Conjunctive-Use Frameworks Driven by Surface Water Operations: Integrating Concentrated and Distributed Strategies for Groundwater Recharge and Extraction

Abstract

This study develops a conjunctive-use framework that couples a surface water allocation model with the MODFLOW groundwater model to evaluate the interactions between surface water operations and groundwater recharge and pumping. The framework enables coordinated surface–groundwater management through iterative feedback between allocation decisions and groundwater responses. Three representative managed aquifer recharge cases in Taiwan are examined, each reflecting a distinct operational logic: (1) a space-for-time strategy that extends wet-season benefits through distributed recharge using irrigation surplus; (2) a centralized support–distributed feedback approach in subsidence-prone areas, where concentrated surface water is delivered to targeted zones while maintaining flexibility for upstream allocation; and (3) a time-for-volume mechanism that converts short-duration flood events into stable, long-term baseflow supply. The simulation results show that these strategies reduce downstream irrigation deficit ratios (e.g., from 0.58 to 0.22), raise groundwater levels by up to approximately 3.5 m in subsidence-sensitive zones, and substantially enhance drought resilience by reducing extreme reservoir depletion during prolonged dry periods. Overall, the proposed framework provides quantitative evidence and a practical planning tool for surface water-oriented conjunctive use, supporting more sustainable and resilient multi-source water management.

1. Introduction

Groundwater provides a natural buffer against the seasonal or even annual variability in surface hydrology, serving as a decentralized water source that can be accessed locally when surface water becomes scarce. Groundwater generally maintains stable quality, requires less infrastructure for development, and is less affected by sedimentation or evaporation losses. The aquifer stores water through natural or artificial recharge during wet seasons, and this stored groundwater can be extracted during dry periods, thereby functioning as a natural subsurface reservoir that stabilizes water supply [1]. In recent years, Taiwan has actively promoted managed aquifer recharge (MAR) and artificial lake projects to enhance water supply resilience during droughts. Such efforts are consistent with international practice in MAR, which seeks to augment aquifer storage, secure drought-period supplies, and mitigate the adverse impacts of over-abstraction [2]. In addition to research on hydrogeological characteristics, recharge potential, infiltration properties, and groundwater flow dynamics, the sustainable utilization of groundwater also requires a study of its integration into the broader water resource system. Such integration ensures that groundwater use aligns with practical supply operations involving multi-source management across space and time. The ultimate goal is to achieve a conjunctive-use strategy that is reliable, efficient, economically viable, and environmentally sustainable.

From an engineering standpoint, the stable operation of a surface reservoir depends on its effective storage capacity and the controllable inflow that can be regulated for supply. When an aquifer is regarded as an operational storage body, its conceptual framework resembles that of a surface reservoir, yet the physical constraints differ. Excessive drawdown of groundwater levels may cause irreversible effects such as land subsidence or seawater intrusion [3], whereas overly high groundwater levels may induce soil liquefaction. Accordingly, the aquifer can be characterized by an operational groundwater level range (OGLR), defined by its top and bottom groundwater levels. The OGLR delineates the effective portion of the aquifer that can be managed for recharge and withdrawal, representing its usable storage capacity. Unlike a surface reservoir that is centrally operated by a single management authority, groundwater systems are developed and utilized by multiple users. Therefore, conjunctive operation must distinguish between the existing groundwater withdrawals and the additional usable volume that can be developed. The controllable inflow in this context corresponds to the net recharge, defined as the total recharge minus existing withdrawals. Because natural recharge processes—such as rainfall infiltration and irrigation return flow—are usually estimated rather than directly observed, and because pumping records from dispersed users are often incomplete, it is necessary to employ numerical modeling for calibration. By adjusting simulated groundwater levels to match historical observations, an inverse modeling approach via optimization can be used to determine a reasonable estimate of net recharge, which then serves as the quantitative basis for assessing further groundwater development.

The core purpose of the conjunctive use of groundwater and surface water lies in dynamically adjusting aquifer recharge and extraction according to the operational conditions within the surface water system, such that the two components mutually support each other in maintaining a supply–demand balance [4,5]. In addition to storage, the operation of a surface reservoir also requires coordination with intake and conveyance facilities. By analogy, in a groundwater system, the “intake structure” corresponds to pumping wells, while the “conveyance facilities” can vary depending on local hydrogeological conditions. Water may be naturally transmitted through aquifer pores to adjacent demand areas or, alternatively, water can be extracted from wells located near recharge zones and conveyed through artificial channels or pipelines. In a multi-source water supply system, groundwater recharge and extraction should therefore not be treated as independent operations but should be integrated within the overall framework of surface water regulation and infrastructure configuration, forming a surface water-oriented strategy for groundwater utilization.

In the application of integrated surface–groundwater modeling, the function of a model is not merely to reproduce or reconstruct historical hydrological processes, but to serve as a tool for planning and decision-making. This concept corresponds to the final stage in the threefold purpose of modeling [6]—understanding the past, predicting the future, and ultimately changing the future. For groundwater development and MAR projects, the value of modeling lies in its ability to explore—prior to implementation—how different configurations may affect the long-term performance and benefits of the water resources system.

Compared with recent integrated modeling studies [7,8,9], the conjunctive-use framework developed in this study places particular emphasis on how surface water operational conditions shape surface–groundwater exchanges and how groundwater levels, in turn, impose constraints on surface water operation strategies. Figure 1 illustrates a conceptual schematic of the proposed framework. The methodological contributions are threefold:

• Cross-timescale coordination. Because groundwater responds more slowly than surface water, the two subsystems are simulated in management-consistent time steps: daily for surface water allocation and at ten-day intervals for groundwater dynamics. Daily surface water operations determine the pumping and recharge decisions for each ten-day period, after which the groundwater model simulates the corresponding water-level responses and feeds them back to guide the next surface water operation cycle.
• Surface water-driven pumping and recharge rules. Pumping and recharge are activated based on operational conditions within the surface water system, such as the availability of surplus flows, reservoir storage exceeding rule curves, or unmet demand surpassing specified thresholds. This structure reflects realistic operational flexibility and priority-based decision-making in multi-source water supply systems.
• Explicit incorporation of groundwater-level constraints. The framework integrates the OGLR that restricts pumping or recharge when groundwater levels fall below or exceed designated thresholds. This mechanism ensures that conjunctive-use strategies remain consistent with long-term aquifer safety and practical management considerations.

