JDec: Joint decision models for citizens, crops, and environment

This project will adapt decision-theoretic tools to agri-environmental management, a domain that has been underserved by mathematical methodology. The process of decision-making within an agricultural context is complex, because it spans multiple interdependent stages, and involves many risks along the way. Decisions - when to apply pesticides, how much to apply, when to prune, when to water, even when to harvest - can affect crucially the produce quantity and quality, and hence the short-term success of the enterprise. The decisions will also determine the extent of environmental harm, which have been challenging to define, as is the value of "services" provided by the ecosystem. To facilitate their inclusion in decision-making we develop models that are more flexible and more holistic than common frameworks in operational research.

First, outcomes need to be valued by utility functions that reflect costs and benefits comprehensively. Among other things, they need to be evaluated along decision trajectories, including appropriate levels of memory and foresight, and interdependencies along the way. For example, a herbicide treatment may look effective only as long as its indirect effect is ignored on the wild pollinators that had visited the weeds, and whose loss will need to be compensated with new costs.

Second, in an agricultural-environmental context, decisions are not taken by humans alone. A modelling approach looking at decisions being taken jointly by all three --- the farmer, the crop and the environment --- opens the flexibility needed to deal with interactions. We further allow for a higher level of uncertainty, in that the influence each of these agents has may itself depend on random events.

Third, our models acknowledge the temporal dimension and potential resource allocation constraints. In a large, interconnected, multi-stage system of land and resource management, past actions influence future decisions. Adding rapidly changing environment, with extreme weather events increasing in frequency, shifting pest and pollinator ranges, and resource depletion, we need to take account of the need for robust approximate solutions in model development. In other words, the challenges of having to make decisions in the "real world in real time" requires a paradigm for "good enough" decision-making, and a conceptualisation of the gap it has to optimal solutions.
Our major objective is to build the mathematical and statistical framework for decision modelling that covers these three aspects. Our work extends existing approaches by building in more flexible mechanisms for uncertainty and interdependencies. Key ideas from behavioural sciences will move us beyond a narrow rationality framework. Subject to data availability, our resulting theory will be applicable to both small and large landscape scales.

We will explore these ideas in two case studies. The first is a system of wild pollinators in apple orchards, a particularly suitable testing ground for understanding indirect effects at the frontier between managed land and its surrounding landscape. The second case study explores the use of decision modelling in a large farm scale experiment with 4 crops and multiple intervention methods. It provides a rich data set for comparing decision strategies.

Our work can directly benefit many citizens: not only crop scientists and land managers, but also ecologists, conservationists, local authorities, charities and policy-makers. The tools we are designing will open up to the field sciences an approach that has been used with great success in a variety of other disciplines. With better tools, such as the ones we are proposing, environmentally-conscious actions taken to feed a growing population in a changing climate can be dynamic, adaptable, and sustainable.

The proposed work aims primarily to expand decision-theoretic concepts and make them applicable to an agri-environmental context. This, we believe, will be of benefit to both academic and non-academic audiences. The non-academic category includes the many different types of stakeholders involved in decision-making about resources and management (e.g. allocation, landscape use and crop management decisions) within an existing legal framework. These stakeholders include land managers, land owners, local authorities, and charities. It also includes those who make recommendations or decisions about policies itself, such as politicians at regional, national and European level.

A popular framework used by policy makers for assessing and comparing the impact of alternative policies is multi-criteria decision analysis (MCA). It gives priority to assessing all three dimensions, economic, social, and environmental and arriving at solutions through negotiation processes. Another such framework is multi-level perspective (MLP), which is frequently applied to scenarios involving technological transitions. Neither frameworks is a priori quantitative, but both draw on other frameworks to conduct that part of the analysis and integrate them into a bigger picture afterwards. This role could also be taken up by our proposed models and the tools built from them, enhancing MCA and MLP with a more flexible and holistic decision-making approach.

A primary output of the work will be the design of tools (both as mathematical models and as software) for the application of our theoretical developments. These can be applied by crop scientists, ecologists, or conservationists to agri-environmental datasets, or datasets that have a similar structure. We intend to present these findings in audience-appropriate scientific publications, with the associated code publicly available. Furthermore, we intend to have worked examples distributed with the code, in order to ensure that users can confidently apply our procedures. Within the research community, our work can be used as the basis for a range of further developments, from crop development to projects for restoration of old agricultural land.

As a by-product of our work, we will be putting together standardised protocols for data quality assessment of agri-environmental datasets, and primary exploratory and data-handling procedures. Since transparency and repeatability are core values of our research, we will make all of these protocols available, and we will associate with them tutorials that can guide learners through the fundamentals in an accessible manner. To further promote training we intend to create learning units from our tutorials, that can be used as teaching resources for life scientists.

While it is tempting to design decision support tools to help land managers do their jobs, such tools have found low uptake in the past. Similar observations have been made in the clinical context, where there have also been very few successful examples. Hence, the first step for building a successful tool is to elicit user needs through dialogue. Both outputs described above can function as vehicles to create such a dialogue with a potential user community and create trust that they can deliver useful advice.
Rather than providing closed or even proprietary software tools, we will provide an accessible theoretical and conceptual framework upon which future researchers and users can build. Realistic quantification of the differences between outcomes of alternative decision strategies and priorities can serve as guidance toward better intuitive agri-environmental decision-making, and may find its way into discussion and future developments of the decision process.