GeoScout 🛰️

AI-powered mineral exploration platform using Sentinel-2 satellite imagery, LangGraph agents, and Claude AI.

Overview

GeoScout analyzes satellite imagery for three major mineral exploration sites:

1. Mountain Pass, CA - Rare Earth Elements (REE) in carbonatite
2. Bingham Canyon, UT - Porphyry copper-molybdenum-gold-silver deposit
3. Bokan Mountain, AK - REE and uranium in peralkaline granite

Features

• 📡 Satellite Data Access - MCP server for Sentinel-2 imagery
• 🤖 AI Agent - LangGraph agent powered by Claude for intelligent analysis
• 📊 Mineral Indices - Calculate iron oxide, clay, ferrous, and vegetation indices
• 🗺️ Interactive UI - Streamlit web app with maps and visualizations
• 🔬 Alteration Analysis - Identify hydrothermal alteration zones

Architecture

geoscout/
├── mcp_server/
│   └── server.py           # MCP server exposing satellite data tools
├── agent/
│   └── mineral_agent.py    # LangGraph agent with Claude AI
├── ui/
│   └── app.py              # Streamlit web interface
└── data/
    └── (Sentinel-2 GeoTIFF files)

Installation

1. Prerequisites

• Python 3.8+
• Anthropic API key (for Claude)
• Google Earth Engine account (for downloading imagery)

2. Install Dependencies

pip install earthengine-api geemap mcp langgraph langchain-anthropic streamlit folium streamlit-folium rasterio

3. Set Environment Variables

export ANTHROPIC_API_KEY="your-api-key-here"

Usage

Step 1: Download Satellite Imagery

First, register your Google Cloud project with Earth Engine:

1. Visit: https://console.cloud.google.com/earth-engine/configuration?project=gen-lang-client-0926459788
2. Register the project for Earth Engine access

Then download imagery:

python download_sentinel2.py

This downloads Sentinel-2 scenes (bands 2,3,4,8,11,12) for all three sites to geoscout/data/.

Step 2: Run the Streamlit UI

cd geoscout/ui
streamlit run app.py

The web interface will open at http://localhost:8501

Step 3: Use the MCP Server (Optional)

The MCP server can be run standalone for programmatic access:

cd geoscout/mcp_server
python server.py

Step 4: Use the AI Agent (CLI)

Run the agent from command line:

cd geoscout/agent
python mineral_agent.py "Analyze Mountain Pass, CA for rare earth minerals"

Streamlit UI Features

1. Map View

• Interactive map showing all three exploration sites
• Click markers to see site details
• View coordinates, geology, and target minerals

2. Mineral Indices

• Calculate spectral indices from Sentinel-2 bands:
  • Iron Oxide Index (Red/Blue): Oxidized minerals, gossans
  • Ferrous Index (SWIR1/NIR): Iron-bearing minerals
  • Clay Index (SWIR1/SWIR2): Hydrothermal alteration
  • NDVI: Vegetation cover
• View statistical summaries and visualizations

3. Alteration Zones

• Identify hydrothermal alteration zones
• Calculate coverage percentages
• Get exploration recommendations
• Find multi-alteration overlap zones (high priority targets)

4. AI Chat

• Ask natural language questions about sites
• Get intelligent analysis from Claude
• Examples:
  • "What mineral indices should I calculate for this site?"
  • "Analyze the alteration patterns at Bingham Canyon"
  • "Compare spectral signatures across different zones"

MCP Server Tools

The MCP server exposes three tools:

1. get_sites

Returns information about all available sites with coordinates, minerals, and data availability.

2. get_image

Get satellite imagery with visualization options:

• rgb - Natural color
• false_color - NIR-Red-Green composite
• iron_oxide - Iron oxide composite
• ferrous - Ferrous minerals composite
• clay - Clay alteration composite

3. calculate_mineral_index

Calculate spectral indices:

• iron_oxide - Iron oxide minerals
• ferrous - Ferrous minerals
• clay - Clay alteration
• vegetation - NDVI

LangGraph Agent Tools

The agent has four tools:

1. get_available_sites() - List all sites with details
2. calculate_mineral_indices(site_key) - Calculate all indices for a site
3. get_spectral_signature(site_key, percentile) - Extract bright pixel signatures
4. analyze_alteration_zones(site_key) - Comprehensive alteration analysis

Sentinel-2 Bands Used

Band	Wavelength	Resolution	Purpose
B2	490 nm (Blue)	10m	Iron oxide index, clay minerals
B3	560 nm (Green)	10m	Vegetation, RGB visualization
B4	665 nm (Red)	10m	Iron oxide, RGB visualization
B8	842 nm (NIR)	10m	Vegetation, ferrous minerals
B11	1610 nm (SWIR1)	20m	Clay, hydroxyl minerals
B12	2190 nm (SWIR2)	20m	Clay, carbonate minerals

