Food supply is affected by a complex nexus of land, atmosphere, and human processes, including short- and long-term stressors (e.g., drought and climate change, respectively). A simulation platform that captures these complex elements can be used to inform policy and best management practices to promote sustainable agriculture. We have developed a tightly coupled framework using the macroscale variable infiltration capacity (VIC) hydrologic model and the CropSyst agricultural model. A mechanistic irrigation module was also developed for inclusion in this framework. Because VIC–CropSyst combines two widely used and mechanistic models (for crop phenology, growth, management, and macroscale hydrology), it can provide realistic and hydrologically consistent simulations of water availability, crop water requirements for irrigation, and agricultural productivity for both irrigated and dryland systems. This allows VIC–CropSyst to provide managers and decision makers with reliable information on regional water stresses and their impacts on food production. Additionally, VIC–CropSyst is being used in conjunction with socioeconomic models, river system models, and atmospheric models to simulate feedback processes between regional water availability, agricultural water management decisions, and land–atmosphere interactions. The performance of VIC–CropSyst was evaluated on both regional (over the US Pacific Northwest) and point scales. Point-scale evaluation involved using two flux tower sites located in agricultural fields in the US (Nebraska and Illinois). The agreement between recorded and simulated evapotranspiration (ET), applied irrigation water, soil moisture, leaf area index (LAI), and yield indicated that, although the model is intended to work on regional scales, it also captures field-scale processes in agricultural areas.
Impacts of climate change on crop yield, as discussed by Kurukulasuriya and Rosenthal (2003), Leakey et al. (2009), Reilly (2002), Rosenzweig et al. (2001) and Rowan et al. (2011).
Projected increases in food demand (Godfray et al., 2010) along with other stressors such as droughts and extreme heat events contribute to threats to global food supply (Wheeler and von Braun, 2013). Despite existing research on food scarcity, there are still unanswered questions about the relationship between food supply and the nexus of water resources, agriculture, and human decisions. For example, how expectations of future climatic conditions influence farmer behaviour, such as capital-intensive switches in technology or cropping systems, is not well understood. Such scenarios require a simulation tool that can capture large-scale hydrologic processes while accurately simulating the impacts of climate, management, and water availability on different crop types. Moreover, regional consequences of decisions intended to mitigate the damages of future stressors are not well understood (Robertson and Swinton, 2005). For example, improvement in the efficiency of irrigation systems may increase consumptive water use and lead to a reduction in return flow from irrigated areas (Causapé et al., 2004; Gosain et al., 2005). Return flow plays a significant role in the water availability of many agricultural regions; e.g., 40 % of the water availability at the Yakima River's Parker Gauge in an average year is generated through return flows from upstream lands (USBR, 2010). Ecosystems and hydroelectric generation are also impacted as return flow changes. These knowledge gaps limit our ability to explore viable adaptation strategies, particularly in understanding unintended consequences. Integrated modeling platforms can contribute to the systems-level understanding of dynamics between agricultural processes, large-scale water resource management decisions, and land–atmospheric interactions.
The overall goal of this study is to develop a computational modeling platform that mechanistically captures the interactions between hydrology, crop growth and phenology, and crop and water resource management decisions in the context of global change. Such a platform allows for investigation around multiple objectives: (1) understanding how climate dynamics and land–atmosphere interactions affect water and agricultural sustainability, and (conversely) (2) exploring the role of agricultural (biophysical and socioeconomic) processes in driving land–atmosphere interactions, including climate feedback mechanisms on larger scales.
While over 800 million people throughout the world suffer from undernourishment (FAO, 2013), global change is expected to exacerbate food security problems. The demand for food is increasing due to population growth and changes in food dietary tendency towards higher consumption of meat products (Long et al., 2015). Food supply, however, may not increase as fast as demand (Wheeler and von Braun, 2013), as it is affected by complicated interactions between climate, the hydrologic cycle, cropping systems, and human decisions. Table 1 shows the variety of ways that climate change can impact crop yield, with some impacts being positive and others negative; the net result is dependent on region, crop, and future time period. Mechanistic integrated modeling platforms are necessary to assess the net impact of global change on crop production.
Although agricultural productivity is affected by disturbances in the regional cycles of water and energy (Pielke Sr. et al., 2007), agriculture itself feeds back to alter the hydrological cycle by changing evapotranspiration (ET) and the magnitude and temporal regime of soil moisture, infiltration, and runoff generation (Haddeland et al., 2006; Harding et al., 2015; Lu et al., 2015; Sorooshian et al., 2012). The impact of irrigated agriculture on energy and water cycles is particularly important (Ferguson and Maxwell, 2011; Lobell et al., 2009; Pokhrel et al., 2016; Puma and Cook, 2010; Scanlon et al., 2007; Sridhar, 2013). Irrigation uses 70 % of total global water withdrawals (Rost et al., 2008) and boosts soil moisture storage available for crop uptake, and ultimately increases ET. Irrigation losses also increase the amount of deep percolation and runoff (Malek et al., 2017).