This study evaluates three representative configurations of conjunctive surface–groundwater use observed in two major groundwater regions in Taiwan: (i) the Gukeng Artificial Recharge Lake in Yunlin; (ii) a managed-recharge strategy for the severe land-subsidence zones in the Choshui River fan apex that combines a recharge lake with deep recharge wells supplied by excess reservoir storage; and (iii) the Dachaozhou Artificial Recharge Lake in Pingtung. The Gukeng site, located at the downstream end of the Douliu Canal in an irrigation-deficient area, adopts a local circulation strategy by co-locating recharge and extraction facilities and utilizing existing irrigation channels to improve water conveyance efficiency. The fan apex strategy represents a centralized support–distributed feedback approach, where surplus water stored in upstream reservoirs is precisely injected through distributed recharge wells within the subsidence zone, thereby improving regional balance and feeding back to upstream water allocation. In contrast, the Dachaozhou site is situated near the Linbian River, where floodwater is diverted for recharge, forming a time-accumulated storage system that enhances baseflow and supports downstream water supply. Together, these three cases illustrate the contrasting spatial and temporal characteristics of managed aquifer recharge (MAR)—from concentrated to distributed sources and from local to regional utilization. To systematically analyze these mechanisms, this study integrates a surface water allocation model with the groundwater flow model to establish an operational surface–groundwater simulation framework capable of representing various recharge–pumping configurations. The results not only quantify the long-term benefits of artificial lake and well-based recharge schemes, but also provide technical references for designing and operating future MAR projects. This framework reinforces the concept of surface water-oriented groundwater utilization, thereby contributing to the resilience and sustainability of regional water supply systems.

2. Materials and Methods

Previous studies on the conjunctive use of surface water and groundwater have commonly employed optimization-based approaches. These methods typically formulate an objective function to minimize the operational costs of groundwater pumping or artificial recharge, subject to constraints such as meeting water demand requirements, maintaining groundwater levels within the OGLR, and satisfying water quality standards throughout the planning horizon [10,11]. Representative optimization formulations include linear and nonlinear programming for cost minimization and resource allocation [12,13], dynamic programming for time sequential decision-making [14], and evolutionary algorithms for solving highly nonlinear or multi-objective problems [15,16].

In optimization applications, the hydrological processes within the analysis horizon are first estimated, after which the optimal operation strategy for the entire event is directly solved. In practical water resources management, such models can be implemented using a rolling horizon approach, in which operation strategies are updated daily or monthly based on real-time hydrological conditions and forecasts. In contrast, simulation-based approaches are more commonly used in water resources planning [17,18,19]. These studies typically simulate long-term, temporally stepwise water allocation processes using observed or synthetically generated hydrological series. They then derive the expected system performance under specific facility configurations and operation rules.

This study adopts a simulation-based approach to develop an integrated analysis framework for conjunctive surface–groundwater use. The framework is implemented using long-term hydrological observations, groundwater-level monitoring records, in situ infiltration test results, and existing hydrogeological investigations as model inputs. In the proposed framework, the operation of the surface water allocation system drives decisions regarding groundwater pumping or recharge, while the groundwater model simulates the subsequent responses of the groundwater system. The feedback from the groundwater responses is then used to adjust the surface water operation strategy, allowing realistic representations of long-term operational dynamics. This framework thus captures the human–system interaction inherent in conjunctive operation, rather than merely coupling the natural flow processes of surface and subsurface water during specific hydrological events.

To provide a clear spatial reference for the subsequent modeling descriptions, Figure 2 presents an overview of the three representative conjunctive-use configurations analyzed in this study. The figure depicts the geographic locations of the Gukeng recharge lake system, the Hushan Reservoir–wellfield MAR configuration, and the Dachaozhou recharge lake system within Taiwan. This consolidated illustration helps to establish the physical context of the study areas before the surface water allocation model and groundwater-flow model are introduced in Section 2.1 and Section 2.2.

2.1. Surface Water Allocation Simulation Model

The WRASIM model developed by Chou and Wu [17] was employed in this study to construct the surface water allocation system of the case study areas. WRASIM is a generalized water resources simulation model based on the minimum-cost network flow (MCNF) formulation for simulating the spatial allocation of water supply and demand. A network flow model represents a system as a set of nodes and directed arcs, where each arc connects two nodes and carries a flow of water. The constraints of the network require flow conservation at each node and each arc is subject to upper flow and lower flow bounds. In addition, every arc is assigned a unit flow cost coefficient. The minimum-cost network flow problem seeks the flow configuration that minimizes the total system cost while satisfying all constraints.

The model first converts a surface water system into a network structure. Figure 3 depicts a generalized water resources network representing the spatial allocation of water within a single simulation time step—one day in this study. In addition to real system elements (e.g., reservoirs, intakes, and demand nodes), the framework also includes virtual arcs, shown as dashed lines, which represent the inflow to or outflow from the system during each time step. By assigning appropriate cost coefficients to virtual demand or storage arcs—such that their relative magnitudes reflect the priority rules defined by water rights or reservoir operation guidelines—the model can generate allocation results consistent with specified distribution priorities. The arc costs here represent the relative priorities for water allocation, following the principles proposed by Chou and Wu [18]. For instance, Figure 3 illustrates the case where the cost coefficients follow the sequence C_d80% < C_d20% < C_s < C_p < 0 < C_t (excluding the recharge lake, recharge and pumping wells). In this configuration, available water is first allocated to satisfy high-priority demands (C_d80%), then to the remaining portion of demands (C_d20%), followed by reservoir storage (C_s) and hydropower usage (C_p). The zero level represents the neutral reference cost, while C_t > 0 corresponds to terminal discharge or downstream transfer, which is executed only when unavoidable. This hierarchical cost structure ensures that the model allocation outcomes strictly adhere to the priority-based water distribution logic embedded in WRASIM.

