A Conceptual Framework for Integration Development of GSFLOW Model: Concerns and Issues Identified and Addressed for Model Development Efficiency

In Coupled Groundwater and Surface-Water Flow (GSFLOW) model, the three-dimensional finite-difference groundwater model (MODFLOW) plays a critical role of groundwater flow simulation, together with which the Precipitation-Runoff Modeling System (PRMS) simulates the surface hydrologic processes. While the model development of each individual PRMS and MODFLOW model requires tremendous time and efforts, further integration development of these two models exerts additional concerns and issues due to different simulation realm, data communication, and computation algorithms. To address these 5 concerns and issues in GSFLOW, the present paper proposes a conceptual framework from perspectives of: Model Conceptualization, Data Linkages and Transference, Model Calibration, and Sensitivity Analysis. As a demonstration, a MODFLOW groundwater flow system was developed and coupled with the PRMS model in the Lehman Creek watershed, eastern Nevada, resulting in a smooth and efficient integration as the hydrogeologic features were captured and represented. The proposed conceptual integration framework with techniques and concerns identified substantially improves GSFLOW model develop10 ment efficiency and help better model result interpretations. This may also find applications in other integrated hydrologic modelings.

system is important in response to climate change. On the basis of available geologic conditions and hydraulic connectivity, the objective of this study was to provide techniques addressing the concerns in model integration from a perspective of MODFLOW development as a groundwater component in the GSFLOW model. The findings from this study are anticipated to provide useful information to modelers/end users regarding the integration of groundwater systems to a surface hydrologic modeling system using GSFLOW model.

5
The current paper is constructed by four compartments. Firstly, the basic components and modeling scope of GSFLOW model are briefly reviewed in Section 2, and it is followed by the description of proposed conceptual framework, in Section 3, where potential concerns and issues during the development of integrated processes in GSFLOW model are identified and addressed. Then, the proposed framework is implemented and demonstrated, in Section 4, through developing a MODFLOW model and integrating it as a groundwater model component for an fully integrated GSFLOW model. Lastly, the discussion 10 and conclusions over the current study were made in Section5. spatial-distributed physical bases, simulating processes from top of vegetative canopy to the bedrock. Based on water balance and energy balance, it particularly focuses on the surface hydrologic processes including canopy interception, snow accumulation/melt, evapotranspiration, surface runoff, and soil-water fluxes. While its groundwater flow is simplified as a stock-and-flow system, a sophisticated groundwater model would improve the modeling performance of integrated water system. MODFLOW is a three-dimensional finite-difference groundwater flow system developed by the USGS (Markstrom et al., 2005). The finite-5 difference method was used to describe the spatial heterogeneity to solve groundwater flow (and contaminants) through porous mediums in three dimensions, by area (e.g., infiltration or evapotranspiration), by line (e.g., streambed infiltration and its water exchange with groundwater), or by point (e.g., water pumping and recharge). It is the most widely used simulation program for groundwater systems throughout the world (Markstrom et al., 2005). By coupling these two models, the major limitation of each model is overcome, as the GSFLOW simulates both surface water and groundwater/subsurface-water simultaneously 10 with dynamic water interacting through saturated and unsaturated subsurface media and through streams and lakes.

A Conceptual Framework for GSFLOW Model Integration Development
In GSFLOW, the integration script was completed by USGS, who developed both PRMS and MODFLOW model. The conceptual framework proposed herein aims to facilitate the model development of GSFLOW from a modeler perspective. Generally, to develop a coupled GSFLOW model, the two models to be integrated are developed separately and have a pre-calibration 15 respectively before the coupling processes (Huntington and Niswonger, 2012;Markstrom et al., 2005). Traditional model development procedures, e.g., model calibration, validation, and initialization, are applicable and required for both individual model.
During these processes, different from an independent model development for non-integration purposes, there are concerns or potential issues that should be aware of or dressed, which would help modelers to better understand the GSFLOW integrated hydrologic model, improve the efficiency of model development, and have better interpretation of simulation results. In the 20 following sections, the main concerns or issues are addressed in the proposed framework: Model Conceptualization (section 3.1), Data Linkages and Function Role Change (section 3.2), and Model Calibration and Sensitivity Analysis (section 3.3).