While farmers can adjust their management decisions to reduce the negative impacts of climate change (e.g., switching to more efficient irrigation technologies, planting more drought-tolerant crop types or varieties with longer growing periods, and implementing precision agriculture), these human decisions can result in unintended impacts on regional water and energy cycles. The consequences of anthropogenic disturbances (e.g., irrigation withdrawal and dam construction) on the regional water cycle can be greater than the impacts of climate change (Haddeland et al., 2014). Irrigation management and changes in cropping patterns are two examples of management decisions influencing the amount of ET, runoff, deep percolation, and soil moisture, all of which can alter timing and magnitude of return flow. In many agricultural basins, the availability of water for downstream users depends greatly on the return flow from upstream lands, which mainly comes from nonevaporative, reusable loss of water through conveyance systems and field-level application of irrigation water. Therefore, regional-scale simulation of the hydrologic cycle is crucial to the analysis of the impacts of water management in large river basins with significant agricultural activities.
VIC–CropSyst provides an advantage over the stand-alone CropSyst model when run over larger scales. Here, we define large-scale results as regionally aggregated responses of agriculture to changes that can impact scales greater than a single cultivated field, such as a policy change (e.g., water law), climate-related impacts (e.g., warming-induced reductions in summer water availability), or development of large-scale infrastructure (e.g., a large reservoir). Allen et al. (2015) interviewed around 20 stakeholders, including governmental and nongovernmental agency staff and producers, to understand their priorities, concerns, and decision-making processes. They found that many of these stakeholders, including individual producers, are interested in local- and basin-scale information about the impacts of climate change, infrastructural developments, and land management practices on the quantity, quality, and temporal regimes of water resources. Therefore, large-scale integrated modeling platforms are also needed to inform regional natural and agricultural resource management policies and actions.
Land surface models (LSMs) are used for regional- to global-scale simulations
of water and energy cycles, often providing terrestrial boundary conditions
to general circulation models (GCMs). Results of modeling studies have
indicated that, despite the tremendous advances in Earth system modeling,
LSMs in their current state are not capable of capturing agricultural processes
in a detailed manner (e.g., Chang et al., 2014; Haddeland et al., 2006; Hansen et al., 2006; Lobell et
al., 2008, 2009; Ozdogan et al., 2010). In many of them, agricultural
processes are similar to natural vegetation (Chang et al.,
2014); due to phenological similarities, agricultural lands are often
represented by grass vegetation (Elliott et
al., 2014). Also, management or harvesting activities as well as CO
Bierkens (2015) reviewed 23 global or large-scale hydrological models (GHMs; e.g., WaterGAP, Verzano et al., 2012; WBMPlus, Wisser et al., 2010; Mac-PDM.09, Gosling and Arnell, 2011; and H08, Hanasaki et al., 2010), LSMs (e.g., VIC, Liang et al., 1994; MATSIRO, Takata et al., 2003; LM3, Milly et al., 2014; NOAH, Liu et al., 2016; JULES, Best et al., 2011; CLM, Fisher et al., 2015; SiB, Baker et al., 2008; and ORCHIDEE, Vérant et al., 2004), and dynamic vegetation models (DVMs; e.g., LPJmL: Biemans et al., 2011; Jägermeyr et al., 2015; Rost et al., 2008). Among these models, H08, MATSIRO and SiB use simple crop growth modules to simulate natural vegetation or generic C3 and/or C4 crops. NOAH, CLM, and LPJmL have more sophisticated crop growth schemes; these are further discussed below.
Using prescribed seasonally and spatially variable leaf area index (LAI) and
root density, Wei et al. (2013) modified aerodynamic and soil deficit
thresholds in the NOAH land surface model, thereby improving the simulation
of warm-season processes. In their model, however, crop growth and
development do not mechanistically respond to climate, CO
Drewniak et al. (2013) enhanced the Community Land Model (CLM) in
agricultural areas by using an improved representation of crop processes,
but CO
Elliott et al. (2014) compared 10 GHMs and 6 global gridded crop models (GGCMs);
they reported that the performance of GHMs is generally poor in the
simulation of future irrigation water demand. Many of them use prescribed
crop growth parameters and did not capture CO
Irrigation is one of the important but underappreciated processes in LSMs (Gordon
et al., 2008; Ozdogan et al., 2010; Pokhrel et al., 2016). Normally,
irrigation processes are treated in LSMs with one of the following
approaches.
Irrigation time and amount are not mechanistically simulated: in most modeling studies, irrigation
requirements are calculated using published irrigation guidelines or a time
series of satellite observations (Pokhrel et al., 2011). In other
models, irrigation water scarcity is not captured (e.g.,
Ozdogan et al., 2010), which can result in less realistic irrigation
management during droughts. Irrigation is included but with unrealistic assumptions of irrigation efficiency: for example, CLM v4
simulates the time of irrigation based on soil deficit but does not consider
irrigation losses (Leng et al., 2013). This
can cause poor representation of hydrologic processes in agricultural areas
and underestimation of irrigation demand. Partitioning of
overall efficiency into different losses through prescribed ratios: Pokhrel et al. (2011) developed
an irrigation module and coupled it to the Minimal Advanced Treatments of
Surface Interaction and Runoff (MATSIRO) model. The irrigation module
considers soil moisture deficit to calculate the time of irrigation, but
their irrigation module did not consider the partitioning of the overall
efficiency into different losses and did not simulate the dynamics between
irrigation losses and the hydrologic cycle. Haddeland et al. (2006) implemented a
simple irrigation module into the VIC model. This irrigation module,
however, was limited to prescribed losses of sprinkler systems. Also,
because the stand-alone VIC model does not mechanistically simulate crop
processes, the timing and amount of irrigation is not responsive to crop
growth, management, and phenology.