The determination of these cost coefficients follows the derivation procedures detailed in Chou and Wu [18], where various priority structures were systematically translated into minimum-cost flow parameters. WRASIM has been applied and verified in multiple planning studies and operational assessments in Taiwan. Its documentation and validation records are available through the Water Resources Planning Institute of the Water Resources Agency of Taiwan.

Although network flow programming is an optimization algorithm, its application in WRASIM serves mainly as a simulation tool that expresses the spatial allocation of water within a single time step based on specified operation rules. The model identifies the optimal water distribution for each period given the defined priority structure, but it does not perform intertemporal optimization of reservoir storage. Instead, WRASIM conducts long-term, stepwise daily simulations to represent the temporal evolution of water allocation under a given configuration and operation strategy. The simulation results are then used to evaluate the system’s expected performances and to further test alternative facility layouts or operational strategies.

2.2. Groundwater Flow Model, Net Recharge, and OGLR

The MODFLOW model developed by the U.S. Geological Survey (USGS) was employed to simulate groundwater flow. MODFLOW numerically solves the three-dimensional saturated flow equation using the finite difference method, accounting for hydraulic conductivity, storage properties, and spatially distributed recharge or discharge fluxes [20]. Through this approach, the direction, velocity, and temporal variations in groundwater flow and levels within each hydrogeological layer can be represented.

The development of the groundwater model begins with constructing a conceptual hydrogeological model based on the site’s geological and hydrological characteristics, followed by defining appropriate boundary conditions. The parameters—such as hydraulic conductivity, storage coefficient, and net recharge—are then calibrated such that simulated groundwater levels closely match historical observations. The calibrated model is subsequently integrated with the surface water simulation framework for conjunctive-use analysis [21]. Because numerous parameters must be estimated within a groundwater system, earlier studies have employed zonation to reduce the number of unknowns or interpolation of parameter values from specific locations to adjacent grid cells [22,23]. Each zone is assigned fixed hydraulic properties and recharge rates, and the calibration seeks to minimize the sum of squared errors between the simulated and observed groundwater levels, typically solved using nonlinear optimization techniques. Comprehensive theoretical discussions can be found in [10,24,25,26,27].

In the conjunctive-use framework, the aquifer is regarded as a functional storage reservoir analogous to a surface reservoir, and the main interaction between the two systems arises from additional recharge or pumping relative to background net recharge conditions. Consequently, simulated groundwater levels may deviate from historical values. However, excessively high or low groundwater levels can induce environmental hazards such as land subsidence, seawater intrusion, or soil liquefaction. To regulate such risks, an operational groundwater level range (OGLR) is introduced as a set of reference limits rather than as a physically derived safety threshold.

Two general approaches are commonly used to define OGLR boundaries. The first is a probabilistic approach, in which long-term groundwater records are analyzed to determine water levels corresponding to specific exceedance probabilities for each calendar period, thereby defining allowable operating ranges. The second is a process-based approach, in which physical models are developed to represent land subsidence behavior or surface–groundwater exchange processes, allowing the OGLR boundaries to be derived from subsidence prevention criteria or environmental flow requirements.

In practical applications, however, the complexity, data requirements, and uncertainty associated with process-based models, together with site-specific geological and hydrological variability, have led most groundwater management practices in Taiwan to adopt probabilistic or empirically based rules. Accordingly, this study applies a simplified and operational form of the OGLR suitable for planning-oriented conjunctive-use analysis. For each well location, the lower bound of the OGLR is defined as the historical groundwater level observed on the same calendar date, and pumping is suspended whenever simulated levels fall below this dynamic threshold. This formulation does not imply that historical groundwater levels represent physical safety limits. Rather, it reflects a conservative management principle aimed at preventing additional drawdown beyond previously experienced conditions. The resulting management response occurs at a periodic (ten-day) scale, consistent with practical groundwater governance, where pumping and recharge strategies are adjusted in discrete operational cycles rather than through real-time control. Importantly, this OGLR formulation is applied only to areas that have not yet exhibited significant land subsidence, where historical groundwater levels correspond to geological conditions that have not undergone irreversible degradation.

To clarify the hydrogeologic conditions represented in the three MODFLOW case studies, brief descriptions of the modeled aquifer stratification are summarized here:

• In the case of the Gukeng Artificial Recharge Lake, the site is located on the southeastern margin of the Choshui River alluvial fan. Recharge from the lake was therefore assigned to the upper unconfined aquifer layer. The layer is composed of coarse sand and gravel with high permeability and an estimated saturated thickness of 100–150 m.
• The Yunlin recharge well is situated in the mid-fan land subsidence zone. The recharge wells inject into the second confined aquifer, which consists mainly of fine sand and silty sand and is overlain by the first regional aquitard. This aquifer occurs at depths of roughly 35–200 m with a thickness of 80–140 m, and is the principal pumping and subsidence-prone layer.
• In the case of the Dachaozhou Artificial Recharge Lake on the Pingtung Plain, the artificial lake overlies a thick, highly permeable unconfined aquifer composed of sand and gravel. Recharge from the lake was accordingly assigned to the upper unconfined layer controlling groundwater mound formation and subsequent baseflow release.

2.3. Integrated Framework and Concept of Conjunctive Surface–Groundwater Operation

To simulate the water exchange between WRASIM and MODFLOW, the nodes representing pumping wells, recharge wells, and recharge lakes were incorporated into WRASIM, as illustrated in Figure 3. The corresponding computational principles governing these nodes are described as follows:

• Pumping wells

  (1)

  Each pumping well node may contain multiple pumping locations. For each location, the pumping capacity and the corresponding aquifer layer and grid coordinates in MODFLOW must be predefined.

  (2)

  The node functions similarly to an inflow node of WRASIM that introduces water into the network. However, unlike the inflow node whose upper and lower bounds are fixed (as shown in Figure 3), this node is associated with flexible flow values. The lower bound of the arc connected to the pumping well node (virtual pumping arc) is set to zero, while its upper bound equals the total pumping capacity.

  (3)

  By assigning different allocation-priority coefficients to the virtual pumping arc, various groundwater-use strategies can be simulated. For example, in Figure 3, when the coefficient Cp is designated such that Cp + C_d80% < 0 and Cp + C_d20% > 0, groundwater pumping serves only as a backup source. It is activated to meet demands of up to 80% of the requirement under the minimum-cost objective function of network flow programming. Conversely, when Cp = 0, groundwater is used routinely and preferentially, replacing reservoir storage releases during normal operation.

