HS2.3.8

EDI
The application of Bayesian approaches in water quality modelling, decision support and risk analysis

Bayesian approaches have become increasingly popular in water quality modelling, thanks to their ability to handle uncertainty comprehensively. This is particularly relevant in environmental decision making where Bayesian inference enables to consider the reliability of predictions of the consequences of decision alternatives, alongside uncertainties related to decision makers’ risk attitudes and preferences, uncertainty related to system understanding and random processes. Graphical Bayesian Belief Networks and related approaches (hierarchical models, ‘hybrid’ mechanistic/data-driven models) can be particularly powerful decision support tools that make it relatively easy for stakeholders to engage in the model building process and inform adaptive water quality management within an uncertainty framework. The aim of this session is to review the state-of-the-art in this field and compare software and procedural choices to consolidate and set new directions for the emerging community of Bayesian water quality modellers. Building on past three years’ success of this session, a specific new emphasize for this year’s session is to explore the utility of Bayesian water quality models in supporting decision making.

We seek contributions from water quality research that use Bayesian approaches to, for example but not exclusively:
• involve stakeholders in model development and maximise the use of expert knowledge
• integrate prior knowledge, especially problematizing the choice of Bayesian priors
• inform risk analysis and decision support using diverse data and evidence
• represent the preferences of the stakeholders in the form of value functions through elicitation, and account for the uncertainty in preferences
• produce accessible decision support tools
• model water quality in data sparse environments
• compare models with different levels of complexity and process representation
• quantify the uncertainty of model predictions (due to data, model structure and parameter uncertainty)
• address the problem of scaling (e.g. disparity of scales between processes, observations, model resolution and predictions) through hierarchical models
• quantify especially model structural error through, for example, Bayesian Model Averaging or structural error terms
• use statistical emulators to allow probabilistic predictions of complex modelled systems
• use machine-learning and data mining approaches to learn from large, possibly high-resolution data sets.

Convener: Miriam Glendell | Co-conveners: Ibrahim Alameddine, Danlu Guo, James E. Sample, Ambuj SriwastavaECSECS
Presentations
| Mon, 23 May, 17:00–18:30 (CEST)
 
Room 2.17

Presentations: Mon, 23 May | Room 2.17

Chairpersons: Miriam Glendell, Ambuj Sriwastava
Decision support and risk analysis
17:00–17:05
|
EGU22-3930
|
ECS
|
On-site presentation
Ambuj Sriwastava and Peter Reichert

Environmental decision support aims to aid decision makers in identifying management alternatives which reflect the societal preferences as close as possible. This requires a representation of societal preferences through describing the preference structure and elicitation of individual preferences. As scientific knowledge is always incomplete, societal preferences aggregate uncertain individual preferences, and both have to be quantified by simplified models, the consideration of uncertainty is of high importance in environmental decision support. Decision analysis, in particular multi-attribute value theory and multi-attribute utility theory, provide a good theoretical basis for such a decision support process. 

Discrete choice experiments are a convenient tool for preference elicitation and their statistical evaluation leads to unbiased estimates of preference model parameters. We demonstrate that by extending discrete choice questions to the elicitation of preference indifference, we can achieve a reduction in the uncertainty of estimated value function parameters by about a factor of three or a reduction in sample size required to achieve the same accuracy by about a factor of ten. This is obtained at the cost of a higher elicitation effort for each question as it involves the provision of preference information through indifference statements. Using synthetically generated data to allow us to analyse potential bias and to perform a sensitivity analysis regarding sample size and uncertainty ranges, we quantitatively compare discrete choice experiments with indifference elicitation regarding the achieved accuracy of parameter estimates. We test these aspects by employing Bayesian inference for parameter estimation for different shapes of the value function using an error model for values as it is often used for the evaluation of discrete choice experiments, and an additional error model for the specification of the indifference point. Through the quantification of the gain in accuracy, our study provides a basis for assessing the trade-off between higher elicitation effort per choice situation and the required sample size. The elicitation of preference indifference opens new perspectives whenever the set of stakeholders from whom preferences have to be elicited is limited, for example in the case of preference elicitation from experts in environmental management. In such cases, the higher elicitation effort may be manageable and results in a similar accuracy of results with about one-tenths of the sample size compared to discrete choice replies or higher accuracy for smaller sample size reductions. 