Mineral Index Interpretation

Iron Oxide Index (Red/Blue)

• High (>1.5): Strong oxidation, gossans, limonite
• Moderate (1.2-1.5): Some oxidation
• Low (<1.2): Minimal oxidation

Clay Alteration Index (SWIR1/SWIR2)

• High (>1.2): Strong hydrothermal alteration (phyllic/argillic)
• Moderate (1.0-1.2): Moderate alteration
• Low (<1.0): Minimal alteration

Ferrous Index (SWIR1/NIR)

• High (>1.1): Iron-bearing minerals (magnetite, pyroxene)
• Useful for mafic/ultramafic complexes

Example Workflows

Workflow 1: Initial Site Assessment

# 1. Download imagery
python download_sentinel2.py

# 2. Launch UI
cd geoscout/ui
streamlit run app.py

# 3. In UI:
#    - View site on map
#    - Calculate mineral indices
#    - Review statistics

Workflow 2: Detailed Exploration Analysis

# 1. Use AI agent for comprehensive analysis
cd geoscout/agent
python mineral_agent.py "Perform complete exploration analysis for Bingham Canyon including alteration zones and spectral signatures"

# 2. Review results and identify targets
# 3. Use UI to visualize specific indices

Workflow 3: Comparative Analysis

# Use AI chat in Streamlit UI:
# "Compare iron oxide signatures across all three sites"
# "Which site shows the strongest hydrothermal alteration?"
# "Recommend priority targets for ground surveys"

Troubleshooting

Earth Engine Registration Error

If you see: Project is not registered to use Earth Engine

Solution:

1. Visit the registration URL in the error message
2. Register your project for Earth Engine access
3. Wait a few minutes for registration to complete
4. Re-run download_sentinel2.py

Missing Satellite Data

If the UI shows "No data available":

Solution:

1. Ensure you've run download_sentinel2.py successfully
2. Check that GeoTIFF files exist in geoscout/data/
3. Verify files are named: {site_key}_sentinel2.tif

API Key Not Found

If you see warnings about ANTHROPIC_API_KEY:

Solution:

export ANTHROPIC_API_KEY="your-key-here"

Or enter the key in the Streamlit sidebar.

Site Details

Mountain Pass, California

• Coordinates: 35.4881°N, 115.5408°W
• Deposit Type: Carbonatite-hosted REE
• Key Minerals: Bastnäsite (REE-fluorocarbonate)
• Geology: Proterozoic carbonatite intrusion
• Exploration Focus: Iron oxide + REE signatures

Bingham Canyon, Utah

• Coordinates: 40.5231°N, 112.1500°W
• Deposit Type: Porphyry Cu-Mo-Au-Ag
• Key Minerals: Chalcopyrite, molybdenite, native gold
• Geology: Tertiary quartz monzonite porphyry
• Exploration Focus: Zoned alteration (potassic core, phyllic halo)

Bokan Mountain, Alaska

• Coordinates: 54.8867°N, 132.1583°W
• Deposit Type: Peralkaline granite-hosted REE-U
• Key Minerals: REE minerals, uraninite
• Geology: Jurassic peralkaline granite
• Exploration Focus: Iron oxide + SWIR anomalies

Technical Notes

Data Processing

• All indices use epsilon (1e-10) to avoid division by zero
• Percentile stretching (2-98%) for visualization
• Statistical analysis uses valid pixels only (excludes NaN/Inf)

Performance

• Image processing is done in-memory with NumPy
• GeoTIFF files are ~50-100MB per site
• Agent responses typically take 2-10 seconds

Coordinate Systems

• Input coordinates: WGS84 (EPSG:4326)
• GeoTIFF files: WGS84 (EPSG:4326)
• Sentinel-2 native: UTM (converted during download)

References

• Sentinel-2 Mission: https://sentinel.esa.int/web/sentinel/missions/sentinel-2
• Mineral Indices: https://www.usgs.gov/spectroscopy-lab
• LangGraph: https://langchain-ai.github.io/langgraph/
• MCP Protocol: https://modelcontextprotocol.io/

License

This project is for educational and research purposes. Sentinel-2 data courtesy of ESA Copernicus Program.

Contributing

Contributions welcome! Areas for enhancement:

• Additional mineral indices (ASTER, Landsat)
• Machine learning classification
• Time series analysis
• Export capabilities (KML, shapefile)
• Integration with geological databases

Contact

Built with Claude Code and Anthropic's Claude AI.