• Recharge Lake: A recharge lake functions similarly to a reservoir node in WRASIM but additionally estimates the infiltration volume entering the aquifer based on a specified infiltration rate (mm day⁻¹) and the simulated daily average water surface area. The infiltration rate is assumed constant and independent of the hydraulic head difference between the lake water surface and the groundwater table. This assumption represents a long-term, time-averaged approximation of field infiltration conditions, as the measured rates inherently reflect the dominant influence of soil permeability and vadose-zone characteristics rather than short-term variations in the hydraulic gradient. Such simplification is considered appropriate for long-term, water balance-oriented simulations. The corresponding spatial domain of a recharge lake is linked to the grid cells and aquifer layers in MODFLOW. The infiltration coefficient determines the recharge volume per time step, which is restricted to the uppermost unconfined aquifer layer.
• Recharge Wells: A recharge well node operates analogously to a demand node. The recharge volume supplied to each well in WRASIM is then injected into the specified aquifer layer and grid cells defined in MODFLOW. Each recharge well must therefore be assigned its recharge capacity, spatial extent, and target aquifer in WRASIM. The recharge fluxes are imposed as specified inflows to the designated aquifer layer.
• By adjusting the cost coefficients C_r of the virtual storage arcs for recharge lakes and C_a of the virtual demand arcs for recharge wells, the priority relationship between surface water storage and groundwater recharge can be represented. For instance, if groundwater recharge is intended to utilize only surplus, the corresponding recharge arcs are assigned lower priority (i.e., higher cost coefficients) than those of other surface water uses or reservoir storages. The priority sequence for the case in Figure 3 follows the order C_d80% < C_d20% < C_s < C_p < C_r or C_a < 0 < C_t.

Within the coupled simulation framework, WRASIM simulates the surface water allocation process and determines the pumping and recharge quantities transferred to the groundwater system. MODFLOW then simulates the resulting groundwater-level responses, which are used to evaluate whether the surface water strategies require adjustment. WRASIM operates on a daily time step, whereas MODFLOW uses a ten-day time step to balance simulation accuracy and computational efficiency. This decadal interval reflects the much slower response time of groundwater compared with surface water systems and aligns with the practical frequency at which groundwater pumping or recharge operations are adjusted in real management settings. Therefore, WRASIM accumulates daily groundwater withdrawals and recharge volumes over each ten-day period before invoking MODFLOW. MODFLOW then simulates groundwater flow and level variations for the same period, based on the total recharge and pumping determined by WRASIM. The simulated groundwater levels are compared with the OGLR, and the results are used to update the surface water operation strategy for the next period. The abovementioned procedure of the conjunctive simulation is summarized in Figure 4 and detailed as follows:

• Evaluate groundwater status: At the beginning of each ten-day period, groundwater levels simulated from the previous period are assessed for all pumping and recharge locations to determine whether they remain within the OGLR.

  (1)

  If groundwater levels at a pumping location fall below the bottom of the OGLR, pumping is halted for that period by setting the pumping upper bound in WRASIM to zero. Otherwise, pumping is allowed up to the equipment capacity.

  (2)

  If groundwater levels at a recharge location exceed the top of the OGLR, recharge operations are suspended or deprioritized by assigning higher cost coefficients to the corresponding recharge arcs.

• Simulate surface water allocation: WRASIM performs daily water-allocation simulations for the next ten-day period, computing the total planned groundwater pumping and recharge volumes.
• Simulate groundwater flow: MODFLOW is executed using the allocated pumping and recharge data to simulate groundwater flow and water-level variation within the analysis domain for the current period.
• Update and proceed: The groundwater levels at the end of the ten-day period are used as initial conditions for the next iteration of the coupled WRASIM–MODFLOW simulation.

This integrated modeling framework emphasizes the coupling of water-allocation dynamics and groundwater flow simulation, enabling the realistic evaluation of conjunctive-use strategies under controlled groundwater responses. The operation is driven by surface water allocation, while the groundwater model quantifies the hydrological feedback, forming a comprehensive system for assessing the long-term effectiveness and sustainability of surface water-oriented groundwater utilization.

3. Results of the Case Study

This study presents three cases of conjunctive operation between surface water and groundwater in two major groundwater regions of southern Taiwan—the Choshui River alluvial fan and the Pingtung Plain. Each case adopts a distinct strategy for groundwater recharge and extraction, demonstrating contrasting characteristics of spatial concentration and dispersion in the interaction between surface water systems and groundwater utilization.

3.1. Gukeng Artificial Recharge Lake

The Gukeng Artificial Recharge Lake is located in the mid-to-lower section of the Douliu Canal irrigation district, administered by the Yunlin Irrigation Association within the Choshui River basin of central Taiwan. The facility was constructed to enhance the reliability of the irrigational water supply for the downstream portions of the canal system. Figure 5 presents the surface water system of the Choshui River basin and the associated alluvial-fan groundwater region where the lake is situated. Figure 6 illustrates the surface water resource network of the Choshui River and the irrigation system of the Douliu Canal.

The Gukeng site is situated on the southeastern margin of the Choshui River alluvial fan, where the surface water supply is regulated primarily by the Jiji Weir. Because the diversion flow at Jiji Weir fluctuates substantially within a day, the downstream Douliu Canal—which conveys water to the Gukeng irrigation area—often experiences unstable and intermittent supply. These characteristics make Gukeng an appropriate setting for evaluating a conjunctive-use strategy in which surplus surface water is captured in an artificial recharge lake and subsequently extracted through nearby wells during dry periods.

According to stratigraphic cross sections, the Choshui River alluvial fan can be conceptually divided into four aquifers and two aquitards. The first aquifer extends from depths of approximately 19 to 103 m below the surface. These four aquifers are hydraulically interconnected near the apex of the fan. Within this regional system, the Gukeng area overlies a relatively shallow unconfined aquifer with semi-independent hydraulic behavior. This hydrogeological setting allows a local MAR–pumping scheme to operate while remaining dynamically linked to the fan-wide groundwater responses represented in the model.

