CESAROPS - Civilian Emergency SAR Operations

Note

AMD Lemonade Challenge

CESARops orchestrates local, open-weight models for maritime Search and Rescue (SAR). For the AMD Lemonade challenge, we have integrated Lemonade as a permanent production execution backend running on legacy AMD Polaris hardware via Vulkan, showcasing heterogeneous computing without requiring datacenter-class GPUs.

Interactive Lemonade Demonstration

Project Architecture & Ecosystem Map

CESARops is organized into a modular ecosystem, rather than a single monolithic repository. This ensures maximum flexibility for heterogeneous inference, distributed orchestration, and external integrations (like photogrammetry and drift modeling).

This README covers the core drift modeling system, but the overarching CESARops ecosystem consists of the following modular repositories:

• Orchestration & UI:
  • cesarops-orchestration: Forge, gates, mission control, fleet components (no model weights).
• Inference & AI:
  • cesarops-inference: CESARops distributed inference — scheduler, API, Forge fleet adapters.
  • nauti-inferer: LLM inference engine (native Rust serve path).
  • nauti-inferer-client: Desktop distributed AI inference client.
• Missions & APIs:
  • cesarops-missions: Lake missions — satellite, aeromag, detection CLIs.
  • cesarops-photogrammetry: COLMAP→Rust incremental SfM / dense pipeline.
  • gl-wrecks-api: Great Lakes Shipwreck Database (9,784 wrecks).
• Validation:
  • cesarops-scorecard: Public model×task scorecard.

For an in-depth view of the system in action, the interactive Lemonade page (linked above) provides the overarching architecture, live worker metrics, hardware strategies, and measured benchmarks.

---

Enhanced Drift Modeling System for Search and Rescue Operations

CESAROPS is a free, open-source drift modeling tool designed specifically for Search and Rescue (SAR) volunteer organizations. It provides robust, offline-capable drift predictions using oceanographic data and optional machine learning enhancement.

🌊 Features

Core Capabilities

• Multi-source ocean current data - LMHOFS, RTOFS, HYCOM support
• Robust data fetching - Retry logic, fallback to cached data
• Flexible particle seeding - Circular, line, and custom patterns
• Forward and backward drift modeling - Track objects forward or backtrack from found location
• Offline operation - Cached data and local storage for field use
• Export formats - CSV, KML, and comprehensive reports

Advanced Features

• Machine Learning enhancement - Optional ML corrections to physics models
• Performance tracking - Model accuracy metrics and validation
• Multi-threading - Non-blocking GUI with background processing
• Comprehensive logging - Full activity logging for analysis
• Portable design - Self-contained with minimal dependencies

Physical Factors Modeled

• Ocean/lake surface currents
• Windage effects (surface wind drag)
• Stokes drift (wave-induced transport)
• Configurable object properties

🚀 Quick Start

Windows Installation

1. Download and extract CESAROPS files to a folder
2. Run the installer: Double-click install_cesarops.bat
3. Follow prompts to install optional components
4. Launch: Double-click run_cesarops.bat

Manual Installation

# Clone or download the project
git clone <repository-url>
cd cesarops

# Create virtual environment
python -m venv venv
source venv/bin/activate  # Linux/Mac
# or
venv\Scripts\activate     # Windows

# Install dependencies
pip install -r requirements.txt

# Run application
python cesarops_enhanced.py

📋 System Requirements

Minimum Requirements

• Python 3.8 or later
• 4 GB RAM
• 1 GB disk space
• Internet connection for data fetching

Recommended

• Python 3.10+
• 8 GB RAM (for large simulations)
• SSD storage (faster data processing)
• Reliable internet (for real-time data)

Optional Components

• scikit-learn + joblib - Machine learning enhancement
• simplekml - KML export support
• matplotlib - Enhanced plotting capabilities

🎯 Usage Guide

1. Data Sources Tab

• Select data source (LMHOFS for Great Lakes, RTOFS/HYCOM for oceans)
• Set time range (UTC timestamps)
• Define bounding box (West, East, South, North coordinates)
• Fetch data or use cached data for offline operation

2. Seed Particles Tab

• Circular seeding: Specify center point and radius for area searches
• Line seeding: Create search patterns along transects
• Configure timing: Start/end times and particle release rate

3. Run Simulation Tab

• Set parameters: Time step, duration, windage, Stokes drift
• Enable ML (if available) for enhanced predictions
• Run forward drift or backtrack simulation
• Monitor progress in real-time

4. Results Tab

• View simulation summary and statistics
• Export to CSV for GIS analysis
• Export to KML for Google Earth visualization
• Generate reports for SAR documentation

5. Settings Tab

• Configure data sources and endpoints
• Adjust performance settings
• Save configuration for team standardization

🧠 Machine Learning Enhancement

CESAROPS includes optional ML enhancement that learns from historical drift patterns to improve predictions:

Training the Model

1. Run multiple simulations to build training data
2. Use the "Train ML Model" button
3. Model automatically validates performance
4. Enhanced predictions used in future simulations

Benefits

• Improved accuracy for local conditions
• Learns systematic biases in physics models
• Adapts to regional patterns over time
• Quantified performance metrics

