It was a cold and icy winter in 2025-2026. Lake ice affects everything from snowfall and fishery populations to recreational activities and the multibillion-dollar commercial shipping industry, making Great Lakes ice cover data highly valuable. However, once winter arrives boats and buoys are removed from the water to protect them from ice, so the NOAA Great Lakes Environmental Research Laboratory (GLERL) monitors the Great Lakes by taking to the sky. GLERL has been using a technique called hyperspectral imagery for over a decade to monitor Great Lakes harmful algal blooms in the summer. This method, which uses cameras to collect wide band data from across the electromagnetic spectrum, can also be utilized to monitor lake ice.
One possible method is monitoring ice conditions from the sky, using a camera that collects a wide band of data from across the electromagnetic spectrum. By using a technique called hyperspectral imagery, GLERL has been collecting data over the Great Lakes for over a decade, frequently to monitor harmful algal blooms during the summer. Starting in 2024, our researchers began to explore flying a hyperspectral camera to monitor lake ice during the winter.
What is hyperspectral data?
Hyperspectral data or imagery contains much more information than a visible image. Traditional images capture what a human eye sees (visible light spectra), which consists of three bands: red, green and blue. The ‘hyper’ in hyperspectral means that the data collected captures parts of light beyond what our eyes can see, such as the near-infrared (thermal/heat range). Instead of three bands like our eyes or a traditional camera captures, hyperspectral cameras capture over 150 bands. How does all that extra information fit into an image? Hyperspectral data isn’t analyzed as a flat 2D image, but forms a 3D data cube. Different materials reflect and absorb light in a unique way, creating a spectral fingerprint similar to the uniqueness of a human fingerprint. By collecting hyperspectral data we are able to shift from looking at an object to identifying it by its spectral fingerprint. A spectral signature plot (or spectra plot) provides a graphical representation of a selected pixel in a hyperspectral image. The x-axis represents the wavelength in nanometers while the y-axis represents the intensity, also referred to as brightness, which is a measure of the amount of light being reflected at a particular wavelength. The blue, green and red vertical lines indicate the specific bands used to render the current hyperspectral image; in Figure 2, which shows different hyperspectral images, they were selected to mimic the standard red-green-blue spectrum seen with a human eye.
The first step in hyperspectral data collection begins with outfitting a small aircraft – a Cessna 210 in our case – with the hyperspectral camera (Figure 3). Once the camera is installed, the pilot flies a pre-defined flight path surveying an area of interest. Throughout our most recent ice flight mission, the aircraft maintained a survey altitude of 10,500 ft, capturing high resolution ice data.
Quick Shifts, Complex Conditions
The Great Lakes are known for their propensity to change rapidly. Many know the phrase “if you don’t like the weather, wait five minutes”, and the lakes are no different. Figure 5 below shows three visible webcam images from the same viewpoint highlighting the shift of Lake Erie ice sheets across a three hour timeframe. Winds and currents have a dramatic impact on how quickly the ice moves. This shows how monitoring, prediction, and reporting ice conditions can be challenging.
How is Great Lakes ice tracked and forecast now? A combination of numerical models and satellites that use radar, visible and infrared imagery determine ice conditions. Ice condition information is crucial for shipping in the Great Lakes, as commercial shipping traffic still transects the lakes throughout the winter months when conditions allow – even when ice is present. However, existing methods to monitor and track ice in the Great Lakes have limitations: satellites only provide periodic static images, and numerical models require high-fidelity data for validation. To address these gaps, GLERL is exploring hyperspectral imaging as a way to improve current monitoring systems and improve model predictions.
After hyperspectral data is collected it can be processed multiple ways. One of the products from these ice flights is true color imagery. Figure 4 below consists of four north-to-south passes over Lake Erie when covered with ice. At first glance one can observe a few different features by eye in the visible imagery: wind-blown snow, rough ice, possibly open water and a wispy cloud that created a slight shadow.
Looking Forward
In addition to using airplanes, we can also use aerial drones to capture hyperspectral data. To leverage this emerging technology, GLERL established an aerial drone program in 2024. This provides a significantly different perspective since aerial drones fly much lower to the ground, allowing them to capture much higher resolution images. Additionally, some drones are able to hover in one place while collecting data.
Collecting hyperspectral data from planes and drones is critical to understanding ice cover and ecology in the Great Lakes. By monitoring harmful algal blooms with hyperspectral imagery, we can safeguard drinking water intakes for millions of people, while also gaining insight into the dynamics of ecosystems in the lakes. Capturing hyperspectral ice cover data is fairly new but the potential payoff is big. Providing timely ice cover data at a resolution and accuracy never before available benefits all those that depend on the Great Lakes. Ice impacts everything from wildlife populations and health to the freighters that navigate the ice moving almost 200 million tons of materials each year, keeping our regional economy strong.
The ice features identified in Figure 4 are just the beginning of ice analysis based on hyperspectral data. More detailed analysis can confirm the ice classification (for example, determining different types or ice formation or thickness of ice) with more confidence and accuracy, thanks to the wealth of data that is embedded in a hyperspectral image. Advanced techniques are able to turn big datasets – including hyperspectral data – into valuable information. One technique is called deep learning, which allows an algorithm to study a big data set and then make predictions about that data. In our case, a deep learning algorithm studied 150 bands of an image and then described more about the ice types than we would be able to tell by looking at it with our red-green-blue detecting eyes. This takes trial and error and requires us to have the “ground truth” in order to train and check the algorithm. Finally, GLERL also conducts an analysis of how Great Lakes ice conditions progressed throughout the previous winter. Comparing the past winter’s conditions to historical trends, helps us understand the key factors that influence ice formation, how ice conditions have changed over time, and how we might predict future ice cover. With an increase in both the quality and quantity of ice condition data, GLERL predictions will continue to improve services for a broad range of Great Lakes stakeholders.
