CBET-EPSRC: Efficient Surrogate Modeling for Sustainable Management of Complex Seawater Intrusion-Impacted Aquifers
Lead Research Organisation:
University of Sheffield
Department Name: Civil and Structural Engineering
Abstract
The overarching goal of the proposed research is the sustainable management of water resources in coastal regions with diverse geological, hydro-technical and governance settings. Pressures on water resources in coastal regions are already great and are expected to intensify due to increasing populations, standards of living and impacts from climate change and sea level rise (SLR). We will focus on coastal areas where aquifer over-drafting has caused seawater intrusion (SWI), thus deteriorating groundwater quality, and where SLR is expected to further reduce availability of fresh groundwater. Solutions to these problems will involve combinations of more efficient pumping schemes, demand reduction, and technological interventions such as desalination. However, determining optimal solutions for these problems poses extreme computational demands. This project will greatly advance the development and application of simulation-optimization (SO) by developing computationally efficient, robust, and accurate surrogate models for coastal groundwater systems.
The limited literature on SO and surrogate modeling in SWI problems has focused on simplified hydrogeological settings and mathematical representations of management strategies. However, realistic SWI problems involve hydrogeological complexities, including discrete lithological facies, faults and fractures, saltwater-freshwater mixing zone dynamics, and surface-water groundwater interactions, as well as nonlinear objective functions and continuous and discrete decision variables to represent a wide range of engineering components. We hypothesize that these hydrogeologic and management features determine the building of accurate and efficient surrogates; and accurate surrogate SO models for SWI problems can be at least an order of magnitude faster than full-scale models. The reduced computational cost allows to investigate a broader range of SLR and climate change impacts and a wider range of management responses to these impacts. The innovative aspects of this research are: (a) development of a systematic approach for building robust surrogates by testing against full-scale SO models on simple to complex problems; (b) assessment of tradeoffs between surrogate model accuracy and computational efficiency across a range of hydrogeologic and management settings; (c) identification of robust management schemes for managing coastal groundwater resources in three "end-member" case study aquifers; and (d) collaboration with water management agencies to develop useful scenarios, optimization frameworks, and model output. The three test aquifers (Santa Barbara, California; Biscayne, Florida; and San Salvador Island, The Bahamas) have diverse hydrogeologic and management characteristics and well-calibrated groundwater flow models. The project objectives are: (a) develop SO-SWI-SLR test problems to provide robust evaluation of model surrogates; (b) formulate management objectives and constraints based on management of the test case aquifers, and identify scenarios relevant to the test cases; (c) program, train, and evaluate the performance of "data-driven" and "model-driven" surrogates to identify optimal management schemes for the test case aquifers, a range of SLR rates, climatology, and groundwater demand scenarios. This work will build on our US-UK group's complementary experience simulating SLR and climate impacts on SWI and in developing SO models for other groundwater problems.
The limited literature on SO and surrogate modeling in SWI problems has focused on simplified hydrogeological settings and mathematical representations of management strategies. However, realistic SWI problems involve hydrogeological complexities, including discrete lithological facies, faults and fractures, saltwater-freshwater mixing zone dynamics, and surface-water groundwater interactions, as well as nonlinear objective functions and continuous and discrete decision variables to represent a wide range of engineering components. We hypothesize that these hydrogeologic and management features determine the building of accurate and efficient surrogates; and accurate surrogate SO models for SWI problems can be at least an order of magnitude faster than full-scale models. The reduced computational cost allows to investigate a broader range of SLR and climate change impacts and a wider range of management responses to these impacts. The innovative aspects of this research are: (a) development of a systematic approach for building robust surrogates by testing against full-scale SO models on simple to complex problems; (b) assessment of tradeoffs between surrogate model accuracy and computational efficiency across a range of hydrogeologic and management settings; (c) identification of robust management schemes for managing coastal groundwater resources in three "end-member" case study aquifers; and (d) collaboration with water management agencies to develop useful scenarios, optimization frameworks, and model output. The three test aquifers (Santa Barbara, California; Biscayne, Florida; and San Salvador Island, The Bahamas) have diverse hydrogeologic and management characteristics and well-calibrated groundwater flow models. The project objectives are: (a) develop SO-SWI-SLR test problems to provide robust evaluation of model surrogates; (b) formulate management objectives and constraints based on management of the test case aquifers, and identify scenarios relevant to the test cases; (c) program, train, and evaluate the performance of "data-driven" and "model-driven" surrogates to identify optimal management schemes for the test case aquifers, a range of SLR rates, climatology, and groundwater demand scenarios. This work will build on our US-UK group's complementary experience simulating SLR and climate impacts on SWI and in developing SO models for other groundwater problems.