To conduct the conjunctive operation of the surface and groundwater systems, the surface water component was simulated using hydrological data from 2009 to 2020. A comprehensive numerical groundwater model of the Choshui River alluvial fan was developed, as represented by the three-dimensional finite-difference grid in Figure 6b. The model calibration was refined particularly for estimating the net groundwater recharge in the Gukeng area. For the analysis, the Gukeng region was divided into eight subzones, each associated with nearby groundwater observation wells, totaling nine stations. Consistent with the simulation period of the surface water model, monthly net recharge values were estimated annually for each subzone.

The calibration aimed to minimize the sum of squared differences between observed and simulated groundwater levels across the nine stations, with the monthly net recharge values of the eight subzones serving as decision variables. The nonlinear optimization algorithm BOBYQA [28] was employed to identify the optimal solution. Figure 7 presents the comparison between observed and simulated ten-day average groundwater levels at the seven observation stations near the Gukeng Artificial Recharge Lake. The resulting efficiency coefficient (CE) for these stations ranged from 0.61 to 0.96, with correlation coefficients from 0.80 to 0.98.

The surface water simulation was conducted at two hierarchical levels:

• Basin-scale simulation: The first level focused on the broader Choshui River basin to evaluate the surplus flow available at the Jiji Weir, as shown in Figure 6a. The surplus flow is defined as the remaining volume of natural flow after deducting the entitled withdrawals and diversions of existing users within the water resource system. In the WRASIM model, this condition was represented by introducing a fictitious demand node whose allocation priority was set after all other demand and storage arcs in the system. The simulation results indicate that the annual inflow at the Jiji Weir totals 4.23 × 10⁹ m³ yr⁻¹, of which 1.69 × 10⁹ m³ yr⁻¹ remains as surplus after satisfying all designated water uses. These surplus volumes predominantly occur during the wet season from May to October.
• Canal-scale simulation: The second level independently simulated the water conveyance of the Douliu Canal and its branch networks, as shown in Figure 6b. In addition to the irrigation water allocated to the Douliu Canal from the previous basin-scale simulation, the model allowed the diversion of surplus water from the Jiji Weir during the wet season—subject to the maximum transmission capacity of the canal—to be conveyed to the Gukeng Artificial Recharge Lake for temporary storage. The designed surface area of the recharge lake is 66.5 ha, with a total storage capacity of approximately 4.08 × 10⁶ m³. Based on in situ infiltration tests [29], the average infiltration rate was set at 2.57 m day⁻¹. In the allocation sequence, the recharge lake was assigned a storage priority lower than that of all irrigation branches of the Douliu Canal. Therefore, surplus water was routed into the lake only when it was not yet full and all demands had been fulfilled. Once the lake reached full capacity, diversions were automatically suspended.

To utilize the recharged groundwater, a pumping scheme was assumed to operate directly around the Gukeng Artificial Recharge Lake. Considering the irrigation demand downstream, the pumping capacity was set at 120,000 m³ day⁻¹. The pumped water was conveyed directly into the Douliu Canal, supplying the downstream irrigation area through the existing canal network. The operational rule assumed that pumping would be activated whenever the water released from the Jiji Weir was insufficient to meet the agricultural demand of the Douliu Canal. The bottom of the OGLR is designed to be temporally variable, equal to the historical observation level on the same date corresponding to the simulation period. When the simulated groundwater level at the nearest monitoring well falls below this range, the pumping operation is suspended. The simulation results are summarized as follows:

• Under the condition without additional recharge or pumping, the average annual water diversion to the downstream branches of the Douliu Canal below the Gukeng site was 22.28 × 10⁶ m³, while the annual water deficit reached 29.72 × 10⁶ m³, corresponding to an average annual deficit ratio of 0.58.
• With artificial recharge implemented, the simulated mean annual recharge volume reached 76.80 × 10⁶ m³. After pumping (18.49 × 10⁶ m³ yr⁻¹) and reusing the recharged water, the average annual deficit in the downstream irrigation area decreased to 11.23 × 10⁶ m³, reducing the deficit ratio to 0.22.
• Table 1. Summary of conjunctive-use performance indicators for the baseline and MAR-implemented scenarios for the Gukeng case, including a deficit ratio of downstream irrigation demands, annual recharge volume, and annual pumping volume.

  Scenarios	Deficit Ratio of Downstream Irrigation Demands	Annual Recharge Volume (Million m3/year)	Annual Pumping Volume (Million m3/year)
  Default without MAR	0.58	–	–
  Implementing MAR	0.22	76.80	18.49

  summarizes the simulation results of the abovementioned scenarios. Since the water supplied to the downstream agricultural area accounted for only about 0.53 times the total recharge volume, a substantial remaining capacity for managed abstraction from adjacent irrigation areas existed to further enhance water supply reliability.

3.2. Artificial Recharge in Land Subsidence Areas

Due to insufficient surface water availability during the dry season, groundwater pumping has long been used to supplement water demands in the Yunlin and Changhua regions, which also coincide with the most severe land subsidence zones in Taiwan. A restricted pumping zone has been designated to reduce further subsidence, and several mitigation measures have been implemented, including sealing public wells, expanding surface water supply, promoting water-saving irrigation, and constructing recharge facilities near the fan apex [30].

In the first simulation scenario, surplus flow from the Jiji Weir is diverted to the Choshui River Fan Apex Artificial Recharge Lake for infiltration. With a diversion capacity of 30 m³ s⁻¹ and an infiltration rate of 2.09 × 10⁻⁵ m s⁻¹, the lake provides an annual recharge of approximately 98.88 × 10⁶ m³. Groundwater levels rise significantly near the lake, but the effect attenuates toward the severely subsided inland areas. At the Tuku observation station, located within the maximum-subsidence zone, groundwater levels increase by an average of 1.23 m. A portion of the infiltrated water percolates through the first aquifer and eventually discharges to the Choshui River as baseflow, illustrating the regulatory behavior of the shallow aquifer system.

