Environmental Modelling & Software 73 (2015) 103e116
Contents lists available at ScienceDirect
Environmental Modelling & Software journal homepage: www.elsevier.com/locate/envsoft
A model integration framework for linking SWAT and MODFLOW J.A. Guzman a, *, D.N. Moriasi a, P.H. Gowda b, J.L. Steiner a, P.J. Starks a, J.G. Arnold c, R. Srinivasan d a
USDA Agricultural Research Service, Grazinglands Research Laboratory, El Reno, OK, 73036, USA USDA Agricultural Research Service, Conservation and Production Research Laboratory, Bushland, TX, 79012, USA c USDA Agricultural Research Service, Grassland Soil and Water Research Laboratory, Temple, TX, 76502, USA d Texas A&M, Agriculture and Life Sciences, Spatial Science Laboratory, College Station, TX, 77843, USA b
a r t i c l e i n f o
a b s t r a c t
Article history: Received 28 November 2013 Received in revised form 7 August 2015 Accepted 11 August 2015 Available online xxx
Assessment of long-term anthropogenic impacts on agro-ecosystems requires comprehensive modelling capabilities to simulate water interactions between the surface and groundwater domains. To address this need, a modelling framework, called “SWATmf”, was developed to link and integrate the Soil Water Assessment Tool (SWAT), a widely used surface watershed model with the MODFLOW, a groundwater model. The SWATmf is designed to serve as a project manager, builder, and model performance evaluator, and to facilitate dynamic interactions between surface and groundwater domains at the watershed scale, thus providing a platform for simulating surface and groundwater interactions. Using datasets from the Fort Cobb Reservoir experimental watershed (located in Oklahoma, USA), the SWATmf to facilitate linkage and dynamic simulation of SWAT and MODFLOW models. Simulated streamflow and groundwater levels generally agreed with observations trends showing that the SWATmf can be used for simulating surface and groundwater interactions. © 2015 Elsevier Ltd. All rights reserved.
Keywords: Integrated hydrologic modelling SPELLmap Surface-groundwater interactions SWATmf Water resource management
1. Introduction Assessing the long-term impacts of natural and anthropogenic drivers in watershed dynamics (i.e., hydrological response, transport of contaminants, and ecosystems services) requires integration of knowledge and modelling capacities across biophysical responses, environmental problems, policies, economics, datasets, and computer capabilities (Laniak et al., 2013). The primary goal of model integration is to bridge fragmented cross-disciplinary knowledge to strengthen the quantitative capacity to rigorously evaluate hypotheses and system response under dynamic scenarios (Arnold, 2013). The Soil and Water Assessment Tool (SWAT; Arnold et al., 1998) and the Modular Three-Dimensional Finite-Difference Groundwater Flow (MODFLOW; McDonald and Harbaugh, 1988) models are well-tested and widely-used surface and groundwater models, respectively. However, these models represent the physical world (i.e., model spatial discretization and process simulation) differently and each is limited to its simulation domain, each having
* Corresponding author. E-mail address:
[email protected] (J.A. Guzman). http://dx.doi.org/10.1016/j.envsoft.2015.08.011 1364-8152/© 2015 Elsevier Ltd. All rights reserved.
advantages and disadvantages when simulating biophysical processes and using computational resources. The SWAT model only simulates shallow groundwater dynamics above a restricted layer (SWAT model lower boundary domain). Percolation below the impervious layer, which is set at a maximum value of six m below the ground surface (Neitsch et al., 2011), is flow assumed lost out of the system (Fig. 1). SWAT simulates surface and shallow aquifer processes (Fig. 1) based on hydrological response units (HRUs), which are conceptual units of homogeneous land use, management, slope, and soil characteristics that extend below the surface to a soil profile depth (Arnold et al., 1998). HRUs are modelled as non-geo-located, spatially disconnected representations of spatially derived geolocated polygons belonging to a given sub-basin in the surface domain. Thus, the SWAT model uses a quasi-two dimensional step-by-step budgetary approach at the HRU level to account for changes in the hydrological response. This configuration has both advantages and constraints (Garen and Moore, 2005; Walter and Shaw, 2005). The lumping of spatially geolocated polygons into HRUs speeds up simulation of processes that takes into account land use and land management. However, individual one-dimensional computations at the HRU level are summed up within a sub-basin and routed to the corresponding sub-basin outlet without considering HRU-to-HRU spatial
104
J.A. Guzman et al. / Environmental Modelling & Software 73 (2015) 103e116
Fig. 1. Schematic of the hydrologic cycle and SWAT simulation processes (Neitsch et al., 2011).
interaction. With this configuration, it is not possible to spatially integrate SWAT with gridded groundwater models at the HRU level. MODFLOW simulates flow processes occurring at the continuum volume in the saturated zone defined by three-dimensional cells (groundwater domain) and hydrogeological properties. MODFLOW simultaneously solves the groundwater flow differential equation using the finite difference approach, and integrates groundwater systems with other hydrological sub-system components (e.g. vadose zone, surface drainage, transport phenomena, etc.) through incorporation of “packages” using a gridded spatial discretization. However, it does not directly account for hydrologic processes that occur on the surface or in the root zone. Consequently, a common practice is to assume lumped percolation fluxes as a percentage of precipitation, and then optimize the value during the calibration process. Whereas the groundwater model calibrated for recharge can provide reasonably good groundwater level predictions, it is possible that the user may get the right answer for the wrong reasons (Kirchner, 2006) because this approach fails to account for spatial variability in recharge rates as a result of varying land use, irrigation and agronomic practices implemented on the surface domain. In addition, this approach may misrepresent transport of nutrients moving to the groundwater domain for the same reasons. Therefore, an integrated SWAT and MODFLOW is essential to better spatially represent feedback fluxes within the surface and groundwater domains. This will improve simulation of impacts of long-term stressors, such as climate variability and change (Brown and Funk, 2008; Wheeler and von Braun, 2013), irrigation technology and management (Playan and Mateos, 2006), land use change (Scanlon et al., 2005; Chu et al., 2013a), disturbances (e.g., wildfire; Beeson et al., 2001), transport of nutrients to aquifers in agricultural production systems, and water resources assessment. SWAT has been integrated with other models to improve simulations of riparian buffer zones (SWAT-REMM; Ryu et al., 2011),
sediment and hydrodynamic flow simulation (SWAT-SOBEK; Betrie et al., 2011), storm water management (SWAT-SWMM; Kim et al., 2011) using dedicated approaches or within the OpenMI model integration framework (Gregersen et al., 2007), and surface and groundwater processes (SWAT-MODFLOW; Sophocleous et al., 1999, 2000; Conan et al., 2002; Kim et al., 2008). Although SWAT and MODFLOW have been integrated for specific purposes and applications, no comprehensive modelling framework is available for application in different locations. Model integration is challenging and limited by specific model code, model internal logic, cross-model data formats, and data interchange. To address these issues, model integration frameworks such as OpenMI (Gregersen et al., 2007) and PCRaster (2008) have been developed. The OpenMI standard allows integration of model components that comply with this standard to be configured to exchange data during simulations (computation at run-time). Additional code is necessary at the core of the model logic to allow data synchronization at time step simulations in coupling models where feedback fluxes are important. The OpenMI architecture was designed to be cost-effective, and to enable model migration while providing model developers freedom to adopt it whenever necessary. The PCRaster is a collection of operational and logical tools integrating the temporal dimension targeted at the development and deployment of cell described environmental models (structured grids) in two or three dimensions. For example, Schmitz et al. (2009) coupled MODFLOW and PCRaster to build an integrated model of the “Utrechtse Heuvelrug” watershed to demonstrate model integration without low-level programing knowledge. In this investigation, neither the OpenMI nor the PCRaster modelling frameworks were adopted due to the specific model features (differences in models spatial discretization approaches), constraints associated with model maintenance cycle and development roadmap, and the need for integration/development with other important model components (e.g., river model) in
J.A. Guzman et al. / Environmental Modelling & Software 73 (2015) 103e116
105
the future. Note that HRUs geolocation in the SWAT model are implicit while using the ArcSwat wrapper application during the model parameterization but they are inexistent during execution of the SWAT model. Also, HRUs in the SWAT model are individual units aggregated to the sub-basin level without spatial connectivity. In contrast, spatial discretization of MODFLOW is based on interconnected structured grids with implicit spatial location. The main goal of this study was to develop a framework “SWATmf” to integrate SWAT and MODFLOW models to improve the assessment of impacts of natural and anthropogenic stressors on both surface and groundwater resources at the watershed scale. This integration should allow users to account for spatially variable recharge that takes into account the effects of dynamic land use and land management practices. A case study was conducted to test the capabilities of the SWATmf model integration approach. The case study was carried out using datasets from the Fort Cobb Reservoir experimental watershed (FCREW), a USDA-ARS benchmark watershed located in southwestern Oklahoma (Steiner et al., 2014).
