibhadon AO, Fitzpatrick P (2015) A review of the existing and alternative methods for greener nitrogen fixation. Chem Eng Process Process Intensif 90:24–33. https://doi.org/10.1016/j.cep.2015.02.004
Capdevila-Cortada M (2019) Electrifying the Haber-Bosch. Nat Catal 2:1055–1055. https://doi.org/10.1038/s41929-019-0414-4
European Commission (2020) Communication from the commission to the European parliament, the council, the European economic and social commitee and the commitee of the regions–critical raw materials resilience: charting a path towards greater security and sustainability
Sutton MA, Bleeker A, Howard CM et al (2013) Our nutrient world: the challenge to produce more food and energy with less pollution. Global overview of nutrient management. Centre for Ecology and Hydrology, Edinburgh
Google Scholar
U.S. Department of Agriculture (2022) Impacts and repercussions of price increases on the global fertilizer market. Foreign Agricultural Service, Washington DC
FAO (2022) The importance of Ukraine and the Russian federation for global agricultural markets and the risks associated with the war in Ukraine. Food and Agriculture Organization of the United Nations, Rome
MATH Google Scholar
UNEP (1995) Programme of action for the protection of the marine environment from land based activities
European Commission (2023) Carbon border adjustment mechanism. https://taxation-customs.ec.europa.eu/carbon-border-adjustment-mechanism_en . Accessed 24 Jun 2024
Statista (2024) Consumption of fertilizers worldwide in 2021, by country. In: Statista. https://www.statista.com/statistics/1287852/global-consumption-fertilizer-by-country/ . Accessed 25 Jun 2024
OECD (2022) Agricultural policy monitoring and evaluation 2022: reforming agricultural policies for climate change mitigation. OECD, Paris
Book Google Scholar
Wang S, Li D, Li T, Liu C (2021) Land use transitions and farm performance in China: a perspective of land fragmentation. Land 10:792. https://doi.org/10.3390/land10080792
Lohmar B, Gale F, Tuan F, Hansen J (2009) China’s ongoing agricultural modernization: challenges remain after 30 years of reform. United States Department of Agriculture, Washington, DC
Ritchie H, Roser M, Rosado P (2022) Fertilizers. Our World Data
van Wesenbeeck CFA, Keyzer MA, van Veen WCM, Qiu H (2021) Can China’s overuse of fertilizer be reduced without threatening food security and farm incomes? Agric Syst 190:103093. https://doi.org/10.1016/j.agsy.2021.103093
Shuqin J, Fang Z (2018) Zero growth of chemical fertilizer and pesticide use: China’s objectives, progress and challenges. J Resour Ecol 9:50–58. https://doi.org/10.5814/j.issn.1674-764x.2018.01.006
FAO (2022) 14th five-year national agricultural green development plan. In: Food Agric. Organ. U. N. https://www.fao.org/faolex/results/details/en/c/LEX-FAOC205820/ . Accessed 16 Nov 2022
Hou J, Hou B (2019) Farmers’ adoption of low-carbon agriculture in China: an extended theory of the planned behavior model. Sustainability 11:1399. https://doi.org/10.3390/su11051399
Taizhou’s Bureau of Agriculture and Rural Affairs (2021) 14th five-year plan for agricultural and rural modernization in Taizhou. Taizhou
Brunner PH, Rechberger H (2016) Handbook of material flow analysis: for environmental, resource, and waste engineers, second edition, 2nd edn. CRC Press, Boca Raton
Book MATH Google Scholar
Huangyan District Statistics Bureau (2024) Huangyan district statistical report on national economic and social development in 2023. Huangyan
Taizhou Agricultural Bureau (2020) Overview of agriculture and rural areas. In: Taizhou Agric. Bur. http://nyncj.zjtz.gov.cn/art/2020/5/25/art_1229051573_1919.html . Accessed 22 Jul 2020
Taizhou Municipal Bureau of Statistics (2022) Taizhou 2021 national economic and social development statistical bulletin. Taizhou
FAO (2023) Crops and livestock products
Guo HB, Li SM, Peng J, Ke WD (2007) Zizania latifolia Turcz. cultivated in China. Genet Resour Crop Evol 54:1211–1217. https://doi.org/10.1007/s10722-006-9102-8
U.S. Department of Agriculture (2020) Search result for tangerine, raw. In: FoodData Cent. https://fdc.nal.usda.gov/fdc-app.html#/food-details/1102601/nutrients . Accessed 13 Nov 2020
Moss RH, Schneider SH (2000) Uncertainties in the IPCC TAR: recommendations to lead authors for more consistent assessment and reporting. World Meteorol Organ
Wang Y, Yang X, Chen K, et al (2019) Interview with various members of Huangyan Agriculture and Forestry Bureau
Fang, Lu, Cao, Li (2019) Interview with various members of the Institute of Citriculture at Zhejiang Academy of Agricultural Sciences
Tao Y (2019) Interview with Mr. Tao, Director of the Seed Management Department at Xihe Ecological Agriculture Development Co.