How to cite: Sriwastava, A. and Reichert, P.: Reducing Sample Size Requirements by Extending Discrete Choice Experiments to Indifference Elicitation, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-3930, https://doi.org/10.5194/egusphere-egu22-3930, 2022.

17:05–17:10
|
EGU22-1539
|
Highlight
|
Virtual presentation
Timothy Clark, Carly Greyell, Norah Kates, Jennifer Lanksbury, and Daniel Nidzgorski

Local wastewater and stormwater utilities invest millions to billions of dollars collected from ratepayers to meet regulatory requirements, protect human life and infrastructure, and protect water quality. In their project prioritization and planning efforts, utilities consider many factors, including the benefit to environmental outcomes. Utilities often compare the environmental benefits of potential projects by only evaluating changes in pollutant loads rather than looking at whether those projects will accomplish better environmental outcomes for people and wildlife. Many utilities lack a framework for considering these ultimate outcomes. In an effort for better-informed decision-making in King County, WA, we developed a framework (the Water Quality Benefits Evaluation toolkit) that connects actions to environmental outcomes.

The toolkit is an adaptable framework containing a watershed pollutant loading model, a pollutant-reduction and cost optimization model, and causal models representing systems surrounding specified environmental outcomes. We developed causal models for six endpoints: toxics in edible fish, fecal contamination at shellfish beds, fecal contamination at swimming beaches, algal toxins at swimming beaches, natural-origin Chinook salmon population health, and Southern Resident Killer Whale population health. The causal models include Bayesian networks, narrative conceptual models, and fish bioaccumulation models. The holistic evaluation of environmental outcomes provides better information to decision-makers to consider alongside other factors such as costs to ratepayers and reversing environmental inequities.

This presentation focuses on the development of the causal models and how they can be applied to support King County’s utility planning decisions. The presentation will also provide insight on how the framework can be employed for additional environmental endpoints and may be adapted to include other types of endpoints, such as equity and community health.

How to cite: Clark, T., Greyell, C., Kates, N., Lanksbury, J., and Nidzgorski, D.: Connecting actions to ecological and human health endpoints - evaluating the benefits of wastewater and stormwater projects, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-1539, https://doi.org/10.5194/egusphere-egu22-1539, 2022.

17:10–17:15
|
EGU22-5652
|
ECS
|
Highlight
|
On-site presentation
Kerr Adams, Christopher (Kit) A. J. Macleod, Marc J. Metzger, Nicola Melville, Rachel Helliwell, Jim Pritchard, Katie Edwards, and Miriam Glendell

The cumulative impacts of future climatic and socio-economic change threaten the ability of freshwater catchments to provide valuable socio-ecological services. Stakeholders who manage freshwater resources require decision-support tools that increase their understanding of catchment system resilience and support the appraisal of adaptive management options. Our research aims to address the following question: Can a Bayesian Network (BN) model support stakeholders in the identification and testing of adaptive management options that help increase catchment system resilience to the impacts of cumulative future change? Using the predominantly arable Eden catchment (320km2), in eastern Scotland as a case study, we invited stakeholders from multiple sectors to participate in a series of workshops aimed at addressing water resource issues and achieving good ecological status in the catchment both now and in the future. Outputs of a BN model that simulates both current and future catchment resilience were presented to stakeholders. Outputs informed the identification of adaptive management options which were grouped into five management scenarios. The effectiveness of each management scenario in increasing catchment system resilience was tested using the BN model to support the appraisal of each management scenario by participating stakeholders. Two optimal adaptive management scenarios were identified; the first optimal management scenario focussed on predominantly nature-based management options such as wetland wastewater treatment methods and rural sustainable drainage systems. The second optimal scenario focussed on resource recovery, including phosphorus recovery from wastewater treatment works and constructed lagoons for crop irrigation. Outputs of the model describing the resilience of the catchment initiated conversations about feasible management options that could be applied across sectors to reduce risk and increase catchment resilience. The ability of the BN model to test and compare adaptive management scenarios in a time-effective manner was seen as an advantage in comparison to conventional methods.

How to cite: Adams, K., Macleod, C. (. A. J., Metzger, M. J., Melville, N., Helliwell, R., Pritchard, J., Edwards, K., and Glendell, M.: Identifying and testing adaptive management options to increase catchment resilience using a Bayesian Network., EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-5652, https://doi.org/10.5194/egusphere-egu22-5652, 2022.