Following completion of the Hushan Reservoir on the Beigang River, several public supply wells within the subsidence zone were scheduled for decommissioning. Instead of sealing them, this study evaluated a conjunctive-use alternative in which these wells are converted into recharge wells. When reservoir storage exceeds the upper limit of its rule curve, surplus water is conveyed through the existing raw water pipeline and injected—after treatment—into the deeper second aquifer beneath the subsided area. This flow path is conceptually illustrated in Figure 2. The total recharge capacity is assumed to be 100,000 m³ day⁻¹.

To ensure that diverting surplus reservoir storage for recharge does not compromise municipal supply reliability, an additional set of pumping wells (100,000 m³ day⁻¹ in total) is assumed near the fan-apex recharge lake. Pumping lowers local groundwater levels, thereby enhancing the infiltration potential of the lake. Simulation results show:

• Recharge through the converted wells reaches 13.94 × 10⁶ m³ per year—about 14% of the recharge volume provided by the lake—because surplus storage at Hushan Reservoir occurs mainly during the wet season, limiting the temporal availability for recharge.
• Despite the smaller volume, direct recharge into the subsidence zone raises groundwater levels at the Tuku station by an additional 3.52 m. Within approximately a 10 km radius, groundwater levels at the other three wells also rise by 264–3.44 m (see Figure 8). This indicates that the proposed strategy produces a much more pronounced localized benefit than lake infiltration alone.

3.3. Dachaozhou Artificial Recharge Lake

The Dachaozhou Artificial Recharge Lake is located in the upper reaches of the Linbian River in Pingtung County, southern Taiwan. Hydrogeologically, it belongs to the Pingtung Plain groundwater region, as shown in Figure 9. Administratively, the Pingtung Plain encompasses both Kaohsiung City to the north and Pingtung County to the south. Water use in Kaohsiung primarily depends on surface water and baseflow from the Gaoping River basin, whereas the northern part of Pingtung County relies mainly on groundwater extraction, and the southern area is supplied by the Mudan Reservoir, situated in the Sichong River basin south of the plain.

The Linbian River basin covers a drainage area of 336.30 km². Within the basin, extensive agricultural lands operated by the Taiwan Sugar Corporation dominate the landscape. Irrigation for these farms mainly depends on baseflow diversions through two major irrigation channels in the upstream, while scattered agricultural areas downstream use smaller-scale withdrawals. Although several rivers traverse the Pingtung Plain, the region lacks large surface reservoirs. Moreover, water quality degradation in certain rivers has further limited the use of surface water. As a result, agriculture, industry, and even domestic water supply within the region depend heavily on groundwater resources. Figure 10 illustrates the overall water supply system of the Pingtung area.

The Pingtung Plain, located in southwestern Taiwan, was formed by alluvial deposits from the Gaoping, Ailiao, Donggang, and Linbian Rivers. The lowland area below an elevation of 100 m covers approximately 1130 km², extending roughly 50 km from north to south and 20 km from east to west, with a gentle slope descending from northeast to southwest. The hydrogeological structure within a depth of 220 m consists of four aquifers and three aquitards. The aquifers are highly developed, thick, and laterally continuous across most of the region. In contrast, the aquitards are relatively thin and primarily confined to the southern portion of the plain, resulting in a distinct separation of aquifers in the south, whereas those in the north and east are hydraulically connected.

The first aquifer, forming the shallowest layer of the hydrogeological system, extends across the entire plain, reaching depths of up to about 83.5 m. Its thickness ranges from 23.5 to 83.5 m, with an average of approximately 49.9 m. Because of its shallow depth, this aquifer has been extensively exploited, resulting in numerous domestic and irrigation wells. Shallow groundwater extracted from this layer supports both agricultural and domestic uses. However, intensive groundwater pumping in recent years has caused groundwater levels along the coastal zone to decline below sea level, highlighting the urgent need for managed aquifer recharge in the area.

The Dachaozhou Artificial Recharge Lake was constructed on farmland adjacent to the upper reaches of the Linbian River. During the floods, surplus flow from the Linbian River is diverted into the lake to facilitate groundwater recharge, thereby augmenting the groundwater storage and simultaneously providing the flood diversion and detention capacity to mitigate flood risk. The water intake structure on the Linbian River is designed as a screened river intake facility with a diversion capacity of 116 m³ s⁻¹. The first phase of the recharge lake, which is currently in operation, covers an area of approximately 35 ha. The excavation depth of the lake is about 15 m, with a total storage capacity of 5.25 × 10⁶ m³. The measured infiltration rate at the site ranges between 7.24 and 10.90 m day⁻¹ [31], indicating favorable hydrogeological conditions for rapid infiltration and effective recharge.

To conduct the conjunctive operation of surface and groundwater systems, the surface water component was simulated using hydrological data from 1982 to 2014, focusing on the water resource systems supplying the southern Pingtung region and the Linbian River basin. The groundwater model incorporated not only the hydrogeological structure of the Pingtung Plain, but also the RIV (river) package in MODFLOW to simulate the water exchange between aquifers and river flows. The model further quantified the groundwater discharge contributing to the river baseflow. During the conjunctive-use simulation, both the environmental flow requirements and existing water rights in the downstream reaches of the Linbian River were maintained. The remaining surplus flow was diverted into the Dachaozhou Artificial Recharge Lake, assuming a conservative infiltration rate of 7 m day⁻¹. When the lake became full and infiltration reached saturation, water diversion was immediately halted. The results are summarized as follows:

• Recharge effectiveness: Long-term simulations yielded an average annual recharge volume of 112.7 × 10⁶ m³ yr⁻¹. Incorporating this recharge into the groundwater model produced the simulated groundwater-level responses shown in Figure 11, which shows the differences in groundwater levels and river seepage/exfiltration (i.e., groundwater discharge to the channel) patterns with and without the Dachaozhou Artificial Recharge Lake at the end of the simulation periods. The results indicate that the rise in groundwater levels was most significant between the northern side of the Linbian River and the Donggang River. Some recharge water crossed beneath the Donggang River, influencing its right bank, while some water moved southward across the left bank of the Linbian River. A small portion of the recharge effect also extended toward the coastal zone of Pingtung.
• River exfiltration: In Figure 11, blue zones denote river infiltration, while red zones indicate exfiltration. The elevated groundwater table caused most river segments to shift toward exfiltration, discharging stored groundwater from the aquifer. Throughout the simulation period, the mean annual recharge volume was 112.7 × 10⁶ m³, of which approximately 72% was discharged through river exfiltration, 6% exited through constant-head boundaries, and the remaining 22% contributed to increased system storage.