📊 Data Sources

Supported Ocean Models

• LMHOFS - Lake Michigan Hydrodynamic and Forecast System
• RTOFS - Real-Time Ocean Forecast System (NOAA)
• HYCOM - Hybrid Coordinate Ocean Model

Data Quality Features

• Automatic failover between data sources
• Quality checking and outlier detection
• Caching system for offline capability
• Data validation and error handling

🗂️ File Structure

cesarops/
├── cesarops_enhanced.py    # Main application
├── config.yaml            # Configuration file
├── requirements.txt        # Python dependencies
├── install_cesarops.bat    # Windows installer
├── run_cesarops.bat       # Launch script
├── data/                  # Cached ocean data
├── outputs/               # Simulation results
├── models/                # ML model files
├── logs/                  # Application logs
└── README.md             # This file

⚙️ Configuration

Edit config.yaml to customize:

erddap:
  lmhofs: "https://coastwatch.glerl.noaa.gov/erddap"
  rtofs: "https://coastwatch.pfeg.noaa.gov/erddap"
  hycom: "https://tds.hycom.org/erddap"

drift_defaults:
  dt_minutes: 10        # Time step (minutes)
  duration_hours: 24    # Simulation duration
  windage: 0.03        # Windage factor (0-0.1 typical)
  stokes: 0.01         # Stokes drift factor

seeding:
  default_radius_nm: 2.0  # Default search radius (nautical miles)
  default_rate: 60        # Seeds per hour

🎯 Use Cases

Search and Rescue Operations

• Person overboard - Model drift from last known position
• Missing vessel - Backtrack from debris field
• Aviation SAR - Ocean survival scenarios
• Mass rescue - Multiple casualty drift patterns

Planning and Training

• SAR exercise planning - Realistic scenarios
• Resource positioning - Optimize asset placement
• Training scenarios - Educational drift modeling
• Risk assessment - Evaluate drift hazards

Research Applications

• Ocean dynamics study - Validate circulation models
• Pollution tracking - Spill trajectory analysis
• Marine biology - Larval transport studies
• Climate research - Long-term drift patterns

🔧 Troubleshooting

Common Issues

"No data available"

• Check internet connection
• Verify time range (not too far in past/future)
• Try different data source
• Check bounding box coordinates

GUI freezing

• Large simulations run in background - wait for completion
• Check system memory usage
• Reduce simulation duration or particle count

ML features not working

• Install optional ML dependencies: pip install scikit-learn joblib
• Generate training data by running multiple simulations
• Check logs for ML-specific errors

Export failures

• Ensure output directory exists and is writable
• For KML export, install: pip install simplekml
• Check disk space availability

Performance Optimization

For large simulations:

• Increase time step (reduce computational load)
• Reduce particle count
• Limit simulation duration
• Use cached data when possible

For better accuracy:

• Decrease time step (10 minutes or less)
• Use recent, high-resolution data
• Enable ML enhancement
• Validate with known drift cases

📚 Scientific Background

Drift Modeling Theory

CESAROPS implements Lagrangian particle tracking with the following physics:

dx/dt = u_current + u_wind * windage + u_stokes
dy/dt = v_current + v_wind * windage + v_stokes

Where:

• u,v_current - Ocean/lake surface currents
• windage - Wind drag coefficient (typically 0.01-0.05)
• u,v_stokes - Stokes drift from surface waves

Coordinate System

• Geographic coordinates (latitude, longitude)
• Earth radius: 6,371,000 m
• Projection effects handled for accurate distances

Time Integration

• Forward Euler scheme for particle advancement
• Adaptive time stepping for numerical stability
• Configurable time steps (1-60 minutes typical)

🤝 Contributing

CESAROPS is open source and welcomes contributions:

Development Setup

1. Fork the repository
2. Create virtual environment
3. Install development dependencies
4. Make changes and test thoroughly
5. Submit pull request

Areas for Contribution

• Additional data sources (new ERDDAP endpoints)
• Enhanced ML models (neural networks, ensemble methods)
• Visualization improvements (real-time plotting)
• Mobile interfaces (tablet-friendly GUI)
• Performance optimization (parallel processing)

📄 License

This project is released under the MIT License, making it free for all SAR organizations and humanitarian use.

🆘 Support

For SAR Organizations

• Training support available for teams
• Customization for local conditions
• Integration with existing SAR systems

Getting Help

• Check this README and logs first
• Create detailed issue reports
• Include system information and error messages
• Provide sample data when possible

🏆 Acknowledgments

• NOAA/GLERL - Great Lakes ocean model data
• NOAA/NWS - Ocean forecast systems
• ERDDAP - Data server technology
• SAR communities - Requirements and validation
• Open source contributors - Code and testing

🔮 Roadmap

Planned Features

• Real-time weather integration (wind data)
• Ensemble forecasting (uncertainty quantification)
• Mobile app version (field operations)
• AIS integration (vessel tracking)
• Web-based interface (multi-user access)

Long-term Vision

• AI-powered SAR assistant - Automated decision support
• Global coverage - Worldwide ocean/lake support
• Sensor integration - Real-time drifter data
• Interoperability - Standards-based data exchange

---

CESAROPS - Saving Lives Through Better Science

For emergency SAR operations, always follow established protocols and use CESAROPS predictions as one tool among many in your decision-making process.