These shortcomings, simplifying assumptions, and the lack of a mechanistic way to simulate irrigation processes in LSMs lead to inaccurate ET and water demand simulations (Pokhrel et al., 2011; Sridhar, 2013). Also, because LSMs are often coupled to atmospheric models, this lack of captured mechanistic irrigation processes will cause biases in turbulent heat flux simulations, leading to GCM errors.
Here, we introduce the newly integrated model VIC–CropSyst, which is a coupling between the VIC hydrologic model and the CropSyst crop growth, phenology, and management model. VIC–CropSyst can be used for regional- to global-scale simulations of water and energy cycles over natural and managed terrestrial ecosystems. A process-based irrigation module was also developed to simulate the interactions between irrigation management decisions and the hydrologic cycle in this integrated model (see Malek et al., 2017a, b for further information on the irrigation module).
The VIC model is a process-based large-scale hydrologic model developed
initially by Liang et al. (1994). VIC uses the
variable infiltration capacity curve introduced by Zhao et al. (1980) to
simulate infiltration and surface runoff, and Franchini and
Pacciani's (1991) formula to calculate base flow.
Liang et al. (1996) further developed the model to
represent multiple soil moisture layers (the original version only had two).
Cherkauer et al. (2003) added additional
sophistication for cold-season processes; further information on the
simulation of the snowpack can be found in Andreadis et al. (2009). While
the simulation time step of the stand-alone VIC model can be specified to be
daily, hourly, or subdaily (e.g., 3 h), in the version of
VIC–CropSyst described herein, the simulation time step is currently limited
to daily time steps. Subsequent VIC–CropSyst model developments will allow
for subdaily time steps. VIC also has the flexibility to be implemented
over multiple resolutions (generally at or greater than 1
CropSyst (Stöckle et al., 1994, 2003) is a process-based
cropping system model, capturing water, nitrogen, and carbon cycles as well
as the key processes related to crop phenology, root and shoot growth, and
biomass production and yield. CropSyst simulates field operations including
irrigation, fertilization, tillage, residue management, and crop rotation. It
also captures the effects of CO
This schematic shows how VIC and CropSyst are coupled. VIC provides the availability of water and energy to CropSyst. CropSyst uses this information to grow the crop, produce biomass and yield, and simulate transpiration. CropSyst passes back the information that is needed by VIC (e.g., the distribution of transpiration uptake in different soil layers, LAI, and root depth) to simulate the hydrologic and energy cycle and the scheduling of irrigation.
We coupled the VIC version 4.1.2-e with CropSyst-v4.15, although the coupled model will be updated with new versions of VIC and CropSyst as they become available. In a spatially explicit manner, VIC–CropSyst is able to capture a large variety of crop groups: (1) cereal grains (e.g., winter and spring wheat, corn, barley, oats, and sorghum), (2) vegetables and melons (e.g., dill, radish, mint, broccoli, cauliflower, cabbage, carrot, onion, cucumber, pumpkins, and watermelon), (3) fruits and nuts (e.g., plum, apricot, cherry, grape, walnut, pear, peaches, apples, blubbery, strawberry, and cranberry), (4) root crops (e.g., potato and sugar beet), (5) leguminous crops (e.g., green and dry bean, lentil, chickpea, and pea), (6) forages (e.g., pasture, alfalfa, hay, grass, clover, and grass), and (7) oil seeds (e.g., soybean, mustard, and sunflower). In the tightly coupled VIC–CropSyst model (Fig. 1), all hydrologic processes except for transpiration are handled by VIC, while crop growth, transpiration, phenology, and management are handled by CropSyst. In the following section we explain the structure of the VIC–CropSyst coupling (Fig. 1). Then we discuss some of the changes we have made to each model to support this integration. Finally, we briefly discuss the irrigation module that we have developed and implemented in VIC–CropSyst; full details on the irrigation module can be found in Malek et al. (2017a, b).
Figure 2 shows how VIC–CropSyst handles the water
and energy budgets. VIC first simulates the energy balance (explained by
Cherkauer et al., 2003 and Liang et al., 1994). It estimates available energy per time
step and uses an iterative approach to partition the available energy into
each of the energy components (e.g., snowmelt and sublimation heat fluxes,
ground heat flux, and sensible heat flux). After these terms are calculated,
the remaining energy will be available to potential evapotranspiration (ET
CropSyst is called while VIC is simulating the energy balance, but after ETp is portioned into each of its terms. Following this, potential transpiration and availability of soil moisture are passed to CropSyst (Fig. 2). Actual transpiration depends on the availability of soil water. When the soil does not have enough water to meet crop demand, actual crop transpiration is less than potential. In the coupled model, CropSyst simulates actual transpiration, soil water extraction from each layer, water stress, and crop growth; it then passes the extracted soil water amount to VIC to calculate the water balance. VIC updates soil moisture and simulates the rest of the hydrologic components such as runoff and baseflow.