Cheng (2019) Interview with Mr. Cheng, Head of Fengyan Village, at Huangyan Duanjiang orange cooperative
Zhao GF (2019) Interview with Mr. Zhao, Director of the Agriculture Institute, at Taizhou Vocational College of Science and Technology
Ni, Wu, Ke, Xu (2019) Interview with various members of Taizhou Agricultural Leading Enterprises Association (TALEA)
Ke (2019) Interview with Mr. Ke, General Manager of Taizhou Yiguan Food Co. Ltd.
Huangyan Agriculture and Forestry Bureau (2020) German-Chinese cooperation “Urban-Rural Assembly”. Organic recycling questionnaire
Xu (2019) Interview with Ms. Xu, Vice Head of the Department of Commercial Crops, at Taizhou Agricultural Bureau
Zhang XF (2019) Interview with Mr. Zhang Xue Feng, manager of Luqiao market
(2019) Interview with various sellers at Huangyan market
Piesse M (2017) The Wasteful Dragon: Food Loss and Waste in China. Future Dir Int Pty Ltd
Zhang K (2019) Interview with Mr. Zhang, Project Manager of Vortex Technology Co,.Ltd.
Yu H (2019) Interview with Mr. Yu, engineer at Huangyan Waste Ecological Landfill
China Wastewater Treatment Engineering Network (2017) Huangyan district yuanqiao sewage treatment plant. In: China wastewater Treat. Eng. Netw. https://www.dowater.com/company/2017-05-18/552508.html . Accessed 21 Nov 2020
Leschen W (2018) Thematic background study–genetic resources for farmed freshwater macrophytes: a review
Yan N, Du Y, Liu X et al (2018) Morphological characteristics, nutrients, and bioactive compounds of Zizania latifolia , and health benefits of its seeds. Molecules 23:1561. https://doi.org/10.3390/molecules23071561
Xie (2019) Interview with Mr. Xie, owner of Beiyang Manchurian wild rice farm
Yang L (2019) Interview with Mr. Yang, owner of Liang Jun Manchurian Wild Rice Farm
Tao K (2019) Interview with Mr. Tao, owner of Juyuandi farm
Muñoz I, i Canals LM, Clift R, Doka G (2007) A simple model to include human excretion and wastewater treatment in life cycle assessment of food products. Cent Environ Strategy Univ Surrey Guildf Surrey
Jensen MB, Møller J, Scheutz C (2017) Assessment of a combined dry anaerobic digestion and post-composting treatment facility for source-separated organic household waste, using material and substance flow analysis and life cycle inventory. Waste Manag 66:23–35. https://doi.org/10.1016/j.wasman.2017.03.029
Buckwell A, Nadeu E (2016) Nutrient recovery and reuse (NRR) in European agriculture. A review of the issues, opportunities, and actions. RISE Foundation, Brussels
George E (2020) Consultation about phosphorus flows in tangerine production with Prof. George, Director of Research at the Leibniz Institute of Vegetable and Ornamental Crops (IGZ)
Golomb A, Goldschmidt EE (1987) Mineral nutrient balance and impairment of the nitrate-reducing system in alternate-bearing ‘wiling’ Mandarin trees. J Am Soc Hortic Sci 112:397–401
Bedoić R, Ćosić B, Duić N (2019) Technical potential and geographic distribution of agricultural residues, co-products and by-products in the European union. Sci Total Environ 686:568–579. https://doi.org/10.1016/j.scitotenv.2019.05.219
Zhang QH, Yang WN, Ngo HH et al (2016) Current status of urban wastewater treatment plants in China. Environ Int 92–93:11–22. https://doi.org/10.1016/j.envint.2016.03.024
Aramaki T, Thuy NTT (2010) Material flow analysis of nitrogen and phosphorus for regional nutrient management: case study in Haiphong, Vietnam. In: Sumi A, Fukushi K, Honda R, Hassan KM (eds) Sustainability in food and water: an Asian perspective. Springer, Netherlands, Dordrecht, pp 391–399
Chapter MATH Google Scholar
Pham HG, Harada H, Fujii S et al (2017) Transition of human and livestock waste management in rural Hanoi: a material flow analysis of nitrogen and phosphorus during 1980–2010. J Mater Cycles Waste Manag 19:827–839. https://doi.org/10.1007/s10163-016-0484-1