17:15–17:20
|
EGU22-4092
|
ECS
|
Highlight
|
Virtual presentation
Giulia Leone, Ana I. Catarino, Ine Pauwels, Thomas Mani, Michelle Tishler, Matthias Egger, Marie Anne Eurie Forio, Peter L.M. Goethals, and Gert Everaert


Plastic clean-up technologies deployed in rivers and estuaries can be fundamental to assist in plastic litter management and collection and to mitigate plastic pollution. However, it is vital to supply stakeholders with tools to monitor and minimize possible bycatch, as organic debris and biota provide essential functions to riverine and estuarine environments. Currently, even though some of the clean-up technologies companies perform environmental impact assessments, an independent and objective tool is still missing to assist stakeholders in deploying clean-up mechanisms with a minimal impact on biota. To support stakeholders in making informed decisions about which clean-up technology is best deployed under specific conditions, we suggest using Bayesian Belief Networks (BBNs) as a support tool that would ensure an effective plastic clean-up removal and minimum unintentional bycatch. We have identified four clusters of parameters that account for multiple conditions influencing the chances of bycatch and will form the basis of the BBN. To feed the model, we will acquire data from scientific and grey literature, expert knowledge, and experimental work. The data will include information on (i) the environmental conditions of the river (e.g., river flow), (ii) plastic debris characteristics such as size or buoyancy, (iii) biota traits (e.g., size, buoyancy, adhesiveness), and (iv) mechanism of clean-up technologies (e.g., river booms with conveyor belts, curtains of air bubbles). After the training and validating stages, the model can then be used in different river systems to suggest what type of plastic clean-up mechanism is best suited for the local parameters. This model will enable stakeholders, such as river managers and policymakers, to obtain information on the optimal trade-off between plastic removal and minimal collateral bycatch. 

How to cite: Leone, G., Catarino, A. I., Pauwels, I., Mani, T., Tishler, M., Egger, M., Forio, M. A. E., Goethals, P. L. M., and Everaert, G.:  Assisting stakeholders in their choice of riverine and estuarine plastic clean-up technologies with the aid of Bayesian Belief Networks , EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-4092, https://doi.org/10.5194/egusphere-egu22-4092, 2022.

17:20–17:25
|
EGU22-12783
|
Virtual presentation
Jannicke Moe, Sophie Mentzel, Merete Grung, Roger Holten, and Marianne Stenrød

Weather patterns of Northern Europe are projected to change with increased temperature and precipitation by 2050. These climatic changes can potentially affect the transport and degradation of pesticides in the environment. Moreover, pesticide application patterns are expected to be altered as plant disease and insect pests potentially increase. Other agricultural practices are also expected to change such as crop types and application rate. We have used a Bayesian network model to better integrate these potential direct and indirect climate change effects on pesticide exposure, in a probabilistic approach to pesticide risk assessment. The Bayesian network serves as a meta-model to incorporate the predictions from a pesticide fate and transport model (i.e. WISPE). In this study, we ran the exposure prediction model for specific environmental factors linked to a representative Norwegian study area such as soil and site parameters together with chemical properties, under different scenarios of climate model projections and pesticide application patterns. The Bayesian network links the pesticide exposure predictions derived for this study area to effect distributions derived from toxicity tests to predict the probability distribution of the risk quotient to surrounding aquatic ecosystemsThus, this approach takes into account both direct climate change impacts (on pesticides fate and transport) and indirect effects (on pesticide application). Compared to traditional (deterministic) risk assessment methods, this probabilistic approach can better account for uncertainty associated with climate projections,

How to cite: Moe, J., Mentzel, S., Grung, M., Holten, R., and Stenrød, M.: A Bayesian network approach to environmental risk assessment of pesticides: direct and indirect effects of climate change, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-12783, https://doi.org/10.5194/egusphere-egu22-12783, 2022.