Algorithm used in VIC–CropSyst to partition available energy into different evaporative components. The energy and water balances are handled by the VIC model. CropSyst receives the amount of energy available for transpiration and the availability of water in the soil to determine crop water uptake. VIC needs actual transpiration in different layers of the soil to close the water cycle. Communication between the two models happens for every time step.
Pathways of irrigation water loss simulated in the irrigation
module.
Soil hydrology: in the integrated VIC–CropSyst model,
CropSyst's soil hydrology is turned off, allowing VIC to simulate soil
hydrologic processes, including the movement of water in soil, bare soil
evaporation, and the generation of runoff and baseflow. We did this to
retain consistency in all of the hydrologic processes. Stand-alone VIC and
CropSyst use different soil hydrologic assumptions to simulate processes
related to soil water movement and the generation of runoff and baseflow;
these inconsistencies can lead to an inaccurate simulation of irrigation
demand and crop productivity. Because crop processes are sensitive to soil
moisture availability, we have modified the VIC soil structure. While VIC
previously had the capacity to handle an indefinite number of soil moisture
layers, the majority of VIC applications utilize three layers, where runoff
and baseflow are generated from the top and bottom layers, respectively,
while the middle layer is the root zone where plant water uptake occurs.
Because the availability of water where roots are concentrated is central to
unstressed crop growth, and because the dynamic simulation of root growth is
sensitive to the vertical distribution of soil moisture, VIC's conventional
three-layer system is too coarse to accurately represent this condition,
particularly during droughts and over rain-fed cropland. Therefore, we
expanded the middle layer of VIC to 15 layers. Finally, the minimum soil
moisture in VIC–CropSyst is set to the wilting point (except in the top
evaporative layer where soil evaporation can result in soil moisture levels
below wilting point). Soil file: the conventional versions of VIC
directly read soil properties (e.g., soil hydraulic conductivity, field
capacity, wilting point, and bulk density) from input files. For a more
consistent way (between VIC and CropSyst) of inputting soil input
information, empirical functions developed by Saxton et
al. (1986) were implemented in the model and VIC–CropSyst internally
estimates the necessary soil parameters using soil textural characteristics
(i.e., sand and clay percentages).
The irrigation module (Fig. 3) is briefly explained below, while a more detailed description can be found in Malek et al. (2017a, b). The irrigation module calculates irrigation frequency, amount, and losses.
Currently, VIC–CropSyst simulates four major categories of irrigation systems: surface, center pivot, sprinkler, and drip. Each category includes subcategories. Drip systems include surface and subsurface drip irrigation. In surface drip irrigation, water is applied on the soil surface, while in subsurface drip irrigation, water is applied below the surface and will not lead to any soil evaporative losses. Surface irrigation includes furrow, rill, and border irrigation, and the main difference between these three systems is in their wetted surface area, which is smaller in a furrow system. Center pivots are represented by 18 different types of sprinklers that fall into two subcategories: impact and spray sprinklers. Impact sprinklers generally have a greater discharge rate and wetted radius. Sprinkler systems in VIC–CropSyst include 17 nozzles from three major subcategories: solid set, big gun, and moving wheels. The subcategories differ in terms of discharge, wetted diameter, height, droplet size, and other aspects. The characteristics of these systems have been collected from different scientific papers, reports, and commercial catalogs, including Nelson Co. (2014) and RainBird (2014). This level of detail offers a more accurate representation of irrigation practices, and it will help users to simulate the adaptation of different irrigation and management scenarios.
Evaporation, transpiration, and deep percolation cause reductions in root-zone soil water content. When soil moisture deficit reaches one of the following two thresholds, VIC–CropSyst triggers an irrigation event: (1) capacity of the irrigation system, which sets the maximum amount of water that can be applied in an irrigation event, and (2) the maximum allowable depletion (MAD), which determines what degree of soil dryness causes water stress in each crop. To define crop-specific MADs, we created a table of parameters using FAO-56 (Allen, 1998).