Schaffner M, Bader H-P, Scheidegger R (2011) Modeling non-point source pollution from rice farming in the Thachin river basin. Environ Dev Sustain 13:403–422. https://doi.org/10.1007/s10668-010-9268-2
Taizhou Statistical Bureau (2019) Statistics yearbook 2019 of Taizhou city. Taizhou
Vitousek PM, Aber JD, Howarth RW et al (1997) Human alteration of the global nitrogen cycle: sources and consequenceS. Ecol Appl 7:737–750. https://doi.org/10.1890/1051-0761(1997)007[0737:HAOTGN]2.0.CO;2
de Boer MA, Wolzak L, Slootweg JC (2019) Phosphorus: reserves, production, and applications. In: Ohtake H, Tsuneda S (eds) Phosphorus recovery and recycling. Springer, Singapore, pp 75–100
Chapter Google Scholar
Möller K (2015) Effects of anaerobic digestion on soil carbon and nitrogen turnover, N emissions, and soil biological activity. A review. Agron Sustain Dev 35:1021–1041. https://doi.org/10.1007/s13593-015-0284-3
Wong JWC, Wang X, Selvam A (2017) Improving compost quality by controlling nitrogen loss during composting. In: Wong JWC, Tyagi RD, Pandey A (eds) Current developments in biotechnology and bioengineering. Elsevier, Amsterdam, pp 59–82
International Renewable Energy Agency (2017) Biogas for domestic cooking–technology brief. IRENA, Abu Dhabi
Garmendia-Lemus S, Moshkin E, Hung Y et al (2024) European farmers’ perceptions and intentions to use bio-based fertilisers: insights from the theory of planned behaviour and perceived utility. J Clean Prod 434:139755. https://doi.org/10.1016/j.jclepro.2023.139755
Zhang Q, Xiao H, Duan M et al (2015) Farmers’ attitudes towards the introduction of agri-environmental measures in agricultural infrastructure projects in China: evidence from Beijing and Changsha. Land Use Policy 49:92–103. https://doi.org/10.1016/j.landusepol.2015.07.021
Moshkin E, Garmendia Lemus S, Bamelis L, Buysse J (2023) Assessment of willingness-to-pay for bio-based fertilisers among farmers and agricultural advisors in the EU. J Clean Prod 414:137548. https://doi.org/10.1016/j.jclepro.2023.137548
Neset T-SS, Bader H-P, Scheidegger R (2006) Food consumption and nutrient flows: nitrogen in Sweden since the 1870s. J Ind Ecol 10:61–75. https://doi.org/10.1162/jiec.2006.10.4.61
Liu Y, Lyu Q, He S et al (2015) Prediction of nitrogen and phosphorus contents in citrus leaves based on hyperspectral imaging. Int J Agric Biol Eng 8:80–88
Menino MR, Carranca C, de Varennes A (2007) Distribution and remobilization of nitrogen in young non-bearing orange trees grown under mediterranean conditions. J Plant Nutr 30:1083–1096. https://doi.org/10.1080/01904160701394543
U.S. Department of Agriculture (2019) Search result for tangerines, (mandarin oranges), canned, juice pack. In: FoodData Cent. https://fdc.nal.usda.gov/fdc-app.html#/food-details/169106/nutrients . Accessed 15 Nov 2020
Czech A, Zarycka E, Yanovych D et al (2020) Mineral content of the pulp and peel of various citrus fruit cultivars. Biol Trace Elem Res 193:555–563. https://doi.org/10.1007/s12011-019-01727-1
U.S. Department of Agriculture (2019) Search result for tangerine peel. In: FoodData Cent. https://fdc.nal.usda.gov/fdc-app.html#/food-details/444009/nutrients . Accessed 15 Nov 2020
Soceanu A, Dobrinas S, Birghila S et al (2009) Levels of phosphorus in citrus fruits. Ovidius Univ Ann Chem 20:87
Liu Y, Shi J, Langrish TAG (2006) Water-based extraction of pectin from flavedo and albedo of orange peels. Chem Eng J 120:203–209. https://doi.org/10.1016/j.cej.2006.02.015
U.S. Department of Agriculture (2019) Search result for tangerine juice, raw. In: FoodData Cent. https://fdc.nal.usda.gov/fdc-app.html#/food-details/169925/nutrients . Accessed 15 Nov 2020
Yan N, Zhang Y-L, Xue H-M et al (2015) Changes in plant growth and photosynthetic performance of Zizania latifolia exposed to different phosphorus concentrations under hydroponic condition. Photosynthetica 53:630–635. https://doi.org/10.1007/s11099-015-0149-7