17:25–17:30
17:30–17:35
|
EGU22-7796
|
ECS
|
Highlight
|
Virtual presentation
Marie Anne Eurie Forio, Francis J. Burdon, Felix Witing, Geta Risnoveanu, Benjamin Kupilas, Nikolai Friberg, Martin Volk, Brendan Mckie, and Peter Goethals

Despite the benefits of riparian vegetation, they are limitedly implemented in water management – which is partly due to the lack of information on their effectiveness. In this context, social learning is valuable to inform stakeholders of the efficacy of riparian vegetation in mitigating stream degradation. Tools used in social learning activities are of paramount importance in the learning process. We developed a Bayesian belief network (BBN) model as a learning tool to simulate and assess the reach- and segment-scale effects of riparian vegetation properties and subcatchment-scale land use on instream invertebrates. We surveyed reach-scale riparian conditions, extracted segment-scale riparian and land use information from geographic information system (GIS) data and collected macroinvertebrate samples from four catchments in Europe (Belgium, Norway, Romania and Sweden). We modelled the ecological water quality, expressed as Average Score Per Taxon, as a function of different riparian variables using the BBN modelling approach. The collected data were used to populate the conditional probability table of the BBN model. The model simulations provided insights into the usefulness of both reach- and segment-scale riparian vegetation attributes in enhancing ecological water quality. We assessed the strengths and limitations of the BBN model for application as a learning tool. Despite some weaknesses, the BBN model has great potential in workshop activities to stimulate key learning processes that help inform the management of riparian zones.

How to cite: Forio, M. A. E., Burdon, F. J., Witing, F., Risnoveanu, G., Kupilas, B., Friberg, N., Volk, M., Mckie, B., and Goethals, P.: A Bayesian Belief Network model assessing the multi-scale effects of riparian vegetation on stream invertebrates, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-7796, https://doi.org/10.5194/egusphere-egu22-7796, 2022.

17:35–17:40
|
EGU22-10539
|
Presentation form not yet defined
Claire Lauvernet, Céline Helbert, Zhu Xujia, and Bruno Sudret

Significant amounts of pollutant are measured in surface water, their presence due in part to the use of pesticides in agriculture. One solution to limit pesticide transfer by surface runoff is to implement vegetative filter strips (VFS) along rivers. The sizing of these strips is a major issue, with influencing factors that include local conditions (climate, soil, etc.). The BUVARD modeling toolkit was developed to design VFSs throughout France according to these properties. This toolkit includes the numerical model VFSMOD, which quantifies dynamic effects of VFS site-specific pesticide mitigation efficiency. However, the toolkit is quite complex to use with many input uncertain parameters (quantitative - such as the slope, the Curve Number - or qualitative -such as the soil type or the rainfall event), making it not easy to use for risk management.

In this study, a metamodeling (or model dimension reduction) approach is proposed to ease the use of BUVARD and to help users design VFSs that are adapted to specific contexts. Different reduced models, or surrogates, are compared, based on Bayesian learning approaches or not: Polynomial Chaos Expansions, Mixed-kriging, and Deep-GP. Mixed-kriging is a kriging method that was implemented with a covariance kernel for a mixture of qualitative and quantitative inputs. Kriging and Deep-GP are built by couple of modalities and PCE and Mixed-kriging are built considering mixed quantitative and qualitative variables. As a last step, Finally, we perform a global sensitivity analysis with the help of the two surrogate models with the best accuracy. The results show that they give the same ranking of the importance of the input parameters.

The metamodel is a simple way to provide a relevant first guess to help design the pollution reduction device. In addition, the surrogate model is a relevant uncertainty tool, to visualize the impact that lack of knowledge of some parameters of filter efficiency can have when performing risk analysis and management.

How to cite: Lauvernet, C., Helbert, C., Xujia, Z., and Sudret, B.: Metamodeling approaches to help designing vegetative filter strips and improve the water quality., EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-10539, https://doi.org/10.5194/egusphere-egu22-10539, 2022.

17:40–17:45
|
EGU22-11862
|
ECS
|
Presentation form not yet defined
Hemie Cho, Jae-Ung Yu, Jinyoung Kim, and Hyun-Han Kwon

In Korea, algal blooms are repeated every year in the four major rivers. Especially, the frequency and duration of algal blooms have increased due to climate change (increase in the atmosphere and water temperature) and environmental changes (river development projects and weir construction), which has led to the development of related research. However, it is still difficult to elucidate the cause of algal blooms because of a complicated mechanism. In this study, the concentration of Chlorophyll-a was selected as an indicator of algae occurrence, and representative hydrometeorological factors affecting the algal phenomenon were selected. Then, the optimal marginal distribution for each variable was found. The risk of algal blooms was analyzed through bivariate copula analysis by identifying the relationship between the influencing factors and the concentration of Chlorophyll-a. As a result, it was possible to identify the factors that had the most significant influence on the occurrence of algal blooms. Further, this study will employ the Vine Copula function to improve the complex relationship between variables in the context of multivariate modeling.

