Lichen Air Quality Index With Machine Learning

ISEF Category: Microbiology

Subcategory: Environmental Microbiology  ·  Difficulty: Advanced  ·  Setup: Home Setup  ·  Time: Full Year

The Hook

Lichens can act like living air monitors. They sit on bark and walls, then respond to pollution in ways your eyes can miss. If you can teach a computer to read those patterns, you can turn ordinary sidewalk photos into an air-quality signal.

What Is It?

Lichens are not one organism. They are a partnership between a fungus and a photosynthetic partner, often an alga. That combo makes them sensitive to air conditions, especially pollutants that settle on surfaces or move through the air. Think of them like a slow, living filter that records local environmental stress over time.

Your project asks a simple question with a modern twist. Can you photograph lichen patches, use machine learning to sort species or growth forms, and then compare those patterns with pollution data from nearby monitoring stations? The goal is to build an index, which is just a score that summarizes what the lichen community says about air quality. If your index tracks NO₂ and PM2.5 trends, you have made a field tool that mixes ecology, image analysis, and environmental data science.

Why This Is a Good Topic

This is a strong science fair topic because you can collect your own field data, train a real classifier, and test a clear link to pollution. You do not need a wet lab, but you do need careful sampling, clean labels, and thoughtful statistics. The project connects to urban air quality, which matters for asthma, traffic exposure, and environmental justice. You can also learn computer vision, data cleaning, and correlation analysis in one project.

Research Questions

• How does lichen community composition change with distance from busy roads?
• What is the effect of NO₂ levels on the abundance of pollution-sensitive lichen species?
• Does a CNN trained on smartphone photos classify common urban lichen types better than a simple color or texture rule?
• To what extent do lichen-based index scores correlate with AirNow PM2.5 measurements across neighborhoods?
• Which photo conditions produce the most accurate lichen classification, such as angle, distance, or background contrast?
• How does lichen diversity differ between tree bark, concrete walls, and shaded stone surfaces?

Basic Materials

• Smartphone with a good camera
• GPS-enabled map app
• Notebook or data sheet for field notes
• Measuring tape or estimated distance tool
• Simple field guide to common lichens
• Laptop with spreadsheet software
• Free image annotation tool
• Weather app or local air-quality app.

Advanced Materials

• Smartphone or DSLR camera with consistent manual settings
• External color card for photo standardization
• Laptop or desktop with a GPU, if available
• Python environment with TensorFlow or PyTorch
• ImageJ for image preprocessing
• QGIS for mapping sample sites
• R or Python stats libraries for correlation and mixed models
• Access to iNaturalist records and verified lichen IDs
• EPA AirNow or local monitoring data archive.

Software & Tools

• Python: Helps you train a CNN, clean data, and run statistical tests.
• ImageJ: Lets you crop images, check contrast, and measure visible lichen cover.
• QGIS: Maps sampling sites and compares lichen patterns with pollution gradients.
• RStudio: Runs correlation tests, mixed models, and plots that compare neighborhoods.
• iNaturalist: Helps you check species labels and compare your field photos with community identifications.

Experiment Steps

1. Define the air-quality question you want your lichen data to answer, then pick one pollution metric as your main outcome.
2. Choose a sampling plan that covers a range of traffic, land use, or distance from roads, so your sites do not all look the same.
3. Set a photo protocol that keeps framing, lighting, and scale as consistent as possible across all patches.
4. Build your label set from verified lichen IDs, then decide whether you will classify species, growth forms, or a simpler visual grouping.
5. Plan how you will turn images into numbers, then test whether your index predicts pollution data better than a simple richness count.
6. Design controls that rule out confounders such as shade, bark type, season, and camera angle.

Common Pitfalls

• Mixing photos from different light conditions, which makes the CNN learn brightness instead of lichen features.
• Using unlabeled or uncertain species IDs, which weakens the training set and makes the model noisy.
• Sampling only one neighborhood type, which can hide the pollution gradient you want to test.
• Comparing lichen sites to the wrong AirNow station, which breaks the link between biology and air data.
• Treating every lichen patch as independent, which can inflate your sample size when nearby photos come from the same tree or block.

What Makes This Competitive

A strong version of this project goes beyond simple presence or absence. You can build a carefully validated image classifier, compare multiple index formulas, and test whether your score predicts pollution better than traditional richness counts. The best projects also control for site type, camera conditions, and spatial clustering. If you add uncertainty analysis and test your model on a new neighborhood, your work starts to look like real environmental sensing research.

Project Variations

• Focus on traffic corridors versus parks, then compare how lichen diversity shifts with road proximity.
• Train the model on lichen growth forms instead of species, then test whether a simpler classification still tracks pollution.
• Compare bark, brick, and concrete surfaces to see whether substrate changes the air-quality signal.

Learn More

• USGS Lichens and Air Quality: Search the USGS site for background on lichen bioindicators and pollution sensitivity.
• NOAA Air Quality and Atmospheric Chemistry resources: Search NOAA for explainers and data pages related to air pollutants.
• EPA AirNow: Use local PM2.5 and NO₂ station data for matching your sampling sites.
• iNaturalist: Search verified lichen observations to help you check species names and build labels.
• PubMed: Search for review articles on lichens as bioindicators and urban air pollution.
• MIT OpenCourseWare Introduction to Deep Learning: Use free course materials to learn the basics of CNNs and image classification.