Figure 12 and Figure 13 summarize the simulated increases in river exfiltration along the Linbian and Donggang Rivers, respectively, presented as boxplots over the 33-year simulation period. Because the Linbian River lies closer to the recharge site, its exfiltration response occurred more rapidly and at a higher magnitude. The average exfiltration rates during the wet and dry seasons were about 200,000 and 100,000 m³ day⁻¹, respectively. In the first four years following recharge initiation, the annual mean exfiltration gradually increased from 7900 to 59,400 m³ day⁻¹, reaching 135,400 m³ day⁻¹ by the fifth year, after which it stabilized within this range. In contrast, the Donggang River, located farther from the recharge site, exhibited a delayed response—its exfiltration stabilized approximately seven years after the onset of recharge. The consecutive annual mean increases during the first seven years were 1200, 10,200, 18,300, 22,100, 41,000, 41,400, and 59,500 m³ day⁻¹. Thereafter, the average exfiltration rates during the wet and dry seasons reached about 80,000 and 60,000 m³ day⁻¹, respectively, showing smaller seasonal variation and more stable outflow patterns.

The area surrounding the Dachaozhou Artificial Recharge Lake belongs to the southern Pingtung water supply zone, which is served by the Mudan Reservoir located at the southernmost tip of Taiwan. Under the scenario of standalone reservoir operation, the designed supply yield of the Mudan Reservoir—38.87 × 10⁶ m³ per year (equivalent to a stable daily supply of 106,500 m³)—was adopted as the target demand in the surface water simulation. The results indicate the following:

• The average daily water supply from the Mudan Reservoir was 97,900 m³. Water shortages primarily occurred between the wet season of 1993 and the dry season of 1994. During this hydrological year (1 June 1993–31 May 1994), the annual water deficit ratio reached 37.8%, while no shortages were observed in other years.
• During the above severe drought period, the reservoir storage dropped to zero for 275 days, during which the Mudan water supply system was entirely depleted, indicating that backup water sources were urgently required during this extreme event.

Considering the recharge effectiveness of the Dachaozhou Artificial Recharge Lake, a new baseflow intake of 30,000 m³ day⁻¹ was assumed to be developed in the downstream reach of the Linbian River. The collected baseflow water, after treatment at a newly constructed treatment plant and conveyance through dedicated pipelines, was integrated into the existing regional supply system. The revised simulation results show that the number of empty-storage days at the Mudan Reservoir decreased drastically—from 275 to only 33 days as shown in Figure 14—and the annual water deficit ratio was reduced to 0.4%. This improvement arises entirely from the stable daily withdrawal of baseflow generated by the recharge lake, which directly substitutes for releases that would otherwise be drawn from the Mudan Reservoir. This clearly demonstrates that incorporating the stable baseflow derived from the recharge lake substantially enhances the resilience of the regional water supply system under severe drought conditions.

3.4. Discussion of Case-Specific Characteristics

The three case studies represent distinct operational philosophies for integrating surface water availability with groundwater storage, each shaped by its hydrological context and management objectives. Their recharge sources, conveyance mechanisms, temporal scales, and system-level roles are summarized in

Table 2. Summary of recharge strategies for the three study cases.

Aspect	(a) Gukeng Artificial Recharge Lake	(b) Conversion of Existing Pumping Wells at Water Treatment Plants into Recharge Wells	(c) Dachaozhou Artificial Recharge Lake
Recharge source	Surplus irrigation water	Excess reservoir storage	River floodwater
Recharge measure	Recharge lake	Recharge wells	Recharge lake
Conveyance to recharge area	Existing irrigation canals	Dedicated transmission pipelines	Direct diversion of river floodwater
Temporal and spatial scale of recharge	Operated within irrigation season with intra-annual cycling	Concentrated during wet seasons with seasonal response	Few days during flood events with long-term aquifer response
Use of recharged water	Pumped near the recharge area and conveyed back to the irrigation system	No direct withdrawal; used indirectly by increasing upstream weir withdrawal	Retained in aquifer to sustain baseflow for subsequent surface water abstraction
Representative quantified outcomes	Deficit ratio reduced from 0.58 to 0.22	Groundwater-level increase of ~2.64–3.52 m in the severe land subsidence zone	The number of empty-reservoir days at Mudan Reservoir reduced from 275 to 33 days.
Additional benefits or highlights	“Trading space for time” strategy: utilizing fallow fields as infiltration areas to rapidly capture surplus water	Policy-oriented recharge: releasing agricultural water rights during dry periods and improving land stability	Improved downstream river water quality and increased available intake volume

and

Table 3. Conceptual comparison of recharge strategies across the three study cases.

Case	Operational Philosophy	Recharge Characteristics (From Concentrated Source to Distributed Infiltration)	Post-Recharge Utilization Characteristics (From Concentrated to Distributed Use)	Temporal Characteristics	Typical Mechanism	System Benefits
Gukeng Artificial Recharge Lake	Trading space for time	Surplus flow at the Jiji Weir (concentrated) distributed through canal networks to mid–downstream recharge lakes	Locally concentrated pumping at recharge lakes redistributed to downstream irrigation areas	Short-term	Surplus irrigation flow → distributed recharge	Expands irrigated area
Dachaozhou Artificial Recharge Lake	Accumulating volume over time	Floodwater concentrated at a single site infiltrates to the aquifer and emerges as a distributed baseflow or underflow accessible at multiple points	River abstractions incorporated into a distributed municipal supply network	Long-term	Flood → concentrated recharge → delayed baseflow	Enhances resilience of public water supply system
Recharge wells converted from existing water treatment plant wells	Centralized support–distributed feedback	Excess storage above reservoir rule curve (centralized) is conveyed via pipelines to multiple distributed wells for precise deep injection	Recharge in subsidence areas indirectly feeds back to the upstream weir, enabling increased groundwater use or reallocation at the Jiji Weir	Stable	Centralized recharge water resources → subsidence-zone injection	Improves regional balance and upstream diversion flexibility

. The discussion below focuses on the distinguishing the strategic principles and management implications of each approach.

