Spatial and temporal optimization of potato planting based on on-farm collected data and field experiments

Context: Managing large farms with many different heterogeneous fields is a complex task. To maximize profits, farmers have to make trade-offs in their management strategy that take into account costs, constraints and the expected yield. A particularly challenging management task is planning the potato planting period, because the decisions within this period highly influence potato yield. These decisions pertain to the planting distance, seed size and the planting date, among other variables. However, it is not straightforward to determine how large the influence of these decisions actually is, especially given the diversity in soil conditions within a farm. Objective: With an increasing number of farmers that collect data, opportunities arise to optimize the decisions in the planting period: the effect of these decisions can be quantified under farmer's conditions, which can then be used to provide farm-specific guidance for this specific challenge. In this paper we propose a flexible data-driven approach to optimize decisions in the planting period such that farmer's profit is maximized. Methods: This approach is tailored to an important case study of a large potato farm in The Netherlands, comprising a total of 600 ha, and its main principles can be transferred to other use cases. The approach consists of three steps: (i) formulation of the initial optimization problem by identifying function parameters and constraints, and using these to construct an objective function, (ii) estimation of objective function parameters by first identifying knowledge and data gaps due to selection bias in the on-farm collected data. Based on this identification, field experiments are set up and analyzed, and on-farm collected data are analyzed to obtain estimates of the parameters, and (iii) optimize the farm management task, which is the planting period. From the data analysis we conclude that the cost function for the optimization in (iii) can be simplified and, accordingly, the proposed optimization takes such a simplified cost into account. Results and conclusions: When using the optimized strategy for the planting period, the farmer can gain an additional 1.5% profit in a dry year and 2.5% in a wet year compared to the farmer's strategy. Significance: This indicates that using optimization techniques combined with data science and agronomic knowledge can result in locally relevant and practical guidance for farmer, illustrating the scientific and practical potential of this cooperation between these different domains.