How to cite: Cho, H., Yu, J.-U., Kim, J., and Kwon, H.-H.: Risk Analysis of Algal Blooms Using the Conditional Copula Model, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-11862, https://doi.org/10.5194/egusphere-egu22-11862, 2022.

17:45–17:50
|
EGU22-941
|
ECS
|
Virtual presentation
Xizhi Nong, Chi Zhang, Dongguo Shao, Hua Zhong, Yuming Shang, and Jiankui Liang

The variation of water quality in long-distance water supply projects is often different from that in natural water bodies, especially the environmental risks of artificial open canal water transfer projects are seldom studied. The spatial heterogeneity of algae growth and the absence of universal reference standards for algae control often lead to water quality problems in these projects. The Middle Route (MR) of the South-to-North Water Transfer Project of China (SNWTPC), the world’s largest inter-basin water transfer project, has operated stably for six years. Its water resources have benefited more than 60 million people and an ecosystem cover more than 160,000 km2. To understand the water environment risk of this mega hydro-project, this study focused on the relationships among three key parameters: water temperature (T), water discharge (Q), flow rate (V) and analyzed the spatial–temporal variation characteristics of the algal cell density (ACD) in the MR of the SNWTPC from January 2016 to December 2018, 36 months in total. The Copula functions were applied to identify and evaluate the multivariate risk variation of the water environment. Our result demonstrated that there was a significant positive correlation between T and ACD, and the ACD at the downstream has a 50% risk higher than 700 × 104 cell/L in summer. Overall, the water quality status of the MR of the SNWTPC is quite well, and the ACD kept at an average level of 106 cells/L during the monitoring period. Additionally, the ACD increased from upstream to downstream, showing the relatively higher ACD in summer and autumn than in spring and winter, with the ranges of 500 ~ 700 × 104 cells/L and 200 ~ 300 × 104 cells/L, respectively. The water temperature affected the ACD over the early-warning thresholds at different canal sections were as follows: 29, 26, and 21℃ from upstream to downstream, respectively. The influences of the hydrodynamic factors, water discharge and flow rate, impact on the ACD variation were analyzed to achieve the purpose of specific algae control for different canal reaches. Our study verified that the growth probabilities of the ACD under higher water temperature and water discharge in the MR of the SNWTPC than other natural water bodies.

    How to cite: Nong, X., Zhang, C., Shao, D., Zhong, H., Shang, Y., and Liang, J.: Multivariate water environmental risk analysis in long-distance water supply project: A case study in China, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-941, https://doi.org/10.5194/egusphere-egu22-941, 2022.

    17:50–17:55
    Improving modelling and understanding
    17:55–18:00
    |
    EGU22-8907
    |
    Presentation form not yet defined
    Ibrahim Alameddine, Jessica Sawma, and Hassan Khalife

    Surface water bodies serve a critical role in preserving ecological systems and maintaining biodiversity. Anthropogenic eutrophication of fresh water ecosystems is one of the main causes of surface water quality degradation. Excessive nutrient loading to freshwater bodies is a driving cause of water quality impairments worldwide. Accurately estimating riverine nutrient loads remains an imperative step towards mitigating and managing impairments. Yet, load estimation is often hindered by the sporadic and infrequent monitoring of nutrient concentrations. Several modelling approaches have been proposed and implemented over the years to estimate pollutant loads; yet most suffer from biases and/or from their capabilities to transparently quantify uncertainties. In this work, we propose a spatio-temporal Bayesian hierarchical ratio-estimator model to predict the annual total phosphorus loads between 2005 and 2020 for six intensively monitored watersheds discharging in Lake Erie and the Ohio River-USA. The integration of higher-level Land-Use-Land-Cover predictors proved successful in capturing inter-station variabilities in phosphorus loading. Meanwhile, accounting for annual climatic variability partially helped explain temporal changes in the flow-weighted nutrient concentrations across the six watersheds. The performance of the model was tested against different levels of data censorship. Results showed that under a weekly sampling program, the load estimates from the proposed Bayesian Hierarchical spatio-temporal model were within -19 and 31 % (mean difference of 0.3% across stations and years) from the true loads calculated for years with uninterrupted concentration measurements. Predictions from traditional load estimation methods were found to vary between -56% and 73% from the true loads. Meanwhile, failing to account for the spatio-temporal hierarchical structure of the proposed model, either by adopting a completely pooled or an unpooled model, resulted in a significant drop in the accuracy of the predicted loads and inflated the associated uncertainties.

