Hyperspectral images reveal the health status of vegetation in Antarctica

An Australian study reveals the importance of aerial survey based on a more advanced technique to assess the health and density of vegetation in an area with limited spectral range such as Antarctica. The highest accuracy exceeds 99%

di Matteo Cavallito

 

The use of advanced technology and artificial intelligence now enables better detection of the health status of mosses and lichens in Antarctica. In other words, a more accurate assessment of the effects of climate change on the area’s only form of vegetation. This is the claim of researchers from the Queensland University of Technology (QUT) in Brisbane, Australia, in a study published in the journal Scientific Reports. The authors described the use of a new system for monitoring and classifying plant species. According to them, this approach has provided more accurate surveys, carried out more quickly and at lower cost compared to the past.

An indicator of Antarctica’s health

“Frost-tolerant vegetation like mosses and lichens in Antarctica are vital to biogeochemical cycles, soil insulation and support of biodiversity,” said Juan Sandino, researcher at the School of Electrical Engineering & Robotics at QUT and co-author of the study, underlining the role of mosses and lichens as indicators of climate stress. These species, he added, “drive nutrient cycles and underpin Antarctica’s ecosystems, yet they are the first to suffer from warming, extreme weather and human trampling.”

For this reason, in short, “Keeping track of their health is vital but extremely difficult in subzero field conditions.”

Observation activity typically makes use of drones, but there are some limitations. “Detecting moss and lichen using conventional red, green, blue (RGB) and multispectral sensors remains challenging,” the research explains. The study, consequently, investigated “the potential of hyperspectral imaging (HSI) for mapping cryptogamic vegetation and presents a workflow combining UAVs, ground observations, and machine learning (ML) classifiers.”

The system records hundreds of shades for each pixel

To overcome the difficulties of traditional models, researchers used a drone equipped with a camera that allows images to be collected in many different bands, identifying materials and vegetation based on their different abilities to reflect light. Specifically, the system records hundreds of shades for each pixel, combining them with a satellite navigation mechanism to accurately assign each image fragment to its exact location. Other high-resolution RGB photos were also added.

The data was used to train machine learning models which, according to the scientists, outperformed traditional metrics. “This new integrated system surpasses conventional digital images (red-green-blue or RGB) and also the satellite-based Normalised Different Vegetation Index (NDVI) that is being used to assess vegetation health and density,” Sandino points out. The results, the study says, “show that common indices are inadequate for moss and lichen detection, while novel spectral indices are more effective.”

An accuracy rate of 99%

In total, the researchers compared 12 different AI models for classifying vegetation: the best solutions achieved an accuracy rate of around 99%. “Full models achieved high performance, with CatBoost and UNet reaching 98.3% and 99.7% weighted average accuracy, respectively,” the study explains. “Light models using eight key wavelengths performed well, with CatBoost at 95.5% and UNet at 99.8%, demonstrating suitability for preliminary monitoring of moss health and lichen.”

Finally, the tests carried out by the researchers showed that flights at higher altitudes, between 30 and 70 meters, allow larger areas to be mapped to obtain regional overviews, while lower altitudes capture more precise details.

In short, the study confirmed the validity of the monitoring method for both small plots and larger areas. It also highlighted the importance of acquiring this type of image, especially in geographical regions with a reduced spectral range, such as Antarctica.