High ammonium concentrations in groundwater (GW) pose substantial challenges to existing GW treatment systems, especially simplified systems like household sand filters (HSFs), which are commonly used for heavy metal removal in developing countries. We aimed to answer how the complex interplay of physical and biochemical processes influences ammonium dynamics in HSFs, thus providing insights to optimize HSFs' future operation. We previously conducted a series of column experiments mimicking HSFs. These experiments revealed limited nitrification, thus fluctuating ammonium removal, highlighting the need to identify and better manage the controlling processes in HSFs. Thus, we established a one-dimensional advective-dispersive-reactive model conditioned on data from column experiments under laboratory and field conditions. The model accounts for temporal variations in nitrification reaction kinetics, transport processes, and a previously unconsidered nitrate leaching process. The modeled breakthrough curves captured the complex dynamics of ammonium, nitrite, and nitrate concentrations in laboratory and field column experiments. We showed that potential lags in the biochemical reactions caused by multi-factor induced inhibitions (e.g., heavy metals) lead to limited nitrification. This was supported by the strong hysteresis of the kinetic rates of nitrification in response to ammonium and nitrite concentrations. The parameterized leaching process leads to the verified nitrate contamination. Our scenario analysis further confirmed the limited nitrification efficiency at current setting suggesting modification is needed to eliminate the inhibition effects.

Running detailed, physics-based numerical simulations of subsurface transport is often computationally expensive. This becomes a challenge when calibrating models against observed data using methods that require a large number of model runs, such as Bayesian inference. To address this challenge, surrogate models are frequently used to approximate simulation outputs. Surrogates are trained using input-output pairs generated by the physics-based model. Traditional approaches typically rely on space-filling designs that uniformly cover the entire parameter space. However, for high-dimensional problems, this becomes impractical and tends to waste computational resources on parameter regions that are either physically irrelevant or contradict available measurement data.To overcome these limitations, we utilize a Bayesian Active Learning (BAL) framework that iteratively selects training points most informative for Bayesian inference given available measurements. We employ Gaussian Processes and Bayesian-Polynomial Chaos Expansions as surrogates, which provide probability distributions for their predictions. Our approach takes advantage of these predictive distributions to evaluate candidate training points using information-theoretic criteria. To account for measurement uncertainty and prevent the algorithm from over-sampling local likelihood maxima, we investigate different strategies for representing observations within the selection process. These criteria are integrated into a multi-objective scoring function that balances global exploration (reducing surrogate uncertainty) with targeted exploitation (refining high-likelihood regions). Additionally, we demonstrate how observations from early time steps can iteratively guide the selection of training points to improve predictive accuracy for later, critical periods of the transport process.We test this method on analytical benchmarks and on subsurface transport models. The framework is evaluated in terms of convergence speed and posterior accuracy relative to existing active learning strategies and reference solutions derived from the full physics-based model. Overall, the proposed goal-oriented strategy aims to reduce the number of expensive model evaluations required for surrogate training, improving the efficiency of subsurface characterization, model calibration and predictive modeling.

Different perspectives of risk and approaches to risk assessment exist, which complicate communication, methodological advances and management across disciplines. To help building bridges, we have developed, tested and applied a generalized mathematical framework for risk assessment in a unique community effort, involving researchers from various fields. We present the derivation of the risk equation, which is tailored to civil and environmental engineering risk assessment of spatially-distributed and dynamic systems. We start off with a general framing and then refine individual parts of the equation as much as needed. The individual terms of the unified risk equation explicitly relate to concepts of frequency, intensity, duration, exposure, vulnerability and asset worth. Our approach takes a new perspective on facilitating communication across disciplines by exploiting mathematical formalism: filling our equation with life enforces a clear definition of the relevant terms and thereby helps in ‘translating’ between different terminology in the involved disciplines. For the sake of clarity and accessibility, we keep the framework simple in terms of additive effects and neglect failure cascades or nonlinear multi-hazard impact functions. Hence, while the framework is not designed to provide a one-fits-all-mathematical solution, it can help carve out the specific properties of a system that potentially violate these assumptions, and this is very valuable when talking risk assessment across disciplines. We demonstrate the utility of our proposed framework with ten selected examples from various domains, ranging from groundwater protection through seismic risk assessment to reliability analysis of critical infrastructure. The structured discussion between all involved researchers has greatly improved mutual understanding, which makes us confident that the proposed framework can serve as a catalyst for interdisciplinary advances in communicating and treating risk.

