Ozone Spike Mapping with Air Quality Data

ISEF Category: Chemistry

Subcategory: Environmental Chemistry  ·  Difficulty: Advanced  ·  Setup: School Lab  ·  Time: Full Year

The Hook

Ozone can jump in one neighborhood while the next block looks normal. That makes air pollution feel a lot like weather, local, messy, and hard to pin down. You can use open data and sensor readings to ask where those spikes come from and when they happen. That gives you a real science problem with real stakes for health.

What Is It?

This project asks you to track ozone, a reactive gas that can irritate lungs and damage plants. Ozone near the ground does not come straight from a tailpipe. Sunlight helps precursor chemicals, especially VOCs, react in the air and form ozone later, often away from the original source.

Think of it like a recipe that finishes cooking after the ingredients have left the kitchen. You are not just measuring the final dish. You are trying to connect the timing, location, and weather conditions to the ingredients that likely caused it.

You would use EPA Air Quality System data, then build a model that looks for patterns across space and time. You can compare those patterns with readings from a low-cost PurpleAir sensor and a Pocket-Lab style sensor setup, then see how well your model matches nearby measurements.

Why This Is a Good Topic

This is a strong science fair topic because the question is measurable, the data are public, and the analysis can go far beyond a simple graph. You can study how ozone changes with weather, traffic patterns, and nearby VOC-rich sources. You also get practice with data cleaning, model testing, and sensor validation, which are real research skills. A student can make progress even without a wet lab, as long as they are willing to learn data analysis carefully.

Research Questions

• How does neighborhood location affect the timing and size of ozone spikes?
• What is the effect of temperature and sunlight on ozone spike frequency?
• Does adding wind direction improve a model that predicts ozone spikes from EPA data?
• To what extent do low-cost sensor readings agree with EPA ozone trends after calibration?
• Which weather variables best explain short-term ozone peaks in one neighborhood compared with another?
• What is the effect of nearby VOC source proxies, such as traffic density or land use, on predicted ozone spikes?
• To what extent does a spatiotemporal model outperform a simple time-only baseline for ozone prediction?

Basic Materials

• Laptop with internet access.
• EPA Air Quality System data access.
• PurpleAir sensor or another low-cost air quality sensor with downloadable data.
• Pocket-Lab style air quality sensor or equivalent basic field sensor.
• Spreadsheet software for cleaning and plotting data.
• Python installed with pandas, matplotlib, scikit-learn, and statsmodels.
• Map data for your study area, such as census tract or ZIP code boundaries.
• Notebook for tracking data sources, model choices, and results.

Advanced Materials

• University or school lab access for sensor calibration checks.
• Reference ozone monitor data if available through a local air district or research partner.
• Portable weather station for local temperature, humidity, wind, and pressure logging.
• Gas-phase sampling or sorbent tube access for VOC proxy comparison.
• GIS software for spatial mapping and interpolation.
• Python with geospatial packages such as geopandas, rasterio, and xgboost or another ML library.
• Data storage and version control system for large multi-source datasets.

Software & Tools

• Python: Cleans EPA and sensor data, fits models, and runs validation tests.
• pandas: Organizes time-stamped pollution and weather records into analysis-ready tables.
• matplotlib: Plots ozone trends, sensor comparisons, and model errors.
• scikit-learn: Trains and cross-validates baseline and machine learning models.
• QGIS: Maps monitoring sites, land use, and neighborhood patterns.

Experiment Steps

1. Define the exact ozone pattern you want to predict, then choose one neighborhood-scale outcome that can be measured consistently.
2. Gather EPA, sensor, weather, and location data, then decide how you will align them by time and place.
3. Build a simple baseline first, so you can test whether the machine learning model really adds value.
4. Choose the features that represent possible precursor sources, weather drivers, and local context.
5. Plan a validation method that compares predictions against held-out dates, sites, or neighborhoods.
6. Decide how you will test sensor agreement, handle missing data, and report uncertainty.

Common Pitfalls

• Mixing ozone data with PM2.5 sensor data, which makes the target variable unclear.
• Comparing readings that are not time-aligned, which can hide the real spike pattern.
• Using too many correlated weather features, which can make the model look better than it really is.
• Treating PurpleAir as a direct ozone monitor without checking what the sensor can and cannot measure.
• Ignoring neighborhood boundaries and source proxies, which turns a spatial project into a generic air quality chart.

What Makes This Competitive

A strong project will do more than predict pollution levels. It will separate weather effects from source effects, test a simple baseline against a smarter model, and explain why the model works or fails in specific places. Good entries also check uncertainty, sensor bias, and spatial transfer, not just overall accuracy. If you can compare neighborhoods with different land use patterns, your project starts to look like real environmental research.

Project Variations

• Focus on one city and compare ozone spikes across neighborhoods with different traffic density.
• Swap in wildfire smoke or heat wave periods and test whether the model still predicts ozone patterns well.
• Compare a machine learning model with a classic regression model to see which gives better neighborhood-level ozone forecasts.

Learn More

• EPA Air Quality System (AQS): Search the EPA site for official air pollution monitoring data and metadata.
• EPA AirNow: Use this site to understand live air quality reporting and ozone basics.
• NOAA Climate Data Online: Find weather data that can help explain ozone formation and transport.
• NASA Earthdata: Explore satellite and atmospheric datasets that may help with regional context.
• PubMed: Search for review articles on ground-level ozone, VOCs, and air pollution exposure.
• USGS National Map: Find boundary and land use layers for mapping neighborhoods and monitoring sites.