A review of conjunctive GW-SW management by simulation–optimization tools

The conjunctive use of groundwater and surface water (GW-SW) resources has grown worldwide. Optimal conjunctive water use can be planned by coupling hydrologic models for the simulation of water systems with optimization techniques for improving management strategies. The coupling of simulation and optimization methods constitutes an effective approach to determine sustainable management strategies for the conjunctive use of these water resources; yet, there are challenges that must be addressed. This paper reviews (1) hydrologic models applied for the simulation of GW-SW interaction in the water resources systems, (2) conventional optimization methods, and (3) published works on optimized conjunctive GW-SW use by coupling simulation and optimization methods. This paper evaluates the pros and cons of GW-SW simulation tools and their applications, thus providing criteria for selecting simulation–optimization methods for GW-SW management. In addition, an assessment of GW-SW simulation–optimization tools applied in various studies over the world creates valuable knowledge for selecting suitable simulation–optimization tools in similar case studies for sustainable water resource management under multiple scenarios.


INTRODUCTION
The conjunctive use of groundwater and surface water (GW-SW) resources is an effective approach to manage water resources systems ( Alluvial aquifers are a significant source of water supply in numerous hydrogeological regions extending along rivers (Rosenshein ; Larkin & Sharp ; Barlow et al. ). Alluvial aquifers occur adjacent to rivers and in buried channels. These aquifers have variable thickness and contain layers of sand and gravel deposited by networks of streams over geologic time (Heath ). Alluvial groundwater has a strong hydraulic connection with natural surface streams (rivers), surface runoff, lakes, and reservoirs. Therefore, groundwater pumping in these aquifers leads to streamflow depletion and reduction of lake and reservoir storage. Correct planning of groundwater withdrawal in alluvial aquifers requires knowledge of GW-SW interactions (Sophocleous ; Rossetto et al. ). Spatial and temporal interactions of GW-SW resources can be assessed by the integrated, dynamic coupling of GW-SW resources employing mathematical models (Ghordoyee Milan et al.

).
Optimal operational rules for GW-SW management have a high priority in alluvial basins, which commonly feature productive aquifers (Brookfield & Gnau ). System analysis techniques for water resource planning and management have achieved remarkable advances over the last few decades (Ahmadianfar et al. a). System analysis techniques may be grouped into simulation, optimization, and simulation-optimization methods (Loucks et al. ).
Coupling planning/operation models with hydrologic models (e.g., MODSIM-MODFLOW) are currently introduced as a powerful tool in integrated water resource management (Morway et al. ). Various optimization tools have been widely applied to tackle real-world problems in water resource science (Ahmadianfar et al. b), in which the result of some novel optimizers such as the gradient-based method has been more promising than evolutionary optimization techniques such as the genetic algorithm (GA) and the differential evolution (DE) algorithm (Ahmadianfar et al. ). Thorough recognition of the GW-SW models and available coupling optimization tools allow the experts to choose the best combination to either solve or predict the problems in watersheds. Simulation and optimization techniques are combined to find possible alternatives for the conjunctive use of GW-SW resources (Karamouz et al. ).  (Soltanjalili et al. ), this evaluates simulation and optimization methods and their many applications. It is noteworthy that a few reviews regarding GW-SW models have been published, which focused on the use of partial differential equations to model the components of coupled systems (Paniconi & Putti ), discussed the physical and numerical alternatives of several models (Furman ), and covered the conceptual approaches to GW-SW interaction (Sophocleous ). However, the combination of GW-SW simulation models and related optimization techniques have not been comprehensively evaluated in review articles. Therefore, this work attempts to fill the following gaps: • determine operational rules for modeling the conjunctive use of GW-SW resources, • present an outlook of the current challenges and trends in the conjunctive use of GW-SW resources, and • overview optimization methods and their applications in coupled simulation-optimization of GW-SW resources management.
This work reviews conjunctive GW-SW models coupled with optimizers; yet the large volume of existing publications does not allow a complete coverage of all published works, but, rather, a selective review of the main types of approaches following in the subject matter of GW-SW simulation-optimization.

HYDROLOGIC SIMULATION MODELS
A systematic understanding of GW-SW interactions is necessary to manage scarce water resources in alluvial aquifers in arid and semi-arid regions (Wu et al. ).
Understanding such interactions can be achieved through field observations, which may be severely limited in many cases. Simulation modeling has arisen in the last few decades as an efficient method to complement field investigations for understanding hydrologic processes.
Simulation models are physically based hydrological models that simulate the terrestrial portion of the hydrologic cycle. The simulation output can determine discharge at the basin's outlets and along stream networks, hydrologic responses to land use and land cover changes, landatmosphere interactions, water quality changes, aquifer response to groundwater extraction, and many other phenomena (Markstrom et al. ). These models calculate surface water and saturated/unsaturated subsurface water flow and other biogeochemical processes. A physically based hydrological model consists of mathematical descriptions of the surface process, subsurface process, external boundary conditions, internal boundary conditions, and initial conditions of a hydrologic system (Furman ). There are multiple categorizations of hydrologic models according to characteristics such as the number of hydrologic processes that are simulated, the types of mathematical conceptualization of hydrologic process, and the type of spatial representations of hydrologic elements and components (distributed, lumped, conceptual, etc.) (Barthel & Banzhaf ). This paper considers characteristics of hydrologic models related to their governing equations, the coupling of the governing equations, solution techniques, and numerical approaches. The governing equations for simulating the hydrological cycle are mostly based on mass and momentum conservation principles for surface and subsurface flow. The partial differential equations that are solved numerically express their complex nonlinear nature (Paniconi & Putti ).  Equations governing hydrologic processes are generally based on systems of partial differential equations. Hydrologic models employ numerical solutions to solve those systems of equations. These numerical approaches can be the finite element method, the finite difference method, the finite volume method, or combinations of these methods.
There are a large number of hydrologic models, such as    The MIKE SHE's code is found in the user's guide.

