Gemini GeoFlow: AI-Powered Archaeological Site Extraction

An intelligent web application that combines Gemini 3 Flash, Google Earth Engine, and Google Maps to automate archaeological site extraction from academic papers and provide comprehensive satellite-based analysis.

---

🎯 What It Does

Upload an archaeological research paper (PDF), and the system:

1. Extracts site information using Gemini 3's multimodal capabilities
2. Analyzes satellite imagery with Gemini 3 Vision to identify archaeological features
3. Performs contextual research using Gemini 3 with search grounding to find related sites and recent findings
4. Downloads satellite data from Google Earth Engine for further analysis

All powered by Gemini 3 Flash with advanced reasoning, multimodal understanding, and real-time search capabilities.

---

🚀 Gemini 3 Integration

Core Features Used

1. Multimodal Input Processing

• PDF Analysis: Gemini 3 processes entire research papers to extract structured archaeological data
• Satellite Image Analysis: Vision capabilities analyze 6-panel satellite composites (RGB, NDVI, NDWI, BSI, DEM, Slope)
• Text + Image Reasoning: Combined modalities for comprehensive site assessment

2. Search Grounding

• Real-time web search integration for contextual enrichment
• Finds similar archaeological sites globally
• Retrieves recent research and conservation status
• All responses include verified source citations

3. Structured Outputs

• Complex JSON schema extraction from unstructured PDFs
• Nested archaeological metadata (coordinates, dating, site characteristics)
• Satellite analysis results with confidence levels and recommendations

4. Advanced Reasoning

• Coordinate format recognition and conversion (DMS, decimal degrees)
• Archaeological feature identification in satellite imagery
• Cross-referencing site data with regional archaeological patterns

Why Gemini 3 Flash?

• Speed: Sub-second response for PDF extraction and image analysis
• Multimodal: Seamlessly handles text, images, and search results
• Structured: Native JSON output with complex nested schemas
• Grounded: Search integration ensures factual accuracy with citations
• Cost-effective: Flash model provides enterprise-grade intelligence at scale

---

🛠️ Architecture

PDF Upload → Gemini 3 (Text) → Structured Site Data
                ↓
    Site Coordinates → Google Earth Engine → Satellite Data
                ↓
    Satellite Images → Gemini 3 (Vision) → Archaeological Analysis
                ↓
    Site Metadata → Gemini 3 (Search) → Contextual Research
                ↓
    Integrated Results → Google Maps Visualization

Google Ecosystem Integration

• Gemini 3 Flash: All AI processing (extraction, analysis, research)
• Google Earth Engine: Sentinel-2 imagery, SRTM DEM, spectral indices
• Google Maps: Interactive site visualization and navigation
• Google Cloud: Service account authentication for GEE

---

📋 Example Workflow

Input: Upload "Archaeological Survey of Brazilian Amazon.pdf"

Gemini 3 Extraction Output:

{
  "sites": [
    {
      "site_name": "Fazenda Colorada",
      "coordinates": {
        "raw_text": "9°57'38.96\"S, 67°29'51.39\"W",
        "latitude": -9.960822,
        "longitude": -67.497608
      },
      "temporal": {
        "dating": "1200-1400 CE",
        "cultural_period": "Late Pre-Columbian"
      },
      "characteristics": {
        "site_type": "Settlement mound",
        "features": ["circular plaza", "defensive earthworks"]
      }
    }
  ]
}

Gemini 3 Vision Analysis (on satellite imagery):

{
  "visual_features": [
    {
      "feature_type": "Circular earthwork",
      "confidence": "high",
      "location": "Center-west quadrant",
      "reasoning": "Distinct circular pattern in NDVI suggests buried structure with altered vegetation"
    }
  ],
  "archaeological_potential": "high",
  "recommendations": [
    "Ground-penetrating radar survey recommended",
    "Darker soils in BSI indicate possible terra preta (anthropogenic soil)"
  ]
}