dimensional cell-to-cell (e.g., gridded HRUs and MODFLOW cells) and point-to-cell (e.g., groundwater wells to HRU/MODFLOW grid cells) model spatial discretization indexation (Fig. 2b). Lookup tables are used by the SWATmf to code/decode necessary data for well extractions, percolation, aquifer-river fluxes, and lake and reservoir infiltration fluxes. The SWATmf project builder application (i.e., SWATmf-app), was integrated with SPELLmap (Guzman et al., 2013) to provide a unified spatio-temporal context for SWAT and MODLOW models (Fig. 2a). SPELLmap is a standalone application capable of creating and manipulating geodata and time series linked to point networks. The SWATmf-app works as the MODFLOW project builder while providing capabilities to edit/visualize SWAT-projects. Also, SWATmf-app allows for retrieval of model outputs, rapid visualization, metrics computation, and serves as the SWAT-MODFLOW project manager.
2. SWATmf e an integrated hydrologic modelling framework
The interfacing routines (Fig. 2a) were developed to setup the required MODFLOW packages during the model execution to: (1) define specific SWAT or MODFLOW control variables (i.e., integrated model configuration), (2) support the unified spatiotemporal framework (due to the differences in spatial discretization for these models), and (3) provide and support the necessary flow of data (model inputs, model set-up, perform spatial domain transformations, etc.) between the two models at daily time steps (models data interchange). These routines were developed in object-oriented FORTRAN and integrated within the modified SWAT and MODFLOW codes. Because the HRUs in the SWAT model are not geolocated, the ArcSWAT-derived HRUs must first be spatially represented in a grid format (raster; converting the HRUs shape file created by ArcSWAT to a raster using the HRUs feature class without the refinement option), and then linked with the corresponding MODFLOW domain through a lookup table (one-tomany cell indexing; Fig. 2b). A first call to the interfacing routines (Fig. 3) performs assignment of MODFLOW “files handles” (e.g., name file; NAM), read lookup tables and contextual data (e.g., well network, HRUs and MODFLOW extend grids), and setup the MODFLOW output control package (e.g., OC). During the SWAT daily simulation, calls to the interfacing routines are invoked (Fig. 2) prior to each MODFLOW stress period simulation to estimate spatially distributed percolation fluxes (i.e., MODFLOW recharge package; RCH) and well extraction volume (i.e., well package; WEL) from SWAT simulations when using the auto-irrigation model. The interfacing routines are responsible for decoding and inputting SWAT-simulated percolation fluxes into MODFLOW.
Due to the differences in the spatial discretization between the two models (SWAT and MODFLOW), there was a need to first geolocate the SWAT model derived HRUs, and then to develop a dynamic linkage (spatial connectivity) to the MODFLOW spatial discretization (i.e., structured grids). The unified spatial framework allowed models to interchange fluxes such as the deep seepage estimations from SWAT simulations (percolating fluxes below the shallow aquifer/root zone boundary) at the HRU level as well as to propagate the estimated SWAT irrigation fluxes to the MODFLOW domain. Additionally, this unified spatial framework was incorporated in SWATmf for data preparation (spatial and non-spatial data), model parameterization, and models output data processing. The goal was to facilitate models interoperability for specific applications or as part of future developments (e.g., Incorporation of other agro-ecosystem modelling elements, Kalaugher, et al., 2013; groundwater chemical transport model; river model). Implementation of SWATmf was carried out in four steps: (1) development of the conceptual framework, (2) development and integration of a unified spatio-temporal framework with capabilities to create and manipulate cell oriented maps (i.e. grids) and time series with SPELLmap (Guzman et al., 2013), (3) development and insertion of hard-coded routines to facilitate interactions between the two models, and (4) development of integrated model by combining the SWAT, MODFLOW, and new developed routines' source codes. 2.1. Model integration The Newton formulation for MODFLOW 2005 (NWT v. 1.0.5, Niswonger et al., 2011) was integrated with SWAT (v. 579; Arnold et al., 1998). Fig. 2 summarizes the conceptual modelling framework and interfacing routines. The MODFLOW-NWT and SWAT codes were modified, combined, and then compiled with the newly developed interfacing routines to develop SWATmf-model as a standalone application in a Windows (Microsoft Inc., Redmond, WA) environment. In order to maintain the models' integrity, and independence, and to allow for future updates, the MODFLOW code was established as three modules that control the launch, execution, and closing of the simulation. The logic that controls the integrated model was incorporated in the SWAT code to trigger MODFLOW stress period (i.e., computational time intervals) simulations within the SWAT daily cycle (Fig. 3) and input/output transformations within the interfacing routines. Moreover, new data structures (i.e., lookup tables) were developed for two-
2.2. SWAT and MODFLOW interfacing routines
2.3. SWATmf application (SWATmf-app) The SWATmf-app works as a project manager, project builder, and model performance evaluator. As a project manager SWATmfapp allows simulation of single surface watershed (single SWAT project) contributing to a single aquifer system (single MODFLOW project) (i.e., coupled; Fig. 2a) or multiple surface watersheds (multiple SWAT projects) contributing to a single aquifer system (single MODFLOW project) (i.e., integrated; Fig. 4). In the coupled approach, the SWATmf-model propagates simulated data between SWAT and MODFLOW at daily time steps while in the integrated approach the SWATmf-model creates the MODFLOW recharge and well extraction packages without triggering MODFLOW execution, and then consolidated the various SWATmf-model outputs into a MODFLOW master project (Fig. 4). The SWATmf-app has the capability to setup MODFLOW
106
J.A. Guzman et al. / Environmental Modelling & Software 73 (2015) 103e116
arcSWAT HRU polygon domain ArcSWAT
SWAT
SPELLmap
• • • •
HDS me series HDS daily maps EXT me series Metrics
A
Interface
SWAT-MODFLOW domains indexa on (Lookup table)
C D
MODFLOW Raster Time series
A
B
E
B→13s→ 1m→ Areafrac B→14s→ 1m → Areafrac B→22s→ 1m → Areafrac B→23s→ 1m → Areafrac B→24s→ 1m → Areafrac B→31s→ 5m → Areafrac B→32s→ 5m → Areafrac A→33s→ 5m → Areafrac A→34s→ 5m → Areafrac A→41s→ 5m → Areafrac A→42s→ 5m → Areafrac A→43s→ 5m → Areafrac A→52s→ 5m → Areafrac
† SWATmf-model SWATmf-app 0s 1s
SWAT project
(SPELLmap)
9s
10s
• MODFLOW outputs (LIST, RCH-S, HDS, WEL-S) • SWAT outputs
HRU Feature class
xll,yll
19s
60s 70s
78 79
SWAT HRU grid domain • NAM, BAS, DIS, NWT, LIST, UPW, LPF, OC, RCH, WEL, DRN, CHD, GHB, RES • Log file • Update file handles
• Lookup tables • Contextual data
3m
1w
4m
SWATmf-model † xll,yll
MODFLOW project
1m
0m
7m
2w
4w
10m
11m
MODFLOW cells grid domain MODFLOW wells domain
(a)
SWAT-MODFLOW project
8m
3w
(b)
Fig. 2. A conceptual modelling framework for integrating SWAT and MODFLOW models a) model integration and flow of data, and b) unified spatial frame where xll and yll stand for the x or y left lower corner grid coordinates. Dotted lines in (a) represent run-time flow of data and solid lines represent batch processes.