    How to cite: Alameddine, I., Sawma, J., and Khalife, H.: A Bayesian hierarchical spatio-temporal ratio-estimator approach to model phosphorus loading in six Ohio watersheds: the importance of accounting for inter-annual and inter-basin variabilities, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-8907, https://doi.org/10.5194/egusphere-egu22-8907, 2022.

    18:00–18:05
    |
    EGU22-11003
    |
    ECS
    |
    Virtual presentation
    Shuci Liu, Danlu Guo, Camille Minaudo, Anna Lintern, Rémi Dupas, Ulrike Bende-Michl, Kefeng Zhang, and Clément Duvert

    Investigations of concentration (C) and discharge (Q) relationships (C–Q relationships) at the catchment scale are commonly used to characterize export regimes of instream particulates and solutes. C–Q relationships also provide insights on spatial and temporal variability in pollutant export, allowing identification of the sources and transfer pathways of pollutants. Previous studies have shown that several key catchment attributes control the export of sediment and dissolved nutrients within catchments. These catchment attributes include land use, topography, geology and soils. However, only few studies have investigated the relative importance of multiple catchment attributes over large spatial scales (e.g., at the continental scale) and between different climate zones. This is mostly due to either a limited number of catchments that have been monitored or a strong focus on temperate catchments. Therefore, our current understanding of key controls on spatial variability and export regimes across different climates is still limited. In this study, we investigated spatial differences and the C–Q relationships of six commonly monitored constituents (i.e., total suspended solid – TSS, total nitrogen – TN, sum of nitrate and nitrite – NOx, total phosphorus – TP, soluble reactive phosphorus – SRP and electrical conductivity – EC) from 507 catchments across the Australian continent. These catchments represent five main climate zones in Australia (i.e., arid, Mediterranean, temperate, subtropical and tropical). We used a hierarchical Bayesian multi-model averaging approach to 1) identify key catchment attributes (e.g., land use, topography, geology and hydrology) driving the spatial variability of mean concentration and export regimes (CQ relationship) for individual constituents; 2) understand the role of climatic gradients in determining the magnitude and direction of the key controls, and 3) use the key controls identified to predict the mean concentration and CQ relationship in multiple catchments across Australia.

    The proposed Bayesian modelling framework provided a higher predictive capability for mean concentrations (Nash-Sutcliffe efficiency (NSE) ranging from 0.58 for SRP to 0.86 for EC), compared to log(C) – log(Q) slopes (NSE ranging from 0.25 for NOx to 0.39 for TP). For mean concentrations, land use (e.g., agriculture and urban) has a significantly positive effect on nutrients (i.e., TN, NOx, TP and SRP), particularly in the Mediterranean, subtropical and tropical regions, indicating that land use is a key driver for these constituents. For log(C) – log(Q) slopes, catchment topographical characteristics (e.g., slope and maximum flow pathway) have relatively high impacts on TSS, TP and EC, indicating export of sediments and solutes in catchments largely controlled by mobilization (sediment) and surface-subsurface flow interaction (solutes). Findings from our study provide a data-driven understanding of key controls on riverine water quality across multiple climate types and can inform future water quality management strategies.

    How to cite: Liu, S., Guo, D., Minaudo, C., Lintern, A., Dupas, R., Bende-Michl, U., Zhang, K., and Duvert, C.: Key controls of catchment attribute on spatial differences and export regimes in riverine water quality: a study across the Australian continent using a Bayesian approach, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-11003, https://doi.org/10.5194/egusphere-egu22-11003, 2022.