Gemini 3 Search Grounding (contextual research):

{
  "similar_sites": [
    {
      "name": "Acre geoglyphs",
      "location": "Acre, Brazil",
      "similarity": "Geometric earthworks from same cultural period",
      "source_url": "https://www.nature.com/articles/...",
      "source_title": "Nature - Ancient Amazonian earthworks"
    }
  ],
  "recent_research": [
    {
      "summary": "LiDAR reveals extensive pre-Columbian settlements in Amazon",
      "year": "2024",
      "source_url": "https://science.org/..."
    }
  ]
}

---

🎨 Key Capabilities Demonstrated

1. Advanced Document Understanding

• Extracts precise coordinates from various formats (DMS, decimal, UTM)
• Identifies ambiguous or withheld location data
• Preserves confidence levels and source references

2. Multimodal Reasoning

• Analyzes 6 different satellite data products simultaneously
• Correlates vegetation patterns (NDVI) with terrain (DEM/Slope)
• Identifies archaeological indicators invisible to single-sensor analysis

3. Grounded Research

• Every factual claim includes source URL and title
• Distinguishes between academic, database, and general sources
• Provides reliability assessment of information quality

4. Practical Intelligence

• Converts academic text to actionable GIS data
• Generates researcher-focused recommendations
• Flags uncertainties and data quality issues

---

🔧 Setup & Deployment

Prerequisites

1. Gemini API Key

   • Get from: https://aistudio.google.com/apikey

2. Google Earth Engine Service Account

   • Create GCP project: https://console.cloud.google.com/
   • Enable Earth Engine API
   • Create service account with JSON key
   • Register at: https://code.earthengine.google.com/

Environment Configuration

Create .env file:

GEMINI_API_KEY=your-gemini-api-key
GEE_PROJECT_ID=your-gcp-project-id

Place GEE service account JSON:

# Rename to gee_service_account.json in project root
mv ~/Downloads/your-project-xxxxx.json ./gee_service_account.json

Docker Deployment

# Build and run
docker-compose up -d

# Access at http://localhost:8088

Local Development

# Install dependencies
pip install -r requirements.txt

# Run application
python app.py

---

📁 Project Structure

gemini-geoflow/
│
├── app.py                              # Flask application entry point
├── config.py                           # Configuration and GEE settings
├── requirements.txt                    # Python dependencies
├── Dockerfile                          # Container configuration
├── docker-compose.yml                  # Docker orchestration    
│
├── prompts/                            # Gemini 3 prompt templates
│   ├── extraction_prompt.py            # PDF site extraction instructions
│   ├── satellite_analysis_prompt.py    # Vision analysis guidelines
│   └── contextual_enrichment_prompt.py # Search grounding prompts
│
├── services/                           # Core business logic
│   ├── llm_service.py                  # Gemini 3 API integration
│   ├── gee_service.py                  # Earth Engine data extraction
│   └── visualization_service.py        # Satellite image rendering
│
├── routes/                             # Flask API endpoints
│   ├── extraction_routes.py            # /extract - PDF processing
│   ├── gee_routes.py                   # /preview_gee, /download_gee
│   └── analysis_routes.py              # /ai_analysis - Gemini vision + search
│
├── utils/                              # Utility functions
│   ├── coordinate_parser.py            # DMS/decimal conversion
│   └── pdf_processor.py                # PDF text extraction
│
├── templates/                          # Frontend HTML
│   └── index.html                      # Web interface
│
└── static/                             # Frontend assets
    ├── css/
    │   └── style.css                   # Application styling
    └── js/
        ├── app.js                      # Application initialization
        ├── api.js                      # Backend communication
        ├── ui.js                       # UI rendering
        ├── eventHandlers.js            # User interactions
        └── mapHelpers.js               # Google Maps integration

---

🔐 Security Notes

Never commit these files:

• .env (contains API keys)
• gee_service_account.json (GCP credentials)

Use .gitignore to exclude sensitive data.

---