The systems approach has several important implications for second generation models

While there was a sense that “decision support” was important, the model developments nevertheless began with research tools that were motivated primarily to better understand basic processes and effects on system performance. As long as model development is motivated primarily by academic and research outcomes, it will remain only loosely connected to user needs. Therefore, to re-orient model development towards user needs, a new set of institutional arrangements and incentives is needed. Fig. 1 presents a diagram of how these new arrangements might be organized. The figure shows the linkages between a “pre-competitive space” of basic science and model development, and the “competitive space” of knowledge product development. The concept of “pre-competitive space” grew out of the efforts of the pharmaceutical industry to collaborate on basic research while competing in product development. The arrows between these two “spaces” point both ways to represent the inevitable and important give-and-take. The model development approach that now exists is largely missing the competitive space component shown in Fig. 1.

To the extent that such a competitive space does exist,round pots it is in the private sector where proprietary management support is being provided, and linkages in Fig. 1 from competitive knowledge product development back to data and model development are largely missing. In Fig. 2 we show how this link from private decision makers to models and public data could be made by connecting on-farm decision support tools to databases that could be used for model development and analysis . Facilitating a pre-competitive environment is likely to require innovations in the way research organizations operate, and may need to involve public-private partnerships that clearly delineate boundaries and roles in creating specific NextGen products. PPPs are one way that science and industry can collaborate to generate new applied knowledge that can feed into the creation of new business and services. In PPPs it is common that both private and public partners provide funding and jointly formulate the research questions that can subsequently be tackled by research institutes and universities. There are a number of challenges in structuring PPPs. For example, in the European Union PPPs have been regulated to avoid unfair competition. The EU regulations stipulate that there always has to be more than one private partner involved and intellectual property rights of the knowledge developed belong to the research partner, which can then license the use to private partners for commercial purposes. An important aspect for a NextGen community of practice is openness. Open here means: first, inviting and engaging others to join and become involved; second, being ready to set priorities jointly with a broader stakeholder community ; and third, being transparent for scientific and public scrutiny of methods, tools and results through not-solely scientific venues.

Only a few of the agricultural systems models and economic models now in use can be said to be “open” in the sense that both the model equations and programming code are fully documented and freely available to the community of science. Establishing an open approach consistent with the principles of good science, including sufficient documentation and sharing of code to allow replication of results with reasonable effort, should be a priority of the practitioner community. Such an approach would facilitate model improvement through peer review, model inter-comparison and more extensive testing, new modes of model improvement and development such as crowd sourcing, and education of the next generation of model developers and users. Creating this open approach will also raise challenges related to incentives and intellectual property that would need to be addressed. The recent experience with the Agricultural Model Inter-comparison and Improvement Project , a new community of science dedicated to an open approach, suggests that researchers are now more willing to participate, but also has identified some of the challenges to an open collaborative approach. For example, obtaining funding for collaborative activities creates coordination issues among research institutions and funding agencies that need to be addressed. Another advantage of an open approach is that it will encourage the emergence of competing models and modeling approaches, rather than a single “super-model.” One dominant “super-model” could eventually emerge, but the only way to know that such a model is desirable is to allow a multi-model environment to flourish.

We also expect to see alternative approaches emerge as modelers tackle challenging features such as representation of heterogeneity and dynamics and linkages across scales. For models to be tractable, trade offs have to be made, and an open approach is needed to facilitate the testing of alternative solutions. There are important examples of recent efforts at creating a more open approach to agricultural model development. The bio-economic farm model FSSIM was made available as open source in 2010 after completion of its main project-related development and published with a license that allowed further use and extension. It is notable that the open sourcing of the model was combined withtraining sessions, but this did not lead to spontaneous community uptake and large-scale development of this relatively complex and data demanding model. The DSSAT crop modeling community is undertaking an effort to make its code open-source with the participation of more than 20 developers. The Global Trade and Analysis Project has provided extensive documentation of its model and data and allows user-modification of its standard model , and there is a large number of users of the model globally. The IMPACT model developed by the International Food Policy Research Center is publicly documented and available to other researchers . The TOA-MD model for technology adoption and sustainability assessment of agricultural systems was developed based on experience which showed that potential users needed a user-friendly, transparent tool for impact assessment. The TOA-MD model is available to users with documentation and a self-guided learning course and there is a growing community of users . To achieve the goal of demand-driven model development, it will be necessary to strengthen the linkages between the pre-competitive space of model development and the competitive space of knowledge product development. The current state of affairs appears to be that, on the one hand, the modeling community is strong on analytical capability but weak on linkage to user demand; while on the other hand, the developers of user-related farm-level products are weak on analytics. Thus, there appears to be the opportunity for “gains from trade” by facilitating more interaction between the two communities. An important part of this interaction has to be to identify the key research that could enable better service delivery to knowledge-product users.