Dibar DT, Zhang K, Yuan S et al (2020) Ecological stoichiometric characteristics of Carbon (C), Nitrogen (N) and Phosphorus (P) in leaf, root, stem, and soil in four wetland plants communities in Shengjin Lake. China PLOS ONE 15:e0230089. https://doi.org/10.1371/journal.pone.0230089
Chen C, Wang F, Hong Y et al (2019) The biomass accumulation and nutrient storage of five plant species in an in–situ phytoremediation experiment in the Ningxia irrigation area. Sci Rep 9:11365. https://doi.org/10.1038/s41598-019-47860-8
U.S. Department of Agriculture (2019) Search result for wild rice, raw. In: FoodData Cent. https://fdc.nal.usda.gov/fdc-app.html#/food-details/169726/nutrients . Accessed 14 Nov 2020
Surendiran G, Alsaif M, Kapourchali FR, Moghadasian MH (2014) Nutritional constituents and health benefits of wild rice ( Zizania spp.). Nutr Rev 72:227–236. https://doi.org/10.1111/nure.12101
Jain A, Singh HB, Bhattacharyya PR (2012) The ethnobotany and nutritional values of wild rice [ Zizania latifolia (Griseb.) Turcz. ex Stapf] (Poaceae) in Manipur. NISCAIR-CSIR, New Delhi
Zhai CK, Lu CM, Zhang XQ et al (2001) Comparative study on nutritional value of Chinese and North American wild rice. J Food Compos Anal 14:371–382. https://doi.org/10.1006/jfca.2000.0979
U.S. National Center for Biotechnology Information (2020) PubChem Compound Summary for CID 1176, Urea. https://pubchem.ncbi.nlm.nih.gov/compound/1176 . Accessed 16 Nov 2020
Roy RN, Food and Agriculture Organization of the United Nations (2006) Plant nutrition for food security: a guide for integrated nutrient management. Food and Agriculture Organization of the United Nations, Rome
Download references
We gratefully acknowledge the support of the Open Access Publication Fund of the Technische Universität Berlin.
Open Access funding enabled and organized by Projekt DEAL. This research was funded by the German Ministry for Research and Education (BMBF) in the context of the research and development project Urban–Rural Assembly (URA)–Managing inclusive transformation-to-sustainability processes at the urban–rural interface of the Huangyan-Taizhou region in China (Grant No. 01LE1804A). Julia Santolin is a holder of a Ph.D. fellowship strategic basic research from the Research Foundation—Flanders (Grant No. 1S57222N).
Authors and affiliations.
Department of Environmental Technology, Chair of Circular Economy and Recycling Technology, Technische Universität Berlin, 10623, Berlin, Germany
Julia Santolin, Oliver Christopher Larsen, Albrecht Fritze, Bing Xue, Zheng Yang & Vera Susanne Rotter
Biobased Sustainability Engineering (SUSTAIN), Department of Bioscience Engineering, University of Antwerp, Groenenborgerlaan 171, 2020, Antwerp, Belgium
Julia Santolin
You can also search for this author in PubMed Google Scholar
Correspondence to Julia Santolin .
Conflict of interest.
The authors have no conflicts of interest to declare that are relevant to the content of this article.
Publisher's note.
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Below is the link to the electronic supplementary material.
Rights and permissions.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/ .
Reprints and permissions
Santolin, J., Larsen, O.C., Fritze, A. et al. Reaching China’s fertilizer reduction goals through nitrogen and phosphorus recovery: a substance flow analysis case study. J Mater Cycles Waste Manag (2024). https://doi.org/10.1007/s10163-024-02067-6
Download citation
Received : 07 March 2024
Accepted : 02 September 2024
Published : 16 September 2024
DOI : https://doi.org/10.1007/s10163-024-02067-6
Anyone you share the following link with will be able to read this content:
Sorry, a shareable link is not currently available for this article.
Provided by the Springer Nature SharedIt content-sharing initiative
Help | Advanced Search
Title: exploring 3d face reconstruction and fusion methods for face verification: a case-study in video surveillance.
