Harmful Algal Bloom Detection in Inland Lakes

ISEF Category: Earth and Environmental Sciences

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

The Hook

A lake can look calm from the shore and still be changing fast on satellite images. Harmful algal blooms can spread before people notice green water or a bad smell. That makes remote sensing a real tool, not just a map trick. You can test whether satellites can spot those changes early in inland lakes.

What Is It?

This project asks you to use satellite images to estimate how much chlorophyll-a is in a lake. Chlorophyll-a is the green pigment in algae and plants. More chlorophyll-a often means more algae, and sometimes that means a harmful bloom. Think of it like using the color of a crowd to guess how packed a stadium is.

Sentinel-2 is a pair of European satellites that take detailed color images of Earth. They do not measure chlorophyll-a directly. Instead, you build a model that connects reflectance, the light a surface sends back, to known lake data. A random forest model is a machine learning method that combines many decision trees to make a prediction. In this project, you would train that model on lake samples and see how well it predicts chlorophyll-a in inland lakes.

Why This Is a Good Topic

This topic works well for science fair research because you can ask a focused question, find real public data, and test a clear prediction. It connects to water quality, public health, and climate-related changes in lakes. You can learn remote sensing, data cleaning, model training, and validation without needing to grow algae in a lab. A strong project here shows you can turn messy environmental data into a careful scientific comparison.

Research Questions

• How does Sentinel-2 surface reflectance predict chlorophyll-a in inland lakes across different seasons? ?
• What is the effect of lake size on random-forest chlorophyll-a prediction accuracy? ?
• Does adding red-edge bands improve chlorophyll-a retrieval compared with visible bands alone? ?
• To what extent do turbidity and chlorophyll-a together change model error in eutrophic lakes? ?
• Which Sentinel-2 band combinations produce the best chlorophyll-a estimates for lakes with low cloud cover? ?
• How does model performance change when you validate against EPA NLA samples from different regions? ?

Basic Materials

• Laptop with enough memory to handle satellite data.
• Internet access for downloading Sentinel-2 imagery and EPA NLA records.
• Spreadsheet software for cleaning sample metadata.
• Python or R for data analysis and modeling.
• GIS software such as QGIS for viewing lake masks and image footprints.
• External storage or cloud drive for large raster files.

Advanced Materials

• Access to university computing or a powerful workstation.
• Cloud-free Sentinel-2 surface-reflectance scenes.
• EPA National Lakes Assessment water-quality records.
• Lake boundary shapefiles or hydrography layers.
• Google Earth Engine or similar geospatial analysis platform.
• Python libraries for machine learning, geospatial processing, and plotting.

Software & Tools

• Python: Fits the random-forest model, cleans the data, and tests prediction accuracy.
• QGIS: Helps you inspect lake boundaries, satellite scenes, and sample locations.
• Google Earth Engine: Lets you filter Sentinel-2 images and build band features at scale.
• R: Runs statistical tests, plots model error, and compares lake groups.
• ImageJ: Can help if you need simple image-based checks on water color in sample scenes.

Experiment Steps

1. Define the lake group you want to study and set strict rules for cloud cover, season, and sample quality.
2. Match Sentinel-2 scenes to EPA NLA records so each image has a real chlorophyll-a target.
3. Choose the predictor bands and environmental variables you will test first.
4. Split your data into training and validation sets in a way that avoids leaking nearby lakes into both groups.
5. Build a baseline model, then compare it with a random-forest model and a simpler regression model.
6. Check where the model fails, then test whether those errors cluster in certain lake types or conditions.

Common Pitfalls

• Mixing satellite dates and field sample dates, which breaks the link between what the sensor saw and what the lake chemistry was.
• Using lakes with too much cloud cover or haze, which can distort surface reflectance values.
• Training and testing on samples from the same lake cluster, which makes the model look better than it really is.
• Ignoring shallow-water shoreline pixels, which can add land signal and corrupt chlorophyll estimates.
• Treating chlorophyll-a as the only bloom signal, which misses cases where turbidity or dissolved organic matter changes the reflectance pattern.

What Makes This Competitive

A competitive project does more than run a model. You need careful validation, a clear error analysis, and a reason for your design choices. Strong projects compare lakes by region, season, or water clarity, then explain where the method works and where it fails. You can also improve the work by testing whether one band set, one model type, or one validation strategy gives a better scientific result.

Project Variations

• Focus on one state or watershed and compare bloom prediction across lakes with different land-use patterns.
• Replace chlorophyll-a with Secchi depth or total suspended solids and test which water-quality target Sentinel-2 predicts best.
• Compare random forest with linear regression, support vector machines, or gradient boosting to see which model handles lake reflectance best.

Learn More

• NASA Earthdata: Search for Sentinel-2 guides, lake remote-sensing tutorials, and surface-reflectance documentation on Earthdata pages.
• EPA National Lakes Assessment: Find lake water-quality records, sampling methods, and metadata through EPA NLA reports and datasets.
• Sentinel-2 User Handbook: Read the mission documentation for band meanings, spatial resolution, and surface-reflectance products.
• USGS Water Data: Use USGS background material on lake and water-quality monitoring to understand field measurements and limits.
• PubMed: Search for review articles on remote sensing of chlorophyll-a and harmful algal blooms in inland waters.