Systematic Conservation Planning for Sustainable Land-use Policies: A Constrained Partitioning Approach to Reserve Selection and Design.

Dimitri Justeau-Allaire, Philippe Vismara, Philippe Birnbaum, Xavier LorcaIJCAI 2019 · DOI

TL;DR

This paper introduces a novel Constraint Programming (CP) approach for systematic conservation planning, enabling the flexible design of protected areas with complex spatial constraints, including buffer zones. By framing the problem as a partitioning of land into multiple regions with customizable rules, it offers a more comprehensive and adaptable tool for balancing conservation needs with socioeconomic development.

What Problem Does This Paper Address?

Environmental managers face the critical challenge of designing sustainable land-use policies that simultaneously promote socioeconomic development and conserve natural habitats. This often involves establishing protected areas like nature reserves. However, traditional approaches struggle with integrating complex spatial requirements, such as ensuring connectivity between protected areas, mitigating edge effects through buffer zones, and managing multiple levels of protection. Existing models often rely on heuristics or mixed-integer linear programming (MILP), which can be less flexible, struggle with more than a few regions, or lack explicit control over spatial attributes. This paper addresses the need for a more unified, flexible, and expressive framework for reserve selection and design that can handle arbitrary numbers of regions and complex spatial constraints, particularly the crucial buffer zone constraint.

Methods

The authors propose a Constraint Programming (CP) approach based on a partitioning formulation. Instead of focusing on a single reserve, their model partitions the entire geographical space into an arbitrary number of regions, each with its own set of constraints.

Key methodological contributions include:

  • Generic Partitioning: The core idea is to define decision variables that assign each grid cell (site) to one of n regions. This allows for any number of land-use policies (regions) to be defined simultaneously.
  • Set and Graph Variables: To efficiently model complex spatial relationships, the authors leverage set variables (for defining regions as sets of sites) and graph variables (for representing connectivity and spatial structure within regions).
  • Novel Buffer Zone Constraint: A significant contribution is the first CP formulation of a generic buffer zone constraint. This constraint ensures a specified area (the buffer) separates two other regions, and it can be composed with other constraints to create more intricate spatial arrangements. The paper also analyzes the consistency and time complexity of this constraint.
  • Extensibility: The framework is designed to be highly extensible, allowing for the seamless integration of various coverage and spatial constraints (e.g., feature coverage, connectivity, size constraints) onto any of the defined regions.

Datasets

The paper evaluates their approach using a real-world dataset from New Caledonia, a biodiversity hotspot. This dataset includes:

  • A 46 × 75 square grid representing a 60km² sensitive area.
  • Species Distribution Models (SDMs) for 173 tree species, providing information on habitat suitability.
  • Information on 50 endangered species for which SDMs were not available.
  • Data on forest fragments within the study area.

This dataset was previously used in [Justeau-Allaire et al., 2018], allowing for a comparison of their new approach.

Key Insights and Ideas

  • Flexibility is Key: The partitioning approach unlocks unprecedented flexibility in defining conservation strategies. Instead of pre-defining a single reserve, managers can now conceptualize and test multiple, interconnected land-use policies simultaneously.
  • Buffer Zones are Crucial and Modelable: The explicit and generic modeling of buffer zones is a major step forward. This allows for the practical implementation of strategies to mitigate edge effects and promote ecological connectivity, which are vital for long-term conservation success.
  • CP as a Powerful Tool for Complex Spatial Problems: The paper demonstrates that CP, with its expressive power and ability to handle complex constraints, is well-suited for the intricate challenges of conservation planning, often outperforming or complementing traditional optimization methods.
  • Progressive Scenario Exploration: The model facilitates an exploratory process, allowing researchers and managers to progressively refine scenarios, starting with simpler configurations and gradually adding complexity (like buffer zones and multiple protection levels).

Key Results

The paper presents results from several scenarios applied to the New Caledonia dataset:

  • SC1 (Original Scenario): Replicated a previous study's single-reserve design, achieving comparable results in terms of solution quality and solving time. However, it highlighted a limitation of that approach where the selected reserve was heavily concentrated on the edge of the study area.
  • SC2 & SC3 (Extended Scenarios): Introduced buffer zones to protect against edge effects. These scenarios demonstrated the model's ability to handle multi-region partitioning and buffer constraints, leading to more ecologically relevant reserve designs. SC2 (optimistic buffer) resulted in 14 connected components (CCs) for the buffer zone, while SC3 (pessimistic buffer) showed the trade-offs between core area coverage and buffer zone fragmentation.
  • SC4 (Final Scenario): Explored a four-region scenario with nested buffer zones, aiming to maximize the core protected area. This scenario yielded a solution with 9 CCs and a significantly higher proportion of the protected area in the core (22.6%) compared to SC3.
  • Performance: The CP model, implemented with the Choco-solver, was able to find solutions within a 4-hour time limit for complex scenarios, demonstrating its practical applicability for exploring numerous constraint configurations.

Limitations and Future Work

  • Optimality Proofs: While the model finds good solutions quickly, proving optimality for complex scenarios remains a challenge and is identified as a prospect for future work.
  • Integration into Decision Processes: The authors emphasize the need to work closely with conservation scientists and managers to refine the model and integrate it into real-world decision-making processes.
  • Scalability: While the current implementation shows promise, further investigation into the scalability of the model for even larger and more complex datasets might be beneficial.
  • Alternative Neighborhood Definitions: The paper mentions that other tessellation types and neighborhood definitions could be used, suggesting future work in exploring these variations.

Summary updated

BibTeX
@inproceedings{justeauallaire2019systematic,
  author = {Dimitri Justeau-Allaire and Philippe Vismara and Philippe Birnbaum and Xavier Lorca},
  title = {{Systematic Conservation Planning for Sustainable Land-use Policies: A Constrained Partitioning Approach to Reserve Selection and Design.}},
  booktitle = {International Joint Conference on Artificial Intelligence},
  year = {2019},
  doi = {10.24963/ijcai.2019/818},
}
Figure