Supporting Sustainable Agroecological Initiatives for Small Farmers through Constraint Programming

Margot Challand, Philippe Vismara, Dimitri Justeau-Allaire, Stéphane de TourdonnetIJCAI 2023 · DOI

TL;DR

This paper introduces a flexible Constraint Programming (CP) model to help small, diversified farms plan their crop allocation. By integrating agroecological, operational, and pedoclimatic constraints, the model offers a powerful tool for designing sustainable cropping plans that were previously too complex for farmers to manage manually.

What Problem Does This Paper Address?

Meeting global food security and nutrition goals while transitioning to sustainable agriculture is a major challenge. Agroecology offers a promising path, but its practical implementation on small farms, which produce a significant portion of the world's food, is hindered by a knowledge gap between principles and application. Specifically, small farms often feature highly diversified cropping systems (like mixed fruit tree-vegetable systems) where crop allocation is a complex, combinatorial problem. Existing decision-support tools, often based on Mixed Integer Linear Programming and focused on maximizing profit, are too simplistic for these farms. They fail to account for the intricate web of constraints that small farmers must balance, including crop rotations, beneficial/detrimental plant interactions, soil conditions, and operational limitations like labor and proximity to equipment. This paper aims to bridge this gap by developing a flexible AI-based tool to assist these farmers in designing sustainable and operationally feasible crop allocation plans.

Methods

The core of this paper is a flexible Constraint Programming (CP) model designed to solve the crop allocation problem for small, diversified farms. The approach involves:

  • Modeling the Problem: The crop allocation problem is formalized as assigning each crop from a given cropping calendar to a specific vegetable bed within a micro-farm. This involves defining vegetable beds, gardens, and the spatial adjacency between beds.
  • Constraint Catalogue: A comprehensive catalogue of constraints is developed, categorized into:
  • Agroecological: Return delays between crops of the same botanical family, avoiding negative interactions between adjacent crops, and spatial dilution of crops from the same species.
  • Pedoclimatic: Ensuring crops are allocated to beds with suitable light requirements.
  • Operational: Grouping identical crops on adjacent beds, and allocating crops with high care or monitoring needs to easily accessible beds.
  • Constraint Programming (CP) Model: The problem is encoded using CP, leveraging its expressiveness and flexibility. Key CP techniques include:
  • Decision Variables: Integer variables representing the vegetable bed assigned to each crop.
  • ALLDIFFERENT Constraints: Applied to maximal cliques of the crop interval graph to ensure no two intersecting crops are in the same bed.
  • TABLE Constraints: Used for handling negative interactions and grouping constraints.
  • Domain Reduction: For pedoclimatic constraints (like light requirements).
  • Optimization Objective: An example objective is introduced to maximize beneficial interactions between adjacent crops.
  • Implementation: The model is implemented using the Choco-solver, an open-source Java CP solver.

Datasets

The paper uses a real-world case study from an organic micro-farm in the Var department, South of France. This farm is 1 hectare and includes 80 vegetable beds grouped into 8 gardens. The cropping calendar used for the study was the list of 42 vegetable crops planned by the market gardeners for the year 2022. This case study serves as the benchmark for evaluating the model's performance under different constraint scenarios.

Key Insights and Ideas

  • CP's Suitability for Complex Agricultural Problems: The paper strongly advocates for CP as a powerful tool for agricultural decision support, particularly for small, diversified farms. Its flexibility and expressiveness are crucial for handling the multitude of context-specific constraints that traditional optimization methods might struggle with.
  • Bridging the Agroecological-Operational Divide: The model successfully integrates both agroecological principles (like beneficial plant interactions and crop rotation) and practical operational needs (like ease of access and labor management). This holistic approach is essential for making sustainable practices viable for farmers.
  • Empowering Farmers with Choice: The CP model doesn't just provide a single solution; it allows farmers to explore diverse cropping scenarios by adjusting the set of constraints. This empowers them to make informed decisions that align with their specific farm context and priorities.
  • The Complexity of Small Farm Systems: The study highlights that small farms, despite their size, can have incredibly complex cropping systems due to high diversification. This complexity necessitates advanced computational tools that go beyond simple profit maximization.
  • Potential for Disruptive Thinking: By automating the complex task of crop allocation, the tool can free up farmers to think more critically about their practices and explore innovative, potentially more sustainable, approaches that they might not have considered due to computational limitations.

Key Results

The paper presents results from three scenarios run on the case study farm:

  • Scenario 1 (Interactions): Focused on negative interactions and maximizing positive ones. The solver struggled to find an optimal solution within the 1-hour time limit, suggesting the complexity of this scenario.
  • Scenario 2 (Rotations & Dilution): Incorporated agroecological constraints like crop rotation by botanical family and spatial dilution. The constraint satisfaction problem was solved quickly on a single thread.
  • Scenario 3 (Facilitating Management): Included operational constraints for easier crop management. This was also solved quickly on a single thread.

Qualitatively, the model's outputs for the scenarios differed significantly from the market gardeners' existing plan. The farmers' plan, while grouping crops by botanical family and simplifying rotations, overlooked many agroecological interactions and operational efficiencies that the CP model could incorporate. For instance, the model avoided negative interactions between adjacent crops, which were present in the farmers' plan.

Limitations and Future Work

  • Solver Performance: For Scenario 1, the solver did not find an optimal solution within the given time limit, indicating that some constraint combinations can still be computationally challenging.
  • Constraint Catalogue Expansion: The authors acknowledge that the current constraint catalogue is under development and validation. They anticipate discovering many additional constraints and optimization objectives as they engage with more farmers in different contexts.
  • Participatory Validation: The current work is presented as a first step, with a crucial next phase involving participatory validation with farmers and agroecological farm designers to ensure the tool's practical applicability and appropriation.
  • Availability and Scalability: The paper notes that making such advanced tools available and accessible to small farmers globally is a significant challenge for achieving SDG 2.
  • Further Optimization Objectives: While one optimization objective (maximizing beneficial interactions) was introduced, the authors suggest exploring other objectives to further enhance the tool's utility.

Summary updated

BibTeX
@inproceedings{challand2023supporting,
  author = {Margot Challand and Philippe Vismara and Dimitri Justeau-Allaire and Stéphane de Tourdonnet},
  title = {{Supporting Sustainable Agroecological Initiatives for Small Farmers through Constraint Programming}},
  booktitle = {International Joint Conference on Artificial Intelligence},
  year = {2023},
  doi = {10.24963/ijcai.2023/657},
}

Loading related content...

Figure