SWATmf-model *.cio SWAT data
SWAT-code
1 MODFLOW-code launch
daily loop
yes coupled
yes
no Lookup tables Contextual data
1
no
simula on
2 • Setup NAM, OC • Log file • Update file handles
no
no
coupled yes
MODFLOW-code Stress period simula on
yes Day =Last
yes
2
Interfacing Rou nes
• Updated RCH • Updated WEL • Log file
SWAT-code
Interfacing Rou nes
BAS, DIS, NWT, LIST, UPW, LPF, OC, RCH, WEL, DRN, CHD, GHB, RES
HRUs percola on HRUs irriga on
Day =First
MODFLOW-code
BAS, DIS, NWT, LIST, UPW, LPF, OC, DRN, CHD, GHB, RES
close
coupled
no
SWAT-code close
END
Fig. 3. Integrated SWAT and MODFLOW models and interfacing routines indicating models calls during the SWAT daily cycle simulation and associated major model inputs/outputs.
packages such as the basic (BA6), discretization (DIS), Newton solver (NWT), and other optional packages (e.g., recharge packageeRCH, upstream weight packageeUPW, output control-OC, list file-LST, drainage-DRN, wells-WEL, time-variant specific headsCHD, flow and head boundary-FHB, and layer property flow packageeLPF). Additionally, SWATmf-app builds the lookup tables (spatial domain indexing) based on SWAT and MODFLOW project
definitions. Moreover, the SWATmf-app can directly edit the SWAT project and change parameters at the watershed, sub-basin and HRU level. As a model performance evaluator, the SWATmf-app has capabilities to extract SWATmf-model outputs and evaluate model performance (metrics) at specific points in both the surface and groundwater domains as well as extract gridded data from MODFLOW outputs (e.g., LIST file).
J.A. Guzman et al. / Environmental Modelling & Software 73 (2015) 103e116
ArcSWAT
SPELLmap
Raster Time series
• SWAT project
START
SWATmf-model slave 1
End
SWATmf-model slave 2
End
SWATmf-model slave 3
End
SWATmf-model slave n
End
107
yes
yes
yes
SWATmf-app (SPELLmap)
• NAM, BAS, DIS, NWT, LIST, UPW, LPF, OC, RCH, WEL, DRN, CHD, GHB, RES • Log file • Update file handles
SWATmf-app
• NAM, BAS, DIS, NWT, LIST, UPW, LPF, OC, RCH, WEL, DRN, CHD, GHB, RES • Log file • Update file handles
• Lookup tables • Contextual data
SWATmf-app project
yes
SWATmf-model slave projects
SWAT-MODFLOW master project
Fig. 4. Integrated SWATmf modelling framework with the capability to simulate multiple surface watersheds contributing to a single aquifer system.
3. Case study: Fort Cobb Reservoir experimental watershed 3.1. Study area The Fort Cobb Reservoir experimental watershed (780 km2) is located in southwestern Oklahoma, USA, and overlays a portion of Rush Spring aquifer (RSA) (Fig. 5a, b). The Fort Cobb Reservoir, built in 1958 (US Bureau of Reclamation Washita Basin project) for flood control, water supply, and recreation purposes, is located close to the watershed outlet (http://www.usbr.gov/projects/). Three major streams feed the reservoir: Willow Creek, Lake Creek, and Cobb Creek with USGS stream gauges located at the lower ends of each sub-watershed (Fig. 6b). The land use is predominantly mixed cropland and grazing land. The topography in the FCREW is hilly and elevations ranges from 380 to 560 m above mean sea level (Fig. 5c). Soils are heterogeneous and erosive with fine sandy loam in the eastern part, fine sandy loams and loamy soils in the north central and south central portions of the watershed, and silt loams in the western portion (Steiner et al., 2008, 2014). The RSA encompasses more than 6200 km2 of west-central Oklahoma (Fig. 5a) and is mainly an unconfined aquifer constituted of fine-grained cross-bedded sandstone with irregular dolomite or gypsum lenses (OWRB, 1965, 1966; Becker, 1998). Thickness of the aquifer ranges from approximately 75 to 90 m and is the main source of water for irrigation, domestic, and public supply in the FCREW. The Marlow formation (Fig. 5d) underlies the RSA and is a moderately to well-cemented unit acting as a boundary to vertical flow due to its low permeability (Becker, 1998). 3.2. Available data The SWAT model requires three GIS data layers, namely digital elevation model (DEM), soils, and land use data in addition to the weather data. Streamflow data are used to calibrate the model for hydrology. A 30 30 m DEM (USGS Seamless Data Distribution System (http://seamless.usgs.gov/viewer.htm) data were used in this study. The DEM was used to calculate sub-basin parameters, such as slope, slope length, and to define the stream network for
the SWAT model. The resulting stream network was used to define the layout of the sub-basins. The DEM was also used to obtain the stream network characteristics, such as channel slope, length, and width. The physical and chemical soils data required by SWAT were extracted from the Soil Survey Geographic (SSURGO; USDA, 1995). These data include soil hydrologic group, maximum rooting depth, soil profile depth, moist bulk density, available water capacity of the soil layer, saturated hydraulic conductivity, and soil texture data (% clay, sand, silt, and rock fragment content) that are required in streamflow computations. Also, these datasets serve to parameterize other SWAT submodels such as the universal soil loss equation (USLE; soil erodibility K factor) required to compute sediment yield. Chemical properties of soil include the fraction of porosity (void space) from which anions are excluded, organic carbon content (% soil weight), and initial concentrations of chemicals. A 30-m land use map from a study conducted in the area in 2005 (Starks et al., 2011) was used. In general, the land use is predominantly cropland (56%, about one-third is irrigated) followed by pasture and rangeland (33%), forest and shrub (5%), water (2%) and miscellaneous uses (4%), with low density homesteads, roads, and small communities. General crop management operations were taken from various crop guides, information provided by farmers, agronomists, animal scientists, and other farming specialists either in or familiar with the study area. Daily weather data (minimum and maximum air temperatures, precipitation, relative humidity, solar radiation, and wind speed) inputs required by the SWAT model were obtained a network of fifteen USDA ARS climate observation sites (MICRONET; Guzman et al., 2014; Starks et al., 2014) and three Oklahoma Mesonet stations (Brock et al., 1995; McPherson et al., 2007) (Fig. 5c). Daily stream flow from US Geological Survey sites (Fig. 6b; USGS, Water Science Center, OK; http://ok.water.usgs.gov/) at the Cobb, Lake and Willow Creeks (e.g., gauging sites 800, 850 and 860; Fig. 6b; Moriasi et al., 2014) were used to calibrate SWAT for streamflow. Core samples from USGS (Oklahoma water science center http:// ok.water.usgs.gov/) and reports from Becker (1998), Penderson (1999), and Magers (2011) were used to define the MODFLOW
108
J.A. Guzman et al. / Environmental Modelling & Software 73 (2015) 103e116
Fig. 5. Study area showing, a) Fort Cobb Reservoir experimental watershed (FCREW) location, b) Surface and groundwater watersheds indicating the USGS real time groundwater monitoring wells and projected aquifer extend on a surface digital elevation model, c) FCREW digital elevation model and MICRONET network, and d) AeA0 cross section indicating the major geological formations and SWAT and MODFLOW domains boundaries.