Abstract: 3D face reconstruction (3DFR) algorithms are based on specific assumptions tailored to distinct application scenarios. These assumptions limit their use when acquisition conditions, such as the subject's distance from the camera or the camera's characteristics, are different than expected, as typically happens in video surveillance. Additionally, 3DFR algorithms follow various strategies to address the reconstruction of a 3D shape from 2D data, such as statistical model fitting, photometric stereo, or deep learning. In the present study, we explore the application of three 3DFR algorithms representative of the SOTA, employing each one as the template set generator for a face verification system. The scores provided by each system are combined by score-level fusion. We show that the complementarity induced by different 3DFR algorithms improves performance when tests are conducted at never-seen-before distances from the camera and camera characteristics (cross-distance and cross-camera settings), thus encouraging further investigations on multiple 3DFR-based approaches.
Comments: | Accepted at T-CAP - Towards a Complete Analysis of People: Fine-grained Understanding for Real-World Applications, workshop in conjunction with the 18th European Conference on Computer Vision ECCV 2024 |
Subjects: | Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI) |
Cite as: | [cs.CV] |
(or [cs.CV] for this version) | |
Focus to learn more arXiv-issued DOI via DataCite (pending registration) |
Access paper:.
Code, data and media associated with this article, recommenders and search tools.
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs .
Ui - skripsi membership :: kembali.
file digital: 1.
IMAGES
VIDEO
COMMENTS
Practical application: Case study research can inform practice or policy by identifying best practices, lessons learned, or areas for improvement. Contextualization: Case study research takes into account the specific context in which the case is situated, which can help to understand how the case is influenced by the social, cultural, and ...
For example, the case study quotes the social media manager and project manager's insights regarding team-wide communication and access before explaining in greater detail. Takeaway: Highlight pain points your business solves for its client, and explore that influence in greater detail. 3. EndeavourX and Figma.
A case study is a detailed study of a specific subject, such as a person, group, place, event, organization, or phenomenon. Case studies are commonly used in social, educational, clinical, and business research. A case study research design usually involves qualitative methods, but quantitative methods are sometimes also used.
The multiple case studies used in this article as an application of step-by-step guideline are specifically designed to facilitate these business and management researchers. This article presents an easy to read, practical, experience-based, step-by-step guided path to select, conduct, and complete the qualitative case study successfully.
The Case Study Must Display Sufficient Evidence 246 The Case Study Must Be Composed in an Engaging Manner 247 Notes to Chapter 6 248 Application 10: A Multiple-Case Study Integrating Qualitative and Quantitative Data: Proposal Processing at 17 Universities 249 Appendix A. A Note on the Uses of Case Study Research in Psychology 258 Appendix B.
Application of a Case Study Methodology. by. Winston Tellis +. The Qualitative Report, Volume 3, Number 3, September, 1997. Abstract. In the preceding article ( Tellis, 1997), the goals and ...
A case study research paper examines a person, place, event, condition, phenomenon, or other type of subject of analysis in order to extrapolate key themes and results that help predict future trends, illuminate previously hidden issues that can be applied to practice, and/or provide a means for understanding an important research problem with greater clarity.
The purpose of case study research is twofold: (1) to provide descriptive information and (2) to suggest theoretical relevance. Rich description enables an in-depth or sharpened understanding of the case. It is unique given one characteristic: case studies draw from more than one data source. Case studies are inherently multimodal or mixed ...
A case study protocol outlines the procedures and general rules to be followed during the case study. This includes the data collection methods to be used, the sources of data, and the procedures for analysis. Having a detailed case study protocol ensures consistency and reliability in the study.
The Application of Case Study Evaluations. ERIC/TM Digest. This article examines six types of case studies, the type of evaluation questions that can be answered, the functions served, some design features, and some pitfalls of the method. Deutch, C. E. (1996). A course in research ethics for graduate students. College Teaching, 44, 2, 56-60.
An application of multiple case study methodology. American Journal of ... An overview of the types of case study designs is provided along with general recommendations for writing the research ...
Case study examples. While templates are helpful, seeing a case study in action can also be a great way to learn. Here are some examples of how Adobe customers have experienced success. Juniper Networks. One example is the Adobe and Juniper Networks case study, which puts the reader in the customer's shoes.
With the integration of 11 applications in this edition, the book gives readers access to exemplary case studies drawn from a wide variety of academic and applied fields. Ultimately, Case Study Research and Applications will guide students in the successful use and application of the case study research method. Available formats.