Additionally, as emphasized in the NextGen Use Cases, there is a public good value to enhancing a broader community that can provide both data and analytics for public investment and policy decision-making. These ideas are further explored in the paper by Janssen et al. and by Capalbo et al. .The explosion in the availability of many kinds of data and the capability to manage and use it create new opportunities for systems modeling at farm and regional or landscape scales. Fig. 2 presents an example of the possible types of private and public data that could be generated and used for both farm-level management Use Cases and for landscape scale investment and policy analysis Use Cases. Some of these data would be generated and used at the farm-level,garden pots plastic others would be generated and used for landscape-scale analysis to support investment decision-making and science-based policy-making. While farm-level decision making and landscape-scale analysis have different purposes, they both depend on two kinds of data: private data, including site and farm-specific characteristics of the land and the farm operation, and the site- and farm-specific management decisions that are made; and public data, i.e. weather, climate, soils, and other physical data describing a specific location, as well as prices and other publicly available economic data. Many farm-level data and decision tools from private and public sources are currently in use, and are evolving rapidly . The left-hand side of Fig. 2 presents the generic structure of these tools, the data they use as inputs, and the outputs that are generated. The right hand side of Fig. 2 shows the general structure of the data and models needed to carry out landscape-scale research and policy analysis. A key feature of landscape-scale models is that they use public data for prices, weather forecast, and policy information, private site and farm-specific input use data, and outcome based data that are useful for both farm-level management decisions and landscape-scale policy decisions. There are three broad categories of landscape-scale data: publicly available bio-physical data, including down-scaled climate and soils data; publicly available economic data, including prices and policy information; and the confidential site- and farm-specific data obtained from producer- and industry-generated databases. Landscape management and policy analysis models require spatially and temporally explicit data that are statistically representative of the farms and landscapes in a geographic region in order to provide reliable information about economic and environmental impacts and trade offs. Such data are not typically available in most parts of the world. As a result, implementation of these models relies on the publically available information on farm management collected periodically through special-purpose surveys. Currently available data are inadequate for various reasons. Many of these data are collected with samples that are not statistically representative of relevant regions or populations for landscape-scale analysis; many data are not spatially or temporally explicit, are only available after substantial aggregation, thus limiting their usefulness, and are often available with long time lags between when the land management decisions are made, the data are collected, and when they become available for research or policy purposes.

Longitudinal data that provide observations of the same farms over time are particularly important for policy research, but there are few such data available. A key implication of the framework presented in Fig. 2 is the complementarity between knowledge product design, agricultural system models, and farm-level data collection. We return to this issue in Section 4.The NextGen Use Cases show clearly the need for whole-farm system approaches. Agricultural systems are managed ecosystems comprised of biological, physical and human components operating at various scales . Farms are embedded within larger ecological and human systems operating at regional scales , as well as larger scales. The need for a system-level understanding, however, should not be construed as meaning that there is not a need for component-level tools as well. Indeed, particularly in the more specialized, industrial systems, there will be a growing demand for tools to improve management of soil fertility, pests and diseases, and other elements of on-farm management. Nevertheless, until these components are integrated into a wider systems approach, it they will not be able to achieve goals of sustainable management. For example, nutrient and pesticide use cannot be managed effectively to account for potential off-farm impacts on water quality without a systems approac.Within each system level, a set of interacting subsystems is involved. This suggests the possibility of constructing models of large, complex systems by combining models of modular sub-systems. The level at which modularization may be possible remains an important question, and this in turn has implications for software engineering. For example, as discussed in Jones et al. , many crops are now modeled individually and separate from livestock. Systems with multiple interacting crops , livestock, and crop-livestock interactions, are needed for various Use Cases, showing the need for these interacting components to be incorporated in a modular “plug and play” system. Also, these biophysical production system components interact with economic-behavioral components and environmental components. These interactions among sub-systems show the need for standard ways to link inputs and outputs among sub-systems. As we noted above, several more complex systems models have been developed , but as yet each modeling system uses its own approach to model linking and model components from different developers cannot easily communicate with each other. Another important issue raised by the systems approach is the appropriate level of complexity for Use Cases, an issue discussed further in Section 3.8. Research in environmental modeling indicates there are often diminishing returns to complexity. Similarly, experience with economic modeling has shown the value to “minimum data” or “parsimonious” approaches . The need for both modularity and parsimony also relate to the need for generic approaches, particularly for complex agricultural systems models and economic models, so that model developers can move away from models that are application-specific.