SylvaSense 🌲🛰️

Planetary Earth Observation & Forest Carbon Intelligence Platform

---

📖 Overview

SylvaSense is an end-to-end Earth Observation (EO) and Geospatial AI platform designed for high-resolution forest canopy monitoring, automated tree crown detection, above-ground biomass (AGB) estimation, and verifiable IPCC Tier-1 carbon stock accounting.

By fusing multi-spectral optical data (Sentinel-2 L2A), Synthetic Aperture Radar backscatter (Sentinel-1 SAR C-Band), Global Ecosystem Dynamics Investigation (ETH Global Canopy Height / GEDI), and digital surface elevation (Copernicus DEM 30m), SylvaSense bridges the gap between raw orbital telemetry and actionable environmental intelligence.

                  ┌────────────────────────────────────────┐
                  │      Planetary Multi-Sensor Inputs     │
                  │   Sentinel-2 (Optical/NDVI)            │
                  │   Sentinel-1 (SAR C-Band VV/VH)        │
                  │   ETH / GEDI 10m Canopy Height         │
                  │   Copernicus DEM 30m & Hansen GFC      │
                  └──────────────────┬─────────────────────┘
                                     │
                                     ▼
                  ┌────────────────────────────────────────┐
                  │       AIML & Biomass Processing        │
                  │  • DeepForest Canopy Cluster Detection │
                  │  • Chave et al. Allometric Modeling    │
                  │  • Trained XGBoost Biomass Regressor   │
                  │  • IPCC Tier-1 Carbon & CO2-eq Math    │
                  └──────────────────┬─────────────────────┘
                                     │
                                     ▼
       ┌─────────────────────────────┴─────────────────────────────┐
       │                                                           │
       ▼                                                           ▼
┌───────────────────────────────┐           ┌───────────────────────────────┐
│     FastAPI Geo Microservice  │           │   React GIS Command Center    │
│  • Zonal Statistics & Coverage│           │  • Interactive Freehand Draw  │
│  • Dynamic Tile Server        │◄─────────►│  • Live Layer Compositor      │
│  • Real-Time GeoJSON Delivery │           │  • Side-by-Side A/B Analysis  │
│  • Executive Report Exporter  │           │  • AI Environmental Copilot   │
└───────────────────────────────┘           └───────────────────────────────┘

---

✨ Key Capabilities & Features

1. Multi-Sensor Earth Observation Telemetry

   • Sentinel-2 L2A: 10-meter true color (RGB) and Normalized Difference Vegetation Index (NDVI) layers for photosynthetic activity and canopy vigor.
   • Sentinel-1 SAR: Dual-polarization C-band backscatter ($VV$ and $VH$) capturing canopy structure, surface roughness, and moisture independence from cloud cover.
   • ETH Global Canopy Height (10m): GEDI-calibrated spaceborne LiDAR canopy elevation model.
   • Hansen Global Forest Change: 2000–2023 loss/gain masks for tracking historical deforestation trajectories.

2. AI Tree Crown & Canopy Cluster Detection

   • Pretrained DeepForest neural architecture deployed over satellite and aerial orthomosaics.
   • Extraction of crown centroids, estimated canopy bounding geometries, and spatial density vectors delivered via standard GeoJSON.

3. Scientifically Grounded Biomass & Carbon Accounting

   • Allometric Crown Modeling: Implementation of the peer-reviewed Chave et al. pan-tropical equation relating crown diameter and canopy height to Aboveground Biomass (AGB).
   • Trained Gradient Boosted Regressor: Custom XGBoost model incorporating multi-band reflectance and radar backscatter to estimate wood density and biomass variation.
   • IPCC Tier-1 Standards: Exact conversion factors yielding total Dry Biomass (Tons), Carbon Stock ($0.47 \times \text{Biomass}$), and $\text{CO}_2$ Equivalent ($\frac{44}{12} \times \text{Carbon}$).

4. Dynamic User-Drawn Polygon Analysis

   • Built-in Leaflet-Draw toolbar allowing operators to trace custom freehand polygons or bounding boxes anywhere across supported forest zones.
   • Instant computation of clipped acreage, total tree counts, mean NDVI, biomass density (Tons/Ha), and sequestered carbon.

