Wildfire Smoke Forecast Fusion for School Air Alerts

ISEF Category: Earth and Environmental Sciences

Subcategory: Other  ·  Difficulty: Advanced  ·  Setup: University Lab  ·  Time: Full Year

The Hook

Wildfire smoke can turn a normal school day into a health problem fast. The tricky part is that a forecast alone can miss what people are actually breathing outside. You can build a system that combines weather model predictions with live sensor data to make better air-quality alerts. That turns your project into a real decision tool, not just a graph.

What Is It?

This project asks a simple question with a hard answer, can you predict when smoke will affect school-day air quality well enough to send useful alerts? HRRR-Smoke is a computer model that forecasts where smoke will move. PurpleAir sensors measure tiny particles in the air, which are a common sign of smoke pollution. A Kalman filter is a math method that blends the forecast and the sensor reading, then updates the estimate when new data arrives.

Think of it like a weather app that checks both the map and the view out your window. The map tells you where smoke should go. The window tells you what is happening right now. By combining both, you can get a better estimate than either one alone, especially when local conditions shift fast.

Why This Is a Good Topic

This is a strong science fair topic because you can test it with real data, compare multiple prediction methods, and measure success with clear metrics like error, false alarms, and missed alerts. It connects to a real problem, school safety during smoke events, and it matters to families, teachers, and districts. You can also build real research skills, like time-series analysis, sensor validation, and forecast evaluation, without needing to invent a brand-new instrument.

Research Questions

• How does Kalman-filter fusion change forecast error compared with HRRR-Smoke alone?
• What is the effect of adding PurpleAir nowcasts on same-day smoke alert accuracy?
• Does the fused model reduce false alarms during low-smoke days?
• To what extent does fusion improve predictions at different distances from wildfire plumes?
• Which alert threshold best balances missed smoke days and unnecessary warnings?
• How does sensor location near roads or buildings affect agreement between PurpleAir and forecast data?
• What is the effect of using one sensor versus several sensors in the fusion model?

Basic Materials

• Laptop or desktop computer with internet access.
• Access to HRRR-Smoke forecast data through NOAA or a research archive.
• Access to PurpleAir public sensor data.
• Spreadsheet software or Python for cleaning and plotting time-series data.
• Free SMS messaging platform with an API or email-to-text option.
• School calendar or local class schedule for mapping alerts to decision times.
• Notebook for tracking forecast days, sensor locations, and alert outcomes.

Advanced Materials

• Python with pandas, NumPy, SciPy, and matplotlib.
• Jupyter Notebook for reproducible analysis.
• Kalman filter library or custom Python implementation.
• Access to multiple PurpleAir sensors across the study area.
• Archived HRRR-Smoke fields from NOAA or a university data portal.
• Ground truth air-quality records from EPA AirNow or nearby monitoring stations.
• GIS software or QGIS for mapping plume paths and sensor locations.
• Server or cloud function for automated alert testing.

Software & Tools

• Python: Cleans time-series data, runs the fusion model, and compares forecast errors.
• Jupyter Notebook: Keeps your analysis organized and easy to rerun.
• QGIS: Maps smoke paths, sensor sites, and school locations.
• ImageJ: Not needed for the core project, but useful if you later analyze plume images or screenshots.
• PubMed: Helps you find review articles on smoke exposure, particulate matter, and sensor validation.

Experiment Steps

1. Define the decision you want to support, such as whether school-day air quality is likely to cross a safety threshold.
2. Choose the one comparison that matters most, such as HRRR-Smoke alone versus fused forecast and sensor data.
3. Select your ground truth, then decide how you will judge success with error metrics, hits, misses, and false alarms.
4. Plan how you will align forecast timestamps, sensor timestamps, and local school hours so the data speak to the same decision window.
5. Build a baseline model first, then add the Kalman filter and test whether the fused estimate improves performance.
6. Design an alert rule that balances being early enough to help with being accurate enough to trust.

Common Pitfalls

• Using a PurpleAir sensor too close to traffic, which can make exhaust pollution look like wildfire smoke.
• Comparing forecasts and sensor readings at different times of day, which creates fake errors from bad alignment.
• Treating one smoke event as enough proof, which makes the result too narrow to trust.
• Ignoring nearby weather shifts, which can move smoke away from the forecast path and confuse the fusion model.
• Choosing an alert threshold without testing false alarms, which can make the SMS system too sensitive or too quiet.

What Makes This Competitive

A competitive version of this project does more than compare two lines on a graph. It tests several fusion rules, checks performance across multiple smoke events, and uses strong metrics like precision, recall, and timing error. You can also make it more serious by comparing schools in different terrain, or by asking whether some sensor locations help more than others. The best versions explain not just what worked, but when, where, and why it worked.

Project Variations

• Compare indoor and outdoor PurpleAir sensors to see which one predicts school-day decisions better.
• Test whether a simple weighted average can match or beat the Kalman filter for smoke alerts.
• Focus on one wildfire season and compare alert performance across urban, suburban, and rural school sites.

Learn More

• NOAA Air Resources Laboratory: Search for HRRR-Smoke documentation and fire weather modeling resources.
• EPA AirNow: Find air-quality data, alert thresholds, and background on particulate pollution.
• USGS Wildfire Science: Read government summaries on wildfire smoke, transport, and impacts.
• NIH PubMed: Search for review articles on wildfire smoke exposure and fine particulate matter.
• MIT OpenCourseWare: Look for free material on probability, estimation, and state-space models that support Kalman filtering.
• NASA Earthdata: Explore satellite products and tutorials related to smoke plumes and atmospheric observation.