Fig. 6. SWAT model spatial discretization and monitoring sites, a) SWAT model derived watersheds indicating the centroids locations. Spatial location of a single HRU from the SWAT sub-basin 78 is indicated in blue navy colour, b) surface sub-watersheds streamflow monitoring sites, major surface drainage features, and location of the Fort Cobb reservoir, c) location of the automatic groundwater monitoring levels. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
groundwater domain discretization (e.g., layer geometry and boundary conditions) and hydrogeological parameterization (e.g., saturated hydraulic conductivity, specific yield and specific storage; Fig. 7) of cells in the aquifer domain. Daily measured water level elevation data from five collaborative USGS and Oklahoma Water
Resources Board groundwater monitoring wells (e.g., Alfalfa, Core2, Eakly, Hinton, and Gracemont; Figs. 5b and 6c; OWRB; http://www. owrb.ok.gov/) were used to estimate initial groundwater conditions and to evaluate the groundwater levels simulated by the coupled SWAT-MODFLOW model.
J.A. Guzman et al. / Environmental Modelling & Software 73 (2015) 103e116
4. Model conceptualization, parameterization, and calibration The areal extent of the study area (surface and groundwater) is described in Section 3.1 of this paper and covers an area of 780 km2 (Fig. 5b) overlaying a portion of the unconfined RSA, which encompasses more than 6200 km2 of west-central Oklahoma (Fig. 5a). In SWAT, the watershed is divided into sub-basins and HRUs using the ArcSWAT GIS interface. Sub-basin delineation should be detailed enough to capture significant variability within the watershed. In this study, the watershed was divided into 79 subbasins in which at the outlet points at three sub-basin there are USGS stream gauge stations (i.e., stations 7325800, 7325850, and 7325860) for the Cobb Creek, Lake Creek, and Willow Creek subbasins, respectively (Moriasi et al., 2014) or a reservoir on the stream channel (Fig. 6a, b). Each sub-basin was further sub-divided into HRUs. The multiple HRU method was used with threshold levels of zero and 5% for soils and land use, respectively, resulting in a total of 1001 HRUs (Fig. 7). These HRUs were represented in a 1193 x 1440 grid with 30 30 m cell resolution (889,980 active cells). The ArcSWAT interface was used to generate default parameter values from primary DEM, soil, and land use layers required for the model to run. Then available weather and land management data were used to parameterize the rest of the required parameters. Using the USDA-ARS Micronet and nearby Oklahoma Mesonet data (Guzman et al., 2014; Starks et al., 2014, Fig. 5c), daily precipitation
109
was estimated at each of the centroids of the 79 SWAT sub-basins using the optimum (minimum average error) inverse distance weighting interpolator (Chu et al., 2013b; Guzman et al., 2013). Values for the most sensitive streamflow parameters that are not easy to measure were determined through the calibration process. The maps required for the MODFLOW project were developed using SPELLmap (Guzman et al., 2013) (Fig. 6) and included the geological layer DEMs, the aquifer hydrogeological properties, setup of the well extraction network, the river drainage definition and properties. These maps were then integrated using the SWATmfapp to develop the MODFLOW project including time-series data linkage such as the Fort Cobb reservoir water level in the reservoir package. The groundwater model represents the unconfined RSA within the FCREW surface extent (SWAT model extent) based on two geological layers or strata and 300 300 m cell size. The large number of existing extraction wells in the FCREW (530 were identified) was a major consideration in defining the cell-size for groundwater simulation with MODFLOW. This spatial discretization resulted in 120 x 144 grid and 9171 active cells. Two unconfined layers were considered in the groundwater model. The top boundary was assumed to be six meters below the surface at the lower SWAT model boundary (Fig. 5d). The river network was represented in MODFLOW as a drainage system (Fig. 8e). Infiltration fluxes from the Fort Cobb reservoir were simulated using the MODFLOW reservoir package (Fig. 8f). Due to lack of data, horizontal non-flow boundary conditions were assumed for the
Fig. 7. Real world to model conceptualization illustrating the SWAT-MODFLOW linkage through lookup tables, a) HRUs to MODFLOW cells linkage, b) river network to HRUs linkage, c) HRUs to wells linkage. Image from Google maps.
110
J.A. Guzman et al. / Environmental Modelling & Software 73 (2015) 103e116
MODFLOW domain (Fig. 5b) with the exception of the southern part in which a drainage boundary condition was allowed. These conditions were set by allowing flow movement outside the aquifer based on the groundwater surface (distributed levels) gradient at this location. Data from several sources were processed using SPELLmap and used to parameterize the MODFLOW model (Fig. 8). The vertical hydraulic conductivities for the reservoir were determined from clay soil properties, sediment thickness from unpublished reports, and the reservoir bottom boundary from the DEM topography. Spatially distributed hydraulic conductivities reported in Penderson (1999) were initially used assuming isotropic conditions while the initial river network “conductance” values were assigned based on the river network soils properties. Data reported by and Becker (1998) were used to initialize specific storage, while the rest of the parameter values were estimated based on model inputs due to the lack measured data. Initial groundwater levels were estimated using spatial interpolation (inverse distance weighted) of data from five USGS groundwater wells one day prior to the first day of model simulations. The initial estimated groundwater surfaces was then cross-validated with the DEM to ensure that the interpolated surface was never over the DEM, and in cases where it was, the initial interpolated groundwater surface was aligned with the DEM. The final adjusted interpolated surface was then improved through an invariant MODFLOW simulation before used in the coupled model. Due to unavailability of well extraction records, irrigated
volumes were estimated using the SWAT auto-irrigation model. The SWAT auto-irrigation model was set-up based on the water plant demand (WSTRS_ID ¼ 1) by setting the water stress threshold (AUTO-WSTRS), which is defined as a fraction of potential plant growth (Neitsch et al., 2011). Although, AUTO_WSTRS varies between 0.00 and 1.00, it is usually set between 0.90 and 0.95, and in this study it was set at 0.90. When AUTO_WSTRS was below 0.90, auto-irrigation was activated. The amount of irrigation water applied each time the auto-irrigation was triggered (IRR_MX; Neitsch et al., 2011), was set at 25 mm for this study. Estimated irrigation needs were associated with the nearest extraction well using auto-generated Thiessen polygons (Fig. 7c; SPELLmap) and daily extraction volumes were then computed by SWATmf. Daily percolation fluxes (deep aquifer recharge) simulated by the calibrated SWAT model were spatially represented at the HRU level and propagated to the aquifer domain (e.g., MODFLOW recharge package) though the lookup tables generated by the SWATmf tool. Once the SWAT and MODFLOW model projects were developed and parameterized, they were coupled using SWATmf-app (Fig. 6). Recharge values were exchanged between SWAT and MODFLOW at the interface of the bottom boundary of SWAT (top MODFLOW boundary; Fig. 5d) during subsequent model simulations. All available data were used to parameterize respective models as described before. . Optimal values for most sensitive groundwater model parameters (e.g., hydraulic conductivity, specific storage, specific yield, and conductance's) were established through the calibration process.
Fig. 8. MODFLOW model conceptualization and initial parameterization, a) MODFLOW top boundary DEM, b) MODFLOW top layer and SWAT lower boundary; SWAT-MODFLOW interface boundary, c) Bottom MODFLOW layer boundary; Marlow formation, d) hydraulic conductivity from Penderson (1999), e) River network definition, and f) Reservoir definition.
