Max Gustav Rudolph
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Max Gustav Rudolph has a total of 19 co-authors. Here are the top co-authors by number of shared publications:
- Andreas Hartmann
Institute of Groundwater Management, TU Dresden, Dresden, Germany
9 shared publications - Thomas Wöhling
Technische Universität Dresden
7 shared publications - Thorsten Wagener
Institute of Environmental Science and Geography, University of Potsdam, Potsdam, 14476, Germany
7 shared publications Alireza Kavousi
Institute of Groundwater Management, Technische Universität Dresden, Dresden, Germany
2 shared publicationsAndreas Hartmann
Institute of Groundwater Management, Technische Universität Dresden, Dresden, Germany
2 shared publicationsMarkus Giese
Department of Earth Sciences, University of Gothenburg, Gothenburg, Sweden
2 shared publications
Latest Publications
- Open Access
- 23 October 2026
- Published by American Geophysical Union (AGU) in Journal of Geophysical Research: Biogeosciences
- Vol. 131 (10)
Direct Flux Measurements Reveal CO 2 Sinks in Icelandic Glacial Streams Driven by Weathering and Hydrological Mixing
Glacier‐fed rivers are increasingly recognized as important components of high‐latitude carbon cycling, yet direct observations of CO 2 exchange in proglacial streams remain limited, especially in young basaltic catchments. Here, we quantify spatial and seasonal patterns of CO 2 exchange along a 3‐km reach of the glacier‐fed stream Virkisá, originating from Falljökull, and complement these with measurements from four glacial streams draining the Öræfajökull ice cap in Iceland. Across four seasons, we conducted 154 direct CO 2 flux measurements at six sites along Virkisá, combined with hydrochemical data, dissolved organic carbon (DOC) concentrations, and stable isotope‐based water source apportionment. The stream consistently acted as a net atmospheric CO 2 sink, with the strongest uptake near the glacier outlet and progressively weaker uptake downstream. pH and distance from the glacier were the strongest predictors of CO 2 flux. Elevated pH and CO 2 uptake near the glacier outlet reflect rapid weathering of freshly eroded basaltic sediments generating CO 2 ‐undersaturated meltwater in which dissolved CO 2 is predominantly converted to bicarbonate. Uptake declined downstream as the water progressively equilibrated with the atmosphere, although net uptake persisted throughout the reach. The chemical signature of meltwater, shaped by the mixing of glacier ice melt, snowmelt, supraglacial runoff, and groundwater, further modified water chemistry and CO 2 exchange. Despite seasonal variability in discharge composition, CO 2 fluxes showed no significant seasonal or diurnal differences although peak uptake occurred during spring melt. Therefore, our findings identify proglacial streams as persistent CO 2 sinks and highlight that water chemistry associated with basalt weathering—not biological activity—primarily controls CO 2 dynamics.
- 23 October 2026
- Published by American Geophysical Union (AGU) in Journal of Geophysical Research: Biogeosciences
- Vol. 131 (10)
Open AccessDirect Flux Measurements Reveal CO 2 Sinks in Icelandic Glacial Streams Driven by Weathering and Hydrological Mixing
Glacier‐fed rivers are increasingly recognized as important components of high‐latitude carbon cycling, yet direct observations of CO 2 exchange in proglacial streams remain limited, especially in young basaltic catchments. Here, we quantify spatial and seasonal patterns of CO 2 exchange along a 3‐km reach of the glacier‐fed stream Virkisá, originating from Falljökull, and complement these with measurements from four glacial streams draining the Öræfajökull ice cap in Iceland. Across four seasons, we conducted 154 direct CO 2 flux measurements at six sites along Virkisá, combined with hydrochemical data, dissolved organic carbon (DOC) concentrations, and stable isotope‐based water source apportionment. The stream consistently acted as a net atmospheric CO 2 sink, with the strongest uptake near the glacier outlet and progressively weaker uptake downstream. pH and distance from the glacier were the strongest predictors of CO 2 flux. Elevated pH and CO 2 uptake near the glacier outlet reflect rapid weathering of freshly eroded basaltic sediments generating CO 2 ‐undersaturated meltwater in which dissolved CO 2 is predominantly converted to bicarbonate. Uptake declined downstream as the water progressively equilibrated with the atmosphere, although net uptake persisted throughout the reach. The chemical signature of meltwater, shaped by the mixing of glacier ice melt, snowmelt, supraglacial runoff, and groundwater, further modified water chemistry and CO 2 exchange. Despite seasonal variability in discharge composition, CO 2 fluxes showed no significant seasonal or diurnal differences although peak uptake occurred during spring melt. Therefore, our findings identify proglacial streams as persistent CO 2 sinks and highlight that water chemistry associated with basalt weathering—not biological activity—primarily controls CO 2 dynamics.