Planned Impact
This work will contribute to solving the critical sustainability problem of water management in stressed coastal regions subject to uncertain SLR scenarios; and advance surrogate modeling for ultra-dimensional and highly nonlinear systems that can be applied to a wide range of engineering problems; and collaborative interactions with decision-makers and stakeholders leading to improved science communication and science-driven policy. Project results will be disseminated by publications in peer-reviewed engineering journals, presentations at scientific conferences and meetings involving coastal water resources policymakers, administrators and stakeholders. The project collaborators are internationally known for their research contributions to groundwater modelling and monitoring, optimization applied to surface and groundwater management. This will strengthen both UK and US's leaderships in the areas of simulation-optimization applications to water management.
People |
ORCID iD |
Domenico Bau (Principal Investigator) |
Publications
BaĆ¹ D
(2022)
Land subsidence surrogate models for normally consolidated sedimentary basins
in Geomechanics for Energy and the Environment
Yu W
(2023)
Investigating the Impact of Seawater Intrusion on the Operation Cost of Groundwater Supply in Island Aquifers
in Water Resources Research
Yu W.
Investigating the impact of seawater intrusion on the operation cost of groundwater supply in island aquifers
in Water Resources Research
Geranmehr M
(2023)
Variable Density Groundwater Flow Simulation by Reduced Order Models
Title | Efficient Approaches for Offline Training of Gaussian Process Models in Coastal Groundwater Management. Presented at 1st IACRR International Conference on Coastal Reservoirs and Sustainable Water Management, Nanjing, China |
Description | We focuson the use of mathematical optimization techniques to manage freshwater demand and control saltwater intrusion (SWI) in coastal aquifers. Traditional methods involve linking variable density flow models with optimization algorithms, but these can be computationally expensive. We propose a more efficient approach using data-driven surrogates within the optimization loop, specifically focusing on a novel iterative search algorithm to select training points for surrogate model development.The study applies this approach to a 2D model of the San Salvador Island aquifer in the Bahamas, considering trade-offs between groundwater supply cost, the produced groundwater rate, and SWI. The optimization problem aims to minimize operation cost and maximize fresh groundwater supply while adhering to constraints on aquifer drawdown and salt mass increase caused by pumping.Gauss Process (GP) regression is employed to train surrogates for various parameters in the decision variable space, which includes pumping depth, distance from shoreline, and groundwater abstraction rate. Three GP model training strategies are proposed: iterative methods based on maximum distance, maximum gradient, and a score function. Results and discussion focus on the efficiency of the proposed GP training strategies, comparing their performance in terms of the average probability of Pareto-optimality for pumping schemes. We conclude that the iterative strategy using the score function outperforms other methods in terms of computational costs and Pareto-optimal solutions. |
Type Of Art | Film/Video/Animation |
Year Produced | 2024 |
URL | https://orda.shef.ac.uk/articles/presentation/Efficient_Approaches_for_Offline_Training_of_Gaussian_... |
Title | Efficient Approaches for Offline Training of Gaussian Process Models in Coastal Groundwater Management. Presented at 1st IACRR International Conference on Coastal Reservoirs and Sustainable Water Management, Nanjing, China |
Description | We focuson the use of mathematical optimization techniques to manage freshwater demand and control saltwater intrusion (SWI) in coastal aquifers. Traditional methods involve linking variable density flow models with optimization algorithms, but these can be computationally expensive. We propose a more efficient approach using data-driven surrogates within the optimization loop, specifically focusing on a novel iterative search algorithm to select training points for surrogate model development.The study applies this approach to a 2D model of the San Salvador Island aquifer in the Bahamas, considering trade-offs between groundwater supply cost, the produced groundwater rate, and SWI. The optimization problem aims to minimize operation cost and maximize fresh groundwater supply while adhering to constraints on aquifer drawdown and salt mass increase caused by pumping.Gauss Process (GP) regression is employed to train surrogates for various parameters in the decision variable space, which includes pumping depth, distance from shoreline, and groundwater abstraction rate. Three GP model training strategies are proposed: iterative methods based on maximum distance, maximum gradient, and a score function. Results and discussion focus on the efficiency of the proposed GP training strategies, comparing their performance in terms of the average probability of Pareto-optimality for pumping schemes. We conclude that the iterative strategy using the score function outperforms other methods in terms of computational costs and Pareto-optimal solutions. |
Type Of Art | Film/Video/Animation |
Year Produced | 2024 |
URL | https://orda.shef.ac.uk/articles/presentation/Efficient_Approaches_for_Offline_Training_of_Gaussian_... |
