Satellite Ozone Regimes and Action-Day Prediction

ISEF Category: Earth and Environmental Sciences

Subcategory: Atmospheric Science  ·  Difficulty: Advanced  ·  Setup: Home Setup  ·  Time: Full Year

The Hook

Air quality maps can look green at one scale and alarming at another. A single neighborhood can sit under a plume that satellite pixels barely resolve. You can use satellite chemistry data to ask which areas are set up for ozone spikes, and which ones are not.

What Is It?

Tropospheric ozone is not emitted directly, it forms in sunlight from nitrogen oxides, or NOx, and volatile organic compounds, or VOCs. If you think of ozone as a recipe, NOx and VOCs are two key ingredients. Change the mix, and you change how much ozone forms.

TROPOMI is a satellite instrument that measures gases in Earth’s atmosphere. It can estimate column amounts of formaldehyde, or HCHO, and nitrogen dioxide, or NO₂. Scientists use the HCHO/NO₂ ratio as a clue about local ozone chemistry. A higher ratio often points toward NOx-limited conditions, while a lower ratio often points toward VOC-limited conditions. That means the same pollution controls will not work the same way everywhere.

Your project can combine satellite data, census-tract geography, and air-quality records to test whether the HCHO/NO₂ ratio helps predict ozone action days. You are not just making a map. You are testing whether a chemistry signal from space lines up with a real health and policy outcome on the ground.

Why This Is a Good Topic

This topic works well because you can turn large public datasets into a clear test. You can compare a satellite-based chemistry indicator with ground-level ozone records, then ask whether the indicator predicts exceedances better than chance. The project connects to real air-quality planning, since ozone action days affect school schedules, outdoor work, and public health. You can also learn data cleaning, geospatial matching, and model evaluation, which are valuable research skills for later projects.

Research Questions

• How does the TROPOMI HCHO/NO₂ ratio vary across census tracts with different ozone action-day histories?
• What is the effect of using monthly averages versus seasonal averages of HCHO/NO₂ on exceedance prediction accuracy?
• Does adding weather variables improve a gradient-boosted model that predicts ozone action-day exceedances from satellite chemistry alone?
• To what extent does the HCHO/NO₂ ratio separate tracts that frequently exceed ozone thresholds from tracts that rarely do?
• Which feature set, satellite only, satellite plus weather, or satellite plus weather plus land use, gives the best predictive performance?
• How does model performance change when you predict exceedances one county at a time instead of across an entire region?

Basic Materials

• Laptop with at least 8 GB RAM and reliable internet access.
• Spreadsheet software for organizing tract-level data.
• Python installed through Anaconda or a similar free distribution.
• Access to public ozone monitor data from EPA AirData or state air-quality networks.
• Access to TROPOMI HCHO and NO₂ products from NASA or Copernicus data portals.
• Census tract boundary files from the U.S. Census Bureau.
• Basic mapping software such as QGIS.
• Notes file for tracking data sources, dates, and joins.

Advanced Materials

• Laptop or desktop with at least 16 GB RAM for larger geospatial joins.
• Python environment with pandas, geopandas, scikit-learn, xgboost, or lightgbm.
• GIS software for spatial overlays and tract aggregation.
• NetCDF or HDF data reader for satellite products.
• AirNow or EPA ozone data for validation against monitored concentrations.
• Land use or emissions datasets from EPA, NOAA, or USGS.
• High-resolution census and boundary layers for population-weighted analysis.
• Optional cloud storage or version control for reproducible workflow files.

Software & Tools

• Python: Cleans satellite, weather, and ozone datasets, then builds the predictive model.
• QGIS: Maps census tracts, monitor locations, and satellite-derived chemistry patterns.
• Google Earth Engine: Helps you inspect large geospatial datasets and filter satellite scenes.
• scikit-learn: Trains and evaluates gradient-boosted models and related baselines.
• ImageJ: Not needed for this topic, but useful if you later compare the project with image-based environmental data.

Experiment Steps

1. Define the ozone outcome you will predict, such as an action-day exceedance, a threshold crossing, or a monitor-based flag.
2. Choose the spatial unit you will analyze, then decide how you will assign satellite pixels and monitor data to each tract.
3. Build a clean data table that matches HCHO, NO₂, weather, and ozone records by place and time.
4. Create a baseline model first, then add the HCHO/NO₂ ratio so you can measure its real contribution.
5. Test whether the ratio works better alone, with weather, or with other land-use features.
6. Plan an error check that compares predicted hot spots with held-out data from a different season or county.

Common Pitfalls

• Using satellite pixels that do not line up well with census tracts, which blurs the chemistry signal.
• Mixing up surface ozone with ozone action-day records, which can turn the target variable into the wrong outcome.
• Ignoring cloud cover or missing retrieval flags, which leaves bad satellite values in the dataset.
• Training and testing on the same season or same county, which makes the model look better than it really is.
• Treating the HCHO/NO₂ ratio as a direct ozone measurement, which overstates what the satellite can prove.

What Makes This Competitive

A strong version of this project goes past a simple map. You would compare multiple feature sets, hold out a real geographic region for testing, and report metrics like precision, recall, and calibration, not just accuracy. You could also test whether the chemistry signal adds value beyond weather and land use, which makes the analysis much stronger. A careful uncertainty analysis, plus a clear explanation of when the model fails, can push the project into serious research territory.

Project Variations

• Use county-level instead of census-tract-level aggregation to see whether coarser geography changes the signal.
• Replace ozone action-day flags with daily maximum 8-hour ozone concentration and model the actual concentration value.
• Compare TROPOMI HCHO/NO₂ with land-use or traffic density to test whether chemistry or human activity better predicts exceedances.

Learn More

• NASA Earthdata: Search for TROPOMI tutorials, product guides, and atmospheric composition datasets through NASA Earthdata or the Earthdata search portal.
• Copernicus Sentinel-5P: Read the mission overview and product documentation for TROPOMI HCHO and NO₂ on the Copernicus data pages.
• EPA AirData: Find hourly and daily ozone monitor records, station metadata, and download tools on EPA AirData.
• NOAA Air Resources Laboratory: Look for ozone and atmospheric chemistry background materials, plus model resources, through NOAA ARL pages.
• PubMed: Search review articles on tropospheric ozone formation, HCHO/NO₂ ratios, and satellite validation studies.
• MIT OpenCourseWare: Search atmospheric chemistry and remote sensing course materials for free lecture notes and background reading.