The FInite Volume Neural Network (FINN) is an advanced, physics-aware machine learning (ML) framework that merges the structure and accuracy of traditional numerical methods with the flexibility of artificial neural networks. Its goal is to learn the unknown components (terms or parameters) of only partially known partial differential equations in order to advance the scientific understanding of investigated systems. While FINN enhances flexibility and interpretability, it also introduces significant complexity due to its integrated, highly parameterized neural networks. This makes accurate uncertainty quantification (UQ) of the learned PDE terms increasingly challenging and expensive. For instance, Bayesian UQ methods for physical models, yielding credible intervals, often rely on resource-intensive Markov Chain Monte Carlo sampling, which is often impractical for such complex and highly parameterized problems. However, reliable and efficient UQ is crucial for assessing confidence in new scientific insights and subsequent predictions, particularly when data are scarce. In response, we introduce a novel, computationally efficient ML-assisted inference-with-UQ method, designed to provide confidence intervals for the components of governing equations learned by FINN. This is achieved by performing data-driven bootstrapping of the available observations, followed by repeated FINN training on the resampled data. We demonstrate the effectiveness and speed of our method by computing confidence intervals of the retardation factor of a diffusion-sorption system, showing that our approach provides an efficient and fast tool for obtaining confidence intervals in machine-learned constituents of governing equations.

The Finite‑Volume Neural Network (FINN) merges the rigor of classical numerical discretizations with the flexibility of artificial neural networks (ANNs) to uncover unknown terms or parameters in partially unknown partial differential equations (PDEs). While this hybrid framework enhances flexibility and interpretability, the highly parameterized ANN makes uncertainty quantification (UQ) of the identified PDE components both demanding and computationally expensive, especially when conventional Bayesian approaches rely on costly Markov Chain Monte Carlo sampling. To address this, we introduce a computationally efficient, Machine Learning (ML)‑assisted inference‑with‑UQ scheme that yields confidence intervals for the PDE components learned by FINN. The procedure consists of data‑driven bootstrapping of the available observations and repeated training of FINN on each resampled set. We illustrate the method on the retardation factor of a diffusion‑sorption PDE, showing that it produces trustworthy interval estimates while markedly lowering the computational burden.

Ordinary differential equations (ODEs) are a conventional way to describe the observed dynamics of physical systems. Scientists typically hypothesize about dynamical behavior, propose a mathematical model, and compare its predictions to data. However, modern computing and algorithmic advances now enable purely data-driven learning of governing dynamics directly from observations. In data-driven settings, one learns the ODE's right-hand side (RHS). Dense measurements are often assumed, yet high temporal resolution is typically both cumbersome and expensive. Consequently, one usually has only sparsely sampled data. In this work we introduce ChaosODE (CODE), a polynomial chaos ODE expansion in which we use an arbitrary polynomial chaos expansion (aPCE) for the ODE's RHS, resulting in a global orthonormal polynomial representation of dynamics. We evaluate the performance of CODE in several experiments on the Lotka-Volterra system, across varying noise levels, initial conditions, and predictions far into the future, even on previously unseen initial conditions. CODE exhibits remarkable extrapolation capabilities even when evaluated under novel initial conditions and shows advantages compared to well-examined methods using neural networks (NeuralODE) or kernel approximators (KernelODE) as the RHS representer. We observe that the high flexibility of NeuralODE and KernelODE degrades extrapolation capabilities under scarce data and measurement noise. Finally, we provide practical guidelines for robust optimization of dynamics-learning problems and illustrate them in the accompanying code.

Development of new multiscale mathematical models often entails considerable complexity and multiple undetermined parameters, typically arising from closure relations. To enable reliable simulations, one must quantify how uncertain physical parameters influence model predictions. We propose surrogate-assisted global sensitivity analysis that combines computational efficiency with a rigorous assessment of parameter influence. In this work, we analyze the recently proposed hybrid-dimensional Stokes–Brinkman–Darcy model, which describes fluid flows in coupled free-flow and porous-medium systems with arbitrary flow directions at the fluid–porous interface. The model results from vertical averaging and contains several unknown parameters. We perform surrogate-assisted global sensitivity analysis using Sobol' indices to investigate the sensitivity of the model to variations of physical parameters for two test cases: filtration and splitting flows. However, constructing surrogates for higher-dimensional random fields requires either many training runs or sophisticated sampling strategies. To address this, we compare polynomial chaos surrogates, including sparse and multi-resolution representations, for their efficiency in global sensitivity analysis, using a predefined Sobol' sequence of training samples. Across the tested cases, multi-resolution approach delivers the most accurate estimation of Sobol' indices.

