Model-based optimisation of agricultural profitability and nutrient management: a practical approach for dealing with issues of scale.

To manage agricultural landscapes more sustainably, we must understand and quantify the synergies and trade-offs between environmental impact, production, and other ecosystem services. Models play an important role in this type of analysis as generally it is infeasible to test multiple scenarios by experiment. These models can be linked with algorithms that optimise for multiple objectives by searching a space of allowable management interventions (the control variables). Optimisation of landscapes for multiple objectives can be computationally challenging, however, particularly if the scale of management is typically smaller (e.g. field scale) than the scale at which the objective is quantified (landscape scale) resulting in a large number of control variables whose impacts do not necessarily scale linearly. In this paper, we explore some practical solutions to this problem through a case study. In our case study, we link a relatively detailed, agricultural landscape model with a multiple-objective optimisation algorithm to determine solutions that both maximise profitability and minimise greenhouse gas emissions in response to management. The optimisation algorithm combines a non-dominated sorting routine with differential evolution, whereby a 'population' of 100 solutions evolves over time to a Pareto optimal front. We show the advantages of using a hierarchical approach to the optimisation, whereby it is applied to finer-scale units first (i.e. fields), and then the solutions from each optimisation are combined in a second step to produce landscape-scale outcomes. We show that if there is no interaction between units, then the solution derived using such an approach will be the same as the one obtained if the landscape is optimised in one step. However, if there is spatial interaction, or if there are constraints on the allowable sets of solutions, then outcomes can be quite different. In these cases, other approaches to increase the efficiency of the optimisation may be more appropriate-such as initialising the control variables for half of the population of solutions with values expected to be near optimal. Our analysis shows the importance of aligning a policy or management recommendation with the appropriate scale.

'Small Data' for big insights in ecology.

Big Data science has significantly furthered our understanding of complex systems by harnessing large volumes of data, generated at high velocity and in great variety. However, there is a risk that Big Data collection is prioritised to the detriment of 'Small Data' (data with few observations). This poses a particular risk to ecology where Small Data abounds. Machine learning experts are increasingly looking to Small Data to drive the next generation of innovation, leading to development in methods for Small Data such as transfer learning, knowledge graphs, and synthetic data. Meanwhile, meta-analysis and causal reasoning approaches are evolving to provide new insights from Small Data. These advances should add value to high-quality Small Data catalysing future insights for ecology.

'Raising the Pulse': The environmental, nutritional and health benefits of pulse-enhanced foods.

Diet is a key modulator of non-communicable diseases, and food production represents a major cause of environmental degradation and greenhouse gas emissions. Yet, 'nudging' people to make better food choices is challenging, as factors including affordability, convenience and taste often take priority over the achievement of health and environmental benefits. The overall 'Raising the Pulse' project aim is to bring about a step change in the nutritional value of the UK consumers' diet, and to do so in a way that leads to improved health and greater sustainability within the UK food system. To achieve our objectives, UK-specific faba bean production systems that optimise both end users' diets and environmental and economic sustainability of production will be implemented in collaboration with key stakeholders (including industry, the retail sector and government). Palatable faba bean flours will be produced and used to develop 'Raising the Pulse' food products with improved nutritional profile and environmental value. Consumer focus groups and workshops will establish attitudes, preferences, drivers of and barriers to increased consumption of such products. They will inform the co-creation of sensory testing and University-wide intervention studies to evaluate the effects of pulses and 'Raising the Pulse' foods on diet quality, self-reported satiety, nutritional knowledge, consumer acceptance and market potential. Nutrient bioavailability and satiety will be evaluated in a randomised-controlled postprandial human study. Finally, a system model will be developed that predicts changes to land use, environment, business viability, nutrition and human health after substitution of existing less nutritionally beneficial and environmentally sustainable ingredients with pulses. Government health and sustainability priorities will be addressed, helping to define policy-relevant solutions with significant beneficial supply chain economic impacts and transformed sustainable food systems to improve consumer diet quality, health and the environment.

Multi-objective optimization as a tool to identify possibilities for future agricultural landscapes.

Agricultural landscapes provide many functions simultaneously including food production, regulation of water and regulation of greenhouse gases. Thus, it is challenging to make land management decisions, particularly transformative changes, that improve on one function without unintended consequences for other functions. To make informed decisions the trade-offs between different landscape functions must be considered. Here, we use a multi-objective optimization algorithm with a model of crop production that also simulates environmental effects such as nitrous oxide emissions to identify trade-off frontiers and associated possibilities for agricultural management. Trade-offs are identified in three soil types, using wheat production in the UK as an example, then the trade-off for combined management of the three soils is considered. The optimization algorithm identifies trade-offs between different objectives and allows them to be visualised. For example, we observed a highly non-linear trade-off between wheat yield and nitrous oxide emissions, illustrating where small changes might have a large impact. We used a cluster analysis to identify distinct management strategies with similar management actions and use these clusters to link the trade-off curves to possibilities for management. There were more possible strategies for achieving desirable environmental outcomes and remaining profitable when the management of different soil types was considered together. Interestingly, it was on the soil capable of the highest potential profit that lower profit strategies were identified as useful for combined management. Meanwhile, to maintain average profitability across the soils, it was necessary to maximise the profit from the soil with the lowest potential profit. These results are somewhat counterintuitive and so the range of strategies supplied by the model could be used to stimulate discussion amongst stakeholders. In particular, as some key objectives can be met in different ways, stakeholders could discuss the impact of these management strategies on other objectives not quantified by the model.

The landscape model: A model for exploring trade-offs between agricultural production and the environment.

We describe a model framework that simulates spatial and temporal interactions in agricultural landscapes and that can be used to explore trade-offs between production and environment so helping to determine solutions to the problems of sustainable food production. Here we focus on models of agricultural production, water movement and nutrient flow in a landscape. We validate these models against data from two long-term experiments, (the first a continuous wheat experiment and the other a permanent grass-land experiment) and an experiment where water and nutrient flow are measured from isolated catchments. The model simulated wheat yield (RMSE 20.3-28.6%), grain N (RMSE 21.3-42.5%) and P (RMSE 20.2-29% excluding the nil N plots), and total soil organic carbon particularly well (RMSE3.1-13.8%), the simulations of water flow were also reasonable (RMSE 180.36 and 226.02%). We illustrate the use of our model framework to explore trade-offs between production and nutrient losses.