J.A. Guzman et al. / Environmental Modelling & Software 73 (2015) 103e116
4.1. Model calibration First, SWAT model was calibrated for streamflow and then the coupled-SWAT-MODFLOW model was then calibrated for groundwater levels. The main goal for an initial SWAT calibration was to ensure that the surface model reasonably represented observed streamflow data and recharge estimates to the deep aquifer (Fig. 1), which is a critical input for MODFLOW. The parameters calibrated in SWAT for streamflow include the curve number (CN2), soil evaporation compensation coefficient (ESCO), aquifer percolation coefficient (RCH_DP), plant uptake compensation factor (EPCO), effective hydraulic conductivity in tributary channel alluvium (CH_K1), and surface runoff lag coefficient (SURLAG) which were the most sensitive parameters in SWAT for this study (Table 1). Although soil available water capacity (SOL_AWC) and soil saturated hydraulic conductivity (SOL_K; mm hr1) were also determined to be sensitive, they were not calibrated to avoid introducing biases from the SSURGO soil datasets used. RHC_DP describes the fraction of percolation from the root zone (deep aquifer recharge) and varies from 0.0 (no percolation) to 1.0 (all the water percolating from the root zone reaches the deep aquifer). The definition and ranges of values for the rest of these parameters are given in the SWAT theoretical documentation (Neitsch et al., 2011). For SWAT, manual streamflow calibration was conducted (January 2005 to December 2012; with 2005 used as a warm up period) by adjusting values of the calibration parameters within the recommended ranges (Neitsch et al., 2011) until the calibration objective functions described below were met; default values were used for the rest of the parameters. Because observed values were not available for each of the water balance components, the calibrations were also constrained such that the simulated ET and biomass values were realistic and representative of the study area in order to minimize the potential for obtaining good statistics for the wrong reasons (Kirchner, 2006). The biomass values were obtained by calibrating the biomass-energy ratio (BIO_E) ((kg/ha)/ (MJ/m2)). BIO_E is the amount of dry biomass produced per unit intercepted solar radiation in ambient CO2 and varies between 10 and 90, inclusive. The greater the BIO_E the greater the potential increase in total plant biomass on a given day (Neitsch et al., 2011). According to Hanson (1991), the mean actual annual ET of this region during the study period was about 88% of precipitation. A target range was set for ET values within 10% of the regions' mean annual ET. Biomass production ranges in metric tons (US tons) used in this study were: 1.8e2.7 (2.0e3.0) for cotton, 4.4e6.6 (4.9e7.3) for sorghum, 8.1e9.1 (8.9e10.0) for peanuts, 4e6 (4.4e6.6) for winter wheat, 3e7 (3.3e7.7) for pasture/grassland, and 5e10 (5.5e11.0) for forest (Starks and Moriasi, 2009; Moriasi and Starks, 2010). No other constraints were placed on the model during calibration. The calibrated values for the most sensitive SWAT flow parameters are presented in Table 1. There were no or minor differences for ESCO, EPCO, SURLAG, and to CN2 within the sub-watersheds. There were large CH_K1 values were obtained for Lake and Willow Creek sub-watersheds, in contrast to a very low value for Cobb Creek sub-watershed. These
111
higher apparent transmission losses were allowed in order to get simulated streamflow to fit measured streamflow in both Lake Creek and Willow Creek sub-watersheds. The calibrated RCH_DP values decreased from east to west gradient (Table 1). This means that the calibrated RCH_DP value for Willow Creek sub-watershed allowed for most recharge to the groundwater system. Whereas the SWAT model was calibrated for streamflow for the 2005e2012 period using a one year “warming up” period (i.e., 2005), the coupled SWAT-MODFLOW model was calibrated for groundwater levels for only 823 days (October 1, 2010 to December 31, 2012). This difference in calibration period was due to the lack of continuous groundwater records in the FCREW. The saturated hydraulic conductivities, specific yield, specific storage, and river network conductance parameters that affect the groundwater level dynamics and groundwater surface of the coupled model were improved through the calibration process. The calibrated values were determined by optimizing the Nash-Sutcliffe efficiency coefficient (NSEC; Nash and Sutcliffe, 1970) when comparing simulated and available observed groundwater levels on a daily time step. In addition, the river network was used to verify that the simulated groundwater surface did not exceed the river network cell elevations. Fig. 9 summarizes the calibrated parameters for the groundwater domain. Although a two layer definition was set for the MODFLOW discretization package, no differences in parameter values were important in neighbouring vertical cells but horizontal clustered cells. Specific yield varied from 0.02 102 to 3.2 102, specific storage from 0.08 107 to 1.00 107, while saturated hydraulic conductivity and river conductance were in the range of 0.02 102 to 3.0 102 and 0.00e0.50 m d1, respectively. Note that as there were not confining layers definition in the MODFLOW discretization package, and thus the specific storage parameter did not play an important role during simulations. Model performance, defined herein as the ability of a model to reproduce field observations during the calibration/validation period, was evaluated using both graphical comparisons and statistical measures (Moriasi et al., 2007). Time series graphs as well as NSEC were used. The threshold NSEC value used in this study for both streamflow and groundwater level simulations was NSEC >0.50 (Moriasi et al., 2007). 5. Results and discussion 5.1. SWATeMODFLOW coupled models Minimal changes in the SWAT and MODFLOW codes were necessary to couple these models. The use of the SWATmf takes advantage of the capabilities of the SWAT model to simulate daily distributed percolation fluxes (Fig. 10a) that are sensitive to land use and management while MODFLOW simulates groundwater heads in 3-dimensions including the effects of well extractions for irrigation (Fig. 10b). Return flow from aquifers (e.g., base flow) to the surface domain was not automatically accounted for within the modelling framework due to the lack of a surface river model in
Table 1 Calibrated parameter values obtained in Cobb Creek, Lake Creek, and Willow sub-watersheds. CN2 ¼ Curve number condition II, ESCO ¼ soil evaporation compensation coefficient, RCH_DP ¼ aquifer percolation coefficient, EPCO ¼ plant uptake compensation factor, CH_K1 ¼ effective hydraulic conductivity in tributary channel alluvium, and SURLAG ¼ surface runoff lag coefficient. Sub- watershed
Cobb Creek Lake Creek Willow Creek
Calibration parameter values CN2
ESCO
RCH_DP
EPCO
CH_K1 (mm hr1)
SURLAG
35e66 35e68 35e65
0.9 0.9 0.9
0.05 0.20 0.80
1.0 1.0 1.0
0.5 100.0 150.0
1 1 1
112
J.A. Guzman et al. / Environmental Modelling & Software 73 (2015) 103e116
Fig. 9. Calibrated groundwater parameters, a) MODFLOW hydraulic conductivities (102) for the two layers, b) Specific yield for the two layers (102), c) Specific storage for the two layers (107), and d) River conductance.
