Toward systems agroecology: Design and control of intercropping.

In view of changing climatic conditions and disappearing natural resources such as fertile soil and water, exploring alternatives to today's industrial monocrop farming becomes essential. One promising farming practice is intercropping (IC), in which two or more crop species are grown together. Many experiments have shown that, under certain circumstances, IC can decrease soil erosion and fertilizer use, improve soil health and land management, while preserving crop production levels. However, there have been no quantitative approaches to predict, design, and control appropriate IC implementation for given particular environmental and farming conditions, and to assess its robustness. Here, we develop such an approach, based on methods and concepts developed in data science and systems biology. Our dataset groups the results of 2258 IC experiments, involving 274 pairs of 69 different plants. The data include 4 soil characteristics and 5 environmental and farming conditions, together with 8 traits for each of the two intercropped plants. We performed a dimensional reduction of the resulting 25-dimensional variable space and showed that, from a few quantities, one can predict IC yield relative to sole cultivation with good accuracy. For given environmental conditions, our computational approach can help to choose a companion plant and appropriate farming practices. It also indicates how to estimate the robustness of IC to external perturbations. This approach, together with its results, can be viewed as an initial step toward "systems agriculture," which would ultimately develop systems of multiple plant grown together in appropriately designed and controlled settings.