- Open Access
- 14 March 2026
- Published by Copernicus GmbH
Towards a Virtual Laboratory for Karst Groundwater Modelling
Groundwater modelling serves as an important tool for the study and management of karst water resources. In such context, and especially when we try to obtain karst system understanding through groundwater models, it is not only important to get the right answers (e.g., a well-fitted spring discharge simulation), but we have to get those answers for the right reasons (i.e., because our model realistically reflects the underlying system). This problem has been addressed before, e.g., via model evaluation upon multiple system signatures such as spring discharge and solute concentrations. However, because the conduit network and the geological structure of karst systems is usually largely unknown, lumped models are often employed, which simplify the karst system and simulate the processes in a spatially aggregated manner. Due to those simplifications, assessing model (structural and process related) realism beyond a model’s ability to simulate system output signals is often impossible. Yet, using models to better understand karst system functioning is vital for sustainable karst water resources management. This discrepancy necessitates the study of karst systems and corresponding (lumped or spatially aggregated) models on the basis of systems that are fully known. For this purpose, we develop a virtual laboratory for karst groundwater modelling. This virtual laboratory facilitates the generation and simulation of synthetic karst systems and system signatures such as spring discharge using state-of-the-art spatially distributed numerical modelling. The virtual laboratory makes the generation of synthetic systems more accessible to a wider hydro(geo)logical community, taking a step towards more realistic process representation in (lumped) karst models in the future. Building on our previous work, this combination allows for the stochastic generation of conduit networks using pyKasso [1] and their subsequent embedding and simulation with the spatially distributed discrete-continuum model MODFLOW CFPv2 [2] via the CFPy package for the Python programming language [3]. We demonstrate the capabilities of the virtual laboratory for a number of cases, show upcoming development goals, and discuss opportunities for future applications. [1] Miville, F., Renard, P., Fandel, C., & Filipponi, M. (2025). pyKasso: An open-source three-dimensional discrete karst network generator. Environmental Modelling & Software, 186, 106362.[2] Reimann, T., Giese, M., Geyer, T., Liedl, R., Maréchal, J. C., & Shoemaker, W. B. (2014). Representation of water abstraction from a karst conduit with numerical discrete-continuum models. Hydrology and Earth System Sciences, 18(1), 227-241.[3] Reimann, T., Rudolph, M. G., Grabow, L., & Noffz, T. (2023). CFPy—A python package for pre‐and postprocessing of the conduit flow process of MODFLOW. Groundwater, 61(6), 887-894.