In the drip and surface categories, evaporative losses happen only from the
soil surface because irrigation happens below the canopy level. Irrigation
takes place above the canopy in sprinkler and center-pivot systems;
therefore, evaporation from canopy-intercepted water (
The following formulas are used to calculate Evaporation of irrigation water intercepted by the crop canopy ( Evaporation from irrigation droplets ( Malek et al. (2017a, b): Playán et al. (2005):
VIC–CropSyst-v1.0 was originally developed and used to forecast the impact of climate change on CRB water supply and irrigation water demand (Yorgey, et al., 2011; Rajagopalan et al., 2017). This version was created using VIC (v4.0.7) and CropSyst (v4.15). This version is a lower coupling in terms of hydrology, i.e., both models simulate their own soil moisture with different soil parameters and soil layers. While VIC provides the water and cropping information and available energy for ET, partitioning of energy to different evaporative losses (i.e., evaporation from soil and transpiration) is separately done in each model and irrigation evaporative losses are not considered in VIC's energy balance. The irrigation efficiencies were hard-coded in this earlier version. VIC–CropSyst-v1.1 was slightly modified and used by Liu et al. (2013). Rajagopalan et al. (2017) also used VIC–CropSyst-v1.1 to evaluate the impact of climate change on agricultural productivity in the CRB. This paper describes the fully coupled version of the VIC–CropSyst model (version 2). This version is tightly connected in terms of which VIC handles all of the soil hydrologic processes; to do this, some VIC soil processes were altered to be more compatible with CropSyst. Furthermore, the influence of crop transpiration on energy balance is captured in this new version. Finally, this version mechanistically simulates irrigation processes and losses (e.g., irrigation evaporative losses) and is able to apply deficit irrigation.
VIC–CropSyst's simulated soil moisture, ET, yield, and irrigation water demand were compared to observed data obtained from the FLUXNET network (Baldocchi et al., 2001). Simulated LAI was evaluated against Moderate Resolution Imaging Spectroradiometer (MODIS) remote sensing observations (Cohen et al., 2006). We also evaluated regional performance of VIC–CropSyst in simulation of ET over the US Pacific Northwest, including the states of Washington, Idaho, and Oregon. Other studies such as Malek et al. (2017a, b), Rajagopalan et al. (2017a, b), Barik et al. (2017), Hall et al., 2017), and Yorgey et al. (2011) evaluated VIC–CropSyst in terms of its capability to capture regional irrigation demand, naturalized streamflow, observed streamflow, county-level yield, snow water equivalent, and irrigation efficiency.
Location of the two flux tower sites in the US, both of which are in agricultural fields. Mead irrigated site (NE) is located in Nebraska; the Fermi National lab site (IL) is located in Illinois; NE is irrigated and the IL is a nonirrigated agricultural site.
The flux tower stations considered in this study are located in the two US states of Nebraska (NE) and Illinois (IL) (Fig. 4). Available environmental and agricultural information include latent heat, soil moisture and meteorological data, crop type, LAI, and biomass production. The towers are all in agricultural fields and have relatively long periods of available data. The station in the IL is not irrigated and the site in NE is irrigated with recorded irrigation frequency and amount.
Daily meteorological data were acquired from the DAYMET
(Thornton et al., 2012) gridded data source. Soil files
were taken from Maurer et al. (2002) for associated
grid cells. We replaced its sand content with data available at the study
site. We also added the clay percentages to Maurer et al.'s (2002) soil
file. In our simulation, VIC–CropSyst reads the sand and clay content and
uses pedotransfer functions developed by Saxton et al. (1986) to generate
saturated hydraulic conductivity, bulk density, air entry potential, the
As with other hydrological models, the VIC model needs to be calibrated for optimized performance over a specific region. Table 3 shows VIC's key calibration parameters; more information on calibration parameters and methods can be found in past VIC studies (e.g., Elsner et al., 2010; Liang et al., 1994; and Maurer et al., 2002). We used calibrated parameters determined by Maurer et al. (2002) for each flux tower station (the last two columns of Table 3). We also tested the sensitivity of soil moisture content, crop growth, and irrigation demand and losses to different calibration parameters using the ranges available in the third column of Table 3; differences were negligible.
Two flux tower stations used for evaluation of the VIC–CropSyst. The Nebraska site is irrigated using a center-pivot system and the Illinois flux tower station is rain-fed.
Thermal accumulation time in CropSyst is used to represent crop phenological development and the rate of biological activity (McMaster and Wilhelm, 1997). Specifically, the sum of growing degree days (GDDs) is used to specify the time needed to reach specific phenological periods. We parameterized VIC–CropSyst for each site using published dates of crop growth stages (Table 4); meteorological information was used to convert calendar days to GDDs. Peak LAI was acquired from the MODIS LAI product (Cohen et al., 2006). Missing phenological information was estimated from the MODIS-derived peak LAIs as follows: (i) flowering is 2–7 days after peak LAI, (ii) filling starts 7–14 days after flowering, and (iii) maturity happen 30–45 days into the filling period. Table 4 shows estimated or observed dates of the growing stages.
Calibration parameters used for VIC–CropSyst over the two study sites (columns 5 to 6) and over the Columbia River Basin (CRB). Column 3 represents the ranges of these parameters used for the sensitivity studies.
Estimated calendar days correspond to each of the growing stages in two study sites. Some of the information is from references listed for each site.