Model Conceptualization
While two models, PRMS and MODFLOW, could have two independent approaches of model conceptualization when for separate studies, aiming for a smooth and successful coupling development for a GSFLOW model, these two ways of model structurally deficient with none driving inputs/forces. Especially on the level of spatially-discretized hydrologic response unit, a structure connection is required for data communication to assure vertical flows (e.g., gravity drainage) between PRMS soil zone and MODFLOW groundwater system; or else, such structure connection needs to be externally defined (Markstrom et al., 2015). Also, as temporal unit of model simulation, time step is another concern of importance. Due to different study interests, surface water and groundwater may have different time steps in terms of hours, days, or months, depending on varied study 5 purposes. Nevertheless, PRMS model only supports daily time step for a PRMS-IV simulation (Markstrom et al., 2015). This limits GSFLOW model simulation to a daily basis and so does the MODFLOW model component for the compatibility reason.

Data Linkages and Function Role Change
Leveraging the future data transferences during the model development facilitates the efficiency and effectiveness of the integrated modeling. As the groundwater component in GSFLOW model, MODFLOW interacts with the surface water system 10 mainly through three data linkages, including: -1) Water percolation, resulting from the surface-water system and driving the groundwater system; -2) Evapotranspiration, composed by shallow ET and deep-root ET simulated by two sub-model respectively; -3) Streamflow, contributed by both surface runoff and dynamic water interacted with groundwater system.

15
As driving input of groundwater model, water percolation determines groundwater system behavior and model performance.
The gravity drainage, resulting from PRMS model simulation, is a portion of infiltration, after the fulfillment of shallow soilwater flow, vertically percolates into and recharges the groundwater system. The spatial distribution and value scale of magnitude of long-term percolation is determinately correlated with those of hydraulic properties in subsurface medium. As results of PRMS surface hydrologic simulation, the value scale and spatial distribution of the gravity drainage make a well correlation 20 between the flow rate and soil type. This well-suited correlation reflects as the driving inputs and hydraulic propertied of MOD-FLOW. Inherently, scale and use PRMS simulation gravity drainage to saves considerable efforts and time resulting a speed up for the model development in terms of initialization. Typical groundwater MODFLOW model simulation requires an initial condition set up for purposes of accurate simulation performance and a successful numerical solution (Bear, 2012;Franke et al., 1987). Instead of initiating an independent groundwater model using numerically expensive approaches, i.e., draining 25 test/spin-up test (Ajami et al., 2014a, b;Seck et al., 2015), directly using the PRMS model output to drive MODFLOW model initiates the data communication between models and leads a heads-up of a GSFLOW model simulation.

ET -leveraging ET simulation in both PRMS and MODFLOW
As one of the most important processes in integrated hydrologic system, ET is considered both in the soil zone of PRMS in areas where the deep ET is active and has great influences on the seasonal variation of the water cycle. In cases while total ET were considered during the initial PRMS model development, the deep ET portion should be split out during the coupling process as to capture active variabilities. In both PRMS and MODFLOW model simulation, high variation of deep ET raises great influences in water dynamics within each sub-system, e.g., soil infiltration, soil water thresholds, soil water discharges to or absorbed from unsaturated zone, groundwater-level, and GW storage. 1) the streamflow receives from /discharges to the groundwater system; 2) the overland flow that enters each stream segment.