The usability of enzymatically induced calcium carbonate precipitation (EICP) as a method for altering porous media properties, soil stabilization, or biocementation depends on our ability to predict the spatial distribution of the precipitated calcium carbonate in porous media. While current REV-scale models can reproduce the main features of laboratory experiments, they neglect effects like the formation of preferential flow paths and the appearance of multiple polymorphs of calcium carbonate with differing properties. We show that extending an existing EICP model by the conceptual assumption of a mobile precipitate, amorphous calcium carbonate (ACC), allows for the formation of preferential flow paths when the initial porosity is heterogeneous. We apply sensitivity analysis to understand the influence of characteristic parameters of ACC that are uncertain or unknown, and compare two model variations based on different formulations of the ACC detachment term to analyze the plausibility of our hypothesis. An arbitrary polynomial chaos (aPC) surrogate model is trained based on the full model and used to reduce the computational cost of this study.

Understanding and predicting groundwater contaminant transport is inherently challenging due to uncertainties in both field-specific properties and contaminant-related parameters. These uncertainties pose challenges for effective environmental management, including project planning, non-invasive long-term monitoring, and remediation efforts. To address this, we propose a framework that combines geophysical monitoring, surrogate-assisted Bayesian inference, and dimensionality reduction techniques to quantify and reduce these uncertainties and aid in decision making processes. For the implementation of Bayesian inference, our work focuses on electrical resistivity tomography, a geophysical method that is particularly well-suited for the abovementioned purpose due to its sensitivity to variations in fluid content and temperature.The proposed approach addresses two major computational challenges. First, Bayesian inference requires extensive model runs, which can become computationally prohibitive for large domains with fine grids, multiple processes, and multiple time steps. To mitigate this, we use surrogate models that approximate the full physics-based model using input-output data pairs, significantly reducing computational costs. Second, the high-dimensional nature of ERT data complicates both surrogate training and Bayesian inference. High output dimensions lead to increased training times, larger data requirements, and difficulties in likelihood estimation due to the "curse of dimensionality." To overcome this, we incorporate dimension reduction techniques into the framework.Our main focus is to evaluate how surrogate modeling approximations and dimension reduction strategies influence the accuracy and efficiency of Bayesian inference when using ERT measurements for contaminant transport applications. We apply our framework on a 2D synthetic non-reactive contaminant transport scenario, integrating ERT measurements while accounting for uncertainties in both field-specific and contaminant-related parameters. This methodology provides a practical tool for subsurface engineering, offering improvements in planning, parameter estimation, and long-term monitoring to enhance contaminant transport predictions and remediation strategies.

Given the importance of ensuring the safe disposal of radioactive waste, it is vital to understand the targeted subsurface systems and to build physics-based models to predict their dynamic responses to human interventions. Constructing robust predictive models, however, is very challenging due to the systems’ complexity as well as the scarcity and cost of geophysical data acquisition. Optimal matching of data acquisition and predictive simulations is therefore necessary and can be achieved via integrating predictive process modeling, Bayesian parameter estimation, and optimal experimental design into a modular workflow. This allows to quantify the information content of measurement data and therefore enables optimal planning of data acquisition and monitoring strategies. Conducting such data-integrated simulation studies, however, requires a robust workflow management that ensures reproducibility, error management, and transparency. To meet this demand, we established a data-centric approach to workflow control combining error-managed simulations with a functional data hub, providing simulations with direct access to a database of essential material properties. The latter are being made available as site specific scenario compilations along with uncertainty margins and meta information. The data hub serves as an interface facilitating seamless data and simulation exchange to support subsequent model-driven decision-making processes and guarantees that simulations are conducted using manageable, comparable, and reproducible test cases. Furthermore, it ensures that the simulation results can be readily transferred to a designated repository allowing for real-time updates of the model. The implementation of the data hub is based on a Python-based framework for two different use cases:1) GUI-based use case: The graphical user interface (GUI) facilitates data import, export, and visualization, featuring distinct sections for geographic data representation, structured table organization, and comprehensive visualization of physical properties in varying dimensions.2) Module-based use case: Built on the YAML-based data-hub framework, it enables direct integration of simulation modules storing measurements and model parameters in the YAML data format.The data is systematically organized to furnish a versatile data selection framework that allows information to be extracted from a variety of references, including specific on-site measurements, laboratory measurements and other references, thereby enabling a comprehensive exploration of different reference-oriented scenarios.This study showcases the data hub as a management infrastructure for executing a modular workflow. Multiple models—such as process and impact models as well as their surrogates and geophysical inverse models—are generated within this workflow utilizing scenarios provided by the data hub. Our study shows that adopting a data-centric approach to control the simulation workflow proves the feasibility of conducting different data-integrated simulations and enhances the interchangeability of information across different stages within the workflow. The paradigm of sustainable model development ensures reproducibility and transparency of our results, while also offering the possibility of synergetic exchange with other research areas.