A case study is an in-depth, detailed analysis of a specific real-world situation. For example, a case study can be about an individual, group, event, organization, or phenomenon. The purpose of a case study is to understand its complexities and gain insights into a particular instance or situation. In the context of a business, however, case ...
With the integration of 11 applications in this edition, the book gives readers access to...case studies drawn from a wide variety of academic and applied fields. Ultimately, [this book] will guide students in the...design and use of the case study research method. New to this edition: includes 11 in-depth applications that show how researchers ...
Applications of Case Study Research. 2012. This book by Robert K. Yin aims to help researchers and graduate level students to improve their knowledge and skills in conducting case study research. 21 individual applications of case study research are included that allow readers to see how the theory can be applied in practice.
A case study is a document that focuses on a business problem and provides a clear solution. Marketers use case studies to tell a story about a customer's journey or how a product or service solves a specific issue. Case studies can be used in all levels of business and in many industries. A thorough case study often uses metrics, such as key ...
To save you time and effort, I have curated a list of 5 versatile case study presentation templates, each designed for specific needs and audiences. Here are some best case study presentation examples that showcase effective strategies for engaging your audience and conveying complex information clearly. 1. Lab report case study template.
A case study is one of the most commonly used methodologies of social research. This article attempts to look into the various dimensions of a case study research strategy, the different epistemological strands which determine the particular case study type and approach adopted in the field, discusses the factors which can enhance the effectiveness of a case study research, and the debate ...
Most resources tell you that a case study should be 500-1500 words. We also encourage you to have a prominent snapshot section of 100 words or less. The results and benefits section should take the bulk of the word count. Don't use more words than you need. Let your data, images, and customers quotes do the talking.
Purpose of case study methodology. Case study methodology is often used to develop an in-depth, holistic understanding of a specific phenomenon within a specified context. 11 It focuses on studying one or multiple cases over time and uses an in-depth analysis of multiple information sources. 16,17 It is ideal for situations including, but not limited to, exploring under-researched and real ...
Money Transfer Mobile Application - Case Study. Multiple Owners. 1.5k 16.9k. Save. Food Delivery App- Mobile UI/UX Case study. Sameer Ahmed. 2.1k 69.6k. Save. Healthcare App Design. Oyasim Ahmed. 226 707. Save. iOS 18 Widgets (Case Study) Yohan Ribeiro. 124 534. Save. ↗️MOBILE APP DESIGN for EDUCATION (E-LEARNING) WEBSITE.
The personal assets framework suggests that dynamic elements of (a) personal engagement in activities, (b) quality social dynamics, and (c) appropriate settings will influence an athlete's long-term outcomes of performance, personal development, and continued participation in sport. The aim of the present study was to conduct a case study of a Norwegian age-restricted team that was ...
Application of a mobile health data platform for public health surveillance: A case study in stress monitoring and prediction Digit Health. 2024 Jun 8: ... Methods: A pilot study was conducted with 45 participants over 2 weeks, using the MHP to collect stress-related data from Apple Health and perceived stress self-reports. Apple, Withings and ...
Design the Case Study Protocol. The first stage in the case study methodology recommended by Yin (1994) is the development of the case study protocol. This stage is composed of two subheadings: Determine the Required Skills and Develop and Review the Protocol. These are presented in the following discussion.
The average AUC value of over 0.98 confirms the model's accuracy in predicting ground subsidence in our study and underscores its potential application for non-binary classification tasks.
The goal of the application is to provide a simple user interface for teaching staff and students to visualise and interact with probability density functions. ... Investigating the effectiveness of using a technological approach on students’ achievement in mathematicsâ€"case study of a high school in a Caribbean country ...
Linear agricultural nutrient regimes are the principal cause for perturbation of the geochemical cycles for nitrogen (N) and phosphorus (P) and other planetary boundaries. Nutrient cycles are characterized by high spatial disparity and China is a hotspot due to high fertilizer application rates. Using substance flow analysis, this study identified and quantified nutrient flows from ...
In the present study, we explore the application of three 3DFR algorithms representative of the SOTA, employing each one as the template set generator for a face verification system. The scores provided by each system are combined by score-level fusion. ... A Case-Study in Video Surveillance, by Simone Maurizio La Cava and 7 other authors. View ...
Furthermore, with the MBKM program encouraging participants to fully engage in work, these habitus can be more maturely developed. Previous studies tend to show that internships are merely a transitional preparation program for entering the workforce without specifically analyzing the social process of adaptation to the professional world.