Soil, climate, vegetation and crop information used for regional
evaluation of VIC–CropSyst over the US Pacific Northwest. The resolution
of the input data was 1
We used the gridded historical climate data developed by Abatzoglou and Brown (2012), including precipitation, minimum and maximum temperature, and wind speed (Table 5). The soil input file was developed using the STATSGO dataset (Schwarz and Alexander, 1995); to develop the soil file we used the same parameters as Elsner et al. (2010) except we added the clay percentage because, as mentioned earlier, VIC–CropSyst uses Saxton et al.'s (1986) pedotransfer functions and can internally calculate the soil parameters (e.g., hydraulic conductivity, field capacity, and bulk density). The calibration parameters (Table 3) used for simulation of ET over the Pacific Northwest were taken from Yorgey et al. (2011). Crop distribution information over the region was developed using data from Washington State's Department of Agriculture for Washington State and the US Department of Agriculture (USDA)'s cropping information for outside of Washington State (Boryan et al., 2011). More information on crop types and crop input parameters (e.g., phenological periods, radiation use efficiency, transpiration use efficiency, and maximum LAI) can be found in Barik et al. (2017), Hall et al. (2017), and Rajagopalan et al. (2017).
Figure 5 compares recorded and simulated irrigation
water (mean error
Simulated versus recorded total seasonal irrigation water in an irrigated corn field at the NE flux tower site.
Comparison of simulated and observed corn evapotranspiration
(ET; mm day
Figure 6 depicts the comparisons between monthly
simulated and observed ET over irrigated and nonirrigated sites. While the
model tends to overestimate ET, particularly during the month with larger
ET, simulations are more accurate at the NE irrigated site. Root mean
square errors (RMSEs) for the NE and IL stations were 0.8 and 1.0 (mm day
Comparison of simulated and observed corn yield at two flux tower sites for the years during which yield observations were recorded.
Figure 7 compares simulated and observed corn yield over the two sites. The mean error of simulated yield for NE (irrigated) and IL (nonirrigated) were 9 and 3 %, respectively. Although Fig. 7 does not show a systematic overestimation by the model, a combination of inaccurate meteorological data, missing processes (e.g., lack of VPD feedback, as discussed in Sect. 3.1.2), and unrecorded conditions such as insufficient irrigation water or heat stress may contribute to these discrepancies. The fact that the error is smaller over the nonirrigated site can be explained by the fact that irrigation management did not have to be simulated, thereby reducing the opportunity for introducing model error.
Figure 8 compares simulated and observed soil moisture over the two sites. Because the soil moisture sensors were placed at 10 and 25 cm depths at the NE site and at 2.5 and 10 cm depths at the IL site, we aggregated the first three VIC soil moisture layers (for a total thickness of 30 cm) for comparison against observations at the NE site. We compared just the first VIC soil moisture layer (10 cm depth) against observations at the IL site. The mean errors were 18 and 16 % for the NE and IL sites, respectively. As with crop yield, soil moisture simulations are better for the nonirrigated site, particularly in terms of variability. The discrepancies may relate to the use of Pedotransfer functions that convert soil textural characteristics to soil hydraulic properties (e.g., field capacity, permanent wilting point, and hydraulic conductivity) for use in VIC–CropSyst (Pachepsky and Rawls, 1999; Tietje and Hennings, 1996). Also, scale discrepancies between the sensors' point-scale observation and the grid-scale simulation (Crow et al., 2012; Robinson et al., 2008) as well as inaccuracy of meteorological and soil data can be other sources of error. Additionally, imperfections in model processes such as soil water movement, ET, and irrigation loss calculation can contribute to the error.
Figure 9 shows that VIC–CropSyst is able to capture the magnitude and seasonality of observed LAI, with a slight underestimation of peak LAI. The information we used for calibration of phenological periods is not specifically collected for the two study sites, but instead was based on state-scale studies and reports; this is a potential source of error in the simulation of LAI. Because of limited information at flux tower sites, we did not consider all of the crop-related parameters (e.g., radiation use efficiency, maximum crop coefficient, and maximum crop coverage) during calibration, which can also lead to some discrepancies (e.g., Jalota et al., 2010; Klein et al., 2012).
Comparison of simulated and observed soil moisture at the flux
tower sites located in IL
Comparison of simulated and observed corn LAI over two flux tower
sites located in IL
We used VIC–CropSyst to simulate ET over the CRB portions of three states:
Washington, Idaho and Oregon (Fig. 10). Simulated
ET was aggregated from the original model resolution of 1
Rajagopalan et al. (2017) performed an evaluation of county level
aggregated irrigated crop yields against NASS crop yield statistics, and a
comparison of average modelled irrigation demands from the Columbia Basin
Project area to irrigation diversions. The mean annual yields between
observed and simulated values are in agreement with relative errors less
than
Farmers adapt their agricultural management to minimize unfavourable impacts of stressors such as climate change (Kurukulasuriya and Rosenthal, 2003). Possible agricultural adaptation strategies have been discussed (e.g., Anwar et al., 2013; Howden et al., 2007; Kurukulasuriya and Rosenthal, 2003; Smit and Skinner, 2002; Smith et al., 2000). However, lack of appropriate simulation tools to assess the effectiveness of an adaptation decision while capturing complex regional impacts is a significant shortcoming. VIC–CropSyst simulates common adaptation strategies used by farmers, and captures the consequences of these adaptation strategies on local and regional hydrology and land–atmosphere interactions. Table 6 shows a list of adaptation decisions that can be handled by VIC–CropSyst. These decisions range from short-term tactical (T) to long-term strategic (L) decisions.