    18:05–18:10
    |
    EGU22-6847
    |
    ECS
    |
    Virtual presentation
    Danlu Guo, Camille Minaudo, Anna Lintern, Ulrike Bende-Michl, Shuci Liu, Kefeng Zhang, and Clement Duvert

    Understanding concentration-discharge (C-Q) relationships is critical to inform catchment export processes for solute and particulates. The contribution of baseflow to streamflow has been found to affect C-Q relationships in some catchments in previous studies. Current understanding on the effects of baseflow contribution in shaping the C-Q patterns is largely limited to temperate catchments, but we still lack quantitative understanding of these effects across a wide range of climates (e.g., arid, tropical and subtropical). The study aims to assess how baseflow contributions within individual catchments influence C-Q slopes across Australia. The wide range of hydro-climatic regimes and land use/land cover conditions in Australian catchments make this continent the ideal experimental field to gain such an understanding. We analyzed 157 catchments in Australia spanning five climate zones, for six water quality variables: electrical conductivity (EC), total phosphorus (TP), soluble reactive phosphorus (SRP), total suspended solids (TSS), the sum of nitrate and nitrite (NOx) and total nitrogen (TN). The impact of baseflow contributions was defined by the median and the range of daily baseflow indices (BFI_m and BFI_range, respectively) for each catchment. A novel Bayesian hierarchical model was developed to synthesize these effects for individual catchments across the continent.  

    Sediments and nutrient species (TSS, NOx, TN and TP) generally show positive C-Q slopes for most catchments, suggesting a dominance of mobilization export patterns. Further, TSS, NOx and TP show stronger mobilization (i.e., steeper positive C-Q slopes) in catchments with higher values in both the BFI_m and BFI_range, while these two metrics are also positively correlated for most catchments. The enhanced mobilization in catchments with higher BFI_m or BFI_range might be explained by more variable flow pathways in catchments with higher baseflow contributions. In such catchments, the more variable flow pathways can lead to higher concentration gradients between low flows and high flows. These gradients are due to  different dominant flow pathways and contributions of groundwater/slow subsurface flow and surface water sources. Our results highlight the need for further studies focusing on identifying and quantifying: a) the influences of temporal variations of baseflow contributions on flow pathways, and b) the impacts of variable flow pathways on catchment C-Q relationships.

    How to cite: Guo, D., Minaudo, C., Lintern, A., Bende-Michl, U., Liu, S., Zhang, K., and Duvert, C.: How does baseflow contribution affect catchment C-Q relationships? A continental synthesis using a Bayesian Hierarchical Model, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-6847, https://doi.org/10.5194/egusphere-egu22-6847, 2022.

    18:10–18:15
    |
    EGU22-12525
    |
    ECS
    |
    Virtual presentation
    Anneli Guthke, Han-Fang Hsueh, Thomas Wöhling, and Wolfgang Nowak

    Bayesian mechanistic modeling often suffers from overconfident and biased posterior distributions for parameters and predictions. This phenomenon arises because the fundamental assumption of Bayesian Model Analysis is violated: the underlying model is assumed to be true, but in fact, it is a simplification of reality with structural errors that show at least during some periods of the modeled time span. As a result, a compromise solution in parameter space is identified that can formally fit the full data set best, but this parameter set will not be representative of the true system state. Neither will it be representative of the “compensation mode” in which the model is whenever structural error kicks in. As a logical consequence, predictions will be biased and their intervals too narrow. The longer the data set used for calibration, the stronger the misleading effect. Typical sources of severe structural deficits that produce dynamically occurring errors are missing or misspecified processes in the model. 

    We propose a formal time-windowed Bayesian analysis to overcome this general problem. When performing Bayesian updating on shorter time windows, the assumption of a (quasi-) true model becomes more plausible, and by sliding this window through the calibration time series, we let the model adjust its posterior parameter distributions according to the current strength of error. These time-shifting parameter distributions allow us to (1) identify periods of statistically significant model error occurrence via measuring time-varying Bayesian model evidence, (2) diagnose potential sources of model error by understanding the time-varying parameter compensation mechanisms, and (3) predict with more realistic uncertainty intervals by distribution averaging. 

    We demonstrate the proposed method on a set of synthetic and real-world scenarios of soil moisture modeling. With this example, we also highlight its usefulness to analyze dynamic systems in a wide range of disciplines, such as water quality modeling, decision support, and risk assessment. Results show that the time sequence of posterior parameter distributions (and dependent model mechanisms such as water retention curves and unsaturated hydraulic conductivity functions) provides valuable insights into the model’s weaknesses, and it also provides guidance for model improvement.