- 14 March 2026
- Published by Copernicus GmbH
Open AccessTowards a Virtual Laboratory for Karst Groundwater Modelling
Groundwater modelling serves as an important tool for the study and management of karst water resources. In such context, and especially when we try to obtain karst system understanding through groundwater models, it is not only important to get the right answers (e.g., a well-fitted spring discharge simulation), but we have to get those answers for the right reasons (i.e., because our model realistically reflects the underlying system). This problem has been addressed before, e.g., via model evaluation upon multiple system signatures such as spring discharge and solute concentrations. However, because the conduit network and the geological structure of karst systems is usually largely unknown, lumped models are often employed, which simplify the karst system and simulate the processes in a spatially aggregated manner. Due to those simplifications, assessing model (structural and process related) realism beyond a model’s ability to simulate system output signals is often impossible. Yet, using models to better understand karst system functioning is vital for sustainable karst water resources management. This discrepancy necessitates the study of karst systems and corresponding (lumped or spatially aggregated) models on the basis of systems that are fully known. For this purpose, we develop a virtual laboratory for karst groundwater modelling. This virtual laboratory facilitates the generation and simulation of synthetic karst systems and system signatures such as spring discharge using state-of-the-art spatially distributed numerical modelling. The virtual laboratory makes the generation of synthetic systems more accessible to a wider hydro(geo)logical community, taking a step towards more realistic process representation in (lumped) karst models in the future. Building on our previous work, this combination allows for the stochastic generation of conduit networks using pyKasso [1] and their subsequent embedding and simulation with the spatially distributed discrete-continuum model MODFLOW CFPv2 [2] via the CFPy package for the Python programming language [3]. We demonstrate the capabilities of the virtual laboratory for a number of cases, show upcoming development goals, and discuss opportunities for future applications. [1] Miville, F., Renard, P., Fandel, C., & Filipponi, M. (2025). pyKasso: An open-source three-dimensional discrete karst network generator. Environmental Modelling & Software, 186, 106362.[2] Reimann, T., Giese, M., Geyer, T., Liedl, R., Maréchal, J. C., & Shoemaker, W. B. (2014). Representation of water abstraction from a karst conduit with numerical discrete-continuum models. Hydrology and Earth System Sciences, 18(1), 227-241.[3] Reimann, T., Rudolph, M. G., Grabow, L., & Noffz, T. (2023). CFPy—A python package for pre‐and postprocessing of the conduit flow process of MODFLOW. Groundwater, 61(6), 887-894.
- Open Access
- 29 December 2025
- Published by American Geophysical Union (AGU) in Water Resources Research
- Vol. 62 (1)
The Effect of Available Data on the Worth of Future Observations for Groundwater Modeling
Groundwater model parameters need to be inferred on the basis of limited observation data, resulting in prediction uncertainty. The reduction of this uncertainty via future complementary observations is of high importance for many problems and can be guided by Bayesian experimental design. We employ a novel combination of Bayesian inversion, accelerated via multilevel methods, and Bayesian experimental design for this purpose. For a synthetic aquifer, we analyze the effect of including or excluding environmental tracer observations besides groundwater heads in two scenarios. In both scenarios, we study the effect of available data on distributions of model predictions after Bayesian inversion and subsequent experimental design. We demonstrate that posterior samples from Bayesian inversion can be reused to perform experimental design without additional model evaluations. In both scenarios, uncertainties and biases of flux‐related and groundwater age‐related predictions are substantially reduced through experimental design. Compared to the scenario with groundwater heads alone, including environmental tracer data in the observation data set leads to less uncertainty and bias in model outputs after Bayesian inversion, greater reduction of uncertainty and bias through experimental design, and reduced overestimation of complementary observation worth. Including environmental tracer observations at the beginning of combined Bayesian inversion and experimental design leads to more reliable predictions and more effective future data acquisition.
