Forecasting Nocturnal Wind Jets and Air Quality

ISEF Category: Earth and Environmental Sciences

Subcategory: Atmospheric Science  ·  Difficulty: Advanced  ·  Setup: University Lab  ·  Time: Full Year

The Hook

Some of the worst air you breathe can arrive after sunset. A quiet wind layer high above the ground can mix pollution downward and change what happens by morning. That makes nocturnal low-level jets a sharp target for a data science project. You can turn weather profiles and air-quality records into a forecast.

What Is It?

A nocturnal low-level jet is a fast ribbon of wind that forms above the ground at night. Think of it like a moving conveyor belt in the lower atmosphere. It can shift heat, moisture, and pollution across a region without you feeling much at street level.

This project asks you to detect those jets in vertical wind profile data, then predict when they will appear the next day. ERA5 is a global reanalysis dataset, which means it blends weather observations with a physics-based model to estimate past atmospheric conditions. You can use those profiles as inputs to a transformer model, a machine learning model that can learn patterns in sequences.

You can then compare your jet predictions with surface PM and ozone data. PM means fine particles in the air, and ozone at ground level forms through sunlight-driven chemistry. If the model flags jet onset before pollution rises, you have a strong link between upper-air structure and surface air quality.

Why This Is a Good Topic

This is a strong science fair topic because you can define the pattern, measure it, and test predictions with public data. You do not need a wet lab, but you still work with real atmospheric science and real environmental impact. The topic connects weather, air pollution, and machine learning, so you can show both scientific understanding and data analysis skill.

Research Questions

• How does nocturnal low-level jet strength relate to next-day surface PM spikes?
• What is the effect of vertical wind shear on the chance of jet onset the next morning?
• Does adding humidity and temperature profiles improve transformer forecasts of jet occurrence?
• To what extent does the timing of jet formation predict ozone changes later that day?
• Which ERA5 pressure levels give the best signal for identifying low-level jets?
• How does forecast skill change when you train on one agricultural region and test on another?

Basic Materials

• Laptop with enough storage for weather and air-quality data
• Internet access for downloading ERA5 and surface monitoring data
• Spreadsheet software for quick checks and summaries
• Python installed on your computer
• Jupyter Notebook or Google Colab for code and notes
• External hard drive or cloud storage for large files.

Advanced Materials

• Access to a university cluster or high-memory workstation
• Python environment with machine learning libraries
• NetCDF tools for handling ERA5 files
• GIS software for mapping station and grid locations
• Data from surface PM and ozone monitors
• Version control system such as Git for tracking model changes.

Software & Tools

• Python: Cleans ERA5 data, builds features, and trains the prediction model.
• Jupyter Notebook: Keeps code, plots, and notes in one place.
• pandas: Organizes time series and station data for analysis.
• scikit-learn: Builds baseline models for comparison with the transformer.
• TensorFlow: Trains sequence models for forecasting jet onset.

Experiment Steps

1. Define how you will label a low-level jet from ERA5 wind profiles.
2. Choose the region, season, and date range that match your research question.
3. Build a baseline forecast first so you can judge whether the transformer helps.
4. Decide which atmospheric features will go into the model and which will stay out as controls.
5. Plan how you will link jet predictions to surface PM and ozone without mixing up cause and timing.
6. Set up a validation scheme that tests the model on dates or locations it never saw during training.

Common Pitfalls

• Using too few pressure levels, which can hide the wind maximum that defines the jet.
• Mixing forecast hours and calendar dates, which can make next-day labels wrong.
• Training on all years at once, which causes data leakage and inflates accuracy.
• Comparing air-quality data from stations too far from the ERA5 grid point, which weakens the link to local jets.
• Skipping a simple baseline model, which makes it hard to prove the transformer adds value.

What Makes This Competitive

A competitive project would do more than predict a yes-or-no jet flag. You would test several label definitions, compare against strong baselines, and report skill with metrics that fit an imbalanced forecast problem. You could also check whether the model generalizes across seasons or counties, then connect the forecast to air-quality spikes with a careful lag analysis. That kind of design shows you understand both the atmosphere and the data.

Project Variations

• Use only summer months over one plains region and test whether jet detection improves ozone forecasts.
• Swap ERA5 for radiosonde archives where available and compare the model against reanalysis-based labels.
• Change the target from jet onset to jet intensity and see whether stronger jets line up with larger PM increases.

Learn More

• ERA5 Reanalysis from the European Centre for Medium-Range Weather Forecasts: Search the ECMWF site for dataset guides and variable descriptions.
• NOAA Air Quality Data: Search NOAA and AirNow resources for surface PM and ozone monitoring data.
• NASA Earthdata: Search Earthdata for atmospheric data tutorials and remote sensing support material.
• PubMed: Search for review articles on low-level jets, pollution transport, and boundary-layer mixing.
• MIT OpenCourseWare Atmospheric Science courses: Look for free lecture notes on boundary layers, winds, and weather prediction.
• USGS Water and climate data portals: Search USGS resources for background on regional environmental conditions and data handling.