Holzworth et al. discussed advances in capabilities and applications over time. Basso et al. reviewed the performance of CERES maize , wheat and rice models compared to measured data over the last 30 years in 43 countries. They reported that model performance, using site-specific inputs, was outstanding for the variables compared . Models of cropping and grassland systems share the same fundament characteristics: both describe crop or grassland agro-ecosystem growth and yield responses to climate, soil, plant species characteristics, and management. However, several aspects of grassland/rangeland modeling present unique challenges. Many of these challenges stem from the requirement that grassland models represent several interacting species, including perennial and woody species of grasses. Persistence of plants over multiple years forces the models to consider residual effects over time. Dependency on soil-derived nutrients or human-induced disturbances like fire reinforce the longer-term perspective needed for grassland modeling. Thus, although most biophysical processes are similar additional factors are considered when modeling grasslands. 2.1.1. Model-simulated responses of interest to users The most common response variable modeled for cropping systems is yield, whether of grain, tuber, or forage biomass yield. This yield is harvested at a single point in time for determinate annual crops, while indeterminate crops and grasslands may be harvested multiple times. Although statistical models may be useful for predicting these biological yields in response to some combination of weather conditions, nutrient levels, irrigation amounts, etc. , they do not predict responses to non-linearities and threshold effects outside the range of conditions in data used to develop them. In contrast, dynamic cropping and grassland system models may simulate these biological yields and other responses important to analysts, such as crop water use, nitrogen uptake, nitrate leaching, soil erosion, soil carbon, greenhouse gas emissions,stackable planters and residual soil nutrients.

Dynamic models can also be used to estimate responses in places and for time periods and conditions for which there are no prior experiments. They can be used to simulate experiments and estimate responses that allow users to evaluate economic and environmental trade offs among alternative systems. Simulation experiments can predict responses to various climate and soil conditions, genetics, and management factors that are represented in the model. “Hybrid” agricultural system models that combine dynamic crop simulations with appropriate economic models can simulate policy-relevant “treatment effects” in an experimental design of climate impact and adaptation .One innovation of early crop modeling pioneers was to categorize the crop production situation being modeled to narrow down the many factors that are needed by crop models . Fig. 1 summarizes three crop production levels and factors that influence each. Potential production is defined as crop production that is determined completely by defining factors of CO2, radiation, temperature, and crop characteristics . Potential production models also include partitioning of biomass growth into grain and other plant parts, with defining factors modeled to affect these processes. This potential production level is rarely achieved in real production situations, although under highly intensive management , production approximates the potential level for the specific CO2, temperature, radiation, genetics, and canopy architecture used. For example, crops grown in greenhouses or in intensively managed fields in some regions produce yields that are at or near potential levels. The next production situation is referred to as water-limited and/or nutrient-limited production . At this level, the defining factors are still important, but there may also be limitations in the water and/ or nutrients needed to achieve full growth potential. Crop models that simulate water and/or nutrient-limitations must include soil water and nutrient component modules to simulate the time-varying availability of water and nutrients, the uptake of these resources, and reductions in growth and development if they are not adequate to meet potential growth demands.

Most cropping and grassland system models contain component modules that simulate soil water, nitrogen, and carbon dynamics because of the global importance of these resources in determining yield. Although some models include phosphorus, most of them do not simulate responses to phosphorus, potassium, or micro-nutrients. Models that represent soil water, nitrogen and carbon dynamics are complicated not only because of the physical and chemical processes that occur in soils, but also because of the complexities in management practices used for these resources . Finally, actual production includes additional factors that may reduce growth and yield . Whereas some crop models have capabilities to introduce damage by diseases and insects , the modeling of these reducing factors has not kept up with other advances in crop modeling for a review of recent progress. Most groups modeling cropping and grassland systems do not include these factors. Thus, few current models simulate responses to pest or disease damage or to their management using resistant varieties, agro-chemicals, or other approaches. This is a major limitation for some applications.Dynamic crop models generally include factors at the potential yield level in addition to water- and nitrogen-limited production level. However, the ways that different models include those factors vary. Fig. 2 shows a schematic of the components in the Cropping System Model that incorporates the CERES , CROPGRO, and other models in DSSAT . The CSM models can include soil water, nitrogen, carbon, and phosphorus dynamics and can introduce pest and disease damage into some crops using the concept of coupling points . It also can simulate multiple seasons so that carry-over changes in soil water, N, and P are simulated to represent longer-term changes in soil resources in response to different management systems . A number of other cropping and grassland system models have similar components and capabilities , although most models do not simulate impacts of pests and diseases unless coupled externally with time-series input data or pest models like in DSSAT CSM . Some models have an ability to simulate intercropping .