- 29 December 2025
- Published by American Geophysical Union (AGU) in Water Resources Research
- Vol. 62 (1)
Open AccessThe Effect of Available Data on the Worth of Future Observations for Groundwater Modeling
Groundwater model parameters need to be inferred on the basis of limited observation data, resulting in prediction uncertainty. The reduction of this uncertainty via future complementary observations is of high importance for many problems and can be guided by Bayesian experimental design. We employ a novel combination of Bayesian inversion, accelerated via multilevel methods, and Bayesian experimental design for this purpose. For a synthetic aquifer, we analyze the effect of including or excluding environmental tracer observations besides groundwater heads in two scenarios. In both scenarios, we study the effect of available data on distributions of model predictions after Bayesian inversion and subsequent experimental design. We demonstrate that posterior samples from Bayesian inversion can be reused to perform experimental design without additional model evaluations. In both scenarios, uncertainties and biases of flux‐related and groundwater age‐related predictions are substantially reduced through experimental design. Compared to the scenario with groundwater heads alone, including environmental tracer data in the observation data set leads to less uncertainty and bias in model outputs after Bayesian inversion, greater reduction of uncertainty and bias through experimental design, and reduced overestimation of complementary observation worth. Including environmental tracer observations at the beginning of combined Bayesian inversion and experimental design leads to more reliable predictions and more effective future data acquisition.
- Open Access
- 25 August 2025
- Published by Wiley
The Effect of Available Data on the Worth of Future Observations for Groundwater Modelling
Models for groundwater flow and solute transport are employed for many problems in research and practice, to guide decisions, and to predict quantities of interest. Following initial model development, Bayesian statistical methods are often used to infer distributions of model parameters from observation data. Reducing the resulting uncertainty in model outputs is of high importance in many contexts and Bayesian experimental design can be used for this purpose. This approach is computationally expensive, which hampers practical applicability. To solve the computational challenges, we use a novel combination of Bayesian inversion, accelerated via multilevel methods, and Bayesian experimental design. For a synthetic aquifer, we then analyze the effect of environmental tracer observations (besides groundwater head observations) in the initial observation dataset on the distributions of various quantities of interest after Bayesian inversion and subsequent experimental design. Our results demonstrate that posterior samples from Bayesian inversion can be re-used to perform experimental design without additional model evaluations. In both scenarios, uncertainties with respect to flux-related and groundwater age-related quantities are substantially reduced through experimental design and complementary observations of tracer concentrations are associated with the highest individual utility. The presence of environmental tracer data in the initial observation dataset leads to reduced uncertainty and bias in model outputs after Bayesian inversion and reduced overestimation in complementary observation utility.
- 25 August 2025
- Published by Wiley
Open AccessThe Effect of Available Data on the Worth of Future Observations for Groundwater Modelling
Models for groundwater flow and solute transport are employed for many problems in research and practice, to guide decisions, and to predict quantities of interest. Following initial model development, Bayesian statistical methods are often used to infer distributions of model parameters from observation data. Reducing the resulting uncertainty in model outputs is of high importance in many contexts and Bayesian experimental design can be used for this purpose. This approach is computationally expensive, which hampers practical applicability. To solve the computational challenges, we use a novel combination of Bayesian inversion, accelerated via multilevel methods, and Bayesian experimental design. For a synthetic aquifer, we then analyze the effect of environmental tracer observations (besides groundwater head observations) in the initial observation dataset on the distributions of various quantities of interest after Bayesian inversion and subsequent experimental design. Our results demonstrate that posterior samples from Bayesian inversion can be re-used to perform experimental design without additional model evaluations. In both scenarios, uncertainties with respect to flux-related and groundwater age-related quantities are substantially reduced through experimental design and complementary observations of tracer concentrations are associated with the highest individual utility. The presence of environmental tracer data in the initial observation dataset leads to reduced uncertainty and bias in model outputs after Bayesian inversion and reduced overestimation in complementary observation utility.