Soil temperature and soil moisture in the unsaturated zone depend on each other and are influenced by non-stationary hydro-meteorological forcing factors that are subject to climate change. The transport of both heat and moisture are crucial for predicting temperatures in the shallow subsurface and, as consequence, around and in drinking water supply pipes. Elevated temperatures in water supply pipes (even up to 25°C and above) pose a risk to human health due to increased likelihood of microbial contamination. To model variably saturated flow and heat transport, a partial differential equation (PDE)-based integrated hydrogeological model has been developed and implemented in the DuMuX simulation framework. This model integrates the hydrometeorological forcing functions via a novel interface condition at the atmosphere-subsurface boundary. Relevant soil properties and their dependency on temperatures have been measured as time series at a pilot site at the University of Stuttgart in detail since 2020. Despite these efforts on measurements and model enhancement, some uncertainties remain. These include capillary-saturation relationships in materials where they are difficult to measure, especially in the gravel-type materials that are commonly used above drinking water pipes. To enhance our understanding of the underlying physical processes, we employ Bayesian inference, which is a well-established approach to estimate uncertain or unknown model parameters. Computationally cheap surrogate models allow to overcome the limitations of Bayesian methods for computationally intensive models, when such surrogate models are used in lieu of the physical (PDE)-based model. Here, we use the arbitrary polynomial chaos expansion equipped with Bayesian regularization (BaPC). The BaPC allows to exploit latest (Bayesian) active learning strategies to reduce the number of model-runs that are necessary for constructing the surrogate model. In the present work, we demonstrate the calibration of a PDE-based integrated hydrogeological model using Bayesian inference on a BaPC-based surrogate. The accuracy of the calibrated and predicted temperatures in the shallow subsurface is then assessed against real-world measurement data.

Machine learning, surrogate modeling, and uncertainty quantification pose challenges in data-poor applications that arise due to limited availability of measurement data or with computationally expensive models. Specialized models, derived from Gaussian process emulators (GPEs) or polynomial chaos expansions (PCEs), are often used when only limited numbers of training points are available. The PCE (or its data-driven version, the arbitrary polynomial chaos) is based on a global representation informed by the distributions of model parameters, whereas GPEs rely on a local kernel and additionally assess the uncertainty of the surrogate itself. Oscillation-mitigating localizations of the PCE result in increased degrees of freedom (DoF), requiring more training samples. As applications such as Bayesian inference (BI) require highly accurate surrogates, even specialized models like PCE or GPE require a substantial amount of training data. Bayesian3 active learning (B3AL) on GPEs, based on information theory (IT), can reduce the necessary number of training samples for BI. IT-based ideas for B3AL are not yet directly transferable to the PCE family, as this family lacks awareness of surrogate uncertainty by design. In the present work, we introduce a Bayesian regularized version of localized arbitrary polynomial chaos to build surrogate models. Equipped with Gaussian emulator properties, our fully adaptive framework is enhanced with B3AL methods designed to achieve reliable surrogate models for BI while efficiently selecting training samples via IT. The effectiveness of the proposed methodology is demonstrated by comprehensive evaluations on several numerical examples.

