Volcanic SO₂ Plume Detection With Sentinel-5P

ISEF Category: Earth and Environmental Sciences

Subcategory: Geosciences  ·  Difficulty: Advanced  ·  Setup: Home Setup  ·  Time: Full Year

The Hook

A volcano can send up a warning signal long before lava appears. Satellites can catch part of that signal as sulfur dioxide, or SO₂, drifting above the summit. You can turn those space-based measurements into a detector that spots unusual gas plumes and checks whether they appear before eruptions.

What Is It?

Volcanic SO₂ is a gas that many volcanoes release before, during, and after eruptions. Think of it like a smoke alarm for the ground, except the alarm is invisible and floats high in the atmosphere. Sentinel-5P is a satellite that measures gases in Earth’s atmosphere, so you can track SO₂ over time and look for unusual spikes near arc volcanoes, which are volcanoes along plate boundaries.

A CNN, or convolutional neural network, is a machine learning model that finds patterns in images or grids of numbers. In this project, the input can be a map or a time series of satellite observations. The model learns what normal SO₂ behavior looks like, then flags days or regions that look unusual. You can also compare the anomaly signal to eruption dates to see whether SO₂ peaks show up early enough to matter as a warning sign.

Why This Is a Good Topic

This topic works well because satellite data is free, the signal is measurable, and you can test clear questions about timing, geography, and model performance. It connects to real hazards, since volcano monitoring helps protect people and aviation. You can learn remote sensing, data cleaning, anomaly detection, and basic machine learning without needing a wet lab.

Research Questions

• How does Sentinel-5P SO₂ anomaly strength change in the days and weeks before eruptions at arc volcanoes?
• What is the effect of volcano location, such as tropical versus midlatitude sites, on SO₂ detection quality?
• Does a CNN detect pre-eruption SO₂ patterns better than a simple threshold method?
• To what extent do eruption lead times differ when you define anomalies by single-day spikes versus multi-day trends?
• Which volcanoes show repeated SO₂ precursor signals across multiple eruptions?
• How does cloud cover or data gap frequency affect the reliability of satellite SO₂ anomaly detection?

Basic Materials

• Laptop or desktop computer with internet access.
• Google Earth Engine account access or another platform for Sentinel-5P data.
• Spreadsheet software for organizing eruption dates and outputs.
• Python installed with Jupyter Notebook.
• Python libraries for data analysis, such as pandas, numpy, matplotlib, and scikit-learn.
• Public eruption catalog from the Smithsonian Global Volcanism Program.
• Sentinel-5P SO₂ data from NASA, ESA, or Google Earth Engine.

Advanced Materials

• Workstation with a dedicated GPU for CNN training.
• Python scientific stack with TensorFlow or PyTorch.
• NetCDF or HDF5 files for gridded satellite data.
• GIS software such as QGIS for map-based inspection.
• Earth Engine Python API for large-scale data pulls.
• A curated eruption dataset with timestamps, volcano coordinates, and eruption style labels.
• Cloud storage for intermediate data products and model checkpoints.

Software & Tools

• Google Earth Engine: Pulls Sentinel-5P SO₂ data and helps you filter by volcano, date, and region.
• Python: Runs preprocessing, feature extraction, anomaly detection, and model evaluation.
• Jupyter Notebook: Lets you document code, plots, and notes in one place.
• QGIS: Helps you inspect satellite maps and compare plume locations with volcano points.
• ImageJ: Can help if you turn gridded outputs into image-like inputs for quick visual checks.

Experiment Steps

1. Define the volcano set, the time window, and the eruption records you will treat as ground truth.
2. Decide how you will turn satellite SO₂ measurements into a clean time series or image grid.
3. Build a baseline detector first, so you have a simple method to compare against the CNN.
4. Plan how you will label normal periods and anomaly periods without leaking eruption dates into training.
5. Choose evaluation metrics that reward early, accurate detection, not just overall accuracy.
6. Set up a lead-time analysis that measures how far ahead of eruption your detector first flags a signal.

Common Pitfalls

• Using eruption dates without checking whether the SO₂ signal started before, during, or after the event, which breaks lead-time analysis.
• Mixing volcanoes with very different background SO₂ levels, which makes one model look better than it really is.
• Treating cloudy or missing satellite pixels as true zero values, which creates fake dips in the time series.
• Training and testing on overlapping time windows from the same eruption, which causes data leakage.
• Calling any SO₂ spike a precursor, even when wind, fires, or data noise can produce a similar pattern.

What Makes This Competitive

A stronger project goes beyond spotting spikes. You can compare multiple volcanoes, multiple eruption types, and multiple anomaly methods, then test whether one approach predicts earlier or more reliably. Good entries also handle missing satellite data, quantify uncertainty, and explain false alarms. If you add a careful lead-time study and a fair baseline comparison, your project starts to look much more like real research.

Project Variations

• Focus on a single volcanic arc, such as the Andes or Indonesia, and compare precursor signals across nearby volcanoes.
• Replace the CNN with a simpler anomaly method, such as isolation forest or seasonal thresholding, and test whether performance changes.
• Add other satellite variables, such as aerosol index or thermal anomaly data, to see whether a multi-sensor model improves eruption warning.

Learn More

• NASA Earthdata: Search for Sentinel-5P tutorials, SO₂ products, and user guides for satellite remote sensing.
• NOAA Global Monitoring Laboratory: Read background material on atmospheric sulfur dioxide and gas transport.
• Smithsonian Global Volcanism Program: Find eruption dates, volcano histories, and activity summaries for your case study sites.
• ESA Sentinel-5P mission pages: Review the instrument basics and product descriptions for TROPOMI observations.
• PubMed: Search for review articles on volcanic gas monitoring, satellite SO₂ detection, and eruption forecasting.
• Google Earth Engine Data Catalog: Look up Sentinel-5P SO₂ datasets and learn how researchers access global time series.