Significant advances in computing power and numerical model development provide the capability to simulate complex physical processes of surface and subsurface flows. As a result, important modelling tasks including design optimization, uncertainty/sensitivity analysis or model calibration can be implemented via high-fidelity computer models of the system under study. As we strive for high-quality insights by employing global optimization methods for water resources management or Monte-Carlo based sensitivity/uncertainty analysis, there are two common obstacles that pose computational difficulties: computational cost of the high-fidelity numerical simulations and curse of dimensionality.
Over the last two decades, the use of surrogate models in water resources has paved the way for enabling computationally demanding tasks to take place without sacrificing high-fidelity analysis. However, time-intensive numerical simulations render the application of single-fidelity surrogate models a challenging task. This is particularly true when the increased dimensionality of a water resources optimization problem or a global sensitivity analysis might require a large set of high-fidelity training data to ensure accuracy with the single-fidelity surrogate model. An alternative and promising approach which has received less attention so far in water resources modelling is the development of multifidelity surrogate models. Various subsurface and surface models are usually available to simulate a process at different levels of fidelity. For example, lower-fidelity models of the system may occur by omitting parts of the physics, reducing model dimensionality, applying less strict spatial/temporal discretization or convergence relaxation and other possible ways to provide faster calculations for the quantities of interest. The general aim is to exploit the inherent knowledge of the physical system that lies within models of lower fidelity and combine this information with only limited data from the high-fidelity computationally intensive model.
The present Research Topic invites submissions of studies that employ multifidelity modelling to optimize and analyze surface and subsurface flow systems. Topics of particular interest include, but are not limited to:
1) Surrogate-based multifidelity optimization for problems related to water resources management;
2) Sensitivity analysis and uncertainty quantification for computationally expensive models (hydrodynamic modelling, density-driven flow models, sediment transport, flow and heat transport models, reactive transport, computational fluid dynamics models);
3) Multifidelity methods for multi-dimensional input-output problems in water resources;
4) Comparisons between high-fidelity and low-fidelity models and implications for computationally expensive tasks; and,
5) Multifidelity model analysis for climate change impacts on water resources.
It is highlighted that mutlifidelity approach is considered explicitly in the present topic as the coupling of physical or empirical or conceptual models of different levels of fidelity. Therefore, the Research Topic focuses more on that scientific area than the single-fidelity data-driven models.
Significant advances in computing power and numerical model development provide the capability to simulate complex physical processes of surface and subsurface flows. As a result, important modelling tasks including design optimization, uncertainty/sensitivity analysis or model calibration can be implemented via high-fidelity computer models of the system under study. As we strive for high-quality insights by employing global optimization methods for water resources management or Monte-Carlo based sensitivity/uncertainty analysis, there are two common obstacles that pose computational difficulties: computational cost of the high-fidelity numerical simulations and curse of dimensionality.
Over the last two decades, the use of surrogate models in water resources has paved the way for enabling computationally demanding tasks to take place without sacrificing high-fidelity analysis. However, time-intensive numerical simulations render the application of single-fidelity surrogate models a challenging task. This is particularly true when the increased dimensionality of a water resources optimization problem or a global sensitivity analysis might require a large set of high-fidelity training data to ensure accuracy with the single-fidelity surrogate model. An alternative and promising approach which has received less attention so far in water resources modelling is the development of multifidelity surrogate models. Various subsurface and surface models are usually available to simulate a process at different levels of fidelity. For example, lower-fidelity models of the system may occur by omitting parts of the physics, reducing model dimensionality, applying less strict spatial/temporal discretization or convergence relaxation and other possible ways to provide faster calculations for the quantities of interest. The general aim is to exploit the inherent knowledge of the physical system that lies within models of lower fidelity and combine this information with only limited data from the high-fidelity computationally intensive model.
The present Research Topic invites submissions of studies that employ multifidelity modelling to optimize and analyze surface and subsurface flow systems. Topics of particular interest include, but are not limited to:
1) Surrogate-based multifidelity optimization for problems related to water resources management;
2) Sensitivity analysis and uncertainty quantification for computationally expensive models (hydrodynamic modelling, density-driven flow models, sediment transport, flow and heat transport models, reactive transport, computational fluid dynamics models);
3) Multifidelity methods for multi-dimensional input-output problems in water resources;
4) Comparisons between high-fidelity and low-fidelity models and implications for computationally expensive tasks; and,
5) Multifidelity model analysis for climate change impacts on water resources.
It is highlighted that mutlifidelity approach is considered explicitly in the present topic as the coupling of physical or empirical or conceptual models of different levels of fidelity. Therefore, the Research Topic focuses more on that scientific area than the single-fidelity data-driven models.