SWAT. Note that SWAT do not have the capability to model streamflow to the level of a river unit, instead, the SWAT model incorporate hydrological flow aggregation from HRUs to “river nodes” derived from the sub-basin spatial discretization in ArcSwat. This limitation constrained geolocated feedback fluxes from MODFLOW to SWAT and therefore future investigation is needed. However, spatially distributed daily percolation and groundwater extractions simulated by SWATmf provide modelers additional information to improve SWAT parameterization using the surface and groundwater model performance objective functions simultaneously. In addition, this integrated model framework provides a platform to integrate other models and to use additional geolocated data (e.g., irrigation wells, river network). For application purposes, it is recommended that modelers initially build SWAT and MODFLOW model projects separately before using the integrated/coupled SWATmf model. SWATmf is then calibrated to refine the most sensitive parameters using available measured flow data based on desired objective functions. Once calibrated, simulations are run and outputs of interest are extracted in grid or time series formats. Modelers need to evaluate the trade-offs between the spatial variability representation (e.g., SWAT HRUs and MODFLOW GRID resolutions, observations, and physical conceptualization) and model output accuracy, computational resources, and model response time. 5.2. Surface water model results Fig. 11 (a, b) summarizes the SWAT model calibration (January
2006 to December 2012) on a monthly basis at the three streamflow monitoring sites in the FCREW (e.g., 800-Cobb, 850-Lake and 860Willow Creeks; Fig. 6b). Computed Nash-Sutcliffe efficiency coefficient (NSEC) were 0.58, 0.69, and 0.55 at the Cobb creek, Lake Creek, and Willow creek stream gauges, respectively. Although the surface model was able to represent streamflow responses, low flow conditions were poorly represented (e.g., base flow) especially during the presence of prolonged dry periods (Fig. 11a). Note that Oklahoma experienced an exceptional wet year in 2007 followed by two wet years (2008 and 2009), and afterwards a dry period in 2010 followed by extreme drought in 2011 and 2012 (http://climate.ok. gov). The model was responsive to this climate variability as indicated in the cumulated monthly average streamflow (Fig. 11b; decrease in flow related to increase in time). One of the possible reasons for low streamflow underestimation was the unaccounted feedback fluxes from the deep aquifer in the SWAT model. This streamflow underestimation was more pronounced for Cobb Creek (Fig. 10b) because of the larger surface contact with the RSA (342 km2) compared to the Lake and Willow watersheds (154 and 75 km2, respectively), which resulted in larger feedback baseflow contributions. Also, soil characterizations in the Cobb Creek (higher clay content) were hypothesized to limit infiltration fluxes; thus limiting seepage flows to the sub-basins outlet (Fig. 11a). In Fig. 11b, departure from the perfect fitting (e.g., 1:1 line) indicates over or under prediction of cumulated streamflow. At the end of 2009, accumulated streamflow underestimations were evident and indicated as the negative departure from the perfect fit (Fig. 11b) resulting in a prolonged period with a negative water balance.
J.A. Guzman et al. / Environmental Modelling & Software 73 (2015) 103e116
113
Fig. 10. Coupled model distributed time-variant simulations, a) Simulated percolation fluxes (mm d-1; see colour scale) at the SWAT and MODFLOW domains interface (SWAT lower boundary and MODFLOW top layer) for the day 15 of each month during the year 2012, b) Five days simulated irrigation fluxes indicated as a 25 mm d1 per 30 30 m cell, c) SWAT model sub-basins identification; see Fig. 7 for location of extraction wells. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
5.3. Groundwater model results Fig. 12 summarizes groundwater level (aquifer domain) simulations by the integrated model. Cross sections in Fig. 12a indicate groundwater levels on 17 July 2011 while Fig. 12b represents the watershed surface (top) and aquifer (bottom) DEM. Fig. 12c shows the timeevariant comparison between simulated and observed groundwater levels at the Core2, Alfalfa, and Eakly monitoring wells. Predicted trends of groundwater levels were in agreement with observations and the computed NSEC were 0.87, 0.91, and 0.14 at Alfalfa, Core2 and Eakly, respectively (daily time step simulations). Although, NSEC for the Alfalfa and Core2 sites indicated good agreement with observations, simulations poorly replicated the timing and magnitude of aquifer extractions. For example, simulated groundwater level depletions started approximately at the end of June, 2011 with a lag time of few to several months when compared to observed groundwater depletions (Fig. 12c). Moreover, the rate of depletion (indicated by the slope of simulated groundwater levels in Fig. 10c) was in disagreement with observations at Core2 and Eakly sites. One possible reason was due to errors in the predicted extraction volumes obtained from the SWAT autoirrigation model, which was based on static land use for the simulation period. Moreover, disagreements between distributed percolating fluxes and simulated fluxes can explain the differences between observed and simulated groundwater levels between July 2011 and June 2012 at Core2 and Eakly sites. At the Alfalfa site, these disagreements were less pronounced due to its proximity to
the Fort Cobb reservoir controlling groundwater levels at the monitoring well (Fig. 12c). The coupled model provides MODFLOW with estimated well extractions from simulated irrigation needs (Fig. 10b) on daily basis. This integrated model capability is important when assessing the impact of groundwater extractions on agricultural lands without records of well extractions. However, improved model algorithms and collection of more data to parameterize the groundwater model domain are needed. The specific model development needs include revising the auto-irrigation model in SWAT to better represent irrigation and an improved algorithm to represent the areal irrigation to extraction volume from wells. Data needs include more field observations to support hydrogeological aquifer and conductance characterization. Furthermore, investigation on measured distributed recharge rates is needed to properly evaluate the distributed model simulations. 6. Summary and conclusions A comprehensive hydrologic modelling framework for integrating SWAT and MODFLOW models was conceptualized, developed and tested. Numerous routines for interfacing SWAT and MODFLOW were implemented to address differences in spatial discretization and representation of the two models so that minimal changes in the codes of the two models were necessary. The SWATmf was developed to work as a project builder and manager, and model performance evaluator. The SWATmf also assists in
J.A. Guzman et al. / Environmental Modelling & Software 73 (2015) 103e116
1 0.1 0.01 0.001
NSEC = 0.55
1
0.1
0.01
0.001
Jan-06 May-06 Sep-06 Jan-07 May-07 Sep-07 Jan-08 May-08 Sep-08 Jan-09 May-09 Sep-09 Jan-10 May-10 Sep-10 Jan-11 May-11 Sep-11 Jan-12 May-12 Sep-12
Willow (m3 s-1)
10
2012
2008
2006
0
60
50
100
150
200 2012
0
40 20
0 0 30
20
40
60 2012
NSEC = 0.69
Jan-06 May-06 Sep-06 Jan-07 May-07 Sep-07 Jan-08 May-08 Sep-08 Jan-09 May-09 Sep-09 Jan-10 May-10 Sep-10 Jan-11 May-11 Sep-11 Jan-12 May-12 Sep-12
Lake (m3 s-1)
10
50
2008
0.001
100
2008
0.01
150
2006
0.1
200
2006
1
Simulated (m3 x 1x106)
NSEC = 0.58
Simulated (m3 x 1x106)
b
10
Jan-06 May-06 Sep-06 Jan-07 May-07 Sep-07 Jan-08 May-08 Sep-08 Jan-09 May-09 Sep-09 Jan-10 May-10 Sep-10 Jan-11 May-11 Sep-11 Jan-12 May-12 Sep-12
Cobb (m3 s-1)
a
Simulated (m3 x 1x106)
114
20 10
0 0
10 20 30 Observed (m3 x 1x106)
Fig. 11. Observed (blue) and simulated (red streamflow, a) monthly streamflow at USGS sites 800-Cobb, 850-Lake, and 860-Willow Creeks outlets, b) cumulated flow; simulated and perfect fit. NSEC ¼ Nash Sutcliffe efficiency coefficient. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
processing and evaluating model outputs for measuring model performance. The developed framework utilizes spatially located HRUs from SWAT projects created using the ArcSWAT application along with grids, time series, and lookup tables to provide a unified spatio-temporal frame for SWAT and MODFLOW models. In addition, this application has capabilities to create the BAS, DIS, NWT, OC, LPF, UPW, RCH, RES, DRN, WEL, CHD, FHB and GHB packages in MODFLOW and the lookup tables for indexing HRUs from SWAT and MODFLOW cells, wells to HRUs, and time series data for wellextractions. SWAT was calibrated for streamflow at three stream gauges in the FCREW to check the surface water balance to ensure that the deep aquafer recharge into SWATmf was reasonable. The most sensitive parameters for the ET, surface runoff, aquifer percolation, and stream channel flow processes were calibrated. The model represented surface flow processes well with NSEC values of 0.55 or greater at the three gauging stations. The deep aquifer recharge values simulated by SWAT were then used as input for the SWATmf model. SWATmf was calibrated and tested with data from three groundwater monitoring wells within FCREW to assess its ability to run dynamically linked SWAT-MODFLOW models. The calibrated parameters included saturated hydraulic conductivities, specific yield, specific storage, and river network conductance parameters that affect the groundwater level dynamics of the coupled model. The SWATmf model was able to represent groundwater levels fairly well with a NSEC greater than 0.5 in all cases except at the Eakly