15
Understanding these two sources as the most critical determinant elements in the streamflow would of great help during the model calibration, which is discussed in the following section. As listed above, these three data linkages summarize the keys of simulating dynamic water interactions across the two sub-systems occurred in two critical realms: soil and stream. The smooth data communication is companied by algorithm changes with different module/packages used in both PRMS model and MODFLOW model (Table A1,A2). Especially, the critical integration process is determined by two modules in GSFLOW 20 (Markstrom et al., 2005;Regan et al., 2016): gsflow_prms2mf and gsflow_mf2prms. The gsflow_prms2mf module is used to direct PRMS outputs to MODFLOW model, which includes distributing gravity drainage and unsatisfied ET to MODFLOW and allocating surface runoff (i.e., overland flow, Dunnian runoff, and Hortonian runoff) and subsurface interflow to stream segments (Related and Tables, 2015). The gsflow_mf2prms module is used to distribute groundwater discharges from MOD-FLOW cells to PRMS hydrologic response units (HRUs) when condition met. Additional parameters, which were required for 25 these two modules, were summarized in Table A2.

Model Calibration and Sensitivity Analysis
Typical groundwater MODFLOW model simulation requires an initial condition set up for purposes of accurate simulation performance and a successful numerical solution (Bear, 2012;Franke et al., 1987). Instead of initiating an independent groundwater model using numerically expensive approaches, i.e., draining test/spin-up test (Ajami et al., 2014a, b;Seck et al., 2015),

30
Lehman Creek watershed is located in east-central Nevada close to Nevada-Utah boundary and encompasses the Great Basin National Park (Fig.2). Defined by the surface topographical conditions, it covers an area of 23.6 km 2 , elevating from 2040  (Homer et al., 2015)) make the climate dry hot at the lower plain area and the humid cool in high-elevation regions. More than 60% of the precipitation falls as snow in the mountainous areas (Volk, 2014). The Lehman Creek initiates at the glacial deposits that overlay older undifferentiated argillite, quartzite, and shale (Unp). In the cross-section shown in Fig.2, the granite and shale intrusion separate the quartzite upstream and the limestone formation downstream, where the groundwater discharges as 5 Cave Springs (Elliott et al., 2006). The groundwater outflows the watershed boundary, passing through the dissolute limestone formation, joining the adjacent Baker Creek (Halladay and Peacock, 1972;Elliott et al., 2006).

GW System Conceptualization
According to Prudic et al. (2015), in the study area where the geology is dominated by quartzite and glacial deposits (Fig.2), 10 most of precipitation forms into surface runoff, with minor groundwater flow occurring. Groundwater flow receives a recharge from macrofractures as well as coarse sediment in the glacial deposits and alluvium with small storativities. Impervious quartzite and granite impede the groundwater flow and force it into the spring discharge (Fig.2). In the area between the intrusion and the downstream watershed boundary, the losing-stream recharges the groundwater through both glacial and alluvial deposits as well as the underlying karst limestone. Also, the groundwater interacts with the neighboring Baker Creek 15 watershed at southeastern boundary (Prudic et al., 2015). To couple the MODFLOW with surface hydrologic PRMS model in a simple and straightforward approach, the identical modeling area and grid mash as used in the PRMS model were applied in the MODFLOW model to ensure the data communication between two sub-systems on both region level and grid level. Yet, it resulted in adjustments in boundary conditions to compensate the imbalanced water cut-off due to the different "watershed" definitions in surface water and groundwater system. Herein, the spring discharges and the groundwater outflows 20 were considered on the basis of water balance estimation, as the boundary conditions. As the Table 1 shows, the water balance estimation includes vertical infiltration as inflow (1010 m 3 /d), derived from the PRMS model; the system outflows of baseflow (450 m 3 /d; (Prudic et al., 2011(Prudic et al., , 2015), spring discharge (245 m 3 /d; (Halladay and Peacock, 1972;Prudic and Glancy, 2009)), and groundwater outflows at an estimation of 315 m 3 /d. Fig. 3 shows the position where the boundary flux occurred. A twolayer groundwater flow system was defined, based on hydrogeologic features (Maxey, 1964;Seaber, 1988). Layer 1 consisted 25 of glacial and alluvial deposits and Layer 2 consisted of fractured quartzite at the upstream, limestone at the downstream, split by granite and shale intrusions (Fig.3). The granite and shale that underlie the fractured quartzite only was represented at the intrusion as model bottom was considered as no-flow boundaries in this model (Fig.3).