Accurately calibrated models of water distribution networks (WDNs) are crucial in predicting system behavior. However, besides imperfect calibration, more fundamental inconsistencies between WDN models and real WDNs may come from unforeseen factors between model and reality, such as topology errors from missing or excess pipe connections in the network plan, valves in unexpected settings, misspecified pipe diameters by design, incrustation or pipe aging, and leakages. Testing WDN models for inconsistencies and localizing their error sources for model improvement is crucial. Bayesian model evidence (BME) is a statistically motivated quantity that can be used to test models even in the face of uncertain parameters and uncertain data. Yet, BME cannot identify and localize inconsistencies within the network. Therefore, the current paper introduces a novel criterion called local model evidence (LME), which identifies and localizes model-to-data mismatches at specific nodes within a WDN model using Bayesian analysis principles. Based on Bayesian principles, it determines whether any given model-to-data mismatch at a WDN node is plausibly explainable by combining parameter uncertainty and measurement errors or points toward an inconsistency. We first tested the introduced LME criterion for identifying pipe failures on a widely used EPANET benchmark WDN model and then applied it to a real-life WDN. Both the test and application cases successfully demonstrated LME's capability to identify pipe failures. In the fully controlled benchmark case, LME detected partially closed valves, while in the real-life WDN, it identified aging pipes and dimension errors. Overall, we found the LME-based Bayesian framework essential for pinpointing inconsistencies between the WDN model and real-world conditions during the precalibration stage of modeling.

Machine learning approaches have gained high notoriety to approximate computationally-expensive models in the geosciences. Surrogate models are trained using input-output pairs to emulate the numerics of full complexity models. These fast models then assist in forward and inverse uncertainty quantification for various applied problems. However, large input dimensions, typically found in groundwater modelling for very heterogeneous environments, present a challenge for surrogate models. Input dimension reduction (IDR) methods, such as the Karhunen-Loéve expansion (KLE), are known to reduce the number of input parameters used to train surrogate models, while also generating stochastic realizations of the input random fields for groundwater modelling applications. Traditionally, KLE truncates the input parameters such that 90% of the input variance is considered. However, in some applied cases, this dimension remains too large for reliable surrogate model training. Specifically, using a smaller number of input parameters (considering a smaller percentage of the input variance) may introduce IDR-associated errors in the surrogate output. These errors are often overlooked when assessing uncertainty in surrogate model outputs and could be particularly significant in Bayesian inverse modelling. We are offering a surrogate modelling framework tailored for high-dimensional problems that accounts for IDR-induced errors in the context of Bayesian inverse modelling. Our framework allows for more informed decision-making when using surrogate models as approximators and to widen the scope in which surrogates can be used in heterogeneous media applications. We demonstrate the introduced approach using a groundwater flow and transport model with a heterogeneous hydraulic conductivity field to estimate contaminant concentrations and pressure head values.

Spatiotemporal partial differential equations (PDEs) find extensive application across various scientific and engineering fields. While numerous models have emerged from both physics and machine learning (ML) communities, there is a growing trend towards integrating these approaches to develop hybrid architectures known as physics-aware machine learning models. Among these, the finite volume neural network (FINN) has emerged as a recent addition. FINN has proven to be particularly efficient in uncovering latent structures in data. In this study, we explore the capabilities of FINN in tackling the shallow-water equations, which simulates wave dynamics in coastal regions. Specifically, we investigate FINN's efficacy to reconstruct underwater topography based on these particular wave equations. Our findings reveal that FINN exhibits a remarkable capacity to infer topography solely from wave dynamics, distinguishing itself from both conventional ML and physics-aware ML models. Our results underscore the potential of FINN in advancing our understanding of spatiotemporal phenomena and enhancing parametrization capabilities in related domains.

The Federal Company for Radioactive Waste Disposal (BGE mbH) is tasked with the selection of a site for a high-level radioactive waste repository in Germany in accordance with the Repository Site Selection Act. In September 2020, 90 areas with favorable geological conditions were identified as part of step 1 in phase 1 of the Site Selection Act. Representative preliminary safety analyses are to be carried out next to support decisions on the question, which siting regions should undergo surface-based exploration. These safety analyses are supported by numerical simulations building on geoscientific and technical data. The models that are taken into account are associated with various sources of uncertainties. Addressing these uncertainties and the robustness of the decisions pertaining to sites and design choices is a central component of the site selection process. In that context, important research objectives are associated with the question of how uncertainty should be treated through the various data collection, modeling and decision-making processes of the site selection procedure, and how the robustness of the repository system should be improved. BGE, therefore, established an interdisciplinary research cluster to identify open questions and to address the gaps in knowledge in six complementary research projects. In this paper, we introduce the overall purpose and the five thematic groups that constitute this research cluster. We discuss the specific questions addressed as well as the proposed methodologies in the context of the challenges of the site selection process in Germany. Finally, some conclusions are drawn on the potential benefits of a large method-centered research cluster in terms of simulation data management.