VIC–CropSyst can be used with other modeling frameworks such as river routing, water management, atmospheric, and socioeconomic models. Many of these integrations simulate the human–land–climate nexus and provide scientists, stakeholders and policy makers with a broader understanding of the interactions of and feedbacks between human decisions and the Earth system. VIC–CropSyst has been already used and implemented in various projects; examples are as follows.
Comparison of simulated and empirically derived ET over the
US Pacific Northwest. The simulation and observation period is 1982–2008.
Panel
Summary of adaptation strategies that can be handled by VIC–CropSyst: the modeling platform is able to simulate the impacts of local decisions on agricultural productivity and at the same time capture the impacts of these decisions on regional land–atmosphere interactions and surface water availability in the basin.
VIC–CropSyst has been used in conjunction with reservoir operation models in
the CRB. For example, Rajagopalan et al. (2017)
utilize such a platform to assess the impacts of climate change on
agricultural production; this includes both the direct impacts of climate
change (precipitation, temperature, and CO
VIC–CropSyst has been used to investigate different scenarios for renegotiation of the Columbia River Treaty (Rushi et al., 2016). Existing modeling efforts to date have focused primarily on the impact that treaty renegotiation would have on flood risk, hydropower generation, and environmental flows (Cosens, 2010; Hamlet and Lettenmaier, 1999); assessment of the impact of CRT changes on irrigated agriculture along the Columbia main stem is a knowledge gap. Therefore, Rushi et al. (2016) applied VIC–CropSyst linked to ColSim (Hamlet and Lettenmaier, 1999) to simulate the complex impacts of climate change and the Columbia River Treaty on hydrology and agriculture in the river basin and concluded that climate change (i) shifts water supply towards earlier in the season, (ii) reduces flood risk in the upper CRB while increasing frequency and magnitude of floods in the middle and lower parts of the basin, (iii) shifts water demand to earlier in the season in some locations with mixed effects on water rights curtailment risk, and (iv) reduces hydropower generation. The authors found that the considered CRT scenarios can improve power generation and agricultural water demand while preventing floods in an altered climate.
VIC–CropSyst is an effective tool for studying the large-scale aggregated impacts of local management decisions and phenomena. For example, VIC–CropSyst was applied by Malek et al. (2017a, b) who found that climate change-induced increases in evaporative (consumptive) losses from irrigation systems and decreases in nonevaporative irrigation losses (i.e., runoff and deep percolation) would lead to a decrease in reusable return flow, which would negatively affect basin-wide water availability and productivity.
VIC–CropSyst has also been used over the Yakima River Basin (YRB) to evaluate the impacts of climate change on decisions related to investment in irrigation technology (Malek et al., 2017a, b). Economic damages of future, more frequent droughts (Vano et al., 2010) are considered the main incentive to invest in more efficient irrigation technology (Berger and Troost, 2014). To analyze future changes in regional irrigation patterns, Malek et al. (2017a, b) used VIC–CropSyst in conjunction with an economic model and the RiverWare model (Zagona et al., 2001). Figure 11 shows a result of this integration to simulate historical (1981–2006) drought frequency and severity, and the percentage of the YRB's perennial crop growers who are simulated to switch to more efficient irrigation systems to minimize the negative consequences of droughts during the two decades of 1990–2000 and 2050–2060. Also, any changes in agricultural activities (e.g., switching to a new irrigation system) directly impacts the hydrology of agricultural fields, thus changing return flow timing and magnitude and the availability of water for downstream users; these downstream consequences can also be simulated by this modeling platform. This is an example of how the human–land–climate nexus can be captured through a modeling framework that simulates large-scale hydrologic processes and regional water availability in a highly cultivated basin, while capturing the dynamics of farm-level irrigation decisions.
Regional application of VIC–CropSyst in conjunction with a river
system model (YAK-RW; Hubble, 2012;
Zagona et al., 2001) and an economic model to simulate historical
(1981–2006) drought frequency
Irrigation and other agricultural decisions modify local to regional climate through changes in land surface conditions such as temperature, water vapor content, and albedo (Fernández et al., 2001; Liu and Kang, 2006). This phenomena can be used to compensate for the negative impacts of heat stress (Lobell et al., 2008), which will be especially important in the future if there are more severe and frequent extreme events related to climate change (Long and Ort, 2010). These management decisions will also impact the regional water cycle, potentially leading to disruption in water availability (Adamson and Loch, 2014) and modifying fluxes of water to the atmosphere (Pielke Sr. et al., 2007). As a part of the BioEarth platform (Adam et al., 2014), VIC–CropSyst is being coupled to an atmospheric model, the Weather Research and Forecast model (WRF; Michalakes et al., 2005; Skamarock et al., 2008), that can be used to quantify the impacts of irrigation and other agricultural management on atmospheric processes, as well as to assess how irrigation management can be used to mitigate heat stress.