- Open Access
- 18 March 2025
- Published by Copernicus GmbH
Where and What to Sample Next? Bayesian Data-Worth Analysis for Regional Groundwater Models Using Multilevel GLUE
Highly parameterized numerical models of groundwater flow and contaminant transport play a central role in water resources management. Quantifying and analysing uncertainties associated with such models is a key challenge for decision-making, especially under the impacts of climate change. Furthermore, an important question often being overlooked in groundwater model applications is where the next observation point should be located and which state variable should be observed in order to reduce (predictive) uncertainty. We utilize the recently introduced Multilevel Generalized Likelihood Uncertainty Estimation methodology (MLGLUE; DOI: 10.1029/2024WR037735) to perform Bayesian inversion, accelerated by exploiting different spatial model resolutions. For a given model we consider two scenarios; in one scenario we utilize all available state observations while we remove environmental tracer observations from the dataset in a second scenario. We analyse the intrinsic data-worth of environmental tracer observations with respect to simulation uncertainty, especially regarding the estimates of quantities of interest derived from model outputs. Besides simulated observation equivalents we let the computational model also return potential future observations during inversion. We then use measures from information theory to select potential future observations which will result in the most substantial reduction of uncertainty regarding quantities of interest in both scenarios. We apply the combined methodology to a synthetic example as well as a previously developed steady-state regional groundwater flow and transport model. Our results demonstrate that the worth of environmental tracer observations is substantial to reduce model output uncertainty and to increase model accuracy. We show that future environmental tracer observations are especially relevant to better constrain estimates of quantities of interest when sampled at informative locations. The approach promises to improve the capabilities of groundwater models used for decision support and water resources management.
- 18 March 2025
- Published by Copernicus GmbH
Open AccessWhere and What to Sample Next? Bayesian Data-Worth Analysis for Regional Groundwater Models Using Multilevel GLUE
Highly parameterized numerical models of groundwater flow and contaminant transport play a central role in water resources management. Quantifying and analysing uncertainties associated with such models is a key challenge for decision-making, especially under the impacts of climate change. Furthermore, an important question often being overlooked in groundwater model applications is where the next observation point should be located and which state variable should be observed in order to reduce (predictive) uncertainty. We utilize the recently introduced Multilevel Generalized Likelihood Uncertainty Estimation methodology (MLGLUE; DOI: 10.1029/2024WR037735) to perform Bayesian inversion, accelerated by exploiting different spatial model resolutions. For a given model we consider two scenarios; in one scenario we utilize all available state observations while we remove environmental tracer observations from the dataset in a second scenario. We analyse the intrinsic data-worth of environmental tracer observations with respect to simulation uncertainty, especially regarding the estimates of quantities of interest derived from model outputs. Besides simulated observation equivalents we let the computational model also return potential future observations during inversion. We then use measures from information theory to select potential future observations which will result in the most substantial reduction of uncertainty regarding quantities of interest in both scenarios. We apply the combined methodology to a synthetic example as well as a previously developed steady-state regional groundwater flow and transport model. Our results demonstrate that the worth of environmental tracer observations is substantial to reduce model output uncertainty and to increase model accuracy. We show that future environmental tracer observations are especially relevant to better constrain estimates of quantities of interest when sampled at informative locations. The approach promises to improve the capabilities of groundwater models used for decision support and water resources management.
- Open Access
- 20 January 2025
- Published by Copernicus GmbH
Accelerating Hydrological Model Inversion: A Multilevel Approach to GLUE
Inverse problems play a pivotal role in hydrological modelling, particularly for parameter estimation and system understanding, which are essential for managing water resources. The application of statistical inversion methodologies such as Generalized Likelihood Uncertainty Estimation (GLUE) is often obstructed, however, by high model computational cost given that Monte Carlo sampling strategies often return a very small fraction of behavioural model runs. There is a need, however, to balance this aspect with the demand for broadly sampling the parameter space. Especially relevant for spatially distributed or (partial) differential equation based models, this aspect calls for computationally efficient methods of statistical inference that approximate the “true” posterior parameter distribution well. Our study introduces multilevel GLUE (MLGLUE), which effectively mitigates these computational challenges by exploiting a hierarchy of models with different computational grid resolutions (i.e., spatial or temporal discretisation), inspired by multilevel Monte Carlo strategies. Starting with low-resolution models, MLGLUE only passes parameter samples to higher-resolution models for evaluation if associated with a high likelihood, which poses a large potential for substantial computational savings. We demonstrate the applicability of the approach using a groundwater flow model with a hierarchy of different spatial resolutions. With MLGLUE, the computation time of parameter inference could be reduced by more than 60% compared to GLUE, while the resulting posterior distributions are virtually identical. Correspondingly, the uncertainty estimates of MLGLUE and GLUE are also very similar. Considering the simplicity of the implementation as well as its efficiency, MLGLUE promises to be an alternative for statistical inversion of computationally costly hydrological models.