Penn State Integrated Hydrologic Model
PIHM is described as a fully coupled hydrologic model with two multi-scale and multiprocessing features. PIHM uses a semi-discrete finite volume approach for coupling groundwater, land surface components, and surface water to make fully coupled ordinary differential equations from flow systems described by partial differential equations showed that GEOtop might be easily interfaced with climate and bounded-area meteorological models to simulate river basin hydrology.
The HydroGeoSphere model In addition, HGS has interfaces with GIS tools such as Arc-View and ArcInfo for enhanced spatial data analysis.
The PARFLOW model (1) temperature (2) precipitation, and (3)   Rather, coupling optimization methods with hydrological flow models seem to be necessary to achieve optimal management policies. The optimization algorithms adjust the model's parameters to improve an objective function, thus yielding optimal solutions to conjunctive management problems.

OPTIMIZATION METHODS
Optimization methods have been widely used in water resources systems to find optimal management strategies for various purposes such as controlling non-point agricultural contamination, determining optimal location of low impact The GA is the pioneering evolutionary optimization algorithm introduced by Holland in 1975. They reach near globally optimal solutions with acceptable errors in problems that some classical methods cannot solve, such as nonlinear, non-differentiable, discontinuous, mixed discrete/real variables, non-convex, deterministic, stochastic, single objective, and multi-objective formulations. Evolutionary and metaheuristic algorithms are less likely to converge to local optima than classical methods (gradientbased NLP methods). However, they require numerous evaluations of the objective function, which may introduce a heavy computational burden solving coupled physically based simulation models and evolutionary algorithms. In addition, most evolutionary algorithms require the specification of parameters that may require calibration.
The optimization with evolutionary and metaheuristics algorithms begins with a randomly generated population of management policies. The water resources system is simulated with hydrologic-economic models based on the current population of generated management policies. The performance of the water resources system is evaluated with single objective or multi-objective functions. The evolutionary algorithm then creates an improved population of management policies with which to simulate anew the performance of the water resources system. This iterative process continues until the performance of the water resources system cannot be improved any further. At that point, the simulation-optimization method has identified an optimal management policy and the corresponding response of the water resources system.

Surrogate-based modeling
This modeling approach converts complex functions with much simpler ones in an iterative model assessment pro-   Table 3.

Simulation
In addition to the works listed in Table 3  PO-GA, preference order genetic algorithm; LP, linear programming; KRG, kriging; and CRF, cubic radial basis functions.
optimize complex conjunctive use problems. The latter authors proposed a surrogate-based optimization for integrated surface water-groundwater modeling, which constitutes an integrated model for optimizing conjunctive river-aquifer management.

CONCLUSION
The water resources system literature indicates the importance of searching for reservoir operational rules and groundwater extraction plans to optimize conjunctive GW-SW resources management. Optimal management of GW-SW resources, particularly in areas dependent on groundwater resources, must rely on coupled simulationoptimization linking hydrologic models with optimization algorithms that accurately represent real-world conditions.
A review of coupled physical-based hydrologic models and optimization models to achieve optimal conjunctive management GW-SW resources is presented in this paper. This paper's review of the most used GW-SW models, their features, and suitable optimization methods provides valuable clues for selecting among them for the purpose of research, planning, and policymaking. The application of simulationoptimization has clear benefits for planners, such as determining: (1) the optimal conjunctive management of GW-SW resources, (2) the best possible management of reservoirs and groundwater withdrawal during droughts, (3) the optimal withdrawal to prevent seawater intrusion in coastal aquifers, (4) the indices of reliability, resiliency, and vulnerability that optimize the conjunctive management of GW-SW resources, and (5) the reservoir releases that optimize hydropower production as well as groundwater and streamflow interactions downstream of reservoirs. There are few published articles in the aforementioned categories, and new optimization methods are not commonly applied.
Other fields in water resource management could benefit from simulation-optimization tools. The following fields of inquiry are herein identified as priorities for future research: • Most previous studies have focused on surrogate-based modeling and evolutionary algorithms; however, application of gradient-based methods, coupling the GW-SW simulation models with different optimizers, and comparing the results of various optimizers has not been adequately covered. Investigation on this subject would provide helpful insights for watershed planers.
• The existing literature is predominantly abundant with coupled GW-SW simulation-optimization tools focused on irrigation management, hydropower production planning, and coastal aquifer management, with less attention given to stormwater management considering groundwater quality impacts. GW-SW simulation-optimization tools can be applied to optimize stormwater control strategies and assess water quality and quantity effects on the conjunctive use of surface water and groundwater. First received 10 July 2020; accepted in revised form 8 January 2021. Available online 8 February 2021