Environmental Monitor

A geospatial ML pipeline for monitoring and forecasting spectral indices derived from Sentinel-2 satellite imagery. The time series of NDVI, BSI, NDMI, and NBR are computed and stored as parquet files in S3. A web app hosted on Hugging Face Spaces visualizes the time series for registered areas of interest (AOIs), with XGBoost forecasts and model metrics.

➡️ Check out the slides from a recent presentation:

➡️ View the live dashboard.

Architecture

GitHub Actions (every 2 weeks)
    └── pipeline.py
            ├── data_download.py  →  S3: {country}/{aoi}/ts/*.parquet
            ├── forecast_ts.py    →  S3: {country}/{aoi}/ml/model_*.pkl
            │                        S3: {country}/{aoi}/ml/metrics_*.json
            │                        S3: {country}/{aoi}/ml/forecast_*.parquet
            └── aois.json         →  S3: aois.json

Hugging Face Spaces (always on)
    └── app.py  ←  reads S3 on page load via DataReader

Indices

Index	Description
NDVI	Normalized Difference Vegetation Index
BSI	Bare Soil Index
NDMI	Normalized Difference Moisture Index
NBR	Normalized Burn Ratio

Adding a new AOI

from scripts.pipeline import Pipeline

p = Pipeline(country="syria", aoi_name="Aleppo", bbox=[...])
p.run(lat=36.2021, lon=37.1343, rad=1000)

The AOI is registered in s3://environment-monitor/aois.json and will appear in the dashboard dropdown on next page load. Future pipeline runs will include it automatically.

Environment variables (set as Space secrets)

Variable	Description
AWS_ACCESS_KEY_ID	IAM user access key
AWS_SECRET_ACCESS_KEY	Corresponding secret
AWS_DEFAULT_REGION	e.g. us-east-1
BUCKET_NAME	S3 bucket name

Local development

cp .env.example .env          # fill in your credentials
pip install -r requirements-dashboard.txt
python app.py                 # opens on http://localhost:7860

Pipeline (separate from dashboard)

pip install -r requirements.txt
python run_pipeline.py                  # runs all AOIs
python run_pipeline.py --aoi Damascus  # runs one AOI