Methodological considerations for redesigning sustainable cropping systems: the value of data-mining large and detailed farm data sets at the cropping system level

Martin Lechenet,Nicolas Munier-Jolain

doi:10.15302/j-fase-2019292

Abstract

Redesigning cropping and farming systems to enhance their sustainability is mainly addressed in scientific studies using experimental and modeling approaches. Large data sets collected from real farms allow for the development of innovative methods to produce generic knowledge. Data mining methods allow for the diversity of systems to be considered holistically and can take into account the diversity of production contexts to produce site-specific results. Based on the very few known studies using such methods to analyze the crop management strategies affecting pesticide use and their effect on farm performance, we advocate further investment in the development of large data sets that can support future research programs on farming system design.

Highlights

Global agricultural production has increased over the last five decades, increasing the food availability per capita[1]
Perfect simulation models will probably never be available for supporting cropping system design, as model developers have to find a compromise in the tradeoff between (1) the number of processes simulated in the model, which is required for covering a large range of environmental situations and technical options, and (2) the risk of over-parameterization that would impair the predictive robustness of the model[30]
Considering the limitations of experimental and modeling approaches, data mining making use of large data sets collected from real cropping systems in real farms within participatory research projects could be a promising approach when addressing the issue of cropping system design for enhanced sustainability

Summary

Introduction

Global agricultural production has increased over the last five decades, increasing the food availability per capita[1]. The objective of this paper is to discuss the potential of methods based on data mining in large data sets compiling details of crop management at the farm level to produce valuable knowledge about systems sustainability. Factorial experiments can only manipulate a very limited number of factors at the same time, and are not always useful in terms of promoting a holistic view of agricultural issues and to adapt all components of cropping systems in a consistent way, so as to improve the overall sustainability. The factors studied are not the individual technologies but the consistent cropping systems designed with complex combinations of technologies This approach requires scrapping the rule of “all other things being equal” at the technology level, and scrapping the objective of demonstrating the effects of individual technologies––a point that might be difficult to acknowledge for some scientists. Perfect simulation models will probably never be available for supporting cropping system design, as model developers have to find a compromise in the tradeoff between (1) the number of processes simulated in the model, which is required for covering a large range of environmental situations and technical options, and (2) the risk of over-parameterization that would impair the predictive robustness of the model[30]

Data mining

Findings

Lessons learned