An unfortunate feature of current crop and grassland models is that modules from one set of models are not compatible with other models. For example, APSIM’s intercropping capabilities are deeply embedded in the system architecture and cannot be simply moved to other models like DSSAT CSM. Moving pest and disease damage modules from DSSAT CSM to APSIM is possible but requires coding of module “wrappers” to handle inter-model communications – a non-trivial task.Most “cropping system” models have evolved as elaborations of component crop and soil models and the focus has been on modeling a single “point” in space over time to explore variability in crop responses to soil, management and weather. A typical structure for this pedigree is shown in Fig. 2. Most operate on daily or hourly time-steps. Some include hourly time steps for computing rates of photosynthesis and other processes but also use daily steps to update state variables such as phenological development, and biomass of plant organs. These time steps are also used to compute changes in soil water, soil nitrogen, and crop biomass that result from soil-water processes including rainfall, infiltration, runoff, percolation, redistribution, and plant uptake, and changes in soil nitrogen. Details of how different growth, hydrology, and soil nutrient processes are represented vary among models. Models may be either functional or mechanistic, with the choice of approach depending on the modeling team’s knowledge of the system, their purpose, the availability of data for parameterization, and their experience in developing and evaluating models. These differences lead to different models producing different responses when used to simulate the same experiment . Most models use simplified functional equations and logic to partition simulated biomass into various plant organs. Functional models also primarily use “capacity” concepts to describe the amount of water stored in a soil that is available to plants; mechanistic models, in contrast, use the potential energy of soil water and “instantaneous rate” concepts from soil physics. In capacity-based functional models, it is the difference between the upper and lower limits of soil water-holding capacity that determine the amount of water available to plants. In this type of soil water model, water movement and its availability for crop growth are represented by functional equations on a daily time step, even though infiltration and runoff processes may be computed with smaller time steps. Some modeling systems can operate with either capacity based or energy based soil water modules and ideally a flexible agro-ecosystem simulation engine or platform will be able to work with component modules specified to different degrees of “mechanism”. Although some models include input information on plant genetics ,stacking pots these are few in number and not yet in widespread use. Most models are not genetic-based, which is one reason that calibration of models using field data is widely practiced to obtain genotype-specific parameters. Some modeling platforms while utilizing crop and soil components such as shown in Fig. 2, have focused more strongly on “agricultural system” features, with capabilities of instantiation that facilitates the simulation of systems features such as multiple paddocks, intercropping, weeds, tree – crop interactions, livestock operations and even non-biological features of farms such as water storage structures.

APSIM is the best known example of this farming systems “platform”. It sits at the interface of the crop-soil systems models typified by Fig. 2 and the whole farm optimization models discussed elsewhere in this paper. Holzworth et al. outlines in full these “agricultural systems” features of the APSIM approach .Grasslands are usually mixed stands comprised of a variety of grasses and forbs, including legumes and sometimes woody species . Unlike croplands, the diversity of species generally precludes use of a single-species parameterization, since species vary in their ability to compete for space, water, nutrients , and light. Grassland models generally represent plant behavior and competition among herbaceous plants using one of: a set of species, each independently parameterized; amalgamations of plants into parameters for plant functional types ; or community-averaged parameterizations . While requiring more effort for parameterization, these amalgamated approaches enable representation of changes in plant community composition over time, for example in response to climate change, competition among plant populations, and mortality. Trees are dynamic components of the world’s native grazing lands and can have significant impacts on ecosystem function . Representing tree/grass competition is challenging because trees respond differently to various drivers and depend on plant population characteristics . Shifts in plant community composition can be self-reinforcing due to co-occurring population and biophysical changes . Dynamic vegetation modeling approaches are used to represent competition among herbaceous and woody types for water, nitrogen, light, and space. Dynamic rangeland vegetation models and state-and-transition models identify a set of plant communities that tend to resist change due to disturbance, but also describe drivers that lead to a transition to another quasi-stable plant community . Expansion of woody species and increases in woody cover are widespread phenomena that under many but not all environmental conditions lead to the transition of early successional communities dominated by grasses and forbs to forests . Studying woody encroachment and understanding the importance of competing drivers has been challenging, in part because of the slow rates of the processes driving changes . These slow changes are reflected in the drivers of transitions in state-and-transition models and contribute to uncertainty in our ability to represent longer-term changes in the tree-grass balance. Ecological succession has been studied by plant ecologists since pioneering work before 1945. More interactions among agricultural and ecological modelers are likely to be mutually beneficial. Grazing animals of all kinds have an impact on plant productivity by removing photosynthesizing tissues, altering light transmission through the canopy, influencing nutrient cycling and affecting plant allocation patterns and differentially influencing species mortality and recruitment rates in grasslands . Such changes to groups of plants can drive changes in the competitive balance and thus plant community composition. Whereas grassland models incorporating species or plant functional types can represent grazing-induced changes in the competitive balance, such models that represent plants with a set of community-wide parameters usually rely on some combination of LAI -driven reduction in production potential along with grazing response curves. Ingrasslands/rangelands, grazing removes some plant productive capacity, and thus models cannot rely upon deterministic growth curves, but must be able to forecast growth for plants with an amount of biomass or leaf area that varies independent of the time of year or climate. There can also be significant differences in growth rates among and even within species after a grazing event .