- 20 January 2025
- Published by Copernicus GmbH
Open AccessAccelerating Hydrological Model Inversion: A Multilevel Approach to GLUE
Inverse problems play a pivotal role in hydrological modelling, particularly for parameter estimation and system understanding, which are essential for managing water resources. The application of statistical inversion methodologies such as Generalized Likelihood Uncertainty Estimation (GLUE) is often obstructed, however, by high model computational cost given that Monte Carlo sampling strategies often return a very small fraction of behavioural model runs. There is a need, however, to balance this aspect with the demand for broadly sampling the parameter space. Especially relevant for spatially distributed or (partial) differential equation based models, this aspect calls for computationally efficient methods of statistical inference that approximate the “true” posterior parameter distribution well. Our study introduces multilevel GLUE (MLGLUE), which effectively mitigates these computational challenges by exploiting a hierarchy of models with different computational grid resolutions (i.e., spatial or temporal discretisation), inspired by multilevel Monte Carlo strategies. Starting with low-resolution models, MLGLUE only passes parameter samples to higher-resolution models for evaluation if associated with a high likelihood, which poses a large potential for substantial computational savings. We demonstrate the applicability of the approach using a groundwater flow model with a hierarchy of different spatial resolutions. With MLGLUE, the computation time of parameter inference could be reduced by more than 60% compared to GLUE, while the resulting posterior distributions are virtually identical. Correspondingly, the uncertainty estimates of MLGLUE and GLUE are also very similar. Considering the simplicity of the implementation as well as its efficiency, MLGLUE promises to be an alternative for statistical inversion of computationally costly hydrological models.
- Open Access
- 17 October 2024
- Published by American Geophysical Union (AGU) in Water Resources Research
- Vol. 60 (10)
Extending GLUE With Multilevel Methods to Accelerate Statistical Inversion of Hydrological Models
Inverse problems aim at determining model parameters that produce observed data to subsequently understand, predict or manage hydrological or other environmental systems. While statistical inversion is especially popular, its sampling‐based nature often inhibits its application to computationally costly models, which has compromised the use of the Generalized Likelihood Uncertainty Estimation (GLUE) methodology, for example, for spatially distributed (partial) differential equation based models. In this study we introduce multilevel GLUE (MLGLUE), which alleviates the computational burden of statistical inversion by utilizing a hierarchy of model resolutions. Inspired by multilevel Monte Carlo, most parameter samples are evaluated on lower levels with computationally cheap low‐resolution models and only samples associated with a likelihood above a certain threshold are subsequently passed to higher levels with costly high‐resolution models for evaluation. Inferences are made at the level of the highest‐resolution model but substantial computational savings are achieved by discarding samples with low likelihood already on levels with low resolution and low computational cost. Two example inverse problems, using a rainfall‐runoff model and groundwater flow model, demonstrate the substantially increased computational efficiency of MLGLUE compared to GLUE as well as the similarity of inversion results. Findings are furthermore compared to inversion results from Markov‐chain Monte Carlo (MCMC) and multilevel delayed acceptance MCMC, a corresponding multilevel variant, to compare the effects of the multilevel extension. All examples demonstrate the wide‐range suitability of the approach and include guidelines for practical applications.