Although the results presented in this article do not include results related to deficit irrigation during times of water shortage, VIC–CropSyst is able to simulate the impacts of deficit irrigation on hydrologic and cropping systems. VIC–CropSyst's deficit irrigation module requires two main inputs: (a) a first approximation to the irrigation water demand obtained by generating time series of irrigation in a zero water stress condition using VIC–CropSyst and (b) deficit fractions that indicate the actual water availability as a function of the crop water requirement. VIC–CropSyst then reads the amount of recorded irrigation from step one and applies the deficit fraction to simulate the agricultural and hydrologic processes under realistic water deficit conditions. The deficit fraction can be either homogenously applied across the entire basin or separately specified for each farmer depending on water rights or other considerations. Also, VIC–CropSyst can apply the deficit fraction during different times of the year. For example, if the water deficit happens later in the season, VIC–CropSyst can adjust irrigation amounts according to the timing of water shortage.
VIC–CropSyst has also been used in conjunction with reservoir models (e.g., ColSim: Hamlet and Lettenmaier, 1999, and YAK-RW: Zagona et al., 2001) to calculate the deficit irrigation fraction (e.g., Barik et al., 2017; Malek, et al., 2017a, b; and Rajagopalan et al., 2017). In general, the following six steps can be used to calculate and apply a deficit fraction: (1) VIC–CropSyst simulates the hydrologic states such as runoff and base flow as well as the irrigation water demand, (2) a routing model (Lohmann et al., 1998) is used to simulate streamflow, (3) simulated flow is bias-corrected against observed flow, (4) a river system model is used to include operation of dams and reservoir and estimate water availability, (5) the availability of water is compared with water demand, and (6) a deficit fraction is calculated and VIC–CropSyst is run to simulate the impacts of an irrigation deficit on the hydrologic cycle and on crop yields.
Which model to apply for a specific research question at hand is dependent on a variety of factors, including geographical considerations but also the level of sophistication needed to address the question. For example, areas with significant irrigation activities can be more precisely simulated with mechanistic irrigation models, or areas with cold climate would necessitate models with more sophisticated cold-season processes. Also, regional agricultural economic studies require a reliable simulation of crop yield for economically significant crops grown in the region. Therefore, models that simulate generic C3 and/or C4 crops are not the best option for this type of question. VIC–CropSyst and LPJML are two examples of models that can be used to answer this type of question. Moreover, some of the models have been already tested and used for a particular region and resolution, which naturally makes them more reliable for that specific situation.
Meeting future food demand will require an extensive understanding of the interactions between agricultural and other systems, such as water resource planning and management as well as socioeconomic and atmospheric processes. The main purpose of this study was to develop the VIC–CropSyst platform that provides tightly integrated and mechanistic representation of both cropping systems and water and/or energy cycles on regional to global scales. Tight integration between VIC and CropSyst necessitated modification of both models, including how the models handle soil movement and vertical distribution, transpiration, LAI, and irrigation. Evaluation of VIC–CropSyst over two flux tower sites shows that the coupled model captures key agronomic and hydrologic states and fluxes on the field scale. Furthermore, implementation of VIC–CropSyst over the US Pacific Northwest region reduced ET simulation error by 40 % over irrigated landscapes.
The VIC–CropSyst platform enables the land surface modeling community to investigate a variety of agricultural management decisions, including crop choice, planted acreage, planting and harvesting date, and multiple irrigation management options. In particular, the new mechanistic irrigation model, which is tightly coupled with both the energy and water cycles, can be used to address questions related to the interaction of climate, hydrology, river basin water management, and irrigation management strategies.
VIC–CropSyst can be integrated with different modeling platforms to capture
the dynamics of the human–land–climate nexus. This can potentially improve
the understanding of environmental processes in highly cultivated basins and
can be used to investigate best management practices to promote future
sustainability of agricultural production while preserving water resources
and minimizing the negative intended and unintended consequences of human
actions. Some examples of these implementations are as follows:
VIC–CropSyst is being used in Earth system models (EaSMs) such as BioEarth
(Adam et al., 2014) and can be
implemented in other EaSMs, such as the Platform for Regional Integrated
Modeling and Analysis (PRIMA; Kraucunas et al.,
2014). Implementation of VIC–CropSyst in EaSMs facilitates a powerful
representation of large-scale interactions between different biophysical and
socioeconomic components over areas with significant agricultural
activities. This is a transformational step in the understanding of the
food–energy–water nexus which can lead to efficient and more sustainable
management decisions that balance and benefit all three sectors.
The VIC–CropSyst is a freeware open-source community model; source codes, user manual, and test cases can be distributed by request to Keyvan Malek (keyvan.malek@wsu.edu), Jennifer Adam (jcadam@wsu.edu), and Mingliang Liu (mingliang.liu@wsu.edu).
The authors declare that they have no conflict of interest.
We would like to thank three anonymous reviewers for their constructive comments and suggestions. We would also like to thank the editor Chritopher Müller for his insightful comments and valuable suggestions. This research was funded by Washington State Department of Ecology's Office of Columbia River Basin and the US Department of Agriculture's National Institute of Food and Agriculture, grant number 2011-67003-30346 (Biosphere-relevant Earth System Model; BioEarth). This research is also financially supported by Washington State University's Graduate School. Edited by: Christoph Müller Reviewed by: three anonymous referees