5. A/B Degradation Comparison Engine

   • Direct side-by-side comparison between intact primary forest cores and adjoining degraded/edge zones (e.g., Tapajós National Forest Core vs. Fragmented Edge).
   • Instant delta quantification showing percentage loss of canopy cover, net carbon emissions, and biomass displacement.

6. Executive GIS Command Center UI

   • Clean, modern, dark-mode command interface built with React, Tailwind CSS, and Lucide icons.
   • Live layer toggles (Optical RGB, False-Color Infrared, NDVI, Sentinel-1 SAR, Canopy Height, Tree Clusters).
   • Integrated AI Forest Assistant for instant contextual querying and anomaly diagnosis.
   • One-click executive PDF / visual report generator.

---

🔬 Scientific Rigor & The "Honesty Metric"

Important

Resolution Transparency: DeepForest and similar computer-vision crown detectors are typically trained on sub-meter aerial / UAV drone imagery (~0.1 m/pixel). Standard Sentinel-2 optical data provides 10-meter spatial resolution.

In accordance with remote sensing best practices:

• Detections over 10m Sentinel-2 data are calibrated and presented as canopy clusters / tree group centroids.
• Aboveground Biomass and Carbon values utilize calibrated empirical models backed by spaceborne LiDAR (GEDI/ETH 10m height) and validated against benchmark study plots.
• High-resolution aerial validation tiles demonstrate individual-tree delineation accuracy where high-resolution ground truth is available.

---

📸 Interface Preview

GIS Command Map & Detections	Dynamic Drawing & Zonal Analysis

Biomass & Carbon Accounting Engine	Side-by-Side A/B Degradation Comparison

---

🗺️ Pre-Configured Areas of Interest (AOIs)

SylvaSense includes fully cached, high-fidelity baseline packages across iconic global biomes:

AOI ID	Name	Biome / Description	Bounding Coordinates
aoi_mudumalai	Mudumalai Tiger Reserve (Tamil Nadu, India)	Nilgiri Biosphere tropical deciduous & dry thorn forest corridor	[76.50, 11.55, 76.65, 11.65]
aoi_1	Kudremukh National Park (Western Ghats, India)	Shola-grassland tropical montane wet evergreen rainforest	[75.20, 13.15, 75.30, 13.25]
aoi_2	Tapajós National Forest Core (Amazon Basin, Brazil)	Primary dense humid Amazonian tropical rainforest	[-55.05, -3.10, -54.95, -3.00]
aoi_3	Tapajós Agricultural Edge (Amazon Basin, Brazil)	Fragmented forest fringe adjacent to agricultural clearance	[-55.20, -3.25, -55.10, -3.15]

---

🏗️ Repository Architecture

SylvaSense/
├── aiml/                               # AI/ML & Remote Sensing Pipeline
│   ├── evaluate_accuracy.py            # Model precision & recall benchmark suite
│   ├── evaluate_generalization.py      # Cross-biome generalization tests
│   ├── gee_engine.py                   # Google Earth Engine telemetry integration
│   ├── model_zoo.py                    # Model architectures & weights catalog
│   ├── run_biomass.py                  # Allometric biomass & carbon estimation
│   ├── run_depth.py                    # Canopy height & vertical profiling
│   ├── run_detection.py                # DeepForest tree crown inference
│   ├── run_pipeline.py                 # Multi-sensor ingestion & preprocessing
│   ├── spatial_engine.py               # Vector topology & CRS coordinate math
│   ├── train_biomass_model.py          # XGBoost biomass model training script
│   └── model_cache/                    # Pretrained XGBoost models & metric weights
├── backend/                            # FastAPI Geospatial Microservice
│   ├── main.py                         # REST API endpoints & tile streaming
│   ├── requirements.txt                # Python backend dependencies
│   ├── test_api.py                     # API integration test suite
│   └── gee-service-account.template.json # Template for Earth Engine credentials
├── contracts/                          # Contract-Driven Integration Specs
│   ├── aoi_config.json                 # AOI definitions, extents, and centers
│   ├── api_spec.yaml                   # OpenAPI 3.0 specification
│   ├── geojson_schema.md               # Standardized detection schema
│   └── stats_schema.json               # JSON schema for biomass/carbon responses
├── data/                               # Cached Datasets & Reports
│   ├── processed/                      # Precomputed GeoJSONs, stats, depth profiles
│   └── thumbnails/                     # AOI preview imagery
├── docs/                               # Documentation & Visual Assets
│   └── screenshots/                    # High-resolution dashboard screenshots
└── frontend/                           # React + Leaflet Web Application
    ├── src/
    │   ├── components/                 # UI components (Map, Sidebar, Panels)
    │   ├── App.jsx                     # Master application shell & state
    │   └── main.jsx                    # Application entrypoint
    ├── package.json                    # Frontend dependencies
    └── vite.config.js                  # Vite configuration