- 17 October 2024
- Published by American Geophysical Union (AGU) in Water Resources Research
- Vol. 60 (10)
Open AccessExtending GLUE With Multilevel Methods to Accelerate Statistical Inversion of Hydrological Models
Inverse problems aim at determining model parameters that produce observed data to subsequently understand, predict or manage hydrological or other environmental systems. While statistical inversion is especially popular, its sampling‐based nature often inhibits its application to computationally costly models, which has compromised the use of the Generalized Likelihood Uncertainty Estimation (GLUE) methodology, for example, for spatially distributed (partial) differential equation based models. In this study we introduce multilevel GLUE (MLGLUE), which alleviates the computational burden of statistical inversion by utilizing a hierarchy of model resolutions. Inspired by multilevel Monte Carlo, most parameter samples are evaluated on lower levels with computationally cheap low‐resolution models and only samples associated with a likelihood above a certain threshold are subsequently passed to higher levels with costly high‐resolution models for evaluation. Inferences are made at the level of the highest‐resolution model but substantial computational savings are achieved by discarding samples with low likelihood already on levels with low resolution and low computational cost. Two example inverse problems, using a rainfall‐runoff model and groundwater flow model, demonstrate the substantially increased computational efficiency of MLGLUE compared to GLUE as well as the similarity of inversion results. Findings are furthermore compared to inversion results from Markov‐chain Monte Carlo (MCMC) and multilevel delayed acceptance MCMC, a corresponding multilevel variant, to compare the effects of the multilevel extension. All examples demonstrate the wide‐range suitability of the approach and include guidelines for practical applications.
- Open Access
- 7 October 2024
- Published by Springer Nature in Grundwasser
- Vol. 29 (4) , 197-198
Zukunft gestalten: Ein junges Netzwerk für Hydrogeologie in Deutschland
- 7 October 2024
- Published by Springer Nature in Grundwasser
- Vol. 29 (4) , 197-198
Open AccessZukunft gestalten: Ein junges Netzwerk für Hydrogeologie in Deutschland
- Open Access
- 19 April 2024
- Published by Wiley
Extending GLUE with Multilevel Methods to Accelerate Statistical Inversion of Hydrological Models
Inverse problems are ubiquitous in hydrological modelling for parameter estimation, system understanding, sustainable water resources management, and the operation of digital twins. While statistical inversion is especially popular, its sampling-based nature often inhibits its application to computationally costly models, which has compromised the use of the Generalized Likelihood Uncertainty Estimation (GLUE) methodology, e.g., for spatially distributed (partial) differential equation based models. In this study we introduce multilevel GLUE (MLGLUE), which alleviates the computational burden of statistical inversion by utilizing a hierarchy of model resolutions. Inspired by multilevel Monte Carlo, most parameter samples are evaluated on lower levels with computationally cheap low-resolution models and only samples associated with a likelihood above a certain threshold are subsequently passed to higher levels with costly high-resolution models for evaluation. Inferences are made at the level of the highest-resolution model but substantial computational savings are achieved by discarding samples with low likelihood already on levels with low resolution and low computational cost. Two example inverse problems, using a rainfall-runoff model and groundwater flow model, demonstrate the substantially increased computational efficiency of MLGLUE compared to GLUE as well as the similarity of inversion results. Findings are furthermore compared to inversion results from Markov-chain Monte Carlo (MCMC) and multilevel delayed acceptance MCMC, a corresponding multilevel variant, to compare the effects of the multilevel extension. All examples demonstrate the wide-range suitability of the approach and include guidelines for practical applications.