The finite volume neural network (FINN) is an exception among recent physics-aware neural network models as it allows the specification of arbitrary boundary conditions (BCs). FINN can generalize and adapt to various prescribed BC values not provided during training, where other models fail. However, FINN depends explicitly on given BC values and cannot deal with unobserved parts within the physical domain. To overcome these limitations, we extend FINN in two ways. First, we integrate the capability to infer BC values on-the-fly from just a few data points. This allows us to apply FINN in situations, where the BC values, such as the inflow rate of fluid into a simulated medium, are unknown. Second, we extend FINN to plausibly reconstruct missing data within the physical domain via a gradient-driven spin-up phase. Our experiments validate that FINN reliably infers correct BCs, but also generates smooth and plausible full-domain reconstructions that are consistent with the observable data. Moreover, FINN can generate precise predictions orders of magnitude more accurate compared to competitive pure ML and physics-aware ML models - even when the physical domain is only partially visible, and the BCs are applied at a point that is spatially distant from the observable volumes.

Abstract Three‐dimensional (3d) numerical models are state‐of‐the‐art for investigating complex hydrodynamic flow patterns in reservoirs and lakes. Such full‐complexity models are computationally demanding and their calibration is challenging regarding time, subjective decision‐making, and measurement data availability. In addition, physically unrealistic model assumptions or combinations of calibration parameters may remain undetected and lead to overfitting. In this study, we investigate if and how so‐called Bayesian calibration aids in characterizing faulty model setups driven by measurement data and calibration parameter combinations. Bayesian calibration builds on recent developments in machine learning and uses a Gaussian process emulator as a surrogate model, which runs considerably faster than a 3d numerical model. We Bayesian‐calibrate a Delft3D‐FLOW model of a pump‐storage reservoir as a function of the background horizontal eddy viscosity and diffusivity, and initial water temperature profile. We consider three scenarios with varying degrees of faulty assumptions and different uses of flow velocity and water temperature measurements. One of the scenarios forces completely unrealistic, rapid lake stratification and still yields similarly good calibration accuracy as more correct scenarios regarding global statistics, such as the root‐mean‐square error. An uncertainty assessment resulting from the Bayesian calibration indicates that the completely unrealistic scenario forces fast lake stratification through highly uncertain mixing‐related model parameters. Thus, Bayesian calibration describes the quality of calibration and correctness of model assumptions through geometric characteristics of posterior distributions. For instance, most likely calibration parameter values (posterior distribution maxima) at the calibration range limit or with widespread uncertainty characterize poor model assumptions and calibration.

Modeling reservoir sedimentation is particularly challenging due to the simultaneous simulation of shallow shores, tributary deltas, and deep waters. The shallow upstream parts of reservoirs, where deltaic avulsion and erosion processes occur, compete with the validity of modeling assumptions used to simulate the deposition of fine sediments in deep waters. We investigate how complex numerical models can be calibrated to accurately predict reservoir sedimentation in the presence of competing model simplifications and identify the importance of calibration parameters for prioritization in measurement campaigns. This study applies Bayesian calibration, a supervised learning technique using surrogate-assisted Bayesian inversion with a Gaussian Process Emulator to calibrate a two-dimensional (2d) hydro-morphodynamic model for simulating sedimentation processes in a reservoir in Albania. Four calibration parameters were fitted to obtain the statistically best possible simulation of bed level changes between 2016 and 2019 through two differently constraining data scenarios. One scenario included measurements from the entire upstream half of the reservoir. Another scenario only included measurements in the geospatially valid range of the numerical model. Model accuracy parameters, Bayesian model evidence, and the variability of the four calibration parameters indicate that Bayesian calibration only converges toward physically meaningful parameter combinations when the calibration nodes are in the valid range of the numerical model. The Bayesian approach also allowed for a comparison of multiple parameters and found that the dry bulk density of the deposited sediments is the most important factor for calibration.