Description | Managing groundwater in island aquifer settings, exemplified by San Salvador (Bahamas), necessitates a delicate equilibrium between the costs of groundwater supply (pumping and treatment) and indicators of seawater intrusion (SWI) intensity. The definition of these indicators significantly shapes pumping strategies. When assessing SWI based on increased salt mass from groundwater pumping, optimal strategies involve pumping midway between the island center and shoreline at shallow depths. Conversely, minimizing SWI by restricting freshwater depletion, determined by aquifer drawdown, demands pumping towards the island center and as deep as possible. Both approaches escalate supply costs, yet prioritizing freshwater depletion prevention emerges as a conservative choice, resulting in higher supply costs. Simulation-optimization plays a pivotal role in coastal groundwater resource management, demanding computationally intensive variable density groundwater flow models. To enhance efficiency, we developed methods utilizing Gaussian Process techniques to create data-driven surrogates. Surrogate training can occur offline, preceding optimization, or online, during optimization. Results indicate that while the online approach generally provides better efficiency, yielding reliable optimal solutions at a lower computational cost, the offline method offers greater flexibility. It allows the exploration and comparison of optimal strategies under diverse constraints on SWI indicators, a crucial feature for decision-making by water administrators and stakeholders. An alternative surrogate modeling approach for variable density flow involves reduced order models using "projection" methods. These models express the solution of full-scale models as a linear combination of orthogonal basis functions with time-dependent coefficients, significantly reducing the size of expensive full-scale models at the expense of a reasonably small accuracy loss. This reduction is advantageous for repeated simulation runs in simulation-optimization (SO) frameworks and sensitivity analyses. Two reduced order models have been developed as part of this effort. One is based on a finite-difference formulation of variable density groundwater flow equations, accommodating basis functions from popular models like USGS's SEAWAT and MODFLOW 6. This model is currently employed for sensitivity analyses, evaluating factors such as the impact of aquifer heterogeneity on SWI vulnerability indicators. Another reduced order model, a variant of SEAWAT, extends the approach to cases where a variable density model is developed, calibrated, and validated, such as the Santa Barbara (California) and Biscayne (Florida) aquifers. |
Exploitation Route | We have planned on organizing a two-day public event involving primarily water administrators of the Biscayne and Santa Barbara aquifers, which are the sites on which our project has focused, along with other USGS scientists that have provided in-kind support to our work. The event should take place in Florida tentatively in May 2024 and will be an ideal way of maximizing the broader impacts of the project and transfer the findings of our research to the interested stakeholders and communities. |
Sectors | Digital/Communication/Information Technologies (including Software) Environment Other |
Title | Investigating the impact of seawater intrusion on the operation cost of groundwater supply in island aquifers |
Description | This is the simulation code for modelling seawater intrusion in a 2D cross-section island aquifer using SEAWAT. |
Type Of Material | Database/Collection of data |
Year Produced | 2023 |
Provided To Others? | Yes |
URL | https://figshare.com/articles/dataset/Investigating_the_impact_of_seawater_intrusion_on_the_operatio... |
Description | University of Texas at El Paso |
Organisation | University of Texas, El Paso |
Country | United States |
Sector | Academic/University |
PI Contribution | This is a collaborative project between the University of Texas at El Paso, USA, and the University of Sheffield, UK. The US PI is Dr Alex S Mayer, whereas the UK PI is Dr Domenico Bau. |
Collaborator Contribution | The overarching goal of this research is the sustainable management of water resources in coastal regions with diverse geological, hydro-technical and governance settings. Pressures on water resources in coastal regions are already great and are expected to intensify due to increasing populations, standards of living and impacts from climate change and sea level rise (SLR). We will focus on coastal areas where aquifer over-drafting has caused seawater intrusion (SWI), thus deteriorating groundwater quality, and where SLR is expected to further reduce availability of fresh groundwater. Solutions to these problems will involve combinations of more efficient pumping schemes, demand reduction, and technological interventions such as desalination. However, determining optimal solutions for these problems poses extreme computational demands. This project will greatly advance the development and application of simulation-optimization (SO) by developing computationally efficient, robust, and accurate surrogate models for coastal groundwater systems. |
Impact | collaborative research |
Start Year | 2020 |
Description | Conference Presentation |
Form Of Engagement Activity | A talk or presentation |
Part Of Official Scheme? | No |
Geographic Reach | International |
Primary Audience | Professional Practitioners |
Results and Impact | title: 'Comparison of off-line and on-line trained Gaussian process models for island groundwater management' authors: Weijiang Yu, Domenico Bau, Mayer Alex, Yipeng Zhang, Lauren Mancewicz, Mohammadali Geranmehr session selected: 'Advancing the State-of-the-Science of Water Resources Modeling - Community Development at the Intersection of Domain, Data, and Computer Sciences' |
Year(s) Of Engagement Activity | 2022 |
URL | https://www.agu.org/FIHM |
Description | Univ of Sheffield - Dept of Civil and Environmental Engineering - PGR Conference 2021 |
Form Of Engagement Activity | A talk or presentation |
Part Of Official Scheme? | No |
Geographic Reach | Local |
Primary Audience | Postgraduate students |
Results and Impact | Poster Title: Studying the impact of seawater intrusion (SWI) on the cost of groundwater supply in island aquifers Authors: Weijiang Yu, Domenico Baù, Alex S. Mayer, Yipeng Zhang, Lauren Mancewicz |
Year(s) Of Engagement Activity | 2021 |