- 19 April 2024
- Published by Wiley
Open AccessExtending GLUE with Multilevel Methods to Accelerate Statistical Inversion of Hydrological Models
Inverse problems are ubiquitous in hydrological modelling for parameter estimation, system understanding, sustainable water resources management, and the operation of digital twins. While statistical inversion is especially popular, its sampling-based nature often inhibits its application to computationally costly models, which has compromised the use of the Generalized Likelihood Uncertainty Estimation (GLUE) methodology, e.g., for spatially distributed (partial) differential equation based models. In this study we introduce multilevel GLUE (MLGLUE), which alleviates the computational burden of statistical inversion by utilizing a hierarchy of model resolutions. Inspired by multilevel Monte Carlo, most parameter samples are evaluated on lower levels with computationally cheap low-resolution models and only samples associated with a likelihood above a certain threshold are subsequently passed to higher levels with costly high-resolution models for evaluation. Inferences are made at the level of the highest-resolution model but substantial computational savings are achieved by discarding samples with low likelihood already on levels with low resolution and low computational cost. Two example inverse problems, using a rainfall-runoff model and groundwater flow model, demonstrate the substantially increased computational efficiency of MLGLUE compared to GLUE as well as the similarity of inversion results. Findings are furthermore compared to inversion results from Markov-chain Monte Carlo (MCMC) and multilevel delayed acceptance MCMC, a corresponding multilevel variant, to compare the effects of the multilevel extension. All examples demonstrate the wide-range suitability of the approach and include guidelines for practical applications.
- Open Access
- 26 October 2023
- Published by Wiley
Extending GLUE with Multilevel Methods to Accelerate Statistical Inversion of Hydrological Models
Inverse problems are ubiquitous in hydrological modelling for parameter estimation, system understanding, sustainable water resources management, and the operation of digital twins. While statistical inversion is especially popular, its sampling-based nature often inhibits the inversion of computationally costly models, which has compromised the use of the Generalized Likelihood Uncertainty Estimation (GLUE) methodology, e.g., for spatially distributed (partial) differential equation based models. In this study we introduce multilevel GLUE (MLGLUE), which alleviates the computational burden of statistical inversion by utilizing a hierarchy of model resolutions. Inspired by multilevel Monte Carlo, most parameter samples are evaluated on lower levels with computationally cheap low-resolution models and only samples associated with a likelihood above a certain threshold are subsequently passed to higher levels with costly high-resolution models for evaluation. Inferences are made at the level of the highest-resolution model but substantial computational savings are achieved by discarding samples with low likelihood already on levels with low resolution and low computational cost. Two test problems demonstrate the similarity of inferred parameter posteriors and uncertainty estimates of MLGLUE and GLUE as well as increased computational efficiency. Findings are furthermore compared to inversion results from Markov-chain Monte Carlo (MCMC) and from multilevel delayed acceptance MCMC. The computation time of inversion of a groundwater flow model was decreases by ≈45% and ≈57% when using MLGLUE instead of conventional formulations of GLUE and MCMC, respectively.
- 26 October 2023
- Published by Wiley
Open AccessExtending GLUE with Multilevel Methods to Accelerate Statistical Inversion of Hydrological Models
Inverse problems are ubiquitous in hydrological modelling for parameter estimation, system understanding, sustainable water resources management, and the operation of digital twins. While statistical inversion is especially popular, its sampling-based nature often inhibits the inversion of computationally costly models, which has compromised the use of the Generalized Likelihood Uncertainty Estimation (GLUE) methodology, e.g., for spatially distributed (partial) differential equation based models. In this study we introduce multilevel GLUE (MLGLUE), which alleviates the computational burden of statistical inversion by utilizing a hierarchy of model resolutions. Inspired by multilevel Monte Carlo, most parameter samples are evaluated on lower levels with computationally cheap low-resolution models and only samples associated with a likelihood above a certain threshold are subsequently passed to higher levels with costly high-resolution models for evaluation. Inferences are made at the level of the highest-resolution model but substantial computational savings are achieved by discarding samples with low likelihood already on levels with low resolution and low computational cost. Two test problems demonstrate the similarity of inferred parameter posteriors and uncertainty estimates of MLGLUE and GLUE as well as increased computational efficiency. Findings are furthermore compared to inversion results from Markov-chain Monte Carlo (MCMC) and from multilevel delayed acceptance MCMC. The computation time of inversion of a groundwater flow model was decreases by ≈45% and ≈57% when using MLGLUE instead of conventional formulations of GLUE and MCMC, respectively.