(a)

Gukeng Artificial Recharge Lake—distributed capture of short-duration surplus flows

The Gukeng case reflects a spatially distributed, surface water-oriented strategy that captures short but frequent surplus flows from the Choshui River system. By using fallow fields and irrigation channels as temporary infiltration spaces, the system effectively converts wet season surplus into intra-seasonal supply. This “trading space for time” mechanism relies on rapid operational adjustments and leverages existing canal networks. Its effectiveness lies in spatial flexibility and responsiveness to short-duration hydrological opportunities (Table 2 and Table 3).

(b)

Conversion of decommissioned public wells to recharge wells—centralized injection with policy-driven feedback

The recharge well strategy exemplifies a centralized support–distributed feedback framework. Surplus storage from the Hushan Reservoir is conveyed through existing pipelines and injected directly into confined aquifers within the subsidence zone. Despite a smaller recharge volume than the lake-based system, direct injection yields a substantial groundwater-level increase (up to 3.52 m at Tuku and 2.64–3.44 m at surrounding wells within ~10 km). This approach strengthens the feedback loop between groundwater levels and upstream reservoir operation, improving institutional flexibility and enhancing resilience to land subsidence. As summarized in Table 2 and Table 3, the key contribution of this strategy lies in its policy-oriented controllability and its ability to translate centralized surface water surplus into targeted subsurface recovery.

(c)

Dachaozhou Artificial Recharge Lake—temporal accumulation and delayed release

Dachaozhou operates as a temporal regulation strategy in which short-duration floodwaters are rapidly infiltrated into thick alluvial aquifers. Stored groundwater subsequently contributes to sustained baseflow and enhances downstream surface water reliability. This “accumulating volume over time” approach provides long-term drought buffering and reduces reliance on upstream reservoirs during extended dry periods. Its system benefit arises primarily from delayed release and aquifer-mediated regulation, as opposed to immediate reuse (Table 2 and Table 3).

In summary, the three cases collectively demonstrate the complementarity between centralized and distributed operations in conjunctive surface–groundwater management. By viewing groundwater as both a storage medium and a regulatory interface between upstream and downstream systems, these cases illustrate complementary pathways for enhancing regional water supply resilience under varied hydrological and institutional environments.

3.5. Limitations and Future Extensions

While the simplified OGLR formulation adopted in this study provides a practical and operationally consistent constraint for long-term conjunctive-use planning, it also entails inherent limitations that warrant further investigation. In particular, the OGLR serves as a managerial reference rather than a physically based indicator of subsidence risk or aquifer safety. Its effectiveness therefore depends on the hydrogeological context, and is limited to areas where historical groundwater conditions reasonably represent pre-degradation states.

Future research could enhance this framework by replacing or supplementing the current OGLR with physically based criteria, such as subsidence thresholds derived from coupled groundwater–geomechanical models. Integrating land subsidence simulations would allow a more explicit linkage between groundwater-level management, recharge strategies, and deformation risk, especially in subsidence-prone regions such as those addressed in the Hushan Reservoir–recharge-well case. Such extensions would improve the interpretability of conjunctive-use benefits in terms of both water supply resilience and geotechnical safety.

From an engineering perspective, the long-term sustainability of artificial recharge systems also depends on maintaining infiltration efficiency. Recharge lakes may experience gradual clogging of the lake bed due to fine sediment deposition, biological growth, or chemical precipitation. Sustained system performance therefore requires periodic lake bed maintenance, such as surface scraping or soil turning, as well as strict control of recharge water quality to prevent the inadvertent injection of pollutants into the aquifer. These operational constraints are not explicitly represented in the present simulations but are critical for translating modeled benefits into durable real-world outcomes.

In addition, the current analysis focuses primarily on water quantity and supply reliability. Incorporating water quality modeling—particularly for river–aquifer exchange and baseflow contributions—would allow explicit assessment of co-benefits related to pollutant dilution, river restoration, and downstream intake suitability. Such integration is especially relevant for cases like Dachaozhou, where enhanced baseflow directly affects surface water availability.

Finally, methodological extensions are needed to improve computational efficiency and broaden applicability. Machine learning-based surrogate models could be developed to approximate groundwater responses, enabling rapid scenario screening and adaptive planning without sacrificing key hydrological feedbacks. Coupling the framework with climate change scenarios would allow for further systematic evaluation of how surface water-oriented conjunctive-use strategies can buffer increasing hydrological variability and extreme events, thereby strengthening long-term water resources resilience under future uncertainty.

4. Conclusions

This study proposes a surface water-oriented conjunctive-use framework that integrates surface water allocation with groundwater recharge and pumping to enhance regional water supply resilience. Unlike conventional conjunctive-use approaches that treat groundwater as an independent or auxiliary resource, the proposed framework explicitly positions groundwater as an operational storage component governed by surface water availability, infrastructure configuration, and management priorities.

Its application to three representative cases in Taiwan demonstrates how different conjunctive-use mechanisms operate across spatial and temporal scales. In the Gukeng case, distributed recharge using surplus irrigation flows reduced the downstream irrigation deficit ratio from 0.58 to 0.22, illustrating how short-term surface water surplus can be converted into intra-seasonal supply. In land subsidence-sensitive areas, centralized recharge through converted wells increased groundwater levels by up to 3.52 m within targeted zones, highlighting the effectiveness of spatially focused recharge strategies. In the Dachaozhou case, floodwater-driven recharge enhanced long-term baseflow contributions, allowing a stable withdrawal of 30,000 m³ day⁻¹ to substitute reservoir releases and reduce the number of empty-storage days at the Mudan Reservoir from 275 to 33 during an extreme drought year.

These results indicate that conjunctive-use performance depends not only on recharge volume but also on the timing, spatial placement, and operational integration of recharge and extraction facilities. The primary contribution of this study lies in articulating and quantifying a surface water-oriented conjunctive-use philosophy, in which groundwater functions as a regulated and responsive storage system that complements surface water operations rather than competing with them. This perspective provides a practical and transferable basis for designing conjunctive-use strategies that improve water supply reliability under hydrological variability.