groundwater monitoring well. However, the integrated model was able to capture trends in the aquifer depletion as a function of seasonal climate variability. Areas that need further improvements include: modifications of SWAT auto-irrigation algorithm to better represent irrigation needs, addition of new algorithms to better spatially link irrigated areas associated with well extractions, integration of a hydrodynamic river model into SWAT, and incorporation of deep aquifer feedback fluxes into the SWAT model. Future field investigation is needed to improve hydrogeological characterization of the RSA, monitoring of well extraction volumes, and assess distributed percolating fluxes. In order to improve the integrated model performance, development of tools allowing unsupervised parameterization of HRUs, and a linkage with distributed evapotranspiration and a river model is fundamental. The integrated modelling framework is expected to improve watershed-scale model simulations and provide a modelling platform to better understand linked surface-subsurface hydrologic processes and associated transport phenomena under time-variant conditions. 7. Software availability The SWATmf-model and SWATmf-app were developed at the USDA Agricultural Research Service (ARS), Grazinglands Research Laboratory, El Reno, Oklahoma and were first released to the public in January 2013 as free software. Latest releases, documentation,
J.A. Guzman et al. / Environmental Modelling & Software 73 (2015) 103e116
115
Fig. 12. Observed and simulated groundwater levels, a) Simulated groundwater cross sectional levels on 17 July 2011, b) Topography (top) and groundwater (bottom) digital elevation models indicating the cross sectional profiles, and groundwater real-time sites locations, c) simulated and observed daily time series at Eakly, Core2 and Alfalfa groundwater monitoring wells.
and examples can be found in http://jguzman.info. The SWAT and MOFLOW-NWT codes were developed by USDA ARS Grassland, Soil and Water Research Laboratory, Temple, TX (http://swat.tamu.edu/ software/swat-model/), and the US Geological Survey (USGS) (http://water.usgs.gov/nrp/gwsoftware/modflow_nwt/ ModflowNwt.html), respectively. Both SPELLmap and SWATmf were developed in object Pascal Delphi (Embarcadero, San Francisco, CA1) for 32-bit or 64-bit Window (Microsoft Corporation) using RAD studio XE2. The coupled SWATmf-model was developed using the Intel visual FORTRAN composer XE 2011 compiler version 12.1 (Intel Corporation, Santa Clara, CA1) under Visual Studio 2010 (Microsoft Corporation, Redmond, WA1). Developer: Jorge A. GUZMAN USDA Agricultural Research Service Grazinglands Research Laboratory, El Reno, OK, 73036 e-mail:
[email protected] Phone: þ1-(405)-262-5291 ext. 268; FAX: þ1-(405)-262-0133] Required software: Windows First year available: 2013 Availability: Download available under a USDA-ARS material transfer agreement from the website: http://jguzman.info
1 Reference to any non-government products in this paper does not constitute an endorsement by the United States Department of Agriculture (USDA).The USDA does not affiliate itself with, nor does it warranty in any way the respective software or files created by these applications.
References
Acknowledgements Development of the SWAT-MODFLOW model framework was funded by U.S. Department of Agriculture and the Agricultural Research Service. The authors would like to thank Shana Mashburn and Mark Becker from the U.S. Geological Survey, and Christopher R. Neel and Jessica S. Magers from the Oklahoma Water Resource Board for data provided, access to core samples and general overview of the study area. We are also grateful to Alan Verser for his contribution to the SWAT model setup.
Arnold, J.G., Srinivasan, R., Muttiah, R.S., William, J.R., 1998. Large area hydrologic modeler and assessment part I: model development. J. Am. Water Resour. As.
116
J.A. Guzman et al. / Environmental Modelling & Software 73 (2015) 103e116
34, 73e89. Arnold, T.R., 2013. Procedural knowledge for integrated modelling: towards the modelling playground. Environ. Modell. Softw. 39, 135e148. Beeson, P.C., Martens, S.N., Breshears, D.D., 2001. Simulating overland flow following wildfire: mapping vulnerability to landscape disturbance. Hydrol. Process. 15, 2917e2930. Becker, M., 1998. Steady-state Simulation of Ground-water Flow in the Rush Spring Aquifer, Western Oklahoma. Water-resources Investigations Report 98-4082. USGS, Denver, CO. Betrie, G.D., van Griensven, A., Mohamed, Y.A., Popescu, I., Mynett, A.E., Hummel, S., 2011. Linking SWAT and SOBEK using open modeling interface (OpenMI) for sediment transport simulation in the blue nile river basin. Trans. ASABE 54 (5), 1749e1757. Brock, F.V., Crawford, K.C., Elliott, R.L., Cuperus, G.W., Stadler, S.J., Johnson, H.L., Eilts, M.D., 1995. The Oklahoma Mesonet: a technical overview. J. Atmos. Ocean. Technol. 12, 5e19. Brown, M.E., Funk, C.C., 2008. Food security under climate change. Science 319, 580e581. Chu, M.L., Knouft, J.H., Ghulam, A., Guzman, J.A., Pan, Z., 2013a. Impacts of urbanization on river flow frequency: a controlled experimental modeling-based evaluation approach. J. Hydrol 495, 1e12. Chu, M.L., Ghulam, A., Knouft, J.H., Pan, Z., 2013b. Exploring the trends and shifts in rainfall and runoff patterns: a hydrologic data screening procedure. J. Am. Water Resour. Assoc. 50, 928e942. Conan, C., Bouraoui, F., Turpin, N., de Marsily, G., Bidoglio, G., 2002. Modeling flow and Nitrate Fate at Catchment scale in Brittany (France). J. Environ. Qual. 32 (6), 2026e2032. Garen, D., Moore, D.S., 2005. Curve number hydrology in water quality modeling: use, abuses, and future directions. J. Am. Water Resour. As. 41 (2), 377e388. Gregersen, J.B., Gijsbers, P.J.A., Westen, S.J.P., 2007. OpenMI: open modelling interface. J. Hydroinformatics 9, 175e191. Guzman, J.A., Moriasi, D.N., Chu, M.L., Starks, P.J., Steiner, J.L., Gowda, P.H., 2013. A tool for mapping and spatio-temporal analysis of hydrological data. Environ. Modell. Softw. 48, 163e170. Guzman, J.A., Chu, M.L., Starks, P.J., Moriasi, D.N., Steiner, J.L., Fiebrich, C.A., McCombs, A.G., 2014. Upper Washita River experimental watersheds: data screening procedure for data quality assurance. J. Environ. Qual. 43 (4), 1260e1261. http://dx.doi.org/10.2134/jeq2013.08.0325. Hanson, R.L., 1991. Evapotranspiration and droughts. In: Paulson, R.W., et al. (Eds.), National Water Summary 1988-1989, Hydrologic Events and Floods and Droughts. 99-104. USGS Water-Supply Paper 2375. Kalaugher, E., Bornman, J.F., Clark, A., Beukes, P., 2013. An integrated biophysical and socio-economic framework for analysis of climate change adaptation strategies: the case of a New Zealand dairy farming system. Environ. Modell. Softw. 39, 176e187. Kirchner, J.W., 2006. Getting the right answers for the right reasons: Linking measurements, analyses, and models to advance the science of hydrology. Water Resour. Res. 42, W03S04. http://dx.doi.org/10.1029/2005WR004362. Kim, N.W., Chung, I.M., Won, Y.S., Arnold, J.G., 2008. Development and application of the integrated SWATeMODFLOW model. J. Hydrol. 356, 1e16. Kim, N.W., Wong, Y.S., Lee, J., Lee, J.E., Jeong, J., 2011. Hydrological impacts of urban imperviousness in White Rock Creek watershed. Trans. ASABE 54 (5), 1759e1771. Laniak, G.F., Olchin, G., Goodall, J., Voinov, A., Hill, M., Glynn, P., Whelan, G., Geller, G., Quinn, N., Blind, M., Peckham, S., Reaney, S., Gaber, N., Kennedy, R., Hughes, A., 2013. Integrated environmental modeling: a vision and roadmap for future. Environ. Modell. Softw. 39, 3e23. Magers, J.S., 2011. Occurrence of Arsenic in the Rush Spring Aquifer Sandstone and its Implications on Groundwater Chemistry: Caddo County, Oklahoma. Master thesis. Oklahoma State University, OK. McDonald, M.G., Harbaugh, A.W., 1988. A Modular Three-dimensional Finite-difference Ground-water Flow Model. Techniques of Water-resources Investigations of US Geological Survey. Book 6 CH. A1, Reston, Virginia. McPherson, R.A., Fiebrich, C., Crawford, K.C., Elliott, R.L., Kilby, J.R., Grimsley, D.L., Martinez, J.E., Basara, J.B., Illston, B.G., Morris, D.A., Kloesel, K.A., Stadler, S.J., Melvin, A.D., Sutherland, A.J., Shrivastava, H., 2007. Statewide monitoring of the mesoscale environment: a technical update on the Oklahoma Mesonet. J. Atmos. Ocean. Tech. 24, 301e321.
