Toward Systems Agroecology: Design and Control of Intercropping

Academic Background

With changing climatic conditions and the gradual depletion of natural resources such as fertile soil and water, exploring alternatives to today’s industrial monocrop farming has become essential. Intercropping (IC), a promising agricultural practice, involves growing two or more crop species together on the same piece of land. Numerous experiments have shown that, under certain conditions, intercropping can reduce soil erosion and fertilizer use, improve soil health and land management, while maintaining crop production levels. However, there is currently a lack of quantitative methods to predict, design, and control intercropping implementation under specific environmental and farming conditions, and to assess its robustness. This paper develops a quantitative approach based on methods and concepts from data science and systems biology to address these challenges.

Paper Source

This paper is co-authored by Sirio Belga Fedelia and Stanislas Leibler, affiliated with the Institute for Advanced Study and The Rockefeller University, respectively. The paper was submitted on July 29, 2024, accepted on November 13, 2024, and published on December 16, 2024, in PNAS (Proceedings of the National Academy of Sciences).

Research Process

Dataset Construction

The study first constructed a publicly available dataset containing the results of 2,258 intercropping experiments, involving 274 pairs of 69 different plant species. The data include four soil characteristics, five environmental and farming conditions, and eight traits for each of the two intercropped plants. Using dimensionality reduction techniques, the researchers simplified the 25-dimensional variable space into a few key variables, enabling accurate predictions of intercropping yields relative to monoculture.

Machine Learning and Prediction

The researchers employed a machine learning algorithm based on the Random Forest Regressor (RFR) to train on 80% of the experimental data and make predictions on the remaining 20%, achieving an R² value of 0.7. Through SHAP (Shapley Additive Explanations) value analysis, the researchers assessed the contribution of each variable to intercropping yield and identified seven major variables that significantly influenced yield predictions.

Dimensionality Reduction and Robustness Analysis

To simplify the analysis of the intercropping space, the researchers used Multidimensional Scaling (MDS) to reduce the 25-dimensional intercropping space to a 2-dimensional “Reduced Intercropping Space” (RIC Space). Through this dimensionality reduction, the researchers were able to evaluate the robustness of intercropping to external perturbations (e.g., climate change) and proposed methods to control intercropping yields by altering adjustable variables such as soil pH and planting density.

Main Results

Prediction of Intercropping Yields

The results show that a few key variables can accurately predict intercropping yields relative to monoculture. In particular, variables such as planting density, soil pH, and solar radiation significantly influence yield predictions.

Robustness of Intercropping

Using the reduced RIC space, the researchers assessed the robustness of intercropping to external perturbations. For example, when solar radiation or precipitation unexpectedly changes, the predicted intercropping yields also change accordingly. The results indicate that intercropping exhibits some robustness to external perturbations under certain conditions, but yields may significantly decline under extreme conditions.

Control of Intercropping

The researchers proposed methods to control intercropping yields by altering adjustable variables such as soil pH and planting density. By adjusting these variables, intercropping yields can be increased from low to high levels. For instance, increasing the planting density of the main crop or selecting companion crops with specific traits can effectively enhance intercropping yields.

Conclusion and Significance

This study provides a quantitative approach to the design and control of intercropping, marking an important step toward “systems agriculture.” By applying methods from data science and systems biology, the researchers were able to predict intercropping yields, assess their robustness, and propose specific measures to control yields. This research not only holds significant scientific value but also offers new insights for the development of sustainable agriculture in the future.

Research Highlights

1. Quantitative Prediction of Intercropping Yields: Using machine learning algorithms, the researchers accurately predicted intercropping yields relative to monoculture.
2. Assessment of Intercropping Robustness: Through dimensionality reduction, the researchers evaluated the robustness of intercropping to external perturbations, providing a scientific basis for addressing climate change.
3. Proposed Methods for Intercropping Control: The researchers proposed specific measures to control intercropping yields by altering adjustable variables, offering practical guidance for agricultural production.
4. The Beginning of Systems Agriculture: This study lays the foundation for the design and control of future multi-plant agricultural systems, marking an important step toward systems agriculture.

Other Valuable Information

The study also revealed the influence of plant traits (e.g., root length and genome size) on intercropping yields, providing new perspectives for further research on plant interactions. Additionally, the researchers emphasized the need for more quantitative experimental data in future studies to validate and refine the computational methods proposed in this paper, ultimately applying them to real-world agricultural production.

This research provides a significant scientific basis for the quantitative design and control of intercropping, marking an important step toward a systematic, data-driven approach to agroecology.