Model Development
Apart from the fundamental MODFLOW model development, e.g., model package setup and parameterization (Chen et al.,    Value Ranges of selected rocks (Heath 1983) loss in the water balance, which includes the Cave Spring and groundwater outflows to the adjacent Baker Creek drainage (Volk, 2014). The parameter (jh_coef ) determining the potential evapotranspiration in the soil was adjusted with a reduction, and the compensation was made by deep-root evapotranspiration simulated by the MODFLOW. Secondly, the gravity drainage from PRMS was the MODFLOW model-driving inflow and was in balance with groundwater outflows that were not considered in the PRMS model, including spring discharges and boundary outflows, by adjusting the parameter (ssr2gw_rate, ssr2gw_exp) to 5 modify the exponential curve that determining the gravity drainage rate. In terms of role exchange, all the routing processes and related parameters in the previous PRMS model were forfeited, and a new module, as a role replacement in the integrated GS-FLOW model, StreamFlow Routing packages (SFR) was used to present the streamflow routing process from stream originate to the outlet of the watershed and to account for the streamflow-groundwater interactions. Streambed thickness and hydraulic conductivity were estimated for each specified featured hydrogeologic formation, according to piezometer measurements and 10 literature studies (Prudic and Glancy, 2009;Allander and Berger, 2009) Detailed changes in modules and related parameters can be found in appendix.

Model Calibration and Sensitivity Analysis
The calibration procedure for MODFLOW, as a component of an integrated hydrologic model GSFLOW, includes steady-state and transient-state model calibration, which were performed for both MODFLOW_only simulation and integrated GSFLOW Geosci. Model Dev. Discuss., https://doi.org/10.5194/gmd-2018-268 Manuscript under review for journal Geosci. Model Dev.

Modeling Results
By applying the conceptual framework and having potential concerns addressed, the MODFLOW development playing a componential role in the integrated system turns out successful, as the transition from an independent model to a system component

Discussion and Conclusion
The primary objective of the current study is to propose a conceptual integration framework with techniques and concerns identified and addressed that can improve GSFLOW model development efficiency and help better simulation interpretations. Focusing on the main elements in modeling procedures, Model Conceptualization, Data Linkage and Function Role Change, and Calibration and Sensitivity Analysis, the proposed conceptual framework identified the keys for a successful 30 model communication between two sub-models, i.e., PRMS and MODFLOW, within GSFLOW model. The tackling strategies and techniques were proposed correspondingly. As a demonstration, the proposed framework was applied to a study in the well estimated the hydraulic conductivities and storativities of the defined stratigraphic units, which kept the water balance estimation and captured the hydrogeologic features with spring discharges and groundwater outflows. In this study, the main 5 conclusions drawn from this study are: -Keeping a consistency in spatial and temporal discretization of two sub systems is important to the GSFLOW model development, while such consistency restrains the implementation of GSFLOW model due to temporal scale and raises extra requirements for boundary conditions due to spatial definition differences; -Leveraging three active data linkages, vertical percolation, deep-root ET uptake, and streamflow-aquifer interactions, in 10 the integrated model development is critical for successful data communication and subsequent dynamics within two sub system and inherently the integrated system; -Using the Gravity Drainage result of PRMS model to drive MODFLOW model is an efficient technique to: 1) fast converge the groundwater modeling as it keeps the soil texture in surface hydrologic simulation align with hydraulic properties in groundwater system simulation; 2) debug the initialized GSFLOW at its early-stage;