---

🚀 Quick Start Guide

Prerequisites

• Python 3.10+
• Node.js 18+ & npm
• (Optional) A Google Earth Engine active service account or account credential for live planetary tile streaming. (SylvaSense runs seamlessly in offline cached mode out of the box).

---

1. Backend Setup

# Navigate to the backend directory
cd backend

# Create and activate a Python virtual environment
python -m venv venv

# On Windows:
.\venv\Scripts\activate
# On macOS/Linux:
# source venv/bin/activate

# Install dependencies
pip install -r requirements.txt

# (Optional) Setup GEE credentials if running live planetary queries
# Copy gee-service-account.template.json to gee-service-account.json and fill in your keys

# Start the FastAPI server
uvicorn main:app --reload --host 0.0.0.0 --port 8000

The backend documentation will now be available at http://localhost:8000/docs.

---

2. Frontend Setup

# In a new terminal window, navigate to the frontend directory
cd frontend

# Install dependencies
npm install

# Start the development server
npm run dev

Open your browser and navigate to http://localhost:5173 to access the interactive GIS Command Center.

---

3. Running Verification Tests

To verify that all API endpoints, cached AOIs, GeoJSON layers, and analytical calculations are operational:

cd backend
python test_api.py

---

🔌 API Reference Overview

Method	Endpoint	Description
GET	/api/health	Service health status and list of ready cached AOIs
GET	/api/aois	List all configured Areas of Interest with bounding boxes
GET	/api/aois/{aoi_id}/stats	Fetch tree count, AGB (Tons), Carbon, and $\text{CO}_2$ metrics
GET	/api/aois/{aoi_id}/detections	Fetch canopy crown features in GeoJSON format
GET	/api/aois/{aoi_id}/layers/{layer}	Retrieve optical RGB, NDVI, or SAR radar raster assets
GET	/api/aois/{aoi_1}/compare/{aoi_2}	Calculate comparative deltas (degradation %, biomass loss)
GET	/api/coverage	Supported satellite capture corridors footprint (GeoJSON)
POST	/api/analyze/polygon	Compute real-time biomass & tree counts for user-drawn geometries
GET	/api/gee/status	Current status of Google Earth Engine authentication

Full OpenAPI specification is available in contracts/api_spec.yaml or at /docs.

---

📊 Evaluation & Benchmarks

SylvaSense includes automated evaluation suites to benchmark detection quality and biomass estimation:

• Accuracy Benchmarks (aiml/evaluate_accuracy.py): Measures IoU, precision, recall, and F1 score against annotated forest canopy plots.
• Generalization Tests (aiml/evaluate_generalization.py): Validates model stability across diverse forest strata (evergreen rainforest vs. dry deciduous vs. fragmented margins).
• Results are automatically generated and preserved in data/processed/evaluation_report.json.

---

👥 Contributors & Hackathon Submission

Developed for Orion 1.0 '26 Hackathon — Problem Statement PS-03 (Earth Observation AI Track).

• Arul Amudhan G (@Starmann1)
• Team SylvaSense

---

📄 License

This project is licensed under the MIT License — see the LICENSE file for details.