🌳 Amazon Deforestation Monitoring – Sub-Region Analysis (Rondônia Hotspot)

Monitoring deforestation in the Brazilian Amazon using Sentinel-2 imagery and Google Earth Engine

Time period: 2023 vs 2024

Focus area: Southern Rondônia / Northern Mato Grosso / Amazonas

Tech stack: Google Earth Engine, Sentinel-2 SR, Leaflet.js, GitHub Pages, Git LFS

🔗 Live map: 👉

---

Animated Change:

---

Project Overview

This project demonstrates a complete end-to-end remote sensing workflow for detecting and visualizing potential deforestation in the Brazilian Amazon.

It combines:

• Cloud-scale satellite processing in Google Earth Engine
• NDVI-based change detection**
• Vectorization of forest-loss hotspots
• An interactive Leaflet web map for public dissemination

Due to computational and export constraints of the full Amazon basin, the analysis focuses on a high-deforestation sub-region (~300,000 km²) within the Arc of Deforestation.

---

Objectives

• Detect potential forest loss between 2023 and 2024
• Quantify the approximate forest loss area
• Visualize results in an interactive web map

Deliverables

• Public Google Earth Engine script

• Forest loss summary statistics

• Animated year-to-year visualization

• Interactive Leaflet map with:

  • Year toggle (2023 / 2024)
  • Independent deforestation hotspots layer
  • Clickable polygons with area estimates

---

Study Area

AOI: Southern Rondônia, northern Mato Grosso, and Amazonas

• Bounding box: [-65°, -13°] to [-60°, -8°]
• Approximate area: ~300,000 km²
• Known region of accelerated deforestation

Defined in GEE as:

var testAOI = ee.Geometry.Rectangle([-65, -13, -60, -8]);

Amazon Basin

---

Area of Interest:

Data & Preprocessing

Dataset

• Sentinel-2 Surface Reflectance (COPERNICUS/S2_SR_HARMONIZED)

Key parameters

• Cloud cover < 50%
• QA60 cloud & cirrus masking
• NDVI = (B8 − B4) / (B8 + B4)

Annual composite

• Per-pixel maximum NDVI quality mosaic
• Prioritizes cloud-free, healthy vegetation observations

---

Change Detection Method

1. Generate max-NDVI mosaics for 2023 and 2024

2. Compute NDVI difference:

   NDVI_2024 − NDVI_2023
   

3. Flag potential deforestation where:

   • NDVI drop ≥ 0.20
   • 2024 NDVI < 0.55

Estimated potential forest loss -> ~579 km² (coarse 100 m estimate)

---

Summary Statistics

Metric	Value
Sentinel-2 images processed (2023)	2,684
Sentinel-2 images processed (2024)	2,518
Mean NDVI (2023)	0.832
Mean NDVI (2024)	0.835
Estimated potential forest loss	~579 km²

---

Visualization Outputs

In Google Earth Engine

• False-color composites (NIR–Red–Green)
• NDVI layers
• NDVI change map
• Binary forest-loss mask

Vectorized Hotspots

• Extracted using reduceToVectors
• 60 m scale, simplified geometries
• Exported as GeoJSON (Git LFS)

Interactive Web Map

🔗 Click Here

Features

• Toggle between 2023 & 2024 imagery
• Independent deforestation hotspots checkbox
• Clickable polygons with area (ha)
• Legend and contextual info panel

---

Technical Challenges & Solutions

Challenge	Solution
Full Amazon too large	Scoped to high-priority sub-region
Export limits	Raster exports at 30 m, vectors at 60 m
GeoJSON >25 MB	Git LFS + media.githubusercontent.com fetch
Local CORS issues	Local HTTP server for development
Web performance	Geometry simplification + lazy loading

---

Limitations

• Annual comparison only (no seasonal dynamics)
• Conservative NDVI thresholds
• Coarse area estimates
• Sub-region not representative of the entire Amazon
• Small patches may be merged or lost during simplification

---

Future Improvements

• Monthly or seasonal NDVI time series
• Expanded coverage (Arc of Deforestation)
• Validation with MapBiomas / Hansen GFC
• Time slider animation in Leaflet
• Area-weighted statistics and dashboards
• Vector tiles (PMTiles) for large-scale performance

---

Conclusion

This project showcases a scalable, reproducible, and visually intuitive workflow for monitoring deforestation using open satellite data and modern web mapping tools.