    How to cite: Guthke, A., Hsueh, H.-F., Wöhling, T., and Nowak, W.: Bayesian updating despite model errors? A sliding time-window approach to rescue , EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-12525, https://doi.org/10.5194/egusphere-egu22-12525, 2022.

    18:15–18:20
    |
    EGU22-11975
    |
    ECS
    |
    Virtual presentation
    Ruiyi Yang, Jiping Jiang, Tianrui Pang, Yunlei Men, and Yi Zheng

    The source identification of surface water pollution is the key issue in environmental management and has important practical needs. Model-based numerical inversion methods have received widespread attention that there are many research reports. However, most of the existing research on the pollution source identification (PSI) problem focuses on the combination innovation and theoretical analysis of the inversion algorithm, and does not consider the urgent time constraints of the emergency response process, which has become an important technical bottleneck. To this end, this study focuses on the timeliness and operability of numerical inversion methods in the emergency response process, and makes full use of multi-source information of pollutants to explore robust and fast sampling and numerical inversion methods in practical operations. The study adopts the Adative Metroplis Monte Carlo (AM-MCMC) Bayesian method as the basic source identification inversion framework, and takes the USGS tracer test in Truckee River from 2006 to 2007 as the basic scenario to carry out numerical experiments. Through the data assimilation method, the pollution source information is dynamically updated. With the input of new monitoring data, the accuracy of the inversion results is gradually improved; By integrating multiple pollutants information, greatly improves the robustness and practical ability of the numerical source identification technology. The study establishes the best practice method of parallel sampling, which can achieve reliable numerical inversion accuracy in the early stage of sampling. The quantitative design criteria of the minimum sampling cost required for inversion to meet a certain error limit under different river hydrological conditions are discussed, and the relative critical time Λ of sampling and Peclet number (Pe) that characterizes river hydrodynamics are found as follow relation: Λ=-0.816×Pe-1/2-0.978×Pe-1/2lnPe-1/2+0.554, R2=0.938, and has been proved by information entropy theory. The precise design process of the emergency PSI monitoring scheme with the timeliness as the optimization goal is further proposed. This study provides important theoretical guidance for the innovative application of new monitoring methods in scenarios such as leakage detection and emergency monitoring under the background of environmental Internet of Things.

    How to cite: Yang, R., Jiang, J., Pang, T., Men, Y., and Zheng, Y.: New insights on the practical significance of numerical methods for surface water pollution source identification, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-11975, https://doi.org/10.5194/egusphere-egu22-11975, 2022.

    18:20–18:25
    |
    EGU22-6062
    |
    ECS
    |
    Presentation form not yet defined
    Holger Virro, Alexander Kmoch, Marko Vainu, and Evelyn Uuemaa

    Water quality modeling plays an important role in better understanding the magnitude and impact of water quality issues and in providing evidence for policy-making and implementing measures to mitigate water pollution. Process-based nutrient models are very complex, requiring a lot of input parameters and computationally expensive calibration. Often there is also a lack of high spatial and temporal resolution water quality data because water sampling is expensive and river water quality can’t be measured using remote sensing. Machine learning approaches have been shown to achieve similar accuracy to the physically-based models and even outperform them when describing nonlinear relationships. We used 242 observation sites located at 139 streams in Estonia, amounting to 469 yearly total nitrogen (TN) and 470 total phosphorus (TP) measurements covering the period 2016–2020 to train random forest models for predicting N and P concentrations. We used a total of 82 predictor variables, including land cover, soil, climate, and topography parameters, and applied a feature selection strategy to reduce the number of dependent features in the model. The models resulted in an accuracy of 82% in the case of TN and 54% for TP. The SHAP (SHapley Additive exPlanations) values used to explain the models showed that the most important features for predicting TN were arable land proportion, soil rock content, and hydraulic conductivity, while the main features affecting TP concentration were the urban and grassland proportion in the catchment. The results indicate that the TN model is a viable alternative to process-based models in Estonia. In the case of TP, the derived feature importances and feature interactions can potentially help improve the corresponding model in the future. 

    How to cite: Virro, H., Kmoch, A., Vainu, M., and Uuemaa, E.: Machine learning-based water quality modeling at national level in data-scarce region , EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-6062, https://doi.org/10.5194/egusphere-egu22-6062, 2022.

    18:25–18:30