Moriasi, D., Arnold, J., Van Liew, M., Bingner, R., Harmel, R., Veith, T., 2007. Model evaluation guidelines for systematic quantification of accuracy in watershed simulations. Trans. ASABE 50 (3), 885e900. Moriasi, D.N., Starks, P.J., 2010. Effects of soil dataset resolution on SWAT2005 streamflow calibration parameters and simulation accuracy. J. Soil Water Conserv. 65 (2), 63e78. Moriasi, D.N., Starks, P.J., Guzman, J.A., Garbrecht, J.D., Steiner, J.L., Stoner, J.C., Allen, P.B., Naney, J.W., 2014. Upper washita river experimental watersheds: reservoir, groundwater, and streamflow data. J. Environ. Qual. 43, 1262e1272. Neitsch, S.L., Arnold, J.G., Kiniry, J.R., Williams, J.R., 2011. Soil and Water Assessment Tool Theoretical Documentation Version 2009. Texas Water Resources Institute Technical Report No. 406. College Station, Texas. Niswonger, R.G., Panday, Sorab, Ibaraki, Motomu, 2011. MODFLOW-NWT, a Newton Formulation for MODFLOW-2005: U.S. Geological Survey Techniques and Methods, p. 44, 6eA37. OWRB., 1965. Ground Water in the Rush Spring Sandstone Caddo County Area. Oklahoma Water Resource Board, Oklahoma City, OK, p. 1965. Publication No 11. OWRB., 1966. Ground Water in the Rush Spring Sandstone Caddo County Area. Oklahoma Water Resource Board, Oklahoma City, OK, p. 1966. Publication No 15. PCRaster, 2008. PCRaster internet site. URL. http://pcraster.geo.uu.nl/. Penderson, R.L., 1999. St'eady-state Simulation of Ground-water Flow in the Rush Spring Aquifer, Cobb Creek Basin, Caddo County, Oklahoma. Master thesis. Oklahoma State University. Playan, E., Mateos, L., 2006. Modernization and optimization of irrigation systems to increase water productivity. Agr. Water Manage. 80, 100e116. Ryu, J., Cho, J., Kim, I.J., Mun, Y., Moon, J.P., Kim, N.W., Kim, S.J., Kong, D.S., Lim, K.Y., 2011. Enhancement of SWAT-REMM to simulate reduction of total nitrogen with riparian buffer. Trans. ASABE 54 (5), 1791e1798. Scanlon, B.R., Reedy, R.C., Stonestrom, D.A., Prudic, D.E., Dennehy, K.F., 2005. Impact of land use and land cover change on groundwater recharge and quality in the southwestern US. Glob. Change Biol. 11, 1577e1593. Schmitz, O., Karssenberg, D., van Deursen, W.P.A., Wesseling, C.G., 2009. Linking external components to a spatio-temporal modelling framework: coupling MODFLOW and PCRaster. Environ. Modell. Softw. 24, 1088e1099. Sophocleous, M.A., Koelliker, J.K., Govindaraju, R.S., Birdie, T., Ramireddygari, S.R., Perkins, S.P., 1999. Integrated numerical modeling for basin-wide water management: the case of the Rattlesnake Creek basin in south-central Kansas. J. Hydrol. 214, 179e196. Sophocleous, M., Perkins, S.P., 2000. Methodology and application of combined watershed and ground-water models in Kansas. J. Hydrol. 236, 185e201. Starks, P.J., Moriasi, D.N., 2009. Spatial resolution effect of precipitation data on SWAT calibration and performance: implications for CEAP. Trans. ASABE 52 (4), 1171e1180. Starks, P.J., Rao, M., Siewe, S., 2011. Land-cover analysis for the Fort Cobb Reservoir watershed, southwestern Oklahoma, chapter 5. In: Becker, C.J. (Ed.), Assessment of Conservation Practices in the Fort Cobb Watershed, Southwestern Oklahoma, Unites States Geological Survey Scientific Investigations Report 2010-5257, p. 8. Starks, P.J., Fiebrich, C.A., Grimsley, D.L., Garbrecht, J.D., Steiner, J.L., Guzman, J.A., Moriasi, D.N., 2014. Upper Washita River experimental watersheds: meteorologic and soil climate measurement networks. J. Environ. Qual. 43 (4), 1239e1249. http://dx.doi.org/10.2134/jeq2013.08.0312. Steiner, J.L., Starks, P.J., Daniel, J.A., Garbrecht, J.D., Moriasi, D., McIntyre, S., Chen, S., 2008. Environmental effects of agricultural conservation: a framework for research in two watershed in Oklahoma's upper Washita river basin. J. Soil Water Conserv. 63, 443e452. Steiner, J.L., Starks, P.J., Garbrecht, J.D., Moriasi, D.N., Zhang, X., Schneider, J.M., Guzman, J.A., Osei, E., 2014. Long-term environmental research: the upper washita river experimental watersheds, Oklahoma, USA. J. Environ. Qual. 43 (4), 1227e1238. http://dx.doi.org/10.2134/jeq2014.05.0229. USDA., 1995. Natural Resources Conservation Services (NRCS). Soil Survey Geographic (SSURGO) Data Base: Data Use Information. National Coil Survey Center, Lincoln, NE. Walter, M.T., Shaw, S.B., 2005. Curve number hydrology in water quality modeling: use, abuses, and future directions” by David C. Garen and Daniel S. Moore. J. Am. Water Resour. As. 41 (6), 1491e1492. Discussion. Wheeler, T., von Braun, J., 2013. Climate change impacts on global food security. Science 341, 508e513.
