CAPTAIN v.3.0 – Preview

Conservation Area Prioritization Through Artificial INtelligence

What is CAPTAIN?

CAPTAIN is a reinforcement learning system for optimizing conservation and restoration strategies in space and time.

Key Features in v.3

⚙️ Engineered for Scale: Complete rewrite for high performance. Full GPU support and optimized sparse matrix operations allow for analysis at larger scales and finer resolutions.

🌱 Dynamic Environment Scenarios: Advanced support for time-evolving scenarios, incorporating climate change projections and dynamic implementation costs.

🧩 Modular & Customizable: A flexible architecture allowing for highly tailored conservation policies and seamless integration of custom spatial data.

🎯 Multi-objective Optimization: Full support to quantify synergies and trade-offs between competing conservation and restoration targets.

⚠️ Notice

This repository contains a preview of CAPTAIN v.3.

• Status: This is an active beta. The core simulation engine and training loops are functional, but the API and internal architecture are subject to change.
• Development: Primary development occurs in a private repository. This public mirror is provided for demonstration, testing, and early-access feedback.
• Support: Please report bugs via GitHub Issues, but note that feature requests may be prioritized based on our internal development roadmap.

Installation

1. Install uv

uv is a high-performance Python package manager that ensures consistent environments across Windows, macOS, and Linux.

• macOS / Linux:

  curl -LsSf [https://astral.sh/uv/install.sh](https://astral.sh/uv/install.sh) | sh

• Windows (PowerShell):

  powershell -ExecutionPolicy ByPass -c "irm [https://astral.sh/uv/install.ps1](https://astral.sh/uv/install.ps1) | iex"

2. Set up the Environment

Once uv is installed, clone the repository and sync the dependencies. This will automatically create a virtual environment (.venv) and install the correct version of PyTorch.

# Clone and enter the repository
git clone https://github.com/captain-project/captain3preview
cd captain3preview

# Sync dependencies and create virtual environment
uv sync

🛠️ Troubleshooting - Windows-Specific Setup

If you are on Windows, you may need to perform two quick steps to ensure uv and Python work correctly:

Enable Long Paths: Windows has a default 260-character limit for file paths. High-performance libraries like PyTorch often exceed this.

Run PowerShell as Administrator and execute:

New-ItemProperty -Path "HKLM:\SYSTEM\CurrentControlSet\Control\FileSystem" -Name "LongPathsEnabled" -Value 1 -PropertyType DWORD -Force

Execution Policy: If PowerShell blocks the uv command, run:

Set-ExecutionPolicy -ExecutionPolicy RemoteSigned -Scope CurrentUser

Examples

Working example scripts are in the examples folder

• plot_input_data.py - Visualize spatial and time-varying input data
• train_policy.py - Full training loop with real data
• run_inference.py - Load a trained model and perform optimization

Example data can be downloaded here. A pre-trained model is available here.

Project Structure

captain3preview/
├── captain/            # Main package
│   ├── agents/         # Policy network, feature extraction, rewards
│   ├── algorithms/     # Evolution strategies trainer, episode runner
│   ├── data/           # SpatialData, ExtinctionRisk classes
│   ├── environment/    # BioEnv simulation engine
│   └── utils/          # Utilities, data loading
└── examples/           # Usage examples

Citation

If you use CAPTAIN v.3, please cite:

@software{captain_3_2026,
  title = {CAPTAIN v.3 beta: Conservation Area Prioritization Through Artificial INtelligence},
  year = {2026},
  url = {https://github.com/captain-project/}
}

@article{silvestro2022improving,
  title={Improving biodiversity protection through artificial intelligence},
  author={Silvestro, Daniele and Goria, Stefano and Sterner, Thomas and Antonelli, Alexandre},
  journal={Nature sustainability},
  volume={5},
  number={5},
  pages={415--424},
  year={2022},
  publisher={Nature Publishing Group UK London}
}

@article{silvestro2025using,
  title={Using artificial intelligence to optimize ecological restoration for climate and biodiversity},
  author={Silvestro, Daniele and Goria, Stefano and Rau, E-ping and Ferreira de Lima, Renato Augusto and Groom, Ben and Jacobsson, Piotr and Sterner, Thomas and Antonelli, Alexandre},
  journal={bioRxiv},
  pages={2025--01},
  year={2025},
}

License

This project is licensed under Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International, see full license for detail.

For commercial licensing inquiries or permission to deviate from these terms, please contact the development team.